The Oxford Handbook of Ellipsis 0198712391, 9780198712398

Citation preview

Oxford Handbooks in Linguistics

Oxford Handbooks in Linguistics   Edited by Jeroen van Craenenbroeck and Tanja Temmerman The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019

(p. ii)

Oxford Handbooks in Linguistics

Recently published The Oxford Handbook of Inflection Edited by Matthew Baerman The Oxford Handbook of Historical Phonology Edited by Patrick Honeybone and Joseph Salmons The Oxford Handbook of Lexicography Edited by Philip Durkin The Oxford Handbook of Names and Naming Edited by Carole Hough The Oxford Handbook of Developmental Linguistics Edited by Jeffrey Lidz, William Snyder, and Joe Pater The Oxford Handbook of Information Structure Edited by Caroline Féry and Shinichiro Ishihara The Oxford Handbook of Modality and Mood Edited by Jan Nuyts and Johan van der Auwera The Oxford Handbook of Pragmatics Edited by Yan Huang

Page 1 of 2

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Oxford Handbooks in Linguistics The Oxford Handbook of Universal Grammar Edited by Ian Roberts The Oxford Handbook of Ergativity Edited by Jessica Coon, Diane Massam, and Lisa deMena Travis The Oxford Handbook of Polysynthesis Edited by Michael Fortescue, Marianne Mithun, and Nicholas Evans The Oxford Handbook of Evidentiality Edited by Alexandra Y. Aikhenvald The Oxford Handbook of Persian Linguistics Edited by Anousha Sedighi and Pouneh Shabani-Jadidi The Oxford Handbook of Lying Edited by Jörg Meibauer The Oxford Handbook of Taboo Words and Language Edited by Keith Allan The Oxford Handbook of Morphological Theory Edited by Jenny Audring and Francesca Masini The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman For a complete list of Oxford Handbooks in Linguistics please see pp. 1124–6.

Page 2 of 2

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Copyright Page

Copyright Page   Edited by Jeroen van Craenenbroeck and Tanja Temmerman The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019

(p. iv)

Copyright Page

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © editorial matter and organization Jeroen van Craenenbroeck and Tanja Temmer­ man 2019 © the chapters their several authors 2019 The moral rights of the authors have been asserted First Edition published in 2019 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press Page 1 of 2

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Copyright Page 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2018940964 ISBN 978–0–19–871239–8 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

Page 2 of 2

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

List of Figures and Tables

List of Figures and Tables   Edited by Jeroen van Craenenbroeck and Tanja Temmerman The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019

(p. ix)

List of Figures and Tables

Figures 7.1 Contrast (left) and elaboration (right) 166 9.1 Processing John upset Mary in DS 214 9.2 Unfolding structure for John upset… 217 9.3 Result of parsing John, who smokes, left 218 9.4 DS parsing context as a graph: Actions (edges) are transitions between partial trees (nodes) 220 9.5 Substitution from context at the ellipsis site of (24): Pronominal anaphora (top) and VP-ellipsis (bottom) 221 9.6 Action replay from context at the ellipsis site 222 9.7 A short answer with binding restrictions 224 9.8 Incremental development of Mary’s/Bob’s context via processing words 226 9.9 Processing Chorlton? in ‘A: the doctor B: Chorlton?’ 226 9.10 Incremental interpretation of self-repair by replaying DS actions in the Context DAG 229 9.11 Successful processing of John interviewed every student who Bill had 231 9.12 Ungrammaticality of (39) as impossibility to unify unfixed node with object of interview in second relative clause 231 15.1 Pitch extraction analysis of the VPE in (1): Distribution of pitch accents and prosodic boundaries 359 15.2 Pitch extraction analysis of gapping in (5): Parallel contrastive accents and prosodic boundaries 361 15.3 Pitch extraction analysis of (8c): Violation of pairwise contrast in gapping 363 15.4 Pitch extraction contour of RNR in (22): Licensing pitch accents and boundary tones 367 15.5 Pitch extraction contour of extraction from VPE 382 15.6 Pitch extraction contour of topicalization from VPE in a relative clause 385 17.1 Experimental stimuli 434 Page 1 of 3

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

List of Figures and Tables 20.1 A partial taxonomy of sluicing, based on the underlying syntax of the sluice 483 29.1 FinSL verbals with the meanings (from left to right) ‘know’ (the finger pads of the open hand touch the forehead twice), ‘teach [someone in front of the signer]’ (the two hands move forward twice in the shown configuration), and ‘an oblong vehicle (e.g. a bicycle) drives forward over a mound-like location’ (the dominant hand articu­ lates an arc-shaped movement over the stationary non-dominant hand). The verbals represent Type 1, Type 2, and Type 3, respectively (see also Figure 29.2) 767 29.2 FinSL Type 2 verbal TEACH as used in the elliptical clause (7). Note that the signer also employs constructed action to show the imaginary locations of the refer­ ents 775 29.3 Video frames showing the production of the sentence (17) 783 (p. x)

Tables 1.1 Overview of Part I of the handbook: Abstract structure, recoverability, and licens­ ing 8 1.2 Overview of Part IV of the handbook: Cross-linguistic distribution of the main el­ lipsis types 13 1.3 Overview of the cross-linguistic distribution of various subtypes of predicate el­ lipsis 14 2.1 Some previous research on the two ellipsis questions 22 4.1 A taxonomy for clarification requests (Purver 2006) 84 4.2 Dialogue gameboard 104 8.1 Commonly discussed constructions that involve ellipsis 189 8.2 Non-elliptical versions of the attested examples of ellipsis in Table 8.1 190 8.3 Less often discussed constructions that involve ellipsis, with examples and refer­ ences 198 12.1 Average percentages of ‘yes’ responses broken down over condition and lan­ guage from Wijnen et al. (2003) 308 17.1 Test conditions: VPE sentence + images 432 17.2 Control condition: Coordination + images 432 17.3 Experimental conditions 433 17.4 Patients’ demographic and clinical profiles 434 17.5 Number of accurate responses per condition for each patient on the VPE test 434 (p. xi) 17.6 Examples of spontaneous production of VPE by children 439 17.7 Test Condition: VPE + images 440 17.8 Control condition: Coordination + images 441 17.9 Experimental results 442 29.1 Ellipsis of A and P core arguments in the sample of 381 transitive clauses con­ taining a Type 1 or Type 2 verbal predicate 770

Page 2 of 3

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

List of Figures and Tables

Page 3 of 3

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors

The Contributors   Edited by Jeroen van Craenenbroeck and Tanja Temmerman The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019

(p. xii)

The Contributors

Klaus Abels

received his PhD from the University of Connecticut in 2003. He is a Reader in Lin­ guistics at University College London and co-editor of the journal Syntax. His main in­ terests relate to movement, constraints on movement, interactions of movement types, the formal modeling of movement, and the role of movement in deriving word order typology.

Lobke Aelbrecht

obtained her PhD at the Catholic University of Brussels in 2009 with a thesis entitled ‘You have the right to remain silent: The syntactic licensing of ellipsis’. Her main re­ search interests are ellipsis, VP topicalization and VP pronominalization, and the Dutch adpositional domain. In 2010, she published the monograph The syntactic li­ censing of ellipsis (John Benjamins).

Scott AnderBois

is Assistant Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. His primary research focus is on the ways in which utterances interact with the discourse “scoreboard,” with a particular focus on to what extent and in

Page 1 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors what ways these interactions are encoded conventionally as part of the sentence meanings, as opposed to arising from pragmatic reasoning. Specific topics of interest include apposition, discourse particles, disjunction, ellipsis, evidentiality, indefinite­ ness, mirativity, topics, and questions. He has explored these issues through primary fieldwork including Yucatec Maya, A’ingae, English, and Tagalog.

John Frederick Bailyn

is Professor of Linguistics at Stony Brook University, where he directs the Linguistics PhD program. His research interests include theoretical syntax, comparative Slavic syntax, binding, case, word order, scrambling, and ellipsis. His other interests include musical cognition and early Soviet history. He also co-directs the NY-St Petersburg In­ stitute in Linguistics, Cognition and Culture (NYI) held every July in St Petersburg, Russia.

Tatiana Bondarenko

received her Bachelor and Master degrees in Linguistics from Lomonosov Moscow State University, and is currently a PhD student in Linguistics at the Massachusetts Institute of Technology. Her research interests include syntax, semantics, morpholo­ gy, argument structure and event structure, ellipsis phenomena, embedded clauses (raising, control, restructuring), structures with dative and applicative arguments, as­ pectual systems, and fieldwork.

Norbert Corver

is the Chair-Professor of Dutch Linguistics in the Department of Languages, Litera­ ture and Communication of Utrecht University, and affiliated with the research insti­ tute UiL-OTS and the research group Language Structure: Variation and (p. xiii) Change. He received his PhD in Linguistics, entitled ‘The syntax of left branch extrac­ tions’, from Tilburg University in 1990. His main research interests are located in the areas of Dutch syntax, comparative syntax, and the interaction between language (morphosyntax) and affect. Specific topics he has been working on include displace­

Page 2 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors ment and locality, the morphosyntax of functional categories, the syntax of adverbs, NP-ellipsis, and the morphosyntactic encoding of affect.

Jeroen van Craenenbroeck

is Associate Professor of Dutch Linguistics at KU Leuven, where he is also Vice-Presi­ dent of the Center for Research in Syntax, Semantics, and Phonology (CRISSP). He is the author of The syntax of ellipsis (OUP) and general editor of the journal Linguistic Variation (John Benjamins). His research interests include ellipsis (sluicing, swiping, spading, VP-ellipsis), expletives, verb clusters, and the left periphery of the clause.

Peter W. Culicover

is Distinguished University Professor at the Ohio State University. He was awarded the Humboldt Research Award in 2006. His primary research has been in syntactic theory. He has been concerned with exploring the cognitive and computational fac­ tors that underlie the foundations of syntactic theory. Most recently he has been pur­ suing an evolutionary account of the origin of grammars from a constructional per­ spective.

Anne Dagnac

is Assistant Professor in French and Romance Linguistics at the University of Toulouse Jean Jaurès (France). Her research, which she conducts within CLLE (CN­ RS /University of Toulouse), focuses mainly on French syntax, and on the microsyn­ tactic variation of the underdescribed French Romance dialects, in particular Picard.

Marcela Depiante

received her PhD from the University of Connecticut in 2000. She is an Associate Pro­ fessor in the Department of Languages at the University of Wisconsin, Eau Claire. Her research interests include comparative syntax and morphology, in particular the

Page 3 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors syntax of ellipsis, as well as the grammar of Heritage Spanish Speakers and Spanish L1 attriters in the US.

Isabelle Deschamps

is a postdoctoral research scholar in the Department of Rehabilitation at Laval Uni­ versity. Previously she received her bachelor’s degree in Psychology and Linguistics as well as her PhD in Communication Sciences and Disorders from McGill University. Her research interests focus on issues pertaining to phonological processes during speech perception and production. In addition, her research aims to understand the relationship between phonological processes and other cognitive functions such as verbal working memory.

Arash Eshghi

is a researcher in Computational Linguistics at Heriot-Watt University. He did his PhD in Psycholinguistics at Queen Mary, University of London. Ever since, he has been one of the main developers of the Dynamic Syntax computational implementation, and is currently exploring its technological applications, e.g., in building more human-like conversational systems.

(p. xiv)

Catherine Fortin

is Associate Professor of Linguistics at Carleton College. She received her PhD in Lin­ guistics from the University of Michigan in 2007. Her primary research interests are the syntax and morphosyntax of Indonesian and Minangkabau, including ellipsis phe­ nomena, clause structure, and argument structure.

Lyn Frazier

is a Professor in the Department of Linguistics at the University of Massachusetts Amherst. Specializing in psycholinguistics, her research spans a range of topics in

Page 4 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors syntactic processing, primarily concerning phrase structure parsing and the parsing of movement dependencies, and issues at the syntax–discourse interface. The latter include research on the role of prosody in sentence processing, ellipsis, processing of not-at-issue content, and the role of implicit and explicit Questions-Under-Discussion (QUDs) in organizing discourse.

Teruhiko Fukaya

is Professor in the Faculty of International Communication at Gunma Prefectural Women’s University in Japan. He is interested in the investigation of the language fac­ ulty through the studies of ellipsis phenomena, such as sluicing, stripping, and frag­ ment answers, in Japanese and English.

Kenshi Funakoshi

is Lecturer in Linguistics in the Department of English at Dokkyo University. He spe­ cializes in generative syntax with an emphasis on ellipsis, verb movement in SOV lan­ guages, and Japanese. His publications include ‘On headless XP-ellipsis/move­ ment’ (2012), ‘Verb-stranding verb phrase ellipsis in Japanese’ (2016), and ‘Backward control from possessors’ (2017).

Jonathan Ginzburg

is Professor of Linguistics at Université Paris-Diderot (Paris 7). He has held appoint­ ments at the Hebrew University of Jerusalem and King’s College, London. He is one of the founders and editor-in-chief (emeritus) of the journal Dialogue and Discourse. His research interests include semantics, dialogue, language acquisition, and musical meaning. He is the author of Interrogative investigations (CSLI Publications, 2001, with Ivan A. Sag) and The interactive stance: Meaning for conversation (Oxford Uni­ versity Press, 2012).

Adele E. Goldberg

Page 5 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors

is a Professor of Psychology at Princeton University where she is also affiliated with the linguistics and cognitive science programs. Goldberg’s work focuses on the psy­ chology of language, particularly on how grammatical constructions are represented, learned, and processed. A more specific interest is in the functions of constructions and how those functions explain facts that are often assumed to be purely syntactic.

Kay González-Vilbazo

is Associate Professor at the University of Illinois at Chicago and Co-Director of the UIC Bilingualism Research Lab. His research focuses on linguistic-theoretical aspects of bilingual phenomena, mostly the grammar of codeswitching. Current projects of his include gender agreement and concord, phase theory, wh-dependencies, ellipsis, pro-drop, case theory, the structure of PF, the phonology of codeswitching, and the theory of the bilingual lexicon.

Eleni Gregoromichelaki

is a Research Fellow at the Philosophy Department, King’s College London and the Cognitive Science Institute, Osnabrueck University. She works within the Dynamic Syntax and Computational Linguistics research groups, exploring analyses of (p. xv) syntactic/semantic natural language phenomena within psycholinguistically informed formalisms. Her principal research interests lie in the language–cognition interface: in particular, the formal/computational and psychological/philosophical implications of various cognitive modeling perspectives on natural language.

Yosef Grodzinsky

is Professor of Neurolinguistics at the Edmond and Lily Safra Center for Brain Re­ search, the Hebrew University of Jerusalem, and a Visiting Professor at the Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich. His research has focused on the neural basis, acquisition, and processing of syntactic and semantic knowledge. At present, his work explores the neural bases of overt and covert nega­ tion. Previously, Grodzinsky was a Professor and Tier-I Canada Research Chair of

Page 6 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors Neurolinguistics at McGill University, a Professor of Psychology at Tel Aviv University, and a Research Professor of Neurology at the Boston University School of Medicine. He is the recipient of several awards, and his research has been funded by govern­ ment agencies in the US, Canada, Israel, and Germany.

Alison Hall

is a Lecturer in English Language at De Montfort University, Leicester, UK. She re­ ceived her PhD in Pragmatics from University College London and has been a post­ doctoral researcher at UCL and at Institut Jean Nicod, Paris. She has published on linguistic underdeterminacy, lexical pragmatics, and the debate between contextual­ ism, indexicalism, and semantic minimalism.

Daniel Hardt

is Associate Professor in the Department of Management, Society, and Communica­ tion at Copenhagen Business School, and is Visiting Research Associate at the Univer­ sity of California, Santa Cruz. He holds a PhD in Computer Science from the Universi­ ty of Pennsylvania. His research deals with both theoretical and computational lin­ guistics, with a particular interest in ellipsis and other matters involving semantics, anaphora, and discourse. He has published articles on ellipsis in journals such as Lin­ guistics and Philosophy, Journal of Semantics, and Computational Linguistics.

William Harwood

obtained his PhD at Ghent University in 2013 with a thesis entitled ‘Being progres­ sive is just a phase: Dividing the functional hierarchy’. His main research interests in­ clude ellipsis, phase theory, idiomatic expressions, auxiliary verbs, aspect, verb move­ ment, VP fronting, existential constructions, and relative clauses.

Julian Hough

Page 7 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors is a Lecturer in the Cognitive Science group in the School of Electronic Engineering and Computer Science at Queen Mary, University of London (QMUL). He received his PhD at QMUL before working at Bielefeld University as a post-doc. He researches di­ alogue modelling and dialogue systems, with a focus on incremental processing and disfluency.

Ray Jackendoff

is Seth Merrin Professor Emeritus at Tufts University and a Research Associate at the MIT Department of Brain and Cognitive Sciences. He was the 2014 recipient of the David Rumelhart Prize in Cognitive Science. His principal research goal at present is the Parallel Architecture, a theory of linguistic structure that incorporates (p. xvi) se­ mantics, syntax, morphology, and phonology, and that integrates language with the rest of the mind.

Pauline Jacobson

is Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. Her research centers on formal semantics and its interaction with syntax, and she is the author of the semantics textbook Compositional semantics: An introduction to the syntax/semantics interface (Oxford University Press, 2014). Her research centers on the hypothesis of Direct Compositionality and on the related hypothesis that the se­ mantics makes no use of variables. Her research program applies these hypotheses to a rich set of natural language phenomena.

Tommi Jantunen

holds an MA in General Linguistics from the University of Helsinki, Finland, and a PhD and a degree of Docent (Adjunct Professor) in Finnish Sign Language (FinSL) from the University of Jyväskylä (JyU), Finland. He is currently affiliated as an Acade­ my Research Fellow at the Sign Language Centre in JyU, Department of Languages, and in his research he investigates FinSL grammar and phonetics as well as sign lan­ guage technology.

Page 8 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors Kyle Johnson

is Professor in the Linguistics Department at the University of Massachusetts Amherst. His research focuses on the connection between syntax and semantics.

Andrew Kehler

is a Professor in the Department of Linguistics at the University of California, San Diego. His primary research foci are discourse interpretation and pragmatics, studied from the perspectives of theoretical linguistics, psycholinguistics, and computational linguistics. His publications include Coherence, reference, and the theory of grammar (2002) and numerous articles on topics such as ellipsis, discourse anaphora, and dis­ course coherence.

Ruth Kempson

FBA is Emeritus Professor of Linguistics at King’s College London and Honorary Re­ search Associate with SOAS and the Cognitive Science research unit of QMUL, Lon­ don. She is the lead developer of the Dynamic Syntax framework, and has a long-term research interest in the interface of syntax and pragmatics.

Marjo van Koppen

is Professor in Dutch Variation Linguistics in the Department of Languages, Litera­ ture, and Communication at Utrecht University, and senior researcher at the Meertens Institute in Amsterdam. She received her PhD, entitled ‘One probe, two goals: Aspects of agreement in Dutch dialectics’, from Leiden University in 2005. Her main research interest is the morphosyntactic variation within Dutch dialects and the older stages of Dutch. The theoretical framework of her research is generative syn­ tax.

Page 9 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors Howard Lasnik

is Distinguished University Professor in the Department of Linguistics at the Universi­ ty of Maryland. He specializes in generative syntax and the formalization of syntactic theories. Among the specific topics he has worked on are phrase structure, anaphora, ellipsis, verbal morphology, case, and locality constraints on movement. His publica­ tions include eight books and over a hundred articles.

(p. xvii)

Winfried Lechner

is Associate Professor for German Linguistics and Theoretical Linguistics at the Na­ tional and Kapodistrian University of Athens, Greece. His main academic interests are located in the areas of syntax, semantics, and the interaction between these two com­ ponents. He has been working on the logical syntax of scope and reconstruction, re­ flexivization, comparatives, ellipsis, the cross-linguistic typology of same/different, ad­ ditive and scalar focus particles, Duke of York opacity, and the architecture of the grammar.

Anikó Lipták

is an Assistant Professor at the Leiden University Centre for Linguistics (LUCL), Lei­ den University. Her main field of research is comparative syntax and Hungarian, and she has published extensively on elliptical phenomena. She is currently researching issues concerning ellipsis identity, the interaction between ellipsis and morphology, and between ellipsis and intonation.

Sophie Manus

is an Associate Professor in Lyon, France, at the Laboratoire Dynamique du Langage, CNRS/Université Lyon 2. She completed her PhD in Linguistics at the Institut Nation­ al des Langues et Civilisations Orientales in Paris. Since then, she has studied, taught, and directed research in Bantu languages, tone, morphology, and fieldwork

Page 10 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors methods applied to underdescribed/endangered languages, and she has more recent­ ly started working on tonal Chibchan languages spoken in Central America.

Jason Merchant

is the Lorna P. Straus Professor of Linguistics and Vice Provost at the University of Chicago. He is the author of The syntax of silence, the co-editor of Sluicing: Cross-lin­ guistic explorations, and the author or co-author of more than two dozen articles on a wide variety of elliptical phenomena.

Philip Miller

teaches English and General Linguistics at the Université Paris Diderot (Paris 7). He was recently Visiting Professor at the Universidade de São Paulo. He is the author of two monographs, Clitics and constituents in phrase structure grammar (Garland Pub­ lications, 1992) and Strong generative capacity (CSLI Publications, 1999). He has worked on clitics and on perception verbs. His current work is centered on ellipsis and anaphora, with specific interest in verbal ellipsis and verbal anaphora.

Joanna Nykiel

is Visiting Professor in the Department of English Language and Literature at Kyung Hee University, Seoul. Her research interests center on elliptical constructions, syn­ tactic variation and modern quantitative methods of data analysis, the history of the English language, and, most recently, language processing. She has published arti­ cles in English Language and Linguistics, Language Variation and Change, and Lin­ gua, among others. She is currently preparing a volume called Syntactic variation for publication in the Cambridge University Press series Key Topics in Syntax.

Timothy Osborne

Page 11 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors is an Assistant Professor of Linguistics at Zhejiang University in Hangzhou, China. His research focus is on the theory of syntax, especially on the notions of constituen­ cy and dependency in syntactic analysis. Particular phenomena of syntax that he has explored are diagnostics for constituents, coordination, comparatives, ellipsis, and idiosyncratic meaning.

(p. xviii)

Cédric Patin

completed his PhD in Linguistics at Université Paris 3 in 2007. His thesis examined the tonal system of the Bantu language Shingazidja. After a postdoctorate at the Lab­ oratoire de Linguistique Formelle (CNRS/Université Paris 7), he accepted the position of Maître de conférences en phonétique et phonologie du français at the University of Lille in 2009. His work focuses on the phonology of Bantu languages, with emphasis on the prosody–syntax interface.

Florent Perek

is a Lecturer in Cognitive Linguistics at the University of Birmingham in the Depart­ ment of English Language and Applied Linguistics. His main research interests lie in the study of grammar from a cognitive and corpus linguistic perspective, with a par­ ticular focus on how syntactic constructions are mentally represented, how they are learned, and how they change over time.

Sergio E. Ramos

is a graduate student at the University of Illinois at Chicago. His main research inter­ ests lie at the intersection of bilingualism and linguistic theory. These include codeswitching, identity, and second language acquisition as well as topics of linguistic theory such as nominal ellipsis, sluicing, case, and Romance linguistics more broadly.

Tom Roeper

Page 12 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors works in generative grammar on acquisition theory and experimentation and on syn­ tactic morphology. Acquisition topics include long-distance movement, quantification, binding theory, passives, V2, small clauses, aspect, ellipsis, implicatures, and theoreti­ cal work on Multiple Grammars, Strict Interfaces, Subset Theory, and African-Ameri­ can English dialect. In morphology he has worked on compounds, gerunds, produc­ tive affixation, and theoretically on lexical transformations, implicit arguments, and labeling theory. In addition, he has written two popular books, most recently The prism of grammar, five co-edited books, including Recursion: Complexity in cognition; co-authored an assessment test, Diagnostic evaluation of language disorders; and coedited Language acquisition and studies in theoretical psycholinguistics. Currently he is working with collaborators on the emergence of recursive self-embedding and its theoretical implications in English, German, Dutch, Japanese, Romanian, Hungarian, and Pirahã.

Andrés Saab

studied Literature and Linguistics at the Universidad de Buenos Aires and at the Uni­ versidad Nacional del Comahue (Argentina). In 2009, he defended his doctoral disser­ tation on the theory of ellipsis. Currently, he is Associate Researcher at the Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET) and Associate Profes­ sor at the University of Buenos Aires. His main research areas are ellipsis, copy theo­ ry of movement, null subjects and, more broadly, the syntax–interface connection.

Lewis P. Shapiro

is an Emeritus Professor at San Diego State University. Research interests include charting the moment-by-moment unfolding of language and cognitive processing in neurologically healthy adults and those with brain damage; brain–language relations through lesion analyses and brain imaging; and the efficacy and neurological implica­ tions of treatment for adults with language disorders. Dr Shapiro’s work has been funded continuously through the US National Institutes of Health since 1988.

(p. xix)

Tanja Temmerman

Page 13 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors is Assistant Professor of Dutch Linguistics at Université Saint-Louis—Bruxelles (Bel­ gium), where she is also Lecturer in English Language and Head of the English De­ partment. She obtained her PhD from Leiden University in 2012 with a dissertation entitled ‘Multidominance, ellipsis, and quantifier scope’. Her principal research foci lie in (generative) syntax, issues at the syntax–phonology and syntax–semantics inter­ faces, Dutch dialectology, and comparative Germanic syntax. Specific topics of inter­ est include ellipsis, the internal and external syntax of idioms, phase theory, long-dis­ tance dependencies, island effects, phrase structure, modals, and negation.

Gary Thoms

is Assistant Professor of Linguistics at New York University. His main research inter­ ests include ellipsis, reconstruction, dialectal and intraspeaker variation in English, Celtic syntax, polarity phenomena, predicate fronting, and the language of poetry.

Maziar Toosarvandani

is Associate Professor in the Department of Linguistics at the University of California, Santa Cruz. His primary research interests lie in syntax and semantics, primarily in Northern Paiute, Persian, and Zapotec. He has published in Language, Linguistic In­ quiry, Natural Language and Linguistic Theory, Semantics and Pragmatics, and the International Journal of American Linguistics.

Luis Vicente

was a lecturer and researcher in the Department of Linguistics at the University of Potsdam, Germany. He earned a BA in English Philology from the Universidad de Deusto, Spain; a PhD in Linguistics from Leiden University, the Netherlands; and af­ ter a short lecture engagement at the University of Amsterdam, he was a postdoctor­ al researcher at the University of California Santa Cruz, supported by a prestigious scholarship from the Basque Government. He successfully obtained his Habilitation in July 2016 at the University of Potsdam. He was a prolific researcher who published a large number of important and influential papers on various phenomena, mainly fo­ cusing on the interaction between syntax and semantics. He passed away on 6 Febru­

Page 14 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

The Contributors ary 2018 at the age of 38 in Freiburg im Breisgau, Germany. His absence is a great loss to academia in general and linguistics in particular. He is sorely missed by us all.

Chris Wilder

is Professor of English Linguistics at the Norwegian University of Science and Tech­ nology in Trondheim, where he has been employed since 2005. Prior to that he spent fifteen years as a researcher and lecturer in Germany and the USA. His research in­ terests include English and German syntax, comparative syntax in general, and ellip­ sis and constituent-sharing phenomena in particular.

Susanne Winkler

is Professor of English Linguistics at the University of Tübingen (Germany). She has a long-standing research interest in syntactic theory, information structure, and the syntax–prosody interface. She has written extensively on the information structure of elliptical constructions and focus constructions. She is the author of Ellipsis and focus in generative grammar (Mouton de Gruyter, 2005) and Focus and secondary predica­ tion (Mouton de Gruyter, 1997), and of papers in a variety of volumes and journals. She directs a DFG-funded research project on Focus and Extraction in Complex Con­ structions and (p. xx) Islands (SFB 833) and co-directs an interdisciplinary DFG-fund­ ed research training group (RTG 1808) on Ambiguity: Production and Preception.

Masaya Yoshida

is an Associate Professor in the Department of Linguistics at Northwestern Universi­ ty. His research interests are online sentence processing and syntax with special fo­ cus on the syntax and processing of ellipsis. He has published articles in a number of linguistics and psycholinguistics journals such as Linguistic Inquiry, Natural Lan­ guage and Linguistic Theory, and Journal of Language, Cognition, and Neuroscience.

Page 15 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Ellipsis In Natural Language: Theoretical and empirical perspectives

Oxford Handbooks Online Ellipsis In Natural Language: Theoretical and empirical perspectives   Jeroen van Craenenbroeck and Tanja Temmerman The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.1

Abstract and Keywords This chapter introduces the general topic of the handbook: the linguistic study of elliptical phenomena in natural language. In addition, it outlines the general structure of the handbook, with a subdivision into four parts (the theory of ellipsis; ellipsis as a diagnostic tool; the taxonomy of elliptical constructions; case studies exploring the elliptical inventory of a single language), and explains the philosophy behind and rationale for this structure. Finally, the chapter also highlights (a selection of) the main results and generalizations that emerge from contributions to the handbook, and briefly reflects on the future of linguistic research into ellipsis. Keywords: ellipsis, structure, recoverability, licensing, ellipsis as a diagnostic, elliptical constructions, crosslinguistic variation

Jeroen van Craenenbroeck and Tanja Temmerman

1.1 Introduction NATURAL language abounds in elliptical expressions, i.e. expressions that seem to leave certain aspects of their meaning unexpressed. Consider a random sample in (1): (1)

Page 1 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

What all of these examples have in common is the fact that they feel incomplete in some sense. For instance, the intended meaning of (1a) is ‘Ed invited someone to his furniture shop, but I don’t know who Ed invited to his furniture shop’, but the final portion of this sentence is missing. Similarly, even though a surgeon saying (1b) only utters the noun scalpel, she conveys a directive to the effect that she be handed a scalpel. Similar observations can be made for the data in (1c)–(1f). This discrepancy between what is overtly expressed and what is intended poses great challenges for theories of sound–meaning correspondence. According to the principle of compositionality usually attributed to Frege, the meaning of a complex utterance is a function of the meaning of its subparts and the way they are combined. In ellipsis, this principle appears to break down. For instance, the meaning of the surgeon’s utterance in (1b) cannot simply be said to be a function of its subparts: there is meaning, but there is no corresponding sound. Given that ellipsis raises such fundamental questions about language (p. 2) in particular and cognition in general, it should come as no surprise that it has garnered considerable linguistic and philosophical interest over the years. At the same time, the examples listed in (1) already suggest that ellipsis is by no means a unified phenomenon. To illustrate, while constructions such as those in (1a), (1d), and (1f) can occur fairly freely in both written and spoken discourse, the remaining elliptical expressions are highly context- and/or register-dependent. For instance, an example like (1c), in which the subject of the sentence (typically ‘I’) is left unexpressed, is only allowed in very specific registers of English, of which diary entries are a prime example. Moreover, variation in elliptical constructions is not only attested within but also across languages. Consider as a case in point the English elliptical example in (2)—where the verbal predicate read War and Peace is missing from the second clause—and its close correlates from Dutch, French, and German in (3a), (3b), and (3c), respectively. (2)

(3)

Page 2 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

While (2) is perfectly acceptable in English, attempts at a word-for-word translation of this elliptical sentence in closely related languages lead to ungrammaticality, showing how ellipsis is subject to cross-linguistic variation. It is observations such as these that have led to very detailed investigations of specific elliptical phenomena in the linguistic literature in recent years. Summing up, ellipsis is a topic that on the one hand raises general and fundamental questions about the workings of grammar and cognition, while on the other it is a veritable treasure trove of detailed and fine-grained points of inter- and intralinguistic variation. It is against this dual backdrop that the current Oxford Handbook of Ellipsis should be situated. As we make clear in the remainder of this introductory chapter, the handbook devotes attention both to fundamental theoretical questions and analyses surrounding elliptical phenomena, and to the empirical richness of this domain. This chapter is organized as follows. We first outline the general structure of the handbook and elaborate on (the rationale behind) its subdivision into four parts (section 1.2). Next, in section 1.3, we highlight some of the main results and generalizations that emerge from the contributions to the handbook, and in section 1.4 we conclude and provide a brief outlook on future research on ellipsis.

1.2 Structure of the handbook This handbook is subdivided into four parts, each of which highlights a specific aspect of the linguistic study of ellipsis. The first part (discussed in subsection 1.2.1) focuses on the theory of ellipsis, and explores the analytical approach taken towards ellipsis both in (p. 3) various linguistic theoretical frameworks and in a number of subfields of linguistics. In the second part (subsection 1.2.2) the perspective is reversed, and ellipsis is construed not as the object but as the instrument of inquiry. The central question in the chapters of Part II is to what extent the study of ellipsis can shed new light on other research domains within linguistics. Part III (subsection 1.2.3) focuses on the traditional taxonomy of elliptical constructions known from the literature and explores the state of the art for each of them. Finally, Part IV (subsection 1.2.4) contains eleven case studies, each of which explores the elliptical inventory of a single language (or a set of closely related languages or language varieties), thus bearing witness to the empirical richness surrounding the phenomenon of ellipsis.

Page 3 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

1.2.1 Part I: The theory of ellipsis The chapters making up the first part of the handbook can be divided into three sets. The first is the singleton consisting of the contribution by Jason Merchant entitled “Ellipsis: A survey of analytical approaches” (Chapter 2). This chapter lays the groundwork for the rest of Part I: it defines the phenomena under investigation, lays out the central research questions, presents a taxonomy of approaches based on how they address those research questions, and weighs some of the evidence presented in favor of and against the various perspectives. The second set consists of Chapters 3–10. These eight chapters discuss the analytical approach taken towards ellipsis in a specific theoretical framework. The frameworks in question are Transformational Grammar (Chapter 3, Howard Lasnik and Kenshi Funakoshi), Head-Driven Phrase Structure Grammar (Chapter 4, Jonathan Ginzburg and Philip Miller), Categorial Grammar (Chapter 5, Pauline Jacobson), Dependency Grammar (Chapter 6, Timothy Osborne), Simpler Syntax (Chapter 7, Peter W. Culicover and Ray Jackendoff), Construction Grammar (Chapter 8, Adele E. Goldberg and Florent Perek), Dynamic Syntax (Chapter 9, Ruth Kempson, Eleni Gregoromichelaki, Arash Eshghi, and Julian Hough), and Inquisitive Semantics (Chapter 10, Scott AnderBois). The third and final set of Part I is composed of Chapters 11–15. These explore the theory of ellipsis not from the point of view of a linguistic framework, but with respect to a specific subfield or subdiscipline of linguistics. The areas covered in these five chapters are psycholinguistics (Chapter 11, Lyn Frazier), acquisition (Chapter 12, Tom Roeper), discourse (Chapter 13, Andrew Kehler), computational linguistics (Chapter 14, Daniel Hardt), and prosody (Chapter 15, Susanne Winkler). While ellipsis frequently surfaces as a topic of investigation in all of these frameworks and subfields of linguistics, only very rarely does this lead to an explicit comparison or evaluation of the various assumptions, arguments, and analyses. The current handbook wants to remedy this, and to this end we asked all authors of the chapters in Part I to focus on the same three theoretical issues. On the one hand, this ensures a high degree of thematic consistency across these chapters, while on the other it allows for a direct form of inter-chapter comparison. The three issues under investigation are (i) the abstract structure of the ellipsis site, (ii) recoverability/ellipsis identity, and (iii) licensing. We now briefly introduce each of these topics (see also Chapter 2 for further, more detailed, discussion). The conundrum regarding the sound–meaning correspondence in ellipsis raised earlier (see section 1.1) can be paraphrased as a tension between semantics (meaning) on the one hand and phonology (sound) on the other. A central question in the study of ellipsis concerns the role of (p. 4) syntax in this dichotomy: to what extent does an ellipsis site contain (unpronounced) syntactic structure? This is an issue of great contention in the literature: hypotheses range from the position that there is no (‘hidden’) structure whatsoever (e.g. Culicover and Jackendoff 2005) all the way to the other extreme, i.e. that there is full-fledged syntactic structure in an ellipsis site (e.g. Merchant 2001). From the latter point of view, the only difference between an elliptical sentence and a non-elliptical one is the lack of pronunciation of part of the former. Given that different theoretical Page 4 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives frameworks and subdisciplines take radically opposing positions in this debate, the topic of “abstract structure of the ellipsis site” is ideally suited for comparison and evaluation across frameworks and subdisciplines. The second recurring topic in all chapters of Part I is recoverability, which concerns the question of how an ellipsis site gets its meaning, or more specifically, ellipsis identity, which concerns the question of the identity relation between an ellipsis site and its antecedent. Consider again the example in (1a), repeated below as (4). It is clear that we interpret the missing part of this sentence as ‘Ed invited to his furniture shop’ because the first half of this example contains exactly these words. Put differently, the meaning of an ellipsis site is recovered by virtue of an antecedent, with which it stands in a certain identity relation. However, the question of whether this identity relation is syntactic, semantic, morpholexical, pragmatic, etc., is far from settled, and one could even point to examples such as (1b), (1c), or (1e) (repeated below as (5a), (5b), and (5c), respectively) to question the very assumption that an antecedent is required in the first place. Once again, the literature on ellipsis contains many different answers to these questions, and the first part of the handbook provides a clear picture of the various arguments and positions. (4)

(5)

The third central theme for the chapters of Part I is licensing, a cover term referring to restrictions on ellipsis (typically syntactic in nature) that are not related to recoverability or ellipsis identity. Consider for instance the following ungrammatical English example: (6)

In the second part of this sentence the noun bike is left unpronounced. Although it is abundantly clear from the context what the sentence should mean—i.e. there is no problem of recoverability—the instance of ellipsis illustrated here is ruled out in English. As shown in (7), though, the Dutch analogue of (6) is perfectly acceptable. This shows that there are restrictions on ellipsis (within and across languages) above and beyond those related to recoverability or ellipsis identity. These restrictions typically go under the rubric of ‘licensing’ in the ellipsis literature. (7) Page 5 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

Although the issue of licensing should be addressed in any comprehensive theory of ellipsis, only rarely does it explicitly feature in the discussion (notable exceptions are Lobeck 1995 and Aelbrecht 2010). By adding licensing to the list of topics that every chapter in Part I addresses, the handbook aspires to put this theoretical notion firmly on the research agenda. (p. 5)

1.2.2 Part II: Ellipsis as a diagnostic tool The study of ellipsis is not only interesting in and of itself. Anyone who has ever taken an intro class in general linguistics, more specifically in constituency, knows that elliptical constructions can be used as a diagnostic tool to answer non-ellipsis-related research questions (in this simple case: the question of whether or not a string of words forms a constituent). Part II of the handbook takes precisely this reversed perspective: ellipsis is now no longer (or at least not exclusively) the object of study, but rather the means or the tool through which the study is carried out. The topics under investigation in this manner are movement and islands (Chapter 16, Klaus Abels), aphasia and acquisition (Chapter 17, Yosef Grodzinsky, Isabelle Deschamps, and Lewis P. Shapiro), parsing strategies (Chapter 18, Masaya Yoshida), and codeswitching (Chapter 19, Kay González-Vilbazo and Sergio E. Ramos). Each of these topics represents an active research area where elliptical phenomena have the potential of shedding important new light on the central research questions. For instance, certain types of ellipsis seem to bleed island effects. In Transformational Grammar, one of the central issues surrounding islands is whether these phenomena should be located at the conceptual-intentional or the articulatory-perceptual interface of the language module. The bleeding of island effects by ellipsis has been taken as an argument in favor of the latter position. Processing-based accounts, on the other hand, use island effects in ellipsis to examine the interplay between the syntactic and the discourse processor. Based on the lack of island effects in certain elliptical sentences, it is concluded that only the latter type of processor may violate islands. Similarly, the ability of aphasia patients to comprehend elliptical, i.e. incomplete, sentences can lead to a deeper understanding of the nature of their language deficit. These two examples are representative of the line of thinking that is developed in the chapters of Part II: in each case, the central research question does not concern ellipsis per se, but the study of elliptical phenomena brings us closer to answering that question.

1.2.3 Part III: Elliptical constructions The third part of the handbook is devoted to detailed studies of specific elliptical constructions. Generally speaking, ellipses appear to cluster at the clausal, predicate, and nominal level, corresponding to, respectively, clausal ellipsis, predicate ellipsis, and Page 6 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives nominal ellipsis. Part III of the handbook starts off from this trichotomy, discussing first the most well-known and best-investigated representative of clausal ellipsis, i.e. sluicing (Chapter 20, Luis Vicente), illustrated in (8), followed by predicate ellipsis (Chapter 21, Lobke Aelbrecht and William Harwood), (9), and nominal ellipsis (Chapter 22, Andrés Saab), (10). (p. 6)

(8)

(9)

(10)

The remaining five chapters of this part of the handbook cover the following elliptical constructions: gapping and stripping (Chapter 23, Kyle Johnson), fragments (Chapter 24, Alison Hall), comparative deletion (Chapter 25, Winfried Lechner), Null Complement Anaphora (Chapter 26, Marcela Depiante), and Conjunction Reduction and Right-Node Raising (Chapter 27, Chris Wilder). Each chapter thus zooms in on one particular subtype—or very often, family of subtypes— of ellipsis, and provides a systematic and detailed overview of its basic properties and distinctive characteristics. In addition to covering the central empirical generalizations, each chapter presents a survey of the main theoretical concerns these elliptical constructions raise and discusses different analytical approaches to them. As such, this part of the handbook is meant to serve as a reference work for anyone interested in a particular subtype of ellipsis.

1.2.4 Part IV: Case studies The fourth and final part of the handbook presents case studies of ellipsis in specific languages. The languages under investigation are: Dutch (Chapter 28, Norbert Corver Page 7 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives and Marjo van Koppen), Finnish Sign Language (Chapter 29, Tommi Jantunen), French (Chapter 30, Anne Dagnac), Hungarian (Chapter 31, Anikó Lipták), Indonesian (Chapter 32, Catherine Fortin), Japanese (Chapter 33, Teruhiko Fukaya), Kiswahili and Shingazidja (Chapter 34, Cédric Patin and Sophie Manus), Persian (Chapter 35, Maziar Toosarvandani), Polish (Chapter 36, Joanna Nykiel), Russian (Chapter 37, John Frederick Bailyn and Tatiana Bondarenko), and varieties of English (Chapter 38, Gary Thoms). These languages were selected on the basis of two criteria: (a) the fact that they exhibit elliptical phenomena that were previously unattested and/or that shed new light on some of the more mainstream generalizations and theories, and (to a lesser extent) (b) typological spread. All of the chapters in Part IV have a double orientation. Firstly, they inventory which of the elliptical constructions discussed in Part III are attested in the language and which aren’t, and they describe their properties. This provides typological depth to the chapters in Part III. Moreover, this offers ample opportunities for inter-chapter comparison in Part IV: (p. 7) these language-specific studies contribute meaningfully to our understanding of the ways in which the elliptical phenomena under scrutiny manifest themselves crosslinguistically, and they open the door to formulating a more systematic cross-linguistic theory of the distribution of ellipsis types. Secondly, the chapters in Part IV focus on facts and observations that are new, previously undiscussed, and/or off the beaten track, and that therefore question or put in a different light the hypotheses and theories based on better-known languages. To name but a few concrete examples: (i) Chapter 29 considers the role that gesture and mime play in elliptical phenomena in Finnish Sign Language; (ii) Chapter 33 shows that in many Japanese ellipsis phenomena (including sluicing, fragments, and stripping), the presence or absence of a case marker on the ellipsis remnant plays a crucial role, with casemarked and non-case-marked fragments being analyzed as instances of surface and deep anaphora, respectively; and (iii) Chapter 32 considers in detail the fact that Indonesian permits prepositions to be omitted in certain elliptical contexts, despite preposition stranding being otherwise prohibited in the language. This is unexpected in light of the cross-linguistically robust generalization that preposition omission under ellipsis tracks preposition stranding in non-elliptical contexts (Merchant 2001).

1.3 Results and Generalizations In this section we highlight some of the main results and generalizations that emerge from the contributions to this handbook. Needless to say, it is neither realistic nor feasible to do full justice to a 1000+ page volume in an introductory chapter, so we needed to be selective in what we present in this section. The common thread throughout the discussion is the question to what extent the structure and goals set out for the handbook (as described in section 1.2) have yielded interesting results and generalizations. Accordingly, this section is structured parallel to the preceding one, with one subsection devoted to each part of the handbook. Page 8 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

1.3.1 Part I: The theory of ellipsis Given that the chapters in Part I all address the same three theoretical questions (abstract structure, recoverability/ellipsis identity, and licensing; see subsection 1.2.1), it speaks to reason to use these three questions as the structuring principle for a summarizing overview of those chapters. Table 1.1 lists for each chapter (a) whether or not it assumes that there is abstract syntactic structure in the ellipsis site, (b) which linguistic submodule is responsible for providing an elliptical expression with an interpretation, or more specifically, which module of the grammar is relevant for defining the identity relation between an ellipsis site and its antecedent (recoverability/ellipsis identity), and (c) what mechanism licenses ellipsis beyond the issue of recoverability. Before turning to the table in more detail, let us make explicit three ground rules we adhered to in creating this overview table. First, the values listed in the cells represent the approach favored by the authors in these chapters. As such, they do not necessarily represent the views of the theoretical framework or linguistic (p. 8) subdiscipline as a whole.1 Very often, the authors point out that even within a certain framework or discipline there is disagreement with respect to these three issues, but they then proceed to express a preference for a particular position in the debate. It is this position that is represented in Table 1.1. Second, a “yes” in the column abstract structure is meant to cover both accounts that assume ellipsis sites contain a fully fledged syntactic (p. 9) representation and analyses that assume only a minimal instantiation of such structure (typically represented as a pro-form). Finally, note that the first chapter of Part I, Chapter 2, is missing from this table: given that it is a survey chapter that is explicitly intended to transcend individual analyses and approaches, we did not include it here.

Page 9 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives Table 1.1 Overview of Part I of the handbook: Abstract structure, recoverability, and licensing Chapter

Abstract structure

Recoverabili ty

Licensing

3

Lasnik and Funakoshi

yes

syntactic/ semantic

local structural relation with licensing head

4

Ginzburg and Miller

noa

discourse semanticb

QUD-based, supplemented with constructionspecific restrictions

5

Jacobson

no

discourse

type-shifting

semantic

rulec

6

Osborne

yes

syntactic/ semantic

only catenae can be elided

7

Culicover and

no

semantic

construction-

Jackendoff

specific restrictions

8

Goldberg and Perek

no

semantic

constructionspecific restrictions

9

Kempson, Gregoromiche laki, Eshghi, and Hough

mixedd

syntactic/ semantic

morphosyntac tic constraints imposed by ellipsis remnants

10

AnderBois

yes

syntactic/ semantic

no frameworkspecific constraints

Page 10 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives 11

Frazier

yes

syntactic

constraints on establishing discourse coherence

12

Roeper

yes

syntactic/ semantic

local structural relation with licensing head

13

Kehler

yes

syntactic/ semantic

constraints on establishing discourse coherence

14

Hardt

agnostice

syntactic/ semantic

[not directly addressed in the computational literature]

15

Winkler

yes

syntactic/

contrastive

semantic/ prosodic

accent on remnants and deaccenting of given material

(a) An exception is made for certain varieties of left- and right-peripheral ellipsis (including Right-Node Raising), where full structure is present, but partly unpronounced. (b) The term “discourse semantic” is meant to refer to approaches where the ellipsis antecedent is located in the discourse context. (c) This does not apply to fragment answers, which are licensed by being a part of a Qu(estion)–Ans(wer) unit. (d) This chapter adopts multiple mechanisms to derive ellipsis, only some of which generate structure inside an ellipsis site. (e) This chapter presents two main computational approaches, one of which assumes abstract structure, and one of which does not.

Page 11 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives What can we learn from this table? With respect to the first question—the presence or absence of abstract syntactic structure—it is clear that this issue remains, to this day, a very contentious one. Proponents of the ‘no structure’ approach often adopt an Occam’s razor-style position, which is backed up by cases where elliptical remnants fail to show connectivity effects with their purported syntactically fully represented source. Compare and contrast in this respect the elliptical reply in (11) with the non-elliptical example in (12) (both examples are from Ginzburg and Miller, this volume). (11)

(12)

If B’s elliptical reply in (11) contained (an abstract version of) the non-elliptical utterance in (12), it should be as ungrammatical as that example, quod non. Hence, adopting abstract syntactic structure in the case of (11) makes incorrect predictions and should be avoided. On the other hand, proponents of the structural approach also make use of connectivity effects in their argumentation, but they cite different types of data, such as the sluicing example in (13) (from Lasnik and Funakoshi, this volume, but orginally from Ross 1969b). (13)

The fact that the sluiced wh-phrase necessarily bears dative case—rather than, for example, the accusative that would be assigned by the immediately governing verb wissen ‘to know’—is argued to show that there must be an unpronounced copy of the dative-assigning verb schmeicheln ‘to flatter’ inside the ellipsis site, i.e. there must be abstract syntactic structure. The contrast between (11)/(12) on the one hand and (13) on the other shows that the standoff between structural and non-structural approaches to ellipsis largely boils down to which set of data should be given primacy. It seems, then, that more systematic and extensive data inventories are needed before this issue can be settled. The chapters in Part I of the handbook can be seen as a first step in that direction. The next column in Table 1.1 concerns recoverability, and more specifically which module of the grammar is relevant for defining the identity relation between an ellipsis site and its antecedent (ellipsis identity). Contrary to the previous issue, there is a greater degree of agreement regarding this second question, in that there seems to be a (near-)general Page 12 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives consensus that there is at least a semantic component to recoverability, above and beyond any morphosyntactic or lexical requirements there might be. For the ‘no structure’ analyses, this position is a logical necessity (as is also pointed out by Merchant, (p. 10) this volume), in that the absence of syntactic structure inside an ellipsis site makes it impossible to compare that structure to that of the antecedent. For the other accounts, though, it should be pointed out that the label “syntactic/semantic” used in Table 1.1 hides a fair amount of variation that exists between the individual accounts. While some authors (Lasnik and Funakoshi, AnderBois, Winkler) argue that recoverability intrinsically has both a syntactic and a semantic component, others claim that the recoverability mechanism can be more (or exclusively) semantic in some cases and more (or exclusively) syntactic in others, depending on the type of elliptical construction (Osborne), the stage in the acquisition process (Roeper), or the type of mechanism used to derive ellipsis (Kempson et al., Kehler, Hardt). Finally, we turn to the issue of licensing. As was pointed out in subsection 1.2.1, licensing is a relative newcomer to the stage of theoretical ellipsis research. This is reflected in the final column of Table 1.1: several frameworks and subdisciplines either make no specific claims with respect to this issue, or they lack a general, overarching licensing mechanism. Several chapters even—implicitly or explicitly—question the need for such an overarching theory, arguing instead that licensing is inherently construction-specific and hence sui generis for each individual ellipsis phenomenon (Ginzburg and Miller, Culicover and Jackendoff, Goldberg and Perek, Kempson et al.). The point is well-taken: to what extent is it possible to reduce the great diversity of elliptical constructions—recall also the sample in (1) in section 1.1—to a single licensing mechanism? Another issue that emerges from the chapters in Part I is that, contrary to what is commonly proposed in the literature (see in particular Lobeck 1995; Aelbrecht 2010), licensing is not necessarily strictly syntactic in nature. Ginzburg and Miller (this volume) propose that ellipsis is licensed when it provides an answer to the Q(uestion) U(nder) D(iscussion) (see also AnderBois, this volume, though for him it is part of recoverability/ellipsis identity), while both Frazier (this volume) and Kehler (this volume) discuss constraints on establishing discourse coherence under the rubric of licensing. In short, licensing is by no means a side character in the study of ellipsis, and we expect it will grow into a full-fledged research track of its own.

1.3.2 Part II: Ellipsis as a diagnostic tool Recall the starting point of Part II of the handbook: to what extent can ellipsis be used as a tool or instrument in the study of other linguistic phenomena? Before reflecting on this part in more general terms, we first provide a brief overview of each individual chapter. Abels (this volume) discusses the interaction between ellipsis on the one hand and movement and islands on the other. As was first discovered by Ross (1969b), sluicing can ameliorate island effects. That is, an A′-dependency that is illicit because it crosses an island boundary can become licit when the offending structure is elided. This fact, when taken at face value, can provide valuable insight into the nature of island constraints: if simply not pronouncing an island can bleed its effect, whatever is causing that effect Page 13 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives must be of a phonological nature (i.e. operative at the PF-interface). As Abels points out, however, once one looks more closely at the facts, they are much more complicated than they seem to be at first sight. He concludes that while ellipsis certainly has the potential of leading to new insights regarding islands (and movement, the second topic of the chapter), it cannot yet live up to that potential. Grodzinsky et al. (this volume) focus on comprehension of VP-ellipsis in (p. 11) speakers with an incomplete language faculty: among others, patients with Broca’s aphasia. The fact that patients with Broca’s aphasia are relatively successful in comprehending sentences with VP-ellipsis suggests that Broca’s aphasia is not a general failure of working memory, i.e. a difficulty of dealing with “syntactically complex” structures, but that it is a much more specific impairment, possibly one that specifically targets movement dependencies. This in turn might have repercussions for the proper analysis of ellipsis, Grodzinsky et al. argue, in that a movement-based approach towards VP-ellipsis (see e.g. Johnson 2001b) seems unlikely. Yoshida (this volume) uses psycholinguistic experiments involving elliptical constructions to gain more insight into the nature of the human parser. He ends up concluding that “the parsing strategies that the human parser employs achieve incremental, rapid, and grammatically detailed structure building,” which is in line with the results of sentence processing studies in other domains. Conversely, the evidence he discusses also suggests that ellipsis sites contain abstract syntactic structure, in that the parser builds the structure of ellipsis sites by copying the structure of the antecedent site. Finally, González-Vilbazo and Ramos (this volume) review the interaction between ellipsis and codeswitching. They argue, on the basis of German–Spanish elliptical codeswitching data, in favor of a constraint-free theory of codeswitching, i.e. a theory without codeswitchingspecific rules. At the same time, their data is also very informative about the theory of ellipsis itself: given that it is possible to codeswitch inside an ellipsis site, such facts can shed important new light on the nature of the identity relation between the ellipsis site and its antecedent (see subsections and 1.2.1 and 1.3.1). Overlooking the four chapters that make up Part II of the handbook, it becomes clear that the dichotomy we started out with about ellipsis being the object or the instrument of inquiry is to a certain extent a false one. All four chapters make clear that there is a fruitful two-way interaction between the theory of ellipsis on the one hand, and that of whatever other phenomenon is under investigation on the other. It is only based on a specific theory of ellipsis that one can draw conclusions about other domains of grammar and vice versa. At the same time, though, we believe the approach developed in this second part is a potentially very fruitful one, and hence one that should be continued to be explored in future research. In particular, there are various other domains where we can see the study of ellipsis having a real impact, from constituency (Depiante and Hankamer 2005; Sailor and Thoms 2014), to the proper definition of the QUD (Kehler, this volume), to the workings of memory (Martin and McElree 2009).

Page 14 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

1.3.3 Part III: Elliptical constructions As pointed out above (subsection 1.2.3), in structuring Part III of the handbook, we started out from a taxonomy that is quite common in the ellipsis literature. Needless to say, such an approach—like any attempt at categorization—risks not capturing phenomena that sit in-between different categories in the taxonomy and thus fall between the cracks. Indeed, as noted by Merchant (this volume): “There are many other kinds of phenomena that go under the rubric of ellipsis as well, some better investigated than others, including argument drop, article drop, haplology, diary language and headlinese, subjectless infinitivals, copula drop, situational ellipses, small clauses, and many more; some are context-sensitive, and some are not. For various (and still incomplete) taxonomies of the missing, see Klein 1985 (p. 12) and Hennig 2013: 447–8” (and see McShane 2005 for a similarly broad view of what qualifies as ellipsis). Interestingly, though, when taking a look at the chapters (in all four parts) of this handbook, one is struck by the fact that some of these ‘non-conventional’ types of ellipsis have found their way into one or more chapters, while others remain unmentioned. Two notable examples of the former type are argument drop (or pro-drop) and copula drop, which are discussed in several chapters of this handbook. Highly context-dependent types such as diary language or headlinese, however, have not made their way in. This might suggest that there is, after all, a natural subdivision of elliptical phenomena into meaningful subgroups (see also Merchant 2016 for related discussion). A partially related observation is that in the chapters of this handbook, the boundaries between the various (sub)types of ellipsis are becoming increasingly blurred. For instance, when considering both the empirical descriptions and the theoretical approaches to sluicing and fragments in different chapters (and even different parts) of this handbook, one could consider not analyzing them as separate types of ellipsis, but rather as instantiations of one and the same elliptical phenomenon (as suggested, for instance, by van Craenenbroeck and Lipták 2006, 2013, and Temmerman 2013). Similarly, it becomes apparent in a number of chapters (both in Part III and in Part IV) that distinguishing between nominal ellipsis and argument drop (pro-drop)—or even between argument drop and (verb-stranding) VP-ellipsis—is not a trivial (and therefore, perhaps an unnecessary) task (cf. also Ginzburg and Miller, this volume, for similar remarks). We find such developments promising, given that they steer us away from the focus on individual constructions, a line of thinking that is reminiscent of the early construction-specific days of generative grammar (see also Johnson 2008a: 3). It seems fruitful to us to not think of specific elliptical (sub)types as construction-specific phenomena, but rather, to try to formulate empirical and theoretical generalizations that transcend the traditional taxonomy.

1.3.4 Part IV: Case studies The language-specific chapters of Part IV inventory which of the elliptical phenomena discussed in Part III are attested (and which ones are not) in the language and describe their properties. Table 1.2 presents an overview of the different languages and elliptical Page 15 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives phenomena under scrutiny. Before discussing the contents of the table, let us first make two observations about how it came about. First off, it should be noted that this overview is crucially based on surface patterns: whether a given construction can or should be analyzed as involving ellipsis or not—a matter discussed in detail in most of the chapters of Part IV—is not taken into account here. For example, a pattern such as the Persian one in (14) (taken from Toosarvandani, this volume) leads to the value “yes” in the SL-column for Persian because it displays the relevant surface pattern (a constituent question reduced to its wh-phrase). Secondly, the use of gray shading indicates that these particular elliptical phenomena are not discussed in the relevant chapters. Table 1.2 Overview of Part IV of the handbook: Cross-linguistic distribution of the main ellipsis types

Abbreviations: SL = sluicing, PE = predicate ellipsis, G = gapping, STR = stripping, FR = fragments, NPE = noun phrase ellipsis, CD = comparative deletion, CR = Conjunction Reduction, RNR = Right-Node Raising a

We include in CR the phenomenon sometimes referred to as non-constituent coordination, e.g. John gave a book to Mary and a car to Bill. Page 16 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives b

The set of verbs that can occur in this pattern seems to be very limited.

c

We have broadened the scope of Chapter 38 from “varieties of English” to English in general. Thoms (this volume) mostly focuses on predicate ellipsis, pointing out that little is known about dialectal variation in the domain of clausal or nominal ellipsis. However, the standard English facts in these domains are well-known from the literature. d

Whether Indonesian exhibits RNR is somewhat controversial. Moreover, there seems to be speaker variation. e

Only marginally possible, subject to speaker variation. (14)

One thing that jumps out from this table is that (surface patterns of) elliptical constructions are cross-linguistically widespread: most if not all of the ellipsis phenomena (p. 13)

are attested in each of the twelve languages studied.2 While this is an interesting (p. 14) observation in and of itself, it might also be partly illusion, caused by the fairly coarse granularity of the phenomena listed in the table (as well as the criteria used to describe the phenomena, i.e. our reliance on surface patterns). This suspicion seems to be confirmed once we increase the level of detail in our overview. Table 1.3 takes one phenomenon from our first table—the most intensively studied one, predicate ellipsis— and splits it up into five subtypes. The resulting picture is considerably more varied than could be gleaned from Table 1.2. Predicate ellipsis is a cover term for several types of ellipsis targeting the (verbal) predicate. Different instantiations include auxiliary-stranding VP-ellipsis, modal-stranding VP-ellipsis, and main-verb-stranding VP-ellipsis. Table 1.3 clearly shows that the twelve languages studied in this part of the handbook do not all behave similarly. At first glance, the following picture seems to emerge: •

(p. 15)

When a language exhibits AuxVPE, it also has ModVPE (but not vice versa).3

• When a language exhibits AuxVPE (and hence also ModVPE), it is possible that it also has VVPE (as in Hungarian), but this is by no means required (see English and Indonesian). • A language can exhibit only ModVPE (as in Dutch or French), or only VVPE (as in Persian). Table 1.3 Overview of the cross-linguistic distribution of various subtypes of predicate ellipsis Page 17 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives

Abbreviations: AuxVPE = auxiliary-stranding VP-ellipsis, VVPE = main-verb-stranding VP-ellipsis, ModVPE = modal-stranding VP-ellipsis, PG = pseudogapping, ACD = Antecedent-Contained Deletion a

Marginally possible in comparatives.

b

Single example, judgments unclear.

In addition to these three types of VP-ellipsis, there is also pseudogapping and Antecedent-Contained Deletion (ACD), two elliptical subtypes which also do not show a uniform cross-linguistical distribution. It is not straightforward to come up with a generalization that describes the (non-)occurrence of pseudogapping or ACD in a given language. For instance, the possible generalization that languages that allow AuxVPE also exhibit pseudogapping (cf. Hungarian, English, and Indonesian) is contradicted by the fact that Polish has AuxVPE, but does not exhibit pseudogapping. It could be the case, then, that AuxVPE is a necessary but not a sufficient condition for pseudogapping. Adding to the variation in Table 1.3 is the fact that several languages exhibit predicate ellipsis phenomena that seem to be sui generis, i.e. that are specific to a single language (usually because it is dependent on other material that is specific to that language). For example, Hungarian also exhibits preverb-stranding VP-ellipsis, British English has a type of predicate ellipsis that looks identical to VP-ellipsis save for the addition of a non-finite

Page 18 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives form of the verb do (known as British English do), and verb-stranding VPE in Persian is actually v-stranding VPE. The contrast between (the second column of) Table 1.2 and Table 1.3 suggests that were one to do the same exercise for, say, sluicing or nominal ellipsis, similar patterns of variation would emerge. Possible sources of variation that come to mind are (i) the different licensers of nominal ellipsis across languages (determiners, demonstratives, adjectives, possessives, etc.—recall the contrast between (6) and (7)), (ii) the crosslinguistic distribution of different types of sluicing (sprouting, swiping, spading, multiple sluicing, etc.), (iii) the cross-linguistic distribution of polarity ellipsis, (iv) variation in the number of remnants allowed by gapping,4 and (v) variation in connectivity effects in elliptical constructions (islands, preposition stranding, case marking…). In short, the empirical picture is complex and nuanced, leading to the all too familiar tension between empirical coverage and theoretical parsimony: to what extent is it feasible and/or desirable to try to construct a general, overarching theory of ellipsis? Given that this very same issue has cropped up in the previous subsections as well, it seems to us that this is one of the central questions in the linguistic study of ellipsis. (p. 16)

1.4 Conclusion Taken together, the four parts of this handbook present a comprehensive, in-depth, and balanced discussion of the phenomenon of ellipsis in natural language. They devote space to well-established theories of and generalizations about ellipsis, but at the same time leave room for cutting-edge research that broadens the scope of the investigations, opening up exciting new prospects, empirically as well as theoretically. One example of the latter is the role that the notion of Question Under Discussion (QUD) plays in much current ellipsis research, as is evidenced by the fact that it shows up in a number of chapters, across different theoretical frameworks and across the different parts of the handbook. With respect to the empirical study of elliptical phenomena, it is clear that more work is needed, not only on more constructions in more languages and in more detail and depth, but also on more types of data (see, for example, the role played by corpus material in Ginzburg and Miller, this volume). In short, the work is far from done, but the future of the linguistic study of ellipsis looks very bright: it is a thriving subfield of linguistics that offers exciting prospects for new discoveries and breakthrough developments, both empirically and theoretically. It seems only fitting that we would end this chapter on an elliptical note, and so we are.

Acknowledgements We would like to thank the authors of the chapters, without whom there simply would be no handbook, for their interest in this project and their valuable contributions. A special

Page 19 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives mention goes out to our dear colleague and friend Luis Vicente, author of Chapter 20, who unfortunately passed away before he could see the completion of this handbook. We also want to express our gratitude to the following reviewers, who invested time and effort in assisting in peer-reviewing the chapters, providing excellent comments and feedback to the authors that substantially helped improve the quality of the handbook: Lobke Aelbrecht, Doug Arnold, Marc Authier, Julia Bácskai-Atkári, Matt Barros, Anne Bezuidenhout, Rajesh Bhatt, Željko Bošković, Hans Broekhuis, Ronnie Cann, Carlo Cecchetto, Simon Charlow, Rui Chaves, Ivano Ciardelli, Barbara Citko, Charles Clifton, Peter W. Culicover, Sonia Cyrino, Anne Dagnac, Marcel den Dikken, Michael Yoshitaka Erlewine, Hans-Werner Eroms, Frank van Eynde, Caroline Féry, Klaus Fischer, Lyn Frazier, Adele E. Goldberg, Lydia Grebenyova, Vera Gribanova, James Griffiths, Alison Hall, Daniel Hardt, Katharina Hartmann, William Harwood, Angeliek van Hout, Kyle Johnson, Ruth Kempson, Chris Kennedy, Katalin É. Kiss, Greg Kobele, Marjo van Koppen, Diane Lillo-Martin, Anikó Lipták, Amalia Lombart, Andrea E. Martin, Jonathan McDonald, Jeff McSwan, Jason Merchant, William O’Grady, Timothy Osborne, Victor Pan, Dan Parker, Eric Potsdam, Esther Ruigendijk, Jeffrey Runner, Craig Sailor, Mamuro Saito, Ana Lúcia Santos, Yosuke Sato, Petra Sleeman, Jennifer Spenader, Jon Sprouse, Mark Steedman, Anna Szabolcsi, Adam Szczegielniak, Daiko Takahashi, Kensuke Takita, Azita H. Taleghani, Maziar Toosarvandani, Luis Vicente, Mark de Vries, Jenneke van der Wal, Andrew Weir, Masaya Yoshida, and Jochen Zeller. Lastly, we are very grateful to Julia Steer (Linguistics Commissioning Editor), Vicki Sunter (Senior Assistant Commissioning Editor, Linguistics), Karen Morgan and Verity Rimmer (Editorial Administrators) at OUP, Manikandan Chandrasekaran (Project Manager at SPi Global), and Jess Smith (copyeditor), for their patience and assistance throughout this (long!) editing process.

Notes: (1) This is also why we represent the chapters by their author names in Table 1.1, rather than referring to the theoretical framework or linguistic subdiscipline that is discussed in the chapter. (2) Comparative deletion might be the odd one out: although data is lacking from four of our twelve languages, two of the remaining eight lack this elliptical construction. (3) Note that we cannot be sure of this generalization for Polish, as Chapter 36 does not contain an example of ModVPE. (4) See, for instance, Patin and Manus (this volume), who show that Kiswahili allows for gapping with multiple remnants, while in the closely related Bantu language Shingazidja there is an upper limit of two gapping remnants.

Jeroen van Craenenbroeck

Page 20 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis In Natural Language: Theoretical and empirical perspectives Jeroen van Craenenbroeck is Associate Professor of Dutch Linguistics at KU Leuven, where he is also Vice-President of the Center for Research in Syntax, Semantics, and Phonology (CRISSP). He is the author of The Syntax of Ellipsis (OUP) and general editor of the journal Linguistic Variation (John Benjamins). His research interests include ellipsis (sluicing, swiping, spading, VP-ellipsis), expletives, verb clusters, and the left periphery of the clause. Tanja Temmerman

Tanja Temmerman is Assistant Professor of Dutch Linguistics at Université SaintLouis—Bruxelles (Belgium), where she is also Lecturer in English Language and Head of the English Department. She obtained her PhD from Leiden University in 2012 with a dissertation entitled ‘Multidominance, ellipsis, and quantifier scope’. Her principal research foci lie in (generative) syntax, issues at the syntax–phonology and syntax–semantics interfaces, Dutch dialectology, and comparative Germanic syntax. Specific topics of interest include ellipsis, the internal and external syntax of idioms, phase theory, long-distance dependencies, island effects, phrase structure, modals, and negation.

Page 21 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: Goldsmiths College, University of London; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Oxford Handbooks Online Ellipsis: A survey of analytical approaches   Jason Merchant The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.2

Abstract and Keywords This chapter surveys a variety of approaches to ellipsis, classifying them by how they answer the questions of structure (‘Is there structure inside ellipsis sites?’) and identity (‘What kind of identity, if any, holds between an ellipsis and its antecedent?’). It reviews the evidence for and against structure in ellipsis, from lower infinitivals and predicate answers. While the results are mixed, on the whole, the preponderance of the evidence suggests that the analysis of the full range of facts of ellipsis requires reference to syntactic structure. Keywords: ellipsis, sluicing, VP-ellipsis, gapping, locality, identity, structure

Jason Merchant

(p. 19)

2.1 Introduction: The phenomena

THE term ellipsis has been applied to a wide range of phenomena across the centuries, from any situation in which words appear to be missing (in St Isidore’s definition), to a much narrower range of particular constructions. Ellipsis continues to be of central interest to theorists of language exactly because it represents a situation where the usual form/meaning mappings, the algorithms, structures, rules, and constraints that in nonelliptical sentences allow us to map sounds and gestures onto their corresponding meanings, break down. In fact, in ellipsis, the usual mappings seem to be entirely absent. In ellipsis, there is meaning without form. VP-ellipsis and sluicing are two of the best-investigated instances of ellipsis and generally show remarkable similarities in the demands they make of the discourse, both usually necessitating some equivalent antecedent which is subject to some kind of parallelism. It is no exaggeration to say that debates over the nature of this parallelism have formed the core of most of the generative work on ellipsis over the last forty years. Almost all Page 1 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches conceivable positions on the parallelism question have been explored and advanced, and these debates are important exactly because they are often used to argue for the necessity of one or another kind of linguistic representation. Most of the debate is located in the arena of semantics and abstract syntactic structures—it is clear that surface syntactic or phonological parallelism is not at stake—and as such, elliptical structures often play an important role in fundamental ontological debates in linguistics. The logic is clear: if the parallelism or identity conditions found in ellipsis resolution require reference to certain kinds of objects, then our theories of linguistic competence must countenance objects of that kind. In generative linguistics, research has focused largely on two sets of constructions.1 Central examples of the first set, drawn from English, include sluicing as in (1), predicate or (p. 20) verb phrase ellipsis (VP-ellipsis) as in (2), and NP-ellipsis (or N′-ellipsis, or more broadly, nominal ellipsis) as in (3). Ellipsis of at least one of these kinds seems to be found in every language in which it has been looked for, though a systematic cross-linguistic theory of the distribution of ellipsis types remains to be formulated. (1)

(2)

(3)

In each case, the second clause can be understood as in (4)–(6). (4)

(5)

(6)

These three kinds of ellipsis are distinguished as well by the fact that distributional facts lead us to expect to find structural elements corresponding to the perceived interpretations: wh-phrases as in (1) require clausal sources, auxiliaries like can in (2) take VP complements, and determiner-like elements such as six in (3) require nominal complements. In other words, selectional or subcategorizational properties of particular Page 2 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches elements require us to posit elided structures in (1)–(3), if we adopt the null hypothesis that these properties are uniform across (1)–(6). A second set of constructions in which ellipsis has been invoked include stripping (or ‘bare argument ellipsis’) in (7), gapping in (8), fragment answers in (9), as well as a host of other cases that fall under the general rubric of ‘conjunction reduction’: (7)

(8)

(9)

All of these structures2 have been the focus of intense theoretical interest over the past four decades, and vast bibliographies can be compiled for each of the above phenomena. I can (p. 21) make no pretense of bibliographic completeness here, and refer the reader to excellent recent surveys for a more detailed treatment of the literature, especially Hartmann (2000), Johnson (2001b), Winkler and Schwabe (2003), Goldberg (2005), Winkler (2005), van Craenenbroeck (2010b), Reich (2011), van Craenenbroeck and Merchant (2013), and the introduction to Johnson (2008b). In what follows, I will examine some representative examples of approaches to the above and discuss their relative merits. In analyzing ellipsis, three questions have occupied much of the literature. The first is given in (10), what I will call the structure question. (10)

The answer that is given to (10) has far-reaching implications for the theory of grammar. If the answer is positive, we must countenance theories of grammars that permit unpronounced phrases and heads. If the answer is negative, there is the possibility that syntax may be ‘wyhiwyg’ (‘what you hear is what you get’), with no unpronounced elements. The debate on this question bears some resemblance to debates in the midtwentieth century about the nature of abstractness and the phoneme: there are good reasons to prefer a parsimonious theory of any domain of data, but not at the expense of coverage of the facts. Some of the various strands of evidence that have been brought to Page 3 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches bear in attempts to answer (10) are laid out in the next section and compared in detail in sections 2.3 and 2.4. The second major question is what I will call the identity question: (11)

This question has generally been answered in terms of various kinds of posited identity relations: elided material (call it XPE) must be identical or parallel to or resolvable by some antecedent phrase (YPA), where the identity (or parallelism, or resolution) may be semantic or syntactic, or some mix of the two. The various approaches to the identity question are addressed in section 2.6. Putting these first two questions schematically, then, we have the following: 1. Is there syntax internal to the ellipsis site? 2. The understood material is identical to some antecedent. Is the relevant kind of identity syntactic (defined over phrase markers or syntactic derivations of some sort) or semantic (defined over semantic representations or computations of some sort)? Table 2.1 organizes a representative selection of the literature by the answers it proposes to these two questions; few if any of the works deal with all kinds of ellipsis: the table assigns them to the various categories based on the kinds of ellipsis they do discuss, though this should not impute to any of them necessarily a uniform theory of ellipsis (most deal only with VP-ellipsis in English). Further, the row labeled ‘both’ includes theories that are hybrid in various ways involving an admixture of syntactic and semantic requirements, sometimes uniformly, and sometimes varying by construction or context. Table 2.1 Some previous research on the two ellipsis questions

Page 4 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

(p. 22)

A third major question, which so far has not attracted quite the attention the above

two questions have, is the licensing question: (12)

The licensing question was traditionally addressed by writing the structural description of a deletion transformation to be sensitive to whatever conditions the theorist thought relevant (for example, in Ross 1969b, Sag 1976a, and Hankamer 1979). It has been addressed in work that does not assume a transformation of deletion by Zagona (1982), Lobeck (1995), Johnson (2001b), Merchant (2001), Aelbrecht (2010), van Craenenbroeck (2010b), van Craenenbroeck and Lipták (2013), and Miller (2011b); all but the last owe a great debt to Lobeck (1995), whose approach is based on a kind of Empty Category Principle applied to a null pro-like element.

2.2 Approaches to the syntax of ellipsis The reasons for theoretical interest in elliptical structures are obvious: in each case, the usual form–meaning correspondence appears to break down—there is meaning in ellipsis without form. In broad terms, there have been two answers to the puzzle posed by ellipsis structures: the non-structural and the structural. The non-structural approach responds by (p. 23) supplementing the theory of meanings, creating or exploiting devices that can generate meanings in the absence of syntactic structure. The structural approach places the burden on the syntax, and claims that the meanings are derived by (ideally all and Page 5 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches only) the mechanisms at play in other contexts; it distinguishes itself from the nonstructural approach by positing structure which is not pronounced. Within structural approaches, two main lines of investigation can be distinguished: those that posit essentially ordinary syntax, subject to some kind of ‘deletion’ to render the syntax unpronounced, and those that posit a null lexical element which is replaced or identified at some level of representation not relevant to the pronunciation (at LF or in some semantic/pragmatic component). Schematically, these various tacks can be distinguished by their answers to the following questions (see also Winkler and Schwabe 2003 and Stainton 2006b for more detailed taxonomies): (13)

Recent advocates of non-structural approaches to ellipsis include Ginzburg and Sag (2001) and Culicover and Jackendoff (2005). Concretely, they propose that e.g. a sluicing example like that in (4) contains no syntactic material corresponding to the usual clausal source for wh-phrases. Instead, the wh-phrase is the sole daughter of an S node which is the complement to know: (14)

Page 6 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches The S node in this account, which has the construction type sluiced interrogative clause, is endowed with featural machinery designed to account for the observed interpretation (among other things). Culicover and Jackendoff’s approach is similar: for them, the S node is notated ‘IL’ (for indirect licensing) and the wh-phrase is an orphan; the semantics then is constructed with a free variable F whose value is constructed from the context via ‘indirect licensing’. (p. 24)

(15)

Among structural approaches, those that do not implement deletion on the PF side of the derivation either posit null elements in the syntax or deletion of syntactic elements. On theories with null elements, either the null element is a single, designated terminal, as in Hardt (1993) and Lobeck (1995), or there are a plethora of null elements, as in Wasow (1972) and Ludlow (2005). These two options assign the structures in (16a,b) to examples like (4) as the representations that feed pronunciation. (16)

The null elements are either replaced by an operation of structure copying before the structure is interpreted, yielding (17) (as in Shopen 1972, Wasow 1972, Williams 1977b, Fiengo and May 1994, Chung et al. 1995, Lappin 1999, and Fortin 2007b), or are interpreted by non-syntactic algorithms for anaphoric elements (as in Hardt 1993 and Merchant 2014a).3 (17)

Finally, we find the traditional generative solution to ellipsis, in which syntactic structures are subject to non-pronunciation, either as the result of some operation of deletion (which operates either in the syntax before Spell-Out or after Spell-Out in the derivation to PF, as variously in Ross 1969b, Sag 1976a, Hankamer 1979, Lasnik 2001, Saab 2009, Baltin 2012, Kobele 2012a, Merchant 2015b) or as a phonological reflex of prosodic algorithms (in the PF ⇝ phonology mapping or in the phonology sensu stricto, as in Merchant 2001, Johnson 2004b, Aelbrecht 2010, van Craenenbroeck 2010b, and others). Under such approaches, a sluiced clause appears as follows, where angled brackets enclose ‘deleted’ or, more neutrally, unpronounced material: (18) Page 7 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

For the most part, the differences between the various implementations seem to be fairly minor, though important for certain architectural deliberations. These differences are taken up in more detail in section 2.5. It is also important to note that several authors have proposed revisiting and maintaining the ‘surface’ vs ‘deep’ anaphoric dichotomy of Hankamer and Sag (1976), allowing for the possibility that certain ‘surface’ anaphors (including relatives of VP-ellipsis and perhaps (p. 25) VP-ellipsis itself) may be due to deletion or replacement at PF alongside ‘deep’ anaphors (null or overt), such as those found in Null Complement Anaphora or exophoric fragments (see Sag and Hankamer 1984, Merchant 2010, Baltin 2012, Bentzen et al. 2013, Miller and Pullum 2013, and Merchant 2013c, 2014a).

2.2.1 Structural and non-structural approaches compared How does one decide whether some piece of syntactic structure is or isn’t there, particularly when that structure in any case does not lead to any pronounced difference? Indirectly, by necessity. Detecting and arguing for such ‘missing’ structures is analogous to searching for and determining the properties of a black hole: one can tell it’s there only by its effects on surrounding material. The logic of the hunt for elided structure is similar. If one finds effects that seem to be due to missing material, there is an argument that such structure exists. In other words, if effects are found which we would otherwise attribute to properties of structure X in similar, non-elliptical, cases, but structure X is, by hypothesis, internal to the ellipsis site, then X exists. If, on the other hand, expected properties are missing, one could conclude that structure X is absent. Structural approaches are based on what I call connectivity effects; non-structural approaches take their lead from non-connectivity effects. Connectivity effects occur when some part of the clause that contains the ellipsis shows ‘connectivity’ to some other, supposed, unpronounced part; non-connectivity is when this does not occur, despite a prior expectation that it would. In what follows, the main lines of evidence for each approach are presented.

Page 8 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

2.3 Evidence for structure in ellipsis There are more than thirteen sets of facts which have been used to argue for unpronounced structure in ellipsis: lower origin effects, locality effects, P-stranding effects, case-matching effects, agreement effects, the distribution of complementizers, of infinitivals, and of predicate answers, binding theoretic effects (Ott 2014; Ott and de Vries 2016), the presence of intermediate reconstruction effects in sluicing (AgüeroBautista 2007), the facts of ‘spading’ (which shows evidence for an underlying cleft, as van Craenenbroeck 2010b argues), the licensing of parasitic gaps inside ellipses (Yoshida et al. 2015a), the existence of syntactic priming effects (Xiang et al. 2014), and others. In the remainder of this section, I briefly illustrate the facts from the first eight sets, and refer the reader to the literature just cited for the last five and others.

2.3.1 Lower origin effects A variety of elements can be moved out of putative ellipsis sites. However such dependencies are analyzed, they involve an origin site (or tail to a chain, or an equivalent device) inside the ellipsis site. Such moved elements include A′ elements (interrogative phrases, relative pronouns, topicalized phrases, comparative operators, etc.), A-moved elements (p. 26) (subjects of passives, raising verbs, and unaccusatives), and head elements. Examples of the first are plentiful in sluicing, antecedent-contained deletions, and comparative ellipsis; examples of A-moved elements are equally easy to find (see Aelbrecht 2010). The following is an example of a topicalized phrase that is across-theboard extracted from an elided VP (from Wallace Stegner’s 1987 novel Crossing to Safety): (19)

For head movement, careful work on a series of constructions has shown that certain languages with verb movement also have verb phrase ellipsis, resulting in structures in which the verb is stranded outside the VP (verb-stranding verb phrase ellipsis), or more generally, X-stranding XP-ellipsis, as dubbed by Lipták and Saab (2014). Representative work includes Huang (1988b), McCloskey (1991a), Cyrino and Matos (2002, 2005), Goldberg (2005), Gribanova (2013a, 2013b), Lopes and Santos (2014), Thoms (2014), and Bennett et al. (2015). The clearest kind of example comes from a language that lacks prodrop, topic drop, or other processes that license null arguments of finite verbs, yet still have verb movement; the best-known such case is that of Irish (example from McCloskey 1991a: 273), which is VSO in finite clauses, with the verb moving to a position outside the verbal projection whose specifier hosts subjects: (20)

Page 9 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

As McCloskey shows, it is extraordinarily difficult to reconcile such examples with what is known of the grammar of Irish verbs and arguments without positing an ellipsis that targets a verbal projection whose head has been extracted to a position outside the ellipsis site.

2.3.2 Locality effects The evidence from locality effects is distributed across a number of domains, but all of it has the same basic form: some kind of locality constraint (typically but not exclusively island constraints) are observed to hold of elements whose putative origin site is inside the understood missing material. If any of these island constraints are due to restrictions on syntactic (broadly speaking) representations, then their presence in elliptical structures argues that those representations must be present.

2.3.2.1 VP-ellipsis The first set of locality effects come from VP-ellipsis, where relative operators, wh-phrases, topicalized phrases, parasitic gap operators, and comparative operators all show sensitivity to islands, even when the tail of the dependency is inside an ellipsis site. The examples below are culled from and discussed in and Sag (1976a), Haïk (1987), Postal (1994), Kennedy and Merchant (2000a), Lasnik (2001), Merchant (2001, 2008b), and Fox and Lasnik (2003). (p. 27)

(21)

4

2.3.2.2 Fragment answers Similar effects are found in some fragment answers to implicit salient questions, as discussed in Morgan (1973), Merchant (2004a), Arregi (2010), Temmerman (2013),

Page 10 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches Griffiths and Lipták (2014), Weir (2014), and Barros et al. (2015) (though see section 2.4.1.2 for complications). (22)

(23)

2.3.2.3 Stripping/Bare Argument Ellipsis Examples of ‘stripping’, whose analysis appears to have much in common with that of fragment answers, show a locality effect between the correlate and the ‘bare argument’ (in Reinhart’s 1991 term; see Depiante 2001, Lechner 2001, Merchant 2009, Yoshida et al. 2015b, and Wurmbrand 2017). (24)

2.3.2.4 Gapping Gapping, which is probably just a version of stripping with more than one remnant, unsurprisingly behaves like stripping in this regard as well (Coppock 2001; Johnson 2004a, 2009; Winkler 2005; Toosarvandani 2013a): (25)

(p. 28)

(26)

Page 11 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches 2.3.2.5 Contrast sluicing Finally, note that even when a sluiced wh-phrase has an explicit correlate, we still find locality effects when the relation between the correlate and the wh-phrase is one of contrast, as originally noted in Merchant (2008b: 148): “Sluicing with indefinite correlates repairs islands, but Sluicing with focused correlates does not.” (27)

See also Merchant (2001), Vicente (2008), Griffiths and Lipták (2014), and Barros et al. (2015) (and note that accounting for the full range of facts requires something like MaxElide, as Merchant 2008b: 152 and Fox and Lasnik 2003: 153 n. 10 point out, pace Messick and Thoms 2016).

2.3.3 The P-stranding generalization The third major strand of evidence for structure internal to ellipsis sites comes from the distribution of preposition stranding under wh-movement out of putative ellipsis sites cross-linguistically. Both under sluicing and in fragment answers, there is a strong (if not always perfect, apparently not random or accidental) correlation between languages that allow P-stranding in non-elliptical contexts and in sluicing/fragment answers. If what regulates P-stranding cross-linguistically is some kind of morphosyntactic condition, and not due to differing semantics across languages (an assumption I know of no serious challenge to; see Abels (2003) for insightful discussion), then the fact that this correlation holds in seemingly elliptical contexts is quite telling. Sluicing data from representative languages is given here, reproduced from Merchant (2001); (28) represents P-stranding languages (as seen in the (b) controls), while (29) illustrates non-P-stranding languages. (28)

(29)

Page 12 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

This parallelism is expected on structural approaches, since the grammatical constraints that govern preposition stranding will be operative in these (elliptical) structures as well. (p. 29)

The parallelism, however, is far from perfect, and numerous empirical caveats to this generalization have been raised in the recent literature: see section 2.4.3.

2.3.4 Case matching As first pointed out in Ross (1969b), case-matching effects found in sluicing (and fragment answers, Merchant 2004a, and contrastive left-dislocation, Ott 2014) are straightforwardly accounted for if the relevant case assigners are syntactically present, though unpronounced. Ross’s particular example comes from German, where schmeicheln ‘flatter’ assigns dative, while loben ‘praise’ assigns accusative: (30)

See Barros (2014b) for fuller discussion of case matching that appears to hold even in certain cases where we would expect a cleft or other non-isomorphic source.

2.3.5 Complementizer deletion If fragment answers involve ellipsis (Morgan 1973) preceded by movement of the fragment out of an elided clause (Merchant 2004a; Merchant et al. 2013; see also Thoms 2016), then the following pattern is accounted for, given that displaced complementizer phrases require overt complementizers. (The well-formed response variant that lacks that

Page 13 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches initial complementizer in (32) is not elliptical: it is an indirect answer, not a direct answer; see Merchant et al. 2013 for discussion of this difference.) (31)

(32)

(p. 30)

2.3.6 Infinitivals: Raising vs control

A similar distinction is found in the distribution of short answers using infinitival clauses: only control infinitivals can be clefted, and only control infinitivals can serve as fragment answers (Merchant 2004a). (33)

(34)

This is expected if the control CP can be fronted, but the raising TP cannot be; see Landau (2013).

Page 14 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

2.3.7 Predicate answers Finally, predicate answers show a distribution which is somewhat puzzling if the connection between question and answer is mediated solely by some semantic/pragmatic relation, and not by syntactic structure (Hankamer 1979; Merchant 2004a). (35)

As Hankamer (1979) pointed out using slightly different terms, the minimal fragment answer to a question whose semantics demand only an element of type , for example, cannot be answered with a simple verb (of type ). Instead, the minimal fragment must be a VP (of type ), despite therefore necessarily including redundant, given information (in the form of the pronoun). This pattern is expected if the short answer involves movement of a phrase to a clause-peripheral position: since English lacks long head movement (and also remnant topicalization of VPs, as Müller 1998 discusses), only the entire VP will be available to form the answer. On non-structural approaches that base-generate words subject only to semantic or pragmatic answerhood conditions, this restriction is mysterious. Equally mysterious for theories that impose category matching between the wh-phrase and the answer (such as Jacobson 2016b) is well-formedness of a tensed VP (p. 31) funded her where such a VP is ill-formed as the complement of a tensed do, shown in (35b). Overall, these facts point to the conclusion that constraints on form— mediated by structure—are active in elliptical constructions.

2.3.8 Agreement triggers Targets of agreement can be controlled by elements internal to putative ellipsis sites: this is well documented for predicate ellipses and certain kinds of nominal ellipsis (see Saab 2009; Merchant 2013a, 2014a; Saab and Lipták 2016) and can be seen also in the following set of data. Subject–verb agreement in English for number does not always track notional or semantic number: pluralia tantum such as nuptials trigger plural agreement on the verb, unlike its singular synonym wedding: (36)

Nominal ellipsis preserves the syntactic properties of agreement: (37)

Page 15 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

(38)

Agreement thus appears to be sensitive to unpronounced structure: the grammatical plural feature on the antecedent of the ellipsis, nuptials, must be encoded locally in the NP subject to ensure plural agreement. As Culicover and Jackendoff (2005: 11 n. 8) put it, the presence of these kinds of connectivity effects would represent “impressive evidence of the reality of the invisible structure” (while reporting that they don’t find consistent island effects in cases like (23b), they don’t consider the remaining facts). The conclusion pointed to by the above kinds of data would seem to be that there is (regular but unpronounced) syntactic structure inside ellipsis sites.

2.4 Evidence against structure in ellipsis Some kinds of data, however, seem to point to the opposite conclusion: that there is no structure inside ellipsis sites (at least no structure that has the properties of its putative non-elliptical counterpart). (p. 32)

2.4.1 Absence of locality effects

The strongest piece of evidence in favor of the non-structural approaches comes from the lack of island effects in certain ellipsis contexts, such as in many sluicing structures, in certain fragment answers, possibly in certain gapping examples, and in certain kinds of comparative ellipsis.

2.4.1.1 Sluicing As Ross (1969b) famously first observed, the putative wh-extraction out of ellipsis sites in sluicing appears insensitive to islands: (39)

(40)

Page 16 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches Though this observation holds in the first instance for cases in which the wh-phrase corresponds to an overt indefinite, Culicover and Jackendoff (2005: 258 n. 17) produce one example with a merely implicit correlate which they judge acceptable (example modeled on one from Chung et al. 1995, where the opposite judgment is reported): (41)

Since this relative clause is veridical, this is not a counterexample; such sluices are discussed in chapter 5 of Merchant (2001), where they are given a non-island source, with an E-type pronominal subject, equivalent here to…but I’m not sure with what he [=that plumber Bob found] fixed the sink. See AnderBois (2014) for additional discussion of the properties of such sluices, and Griffiths and Lipták (2014) for a comprehensive attempt to distinguish apparently island-sensitive from non-island-sensitive sluices, following Merchant (2008b), as well as Barros et al. (2015).

2.4.1.2 Fragment answers Similar observations have been made for certain fragment answers (in Culicover and Jackendoff 2005: 244ff.; Stainton 2006b). (42)

(43)

Interpreting these data requires some care, however. First, sometimes bound prefixes can appear without their hosts, as in (44). Second, the interpretation of the fragment in (43B) is (p. 33) equivalent to the paraphrases given in (45a–c) (readings which the presumably non-elliptical (45d,e) can have as well, in this context), and the fragment in (43B) does not appear to have the expected ‘island-violating’ reading given in rough paraphrase by (45f). While this set of facts is expected on the structural approach, it is not clear how the non-structural approach rules out the interpretation in (45f) for (43B). (44)

(45)

Page 17 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Casielles (2006) and Stainton (2006b) also adduce fragment answer examples out of islands that seem quite acceptable. It is also true, as Progovac et al. (2006b) point out, that without a comprehensive theory of islands it may be difficult to properly assess the importance of island sensitivities (they suggest, following others, that perhaps some islands are semantic or pragmatic in nature, not syntactic); much more work is needed to ascertain the full empirical lay of the land in this domain as well. See Merchant et al. (2013) for some experimental investigation of fragments.

2.4.1.3 Gapping Culicover and Jackendoff (2005: 273) also adduce one example, in (46), for which they claim acceptability; to their example I add the attested examples in (47). (46)

(47)

2.4.1.4 Ellipsis in comparatives Kennedy and Merchant (2000a) argue that examples like (48a) involve a degree phrase extracting from a left-branch (here, attributive) position within a noun phrase, structurally parallel to (48b). (p. 34)

Page 18 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches (48)

Non-structural approaches have a ready explanation for this state of affairs, if one assumes that island effects come about only in movement structures. By parity of reasoning, one could have a structural account which eschews movement in these particular structures, as Chung et al. (1995) and Lobeck (1995) pursue, which derives the same effect. These structures are difficult only for ‘deletion’ approaches that fall under (13b.ii) and those null-structure accounts like Wasow’s and Williams’s that posit regular null structures as well.

2.4.2 Case mismatches Certain short NP answers display an unexpected case, given a simple equivalence between elliptical and non-elliptical structures. English subject questions can be answered with fragments in the accusative, where such pronouns would be ill-formed in non-elliptical sentences: (49)

Such mismatches are discussed in Morgan (1973), Barton (1990, 2006), and Progovac et al. (2006a). Note that while such mismatches are not found as robustly in sluicing, as Barros (2014b) in particular discusses, there do seem to be a handful of problematic cases (see the list in Vicente 2015).

2.4.3 Exceptions to the P-stranding generalization Although the P-stranding generalization seemed to hold across a substantial set of data, Merchant (2001) did note some apparent exceptions, such as that in (50) from Italian, remarking that “[i]n some cases and in some languages, it seems that speakers are willing to accept a bare wh-phrase in place of the PP, though I have not yet determined with sufficient clarity under what conditions this is possible, or whether or not this is a systematic property of a class of prepositions or languages” (Merchant 2001: 100). (50)

Page 19 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

This sub-area has been explored in more detail in recent years, with investigations of data in Serbo-Croatian by Stjepanović (2008, 2012), Brazilian Portuguese by Almeida and Yoshida (2007), a variety of Romance languages by Vicente (2008) and Rodrigues et al. (2009), Indonesian by Fortin (2007b), Polish and others by Szczegelniak (2006) and Nykiel and Sag (2008), and in several languages by van Craenenbroeck (2010a), Barros (2014b), and Barros et al. (2015). While Nykiel and Sag (2008) take such non-connectivity effects as in (50) to be straightforward support for a non-structural approach to ellipsis, most of these authors attempt to find some pattern in the putative counterexamples such that these examples are subject to a different analysis (and thus not undermining the structural account). Stjepanović (2008, 2012), Vicente (2008), Rodrigues et al. (2009), van Craenenbroeck (2010a), for example, variously point out that there are environments (even in languages like Spanish that otherwise seem to allow P-less wh-phrases in ‘regular’ sluices) which strictly enforce the P-stranding ban—that is, where the preposition becomes obligatory, as expected on a structural account. These environments include the remnants in gapping and pseudogapping, the counterweight to pseudocleft clauses, fronted CPs, and in sluices with else-modification (only the latter illustrated here): (51)

As these authors point out, it is exactly in these contexts that a copular source for the elided clause—what we may call pseudosluicing5—is unavailable, as seen in (52a); for this reason they suggest (in agreement with Szczegelniak 2006 for Polish) that the P-less ‘sluices’ in fact derive from a copular or reduced cleft-like source, as in (52b) (where material in angled brackets is elided): (52)

Page 20 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Positing cleft and copular sources for fragments and sluices has been successful in accounting for other restrictions as well, such as those found with non-intersective adjectives (from Barros 2014b: 31 and Ueno 2015: 119, respectively): (p. 36)

(53)

(54)

One kind of apparent exception to the P-stranding generalization is problematic in particular for the system of Jacobson (2016b), in which matching is enforced between the category of the correlate and that of the remnant. But in pairs like (55), the wh-phrase is a simple NP, while the fragment answer is a PP; crucially, the preposition at here is semantically vacuous. (55)

As I wrote in Merchant (2016), “while a movement-based deletion theory of ellipsis can make use of the usual mechanisms for handling the syntax/semantics mismatches that go under the rubric of reconstruction (here, the preposition at in the answer reconstructs), theories like that of Jacobson (2016)…predict that such pairs will be ill-formed” (as Jacobson herself acknowledges). In sum, the data from preposition stranding under sluicing (and fragment answers and elsewhere) are quite complex, and have become the subject of a rich vein of work. Whether or not the data uniformly support a structural analysis, clearly the subject forms an important area of debate. For analysts who pursue non-structural approaches, mismatches (or non-connectivity) are very welcome, since such mismatches would indicate that the constraints on preposition stranding in non-elliptical dependency Page 21 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches structures are independent of the constraints on the presence of prepositions in ellipsis. But one last point remains to be made in this respect, one which has not been made in the literature on these questions to my knowledge: if the conditions on P-stranding under non-elliptical wh-movement and those on prepositions in elliptical environments are in fact independent, what prevents us from expecting to find a language like the one below, call it ‘reverse-English’? (56)

‘Reverse-English’ would be like child spoken English (and many adult varieties, too) in requiring P-stranding in non-elliptical questions, but like some strict version of German in requiring the presence of the P in sluices in which the correlate of the wh-phrase is governed by a P. If these constraints are truly independent, such a language should strike us as just as natural as the Spanish that allows P-less sluiced wh-phrases. Since the nonstructural analyses make no reference to the conditions on wh-displacement, writing constraints that impose such a requirement should be simple. In total, this kind of evidence seems to favor structural approaches to some kinds of ellipsis. If these approaches are correct, we must search elsewhere for an explanation of the (p. 37) sometime lack of island effects (see Merchant 2008b, Abels 2009, Temmerman 2013, Bošković 2014, Griffiths and Lipták 2014, and Barros et al. 2015 for some recent approaches), and other non-connectivity effects that are sometimes adduced (such as the so-called ‘vehicle change’ effects of Fiengo and May 1994; Merchant 2013b). Several proposals have been advanced to account for these effects, which have been discussed extensively in the recent literature; with respect to deciding whether or not structure must be posited internal to the ellipsis site, it seems clear that it is completely unimportant which particular proposal for island and other repair effects is correct. While one can imagine many possible ways to account for repair effects, or the absence of expected grammatical sensitivities (one influential strand of thinking ties them all to properties of the syntax–phonology interface systems), it is essentially impossible to imagine an account of island effects and P-stranding that would make their presence in the elliptical structures seen here accidental or orthogonal to their presence in nonelliptical structures. At present, I see little prospect for building such a theory that would not essentially have to reimport the constraints needed for non-elliptical structure.

2.5 Null anaphora and ‘deletion’ Within structural approaches, two tacks can be discerned: the null anaphora approach and ‘deletion’. The latter ranges from the traditional formulation of a deletion transformation (as in Ross 1969b and Hankamer 1979 among many others) to more Page 22 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches recent proposals with implementational details more consistent with recent views of syntax and morphology. On all these approaches, the syntax of an ellipsis site is in general just the same as the syntax of its non-elliptical counterpart, but subject to some kind of operation or constraint which results in no pronounced material. In modern incarnations, the difference between an elliptical and non-elliptical XP, for example, is often cast solely in terms of the presence or absence of a feature in the structure which signals to the phonology that the phonological value of the XP is null (Merchant 2001; van Craenenbroeck and Lipták 2006; Vicente 2006; Ha 2008a; Toosarvandani 2008, 2009; Aelbrecht 2009; Corver and van Koppen 2010, 2011; van Craenenbroeck 2010b), or that Vocabulary Insertion does not take place at the level where the morphological structure is computed (Saab 2009; Saab and Zdrojewski 2012; Temmerman 2012; Merchant 2015a). Such a feature—call it the E-feature for ‘ellipsis feature’—should, ideally, be the sole repository of all information about the ellipsis. That is, it should have a syntax, a semantics, and a phonology. The syntax of this feature should serve to delimit what heads or other structures can host it (the ‘licensing’ question), the semantics could be used to impose an identity condition (see section 2.6 below: the Efeature is an anaphoric device that introduces a pointer that is resolved by re-using a derivation or its output, or triggering a search for an already constructed derivation or structure—e.g., anaphora to a meaning), and the phonology would be a trigger for a rule or constraint syncopating the phrase’s phonological value (or triggering non-Insertion on all dominated terminal nodes). There are several ways to imagine implementing such a feature, and different versions are pursued by different authors. The main advantage to such accounts is that nothing more need be (p. 38) said about the syntax, and all connectivity effects follow straightforwardly. For example, the origin site of the displaced wh-phrase in a sluicing example like (57) is inside the unpronounced sentential node: no additional, sluicing-specific mechanism need be employed to base-generate the phrase in specCP, nor to account for its selectional properties (the fact that angry idiosyncratically selects a PP headed by at): (57)

The [E] feature can be added to the feature matrix of the ‘licensing’ head (certain Cs for sluicing, certain Ts for VP-ellipsis, etc.), as in the following structures, or more complex relations between E and the elided material can be entertained (as in van Craenenbroeck and Lipták 2006 and Aelbrecht 2010). (58)

Page 23 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

(59)

The alternative is to insert an empty placeholder node in the structure, which acts like a null anaphor and which must be replaced at LF by full structure (on LF-copy approaches like Chung et al. 1995) or otherwise filled in or interpreted. One advantage of this family of approaches is that it assimilates the local licensing conditions on null VPs, TPs, and NPs as they appear in elliptical constructions to the more general licensing conditions on null elements; Lobeck (1995) pursues this line, as does Johnson (2001b) with a different emphasis. But this advantage may be entirely illusory, since the equivalent deletion approach using a feature can capture these restrictions with arguably the same level of sophistication: we simply place the same conditions on appearance on E that we did on pro. That is, we can call E an anaphoric element subject to the same licensing and discourse constraints on appearance that adhere to pro on theories that employ pro. It is no more nor less explanatory to claim that pro is subject to the Binding Theory by virtue of a featural specification [+pronominal, −anaphoric] than it is to say that E is subject to the same constraints by specifying it as [+pron.,−ana.]: this is just to claim that E is a discourse-anaphoric element with local syntactic requirements on its appearance. Lobeck’s attempt to (p. 39) reduce the distribution of elliptical e entirely to that of pro failed: that analysis was forced to supplement the conditions on pro with an extra codicil specifying that e must appear in a context of ‘strong agreement’, but the definition of strong agreement does not track anything independent: English each but not every must Page 24 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches be a ‘strong agreer’. In the end, the ‘strong agreers’ were all and only the heads that licensed ellipsis: obviously, this approach has nothing to recommend it over an equivalent listing of the heads that E can appear on (and the latter approach at least places interand intralinguistic variation in the class of head licensers in the most plausible domain: that of the lexicon).

2.6 The identity conditions on ellipsis The second major question arising in ellipsis concerns the identification of the elliptical material. That is, how does one calculate what the ‘missing’ material means? Clearly ellipsis is anaphoric, broadly speaking, and depends on its context to get its meaning: an ellipsis site apparently has no intrinsic lexical content at all. Theories that derive the attested meanings for ellipses do so on the basis of identity of the ellipsis to some antecedent, or of ‘parallelism’ or ‘resolution’ of the ellipsis, terms that are partially overlapping but generally equivalent for our purposes here. It is generally assumed that ellipsis requires an antecedent (perhaps only implicit or inferred, in the case of exophoric ellipses), on the basis of which the meaning is derived. But what kind of antecedent does ellipsis need? And what is the relation that must hold between an ellipsis and its antecedent? There are broadly three kinds of answers to these questions: ones that posit that the relation between the ellipsis and its antecedent involves a kind of identity of (or anaphora to) meaning, ones that posit a kind of identity of structure, and ones that use a bit of both. This terminology is meant to be neutral between theories that take the relation between an ellipsis site and its antecedent to be one of anaphoric ‘resolution’, recoverability, ‘parallelism’, or identity: for most purposes, these terms are picking out the same relation. Clearly what is not at stake is anything like surface identity, given examples like the following: (60)

(61)

(62)

Page 25 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Since the bare form of the verb following to in (60b) is not surface-identical to the past form of the verb in the antecedent VP, any identity relation that elided such a verb phrase in (60a) based on morphological or phonological identity with its antecedent would be clearly wrong. Likewise for the imperative and the progressive participle, as the pair in (61) shows. Finally, this point can be seen in an even more striking way when we consider ellipsis licensed across speakers using different languages, as in (62), which reports a conversation between two bilingual speakers of English and Greek; we must assume that the English ellipsis in B’s response in (62) is sensitive not to the overt form of the antecedent in Greek, but rather to more abstract properties not immediately obvious in the ‘surface’ form of the Greek (see Merchant 2015b). (p. 40)

2.6.1 Semantic identity and information structure Although the vast majority of the generative research on ellipsis in the years from 1965 to the mid 1990s (e.g. Chomsky 1965, Ross 1969b, Hankamer and Sag 1976, Sag 1976a,Williams 1977b, Hankamer 1979, Chao 1987, Lappin 1992, Rooth 1992b, Fiengo and May 1994, Chung et al. 1995, Lappin 1996, and many others) worked with the assumption that the identity relation was to be stated over phrase markers (whether Dstructure, deep structure, LF, or something else—often, it should be noted, faute de mieux), since the early 1990s ever more proposals have been made that state the identity relation over semantic representations or which take the resolution of ellipsis to be essentially semantic (Dalrymple et al. 1991, Jacobson 1992a, Hardt 1993, 1999, Gardent et al. 1998, Kempson et al. 1999, Asher et al. 2001, Ginzburg and Sag 2001, Merchant 2001, Hendriks 2004, Hendriks and Spenader 2005, van Craenenbroeck 2010b, Yoshida 2010, and many others; perhaps the earliest analysis in this vein is Keenan 1971, with Sag and Hankamer 1984 an important precursor as well). Sometimes the proponents of semantic approaches base their choice on the ability of these approaches to deal more directly with scopal interactions in ellipsis, and the distribution of strict and sloppy readings of pronouns (Dalrymple et al. 1991 is one such example). But such interactions are not necessarily a direct argument for a semantic identity relation, despite first appearances. First, these effects have been dealt with in syntactic identity approaches as well, sometimes with greater empirical success (see Fox 2000, for example). Second, taking such phenomena as arguing for or against any version of an identity condition on ellipsis is misguided. Tancredi (1992) showed conclusively that the problem of delimiting a number of phenomena traditionally thought to belong solely to the domain of ellipsis in fact formed merely a subpart of the problem of structuring discourse coherently, in particular with respect to focus and deaccenting. Thus traditional Page 26 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches concerns of strict vs sloppy identity, Tancredi showed, could not be addressed merely by looking at elliptical structures, but had to be approached from deaccented structures (similarly for scopal parallelism effects, the Dahl (many pronouns) puzzles, and the many clauses puzzles; see Fiengo and May 1994). Tancredi’s great contribution, and one whose impact is sadly often underestimated, was to show that theories of ellipsis per se did not have to deal with these (p. 41) phenomena at all, and that any theory of, say, the distribution of strict/sloppy readings that made reference to ellipsis was mistaken.6 Instead, the best arguments for semantic identity theories come from a large set of mismatches between the syntactic structure of the antecedent and that of the purported elided phrase. Some of these were the focus of Dalrymple (1991) and were enumerated at greater length by Fiengo and May (1994), who dubbed them ‘vehicle change’ effects. Although Fiengo and May (1994) use the term ‘vehicle change’ for about a dozen phenomena, I will illustrate only two here: pronoun/name equivalences, and polarity item/ non-polarity item equivalences. Important additional recent mismatch discoveries include Malagasy voice mismatches (Potsdam 2007), spading (van Craenenbroeck 2010b) and various pseudosluice phenomena (Barros 2014b), category switches (Fu et al. 2001; Johnson 2001b; Merchant 2013b), sprouted implicit arguments and adjuncts (Chung et al. 1995, 2011; Merchant 2001; AnderBois 2014), missing expressives (Potts et al. 2009), as well as the other phenomena discussed in chapter 1 of Merchant (2001) involving finiteness mismatches, word order, clitics, and other issues. Pronoun/name ‘vehicle change’ is illustrated by the following data (I illustrate only with names, though the problem is fully general and extends to all R-expressions); although sluicing and VP-ellipsis are licit in (63), the indicated coreference between the pronoun and c-commanded name in the presumptive non-elliptical equivalents in (64) is ruled out. (63)

(64)

A similar mismatch between grammatical ellipses and their ungrammatical putative nonelliptical counterparts is found with polarity items, as noted in Sag (1976a: 157–8) (and discussed in Merchant 2013b): (65)

Page 27 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

(66)

While the semantics of names and pronouns on the one hand, and polarity and non-polarity indefinites on the other, can reasonably be construed as equivalent (under a single assignment function g, if 〚he3〛g = Alex, then any proposition containing he3 evaluated with respect to g will have the same truth conditions as that proposition where (p. 42)

Alex replaces he3; likewise for the basic semantic contributions of polarity items), but it is difficult to see how he and Alex could be syntactically equivalent.

2.6.2 Syntactic identity Despite the success of semantic theories of elliptical identity, there are several sets of data that seem to require some syntactic identity. The first set of evidence I will mention here comes from the uneven distribution of voice mismatch effects in ‘big’ vs ‘small’ ellipses, and the second from certain morphological facts; see also Chung (2013) for an important set of facts from English and Chamorro sprouting.

2.6.2.1 Voice mismatch under ellipsis In ‘big’/high ellipses—viz., sluicing, fragment answers, gapping, and stripping—elided material and antecedent phrase must match in voice: if the antecedent clause is in the passive, then the elided clause must also be in the passive, and likewise for the active, mutatis mutandis. This is illustrated for sluicing in (67) (see Merchant 2013d for the other ellipsis types and data from additional languages). (67)

Page 28 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches In contrast to big ellipses like sluicing, ‘low’ or little ellipses allow voice mismatches: the relevant ellipsis type is VP-ellipsis in English (see Merchant 2008b for discussion of pseudogapping, which I omit here). The first attested example is from Hardt (1993); for further examples and discussion see Sag (1976a), Dalrymple et al. (1991), Fiengo and May (1994), Johnson (2001b), Kehler (2002), Arregui et al. (2006), Frazier (2008), Kim et al. (2011), San Pietro et al. (2012), and Merchant (2013d). (68)

The uneven distribution of these voice-matching effects does not seem to be arbitrary (in other words, it would be unexpected to find a language showing the reverse (p. 43)

pattern of English), and can be fairly straightforwardly understood given recent proposals for the syntax of voice following Rivero (1990) and Kratzer (1996) which separate a Voice head from the rest of the VP. This separation allows for the differentiated targeting of nodes for ellipsis: in high ellipses (sluicing, etc.), a clausal node that necessarily includes Voice; in low ellipses (VP-ellipsis), the verbal projection that is complement to (or inside the complement of) Voice. The structure for a representative example is given in (69). If the elided phrase XPE and its antecedent YPA must be identical, it’s obvious why ellipsis fails in the sluicing case in (69) (since TP deletion includes the Voice head and therefore TPA ≠ TPE) but succeeds with the articulated syntax in an ellipsis targeting a verbal projection (vP or VP) dominated by VoiceP (since the ellipsis excludes the Voice head, and so VPA = VPE). (69)

Page 29 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

It’s far less clear how current semantic identity proposals would handle this uneven distribution: most of them are designed to allow active/passive mismatches (such as Dalrymple et al. 1991 and Hardt 1993) and consider only VP-ellipsis data. Once the sluicing data are also brought into the picture, a uniform semantic analysis becomes harder to support.

2.6.2.2 Auxiliary form matching A second argument for syntactic identity in ellipsis comes from the exceptional behavior of be under ellipsis (Warner 1985; Lasnik 1995a; Potsdam 1997a; Roberts 1998b; Merchant 2015b). In general, verbs (both regular and irregular) don’t require morphological identity: (70)

(p. 44)

But forms of be do require morphological identity:

(71)

Page 30 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Lasnik (1995a) accounts for this distribution by positing that forms of be are inserted into the derivation fully inflected, while other verbs get their inflection in the course of the derivation. The syntactic identity therefore is met before the inflection of most verbs, but can never be met for differing forms of be, since they differ at every level of representation. This explanation, appealing as it may be, fails to account for the fact, noted in Merchant (2015b), that the identical effect appears when the antecedent is in Greek (or Spanish, or German); that paper proposes that the examples are ruled out by a non-parallel representation of the binding of the tense variables on be and on the predicate, precisely as long-distance and mixed bound readings are ruled out in the Dahl puzzle cases discussed in Fiengo and May (1994) and Fox (2000).

2.6.3 Hybrid theories There is a large amount of data that any theory of ellipsis needs to account for. Some of that data seem more amenable to a semantic treatment, and some to a syntactic one. For this reason, some researchers have proposed hybrid theories that incorporate both semantic and syntactic identity conditions, but impose them under differing conditions or selectively. Examples of such proposals include Kehler (2002) (though see Frazier and Clifton 2006 for critical discussion), Chung (2006, 2013), van Craenenbroeck (2010a), Chung et al. (2011), and Merchant (2013d). Some of these seek to embed a very narrow amount of syntactic sensitivity (perhaps just the case-assigning heads, or the Voice heads, or the heads that determine argument structure) into a broader semantic theory, others to make different elliptical constructions sensitive to different conditions. The proposal of Chung (2013), for example, includes the following conditions: (72)

Page 31 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

All of the new crop of hybrid theories promise to contribute to our growing understanding of what is meant by recoverability, and of how far the grammar can go in accommodating antecedents that fail to share local grammatical properties with their elliptical counterparts without simply letting any and every possible conceivable linguistic expression serve as a source for the computation of the anaphoric identity or resolution found in ellipses. (p. 45)

2.7 Conclusions Ellipsis continues to fascinate because its analysis goes directly to the heart of the main reason we study syntax: to discern the nature of the form/meaning correspondence. Theorizing in this domain requires one to tackle questions of basic ontology, and to make decisions about the nature of arguments for linguistic representations. Much work on ellipsis has taken it for granted that elliptical structures (and the way we derive their meanings) should be parallel to non-elliptical structures, and that theorizing about the two should be uniform. Indeed, this imperative underlies much work in theoretical linguistics more generally, and has been named the ‘structural uniformity’ assumption: Structural Uniformity An apparently defective or misordered structure is regular in underlying structure and becomes distorted in the course of derivation. (Culicover and Jackendoff 2005: 7) Culicover and Jackendoff (2005) take this assumption to task and argue that in the domain of ellipsis in particular, it leads to unnecessary positing of unpronounced structures. One may respond that their own proposals, which eschew any kind of unpronounced structure at all, are an instance of a different kind of uniformity assumption: Analytical Uniformity If a certain kind of meaning or use can be made in the absence of syntactic guides to that meaning or use, then syntactic guides are never needed for computing that meaning or use.7

Page 32 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches It is clear that, given the richness of the empirical database in ellipsis, the complexity of the analytical problems to which ellipsis gives rise, and the nature of the conclusions one can and must draw from the analysis of ellipsis, theorizing in this domain would do well to beware the ‘uniforms’ of any stripe.

Acknowledgements Great thanks to the excellent reviewers, and to the editors for their encouragement and patience, and to Jeroen van Craenenbroeck for many years of productive discussion of these issues.

Notes: (1) There are many other kinds of phenomena that go under the rubric of ellipsis as well, some better investigated than others, including argument drop, article drop, haplology, diary language and headlinese, subjectless infinitivals, copula drop, situational ellipses, small clauses, and many more; some are context-sensitive, and some are not. For various (and still incomplete) taxonomies of the missing, see Klein (1985) and Hennig (2013: 447– 8.). (2) I’ve omitted pseudogapping, a construction that seems to mix properties of gapping and VP-ellipsis; see Gengel (2013). (3) Kobele (2015) presents a system where the parser looks for memo-ized antecedents; the grammar, however, generates elided structures with an operation of deletion. (4) As Postal (2001) discusses, it is important to have a parasitic gap in this example, as it forces the missing VP to host a wh-trace, yielding the observed island effect. Without the parasitic gap, this example would be acceptable, as an example of ‘vehicle change’ (see Fiengo and May 1994). (5) The term pseudosluicing, as originally used in Merchant (1998) and van Craenenbroeck (2010b: 79–81), covered complex copular sources that included a cleft or cleft-like clause like it is X that…, with independent null subjects and copulas, but we can extend the term to cover ellipsis of simple copular clauses that lack the relative-clauselike part as well, such as simple it is X, irrespective of whether null variants of it and be are available in the language. (6) And note that a semantic identity theory need not accept the claim that there is no unpronounced syntactic structure: it’s perfectly consistent to claim that while ellipsis sites have syntactic structures, the fact that they are unpronounced is due to a semantic/ pragmatic requirement being satisfied. The structure question and the identity question are partially independent; see Table 2.1. (7) Alternatively, if some device D can relate a form F and meaning M, then whenever we have M, D is being used. Page 33 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis: A survey of analytical approaches

Jason Merchant

Jason Merchant is the Lorna P. Straus Professor of Linguistics and Vice Provost at the University of Chicago. He is the author of The Syntax of Silence, the co-editor of Sluicing: Cross-linguistic Explorations, and the author or co-author of more than two dozen articles on a wide variety of elliptical phenomena.

Page 34 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

Oxford Handbooks Online Ellipsis in Transformational Grammar   Howard Lasnik and Kenshi Funakoshi The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.3

Abstract and Keywords This chapter examines three themes concerning ellipsis that have been extensively discussed in transformational generative grammar: structure, recoverability, and licensing. It reviews arguments in favor of the analysis according to which the ellipsis site is syntactically fully represented, and compares the two variants of this analysis (the deletion analysis and the LF-copying analysis). It is concluded that the deletion analysis is superior to the LF-copying analysis. A discussion of recoverability follows, which concludes that in order for elided material to be recoverable, a semantic identity condition must be satisfied, but that is not a sufficient condition: syntactic or formal identity must be taken into account. The chapter finally considers licensing. It reviews some proposals in the literature about what properties of licensing heads and what local relation between the ellipsis site and the licensing head are relevant to ellipsis licensing. Keywords: ellipsis, transformational grammar, structure, recoverability, licensing

Howard Lasnik and Kenshi Funakoshi

3.1 Introduction ELLIPSIS has played a role in transformational generative grammar (TGG) since the inception of the approach in the 1950s. In the earliest model (Chomsky 1955) the vast majority of descriptive work was done by syntactic transformations. Phrase structure rules created basic monoclausal structures (often represented as trees), then two kinds of transformations dealt with these trees: Generalized transformations combined separate trees into one (sometimes altering them in the process); and singulary transformations operated on single trees, both basic ones and those created via generalized transformations. These singulary transformations inserted material, rearranged material, or deleted material, this latter kind of operation—labeled ‘deformation’ by Chomsky— being responsible for ellipsis. There have been many changes over the decades in the Page 1 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar technology of TGG but the core ideas survive: Sentences have a basic underlying structure (or set thereof) and a transformationally derived ‘surface’ structure more immediately related to phonological form. Questions about ellipsis have persisted through the various theoretical changes. ‘Recoverability’ is a long-standing theme. In classical TGG, the question concerned the precise nature of the identity requirement governing deletion, since it was known that total identity was not required. Chomsky (1965) and Ross (1969b) especially contributed to this discussion. In slightly later work, it was often assumed that an ellipsis site is empty in initial structure, and then the antecedent is copied in (the inverse of a deletion transformation) to create a fully structured Logical Form (LF) representation. The recoverability question then concerns the extent to which the antecedent can be altered in the course of being copied. Williams (1977b) and Fiengo and May (1994) were influential contributors to this perspective. The recoverability issue obviously intimately relates to the question of the exact nature of the ellipsis site. In the earliest work, it was simply assumed that the ellipsis site is syntactically fully represented, prior to deletion. But by the late 1960s, two alternatives began to be (p. 47) pursued, one involving syntactic structure (exactly how much varied), but no lexical items, in the ellipsis site, the other having no structure whatsoever (a possibility resurrected by Culicover and Jackendoff 2005). Ross (1969b) already presented strong arguments against this second possibility and at least some versions of the first. In addition to recoverability, ‘licensing’ is a requirement for ellipsis. Not just any XP can be elliptical (even in the presence of a suitable antecedent), and even an XP that can be elliptical can usually only be so in specific licensing contexts. This factor, unexplored in early work on ellipsis, became a significant research topic beginning in the 1980s, especially under the influence of Zagona (1982, and much subsequent work), Lobeck (1990), and Saito and Murasugi (1990a). Licensing still remains rather mysterious. In this chapter we will examine these three themes of structure, recoverability, and licensing as they developed in transformational grammar.

3.2 Structure Analyses of ellipsis can be classified into two fundamental approaches: (i) ones that assume that there is full internal structure in the ellipsis site at certain levels; (ii) ones that assume that there is neither internal structure nor an empty category in the ellipsis site. Let us call (i) the structural analysis and (ii) the non-structural analysis. In order to illustrate these analyses, let us consider a sluicing example like (1). (1)

Page 2 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar Sluicing was first explored by Ross (1969b), and has been discussed by van Riemsdijk (1978), Chao (1987), Lobeck (1991, 1995), Chung et al. (1995), Romero (1998), Ginzburg and Sag (2000), Lasnik (2001), Merchant (2001), and Culicover and Jackendoff (2005), among others. In the second conjunct in (1), the embedded wh-question clause (the sluiced clause) is missing except for the wh-phrase who. Under the structural analysis, the missing part of the sluiced clause is fully represented at certain levels. Thus, the embedding verb know takes CP as its complement, as illustrated in (2a). On the other hand, under the non-structural analysis, there is nothing in the missing part, and know directly selects the wh-phrase, as illustrated in (2b). (2)

In this section, we will review arguments in favor of the structural analysis over the nonstructural analysis, and then we will compare two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We will see that there are arguments for the deletion analysis, and the potential problems for the deletion analysis can be solved under specific analyses of sluicing.

(p. 48)

3.2.1 Arguments for the structural analysis 3.2.1.1 Case-matching effects The first argument for the existence of syntactic structure in the ellipsis site is based on case-matching effects in German sluicing, discussed by Ross (1969b), who credits George Williams, and explored in great detail by Merchant (2001). Consider the following examples: (3)

(4)

The verb schmeicheln ‘flatter’ assigns dative case while loben ‘praise’ assigns accusative case. (3) and (4) show that the wh-phrase in the sluiced clause (the remnant wh-phrase) must be assigned the same case that the verb in the antecedent clause assigns. This fact Page 3 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar can easily be accounted for if there is syntactic structure in the ellipsis site in sluiced clauses and an identity requirement obtains (see section 3.3), as the elided site then contains the same verb as the antecedent clause. On the other hand, if there is no structure in the ellipsis site, it is not obvious how to account for the matching requirement.

3.2.1.2 Agreement Another argument, also due to Ross (1969b), comes from subject–verb agreement. The verb that takes a sluiced clause as its subject exhibits singular agreement irrespective of the number marking of the remnant wh-phrase itself, as illustrated by (5). (5)

The remnant wh-phrase in (5a) is singular while that in (5b) is plural. In spite of this difference, the verb must exhibit singular agreement in both clauses. Such obligatory singular agreement is a general property of a clausal subject, as shown by (6). (6)

Example (6) shows that a clausal subject requires singular agreement on the verb (though there are some exceptions, as discussed by McCloskey 1991b). The similarity between (5) and (6) can be easily accounted for if there is syntactic structure in the ellipsis site. (p. 49)

3.2.1.3 Subcategorization The verb wonder takes an interrogative clause as its complement while it cannot take a DP, as shown in (7). (7)

Given this, Ross (1969b) argues that (8) indicates that there is syntactic structure in the ellipsis site in a sluiced clause.

Page 4 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar (8)

In the sluiced clause in (8), wonder is followed by a DP, how many men. If there is no structure in the ellipsis site and wonder takes the DP as its complement in (8), the acceptability of (8) is a mystery. On the other hand, if there is structure in the ellipsis site, and wonder in (8) takes an (unpronounced) interrogative CP complement, the acceptability of (8) is accounted for.

3.2.1.4 Preposition stranding Ross (1969b) observes that the possibility of P(reposition)-stranding in the sluiced clause correlates with that in non-ellipsis clauses. (9) shows that P-stranding is obligatory when the wh-phrase is the complement of with in a do-away-with construction.1 (9)

When the antecedent clause is the do-away-with construction, the remnant wh-phrase cannot appear along with the preposition with, as (10) shows. (10)

Under the structural analysis, the contrast between (10a) and (10b) and that between (9a) and (9b) can be treated in the same way since the sluiced clause in (10a) involves Pstranding while that in (10b) does not under the structural analysis, as illustrated by the following: (p. 50)

(11)

Merchant (2001) examines cross-linguistic data concerning the correlation between the possibility of P-stranding in sluicing and that in non-ellipsis contexts, and concludes that P-stranding is allowed under sluicing only if P-stranding is allowed in regular wh-movement (in non-ellipsis contexts).2,3 Merchant (2001: 92–100) shows that English, Frisian, Swedish, Norwegian, Danish, and Icelandic allow P-stranding both in sluicing and in regular wh-movement while Greek, German, Dutch, Yiddish, Russian, Polish, Czech, Bulgarian, Serbo-Croatian, Slovene, Persian, Catalan, Spanish, French, and Italian do not Page 5 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar allow it in either context. This cross-linguistic P-stranding correlation can also easily be accounted for under the structural analysis.

3.2.1.5 Idiom reconstruction Rottman and Yoshida (2013) provide further evidence for the structural analysis, based on idiom chunks. A part of an idiom chunk can undergo wh-movement, as shown in the following: (12)

In (12b), which strings, which is a part of the idiom chunk (pull strings), undergoes wh-movement. The sentence retains its idiomatic meaning. Rottman and Yoshida (2013) observe that a part of an idiom chunk can be a remnant of sluicing, as shown in the following: (13)

The sluiced clause in (13) retains its idiomatic meaning. If there is no structure in the ellipsis site, it is a mystery how it yields the idiomatic meaning.4 On the other hand, under the structural analysis, the sluiced clause in (13) is directly analogous to (12b): the mechanism via which the idiomatic meaning obtains in (13) is reduced to whatever mechanism induces the idiomatic meaning in ordinary wh-question sentences with overt structures like (12b). (p. 51)

3.2.1.6 Parasitic gaps Yoshida et al. (2015a) provide a new argument for the structural analysis of sluicing. Their argument is based on the parasitic gap construction, which is exemplified by (14) (Chomsky 1982, 1986; Engdahl 1983; Culicover and Postal 2001; Nunes 2004). (14)

This sentence involves two gaps. The first gap (ti) is called a true gap since it is in a position where extraction is normally permitted. The true gap is basically a trace left by overt A′-movement. The second gap (__PG) is called a parasitic gap. A parasitic gap is in a position where extraction is normally not allowed (e.g., adjunct islands and subject islands). A parasitic gap needs to be licensed by a true gap in its local domain. Thus, the sentence becomes unacceptable when there is no true gap, as shown in the following: (15) Page 6 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

The sentence also becomes unacceptable when there is a true gap but the true gap is not in a local configuration with the parasitic gap, as illustrated by the following example: (16)

Yoshida et al. (2015a) observe that a parasitic gap can appear in a sluiced clause, as shown by the following example:5 (17)

In this sentence, the antecedent clause contains a parasitic gap, which is licensed by the local true gap. What is crucial in this example is that the sluiced clause also involves a parasitic gap, which is contained in the remnant wh-phrase. Yoshida et al. (2015a) argue that the parasitic gap in the sluiced clause is licensed by a gap inside the syntactic structure that undergoes ellipsis. If there is no structure in the ellipsis site, it is unclear how the parasitic gap in the ellipsis clause is licensed. (p. 52)

One might argue that the relevant parasitic gap is licensed by the true gap not in

the elided structure but in the antecedent clause. However, the following example indicates that this is not the case: (18)

(18) does not involve sluicing. The gap in question that is contained in the wh-phrase is followed by overt material. If it were the true gap in the antecedent clause that licenses the parasitic gap in the sluiced clause in (17), (18) should be acceptable. To sum up, in order to license the parasitic gap in the sluiced clause in (17), we need to assume that there is a true gap in the sluiced clause, which means that there is structure in the ellipsis site in sluicing.

3.2.1.7 Extraction out of VPE site Williams (1977b) claimed that wh-movement out of an elided VP is generally impossible, giving the following example: (19) Page 7 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

However, in our experience, native speakers find this example to be fine. Many parallel examples appear in the literature, almost invariably labeled grammatical. Fiengo and May (1994) give the following: (20)

(21)

(22)

If there is internal structure in the elided VP, the source of the wh-phrase is straightforward, as is the existence of a trace for it to bind.

3.2.2 Deletion vs LF-copying We have seen that there is substantial evidence for the structural analysis over the nonstructural analysis. In this section, we will look at two variants of the structural analysis: the deletion analysis (Ross 1969b; Lasnik 2001; Merchant 2001) and the LF-copying analysis (Williams 1977b; Chao 1987; Lobeck 1991, 1995; Fiengo and May 1994). Both analyses assume that there is syntactic structure in the ellipsis site. They have different views regarding at which level of derivation structure appears. Under the deletion analysis, there is structure in the elided site at the beginning of the derivation, but it is deleted at PF. On the other hand, under the LF-copying analysis, a null pronominal-like element occupies the ellipsis site at first, and this null element is replaced at LF by syntactic (p. 53) structure, which is copied from some appropriate linguistic antecedent (see section 3.3 for what kind of linguistic antecedent is appropriate for this purpose). In section 3.2.2.1, we will see arguments in favor of the deletion analysis. In section 3.2.2.2, we first review two phenomena that are potentially problematic for the deletion analysis, and then we discuss possible analyses of them that are compatible with the deletion analysis.

3.2.2.1 Arguments for the deletion analysis As Ross (1969b) and Merchant (2001) contend, the case-matching effects that we presented in section 3.2.1.1 as an argument for the structural analysis also tend to argue for the deletion analysis of sluicing. Under the deletion analysis, the deletion operation applies to IP after the remnant wh-phrase undergoes wh-movement.6,7 Therefore, the case that is assigned to the remnant wh-phrase is naturally maintained after the wh-phrase moves. On the other hand, under the LF-copying analysis, where the ellipsis Page 8 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar site is empty throughout overt syntax, the remnant wh-phrase is base-generated in specCP. Thus, some extra mechanism is required to account for the case-matching effects. The P-stranding generalization also favors the deletion analysis over the LF-copying analysis, since it straightforwardly follows from the deletion (plus movement) analysis while it does not from the LF-copying (plus base generation) analysis.

3.2.2.2 Potential arguments against the deletion analysis However, there are potential problems for the deletion analysis. In this section, we first review two of them, and then we discuss possible approaches to them that are compatible with the deletion analysis (section 3.2.2.3). Island violation repair by ellipsis. Islands effects are not observed (or at least are significantly lessened) under sluicing in various circumstances. For example, Ross (1969b) observes that sluicing is not as sensitive to Complex NP islands as regular wh-movement is, as shown in (23). (23)

As Culicover and Jackendoff (2005) argue, this is problematic for the deletion (plus movement) analysis. Under the deletion analysis, the ellipsis clause in (23b) involves wh-movement out of a complex NP island, as illustrated by (24). (24)

Thus, under the deletion analysis, it might be predicted that (23b) is as unacceptable as (23a), since the derivation involves an island violation. On the other hand, the LF-copying analysis correctly, or nearly correctly, predicts that (23b) is acceptable since the remnant wh-phrase is base-generated in specCP (i.e., wh-movement is not involved). As Ross (1969b), Chung et al. (1995), Lasnik (2001), and Merchant (2001) discuss, this island-repair-by-ellipsis phenomenon is widespread, applying to several other islands such as wh-islands, left-branch islands, derived position islands, and coordinate structure islands.8 We will discuss a possible explanation of island violation repair phenomena under the deletion analysis in section 3.2.2.3. (p. 54)

Multiple sluicing. Nishigauchi (1998), Richards (2001), and Lasnik (2014) discuss another kind of repair by ellipsis. Unlike languages like Bulgarian and Serbo-Croatian, English does not allow multiple wh-fronting: (25) Page 9 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

In the second sentence in (25), the two wh-phrases (which and from which) are fronted, yielding the unacceptability. Surprisingly, the multiple wh-fronting sentence becomes acceptable once sluicing applies to it, as shown in (26). (26)

Nishigauchi (1998) suggests that sentences like (26) are similar to gapping, proposing that the first wh-phrase is in specCP while the second one is in some other position. However, Richards (2001) points out that multiple sluicing in English exhibits different properties from gapping. For example, in gapping constructions, the gapped clause must be conjoined with the corresponding non-gapped clause, in fact IP with IP, as illustrated by the following examples: (27)

In these sentences, the second bracketed clauses are the gapped clauses and the first ones are the corresponding non-gapped clauses. In (27a) and (27b), the gapped IPs are conjoined with the corresponding non-gapped IPs, yielding the acceptable gapped clauses. On the other hand, in (27c) and (27d), the two IPs are not conjoined, resulting in illegitimate gapped clauses. (p. 55)

As (28) shows, multiple sluicing is not subject to such a strict locality condition. (28)

This indicates that multiple sluicing cannot be reduced to gapping. If a sentence like (28) is a genuine instance of multiple sluicing, as Merchant (2001) and Richards (2001) argue, this poses a potential problem for the deletion (plus movement) analysis of sluicing; it is a Page 10 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar mystery why multiple wh-fronting is allowed in English only under sluicing. We will discuss a possible account for multiple sluicing that is compatible with the deletion analysis in section 3.2.2.3.

3.2.2.3 Explanations under the deletion analysis Explanations of island repair.9 Ross (1969b: 277) takes island violation repair as evidence that “the theoretical power of [global] derivational constraints is needed in linguistic theory”; an ungrammatical sentence obtains if a syntactic island is crossed by movement, and the island appears in surface structure while a lesser degree of the ungrammaticality results if the island does not appear in surface structure as a result of a deletion transformation. Rejecting such global derivational constraints, Chomsky (1971) presents a representational constraint to account for island violations and their repair by ellipsis. He proposes that an island is marked by * (# in Chomsky’s 1972a presentation) when it is crossed by movement, as illustrated by the following, where a complex NP island is crossed. (29)

An output condition rules out a representation with * in surface structure, explaining the unacceptability of standard island violations. Chomsky (1972a) further proposes that if a later operation deletes a category containing *, the derivation is salvaged, explaining island violation repair by ellipsis.10 Merchant (2001) gives an empirical argument against the *-marking analysis. He points out that island violations are not always repaired by ellipsis. For example, in (30), which moves out of a complex NP island, and the crossed island is deleted by VP-ellipsis. In spite of this, the sentence is unacceptable (i.e., the island violation is seemingly not repaired). (p. 56)

(30)

This contrasts with sluicing on the face of it, as shown in the following: (31)

Under the *-marking analysis, the contrast between VP-ellipsis and sluicing in this respect is not expected.11

Page 11 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar Merchant (2001) argues that all ellipsis is a deletion operation at PF and classifies islands into two types: PF islands (e.g., left-branch islands, COMP-trace effects, and derived position islands) and propositional islands (e.g., complex NP islands, adjunct islands, and sentential subject islands). PF islands are constraints that apply to PF representations. Therefore, ellipsis as a deletion operation at PF can get rid of illegitimate PF representations. Merchant (2001) argues that the case of apparent repair of propositional islands involves a smaller subpart of the antecedent that does not contain an island. For example, (31) is reanalyzed as in the following: (32)

In contrast with sluicing, the case of VP-ellipsis (i.e., (30)) cannot be analyzed as involving an alternative short source that does not contain an island. Thus, the contrast between (30) and (31) follows.12 Explanation of Multiple Sluicing. Lasnik (2014) argues that multiple sluicing in English involves wh-movement of the first wh-phrase and rightward movement of the second wh-phrase, as illustrated by the following: (33)

Lasnik’s (2014) arguments for the rightward movement analysis are based on the observation that multiple sluicing in English is subject to constraints on rightward

(p. 57)

movement in general, rather than those on wh-movement. First, the second wh-phrase in multiple sluicing prefers to be a PP rather than a DP, as illustrated by the following: (34)

The same preference for PP over DP is found in rightward movement, as shown in the following: (35)

Second, it is well known that rightward movement of DP improves when the DP is made ‘heavier’, as illustrated in the following (see Ross 1967 and Fiengo 1980 among others): (36) Page 12 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

Likewise, multiple sluicing involving DP as the second wh-phrase improves when the DP is made heavier, as shown in the following: (37)

Third, rightward movement does not allow P-stranding, as illustrated by the following (see Ross 1967 and Drummond et al. 2010): (38)

The second wh-phrase in multiple sluicing is subject to the same constraint: (39)

Finally, rightward movement is subject to the Right Roof Constraint (Ross 1967) that rightward movement out of an embedded clause is prohibited, as shown in the following: (40)

Likewise, the first wh-phrase and the second one in multiple sluicing cannot be separated by a clause boundary, as (41) shows. (p. 58)

(41)

Given these similarities between rightward movement and multiple sluicing, Lasnik (2014) concludes that multiple sluicing in English involves rightward movement of the second wh-phrase. This analysis is compatible with the deletion analysis of sluicing. Thus,

Page 13 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar multiple sluicing ceases to be problematic for the deletion analysis under Lasnik’s (2014) account.

3.2.3 Summary of section 3.2 In this section, we reviewed arguments in favor of the structural analysis of ellipsis over the non-structural analysis, and then we compared the two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We saw that there are arguments for the deletion analysis, and that potential problems for the deletion analysis (island repair and multiple sluicing) can be solved under Merchant’s (2001) and Lasnik’s (2014) proposals.

3.3 Recoverability Elided material must be recoverable somehow. It is generally agreed that elided material must be identical to its antecedent in order to guarantee the recoverability of ellipsis. However, the precise formulation of the identity condition of ellipsis is controversial. Some argue that elided material and its antecedent must be syntactically or formally identical (Chomsky 1964, 1965; Sag 1976a; Williams 1977b; Kitagawa 1991; Fiengo and May 1994; Chung et al. 1995; Lasnik 1995b; Fox 2000; Fox and Lasnik 2003; Chung 2006; Merchant 2008a; among others) while others argue that they must be semantically or pragmatically identical (Dalrymple et al. 1991; Hardt 1992c; Prüst 1993; Ginzburg and Sag 2000; Merchant 2001; Yoshida 2010; Weir 2014, among others). In this section, we review some proposals under syntactic identity and semantic identity. First, we will present arguments for syntactic identity in section 3.3.1, and then we will look at arguments against syntactic identity and for semantic identity in section 3.3.2.

3.3.1 Syntactic identity The first relevant discussion of ellipsis appears in Chomsky (1955). He begins by considering the following examples (p. 423): (42)

For Chomsky, the derivation of such sentences involves a generalized transformation (conjunction), a permutation (exactly the same one used in simple yes–no questions), and a ‘deformation’ (deletion operation) eliminating the material after the first auxiliary. The first mention of an identity requirement appears here, but it is not the (p. 59)

Page 14 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar requirement one might expect (material to be deleted identical to an antecedent). Rather, Chomsky suggests that the auxiliaries must be identical: (43)

“…the auxiliary phrases must be identical in the two conjuncts…for the transformation… to apply.” That is, material outside the ellipsis site must be identical, an interesting possibility discussed briefly much later by Fox and Lasnik (2003) and Lasnik and Park (2013). Simple VP-ellipsis also comes under investigation, including now classic instances such as: (44)

Again, Chomsky notes that “these appear to be ‘truncated’ sentences, with a verb phrase ‘understood’” and he indicates that formal analysis leads to the same conclusion, thus “providing grounds in formal linguistic structure for this intuition.” In particular, precisely the same analysis of the English verbal system that Chomsky had motivated to account for yes–no question alternations is efficacious in describing the ellipsis patterns. Chomsky does point out an interesting problem that arises, an apparent ordering paradox. In the derivation of (44c), have and the perfect morpheme en were introduced together in the Aux. So to derive the correct output, it must be that the rule separating en from have and attaching it to the following verb (called Affix Hopping in later work) must precede the deletion rule, since that affix is included in the deletion site. (And likewise for (44b).) On the other hand, the tense affix in (44d) survives the deletion, so in this case, deletion must precede Affix Hopping. Chomsky’s solution is to propose that there are two Affix Hopping rules, one preceding deletion, for en and ing, and another for tense, which follows deletion. Another possibility is that Affix Hopping is the right account for tense morphology, but not for aspectual morphology, the latter being ‘lexicalist’. This is not unlikely since the strong evidence for syntactic separation of tense does not obtain for other pieces of inflectional morphology. More on this possibility, suggested by Lasnik (1995b), later in this subsection. A syntactic recoverability condition on elided material is first explicitly articulated by Chomsky (1964). There are two instances. First, a “designated representative of a category” may be deleted. For instance, Chomsky hints that indefinite pronouns (‘designated representatives’ of the category NP) underlie interrogative ones and thus Page 15 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar are deletable by the interrogative transformation. The second instance is more familiar: apart from a designated representative, “a transformation can delete an element only if… the structural condition that defines this transformation states that the deleted element is structurally identical to (p. 60) another element of the transformed string” (p. 41). This formal identity condition guarantees that “A deleted element is…always recoverable.” Chomsky (1965) explores this recoverability condition further. He first suggests, subject to later refinement, that “an erasure operation can use the term X to delete Y just in case X and Y are identical” (p. 145). Chomsky then considers the notion ‘identity’, indicating that it might be taken to mean “strict identity of feature composition” (p. 177). But he then gives a pair of examples showing that this cannot be quite right: (45)

(46)

Assuming a clausal source for comparatives, (45) works as expected, with is clever deleted under complete identity. But, Chomsky observes in (46) the antecedent for deletion of is clever is are clever, which is not fully identical. The problem is even more evident in the French analogue, where the deleted adjective is not completely identical to its antecedent, differing in the feature ±plural, like the copula, and also in being +feminine: (47)

Based on this, Chomsky makes the intriguing suggestion that features added by agreement rules do not have the same status as features that are inherent parts of lexical items: …in the case of Adjectives and the copula (also Verbs, which take part in similar rules) the inflectional features that are added by agreement transformations are apparently not considered in determining whether the item in question is strictly identical with some other item. (p. 180) Even the plural feature of a nominal expression is not always inherent. Predicate nominals show the pattern of verbs and adjectives, as seen in (48)—reasonable since in this instance as well, the feature is triggered by agreement.13 (48)

Page 16 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar Based on this, Chomsky offers two conclusions: First, features introduced by transformation into lexical formatives are not to be considered in determining when deletion is permitted; a formative, in other words, is to be regarded as a pair of sets of features, one member of the pair consisting of features that are inherent to the lexical entry or the position of lexical insertion, the second member of the pair consisting of features added by transformation. Only the first set is considered in determining legitimacy of deletion in the manner previously described. Second, what is involved in determining legitimacy of deletion is not identity but rather nondistinctness in the sense of distinctive feature theory…Thus consider once again the case of ‘I know several lawyers’—‘Bill is a (p. 61) lawyer.’ The Predicate-Nominal of the latter is not singular, in the base structure; rather, it is unspecified with respect to number exactly as the nasal is unspecified with respect to point of articulation in the lexical representations of the formatives king, find, lamp, etc. Hence, it is not identical with the corresponding nominal element of ‘I know several lawyers’; it is, rather, nondistinct from it, and the example suggests that this is sufficient to permit deletion. (p. 181) Finally, Chomsky argues that this analysis need not be stated or even mentioned in any specific rules of grammar, since: …it is, apparently, determined by a general convention regarding the form of grammar. In other words, we are tentatively proposing it for consideration as a linguistic universal, admittedly, on rather slender evidence…If this proposal is a correct one, then the analysis of formatives that we have suggested is a general condition on the functioning of erasure transformations. (pp. 181–2) In a further consideration of formal identity, Lasnik (1995b) presents an account of a surprising asymmetry in English VP ellipsis first noticed by Warner (1986) and rediscovered by Colbert (2007). Lasnik’s account is based on a ‘hybrid’ approach to English verbal morphology, initially motivated by the divergent behavior of auxiliary and main verbs with respect to negative placement and Subject-Aux Inversion. In this hybrid approach, main verbs come to be associated with their verbal morphology (particularly their tense morphology) via Affix Hopping. Auxiliaries, on the other hand, are ‘lexicalist’, introduced into the syntactic derivation already fully inflected. The core of the ellipsis asymmetry is that finite forms of main verbs, but not of be, can apparently antecede deletion of infinitival forms: (49)

(50)

Page 17 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar If, as Lasnik argued, slept is underlyingly sleep with a syntactically independent tense morpheme (as in Chomsky 1955 and 1957), there is a stage in the derivation where slept and sleep are completely identical. On the other hand, if was is was throughout the derivation, there is no stage where it is identical to be, thus excluding (50), if there is a strict formal identity requirement on ellipsis:14 (51)

In further motivation of a formal identity requirement, Chung (2006) observes an extremely interesting constraint on ‘sprouting’, a type of sluicing where there is no (p. 62)

apparent antecedent for the wh-trace (see Chung et al. 1995): (52)

(53)

It should be noted at the outset that sprouting raises non-trivial difficulties for any identity requirement, whether syntactic or semantic, as noted by van Craenenbroeck and Merchant (2013). One possibility briefly suggested by van Craenenbroeck and Merchant (2013) is that there is a null adjunct or null argument that does serve as an antecedent. We briefly return to this question in 3.3.2.1. Chung’s discovery is that when the missing antecedent would have been a PP, the sluicing fragment must also be a PP, and not a mere DP: (54)

(55)

(56)

(57)

It is worthy of note that even when the preposition is predictable, as in some of the above examples, it still must show up in the sprouting construction. This is in sharp contrast Page 18 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar with standard sluicing (in a language allowing preposition stranding, like English). In these instances, the preposition is optional (and probably even dispreferred) in the fragment: (58)

(59)

The presence of some prepositions seems entirely formally motivated, as in (58) but even these must show up in sprouting. Significantly, this pattern in Chung’s examples is not unique to English. It shows up, as Chung points out, in other languages with preposition stranding, such as Danish and Norwegian. Chung concludes that we must look beyond semantics and pragmatics to account for the contrasts. That is, there is a formal identity condition at work here. Merchant (2008a) presents another argument for a syntactic identity requirement on ellipsis. Merchant observes that sluicing doesn’t tolerate active–passive mismatches:15 (60)

(p. 63)

(61)

On the face of it, this is rather strong evidence for a syntactic identity requirement, since actives and corresponding passives are very close to semantically parallel. But there is evidently a glitch, since VP ellipsis does seem to tolerate active–passive mismatches: (62)

(63)

Merchant’s insightful solution to this nearly paradoxical situation reaffirms syntactic identity. He proposes that the functional head responsible for active and passive voice is included in a clausal ellipsis site, but is outside of a VPE site. Thus in the case of VPE, antecedent and ellipsis site are syntactically identical (neither active nor passive), while Page 19 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar in the case of sluicing they are not, thus accounting for the contrast between (60), (61) on the one hand and (62), (63) on the other.16

3.3.2 Semantic identity 3.3.2.1 Arguments against (strict) syntactic identity Merchant (2001: 19–25) provides arguments against syntactic identity. The first argument, alluded to in section 3.3.1, concerns sprouting: (64)

In these examples, there is no overt element in the antecedent clauses that corresponds to the remnant wh-phrase (what in (64a) and when in (64b)). The IPs in the antecedent clauses and those in the sluiced clauses are therefore structurally different. Thus, examples like these should be unacceptable under a strict syntactic identity condition. Merchant (2001) also argues that sluicing in Romanian poses a problem for syntactic identity. In Romanian, a D-linked wh-phase such as care ‘which’ must be accompanied by a clitic, as illustrated by the following (see Dobrovie-Sorin 1993; Grosu 1994: 224): (65)

In this example that involves the D-linked wh-phrase pe care băiat ‘ACC which boy’, the clitic l- ‘him’ is required. Merchant (2001) observes that a D-linked remnant wh-phrase in sluicing does not require a clitic in the antecedent clause, as shown in the following: (p. 64)

(66)

There must be a clitic associated with the D-linked remnant wh-phrase in the sluiced clause. In spite of this, there is no clitic in the antecedent clause. This means that the elided IP and the antecedent IP are structurally different. Thus, (66) should be unacceptable under the syntactic identity condition, contrary to fact. Finally, Merchant (2001) argues that what Fiengo and May (1994) call vehicle change phenomena pose a problem for a strict syntactic identity condition, which incorrectly

Page 20 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar predicts that the following sentence should be unacceptable due to a Condition C violation: (67)

Under a strict syntactic identity condition, the ellipsis clause headed by though in (67) must be as in the following: (68)

This sentence contains an R-expression Alex that is bound by he, violating Condition C. Therefore, (67) should be unacceptable under a strict syntactic identity condition.17 A further argument against syntactic identity comes from a growing body of literature that provides evidence that a ‘cleft’ can be an underlying structure for ellipsis when the antecedent does not contain a cleft18 (Szczegelniak 2006, 2008; Vicente 2008; Rodrigues et al. 2009; Barros and van Craenenbroeck 2013; see also van Craenenbroeck 2010a). (p. 65) For example, Barros and van Craenenbroeck (2013) argue that tag questions provide evidence for the possibility of clefts as an underlying source for ellipsis. As (69) indicates, tag clauses and assertion clauses that are followed by them are subject to a strict identity condition. (69)

Given this property of tag questions, Barros and van Craenenbroeck (2013) argue that the ellipsis site of the fragment answer in (70B) has a cleft structure rather than a structure isomorphic with the antecedent. (70)

This is problematic with a strict syntactic identity condition, under which the ‘wasn’t it’ version of (70B) is expected to be unacceptable.19

3.3.2.2 Merchant’s semantic identity Given the empirical problems with syntactic identity, Merchant (2001) argues that there is no syntactic identity requirement at all, and instead proposes a semantic identity Page 21 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar condition on ellipsis. Informally, Merchant’s (2001) semantic identity condition amounts to requiring that the antecedent entail the elided element, abstracting away from focused material that it contains, and the elided element also entail the antecedent, again abstracting away from (p. 66) focused material.20 With this condition, let us consider (64a), the representation of which under the deletion analysis is given in (71): (71)

There are no focused portions either in the antecedent IP or the elided IP. Given that the trace of wh-movement is translated as an existentially bound variable, the elided IP roughly means that there is something Abby was reading. The antecedent IP also entails that there is something Abby was reading. The elided IP and the antecedent IP entail each other, satisfying Merchant’s semantic identity condition. Therefore, the semantic identity condition can account for sentences like (64a). Merchant’s semantic identity condition also straightforwardly accounts for the Romanian example with a D-linked wh-phrase (66) because the antecedent and the elided element in this example are semantically equivalent, hence entailing each other. The vehicle change phenomena can also be accommodated under Merchant’s semantic identity. Suppose that the elided VP in (67) contains the pronoun him rather than the Rexpression Alex, as illustrated by the following: (72)

This representation does not violate Condition C. The semantic identity condition allows ellipsis like this because the antecedent VP and the elided VP are semantically equivalent when the pronoun him refers to Alex. Yoshida (2010) provides further evidence for semantic identity on the basis of what he calls the sluicing in adjunct (SIA) construction, exemplified by the following: (73)

In these examples, the adjunct clauses undergo sluicing. Yoshida (2010) argues that in the SIA construction, the antecedent for the elided IP in the adjunct clause is the matrix VP (p. 67) rather than the matrix IP. If this is correct, the SIA construction is problematic for syntactic identity because a VP and an IP are evidently different in structure. In what follows, we review Yoshida’s (2010) argument that the antecedent in the SIA construction is the matrix VP rather than the matrix IP.

Page 22 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar The only available interpretation for the adjunct clauses headed by without in (73) is (74). (74)

This means that the elided material in (73) does not contain negation or must. This contrasts with ordinary sluicing. As shown in (75), the only available interpretation is the one in which negation and must are interpreted in the ellipsis sites. (75)

That the elided IPs in (73) do not contain negation or must means that the matrix IPs do not serve as the antecedents for the elided IPs because IPs contain negation and modal verbs. If the antecedents for the elided IPs were the matrix IPs, negation or must would be interpreted in the ellipsis site both under the syntactic identity approach and under the semantic identity approach. On the other hand, if the matrix VPs rather than the IPs serve as the antecedents, the SIA construction ceases to be problematic under the semantic identity condition. Suppose that the adjunct clauses headed by without in (73) are adjoined to VP,21 and that subjects are base-generated in specVP (the VP-internal subject hypothesis: Zagona 1982; Fukui and Speas 1986; Kitagawa 1986; Kuroda 1988; Koopman and Sportiche 1991), and that they leave copies in that position when they raise to specIP. Then, in (73b), for example, the interpretation of the matrix (lower segment of) VP is ∃x. John loves x, and that of the elided IP is ∃x. he loves x. They are semantically equivalent.22 Thus, sluicing in (73b) satisfies Merchant’s semantic identity. Note that this account of the SIA construction is possible only under the semantic identity approach since VP and IP can be identical only semantically, not syntactically. (p. 68) Therefore, the SIA construction provides an argument for the semantic identity approach.

3.3.3 Summary of section 3.3 To sum up, there are arguments for syntactic identity and semantic identity. Given Lasnik’s (1995b) English verbal morphology argument, Chung’s (2006) sprouting argument, and Merchant’s (2008a) voice mismatch argument, a semantic identity condition is not a sufficient condition on ellipsis. Furthermore, Merchant’s (2001) sprouting, Romanian sluicing, and vehicle change arguments and Yoshida’s (2010) SIA construction argument indicate that a syntactic identity condition should not be too strict, as Chomsky (1965) already pointed out.

3.4 Licensing It is not the case that ellipsis is possible whenever an identity condition is satisfied. Consider the following sluicing examples: Page 23 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar (76)

In (76a) the elided IP follows the remnant wh-phrase, while in (76b) it follows the complementizer. In both examples, (both syntactic and semantic) identity is satisfied. In spite of this, (76a) is acceptable while (76b) is not. Therefore, there is something other than an identity condition to satisfy, in order for ellipsis to be licensed. There seems to be a near consensus that ellipsis requires the presence of a licensing head (Zagona 1982, 1988a, 1988b; Chao 1987; Lobeck 1990, 1993, 1995; Saito and Murasugi 1990a; Johnson 2001b; Martin 2001; Merchant 2001, 2004a; Gergel 2006; Aelbrecht 2009): in order to be licensed, the ellipsis site must be in a local relation with a head with specific morphosyntactic properties. In this section, we review some proposals in the literature about what properties of licensing heads and what local relation between the ellipsis site and the licensing head are relevant to ellipsis licensing.

3.4.1 Licensing heads 3.4.1.1 Agreeing heads Lobeck (1990, 1995) and Saito and Murasugi (1990a) argue that agreeing functional heads license ellipsis of their complements, based on Fukui and Speas’ (1986) notion of Kase. (p. 69) C with the feature [+Wh], I with the feature [+Tns/+Agr], and D with the feature [+Poss] or [+Plural] assign Kase to their specifier, as summarized in the following table: (77)

Kase assigners are agreeing functional heads while non-Kase assigners are non-agreeing functional heads. D with [+Poss] or [+Plural] licenses ellipsis of its complement (i.e., NP), as illustrated by the following examples (this is an instance of so-called N’-ellipsis or NPellipsis; see Saito and Murasugi 1990a):23 (78)

Page 24 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

Given that genitive case is assigned by D with [+Poss], NP-ellipsis in (78a) is licensed by D with [+Poss]. A plural demonstrative is a D specified with [+Plural]. Thus, the D in (78b) licenses NP-ellipsis. On the other hand, NP-ellipsis is not licensed by a non-agreeing functional head, as illustrated by (79). (79)

Neither the, a, nor this is specified with [+Poss] or [+Plural], hence are not agreeing heads. Thus, NP-ellipsis in (79) is not licensed. Furthermore, the contrast between (76a) and (76b), repeated here as (80a) and (80b) respectively, can be attributed to the distinction between agreeing heads and nonagreeing heads. (p. 70)

(80)

Given that C whose specifier is filled with a wh-phrase is specified with [+Wh] while C filled with a complementizer is not, the elided IP is licensed in (80a) but not in (80b). This is why the elided IP must follow a wh-phrase. An elided VP must be licensed by I with [+Tns/+Agr]. This is illustrated by the following:24 (81)

Page 25 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar Given that only I that is lexically filled is specified with [+Tns/+Agr], I filled with a modal verb or with pleonastic do licenses VP-ellipsis while I that is not lexically filled does not. Non-finite I in raising constructions is not an agreeing head because it lacks [+Tns/+Agr]. Thus, it does not license VP-ellipsis, as shown in the following examples: (82)

However, as Saito and Murasugi (1990a) and Takahashi (1994) observe, VP-ellipsis is licensed in control infinitivals, as illustrated by the following: (83)

According to Martin (2001), control to, unlike raising to, is an agreeing functional head since it assigns null case to PRO in its specifier. Given this, the difference between raising to and control to can be reduced to the agreeing/non-agreeing distinction.

3.4.1.2 Heads with strong features Martins (1994) argues that functional heads with a strong feature, which triggers overt verb movement, license VP-ellipsis in Romance languages, based on her observation that there is a correlation between the possibility of enclisis in finite clauses and that of VPellipsis in (p. 71) Romance languages. Portuguese and Galician allow enclisis in finite clauses while Spanish, Catalan, French, and Italian do not, as shown in the following examples: (84)

(85)

Page 26 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

Martins (1994) observes that the cross-linguistic difference in terms of the possibility of enclisis is correlated to another cross-linguistic difference in terms of the possibility of VP-ellipsis: Languages that allow enclisis also allow VP-ellipsis while languages that do not allow enclisis do not allow VP-ellipsis. By VP-ellipsis, Martins (1994) means Vstranding VP-ellipsis. V-stranding VP-ellipsis is VP-ellipsis that applies to a VP whose head has moved out of it (see Otani and Whitman 1991; Lasnik 1997; Potsdam 1997a; Ngonyani 1998; Doron 1999; Cyrino and Matos 2002; Goldberg 2005; Toosarvandani 2009; Funakoshi 2012, 2014, 2016; Gribanova 2013b; Abe 2014; Hayashi 2015; a number of these works show that mere object drop does not suffice to account for all the observed phenomena). When it applies to a transitive sentence, V-stranding VP-ellipsis yields a null object sentence, as illustrated by the following: (86)

As (87) shows, VP-ellipsis is possible in Portuguese and Galician, which allow enclisis. (87)

On the other hand, languages that do not allow enclisis do not allow VP-ellipsis, as shown in (88). (p. 72)

(88)

Page 27 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

In order to account for the correlation between enclisis and VP-ellipsis, Martins (1994) proposes that enclisis results when a finite verb moves to a functional head (∑ in her terms; see also Laka 1990), and verb movement is triggered by a strong feature on the functional head. This is illustrated by the following schematic structure (Cl stands for a clitic):25 (89)

The correlation follows if we further assume that VP-ellipsis must be licensed by a functional head with a strong feature.26

3.4.1.3 Potential problems We have seen that ellipsis requires a licensing head, and a licensing head must be specified with certain features (Kase in English and a strong feature in Romance languages).27 This properly describes some facts, but not all, about ellipsis. For example, negation licenses VP-ellipsis in English, as Potsdam (1997a) points out (see also Baltin 1993 and Lobeck 1995): (90)

Even if we assume, following Pollock (1989), that sentential negation is the head of NegP, it is hard to assume that it is specified with Kase (i.e., it is an agreeing head). This means that Kase is not a necessary condition for a licensing head, as we already indicated in n. 23. Furthermore, a strong feature is not a sufficient condition on licensing V-stranding VP-ellipsis. In Danish, a V-2 language, verb movement is triggered by a strong feature. In spite of this, V-stranding VP-ellipsis is not allowed, as shown in the following examples: (p. 73)

(91) Page 28 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

(91b) shows that ordinary (headed) VP-ellipsis is permitted in Danish. However, (91a) indicates that V-stranding VP-ellipsis is not allowed.28 A more serious and general problem with these approaches is that there is no clear criterion on what counts as an agreeing head. We do not have ellipsis-independent evidence that heads with [+Wh], [+Tns], [+Poss], or [+Plural] are agreeing heads while that, if, for, and the non-control to are not. To put it more generally, it is not clear why a certain functional head with a certain feature is required to license ellipsis. Why are an agreeing functional head or ∑ with the strong feature relevant to licensing ellipsis? As far as we know, a principled explanation of this has not yet been proposed.

3.4.2 Relation between licensing heads and ellipsis sites It is generally assumed in the literature that it is not sufficient for a licensing head to have a certain feature specification in order to license ellipsis: a licensing head must be in a specific relation with the ellipsis site. It has been proposed (without much empirical evidence) that the relevant relation is government (Zagona 1982; Lobeck 1990, 1995; Saito and Murasugi 1990a), the head–complement relation (Merchant 2001), or ccommand (Aelbrecht 2009). Aelbrecht (2009) provides an argument that the relevant relation is not an extremely local one such as the head–complement relation. Consider the following VP-ellipsis sentence: (92)

Aelbrecht (2009) argues that ellipsis of thinking about that in this example is licensed by the finite auxiliary should rather than the non-finite been, which is in the head– complement relation with the elided VP. That non-finite auxiliary verbs cannot license VPellipsis can be confirmed by the unacceptability of (93a). (93)

Page 29 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar The acceptability of the non-ellipsis counterpart of (93a) (i.e., (93b)) suggests that the unacceptability of (93a) is due to (unlicensed) ellipsis. Given that non-finite auxiliary verbs cannot license VP-ellipsis, Aelbrecht (2009) concludes that it is should rather than been that licenses VP-ellipsis in (92). In (92), the licensor should is not in the head– complement relation with the ellipsis site. Therefore, (92) indicates that a licensor does not have to be in the head–complement relation with an ellipsis site. Given this, Aelbrecht (2009) argues that the relevant relation for ellipsis licensing is c-command (or the Agree relation, which is defined based on c-command). (p. 74)

3.5 Conclusion This chapter has discussed three issues related to ellipsis (i.e., the structure of the ellipsis site, recoverability, and licensing) from the point of view of transformational grammar. In section 3.2, we reviewed arguments in favor of the structural analysis of ellipsis that mandates full internal structure in the ellipsis site at certain levels, and compared the two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We concluded that there are arguments for the deletion analysis, and the potential problems for the deletion analysis (island repair and multiple sluicing) can be addressed under Merchant’s (2001) and Lasnik’s (2014) proposals. Section 3.3 concerned recoverability. We concluded that in order for elided material to be recoverable, a semantic identity condition must be satisfied, but that is not a sufficient condition; syntactic or formal identity must be taken into account (Lasnik’s 1995b English verbal morphology argument, Chung’s 2006 sprouting argument, and Merchant’s 2008a voice mismatch argument). However, we also saw that a syntactic identity condition should not be too strict, as Chomsky (1965) already pointed out (Merchant’s 2001 sprouting, Romanian sluicing, and vehicle change arguments and Yoshida’s 2010 SIA construction argument). Finally, we discussed licensing in section 3.4, where we reviewed some proposals in the literature about what properties of licensing heads and what local relation between the ellipsis site and the licensing head are relevant to ellipsis licensing.

Notes: (1) See Ross (1967: 196–240) for discussion of this phenomenon. (2) However, potential counterexamples to this generalization have been reported in the literature (Almeida and Yoshida 2007; Sato 2011b). Although Stjepanović (2008), Vicente (2008), and Rodrigues et al. (2009) also discuss counterexamples, they conclude that the relevant data are in fact consistent with the generalization, once alternative sources, especially cleft-like sentences, are considered. See van Craenenbroeck (2010a) for extensive relevant discussion. (3) Although Merchant (2001: 92) presents the generalization as a necessary and sufficient condition, it should be just a necessary condition because a language might

Page 30 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar allow wh-movement to strand a P while ellipsis of IP is prohibited for independent reasons. (4) As a reviewer points out, this argument presupposes that idiomatic readings require the relevant parts to form a constituent at some point in the derivation. (5) Interestingly, the non-ellipsis counterpart of this example is unacceptable: ((i))

Yoshida et al. (2015a) argue that this indicates that sluicing can repair some island violations. (6) However, Kimura (2010) proposes a deletion analysis of sluicing in which the remnant wh-phrase does not undergo movement. (7) As discussed by van Craenenbroeck and Lipták (2005, 2006, 2013) and van Craenenbroeck (2012a), sluicing could be a deletion operation that applies to a CP projection rather than IP given that CP constitutes several separate projections (see Rizzi 1997). (8) However, Barros et al. (2014) argue that island repair is an illusion since sluicing becomes island-sensitive once we properly control for non-island construals for ellipsis sites. (9) The content of this section is based on Lasnik (2011). (10) Lakoff (1972) argues that Chomsky’s (1972a) *-marking analysis is conceptually problematic because it requires that a new element be introduced, perhaps foreshadowing inclusiveness condition violations in modern terms. Kitahara (1999) explicitly appeals to inclusiveness in rejecting *-marking for a different but related phenomenon, with * on a trace left behind by an overly long movement, as in Lasnik and Saito (1984) and Chomsky (1986). Lasnik (2001) suggests a technical solution to this technical problem: Everything is “born” with a √. When a violation occurs, the √ is erased. A representation with a category without a √ yields an unacceptable sentence. Also see Temmerman (2013), developing ideas of Merchant (2008b), for an argument for *marking of traces. (11) However, Merchant (2008b) proposes a variant of the *-marking analysis that can accommodate this contrast between VP-ellipsis and sluicing. (12) Lasnik (2001) argues that propositional island violations can be repaired even if an alternative short source is not available (see Barros et al. 2014 for arguments against Lasnik’s arguments). Furthermore, Lasnik (2001) also argues that apparent failure of repair in VP-ellipsis obtains even when Merchant’s (2001) PF islands are involved. See Page 31 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar Fox and Lasnik (2003) for detailed discussion about the contrast between sluicing and VPellipsis in terms of island repair. (13) See also Merchant (2014a). (14) See Rouveret (2012) for extensive commentary on Lasnik’s (1995b) account, and an alternative that relies on another sort of formal identity requirement, due to Chung (2006). Immediately below, we will briefly discuss Chung’s proposal. Also see Potsdam (1997a) for a challenge to Lasnik’s (1995b) account, and Lasnik (1997) for a partial response. (15) Potsdam (2007) argues that voice mismatches are tolerable in Malagasy. (16) See Kertz (2010) for an information-structure-based alternative account of the asymmetry between sluicing and VP-ellipsis in voice mismatches. (17) The other argument against syntactic identity by Merchant (2001) concerns English verbal morphology in ellipsis. The verbal morphology in the sluiced clause does not have to match with that in the antecedent clause, as illustrated by the following (and as already noted by Ross 1969b): ((i))

However, this largely ceases to be problematic under Lasnik’s (1995b) analysis of English verbal morphology. This is so because as we saw in section 3.3.1, there is a stage in the derivation where decorating and decorate in (ia) and meeting and met in (ib) are completely identical. The deletion of to in (ia) might remain a problem. (18) The relevant literature almost invariably calls these potential ellipsis sources ‘clefts’, but that usage is in conflict with tradition. Thus, a cleft version of (69a) is really (i), which contains all the elements of the original sentence: ((i))

(19) Another potential problem for syntactic identity concerns antecedent-contained deletion (ACD; Bouton 1970, May 1985, Hornstein 1995a, and Kennedy 1997, among others). A typical example of ACD is given in the following: ((i))

Page 32 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar In (i), the elided VP is contained in its antecedent VP. Thus, if ellipsis is subject to a (strict) syntactic identity requirement, (i) has the representation shown in (ii). ((ii))

(ii) is uninterpretable since it contains an infinitely large VP. Thus, it is a mystery under syntactic identity why (i) is interpretable. However, May’s (1985) analysis makes ACD compatible with syntactic identity. Assuming the operation Quantifier Raising (QR) in LF, which is A′-movement that adjoins a quantificational phrase to a dominating node (IP or VP), May (1985) argues that (i) has the following LF representation: ((iii))

In (iii), the quantificational DP undergoes QR, adjoining to the matrix IP. As a result, the elided VP is no longer contained in its antecedent VP, deriving the well-formed LF representation. Thus, ACD becomes unproblematic for syntactic identity (i.e., at the level of LF) under May’s QR analysis. (20) The formal definitions are given below: ((i))

((ii))

((iii))

((iv))

Page 33 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar (21) Yoshida (2010: 351) argues that the following constituency tests suggest that the adjunct clause is adjoined to VP: ((i))

However, we cannot conclude from these data that the adjunct clause must adjoin to VP since they can be accounted for even if we just assume that the adjunct clause can adjoin to VP. (22) As a reviewer points out, the elided IP and the antecedent VP are not semantically equivalent if we take the tense information in the matrix clause into account. Noticing this potential problem, Yoshida (2010) suggests that the elided IP does not contain a tense because the tense in the adjunct PP depends on the matrix tense. (23) A quantifier also licenses NP-ellipsis, as Lobeck (1990: 350) observes: ((i))

Lobeck (1990) argues that quantifiers are functional categories that head QP, taking NP as their complements, and they license ellipsis although they are not Kase assigners. This means that Kase is not a necessary condition for a licensing head. See section 3.4.1.3. (24) Though this specific argument is widely cited, there appear to be interfering factors. First, if we drop the shouldn’t in (81a), the resulting sentence, whether elliptical or not, is a contradiction: Because she smokes, Mary doesn’t smoke. Second, in both (81a) and (81b), deleting the VP should result in a stranded affix, triggering do-support. See the Chomsky (1955) argument in section 3.3.1. (25) Martins (1994) assumes that a clitic adjoins to AgrS. (26) Thoms (2011b) proposes a different approach to ellipsis licensing from the Lobeck– Martins approach. According to him, ellipsis is licensed by movement. Interestingly, this movement approach to ellipsis licensing makes similar predictions about V-stranding VPellipsis as Martins’ (1994) approach. (27) Matos (1992) proposes that the set of licensing heads can differ from one language to another (see also Martins 1994: 187). (28) See Funakoshi (2012, 2014) for other conditions on headless XP-ellipsis, of which Vstranding VP-ellipsis is a subtype. Page 34 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Transformational Grammar

Howard Lasnik

Howard Lasnik is Distinguished University Professor in the Department of Linguistics at the University of Maryland. He specializes in generative syntax and the formalization of syntactic theories. Among the specific topics he has worked on are phrase structure, anaphora, ellipsis, verbal morphology, case, and locality constraints on movement. His publications include eight books and over a hundred articles. Kenshi Funakoshi

Kenshi Funakoshi is Lecturer in Linguistics in the Department of English at Dokkyo University. He specializes in generative syntax with an emphasis on ellipsis, verb movement in SOV languages, and Japanese. His publications include ‘On headless XP-ellipsis/movement’ (2012), ‘Verb-stranding verb phrase ellipsis in Japanese’ (2016), and ‘Backward control from possessors’ (2017).

Page 35 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

Oxford Handbooks Online Ellipsis in Head-Driven Phrase Structure Grammar   Jonathan Ginzburg and Philip Miller The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.4

Abstract and Keywords The chapter provides an overview of the types of analyses of elliptical phenomena that have been proposed in the literature on Head-Driven Phrase Structure Grammar (HPSG). First, it explains the central insights behind the analyses, which have concentrated on three classes of phenomena: (i) non-sentential utterances, (ii) argument or predicate ellipsis, and (iii) constructions involving unpronounced syntactic structure. HPSG is crucially non-modular. Constraints involving various levels can be easily stated, which benefits the framework for the analysis of ellipsis, because it allows one to express simultaneous semantic and syntactic constraints on ellipsis (explaining for instance the connectivity effects among non-sentential utterances) and provides means to integrate non-semantic information-information about the realization of utterances-into context. A more detailed discussion of the theory then follows. The chapter provides a more technically precise account of the syntax of argument ellipsis, an area which has received a considerable amount of analysis in HPSG, for which there is a broad consensus. Subsequently, more recent developments are introduced and a detailed analysis of nonsentential utterances is provided. In particular, we introduce an alternative version of HPSG, which allows the grammar to directly interface with dialogue context as conceived in the framework of KoS. Keywords: Head-Driven Phrase Structure Grammar, non-sentential utterances, argument or predicate ellipsis, KoS, dialogue

Jonathan Ginzburg and Philip Miller

Page 1 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.1 Introduction 4.1.1 General properties of ellipsis THE characterizing feature of ellipsis is that elements of semantic content are obtained in the absence of any corresponding syntactic form. The syntax thus appears to be incomplete. More specifically, the implicit semantic content is recovered from elements of the linguistic and non-linguistic context. In this sense, ellipsis is similar to anaphora, except that there is no overt anaphoric element involved. The elements present in an elliptical clause are predicates, arguments, or adjuncts of what is omitted. It is the presence of these elements that makes it possible to recognize the ellipsis. It is in principle always possible to reanalyze any elliptical phenomenon as a case of anaphora, either by hypothesizing an unpronounced pro-form in the elliptical site (zero anaphora) or by considering the licenser to be anaphoric (e.g., considering the preelliptical auxiliary to be an anaphor in verb phrase ellipsis (VPE); cf. Schachter 1978; Hardt 1993). It is an interesting point about the structure of the present handbook that there is no chapter on unrealized subjects of finite verbs, often treated in terms of ‘pro’, i.e., zero anaphora, in mainstream generative grammar. There is no a priori reason not to treat these simply as cases of ellipsis, just as there is no a priori reason not to analyze VPE as a case of anaphora. Decisions on these questions can only be made on the basis of further theoretical considerations. As is the case for overt anaphors, ellipsis can be anaphoric or exophoric. Resolution of ellipsis thus requires a theory of context, and specifically a theory of how both nonlinguistic and non-semantic material—information about the phonetic, phonological, and morphosyntactic realization of utterances—gets incorporated into context. More generally, ellipsis resolution requires a theory of conversation if one is to go beyond isolated examples and monologue. We will explain in this chapter how Head-Driven Phrase Structure Grammar (HPSG) allows one to account for context and for conversation. (p. 76) On the other hand, most mainstream generative approaches fail to take these phenomena into account in sufficient depth. HPSG typically takes the view that there is no single mechanism underlying ‘ellipsis’. Rather, the term covers a variety of cases involving contextually based resolution. Three central cases can be distinguished. The first two of these involve constructional analyses of ellipsis (see section 4.2.1); the third involves non-canonical correspondence between the phonology of a phrase and its daughters, and is much more restricted and limited in its scope (see section 4.2.2). Case 1: Non-sentential utterances(NSUs), e.g., Bare Argument Ellipsis and sluicing (see sections 4.2.1.2 and 4.7.4), which acquisition data shows are acquired over a span of years (see, e.g., Ginzburg and Kolliakou 2009). These are analyzed as specific

Page 2 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar constructions typically characterized by the fact that a phrase is taken to be the only constituent of a clausal structure. Case 2: Argument or predicate ellipsis, e.g., null subjects and null objects, VPE, nullcomplement anaphora, etc. (see sections 4.2.1.1 and 4.6). These are again analyzed in terms of specific elliptical constructions, typically involving a non-canonical correspondence between argument structure (ARG-ST) and syntactic subcategorization (SPR and COMPS lists), so that arguments that are semantically present can be syntactically absent. Thus, in these two cases, there is no unpronounced syntactic structure. Case 3: Constructions involving unpronounced syntactic structure, e.g., certain varieties of left- or right-peripheral ellipsis, in particular Right-Node Raising. These are analyzed by having the PHON (phonology) feature of the mother not simply be a concatenation of the PHON features of all of its daughters, as is normally the case. The latter analysis is rather similar in spirit to various mainstream analyses involving unpronounced syntactic structure, especially to analyses involving PF-deletion. However, HPSG studies have argued that this type of analysis is only applicable in a very limited number of cases characterized by specific properties, as discussed in section 4.2.2. In the following subsection, we will present HPSG in general and in section 4.2, we will present an overview of the analyses of elliptical phenomena in HPSG.

4.1.2 Head-Driven Phrase Structure Grammar Head-Driven Phrase Structure Grammar (HPSG) is a constraint-based linguistic theory initially proposed by Pollard and Sag (1987) and further developed in Pollard and Sag (1994) and numerous other papers. It is impossible to provide a detailed description of HPSG within the scope of this chapter.1 We will limit ourselves to sketching the essential elements of the framework that are crucial to understanding its approach to ellipsis. In HPSG, linguistic objects are modeled in terms of typed feature-value matrices. The value of a feature can be complex and even recursive. A language consists of a set of signs. These are abstract entities that are the locus of constraints on the interface between form, meaning, and use. A grammar is a system of constraints that conjointly define the signs of a (p. 77) given language.2 The theory uses types, organized in inheritance hierarchies, making it possible to express generalizations of varying granularity across linguistic levels.3 Because signs combine information about all levels of representation (phonology, morphology, syntax, semantics, pragmatics, discourse, and usage), HPSG is crucially non-modular. Constraints involving various levels can be easily stated. We will see that this is a powerful property of the framework for the analysis of ellipsis, as it allows one to express simultaneous semantic and syntactic constraints on ellipsis (explaining for instance the connectivity effects among non-sentential utterances) and provides means to integrate non-semantic information—information about the

Page 3 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar realization of utterances—into context. Thus, HPSG allows one to avoid two tendencies that are dominant in the literature: • ignoring linguistic complexity, as in logically-based models which abstract away from a linguistic level. • ignoring contextual complexity, as in generative accounts that abstract away from dialogue context. As we will see in more detail, HPSG initiated the constructional approach to ellipsis resolution, making central use of the notion of Question Under Discussion (QUD); cf. Ginzburg and Sag (2000). Furthermore, HPSG aims to be psycholinguistically plausible, specifically by providing a theory that is compatible with incremental processing.4 Finally, work in HPSG typically aims at strong empirical grounding, based on corpus research, fieldwork, acceptability experiments, etc. As in the Simpler Syntax framework (see Chapter 7 of this volume), no distinction is made between core and periphery. Specifically, it is unclear that phenomena relegated to the ‘periphery’ are less complex than ‘core’ phenomena, so that nothing is gained by the distinction in explaining acquisition: learning the periphery raises the same problems as learning the core. The ambition is to provide a large-scale description for a broad range of empirical data that is concise, formally precise (and thus falsifiable), and insightful. In the following section, we will introduce the two central types of analyses used to characterize ellipsis in HPSG, namely (i) constructional analyses (section 4.2.1), and (ii) non-canonical correspondence between the phonology of a phrase and its daughters (section 4.2.2). We conclude with some discussion of corpus studies of elliptical phenomena (section 4.2.3). (p. 78)

4.2 An overview of ellipsis in HPSG Our purpose in this section is to provide an overview of the types of analyses of elliptical phenomena that have been proposed in the HPSG literature. We attempt to convey the central insights behind the analyses, providing only the minimal formal details necessary so that readers unfamiliar with the framework can grasp the central intuitions and thus understand the three following sections which develop the HPSG positions on (i) the structure of the ellipsis site (section 4.3); (ii) recoverability (section 4.4); (iii) licensing (section 4.5). In section 4.6, we will provide a more technically precise account of the syntax of argument ellipsis, an area for which there has been a considerable amount of work in HPSG and for which there is a broad consensus on the analysis. Finally, in section 4.7, we will introduce more recent developments and provide a detailed analysis of nonsentential utterances. In particular, we will introduce an alternative version of HPSG, one which allows the grammar to directly interface with dialogue context as conceived in the framework of KoS (Ginzburg 2012).5 As for the cases of constructions with unpronounced Page 4 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar syntactic structure, for reasons of space, we will not be able to go into more detail than the brief introduction provided in section 4.2.2.

4.2.1 Constructional analyses of ellipsis In this section we will introduce two central types of constructional analyses of ellipsis, namely cases where there is a non-canonical correspondence between argument structure and valence and cases of non-sentential utterances.6

4.2.1.1 Non-canonical correspondence between argument structure and valence The feature structures describing lexical items in HPSG contain a feature called ARG-ST (argument structure), whose value is a list of feature structures describing the syntactic and semantic properties of the items they subcategorize for. Canonically, the ARG-ST list is the concatenation of the values of the two valence features, SPR (Specifier) and COMPS (Complements), which respectively govern the combination of the lexical item with its specifier and complements. (1) provides some typical examples of ARG-ST, SPR, and COMPS values. (p. 79)

(1)

The ARGUMENT REALIZATION PRINCIPLE (ARP; see (41) for a more precise formulation) states that, canonically, the value of ARG-ST for a given lexical item is the concatenation of the values of the SPR and COMPS features, as illustrated in (1). Given this set-up, we can provide an analysis of the ellipsis of complements or specifiers simply by allowing non-canonical constructions in which items mentioned in the ARG-ST do not appear in the valence features (such constructions are non-canonical in the sense that they do not respect the general formulation of the ARP). For instance, we can account for verb phrase ellipsis7 by a constraint allowing auxiliaries to have alternate non-canonical lexical entries where their VP argument is specified as pronominal in the ARG-ST list and there is no corresponding VP in the COMPS list (for details, see (39a) and (39b)). This means that the VP meaning will be recovered anaphorically or exophorically from the context, ensuring appropriate semantic interpretation (see, e.g., Schachter 1978; Hardt 1993; Kim 2006), but will not be realized in the syntax, since it is not present Page 5 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar on the COMPS list. Thus, a sentence like Kim will. receives a very simple constituent structure where the VP has only a single daughter, the V[AUX] will (see (40)). We can provide an analysis for Null Complement Anaphora (NCA), illustrated in (2a), along the same lines. However, it is well-known that NCA differs from VPE in that there is apparently no independently motivated subclass of verbs to which it is applicable (as shown by the classical contrast between (2a) and (2b)). Thus, specific idiosyncratic lexical constraints will be required for verbs like try which allow NCA, as opposed to the general constraint which applies in the case of English auxiliaries for VPE. (2)

This type of analysis can also be used for languages that allow null pronominal subjects and/or objects (sometimes called ‘pro-drop’; see, e.g., Manning and Sag 1999; Melnik 2007). In cases where they are systematically available, e.g., null objects in Brazilian Portuguese and null subjects in Spanish, a general constraint is required, similar to that suggested for English auxiliaries. Null objects, as in the Brazilian Portuguese example (3), will be treated by having a pronominal NP (rather than a VP as in English VPE) as the second item of the (p. 80) ARG-ST list, with no corresponding item on the COMPS list. Null subjects, as in the Spanish example (4), are similar except that they require a pronominal NP as first item of the ARG-ST list with no corresponding item on the SPR list. This leads to constituent structures where the VP contains only the verb and the prepositional object, as in (3), or the S contains only a VP daughter, as illustrated in (4). Thus the constituent structures for these sentences contain no unpronounced positions in the syntax (see section 4.6.2 for further detail). (3)

(4)

In certain languages, null subjects and objects are subject to various restrictions, requiring complexification of the relevant constraints (see, e.g., Melnik 2007 for a discussion of null subjects in Hebrew).

Page 6 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar In all of these cases the resolution of the ellipsis reduces to the resolution of the anaphoric element on the ARG-ST list. We will provide further detail of how this works for VPE in section 4.6.2.

4.2.1.2 Non-sentential utterances: Sluicing, short answers, and gapping In this section we will briefly introduce the analyses of non-sentential utterances (NSUs), more specifically SLUICING and SHORT ANSWERS, which have been proposed in HPSG (see in particular Ginzburg and Sag 2000: ch. 8; Ginzburg 2012). We will also briefly present the analysis of GAPPING proposed by Abeillé et al. (2014), in which the gapped clause is treated as a cluster of NSUs. Both sluicing and short answers involve XP ‘fragments’ which receive a ‘sentential’ interpretation (e.g., as propositions and questions). Following Ginzburg and Sag (2000), we will analyze these as specific sentential constructions where the verbal projection is reduced to a single phrase. Despite their sentential semantics, treating NSUs as sentences in the sense of verbal projections may not be the best analysis in all cases. For certain cases, e.g., sluices, it clearly is, since it is well-known that they have a sentencetype distribution, rather than an NP or PP distribution (see, e.g., Merchant 2001: section 2.1 and Culicover and Jackendoff 2005: 268–70). However, for other cases, there is crosslinguistic variation in this respect. For instance, oui (‘yes’), non (‘no’), si (‘yes’) can be embedded under the complementizer que in French (Je crois que oui/non/si. ‘I think that yes/no/yes.’), whereas this is much more difficult in English. Similarly, NP short answers can be embedded in the Brazilian Portuguese example in (5). This is clearly much less natural in English, as shown by the glosses.8 (p. 81)

(5)

As will be discussed in more detail in sections 4.3.1 and 4.7, the form and interpretation of these NSUs are determined by elements of the previous context. Gapping, illustrated in (6a), has received much attention in the literature on ellipsis (see also Chapter 23 of this volume). Many analyses, following Ross (1967), invoke some form of a deletion process, which, in more recent minimalist versions, is sometimes fed by movement of the remnants to the left periphery (see among others Sag 1976a; Merchant 2004a; Hartmann 2009, as well as Chaves 2009 within an HPSG framework). Other studies have proposed a specific construction mapping a headless structure into a propositional meaning, accounting, among other properties, for the fact that the order of remnants in the gapped clause does not necessarily parallel that of their correlates in the antecedent clause, as illustrated in (6b) (see among others Sag et al. 1985; Steedman Page 7 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar 1990; Gardent 1991; Culicover and Jackendoff 2005; Mouret 2007a; Bîlbîie 2011). Following this tradition, Abeillé et al. (2014) provide a series of further arguments against deletion-based analyses (based on French and Romanian data) and propose a construction-based approach set in an HPSG framework. Specifically, they show that syntactic parallelism is less strict than is usually assumed: each remnant in gapping is required to match some subcategorization frame of the verbal predicate its correlate depends on. This is illustrated for Romanian in (6c) (Abeillé et al. 2014: 251 (33b)), where the second remnant, tuturor copiilor, is a dative NP whereas its correlate in the antecedent, la trei dintre copii is a PP. On the other hand, discourse parallelism is required. The gapped clause is analyzed as a cluster of major fragments.9 (6)

4.2.2 Non-canonical correspondence between the phonology of a phrase and its daughters Right-Node Raising (RNR), also known as Right-Peripheral Ellipsis has been the subject of numerous studies (see, e.g., Culicover and Jackendoff 2005; Hartmann 2009; Johnson 2009). Within the HPSG literature, Crysmann (2003); Yatabe (2003); Beavers and Sag (2004); Chaves (2008, 2014); Abeillé et al. (2015) have insisted that it has radically different properties from the constructions discussed in sections 4.2.1.1 and 4.2.1.2, which warrant a different type of analysis. For reasons of space, we will not be able to discuss the HPSG analyses of RNR beyond the brief (and somewhat terse) summary given here of the most recent developments. Specifically, Chaves (2014) has argued that many of the complexities in the vast literature on RNR can be eliminated if it is recognized that many cases of apparent RNR also have analyses either as cataphoric VPE or N’-ellipsis or as ATB Extraposition. If one restricts oneself to cases which are unambiguously RNR, it appears that RNR “(i) can target any peripheral string of words that can form an independent prosodic unit, and (ii) imposes morphological form identity” (Chaves 2014: 864).10 Chaves furthermore argues that certain cases of RNR cast doubt on the possibility of maintaining classical accounts of RNR, in which the RNRaised string undergoes movement.11 (p. 82)

(7)

Page 8 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

In (7a) the putatively RNRaised string is an idiom chunk; in (7b) it is a word-part; in (7c) it is very difficult to argue that it could be a constituent. In all three cases, there is no reason to believe that the RNRaised string is a syntactically movable entity (as shown by the ill-formedness of the variants in parentheses). In order to account for these properties, Chaves proposes an analysis which might be considered to be close in spirit to PF-deletion. In general, in HPSG, when signs are combined, the phonology of the mother is assumed to be the concatenation of the phonology of its daughters. The construction for RNR, on the other hand, is exceptional in that it allows one to transfer only the phonology of the second of two sequences of morphophonological constituents that are identical up to morphological forms. For technical details, see Chaves (2014), especially pp. 874ff. The following properties characterize elliptical phenomena that must be analyzed in this way: (i) they require strict syntactic identity;12 (ii) they are strictly intrasentential; (iii) they are never exophoric. (p. 83)

4.2.3 Elliptical phenomena: Evidence from corpora Corpus studies are important since they provide a “lower bound” for what the grammar needs to characterize; due to sparsity issues, admittedly, some rare phenomena (or at least rare in the settings sampled by a given corpus) might be missed. Hence, corpusbased generalizations need to be made on the basis of corpora with a wide range of genres and it is important to combine this source of data gathering with an experimental program. In this section, we review some studies that form an important backdrop to the HPSG/KoS work described below. We begin with a presentation of the corpus-based classification of clarification requests developed by Purver et al. (2001), because it provides an example of how the extensive discussions of certain elliptical constructions in the literature (in particular sluicing and Bare Argument Ellipsis) have ignored important aspects of the relevant range of data, which are brought to light by corpus investigation. We will then mention more briefly other corpus-based work on ellipsis in HPSG. Successful coverage of the empirical phenomena presented in these papers, we believe, constitutes a significant test of adequacy for any theory of ellipsis. Purver et al. (2001) propose a classification of the range of form/contents of clarification requests (CRs) in the British National Corpus (BNC). These can take many forms, as illustrated in Table 4.1, all but the first three being elliptical. The examples in this table Page 9 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar are to be understood as clarification requests in a discourse context where A has just uttered ‘Did Bo leave?’ (except for the last case, Filler, where A utters the incomplete utterance ‘Did Bo…’).13 Table 4.1 A taxonomy for clarification requests (Purver 2006) Category name of CR:

Example of CR:

Explicit

B: Did you say ‘Bo’? / What do you mean ‘leave’?

Literal reprise

B: Did BO leave? / Did Bo LEAVE?

Wh-substituted Reprise

B: Did WHO leave? / Did Bo WHAT?

Wot

B: Eh? / What? / Pardon?

Reprise Sluices

B: Who? / What? / When?

Reprise Fragments

B: Bo? / Leave?

Gap

B: Did Bo…?

Filler

A: Did Bo…B: Win?

Providing explicit formal analyses of just about any of these classes is a formidable challenge for any approach where ellipsis resolution is not embedded within a theory of dialogical interaction; for detailed discussion, see Ginzburg and Cooper (2004); Ginzburg (2012), who argue that neither deletion-type accounts, nor higher-order unification-type accounts can deliver the requisite readings. Indeed, to date, we are not aware of analyses of these phenomena in existing formal grammatical frameworks apart from HPSG. We highlight just several of the most significant issues. The first point to note is that a number of these forms are ones whose sole analysis is as clarification requests—this applies to the classes Wh-Substituted Reprise and to Gap. (p. 84) These constitute instances of forms whose meanings cannot be analyzed in interaction-free grammar. A second point relates to cross-turn parallelism. Ginzburg and Cooper (2004); Ginzburg (2012) argue in detail that reprise fragments have two main classes of uses, one to request confirmation about the content of a previous sub-utterance; the other to find out about the intended content of a previous sub-utterance. Both uses have strong parallelism requirements, which require semantic and structural information to be projected across multiple turns into context. Thus, the former requires identity of morphosyntactic category between source and target, as illustrated in (10a) and (10b); the latter requires

Page 10 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar segmental identity between source and target, as exemplified in (10c). Parallelism of the latter kind seems needed also for the Gap class of CRs: (10)

More generally, Fernández and Ginzburg (2002); Schlangen (2003) characterize the range of NSU types in the BNC, showing that they can be reliably classified into a small number of categories, revolving around the commonality in semantic resolution process.14 Subsequently, detailed semantic and syntactic accounts of these classes have been developed in Schlangen (2003), Fernández (2006), and Ginzburg (2012), some of which we sketch in section 4.7. These classifications have been extended crosslinguistically, with minor modifications: French (Guida 2013), Chinese (Wong and Ginzburg 2013), Spanish (Marchena 2015), and to other genres (see, e.g., Filtopoulos 2015). Focusing on one subclass of NSUs, which has attracted much attention, namely sluicing, the large generative literature on the topic (see e.g. Ross 1969b; Chung et al. 1995; Merchant 2001, (p. 85) and many chapters in this volume) has, with very few exceptions, ignored the fact that bare wh-phrases are systematically ambiguous and concentrated almost exclusively on what Ginzburg and Sag (2000) dubbed direct uses of sluicing. Fernández et al. (2007) propose the existence of a four-way ambiguity, an ambiguity they demonstrate to be reliably coded by human subjects: • Direct The utterer of the sluice understands the antecedent of the sluice without difficulty. The sluice queries for additional information that was explicitly or implicitly quantified away in the previous utterance.15 (11)

• Reprise The utterer of the sluice cannot understand some aspect of the previous utterance which the previous speaker assumed as presupposed: (12) Page 11 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

• Repetition The sluice is used to ask for repetition of the previous utterance as a whole. (13)

• Wh-anaphor The utterer of the sluice cedes the turn to the previous speaker, who has indicated his wish to answer a (possibly embedded) wh-question s/he has just uttered: (14)

‘Repetition what’ can be produced after any utterance, regardless of its content, and is indeed the highest-frequency type of sluice in the BNC; wh-anaphor uses are significantly more selective and complex in that they both presuppose that the previous utterance embeds a question and involve phonological parallelism between the fragment and the (p. 86) wh-phrase in the prior utterance. We discuss direct and reprise sluicing in 4.7.4.3 and 4.7.4.4, whereas the other two classes are analyzed in Ginzburg (2012). Beyond the cases of clarification requests and NSUs just discussed, other work in HPSG has been based on detailed corpus investigations, e.g. Bîlbîie (2011) on gapping in Romanian and French, Abeillé et al. (2014) on gapping in Romance, Abeillé et al. (2015) on peripheral ellipsis in French, Miller and Pullum (2014) on exophoric VPE, Miller (2014) on pseudogapping.16

4.3 The structure of the ellipsis site Overall, the default assumption in HPSG is that there are no empty morphemes or unpronounced syntactic structures. This is a consequence of a general principle of economy (Occam’s razor, cf. Miller 1997). See Arnold and Spencer (2015) for recent explicit argumentation against empty morphemes and Müller (2015a: 553–72) for extensive discussion of HPSG analyses and a comparison with other grammatical Page 12 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar frameworks. For elliptical constructions this means that, in the absence of strong evidence to the contrary position, there should be no syntactic structure at the ellipsis site, as mentioned above for argument ellipsis and NSUs in section 4.2.1. Similarly in long-distance filler-gap constructions, it is assumed that there is no empty syntactic node at the foot of the construction (i.e. there is no “trace”; cf. Sag and Fodor 1994).17 On the other hand, HPSG does not make the assumption that all forms of ellipsis should necessarily be treated by a single formal mechanism, so that unpronounced structure can be assumed if warranted in a given case, as was argued for RNR in section 4.2.2. With respect to unpronounced syntactic structure, HPSG is in clear disagreement with much mainstream generative syntax. Numerous publications have argued in detail that such unpronounced structure is necessary to provide a motivated explanation for various constraints on ellipsis. It is thus important in the context of this chapter to review the central types of evidence that have been brought to bear on this question and to present the counterarguments that have been advanced on the HPSG side (in line with numerous other studies, see e.g. Chapters 5, 7, 8, 9 of this volume). We will discuss connectivity, locality effects, and the accessiblity of missing referents in turn. (p. 87)

4.3.1 Connectivity effects Since Ross’s and Morgan’s seminal papers on sluicing and short answers (Ross 1969b; Morgan 1973) it has been known that fragment answers and sluices exhibit connectivity effects, i.e., fragments typically appear in the form that would be theirs in the corresponding understood full clause. Various types of connectivity can be distinguished. In languages with morphological case, NP fragments typically appear in the case that they would have in the understood complete sentence, as in the classical examples (15) from Ross (1969b). Similarly, PP fragments appear with the preposition that they would have, as in (16). (15)

(16)

Page 13 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Along the same lines, the form of pronoun fragments is frequently the one you would find in the corresponding complete sentence and thus appears to be explained by the usual application of the binding theory: (17)

The apparent generalization is that fragments appear in a form that would be appropriate in the corresponding understood full sentence. On the face of it, this might seem to be a convincing argument for hidden syntactic structure, since the presence of such structure would explain why the fragment has the form it has, rather than some other. However, a closer look at the data shows the limits of the putative generalization and makes such an analysis less plausible than it initially seems. First, there are numerous cases where fragments appear in a form that would be impossible in a complete sentence. Specifically, in English, only pronouns have a nominative/accusative distinction. Possessive case, on the other hand, appears both with pronouns and with NPs (realized as an NP final ’s). Possessives exhibit typical connectivity effects as in (18a). However, in all cases where a possessive is not expected, it is the accusative pronouns that appear in elliptical utterances, not only as expected by connectivity in (18b), but also when a nominative would be necessary in a full clause, as in (18c), and when there is no obvious full clausal structure to be recovered, as in (18d). (18)

Similarly, for binding effects, it turns out that pronouns often appear in forms that would not be licensed in a full clause. (p. 88)

(19)

Page 14 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar It thus appears that although connectivity is a common situation, there are various cases where fragments could not be reintegrated as such into a well-formed complete clause. Because of the non-modularity of the sign, HPSG has no problem in accounting for connectivity effects without reference to hidden structure. In section 4.7.2, we sketch an account of parallelism in NSUs originating in Ginzburg and Sag (2000), which amplifies QUDs to include a restricted amount of non-semantic information. Syntactic parallelism is obtained by imposing matching conditions between a fragment and a prior utterance. Thus, prior syntactic structure is not used in constructing the content of a fragmentary utterance; it functions merely in establishing the appropriateness of the fragment in the given context (hence the # judgment on the unacceptable like (15), (16), (17), etc., which we claim are inappropriate in the discourse context provided, rather than ungrammatical). In embedding the antecedent information within QUD, Ginzburg and Sag (2000: 301) make strong predictions about the extent of categorial parallelism. Specifically, only properties of the maximal element of QUD (MAX-QUD) can trigger syntactic or phonological parallelism.18 On the other hand, examples of non-connectivity such as (18) and (19) raise problems for any analysis that wishes to explain the cases where connectivity holds in terms of unpronounced syntactic structure. They require various ad hoc assumptions to account for the absence of the expected connectivity effects. For detailed exemplification, see e.g. Ginzburg (1999, 2012); Stainton (2006b); Nykiel (2013). Furthermore, because deletion or reconstruction-based analyses usually do not offer an explicit theory of discourse context, (p. 89) even in those cases where connectivity does hold, they almost invariably fail to offer an explicit account of how the reconstructed syntactic material is accessed in the context and introduced into syntactic structure. This is true even in cases where the antecedent is anaphorically retrieved, but is even more problematic in exophoric cases.19 It should also be noted that cases where connectivity holds (as it typically does crosslinguistically with case morphology, for instance) raise the symmetrically inverse challenge for analyses whose architecture enforces the assumption that the context used for interpreting fragments is purely semantic e.g. higher-order unification accounts (Pulman 1997) and Categorial Grammar (see, e.g., Jacobson’s contribution in this volume). Jacobson (2016b) opts for an account that closely resembles the one proposed by Ginzburg and Sag (2000) for short answers, by postulating a QuAns construction which enables case-matching phenomena to be captured and offers much insightful discussion of issues relating to connectivity.

4.3.2 Locality effects Locality effects have provided another classical motivation for unpronounced syntactic structure. Specifically, it has been argued that various cases of ill-formedness in elliptical constructions can be explained in terms of island violations involving such structure. Consider the following examples.

Page 15 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Sentence (20a) is claimed to be ungrammatical because it exhibits the same island violation in unpronounced structure as its non-elliptical counterpart (20b), an analysis which can only be expressed if there is unpronounced structure (cf. the similar examples provided by Chung et al. 1995: 275). (20)

However, these facts raise a number of problems. First, as is known since Ross (1969b), sluicing appears to absolve island violations, as illustrated by the contrast between (20) and (21). (21)

Second, as shown in (23), there are cases where what would be putative island violations in unpronounced structure do not lead to ungrammaticality (though some (p. 90)

speakers may find acceptability slightly reduced, their status is not at all comparable to that of (20a)):20 (23)

In sum, the factors that constrain the interaction of wh-movement and VPE are at present not completely understood and it is hard to see how a strong argument for unpronounced structure can be made from these facts. Similar arguments have been made, for instance, for Bare Argument Ellipsis (see, e.g., Merchant 2004a) and gapping (see, e.g., Coppock 2001), purporting to show that they are affected by similar island violations in unpronounced structure. However, once again, it has been argued that similar examples which should exhibit the same island violations can also be perfectly acceptable, e.g. by Culicover and Jackendoff (2005) as well as Chapter 7. Thus, overall, the evidence appears to be mixed and does not at present allow one to draw strong conclusions either in favor of or against unpronounced structure. Our expectation is that a better understanding of the discourse conditions on the elliptical Page 16 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar constructions in question and their interaction with wh-movement might lead to a more satisfactory explanation of the data.

4.3.3 The accessibility of missing referents To conclude this section, we briefly present one final argument against hidden syntactic structure which can be derived from Arnold and Borsley (2008), whose central point is to argue against the orphan view of non-restrictive relative clauses. Specifically, they observe that fragment answers do not allow non-restrictive relative clauses connected with part of the “missing” material, whereas anaphoric it can recover the same referents despite the fact that they are not expressed. This is illustrated in (24). (24)

As shown in (24a), a non-restrictive relative clause can have as its antecedent the whole proposition expressed by the fragment, but it cannot have an understood but syntactically unrealized NP as its antecedent, as (24b) shows. In this, which differs from the pronoun it which can access such a referent, as evidenced in (24c). They argue that this shows that non-restrictive relatives are syntactically integrated and require a syntactically represented antecedent, whereas anaphoric it simply requires that its antecedent be accessible in conceptual structure. Assuming the presence of unpronounced syntactic structure in such cases (something like: Kim owns a dog, which is a dachshund. for (24b)) makes it much more difficult to distinguish the cases in a relevant way, since the unpronounced referent is syntactically accessible. On the other hand, as shown by Arnold and Borsley (2008), the analysis for fragments proposed by Ginzburg and Sag (2000) (presented below in section 4.7) immediately predicts the data. (p. 91)

4.4 Recoverability The idea that ellipsis requires some form of recoverability is inherent to the concept itself. The current technical debate on recoverability originates with Chomsky, who proposed a “convention to guarantee recoverability of deletion” which allows (among other cases) deletion of elements that are “otherwise represented in the sentence in a fixed position” (1965: 144–5). This is the source of the position that frames recoverability as a requirement for the presence of a syntactically identical antecedent, allowing recovery of the ellipted material. Following Sag and Hankamer (1984), HPSG studies Page 17 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar have assumed that recoverability is governed by discourse semantics, rather than syntax: the central condition is the availability of an appropriate antecedent in the discourse context (which combines both linguistic and non-linguistic information). Providing formal conditions on recoverability therefore requires an articulated theory of context, and in particular of conversational context, which is sketched in Ginzburg and Sag (2000) and elaborated in detail in Ginzburg (2012). Of course, as discussed in the preceding section, HPSG also provides means for accounting for well-known syntactic constraints on the relationship between the elliptical material and its antecedent-trigger,21 though these are argued to be less numerous than is usually thought. Specifically, it has been argued that many of the classical cases where acceptability is reduced in the absence of a syntactically identical antecedent can in fact receive an alternative explanation, namely that the reduced acceptability is due to violations of (p. 92) general discourse conditions (cf. Kehler 2002; Kertz 2013) or of construction-specific discourse conditions (cf. Miller and Hemforth 2014a; Miller and Pullum 2014). We will discuss in turn: (i) split antecedents and syntactic mismatches; (ii) exophoric uses; and (iii) incrementality.

4.4.1 Split antecedents and syntactic mismatches A first class of cases that argue against the idea that recoverability should be based on syntactic identity are those where the previous linguistic context provides no identical antecedent-trigger (we ignore here cases of morphological mismatch known not to affect grammaticality). These include the well-known cases of split antecedents and those where there is some form of syntactic mismatch between the antecedent-trigger and the ellipted material. We will illustrate these in turn. VPE with split antecedents, first discussed by Webber (1979: 4–39), is illustrated in (25), where they hadn’t is understood as meaning ‘Willem hadn’t stayed in Wyoming and become a ranch hand himself and Jude hadn’t wound up in prison, or in a hospital, or dead, or worse.’: (25)

In fact, split antecedents are a special case of a more general situation where the antecedent is recoverable from the contents of the previous discourse without actually being present in it in any appropriate form. Consider the following examples: (26)

Page 18 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

In (26a), the antecedent ‘goes running’ is inferrable from the previous sentence. Similarly, in (26b), the antecedent ‘have the shelves’ is inferrable from ‘wanted the shelves’. As for (26c), (p. 93) the previous context makes it clear that the elliptical clause is intended to mean something like ‘I will come back to meet your sister who sings’. Again, this content appears nowhere as such in the conversational context; but it can be inferred from the dialogue as a whole.22 Similarly, in the second instance of sluicing in (27), the intended antecedent is recovered by inference from the discourse context. (27)

Cases of syntactic mismatch between the antecedent and the elliptical sentence (e.g., voice mismatches, category mismatches) have been extensively discussed in the literature, especially for the cases of VPE and sluicing (see, among many others, Hardt 1993; Merchant 2001, 2013d; Kehler 2002; Kertz 2013). For sluicing, the great majority of the literature has agreed that mismatches are impossible. For VPE, the situation is much more complex because of the fact that there is a wide range of different acceptabilities for different cases of VPE with mismatched antecedents, ranging from the perfectly acceptable to the very unacceptable. This has led to two different positions. Either one assumes that the syntax disallows mismatches, in which case acceptable cases of mismatch are analyzed as ungrammatical but repaired. This position has been defended by Lyn Frazier and her colleagues in a series of studies (see, e.g., Arregui et al. 2006; Page 19 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Grant et al. 2012). The other is to assume that mismatches are always grammatical, and that unacceptable instances arise from the violation of independent discourse constraints (e.g., Kehler 2002; Kertz 2013).23 Within this debate, HPSG studies have followed the lead of Kehler and Kertz, but have focused on developing construction-specific discourse constraints. For VPE, Miller (2011a), Miller and Hemforth (2014a), and Miller and Pullum (2014) distinguish two subtypes of non-comparative VPE, which they call Auxiliary Choice and Subject Choice respectively, and propose the following constraints.

Type 1: Auxiliary Choice VPE FORMAL CHARACTERISTICS: The subject of the antecedent is identical with the subject of the VPE construction and the auxiliary is (at least weakly) stressed, signaling a new choice of tense, aspect, modality, or (in the most overwhelmingly frequent case) polarity. DISCOURSE REQUIREMENT: A choice between the members of a jointly exhaustive set of alternative situations must be highly salient in the discourse context, and the point of the utterance containing the VPE is limited to selecting one member of that set. (p. 94)

Type 2: Subject Choice

FORMAL CHARACTERISTICS: The subject of the antecedent is distinct from the subject of the VPE construction, and stressed if it is a pronoun. DISCOURSE REQUIREMENT: A particular property must be highly salient in the discourse context, and the point of the utterance containing the VPE must be limited to identifying something or someone possessing that property.24 Importantly, they point out that when these discourse constraints are satisfied, VP anaphors like do it are dispreferred, even with mismatched antecedents. This is illustrated in (28a) for an Aux-Choice case. (28b) provides a Subj-Choice case. (28)

For VPE with nominal antecedents, Miller and Hemforth (2014a) show how these discourse constraints can explain the unacceptability of most cases. Specifically, for the case of Aux-Choice VPE, nouns are typically incapable of making an alternative salient. However, they point out the existence of a small class of ‘polar nouns’ that can be interpreted as concealed polar questions, as is the case for survival in (29a), which can be paraphrased by an indirect polar interrogative as shown in (29b). (29)

Page 20 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

Miller and Hemforth show that when a polar noun appears in a discourse context that makes its interpretation as a concealed polar question salient, as is the case in (29a), it can serve as a highly acceptable antecedent for VPE. These observations are corroborated through a series of acceptability experiments. On the other hand, NPs are incapable of making open propositions salient, so that Subj-Choice VPE is never acceptable with a nominal antecedent. Similarly, though no detailed study has been conducted, we expect that the discourse conditions on sluicing (namely that it requires that a quantified proposition be at issue, cf. n. 15 and section 4.7.4.3) can account for the fact that voice mismatches are unacceptable. On the other hand, examples such as (30), due to Beecher (2008), illustrate the possibility of sluicing with a nominal antecedent. In both cases, there exists an entailment of the requisite quantified proposition and, despite the lack of a clausal (verbal) antecedent, the sluice seems acceptable, as expected from an account based on the above discourse condition: (p. 95)

(30)

4.4.2 Exophoricity Let us now consider exophoric uses. In these, there simply is no linguistic antecedent at all. Examples of exophoric VPE are provided in (31). (31)

Page 21 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Miller and Pullum (2014) argue that exophoric VPE is in fact freely available syntactically (contrary to Hankamer (1978) who claims that the rare cases of exophoric VPE are idioms) but that it is relatively rare because the discourse conditions on VPE are hard to satisfy through extralinguistic context. Specifically, for Aux-Choice VPE, they argue that, in general, it is difficult for non-linguistic context to make an alternative salient. The most usual means for doing so are polar questions and assertions (see section 4.7), but these typically cannot be performed non-linguistically. Thus, only less frequent strategies for establishing an alternative as the QUD can give rise to exophoric VPE. Directives, for example, can make an alternative salient (whether or not to comply), whether they are performed linguistically or deictically, as in (31a). Similarly uttering a reproach, as in (31b), forces accommodation of the alternative between doing or not doing the incriminated behavior, satisfying the discourse conditions. Crucially, they point out that when the discourse conditions on VPE are satisfied, do it is less felicitous, as shown in the variants of (31). On the other hand, they suggest that it is almost impossible for nonlinguistic context to make an open proposition salient, so that exophoric Subj-Choice cases are almost impossible. Similarly, though exophoric sluicing is rare in corpora, it is clearly attested, as shown by examples like (32). (32)

Once again, we assume that the relative rarity of exophoric sluices is due to the discourse conditions on the use of the construction: sluicing requires that a quantified proposition be at issue (see discussion of example (60) and Ginzburg 2012: section 7.8).25 This is difficult to construct non-linguistically, in a way parallel to what is proposed for VPE. (p. 96)

Another class of exophoric ellipsis is declarative fragments tied to various interactional genres, as in (33): (33) Page 22 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

Finally, we note here a class of non-sentential utterances, for which exophoric resolution is close to being the rule. This involves the little-discussed phenomenon of exclamative sluices (Ginzburg 2012), exemplified in (34): (34)

Example (34a) illustrates a case of an anaphoric exclamative sluice that is parallel to typical interrogative sluices. As for examples like (34b), they are by far the most frequent and are typically exophoric. This shows that any account of sluicing based on the longstanding assumption that wh-binding requires a linguistic antecedent is not viable. It is important to emphasize the real difficulty these exophoric cases pose for analyses that claim that recoverability must be stated in terms of any kind of syntactic identity (including LF-identity). It is completely unclear how such analyses can make sense of exophoric uses at all, barring the assumption that a speaker constructs a specific syntactic representation of the relevant extralinguistic properties, which could then be used as an ‘antecedent’. However, this idea defeats the basic purpose of the idea of ‘recoverability’ of deletions, as initially proposed by Chomsky, since it then becomes unclear why there are any constraints at all on exophoric uses.26 (p. 97)

4.4.3 Incrementality An additional argument against viewing a construction such as sluicing as S-deletion (or similar) is that it is possible as soon as its nominal antecedent has been introduced, as in (35); it need not occur as part of a completed sentential construction. This can occur either in monological examples such as (35a) or in dialogical cases such as (35b) and (35c). Once one adopts an incremental view of semantic processing, accommodating such data becomes straightforward, as we will show in section 4.7.4.3, based on Ginzburg et al. (2017). (35)

Page 23 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.5 Licensing As has been mentioned above and as will be developed in much more detail in section 4.7, the HPSG approach to ellipsis resolution makes central use of the notion of Question Under Discussion (QUD), which is augmented to include restricted amounts of nonsemantic information. In a sense then, HPSG assumes that QUD is the central factor in the licensing of ellipsis. This makes sense under a more general view of anaphora resolution involving some notion of accessibility or givenness (see, e.g., Ariel 1990; Gundel et al. 1993), as it is known that, in general, phonologically weaker anaphors require more accessible antecedents. Since ellipsis can be seen as the phonologically weakest form of anaphor, it requires antecedents that are most accessible, namely, antecedents involved in the most salient QUD at any moment in the discourse. This feature of the model thus makes strong general predictions about the syntax, phonology, semantics, and pragmatics of elliptical constructions, namely that only properties of the most salient QUD can be relevant in determining their form and interpretation. Beyond this, HPSG does not assume that there is a single mechanism underlying ellipsis. Rather, elliptical constructions each have their own specific licensing conditions. To the extent that generalizations can be made, they can be captured through the hierarchy of types. For example, it is known that different subtypes of VPE have subtly different properties (see, e.g., Miller 2011a; Kertz 2013). The general properties of VPE can be attributed to the construction-type VPE, whereas the various subconstructions can be specified for their specific properties, while inheriting the general properties from the more general construction. Similarly, it would be possible to represent the generalizations available for different types of argument ellipsis (see section 4.2.1.1 and 4.6), including VPE, through the type hierarchy, and more generally to associate the general properties of ellipsis involving QUD to an overarching ellipsis-type, though to our knowledge this has not yet been attempted. In sum, the basic methodological assumption of HPSG is that one should not move too quickly in assuming that properties of constructions are generalizable. The complexity of the varieties of subconstructions must first be established in as much detail as possible before meaningful generalizations can be made. Throughout this chapter, the discussion of the various elliptical constructions highlights their specific licensing properties. We have distinguished three central cases: (i) certain constructions license non-sentential constituents as complete utterances (e.g. Bare Argument Ellipsis, sluicing, gapping); (ii) certain constructions license noncanonical correspondence between ARG-ST and valence features (e.g., null subjects and objects, VPE, NCA); (iii) certain constructions license a non-canonical correspondence between the PHON of a mother and its daughters (e.g., peripheral ellipsis, RNR). More specific constructions involve more specific licensing conditions, e.g. VPE is licensed by the presence of an auxiliary in English, whereas sluicing requires a wh-phrase. (p. 98)

Page 24 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.6 The syntax of argument ellipsis In this section we provide a more explicit discussion of the constructional analysis of argument ellipsis which was sketched in section 4.2.1.1. As mentioned there, the crucial insight underlying the analysis is that there is a non-canonical correspondence between the argument structure of a head and its valence. More specifically, the value of the ARGST feature is not simply the concatenation of the values of the valence features SPR and COMPS.

4.6.1 Phrase structure and valence features in HPSG In HPSG, the combination of heads with their sisters is governed by various rules, two of which are crucially relevant in the present context. The HEAD-COMPLEMENT RULE combines the head (H in (36a)) with its complements (1 to n in (36a)). The resulting phrase inherits the HEAD value of its head-daughter, but crucially, not the value of the COMPS feature whose specification has been satisfied, so that it has an empty list as the value for COMPS. Similarly, the HEAD-SPECIFIER RULE combines the head with its specifier (1 in (36b)), satisfying the valence requirement. (36)

This is illustrated in tree (37). In order to make these rules and the tree fully clear, we need to introduce the way STRUCTURE SHARING works in HPSG. It is indicated by boxed numbers. Putting the same boxed number in several places in a feature structure or rule, indicates that the corresponding values are token identical. Thus, in tree (36), the 2 expresses the fact that (p. 99) the feature structure which is the sole element in the value of the COMPS list of saw is token identical to the feature structure describing the NP Pat. This means that it would be entirely redundant to replace ‘COMPS〈2〉’ by ‘COMPS 〈2NP〉’ (of course, the tree would have expressed exactly the same property if we had had the latter above saw and only the former above Pat). Similarly, in (36a), the shared boxed

Page 25 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar numbers indicate that the feature structures describing the items on the COMPS list and those describing the corresponding daughters are token identical.27 (37)

Similarly, tree (38) provides an example with the auxiliary will. (38)

Page 26 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.6.2 The Argument Realization Principle and ellipsis Given this set-up, we can provide an analysis of the ellipsis of complements or specifiers simply by allowing non-canonical constructions for heads, in which items mentioned in the ARG-ST do not appear in the valence features (such constructions are non-canonical in the sense that they do not respect the general formulation of the ARGUMENT REALIZATION PRINCIPLE provided in (41)). For instance, we can account for verb phrase ellipsis by constraint (39a) which allows for auxiliaries with a noncanonical relation between the COMPS and ARG-ST lists. Constraint (39a) will be inherited by will (since it is an auxiliary), providing it with the alternate COMPS list given in (39b), and thus giving rise to a structure like (40) (see, e.g., Kim 2006).28 (p. 100)

(39)

Page 27 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

(40)

Since (39a) applies to all auxiliaries, we correctly express the classical generalization that VPE is systematically available behind auxiliaries. Specifying that the VP argument of pre-elliptical auxiliaries is pronominal essentially captures the intuition of (p. 101)

Page 28 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Schachter (1978) and means that it will be interpreted through general principles for resolving anaphoric and/or exophoric dependencies (cf. e.g. Hardt 1993).29 In the light of this example, we can state the ARGUMENT REALIZATION PRINCIPLE as follows: (41)

In (41), ⊕ is the list concatenation operator. For the purposes of this chapter, noncanonical items on the ARG-ST list will always correspond to null pronominals.30 We can now more precisely indicate the analysis for null pronominal subjects and objects, as in the Spanish and Brazilian Portuguese examples (3) and (4), repeated here. (42)

(43)

The constraint in (41a) allows a pronominal subject argument to appear on the ARG-ST list but not on the SPR list, giving rise to (45) as the analysis for (43). (44b), on the other hand, allows for a pronominal object argument that does not appear on the COMPS list, and provides (42) with an analysis similar to that given for will in (40).31 (p. 102)

(44)

Page 29 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

In (44a), B is the COMPS list which is concatenated to a list consisting of a non-canonical nominative personal pronoun NP in the ARG-ST. In (44b), this non-canonical argument is missing in the COMPS list (list B may of course be empty). (45)

Page 30 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.7 HPSG ellipsis resolution in a dialogue setting Dialogue is the primary setting for language, and elliptical language, where content is resolved largely based on context, is particularly characteristic of this setting. In this section we offer an explicit model of dialogue context and show how this can be used in the analysis of various elliptical constructions. In so doing we also introduce an alternative version of HPSG, using the formalism of Type Theory with Records. As we argue in 4.7.3, this version (p. 103) of HPSG, HPSGTTR—or at least its defining characteristics—is needed for an analysis that strives to deal with the challenges of spoken language.32

4.7.1 A dialogical working example We will illustrate how HPSG analyzes ellipsis constructions with reference to the constructed example in (46). A poses a wh-question. This triggers clarification interaction concerning the intended reference of the NP ‘Bo’ via a reprise fragment (see section 4.7.4.4); B’s question in (2) is answered via the short answer in (3) (see section 4.7.4.2); the original question posed in (1), three turns earlier, is then the antecedent for the VPE utterance in (4); via inference, (4) triggers a (direct) sluice in (5) (see section 4.7.4.3). B responds with the short answer in (6). This, in turn, triggers the VPE query in (7), which exemplifies a self-correction utterance (see section 4.7.4.5). This gets answered by the polar particle ‘no’ (see section 4.7.4.1). (46)

(46) exemplifies various claims we made earlier: • The role of inference in ellipsis resolution (e.g., in (2) resolution is driven by B’s inability to resolve the reference of ‘Bo’ in (1)—there is no salient linguistic content around to provide the content ‘Who do you mean “Bo”?’ for this NSU; similarly, in (5) the resolution emerges from the possibility that Jack does not constitute the sole answer to (1)). Page 31 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar • The need for resolution to operate across turns (virtually all cases), in some cases multiple turns (e.g., (4)). • The need to project structural (categorial and phonological) information into context in order to capture cross-turn parallelism of varying degrees (e.g., segmental identity in (2), categorial identity in (5)). • The interleaving of self- and other-repair in ellipsis resolution (e.g., (2) and (7)). The account we propose is based on two main components. First, independently motivated processes of dialogue context dynamics—these are described in section 4.7.2, with a brief sketch of the presupposed semantic framework in section 4.7.3. Second, constructional/lexical specification that can interface directly with dialogue context that contains both linguistic and non-linguistic information. These are sketched in section 4.7.4. The two components are combined in an analysis of the above dialogue in section 4.7.5, which concludes the chapter. (p. 104)

4.7.2 The dialogue gameboard

Our account of dialogue context follows that developed within KoS (cf. n. 5 and Ginzburg 1994; Larsson 2002; Ginzburg and Cooper 2004; Fernandéz 2006; Purver 2006; Ginzburg and Fernandez 2010; Ginzburg 2012).33 KoS is a theory that combines an approach to semantics inspired by situation semantics and dynamic semantics with a view of interaction influenced by Conversational Analysis. On the approach developed in KoS, there is actually no single context—rather, analysis is formulated at a level of information states, one per conversational participant. Each information state consists of two ‘parts’, a private part and the dialogue gameboard, inspired by Lewis (1979), that represents information that arises from public interactions. For recent psycholinguistic evidence supporting this partition see, for instance, Brown-Schmidt et al. (2008). The structure of the dialogue gameboard (DGB) is given in Table 4.2. The Spkr and Addr fields allow one to track turn ownership; Facts represents conversationally shared assumptions; VisualSit represents the dialogue participant’s view of the visual situation and attended entities; Pending represents moves that are in the process of being grounded and Moves represents moves that have been grounded; QUD tracks the questions currently under discussion, though not simply questions qua semantic objects, but pairs of entities which we call InfoStrucs: a question and an antecedent subutterance; motivation for this view of QUD is given below. Table 4.2 Dialogue gameboard

Page 32 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

In terms of the DGB, the model of context in Sag and Hankamer (1984) had some version of Facts (a mental model as in Johnson-Laird 1983 or situational model like Zwaan and Radvansky 1998) and of Moves (a ‘propositional textbase’; see Fletcher 1994). The DGB, then, goes beyond such a view of context in at least two significant ways which derive from the fact that it is intended as a contextual resource for dialogical interaction, in contrast to earlier models of context, intended to process text or monologue. The two (p. 105)

fundamental innovations relate to QUD and Pending. Questions get introduced into QUD by a number of processes. These include: querying (asking q makes the question q QUD-maximal); assertion (asserting p makes the (polar) question p? QUD-maximal); accommodation triggered by clarification interaction (e.g., if A’s sub-utterance u is difficult to resolve or involves an error, the issue ‘what did A mean by u?’ can become QUD-maximal); accommodation triggered by interaction in a conversational genre (e.g., in a customer/client interaction, the issue ‘what does the client require?’ can become QUD-maximal). Adopting the assumption that (structural) parallelism typically exhibits a similar time course to the salience of the relevant entity of QUD, we can capture such effects by viewing QUD as tracking not simply questions qua semantic objects, but pairs of entities: a question and an antecedent sub-utterance. This latter entity provides a partial specification of the focal (sub-)utterance, and hence it is dubbed the Focus Establishing Constituent (FEC) (cf. parallel element in higher-order unification-based approaches to ellipsis resolution (e.g. Gardent and Kohlhase 1997); Vallduví (2016) relates the FEC with a notion needed to capture contrast. Thus, the FEC in the QUD associated with a wh-query will be the wh-phrase utterance, the FEC in the QUD emerging from a quantificational utterance will be the NP utterance, whereas the Page 33 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar FEC in a QUD accommodated in a clarification context will be the sub-utterance under clarification. Pending is the contextual resource corresponding to utterances that are still in progress or under clarification. One of the key structuring aspects of conversational interaction is the ubiquitous metacommunicative interaction between the participants observable via periodic verbal and gestural backchannels, and occasionally via clarification questions of various kinds (e.g., ‘What did the speaker mean by “…”’). This cycle of grounding (Clark 1996) and clarification interaction (Ginzburg and Cooper 2004) is also present for a given speech participant monitoring her own speech, overt evidence for which are various disfluencies such as hesitations (‘the next thing to say is problematic’) and selfcorrections (‘the recent sub-utterance needs fixing’). Ginzburg (2012) offers detailed arguments on this issue, including considerations of the phonological/syntactic parallelism exhibited between CRs (clarification requests; cf. 4.2.3) and their antecedents and the existence of CRs whose function is to request repetition of (parts of) an utterance; see (10). Taken together with the obvious need for Pending to include values for the contextual parameters specified by the utterance type, Ginzburg concludes that the type of Pending combines tokens of the utterance, its parts, and of the constituents of the content with the utterance type associated with the utterance. An entity that fits this specification is the locutionary proposition defined by the utterance: in the immediate aftermath of a speech event u, Pending gets updated with a record whose two components are u and Tu, a grammatical type for classifying u that emerges during the process of parsing u. Locutionary propositions are instances of Austinian propositions; see Barwise and Etchemendy (1987). The original Austinian conception was that s is a situation deictically indicated by a speaker making an assertion whose truth involves s being of type T. A locutionary proposition specializes this notion to the case of a speech event u and a grammatical type Tu, in other words, an entity such as the sign in the sense of HPSG. The relationship between u and Tu—describable in terms of the proposition:

pu=[ sit=usit‐typ=Tu] —can be utilized in providing an analysis of grounding/clarification-interaction conditions of an utterance: (p. 106)

(47)

Page 34 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.7.3 Grammar and semantics in Type Theory with Records In this section, we introduce the basic semantic and grammatical notions we use in the rest of this section. KoS is formulated within the framework of Type Theory with Records (TTR) (Cooper 2005; Cooper and Ginzburg 2015), a model-theoretic descendant of Martin-Löf Type Theory (Ranta 1994) and of situation semantics (Barwise and Perry 1983; Ginzburg and Sag 2000). TTR enables one to develop a semantic ontology, including entities such as events, propositions, and questions. With the same means TTR enables the construction of a grammatical ontology consisting of utterance types and tokens and of an interactional domain in which agents utilize utterances to talk about the semantic universe. Ginzburg (2012: ch. 5) provides detailed argumentation for what makes TTR advantageous for a dialogically oriented grammar and, in particular, for the advantages of analyzing signs in terms of record types rather than typed feature structures, as done in standard HPSG (henceforth HPSGTFS) and other ‘unification-based’ frameworks. We summarize the arguments here: • Object level types and tokens: HPSGTTR provides access to both types and tokens at the object level, specifically here to utterance tokens (or speech events) and utterance types (also known as signs). This is crucial in formulating grounding/ clarification-interaction conditions of an utterance, as in (47), which constitutes the basic dynamic process in utterance processing. It is also a crucial ingredient in providing semantics for repair constructions (see, e.g., sections 4.7.4.4 and 4.7.4.5) and for quotation (Bonami and Godard 2008; Ginzburg and Cooper 2014), which are anaphoric to utterance tokens and involve inference processes that make references to utterance types. HPSGTFS specifies a grammar in terms of types but has no corresponding means of directly referring to tokens that constitute speech events.34 • Direct access to semantic entities: HPSGTTR directly provides semantic entities, whereas HPSGTFS simulates them. Thus, whereas HPSGTTR can effect variable binding, function definition, and abstraction, in HPSGTFS these notions are merely coded up.35 To take a concrete example: as we will shortly see, HPSGTTR represents the contextual parameters of a meaning by means of record types. A token of such a type—a record—represents a specific context for an utterance. In HPSGTFS there are ways of representing contextual parameters but there is no explicit way of modeling the fact that these need to be instantiated in context. (p. 107)

For current purposes, the key notions of TTR are the notion of a judgment and the notion of a record. • Typing judgments A typing judgment a:T classifies an object a as being of type T. • Records A record is a set of fields assigning entities to labels of the form (48a); a concrete instance is exemplified in (48b).36 Records are used here to model events and states, including utterances, and dialogue gameboards.

Page 35 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar (48)

• Record Types: A record type is simply a record where each field represents a judgment rather than an assignment, as in (49a). Record types are used to model utterance types (HPSG signs, as in (49b)), as components of semantic entities such as propositions and questions, and to express rules of conversational interaction.37 (p. 108)

(49)

Page 36 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

The basic relationship between records and record types is that a record r is of type RT if each value in r assigned to a given label li satisfies the typing constraints imposed by RT on li. To take a simple example, (50a) classifies the situation in Nome (see (48b)) at a certain point in time, assuming the conditions in (50b) hold. (50)

Page 37 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

To take a more linguistic example, a conversational state r will be a record such that the conditions in (51) hold—it is of the type dialogue gameboards must satisfy; in other words, r should have the make-up in (51b) and the constraints in (51c) need to be met: (p. 109)

(51)

The final two notions we need are propositions and questions. We have already mentioned the notion of an Austinian proposition. Truth conditions for Austinian propositions are defined in (52): (52)

Page 38 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Extensive motivation for the view of questions as propositional abstracts has been provided in Ginzburg (1995) and Ginzburg and Sag (2000)—TTR contributes to this by providing an improved notion of simultaneous, restricted abstraction: A (basic, noncompound) question is a function from records into propositions. Here we assume a minor refinement of this view proposed by Ginzburg et al. (2014a).38 This involves introducing a notion of Austinian questions, defined as records containing a record and a function into record types, the latter associated with the label ‘abstr(act)’. The role of wh-words on this view is to specify the domains of these functions; in the case of polar questions there is no restriction, hence the function component of such a question is a constant function. (53) exemplifies this for a unary ‘who’ question and a polar question. For notational simplicity, we will notate questions in what follows just in terms of the associated abstract. (p. 110)

(53)

We combine all elements introduced in this section in a sketch of how clarification interaction can be specified. In principle, one could have a theory of clarification interaction based on generating all available CRs an utterance could give rise to. But in

Page 39 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar practice, there are simply too many to be associated in a ‘precompiled’ form with a given utterance type. Instead, CRs can be specified by means of a uniform class of conversational rules, dubbed Clarification Context Update Rules (CCURs) in Ginzburg (2012). Each CCUR specifies an accommodated MaxQUD (the maximal element of QUD) built up from a sub-utterance u1 of the target utterance, the maximal element of Pending (MaxPending). Common to all CCURs is a license to follow up MaxPending with an utterance which is co-propositional with MaxQUD.39 (54) is a simplified formulation of one CCUR, parameter identification, which allows B to raise the issue about A’s sub-utterance u0: what did A mean by u0? (‘pre’ expresses the preconditions and ‘effects’ specifies the effect of the rule; ‘MaxPending.sit.constits’ expresses a path): (54)

4.7.4 Non-sentential utterance constructions The detailed theory of context sketched in previous sections enables the development of a grammar of the various types of sentential fragments discussed earlier. The basic strategy adopted in KoS/HPSGTTR to analyze NSUs is to specify construction types where the combinatorial operations integrate the denotata of the fragments with elements of the DGB. We sketch how this can be done with several such construction (p. 111)

Page 40 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar types; a detailed account of a wide variety of sentential fragments analyzed in such terms can be found in Fernández (2006) and Ginzburg (2012).

4.7.4.1 Propositional lexemes yes In its most straightforward use, ‘yes’ affirms the proposition p, where p? is the maximal element in QUD, either because the polar question p? was posed or as a side effect of an assertion of p. Hence, the lexical entry in (55). Categorically ‘yes’ is classified here as an adverbial40 that can only occur as a root utterance (cf. n. 8), whereas semantically its content is the proposition component of a polar question:41 (55)

no ‘No’ has a number of distinct uses, including an exophoric use (on which see Cooper and Ginzburg 2015, where a treatment of negation in TTR is discussed) and a ‘pragmatic denial’ use, for which see Tian and Ginzburg (2017). Here we provide a lexical entry that captures the basic fact about the most prototypical use of ‘no’ in dialogue—its content is always a negative proposition. This can arise either by negating a proposition, when MaxQUD is a positive polar question (e.g., A: Did Bo leave? B: No (= Bo did NOT leave)), or by affirming it when MaxQUD is a negative polar question (e.g., A: Bo didn’t leave? B: No (= Bo did not leave)): (56)

4.7.4.2 Declarative fragments Consider the declarative fragment (57a): (57)

Page 41 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

B’s utterance in (57a) can receive a variety of contents, depending on the context in which it is uttered and the intonation contour it receives: it can be interpreted as a short answer, as in (57b); it can be interpreted without any prior utterance, as in (57c), though in such a case various paraphrases are possible, depending on the conversational genre;42 it can also be interpreted as the (‘metalinguistic’) correction in (57d). (p. 112)

The different mechanisms underlying these resolutions can be uniformly described by the schema in (57e). This indicates that the content of the construction type declarativefragment clause arises by predicating the propositional function constituted by the maximal element of QUD of the content of the bare fragment utterance, a generalization of a rule proposed already in Hausser and Zaefferer (1979). The particular content exhibited in (57b) could arise because the issue ‘What did you buy in the bakery?’ is MaxQUD as a result of A’s query; (57c) arises given that the issue ‘What does the current customer want to buy?’ is a characteristic issue of the BakeryShopping genre (as it is of many related genres); the content in (57d) could arise if B decided not to ground A’s utterance, but using the parameter identification conversational rule (discussed in section 4.7.4.4) to initiate repair interaction, accommodates the issue ‘What did you mean by utterance “four crescents”?’ as MaxQUD. We have also emphasized that different NSU constructions exhibit morphosyntactic and/ or phonological parallelism with their antecedents. In other words, not only the combinatorial semantics of NSU constructions integrates information from the DGB, but this is also potentially true of the morphosyntactic and phonological specifications of such constructions. Here we utilize this to deal with connectivity: we specify that the fragment has to match the categorial specification of the FEC. In light of this, we can write a specification of decl-frag-cl as in (58). Categorially the construction is sentential, as discussed in section 4.2.1.2; it has one DGB parameter—i.e. contextual parameter—the maximal element of QUD, whereas as just mentioned, its content arises by functional application of MaxQUD to the entity denoted by the fragment: (p. 113)

(58) Page 42 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

4.7.4.3 Direct sluicing Direct sluices are analyzed in a similar fashion. In Ginzburg and Sag (2000) declarativefragment clauses and sluices are unified as subtypes of a phrasal type dubbed hd-frag-ph. This construction builds a verbally headed phrase from an NP or PP under the constraint that the context’s FEC is categorially identical to and coindexed with it.43 Three aspects distinguish a direct sluice, specified by the clause type sluice-int-cl from a declarative-fragment clause: the sluice is constructed from a wh-phrase, the sluice clause denotes a question, and, arguably, it has a distinct contextual background. What is the contextual background of a direct sluice? Ginzburg and Sag (2000) and Ginzburg (2012) argued that this involved the QUD-maximality of a quantified polar question (cf. n. 15 and surrounding text). This seems to be accurate in most cases. There are, however, several cases where this is less persuasively so. An alternative, suggested by Ivan Sag (p.c.), is to assume that it is the corresponding wh-question. This, however, requires a somewhat more ‘inferential’ view of QUD since in many cases there will not have been a prior utterance of a wh-interrogative. Nonetheless, since it is clear that QUD needs to be updated by some restricted inferences in a number of cases, as we have already discussed, this is not a qualitative modification. Indeed, as will become clear from the discussion below, from a compositional semantics point of view, it makes little difference. Assuming the view of wh-phrase content sketched in section 4.7.3 and remaining agnostic for the moment as to the QUD-specification of a sluice’s context, we can explicate how Page 43 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar the content of a sluice arises straightforwardly, using exactly the content of the wh-phrase and MaxQUD. As summarized in (59a), the sluice denotes a question (i.e., a function from records into propositions) whose domain is the type denoted by the wh-phrase and whose range is that given by MaxQUD’s proposition where the wh-phrase’s (p. 114) variable is substituted for that associated with the FEC (‘↝’ signifies ‘reduces to by β-conversion’). (59)

We note that this view extends to two tricky cases for deletion-based accounts. First, exophoric cases like (32). In such cases awareness of the genre is what leads QUD to be updated (see Ginzburg 2012 for details). The essential idea of that proposal is that a given genre can be characterized, in part, by a partially ordered set of questions, discussion of which constitutes its defining subject matter. At appropriate points these questions can be accommodated into QUD without being uttered overtly.44 The way this account applies to declarative fragments was explained in the discussion of (57c). This is what licenses exophoric sluices as in (60), illustrating the taxi-cab genre. (60)

Page 44 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar Second, incremental cases like (61a), already discussed in section 4.4.3. Assuming an incremental view of semantic processing (Purver et al. 2011; Rieser and Schlangen 2011; Ginzburg et al. 2014b), the sluice in (61a) is predicted to mean, immediately after it is uttered, ‘Who is that person (that has some as yet uninstantiated property)’, whereas ‘John?’ means ‘Is it John (that has some as yet uninstantiated property)?’. The incremental view merely needs the (incremental context) assumption that QUD can get updated word by word (rather than at turn boundaries), roughly as in (61b):45 (p. 115)

(61)

4.7.4.4 Reprise sluicing and reprise fragments A key component of our analyses of elliptical reprise constructions are the questions accommodated into QUD as a consequence of the clarification interaction process triggered by incomplete understanding of the previous utterance, as described in section 4.7.3. With that in place, the existing grammar can then directly yield the requisite readings for reprise sluicing and for the confirmation readings of reprise fragments. Assume the utterance to be clarified is (62a). B uses the CCUR parameter identification to build a context as in (62b): (62)

Given this, the analysis of the construction is illustrated in (63): the construction declfrag-cl builds the proposition Mean(A,u2,b); the construction polarization builds a polar question from this: (63)

Page 45 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

Intended content readings of RFs involve a complex mix of a prima facie non-transparent semantics and phonological parallelism. Independently of intended content readings, we need to capture the utterance anaphoricity of ‘quotative’ utterances such as (64): (p. 116)

(64)

Page 46 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar We assume the existence of a grammatical constraint allowing reference to a subutterance under phonological parallelism. (65) exemplifies one way of formulating such a constraint: the phon value is type identical with the phon value of an utterance identified with the FEC, whereas the content is stipulated to be the utterance event associated with the focus-establishing constituent:46 (65)

With this in hand, we turn back to consider the issue of how intended content RFs arise grammatically. It is worth emphasizing that there is no way to bring about the desired content using decl-frag-cl, the short-answer/reprise sluice phrasal type we have been appealing to above, regardless of whether we analyze the NP fragment as denoting its standard conventional content or alternatively as denoting an anaphoric element to the phonologically identical to-be-clarified sub-utterance. This is a prototypical instance of appeal to constructional meaning—a complex content that cannot be plausibly constructed using ‘standard combinatorial operations’ (function application, unification, etc.) from its constituents. Thus, one way of accommodating intended content RF is to posit a new phrasal type, qud-anaph-int-cl. This will encapsulate the two idiosyncratic facets of such utterances, namely the MAX-QUD/CONTENT identity and the hd-dtr being an utt-anaph-ph: (66)

Given this, we can offer the following analysis of (67): (67)

B lacks referent for ‘Bo’; uses parameter identification to update MaxQUD accordingly: Page 47 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar (p. 117)

(68)

Using qud-anaph-int-cl yields: (69)

Page 48 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar 4.7.4.5 Disfluency Disfluencies are viewed as a performance phenomenon in most formal grammatical treatments.47 Ginzburg et al. (2014b) provide extensive argumentation for the need to integrate disfluencies in the competence grammar, demonstrating that they participate in semantic and pragmatic processes like anaphora, implicature, and discourse marker content, as well as being subject to cross-linguistic variation and also exhibiting some universals. Ginzburg et al. (2014b) develop their account of disfluencies in KoS by extending the account mentioned in previous sections of the coherence and realization of clarification requests: as the utterance unfolds incrementally there potentially arise questions about what has happened so far (e.g., what did the speaker mean with subutterance u1?) or what is still to come (e.g., what word does the speaker mean to utter after sub-utterance u2?). These can be accommodated into the context if either uncertainty about the correctness of a sub-utterance arises or the speaker has planning or realizational problems. Thus, the monitoring and update/clarification cycle is modified to happen at the end of each word utterance event, and in case of the need for repair, a repair question gets accommodated into (p. 118) QUD. In this way, the coherence of ‘spelled-out’ self-repairs such as (70a) can be explained, but also of elliptical self-repairs such as (70b). The latter is analyzed as a declarative-fragment clause (see section 4.7.4.2), where the antecedent question is the accommodated ‘what did A mean by “earphones”?’ with the utterance ‘earphones’ being the FEC: (70)

4.7.5 A worked example As exemplification of what we have done in this section, we return to the example we introduced earlier, (46), where the numbering henceforth relates to the sequence of utterances in that example. Our illustration involves a sequence of dialogue gameboards that arise sequentially via updates triggered by conversational rules. We will abuse notation somewhat and notate by ‘:=’ cases where a field gets assigned a new value. Another point to note is that in some cases we consider the dialogue from A’s point of view (in which case we have fields such as ‘A.LatestMove’); in other cases from B’s point of view (in which case we have fields like ‘B.QUD’). This is an inevitable consequence of the fact that in dialogue there is not always a context that is identical for the conversational participants. (p. 119)

(p. 120)

(p. 121)

(1)

Page 49 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

(3)

(3)

Page 50 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

(4)

(5)

Page 51 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

(6)

(7)

Page 52 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

48

(8)

Page 53 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar

Acknowledgements This piece is dedicated to the memory of our friend, teacher, and collaborator, Ivan Sag. Ivan was originally intended as a co-author of this piece, but died before it was drafted. His inspiration and many ideas are evident, though the chapter is much the poorer due to his absence. We would like to thank Doug Arnold, Bob Borsley, Rui Chaves, Jong-Bok Kim, Jason Merchant, François Mouret, Stefan Müller, Enric Vallduví, and Tom Wasow, as well as two anonymous reviewers for comments on the first version. We are also very grateful for the help and extreme patience of the editors, Jeroen van Craenenbroeck and Tanja Temmerman, whom we kept waiting endlessly. We acknowledge the support of the French Investissements d’Avenir-Labex EFL program (ANR-10-LABX-0083) and the Disfluency, Exclamations, and Laughter in Dialogue (DUEL) project within the projets francoallemand en sciences humaines et sociales funded by the ANR and the DFG.

Notes: (1) See Sag (2012) and Müller (2015b) for recent synopses, and Sag et al. (2003) for a pedagogical introduction. For further references, see the HPSG bibliography, maintained by Stefan Müller: . (2) The version of HPSG formalized in TTR (Type Theory with Records, cf. Cooper 2012; Ginzburg 2012), presented in section 4.7 refines this view. It takes a dialogue-oriented view of grammar in which a grammar provides types that characterize speech events (for detailed discussion, see Ginzburg and Poesio 2016). Thus, it proposes a linguistic ontology that includes both tokens (speech events, modeled as records) and types that characterize such tokens (signs, modeled as record types). This view of grammatical tokens and types plays an important role in our account of conversational interaction, and is key to understanding various elliptical resolution processes, as discussed in section 4.7. (3) Sag et al. (2012) and Sag (2012) propose an integration between Berkeley-style Construction Grammar and HPSG under the name of Sign-Based Construction Grammar (SBCG). Under this view, constructions (including lexical constructions) are understood to Page 54 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar be constraints on classes of signs and their components, which are organized into latticelike arrays of types and subtypes. (4) In fact, recent work on phenomena such as self-repair and exclamative interjections (e.g., Ginzburg et al. 2014b, 2017), following earlier work in Dynamic Syntax (e.g., Purver et al. 2006 and Chapter 9 of this volume), takes a stronger line, namely that the grammar needs to be stated incrementally. See section 4.7.4.5 for a brief discussion. (5) KoS—a toponym, not an acronym—is a framework for describing dialogue interaction, which we will introduce in section 4.7. (6) A further case of constructional ellipsis, namely nominal ellipsis, has not been discussed in this chapter for reasons of space. See, e.g., Nerbonne et al. (1989, 1990) and Nerbonne and Mullen (2000) for HPSG analyses using a null head, and Branco and Costa (2006) and Arnold and Spencer (2015) for analyses avoiding an empty morpheme. (7) We use the traditional name, rather than the more appropriate ‘Post-Auxiliary Ellipsis’ suggested by Sag (1976a: 53): what is ellipted is not always a (surface) VP (in particular, it can be any predicative complement of the copula be), and it doesn’t even have to be a surface constituent. Furthermore, in examples of NCA like (2a), a VP is ellipted, but the usual criteria show that this is not VPE; see Miller and Pullum (2014: 6). Note also that we are using ‘auxiliary’ in the classical sense in English linguistics, namely a verb that has the NICE properties (cf. Huddleston 1976), so that copular be and some uses of the main verb have are included. (8) Ginzburg and Sag (2000) use the feature ± IC to distinguish embeddable from nonembeddable clauses (as in e.g. (55) and (56)). See van Craenenbroeck and Lipták (2006: section 6.3) and Temmerman (2013) for proposals as to which languages do and do not allow such embedded NSUs. (9) For reasons of space, we have not dealt with coordinate ellipsis in this chapter, beyond what little is said here about the specific case of gapping. Among the HPSG works on the topic are: Sag et al. (1985); Beavers and Sag (2004); Mouret (2006, 2007a); Chaves (2007); Yatabe (2012). (10) Morphological form identity (cf. Booij 1985) imposes identity beyond phonological identity, in particular the same morphemes must be used with the same senses. (11) Chaves uses small caps to signal contrastive focus, with an L+H* tone, and square brackets to identify the RNRaised string. (12) Abeillé et al. (2015) report corpus data, among which (8) and (9), and the results of acceptability experiments on RNR in French. These provide evidence that there can be minor mismatches in content-less material (e.g., determiners and prepositions). Note that in (8) it would be ungrammatical to replace des by de in what corresponds to the ellipted

Page 55 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar segment so that there is not strict identity. Similarly, in (9), parvient requires the preposition à. ((8))

((9))

They consequently suggest an analysis which allows content-less material to be asymmetrically ellipted on the left, whereas content-full peripheral material must be shared on the right. Bîlbîie (2013) reports similar examples in English found in the Penn Tree Bank, e.g. They were also as liberal or more liberal than any other age group in the 1986 through 1989 surveys (WSJ). Experimental work is required to decide whether these English examples have a similar status to that of the French examples above, or whether they should be considered as errors. (13) ‘What?’ is multiply ambiguous. It can be used to signal that the previous utterance was not understood overall (it is then classified using the tag Wot); it can also be used to clarify the inanimate filler of an argument role (one that is implicit in the case of ‘Did Bo leave?’ and makes sense only on certain senses of ‘leave’; it is then classified using the tag Reprise Sluice). (14) Fernández et al. (2007) develop a machine-learning based algorithm for this classification which achieves F-scores of approximately 85 percent. (15) In other words, further elaboration of the referents associated with a previously occurring quantifier NP or with elements which license such a content via inference (e.g., ‘Bo ate.’ implies ‘There is something Bo ate.’ and ‘There is a time t such that Bo ate at t.’). (16) Nielsen (2005); Bos and Spenader (2011) on VPE and Beecher (2008) on sluicing also show how corpus studies allow one to bring to light new theoretically important phenomena. Though these studies are not set in an HPSG framework, their assumptions are largely compatible. (17) Though not assuming empty morphemes is the default position in HPSG, various analyses involving such morphemes have been proposed. Specifically, the analysis of filler-gap constructions in Pollard and Sag (1994) did involve a trace, and, as mentioned in Page 56 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar n. 6, several analyses of nominal ellipsis have involved hypothesizing an empty nominal head. Similarly, one could analyze VPE or nominal argument ellipsis as involving an empty VP or NP position in the syntax, rather than no position at all: instead of having a situation where an element present in the argument structure is missing on one of the valence lists (as proposed in section 4.2.1.1), it is possible to include the element in the valence list, while specifying that it will have an empty phonological realization. (18) What of the putative cross-linguistic generalization known as P-stranding relevant both to sluicing and short answers? For an excellent discussion of the empirical and theoretical status of this generalization see Jacobson (2016b). Building on earlier work by inter alia Sag and Nykiel (2011), Jacobson points to the significant empirical unclarity the generalization faces. Jacobson sketches a potential account of certain facts related to the generalization available to non-deletion-based accounts such as Ginzburg and Sag (2000) and her own Categorial Grammar approach. She further points out challenges faced by both deletion-based and surfacist accounts in providing a complete account of the data. (19) Consider an example like (18a), where the second speaker’s answer Mine is assumed to have a structure like That is mine. The second speaker is assumed to be able to ellipt here through recovery of the contents of the first speaker’s question. However, the question of exactly how this content is recovered is not addressed, presumably being considered to be a problem of performance that is irrelevant to a theory of competence. But this makes it impossible to address the question of constraints on the recovery mechanism, which we have just suggested are linked to the accessibility of MAX-QUD. The question becomes even more difficult in a case like (18d) where it is unclear how the speaker is expected to recover a specific syntactic structure from the exophoric context. (20) As pointed out by Fox and Lasnik (2003), there are cases where VPE interacts with wh-movement triggering unacceptability, despite the absence of island violations in the putative unpronounced structure. For example, neither the non-elliptical variant, nor the sluice in (22a) raise a problem, but the VPE variant in (22b) is ungrammatical: ((22))

See Merchant (2008b) for a proposal aiming to account for the ill-formedness of (22b) in terms of the Max-Elide constraint which “Roughly put, […] states that if ellipsis applies in a structure with a wh-trace, ellipsis should target the largest constituent possible” (p. 141). (21) We take the term ‘antecedent-trigger’ from Cornish (1999), who uses it to designate the segment of text allowing one to construct the antecedent (he uses ‘antecedent’ to

Page 57 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar refer to the discourse-model representation making interpretation of the anaphor or ellipsis possible). (22) Elbourne (2008) proposes an analysis of split antecedents according to which the antecedent of the ellipted VP is constructed from the two VPs present in the discourse. Though his proposals would work for classical split antecedents such as (25), it is unclear how they could deal with the examples in (26). (23) Kim et al. (2011) develop an alternate proposal suggesting that the reduced acceptability of mismatches can be attributed to processing difficulties. Miller and Hemforth (2014a) point out that this is compatible with the proposals of Kehler and Kertz, and suggest that discourse factors and processing factors can cumulatively reduce acceptability. (24) Note that asking a polar question and asserting a proposition are central ways of making a polar alternative salient (see section 4.7.2), whereas asking a wh-question is a central means of establishing the salience of a property. Although we will not attempt to formalize these discourse constraints in this chapter, we note that this could be done quite straightforwardly in terms of (Maximal)-Question Under Discussion, using the KoS framework developed in section 4.7. (25) In section 4.7.5, we analyze the case of sluices of the form ‘Who else?’, which are compatible with a referential antecedent ‘X V’ed’. Nonetheless, their acceptability clearly requires the issue ‘Somebody else apart from X V’ed’ to be worthy of discussion. (26) Merchant (2010) has the most developed discussion of exophoric uses from an unpronounced structure perspective that we are aware of. He proposes that examples of NSUs like (32a), (32b), and (33) should be analyzed using the concept of scripts (in the sense of Schank and Abelson 1977). It is not clear, however, whether an example like (32c) is amenable to such an analysis. And it is even less clear how scripts could account for the wide range of exophoric VPE cases reported in Miller and Pullum (2014) and illustrated here in (31). (27) Signs in HPSG can be lexical or phrasal. In the latter case, formally speaking, the constituent structure is modeled within the feature structures using the features MOTHER, HEAD-DTR (Head-Daughter), and NON-HEAD-DTRS (Non-Head-Daughters). Thus tree representations of the type given in (37) are an informal simplification enhancing readability. (28) Note that the properties [nom] for the NP subject and [base] for the VP are respectively inherited from the basic lexical constraints for auxiliaries in general and for modal auxiliaries in particular. Note also that some recent versions of HPSG would not have the non-branching VP which we have used for simplicity in (40) (a similar comment can be made for the non-branching S node in (45)).

Page 58 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar (29) Further work on VPE in HPSG includes Grover et al. (1995) on strict/sloppy ambiguity; Lappin (1999) and Gregory and Lappin (1999) on antecedent-contained deletion; Egg and Erk (2002) and Arnold and Borsley (2010) on auxiliary-stranding relative clauses; and Lee (2012). For an early HPSG analysis of pseudogapping as a case of VPE where the pre-elliptical auxiliary selects a complement, see Miller (1990). (30) Other non-canonical items include extracted elements and certain types of clitics; see, e.g., Miller and Sag (1997). (31) Brazilian Portuguese allows Ana colocou ele na estante, with the overt strong object pronoun ele, as a variant of (42). These two sentences will receive exactly the same analysis with respect to syntax and semantics, except that the pronominal NP will appear both on the ARG-ST and on the COMPS list in the latter case, causing the object pronoun to be overtly realized. The intended anaphoric link to the object of the previous sentence in the context provided is mediated through the ARG-ST list, which is the same in both cases, so that anaphoric resolution works similarly. Note however that the details of the discourse pragmatics will be different, so that constraint (44b) will also have to impose additional properties at that level. (32) More generally, Ginzburg and Poesio (2016) propose constraints that any formalism striving to deal with spoken language needs to satisfy. (33) For an alternative view on how to integrate pragmatic information from context in HPSG, see Bertomeu and Kordoni (2005). (34) Ginzburg (2012) reviews the history of HPSG’s view of formalizing grammars on this score which were, with one exception, formulated either in terms of tokens or, subsequently, types. Pollard and Sag (1987) suggested thinking about feature structures as ‘partial descriptions of signs (or sign tokens) and other linguistic objects which occur as part of signs’ (p. 28). However, this was violently repudiated by Pollard and Sag (1994): ‘One thing that [language] certainly does not consist of is individual linguistic events or utterance tokens, for knowledge of these is not what is shared among the members of a linguistic community’ (p. 14). Paul King, on the other hand, in his formalizations of HPSG (see. e.g., King 1989, 1996), does view a grammar as a characterization of the class of well-formed utterance tokens. In King 1996, he identifies a token as a pair 〈u, I〉 of an entity u and an interpretation I (in a technical sense King develops). From this, he constructs types as equivalence classes of indiscernable tokens. Whether King’s theory could serve to underpin linguistic description of utterances used in repair or quotation is not straightforward to discern since it is not clear that his formalism allows for types as first-class entities. (35) As Penn (2000: 63) puts it (in discussing a related set of issues), ‘At this point, feature structures are not being used as a formal device to represent knowledge, but as a formal device to represent data structures that encode formal devices to represent knowledge’. (36) Assignment of an entity to a label is noted by ‘=’. Page 59 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar (37) In (49b) we make use of the notation [x=k: T] which is based on the notion of a manifest field (Coquand et al. 2003). This is a shorthand for a judgment x : Tk where Tk is the singleton subtype of T whose only witness is k. For instance, ‘CAT = V[+fin] : syncat’ means cat : syncatV[+fin], so in such a case CAT is restricted to the subtype of the type syncat whose sole witness is V[+fin]. (38) This refinement is motivated in part by the need to enable conjunction and disjunction to be defined for questions; it also enables an account of the interaction between questions and adjectives, as in ‘difficult/quick/philosophical question’. (39) Two utterances u0 and u1 are co-propositional iff the questions q0 and q1 they contribute to QUD are co-propositional. Co-propositionality for two questions q0 and q1 means that their range is not disjoint, i.e., there exists a record r such that q0(r) = q1(r). (40) The rationale behind this proposal is that it makes ‘yes’ resemble its counterparts in other languages (see discussion in 4.2.1.2), as well as other adverbials in English (‘probably’, ‘possibly’) which have a use as a propositional lexeme; for English, one might make other decisions on this score, e.g., classifying ‘yes’ as an interjection. (41) Recall that a polar question p? is, in the semantics we propose, a constant function mapping any record r to the proposition p. Hence, in particular applying p? to the empty record [] yields p. (42) See Wittgenstein (1953); Clark (1996) for discussion and exemplification. (43) Of course, in languages where connectivity properties are not as strict as in English, this constraint can be weakened. See Sag and Nykiel (2011) and Nykiel (2013) for an example of how this can be done for sluicing in Polish. Similar analyses are possible for languages like French or Portuguese where a wh-NP can have a PP correlate (cf. n. 18). See also Kim (2015) on sluicing in Korean. (44) More precisely, what gets accommodated are InfoStrucs, in other words question utterance pairs that therefore include also the specification of an FEC. For some discussion as to how this relates to case specification, see Ginzburg (2012: section 7.10). (45) For a detailed account of this, see Ginzburg et al. (2017). (46) Postulating such an ambiguity goes back to Frege and Quine, who suggested that phrases within quotative operators denoted the string itself. (65) makes one simplifying assumption: identifying the phon value of the focus-establishing constituent with that of the utterance anaphoric phrase. In practice, this should only be the segmental phonological value. (47) Though not by psycholinguists: see, e.g., Levelt (1983); Clark and Fox Tree (2002). (48) Why is q2 MaxQUD? In KoS QUD is not taken to be a stack, but rather a partially ordered set. Ginzburg (2012), motivated in part by multi-party dialogue, proposed that when a question q is pushed onto QUD it doesn’t subsume all existing questions in QUD, Page 60 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Head-Driven Phrase Structure Grammar but rather only those on which q does not depend. Hence, wh-questions do not get outranked by polar questions which are their subquestions.

Jonathan Ginzburg

Jonathan Ginzburg is Professor of Linguistics at Université Paris-Diderot (Paris 7). He has held appointments at the Hebrew University of Jerusalem and King’s College, London. He is one of the founders and editor-in-chief (emeritus) of the journal Dialogue and Discourse. His research interests include semantics, dialogue, language acquisition, and musical meaning. He is the author of Interrogative Investigations (CSLI Publications, 2001, with Ivan A. Sag) and The Interactive Stance: Meaning for Conversation (Oxford University Press, 2012). Philip Miller

Philip Miller teaches English and General Linguistics at the Université Paris Diderot (Paris 7). He was recently Visiting Professor at the Universidade de São Paulo. He is the author of two monographs, Clitics and constituents in phrase structure grammar (Garland Publications, 1992) and Strong generative capacity (CSLI Publications, 1999). He has worked on clitics and on perception verbs. His current work is centered on ellipsis and anaphora, with specific interest in verbal ellipsis and verbal anaphora.

Page 61 of 61

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

Oxford Handbooks Online Ellipsis in Dependency Grammar   Timothy Osborne The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.5

Abstract and Keywords This contribution considers how dependency grammars (DGs) view ellipsis phenomena. For the most part, comprehensive explorations of particular ellipsis phenomena are rare in the tradition of DG, but when ellipsis is examined, there is a tendency to posit the existence of null material/nodes to accommodate aspects of ellipsis. One recent development in the DG tradition that is particularly relevant to the analysis of ellipsis is the catena unit. A catena is a word or combination of words that are linked together by dependencies. Most ellipsis phenomena are eliding catena, whereby these catenae are often not constituents. Given the catena unit, a more economical approach to ellipsis has become possible, one that does not need to resort to movement to vacate the remnant from the encompassing constituent before that encompassing constituent is elided. Keywords: catena, Dependency Grammar, DG, Ellipsis, Gapping, left-edge ellipsis, pseudogapping, stripping, VPellipsis

Timothy Osborne

6.1 Dependency-based syntax DEPENDENCY grammars (DGs) assume that hierarchical relationships between words are immediate; these relationships are not mediated by the intermediate groupings associated with phrase structure grammars (PSGs ≈ constituency grammars). Dependencies are directed; they exist between a head word and its dependents. A brief illustration of a dependency tree is provided next, and the corresponding phrase structure tree is included for comparison: (1)

Page 1 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

1

(p. 143)

A main difference between the dependency tree (1a) and the phrase structure

tree (1b) concerns the presence/absence of the phrasal nodes (NP, PP, VP, S). Dependency trees lack these purely phrasal nodes;2 the hierarchical relationships between syntactic units are therefore direct. The distinction between dependency- and constituency-based approaches to natural language syntax illustrated with trees (1a) and (1b) is discussed in numerous places (e.g. Matthews 1981: 71–95, Fischer 1997: ch. 2.4.5, and Osborne 2005).

6.2 Existing DG accounts of ellipsis When one considers ellipsis through the lens of the theories of syntax that assume dependencies more or less along the lines of (1a), one learns that many established DGs have not been so concerned with ellipsis. There are, though, a number of DGs that have at least touched on ellipsis. Some of them will now be briefly considered. Lucien Tesnière, the French linguist who is widely acknowledged as having produced the first comprehensive dependency-based theory of syntax, addressed ellipsis in his main work Elements of Structural Syntax (1959/2015) mostly only in connection with his subtheory of transfer (Fr. translation). He was hence interested in lexicalized forms of ellipsis, rather than in the productive types of ellipsis that more recent theories of syntax acknowledge and explore. However, Tesnière’s analysis of coordination includes two Page 2 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar examples that allow one to assume that he was willing to posit the existence of null material in (at least some) cases of ellipsis. The relevant examples involve what would be viewed as stripping in our modern times (Stemmas 269 and 271): (2)

Tesnière’s Stemma in this case posits an empty node corresponding to a second occurrence of aime ‘love’. Interestingly, however, Tesnière did not posit a null node in related cases involving gapping—see his Stemma 273. The tendency to assume null nodes in cases of ellipsis is also present in Weber (1992). Weber’s approach to coordination, and to syntax more generally, is based closely on Tesnière’s work. Weber does something that Tesnière did not do, though: he posits empty nodes to accommodate cases of gapping (although Weber does not use the term gapping), as the following example from German demonstrates: (p. 144)

(3)

The [-] marks a null instance of the verb liebt ‘loves’. For Weber this null element is necessary to maintain an account that in this case sees coordination as necessarily coordinating clauses.

Page 3 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar Heringer (1996: 44, 147–8, 202–3) also uses null elements to address aspects of ellipsis. His understanding of these null elements is in terms of deletions (Ger. Tilgungen). He relies heavily on deletions to accommodate aspects of coordination. For instance, he assumes deletions in cases of gapping: (4)

And he assumes deletions in cases of VP-ellipsis, e.g.: (5)

Heringer’s discussion of ellipsis is like other DG accounts, though. These accounts merely touch on aspects of ellipsis, without using established terminology. There is no systematic effort to investigate one or more of the various types of ellipsis and to determine their distributions. In his dependency-based syntax of German, Hans-Werner Eroms (2000: 462ff.) addresses ellipsis also mainly just with respect to certain instances of coordination. He assumes the presence of empty nodes in order to accommodate certain readings, e.g.: (6)

According to Eroms, the null node in this case is necessary to accommodate the only available reading, the one involving two groups of elephants. It is, namely, impossible for an elephant to be both big and black and small and white. (p. 145)

Page 4 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar In his prolific works on the dependency-based Meaning–Text Theory (MTT), Igor Mel’čuk has hardly addressed ellipsis.3 He produces a brief mention of the necessity to posit an empty node in the case of gapping: (7)

Mel’čuk comments that the empty node is necessary in such cases in order to account for the fact that in many languages the elided verb imposes lexical and inflectional choices on its dependents. The dashed line marks an anaphoric relation across the elided verb and its antecedent. Hudson (2007: 172–82, 2010) provides a bit more discussion of ellipsis in his framework Word Grammar. He produces empirical observations that support an approach to ellipsis in terms of “unrealized words.” One of the main empirical considerations he produces is the same point that Mel’čuk makes, and it is in fact the point that appears time and again as justification for the existence of empty structure in many cases of ellipsis: in languages that have morphological case, NPs are marked for a certain case despite the absence of an overt case assigner. This point is illustrated with an example of sluicing from German: (8)

Page 5 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

This example involves an interrogative proform in an embedded clause. The analysis follows the approach developed by Groß and Osborne (2009), which assumes rising. The wh-element wem ‘whom’ has risen to become the root of the embedded interrogative clause. The dashed (p. 146) dependency edge and g subscript indicate the presence of rising.4 The notion of rising is not to be understood literally, but rather it is merely a convenient metaphor for denoting certain constellations, in particular those that contain (what would otherwise be) discontinuities.5 The particular analysis of fronting shown in (8) is secondary to the point at hand. This point is that one can account for the dative form of the interrogative pronoun wem by assuming that there is a covert instance of begegnet present as shown. The verb begegnen ‘run into’ is unlike most transitive verbs in German in that it assigns dative case instead of accusative. Without the covert presence of this case assigner, it would be more difficult to account for the appearance of the dative wem. Plewnia (2003, 2013) has worked directly on ellipsis using a DG approach. In his 2003 dissertation he posits “ghost structures” to accommodate ellipsis. These structures appear in the third dimension on a plane behind the surface plane. In his 2013 article, in contrast, he shows the presence of null structure directly in the dependency tree on the surface plane. He explores a type of ellipsis in German that omits one or more verbs from the right edge of a clause. He takes the following example from Kurzeck’s novel Vorabend ‘Evening Before’ (2011: 213): (9)

Page 6 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

The verb chain schreiben kannst ‘can write’ can be interpreted as elided from the right edge of the embedded clause. Plewnia produces a number of similar examples, all taken from Kurzeck’s novel. These examples allow one to interpret the elided material as verbal in nature. Interestingly, Plewnia does not give the ellipsis mechanism involved a name. The term right-edge ellipsis (REE) might be appropriate. (p. 147)

This brief survey of a few previous approaches to ellipsis in the DG tradition has

established two points. The first is that ellipsis has not been systematically investigated in many established DG frameworks. Detailed accounts of various ellipsis mechanisms (e.g., gapping, stripping, VP-ellipsis, sluicing, pseudogapping, answer fragments, left-edge ellipsis, etc.) are rare in the DG tradition, if they exist at all. The second point is that when DGs do address ellipsis phenomena, there is a strong tendency to posit the existence of empty structure, i.e. null nodes, to accommodate central aspects of ellipsis. This empty structure is sometimes shown in dependency trees as empty nodes, and at other times it is given in strings in terms of deletions. This being said, there has in fact been a recent development in the DG tradition that can provide a solid foundation for the exploration and analysis of ellipsis in general. This development is the catena unit. The catena was proposed by O’Grady (1998) as the basis for his analysis of the syntax of idioms, although O’Grady used the term chain instead of catena. The concept has since been seized upon by others and extended to other phenomena of syntax, including ellipsis (see Osborne 2005 and Osborne et al. 2011, 2012). The claim is that ellipsis mechanisms are eliding catenae, whereby these catenae often fail to qualify as constituents. This point is developed in some detail in section 6.5. Before turning to the catena, however, the discussion considers the nature of the elided material, i.e. the structure of ellipsis sites, and the relationship that can or must obtain between an ellipsis site and its antecedent or postcedent, i.e., recoverability. Page 7 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

6.3 Structure The strong tendency for DGs to assume null nodes in cases of ellipsis is easy to understand, since without the covert nodes, ellipsis would often result in dependency hierarchies that are disconnected. The following two competing analyses of an instance of gapping illustrate this point about connectedness: (10)

6

Example (10a) posits an empty node corresponding to the elided verb arrived, whereas example (10b) assumes that no such empty node is present. By positing the presence of the empty node, the analysis in (10a) can maintain the widespread assumption that dependency hierarchies are collections of connected nodes (words), which in most DGs are assumed to be trees. The difficulty with the analysis in (10b) should be apparent in this regard, since without the null node corresponding to the elided verb arrived, the three subtrees in the gapped clause (Susan, today, and with her children) end up disconnected. Many syntactic structures involving ellipsis would become a seemingly arbitrary mish-mash of disconnected parts, rendering a coherent analysis of ellipsis phenomena more difficult.7 (p. 148)

Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar While there is thus strong motivation for DGs to assume that ellipsis involves null structure, the actual status of this empty structure can be debated, as stated in section 6.2. The approach to ellipsis pursued in this contribution assumes that empty structure is present in the form of null nodes, but these null nodes often do not have the same syntactic status as the corresponding overt words. In other words, the empty nodes enjoy a unique status. The motivation for granting them a unique status can be seen in the following example of an answer fragment: (11)

The object form of the pronoun me is acceptable as an answer fragment, whereas the subject form of the pronoun I is not acceptable. When ellipsis has not occurred, in contrast, the acceptability judgments are reversed: I did it vs *Me did it. Such data show that if answer fragments are construed as involving null nodes, these null nodes must have a special status; they cannot have the same syntactic status as the corresponding overt words. A second observation demonstrating that the null nodes of (at least some) ellipsis mechanisms must enjoy a unique status occurs with negation. Many ellipsis mechanisms are incapable of including clausal negation in the elided material, e.g.: (12)

8

The indicated reading is not available each time because the ellipsis mechanism at hand cannot elide the negation not.9 If the null nodes of ellipsis had the same syntactic status as their corresponding overt counterparts, there would be no reason to expect these acceptability judgments. Apparently, no null node of gapping, stripping, VP-ellipsis, or pseudogapping can be the equivalent of clausal negation.10 (p. 149)

The two points just produced demonstrate that the empty structure of at least some ellipsis mechanisms (answer fragments, gapping, stripping) cannot be construed as having the same syntactic status as the corresponding overt words. Given this state of Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar affairs, an approach to ellipsis is warranted that takes the middle ground. This middle ground views ellipsis as involving empty structure in the form of null nodes, but these null nodes enjoy a special status; they are not syntactically equivalent to the corresponding overt words.

6.4 Recoverability The question of recoverability concerns how an ellipsis site gets its meaning. Many occurrences of ellipsis are such that the meaning of the ellipsis site is retrieved from overt material in the linguistic context. The elided material often stands in an identity relationship with an antecedent or postcedent. Recoverability concerns the nature of this relationship; is it syntactic, semantic, or pragmatic? Since the DG tradition has for the most part not devoted much effort to exploring ellipsis phenomena, as discussed in section 6.2, the answer to this question is not established. However, the stance adopted in this contribution points to the fact that ellipsis mechanisms are a heterogeneous bunch; the one ellipsis mechanism can behave very much unlike the next with respect to recoverability. This point will now be demonstrated by considering a few central traits of three ellipsis mechanisms: gapping/stripping, VP-ellipsis, and left-edge ellipsis (LEE). Dependency-based tree structures are employed for illustration. Gapping and stripping are quite unlike other ellipsis mechanisms in a number of ways. Above all, they occur in the non-initial conjuncts of coordinate structures only, whereby there is always an overt antecedent in the initial conjunct that supplies the meaning of the gapped/stripped material. In other words, there is a clearly defined syntactic relationship that obtains between the ellipsis and its antecedent. For one, there must be an antecedent; postcedents are not possible, and the ellipsis usually matches this antecedent quite closely, e.g.: (p. 150)

(13)

The antecedent did…offend matches exactly the elided material. Some minor variation is possible, for instance concerning number and or person morphology, e.g., He likes Susan,

Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar and you like Susan, too (likes vs like). In general, though, the identity relationship across ellipsis and antecedent is clearly defined in syntactic terms. The relationship between ellipsis and antecedent (or postcedent) is less stringent for VPellipsis. The elided VP usually has an antecedent, but postcedents are also possible. Furthermore, ellipsis and antecedent in cases of VP-ellipsis need not appear in the conjuncts of a single coordinate structure, but rather the one can appear in a clause that is superordinate to the clause containing the other. These points are illustrated with the following examples: (14)

In example (14a) and (14c), the ellipsis has an antecedent, whereas in example (14b), it has a postcedent, and in example (14c), the ellipsis appears in the matrix clause at the same time that its antecedent is located in the subordinate clause.

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar Interestingly, however, there is in fact a solid syntactic restriction on recoverability in cases of VP-ellipsis. The material that provides the content of the ellipsis (i.e., the antecedent or (p. 151) postcedent) must precede or govern the ellipsis. In other words, if the ellipsis has a postcedent, this postcedent must govern the ellipsis. If it does not, VPellipsis is impossible, e.g.: (14)

This instance of VP-ellipsis fails because the elided material governs its postcedent, that is, the elided trying hard (indirectly) governs the overt trying hard. Based on this observation, there is in fact a solid syntactic restriction on the recoverability relationship that obtains between the ellipsis and antecedent/postcedent of VP-ellipsis. Be that as it may, the recoverability relationship for VP-ellipsis is indeed much more flexible than for gapping and stripping. As long as the VP-ellipsis has an antecedent, this antecedent can appear at a considerable distance to the ellipsis; ellipsis and antecedent can even appear in separate sentences that are separated by a good amount of other linguistic material, e.g.: (15)

And it is even possible for the ellipsis to lack a linguistic antecedent altogether. The antecedent can be supplied by the situational context, e.g.: (16)

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

And there is yet a further aspect of VP-ellipsis indicating that the recoverability relationship between ellipsis and antecedent/postcedent is flexible. The ellipsis and antecedent can at times be dissimilar in syntactic category. The widely discussed mismatches in voice (active vs passive) illustrate this flexibility: (17)

The antecedent is passive (the dishes…been scrubbed), whereas the elided material is active (scrub the dishes). This sort of flexibility across ellipsis and antecedent is impossible in cases of gapping and stripping, e.g., *The dishes were scrubbed by Tom, and Bill scrubbed the dishes, too. Moving to a third type of ellipsis, i.e., left-edge ellipsis (LEE), it is quite unlike both gapping/stripping and VP-ellipsis in the ways under discussion. LEE occurs in relaxed registers (everyday conversations, emails, text messages, diary entries, etc.).11 It (p. 152)

elides material from the left edge of an utterance. The elided material is of a sort that can be easily inferred from the situational context. First- and second-person subject pronouns as well as auxiliary verbs are thus typical candidates for omission in terms of LEE, e.g.: (18)

It should be apparent that the question of recoverability receives a much different answer when the focus is on LEE. A linguistic antecedent is typically not present at all, but rather the elided material is inferred from a conventionalized context. In a context where something is being explained, the question Understand? is immediately understood despite its fragmentary nature; the elided material Do you is easily inferred. The same is true of diary entries that elide first-person pronouns. Diary entries are typically about the author’s daily experiences, whereby the first person, i.e. the author, is front and center. In such a conventionalized context, it is not necessary to continually repeat the first-person pronoun I or we. What these observations mean for Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar recoverability in cases of LEE is that it is pragmatic in nature. Situational context is responsible for recovering the elided material, not the presence of a linguistic antecedent or postcedent. This section has considered the question of recoverability using DG tree structures for illustration. The main message has been that ellipsis mechanisms are a heterogeneous bunch. The one ellipsis mechanism can be quite unlike the next in terms of the recovery of the elided material. Gapping and stripping are very stringent in this area. They require the presence of a linguistic antecedent that immediately precedes the elided material and matches it rather closely in syntactic form. VP-ellipsis is more flexible than gapping and stripping regarding recoverability, although it too places a concrete syntactic restriction on the relationship between ellipsis and antecedent/postcedent. LEE, in contrast, is altogether different regarding recoverability. The recovery of elided material in cases of LEE occurs through a conventionalized context, meaning that recovery is pragmatic in nature, not syntactic. In sum, then, there is no one answer to the question of recoverability when addressing ellipsis in general. There are, rather, a number of distinct answers that vary according to the ellipsis mechanism at hand. (p. 153)

6.5 Licensing The area in which DG can make a major contribution to the theory of ellipsis in general concerns licensing. Licensing is understood here as the syntactic constraints that restrict the material that the various ellipsis mechanisms can elide. Consider the following instances of ellipsis in this respect: (19)

(20)

Example (19) is an instance of pseudogapping, and example (20) an instance of sluicing. The elided material in each case—want to say that in (19) and played a prank in (20)— does not qualify as a constituent in most theories of syntax. This situation presents a challenge, since it requires that the theory of ellipsis somehow account for the fact that many ellipsis mechanisms appear to be eliding non-constituent units. Many theories of syntax take the constituent to be the basic unit of syntactic analysis, which means that when they are confronted with data like (19)–(20), they choose to augment the theory of syntax in one way or another to account for the fact that many Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar ellipsis mechanisms appear to be eliding non-constituent units. Perhaps the favored approach in this respect is to assume movement first, followed by ellipsis second (e.g. Sag 1976a; Jayaseelan 1990; Johnson 2001b, 2009; Merchant 2001, 2004a, to name just a few). The remnant—to you in (19) and on whom in (20)—first vacates an encompassing constituent so that the encompassing constituent can then be elided. In this manner, a constituent-based theory of ellipsis can be maintained. There is a significant problem with this sort of approach, however. There is little empirical evidence supporting the assumption that movement of the remnant(s) occurs (and the evidence that is produced is contrived). In the cases of (19) and (20), there is little empirical support for the stance that to you in (19) and on whom in (20) are each first moved out of the relevant encompassing constituent so that this constituent can then be elided. In fact, the movement-first-ellipsis-second approach in general is ad hoc. It bears witness to a flawed interpretation of syntactic structure. The basic unit of syntactic analysis is not the constituent, but rather the catena. As stated above in the introduction, the catena unit was introduced by O’Grady (1998), and it has been developed much further in a series of recent articles (Osborne 2005, 2012, 2014; Osborne et al. 2011, 2012; Osborne and Groß 2012; Groß and Osborne 2013). One of the main empirical insights backing the catena unit is the observation that the elided material of ellipsis mechanisms are catenae. The elided strings want to say that in (19) and played a prank in (20) qualify as non-constituent catenae. The following sections first introduce the catena unit and then demonstrate that ellipsis mechanisms of all sorts are eliding catenae. By acknowledging the role of catenae, a major step in the direction of a coherent and comprehensive theory of the licensing of ellipsis is accomplished. (p. 154)

6.5.1 Catenae

The catena unit is defined as follows: Catena A word or a combination of words that is continuous in the dominance (vertical) dimension.12 The catena stands in contrast to the string, which is defined as follows: String A word or a combination of words that is continuous in the precedence (horizontal) dimension. The catena unit is an inclusive unit of syntactic analysis, much more inclusive than the constituent of phrase structure grammars. In other words, many more word combinations in most sentences qualify as catenae than as constituents.

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar The catena unit is illustrated with the following tree structure. The capital letters serve to abbreviate the words: (21)

Each individual word counts as a catena by definition, and each combination of words is a catena if the words are continuous in the vertical dimension: that is, if they are linked together by dependencies. All the distinct catenae in (21) are listed next: A, B, C, D, E, AB, BC, CE, DE, ABC, BCE, CDE, ABCE, BCDE, and ABCDE. There are hence fifteen distinct catenae in (21). Compare this number with the total number of distinct word combinations in the sentence, 32 (=25−1).13 This means that there are 16 (=31−15) distinct word combinations in (21) that fail to qualify as catenae, e.g. AC, AE, BD, ABD, ABDE, etc. As the number of words present in the sentence at hand increases, the discrepancy between the number of distinct catena and non-catena combinations grows. Hence, while the catena is an inclusive unit of syntactic analysis, it is not overly inclusive. In most sentences, the number of non-catena word combinations far outnumbers the number of catena combinations. The claim put forth has been that the catena unit is inclusive enough to acknowledge those word combinations that are pertinent for the exploration of syntactic phenomena and at the same time, exclusive enough to exclude the word combinations that are not relevant for the exploration of syntactic phenomena. The potential of the catena unit for the theory of ellipsis becomes apparent when one begins to examine the word combinations that typical ellipsis mechanisms elide. The claim put forth is that ellipsis elides catenae; ellipsis mechanisms of all sorts are eliding catenae. This claim is expressed concretely here as the Catena Condition: (p. 155)

Catena Condition An instance of ellipsis elides a catena. The following subsection presents evidence supporting this claim. Examples of numerous types of ellipsis are considered.

Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

6.5.2 The evidence This section produces evidence from gapping, VP-ellipsis, pseudogapping, sluicing (including multiple sluicing), and answer fragments supporting the claim that ellipsis mechanisms are eliding catenae. The examples discussed are consistent insofar as when ellipsis is possible, the elided material has the status of a catena, whereas attempts at ellipsis where the elided material is not a catena are usually bad. A type of exception to the Catena Condition is also acknowledged. Sentence (22) is a well-known example from Ross (1970: 250); it is expanded here to illustrate the Catena Condition: (22)

Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

Examples (22a–d) are all at least somewhat acceptable instances of gapping, and in each of them, the elided material qualifies as a catena. Examples (22e–i), in contrast, are all unacceptable on the readings shown. They are unacceptable because the elided material is not a catena. For instance, the elided word combination wants…to begin to (p. 156)

Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar write is not a catena in (22e), nor is wants…a play in (22i). The same is true of (22f–h); in each of these examples, the elided word combination shown fails to form a catena. Note that examples (22e–h) are actually acceptable instances of gapping if an alternative reading is assumed, one in which just the verb wants has been elided. The point here, however, is that the readings shown in (22e–i) are not available. The elided material of VP-ellipsis is, trivially, always a catena, e.g.: (23)

VP-ellipsis usually elides complete non-finite VPs (here admit her mistakes). On a dependency-based analysis of syntactic structure, non-finite VPs are complete subtrees, and complete subtrees are, trivially, catenae. As complete subtrees, non-finite VPs are a particular subtype of catena, namely ones that include all the words that the root nonfinite verb dominates. Example (23b) is bad precisely because the elided words admit her do not qualify as a complete subtree—they do not even qualify as a catena. The next examples involve pseudogapping: (24)

Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

Pseudogapping is like VP-ellipsis in that the gap is introduced by an auxiliary verb (here will), and it is like gapping in that there is (at least) one remnant present (here me). Note that the elided words in (24a) are not a string; want…to stay is not a string. Despite this (p. 157) fact, it is a catena because want immediately dominates to and to immediately dominates stay. Examples (24b) and (24c) are bad in part because the elided material shown—will…you in (24b) and will…you stay in (24c)—does not form catenae, and they are also bad insofar as they do not actually qualify as attempts at pseudogapping, since the gap of pseudogapping must be introduced by an overt auxiliary verb. Sluicing occurs in matrix and embedded interrogative clauses that are introduced by wh-expressions. In most cases, everything is elided from the sluiced clause except the wh-expression. In certain cases, however, more than just a/the wh-expression can appear in the sluiced clause. These cases are of particular interest here since they provide strong evidence supporting the catena, e.g.: (25)

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

These trees show particular aspects of the approach to discontinuities put forth by Groß and Osborne (2009)—see examples (8) and (14b), and nn. 4 and 5. These aspects are of secondary importance to the point at hand, this point being the fact that the elided material of sluicing qualifies as a catena. In the acceptable instances of sluicing (25a), (25c), and (25d), the elided words they are laughing (at) form a catena, whereas in the unacceptable attempts at sluicing in (25b) and (25e), the elided words they…laughing do not qualify as a catena. Observe the alternative word orders across (25a) and (25c–d). Despite the presence of the overt preposition in (25c), the elided material they are laughing still qualifies as a catena. Examples (25b) and (25e) are bad because the presence of are means that the elided words they…laughing fail to qualify as catenae. Instances of so-called multiple sluicing (Merchant 2001) provide further support for catenae. Multiple sluicing occurs when two (or perhaps more) wh-expressions are present in the sluiced clause. In such cases, only one of them can be fronted; the other remains in situ, e.g.: (p. 158)

Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar (26)

Which of these variants is preferred will likely depend on the register in which the sentence is uttered and the speaker’s choice. Regardless of which variant actually occurs, the elided words he lived form a catena in each case. Answer fragments provide particularly strong support for the role of catenae in ellipsis. A given answer fragment is usually a complete subtree (a phrasal constituent in phrase structure grammars), which means the elided material, which corresponds to the rest of the sentence, is necessarily a catena, e.g.: (27)

Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

14

In each of (27a–e), the answer fragment qualifies as a complete subtree, i.e. a constituent, which means that the elided material is necessarily a non-constituent catena. (p. 159)

This is so even when the elided material is not a string, as in example (27d) where the overt Bill’s interrupts the elided material in the linear dimension. Answer fragments in which the elided material does not qualify as catena are bad, e.g.: (27)

Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

Each of these answer fragments is bad because the elided material is not a catena. Observe that the questions, which are attempting to elicit the answer fragments indicated, are themselves awkward. This awkwardness is an additional indication that the syntax of answer fragments is sensitive to catenae. Forming questions that might elicit non-catena fragment answers is itself difficult to do, if not impossible. While the discussion so far has revealed that many ellipsis mechanisms that have been widely acknowledged and explored in the literature on ellipsis are consistent with a catena-based approach to syntax, there are narrow exceptions involving answer fragments that must be conceded. Certain types of questions are capable of focusing an individual word that does not qualify as a complete subtree of dependency structures (because that word dominates other words), and when this occurs, the answer fragment that is produced can fail to qualify as a DG constituent, which means, in turn, that the elided material no longer qualifies as a catena,15 (p. 160)

(28) Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar

e.g.: The or-question can focus specific words (here buy and rent), whereby these words are not complete subtrees. In such cases, the answer fragment that is produced also need not be a complete subtree, which means the elided material is not a catena. A similar type of exception to the Catena Condition involves the interrogative words who and what. When they appear in an echo question, these words can also focus an individual word that does not qualify as a complete subtree of dependency structures—e.g. You saw Bill’s what? – I saw his Unicorn—which means the elided material is, again, not a catena. The data examined have demonstrated that excepting narrow cases involving answer fragments elicited by or-questions and certain echo questions, the Catena Condition is a necessary condition on the elided material of ellipsis mechanisms in English. The data have not, however, demonstrated that the Catena Condition is sufficient for predicting where and when ellipsis can occur. Indeed, no claim has been made to this effect. In other words, while it appears as though the elided material of ellipsis mechanisms should be a catena (in most cases), being a catena is not sufficient to license ellipsis. There are many further aspects of ellipsis beyond the Catena Condition that influence whether a given attempt at ellipsis succeeds or fails. For instance, gapping, stripping, and pseudogapping require a remnant to stand in contrast to the parallel constituent in the antecedent clause, and there is a condition on many ellipsis mechanisms that requires a preposition to be present if its complement is present (gapping, stripping, pseudogapping, sluicing, LEE).

6.6 Conclusion To conclude this contribution, some of the main ideas expressed above are enumerated here in abbreviated form: 1. Many established DG frameworks have not devoted much effort to explore ellipsis. 2. When DGs do address ellipsis, there is a strong tendency to assume the presence of null material/nodes. 3. The null nodes of ellipsis enjoy a special status; they do not have the same syntactic status as the corresponding overt words.

Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar 4. Aspects of recoverability vary significantly from one ellipsis mechanism to the next. Some ellipsis mechanisms place clear syntactic restrictions on recoverability, whereas for others, recoverability is determined pragmatically. 5. The catena unit serves as the basis for a comprehensive theory of licensing in a DG theory of ellipsis. (p. 161) 6. Typical ellipsis mechanisms are eliding catenae, whereby attempts at ellipsis where the elided material does not form a catena are usually bad. 7. The catena condition on the elided material of ellipsis is a necessary condition, but not a sufficient one. Many other aspects of ellipsis play a role in determining which material a given ellipsis mechanism can and cannot elide. With the importance of the catena established, the door is now open to comprehensive DG accounts of the various ellipsis mechanisms. The restrictions placed on elided catenae and on the remnants of ellipsis can be identified and explored in a principled manner, without recourse to unmotivated stipulations such as movement-first-ellipsis-second.

Notes: (1) The phrase structure tree (1b) is flatter than many modern phrase structure grammars would assume; it allows n-ary branching, as opposed to just binary branching. This ‘flat’ analysis is given here in the interest of promoting the comparison. Dependency hierarchies are relatively flat by their very nature. (2) The qualification ‘purely’ points to the fact that every node in a dependency tree can be viewed as both lexical and phrasal. That is, every node marks the presence of a word as well as the phrase that that word heads; the nodes in DG structures are hence playing two roles at the same time. In contrast, these roles are played by distinct sets of nodes in PSGs. (3) This statement is based on personal communication with Mel’čuk. (4) The dashed dependency edge identifies a head (here wem) that is not the governor of (one of) its dependent(s), and the g subscript marks the governor of the risen catena. In the case of (8), begegnet is the governor of the risen wem. (5) Rising is the means by which Groß and Osborne (2009) and Osborne (2014) address discontinuities (long-distance dependencies). ‘Rising’ denotes a constellation in which a given word takes on another word as its head that is not its governor. In DG parlance, it is the means by which projectivity violations can be avoided in surface syntax. There is unfortunately not room here to introduce and illustrate rising more extensively, so the reader is referred to the articles cited. The reason an example involving rising is given here is that sluicing is a common type of ellipsis that involves a risen wh-expression. More instances of sluicing appear in section 5.2. (6) The arrow dependency edges in (10a) identify adjuncts (as opposed to arguments or complements). The practice of using arrows to identify adjuncts has precedent in the DG Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar literature, although the exact nature of the arrows employed and the exact nature of what they signify varies considerably. See Tesnière (1959: ch. 21); Engel (1994: 43ff.); Eroms (2000: 217ff.), and Mel’čuk (2003: 193). (7) Note that this point about connectedness is not valid for PSGs, since they can and often do posit empty structure independently of whether ellipsis is deemed to have occurred. This is particularly true of the X-bar schema. Often a given head word appears without a specifier or complement, which means the specifier and complement positions are unoccupied. And when movement occurs, the moved unit vacates a syntactic position, leaving that position unoccupied in a sense. (8) Sentence (12a) is actually acceptable, but on a different reading from the one indicated: ‘It is not the case that I called you and you called me’. This reading is such that the negation scopes over the entire sentence. The availability of this other reading is consistent with the claim here, namely that ellipsis cannot be eliding the negation. A similar comment applies to example (12b). (9) In contrast to gapping, stripping, VP-ellipsis, and pseudogapping, other ellipsis mechanisms (sluicing, answer fragments, null complement anaphora) can easily elide a negation, e.g. He did not do it, but he won’t say why he did not do it. (10) The qualification “clausal” is necessary. An editor points out that a negation that is confined to a VP can be elided, e.g. I have deliberately not spoken with Joe, and you have deliberately not spoken with Joe, too. (11) Aspects of the LEE mechanism are addressed in at least five places (Napoli 1982; Wilder 1997; Kay 2002; Merchant 2004a; Fitzpatrick 2006). The term itself, i.e. left-edge ellipsis, is borrowed from Wilder (1997: 77). These accounts of LEE vary significantly. Kay (2002), for instance, is interested only in LEE in cases of tag questions (e.g. It is Difficult, isn’t it?) and Fitzpatrick (2006) is concerned only with the ellipsis of an auxiliary verb (e.g. Is Anyone interested in dessert?). (12) The catena unit is defined over DG trees more formally as follows: Catena (formal definition) Given a dependency tree T, a catena is a set S of nodes in T such that there is one and only one member of S that is not immediately dominated by any other member of S. (13) The total number of distinct word combinations is calculated using the formula 2n−1, where n = the number of words/nodes present. (14) Examples (27a) and (27d) actually involve two ellipsis mechanisms, N-ellipsis (also known as NP-ellipsis or NPE) as well as the answer fragment mechanism. N-ellipsis is responsible for eliding friend in (27a) and music in (27d). Despite the presence of Nellipsis in these two examples, the other elided material, i.e. the other material not elided by N-ellipsis, still counts as a catena, as examples (27b) and (27e) demonstrate. Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dependency Grammar (15) Observe that answer fragments can easily fail to qualify as DG constituents but still be acceptable. This occurs when there are two or more remnants present in the fragment, as in, for example, sentence (26a). In such cases, though, the elided material is a catena.

Timothy Osborne

Timothy Osborne is an Assistant Professor of Linguistics at Zhejiang University in Hangzhou, China. His research focus is on the theory of syntax, especially on the notions of constituency and dependency in syntactic analysis. Particular phenomena of syntax that he has explored are diagnostics for constituents, coordination, comparatives, ellipsis, and idiosyncratic meaning.

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar

Oxford Handbooks Online Ellipsis in Categorial Grammar   Pauline Jacobson The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.6

Abstract and Keywords One of the fundamental tenets of (most versions of) Categorial Grammar is that the syntax and semantics work ‘in tandem’: the syntax proves expressions well-formed while the semantics assigns them a meaning. Under this view (termed Direct Compositionality), it is difficult at best to state a rule deleting or silencing material under identity with some other overt linguistic material in the discourse context, which suggests that the common wisdom that there is ‘silent linguistic material’ is incorrect. This chapter explores an alternative way to view VP-ellipsis without silent linguistic material. Using conventions developed in earlier work within variable-free semantics and Direct Compositionality, it is shown that such an approach extends immediately to Antecedent Contained Deletion (which is just a special case of ‘transitive verb phrase ellipsis’) as well as to pseudogapping. The chapter also briefly explores the analysis of fragment answers to questions without invoking silent linguistic material, and shows that some of the apparent challenges to this view are in fact not real challenges. Keywords: Direct Compositionality, Categorial Grammar, variable-free semantics, VP-ellipsis, Antecedent Contained Deletion, pseudogapping, fragments answers, TVP ellipsis

Pauline Jacobson

5.1 Introductory remarks THIS chapter focuses on ellipsis within the general framework of Categorial Grammar (CG), whose leading ideas and their implications for ellipsis are elucidated in section 5.2. We focus primarily on VP-ellipsis (VPE) and fragment answers as two case studies. In a nutshell, the version of CG endorsed here would make it surprising to find deletion or silencing of linguistic material under identity with some other linguistic expression. This raises several questions. First: what is the structure of the ‘ellipsis site’? The answer here is straightforward: there is no structure; see 5.3 for more explicit discussion. This in turn Page 1 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar raises the question of how the relevant meaning is computed—how is it that material is understood when nothing is there? This is addressed in 5.3 (see also 5.4.1 and 5.4.3) (the answer to this question is somewhat different for fragment answers than it is for VPE). Finally, this raises the question of what ‘licenses’ these kinds of ellipsis. To put that differently, how is it that the syntax specifies that the relevant material is a well-formed expression? Section 5.6 details this with respect to VPE (including Antecedent-Contained Deletion and pseudogapping). For the case of fragment answers, the relevant question can be framed in a slightly different way. That they are well-formed is unsurprising; they are expressions of various categories (they simply are not expressions of category S). But a slightly different although related question arises: What allows particular expressions to serve as ‘answers’ to questions? This is discussed in 5.3.2.

5.2 Categorial Grammar While there are several versions of CG, part of the interest of CG for the analysis of ellipsis stems from one of these versions embodying a rather strong hypothesis about the syntax/semantics interface. This is the hypothesis of Direct Compositionality (hereafter, DC) articulated in, among others, Montague (1970). DC rejects the notion that the syntax and (p. 123) semantics work separately and rejects the idea that the syntax computes a representation which is ‘sent’ to the semantics for interpretation. Rather, the syntax is a system of rules/principles which prove expressions well-formed (often proving larger expressions well-formed on the basis of smaller ones), and the semantics works in tandem, supplying a model-theoretic interpretation of each expression that is ‘built’ (i.e., proven well-formed) in the syntax. This has some interesting consequences. First, there is no use of an intermediate level of representation such as Logical Form (LF) mediating between the actual (pronounced) sentence and its interpretation. Second, each local expression that is well-formed according to the syntax has a meaning. And since the semantics works compositionally, its meaning cannot depend on the presence of other material in the larger sentential or discourse context in which it is embedded.1 We will return to the consequences of this for the analysis of ellipsis.

5.2.1 Some basics of CG First, we develop a rudimentary CG syntax and semantics. Thus every linguistic expression (a word, sentence, or anything in-between) can be seen as a triple of sound, category, and meaning. We use the notation [α] to represent the sound of an expression α (although I use orthography for convenience), and [[α]] to represent the meaning of α. Note that by ‘meaning’ I mean some model-theoretic object, not a string of symbols, although I will use such symbols as a way to name the model-theoretic objects. Following standard assumptions in much formal semantics (including non-CG theories), we take the basic building blocks of ‘meanings’ to be a set of individuals (notated e), a set of worlds (w), a set of times (i), and the two-member set {1,0} which is the set of truth values (t). We take each linguistic expression to have as its semantic value something from one of these sets, or some function built from these sets (including functions which have other Page 2 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar functions as their domain and/or range). Let mean the set of functions from a set a to a set b. (Hence, is the set of functions whose domain is the set of individuals and whose range is the set of truth values.) For the most part I will simplify in this chapter and ignore intensions (i.e., the role of worlds and times). Having made that simplification, the value of a VP like chased the rat is a function in . We thus say that the semantic type of a VP is . Under this conception, the grammar is a system of rules each of which takes one or more expression(s) (i.e., triples) as input and yields a triple as output. The interest in a CG syntax lies in its conception of syntactic categories. As is the case for the semantics, we assume a set of basic syntactic categories; for present purposes let these include S, NP, N, and AP. In addition, we recursively define additional categories as follows: If A is a category and B is a category then A/RB is a category and A/LB is a category.2 The intuition behind these (p. 124) complex categories is that an expression of category A/RB says ‘give me an expression of category B to my right and I’ll return an expression of category A’, and similarly an expression of category A/LB takes a B to its left to yield a new string of category A. Thus, the syntactic categories directly encode the distribution of an expression. Note also that ‘VP’ is not in the list of primitives; this is because it can be recast as S/LNP. I will, however, often use the term ‘VP’ for convenience. We introduce additional categories later. The advantage of adopting this notion of syntactic categories is that it allows for the combinatory principles/rules to be stated in very general terms, and it allows for a transparent mapping between the syntax and the semantics. Thus, assume that all expressions of the same category have the same type of meaning. Let any expression of category A have as its meaning some member of the set a, and let any expression of category B have as its meaning some member of b. Then an expression of category A/B (whether it be a ‘left slash’ or a ‘right slash’) has as its meaning a function from b to a, i.e., it is a function in . For example, let all expressions of category NP denote individuals, and all expressions of category S denote truth values. Then S/NP (which is a ‘VP’) is of type —it denotes (relative to a world) some function from individuals to truth values. Given that, we can state the syntax and the semantics (extensionally) with two very general rule schemas: (1)

3

There is also no reason not to expect unary rules—i.e., rules taking a single triple as input to return a triple as output. The rules that have gone under the rubric of ‘type-shifting’ rules in the literature are a special case of these. We will see some such rules later. Page 3 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar To illustrate this mini-grammar, take a simple sentence like Tabby chased the rat. First we need some lexical entries. Assume that proper nouns like Tabby are listed in the lexicon as being of category NP (and each denotes an individual). Ignoring intensions again, take a basic intransitive verb such as howled to be listed in the lexicon as being of category S/LNP with a meaning of type (thus it characterizes a set of individuals), while a transitive verb such as chased is of category (S/LNP)/RNP and of type . Rat is listed as N and presumably the type of expressions of category N is also (i.e., also a function which characterizes a set of individuals), and let the be listed as NP/RN where its meaning is of type and, more specifically, it maps a set to its unique or unique contextually salient member. Incidentally, the directional slashes should not be stipulated item by item as they are entirely predictable; see Jacobson (2014) for conventions allowing lexical items to be listed in an underspecified form with rules supplying the directional features. Given this, the two rule schemas in (1) allow the to combine with rat to give the rat of category NP (p. 125) (and it will denote an individual); chased to combine with the rat to give the expression chased the rat of category S/LNP whose extension is some set, and that in turn to combine with Tabby to the left to give the S Tabby chased the rat (whose extension will be 1 or 0).

5.2.2 Relevance for ellipsis What is the relevance of all this for ellipsis? In a nutshell, if direct compositionality is correct, we would be surprised to find rules which suppress the phonology of some material under identity (either formal or semantic) with other linguistic material. To see why, take the case of VP-ellipsis. (I use the term ‘verb phrase ellipsis’ in a theory-neutral way, without of course endorsing a view in which linguistic material is deleted or silenced. Moreover, ‘verb phrase’ is too narrow, since adjective phrase complements of be can also be missing.) Suppose we try to think of the ‘silencing’ as a unary rule taking some triple as input and returning as output a triple with the same category and meaning, but where the phonology is empty. We can begin to formulate this as in (2): (2)

The two possibilities listed in (2) are intended to instantiate two different theories of ellipsis under identity: (a) is one in which formal identity is required while (b) requires instead semantic identity. And the rule as stated in (2) is too crude in that obviously not any VP can have its phonology suppressed, but that part can be fixed by appropriate use of features. Hence, there is nothing wrong with the idea of a unary rule that changes the phonology, even suppressing it altogether—a rule that deleted or silenced a constant (such as, perhaps, that) is perfectly compatible with the framework here. But the problem Page 4 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar with (2) for the kind of locality imposed by direct compositionality comes from the ‘provided that’ part: being identical (in some sense) to other material in the discourse context is not a local property of the input expression.4 Yet if there is no actual linguistic material (either at some level, or simply unpronounced) in the position of an ‘ellipsis site’, two questions arise. First, what licenses material which many theories would take to be syntactically incomplete? In the case of VPE, for example, auxiliaries usually select for VPs, so what allows for a VP to be absent? In the case of fragment answers, what allows certain subsentential expressions to stand alone and be construed as answers to questions? The answer to this will be addressed as we present an analysis of each of the domains discussed here. The second question is: how is it that extra material which is not actually present is understood in the context of a fuller discourse? In theories that do posit silent linguistic material (hereafter, SLM) the usual assumption is that the relevant material is understood in the way it is because it is in some sense actually present in the linguistic string, albeit in silent form. We return to challenge this received wisdom in section 5.4.3. Nonetheless, for now one might wonder how a theory without SLM can account for the understanding of ‘ellipsis constructions’. (p. 126)

Thus the remainder of this chapter centers primarily on VPE as a case study, and also briefly touches on fragment answers. Our first task (5.3) is to sketch an answer to the second question above: how, in each of these cases, a listener understands extra material which does not correspond to any actual linguistic string. Our next task (5.4) is to briefly answer some of the many arguments that have been given for the SLM point of view. Space precludes a thorough discussion, but we will show that large classes of these arguments are very theory-dependent and/or are in some cases based on questionable assumptions or logic. Section 5.5 develops some more tools of CG which are used in 5.6 to provide a more detailed analysis of VPE (as well as of ACD and pseudogapping) within a CG framework. Obviously the brief sketch of fragment answers combined with a more detailed analysis of VPE (including ACD and pseudogapping) does not exhaust the landscape of phenomena that have gone under the rubric of ellipsis, but hopefully provides a representative sample of how a CG can easily handle these.

5.3 If not silent material, then what? 5.3.1 VPE I will suggest that VPE (and many other ellipsis constructions) is basically no different than the case of ordinary free pronouns. The obvious apparent differences (e.g., the socalled deep vs surface anaphora distinction) stem from the nature of the object that these pick up. To clarify, take a typical VPE case as in (3): (3)

Page 5 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar Without additional supporting discourse context, the interpretation of the second sentence is that Lindsay will also win a gold medal. Traditional SLM accounts posit that this is because the representation of the full second clause is something like Lindsay will win a gold medal too, where the strikethrough material represents material that is either deleted (but is present at the level that semantic interpretation cares about) or is present but with null phonology. But I will argue there is no silent material. Will is allowed (by mechanisms to be (p. 127) spelled out in 5.6.1) to occur with no complement, and— following (roughly) Hardt (1993) and many since—we assume that the understanding is supplied by the discourse context. Put informally (again the formal details are spelled out in 5.6.1), one can think of the actual meaning of (3) to include a free variable P over properties (i.e., functions of type ); so that the second clause in (3) is the proposition that Lindsay will also have the P property, and where the value of P is supplied by context. (Note that more precisely P is of type —i.e. a timedependent set, since [[will]] maps a time-dependent set into another time-dependent set but we continue to ignore worlds and times.) Crucially, I am not claiming that VPE sometimes involves a contextually supplied property and other times involves some kind of deletion/silencing under identity. Rather, the claim is that VPE always involves picking up a contextually salient property. But it just happens that very often the relevant property is made salient by having been named—hence the illusion of an ‘identical antecedent’. Here the first clause contains a VP whose meaning is [[win a gold medal]] and hence this property is obviously very salient. In other words, there never is a real ‘antecedent’; the terms ‘matching’, ‘identity’, etc. are all inappropriate as there is no formal or grammatical relation between the understood complement of the auxiliary and the VPs whose meaning happens to make the relevant (i.e., understood) property salient. Of course this view immediately runs afoul of much common wisdom beginning especially with Hankamer and Sag (1976) which has shown that in general it is extremely difficult (at best) to supply a contextually salient property unless it is explicitly named. Let me put this objection aside for the moment and return to it in section 5.4.1.

5.3.2 Fragment answers Many ellipsis constructions can be handled similarly (an obvious example would be Null Complement Anaphora). But I don’t wish to claim that everything that has gone under the rubric of ellipsis is a matter of an anaphoric process picking up contextually salient material. Fragment answers to questions, for example, have a tighter linguistic connection to the question. For example, it is well known that in languages with rich case marking on NPs, the case of a fragment NP answer must match the case on the wh-word in the question (Morgan 1973, Merchant 2004a, and many others). But this too can be accounted for—as can the understanding of a fragment answer in a discourse—without SLM. Take a question–answer pair like (4): (4)

Page 6 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar

Obviously the answer here is understood as conveying the proposition that Claribel invited Bozo. But this follows under the view of questions and answers put forth in Groenendijk and Stokhof (1984) and also (in slightly different terms) in Ginzburg and Sag (2000) whereby the grammar contains question–answer pairs as an actual linguistic unit (call it a Qu-Ans). There is, after all, no reason to think that the largest unit the grammar has anything to say about must be a sentence, and if there are larger units then surely a question–answer pair is a good candidate for a full-blown linguistic unit. Thus, we can say that a Qu-Ans has a syntax and a semantics. The syntax is such that the pair constitutes (p. 128) a Qu-Ans only if the category on the fragment ‘answer’ matches the category of the wh-word. Moreover (as pointed out in both of the works above), the case-matching facts follow as long as one makes the simple assumption that this matching requirement includes case marking. The semantics relies on the assumption that the meaning of the question in (4) is a function of type : it is the function λx[Claribel invite x], although obviously it also contains some kind of illocutionary force marker distinguishing this from an expression like was invited by Claribel. The semantics of Qu-Ans specifies that a proposition is derived by applying the meaning of the question to the meaning of the answer. This means that Bozo in the answer in (4) is nothing more than a bare NP whose meaning by itself is just the individual b. However, in conjunction with the question, the full QuAns gives the proposition that Claribel invited Bozo. This is an interesting case in that the propositional information is therefore shared between the questioner and the answerer. Jacobson (2016b) details a number of advantages of this over SLM accounts. While space precludes detailed discussion of this we mention one here. Consider the question–answer pair in (5): (5)

The short answer Jill commits the responder to the ‘presupposition’ that Jill is a mathematics professor. This follows immediately under the Qu-Ans analysis: the meaning of the wh-expression contributes to the meaning of the question as a whole. Hence the semantics of the question is the function λxx E[[math prof]] [x came to the party]—i.e., it is a partial function defined only for mathematics professors. Thus it can combine with the referent of Jill only if that person is a mathematics professor. No such commitment is made by a responder using a ‘long reply’ (which is not technically an ‘answer’ in the sense of the Qu-Ans analysis); thus the long reply Jill came to the party does not commit the responder to this view that Jill is a mathematics professor. (In fact, it is most natural in contexts where the responder is not sure about this, and most natural if preceded by Well, and uttered with the ‘fall–rise’ intonation discussed in Ward and Hirschberg 1985.) Page 7 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar Jacobson (2016b) details that the SLM view has no ready account of this difference between the short answer and the long ‘reply’. While space precludes demonstrating this here, we can certainly note that the Qu-Ans analysis accounts for the case connectivity between the wh-word and the answer, accounts for the meaning of the entire Qu-Ans construction without SLM, and accounts for the presupposition facts in (5). As to what allows particular expressions to stand alone as answers (the ‘licensing’ question): under this analysis they must match the category of the wh-word. Hence it would be those expressions whose category can be fronted wh-expressions in questions.

5.4 Arguments for SLM? Readers familiar with the many arguments for SLM will wonder how they can be addressed under the view here. Obviously space precludes discussing more than but a fraction of these, but a few of direct relevance to the points here are worth commenting on (for further discussion, see, among others, Jacobson 2016b). (p. 129)

5.4.1 Deep vs surface anaphora

The first obvious objection to the view in 5.3.2 with respect to VPE stems from the seminal observations of Hankamer and Sag (1976): quite often it is impossible to retrieve the relevant property just from the discourse context, even when the context is loaded to make a property salient. It is this, perhaps more than anything else, which has driven the idea for decades that the grammar requires an overt antecedent for VPE and this then leads to the questions of the nature of the identity that must hold between that antecedent and the ‘silenced’ VP. The VPE case contrasts vividly with the case of ordinary personal pronouns which of course can refer to entities never having been named (see, e.g., Lasnik 1976). But it has also been known for decades that it is not true that VPE must have an overt linguistic antecedent: there are many examples of VPE picking up a property which has not been named. Such examples come in two varieties. In the first, there is no remotely plausible linguistic antecedent. The literature contains many such examples: see, e.g., Dalrymple, Sheiber, and Pereira (1991), Merchant (2004a), and Miller and Pullum (2014) for a variety of attested naturally occurring examples. The second type are those in which there is linguistic material which can plausibly be seen as some sort of ‘antecedent’ but where—if overt material had been in the ellipsis site—it would not match that ‘antecedent’ either in form or in meaning. Examples include (6) (Webber 1978) (where the understanding is something like neither of themi will go the place that hei wants to go), and the well-known voice mismatch examples such as the one in (7) (Hardt 1993): (6)

(7) Page 8 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar

Notice that mismatch in (7) is not just a matter of mere formal mismatch. I assume that a passive VP such as (be) looked into does not have the same meaning as an active VP such as look into the problem, hence the so-called antecedent and the VP that would have been there had there been pronounced material match in neither meaning nor form. And then there are of course a number of cases involving things like ‘vehicle change’ (Fiengo and May 1994), and Negative Polarity Item mismatches (see Sag 1976a and many since) which surely involve at least formal mismatch; whether or not these involve a semantic mismatch depends on particular analyses. At any rate, though, there are plenty of cases running counter to the Hankamer and Sag generalization. And so, as long as there are any cases whatsoever of an elided VP without an ‘antecedent’ to match, then any theory needs an account of why this is not always possible. However one answers this question, it is surely better to have a single account of VPE that can be supplemented with a theory of why sometimes the relevant properties are difficult to access without prior linguistic mention than it is to posit two entirely separate mechanisms (as seems to be forced under the SLM under identity view) and still need to supplement that with an account of why it is that the mechanism at work when there is no actual antecedent applies only in limited circumstances. The hope of the account here, then, is that VPE is always a case of ‘deep anaphora’—it just picks up a contextually salient function of type —but that, unlike individuals, such (p. 130) functions are fragile objects. They are not easy to access without heavy contextual support, and they therefore like to be made salient by having been recently named. Hence the illusion of there being an ‘antecedent’ just comes from the fact that very often the property is made salient by being named—it is the meaning of a nearby VP. Such functions, unlike individuals, also decay in salience quickly: it is not enough to have the relevant property be named by overt linguistic material; it must also have been named recently (see Sag and Hankamer 1984). Under this view, then, the question still arises as to why sometimes we can infer the property from the context, sometimes it helps to have linguistic material that does not overtly name the property but from which we can infer the relevant property (e.g., in (6) the first sentence makes highly salient the function mapping an individual x to the property of going to the place x wants to visit), while at other times the property needs to be explicitly named (as in the classic examples in Hankamer and Sag 1976). And I do not have any definitive answers to give here. But this reframes the question to be one about processing and how entities (including properties) become sufficiently salient in a discourse. In any case, it is worth stressing again that as long as there are any good cases of a ‘missing’ VP without an overt linguistic antecedent—and such cases surely do exist—then any theory needs to explain why this is limited. So no extra burden of explanation is placed on the theory here. There is one final objection one might raise to the claim that VPE is really an instance of a ‘deep anaphoric process’ and requires no overt linguistic antecedent. If the illusion that VPE does require an overt antecedent is just due to the fact that properties (i.e., functions Page 9 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar of type ) are fragile and not easy to access, then why do we see the contrasts between do it and the case of a missing complement of an auxiliary (including the auxiliary do) that played such a prominent role in Hankamer and Sag? That is, why is (8a) perfect while (8b) is at best quite degraded in the same context? (8)

But there is no reason to think that the referent for it in this or other discourses is the same kind of object as the meaning that needs to be picked up in (9a) (for relevant discussion, see Miller and Pullum 2014). Arguably, the vast amount of work assuming that there is a moral to be drawn from the contrast between do (with no complement) and do it is based on an accident of the homophony between the two different do’s. Auxiliary do and main verb do are different and have different meanings, and so the type of objects that their complements would pick out are also different. For example, we know that main verb do only occurs with agentive predicates, and find contrasts like (9): (9)

Presumably, it is an anaphor over events (or something similar), and in particular over events with agents. The function [[know French]] which in (9b) is made salient by the first conjunct can be picked up as the ‘missing’ property complement of auxiliary do, but not as (p. 131) the complement of main verb do (and so it cannot refer to that function). Thus, there is no reason to think that the way in which events can be made salient is the same as the way in which the relevant functions can be.

5.4.2 Connectivity effects Aside from the fact that silencing under identity would be surprising under DC, there is another way in which the DC hypothesis bears on the analysis of ellipsis. Thus, many of the arguments for the presence of SLM are based on the following reasoning: Premise: There is some phenomenon P which can be stated only with respect to large chunks of representation. Fact: Smaller expressions (such as fragment answers) show the effects of phenomenon P. Conclusion: Therefore, the small expression must hiddenly be surrounded by more material. To use fragment answers as illustration, a perfect example of this comes from ‘bound’ pronouns. Thus, it is often assumed that these must be c-commanded by a binder (perhaps at LF) and that this would be the only way for the semantics to put together the relevant reading. Yet we find question–answer pairs like the following:

Page 10 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar (10)

(The indices are included here just to indicate the intended reading, and are not assumed to be part of the actual grammatical apparatus.) If indeed the pronoun his must be ccommanded by an instance of every third grade boy in order to get the right interpretation, then the answer here must hiddenly be a fuller sentence such as Every third grade boy calls his mother where the strikethrough indicates silent material. But analyses of any phenomenon that rely on constraints on chunks of representation (such as a c-command constraint, Principle A, Principle B, etc.) in any case could not be correct under DC, as the syntactic and semantic rules do not have access to such chunks of representation. Trees are just convenient representations for the linguists and not something that the grammar ‘sees’ and it cannot state such constraints. So according to the DC worldview, the premise in arguments of this sort cannot be correct. Of course, saying that constraints (such as a c-command constraint on binding) could not be stated under DC would be of little consolation if there were no alternative ways to account for the phenomena in question. But indeed there are. Thus the kind of variablefree semantics developed in Jacobson (1999) automatically accounts for the question– answer pair in (10) (note that variable-free semantics is motivated by entirely independent facts). A necessarily very brief sketch is as follows. First, the question is a functional question: it asks for the identity of a function f of type such that every third grade boy is an x who is in the call-relation to f(x) (see Groendijk and Stokhof 1984 and Engdal 1986). That this question asks for a function is not novel in the variable-free approach, but what is novel here is that no extra apparatus is needed in order to get that reading. Second, the answer his mother automatically denotes a function of this type: it is the function mapping each (male) individual onto his mother. As such, it is appropriate as an answer to a functional question. Further details for a related case can be found in Jacobson (1994) and for this case in Jacobson (2016b). We give this here as just one example, but the moral is that arguments for SLM based on connectivity effects are generally irrelevant under the DC worldview. (p. 132)

5.4.3 The meaning comes for free—but does it?

While not always made explicit, one observation which seems to implicitly underlie part of the motivation for the SLM view is that the natural understanding of B’s utterance in a discourse like in (11) is automatic if this utterance really contains an instance of the silent material win a medal too: (11)

Page 11 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar

It certainly seems as if Bode will too MEANS the same as Bode will win a gold medal too, and this is automatic if B’s utterance really hiddenly is just that. But does B’s utterance really mean [[Bode will win a gold medal too]]? Actually, we have no way of knowing that—the only empirical fact is that in the context of a discourse like (11) most speakers (A or some other observer C) would assume that this is what B intends to convey. Notice that the proposition conveyed by Bode will too is not stable and depends entirely on the discourse context (which is unlike other literal contributions to the truth conditions of a sentence which are stable). In fact, a bit of reflection reveals that the claim that B’s utterance hiddenly contains silent material provides no more of an explanation for the empirical facts about a listener’s understanding than does the account here. After all, the listener has no direct access to the silent material, and must infer what it is. And this in turn raises interesting questions about how processing really works under the SLM account. Do we wish to say that the listener, having heard the string [win a gold medal], supplies this phonological material and then decodes the string [Bode will win a gold medal] exactly as if this were what s/he had heard? This is highly unlikely for two reasons. First, since the VP [win a gold medal] has already been processed (i.e., assigned a meaning), it would be an odd processor that had to go back and reprocess it. Second, as known since at least as early as Sag and Hankamer (1984), the situation with indexical elements shows that what has to be supplied by the processor is a meaning and not a form. For example, consider the discourse in (12): (12)

The understanding of the ‘missing’ complement in B’s utterance is the property of voting for A’s mother, not B’s mother, and so it can’t possibly be that what is supplied is the string [vote for my mother] which is then interpreted. And it is not just a matter of a simple mechanism switching first- and second-person pronouns. Thus consider (13): (13)

Here the ‘missing’ material cannot be nominate your mother or nominate my mother. It must be a meaning which is the value of the VP in A’s utterance in the relevant discourse context (i.e., property of nominating B’s mother). No simple recipe can be given to predict (p. 133) these without consulting the meaning (for more extensive discussion in Page 12 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar the context of fragment answers, see Jacobson 2016b). So the listener must supply some meaning as the missing complement. Given that, the way in which ‘missing’ VP meanings are understood in a discourse context provides no evidence for the existence of SLM. For assuming that the listener directly ‘picks up’ the relevant property (and that there is no SLM) works every bit as well to account for the empirical facts about understanding in a discourse. Of course, this does not mean that there could not be other reasons to believe in SLM, but the fact that these elliptical constructions appear to mean something more in a discourse context is not one of these reasons.

5.4.4 Preposition stranding Finally, a brief look at the so-called preposition-stranding (P-stranding) generalization (see, e.g., Merchant 2001) is in order as it has played such a large role in the literature. Since this generalization has received considerable discussion elsewhere (including the question of how robust it is) we will not review the evidence for and against P-stranding as usually framed (see Merchant this volume for relevant discussion, as well as the references cited there, and see Jacobson 2016a for a possible way to capture this generalization—should it survive—without SLM). It is, however, worth pointing out a relevant set of facts discussed in Jacobson (2016a) as these have played less of a role in discussions about the P-stranding generalization; these center on apparent mismatches (in questions and answers) in English. Consider first the question–answer pair in (14): (14)

Note that here (and in Jacobson 2016a) we crucially use a meaningful preposition. These appear at first glance problematic for the Qu-Ans analysis shown above, but a solution to this under Qu-Ans is sketched in Jacobson (2016a) and we simply refer the reader to that discussion. On the other hand, it is not clear how SLM can account for the mismatch here. Under formal identity the trace in (14b) is crucially an NP trace while in the answer it is a PP. Nor is there semantic identity between the question portion of (14) (Alice fall [PP t]) and the apparent silent material Alice fell into t, since the semantic contribution of into is part of the meaning of the silenced part of the answer, but is not part of the meaning of the relevant portion of the question. (One might think that a copy theory of movement could help for this case, since under such a view into remains in the trace position in the question. But the copy theory of movement work wreaks havoc with the identity of questions and answers in many other cases, as the interested reader can verify.) Note incidentally that a semantic (rather than syntactic) identity condition could work for those cases like (14) but where the preposition is meaningless. But crucially (14) shows that these mismatches are allowed even with meaningful prepositions. While this particular case appears to actually give evidence for Qu-Ans over SLM, the situation unfortunately Page 13 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar is more complex. For neither theory can in any obvious way handle the reverse case shown in (15): (15)

Interestingly, Merchant (this volume), also references this type of case from Jacobson (2016a). His discussion, however, deals only with cases analogous to (15) but which involve a meaningless preposition (as in be angry at; Merchant acknowledges that the fact that at here is meaningless is crucial for his account). But (as noted above) Jacobson (2016a) points out that meaningful prepositions behave the same way, and that the goodness of the pairs in (15) therefore presents a problem for both SLM and Qu-Ans. In sum, the P-stranding facts are sufficiently complex—not only cross-linguistically but even the simple case of matching/mismatching in P-stranding languages like English— that at this point there is no reason to believe that SLM handles the full range of facts (p. 134)

better than does the analysis here.

5.5 More background: More Categorial Grammar Let us move away from ellipsis per se to introduce some apparatus within Categorial Grammar which allows for a more explicit treatment of VPE, including AntecedentContained Deletion (ACD) and ultimately also pseudogapping.

5.5.1 Wh-extraction constructions We begin with a proposal of Partee and Rooth (1983) (see also Montague 1974) which we embed into a theory with a CG syntax. Let any ordinary NP which denotes an individual map to a generalized quantifier (in particular, it denotes the set of sets containing that individual) with a corresponding syntactic category change. This is a unary rule; it takes a single triple as input and returns a new triple as output: (16)

This can be stated more generally: the directional slashes on the output need not be stipulated in the rule and any expression which is normally an argument to some other expression may ‘lift’ to a function taking the latter as argument. Such conventions are spelled out in Jacobson 2014, but due to space considerations we content ourselves here with (16). This means that for an ordinary ‘NP’ like Tabby which denotes an individual (call that individual t) there is a homophonous expression of category S/R(S/LNP) with Page 14 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar meaning λP[P(t)]—that meaning is a function characterizing the set of all sets having t as a member. Next, Steedman (1987), Dowty (1988), and others proposed that wh-extraction and related constructions could be accounted for by adding to the grammar additional binary combinatory schemas allowing function composition (in both the syntax and the semantics). I illustrate this informally as I will propose a slight revision momentarily. The idea is that, for example, an expression of category A/RB could take an expression of category B/RC to its right to give an A/RC, and the associated semantics is function composition. The case of ‘left slashes’ is parallel: A/LB can combine with a B/LC to its left to give an expression of category A/LC. (p. 135)

With this, consider a relative clause such as the italicized portion in (17):

(17)

Chased is of category (S/LNP)/RNP. Tabby can lift as above to give S/R(S/LNP) and these two functions compose to give the expression Tabby chased of category S/RNP and whose meaning is the function which characterizes the set that Tabby chased. (We ignore the step combining this with that; nothing hinges on that detail.) Allowing the grammar to compose expressions like Tabby chased as well-formed, meaningful expressions has benefits besides just for the treatment of wh-constructions. For example, it provides a ready account of so-called Right-Node Raising sentences such as Tabby chased and Orinda caught that pesky little mouse that had stolen the cheese. Here Tabby chased and Orinda caught are both of category S/RNP and have meanings of type ; they conjoin as they have the same category (and type of meaning), and then the conjoined expression (also of category S/RNP) takes the mouse-NP to its right. Note too that the final meaning of (that) Tabby chased in (17) is the same as is assumed in most other treatments of relative clauses; it is a set of individuals. But the way in which we arrive at this meaning is different. A fairly common view in movement-based theories is that a (here silent) wh moves and leaves a trace, so that the material following the rat is which (that) Tabby chased t. The t translates as a variable but this variable is lambda-abstracted over (see, for example, Heim and Kratzer 1998) so that at the end of the day the material pronounced [Tabby chased] has the same meaning as it does on the CG account above. The difference is that the CG account makes no use of a variable and no trace; the transitive verb chase directly combines with (lifted) Tabby. This will be crucial in the analysis of ACD in 5.6.2. I will, however, make one modification to the above. Rather than introduce a binary function composition rule, we instead posit a unary rule which, when combined with function application (as in the schemas in (1)), is equivalent to function composition. This rule is often referred to in the CG literature as ‘Geach’ (or Division). Hence, assume a rule (call it g) allowing an expression of category A/RB to map to one (A/RC)/R(B/RC); note that this can then combine with an expression of category B/RC to give A/RC. The Page 15 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar situation is parallel for left slashes: A/LB can map to (A/LC)/L(B/LC). The semantics of g is such that given a function f in , then g(f) is a function in such that g(f) = λX[λcC[f(X(c))]]. The reader can verify that g(h)(f) = h o f. The derivation of (that) Tabby chased is exactly as above, except that rather than lifted Tabby directly function-composing with chased, it first maps to an expression of category (S/RNP)/R((S/ LNP)/RNP). To get the intuition of this category: the normal lifted category wants a VP (to its right) to give S. The ‘Geached’ category wants an incomplete VP to its right—i.e, it wants a transitive verb to its right, and it returns an S that is still missing an NP to its right. Thus Tabby can now take chased as argument, yielding S/RNP just as in the function composition derivation sketched above. The result here too is that the meaning of Tabby chased is the function characterizing the set of individuals chased by Tabby.

5.5.2 Variable-free semantics of pronouns As noted above, this is a variable-free treatment of relative clauses. Jacobson (1999) extends this to pronouns, and since we are claiming that VP-ellipsis involves some kind of anaphora it (p. 136) is relevant to develop that treatment of pronouns here. Thus in the variable-free semantics of Jacobson (1999) there is no use of variables as the meaning of pronouns (or any other expression) and no use of indices in the syntax. Rather, an expression such as he left is a function from individuals to propositions (rather than, as is standard, an assignment-dependent proposition) and his mother is a function from individuals to individuals (i.e., it is the-mother-of function). Quite central to the program is the treatment of ‘binding’ in something like Every man loves his mother but we can skip that here as it plays no role in the remarks below. Consider, then, the meaning of a pronoun. In a standard view with indices and variables, it is an individual relative to an assignment function (where every assignment function is a way to assign values to the variables). Here, there are no variables, no indices, and no assignment functions. Thus the hypothesis is that any expression containing a pronoun ‘unbound’ within that expression—including a pronoun itself—denotes a function from individuals to something else. More generally, we introduce an additional set of categories: If A is a category and B is a category then AB is a category. Like A/B, any expression of this category denotes some function in . The ‘slash’ and the superscript thus have a different syntax, but they have the same semantic effect. Let the lexical category of a pronoun be NPNP and let its meaning be the identity function on individuals—thus it is of type . The next question is how a pronoun (or something containing a pronoun) can be the subject of lost in a simple sentence like He lost (where the pronoun remains free) or in a more complex case like Every boyi fears that hei lost (where he lost is embedded and he is ultimately ‘bound’). Again we skip the mechanism for bound readings (it is effected by a unary rule operating on the category and meaning of fears), but the composition of he lost is straightforward. Simply extend the g rule given above so that something of category A/ B can map not only to (A/C)/(B/C) (call that g-slash) but also to AC/BC (call that g-sup) with the same semantics as given above for g-slash. Intuitively, then, this allows something to Page 16 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar combine with a pronoun (or any expression containing a pronoun unbound within it) and it ‘passes up’ the information that there is a pronoun. The consequence of this is that lost of category S/LNP can map to SNP/LNPNP and can combine with he. The result is he lost of category SNP with meaning (ignoring the contribution of gender here) the same as [[lost]] —a function from individuals to the proposition that that individual lost. So, a free pronoun is not represented in this system as a ‘free variable’ and He lost is not a proposition but rather a function from individuals to propositions. The listener, however, derives a proposition from this by applying this to some contextually salient individual. While it might seem odd to think of [[He lost]] as not a proposition but actually a function of type , it is important to realize that under a system with variables and assignment functions its semantic value is not a proposition either. It is an assignmentdependent proposition—technically, a function from assignment functions (ways to assign values to the variable, or indices) to a proposition, and the listener derives a proposition only by picking some contextually salient assignment. Here the listener needs to supply a contextually salient individual, rather than a contextually salient assignment. Before returning to VPE, we make one further generalization. We have seen that g can introduce either a ‘slash’ or a superscript to a would-be argument of a function, and pass that up to the result category. Let us generalize this so that not only can expressions of (p. 137) category A/B input the g rule but so can AB (Jacobson 2000a shows that this immediately accounts for the fact that pronouns have paycheck pronoun meanings). We won’t formalize this here—the idea simply is that an expression of category A/B can map to one of category (A/C)/(B/C) or one of category AC/BC, and an expression of category AB (BC) can map to (AC) or to (A/C)(B/C). All four of these have the same semantics: the semantics of g shown above.

5.6 VP-ellipsis 5.6.1 Simple VP-ellipsis With this set of tools, we now make more formal the notion that VPE is a case of deep anaphora, where the ‘missing’ complement of an auxiliary is understood as a contextually salient property. To illustrate, take the auxiliary will listed in the lexicon with syntactic category VP/RVP—call this the lexical item will1. (Since will1 takes only VPs with bare stems, assume that there is some additional feature on the argument VP. Moreover, auxiliaries form a distinct class in terms of their syntax, so presumably they all have a feature [AUX].) It is of course oversimplified to talk about the meaning of will while ignoring times, since [[will]] takes a time-dependent set and returns a time-dependent set (hence its type is ). But we nonetheless continue to extensionalize for convenience, and so assign [[will]] the simplified type . Then assume a unary rule whereby auxiliaries—i.e., items of the category VP/VP with the feature [AUX]—can map to another item with the same phonology and the same meaning but with category VPVP. In this sense, they are like other ‘proforms’ (although, with the Page 17 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar possible exception of do, they do not denote identity functions). This then predicts the existence of a second item—call it will2 of category VPVP—which has the same meaning (and phonology) as will1. In other words, what licenses auxiliaries to appear without overt complements is simply the fact that there is a unary rule mapping them to VPVP. Consider then a sentence like Bode will. Here we have the item will2, whose category is VPVP. Bode can lift to be S/R(S/LNP), which (using our convenient abbreviation) is the same as S/RVP. Then by g-sup this maps to SVP/VPVP which takes will as argument. At the end of the day, then, Bode will is of category SVP, and the semantics is such that its meaning is λP[〚will〛 (P)(b)] or, slightly more informally, λP[Bode will P]. And this is supplied by the listener to a contextually salient function of type . In the case of someone hearing speaker B’s utterance in (11), that function is the [[win a gold medal]] function which is made salient in virtue of having been named by speaker A.

5.6.2 Antecedent-Contained Deletion What about other cases of VPE such as Antecedent-Contained Deletion (ACD) discussed first in Bouton 1970 and illustrated in (18)? (18)

(p. 138)

It is generally assumed that this is just an instance of VPE—i.e., that whatever

mechanism one adopts for a case like (11) should automatically extend to this case. Indeed, I assume that that is correct. The question, then, is whether the CG mechanisms sketched above accomplish this. Happily, the answer is yes; the existence of cases like (18) follows immediately. Moreover, ACD has been taken since at least as early as Sag (1976a) to necessitate a level of LF at which quantified objects are raised from their pronounced position. But—as shown in Cormack (1984), Evans (1988) (for a different but related case), Jacobson (1992a,b, 2003, 2008 etc.)—the CG mechanisms above mean that no LF (and no rule of Quantifier Raising) is needed. To illustrate, we need first to consider the parallel case without ellipsis: (19)

Standard wisdom is that the relative clause can be put together only if the material following will is a full VP (of the form read t). From this it follows under SLM that the ‘missing’ or silent material in the ACD case (18) must also be read t. We know that VPE at least certainly likes to have overt linguistic material to supply the relevant meaning (or form), but in (18) there is no material that it can supply unless the object NP is pulled out by QR giving a representation read t. (We have, of course, disputed the claim that there must be an overt antecedent. But since it is indeed difficult to get VPE out of context without this, we agree that ACD cases can be used as a diagnostic for the form and/or Page 18 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar meaning of the ‘antecedent’—i.e., of the linguistic material that makes salient the meaning that is supplied.) But notice that under the CG account of relative clauses in general, no full VP is needed in the parallel non-ellipsis case in (19). Rather, will1 is mapped by g-sl to a version which directly takes a transitive verb (rather than a VP) as argument. And so (18) (the ACD case) can also be seen as a matter of a missing transitive verb (phrase)—we call this TVP ellipsis.5 What is supplied is just the two-place relation 〚read〛 which is made salient by being the lexical meaning of the verb in the first conjunct. To detail how this works, the key point is that will1 can map not only to a ‘proform’ over VP meanings but also one over TVP meanings. Thus recall that it maps to VPVP. But because we have defined the g rule in full generality, that item in turn can map to a (VP/ NP)(VP/NP)—intuitively a proform over transitive verbs. I will not spell out the full semantics (the interested reader can do that), but the ultimate meaning is λR[λx[〚will〛 (R(x))]. And the superscript feature (and argument slot that corresponds to this in the semantics) is passed up, such that the ultimate meaning of (18) is (informally) λR[Sarah will read every x such that x is a newspaper and Katie will R(x)]. Thus, this is a function from two-place relations to propositions, and picks up the contextually salient 〚read〛 relation—made highly salient by being the lexical meaning of read in the first clause. Since the case of TVP ellipsis follows here directly from the mechanisms needed for VPE in general, we should assure ourselves that it—like ordinary VPE—can pick up (p. 139)

the missing meaning across sentences and can also occur with no antecedent. Example (20) (from Evans 1988) shows the first point, and (21) shows the second: (20)

(21)

Finally, we should mention that Bouton (1970) discussed another type of case of ACD exemplified in (22): (22)

Page 19 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar In Jacobson (1992a, 2003, and 2008) it is shown that the conventions here combined with the account of pronominal binding developed there immediately extend to this case too. Moreover, cases like (23) which were discussed later under the rubric of ‘rebinding’ (Merchant 2001) are also shown in Jacobson (1992a) to follow from the variable-free account; interested readers can consult Jacobson (1992a) for details: (23)

5.6.3 Pseudogapping Interestingly, pseudogapping also follows immediately from a generalization of the conventions discussed here. Note first that the lift rule in (16) was formulated to only lift NPs to be of category S/(S/NP). But it can (and should) be stated more generally, with many useful consequences. For example, Dowty (1988) shows that generalizing this allows for various coordination cases like Captain Jack served scallops yesterday and steamers today. Part of Dowty’s analysis involves lifting an object NP (e.g., scallops) over transitive verbs—thus this NP maps to VP/L(VP/RNP). (That then combines with yesterday whose category is VP/LVP by the conventions above generalized to the case of left as well as right slashes.) Again the particular directions of slashes on the output category of lift can follow from general conventions; see Jacobson 2014. Now take a pseudogapping case like: (24)

Let scallops lift as above to be VP/L(VP/RNP). But this can then also map by g sup in such a way as to introduce a VP/RNP superscript feature. To make reading this simpler, we abbreviate VP/RNP as TV. Hence lifted scallops maps to VPTV/LTVTV. The intuition here is that this category says ‘give me an anaphor over TVs to my left (such as will) and I’ll yield a VP with an unbound TV meaning within it’. Thus nothing new needs to be said to in order to license the possibility of a ‘missing’ TV as is found in pseudogapping. Ultimately all of this comes from the unary rule mapping will to a VPVP—all of the rest follows simply from the existence of the generalized g rule. We would thus be unhappy if pseudogapping did not exist. Of course, many researchers might take the fact that pseudogapping comes ‘for free’ to be a disadvantage of the approach here, for it is well known that there are heavy constraints on pseudogapping and it does not show all the same behaviors as does VPE (or ACD). But arguably this does not show that the grammar itself should have a separate category of ‘pseudogapping’. Miller (2014) suggests that these constraints all derive from (p. 140)

Page 20 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar independent principles of information structure and/or discourse and are not constraints imposed by the grammar.

5.7 Other Categorial Grammar accounts of ellipsis We have concentrated here on one version of CG and, within this, one account of VPE and fragment answers. There are, however, other accounts of ellipsis within CG and accounts of additional constructions to which we can only give a few pointers here. First, Szabolcsi (1992)—also within a variable-free version of CG—proposes that VPE can be treated as a kind of ‘binding’ of a missing VP. One difficulty with treating all cases of VPE this way is that it cannot handle VPE across sentences and speakers (although one might try to pursue a dynamic approach to binding to extend to such cases), nor can it handle the cases of VPE without overt antecedents. Charlow (2008), however, points out that this does not mean there are no instances of ‘bound’ VPs, and uses the variable-free apparatus of Jacobson (1999) to extend to VPE in such a way that VPE is similar to pronouns: the grammar makes available both ‘bound’ cases and cases where a contextually salient property is supplied (as in the account here). He argues further that focus provides evidence for having both mechanisms. We have said nothing here about gapping; see Steedman (1990) for one approach within CG, and Kubota and Levine (2012, 2016) provide a somewhat different account within a different variety of CG. See also Kubota and Levine (2017) for an analysis of pseudogapping quite different from the one presented here. Of course, this is by no means an exhaustive list of ellipsis work in CG, but hopefully gives the reader pointers to additional work of relevance.

5.8 Conclusion As noted in the introduction, an important aspect of CG for a theory of ellipsis comes from the hypothesis of direct compositionality, which would make it surprising to find silencing of linguistic material under identity with other overt material in the discourse context. But we have seen that no SLM is necessary; VPE and fragment answers both have rather natural analyses without this. Moreover, the traditional arguments for SLM are not strong: some in (p. 141) fact rely on a non-DC worldview so become irrelevant under DC. Finally, we have seen that given some simple CG apparatus (in particular, a generalized g rule) it is easy to account for not only VPE but also ACD and pseudogapping as special cases. The ACD case is particularly interesting, in that under this view there is also no reason to believe that ACD necessitates a level of Logical Form; its existence and interpretation are entirely compatible with Direct Compositionality.

Page 21 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar

Acknowledgement This research was supported in part by NSF Grant BCS0646081.

Notes: (1) This is oversimplified in that there certainly are expressions whose semantic value depends on the context of utterance—indexicals being the obvious case. A compositional semantics can accommodate these by taking the semantic value of every expression to be a function from speech contexts to something else (hence the value of I live in Providence is a function from contexts to propositions). We will ignore the case of indexicals (and related expressions) for now; see n. 4 for discussion about their relevance to the account of ellipsis. (2) Following much of the tradition within CG, I use NP rather than DP for the category of things like the disobedient husky and Samantha. The reader preferring DP can make the substitution as needed. (3) Recall that we are ignoring intensions here; the true meaning (intension) of expressions is actually a function from worlds and times to other things, and so the rules in (1) are actually slightly more complex. For treatments of intensionality, see any introductory formal semantics textbook such as Dowty, Wall, and Peters (1981), Heim and Kratzer (1998), or Jacobson (2014). (4) This is not to say that there is no way such a rule could conceivably be stated. As noted in n. 1, there are indexicals whose value depends on the speech context, and so one might take VPE as a kind of indexical. But we will not pursue that possibility here as this is both complex and does not resemble other phenomena whose value varies with the speech context. The value of speech-context-dependent expressions usually depends on properties like who is the speaker, who is the hearer, what material is salient in the context, etc., and is generally not dependent on the presence or absence of (or successive) utterances. Note, for example, that deaccenting of ‘old information’ (or, lack of accent on this) does not require overt mention of that information; it can be inferred from other facts about the discourse context or even just be salient in virtue of facts about the world. Of course, the claim that VPE is similar (in picking up a salient property) raises the question as to why VPE is so much ‘fussier’ than deaccenting (see, e.g., Rooth 1992a). But there is a reasonable answer: in the case of deaccenting the processor does not need to do the same work since the deaccented material is actually said. This is not the case for ellipsis constructions. (5) We use the term ‘transitive verb phrase ellipsis’ because it can pick up complex transitive verb meanings. For example, in (i) the two-place relation 〚said that Suzy should read〛 can be picked up as the missing complement of did. Note that the CG conventions allow said that Suzy should read in the first clause to combine together to give a meaningful expression of type . Page 22 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Categorial Grammar ((i))

For discussion, see Cormack (1984) and Jacobson (1992a), who also show that the generalization in Sag (1976a) to the effect that these require a de re interpretation for the object also follows.

Pauline Jacobson

Pauline Jacobson is Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. Her research centers on formal semantics and its interaction with syntax, and she is the author of the semantics textbook Compositional Semantics: An Introduction to the Syntax/Semantics Interface (Oxford University Press, 2014). Her research centers on the hypothesis of Direct Compositionality and on the related hypothesis that the semantics makes no use of variables. Her research program applies these hypotheses to a rich set of natural language phenomena.

Page 23 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

Oxford Handbooks Online ellipsis in simpler syntax   Peter W. Culicover and Ray Jackendoff The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.7

Abstract and Keywords Same-Except is a fundamental domain-general cognitive relation in which entities in proximity to one another are judged to be the Same, Except for some part or property where they differ. This relation can be attested in non-linguistic modalities such as vision, audition, and taste, and it plays an important role in non-linguistic categorization. The chapter shows that this relation is expressed linguistically by means of a wide range of devices, including (a) lexical expressions such as same and except, (b) contrastive stress, (c) anaphora (e.g. definite and indefinite NP anaphora and VP anaphora), (d) ellipsis (e.g. bare argument ellipsis, sluicing, gapping, and VP-ellipsis), and (e) fixed expressions such as vice versa. This approach thereby unifies the semantics of all these phenomena under a common account that is based on a domain-general cognitive principle. The approach is compared with accounts of ellipsis based on syntactic copying or deletion, showing that both approaches have their difficulties. Keywords: ellipsis, gapping, sluicing, anaphora, VP anaphora, contrastive stress

Peter W. Culicover and Ray Jackendoff

7.1 Simpler Syntax SIMPLER Syntax (henceforth SS) seeks to specify the role of syntax in linguistic theory in such a way as to achieve maximum overall explanatory power. The Simpler Syntax Hypothesis is formulated as follows in Culicover and Jackendoff (2005): Simpler Syntax Hypothesis (SSH): The most explanatory syntactic theory is one that imputes the minimum syntactic structure necessary to mediate between phonology and meaning.

Page 1 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax SSH calls for the rigorous application of Occam’s razor to syntactic analysis. It hypothesizes that syntax per se is severely restricted in the sorts of phenomena it can account for, especially in comparison to more conventional Chomskyan syntactic theories (Chomsky 1965, 1973, 1977b, 1981, 1986).1 SS adopts the Parallel Architecture perspective of Jackendoff (2002), in which phonological, syntactic, and semantic structures are subject to their own well-formedness conditions, and are related to one another through correspondence (or interface) rules. Individual words are treated as small-scale interface rules that map pieces of phonology to pieces of syntax and pieces of meaning. SS also countenances meaningful constructions (in the sense of Construction Grammar), defined in terms of similar correspondences, but at a phrasal level. Hence phrasal syntactic structures correspond to larger pieces of meaning, but not necessarily in one-to-one Fregean fashion, since aspects of phrasal meaning can be specified by particular constructions. For instance, the meaning of the comparative correlative construction (the more…, the more) is to some extent a property of the entire construction and cannot be localized in any of its words (Culicover 2013b). (p. 163)

SS assumes no abstract syntactic structure (i.e., no functional heads, no massive

binary branching, etc.), no movement, no invisible constituents (except Aʹ trace), and no (or limited) UG island constraints. We do not assume the stipulations of GB theory and Principles and Parameters theory such as the Theta Criterion, the Projection Principle, the Extended Projection Principle, and so on. To the extent that SS is on the right track, the phenomena that fall under these various devices must find explanation outside of syntactic theory proper, either in semantic well-formedness conditions, in the interface between syntax and semantics, or in processing complexity.

7.2 SSH and elliptical constructions To see the consequences of SSH with respect to elliptical constructions, consider bare argument ellipsis (BAE), illustrated in B’s reply to A in example (1). (1)

B’s reply can convey the same meaning as sentence (2), thus going beyond the meanings of yeah and scotch. (2)

Page 2 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax There are two basic approaches to explaining how the meaning in (2) is conveyed. One is that (1B) is a complete sentence, with covert syntactic structure matched to the portion of (1A) that gets the interpretation ‘Harriet’s been drinking x’; and scotch is substituted for x. The other is that the syntax of (1B) is simply Yeah, scotch, and the meaning in (2) is computed by matching scotch with the unexpressed argument of drinking in the interpretation of (1A). Both approaches have a long and distinguished history, and are well-represented in the chapters of this volume. Other things being equal, SS leads us to prefer the second analysis, because it does not posit invisible syntactic structure. If an account in terms of invisible syntactic structure and one in terms of interpretive rules were empirically equivalent, it could be argued that this preference is simply an aesthetic one. However, they are not empirically equivalent: the two accounts diverge in explanatory power with respect to the full range of phenomena. Specifically, the relation between the elliptical utterance and its antecedent depends not on syntactic identity, but rather on the interpretation of the antecedent. We illustrate with four cases. First, there is no syntactic difference among A’s utterances in (1) and (3), but the interpretation of the antecedent is clearly different. (3)

These differences cannot be explained by an approach to ellipsis that depends only on syntactic structure. And if such an account sorts out the interpretations of Yeah, scotch in various contexts by relying on additional rules of pragmatic inference, we would argue that the latter sort of rule, whatever it may be, is sufficient to account for the facts. (p. 164)

Second, as is well-known, the putative hidden syntactic forms for many examples of ellipsis either are ungrammatical (4Bi and 5Bi) or diverge wildly from the form of the antecedent (4Bii and 5Bii). (4)

(5) Page 3 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

Third, the antecedent can extend over more than one sentence, so the ellipsis cannot straightforwardly be derived from a covert syntactic clause. (6)

Fourth, as seen in (7), the antecedent can be a non-linguistic situation in the environment, in which case there is no linguistic context at all to motivate deletion of elliptic syntactic material. (7)

The claim of SSH, then, is that rules of interpretation and inference are required to account for the full range of ellipsis phenomena. By Occam’s razor (and SSH), if the crucial cases require rules of interpretation and inference, and if these rules are sufficient for the uncontroversial cases, then there is no need to posit invisible syntactic structure as well.

Page 4 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax The rest of this chapter sketches a very general approach to elliptical constructions within an interpretive framework, developed in more detail in Culicover and Jackendoff (2012). In anticipation of the analysis, we summarize our stance on the three major issues in ellipsis sketched in Chapter 1: • Structure: Following the SSH, we assume that for most types of ellipsis, the ellipsis site itself has no hidden structure. For instance, the syntactic structure of Yeah, scotch in (1) is essentially just that. • Recoverability: As we have seen in (1)–(5), the interpretation of the elliptical expression is based in part on the interpretation of its antecedent, with syntax often playing a secondary role. What we take to be most novel in our approach is that the interpretation invokes a domain-general cognitive relation which we call SAMEEXCEPT, to be introduced in section 7.3. • Licensing: We take a constructional approach to the licensing of elliptical expressions. Licensing is not a matter of specifying what can be deleted or what structure can be empty. Rather, each elliptical construction (BAE, VPE, gapping, etc.) specifies a particular type of syntactic fragment that can serve as an elliptical expression, along with a specification of how it is to be interpreted in terms of the SAME-EXCEPT relation. Sections 7.3 and 7.4 show how the SAME-EXCEPT relation is expressed in syntactic structure in a variety of elliptical constructions (including Bare Argument Ellipsis), and section 7.5 shows how it also can be expressed by syntactic constructions not normally thought of as elliptical. Section 7.6 discusses some phenomena that have been adduced as evidence for hidden syntactic structure, and section 7.7 summarizes.

7.3 The framework: same-except The SAME-EXCEPT relation is a domain-general cognitive mechanism. It spontaneously evaluates similar objects, extracting the properties that they share (the SAME) and those that they do not (EXCEPT). Figure 7.1 illustrates two basic cases, pointed out by William James (1890), which we call contrast and elaboration. In the left-hand pair, the two wugs are immediately seen as (nearly) the SAME, EXCEPT for their contrasting head decorations, which stand out as a result. In the Figure 7.1 Contrast (left) and elaboration (right) right-hand pair, they are seen as (nearly) the SAME, EXCEPT for the presence of a head decoration in the second one. This same relation can

Page 5 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax be experienced not only with objects, but also with actions, musical motives, and even food.2 Our hypothesis is that this SAME-EXCEPT relation appears in natural language semantics as well as in all these non-linguistic domains, and it can be expressed in language in a variety of ways, one of which is with the words same and except and their synonyms (see section 7.5).3 When two juxtaposed constituents are sufficiently similar, their interpretations are taken to be the same, except for the part that is overtly different. One of the two juxtaposed (p. 166) expressions (usually the second) may be elliptical, in which case its interpretation is the SAME as the other expression—the antecedent—EXCEPT for the part that contrasts with the antecedent. For example, in (1), B’s response is taken to express an assertion that is the same as A’s assertion Harriet’s been drinking, but with the elaboration scotch. That is, scotch plays a role parallel to the head decoration on the rightmost wug in Figure 7.1. An immediate virtue of this analysis is that the semantics associated with ellipsis is not specifically linguistic: it follows from a general property of human cognition. Before turning to ellipsis in more detail, it is useful to examine the marking of contrast by stress. Consider (8). Contrastive accent is marked in caps. (8)

The two expressions Fred likes fish and Sue likes fish are understood as the SAME, EXCEPT for the pair Fred/Sue. In effect, SUE likes fish means exactly the SAME thing as Fred likes fish, EXCEPT that SUE likes fish is about SUE. Note that Fred and Sue are syntactically parallel. Notice also that the contrastive accent marks the elements that are interpreted as participating in the EXCEPT part of the relation. Sentences like (9) violate this condition. (9)

Here the phonologically contrasted expressions do not support a semantic contrast. That is, Sue likes FISH does not mean the same as Fred likes fish EXCEPT for fish, and FRED likes soup does not mean the same as Fred likes fish EXCEPT for Fred. To keep track of the relationships that hold between the two expressions, we employ the notational device of a tableau, as in (10).4 (p. 167)

(10) Page 6 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

More generally, the form of a tableau is shown in (11). (11)

E1 and E2 are the Entities judged to be the same, e.g. the wugs in Figure 7.1; P1 and P2 are the respective Parts of E1 and E2 that contrast, e.g. the head decorations on the lefthand pair of wugs in Figure 7.1. In the alternative configuration, elaboration, P1 is null and P2 adds a new part, as in the right-hand pair of Figure 7.1. In order to establish a SAME-EXCEPT relation, the two entities to be related must be identified. In the case of Figure 7.1, the items in question are adjacent. In the linguistic case, the constituents expressing the two entities in question are typically adjacent or nearby. For instance, in (8), the two clauses are the entities that are SAME. We use the informal term find for the computational operation that identifies the two entities to be related as SAME. A second operation, which we informally call align, involves finding the relevant features of the entities and the correspondences between them—picking out visual parallelisms or, in the linguistic case, identifying parallel thematic roles and the like. Finally, in order to compute EXCEPT, there must be an operation of identify differences that picks out the aligned parts that are different. For instance, in (8), Fred and Sue are aligned, and different from each other. The most typical environment for linguistic juxtaposition is coordination, as in (8). Coordination offers about the simplest parallelism of structure, making it straightforward to find entities to compare. However, many other syntactic environments also count as ‘juxtaposed’, such as a main and a subordinate clause (12a) and relative clauses within the subject and object of the same verb (12b). (12)

Page 7 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

7.4 Licensing bare argument ellipsis Returning to ellipsis: Again, the basic idea behind our account is that ellipsis constructions express a SAME-EXCEPT relation, in which the antecedent of ellipsis expresses E1, and the elliptical expression itself expresses P2—the part of E2 that is different from E1. The rest of E2 is unexpressed, but it can be inferred from its SAMEness to E1. The interpretation of the ellipsis is recovered by matching the structure and interpretation of the elliptical expression with that of the antecedent, along the lines summarized in the preceding section. We illustrate the approach by applying it to Bare Argument Ellipsis (BAE), as in (13), elaborating the account presented in section 7.3. (13)

In order to formalize the BAE construction, we have to decide on its syntactic structure. Culicover and Jackendoff (2005) argue that BAE is not a clause in the usual sense, because it does not embed freely (*Harriet realizes (that) scotch, etc.). For convenience, we assign it to the category Utterance (of which S is a particular type), dominating a phrase of any syntactic category (XP).5 Simpler Syntax posits that the ellipsis site for BAE has no further syntactic structure aside from the bare XP. Under this account of the syntax, the BAE construction can be characterized as (14), an interface rule correlating syntax and semantics. The coindex 1 matches the syntactic Utterance to the E2 cell in the semantic tableau, and the coindex 2 matches the XP constituent to the P2 cell. The underlining in the E1 and P1 cells indicates that these pieces of the meaning are to be found in the antecedent. (14)

Page 8 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

(14) can be thought of as a meaningful construction of English in the sense of Construction Grammar: it licenses a connection between a particular syntactic structure and a particular semantic interpretation, going beyond the meanings of the words in the syntax. (p. 169)

Let us see how this construction derives the interpretation of scotch in (13aB). First, the coindexation in (14) is applied to derive (15). (15)

Next, E1 and P1 must be found and aligned. In this case they are found in the preceding sentence, but in other cases they might instead come from a non-linguistic situation. The result is (16). (16)

Page 9 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax It remains to specify E2. Informally, we need to subtract P1 (VODKA) from E1 and add P2 in its place. We can notate this operation as the equation (17a), and the resulting tableau as (17b). (17)

This derivation illustrates the ‘contrast’ case of SAME-EXCEPT, in that scotch contrasts with vodka. The other pattern, elaboration, appears in (13b): the elliptical expression supplies an argument (or adjunct) that does not correspond to anything in the antecedent. In this case P1 is null; hence in deriving E2, P2 is simply added without any subtract step.6 Every part of these derivations has counterparts in other theories. In a syntactic theory of ellipsis, the syntactic antecedent must also be located and aligned with the (p. 170)

elliptical expression. More importantly, our informal process of subtracting and adding can be instantiated more formally in terms of lambda abstraction of P1 and application of the resulting property to P2. Subtracting P1 is equivalent to lambda abstraction, while adding P2 is equivalent to applying the lambda abstract to P2. The same basic mechanism has also been used in standard accounts of the semantics of focus and ellipsis (e.g., Merchant 2001; Culicover and Jackendoff 2005). It is based in turn on the earlier approaches of Kraak (1967), Chomsky (1971), Jackendoff (1972), and Akmajian (1973) and developed formally by e.g. Dalrymple et al. (1991), Rooth (1992b), and Lappin (2005). In fact, all approaches to focus and ellipsis that we are aware of assume procedures along the lines of find, identify differences, align, subtract, and add; they differ only in terms of the particular level of representation that the procedures apply to: syntax, LF, or semantics. An important semantic constraint on BAE follows from the character of SAME-EXCEPT. Consider the minimal pair in (18). (18)

Page 10 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax The difference in acceptability between B’s two replies follows from the discourse relations, which are crucial to the interpretation. As before, B’s yeah in (18Ba) indicates that the following utterance is understood as assenting to and elaborating A’s statement. On the other hand, B’s no in (18Bb) marks the utterance as a contrast, and SAMEEXCEPT stipulates that the contrast must be found in P1. But no part of A’s statement can serve as P1, since the time period in five minutes does not bear the same semantic relation to E1 as the location in the kitchen does to E2. Hence there is a failure of align, and the intended reading of (18Bb) is ill-formed. Notice, though, that the two adjuncts are syntactically parallel; it is only in the semantics that alignment fails. This result speaks in favor of a semantically based account of ellipsis. Notice also that the same effect occurs with contrast signaled by words like but and contrastive stress, as our account predicts. (19)

The E2 cell of a tableau for BAE specifies that the Utterance has the semantics of a Situation (i.e., a State or Event), even though the syntax specifies only one of its constituents, namely XP. Since illocutionary force can be assigned to expressions of Situations, BAE may act semantically as an interrogative or an imperative, as well as a declarative. (Similar arguments appear in Stainton 2006b.) (20)

7.5 Licensing other cases of ellipsis In this section we show how the SAME-EXCEPT mechanism licenses and renders recoverable a range of familiar types of ellipsis discussed in the literature: sluicing, gapping, VP-ellipsis, and pseudogapping. Sluicing and gapping are like BAE in that they involve the interpretation of one or more fragments, while VP-ellipsis and pseudogapping involve a bit more complication. We also mention VP-anaphora which is strictly speaking not ellipsis, but appears to employ the same interpretive mechanisms.

7.5.1 Sluicing Sluicing involves a bare wh-phrase in a position that licenses an indirect question. Page 11 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax (21)

(21a,b) are instances of elaboration, like (13b); these cases are called ‘sprouting’ in the sluicing literature (Chung et al. 1995). (21c) is an instance of contrast, where who contrasts with some guy in this class; this case is referred to as ‘matching’ in the sluicing literature. Sluicing is a wh-interrogative counterpart to BAE. We assume as before that the ellipsis site per se has no hidden structure. An important clue to the nature of sluicing is the fact that any type of phrase that can occur in sluicing can also occur in BAE (e.g., parallel to (20b), A: Ozzie’s drinking again. B: What?). Hence an account of sluicing should be parallel to that of BAE insofar as possible. There are, however, two differences. First, the single phrase in sluicing must be a wh-phrase. Second, this phrase counts syntactically as a clause rather than as an Utterance: it appears in positions characteristic of clauses, such as extraposed position (22a); and, like a clause, it conditions singular agreement even when the wh-phrase is plural (22b). (22)

The structure of the sluicing construction is given in (23), licensing an interpretation of a sentence consisting only of a wh-phrase. (23)

Page 12 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax The Syntax part of (23) captures the clausal nature of the sluiced phrase by stipulating that it is an S that contains (only) a wh-phrase. This is where sluicing differs from BAE. (p. 172)

Aside from this syntactic specification, the two constructions are identical and should follow from the same principles. Given that there are BAE examples such as (4), (5), and (6) that have no conceivable syntactic source, we therefore conclude that sluicing cannot be accounted for in syntactic terms, without losing the fundamental parallelism between it and BAE. Derivational accounts of sluicing assume either that this syntactic structure is derived by deleting the rest of the S (e.g. Ross 1969b), or that the wh-phrase is the visible part of a complete clausal structure (e.g. Merchant 2001). Both approaches appear to be at least as stipulative as the constructional one proposed here, since both have to say what is deleted/invisible, and what can be left behind. The present account, however, is not focused on describing what can be left out. Rather, it describes in positive terms the surface configuration and what it means. (See section 7.7 for more discussion of these accounts.)

7.5.2 Gapping Culicover and Jackendoff (2005, 2012) analyze gapping as essentially double BAE: SAMEEXCEPT with two contrasting sites, as in (24a).7 As in the case of BAE and sluicing, the ellipsis site has no hidden structure. Gapping works the same way as full sentences with double contrastive stress such as (24b), except that the non-contrastive parts are left out. (24)

In (24a), Fred and Sue constitute one contrast, and fish and soup another, as in the tableau (25), which has two EXCEPT rows. (25)

Page 13 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

The general form of the gapping construction appears in (26). Given that only the contrast case is possible, there is no option for null P1 and Q1, as there was in BAE and sluicing. (p. 173)

(26)

After finding the antecedents for E1, P1, and Q1, this yields (25) for the structure of Sue soup in (24a). As in previous cases, Sue soup is understood as ‘Sue likes soup’ by virtue of ‘solving for E2’ in (25), using the inferential steps subtract and add. (27)

We note that gapping is syntactically more constrained than prosodic contrast and BAE. For example, in a subordinate clause it is possible to have contrast, but not gapping, as shown by (28a,b); and it is not possible to have gapping in the typical BAE position as an isolated Utterance (28c,d).

Page 14 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax (28)

These contrasts can be thought of as stemming from a difference in where find looks for an antecedent. In gapping, it looks within a quite constrained range: preceding conjuncts and a very small number of other environments. This difference could be a stipulation of individual elliptical constructions, or it could follow from independent principles. We have no evidence one way or the other. We note however that any theory of ellipsis has to make the same distinctions.

7.5.3 VP-ellipsis VP-ellipsis (VPE) is the best-studied variety of ellipsis. The literature on VPE offers many options for the syntax, including VP-deletion under syntactic identity (Sag 1980), a pro-VP (Hardt 1993; Lobeck 1995), a fully structured but phonologically empty VP (Wasow 1972, 1979), and no VP (Culicover and Jackendoff 2005). Because VPE allows non-linguistic antecedents (as in our cartoon in (7)), we can immediately rule out deletion under identity. (p. 174) For convenience, we adopt the pro-VP option here, represented in our notation as a phonologically null VP (though we recognize that other things being equal, the SSH would favor a solution with no VP at all, thereby minimizing syntactic structure). A first approximation of the rule is (29). (29)

Page 15 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax That is, the interpretation of the pro-VP is the SAME as that of the antecedent; contrasting EXCEPT constituents may present, but need not be, as in A: Mary called. B: Yes, she did. Example (30) illustrates the construction in a case where the subjects contrast and the auxiliaries contrast in polarity, hence a double contrast. (30)

To flesh out this analysis, of course, many questions need to be resolved about how VPE in various syntactic environments is mapped into tableaux, and about the degree to which align depends on syntactic configuration. The specifics of the analysis ultimately depend on a number of assumptions about the syntax of VPE, control, raising, and expletive subjects. To pursue all these issues in detail would take us far beyond the scope of the present treatment. Some of them are discussed in Culicover and Jackendoff (2005: ch. 8). Here we mention just two such issues. First, a longstanding problem is “vehicle change,” where the interpretation of the missing VP cannot be derived by copying over the syntax of the antecedent. (31) presents six typical cases. (31)

Such examples are problematic for a theory of VPE that requires syntactic identity. Within the present approach, the question is how much syntactic parallelism is required in order to establish semantic alignment; these cases show that alignment cannot always require syntactic identity. We leave this question open. (p. 175)

Page 16 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax Another puzzle for VPE is the “missing antecedents” problem posed by Grinder and Postal (1971). (32)

The antecedent of it cannot be the overt NP a gorilla, because if but Bill does is omitted, (32) is ill-formed. Rather, it clearly refers to the gorilla Bill has, not to the gorilla John doesn’t have. But this antecedent is not expressed in (32). Grinder and Postal (1971) use this observation to argue that the underlying form of Bill does is Bill has a gorilla, which then undergoes deletion of the VP. In the present approach, Bill’s gorilla emerges in the semantics: (33)

When we “solve” for E2, we get the meaning ‘Bill has a gorilla’, in which the gorilla is asserted to exist. And this gorilla, derived through the inferential process of “solving for” the ellipsis, is what it refers to. Setting these syntactic issues aside, the main point we have tried to establish in this section is that the semantics of VPE can be represented in terms of SAME-EXCEPT. We believe that the difficulties the construction poses are primarily with how its syntax maps into the tableau.

7.5.4 VP-anaphora It is instructive to compare our account of VPE with VP-anaphora (do it or do that), as in (34). There is one difference in the semantics: VPE can denote any Situation—an Event or a State—whereas do it has to denote an Action on the part of the subject. (34)

Page 17 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

Note that in this construction, E1 can be an entity in the visual environment rather than in a previous sentence. (The notation {} indicates meaning that is derived from the non-linguistic context.) (p. 176)

(35)

We thus take do it-anaphora to be an expression of SAME. (36)

Informally, (36) says that do it means ‘X performs an action that is the same as E1’, where X is the external argument of ACT (either an overt or a controlled subject), and E1 is either expressed by an antecedent VP or observed in the non-linguistic context. Comparing (36) with the VPE construction (29), we see that they are essentially the same, except for two critical differences. First, (36) is stated in terms of a VP with the phonological form do it and corresponding syntactic structure, while VPE is stated in

Page 18 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax terms of a phonologically and syntactically empty VP. Second, do it-anaphora is restricted to an ACT interpretation, while VPE is not.

7.5.5 Pseudogapping Consider finally pseudogapping, illustrated in (37).8 (37)

In this construction, the main verb is absent, but the auxiliary is overt. In this respect, pseudogapping is similar to VPE (Hoeksema 2006: 336). The difference is that pseudogapping includes an argument of the missing verb. Culicover and Jackendoff (2005: 295) observe, following Levin (1980), Hoeksema (2006), and Merchant (2008a), among others, that pseudogapping is somewhat less restrictive than gapping. For instance, pseudogapping is marginally possible when the antecedent is in a preceding subordinate clause (38a,b), while gapping is not (38c,d). (p. 177)

(38)

Pseudogapping typically has a stranded NP complement, but it also allows a stranded PP (39a) or S (39b) complement. Moreover, the antecedent of the stranded complement need not be VP-final, as seen in (39c). (39)

Examples such as these suggest a formulation of pseudogapping that incorporates properties of VPE and gapping. (40) offers two possible formulations. They differ in whether the remnant XP is embedded under an otherwise empty VP (40a), or whether it

Page 19 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax is simply dominated by S, concatenated after the Aux (40b). (SSH favors (40b), because it has less structure, but we have no empirical evidence to decide between them.) (40)

(39c) will then have an interpretation equivalent to ‘Sandy put a case of beer in the fridge and Terry did the same, except that it was a can of Coke’. Pseudogapping is most felicitous when there is a further contrast in the sentence, either the subject, the auxiliary, or both. This could be stipulated in rule (40) by adding further EXCEPT rows, or it might follow from, say, principles of information structure. We leave the question open. The main point is that the semantics follows the same pattern as all the other elliptical constructions; the differences lie in the specification of the syntax that can be linked with this semantics.9 (p. 178)

7.6 Other same-except constructions, not normally considered elliptical Our account of ellipsis rests on the semantics of the SAME-EXCEPT relation, and on the operations find, align, identify differences, subtract, and add that are invoked in filling out a tableau. As it turns out, these operations are invoked in the interpretation of non-ellipsis constructions as well, which lends them more motivation than just their utility in an account of ellipsis. This section briefly presents three cases where there is no possibility of positing an invisible constituent.

7.6.1 Anaphora Consider the interpretation of a definite pronoun such as he, she, or it. (41)

Page 20 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

The pronoun expresses complete identity10—there is no EXCEPT. The standard coindexation notation in (41) is the usual way of showing this SAME relation. In our tableau notation, the lexical entry of the pronoun can be notated as in (42). (p. 179)

(42)

This stipulates that the reference of the pronoun must be in the SAME relation with its antecedent. To derive the reading of the pronoun in (41), the underlined E1 says that he has to find an antecedent. When the antecedent John is found, its meaning is plugged into the E1 cell to yield (43). (43)

The standard binding principles A, B, C (or their counterparts in other theories) can be interpreted as constraints on find. A definite pronoun usually denotes token-identity. This distinguishes it from the indefinite pronoun one, which normally expresses identity of sense, i.e. ‘same type’, as in (44a). However, one with a definite determiner, e.g. (44b), can signify token-identity to a nonlinguistic antecedent. (44)

Page 21 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax Parallel to he, one has the lexical representation in (45). (45)

The NP headed by one corresponds to a count individual in the semantics that serves as E2 in the tableau. (The feature COUNT is necessary to rule out mass antecedents, as in *John drank wine, and Bill drank one too.) As in the case of definite pronouns, the antecedent E1 is supplied by find. Unlike definite pronouns, one allows EXCEPT-constituents—differences among tokens of the same type. One way to express EXCEPT is as a modifier of one, as in (46). (p. 180)

(46)

In such cases, one sets up a SAME-EXCEPT relation, which initiates find, align, and identify differences. The contrasting modifiers result in EXCEPT rows being added to the tableau, producing a tableau like (47) for (46b). (47)

Notice that the contrasting modifiers require contrastive stress, as discussed in section 7.3. Notice also that align connects the prenominal modifier onion with the postnominal modifier with mushrooms. This shows that align relies on semantic rather than (or in addition to) syntactic parallelism. (See Culicover and Jackendoff 2012 for more details.) We claim, then, that definite and indefinite anaphora also fall within the ambit of SAMEEXCEPT relations. This reflects back on the ellipsis constructions discussed in sections Page 22 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax 7.4 and 7.5: there should be a uniform account of all of them. Not since the earliest days of generative grammar have definite pronouns been generally considered to be the product of syntactic deletion.11 Thus we conclude that ellipsis constructions should not be either.

7.6.2 Vice versa Next consider another non-ellipsis construction, vice versa (Fraser 1970; McCawley 1970; Kay 1989; Yamauchi 2006). (48)

Vice versa requires an antecedent clause. Within the antecedent, there are two constituents whose places are exchanged in the interpretation of vice versa (hence the synonymous the other way round). The general case looks like (49). Syntactically, it differs from everything we have seen so far, in that all the EXCEPT constituents are anaphoric and are filled in from the antecedents. Yet the semantics is a variation on the same theme. (p. 181)

(49)

The semantics of (49) says that the meaning of vice versa is the same as that of its antecedent E1, except that two parts of the meaning of the antecedent are substituted for one another. Consider how it applies to (48). Find locates the antecedent of vice versa: Kim likes Pat. Align and identify differences match KIM with PAT and PAT with KIM. Solving for E2,

Page 23 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax by subtracting KIM and PAT from KIM LIKE PAT and adding them back in according to the alignment, yields the interpretation ‘Pat likes Kim’. (50)

(51)

Note the similarity of (49) to gapping, for instance in (52), which has the same interpretation as (48). (52)

The difference is that in gapping, P2 and Q2 (here the Utterance Pat, Kim) are overt, whereas in vice versa they are identified as copies of the contrasting constituents in the antecedent clause. Again, not since the earliest days of generative grammar has anyone considered vice versa to be transformationally derived from a partial syntactic copy of its antecedent. The fact that it fits naturally into the SAME-EXCEPT analysis is, we think, important evidence for our approach.

7.6.3 The same, except, and their paraphrases Finally, consider the lexical items the same, except, and their variants, as mentioned in section 7.2. (p. 182)

(53)

Page 24 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

A tableau for (53a) is shown in (54). (54)

Informally (see Culicover and Jackendoff 2012 for more detail), same and its variants set up a SAME row in a tableau, in which the subject (in (53a), this wug) occupies the E2 cell and the entity to which it is being compared (that wug) occupies the E1 cell. The complements of except and its variants (this and crest) supply the P2 cell(s); and if present, the complements of instead (here, that and curl) supply the P1 cell(s). There are numerous syntactic variants. The same can act like a symmetrical predicate, with both E1 and E2 in its subject (55a); the same and except can have plural arguments (55b); they can compare three or more entities (55c) (resulting in multiple columns in the tableau); and the same and instead can be anaphoric, using find to locate their antecedent (55d,e). (55)

Again, nobody to our knowledge has proposed a syntactic account of these constructions that accounts for their interpretation. Yet they share their semantics with all varieties of

Page 25 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax ellipsis. We take the possibility of capturing these generalizations as strong evidence for our approach.

7.7 Structure in the ellipsis site? We have argued that the interpretation of ellipsis constructions does not require invisible syntactic structure: interface rules of a type independently needed for the constructions in sections 7.3 and 7.4 will suffice. This means that ellipsis constructions such as BAE and sluicing are syntactically ‘WYSIWYG’, albeit with different syntactic properties and mappings into semantics. Likewise, our analysis of VPE posits an empty VP with no internal structure, which facilitates statement of the interpretation (and we do not rule out an account lacking a VP altogether). This said, we must still address arguments in the literature for invisible structure in ellipsis, based not on interpretation, but on syntactic distribution.12 These arguments date back to Ross (1969b), who argues against a semantically based theory of sluicing; and they have been extended more recently by e.g. Merchant (2001, 2004a). The arguments are all based on the same principle: the syntactic form of an elliptical construction is subject to those constraints that would have been placed on it, if it had been in a full sentence whose syntax was a copy of the antecedent. Moreover, inasmuch as these constraints are sentence-level constraints, they must be imposed before deleting the elided material. The derivation thus involves movement of a copy of the visible material to a peripheral position in the structure, and deletion of the entire remnant under identity with an antecedent. For instance, consider (56), the simplest example of so-called connectivity effects. (56)

The proper form of B’s BAE response is determined by the fact that it has to be bound to John. But reflexive pronouns are normally bound intrasententially, not across discourse. So the only way to accomplish this binding, it is argued, is to provide an underlying form for B’s reply that contains an antecedent for the reflexive, e.g. himselfi John likes ti a lot, where the material struck out is either deleted in syntax or not pronounced in PF. A similar argument involves case marking, where a BAE fragment carries the case demanded by the verb of which it is understood as the complement. (57), one of Ross’s original examples, illustrates this phenomenon in German. (57) Page 26 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

(p. 184)

Merchant (2004a) adduces similar examples in Korean, Hebrew, Greek, Russian,

and Urdu. Again, the argument is that the case marking of the overt material must be governed by an invisible or deleted verb. Similar examples in English involve the prepositions governed by verbs, as in (58) (from Culicover and Jackendoff 2005: 249). (58)

The verb flirt requires the ‘flirtee’ to be marked by the preposition with. This requirement is apparently responsible for the necessity of the preposition in B’s response in (58a). The very close paraphrases (58b,c) push the point home further: they differ only in that proud requires its complement to use the preposition of, while pride idiosyncratically requires in. The replies, using BAE, conform to these syntactic requirements, just as if the whole sentence were there. Page 27 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax If the antecedent includes the relevant preposition, though, the ellipted expression does not need it. (59)

In contrast, German requires the preposition even when it is present in the antecedent. (60)

Merchant (2006b) presents evidence that this difference correlates cross-linguistically with the ability to strand prepositions: languages that can strand prepositions behave like English, and those that cannot strand behave like German. Thus his explanation of (60) is that B’s reply is derived from a full sentence that has undergone topicalization, which in German must pied-pipe the preposition. Culicover and Jackendoff (2005) propose an alternative approach to this problem called Indirect Licensing (IL). The general idea is that an item related to an antecedent can be (p. 185) syntactically licensed, not by the sentence it is itself in, but via its semantic role in relation to the antecedent. IL can be seen in constructions like topicalization and pseudocleft: for example, in (61). (61)

Consider first (61a). Topicalization is standardly analyzed as ‘movement’, where himselfi is licensed before movement takes place. Alternatively, though, the licensing of himselfi can be explained without movement if we can implement the following: (i) Johni binds the Page 28 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax direct object argument, (ii) the direct object argument is a gap, (iii) the topicalized constituent is identified with the gap, (iv) therefore Johni indirectly binds himselfi, (v) therefore the reflexive is indirectly licensed. (This is a variant of what has been called ‘reconstruction’.) Now consider (61b). There is no plausible movement account in this case. But himselfi can be licensed by the same mechanism envisioned for topicalization: (i) Johni binds the direct object argument, (ii) the direct object argument is a gap, (iii) whoi is identified with the gap, (iv) himselfi is identified with whoi across the copula, (v) therefore Johni indirectly binds himselfi. In both cases, the critical step that permits IL is the identification of the reflexive with a particular constituent of the sentence, here the gap in the direct object. From this it is a minor step to extend IL to the interpretation of ellipsis. On this view, IL permits a phrase XP to be syntactically licensed indirectly through a SAME-EXCEPT relation. The conditions on such licensing are (i) there is an antecedent sentence S that can serve as E1; (ii) XP aligns with the interpretation of some part YP of S, so that XP serves as P2 and YP serves as P1; and (iii) the rest of E2 is implicit.13 In such a configuration, IL says that the syntactic properties of XP are those appropriate for its semantic role in E1. To be more specific, we consider three cases of IL, which differ in how XP (= P2) is related to P1. The first case is Contrast, in which P1 is expressed by YP, a constituent of S, as in (62). (62)

Here IL says that XP, in this case scotch, must have the syntactic features appropriate to the position of YP(= P1), here vodka. This case also includes the German examples in (57): dem Lehrer is P2, so by align it receives the semantic role of the individual being followed, wem (= P1). That role is expressed as the object of folgt. Since folgt licenses dative case on its object, Indirect Licensing says that dative case is licensed for dem Lehrer as well. Binding connectivity (56) works in much the same way. Himself in (56Bi) plays the role of P2, and someone, the direct object of like, plays the role of P1. Someone is locally ccommanded by the subject John, so IL says that himself also counts as locally ccommanded by John, and hence can be coreferential with John. Similarly, the pronoun (p. 186) him in (56Bii) counts as locally c-commanded by John, and therefore cannot be coreferential with John. The second case of IL is where P1 is an implicit argument: that is, it is present in the semantics but not in the syntax, as in (63). (63)

Page 29 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

Here John drinks has an implicit object argument INDEF, not present in the syntax, that serves as P1. B’s response scotch (= P2) aligns in the semantics with INDEF. IL therefore says that scotch acquires syntactic properties appropriate to this role—that is, those of an overt direct object of drink. This case also includes the examples in (58). Intransitive flirt in (58a.A) has an implicit ‘flirtee’ (= P1), and this role aligns with OZZIE (= P2). IL therefore says that Ozzie must satisfy the subcategorization requirements of the verb flirt. When flirt has an overt ‘flirtee’, it requires the governed preposition with, and therefore we must have with Ozzie and not simply *Ozzie. The third case of IL is elaboration, where P1 is semantically as well as syntactically null, for instance in (64). (64)

Because every day (=P2) is a time expression, A’s statement (=E1) has no counterpart, syntactically or semantically, that can serve as P1. Hence A’s statement imposes no syntactic licensing conditions on every day—just semantic conditions such as compatibility with tense and aspect. Therefore, we predict that elaboration is possible when XP is interpreted as an adjunct to S: the syntactic properties of adjuncts, unlike those of arguments, are determined entirely by their semantic role. Based on the evidence in Culicover and Jackendoff (2005: chs 7 and 8), we think this is a correct result.

7.8 Conclusion We have argued that there are two types of ellipsis, neither of which involves hidden syntactic structure. One type, exemplified by BAE, interprets a fragment as being the SAME as that of an antecedent, EXCEPT for the part of the interpretation associated with the fragment. A second type, exemplified by VPE, assigns an interpretation to a sentence with a pro-VP in much the same way: the meaning of the sentence is the SAME as that of the antecedent, EXCEPT for the part of the interpretation associated with the overt material. This analysis fits well with the constructional, interpretive framework called for by the Parallel Architecture and Simpler Syntax, in that the syntactic structure of elliptical utterances is as minimal as possible. Each elliptical construction licenses its own mapping between the SAME-EXCEPT semantics and a stipulated syntactic configuration. These mappings not only license the syntactic structure but also enable the hearer to recover the full intended semantic content. We have shown in addition that this choice of framework is not merely aesthetic: it allows us to express empirical generalizations not available in other approaches. The (p. 187)

Page 30 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax interpretation of both types of ellipsis can be seen as elaborations of the interpretation of anaphora, and also as sharing semantics with non-elliptical constructions such as contrastive stress and with lexical items such as the same, except, and instead. Moreover, this semantic structure is not proprietary to language: it is responsible for judgments in every modality of perception. We take these to be important and unexpected generalizations both within the language faculty and in human cognition as a whole. We recognize that at this point many aspects of our analysis are quite informal. In particular, the tableau notation, the inferential processes find, align, identify differences, add, and subtract, and the mechanism of Indirect Licensing need to be fleshed out in considerably greater detail. However, we have observed that on the one hand these correspond to parallel notions in other theories of ellipsis, and on the other hand, they need to be sufficiently general to apply to non-linguistic as well as linguistic situations. We take this as a challenge for the field.

Acknowledgements Most of this chapter is adapted from Culicover and Jackendoff (2005), which analyzes in more detail each of the cases discussed here. The wug image is used with permission of Jean Berko Gleason. We are grateful to Neil Cohn for the wug variants in Figure 7.1 and the cartoon in (7).

Notes: (1) We exclude the Minimalist Program (Chomsky 1995b), which appears in some respects to converge on SS. (2) A reader suggests the following locution as an explication of the same-except relation: “judged to be overall similar in virtue of identity of all attributes other than the points of contrast.” (3) To see how ubiquitous such expressions are, readers are invited to count the number of times in this chapter that we use locutions such as X is the same as Y, except for Z. (4) In an early incarnation of this work, we attempted to implement a formalism along the lines of lambda abstraction, which in some respects does the sort of work we want. However, we eventually set this formalization aside for two reasons. First, it rapidly became unwieldy, to the point of obscuring the overarching generalizations we were trying to express. Second, SAME-EXCEPT is a domain-general relation, and it is hard to see how to apply lambda abstraction to non-linguistic phenomena such as wugs, melodies, and soups. We therefore have settled on the relatively informal tableau notation, which we find more revealing; it provides a conceptual basis from which more precise formalisms can be developed.

Page 31 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax Similarly, we initially tried to lay out our semantic representations in terms of a more standard predicate–argument sort of notation, of the sort in Jackendoff (1990, 2002). Again, we found that insisting on detailed formalization got in the way of the basic insights, and so we have represented the semantics very informally. (5) Not all non-sentential Utterances are elliptical. For instance, words like hello, ouch, and abracadabra can stand on their own and do not embed. Other non-sentential Utterance types have argument structure, such as How about XP?, What about XP, and P with NP! (e.g. Off with his head!). (6) Culicover and Jackendoff (2012) make a further distinction between “elaboration” and “specification” which we will not pursue here. (7) The idea of gapping as multiple contrast goes back at least to Pesetsky (1982). (8) For a comprehensive derivational treatment of pseudogapping, see Lasnik (1999b). (9) There is no question that this description is largely ad hoc, and we have omitted many interesting details. But under any account, it is necessary to stipulate that it is just the verb and perhaps a complement that are pseudogapped, and not a larger structure. For instance, pseudogapping may not apply to a sequence of verbs, leaving just the inflected AUX. ((i))

An account that assumes movement of the stranded XP and deletion of the remaining VP (e.g. Lasnik 1999b) cannot delete too broadly or strand non-complements, or it will end up deriving ungrammatical sentences like (ii). ((ii))

However, for corpus evidence that pseudogapping is freer than is typically claimed, see Miller 2014. A still more complex construction is comparative ellipsis, which has cases resembling many other types of ellipsis, e.g.: ((iii))

Page 32 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

While the semantics in terms of SAME-EXCEPT is fairly clear, the mapping to syntax, and in particular the variety of available ellipses, prove far less so. (10) We are abstracting away here from exactly what counts as ‘identity’, given the complex issues surrounding the distinction between discourse anaphora and binding of pronouns, strict and sloppy identity, etc. In particular, discourse pronouns such as that in (41) convey token identity, while in sloppy identity, a definite pronoun conveys type identity. (11) A notable exception is Elbourne (2001, 2005). (12) Here we are recapitulating points made in Culicover and Jackendoff (2005, 2012). (13) Culicover and Jackendoff (2005) used the terminology ‘antecedent’ for E1, ‘indirectly licensed’ (IL) for E2, ‘target’ for P1, and ‘orphan’ for P2.

Peter W. Culicover

Peter W. Culicover is Distinguished University Professor at the Ohio State University. He was awarded the Humboldt Research Award in 2006. His primary research has been in syntactic theory. He has been concerned with exploring the cognitive and computational factors that underlie the foundations of syntactic theory. Most recently he has been pursuing an evolutionary account of the origin of grammars from a constructional perspective. Ray Jackendoff

Ray Jackendoff is Seth Merrin Professor Emeritus at Tufts University and a Research Associate at the MIT Department of Brain and Cognitive Sciences. He was the 2014 recipient of the David Rumelhart Prize in Cognitive Science. His principal research goal at present is the Parallel Architecture, a theory of linguistic structure that incorporates semantics, syntax, morphology, and phonology, and that integrates language with the rest of the mind.

Page 33 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in simpler syntax

Page 34 of 34

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar

Oxford Handbooks Online Ellipsis in Construction Grammar   Adele E. Goldberg and Florent Perek The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.8

Abstract and Keywords This chapter emphasizes the shared communicative motivation of ellipsis constructions that leads to cross-linguistic similarities and certain predictable functional constraints. More specifically, ellipsis is licensed by a system of motivated constructions, i.e. learned pairings of form and function. Constructions capture a range of restrictions on form and function, including those related to semantics, discourse context, register, genre, and dialect. Generalizations across constructions are captured by a network of constructions with partially overlapping representations. Semantic recoverability is facilitated not by copying and deleting syntactic structure at the ellipsis site, but by a cognitive process that ‘points’ to information that is, typically but not always, available from the memory trace of an antecedent. This mechanism is independently required for fragments, nonelliptical expressions such as ditto and respectively, and the many examples that do not display the ‘connectivity’ effects that are predicted by copy and deletion accounts. Keywords: constructions, communication, pointer, gapping, VP-ellipsis, French

Adele E. Goldberg and Florent Perek

8.1 Introduction ELLIPSIS constructions are formal patterns in which certain syntactic structure that is typically expressed is omitted. Some of the most commonly discussed ellipsis constructions along with an attested example of each are provided in Table 8.1. All examples in quotes within and following Table 8.1 come from the Corpus of Contemporary American English (COCA) (Davis 2008). Table 8.1 Commonly discussed constructions that involve ellipsis Verb phrase ellipsis

‘French kids eat spinach and ours can too.’

Page 1 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar Sluicing

‘He said that I was “different.” He didn’t say how.’

Gapping

A: ‘You made me what I am today.’ B: ‘And you me.’

Stripping

‘George Greenwell was a patriot but not a fool.’

Comparatives

‘His front teeth seemed to protrude more than Henry remembered.’

Source: Corpus of Contemporary American English (COCA), Davis 2008 Every language balances the need to be expressive with the need to be sufficiently easy to produce. These two major functional principles (described by Goldberg 1995: 67 as Maximize Expressive Power and Maximize Economy) give rise to different networks of learned constructions in different languages via general processes of grammaticalization or constructionization (Paul 1889; Bybee et al. 1994; Hopper and Traugott 2003; Fried 2009; Traugott 2014). This chapter emphasizes the shared communicative motivation of ellipsis constructions that leads to cross-linguistic similarities and certain predictable functional constraints (section 8.2), while we also emphasize the fact that ellipsis is licensed by a system of motivated constructions—i.e., learned pairings of form and function. Specific constructions readily capture a range of restrictions on form and function, including those related to semantics, discourse context, register, genre, and dialect. Constructions’ specific licensing properties, as well as generalizations across constructions, are captured within a network of constructions. On this view, our knowledge of language is a learned system of constructions that are strongly motivated by communicative concerns. Constructionist approaches avoid positing ‘underlying’ levels of syntactic representation or invisible/inaudible structure; instead, semantic recoverability is accounted for by an independently needed psychological ‘pointer’ function (section 8.3). An explicit account of the English GAPPING construction and a discussion of several other constructions that have received less attention in the literature are offered in section 8.4. A brief comparison of French with English in section 8.5 makes clear that there exist cross-linguistic differences (p. 189) even in closely related languages. Finally, some standard arguments against the sort of semantic, surface-based proposal suggested here are addressed in section 8.6.

8.2 Motivating ellipsis This section reviews some general commonalities among ellipsis constructions, before delving into more detailed analyses. In a very general way, the existence of elliptical constructions is clearly motivated by our need to express our messages economically (Paul 1889; Grice 1975; Hankamer and Sag 1976). When part of an intended interpretation is recoverable from context, there is no need for it to be overtly specified Page 2 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (Shannon 1993; Piantadosi et al. 2012). Thus ellipsis constructions likely exist in every language. In fact, while speakers often have the option of redundantly expressing material that could be elided, other times, non-elided counterparts sound quite odd, and ellipsis is required. The non-elliptical counterparts of expressions in Table 8.1 are given in Table 8.2, and while the first three sound fairly acceptable with appropriate intonation, the last two are much less felicitous than their elliptical counterparts (indicated by ‘#’), as they sound quite robotic. Many commonly discussed ellipsis constructions involve a semantic relationship that Culicover and Jackendoff (2005) describe as “SAME-EXCEPT.” That is, what is conveyed by ellipsis constructions is generally a semantic proposition that is the same as one that has been uttered or is otherwise recoverable, except that it differs in some key respect. Culicover and Jackendoff note that the SAME-EXCEPT relationship is independently needed to account for lexical phrases like the same/identical/similar/alike…except/aside from. In order to highlight what is distinct while taking for granted what is the same, it is natural, indeed iconic, to assert only what is distinct.1 Table 8.2 Non-elliptical versions of the attested examples of ellipsis in Table 8.1 French kids eat spinach and OUR kids can eat spinach TOO. He said that I was ‘different.’ He didn’t say HOW I was different. A: You made me what I am today. B: And YOU made me what I am today. # George Greenwell was a patriot but George Greenwell was not a fool. # His front teeth seemed to protrude more than Henry remembered his front teeth protrude.2 Ellipsis constructions require that the omitted information be recoverable, either on the basis of an overt clause or phrase (Chomsky 1964, 1965; Katz and Postal 1964; Hankamer and Sag 1976), or from the non-linguistic context (Dalrymple et al. 1991; Culicover and Jackendoff 2012). That is, in order to note differences, as implied by the SAME-EXCEPT function, whatever is the SAME must be recoverable. Whether the recoverability is based on an overt string or whether non-linguistic context can potentially supply the information depends on the particular ellipsis construction involved. Certain expressions would be impossible to interpret without reference to something uttered in the context (see further discussion below; also (p. 190) Chomsky 1964; Hankamer and Sag 1976; Murphy 1985). For example, a pair of noun phrases (e.g., you, me as in the third example of Table 8.1) is hard to interpret unless it is licensed by a gapping construction, which provides the missing semantic relation.

Page 3 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar The SAME-EXCEPT function implicitly assumes a psychological POINTER mechanism to some overtly expressed linguistic material, or to some relation that is recoverable from the non-linguistic context (see also Tanenhaus and Carlson 1990; Culicover and Jackendoff 2005; Martin and McElree 2008; Abeillé, Bîlbîie, and Mouret 2014). This psychological POINTER mechanism is discussed in section 8.3.

8.3 Recoverability: An independently needed pointer mechanism Just as the SAME-EXCEPT semantic function is part of the meaning of many words and phrases that do not involve ellipsis, a psychological function that ‘points’ to previous linguistic material is likewise required by many words and phrases, such as those underlined in (1)–(5): (1)

(2)

(3)

(4)

(5)

(p. 191)

The only way for the underlined terms in the examples above to be interpreted is

for the listener to understand them to refer to (or presuppose) some overtly specified linguistic material. Thus these terms ‘point’ to another word or phrase, much as anaphoric pronouns do (see also Asher and Lascarides 2003). In examples (1)–(5), respectively and vice versa point to an overt linguistic string with a particular word order.3 On the other hand, secondly presupposes only that there was some ‘first’ statement, without referencing (pointing) to the form of the first statement (as does on the other hand itself). It is a benefit of positing a general pointing function that it allows some constructions to point to a quite specific overt linguistic string, while others only require that a semantic entity or proposition be evoked.

Page 4 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar Debates about the nature of ellipsis typically center around whether detailed inaudible syntactic structure of some sort is actually present at the ellipsis site, and this issue is discussed further in section 8.6. For now, note that the lexical items in (1)–(5) do not lend themselves to such an analysis. Surely no inaudible words or phrases are in any sense ‘within’ these terms. Occam’s razor would suggest that whatever means is used to account for the examples in (1)–(5) should be used to account for standard ellipsis constructions. The same reasoning also applies to the many other examples that do not lend themselves to a derivational account (see section 8.6). In fact, psycholinguistic evidence from Martin and McElree (2008) supports the idea that ellipsis is interpreted via a pointer mechanism instead of by appeal to inaudible syntactic structure at the ellipsis site (see also Tanenhaus and Carlson 1990; Culicover and Jackendoff 2005; Martin and McElree 2011). Specifically, in a task that required participants to determine whether a sentence made sense or not, Martin and McElree (2008) found that increasing the length or complexity of the phrase required for interpretation did not result in an increase in the time it took for comprehension (see also Tanenhaus and Carlson 1990; Frazier and Clifton 2000), while increasing the distance between the antecedent and the site of ellipsis did reduce accuracy, an indication that a memory of the original phrase or its semantic interpretation is retrieved, but is not reconstructed syntactically. For example, it took no longer to decide that (6) made sense when compared with (7); nor did it take longer to decide that (8) did not make sense than it did to make the same determination for (9). Importantly, the first clause has to be processed in order to correctly recognize whether the sentence makes sense or not. Sensical: (6)

(p. 192)

(7)

Nonsensical: (8)

Page 5 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (9)

This finding suggests that the elided phrase is not created anew at the ellipsis site, undermining a copy and deletion mechanism. Martin and McElree (2008) instead suggest that ellipsis involves a pointer to structures in memory, the perspective we endorse here. Phillips and Parker (2014) issue a word of caution on Martin and McElree’s interpretation of their results because the task did not require that the full antecedent be semantically retrieved. While this is true, claims that the antecedent’s structure is copied generally assume that the structure is copied automatically as the sentence is processed, not merely that it can potentially be, if relevant. The lack of complexity effect found by Martin and McElree is then only dismissable under a copying mechanism account, if participants were not actually processing the sentences as they normally would. Xiang, Grove, and Merchant (2014) report that a double-object antecedent followed by either VP-ellipsis or a repeated double-object construction primes the production of a double-object expression, when compared to a double-object construction followed by an intransitive clause.4 But the authors acknowledge that the finding is compatible with the idea that ellipsis leads to the activation of the antecedent phrase in memory rather than the construction of syntax at the ellipsis site: “It is important to note that our goal here is to examine whether syntactic structures are accessed or activated at the ellipsis site, not the narrower question of whether the parser incrementally builds such structures at the ellipsis site” (2014: 3, emphasis added). The pointer mechanism predicts reactivation of a previously mentioned phrase if the phrase is required for interpretation; thus the finding is consistent with the present proposal. To summarize, the psychological pointer mechanism can account for the interpretation of elliptical expressions without the need for inaudible syntactic structure at the site of ellipsis. That is, the pointer mechanism that is required for certain cases (e.g., ditto) can be readily extended for cases of ellipsis as well. The pointer mechanism is also consistent with evidence from psycholinguistic processing. In the following section we outline how ellipsis is licensed by constructions by detailing the English gapping construction. (p. 193)

8.4 Licensing: Gapping and other ellipsis constructions Table 8.1 listed several well-known examples of elliptical constructions. As already observed, all such constructions are motivated in a general way by communicative concerns of efficiency. Yet at the same time, each individual elliptical construction has a distinguishable function in discourse and a corresponding different form (see also Miller 2011a, 2014). For example, gapping (10) and pseudogapping (11a,b) in English are Page 6 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar similar in terms of the amount and type of elided material. In both constructions, the overtly expressed remnant contains two constituents that act as arguments or adjuncts to an unexpressed predicate recovered from an antecedent. In gapping, only the arguments or adjuncts are overtly expressed (10), while the remnant of pseudogapping contains a tensed auxiliary as well (11a,b). (10)

(11)

As Hoeksema (2006) has observed, gapped and pseudogapped fragments bear different semantic relations to their antecedents. In a corpus of naturally occurring examples, he finds that the vast majority of instances of gapping (93 percent) occur in coordination with their antecedent (as in (10)). Moreover, the gapping construction requires contrastive stress (Culicover and Jackendoff 2012: 326). On the other hand, most tokens of pseudogapping (87 percent) occur as the second part of a comparative construction (as in (11a,b)), and the overt tensed auxiliary often provides the contrastive information (see also Miller 2014 who estimates that as many as 97 percent of instances of pseudogapping are comparative). To see that gapping and pseudogapping constructions are not generally interchangeable, note that (10) and (11) become unacceptable if pseudogapping is used instead of gapping (12), or vice versa (13): (12)

5

(13)

In order to understand how elliptical constructions can be represented, we provide the example of the English gapping construction (generalized to include “argument cluster (p. 194) conjunction,” described below). A constructionist account allows us to specify that Page 7 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar the construction involves two conjoined semantic propositions. The first proposition is expressed as a regular clause, while the second proposition is expressed formally only by exactly two filler phrases that designate arguments or adjuncts that contrast in meaning with two in the first clause. We can represent this as follows: (14)

The representation in (14) is best unpacked by considering an example, as in (15): (15)

As captured by the representation in (14), gapping is restricted to formal registers, which predicts that it occurs much more often in written than spoken language (Tao and Meyer 2006; Hurford 2012). The content of the example in (15) makes clear that the context is, in fact, a formal or respectful one. While the words you and me are repeated in the gapped phrase in (15), they are recognized to necessarily contrast with the intended referents in the first sentence; this non-identity requirement is captured in (14) as a requirement on the construction’s function. The understood predicate in (15) is made, here a simple verb in the past tense. The complement, what I am today stands in a ‘sloppy identity’ relationship in the first and second clause, since ‘I’ refers to person A in the first clause and person B in the second clause. The ‘≈’ allows for such sloppy identity. The correspondences required by the gapping construction in (14) for the example in (15) are made explicit below:6 P = made X: you (referring to person B) Y: me (referring to person A) Z: what I(person A) am today X’: you (referring to person A) Y’: me (referring to person B) Page 8 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar Z’: what I(person B) am today The predicate P in our representation of the gapping construction in (14) specifies either an active or a passive construal. That is, active and passive predicates serve distinct functions in terms of information structure, since the actor is topical in an active transitive sentence, while the undergoer is topical in a passive sentence. Thus, the representation in (14) predicts that voice mismatches are not possible in the gapping construction and this prediction is borne out. That is, it is impossible to interpret the elided phrase as passive if the first predicate is in active voice (16a), or vice versa (16b). Here and below, following convention, when we wish to make an elided phrase explicit, it is represented by a crossed-out phrase, although, as already argued, we do not intend that the crossed-out phrase literally exists at the ellipsis site (see also section 8.6). (p. 195)

(16)

The representation in (14) is appropriately general in allowing for cases that are not traditionally considered gapping such as that in (17): (17)

Such cases have been referred to as “argument cluster conjunction,” and assumed to be distinct from gapping since if the verb were expressed in the second clause it would not intervene between the two constituents (Beavers and Sag 2004). But what is important to the gapping construction in (14) is the interpretation of the omitted verb, not its position relative to the two expressed arguments. Thus (17) is naturally accounted for by the representation in (14), where P = visited; X: Jan; and X’= YO; Y: on Monday; and Y’ = on TUESDAY. In this way, the emphasis on surface structure in constructionist approaches leads to a more general formulation than is possible from the derivational account (see also Goldberg 2002). We do not label the grammatical category of the [X’, Y’] phrase in (14) because it is not an instance of any familiar category. Instead of coining a new category label, or stretching an otherwise familiar category to include a unique type of constituent, constructionist approaches allow certain pairings of form and function to be restricted to particular constructions (Croft 2001; Culicover and Jackendoff 2005: 237). In the case of (14), two argument or adjunct phrases correspond to a full propositional meaning. While the semantic interpretation involved in the gapping construction involves an identical interpretation of a predicate, which itself must necessarily be previously expressed, other types of ellipsis are more flexible. For example, the requirement that the elided predicate involves identical voice is weakened in the case of VP-ellipsis in limited Page 9 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar contexts that involve contrastive topics (Kertz 2013) and/or cause–effect interpretations (Kehler 2000), as in (18). (18)

In an empirical judgment study, Kertz finds that examples like (18) are of intermediate acceptability when compared with VP-ellipses that involve matching voice, and those that (p. 196) involve mismatches of voice in different discourse contexts. Sentence fragments (Morgan 1973; Stainton 2006b) fall at the other extreme in that they generally require no overt linguistic antecedent at all as long as the intended interpretation is recoverable (Culicover and Jackendoff 2005: 242–8). Constructionist approaches readily allow for subtle differences of this kind on individual constructions. Individual elliptical constructions can be characterized by syntactic, semantic, discourse, and register properties, or they may underspecify aspects of these dimensions. Attention to these properties serves to undermine the idea that the constructions are simply shorter variants of full-fledged sentence patterns, or that they should all be accounted for in the same way. It instead supports recognizing them as constructions in their own right. Constructionist approaches do not stipulate a distinction between a ‘core’ part of grammar and some sort of ‘residue’ or ‘periphery’. Instead, we aim to account for all form and function correspondences, as is needed for any theory to be descriptively adequate. Thus, elliptical expressions that are restricted in terms of genre or which are not fully productive in terms of lexical options are also treated as ellipsis constructions, as Occam’s razor dictates that they should be, in the sense that they are captured by direct pairings of form and meaning in which key aspects of their semantics are unexpressed. For instance, the examples provided in (19)–(23) are all lacking a main verb and yet each qualifies as a full utterance. (These might be characterized as instances of “deep anaphora” according to Hankamer and Sag 1976.) Each is discussed in turn below. (19)

(20)

(21)

(22)

Page 10 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (23)

The discourse functions of these ellipsis constructions go well beyond simple recovery of some previously mentioned content; in fact, they do not require the pointer function to reference any linguistic material at all. The example (19) (Elise, Casey) can be uttered without a linguistic antecedent as a means of introducing Casey to Elise. This use of paired proper names or descriptions, uttered with a pause and typically a gesture from the first person to the second, is conventionally interpreted to mean, ‘Elise, this is Casey.’ This simple construction is represented in (24): (24)

Several other phrases express a strong emotional response of one sort or another. For example, the Down with construction exemplified in (20) and (25) expresses the speaker’s strong disapproval of whatever is named by the noun phrase. (p. 197)

(25)

This conventional construction involves no linguistic antecedent and in fact has no nonelliptical counterpart, although it is elliptical in the sense of not providing an overt main verb. Notice that its interpretation requires a constructional analysis since its meaning is not compositionally derived from the words themselves. In fact, the same phrase, down with , has an almost opposite meaning in colloquial English if it follows a subject and copula (26). Note, too, that adding a negation as in (27) does not solve the nonequivalence with (25). (26)

(27)

Another expression which has become conventional is Obama’s campaign slogan, Yes we can, which is an idiomatic instance of VP-ellipsis, but which requires no antecedent and conveys that ‘we’ can accomplish some contextually evoked agenda. Page 11 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar The elliptical (Well), I never is a conventional phrase that only exists in a formal register in certain (stereotypically female) dialects. It implies that the speaker is appalled at some event that is contextually recoverable but not stated linguistically (e.g. (28a)). It is infelicitous in contexts that do not lend themselves to expressions of outrage (28b): (28)

As if is another conventional phrase that is limited to certain (younger) dialects. It implies sarcasm indicating that the speaker views the elided, hypothetical event to be highly unlikely as is clear in (29): (29)

In order for any of these constructions to be used with their conventional interpretations, speakers must recognize that they require special interpretations that do not follow from any general principles of composition or deletion. Thus each of these constructions represents a conventional pairing of interpretation with surface form: a construction. A few other specialized constructions involving ellipsis that have been discussed elsewhere are provided in Table 8.3. Table 8.3 Less often discussed constructions that involve ellipsis, with examples and references Name of construction

Example

Let alone construction (Fillmore, Kay, and O’Connor 1988)

It’s no way to run a hotel, let alone a democracy.

Mad Magazine construction (Akmajian 1984; Lambrecht 1990)

Him, a presidential candidate?

Coffee construction (Stainton 2006b; Heine 2011)

Coffee? Tea? Biscuit?

If one wishes to account for all of the nuances of speakers’ knowledge of language, it is an inescapable conclusion that multiple constructions are needed. The following section demonstrates the fact that these constructions, while highly motivated (section 8.2), differ in their specifics cross-linguistically. (p. 198)

Page 12 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar

8.5 Cross-linguistic differences: French and English We expect elliptical constructions to exist in every language, since they are motivated by general communicative pressures. The general motivation also ensures that such constructions are typically not difficult to comprehend, even upon initial encounter. At the same time, constructionist approaches predict that constructions vary in their specifics cross-linguistically (Croft 2001; Haspelmath 2008; Evans and Levinson 2009; Boas 2010), and ellipsis constructions are no exception. Thus, speakers need to learn exactly the nuances of how individual ellipsis constructions are conventionalized in each particular language. For example, at first blush, French contains several very similar elliptical constructions as English, e.g., gapping in (30), sluicing in (31), and not-stripping in (32) (examples from the Frantext corpus).7 (30)

(31)

(32)

Yet, there are striking differences between the two languages with respect to the possibility of VP-ellipsis (Busquets and Denis 2001; Dagnac 2010; Authier 2011, 2012; Abeillé et al. 2014). In particular, only a very limited number of verbs can be used in French VP-ellipsis, such as the modal pouvoir ‘can’ (33) and a few other modal-like verbs. Neither avoir ‘have’ (34), nor être ‘be’ in the simple perfect (‘passé composé’) (35), nor the passive auxiliary, nor copular verbs are allowed, while the corresponding English equivalents are fully acceptable. (p. 199)

(33)

Page 13 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar

(34)

(35)

Moreover, VP-ellipsis in French requires that the subject of the remnant be coreferential with the subject of the antecedent when the ellipsis site is within a relative clause (36a,b), a subordinate clause, or in a comparative construction (Dagnac 2010). English has no such constraint as is evident by the fact that either translation in (36b) is acceptable: (36)

Thus the comparable elliptical construction is much more restricted in French. French has a different means of expressing the function that VP-ellipsis commonly serves in English. In coordinating contexts, French has a specific type of stripping construction, in which a particle appears in the fragment. The particle varies depending on the polarity of both the antecedent and the fragment: aussi ‘too’ (positive/positive: (37)), pas ‘not’ or non ‘no’ (positive/negative: (38)), oui or si ‘yes’ (negative/positive: (39)), and non plus ‘neither/not either’ (negative/negative: (40)). (37)

(38) Page 14 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar

(39)

(40)

The direct translations in English are (marginally) acceptable in the case of a positive antecedent (41a,b), and then only in main clauses, contrary to the French (p. 200)

counterpart (42a,b). The translations involving a negative antecedent sound quite odd in English (43a,b). (41)

(42)

(43)

The French stripping constructions are available for all verbs and in all tenses, but unlike VP-ellipsis, due to the lack of a finite verb, they are unable to convey a distinction in tense or modality from the antecedent.

Page 15 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar In sum, even languages that have been in close contact for hundreds of years differ in the specifics of their ellipsis constructions. Constructionist approaches anticipate such differences, and can readily capture them within each language’s system of constructions.

8.6 Why positing ‘underlying’ structure is problematic As was mentioned earlier, the constructionist approach to ellipsis proposes that while the requisite semantic structure must be recoverable (sections 8.2 and 8.3), there is no ‘underlying’ syntactic structure at the ellipsis site, either unpronounced or deleted. In this section we review the facts that have led many researchers to assume that such underlying syntactic structure is needed, and counter that positing such structure raises more questions than it resolves. Ross (1967), and many others since, have observed certain intriguing “connectivity effects” between an expressed remnant and an antecedent word or phrase (see Merchant 2010 (p. 201) for a review). This is what has led many researchers to assume the existence of an identical, albeit unpronounced word or phrase at the site of the ellipsis. For example, case marking is sometimes determined by the antecedent clause, as illustrated in examples (44) and (45). (44)

(45)

Accounts that assume there is unpronounced syntactic structure predict such connectivity effects (e.g., Ross 1967; Merchant 2004a). In fact, such accounts predict that overt form must be identical with what it would be if there were no ellipsis, because the assumption is that there exists an underlying level of representation in which there is no ellipsis. Yet there exist many examples in which connectivity effects do not hold. For example, in (46b) the accusative me is preferable as a response to the question in (46a), and yet it is clearly unacceptable in the non-elliptical (46b’): Page 16 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (46)

This type of example would seem to provide evidence against deletion in favor of direct interpretation based on surface form and semantic recoverability (Barton 1990). In an effort to defend the deletion account, Merchant (2004a) suggests that the example in (46b) should actually be based on a left-dislocation construction as in (47), where the entire clause following the pronoun me is deleted. (47)

But an analysis in terms of left-dislocation does not work for all relevant cases. For example, most English speakers prefer accusative pronouns in comparative ellipsis ((48a) vs (48b)), even though nominative case is required in the counterpart involving VP-ellipsis (49b). Unlike the example in (47), (48a) cannot readily be analyzed as involving leftdislocation, because the dislocated version is unacceptable (50): (48)

(49)

(p. 202)

(50)

Likewise, the attested gapping example in (51a) involves accusative me, whereas the nonelliptical version involves nominative case (51b). (51)

What happens if we assume that the second clause involves left-dislocation, as in (52)? Page 17 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (52)

Notice that the elliptical example (51a) is most naturally interpreted conjunctively: ‘you don’t have to trust me and I don’t have to trust you.’ That is, the negative (don’t) is naturally interpreted as having wide scope over the disjunction (or) (via De Morgan’s Law). On the other hand, the non-elliptical versions (51b) and (52) do not allow the widescope reading; they are instead interpreted to mean: ‘either you don’t have to trust me or I don’t have to trust you’ (Oehrle 1987; Johnson 2009). Instead, the wide-scope interpretation that is natural in (51a) requires an explicit indication of the intended wide scope as in (53), if there is no ellipsis: (53)

Fiengo and May (1994) recognize certain such cases where connectivity effects are lacking and attribute them to a process of “vehicle change” in which the proposed deleted structure is not identical with an overtly expressed form. This idea is taken a step further in recent work by Barros et al. (2014: 35) who allow the elided material to be wholly distinct from any overt linguistic antecedent. That is, the unpronounced structure in (54) is assumed instead of that in (55): (54)

(55)

Barros et al. (2014) make this move in an effort to avoid the apparent violations of island constraints that expressions like that in (55) would entail if the remnant were assumed to “move out” of the unpronounced structure. But of course the non-identity of deleted structure with an overt antecedent undermines the original argument in favor of deletion, which was based on the claim that the omitted material was necessarily identical with an overtly expressed form. As Merchant (2009: 8) puts it, “structural approaches [to ellipsis] are based on connectivity effects” (emphasis added). Culicover and Jackendoff (2005) provide still other examples in which some linguistic material appears to serve as an antecedent for the elliptical meaning, but the antecedent spans more than one sentence (56), or is discontinuous (57).

Page 18 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar (p. 203)

(56)

(57)

The fact that connectivity effects sometimes exist suggests that they serve to facilitate the correct identification of the relationship between overt and omitted material. That is, case marking often indicates the semantic role of an argument in an event, so the connectivity effects that exist are likely motivated by their tendency to facilitate comprehension. Alternatively, since certain constructions point to or evoke a specific verbal predicate, the overtly expressed arguments associated with that predicate may simply be primed to bear the case marking they normally bear. To summarize, connectivity effects in ellipsis constructions cut both ways. Sometimes the elided phrase aligns with the non-elided paraphrase (44, 45), but other times it does not (46b, 48a, 51a, 54, 56, 57). The examples in which connectivity effects are not in evidence argue against unpronounced syntactic structure in favor of an account that assigns interpretation directly to surface form. That is, one cannot both endorse “vehicle change” and assume “connectivity facts” hold, while maintaining a single rule for ellipsis.

8.7 Conclusion Because ellipsis constructions are motivated by general communicative goals to express our messages efficiently, they are expected to recur across languages. Communicative demands insure that elliptical utterances which require an interpretation that is not recoverable in context will not be felicitous. In addition, constraints associated with individual constructions bear directly on acceptability judgments. This short overview of ellipsis in terms of constructions has emphasized their motivated form, and a range of formal and functional properties. The proposal outlined here has much in common with other surface-based approaches to ellipsis, as these are also essentially constructionist in nature (Murphy 1985; Fillmore, Kay, and O’Connor 1988; Ginzburg and Sag 2000; Culicover and Jackendoff 2005; Osborne 2006a; Miller 2011a, 2014; Sag and Nykiel 2011; Osborne and GroB 2012). A psychological POINTER mechanism is endorsed as a means by which particular overtly expressed linguistic material is evoked by ellipsis (see also Tanenhaus and Carlson 1990; Culicover and Jackendoff 2005; Martin and McElree 2008). This semantic pointer mechanism is independently required for non-elliptical expressions such as ditto and Page 19 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar respectively, elliptical idiomatic expressions without linguistic antecedents (well, I never!), fragments (Ok, Tomorrow.), and examples that cannot be reconstructed by copying syntax from an antecedent (you don’t have to trust me, nor me (*don’t have to trust) you. Moreover, the non-derivational, semantic proposal is consistent with current psycholinguistic evidence, while syntactic copy and deletion accounts are not. Many elliptical expressions have quite distinctive interpretations and restrictions on use. Some ellipsis constructions (e.g., gapping) require a linguistically expressed antecedent; (p. 204) others (e.g., sentence fragments) can occur as long as the intended meaning is recoverable in context. We offer an account of gapping that is both general, in that it can include cases of traditional “argument cluster coordination,” and restrictive in that it includes a constraint on register. In a comparison between English and French, French VP-ellipsis can be seen as much more restricted than English. We conclude that a constructionist approach that emphasizes semantics and surface structure, and which allows for distinctions within and across languages, is preferable to universalist copy and deletion proposals. Each construction posited makes predictions that are then testable against new data. A single general rule is not explanatory because every theory must ultimately account for the subtle differences among constructions that exist within and across languages. Thus invisible formal structure does not license the form, interpretation, and distribution of ellipsis, constructions do.

Acknowledgements We are grateful to three anonymous reviewers as well as to Jeroen van Craenenbroeck and Tanja Temmerman for very helpful comments on an earlier draft of this chapter. We also wish to thank the Einstein Foundation of Berlin for generous funding to AEG.

Notes: (1) Another available option that can be used to emphasize what is distinct is the use of contrastive stress (see acceptable examples in Table 8.2). (2) An anonymous reviewer suggests that this example could be partially addressed if we were to assume that a constituent, his front teeth to protrude, exists before a “raising” operation on the “subject” of that phrase, and that it is this phrase that is copied and deleted by ellipsis. But remember does not allow a VP with to (*Henry remembered his front teeth to protrude), so positing this as an “underlying” form that is then raised and deleted would require that ungrammatical forms are base generated, counter to prevailing assumptions. (3) An anonymous reviewer points out that respective is used to evoke a semantic pairing that is not explicit in the linguistic material in the following example: (a.) The bacteria were classed in their respective genera.

Page 20 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Construction Grammar Interestingly, this does not seem to be possible for respectively: (a.’) ?? The bacteria were respectively classed in their genera. (4) The assumptions behind the design are not entirely clear to us, since it would normally be expected that the double-object construction should be primed in all three cases, assuming structural priming lasts beyond one intervening clause (Bock and Griffin 2000, although see Bernolet, Hartsuiker, and Collina 2016). (5) We use ‘??’ instead of the more traditional ‘*’ in recognition of the fact that judgments are gradient and affected by many factors including context and intonation. (6) See Abeillé et al. (2014) for relevant discussion and formalization of the parallel gapping constructions in French and Romanian. (7) Cf. . The corpus examples were taken from contemporary texts (i.e., written after 1989). An English gloss is provided for all French examples in this section. A translation is also given if it differs from the gloss.

Adele E. Goldberg

Adele E. Goldberg is a Professor of Psychology at Princeton University where she is also affiliated with the linguistics and cognitive science programs. Goldberg’s work focuses on the psychology of language, particularly on how grammatical constructions are represented, learned, and processed. A more specific interest is in the functions of constructions and how those functions explain facts that are often assumed to be purely syntactic. Florent Perek

Florent Perek is a Lecturer in Cognitive Linguistics at the University of Birmingham in the Department of English Language and Applied Linguistics. His main research interests lie in the study of grammar from a cognitive and corpus linguistic perspective, with a particular focus on how syntactic constructions are mentally represented, how they are learned, and how they change over time.

Page 21 of 21

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

Oxford Handbooks Online Ellipsis in Dynamic Syntax   Ruth Kempson, Eleni Gregoromichelaki, Arash Eshghi, and Julian Hough The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.9

Abstract and Keywords In this chapter, we argue that ellipsis is a phenomenon that directly parallels anaphora, hence providing direct evidence of the concept of context on which natural language (NL) processing depends. From this perspective, we argue first that, in failing to give due recognition to the interactive and multimodal nature of NL processing, theoretical linguistics has entered a stalemate situation in which no unitary account of ellipsis is possible. The alternative Dynamic Syntax account that we provide next, however, presents ellipsis as a test case for the view that each NL constitutes a set of mechanisms for situated human interaction, with syntax not as a level of representation, but instead comprising a set of procedures for incrementally and predictively effecting conceptual structure–NL-string mappings. The significance of the extended set of so-called ‘elliptical’ phenomena that are examined from this perspective is that they all provide evidence for the seamless integration of NL structures and processing under domain-general action and perception processes. Keywords: Dynamic Syntax, ellipsis, anaphora, dialogue, correction, clarifications, split utterances, incrementality

Ruth Kempson, Eleni Gregoromichelaki, Arash Eshghi, and Julian Hough

9.1 Language as action and the phenomenon of ellipsis FROM the perspective that takes language as a form of action, ellipsis is (part of) the phenomenon where use of verbal actions by an interlocutor is deemed less efficient for coordination in interaction than relying on the context, either of one’s own recent verbal actions (see both utterances in (1)), somebody else’s verbal actions ((1B), (2B), and (3B)), or of the physical environment (4): (1) Page 1 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

(2)

(3)

(4)

This view of ellipsis, as a phenomenon seamlessly integrated in and enabling the joint action of conversational participants, has been obscured by theoretical linguistic approaches that abstract linguistic phenomena away from interaction and environmental resources and view natural language (NL) instead as a code, to be analysed in the model of formal languages with a specification of sentential syntactic organization and parallel mapping to propositional semantics as the explanatory basis. In such an approach, linguistic strings like those in (1) are analysed as (a) containing syntactically determined (p. 206) ellipsis sites, in the sense that the interlocutor’s presumed silence is attributed some underlying structure, which (b) needs to be ‘licensed’ by special linguistic devices so that (c) a complete proposition including the meaning of the ‘missing’ part is recovered on the basis of identity with some other linguistically derived structure. Assumptions (a) and (b) immediately exclude (2) and (3) because here there is no obvious ellipsis site, nor can there be predefined linguistic rules which determine the interactive meaning of these utterances as continuations, interruptions, or explicit commentary on (aspects of) another’s talk. Assumption (c) immediately excludes the explanation of phenomena like the one in (4) as part of the model because here there is no linguistically derived content to rely on in deriving interpretation. As a result, such accounts of ellipsis restrict their attention to strings like the ones in the individual utterances in (1) where, it is claimed, the specifically linguistic knowledge that people exhibit in interpreting them circumscribes them as a distinct phenomenon. We argue here that this delimitation of the phenomenon, and its commitment only to explaining a subset of the data, show that the standard methodology of abstract sentencedescriptive grammars impedes proper understanding of what a unified account of ellipsis really shows: namely, language users’ know-how to employ and derive significance from multiple resources as they engage moment-to-moment with each other. For this reason, we argue, the standard competence–performance distinction that underlies the standard

Page 2 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax methodology has to be replaced by a different perspective in which the coordinated and context-dependent actions of conversational participants are central to the explanation. In this chapter, we sketch an alternative view of ‘syntax’ as an ability to act jointly, hence displaying properties in common with those that have been assumed up to now as characteristic of semantics, pragmatics, and processing models. From this perspective, we will show how the presumed ‘syntactic’ licensing of an ‘ellipsis site’ as in (1) can be accounted for without the assumption of an autonomous level of syntactic analysis. Under the same terms, we also aim to show how recoverability of ellipsis interpretations, including those without linguistic antecedents such as (4), can be achieved on an incremental dynamic basis without assuming the necessity of sentence–proposition mappings. We do this through modelling the various phenomena within the architecture of Dynamic Syntax (DS, Kempson et al. 2001), an inherently incremental and actionoriented formalism for modelling interaction. We introduce DS in §9.2 and then show how it captures the various interactive ellipsis phenomena in §9.3 after we discuss the challenges for other standard accounts in the next subsections.

9.1.1 The ellipsis stalemate in competing domains of analysis Syntactic accounts of ellipsis rely on models of linguistic ability that postulate distinct types of declarative propositional knowledge as necessarily underpinning human linguistic performance. Having adopted this perspective, they then accordingly partition the phenomena by attributing to some instances causes distinct from those attributed to other intuitively similar cases. This is because such models, by ab initio definition of their remit, exclude in principle certain cases from consideration (see, e.g., Stabler 2013). For example, the cause of the surface appearance of the strings in (1), so-called VP-ellipsis and sluicing, is assumed to be underlying syntactic structure that has remained unpronounced: since Ross (1967) and Sag (1976a), the ellipsis site is analysed as projecting complex sentential structure which is deleted at a late stage of the derivation under conditions of identity with that of an antecedent clause. However, even for this now narrowly circumscribed domain of phenomena, problems arise for such an account, which is symptomatic of the illegitimacy of this fractionating process. Firstly, the particular interpretations naturally recovered for various such ellipsis sites turn out to appear very diverse, which, due to compositionality considerations, then leads to constant differentiation of categories of ellipsis types (e.g. for VP-ellipsis: endophoric, exophoric, null-complement anaphora, tautosentential, discourse-anaphoric, etc., all terms and phenomena illustrated in Miller and Pullum 2014; for various differentiated semantic categories of sluicing, see, e.g., Fernández et al. 2007). Secondly, it has been shown that identity of surface structure is not required for the licensing and recoverability of meaning of ellipsis sites (see, e.g., Hardt 1993). And, lastly, meta-theoretically, the account lacks parsimony, since antecedents which present a single intuitive reading have to be assigned ambiguous underlying structures to allow for the multiple possible ellipsis types. From an empirical point of view, this makes such accounts psychologically unrealistic, since a left-to-right incremental version of them must invoke ambiguity to account for the subsequent interpretation of the ellipsis site. A particularly notorious Page 3 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax such problem is the strict vs sloppy ambiguity (Sag 1976a), where, for a single fixed interpretation of the antecedent clause, the ellipsis site allows for two interpretations, e.g., in (5) his can be taken to denote either John as ‘Bill checked John’s mistakes’ (the strict interpretation), or Bill as in ‘Bill checked his own mistakes’ (the sloppy interpretation): (5)

To avoid this problem, a competing semantic account (Dalrymple et al. 1991) presumes that ellipsis sites are syntactically simple with no hidden structure. The burden of explanation for the strict/sloppy ambiguity is then shifted to an abstraction over semantic content of a clausal antecedent, which involves either abstracting out only the subject content (the strict interpretation), or abstracting out all occurrences of the term occupying that subject position (the sloppy interpretation). Notwithstanding their differences, syntactic and semantic accounts share the property of being sentence-based: the elliptical string is taken to correspond to a sentence which is in some sense incomplete, with interpretation recovered from a clausal antecedent. Both thus confront the difficulty of characterizing sub-sentential utterances which provide no evidence of having a sentential antecedent and yet are able to achieve illocutionary acts, which Stainton (2006b) argues require a pragmatic inference account over and above whatever semantic or syntactic accounts of ellipsis might achieve: (6)

Moreover, there are sub-sentential utterances where no identical, linguistically provided antecedent is available on which the recoverability of some particular interpretation can be based: (7)

(p. 208)

9.1.2 Morphosyntactic licensing of sub-sentential talk

The difficulty confronting both semantic and pragmatic accounts of ellipsis when applied to accounts of sub-sentential utterances, is that the morphosyntactic forms of such utterances remain the same whether they appear as sentential constituents or not (Merchant 2004a; Ginzburg 2012; Gregoromichelaki 2012). For example, they display specific case requirements: (8) Page 4 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

Contra Stainton (2006b), in such case-rich languages, the form has to match whatever is taken as the required interpretation, even in the absence of any sententially, or even linguistically, derived antecedent (Gregoromichelaki 2012, 2016). For example, in a context as in (9), den Arzt needs to be in accusative case:1 (9)

A further problem indicating the need to integrate formal constraints with interpretational ones, is that ellipsis is argued to be subject to island constraints (Ross 1967), until very recently, widely taken to be diagnostic of a syntactic phenomenon (see Merchant, this volume). For example, the so-called Complex NP Constraint is exhibited in so-called Antecedent-Contained Deletion/Antecedent-Contained Ellipsis constructions. In such cases, a dependency needs to be licensed within the restrictor of a quantifying element provided by the adjunction of a relative clause. A further elaboration of the restrictor provided by a second relative clause prevents such a dependency: (10)

(11)

The challenge of defining accounts of sufficient richness to express such phenomena has been addressed in an impressively detailed manner by Ginzburg (2012) and colleagues, (p. 209) within a constructional version of HPSG formulated in the Type Theory with Records (TTR) representational framework (HPSGTTR, Ginzburg 2012; Ginzburg and Miller, this volume). In this model, sub-sentential strings, as in (7), are taken to express paraphrases of what a whole sentence formulation would denote; and so the subsentential elements are mapped directly onto the various semantic/pragmatic functions they are taken to accomplish. However, this attempt to analyse ‘fragments’ as commensurate to whole sentences means that the HPSGTTR framework retains conservative assumptions vis-à-vis syntax: fragments are modelled as ‘constructions’ of sentential type, involving sui generis syntactic mappings of sub-sentential constituents directly to ‘sentences’ paired with necessarily (quasi-)propositional interpretations. For example, one of the grammatically derived interpretations of the string in (7) will be ‘A asks B who A meant by the use of “Billy”’. In consequence, each such sub-sentential Page 5 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax string is assigned multiple such pairings and is by definition characterized to be as spuriously ambiguous as the constructions it is involved in. This means that the problematic aspects of the syntactic accounts, namely ambiguity instead of general mechanisms, is not sidestepped—on the contrary, it is wholly embraced as is natural in a constructional approach. However, as argued in Gregoromichelaki (2012), the interpretational options for such fragments are open-ended, so a constructional approach faces the problem of ignoring the radical flexibility and adaptability of linguistic resources. Additionally, there remains the empirical problem that morphological/syntactic restrictions on ‘fragments’ pervade any occurrence of sub-sentential utterances whether these depend on linguistic or wholly non-linguistic contexts (see (9) and (12)):2 (12)

The defined HPSGTTR constructions apply only to linguistically provided partial antecedents which are clarifiable, so the rules specified by the grammar will not apply to the intuitively similar clarification cases such as these. An account that aims to capture global syntactic generalizations while, at the same time, accounting for licensing idiosyncrasies without an independent level of syntactic representation over strings is the general framework of Minimalist Grammars (MG, Stabler 1997; Kobele 2006). The MG formalism, applied to elliptical phenomena in Kobele 2012b, 2015, (p. 210) requires, firstly, the relation between sentence string and semantics to be mediated by extensive transformations. This is because a copying account of ellipsis recoverability under identity is sustainable only on the basis of a decompositional mapping of surface appearance to semantic interpretation. This mapping is effected through various grammatical operations, like merge, move, and indeed e(llipsis), applying over fine-grained linguistic categories (‘types’) of expression, as constrained by global syntactic constraints such as the Shortest Move Constraint (SMC). This account presents a significant improvement on standard syntactic analyses, since an ellipsis site is considered, as in semantic accounts, syntactically atomic, i.e. no underlying syntactic structure is derived. Instead, ellipsis sites are generated due to particularized grammatical operations mapping empty or fragmentary strings to fine-grained syntactic categories. Because both syntactic categories and operations are so fine-grained, this account allows ellipsis sites to behave like pro-forms (as in Hardt 1993) whose antecedents are syntactically restricted to be particular ‘derivational contexts’, i.e. syntactic derivations that behave as functions on the potentially overt linguistic elements that are mismatched between the antecedent and the ellipsis site.

Page 6 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax Given that this account incorporates global syntactic restrictions like the SMC, it can deal with formal licensing constraints like the island restrictions of (11). Additionally, the fact that it makes use of ‘derivational contexts’ as antecedents, instead of syntactic structure, means that it can deal with cases of mismatched antecedents, as in (5), without having to postulate destructive structural operations or unnecessary ambiguity in the antecedent itself. However, this model still maintains standard assumptions regarding syntax, in that its explanations rely on assigning psychological reality to a view of syntax that is independent and distinct from other cognitive abilities. Syntactic categories, operations, and global constraints, like the SMC, are assumed to be specifically linguistic and not attributable to general cognitive constraints (see also Stabler 2013). This has consequences: for example, in order to fully account for phenomena like (5), the grammatical characterization provided is not adequate. Given that MG is a competence model, it has nothing to say on how antecedents are defined. In order for antecedents to be provided, MG has to be integrated in a performance model, a parser/generator (Kim et al. 2011; Kobele 2012b). This is because the ‘derivational contexts’ invoked by the assumed pro-forms simply represent, as far as the grammar is concerned, decontextualized descriptions of the combinatorial licensing of linguistic elements. So, from our point of view which takes the grammar–processor distinction as unsustainable and an artifact of the view of language as a code, the MG account (a) introduces theoretical redundancy because syntactic restrictions are not modelled on the basis of general cognitive processes directly; (b) in the case of ellipsis, this has the consequence that such phenomena are analysed as involving various cognitively arbitrary category specifications or being explainable on the basis of otherwise general processing explanations; and (c) as a consequence, because the MG grammatical account is confined to linguistic structures, it is unable to account for intuitively analogous non-linguistic antecedents such as (9) and (12), since, even embedded in a performance model, there is no linguistic derivation in the context to be invoked as the antecedent grammatical operation. Finally, along with all the accounts already examined, given that the syntactic derivations of MGs license whole sentential strings paired with propositional interpretations, it will be unable to account for dialogue phenomena like (2), as we will now argue. (p. 211)

9.1.3 Ellipsis as completability Accounts of ellipsis assuming a sententialist methodology struggle to address the challenge of the dynamics of conversational dialogue. The problem posed by conversational exchanges is that people commonly and purposefully do not utter full sentences. Conversations are replete with fragmentary forms which build incrementally on what has gone before, in which participants effortlessly switch between speaker and hearer roles, and commonly no-one in the conversation will have anticipated its overall content in advance: (13)

Page 7 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

If one attempted to preserve the assumption that all fragments are independent incomplete sentences, one might consider data like (13) as at least five elliptical sentences each to be mapped onto representations of propositional content; but it is far from obvious what source to provide to any such posited underpinning sentential structures when the overall structure cannot be anticipated. Even the illocutionary flavour of the subparts may shift as participants are more or less authoritative, or in disagreement: (14)

An account in terms of sentential ellipsis of the above misses the significance of B’s action in (14) as an interruption and the ones in (13) as continuations. In any case, such an ellipsis account cannot be sustained since this split utterance phenomenon can take place at any point in an emergent structure (Gregoromichelaki et al. 2011; Howes et al. 2011), and affects every syntactic/semantic dependency: preposition–complement (15), reflexive–antecedent (16), quantifier–pronoun binding, negative polarity, quantifier–tense (17), even head–complement dependencies (18): (15)

(16)

(17) Page 8 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

(p. 212)

(18)

Therefore, no account that relies on specialized constructions or syntactic types, instead of general mechanisms, can provide an adequate explanation for this phenomenon. Instead, sub-sentential strings need to be licensed by the grammar incrementally and without the requirement that a whole sentence needs to be licensed eventually (as in Kobele 2012b and others). This is because such sub-sentential strings are not only interpreted incrementally but also can be used perfectly naturally to perform speech acts that rely on their very incompleteness (Gregoromichelaki 2013): (19)

But even incremental syntactic accounts (e.g. Poesio and Rieser 2010, or a conceivable extension of Kobele 2012b, 2015) cannot deal with such data if they maintain an independent level of syntactic analysis since, as (16) shows, splicing together the two strings will yield ungrammatical results. Given the rapidly accumulating experimental evidence that parsing and production are both incremental and interdependent in achieving action and representational coordination during interaction (Pickering and Garrod 2013), accounts of sub-sentential utterances as somehow incomplete cannot be sustained. Firstly, as we saw in (9) and (12) for adult interaction, sub-sentential talk underpinned by the context dependency of human interaction can achieve any function that sentential and propositional units perform. Secondly, such utterances form the basis of language acquisition. Children at the one-utterance phase can perform perfectly adequate verbal actions employing subsentential strings relying on the adult caregiver to provide what they lack at that stage, namely, the requisite conceptual breakdown of a holistic situational representation so that new linguistically expressible concepts can be acquired: (20) Page 9 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

Moreover, the split utterance data cannot be excluded from the core remit of the grammar as performance disfluencies since they also constitute effective verbal actions for the language-acquiring child to coordinate with caregivers: (21)

(22)

Instead, in our view, the characterization of sub-sentential data needs to make reference to the dynamics of real-time processing expressed in a vocabulary that allows (p. 213)

contribution from environmental resources and non-linguistic cognitive domains at a subsentential level. Such a model, in our view a grammar, needs to: • include a set of mechanisms for inducing recovery of content from both linguistic and non-linguistic sources; • model word-by-word incremental, semantic licensing; • model the context as an evolving record of all inputs, including words, structures, and procedures that induce such incrementally developing content; • define shared mechanisms for parsing and production. The model we sketch below, Dynamic Syntax (DS), meets these criteria. In DS, the pervasive use of sub-sentential elements across dialogue turns and participants will not only be wholly predicted, but will be seen to serve a key purpose in communication by narrowing down the otherwise mushrooming alternative structural and interpretative analyses through the ongoing interactive coordination of participants (see Gargett et al. 2009; Eshghi et al. 2015). We will argue that this perspective not only illuminates the sharing of linguistic actions in context, as seen in (13)–(22), but also provides a processing approach which reveals a novel and surprising meta-theoretical generalization: namely, the lack of need for conceptual distinction between syntactic vs semantic vs pragmatic procedures in the integration of verbal actions with coordinated joint action. In all three domains of traditional linguistic inquiry the mechanism of jointly developing initially underspecified information operates uniformly. Just as pronominal anaphora construal has been shown not to be definable as confined within sentential, Page 10 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax speaker-turn, or linguistic boundaries, ellipsis needs to be modelled in the same way, as an anaphoric mechanism both relying on and sustaining participant coordination and context-dependent action. This is achieved in DS by modelling both pronominal anaphora and ellipsis through the initial projection of temporarily underspecified contents imposing their resolution as a goal for all conversational participants and thus subsequently resolved by exploitation of either environmental or linguistic resources. Furthermore, we argue, in contrast to any other formalism, this perspective allows us to extend the parallelism beyond ellipsis and pronominal anaphora, to the explication of the function of syntactic mechanisms. Syntactic dependencies introduced by fragments, morphosyntactic constraints introduced by e.g. case morphemes, but also the mechanism underpinning long-distance dependency, will be modelled as introducing initially underspecified structural projections with goals imposed on either participant in a conversation to resolve them either through context or by offering an appropriate subsequent verbal action.

9.2 Dynamic Syntax Dynamic Syntax (DS) is a grammar formalism based on the psycholinguistically inspired action-based modelling of NL string–interpretation mappings in context. NL syntax is reconceptualized to be no longer its own level of representation, but, instead, a set of (p. 214) licensing actions for inducing or linearizing semantic content, incrementally, predictively, and on a word-by-word basis. Parsing and production are interdefined as coordinated activities operating in tandem, each having access to the same set of construction steps. Context in DS is modelled as including an incrementally evolving record of all the actions employed and their outcomes, enabling recoverability of any of these for reuse. The interaction achieved in split utterances emerges as an immediate consequence, since both parsing and production employ predictive actions which make available to each interlocutor, at each sub-sentential step, the role of parser or generator.

9.2.1 String-content mappings To model this action-directed perspective, DS is founded on a dynamic modal logic that defines the transitions among states taken to constitute the current context of processing at each point in a parse/generation task (see Kempson et al. 2001 for formal details). The accessibility relations among these states are defined through actions which license goaldriven, time-linear transitions from state to state. Such states can be taken to model the total context of each processing step, linguistic and non-linguistic, so that the whole system licenses mappings from context to context. In DS, particular conceptualizations of eventualities are taken to involve actions of building (parsing) or linearizing (production) a semantic tree whose nodes incrementally come to reflect the content of some utterance. Keeping in mind that processing can start at any point, if we take the case of aiming to achieve a propositional structure, the first step is a one-node tree that merely states the goal to be achieved, namely, to derive a Page 11 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax formula of propositional type (Figure 9.1, above ⇓), indicated by the requirement ?Ty(t), the query symbol ? indicating this is a goal as yet unrealized. This single node tree is then incrementally enriched as word-by-word processing proceeds, eventually leading to a complete tree of propositional type with no goals left outstanding (Figure 9.1, below ⇓), a result achieved through combining grammatical actions with content derived from context. DS trees are invariably binary, reflecting functor– argument structure (argument node on the left branch, functor node on the right), with a pointer ◊ identifying the node under development). Each node is annotated, not with words, but with terms of a logical language (e.g.

Figure 9.1 Processing John upset Mary in DS

Ty(e)), these terms here being subterms of the

(p. 215)

Mary′), and their type (e.g. resulting propositional

representation at the root node. The representation includes an event/situation argument S of type es, enabling tense/aspect construal (we suppress details here; see Gregoromichelaki 2006; Cann 2011).3 Since event/situation arguments are first-class citizens of the combinatorial structure, predicate assignments reflect this: Ty(e→(es→t)) the intransitive-verb type for combining with a subject first and then the situation argument; and Ty(e→(e→(es→t))), the transitive type.

9.2.2 Formalization of tree structure and incremental tree development In order to talk explicitly about how such structures are constructed incrementally, trees are formally defined, so that their incremental construction can be achieved through actions that induce the requisite tree development. DS adopts a logic with modalities indicating not only what is currently true about the tree but also what is to be true at future developments of the tree (Blackburn and Meyer-Viol 1994). For example, the daughter relation is defined as 〈↓〉: 〈↓〉α holds at a node if α holds at its daughter (with variants 〈↓0〉 and 〈↓1〉 for argument and functor daughters respectively). There is also the inverse mother relation 〈↑〉α with argument/functor variants, 〈↑0〉, 〈↑1〉. Thus a DS grammar includes a set of actions which are procedures for building/ linearizing trees: ‘syntax’ is just the subset of actions which unfold tree structure under the guidance of linguistic elements. Syntax is formulated as the employment of packages (macros) comprising simple atomic actions—make(X) for creating new nodes, go(X) for

Page 12 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax moving the pointer, put(Y) for annotating nodes, where X and Y are tree relations and node annotations (labels) respectively. Computational actions are language-general macros not needing a lexical trigger: these introduce tree relations; move the pointer ◊ around some partial tree under construction (e.g. COMPLETION, which moves the pointer out of a node once its type requirement is satisfied); remove satisfied requirements (THINNING); and perform function-application (BETA-REDUCTION) operations compiling information up a tree on a bottom-up basis once all daughter decorations are suitably provided. Primary amongst these is the construction of ‘unfixed nodes’ (*-ADJUNCTION): such actions are defined using the Kleene star operator * annotating a so-called ‘unfixed’ node with the specification 〈↑*〉 Tn(a), which indicates that this node is related to some node a along a sequence of zero or more mother relations to it. Lexical actions are macros triggered by individual words of the particular NL whose actions also perform the same tree or string development along with introducing semantic content on the nodes of the tree. Partial trees: imposing constraints on update. The underpinning dynamic which drives NL processing is the gradual resolution of initial underspecifications. There are three basic types, each with an attendant requirement for update: (p. 216)

(1)

(2)

(3)

Such underspecifications all allow modal constraints on their resolution, giving rise to a variety of locality restrictions. For example, anaphoric expressions can impose locality constraints associated with limits on recoverability of their antecedent. Locality within a given predicate domain is definable as the tree traversal of an unbroken functor path: arguments local to a given predicate meet the characterization 〈↑0〉〈↑*1〉Tn(a) (i.e. there is a path up one argument relation plus a possibly empty sequence of function-path relations from the current node to the dominating Tn(a) node). Reflexive anaphors are characterized as projecting a restriction on putative antecedents that they be ‘coargument’ terms, i.e. holding along the relation 〈↑0〉〈↑*1〉〈↓0〉Tn(a). Conversely, Page 13 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax pronominals exclude as antecedent any formula at a node standing in such a local relation; and wh-pronouns contribute a metavariable (WH), defined as awaiting substitution by a term, the answer, in some future utterance, a yet further long-distance dependency effect (see Gregoromichelaki 2006, 2013 for a reformulation of Binding Theory in DS terms). Unfixed tree relations vary as to whether or not their resolution is subject to such tight locality restrictions. An unfixed node within some tree is identified by the annotation 〈↑*〉 Tn(a),?∃xTn(x) which defines it as unfixed but within a tree whose tree root is Tn(a). There is also a more specific localized variant defined as 〈↑0〉〈↑1*〉Tn(a), i.e. the tree root now encompasses a local predicate–argument sub-tree. Morphosyntactic specifications also involve locally statable constraints. Accusative case marking, for example, is definable as imposing a requirement on a node of the form ?〈↑0〉 Ty(e→(es→t)), dictating that in the outcome, it must be immediately dominated by a predicate node. In every case, what drives processing is the set of requirements lexically or computationally imposed: such requirements operate as predicted goals for either participant in the joint task to accomplish that constraining of future developments. Of these, it is underspecification of hierarchical position within a tree that lies at the core of the account, underpinning many discontinuities, and expressible through the construction of unfixed nodes. Figure 9.2 shows how the building of a locally unfixed node (LOCAL-*-ADJUNCTION) can feed lexical actions, so that in combination, the word order properties of each particular NL are derived. In some languages this underpins scrambling. In English, this mechanism, in combination with the lexical action of the verb, induces the imposed SVO (subject–verb–object) ordering: the DP before the verb is taken to annotate a locally unfixed node; its (p. 217) position as logical subject (or not) is determined by the action sequence projected by the following verb, which among other actions includes such specification. The left-peripheral nature of long-distance dependency is expressed through the building of an unfixed tree relation without such a single-clause locality restriction, which involves a computational action defined in the presence of a node requiring completion to a propositional type Tn(a),?Ty(t) with a precondition that the node is not already developed in any other way, relative to which the action defined licenses the construction of an unfixed node annotated (p. 218) as 〈↑*〉 Tn(a),Ty(X),?∃xTn(x), i.e. with a resolution of this underspecified tree relation having to be at some possibly later point in the construction of the tree being developed from Tn(a) (this is the final step of MERGE4 in Figure 9.2); hence the potentially ‘long-distance’ discontinuity effect.

Page 14 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax In all tree developments, the closing stages invariably involve modalized functionapplication (beta-reduction) steps which compile up the content of all non-terminal nodes to finally yield a complete tree with no requirements outstanding (see Figure 9.1).

Figure 9.2 Unfolding structure for John upset…

9.2.3 Linking trees through term sharing In addition to the incremental projection of predicate–argument structures, a freely available computational action licenses paired ‘LINKed’ trees, which are in effect correlated through a conventionalized anaphoric device that results in a shared term appearing in both such trees. Formally, a transition from some node in the main tree onto the top node of a new tree is licensed, this new tree-beginning having imposed on its development a requirement that it include the term appearing on the node from which the transition was initiated. Relative clauses are the paradigm case: the relative pronoun triggers actions which reflect the LINK transition and some initial construction within the newly emergent tree of an unfixed node. The WH-metavariable then functions in a fixed anaphoric way to provide the required copy from the main tree, thus creating the resultant shared term in the two trees (see Figure 9.3, which displays the outcome of processing the string John, who smokes, left).5 Such LINKed trees provide opportunities mid-sentence for NL processing to shift temporarily to a distinct structure for purposes of elaboration, expansion, explanation, etc. of terms in the main structure. And Figure 9.3 Result of parsing John, who smokes, left this can happen either within a single individual’s utterance, giving structures like relative clauses, HangingTopic Left-Dislocation, clausal and phrasal adjuncts, etc., or across speakers where the effects include clarifications, confirmations, continuations, etc. (see e.g. Gargett et al. Page 15 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax 2009). This is because LINK- (p. 219) adjunction allows indefinite iteration of adjunct extensions of arbitrary type, a common way of extending utterance exchanges: (23)

The fact that LINKed trees temporarily shift processing to an independent adjunct tree to elaborate a term on the main tree provides the basis for a requirement of anaphoricity from the one structure to the other. It also explains why this form of tree relation is what is needed for modelling sluicing (see e.g. Merchant 2001), amongst other structures which may be sensitive to island constraints as we shall see in detail in §9.3.7 for antecedent-contained ellipsis. In such cases, for example, where relative clauses constitute the linked ‘islands’, the relative pronoun provides the copy that links the newly introduced tree to the main tree. This copy process creating the shared term is NOT island-constrained with respect to its antecedent. Nonetheless, since that copied term has to occur on an unfixed node as the new tree is being introduced, the resolution of its position within that new tree is domain-sensitive: it must occur within the construction of that very tree. In this sense, island constraints in DS emerge from the need to shift processing to another domain along with needing to complete this structure before resuming processing of the one temporarily interrupted.

9.3 The Dynamics of Ellipsis With the formal underpinnings of its semantic tree construction process to hand, we now describe the DS model of ellipsis. We show how the incremental notion of DS context can account for the varying types of elliptical forms found in interaction.

9.3.1 Context in DS: Mechanisms for recovery of content at the ellipsis site The concept of context appropriate for a processing-oriented account is substantially richer than is expressible in either model-theoretic accounts or semantically blind syntactic accounts. For this reason, context in DS is conceived as a dynamic, multimodally induced record of (a) words; (b) conceptual content notated as tree structures as described above; and (c) the sequence of actions in building the emergent strings and trees (Cann et al. 2007; Purver et al. 2010; Eshghi et al. 2015). Since a DS grammar, through the imposition of predicted goals, models the recovery from memory of the range of options available for the next processing steps, this set of options is attributed Page 16 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax psychological reality so it also constitutes part of the model of context. To model context time linearly, we follow Sato (2011a), Purver et al. (2011), and Eshghi et al. (2015) in defining this range of options as a (p. 220) Directed Acyclic Graph (Context DAG) as in Figure 9.4, where each node represents the current (partial) tree and each edge in the graph records the action taken. The entire DAG represents the search space for parsing/ generation, which will also include any conceptual content derived from context. In Figure 9.4, the top path from T0 to T12 is the chosen one given the lexical input, and corresponds to the derivation outlined in Figure 9.2.

Figure 9.4 DS parsing context as a graph: Actions (edges) are transitions between partial trees (nodes)

The context of a partial tree is then the path back to the root of this graph; and actions, as well as conceptual content notated in tree form, are recoverable through a mechanism of backwards search for reiteration/

reuse in creating new construals. As a result, there are three basic mechanisms by which an ellipsis site, being an underspecified element awaiting resolution, can exploit the Context DAG for content recovery: (a) Reuse of content (semantic-formulae) from some (partial) tree on the Context DAG, which can include recoverability of information direct from the utterance scenario, yielding indexical construals. (b) Reuse of sequences of actions from the DAG (sequences of DAG edges). (c) Direct reuse of structure, i.e. extension of some (partial) tree in context.

9.3.2 Content underspecification and recoverability through copying or action replay Forms of the ellipsis site. We take first the familiar cases of English VP-ellipsis. Under DS, these involve an initial content underspecification projected at the ellipsis site, with optional triggers for content recovery in the form of strings such as do so, too, and bare auxiliaries. Auxiliaries are taken, like pronouns, to project a temporary content placeholder of predicate type (e.g. U :e→(es→t)) with an accompanying requirement ? ∃x.Fo(x) which triggers the context search. Recovering predicate content. In such cases, two types of information are recoverable from context: (a) a formula content copied from some tree; (b) a sequence of actions to be reiterated at the ellipsis site.

Page 17 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax The two mechanisms fetching such information to the new processing task through backwards search are implemented by the pair of computational actions Substitution and Regeneration.6 Figure 9.5 displays the identical substitution process for (a) a pronominal and (b) an elliptical expression as in (24): (p. 221)

(24)

Figure 9.5 Substitution from context at the ellipsis site of (24): Pronominal anaphora (top) and VPellipsis (bottom)

Figure 9.6 then displays the regeneration process at an ellipsis site for cases like B’s answer in (25): (25)

Figure 9.6 Action replay from context at the ellipsis site

Page 18 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

The new content in B’s answer is derived through rebinding of the variable projected by the anaphor to the newly introduced subject. This is achieved by searching the DAG and then reusing the actions involved in constructing the question to flesh out the ellipsis site in the answer, as triggered by the metavariable placeholder projected from the auxiliary. The processing of the question in (25) involves parsing of the subject who, constructing a predicate as indicated by the verb, and constructing its object argument; these actions, (p. 222) having been stored as a sequence in context, can be reused in the next stages. But with John’ now annotating the subject node, these actions will now yield a rebinding of the object argument to the new local subject. The effect achieved is the same as the higher-order unification account of Dalrymple et al. (1991) but without anything beyond what has already been used for the processing of the previous linguistic input and, consequently, without any need to assign some distinct type of expression to the elliptical element did or the subject John. All that has to be assumed is that the metavariable U contributed by the anaphoric did can be updated by an action sequence taken from the DAG context. This duality of mechanisms for content recovery provides a way of capturing all strict/ sloppy ambiguities observed in several forms of ellipsis. Either content specifications or sequence-of-action specifications can be searched for and reiterated, the first preserving some previous construal, the other preserving only the pattern of construal: (26)

This gives precisely the right basis for the ambiguity without having to invoke ambiguity of the antecedent. In (26), a strict construal λx.Help′(Students′(of – Bill′))(x), ‘help Bill’s students’, is carried over as the fixed predicate content to be predicated of the subject John′. The sloppy interpretation involves the sequence of actions associated with processing help + his + students in the first conjunct reapplied to the new subject John′. (p. 223)

This is not a mechanism identified specifically for ellipsis: the same type of analysis applies to the DS analysis of quotation (Gregoromichelaki 2017) and two subcases of pronominal anaphora too, i.e. coreferentiality and so-called lazy pronouns. Coreferential construals constitute a replication of some antecedently constructed content of individual type e: (27)

Page 19 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax Lazy construals involve recovery of actions to be rerun at the site indicated by the pronoun to yield some requisite term that is distinct from that picked out by the antecedent. In (28), them is resolved by rerunning actions used to process his keys in the previous clause: (28)

Under this view, neither ellipsis sites nor pronominals deviate in any way from the usual processing mechanisms DS defines: like all other morphosyntactic and lexical specifications, such elements impose predictive goals constraining the subsequent actions of the interlocutors as these unfold in the DAG model of context; general processing mechanisms like backwards search, Substitution, and Regeneration can then be employed, by either interlocutor, to achieve appropriate expansions of such potentially radically underspecified content in ways that reflect their own purposes. Since the DAG unfolds and constrains these future processing options on a word-by-word basis, employing and resolving underspecification can be achieved either through the exchange of propositional contents, as seen above, or through the exchange of sub-sentential elements that exploit verbal or non-verbal actions for becoming integrated in the functioning of the participants’ joint actions.

9.3.3 Licensing sub-sentential utterances: Morphosyntactic constraints Use of sub-sentential utterances in dialogue shows how particular morphosyntactic forms impose goals on interlocutors that curtail the processing paths compatible with further processing or backwards search. For example, as we saw earlier, so-called wh-pronouns project a specialized metavariable requiring instantiation externally to the tree under construction, this in question–answer pairs being the answer to that question. On the other hand, the metavariable provided by reflexives is restricted to obtaining a locally provided antecedent. These two types of constraint can combine leading to interpretations that are very finely constrained: short answers to a wh-question (as in (29)) need to be interpreted within the structure which that question projects, while the local substitution imposed by a (p. 224) reflexive as shown in Figure 9.7 will lead to a cascading effect affecting all occurrences of the WH-metavariable, thus yielding an appropriate interpretation of the answer.

Page 20 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

(29)

Figure 9.7 A short answer with binding restrictions

Sub-sentential utterances may also bear morphological features, which are expressed in DS via imposition of predictive goals to interlocutors to pursue processing paths that conform to some particular specification of the conceptual content being developed. Like indexical pronominals, imposing, for example, gender restrictions in English, accusative case marking, as in (9)–(12), introduces the requirement: ?〈↑0〉Ty(e→(es→t)), a restriction that a node so annotated be immediately dominated by a node of predicate type. Since backwards search through the DAG will not fetch the appropriate content in this case, the imposition of such morphological constraints has the effect of imposing particular conceptualizations of the utterance situation becoming newly available as processing options in the DAG. Since the DS-DAG is not restricted to representing specifically linguistically derived content, such morphologically triggered context enrichment does not presuppose the mediation of any syntactic derivations. Nevertheless, unlike other frameworks (e.g. Ginzburg 2012), any DS rules applicable to linguistically derived content will be able to apply unchanged to the subsequent development of such contextually derived content (as in e.g. (12)).

9.3.4 Licensing split utterances Given that processing in DS is modelled as incremental and predictive, the licensing of sub-sentential utterances is achieved word-by-word without requiring necessary expansion to a propositional structure for a string to be interpreted or produced. So subpropositional strings can be employed to perform speech acts as naturally as propositional ones, as seen in (19) and (22). The DS explanation of the licensing of split utterances without having to assume inference of propositional intentions relies on the DS modelling of the tight coupling of parsing and production (‘mirroring’). Parsing in DS incorporates aspects of production, i.e., the parser does not passively expect the scanning of input to produce structure; instead, the Context DAG is expanded at each word step with the generation of predictions (goals) in order to accommodate what will be encountered next. On the other hand, production in DS employs exactly the same mechanism and DAG as parsing, with the only difference that, instead of scanning input, every step of articulation is licensed by a richer tree, a so-called goal tree. The goal tree Page 21 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax is a tree at least one step forward from the tree-under-construction, and a potential next processing step through the DAG is licensed by the tree-under-construction subsuming the goal tree. Subsumption means that the partial tree that is being developed must be extendible into that goal tree by following the licensed actions of the system (Purver and Kempson 2004). Put simply, whereas parsers have to follow what the speaker offers them, speakers have to have at least some partial idea of what they are going to be communicating at the next step. So given this incrementality and predictivity of processing, first, there is nothing to prevent speakers initially having only a partial structure to convey, i.e. the goal tree may be a PARTIAL tree, perhaps only one step ahead from what is being voiced. Secondly, since the DS-grammar implements a notion of predictivity, if, at some stage during subsentential processing, an interlocutor has the ability to satisfy the projected goals via their own resources, e.g. via lexical access or by extending the current tree with a LINKed tree, it is perfectly sanctioned by the grammar for them to take over and continue extending the partial tree through DAG paths perhaps not foreseen by the original speaker (see e.g. (2) and (14)). Moreover, since DS does not assume the mediation of any connected syntactic derivation in such cases, it is able to deal even with cases where, as we saw in (16), repeated modified here as (30), split utterances can take forms which would be ungrammatical under standard assumptions (*Did you burn myself?): (30)

(p. 225)

Figure 9.8 displays the partial tree induced by processing Mary’s utterance Did

you burn, which involves a substitution of the metavariable projected by you with the term standing for the current addressee, Bob. At this point, Bob can complete the utterance with the reflexive. This is because a reflexive, by definition, just copies a formula from a local co-argument node onto the current node, just in case that formula satisfies the person/number conditions of the expression, in this case, that it designates the CURRENT speaker.

Page 22 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax Since sub-sentential/subpropositional elements can perform various discourse functions (‘speech acts’), as seen in e.g. (19), the DS analysis of split utterances applies equally to cases like (reprise) sluicing, reprise clarifications, corrections, etc. without Figure 9.8 Incremental development of Mary’s/Bob’s assuming that such subcontext via processing words sentential constituents need to be sententially/ propositionally expanded (Kempson et al. 2007; Gargett et al. 2008, 2009; Gregoromichelaki et al. 2009; Eshghi et al. 2015). For example, cases of elliptical clarification interaction, mid-sentence, as in (31), only involve locally attached link structures, with their discourse function modelled only as effects on the DAG (see e.g. Eshghi et al. 2015): (31)

Such clarifications are construed relative to whatever constituent immediately precedes them, as in (31), where what is pertinent to the sub-sentential interruption is (p. 226)

the immediately preceding DP the doctor. So such utterances are analysable in exactly the same terms as split utterances more generally, where the fragment is taken to directly extend its antecedent structure in context, here just the content derived from the doctor through the independently motivated link adjunction. Figure 9.9 shows the result of processing (generating for B, parsing for A) of the utterance Chorlton.7

Figure 9.9 Processing Chorlton? in ‘A: the doctor B: Chorlton?’

Page 23 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

9.3.5 Licensing the splitting of dependencies across turns Having given a unified account of linguistic processing as mappings from context to context, making redundant extra machinery of idiosyncratic constraints applying to structure over strings, it becomes evident that the phenomena ‘anaphora’ and ‘ellipsis’ are really part of the wider phenomenon of incremental licensing of NL strings and interpretations. We therefore expect them to display parallel properties, and this is in fact what we find. (p. 227)

We first note the parallel potential between pronouns and ellipsis for anticipatory forward-looking resolution (‘cataphora’). In English, only some pronouns license this localized anticipatory function, with the clausal sequence following the predicate used to provide the value for the anticipatory expletive pronoun (for details, see Cann et al. 2005).8 Given our modelling of split utterances in terms of DAG options constraining the next processing actions of both interlocutors, as we would expect, the requirement of resolution of such an anticipatory device can be taken over by another party to formulate according to their own context or to serve their own purposes (see also (2)): (32)

(33)

Further, as we show above (e.g. (4)), there is the parallelism between pronominal anaphora and VP-ellipsis in that they are both resolvable, indexically, from the utterance situation. So ellipsis, like pronominal anaphora, is resolvable either through backwards search recovering content or actions from the Context DAG, or by just imposing a constraint for its resolution and restricting the available DAG paths for both interlocutors to those providing content for its resolution, or, indexically, by introducing in the DAG aspects of the utterance situation. Now, a generalization missed by other frameworks (e.g. Kobele 2012b) but revealed through this modelling is that this threefold parallelism extends to ‘syntactic’ mechanisms, for example, those underpinning long-distance dependencies. First, in parallel to cases (32)–(33), the method of initially building an unfixed node and resolving its position later is the standard mechanism used in DS for long-distance dependencies. But the structure that finally resolves the structural underdetermination of such unfixed nodes can equally be provided by backwards DAG search and reiterating actions from

Page 24 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax context. For example, Mary in (34) is uttered in a context where actions are available for repeated application: (34)

Here, triggered by the presence of too, Mary too is interpreted as ‘John upset Mary’ via a sequence of actions in which Mary’ is taken to annotate an unfixed node, later unified to become the internal argument of Upset’, exactly following actions used in interpreting the preceding string. And given that DS parsing and production mirror each other, resolution goals imposed on the DAG by one interlocutor both constrain and are resolvable by the other: (p. 228)

(35)

Completing the parallelism is the resolution of an unfixed node indexically. Of these, striking instances are children’s one-word utterances in early language acquisition, where they rely on their interlocutor to provide some open structure into which their presented fragment can be incorporated, as in (20) repeated here as (36): (36)

Thus, from this perspective, use of fragments in split utterances (see e.g. (15)–(17)) isn’t merely an optional economy measure, it is an essential vehicle for coordination, which is crucially exploited during both language acquisition and in day-to-day human interaction. Moreover, this prediction of a parallel dynamics between pronominal anaphora, ellipsis, and long-distance dependency resolution, inexpressible in other frameworks, provides meta-theoretical foundations supporting not merely the dynamic perspective on ‘syntax’ proposed, but also the potential it provides for context-grounded interaction.

9.3.6 Licensing repair devices: Self-repair and corrections But even further than this, a unified account of linguistic processing as mappings from context to context, aiming to model action coordination during interaction, can accommodate, in the same terms, the phenomena that arguably overtly promote coordination, namely, repair (see e.g. Schegloff 2007), which has traditionally been rendered a performance phenomenon unworthy of interest for theoretical linguistics. For Page 25 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax example, self-repair, a pervasive phenomenon in dialogue (Shriberg 1994), is naturally accounted for given the DS notion of the DAG context and content recoverability employing backwards search. The same backwards-search mechanism described above for various other ellipsis cases applies again in repair cases; however, these are triggered by the current processing state being abandoned and processing resuming again from appropriate prior points in the context (see Hough 2011 and Hough and Purver 2012 for details). This strategy models common types of self-repair phenomena such as short stuttered repeats (I, I go to Paris) and substitutions (John likes, uh, loves Mary), but it can also deal with more complicated licensing phenomena where, as predicted, ellipsis interacts (p. 229) with the repairing of structure. In the following example, ellipsis reconstruction must operate across an interruption point + in a repair:9 (37)

This is a case where the content of London repairs the content derived from Paris on Monday. However, under one interpretation, the speaker continues to describe the event as one occurring ‘on Monday’, and this requires elliptical resolution after abandoning the content of Paris and replacing it with the content of London. To model this, the action sequences triggered by on Monday must be replayed after the substitution of one argument content for another. Schematically, the incremental steps leading to this can be seen in Figure 9.10, where the operation of Regeneration applies after processing has resumed from the point just before the DAG options including Paris are abandoned (characterized as state S4 in the diagram), finally arriving at the revised interpretation at S7. This same backtracking with Regeneration mechanism naturally extends to othercorrections, where the repair is not of one’s own Figure 9.10 Incremental interpretation of self-repair utterance but another’s. It by replaying DS actions in the Context DAG is the fact that actions are first-class citizens in the DS context that allows this straightforward integration of parsing and generation to enable the direct reflection of incremental dialogue-level coordination across interlocutors.10 Nevertheless, as we will see immediately below, such an action-based framework remains capable of expressing even presumed arbitrary formal ‘syntactic’ constraints as those applying to restrict the licensing of various forms of ellipsis, for example, antecedentcontained ellipsis as in (11). (p. 230)

Page 26 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

9.3.7 Licensing in local processing domains: Island restrictions DS takes the grammar to be a model of the joint processing actions of speakers and hearers. From this perspective, standard syntactic constraints are reconceptualized as constraints on processing DAG options, expressed in the same vocabulary, and, as we saw, expected to interact with phenomena like pronominal anaphora and ellipsis. We’ve already seen that standard, presumed sui generis linguistic constraints, like island restrictions as displayed by, for example, relative clauses, can be accounted for through the assumption of self-contained processing domains (see §9.2.3). Processing a relative clause, as we saw in §9.2.3, is taken to involve temporary shifting to a newly introduced tree structure in order to provide the context deemed necessary for proceeding with the processing of the main tree. For this reason, while processing goes on in the new domain, the fixing of an unfixed node initiated in the main tree has to await shifting of the processor back in the main tree (see Kempson et al. 2001; Cann et al. 2005). With this perspective on island constraints, the restrictions on antecedent-contained ellipsis emerge unproblematically: (38)

(39)

Simplifying for reasons of space here, in these cases, the DP is minimally made up of a determiner (every), a nominal (student), and a relative pronoun (who) initiating a relative clause which contains the ellipsis site (had). This relative clause is expected to provide an added restrictor to the variable bound inside the epsilon-term τ,x,Student′(x) which is the content contributed by the DP (see Figure 9.11).11 The DP word sequence has to be processed under the usual principles governing the processing of relative clauses. First, processing the determiner phrase every student involves constructing an abstract λy.τ,y,Student′(y) that will result in binding a variable x introduced by the noun student. Second, a LINKed tree is constructed from the node occupied by x with the requirement to include this variable as one of the arguments of this new LINKed tree in order to furnish it with further restrictions. Because relative pronouns in English appear as ‘left-dislocated’ elements, an unfixed node is introduced initiating the building of this linked tree in order to process the relative pronoun, who. Processing the relative pronoun, through anaphoric means, duly annotates this unfixed node with a second copy of the variable x. It is then the underspecified domination relation associated with the unfixed node, (〈↑*〉Tn(a)), which determines the locality of that initially unfixed node as being within the domain of a single tree. As a result, this then precludes the possibility that this unfixed node could be unified as the argument of a

Page 27 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax further LINKed tree as, by definition, there are no dominance relations holding across linked structures (Figure 9.11): Now, in coming to resolve the metavariable V which the ellipsis site had has contributed inside the relative clause, a sequence of actions from the context has to be retrieved that will result in a sub-tree of Figure 9.11 Successful processing of John Ty(e→(es→ t)). But now the interviewed every student who Bill had choice of which sequence to select is constrained: the selected sequence that will resolve the ellipsis has to conform to the already mentioned independent restriction on unfixed nodes imposed on the partial tree already constructed from the relative pronoun. Hence the variable x can only appear in the local tree and cannot cross further to another LINKed one. This explains the island sensitivity yielding ungrammaticality in (39) where this constraint cannot be satisfied (p. 231)

(Figure 9.12). (p. 232)

Notice the

significance of this result. In other frameworks, island constraints would be articulated within the component of syntax, independent of any interpretation/processing actions, hence not expected to interact with ellipsis/anaphora construal. In DS, however, Figure 9.12 Ungrammaticality of (39) as with syntax defined in impossibility to unify unfixed node with object of interview in second relative clause terms of update of the DAG content, such restrictions, being constraints on local processing domains, are directly predicted also to constrain ellipsis. This is because ellipsis is also modelled as a process of contextual development, therefore required to conform to any restrictions applying independently to such processes.

Page 28 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

9.4 Reflections: Static vs dynamic perspectives compared Stepping back from the details, we can see how the present account of ellipsis compares to standard accounts. Like earlier accounts, the present one also addresses the issues of the nature of the ellipsis site itself, the recovery of its content, and licensing limits on that recoverability. However, the main explanatory tool for addressing standard cases of ellipsis here does not provide descriptions of decontextualized syntax–semantics mappings (derivations, e.g. Kobele 2015). Instead, DS models directly the parser/ generator actions involved in processing the skeletal information projected by some ellipsis sites. Such actions, as in routine anaphora resolution cases, include the ability to introduce, retrieve, and reuse contextual resources. Replacing a competence model embodying propositional knowledge of sentence–proposition mappings, we take grammars instead to model the know-how of interpreting and producing language in context. As such, the grammar in general, and syntax in particular, do not involve sui generis devices or representations; instead, domain-general mechanisms like underspecification and update are the core notions. As in the general domains of perception and action, the conceptualizations afforded by linguistic elements proceed incrementally and predictively, and since participants’ actions in an interaction form each other’s context, actions and conceptualizations can interleave in jointly achieving coordination across sentence boundaries and across turns. From such a perspective, participants do not need to formulate hypotheses as to each other’s intentions, since the goal-directed subpersonal mechanisms of the grammar are adequate for enabling lowlevel coordination without having to explicitly derive propositional descriptions of each other’s actions.

Acknowledgements This chapter is grounded in work done on the ESRC project The Dynamics of Conversational Dialogue (DynDial) (ESRC-RES-062-23-0962), in collaboration with Patrick G. T. Healey, Matthew Purver, Ronnie Cann, Christine Howes, and Wilfried Meyer-Viol, who have each played a central role in the project. We are extremely grateful to them for their input and to many others who have contributed to the way Dynamic Syntax has evolved, in particular, Stergios Chatzikyriakidis, Miriam Bouzouita, Jieun Kiaer, and others such as Graham White and Dov Gabbay, who have supported this work in more general ways. Eshghi’s research in this paper was supported by the EPSRC, under grant number EP/M01553X/1 (BABBLE project) and Hough’s by the DFG-funded Cluster of Excellence, Cognitive Interaction Technology ‘CITEC’ (EXC 277) at Bielefeld University, and the DFG-funded DUEL project (grant SCHL 845/5-1).

Page 29 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

Notes: (1) Note that the claim here is not that a sentential structure or linguistically-based logical form needs to be recovered in order to interpret the case-marked NP (e.g. a verb like rufen in German); instead, as argued later, what is advocated here is that such morphological constraints restrict directly the role of referents within conceptualizations of events/situations (Gregoromichelaki 2006; Hough 2015), the latter conceived as nonlinguistically mediated, geometric-topological structures (as in, e.g., Gärdenfors 2014), or perceptually grounded type classifications (see, e.g., Larsson 2015). There is no necessary one-to-one mapping from such conceptual structures to the words of a particular language. (2) Again the claim here is not that Accusative vs Nominative choice in Greek indicates the presence of linguistically articulated but hidden structure; on the contrary, as in (8)– (9), morphological case, like any other linguistic constraint, restricts the conceptualization of the ongoing or envisaged event/situation, in this instance, the roles of the entities referred to; see also n. 1. Our general claim in this chapter is that exactly this function of linguistic elements is the reason why linguistic structures routinely do not have to be completed up to a sentential level in order to be interpretable. (3) In this exegesis we ignore quantification, merely noting that NPs (DPs) in DS project actions that lead to the construction of arbitrary names denoting a witness of type e, with scope variation expressed as anticipatory statements of term dependency incrementally projected during the construction process (Kempson et al. 2001; Cann et al. 2005). Proper names will be shown here in the abbreviated form John′, Bill′, etc. Tense/aspect/modality specification, defined as sortally restricted event terms, are indicated by Si,j,… (Gregoromichelaki 2006; Cann 2011). (4) This use of the term Merge is a unification step, not to be confused with its use in Minimalism. (5) The arrow linking the two trees depicts the so-called link relation and 〈L−1〉 reads as ‘linked to’. (6) For formal definitions, see Kempson et al. (2015). (7) Simplifying here, with definites taken to contribute anaphoric content needing resolution, the node marked with the box, ‘the doctor’, shows the incorporation of additional information identifying the individual sought to replace the metavariable U, namely, that he is a doctor and named Chorlton, this being the result of the conjunction of the formula annotating the root node of the linked tree, with that of that node—the operation associated standardly with evaluation of linked trees (see §9.2.3). (8) Many languages display anticipatory as well as anaphoric uses, commonly called Pronoun Doubling (Gregoromichelaki 2013).

Page 30 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax (9) The bracketing and symbolization follows Shriberg (1994), where the squarebracketed material to the right of the interruption point + repairs the material to the left of it. (10) For a more detailed account of contextual updates as a result of self-repairs, corrections, clarification interaction and their responses, and other forms of feedback in dialogue, see Eshghi et al. (2015). (11) DS employs the epsilon-calculus (Hilbert and Bernays 1939) with arbitrary names (epsilon-terms) as the content derived from quantified DPs.

Ruth Kempson

Ruth Kempson FBA is Emeritus Professor of Linguistics at King’s College London and Honorary Research Associate with SOAS and the Cognitive Science research unit of QMUL, London. She is the lead developer of the Dynamic Syntax framework, and has a long-term research interest in the interface of syntax and pragmatics. Eleni Gregoromichelaki

Eleni Gregoromichelaki is a Research Fellow at the Philosophy Department, King’s College London and the Cognitive Science Institute, Osnabrueck University. She works within the Dynamic Syntax and Computational Linguistics research groups, exploring analyses of syntactic/semantic natural language phenomena within psycholinguistically informed formalisms. Her principal research interests lie in the language–cognition interface: in particular, the formal/computational and psychological/philosophical implications of various cognitive modeling perspectives on natural language. Arash Eshghi

Arash Eshghi is a researcher in Computational Linguistics at Heriot-Watt University. He did his PhD in Psycholinguistics at Queen Mary, University of London. Ever since, he has been one of the main developers of the Dynamic Syntax computational implementation, and is currently exploring its technological applications, e.g., in building more human-like conversational systems. Julian Hough

Julian Hough is a Lecturer in the Cognitive Science group in the School of Electronic Engineering and Computer Science at Queen Mary, University of London (QMUL). He received his PhD at QMUL before working at Bielefeld University as a post-doc. He researches dialogue modelling and dialogue systems, with a focus on incremental processing and disfluency.

Page 31 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis in Dynamic Syntax

Page 32 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

Oxford Handbooks Online ellipsis in inquisitive semantics   Scott AnderBois The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.10

Abstract and Keywords Work in inquisitive semantics has developed an alternative-rich notion of semantic content which is uniform across questions and assertions. This chapter explores the ramifications of this view for the theory of ellipsis. It examines these issues primarily by focusing on the analysis of a particular ellipsis process in English: sluicing. Empirically, it reviews a number of arguments in favor of an account of sluicing incorporating inquisitive semantics, most notably cases where truth-conditionally equivalent sentences have differential behavior for sluicing. Theoretically, the chapter demonstrates how a theory of ellipsis building on Merchant (2001) but based on an inquisitive semantics helps address this data. Beyond this, it briefly discusses motivations for extending this sort of approach to other ellipsis processes and compares the proposed account with other potential ways of incorporating inquisitive semantics into a theory of ellipsis. Keywords: alternatives, appositives, inquisitive semantics, sluicing, sprouting

Scott AnderBois

10.1 Introduction RECENT years have seen a proliferation of approaches to ellipsis which make crucial reference to the semantic interpretations of ellipsis sites and their antecedents in various ways. At the same time, recent decades have witnessed a sea change within the field of semantics, with many researchers treating sentence meanings not in terms of mere truth conditions, but rather in terms of a broader notion of Context Change Potential (CCP) or Information Exchange Potential. It seems natural, then, to ask the question of how this broader notion of semantic content can be brought to bear in the analysis of ellipsis. In this chapter, we engage this question by focusing on one particular branch of semantic theories with this broader conception of meaning: inquisitive semantics (Groenendijk Page 1 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics 2007; Groenendijk and Roelofsen 2009; AnderBois 2012a inter alia; see Ciardelli et al. 2013 for a recent overview). Inquisitive semantics holds that sentence meanings for both declarative and interrogative sentences consist of (or at least determine) sets of alternative propositions. For the study of ellipsis, then, the hypothesis is that semantic conditions on certain ellipsis processes will (or at least may) make reference to this broader, alternative-rich notion of semantic content rather than to mere truth conditions. The outline for this chapter is as follows: §10.2 introduces inquisitive semantics; §10.3 briefly presents the most fully fleshed-out account of an ellipsis process using inquisitive semantics: AnderBois (2014); §10.4 concludes by addressing the questions of STRUCTURE, RECOVERABILITY, and LICENSING from the perspective of the account in §10.3 and discusses how they might differ under other possible ways of incorporating inquisitive semantics into a theory of ellipsis.

(p. 234)

10.2 Inquisitive semantics

10.2.1 What makes a semantics ‘inquisitive’? The core intuition behind inquisitive semantics is that the meaning/CCP of sentences not only includes truth-conditional information, but also includes the issue(s) that it raises, i.e., its inquisitive content. This has been long recognized, of course, for the CCPs of interrogative sentences. Inquisitive semantics extends this idea to capture the intuition that assertions, especially those containing widest scope disjunctions and indefinites, also raise issues in discourse. For example, (1) introduces two alternatives—‘that it will rain’, ‘that it will snow’—and thereby makes this issue salient in the output context in a way that truth-conditionally equivalent sentences such as ‘it will precipitate’ do not (assuming that rain and snow are the only forms of precipitation). (1)

Inquisitive semantics therefore builds on a number of recent works on disjunction and indefinites (e.g., Kratzer and Shimoyama 2002; Simons 2005; Alonso-Ovalle 2006) in what has been called ‘Hamblin’ or ‘Alternative’ semantics. These works hold that indefinites and disjunctions introduce alternatives into semantic composition. While the name ‘inquisitive semantics’ refers to a family of related semantic/pragmatic theories (see, e.g., Groenendijk 2011 for discussion), there are two fairly consistent ways in which inquisitive semantics differs from Hamblin semantics. First, the two differ in the ways in which alternative-rich meanings are composed and what their formal properties are. These properties have no clear importance for the study of ellipsis, so we refer the reader to Ciardelli et al. (2013) and references therein for detailed discussion. Second, alternativeevoking (or lack thereof) is treated as an aspect of the top-level meaning of assertions and questions alike and therefore of their contribution to discourse as well.1 Page 2 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

10.2.2 Inquisitive semantics across sentence types We turn now to give a concrete version of inquisitive semantics for both questions and assertions. Given our present purposes, the presentation here is necessarily informal (see Groenendijk and Roelofsen 2009 and AnderBois 2012a for more detailed formal presentations). The core formal shift in inquisitive semantics is to treat sentence meanings not as sets of possible worlds (i.e., propositions), as is done classically, but rather as sets of sets of possible worlds (i.e., sets of propositions). This move itself has a precedent in Hamblin’s (1973) semantics for interrogatives, but differs in that expressions other than questions will make (p. 235) use of these richer meanings. In particular, we assume that disjunctions, indefinites, and other forms of existential quantification also contribute alternatives into semantic composition. Following Groenendijk and Roelofsen (2009), we will call a sentence inquisitive if its interpretation contains more than one alternative. The idea, then, is that not only interrogatives, but also declarative sentences may be inquisitive in this sense. Furthermore, if covert existential quantification is also taken to be inquisitive (as we suggest in §10.3.3), the inquisitivity of declaratives will be quite regular. Taking (1) as an example, we illustrate this idea in informal set notation in (2a) and graphically in the diagram (2b). Diagrams like (2b) provide a pictorial representation of the interpretation of the formula in a toy model. We assume a toy model containing four possible worlds (w00, w01, w10, w11), represented visually by the four named circles. The names here correspond to the truth values of two propositions, p and q. In our current example, then, p = ‘that it will rain’ and q = ‘that it will snow’. The boxes, then, represent alternatives in the interpretation of the sentence. (2)

The classical semantics for disjunction is not inquisitive since it produces only a single ‘alternative’, the proposition that it will precipitate. In contrast, an inquisitive semantics for disjunction produces the alternative-rich interpretation with distinct alternatives for the different forms of precipitation mentioned in the sentence. While these two interpretations differ in their inquisitive content, they contain the same informative Page 3 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics content, i.e., determine the same truth conditions. It is the same set of possible worlds which appear in some alternative or other in the interpretation of the two formulas, just structured differently. Inquisitive semantics therefore allows us to distinguish truth-conditionally equivalent formulas on the basis of the alternatives they evoke. To continue with our toy example, then, we assign a sentence with a disjunction like (1) the alternative-rich interpretation on the right of (2), while assigning (3) the different (yet truth-conditionally equivalent) semantic representation on the left of (2). (3)

Extending such a semantics to indefinites and other forms of existential quantification is fairly straightforward at an intuitive level.2 Whereas a disjunction specifies alternatives one (p. 236) by one, an indefinite produces a set containing one alternative per individual in the restrictor set. For example, a sentence like (4) will receive an interpretation with n alternatives, one per each of the n individuals in the interpretation of ‘student’.3 (4)

(5)

In this section, we have introduced the basic conception of sentence meaning in inquisitive semantics with a focus on the two main alternative-evoking elements: disjunction and existential quantifiers such as indefinites. Simple sentences containing these elements make salient a set of alternatives and simultaneously contribute the information that the world of evaluation lies within some alternative(s) in this set. In §10.2.4, we provide a semantics for questions which makes the parallel with indefinites and disjunction explicit, while also capturing the difference between the two classes. Before doing so, however, we introduce a class of operators which interact with the inquisitive component of formulas to which they apply.

10.2.3 Negation and other operators Thus far, we have given an informal introduction to an inquisitive semantics for disjunction and existential quantification. This semantics holds that sentences containing these elements make salient in the discourse a set of alternatives in a way that truthconditionally equivalent sentences may not. While this intuition seems fairly clear for the simple sentences we have looked at thus far, it turns out only to hold for sentences which contain wide-scope disjunctions and existential quantifiers. For example, a disjunction

Page 4 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics within the scope of negation, as in (6a), seems to be no more inquisitive than its nondisjunctive counterpart, (6b). (6)

Moreover, this behavior in fact follows from the way negation is naturally defined in inquisitive semantics (see Roelofsen 2013 and references therein for detailed discussion of its mathematical foundations). Since sentence meanings are sets of alternatives, negation rejects each of these alternatives as in ((7), middle), returning the maximal alternative which does not overlap with any of these. One important consequence of this is that, while it preserves truth conditions, double negation is no longer semantically vacuous since it eliminates the alternative-rich structure of the formula to which it applies, ((7), right). (p. 237)

(7)

It follows, then, that not just any sentence containing an inquisitive element will be inquisitive, but rather those where the inquisitive element takes widest scope. We have illustrated this here for negation, but other operators may also have this property, such as the COMMA operator found in appositive relative clauses. This fact is parallel to the observation (e.g., Chung et al. 1995; Romero 1998) that inner antecedents for sluicing must take wide scope (indeed, licensing of sluicing is often used as a diagnostic for scope of indefinites).4

10.2.4 Questions While inquisitive semantics assigns a more question-like semantics to sentences with wide-scope disjunctions and existential quantifiers, we still need to distinguish these latter elements from questions. The basic approach in inquisitive semantics—at least in matrix clauses—is to differentiate the two in terms of their informative potential. For ease of exposition, we will simply use the term ‘informativity’, though in all cases the relevant Page 5 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics notion is possible informativity rather then actual (see AnderBois 2012a for detailed discussion). The inquisitive semantics for disjunction in the right side of (2) not only introduces a set of alternatives, it also includes the truth-conditional information that some alternative or other holds. That is to say that a declarative with a wide-scope disjunction rules out the possibility that none of the alternatives holds. Intuitively, and perhaps definitionally so, questions are not possibly informative in this way. The literature on inquisitive semantics has seen two ways to cash out this insight, which we can call absolute and relative uninformativity. We can illustate these two approaches for the wh-question in (8), recalling that we keep the pictures to two positive alternatives. The absolute uninformativity approach is exemplified by Groenendijk and Roelofsen (2009), who propose a Q(uestion) operator which adds in the ‘no-one’ alternative, as in the left picture of (9). The second option, due to AnderBois (2012a), is to claim that questions have an existential presupposition and that the alternative set of the (p. 238) question is uninformative only relative to this presupposition. We can indicate this pictorially by shading out the worlds presupposed not to be live options (just world 00 in this case): (8)

(9)

Here, we will follow AnderBois (2014) in adopting this latter option. Ultimately, however, the decision between these two approaches is an empirical one, resting largely on the longstanding question of whether questions contribute existential presuppositions. Beyond the unresolved nature of this question for English, it is of course possible that both options are needed across languages or across question constructions within a single language (e.g., argument vs adjunct wh-questions, wh- vs alternative questions).

Page 6 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

10.3 Sluicing is sensitive to inquisitive content Having introduced inquisitive semantics, we turn now to apply it to the analysis of ellipsis and, in particular, AnderBois’s (2010, 2014) account of both merger and sprouting subtypes of sluicing. Finally, §10.4 will conclude by exploring other possible ways of incorporating the core insights of inquisitive semantics into a theory of ellipsis, drawing on related theories such as dynamic semantics and QUD approaches to discourse.

10.3.1 The need to move beyond truth conditions As a theory of the semantic content of questions and assertions, inquisitive semantics is only of direct relevance to ellipsis to the extent that semantics itself is (or pragmatics which is sensitive to semantics). In principle, it would be consistent for inquisitive semantics to provide an appropriate theory of semantic content, yet for ellipsis to be resolved in a purely syntactic (or LF-syntactic) way (as in, e.g., Sag 1976a; Chung et al. 1995). However, there is a large body of work across many different frameworks arguing that semantics/pragmatics (p. 239) do play a crucial role in ellipsis phenomena (Sag and Hankamer 1984; Hardt 1993; Ginzburg and Sag 2000; Merchant 2001; Culicover and Jackendoff 2005; Chung et al. 2011, among many others). Assuming that this is right in some way, inquisitive semantics naturally raises the question of whether this condition will be sensitive not only to truth-conditional information, but also to the inquisitive aspect of semantic content. There are several different kinds of data which have been argued to support the need for the semantic condition on sluicing to be sensitive to inquisitive content. Except where noted, we focus on data from English, though there is no reason to expect the observations we make to not be more general. In this section, we focus on data from the subtype of sluicing which Chung et al. (1995) dub ‘merger’, i.e., cases where there is an overt ‘inner antecedent’ in the A(ntecedent) clause corresponding to the wh-phrase in the E(lided)-clause. In §10.3.3, we turn to cases of Chung et al.’s (1995) ‘sprouting’, i.e., cases where there is no overt inner antecedent, with the wh-phrase instead corresponding to an implicit argument or adjunct in the A-clause. Perhaps the most fundamental observation supporting the relevance of inquisitive semantics for sluicing is the role played by inquisitive elements as inner antecedents. We see this clearly in the contrasts between the felicitous sluices in (10a), whose antecedents are inquisitive sentences, and the infelicitious ones in (10b), whose antecedents are noninquisitive. Notice that in both cases the corresponding full-clausal versions are felicitous. (10)

Page 7 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics In addition to this basic observation, it is well-known that these elements must take wide scope in order to serve as inner antecedents (e.g., Chung et al. 1995; Romero 1998; Barker 2013). As we will see in a moment, this generalization follows straightforwardly from an inquisitive semantic approach to sluicing. In fact, from the inquisitive semantic perspective, the ability of disjunctions and indefinites to serve as inner antecedents is simply another manifestation of the interrogative-indefinite-disjunction affinity (e.g., Haspelmath 1997; Bhat 2000; Haida 2008). While the privileged role of disjunctions and indefinites is of course quite suggestive, this alone leaves somewhat open the possibility that it is some other aspect of these expressions which is crucial. For example, focusing primarily on indefinites, Chung et al. (1995) argue that it is the logical form of these elements (i.e., the fact that they contribute a variable in the Heimian view, Heim 1982) which is crucial. Merchant (2001) argues that it is the truth conditions following existential closure of the A- and E-clauses which is relevant. One kind of data which are problematic for the former view at least (as Chung et al. 1995 note) are disjunctive inner antecedents where the disjunction is not of arguments, as was the case in (10a), but of entire clauses, as in (11). (11)

(p. 240)

As for the latter view, there are several cases of expressions which are truth-

conditionally equivalent to overt widest-scope indefinites, yet do not license sluicing (i.e., cannot serve as an inner antecedent). Such cases are unexpected if truth conditions are all that the semantic condition cares about, but entirely expected from the view of inquisitive semantics since they plausibly have different inquisitive content. Perhaps the most straightforward case where truth conditions alone prove inadequate are examples like (12) where we see that indefinites and NPIs with double negation and negative quantifiers with single negation (in Standard American English) fail to license sluicing. The would-be A-clauses have counterparts with no negation (e.g., ‘Someone left.’ for (12b)) which are true in the same circumstances but differ in their ability to license sluicing. (12)

Page 8 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics The second case discussed by AnderBois (2014) are indefinites that occur inside appositive (non-restrictive) relative clauses. In contrast to restrictive relative clauses, the content of appositive relative clauses is generally thought of as having sentence-level scope (or perhaps as being in some sense ‘scopeless’), but having no other truthconditional impact. Therefore, a truth-conditional account predicts that inquisitive elements inside of appositive relative clauses ought to readily serve as inner antecedents for sluicing. As seen by the infelicity of (13), based on examples in AnderBois (2014), this prediction is not borne out. (13)

More recently, this generalization has been investigated experimentally by Collins et al. (2014) who argue that such examples can be improved by (i) using a wh-phrase of the form which + NP in the E-clause rather than a bare wh-word like who, and (ii) making the issue raised by the E-clause (or a related one) salient in the discourse preceding the target sentence. Given the space limitations of the present work, we will leave a detailed discussion of these issues to future work.5 However, it is worth noting that both of these manipulations are ones which plausibly raise the salience of the E-issue in the (p. 241) ambient discourse and therefore in our view should not be seen as evidence that the indefinite inside the appositive can in fact serve as the inner antecedent, so much as casting doubt on the logically prior question of whether a linguistic antecedent is strictly necessary in the first place (i.e., whether sluicing is an instance of ‘surface’ or ‘deep’ anaphora in the terms of Hankamer and Sag 1976), as the authors point out. At the same time, the existence of contrasts like (10) make clear that inquisitive elements at least can play a privileged role in licensing sluicing, one which is unexpected for Ginzburg and Sag (2000), as discussed in §10.4. As an anonymous reviewer notes, there is one aspect of Merchant’s (2001) account that one might expect could help capture such data: the requirement that the antecedent be salient. While salience is often not fleshed out in much detail, it seems a priori plausible that double negation and apposition reduce the salience of the antecedent and that this is the reason they impede sluicing. However, we find that other elements typically thought to be sensitive to salience such as pronouns and even VP-ellipsis can readily find an antecedent in these environments as in (14). Therefore, it seems that the effects we are seeing cannot be straightforwardly attributed to salience of the sort relevant for anaphoric processes more generally. (14)

Page 9 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

A third case where truth-conditional equivalence proves insufficient to license sluicing comprises certain cases of bare noun incorporation, such as those discussed in detail by Collins (2013) for Samoan. Collins observes that, unlike other truth-conditionally equivalent indefinite-like expressions in the language, incorporated bare NPs do not license sluicing (nor do they license pronominal anaphora) as seen in (15). Collins (2013) pursues an analysis in the closely related framework of dynamic semantics, though clearly this could potentially be treated as a difference in inquisitivity in the current context. (15)

Thus far, we have seen several cases in which sentences with identical truth conditions to those provided by overt indefinites are unable to license sluicing. We turn now to one further kind of support for the relevance of ‘issues’ to sluicing: the fine-grained patterns of variation across different types of nouns and wh-words investigated by Barros (2013) (see also Dayal and Schwarzschild 2010 and Barker 2013 for related observations). Barros observes that the felicity of sluicing varies depending on complex interactions between the nominal content of the inner antecedent and the properties of the wh-remnant.6 For nouns in the A-clause, Barros claims that the felicity of such sentences depends on whether the noun is a ‘basic-level’ noun or not (in the sense of Brown 1958, Cruse 1977, and others). Basic-level nouns are nouns that have a privileged status tied to their encoding a ‘neutral’ level of specificity (i.e., presumably for non-linguistic reasons). For example, out of the blue it would sound more natural to talk about my ‘cat’ than my ‘mammal’ even though clearly both are equally truthful descriptors for a cat.7 (p. 242)

(16)

Page 10 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics (17)

Beyond variations based on the specificity of the noun itself, Barros argues that the animate wh-word who differs from the inanimate what in allowing more ‘specific’ nouns in the inner antecedent: (18)

Glossing over important details, the basic idea Barros pursues is that wh-words lexically specify a particular level of specificity, that is, they specify an issue whose alternatives have a particular level of granularity. For sluicing to be felicitous, then, the descriptive content of the noun or other inner antecedent material must not be more specific than the level specified by the wh-word. Who and what differ in the level of specificity they specify, leading to the asymmetry seen above. Barker (2013) makes similar observations and proposes the generalization in (19): (19)

We return to these observations in more detail in a moment once we have presented AnderBois’s (2014) account. For now, let us remark that these observations fit naturally in a view where semantic content is alternative-rich. (p. 243)

10.3.2 An account based on symmetric inquisitive entailment Inquisitive semantics is a theory of semantic content, and as such can be implemented within a variety of different theoretical approaches to ellipsis. In this section, we present the most worked-out inquisitive semantic account of an ellipsis process, AnderBois’s (2014) account of sluicing (AnderBois 2010, 2011 present earlier versions of more or less the same approach). AnderBois’s (2014) approach builds on the approach of Merchant (2001), but incorporates the inquisitive semantic conception of semantic content. Given this, we very briefly review the major features of Merchant’s (2001) account. Merchant’s (2001) theory of ellipsis assumes, along with many other authors, that pronounced material in the E-clause (i.e., the wh-remnant), arises from a fully articulated clausal version, as in (20). From this starting point, an additional mechanism specifies the non-pronunciation of the redundant material in this clause, indicated in strikethrough, an approach commonly known as PF-deletion. Page 11 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics (20)

For Merchant, this PF-deletion operation is subject to the condition in (21), which ensures that there is a semantically identical antecedent salient in the surrounding context.8 The existential type-shifting portion of the definition existentially quantifies over missing arguments in order to be able to apply the definition to ellipsis processes which operate over parts of clauses (e.g., existentially quantifying over the subject in order to compute entailment between verb phrases for VP-ellipsis). (21)

(22)

9

(p. 244)

(23)

AnderBois (2014) departs from this basic set-up in two ways, only the latter of which is directly related to inquisitive semantics. First, following Chung (2006) and many subsequent works, the work adopts a ‘hybrid’ approach which supplements the semantic condition with a lexical identity condition in (24). The numeration is the minimalist term for the list (technically a multiset) of lexical items that comprise the sentence. (24)

This lexical identity condition is primarily relevant for handling cases of sprouting (see §10.3.3), but is aimed at capturing the infelicity/ungrammaticality of examples like (25), where the A- and E-clauses clearly have the same truth conditions, yet sluicing is not possible. While it is not impossible to imagine that certain such cases can be captured semantically (e.g., AnderBois 2010 attempts such an account for (25a)), we set aside this Page 12 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics possibility here. It should be noted, however, that even accounts of ellipsis which do not posit silent linguistic material (e.g., accounts of ‘fragment answers’ by Ginzburg and Sag 2000 and Jacobson 2016b) similarly treat such data as arising from a minimal (morpho)syntactic condition regarding the syntactic category of the E-remnant. (25)

The second difference from Merchant (2001) is that the symmetric entailment condition analogous to (21) is formulated with respect to inquisitive semantic contents, rather than just truth conditions. Groenendijk and Roelofsen (2009) give the formal definition for entailment in (26). To unpack this definition a bit, recall first that the interpretations of any given formula will be sets of alternatives. Given this, to see if φ entails ψ, we check to see if each alternative in 〚φ〛 is a subalternative of some alternative or other in 〚ψ〛. Thinking in terms of the pictorial representations above, then, entailment checks to see if each box in the interpretation of φ fits inside some box or other in the interpretation of ψ. Symmetric entailment then, means that the interpretations of two formulas have the same alternatives. (26)

One important thing to note about this definition is that it does not take into account the presuppositions of either formula, just the alternatives in the proposed output state.10 As we saw in §10.2.4, the existential presupposition is the only thing distinguishing the interpretation of an interrogative and a corresponding declarative with (p. 245)

a wide-scope indefinite. This therefore allows for the condition on sluicing to be formulated as follows: (27)

This condition essentially adapts Merchant’s e-Givenness condition to the inquisitive setting (we set aside issues related to F-closure; see n. 9). Given the deep semantic connection between interrogatives and indefinites/disjunctions in inquisitive semantics, however, it can be stated over the entire clause (CP), rather than just the deleted portion, IP. Since the E-clauses in sluicing are necessarily interrogative, the inquisitive entailment Page 13 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics condition in (27) derives what AnderBois (2014) calls the ‘inner antecedent generalization’ in (28). (28)

Beyond this, as argued by Barros (2013), Barker’s ‘answer ban’, (19), follows as a particular case of (27). If the A-clause resolves (or partially resolves) the issue raised by the E-clause, the E-clause by definition does not entail the A-clause and therefore fails the condition in (27). Accounting for all of the data of this sort does require a more finegrained semantics for wh-words than we will give here, so we again refer the reader to Barros (2013) for further details.11 With this in place, we now show how the account tackles basic cases of sluicing as well as the infelicitous cases discussed in §10.3.1. First, let’s consider a basic case where the inner antecedent is an overt indefinite ‘someone’. The inquisitive interpretation of the Aclause someone left will be a set of alternatives of the form ‘x left’, (31, left). The interpretation of the E-clause, who left, consists of the same set of alternatives (31, right). While the E-clause additionally includes a (non-inquisitive) existential presupposition, the entailment condition in (27) ignores this, and so, the symmetric entailment condition is met and (29) is correctly predicted to be felicitous. We can visually verify that (27) is met by looking at the (p. 246) diagrams in (31) and seeing that the alternatives on the two sides are the same. Disjunction behaves the same, differing only in that the specific nature of the antecedent obliges the use of the D-linked which in the E-clause (see Dayal and Schwarzschild 2010; Barros 2013 for discussion of the conditions on which). (29)

(30)

(31)

Page 14 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

Turning to the infelicitous cases, we look first at double negation. In (7), we saw that double negation in inquisitive semantics preserves truth conditions, as one would expect, yet eliminates the inquisitive content of the formula to which it applies. The result is that an attempted sluice in (32) is interpreted as in the picture in (33). Applying the inquisitive entailment condition, we find that the E-clause does entail the A-clause since each of the alternatives of the form ‘that x left’ is a subalternative of some alternative or other in the interpretation of the A-clause (namely, the single alternative ‘that someone or other left’). In the reverse direction, however, the entailment does not hold. Given the single alternative in the A-clause, we cannot find any super-alternative in the E-clause. Since the symmetric entailment condition fails, the account correctly predicts that sluicing will not be possible in this case. (32)

(33)

Page 15 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics Finally, let’s look at the case of the indefinite inside an appositive relative clause. A number of recent works have argued in one form or another that appositive relative clauses have a special discourse status of one sort or another (e.g., Potts 2005; Amaral et al. 2007; Simons et al. 2011; AnderBois et al. 2015). One aspect of this special status is that appositives represent purely informational updates which do not interact with the Questions Under Discussion (QUDs) in any direct way. AnderBois (2014) proposes that in order to capture this aspect of their meaning, appositives ought to be treated as lacking the alternative-rich structure inquisitive semantics assumes for at-issue assertions, and instead be assigned a single classical proposition of type st. Compositionally, this is achieved through the COMMA operator in (34): (p. 247)

(34)

Since inquisitive meanings are captured as non-singleton sets of alternatives, the COMMA operator ensures that the formula to which it applies—as it enters the discourse record—will not be inquisitive regardless of its internal composition. As in the case of double negation, then, appositives deliver truth-conditionally equivalent interpretations, yet lack the alternative-rich meanings needed to meet the symmetric entailment condition. Summing up, we have seen in this section that an account of sluicing based on symmetric entailment defined over inquisitive semantic interpretations captures both the data which motivated Merchant’s (2001) semantic approach (and indeed previous approaches dating back at least to Sag and Hankamer 1984) as well as a number of other sets of data, including several cases where sluicing fails despite truth-conditional equivalence.

10.3.3 Sprouting Having examined cases of ‘merger’ sluicing with overt indefinite or disjunctive inner antecedents, we turn now to cases of so-called ‘sprouting’ where the wh-phrase has no overt correlate in the A-clause. In some cases, of course, there is good reason to believe that, despite the lack of an overt inner antecedent, there is nonetheless an implicit argument present in the A-clause. For example, a large body of literature dating back to Fillmore (1969) holds that apparently intransitive uses of verbs like eat in (35a) include an existential/indefinite implicit argument (see AnderBois 2012b for recent discussion of sluicing and the typology of implicit arguments). Beyond this, there are cases like (35b) where there is clearly an existential entailment, and arguably also (contextually restricted) existential quantification depending on one’s semantics for tense. (35)

Page 16 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics Such data, therefore, have given rise to the claim that in a certain sense, there is no sprouting, but rather that there is always an implicit argument either syntactically (e.g., Fortin 2007b, 2011) or semantically (Merchant 2001). While this approach is potentially viable for the data in (35), Chung (2006) points out that there are other cases where even an existential entailment (p. 248) (let alone a true indefinite) is clearly not present, (36). People can finish projects on their own, and Seth can arrive by car, bike, helicopter, etc. Such cases, therefore, present somewhat the opposite puzzle to what we have seen in §10.3.1 for merger. Here, the A-clause is not even truth-conditionally equivalent to the Eclause, yet sluicing is possible. (36)

For the examples in (36), there is a clear intuition that these instances involve some kind of accommodation. For (36a), this accommodation is pretty easy given the nature of projects. For (36b), the ease of accommodation seemingly depends on various kinds of world knowledge: Is Seth someone who is likely to take the bus? Are there multiple buses which he could have taken? etc. While B’s question in (36c) sounds fairly odd out of the blue, it sounds quite natural in the admittedly unusual context where it is known that Fred only learns languages to impress foreign visitors. Unconstrained, however, accommodation runs the risk of overgenerating and predicting that sprouting should be possible quite generally. However, there are many cases, as in (37), where sprouting remains infelicitous even though it would seem quite plausible given world knowledge. (37)

These sorts of restrictions on sprouting were first noted by Chung et al. (1995), who attribute them specifically to the presence of syntactic islands, claiming that in contrast to the well-known island insensitivity of sluicing more generally, sprouting is sensitive to islands. However, subsequent work by Romero (1998) and Merchant (2001) argues on the basis of pairs like (38) that this difference is not limited to islands (since the non-elliptical control in (38b) is grammatical), and is therefore best captured by appealing to the independently observable narrow scope of implicit existential quantification. For example, whereas an indefinite and clausemate negation ordinarily give rise to a scope ambiguity, an existential implicit argument like the one in ‘Sally didn’t eat’ unambiguously takes

Page 17 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics narrow scope relative to negation (i.e., ‘Sally didn’t eat’ does not have a reading paraphraseable with ‘There is a thing/meal x such that John didn’t eat x’). (38)

For cases like these with an implicit argument, then, the merger account can be extended straightforwardly. Felicitous sprouting as in (35) is possible because the A-clause has an indefinite implicit argument and therefore has an interpretation which is inquisitive. In examples like (37a) and (38a), there is still an indefinite implicit argument, but one which cannot take widest scope. Therefore, the whole sentence’s interpretation is noninquisitive and sluicing is correctly predicted to be infelicitous. What, then, about examples like (36) in which we have seen following Chung (2006) that no implicit argument or even existential entailment is present? Building on the above intuition that at least some cases of sprouting involve accommodation of some sort, AnderBois (2014) proposes an account which is partially semantic and partially (p. 249)

pragmatic. On the semantic side, the proposal extends the inquisitivity that we have thus far associated with overt indefinites and disjunctions to existential quantification quite generally, including covert quantification over neo-Davidsonian eventuality arguments. The meaning of a simple sentence like ‘John left’ not only includes the information that there is some event or other which is a leaving event and of which John is the agent, but also the issue of which event it is that satisfies these requirements. On the pragmatic side, then, the account claims that sluicing is felicitous to the extent that the alternatives in the interrogative E-clause covary with that of the A-clause, an accommodation process AnderBois dubs ‘issue-bridging’. The rest of this section spells out both parts of this proposal a bit more, though we refer the reader to AnderBois (2014) for further details. A central notion in inquisitive semantics is the idea that the kind of indeterminacy we find in indefinites and disjunctions is intimately related (and in some cases compositionally related) to the inquisitivity we find in questions. In both cases, a set of alternatives is made salient, leaving the issue of which alternative(s) in fact hold as at least a safe potential topic for future conversation. For example, the sentence in (39a) introduces a set of alternatives in (39c), makes salient the issue of which of these in fact hold, and conveys the information that at least one does. (39)

Page 18 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

Here, we extend this idea beyond overt indefinites to existential quantification over covert arguments, in this case, the neo-Davidsonian event(uality) argument. As is clear in (40), the proposed semantics for the covert existential is formally entirely parallel to what we have seen in (39). A sentence like (40a) introduces a set of alternatives, (40c), makes salient the issue of which of these alternatives in fact hold, and conveys the information that at least one does. (40)

The issue it makes salient is, however, a somewhat odd one, paraphraseable as ‘Which event is an event of Seth leaving?’. The apparent oddity of this issue, however, is due not to anything about the inquisitive quantification itself, but rather the ontological status of events in the first place (as discussed, for example, by Parsons 1990). Although it is standard to take events to be things in the actual world in more or less the same way that individuals are, it is far less intuitive to do so for events. (p. 250)

Page 19 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics With this semantics in place, we turn now to the pragmatic part of the story. For concreteness, we will work with the example in (36b), repeated as (41). The semantics above holds that the A-clause makes salient a quite fine-grained issue of the form ‘which event?’. The E-clause, on the other hand, makes salient a more coarse-grained issue about some aspect of the event in question, in this case its manner. The inquisitive entailment condition is therefore not met, given the difference in the granularity of these two issues.12 (41)

While these two issues are not identical, the claim is that they are sufficiently similar that the E-clause can be accommodated. AnderBois (2014) calls this accommodation process ‘issue-bridging’, on analogy with bridging definite descriptions like that in (42). The existence of a driver is not simply accommodated directly, but rather by virtue of a salient relationship with something whose existence and discourse salience are already established, a bus. (42)

Rather than bridging to an individual, however, indirect sprouting involves bridging to an issue introduced in the A-clause. Just as a driver is typically an aspect of a bus, times, locations, manners, etc. are typically aspects of events (see also Barros 2014 for a related approach which works directly with these categories rather than events). Concretely, then, the prediction is that sprouting (and indeed sluicing more generally) should be subject to the condition in (43). One important feature of the account to note is that it relies crucially on the presence of inquisitive material in the A-clause, and thus avoids overgenerating and allowing examples like (37) since the event quantification in question does not take wide scope.13 (43)

This section has shown two ways to extend the inquisitive semantic account of sluicing to sprouting—one for cases where an indefinite implicit argument is present, and one for cases where no such argument is found. Both accounts rely on independently known scopal (p. 251) properties of implicit existential quantification to help constrain the account, thus deriving the asymmetries between merger and sprouting first discussed by Chung et al. (1995).

Page 20 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics

10.4 Conclusions: Structure, recoverability, and licensing In this chapter, we have informally introduced inquisitive semantics, reviewed the most fleshed-out inquisitive semantic account of ellipsis to date—AnderBois’s (2014) work on sluicing—and explored various kinds of data consistent with this view. We conclude here by considering how three major questions in the theory of ellipsis—STRUCTURE, RECOVERABILITY, and LICENSING—are answered under this account as well as briefly considering other potential ways of incorporating inquisitive semantics into the theory of ellipsis. By STRUCTURE, we mean the question of what syntactic structure, if any, is found within the ellipsis site itself. RECOVERABILITY refers to the way in which the ellipsis site’s interpretation is arrived at. LICENSING covers any additional constraints or conditions on ellipsis that are not clearly part of the latter two categories. Since inquisitive semantics is a theory of semantic content rather than of the interfaces between semantics and syntax or phonology, it in principle need not impose any requirements on the theory of ellipsis. That said, for inquisitive semantics to play a role in accounting for a given ellipsis process, the condition on RECOVERABILITY must be at least partially semantic in nature. Inquisitive semantics locates alternatives in the interpretation itself, rather than in the LF, and so even an LF-syntactic approach to recoverability such as Chung et al. (1995) will not suffice. A complete theory of ellipsis must of course address the other major questions as well, which we do presently both for the main account described here, AnderBois (2014), as well as briefly discussing how these answers might change under other potential ways of incorporating inquisitive semantics into a more comprehensive theory of ellipsis. Since the account of sluicing in §10.3 builds off of Merchant (2001), it addresses these major issues in largely similar ways. On the question of STRUCTURE, both accounts posit silent linguistic material with ellipsis consisting of PF-deletion. Typically, we assume that the deleted material is full interrogative clauses, though nothing in the analysis rules out other underlying structures, such as clefts, provided that they satisfy the relevant identity conditions (see, e.g., Barros 2014 for a closely related but non-inquisitive approach making use of this option). On the question of RECOVERABILITY, the primary condition is a semantic one: symmetric entailment between A- and E-clauses with entailment crucially being defined over inquisitive semantic representations rather than just truth conditions. Beyond this, we have departed slightly from Merchant (2001), along with Chung (2006) and others in supplementing this semantic condition with a minimal lexico-syntactic one to handle certain issues that arise in sprouting. Ultimately, then, the approach to RECOVERABILITY in §10.3 is a hybrid one, in line with recent works in a variety of otherwise quite different approaches (e.g., Ginzburg and Sag 2000; Chung 2006; Jacobson 2016b).

Page 21 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics Finally, for LICENSING, it does not seem that inquisitive semantics imposes any particular constraints on possible accounts. Given the close parallels with Merchant (2001), we refer (p. 252) the reader to chapter 2 in that book (and Lobeck’s 1995 work cited therein) for discussion of this issue. One place where inquisitive semantics does help shape the range of answers to the LICENSING question (or perhaps RECOVERABILITY) is that it provides a semantic account to certain kinds of cases that one might have thought were due to syntactic or other form-based constraints (e.g., the case of double negation in (7)). However, we again stress that the way the account in §10.3 answers the question of STRUCTURE is not intrinsic to inquisitive semantics per se, and that there are in principle many different frameworks for understanding ellipsis in which inquisitive semantics could be incorporated. Given the tight connection between inquisitive semantic issues and Questions Under Discussion (QUDs), one obvious candidate would be to build on Ginzburg and Sag’s (2000) QUD-based approach. In a nutshell, their approach is a structure-free one which, beyond a minimal condition on form referring to a salient utterance (SAL-UTT), fills in the interpretation of the wh-phrase anaphorically from the maximal QUD (MAX-QUD). On a classical semantics for indefinites and disjunctions, however, this approach offers no clear way to explain the privileged role that these elements play in sluicing (in fact, Ginzburg and Sag 2000 briefly argue against this claim on p. 321). Adding inquisitive semantics to this picture, however, these elements conventionally make salient a possible QUD, thus explaining their privileged role. While such an account is in many ways an attractive one, further work on the various linguistic and non-linguistic ways in which QUDs arise in discourse is needed to make it viable. For example, (44) is a case where a clear QUD is established contextually, it would seem, and yet sluicing appears to be quite bad. (44)

Another alternative would be to seize upon the deep parallels between inquisitive semantics and dynamic semantics and draw upon work that treats ellipsis as discourse reference of a special sort, such as Hardt (1993) (possibly supplemented with a lexicosyntactic condition of some sort). One challenge for such an approach, however, is that individual discourse reference does not exhibit the same interactions with double negation and appositives as does sluicing, as discussed by AnderBois (2014). Nonetheless, such an approach is in principle possible and, again, would give quite different answers to the question of STRUCTURE at least and possibly RECOVERABILITY and LICENSING as well.

Page 22 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics To summarize, inquisitive semantics proposes that the context change potential of sentences containing disjunctions, indefinites, and other existential quantification is ‘alternative-rich’. We expect therefore that we might find ramifications of this richer notion of semantic content in various areas of the grammar. We hope to have shown that this is so for ellipsis and to have shown one way of modeling such effects by focusing primarily on a particular ellipsis process: sluicing. Given the central role of interrogatives, indefinites, and disjunction in sluicing, these issues are naturally most salient here. However, it should be clear that once we adopt alternative-rich sentence meanings, the question arises of whether other ellipsis processes may similarly require reference to inquisitive semantic representations in some form.

Notes: (1) It should be noted that while this was a driving motivation in many early works in inquisitive semantics (e.g., Groenendijk 2007; Groenendijk and Roelofsen 2009; Ciardelli 2009), some more recent works have used the moniker ‘inquisitive semantics’, yet lack this second property (e.g., Farkas and Roelofsen, forthcoming, propose that declarative clauses include a closure operator eliminating this possibility). (2) Formally, it has been argued that this extension is in fact somewhat more fraught in the case of models with non-finite domains. See Ciardelli (2009) for detailed discussion. (3) While in principle they are equally applicable, pictorial representations like (2b) become unhelpful for larger sets of alternatives. (4) While other scope-taking elements such as conjunction and universal quantification are defined by most authors in ways that allow their alternatives to be ‘passed up’ the composition, one could alternatively define them in ways that do not have this effect. The choice ultimately depends on how one wishes to handle scope-taking more generally and we therefore set aside this concern here. (5) We would also note that it is not entirely clear how to interpret felicity judgments for these data. Reading these sentences, it is relatively easy to figure out what such examples were supposed to have meant after the fact. However, it still seems somewhat unlikely in our opinion that speakers in fact produce such sentences frequently and possible that they have difficulty processing them when encountering them in natural speech. Therefore, it seems quite possible to imagine an analysis of this gradient pattern of judgments in which the grammar of appositives and sluicing does not generate such sentences, but rich context and other ‘repair mechanisms’ allow speakers to figure out what they are to have meant when encountered in experimental settings. In any case, further corpus and experimental work is needed in this area. (6) Barros (2013) also notes a further interaction with the presence or absence of exactly in the E-clause which we will set aside here.

Page 23 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics (7) See Barros (2013) and references therein for further discussion and independent linguistic diagnostics of these categories. (8) This condition itself builds on the one proposed by Schwarzschild (1999) for deaccenting. It is worth noting therefore that, as observed by AnderBois (2014), deaccenting appears to be sensitive only to truth-conditions rather than inquisitive content as well. Consider, for example, the contrast between (12) and the following (underlining indicates deaccenting): ((i))

(9) The focus-closure (F-clo) part of the definition is needed primarily to handle two kinds of sluices which we will not discuss here: ones where the wh-phrase contains else, as in (i), and so-called ‘contrast’ sluices like (ii), both examples from Merchant (2001). ((i))

((ii))

We set aside these cases here, while not denying their importance. (10) The idea that such a notion of entailment is relevant for natural language has been independently proposed in the literature on NPI-licensing (von Fintel 1999 et seq.), where it has been dubbed ‘Strawson entailment’. (11) Barker (2013), it should be noted, does briefly address Barros’s (2013) claim that the answer ban follows from the inquisitive entailment condition, but expresses skepticism for two reasons. First, Barker apparently was familiar only with AnderBois (2010), which only analyzed cases of merger, whereas AnderBois (2011) and (2014) also address sprouting (see §10.3.3). Second, Barker notes that Ciardelli et al. (2009) propose an inquisitive-semantics-based account of epistemic possibility modals like might which do not license sluicing: ((i))

However, while it is true that Ciardelli et al.’s (2009) account of might is couched in a version of inquisitive semantics, their approach does not actually claim that ‘John might leave’ has the same semantics as ‘John will leave or John won’t leave’, and in fact does not claim that ‘John might leave’ is inquisitive at all (but instead defines a new category: Page 24 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

ellipsis in inquisitive semantics ‘attentive’). Therefore, even taking this analysis of might at face value, we do not predict (i) to be well-formed. (12) In this particular example, there is a second apparent lack of symmetric entailment: the fact that the E-clause in this example has a presupposition not found in the A-clause. However, this concern is already taken care of given that we have adopted a notion of Strawson entailment. (13) The account raises a number of issues related to scope-taking which we cannot address here for lack of space. See Charlow (2014) for recent discussion of scope-taking and sluicing in a closely related semantics.

Scott AnderBois

Scott AnderBois is Assistant Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. His primary research focus is on the ways in which utterances interact with the discourse “scoreboard,” with a particular focus on to what extent and in what ways these interactions are encoded conventionally as part of the sentence meanings, as opposed to arising from pragmatic reasoning. Specific topics of interest include apposition, discourse particles, disjunction, ellipsis, evidentiality, indefiniteness, mirativity, topics, and questions. He has explored these issues through primary fieldwork including Yucatec Maya, A’ingae, English, and Tagalog.

Page 25 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics

Oxford Handbooks Online Ellipsis and Psycholinguistics   Lyn Frazier The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.11

Abstract and Keywords Psycholinguistic studies bear on the grammar of ellipsis in many ways. They can shore up the empirical base of linguistics, through careful testing of multiple participants and items, and they permit testing of subtle judgments involving interactions. Processing evidence affords new types of arguments, e.g. based on syntactic indices being copied into an elided constituent despite semantic anomaly, and it highlights the fact that linguistic intuitions—judgments of acceptability—are not judgments of grammaticality, raising the possibility that the acceptability of certain ellipsis sentences may be best explained by (repair) operations of the parser, not by the grammar. Experimental results are brought to bear on the issues of the existence of syntactic structure in the ellipsis site, on the nature of the recoverability condition, and on licensing. Keywords: ellipsis structure, syntactic matching, acceptable ungrammaticality, replacives, stripping, comparatives, gapping, verb phrase ellipsis, sluicing, fragment answers

Lyn Frazier

11.1 Introduction THIS chapter begins by looking at a range of ellipsis structures and asking if there are any empirical problems for a naive view of ellipsis processing—one where a structural ‘shell’ is copied from the antecedent clause and then remnants (overt constituents in the ellipsis clause) are lined up with the most similar constituents in the antecedent clause (section 11.1.1). The chapter then turns to the question of whether psycholinguistic evidence supports the claim that there is syntactic structure in the ellipsis site (section 11.2), what psycholinguistic evidence says about recoverability and the nature of the relation between an antecedent and the ellipsis site (section 11.3), and licensing (section 11.4). Section 11.5 concludes.

Page 1 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics Before embarking on substantive discussion, let’s consider what psycholinguistic studies can tell us about the grammar of ellipsis. One traditional approach to linguistics grants intuitions a privileged status, as if linguistic intuitions directly reflect grammatical knowledge and only grammatical knowledge. A grammar, on this view, is just a systematic approach to characterizing linguistic intuitions. On this view, intuitions really are intuitions of grammaticality, perhaps with the caveat that some unprocessable sentences may appear to be ungrammatical (‘bad’) even though they have grammatical structures. An alternative view is that linguistic judgments are judgments of goodness and, like other judgments humans make, they are influenced by a variety of factors. These include the grammar, processing complexity, social factors, and the like. Empirical evidence favors this view. Marginal sentences are more likely to be rejected if the judge is in front of a mirror, because mirrors make people self-conscious, and self-conscious people are more conservative (Carroll et al. 1981; Nagata 1989). Complex sentences are rated as less good than less complex sentences, even when they are uncontroversially grammatical. The fact that processing influences acceptability judgments is perhaps brought home by two examples: ungrammatical sentences are rated more acceptable if they have a temporarily grammatical analysis than if they do not even temporarily permit a grammatical analysis (Fanselow and Frisch 2006); and fully grammatical sentences containing a licensed polarity item are rated less acceptable if they contain a temporary analysis on which the polarity item is unlicensed (p. 254) than if they do not (Frazier and Clifton 2011). Such facts and many others are difficult to reconcile with the view that intuitions or acceptability ratings reflect only the grammar. On this second view, which will be assumed throughout the present chapter, the notion of ‘grammatical’ or ‘ungrammatical’ is a theoretical notion. Linguistic intuitions are judgments about the goodness of a sentence. It is up to the theorist to determine whether utterances that are judged bad do not conform to the grammar in some respect or whether they are difficult to process, for example. Examples that are judged good may be grammatical or they may just be something a native speaker might say or hear, i.e., an utterance that conforms to most of the constraints of the grammar, is transparent in terms of the message–form mapping, and thus might sound good in certain circumstances despite being ungrammatical. So intuitions of acceptability do not wear their interpretation on their sleeve. It is the overall theory of language (grammar plus processor) that must account for data—whether the data be intuitions/judgments or processing time data. This view makes processing studies germane even for those who might be interested only in grammar: without an empirically motivated theory of processing, it will be unclear whether it is the grammar or the processor that should explain some fact or some contrast in the acceptability status of sentences. It is less controversial, perhaps obvious, that psycholinguistic data, in the form of experimentally collected acceptability judgments, may contribute to linguistics by showing that some linguistic generalization reliably holds across a range of items and for a group of speakers. It may be less obvious, but I suspect it is true, that typically it is only experimentally collected acceptability judgments that permit interactions to be tested. Page 2 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics For example, if one were to claim that d-linked phrases are not subject to island constraints then one would be claiming that the difference between d-linked and non-dlinked interrogatives is larger in the context of an extraction from an island than in the case of an extraction not out of an island (and not just that d-linking helps extractions out of islands). But this sort of ‘interaction’ judgment, where the size of the difference between the judged acceptability of one pair of sentences is predicted to be larger than the size of the difference between another pair of sentences, is not easy to make with confidence intuitively, though it is precisely the relevant comparison. Of course assessing this sort of interaction is exactly what experiments with factorial designs can do. In sum, psycholinguistic evidence about ellipsis and processing ellipsis can guide us to better theories of the grammar of ellipsis. Experimentally collected data can shore up the empirical basis for acceptability judgments, especially those that in reality involve predicted interactions, and data about processing ellipsis can offer a more nuanced view of ellipsis in part by helping to determine what should be accounted for in the grammar and what in the processor. Keeping open the possibility that the processor may bear some of the burden of accounting for acceptability judgments will become particularly important in 11.3.3 in the discussion of recycling or repair theories based on the proposal that inputs that are likely syntactic blends may be repaired in comprehension.

11.1.1 Might a naive view of ellipsis processing suffice? Many types of ellipsis exist. Carlson (2002) investigated replacives (1), stripping (2), and comparatives (3) and found strong parallelism effects for all of them. In what follows, these (p. 255) studies will be reviewed and then used to evaluate a simple approach to ellipsis processing where the abstract syntactic structure of the antecedent clause is copied to create a syntactic ‘shell’ into which remnants are mapped, with the similarity between remnants and constituents in the antecedent clause used by the processor to carry out the mapping. A copied ‘shell’ might also apply in the processing of conjoined clauses in general, including those without ellipsis, explaining the processing advantage (priming) observed in processing structurally parallel conjoined clauses (e.g., Frazier et al. 1984). As we will see, this simple approach to processing ellipsis and conjunction would account for the bulk of the studies to be reviewed in this section, but it falls down dramatically in accounting for one robust effect. Carlson (2002) reported the results of written studies in which participants were given a sentence and two paraphrases. They were asked to indicate which paraphrase corresponded to their interpretation of the sentence. The choice of an associate for the remnant (the phrase with which the remnant contrasts) was influenced by the lexical parallelism between remnant and associate in terms of the type of the DP, e.g., whether it was a name or a definite description. For example, participants chose Maude, the matrix subject as associate 63 percent of the time when the correlate was Marjorie in (1a), but only chose Maude as the associate 22 percent of the time when the remnant was a fireman in (1b). In (1)–(3), the percentages are the percent subject response.

Page 3 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics (1)

(2)

(3)

For replacives (1), stripping (2), and comparatives (3), a strong lexical parallelism effect was found: readers chose a phrase as an associate more often when it was the same type of DP as the remnant than when it was not. Carlson further showed that placing a prominent pitch accent on a potential associate in auditory interpretation studies also increased the probability of choosing that phrase as the associate. For example, accenting Maude in (1) increased the probability of choosing Maude as associate relative to a condition where Maude did not receive a prominent pitch accent.1 (p. 256)

Turning to gapping (4a),2 Carlson (2001) conducted a written judgment study

where participants read sentences ambiguous between a gapping analysis and a conjoined VP analysis. Participants were asked to read each sentence, choose the best paraphrase for it, rate the difficulty of the sentence, and finally indicate whether it could have any other meaning. Testing examples like those in (4), she found that a gapping analysis (Josh visited the office during the vacation and Sarah visited the office during the week) was not chosen often unless the potential subject remnant in the gapped clause was lexically parallel only to the subject of the antecedent clause.3 Even in this case (4a), only a third of the responses indicated a gapped as opposed to a conjoined VP interpretation. (4)

Page 4 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics

In short, the gapping analysis was disfavored strongly, though promoted to some extent by similarity between the subject of the antecedent clause and the potential subject remnant. In addition to the interpretation data presented under (4), Carlson also found that gapped structures were rated as being more ambiguous and more difficult than their ungapped alternatives. As with replacives, stripping, and comparatives, placement of a prominent pitch accent on the subject of the antecedent clause and the potential subject remnant also promoted the gapping analysis in auditory interpretation studies. Carlson (2001) placed a prominent pitch accent on the subject or object of the antecedent clause in sentences like (5). (5)

She found significantly more gapping analyses (44 percent) in (5a) than in (5b), where only 27 percent of the responses indicated a gapping analysis. But even a prosody favoring the gapping analysis, (5a), resulted in a majority of non-gapping analyses. Imagine that perceivers copied the syntactic structure of the first conjunct of a conjoined clause sentence, creating a predicted clausal template or shell, and then lined up the remnant(s) of the second conjunct with corresponding lexically parallel or focused constituents in the first conjunct, as illustrated in (6). (p. 257)

(6)

In effect the copied structure would permit constituents in the ellipsis clause to be mapped to empty counterparts of the corresponding constituents in the antecedent clause, e.g., with to the doctor mapped into the empty constituent under to the dentist. Page 5 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics This would permit very simple essentially predictive processing of the ellipsis clause. The existence of strong parallelism effects would be explained. Assuming that a remnant and its associate contrast and thus should be accented, the effect of prominent pitch accents would follow naturally. Assuming that clausal conjuncts always give rise to syntactic shells, a clause containing ellipsis would be predicted to be easier to process than a corresponding unelided clause: fewer constituents need to be parsed. Clausal conjunction would be predicted, correctly, to show strong parallelism effects, i.e., an advantage when the second conjunct is lexically and structurally parallel to the first conjunct (Frazier et al. 1984). In terms of processing, the problem with this view is that it makes the wrong empirical predictions. Consider (4c) Theresa introduced a new policeman to the mayor and Susan to the councilmen. Susan is more similar to Theresa than to other arguments and to the councilmen is most similar to to the mayor. So the processor should map the constituents in the ellipsis/conjoined clause to the positions beneath the most similar constituents, resulting in the gapping analysis (Susan introduced a new policeman to the councilman) of (4c). But the gapping analysis was chosen only 19 percent of the time. The strong dispreference for gapping should not exist on this naive view of ellipsis. Thus, Carlson’s evidence seems to disconfirm the predictions of this simple and in some ways attractive approach to ellipsis. In (4a) then, it must be something like the structural economy of the conjoined VP analysis that results in the preference for the ungapped structural analysis (see discussion of the results of Carlson, Dickey, and Kennedy 2005 in Chapter 18 for further support for the assumption that structural economy is at issue). What does account for the results presented in this section? Carlson (2002) argued that the effects that she observed suggest that the existence of similarities/parallelism between corresponding constituents of two clauses breeds a desire for further similarities so that constituents that are similar to each other preferentially are assigned similar phrase-external syntax and constituents in corresponding syntactic positions preferentially have similar properties to each other phrase-internally. In short, parallelism breeds parallelism. Before turning to a discussion of structure in the ellipsis site, it should be noted that the types of ellipsis discussed so far, gapping (7b), stripping (8b), and comparatives (9b), do not easily embed: (7)

(8)

Page 6 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics (p. 258)

(9)

By contrast, the types of ellipsis that have been most studied, such as VP ellipsis (10) or sluicing (11), readily embed: (10)

(11)

Frazier and Clifton (2000) suggested that the two types of ellipsis, those that readily embed and those that do not, may be processed differently. They argued that a cost-free copying operation is available for the latter, allowing an antecedent to be copied into an elided constituent. As Carlson argued, the similarity of the remnant and its associate seems to be critical in the processing of ellipsis that doesn’t embed, such as the examples discussed in this section. By contrast, it will be argued in 11.4 that information structure notions are critical in processing the types that do freely embed, such as VP ellipsis, e.g., Sam stumbled but Bill didn’t, and sluicing, e.g., Someone left but I don’t know who. Further, it seems to matter for the latter types of ellipsis whether the elided constituent is part of asserted material (e.g., in a conjoined predicate) or is presupposed (in temporal or comparative clauses) (see discussion of Moulton 2007, 2013 in section 11.4.3). To what extent these differences between types of ellipsis, those that embed and those that don’t, are deep, or even will hold up descriptively, is not evident at present.

11.2 The ellipsis site 11.2.1 Syntactic structure in the ellipsis site Theories of ellipsis differ in whether they claim that there is syntactic structure in the ellipsis site. One possibility is that an ellipsis site has no internal structure (Culicover and Jackendoff 2005), on a par with many accounts of pronouns. Another possibility is that the ellipsis site contains structure that is fully articulated but not pronounced (Merchant 2001). Psycholinguistic evidence relevant to this debate is discussed in section 11.2.2. Page 7 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics Examples will be given of various types of evidence or arguments for the existence of syntactic structure in the ellipsis site. There is a class of findings showing that syntactic mismatch effects are stronger, or temporally different, for ellipsis than for anaphora (section 11.2.2.1), and a class where syntactic matching of antecedent and remnant does not suffice because the derivation of the remnant would violate a constraint on movement such as pied-piping in German, or clause fronting in English (section 11.2.2.2). Violations of various types of island constraints and c-command violations are discussed (sections 11.2.2.3–11.2.2.4). We (p. 259) conclude the section with evidence that a syntactic index is copied from an antecedent even when the copied index produces a semantic anomaly (section 11.2.2.5), and suggestive evidence that even phonological information copied from the antecedent may be available in the ellipsis site (11.2.2.6). In addition to the comprehension-based arguments and evidence that will be presented,4 investigators have also argued for the existence of syntactic structure in the ellipsis site in production studies: structure in the elided constituent primes the structure used to describe a picture (see Xiang, Grove, and Merchant 2014, Hogue 2014, and also Cai, Pickering, and Sturt 2013, for the failure to find priming from “VP ellipsis” in Mandarin Chinese).

11.2.2 Empirical evidence for syntactic structure in the ellipsis site 11.2.2.1 Larger penalties of syntactic mismatch for ellipsis than anaphora Hankamer and Sag (1976) contrasted what they called “deep anaphora” and “surface anaphora.” Deep anaphors take a conceptual antecedent; surface anaphors require a linguistic antecedent, which must be syntactically parallel to the ellipsis site. Tanenhaus and Carlson (1990) tested this theory in a series of “makes sense” judgment experiments. Participants were asked to read a context sentence, labeled S1 in (12), read a second sentence (S2), and judge whether S2 made sense in context. In one study, they manipulated voice (passive vs active) and anaphor type: VP ellipsis (a surface “anaphor”) and do it, a deep anaphor, as illustrated in (12). (See also Ward, Sproat, and McKoon 1991.) (12)

What they found in this experiment and several other closely related studies was that “makes sense” judgments (percentage yes) reflected the critical interaction, with the lowest acceptability for surface anaphors in the non-parallel condition. However, response times showed a penalty for non-parallelism for both surface and deep anaphors. Tanenhaus and Carlson explained the response time advantage for parallel structures in deep anaphora in terms of the ease of retrieving an antecedent when it is parallel. Even if Page 8 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics a deep anaphor doesn’t require a linguistic antecedent, identifying a conceptual antecedent may be easiest when a linguistic antecedent is readily identifiable. Since the time of Tanenhaus and Carlson’s pioneering work, it has generally been assumed that ellipsis is not the same as anaphora. There is considerable further work comparing ellipsis and anaphora and finding stronger syntactic mismatch penalties for ellipsis; see Ward, Sproat, and McKoon 1991, for example. Roberts, Matsuo, and Duffield (2013) used eye-tracking during reading, and found different effects of syntactic mismatch for VP ellipsis than VP anaphora and also a different time course for the two types of dependencies. Though the existence of syntactic mismatch penalties for ellipsis is relatively uncontroversial, not everyone agrees that they should be attributed to the grammar. See discussion in 11.3.3 (and discussion of Dickey and Bunger 2011 in Chapter 18 of the present volume). (p. 260)

11.2.2.2 Syntactically matching remnants that violate pied-piping or clause movement conditions One way to test for the existence of syntactic structure in the ellipsis site is to examine whether conditions on movement apply within the ellipsis site. Merchant (2004a) argued that fragment answers to questions involve movement of the answer/new information to a focus projection high in the syntactic tree, and ellipsis of the remainder of the tree. The analysis predicts that (13Bb) is ungrammatical as an answer to (13A). It is ungrammatical for precisely the same reason that (14b) is ungrammatical: movement of a clause requires the clause to have an overt complementizer. (13)

(14)

The analysis explains why only constituents are acceptable fragment answers. It is because only constituents move. Similar to the explanation for the contrast in (13), the account also predicts that in a language where preposition stranding is not allowed, DP answers corresponding to the complement to a preposition will not be allowed (though various potential counterexamples have been noted, e.g., in Polish—see Nykiel 2013). Merchant et al. (2013) confirmed the prediction experimentally. German is a language that does not permit preposition stranding. Hence, it is predicted that (15B) should be a fully acceptable fragment answer, but (16B) should not (since the DP would need to move to a focus projection, stranding a preposition.) In a written German acceptability rating Page 9 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics study using a seven-point scale, testing forty participants at the University of Potsdam, the prediction was confirmed. Fragment answers like (15B) received a mean rating of 5.99 whereas the DP fragment answers like (16B) received a significantly lower rating of 4.76. (Note that it is the difference in ratings that can be interpreted in this sort of study. The absolute numeric value means little apart from the relation of the experimental sentences to the filler sentences.) (p. 261)

(15)

(16)

The result is striking because both answers are constituents, both provide the information requested, and both syntactically match the question. Merchant et al. (2013) also tested the prediction that a clausal fragment answer requires a complementizer. In a written acceptability rating study in English using a scale of 1…5 (=good), the predicted grammatical form received a rating of 4.25, which was significantly higher than the rating for the no-complementizer counterpart (which received a mean of 3.73). (17)

In an eye movement recording study of items with the structure of (17), a reading time penalty for no-complementizer fragment answers was found in first-pass data and number of fixations on the moved clause, as well as in later measures (Clifton and Frazier, in progress).5 This evidence indicates that penalties for ellipsis involving illicit movement to the focus phrase show a processing cost online, and not only in deliberative judgments or offline intuitions. (p. 262) Page 10 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics 11.2.2.3 Islands are obeyed inside the ellipsis site Another way to test for syntactic structure in the ellipsis site is to check whether islands are obeyed inside elided constituents. Frazier and Clifton (2005) reported several written acceptability judgment studies of island violations in sluicing structures including relative-clause islands, subject islands, and adjunct islands (18d). They found significant penalties, drops in acceptability, for island violations inside an elided constituent, in a speeded acceptability judgment task. Interestingly, the island violation imposed a significantly greater cost when sprouting (Chung, Ladusaw, and McCloskey 1995) was required, i.e., in examples like John ate but I don’t know what, where no copied structure is available to house the variable that the interrogative must bind.6 Frazier and Clifton (2005) tested adjunct island violations (18d) in a visual “got it” task, where participants read the sentence and then indicate whether it is or is not a good sentence of their language. Crucially, there was a significant drop in acceptability both for an island violation in an overt constituent ((18b) vs (18a)) and in an elided constituent ((18d) vs (18c)). However, the size of the island violation penalty ((18d) vs (18c)) was smaller when the violation occurred in an elided constituent compared to when it occurred in an overt constituent ((18b) vs (18a)). (The sentences in (18e,f) just show that the baseline sentences are comparable when there is no extraction.) (18)

They accounted for the fact that the adjunct island violation penalty was smaller inside an elided constituent than in a pronounced constituent in terms of the effect of focus on attention allocation during sentence processing. Other things equal, more attention is allocated to focused material than to unfocused material (e.g., Cutler and Fodor 1979), and by hypothesis more to overt material than to covert material. See also Sprouse and Hornstein (2013) for discussion. (p. 263)

Page 11 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics 11.2.2.4 C-command inside the ellipsis site If syntactic structure exists inside an elided constituent, then syntactic relations such as c-command should be defined for dependencies inside elided constituents. Yoshida, Dickey, and Sturt (2013) exploited the gender mismatch effect to investigate whether there are articulated c-command relations inside the ellipsis site in sluicing. Consider (19a). If the sentence ends after himself, then the only analysis of the sentence is one with sluicing, where an interrogative clause has been elided. On the sluicing analysis, the reflexive, e.g., himself in (19a) in the interrogative constituent (which story about himself), will be bound by grandfather, and no gender mismatch will result. However, in (19b) there will be a gender mismatch when himself is bound by grandmother. Consequently, if readers compute the sluiced analysis, there should be a boggle at himself in (19b). The sentences in (19c,d) are not even temporarily open to a sluicing analysis, so they serve as controls. Reading times in (19d) should be no longer than in (19c) since a sluicing analysis is not possible in either of the two sentences. In a self-paced reading study, reading times at himself and the following word confirmed the prediction that there would be a gender mismatch effect only in (19b). (19)

The results suggest that binding of the reflexive takes place inside the ellipsis clause. This in turn suggests that c-command relations are defined in the ellipsis site given that typically only c-commanding antecedents are considered for reflexives (Dillon et al. 2013). That the c-command relations in the elided constituent are assessed so quickly, in circumstances where there is no need to posit ellipsis at all, suggests that the parser automatically computes syntactic relations for material that is copied from the antecedent.

11.2.2.5 Indices are copied into the ellipsis site Investigating the processing of elided constituents whose antecedents contain syntactic indices could be informative with respect to different hypotheses about the grammatical structure in the ellipsis site. Semantic priming has been used to study the processing of the strict interpretation of sentences like (20a), where the fireman defended the policeman. Participants listen to a sentence and at some point during the sentence a letter string, the target for a word/non-word decision, appears on a computer. On the Page 12 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics critical “yes” (word) decisions, a participant’s response times are faster if the word is semantically related to (p. 264) what the participant is processing than if it is semantically unrelated. Shapiro and Hestvik (1995) investigated whether there would be priming for words related to the first clause subject in the ellipsis clause, at the site of the elided verb phrase, in examples like (20a). The question is whether hearers compute the strict interpretation, where the first clause subject binds the reflexive, online. The question is of particular interest because it is assumed that it is the sloppy interpretation (the fireman defended the fireman) that is preferred intuitively. Their results suggested that priming from the strict interpretation does exist. Following up this result, Shapiro et al. (2003) used priming to investigate whether the strict (the fireman defends the policeman) interpretation is computed even in sentences containing a predicate like perjure, which is not felicitous when applied to a different person, i.e., the fireman cannot perjure the policeman. (20)

The results indicated that priming from the first clause subject did take place at the ellipsis site even in (20b). The investigators concluded that, when copying the antecedent VP into the ellipsis site, the index on the reflexive is copied too, and it is processing this index that leads to reactivation of the first clause subject (though see Koeneman, Baauw, and Wijnen 1998 for discussion). In short, if the ellipsis site contains copied syntactic structure, the result makes sense. But without the assumption that syntactic structure exists in the ellipsis site, this result is puzzling. If listeners only copied the semantic property (perjure-self) from the antecedent into the ellipsis site, as expected on a purely semantic theory of ellipsis, then no priming of words related to policeman should have been observed.

11.2.2.6 Phonological information may be present in the ellipsis site Few studies have tried to investigate whether phonological structure is activated in an elided constituent. However, Snider and Runner (2011) is an exception. They investigated the types of information activated at the ellipsis site, comparing VP ellipsis (21a) and VP anaphora (21b). (21)

They conjectured that if ellipsis involves full syntactic structure at the ellipsis site then words should be present, and they should activate their phonological properties. In a Page 13 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics visual world study, they showed that semantic associates of the antecedent object (key) were available for both VP ellipsis and VP anaphora. But at the point at the ellipsis site/ pronoun, effects of phonological neighbors (log) of the antecedent object were activated only for VP ellipsis, but not for VP anaphora. Although any single study of this sort probably cannot (p. 265) show compellingly that syntactic structure is present at the ellipsis site, various types of evidence are now accumulating which converge on this conclusion.

11.3 Identity condition 11.3.1 Recoverability To recover the interpretation of an elided constituent there must be an identity relation of some sort with an antecedent. Theories of ellipsis differ on whether it is identity at some level of syntactic structure or semantic identity. It is to this topic that we turn in section 11.3.2. However, given that certain morphosyntactic features may differ between the antecedent and the ellipsis site, e.g., the value for gender in The boy hurt himself and the girl did too, the term ‘matching’ (identity up to differences in morphological features such as gender or number) will be used in the present chapter rather than ‘identity’. In section 11.3.3 we take up a proposal based on repair of the input that may restore syntactic matching of antecedent and ellipsis site under particular circumstances. Section 11.3.4 takes up accounts that do not require syntactic matching in the grammar, but nevertheless favor matching for processing reasons.

11.3.2 Syntactic structure effects From a psycholinguistic perspective, many of the findings discussed in section 11.2 suggest the existence of a syntactic matching constraint. The reason is that comprehenders (listeners and readers) generally could only identify structure in the ellipsis site by copying it from the antecedent. This copying will result in syntactic matching of antecedent and ellipsis.7 If the relation between antecedent and ellipsis site were one of semantic matching in a purely semantic theory of ellipsis, it would be puzzling how the comprehender managed to intuit the structure in the ellipsis site. Indeed, it would be puzzling why the comprehender would even bother to spell out the structure in the ellipsis site if an interpretation had already been arrived at through the semantic matching constraint and the grammar did not require syntactic structure to be present in the ellipsis site. If the grammatically imposed matching relation were semantic, the question also arises of why identity/matching effects are stronger for ellipsis than for anaphora, why semantically matching DPs (for German fragment answers) are less acceptable than PPs in some contexts (those prohibiting preposition stranding), why gender mismatch effects should appear in (19b) but not (19d), and why priming of the first clause subject should appear at the gap site in the second clause when the corresponding (strict) interpretation would be semantically impossible as in (20b). The hypothesis that syntactic identity is grammatically required also offers a Page 14 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics natural account for effects implicating the phonological content of the empty phrase on the assumption that what is copied includes words. It is possible that a (p. 266) theory embracing a semantic identity/matching constraint might be devised, e.g., one based on activating all properties of the antecedent, including its phonology, when the semantic antecedent is retrieved. However, the challenge would then be to say why the same thing didn’t happen when the antecedent for a pronoun is retrieved. Turning to a hybrid grammatical account, claiming that syntactic structure exists in the ellipsis site but that the identity relation is semantic (Merchant 2001), in principle one might imagine that the processor builds syntactic structure in the elided constituent not by copying it from the antecedent, but from the bottom up, syntactic inference by syntactic inference (assuming each syntactic phrase postulated must be licensed by the grammar). But this incorrectly predicts a general preference for small ellipsis sites and it predicts, counterfactually, that large elided constituents should take longer to process than small ones because more structure must be built (see Frazier and Clifton 2005 and Martin and McElree 2008 for evidence that large elided constituents are no more costly than small ones).8

11.3.3 The relation between the grammar and the processor As noted in the introduction, the observation that it is the grammar together with the processor that account for actual data, including acceptability judgments, is important because it means that intuitions need not be explained by the grammar alone. Arregui et al. (2006) proposed a recycling approach to ellipsis, where a syntactically mismatching antecedent is repaired at LF. In later developments of the approach, it is shown that likely syntactic blends are repaired in general, not just in the case of ellipsis (Frazier 2015). On this approach, it is the processor which accounts for the relative acceptability of some easy-to-repair sentence tokens. Arregui et al. (2006) argued that mismatch ellipsis is an instance of acceptable ungrammaticality. In effect, a sentence that may have been produced as a syntactic blend (a speech error where elements from two different formulations of a message are mixed together) may be repaired in comprehension. In a speeded acceptability judgment study, Arregui et al. (2006) tested passive antecedent-active ellipsis (22a) mismatches and active antecedent-passive ellipsis (22b) mismatches. They predicted that the passive– active mismatches would be more acceptable than active–passive mismatches because passives are misremembered as actives more often than actives are misremembered as passives (Mehler 1963). The prediction was confirmed in a self-paced reading study. (22)

Page 15 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics They proposed that a speaker might utter a passive antecedent clause and then at the ellipsis clause forget the voice of the antecedent clause and utter an active ellipsis clause even if she is using a grammar requiring syntactic matching. Listeners might find such mismatches relatively acceptable because they sound familiar, in the sense of corresponding to an expected output of the human language system. Further, they are easy to repair. Indeed, Arregui et al. also present data that the number of operations implied by a repair correlates inversely with acceptability (see Kim et al. 2011 for further evidence). On this account syntactically mismatching ellipsis examples are ungrammatical but certain tokens of mismatch ellipsis that are motivated by the principles of the production system are nevertheless relatively acceptable (if they are also easy to repair and yield a plausible interpretation).9 (p. 267)

The repair hypothesis predicts that other structures where the speaker has two common ways of syntactically formulating the same message should also give rise to relatively acceptable syntactic mismatches, and to interpretations that would result from the repaired structure. For example, a speaker predicating two predicates of the same individual might express the message using conjoined VPs or using conjoined clauses. This might result in VP ellipsis even when the antecedent contained conjoined clauses: a speaker might utter a conjoined clause antecedent, forget that it was conjoined clauses rather than conjoined VPs, and then utter an elided VP. The comprehender might repair the input and assign it a plausible interpretation. Frazier and Clifton (2011) performed an interpretation study of sentences like those in (23). 84 percent of the responses indicated that readers computed the conjoined VP (George travels in winter and stays home in summer) interpretation of both the conjoined VP and the conjoined clause examples, as if the comprehender repaired the conjoined clause input to a conjoined VP, i.e., to the antecedent structure that the speaker might have chosen but didn’t actually choose. (23)

The speech error repair account was developed on the assumption that the grammar requires syntactic matching of antecedent and ellipsis site. Thus, mismatch ellipsis on this account is always technically ungrammatical, but sometimes it is acceptable. In particular, as already noted, mismatch ellipsis was argued to be acceptable when an input could be the result of a likely production error, one that is easy to repair, and interpreting the repaired input results in a plausible interpretation. Frazier (2015) argues that the assumption that repaired utterances may sometimes be relatively acceptable is needed in general and not just for ellipsis. She argues that there are two interpretive systems for natural language: one that is compositional, type-based, and delivers both plausible and implausible interpretations; and one that makes reference to the grammar and the processor, is token-based, and delivers only plausible interpretations. (See Frazier and

Page 16 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics Clifton 2015 for further evidence of repairing syntactic blends outside the domain of ellipsis.) (p. 268)

11.3.4 Alternative accounts of syntactic mismatches Proponents of ellipsis theories that do not require syntactic matching (Kim et al. 2011; Miller and Hemforth 2014b), or do not require it in certain discourse (Kehler 2002) or information structure (Kertz 2013) circumstances, do not think that experimental demonstrations of a reliable acceptability penalty for syntactic mismatches are problematic for their theories, even though mismatches are predicted on their theories to be fully grammatical. They point out that there might be another explanation for the mismatch penalty—some explanation other than that the grammar requires syntactic matching. For example, Kertz (2013) compared “contrastive topic” sentences (24a,b) and auxiliary focus sentences (24c,d), expecting a syntactic mismatch effect only with “contrastive topic” sentences. She found a syntactic mismatch penalty even for auxiliary focus sentences ((24d) relative to (24c)), where it was not predicted. She noted that sentences without a syntactically matching antecedent may have a small LF mismatch penalty (though she does not elaborate on what kind of penalty this is supposed to be). (24)

Miller and Hemforth (2014b) compared verbal (25b) and nominal (25a) antecedents for VP ellipsis. They found a mismatch penalty in (25a) not expected on their account where the grammar requires only semantic matching of antecedent and ellipsis site. But they note that the matching examples might be easier to process because a matching antecedent (25b) might be easier to identify and retrieve than a syntactically mismatching antecedent, e.g., in (25a).10 (25)

The idea that syntactic matching may facilitate processing or antecedent retrieval seems perfectly reasonable, and it makes a prediction. The effect of syntactic matching should be greater in more complicated sentences than in easy sentences where retrieval of the antecedent should not be difficult. Although this prediction has not been tested experimentally, my own intuitions do not support it. Indeed, they suggest the reverse. It is precisely in short simple unadorned sentences like (26) that a violation of syntactic Page 17 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics matching seems worst despite the fact that there is little processing complexity, ambiguity, or memory load. (p. 269)

(26)

In any case, it is reassuring that the distinct approaches make distinct predictions. In sum, a theory with a semantic identity/matching requirement in the grammar and a memory retrieval account of the (as far as I know) always present syntactic mismatch penalty predicts a larger mismatch penalty in more complex sentences. A theory with a syntactic matching requirement predicts a syntactic mismatch penalty should always be present, even in very simple easy-to-process sentences.

11.4 Licensing 11.4.1 Licensing in the grammar Ellipsis cannot occur just anywhere but instead is restricted. This is usually discussed in terms of licensing conditions (e.g., Lobeck 1995). The ellipsis site must be licensed by a “governor” that permits its complement to be empty (see also Merchant 2001, for example). In an example like (27), the complement of denied could not be elided because there is no licensor for an empty CP (and deny is not a verb that permits null complement anaphora). Consequently, the ellipsis is not licensed and the sentence is ungrammatical. (27)

In the psycholinguistic literature, this type of licensing has not been investigated at all, perhaps because it is difficult to compare the processing of grammatical/licensed and ungrammatical/unlicensed ellipses. Other types of licensing are related to discourse and information structure. Discourse coherence relations will be discussed in 11.4.2, the main assertion hypothesis in 11.4.3, and the role of accommodating presuppositions and its interaction with ellipsis in 11.4.4.

11.4.2 Discourse coherence relations Kehler (2002) proposed an interesting theory of discourse coherence relations in which each clause of a discourse is connected by a discourse coherence relation such as a causal relation or a “resemblance” relation involving contrast (John ate cauliflower. Bill ate peas) or similarity (Amanda ate snails. Joe ate them too). An ellipsis clause connected Page 18 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics by a resemblance (similarity, contrast) coherence relation requires syntactic matching because sub-clausal constituents (John, Bill, cauliflower, peas) must be identified in order to identify the similarities/contrasts that underlie the resemblance coherence relation. In contrast, an ellipsis clause connected by a causal coherence relation does not require syntactic matching on this theory because a causal coherence relation may be established based on propositional information alone, without consulting sub-clausal constituents. See Chapter 13 (this volume) for an enlightening discussion and development of this line of inquiry. The results of psycholinguistic studies of ellipsis and coherence relations (p. 270) (e.g., Frazier and Clifton 2006) have often been less than compelling to some proponents of the theory, because they have not taken into account the observation, clearly supported in Kehler (2002), that connectives may establish more than one discourse coherence relation. To ensure that participants understood the experimental sentences with the intended discourse coherence relation, Clifton (2014) tested examples like those in (28) in a written acceptability judgment task. In (28) there is a question appropriate for either a resemblance (28c,d) or causal (28a,b) construal of the answer. The answer contained verb phrase ellipsis with either a matching (28b,d) or mismatching (28a,c) antecedent. (The word too was present only in the second of two experiments, which were otherwise identical.) (28)

In both experiments, Clifton found a penalty for syntactic mismatches, but one of equal size statistically for resemblance (28c) and for causal (28a) relations. This suggests that a causal coherence relation cannot license syntactically mismatching ellipsis. Finding the same size cost for syntactic mismatches in resemblance coherence relation and causal coherence relation examples cannot be due to ambiguity in the coherence relation in Clifton’s experimental materials.

11.4.3 Main assertion preference Focus is important in choosing an antecedent, as discussed in connection with Carlson (2002) for studies of replacives, stripping, and comparatives. Studies of sluicing show that focused antecedents are also preferred in the case of sluicing (Frazier and Clifton Page 19 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics 1998). In addition to focus, other information structure notions are also important, such as the notion of “main assertion”—typically the content expressed by main clauses, not relative clauses or subordinate clauses.11 Frazier and Clifton (2005) argued that across sentence boundaries there is a tendency for an utterance to relate to the main assertion of the preceding utterance. Consider (29) where the elided VP in the final clause is ambiguous between an interpretation with the embedded VP taken as antecedent and an interpretation based on the matrix VP as antecedent. In written comprehension studies, the VP ellipsis in (29a), the baseline, was interpreted with the matrix VP antecedent 40 percent of (p. 271) the time. By contrast, in (29b) where the only difference is the presence of a sentence boundary, there were significantly more (55 percent) matrix VP antecedents. (29)

In (29) one might be concerned that the main-assertion tendency is really just the absence of the usually observed syntactic low attachment preference. In a self-paced reading study of the sentences in (30), however, the preferred interpretation, as revealed by the answers to questions, was for the matrix antecedent roughly 75 percent (74 percent in (30a), 70 percent in (30b)) independent of the order of main and subordinate clauses. In (30) there is no attachment issue, but the main assertion antecedent was favored nonetheless. (30)

Similarly, Frazier and Clifton (2005) report results from a self-paced reading study of potentially epistemic clauses like I think in (31a), which are open to an analysis where the matrix clause only contributes information about the speaker’s degree of certainty about the assertion expressed in the complement clause. Since (31b) is not open to such an interpretation, the prediction of the main assertion principle is that there should be fewer matrix antecedents in (31a) than in (31b). This prediction was confirmed by the answers to questions about the elided VP, e.g., What does Sam do?, with 32 percent matrix VP responses in (31a) and significantly more (50 percent) in (31b). (As usual, it is the difference between conditions that is of interest to us rather than the baseline level of interpretations of each type.) (31)

Page 20 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics A control study tested verbs like announce, which are not open to an epistemic interpretation, and found that the effect was not due to the presence of a first- vs thirdperson subject per se. One might wonder whether the main assertion principle holds only of ellipsis. Presumably it does not. In a study where participants were given a sentence followed by two words and then had to decide whether it was true that both words appeared in the sentence or not (a “two-word probe study”), Walker, Gough, and Wall (1968) tested items like (32) and found the fastest times for probe words both drawn from the main clause (scouts killed in (32)). In this task participants are simply understanding a sentence and then checking whether the probe words occurred in the sentence. So presumably what the results indicate is that the main assertion of a sentence is what’s most salient after the sentence has been understood. (32)

The results thus suggest that the preferred antecedents for ellipsis are reflecting general properties of memory and information structure, not something specifically tied (p. 272)

to ellipsis.

11.4.4 Accommodation Moulton (2007) proposed a principle of minimizing accommodation. He focused on the information status of different constituents, as illustrated in (33). (33)

In a written interpretation study, he found that in interpreting an elided VP in a coordinated main clause, there was a preference for large antecedents. However, when the elided VP was in a structure that generally conveys presupposed information (34b,c), then there was a preference for small antecedents. (34)

Page 21 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics

In sum, the choice of antecedent is sensitive to whether the elided constituent corresponds to asserted or presupposed content. In presupposed contexts, the processor minimizes accommodation. See Moulton (2007) for studies ruling out alternative syntactic accounts of the results, and Moulton (2013) for additional studies, interpreted in terms of the more general hypothesis that comprehenders prefer to minimize embedded content, specifically content reported in a restrictor clause.

11.4.5 Question Under Discussion (QUD) The idea that discourse is organized by a series of questions has received considerable attention in semantics (Beaver and Clark 2008; Roberts 2012), and the notion has been applied to ellipsis (Ginzburg 2012). Grant, Frazier, and Clifton (2012) proposed that the presence of alternatives that might be taken as the Question Under Discussion (QUD) facilitates mismatch ellipsis if the elided constituent comments on those alternatives/the QUD (see Clifton and Frazier 2012, for the role of alternatives and QUD in processing in general). In their study, it was the presence of a modal that introduced alternatives (needed in (35b)), presumably introducing the (p. 273) implicit question of whether the state of affairs described (the information being released) was actual or not. The modal increased the acceptability of mismatches in written acceptability judgment studies. (35)

In an eye movement recording study, it also resulted in faster reading times in (35b) than (35a) in the region after the ellipsis site (even though in (35)).12 (The slashes in (35) indicate analysis regions, but the slashes were not visible to participants.) The results suggest that minor syntactic deviance can be tolerated better when the offending form addresses the QUD.

Page 22 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics In Miller and Hemforth’s (2014b) study of polar nouns described in connection with (25), they also manipulated whether the context introduced an alternative or not (polar alternatives, e.g., It is impossible to predict whether Andrew will participate/Andrew’s participation in the tournament. He is sure to win if he does/does it.). In addition to a penalty for nominal antecedents compared to verbal antecedents for verb phrase ellipsis, they also observed an interaction whereby having an alternative in context facilitated verb phrase ellipsis more than verb phrase anaphora. One might take this as further indication that once alternatives are under discussion they facilitate ellipsis that comments on those alternatives. Miller and Hemforth’s own conclusion was that there is a special type of verb phrases ellipsis, aux focus ellipsis, which is limited to circumstances where alternatives involving tense, aspect, mood, or polarity are salient, and the entire point of the utterance containing ellipsis is to pick one member from that set. In the end they opt for a semantic theory of recoverability because they think it is simplest.13

11.5 Conclusions Many different types of psycholinguistic arguments and data support the hypothesis that there is syntactic structure in an ellipsis site. Stronger penalties are observed for structural mismatch between an ellipsis clause and its antecedent clause than between a pronoun-containing clause and the clause containing its antecedent. Gender mismatch effects show that c-command relations in the elided constituent exist and are computed rapidly and (p. 274) automatically even when ellipsis is not necessary. Effects of island violations have been established in elided constituents. Indices from an antecedent are copied even when they are incompatible with the semantic constraints of the predicate. Suggestive evidence indicates that phonological content may be copied from an antecedent into an ellipsis clause but not into a clause containing anaphora. Recovering meaning for an ellipsis clause is dependent on there being an identity or matching relation with an antecedent clause. It was argued that this relation is one of syntactic matching. In the absence of mental telepathy between speaker and hearer, the very existence of syntactic structure in the ellipsis site is at least suggestive evidence for a syntactic relation since it is by copying, or sharing, structure with the antecedent clause that the processor could obtain syntactic structure.14 Although in principle the processor might build syntactic structure in the ellipsis site bottom-up, inference by inference, guided by meaning or by properties of the antecedent or by re-parsing the antecedent word string, any of these possibilities would be difficult to reconcile with the fact that large ellipsis sites are often preferred to smaller ones. Further, there is no cost for large elided constituents relative to small ones (Frazier and Clifton 1998, 2005; Martin and McElree 2008). If structure were built bottom-up, this would fly in the face of the parser’s very pervasive structural economy preferences. In sum, given syntactic structure in the ellipsis site, it is probable that it comes into being by copying the syntax of the antecedent clause.

Page 23 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics One central reason that theorists prefer a semantic identity requirement is their belief in Occam’s razor: a semantic identity requirement in the grammar appears to be simpler than a syntactic matching requirement plus a theory of repair (see discussion in Miller and Hemforth 2014b). However, the theory of sentence processing must independently contain a theory of (garden path) repair and an account of the repair of syntactic blends not involving ellipsis (Frazier 2015; Frazier and Clifton 2015). Once Occam’s razor is applied to the entire theory of language, both the grammar and the theory of processing, the claimed simplicity of a grammar with semantic identity vanishes. It was emphasized that it is the grammar together with a theory of processing that makes predictions about actual data such as linguistic intuitions/judgments, reading times, and the like. When comprehenders encounter an ungrammatical sentence, they do not assign an asterisk and stop processing. Rather they assign a best-fit analysis and interpretation. Under certain circumstances such utterances may be acceptable: when the utterance sounds natural (expected given the principles and constraints of the human production system), the input is easy to repair, and the interpretation is plausible, especially if it addresses a tempting Question Under Discussion. The assumption that “acceptable but technically ungrammatical” utterances may be rated as being acceptable relative to other ungrammatical sentences is needed independently (Frazier 2015 and references therein). In the case of ellipsis, it offers a potential explanation for why certain tokens of ellipsis sentences with a syntactic mismatch between antecedent and elided constituent may sound quite good even if the grammar requires syntactic matching of antecedent and ellipsis site. (p. 275)

Discourse factors are also important in ellipsis. In particular, across sentence

boundaries, other things being equal, listeners and readers prefer antecedents that are part of the main assertion. Preferred ellipsis sites in constituents with asserted content tend to be as big as possible. By contrast, preferred ellipsis sites in constituents with presupposed content appear to be as small as possible, presumably to minimize accommodation.

Acknowledgements I am grateful to the editors, two anonymous reviewers, Chuck Clifton, and Tim Hunter for helpful comments on the manuscript.

Notes: (1) See Biezma (2014) for studies of replacives and stripping in Spanish showing that the mere presence of an overt subject in pro-drop languages serves as a focusing device and thereby increases the probability that an overt subject will serve as an associate. See Grant (2012) for the processing of comparatives in English though the work is not focused on comparative deletion per se.

Page 24 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics (2) Gapping is a plausible instance of ellipsis though it has received various analyses in the linguistic literature. In particular, see Johnson (2009) for an analysis based on low coordination and across-the-board movement. (3) This pattern of favoring a particular analysis primarily only when one potential antecedent is accented and the other is not accords well with other findings, such as those reported in Carlson et al. (2009a). It also fits with current views characterizing memory as a cue-based direct-access system, where antecedents are activated based on their degree of match with retrieval cues (Lewis, Vasishth, and Van Dyke 2006), rather than as a (backwards) search mechanism. (4) Evidence from Antecedent-Contained Deletion has also been presented based on quantifier raising and the ease of ACD under conditions where raising has already applied for independent reasons (Hackl et al. 2012). Unfortunately, the reading-time results do not hold up; the acceptability judgments, which do, have been attributed to an alternative source (Gibson et al. 2015). (5) Examples from Clifton and Frazier (in progress) appear below, along with the data from a number of fixations in the penultimate region. Slashes indicate analysis regions but were not visible to the participants. ((i))

(6) Frazier and Clifton (2005) explained the high cost of sprouting island violations in terms of the failure of repair: the syntactic processor cannot make the repair because the syntactic processor does not violate islands, and the discourse processor cannot make the repair in sprouting examples because it cannot build syntactic structure. (7) Copying of the antecedent and structure sharing may be notational variants, though see Martin and McElree (2008, 2009), for an alternative view.

Page 25 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics (8) See Paape (2015) for evidence that in examples where the antecedent contains a garden path there is no parsing difficulty at the ellipsis site. If the parser built structure in the ellipsis site by reparsing the antecedent’s word string, difficulty would have been expected. Note that one early study did find elided constituents with large antecedents to be more difficult to process than ones with small antecedents (Murphy 1985). It’s not entirely clear why, though perhaps the use of small clause antecedents is relevant. (9) One issue is why voice mismatch seems less acceptable in sluicing than in VP ellipsis. The difference probably lies in the stronger need for parallelism when variables must be identified as is always the case in sluicing, where the interrogative must bind a variable. This predicts that voice mismatches should also be worse in VP ellipsis when the elided VP contains a variable, as in ACD cases. (10) Miller and Hemforth (2014b) also manipulated properties of the context, not illustrated here, because it is not of relevance to the present discussion. In the end they propose a theory with a semantic identity requirement. They suggest that Occam’s razor favors this account over one containing repair operations. However, as discussed in section 11.5, if we assume that the theory of language must include both a grammar and a theory of processing, then given the independent evidence that repair operations are needed, e.g., for garden paths, the Occam’s razor argument no longer holds. (11) The notion of main assertion is related to the notion of Question Under Discussion (QUD), assuming the QUD is generally introduced by material in a prominent position in the tree. Elided material preferentially addresses the QUD (Grant et al. 2012); the QUD is preferentially derived from alternatives introduced by material in the main assertion (Clifton and Frazier 2012). (12) Specifically, in the region following the ellipsis site, first fixation durations (271 ms vs 246 ms) and gaze durations (370 ms vs 333 ms) were significantly longer in the nonimplicature (a-) forms than in the implicature (b-) forms. In the final region of the sentence, ‘go past’ times showed the same pattern (1417 ms vs 1280 ms, all p < .01). (13) While Miller and Hemforth’s theory might be simple and adequate when considering only polar nouns, they do not really explain why non-polar nouns are not as good as polar nouns: if the grammar does not require anything except semantic matching, non-polar nouns should also be acceptable antecedents. (14) The listener might posit structure in the ellipsis site by guessing randomly and then try to match it to the antecedent, and then guess again, etc. But this sort of Analysis-bySynthesis model is inefficient at best and has largely been abandoned.

Lyn Frazier

Page 26 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Psycholinguistics Lyn Frazier is a Professor in the Department of Linguistics at the University of Massachusetts Amherst. Specializing in psycholinguistics, her research spans a range of topics in syntactic processing, primarily concerning phrase structure parsing and the parsing of movement dependencies, and issues at the syntax– discourse interface. The latter include research on the role of prosody in sentence processing, ellipsis, processing of not-at-issue content, and the role of implicit and explicit Questions-Under-Discussion (QUDs) in organizing discourse.

Page 27 of 27

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Oxford Handbooks Online Ellipsis and Acquisition   Tom Roeper The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.12

Abstract and Keywords Ellipsis provides a quintessential acquisition challenge—deleted material is an obvious instance of poverty of the stimulus. This chapter reviews the naturalistic and experimental evidence to evaluate how far UG allows immediate correct projection of VP and NP ellipsis, based on Merchant’s (2014) E-features. It develops an acquisition theory whereby a child moves, step by step, from Open Interfaces to isomorphic Strict Interfaces between LF and syntax which entails full reconstruction. The approach both utilizes pro as an initial default and explains why it remains in the final grammar for particularly complex constructions. It follows an economy principle: minimize recovery by contextual inference. The analysis is extended to sluicing, ACD, fragments, and the projection of QUD, and engages English, Portuguese, German, Dutch, Italian, and African-American English. Keywords: licensing, Strict Interfaces, recoverability, E-feature, Strict Identity Hypothesis, partitivity, full reconstruction, inference, pragmatics, learnability

Tom Roeper

12.1 Introduction: What is the acquisition challenge from ellipsis? HOW does a child acquire the special properties of ellipsis which vary from language to language? (See Johnson 2008 for background.) How does she know that (1a) is impossible in English with articles but possible in German (1b) with case marking (ein = nominative)? Furthermore, she must know that demonstratives with noun agreement are fine in English as shown in (1c), where this and that are singular (not these, those), marking a deleted noun as singular: (1) Page 1 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Likewise, English allows verb phrase ellipsis (VPE) but other languages do not. Languages that allow VPE often do not allow a hanging to—a language-particular licenser —but English does. (2)

It will become clear (a) that subtle, language-particular features like agreement license ellipsis, (b) that the child must have one (or more) operations to recover what is missing, and (c) that different degrees of precision are possible for the representation that is created. Sometimes gender, case, and number can be ignored, suggesting a more abstract or incomplete representation, but at times, some but not all adult speakers will experience jarring ungrammaticality if gender (her arm) is altered (to his arm): (3)

Such sentences indicate unresolved ragged edges for ellipsis which continue to perplex linguistic theory but which are present in the child’s everyday linguistic environment. (p. 277)

There is an immediate acquisition consequence here. Where the input data is especially complex and decisive cases are rare, then abstract principles must play an organizing role. Only via narrow and sharp UG principles could a child possibly carry out the induction of where ellipsis is possible/impossible and what meaning must be reconstructed as. In general, we know that our grammar can lead to amazingly exact forms of ellipsis over complex phrases. (4)

Page 2 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition When does a child know a full reconstruction with focus is necessary? This is one of many experiments waiting to be done. What will a 4-year-old child say if we ask, ‘what did Mary wear’? Will it be, ‘a bright red, speckled HAND-made bracelet’ with identical phonology and focus intonation, showing that a full phonetic representation has been recovered? Or, will it be just the head ‘bracelet’? Thus far, no one has done this experiment.1 The entire question of the role of topicalization and focal contrast in ellipsis acquisition remains in early stages.2

12.1.1 General theoretical consequences: An overview of Strict Interfaces We begin with an extensive introduction to orient linguistic theory to the acquisition challenge. As we proceed, we provide small previews of the data which we then navigate in fairly microscopic detail. Our exposition aims to provide an overview of what has been observed about ellipsis for major structures in acquisition while proposing a theoretical approach.3 The core idea is that the child should begin with some universal invariant Strict Interfaces. They function as a teleological goal of acquisition; it is what a child looks for and seeks to impose. The abstract ideal of a Strict Interface is as follows:4 (p. 278)

(5)

An intuitive example of a Strict Interface is the notion of a negative imperative: ‘don’t!’. This is understood by children below 2 years although it is heavily elliptical (and it is hard to be sure where general inference enters). A speech act, in this case an imperative, links (a) syntax: a missing subject and empty verb (you, verb, e.g. push), (b) semantics: illocutionary force relation, (c) phonology: focus intonation, (d) pragmatics: an open inferential variable referring to possible action in a context. Language-specific morphology (an imperative inflection) is possible in some languages, but the core connections are arguably universal for imperatives, and critical to early communication. Although animal communicative powers remain largely unknown, it is plausible to argue that Strict Interfaces are species-specific. Strict Interfaces can, we propose, allow the child to reject grammars that fail to have a tight link between inherently different representational levels. It stands in contrast to what we call Open Interfaces. (6)

Page 3 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition We consider the notion of Strict Interfaces to be implicitly entailed in proposals by Chomsky (2005, 2013, 2014) that interfaces are a crucial dimension of UG. Open Interfaces link naturally to claims about pronominal representations for ellipsis, which have been widespread for over twenty-five years (Hardt 1999; Merchant 2001, this volume; van Craenenbroeck 2017), and which we shall explore. Our goal is to show that the Strict/Open contrast is a natural part of UG that will predict the acquisition path and represent an available default structure for adults. The UG acquisition principle moves the child from a minimal representation (simple pro) with inferences to a maximal one (full syntactic and phonological representation) under the following principle: (7)

This leads to our general acquisition path claim: (8)

We argue that such diverse features of grammar in a language are a necessary and valuable property of UG. Such a theory should both account for intuitional data and capture the acquisition path.

(p. 279)

12.1.2 The acquisition perspective on representations, licensing, and recoverability Representations. UG allows diverse and incompatible representations for elided material which operate with different primitives. If the simplest representation can be an empty pro, while the most complex representation contains phonological, syntactic, and semantic information, then the acquisition path should exhibit the simpler forms first. It is essential to observe that alternative projections of VPE (simple pro and full PF) are not straightforwardly statable as a single rule with options. Where options exist, for instance in parametric choices, they are of the same type (+ or – subject, VO or OV). Therefore, since they are separate representations in a language (pro or full PF), but not collapsible, then, in effect, the language contains Multiple Grammars in miniature. We argue that both options are necessary to make acquisition succeed.5 Those options may in turn be reflections of deep typological differences, like the contrast between “hot” languages that allow contextual reference and “cool” languages that do not (Huang 1982) or NP and DP languages (Bošković 2008, 2016). That is, some languages project bare nouns and a simple NP where others project a full DP. It should not be a surprise that the acquisition path is guided by access to Multiple Grammars rooted in different typologies. Page 4 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition Licensing. Exactly what features must be present to license ellipsis? Merchant (2014a, this volume) argues that a specific E(llipsis)-feature must be present on a head. For the child, the input data should be immediately informative about what carries the E-feature given two UG assumptions: (a) licensing requires a head from certain maximal projections, such as TP, and (b) it may presuppose a moved head under some theories (e.g., Thoms 2011b) but one which is UG-required. The choice of head can be very clear from the input. For instance, tense (e.g., ‘did__’), or ‘to__’. Early child uses of ‘I did’ suggest that licensing is quickly identified from the elliptical input (see more examples below). Where no UG-determined head is available, then the next step is to use verb subcategorization as the only alternative the input offers to know what is elided. Input evidence again should straightforwardly license Null Complement Anaphora (NCA), as in John liked to sing, Bill wouldn’t try => (to sing). Other verbs fail: *Bill wouldn’t like. Notably, NCA is linked to specific lexical items, not syntactic categories, which is a sign of limited generality in the adult grammar. Recoverability. It is clear that full PF recovery is ultimately necessary, which can be accomplished by traditional concepts of copy/delete. Our approach suggests that at earlier stages only minimal syntactic structure will be copied, just heads, and later full maximal projections. Finally, semantic abstractions of antecedent VPs at LF are added. While this acquisition theory is intuitive, there remain a great many empirical gaps, and occasions where children fall back on earlier (pronominal) representations.

(p. 280)

12.1.3 Exposition path: From theory to facts This chapter is organized as follows. We first look at the relevant acquisition logic that demands a heavily constrained theory of interfaces. We carry this out with an eye to the persistent role of general inference. Then we refine the concept of Strict Interfaces, the role of simplicity, and parallels with repair strategies in parsing. We propose an abstract version of the acquisition path it predicts. In section 12.2, we first survey early evidence of non-adult ellipsis, disorders, and the first evidence for ellipsis. We argue that early stages are inferentially based. Then we turn to evidence about VPE acquisition in early Portuguese and then, in section 12.3, to the role of a Strict Interface, based on Kennedy’s (2008) Semantic Identity Hypothesis (SIH). (9)

In sections 12.4 and 12.5 we turn to variants such as antecedent-contained deletion and sluicing which show that wh-movement may be involved in ellipsis. In section 12.6, we turn to argument drop in Japanese and English and in section 12.7, to NCA. Throughout, we argue for a pronominal representation to capture children’s behavior that deviates Page 5 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition from adult grammar. Then we sketch a Strict Interface for noun phrase ellipsis (NPE) (e.g., ‘John has a big cake, and Bill has a small__’) in section 12.8. Here, an LF representation allows definite reference to be imposed. (10)

Then, we develop a Strict Interface based upon an LF that expresses a partitive relation, such that a sentence like ‘I want five’ entails the presence of a larger set of which ‘five’ is a subset. (11)

These serve as examples of what should ultimately be a rich array of interface connections that significantly simplify the acquisition process and explain default resort options in adult grammar. Finally, in section 12.9, we consider the potentially interfering and obscuring role of fragments and shifted Questions Under Discussion (QUD), which a bias toward Strict Interfaces should enable the child to accommodate. We argue that a pronominal alternative is a constant acquisition variant throughout these diverse structures. In sum, the very obscurity of ellipsis in conversation demands a theory of Strict Interfaces to explain how children arrive at highly restricted interpretations for highly complex and often rare structures.

12.1.4 Acquisition theory: Inferences create a learnability challenge (p. 281)

A strong alternative to systematic interfaces has been proposed: Culicover and Jackendoff (2012) claim that ellipsis can be completely governed by extra-grammatical general inference that can be coupled with any form of contextual parallel alignment—an ability not limited to language—to motivate ellipsis (assumed by many psychologists as well): Our approach thereby unifies the semantics of all of these phenomena under a common account that is based on a domain-general cognitive principle. (Culicover and Jackendoff 2012: 305)6 In general, we know that minimal conversations often provide no basis for discovering a clear verbal antecedent for ellipsis, requiring intricate inferences. Consider the following conversation:

Page 6 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (12)

As we shall see, much dialogue with children works this way too. Such inferences could be and often are sufficient to substitute for actual reconstruction, as Culicover and Jackendoff (and also others) argue. However, their approach overgenerates possible interpretations. Our approach seeks to account for how possible and plausible inferred interpretations are excluded. If inferences can be sufficient, we have a learnability challenge: Why use intricate reconstruction if general cognitive inference suffices? The Culicover/Jackendoff approach reduces the “learnability problem” to virtually nothing since it is presumed to be inborn and obviously advanced in the behavior of any 2-year-old. They utilize cognitive parallelism to establish what an antecedent for ellipsis must be. They comment, “we also achieve a gain by eliminating the machinery of deletion under identity,” and ask further, “This might be considered a wash: Why prefer a new untested piece of machinery to the old more or less reliable one?” (Culicover and Jackendoff 2012: 337). This view has intuitive appeal to many and so it needs to be addressed.

12.1.5 The simplicity of mechanisms Culicover and Jackendoff’s approach overlooks a fundamental concept of simplification. The answer to their question is: After it is acquired, a complex system of reconstruction can (p. 282) simplify mental computation. That is, it becomes easier to systematically reconstruct missing information syntactically or phonologically with an automatically rich representation than to undertake a complex guess based on context where a huge number of factors introduce a huge number of possible inferences. For instance, the precision of reconstructing a contrastive element inside a compound ‘HAND-made’ in (4) does not easily follow from general inference, but must follow from total reconstruction. To use an analogy, a very complex object like a computer can be designed for usesimplicity such as Amazon’s one-click ordering system. Likewise, once constructed, the grammar is easier to use than inference. The critical claim is that difficulty of construction—following the acquisition path—is compensated for by ultimate simplicity. To make such a model work, however, the system being built must know where it is going (just like a computer programmer has a programming aim for user-friendly machines). We argue that the child knows that Strict Interfaces should be identifiable in the language input.7

Page 7 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition Once the grammar of reconstruction is established, it operates with such efficiency that it is always preferable to the available repetition of the VP. Looking ahead, even 2-year-olds, who begin with full repetitions (e.g., saying ‘I did eat’ after hearing ‘Did you eat?’) easily move to ‘I did’, ‘I didn’t’, ‘I can’, ‘me, too’ (Renato 2016). Over 100 examples of the fairly rare word ‘either’ with ellipsis are found with 2-year-olds on CHILDES (MacWhinney 2000). (13)

Therefore, prima facie evidence can immediately project missing syntactic material (these [animals] do not [stand up]) with the sophisticated semantics of negative either, although it is still not fully grammatical. Easy use of such kinds of ellipsis suggest that a sophisticated syntax/semantics Strict Interface is what the child is seeking. Nonetheless, a much more refined analysis is necessary before we can grasp an acquisition path for more unusual elliptical constructions.

12.1.6 The repair parallel In a parallel argument, Frazier and Clifton (2006) argue that the parser repairs representations in the process of reconstruction which must likewise presuppose and prefer the same Strict Interfaces. For instance, ellipsis can ignore (active–passive) voice mismatches ‘hotdogs were eaten, but I didn’t want to’, inflectional variation ‘John is playing baseball, but (p. 283) Bill won’t’, and some kinds of referential flexibility. Speakers, along a gradient, consider these worse than PF-perfect forms of reconstruction, as Frazier and Clifton (2006) have shown in their theory of ellipsis repair. Therefore, we assume, in concert with the theory of repair, that Strict Interfaces must be presupposed (see Merchant 2015b).

12.1.7 A pivotal role for inferences Nevertheless, rather than reject general inference, we need to incorporate it as part of the system that produces a plausible acquisition path. Again, we claim that at Stage 1 a child begins with general cognitive inference applied well beyond language, but which step by step replaces inference with grammatical reconstruction. This goal of minimizing pragmatic inference is built into UG. So grammars will vary in where they allow full reconstruction or continue to use pro. Two distinct traditions—pronominal representation and total reconstruction—have sought to characterize approaches to ellipsis without either proving completely adequate. Acquisition logic offers a different approach. Both systems are needed to account for how

Page 8 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition the child projects first representations and then refines them. First we need to look closely at how they contrast. It is obvious that a pronoun will work in an example like (14) where an inference or a reconstruction can arrive at ‘sang’. (14)

But note that a pronominal form, substituting either for the whole VP or a larger or smaller part, often allows other inferences as well. (15)

In (15b) that refers to ‘hated washing dishes’ or it can just refer to ‘washing dishes’. Thus, a simpler grammatical representation introduces a plethora of inferential options. These can only be resolved by applying further inferences to select which is right, adding to overall mental computation. Experiments below show that children are susceptible to just such misinterpretations. In more complex environments, Merchant (this volume) and van Craenenbroeck (2017) have argued that just such a pronominal form reappears for adults. Consider this case from Hardt (1999) and Schwartz (2000) cited by van Craenenbroeck: (16)

He observes (2017: 7) “…the second VPE-site (x2) takes as its antecedent the bracketed [VP in [VP]] in the antecedent clause” [= a variable inside a variable].” Merchant suggests that the second site should be filled with just do that, and Craenenbroeck argues that “the second ellipsis site is pronominal in nature.” (p. 284)

Göksun et al. (2011) provide a suggestive acquisition experiment where children were given both: John wanted to make pancakes, but Bill doesn’t want to__ or Bill doesn’t want some. In both instances they chose a picture of pancakes rather than an action of Page 9 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition cooking, suggesting a missing pro-N rather than a VP, which is compatible with the argument and experiments below. The pro option should not be a random alternative, but a predictable option. In a nutshell, it means that if there is a complex case, involving discourse or quantification, with linked VPEs, then UG may allow reversion to a simple pro at the second ellipsis site with new inferential demands. If the child’s grammar is less advanced—say, in knowledge of quantification—the reversion to pro could happen in other grammatical environments. While the seemingly intractable nature of these theoretical conflicts has led to proposals that there should be hybrid theories (see Merchant this volume and van Craenenbroeck 2017 for a summary), we argue that what seems like a hybrid reflects a layering of systematic representational options so that the term “hybrid” no longer applies. Instead, the availability of these options allows an acquisition micro-path that is just coming into view.8 We propose a rough path through these options. (17)

These proposals remain programmatic until a magnified vision of the acquisition path can be articulated, but they provide the basis for a research program that looks carefully at each form of ellipsis. (p. 285)

Page 10 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

12.2 Early discourse: What do children’s elliptical dialogues look like? Every conversation demands that we add unspoken information. What do we (or a child) think when we hear ‘oh dear’ or just ‘oh’ or ‘uh-oh’? ‘Uh-oh’ is often a child’s first word. They are different from ‘um-hm’ (assent) or ‘un-unh’ (disagree) or just ‘mhm’ (understood), all of which require a verbal context (generally). A child’s life is full of expressive utterances like these. By their fourth year, children control dialogues where both nominal and verbal ellipsis are carried along across lengthy exchanges. Here is a typical dialogue (18), with a 4-year-old (Wijnen et al. 2003): (18)

It has noun phrase ellipsis (NPE), ‘more’, and verb phrase ellipsis (VPE), ‘yes you do’, which reproducing the dialogue must have NPE inside it (i.e., ‘yes you do see more [N]’). The ‘it’ is never revealed, and even ‘mmhm’ requires reconstruction of elided information. The 4-year-old has mastered a great deal beyond what a 2-year-old knows.

12.2.1 Earlier evidence of non-adult ellipsis Very early child moments, when the child seems to assume ellipsis, often fail to communicate. (19)

Page 11 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

The child implicitly assumes a broad inference that gets to ‘I want milk on it’, while the adult needs syntactic precision and never figures it out. (p. 286)

A look at other failed dialogues reveal children trying to use ellipsis inadequately or inappropriately, which give the reader a sense of frustrating real child dialogues (Jensen and Thornton 2008). (20)

Page 12 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (21)

There is no doubt that some representation from the previous sentence is presupposed, but its grammatical characterization is obscure. Often a verb is presupposed ‘(put) on leaf’, ‘(put) up here’, or ‘(go) up in the air’, or silently projected in a way that adult grammar does not allow.

12.2.2 First verb phrase ellipsis: Where should it be? We look first for early stages where UG principles can be directly reflected. We predict: Where ellipsis primarily reflects a straightforward UG principle, then it should be triggered immediately for very young children, with minimal evidence. In fact, young children perceive and use VPE in both Portuguese and English (see section 12.2.4), even though VPE is not universal. Although VPE is language-particular, a dialogue provides dramatic evidence. When someone says ‘I did’, both adult and child see immediately that something missing must be filled in. Given the UG principle that the immediately preceding head must license ellipsis, (p. 287) acquisition is instant with no delay or detour, allowing some form of reconstruction of the missing complement. In addition, as Aelbrecht (2010) shows, some non-adjacent cases must be accommodated, which we can predict will be more challenging. The adjacency claim does not guarantee, as we shall see, that what is reconstructed is entirely adult; it may still contain an invisible pronoun. A cursory survey of ten children between 1;9 and 2;2 years for their use of ‘did’ reveals that in every instance but one, an earlier tensed sentence occurs before the child’s VPE with ‘did’. (22)

(23)

Page 13 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition While I did by an adult provides sufficient information to know that an empty element is present, it does not reveal how richly structured the missing information is. That is, we satisfy the licensing requirement, but the representation is not automatically adult. It could be a general pro projection with little structure, as our model suggests. Jensen and Thornton (2008) show that twenty children (1;10–2;8yrs) control these structures (‘I did’) early but there is an interesting dip in their use of VPE as follows. At first we have wh-questions like below. (24)

The simple answer ‘tomato’, all that a wh-question requires, arises only when children have sluicing sentences like ‘I want to eat, but I don’t know what’, which exhibits A-bar wh-movement and ellipsis (see section 12.5 on sluicing). This correlation provides a hint that when children first use VP-ellipsis, a representation like a simple pro could be present, but when sluicing arises with wh-movement, it triggers a richer projection into the elided position. This grammar shift for the child can explain the dip in ‘I did’ before sluicing arises: children will briefly avoid systematic ellipsis when they are unsure of what should be projected, pro or a full VP.

12.2.3 VP-anaphora In contrast to I did, an expression like I did it occurs as explicit VP-anaphora (including so-called deep anaphora) in a child’s experience. It is found in early files (25a). At the same age we find cases where an explicit pronoun it is linked to a contextual equivalent— an action (25b): (p. 288)

(25)

Page 14 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition The child knows that if no previous VP is available, then the pronoun must be overt, and verb phrase anaphora is required. Although such a study should be done more carefully, the evidence points to the notion that VPE occurs only when a tensed IP is present, which is immediately triggered through child/adult conversation. By contrast, so-called deep anaphora calling for an explicit pronoun ‘did it’ triggers direct reference to context.9

12.2.4 Early Portuguese verb phrase ellipsis In her careful work on early Portuguese, Santos (2009b) finds 20 percent (218 out of 1061 utterances) must be elliptical examples, indicated by the correct tense marking on a single word, echoing the English facts, thus evidence clearly compatible with the view that Tense is a licensing head.10 (26)

Reasoning across English and Portuguese, the licensing principle involving functional categories for ellipsis is apparently available before wh-movement. This is what we would (p. 289) naturally expect if ellipsis is at the core of conversation, but wh-movement is acquired later because of language-specific phenomena.

12.2.5 The movement requirement Thoms’ (2011b) analysis involves an important extra step and a strong constraint (arguably too strong):11 Page 15 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (27)

Thus if an element, like an auxiliary (could in (27b)), moves to a tense domain, then the TP can carry an E-feature which now allows ellipsis: (27)

Failure to move blocks ellipsis because no functional category licenser occurs. Progressives in English do not move, staying in the lexical vP, and therefore do not allow ellipsis. It is difficult to prove absence, but no one as far as I know has reported or seen cases like (28b) in the CHILDES database: (28)

It is raised to a T-position, but the progressive being is projected without movement under an Aspect node. Since being does not license ellipsis, we correctly predict that we find being only with a predicate in acquisition data: (29)

Out of 140 examples on CHILDES, none ends with being and ellipsis, and likewise out of roughly 200 examples of having none occurs with ellipsis. Evidence of such subtlety and precision obtained from very young children should be seen as signal support for Merchant’s theory of E-features and Thoms’ (2011b) theory of movement (also Lobeck 1995). Nevertheless, Aelbrecht and Harwood (2015) discuss evidence that the sub-features on the E-feature can vary from language to language, making the fact of early recognition more remarkable, since it means that the child must be sensitive to specific input as well as UG consequences. (p. 290)

Page 16 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

12.3 Verb phrase ellipsis and bound variables If we now ask about the recoverability of meaning when binding is involved, it calls for a more elaborate syntax/semantics interface, and the role of Strict Interfaces become critical. We utilize the formulation by Kennedy (2008), the Strict Identity Hypothesis (SIH), where LF is shorthand for logical meaning: (30)

We claim this is the Strict Interface that the child seeks; he knows an LF correlate should be projected. It is commonly represented as abstract copied syntactic categories together with linked lambda abstractions (which allows subjects to be treated as abstract extracted variables (x), so that subjects can be switched in the ellipsis site [Oscar => Bert] in (31)) if they are each equal to the variable x as Foley et al. (2003) demonstrate in the following experiment.12 Consider, for example, (31), a scenario where the child could either select the banana from Oscar’s plate for Oscar to eat, or select the banana from someone else’s plate. The child could retain that object for Bert to eat or could select another banana. (31)

They provide sharp evidence for children’s representation of bound variables with a series of experiments systematically testing pronominal, definite, indefinite, and full NP variations with children from 3 to 6 years. (32)

(33)

(34)

Page 17 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

(p. 291)

(35)

(36)

It is clear that the sloppy readings are the preferred option from very early, and all the other options in (35) are excluded by 4 years.13 Still one third of the 3-year-olds do not exhibit sloppy identity and an acquisition path may be involved.14 In what follows, we propose that Strict Interfaces are at the root of the explanation.

12.3.1 Determiner deletion Abdulkarim and Roeper (1997) provide further sharp evidence that favors the presence of a binding variable.15 Consider these results, which are distinctly non-adult, from an actout experiment by Abdulkarim and Roeper (1997): (37)

Page 18 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Thirty-five Arabic- and thirteen English-speaking 3- to 5-year-old children essentially drop the explicit possessive (i.e., John’s truck) and impose a bound variable reading (i.e., x pulls x’s truck (Fred pulls Fred’s truck)); that is, 80 percent of the Arabic children and 35 percent of the English children. (38)

Like English-speaking children, Arabic-speaking children tend to ignore an explicit antecedent (i.e., Mansour’s jeep) and instead reconstruct the VP as having a (p. 292)

bound variable reading (i.e., Mohammed’s jeep). This is consistent with the notion that children prefer binding variable identity even over a simple PF copy with a strict interpretation of a filled possessive. The fact that they will impose the bound variable reading by altering the representation antecedent ‘Bill’s truck’ to ‘Fred1 pulls his1 truck’ fits the claim that they are seeking a Strict Interface like the SIH whenever there is possible contextual support. Why do adults not allow this default as well? The children seem to be able to delete a possessive determiner where the adult English speaker requires it. The answer is not obvious, but if the parametric choice of DP over NP is not fully realized yet, then it is possible that the child retains the NP option with no determiner which can fulfill the Strict Interface for a bound reading, and this motivates determiner deletion in comprehension.

12.3.2 External reference in VPE The matter is not quite settled, though. Vasić, Avrutin, and Ruigendijk (2006a, 2006b) gave children sentences like (39), where the grandfather scratched “a third external person like the uncle’s dog,” quite at odds with adult English, but not Japanese (as we shall see). (39)

Page 19 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition When given a choice between grandfather or uncle, 37 percent of Dutch 5-year-olds took the alternate contextually available reference ‘uncle’ rather than the binding variable reading ‘grandfather’; when given a choice of ‘boy’ or ‘uncle’, 76 percent took the alternate reference ‘uncle’. This interpretation is also compatible with the view that the child reconstructs does that with a pronoun that allows broader inferences. (40)

Acquisition of more complex structures with antecedent-contained deletion points in the same direction.

12.4 Antecedent-contained deletion, parallelism, and locality Our central claim that VPE manifests a Strict Interface between syntax and semantics can be further complicated by a prior operation: quantifier raising. Sentences like (41a) entail an infinite regress if reconstruction is carried out: (p. 293)

(41)

The solution is to raise the quantifier beforehand and leave only the verb to be copied.16 (42)

It involves LF extraction to a higher node prior to VPE. In that sense, it is a complex interface between syntax and LF, and therefore we can expect a more complex acquisition path. In an elegant experiment Syrett and Lidz (2009) demonstrate that 4-year-olds carry out both quantifier raising (QR) and VPE, supporting again the UG-guided nature of the process. (43)

Page 20 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

The correct answer is that Miss Red jumped Frog 2, 3, 4 but not Frog 1, therefore Miss Black jumped over the same subset of frogs. Syrett and Lidz show that QR must be an available UG option because children apply it at the age of 4 years, despite the rarity of antecedent-contained deletion (and other QR environments). In particular, they showed that in (44) children allowed QR to raise both to an intermediate position with embedded scope ‘every book that Goofy read’ or to matrix scope ‘every book that Scooby wanted Goofy to read’. This is important supportive evidence for the presence of QR, but to ascertain that the children have the adult grammar, they must also exclude what adults exclude. The sentences in Experiment 3 were ambiguous, such as: (44)

However, the sentences in their next experiment are typically reported to be unambiguous. This lack of ambiguity is classically taken as evidence that QR is clausebounded. (45)

Notably 65 percent of children allowed exactly the wide-scope QR over tensed complements too, which adults reject (46b). (p. 294)

(46)

As Syrett and Lidz point out, this violates subset theory; learnability is maintained only when you expand the set of readings from evidence—the absence of negative evidence makes it impossible to restrict the set of possible readings (see Fox 2002). They make a variety of suggestions about parsing limits or discourse pressure to explain why children do this. In contrast, we claim that the structure of UG itself should accommodate this variation: Strict Interfaces, once fully applied, will eliminate these default readings. If again, we claim that the child uses ‘and Scooby did that too’, then a wide range of interpretations Page 21 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition become possible, including ‘Scooby said that Goofy read…’. The idea that general inference is invoked is supported by the fact that the meaning reconstructed is more, not less, complex than the adult grammar delivers, and it incorporates as much of context as possible, that is, the larger context of saying, not just the narrower context of reading. Supporting this approach is the fact that adults do precisely the same in a significant minority of cases, as we predict under the default pro option: 19 percent. Moreover, they do something similar in the original experiment. Syrett and Lidz added a further control to see if a simpler form of coordination would allow pure VPE, which is in the spirit of recognizing conjunction as a default alongside the Strict LF we argue is another default, which would translate (43a) into (47). (47)

Their adult controls seem to have made such a substitution. They comment as follows: (48)

This result follows from a larger theory of defaults where conjunction or parataxis plays a central role and which is supported in work for over forty years in acquisition, parsing, and fieldwork (see work on recursion in Amaral, Maia, Nevins, and Roeper 2018 and others). The parallelism would restore the full VP without the operation of raising. The imposition of a default form of conjunction is another example of how principles of grammar continue to govern behavior that is sometimes seen as “performance errors”.

12.5 Sluicing Sluicing entails both wh-movement and ellipsis. We noted Jensen and Thornton’s (2008) claim that wh-answers were correctly minimal (‘what did you eat?’ → ‘tomato’) only when children showed knowledge of sluicing which involves both wh-movement and ellipsis. (p. 295) In fact, early instances of sluicing include those from 2- to 3-year-olds where there are hundreds of ‘know what’ examples. (49)

However, other wh-words such as why and when are also observed. (50) Page 22 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

The case presented in (50c) involves more because the child must convert his question into an elided declarative ‘I know why you can work this thing’, a topic to which we return. Jensen and Thornton claim that simple VPE I did (pro) is available very early but then shifts when sluicing arrives. That is, sluicing entails a full reconstruction of the complement. When it becomes available, then I did pro is shifted to a full VP: I did [VP]. These hints from the naturalistic data are buttressed by experimental evidence from Lindenbergh, van Hout, and Hollebrandse (2015). They demonstrate that 5-year-old Dutch children comprehend sluicing in a contrastive pragmatic environment. They contrasted pictures where a hidden person painted a guitar and another painted a flower, and yet another was holding a flower, then asked: (51)

The accuracy was around 95 percent: the children clearly understood sluicing (only the person painting the flower is intended) and, moreover, they generally produced it as well in an elicited production task. These subtle interactions indicate strongly that children, while they can project a broad pronominal, proceed not only to an embedded IP but to a very precise CP representation linked to a sluiced complement.17 (p. 296)

12.5.1 Verb phrase ellipsis summary In sum, child external references in elided VPs can go beyond what adults allow, which suggests that the apparent binding variable reading as well could have a different pronominal origin that does not require the parallelism of a binding variable at LF entailed in Kennedy’s (2008) SIH, but is just a pronominal choice. How does the child ever get rid of this non-parallel external reference reading? The pronominal path involves Page 23 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition an Open Interface, supplemented by inference. Our answer is that a Strict Interface when recognized will overrule the Open Interface and impose full parallelism for VPE and therefore must be biased against any possible non-parallelism that an Open Interface entails.

12.6 From argument drop to the NP/DP distinction If the argument for a pronominal instead of VP works to explain departures from adult VPE, then we should expect it to be present when just an object, which is less than a VP, is elided. While some grammars allow missing objects which can be filled in by context, in English we have what seems to be a process of intransitivization that allows an indefinite reading, where the object is an unprojected argument that is automatically unspecified, as in ‘John wants to eat’ = eat something. A non-existent empty object could be where the acquisition path starts, but not necessarily. While Saito (2007) argues that there is complete argument drop which effectively converts the sentence into an intransitive, Hoji (1998) argues that some form of pro is available for Asian languages. Huang (1982) provides a typological backdrop. He argues that there are “hot” and “cool” languages where the former give priority to contextual reference, hence favoring the projection of a pro that allows definite reference. That is, in many “hot” languages you can say (52a), while in English one must say (52b): (52)

Languages like English block an elided object, while it would be pragmatically superfluous in Chinese. Pérez-Leroux et al. (2018) argue at length for the presence of an object pro in early grammars that leads to definite reference. They use a wide range of evidence under a principle of Default Transitivity. A sharp example comes from Pérez-Leroux et al. (2008). (53)

Page 24 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

18

If the child uses an object pro, then it should refer to discourse or context and book is the clear, relevant, salient topic.19 The answer would be “no she’s not”, because she is not reading the book, but if an intransitive or an unspecified interpretation is projected, then the object can be anything in general, so the answer should be “yes she is”, because she is reading the newspaper. Thirty percent of English children, in various experiments, answer “no” suggesting object pro. Pérez-Leroux et al. claim the principle of Default Transitivity leads to object pro.20,21

12.6.1 Japanese argument drop The pro option leads to something deeper: a parametric contrast between NP and DP grammars explored by Bošković (2005, 2016, and references therein). Languages that allow empty objects allow empty determiners. A fundamental issue is: a child must decide if determiners will be present or absent. Is it an NP or a DP grammar which separates English and Japanese? Matsuo (2007) tested argument drop with both English and Japanese children using sentences like (54), which children were asked to confirm or deny.22 Fourteen English children between 4;9–6;9 years and nineteen Japanese 3;7–6;11 years heard this scenario. (54)

Page 25 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Japanese adults, Japanese children (70 percent), and a significant subset of English children (30 percent) say “yes”. (p. 298)

(p. 297)

In Japanese, the main verb is repeated and the grammar fits the Pérez-Leroux et al. account of a null object pro. (55)

This produces the indirect potential for apparent VP-anaphora if pro has an open interpretation. The Japanese children used an open pro substantially more than the English children which would be compatible with the view that the Japanese children reconstructed just the noun ‘fish’, which is possible in Japanese but not English. The 30 percent English response remains significant and suggests that they could still have access to that grammar type as well, but it is not the primary grammar. It is another indication of the presence of Multiple Grammars. In another context where the object was totally different, Matsuo found smaller but significant groups in both languages (25 percent English, 40 percent Japanese) said “yes” in this scenario. (56)

This result is again not surprising if the children actually generate an empty pro (did eat pro = eat something), which can be interpreted through context. However, adults do not do this in Japanese, preferring the bound-variable reading. If they choose full VPE identity, the transitive reading is required. The fact that there is a minority of child responses in both languages for non-binding variable interpretation suggests that both transitive and intransitive grammatical options could be available to children.

12.6.2 Strict Interface version of argument drop We have examined the Japanese option under the notion that a pro is projected. Could there be a Strict Interface at LF as well? Bošković (2016) argues that argument deletion is directly linked to formation of a semantic at LF which in turn permits sloppy

Page 26 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition identity. This is presented as an alternative to universal pro in this position. We will just sketch how acquisition connects to this evolving perspective. Bošković shows that languages with no DP permit argument drop which in turn allows sloppy readings. He argues that it is directly linked to an LF-moved NP with the semantic properties of converted into at LF. In his words (p. 13): “Simplifying somewhat, in the syntax itself argumental traditional NPs are then of type e in DP languages and of type in NP languages. D turns NPs of type to in DP languages; while in NP languages this is accomplished via type-shifting.” Then he claims: argument ellipsis affects elements of type . (p. 299)

To which he adds the further claim that:

only elements of type can be copied in LF. In other words, an automatic LF operation does for NP languages what DP does explicitly in the syntax for DP languages.23 While the purpose of his claim is to capture sloppy readings, we suggest that it could extend naturally to capture the definiteness property entailed in contextually supplied objects. We will not explore the details here but simply assume that a parallel argument for LF interpretation leads to the capacity for reference to contextually salient objects if a child has an NP-grammar available. Therefore, children, if they are in an NP-phase or retain that alternative, could spontaneously project an LF even though there is no triggering evidence from adults in their own language. This would then justify saying “no” in Pérez-Leroux et al.’s (2018) experiment because the children have projected an LF which permits the definite reference. This is then the exact same kind of Strict Interface correlation we found in full VP-ellipsis, which entails an LF representation under Kennedy’s Strict Identity Hypothesis, and which we argued could be spontaneously applied because it is a UG goal.

12.6.3 Reproducing the VP acquisition path in VP-ellipsis? Why would English children who themselves use articles, other determiners, possessives, and adjectives not reconstruct them in VPE? An intriguing hypothesis is as follows: (57)

Children all go through an early stage where maximal projections are absent in production and they use just heads [V [N]] instead of [VP [NP]]. They first analyze “want cookie” as [V [N]] not as [VP [NP]]. That option would lead again to the following. Page 27 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (58)

The VP ‘found a blue fish’ can be analyzed as [V [N]] = [find fish] at the ellipsis site if it fits the context. Thus, they might begin VPE reconstruction with minimal representations as well, in effect reproducing the initial acquisition path where first they use V but not VP and N but not NP or DP. Moreover, the bare object representation continues to be needed in adult English as an input to compound formation: make cookie => [[cookie-maker]]; a cookie cannot be incorporated *[[a cookie]-maker]] (see Roeper and Siegel 1978; Harley 2009). Therefore, it should not be surprising that children would retain the non-Maximal Projection grammar as a possible default which applies in VPE contexts. (p. 300)

12.6.4 Triggering full VPE

The full DP form of VPE would have to emerge. What forces it? It could be because, as we have suggested, the acquisition device seeks a Strict Interface which would be a fully parallel VP in the ellipsis site: faithful reconstruction of the VP would entail the full DP form. Again, this ideal Strict Interface is playing a role in learnability by leading to the rejection of forms that fail to honor parallelism. Nevertheless, the child needs an experience which forces the revision of grammar. Suppose the child hears this question/answer between adults. (59)

If the adult says “no”, then this would be evidence that simply reconstructing sweater would be actually wrong, since he should have said “yes” if that were even an option in his grammar. Now the Japanese NP option is blocked. Note again that this approach is independent of statistical frequency and depends on the dialogue experience—more likely than one might imagine—as children naturally pay attention to the focus of adult conversations. Such triggers should be carefully imagined for all complex grammatical decisions.

12.7 Null complement anaphora A host of exceptions to strict reconstruction dominate the current literature (van Craenenbroeck 2017). Many of them are far too arcane to be obvious triggers a child would use to change their grammar. Others, however, as some argue, remain central to grammar and therefore could be the basis upon which the child—or the adult—organizes their entire grammar.

Page 28 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition Consider NCA. It involves contrasts like (60): (60)

(60a)—which a child would never hear—lacks the licenser to but (60b) works without it, by projecting pro into the ellipsis site and coindexing with VP1. Therefore, no isomorphic reconstruction site occurs and no precise functional category licenser is necessary.24 How (p. 301) does the child keep NCA limited to particular lexical items? The child must not decide from a few examples that a licenser like ‘to’ is never necessary because that would generate (60a). The early evidence on ‘try/want’ (see below, this section), but no other verbs, suggests that they do in fact avoid a broader generalization to all verbs when they hear NCA. We can capture this pattern if we assume that UG guides the child to limit the default options to specific lexical items wherever possible. In this case, the option of not reconstructing an ellipsis site in full should make it a first preference for a child because it looks representationally much simpler. We noted above that children never allowed ellipsis with ‘being’, but we do find null complement anaphora with ‘trying’ and we find hundreds of cases among 2-year-olds of all the options ‘try’, ‘try it’, ‘try to’, and ‘start’. This suggests that they honor lexical restrictions from the start: (61)

(MacWhinney 2000) If 2-year-old children have both VPE with strict reconstruction and the anaphoric option, they must be defined in relation to each other in order to prevent overgeneration. Why should we have a difference between ‘refuse’ and ‘attempt’ (i.e. ‘he refused’ vs *‘he attempted’ only ‘he attempted to’)? Note that L2 speakers easily say incorrect things like ‘do you want?’. It is notable however that we simply do not find stranded want in the database for children. This leads to the conclusion that NCA seems to depend on specific Page 29 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition verb-by-verb triggering to which the child is sensitive, if not the L2 speaker. This is not surprising: anything that is lexically specific can allow the presence of another grammartype within the language, just as pro-drop is restricted in English to a few verbs like ‘seem’, ‘look like’ but not ‘happen’, or ‘matter’. Again, if we find an evolution in grammar from pro to more precise reconstruction, we can maintain the view that both options are independently in UG, not written into a single form of VPE. That would lead to this conclusion: NCA is a potentially lexically restricted reflection in the adult grammar of an earlier stage in acquisition which has roots in another typology where to is not necessary.

12.8 Noun phrase ellipsis: Partitivity Our approach will gain force if we can show that NPE also has a pronominal alternate and a Strict Interface at LF. If correct, our first hypothesis should be that NPE may exhibit a syntactically simple pro at the first stage. Whether one projects pro or a copied/ deleted (p. 302) trace-NP, we can ask: is there an LF goal which children seek to impose on ellipsis as a Strict Interface? A number of scholars have argued that children’s NPE is linked to partitivity (see Sleeman and Hulk 2013 and references therein) which again may be linked to an LF projection.

12.8.1 Early evidence of partitivity What is the path for NPE? Among the first dozen words of a child is often ‘more’. While it is hard to prove that there is an invisible N [more [N]] rather than a broadly construed Expressive about an Event, an [N] could be what one would expect if the child pursued a direct relation between invisible information (i.e., pro) and context. If the child projects empty [N], how rich is the representation in the ellipsis site? It is not just anything for adults. For instance, from discourse it must be the highest NP. Take the following sequence. (62)

This refers only to juice, not cake, because only the highest NP1 is relevant in [NP1 juice [PP next to [NP2 cake]]]. An anti-pragmatic experiment could test this, but has not been done: (63)

This question asks only if you want to eat ‘nails’ not ‘cake’, while cake is the pragmatically natural alternative. Only when a child can show this knowledge do we have Page 30 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition clear evidence of syntactic reconstruction that the child obeys despite an anti-pragmatic context.

12.8.2 Licensers: Inflection, possessives, agreement The second question is what licenses NP ellipsis. In English, we find comparatives, numerals, and possessors will license it, and arguably each of those concepts—degree phrase, numeral phrase, possessive phrase—can introduce a functional node which carries a licensing capacity. (64)

Sugisaki (2005) has done a study noting that children never say *‘five ones’, which suggests that from the outset ‘one’ itself is in a numeral phrase, not a possible lexical item in an adjective phrase, which she might think if she hears the one hat. If ‘one’ were an adjective, we would predict that other adjectives could appear, but they do not for either adults or children. (65)

(p. 303)

Under Merchant’s (2001) approach, each licenser must carry an E-feature on a

functional category that specifically licenses a deletion operation. In fact, the possessive licenser seems to be extremely early. Jensen and Thornton note possessive ellipsis from a 2-year-old: (66)

A more interesting case is African-American English (AAE). Conner (2015) has shown that AAE children, whose exposure to ’s is more limited, consistently project ’s in ellipsis environments while it is absent otherwise. That is, children even say recursive things like (67a), or even phrasal cases like (67b) which involve phrasal possessives. (67)

Page 31 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Nevertheless, the ’s suddenly appears before ellipsis. As Green (2011) and Conner (2014) show, AAE children do not say sentences like (68a) but (68b): (68)

Thus, AAE, despite the rarity of possessive ’s, illustrates the requirement of a specific licenser, which has a Merchant-style E-feature, following Conner (2014). This is perhaps the sharpest evidence that a particular morpheme can carry licensing capacity.25 In languages where adjectives carry inflection, we find that ellipsis occurs easily: (69)

Here the child must note the presence of the inflection as a licenser, which is unavailable in English (see Hubert 2009). Now an interesting learnability conundrum arises: at early ages children in inflectional languages appear to be insensitive to them. Thus, a German child could say sentences like (70), lacking inflection -es, which must occur inside an NP marking gender (neuter) for ein blaues Auto. The child putatively hears: (70)

If an adult said (71) but the child does not hear the inflection, then it could think it had evidence that the adjective by itself would license ellipsis: (p. 304)

(71)

Page 32 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

If UG dictates that a licenser must be present, the child must assume its presence until it is phonologically able to identify it. In fact, in Dutch uninflected adjectives do not allow ellipsis in the adult language (Sleeman and Hulk 2013), supporting the licensing argument.

12.8.3 The movement argument Thoms (2011b) claims that ellipsis also requires some form of movement, not just the presence of recognizable inflection. If true, then the child could know from UG that a bare adjective would not license ellipsis, but only one with overt evidence of movement. This view calls for designing the grammar to have movement of an adjective to a higher AGREEMENT node where the inflection resides. However, the fact that there is a corollary acquisition argument—the absence of inflectionless ellipsis errors in AAE— should be seen as striking evidence on behalf of Merchant’s argument for a licenser as well as compatible with Thoms’ proposal that movement to a higher node carrying POSS0 is present. We have shown that when unmoved inflection is present, children do indeed fail to elide *John is being bad, but I’m not being. However, careful work in Dutch and French by Sleeman and Hulk (2013), where they summarize much work beyond their own, shows early stages where exactly the licensing features of agreement are not found in Dutch (and French). (72)

Page 33 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition They comment: “In child L1 Dutch, noun ellipsis is apparently not licensed by inflection (neither is it in the adult language), because adjectival agreement (and correct gender assignment) is not target-consistent before the age of 7 (Blom et al. 2008). The schwa and common gender are the default options until that age.” While it may seem that agreement could not be present here, it is still not unreasonable to suppose that it is still satisfied formally by a schwa, given the cross-linguistic comparison with English. (p. 305)

12.8.4 Semantics and partitivity Sleeman and Hulk (2013), following Valois et al. (2009) and Algueró et al. (2012), argue with sharp early evidence that toddlers’ single words entail a partitive reading. The speaker and the hearer share a common-ground assumption that a set exists—or minimally a focal contrast. Words like ‘other’ directly imply such contrast. Their data show that French children use ellipsis to communicate a partitive meaning at 1;1 years, and Dutch children at 2;4 years (p. 11): (73)

They list a range of words that are typical, such as encore ‘more’, nog ‘still’, ook ‘also’, and many others. How should we define the semantic dimension? Gagnon (2013b) offers this definition of partitivity based upon Barker (1998): Page 34 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (74)

In effect, then there is a presupposed set x of which y is a subset (see n. 27 for discussion of other options and formalism). For our purposes, we will avoid questions about whether LF, information structure, or semantic arguments are involved and simply propose that the child, when empty structure inside the NP arises, will assume a Strict Interface with some restrictive semantic representation.26 (p. 306)

(75)

This Interface allows the child to immediately project the ellipsis/partitive connection. Gagnon argues that the ellipsis involved is a lexically marked argument of the quantifier which explains its absence for some quantifiers like ‘all’, ‘the’, ‘a’, and ‘every’. English evidence supporting this account can be seen in the following cursory exploration of the contrast between ‘want more’ and ‘want all’. We found over 100 examples of ‘want more’ at the 1- and 2-year-old stage. Here is a group from one child Sarah, including assertions and questions (MacWhinney 2000): (76)

We found zero examples of ‘want all’ with ellipsis with any child, but always with a filled NP-argument and often an overt form of partitive. (77)

Page 35 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

An interesting case arises where NPE occurs inside the argument. (78)

This data, once again, occurs remarkably early and with remarkably precise languagespecific constraints.27 (p. 307)

12.8.5 What defines the partitive set? The notion that partitive/ellipsis connection involves a subset/set relation leads to a new question: what determines the relevant set in context? Wijnen et al. (2003) asked specifically about discourse reconstruction when the context offers multiple possible sets. Three kinds of pictures were presented with the following scenario and two sorts of sentences.28 (79)

(80)

Page 36 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Reconstruction requires both a noun and an adjoined PP. There are three possibilities: (81)

The yes/no questions in (80) were asked of each picture for children between 4 and 7 years in English and Dutch.29 (p. 308)

Page 37 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition The results showed an interesting divergence from adult grammar. A few children gave “yes,” “are two __ upside down” for the third picture, but mostly correct “no”. A significant number (45 percent Dutch, 36 percent English) allowed (80b) to be “yes”. In effect, as they argue, the noun kids is obligatorily reconstructed, but not the adjunct PP in the sandbox. Children by 4 years are clearly able to reconstruct ellipsis (N = kids, ≠ adults), against the suggestive pragmatics of the picture (adults who are upside down). See Table 12.1 for detailed comparisons. Table 12.1 Average percentages of ‘yes’ responses broken down over condition and language from Wijnen et al. (2003) 2 kids (outside)

2 kids (inside)

2 adults

Dutch

45.3%

93%

38.7%

English

35.6%

84.4%

26.7%

Both Dutch and English children allow reconstruction of two kids dropping the adjunct as a restrictor (see Wijnen and Kaan 2006 for a semantic representation that goes in the direction of the approach here). Some younger children, but far fewer, allow two adults which shows context overruling syntactic reconstruction. Therefore, most of the children meet the requirement of phonological and syntactic reconstruction. How about partitivity? Given a selection of options, half of the Dutch children and two thirds of the English children are able to reconstruct the PP as well, allowing them to reject the contextually offered sets. That is, it is not a partitive reading from the set of children, but a subset of the set of children in the sandbox. Nonetheless, a significant number seem to be able to adjust the question at issue so that they can limit the restrictor of the reconstructed noun. A smaller number can adjust the obligatory noun as well: They interpret ‘two__’ as ‘two [of [set of people in context]]’. Here, the set of people include adults, who are visually upside down, so they become an acceptable partitive subset of the whole array. Thus, we have evidence of the primacy of phonological/syntactic reconstruction, and the unsurprising residual capacity to effectively adjust the at-issue content to match the (p. 309) context. One can almost see children struggling to obtain a perfect match between all dimensions of the Interface.30

12.9 Overview: The child’s discourse environment The child’s verbal landscape remains daunting: it includes interpretive dimensions that significantly magnify the learnability challenge. First general inference keeps a robust presence and is needed to filter out irrelevant aspects of context. Children’s earliest Page 38 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition utterances like ‘uh, oh’, ‘bye-bye’, ‘hi’—just like adults’—have complex meanings but show no signs of systematic grammar, so they must lean on inference. How far does it reach? Our earlier examples of tomato store or up in the air show that inference is at work where more structured recovery is available to adults. Now consider cases like ‘me, too’ that look like there must be an elided VP or clause. They are found with no surrounding discourse from the ages of 1;10 among many children. Consider adult and child uses of elliptical expressions like ‘me, too’: (82)

Very young children do the same: (83)

It is noteworthy that there are no reports that a child simply says ‘too’, which with inferences might be sufficient. They have already registered the fact that ‘me, too’ is used by adults with missing information. A glance at twenty-five early uses of ‘me, too’ finds that most of them at 1;10 years or 2 years on CHILDES are like these (MacWhinney 2000): (84)

Page 39 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition Here the whole exchange seems to be highly dependent upon obscure inferences about actions, not systematic reconstruction of verbal antecedents. (p. 310)

(85)

Note that ‘me, too’ reaches back to the sentence [go out]. One point is again of critical importance here: if all these cases were syntactically generated, then the prediction should be that they require actual clear reconstruction from full clause antecedents, which would lead us to expect them to be much later. Moreover, if were they introduced through the abstract Kennedy-type formalism, and if that formalism cannot appear until its ingredient structures (i.e., VP, operator, etc.) are in the grammar with full clause antecedents, they should again be late. We are led to conclude that they must involve a simple representation, like pro for the missing clause, and substantial inference.

12.9.1 Fragments To fill out the child’s challenge, we must now reckon with recent theoretical insights that show that these fragments are not interpreted only by inference, as is clear from work on fragments by Merchant (2001) and a comprehensive theory by Weir (2014). Weir uses the example of a child trying to do something and the parent says ‘both hands!’ which entails an elided imperative clause ‘(you must use) both hands’. Children do not balk at such utterances, but regularly seem to understand them, just as 1-year-olds grasp ‘don’t’. Do children show any knowledge of grammar with fragments? Jensen and Thornton (2008) provide this example of an embedded fragment from a 2-year-old: (86)

Weir, following Barbiers (2000) and Temmerman (2013), shows that embedded fragments only occur with bridge verbs which include ‘think’ but not ‘know’. Therefore, we predict Page 40 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition that we will never find sentences like *‘I know just a little real food (= ‘I know that they eat (p. 311) just a little real food’)’. Thus, even early fragments ultimately show the same kinds of UG constraints found elsewhere. We will not explore the details, but this theoretical work shows that fragments submit to complex analyses in terms of focus, movement, an E-feature, and clausal ellipsis which are not identical cross-linguistically.31 For instance, Merchant (2001) shows subtle language-particular effects. He shows that bare nouns may be grammatical only in some environments. While in English one can answer as (87aB), in German, where prepositionstranding is not allowed, the preposition must be supplied, as in (87bB). (87)

In other circumstances, not yet perfectly understood, English also demands the preposition (88aB), just like German (87bB), but not (88bB): (88)

It is not, however, inevitable that children immediately realize these adult uses of fragments. One form of language disorder is a failure to eliminate inference in favor of structure by some children. In fact, disordered children often answer with fragments which are communicatively adequate, but perceived by adults as grammatically imperfect. Consider what many disordered children do in answering (88aA): they answer exactly: ‘broom’ (Roeper and Seymour 1994, Seymour et al. 2005). ‘Broom’ communicates adequately—with an easy inference about instrumentality—but ‘with a broom’ is required from adults to honor the case requirements of full reconstruction (i.e. ‘she swept the room with a broom’). Why do adults allow Prep-deletion with a temporal PP noon, but reject manner Prep-deletion of with for ‘broom’? In fact, E. C. Klein (1993) observes that Page 41 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition it is a frequent L2 error as well, which suggests that it is not clear what kind of disorder it is.

12.9.2 Question Under Discussion A second factor can dramatically tilt discourse in a new direction: Questions Under Discussion (QUD). We (and a child) must always determine what the chatter is about, in other words, what question is at issue. Surprisingly, often adults shift the focus. Consider this commonplace interaction: (p. 312)

(89)

The child must infer an answer to the actual question, ‘no’, and see how the assertion carries this implication. Often the shift is oblique, which requires the inference ‘yes, but’. (90)

Can children handle such shifts? Two of our examples already illustrate the capacity. The child himself performs such a shift in (50) (Adult ‘can’t you work this thing?’ followed by Child ‘I know why’). The child has shifted from a yes/no question to the reconstruction of an elided indirect question which presupposes the truth of the prior question: ‘I know why you can’t work this thing’. In the ‘me, too’ conversations (85), the child ignores the opposite command ‘you stay here’ to a response that presupposes an elided VP from an earlier sentence: ‘I’m going outside’ to which the ‘me, too’ refers. These conversational gymnastics, coupled with the emotional shifts they entail, are impressive evidence that the child is manipulating a complex interface that depends upon both grammatical and social factors. They are far beyond ‘logical’ induction that a computer can mimic. Thus, we might say that we have here—though not fully resolved—a candidate for a Strict Interface which is present from the earliest cases: determine the Question Under Discussion before you reply. One might want to attribute this to the child from the outset and many adult expressions toward children presuppose just this kind of ability to determine what is at issue.

Page 42 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

12.10 Conclusion We have argued that children are able from the outset to license ellipsis, recover missing structure from discourse and context, and generate both an open pro and a full PF representation. The capacity seems to be immediately triggered because we find ‘I did’ from the earliest verbal interactions, though not with perfect adult results. Each of these capacities must be sensitive to language-particular licensers, varying principles of recoverability, and the abstractness of the representations achieved in cases where mismatches are allowed. We have addressed the special ellipsis properties of: VPE with antecedent-contained deletion, sluicing, argument drop, and then NCA and NPE with partitives. We have labored to embed these observations in an updated version of what has been called the Primary Linguistic Data (Chomsky 1965)—the adult input from which children must derive these structures. They involve not only the entire syntactic structure of discourse, but properties of Focus, QUD, and constraints on fragments. An important conclusion follows: To cope with an obscure verbal and pragmatic environment the child must utilize UG Strict Interfaces as goals to isolate what is linguistically and pragmatically relevant. Strict Interfaces substitute for an evaluation metric as a method to exclude grammars that allow non-adult interpretations. (p. 313)

We have argued that (a) Kennedy’s (2008) Strict Identity Hypothesis serves as an

LF interface behind sloppy identity, (b) a semantic notion of definiteness entails an invisible LF-projection for NPE, and (c) a semantic representation of partitivity is immediately available to children. These goals are realized in steps—not yet perfectly visible—which move the child from default pro representations (also used by adults to reduce complexity) to full PF copies in the elided domain. The acquisition path for the elided structure then resembles or perhaps reproduces the acquisition path for child’s first language-specific representations leading to maximal projections of VP and NP. The acquisition process obeys an overarching notion of economy that seeks to minimize use of pragmatic inference and maximize the efficiency of inherently complex linguistic mechanisms that operate both within and across modules. The path which results shows these rough stages that we outlined: (91)

Page 43 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition

Ellipsis constitutes one of the most dramatic examples of poverty of the stimulus in as much as it is built upon silence. While ellipsis is omnipresent, its variety is so intricate when pragmatic and interface factors are added that examples of each type remain very rare in what the child hears. Therefore, it must be built into UG interfaces. This extension of UG to include interfaces adds to the many arguments for the innateness of grammar.

Acknowledgements Thanks to L. Amaral, L. Green, J. Hartman, K. Johnson, J. de Villiers, A. Pérez-Leroux, A. van Hout, A. Weir, L. Renato, editors J. van Craenenbroeck and T. Temmerman, and reviewers and members of the Language Acquisition Research Center (LARC) at UMass. Thanks to Jaieun Kim for proofreading, formatting, and many other improvements. Misinterpretations and errors are mine.

Notes: (1) At least to my knowledge it has not been done. (2) See Renato (2016) who demonstrates some sensitivity to the pragmatics of topic and focus for double-object constructions among 4-year-olds, but resistance to ellipsis, indicating that the computation of topic focus and ellipsis is not initially complete. (3) We will not address comparatives or gapping (see Syrett 2016 and Bryant 2006, respectively, for interesting work which points toward an important role for semantics). (4) A major presupposition here is that interfaces must be biologically stipulated, as they are elsewhere. The links between the heart and lungs (i.e., interfaces) are just as innate as the heart and lungs, or the link between separate three-dimensional representations for eyes and hands needed in eye–hand coordination. The view that interface links are innate stands in contrast to arguments that cognitive third-factor features of language (Chomsky 2005) mean that much of grammar is not innate. See Roeper (2014) for discussion of how the grammatical interface with a cognitive notion of AGENT is modified at the interface with grammar. (5) There is now a substantial literature on Multiple Grammars (see Roeper 1999, 2014; Yang 2002; Amaral and Roeper 2014) and comments and replies in the same volume (Rankin 2014, 2016; Roeper 2016, and others). Page 44 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (6) They add “We compare our approach with accounts of ellipsis based on syntactic copying or deletion, showing that although both approaches have their difficulties, the challenges to the copying/deletion approach are more severe” (Culicover and Jackendoff 2012: 305). (7) The idea that UG has an ideal or perfect form of interfaces which constrain acquisition marks a shift in acquisition history. Early theories always assumed that ‘performance’ obscured the child’s grammar. Most of the work has shown that there are few or no performance constraints that children obey. An example of early advanced knowledge is in Roeper (1982), who found that 3-year-old children grasped the control difference between ‘show the dog jumping’ and ‘show the dog the jumping (where a non-dog can jump)’, which entailed a tight connection between the syntax of determiners and the semantics of control. The experiment indicated that interface aspects of determiners must be represented very early: the semantic constraint on control in verb–noun conversion is signaled to 3-year-olds by the determiner ‘the’. They know that there can be control in ‘John likes singing songs’, but not in ‘John likes the singing’. (8) For instance, it could be the case that sloppy identity was not available for (i) but first appears with quantificational and contextual support as in (ii). ((i))

((ii))

(9) See also Duffield and Matsuo (2001) for experimental evidence that children control this distinction with older children. (10) Other examples in Santos’ collection are ambiguous, raising the possibility that nonTense elements could license ellision (see also Thornton 2011). (11) K. Johnson (p.c.) points out that NPE lexical exceptions are not obviously moved (‘I don’t want any’), or, as J. van Craenenbroeck (p.c.) observes, obligatory ‘to’ is not obviously moved. (12) Foley et al. (2003) have a similar view that lambda abstraction is favored in order to preserve a form of economy. (13) Work by Syrett and Lidz (2009) as well as Thornton and Wexler (1999) supports this view, as well as Kiguchi and Thornton (2004) and others cited therein. (14) See related phenomena in Bever (1970) and Roeper et al.’s (2011) demonstration that children prefer “distributivity”.

Page 45 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (15) The status of all the sloppy-identity results relates to some degree to further evidence of quantifier spreading. See Philip (2011) for a summary, and Roeper et al. (2006). (16) Sugisaki (2007a, 2007b) with intriguing but limited evidence suggests that swiping is also available just over 2 years, which is possibly connected to pied-piping or P-stranding, again suggesting very subtle knowledge, very early: Aran (2;7) ((i))

((ii))

((iii))

((iv))

(17) Like Thornton and Jensen’s (2013) claim that wh-movement in sluicing is linked to VPE, Foley et al. (2003) argue that both sloppy and strict readings involve long-distance wh-extraction which accounts for the fact that children acquire strict and sloppy readings at the same time. (18) Sample picture from Pérez-Leroux et al. (2008). (19) L. Amaral (p.c.) reports that this kind of misunderstanding is very common in L2 English speakers whose L1 is Spanish or Portuguese. (20) We will not address here the complex interaction with clitics and the large literature on object clitic omission. (21) See Roeper (1999, 2003), Yang (2002), and Amaral and Roeper (2014), among others. (22) See Sugisaki (forthcoming) for further evidence of argument drop in early acquisition and a discussion of the complex agreement facts that are a putative part of an ellipsis parameter. (23) Bošković (2016) adds: “It should be noted that Tomioka (2003) (i.e. his property pro) is an important predecessor of the analysis argued for here, which situates the gist of Tomioka’s proposal within a broader perspective.”

Page 46 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (24) Weir (p.c.) pointed out that it is possible to say (i) but we note that it does not work with all verbs as shown in (ii). ((i))

((ii))

We note that (i) with whatever only works with verbs that independently allow an object: wants food/*hopes food. (25) Aelbrecht (2010) (see van Craenenbroeck 2017 for review) claims that the licenser could be non-adjacent with language-specific features. This view significantly complicates the acquisition problem. (26) Gagnon (2013b) argues for a narrower contrast between NPE and partitive ellipsis which ultimately argues that a null pronoun is present equal to ‘of them’. There are many efforts to provide formal definitions of ‘partitivity’, some interestingly related to whether a numeral can apply to a singular noun, which we will not attempt to choose among. The formalisms involved raise many obscure acquisition questions unless one argues, as we do, that the child is able to comprehend them as an inseparable cognitive unit whose parts are not subject to independent steps of acquisition. The topic requires extensive separate discussion. For our purposes, we assume that some formulation is immediately available to the child. Another case where the partitive properties are more explicit comes from Ionin, Matushansky, and Ruys (2011): (a.) PART(x) = def fZ : Z is a partition of xg (b.) A partition of an aggregate x is a set of aggregates Z such that the joint (sum) of all the elements in Z is equal to x (W Z = x) and for any two elements, w and v, in Z, the meet of those two elements is empty (w^v = / 0). 27

Where inflectional licensing is available, such as the agreement marker -e in German, the ellipsis is licensed again for child and adult. ((i))

Page 47 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Acquisition (28) Experiment pictures from Wijnen et al. (2003). (29) The experiment was repeated by Obdeijn (2005) with the same results and with van Weelden’s (2009) eyetracking study (which produced some increase in contextual reference). (30) There is a further variant which promotes such adjustments: sentences with ‘there’ (or er in Dutch). In Dutch er is obligatory, unlike English, but we will not explore the matter here. (31) Merchant and Weir argue that clausal ellipsis is coupled with a moved focus item which satisfies the E-feature.

Tom Roeper

Tom Roeper works in generative grammar on acquisition theory and experimentation and on syntactic morphology. Acquisition topics include long-distance movement, quantification, binding theory, passives, V2, small clauses, aspect, ellipsis, implicatures, and theoretical work on Multiple Grammars, Strict Interfaces, Subset Theory, and African-American English dialect. In morphology he has worked on compounds, gerunds, productive affixation, and theoretically on lexical transformations, implicit arguments, and labeling theory. In addition, he has written two popular books, most recently The Prism of Grammar, five co-edited books, including Recursion: Complexity in Cognition; co-authored an assessment test, Diagnostic Evaluation of Language Disorder; and co-edited Language Acquisition and Studies in Theoretical Psycholinguistics. Currently he is working with collaborators on the emergence of recursive self-embedding and its theoretical implications in English, German, Dutch, Japanese, Romanian, Hungarian, and Pirahã.

Page 48 of 48

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

Oxford Handbooks Online Ellipsis and Discourse   Andrew Kehler The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.13

Abstract and Keywords Despite the considerable attention paid to elliptical phenomena in the literature, the conditions under which a representation of an utterance may serve as a suitable antecedent for interpreting a subsequent ellipsis remain highly contentious and poorly understood. Focusing primarily on VP-ellipsis, this chapter surveys evidence that informs this debate, focusing on issues bearing on the discourse context in its role of making meanings available for ellipsis recovery. It argues that the evidence more strongly favors discourse-based explanations over syntactic ones, although problems for discourse-based theories do remain. Keywords: verb phrase ellipsis, identity constraint, syntactic mismatch, information structure, strict/sloppy reading, missing readings puzzle, focus and deaccenting, coherence relations, event reference, Question-UnderDiscussion (QUD) models

Andrew Kehler

13.1 Introduction WHEN contextual conditions are right, natural languages permit speakers to omit linguistic material that expresses information that they nonetheless intend to communicate to their hearers. This is a significant design feature of language, in that it allows speakers to minimize articulatory effort while still being expressive in communicating their message (Zipf 1949), at least in cases in which the hearer is able to perform the intended meaning recovery. An important class of such cases falls under the cover term of ELLIPSIS. Unlike some other forms of content enrichment (implicatures, etc.), ellipsis phenomena are among those in which the grammar itself provides a tip-off to the hearer that normally required linguistic material is in fact missing. Successful interpretation in such cases is enabled by the fact that context allows for the recovery of

Page 1 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse already activated (precomputed or otherwise predictable, and hence redundant) meanings. There are many types of ellipsis found in the world’s languages, with an impressively diverse set of constraints on their use. One instance that is prevalent in English (less so in other languages) is VERB PHRASE (VP) ELLIPSIS, which will be the primary focus of this chapter: (1)

Here, the stranded auxiliary in the second clause (henceforth referred to as the ELLIPSIS clause) marks a vestigial verb phrase, a meaning for which must be identified to recover the proposition denoted by the sentence. In many cases of VP-ellipsis, the ability to identify the meaning is enabled by the occurrence of another linguistic expression (the ANTECEDENT), which in this case is located in the first clause. In certain cases, an antecedent is capable of making more than one possible meaning available for a VP-ellipsis interpretation. This is in fact the case for (1). Assuming that the assignment of the pronoun his in the antecedent clause leads to a reading in which John loves John’s mother, two possible meanings become available for the ellipsis clause: one in which Bill loves John’s mother (the STRICT reading), and one in which Bill loves his own mother (the SLOPPY reading). As we will soon discover, the facts become more complicated when the antecedent contains multiple coreferring pronouns. Despite the considerable attention paid to VP-ellipsis in the literature (not to mention a wide variety of other types of ellipsis), the conditions under which a (p. 315)

representation of an utterance may serve as a suitable antecedent for interpreting a subsequent ellipsis remain poorly understood. Three prominent views found in the literature that we will focus on include: Syntactic: The recovery of the elided VP meaning is dependent on there being a suitable syntactic VP to serve as an antecedent in the discourse. This need may arise from a constraint that allows for deletion under identity at surface structure or syntactic logical form (e.g., Sag 1976a), or a recovery procedure that reconstructs the missing VP at the ellipsis site at one of these levels (e.g., Williams 1977b; Fiengo and May 1994). Anaphoric: VP-ellipsis is essentially a null proform, and hence interpretation is governed by the same types of processes used to resolve other types of referential expressions such as pronouns (Schachter 1977b; Chao 1987; Dalrymple 1991; Hardt 1992b; Kehler 1993a; Lobeck 1999, inter alia). Deaccenting: VP-ellipsis is essentially deaccenting taken to the point of complete lack of articulation. The constraints on ellipsis are then predicted to be based on some version of the same processes that govern the deaccentuation of overt material (Rooth 1992a; Tancredi 1992; Chomsky and Lasnik 1993; Merchant 2001). Whereas analyses in this Page 2 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse category typically posit the existence of unpronounced syntactic structure at the ellipsis site, they differ from syntactic deletion approaches in that the constraints on elision only enforce semantic identity between the elided VP and its antecedent. Note that these categories are offered only to provide prototypical characterizations of approaches, presented so as to allow comparison of their predictions without getting bogged down in the details of various particular analyses. Indeed, some analyses span multiple types; for instance, Rooth (1992a) offers a deaccenting analysis that is augmented with a syntactic redundancy condition. As the editors explain in the handbook’s introduction, the central questions in ellipsis research can be categorized in terms of three issues: (i) the structure of the ellipsis site (i.e., whether covert syntactic structure exists at the ellipsis site), (ii) recoverability (i.e., how the ellipsis site gets its meaning), and (iii) licensing (i.e., syntactic restrictions on ellipsis beyond those bearing on recoverability). Whereas this chapter will have little to say about the third question, the potential for an article entitled “Ellipsis and discourse” to shed light on the other two is dependent on to what extent discourse considerations play a role in enabling ellipsis. On the syntactic view, the interest level might be pretty low, as such analyses straightforwardly require structure at the ellipsis site (in answer to the first question) and that a suitable VP exist in the immediate context (in answer to the second question). On the other hand, the questions become more interesting on the anaphoric and deaccenting views. We will survey some of the evidence that informs these questions, focusing on issues bearing on the discourse context in its role of making meanings available to enable subsequent ellipsis. Section 13.2 first addresses these questions by analyzing constraints on the linguistic form of the antecedent and whether or not they entail the existence of syntactic structure at the ellipsis site. Section 13.3 then examines these questions by analyzing the role of syntax in shaping the anaphoric dependencies that give rise to strict and sloppy ambiguities. Consideration of the data will run us straight into the arms of fundamental questions about the interaction among grammar, meaning, and processing: To what extent are infelicitous (p. 316) examples due to grammatical constraints, versus semantic or pragmatic factors, versus mere interpretation difficulty? Whereas we will not resolve these questions here, our examination will give rise to a set of adequacy criteria with which future work can engage.

13.2 Constraints on linguistic form Determining whether or not VP-ellipsis requires a syntactically matching antecedent should be straightforward enough. We can simply look to cases in which a salient VP meaning exists in (or is easily constructible from) the discourse context, but the antecedent is not in the required syntactic form. Syntactic approaches predict that such cases should be bad, whereas the anaphora and deaccenting approaches suggest they should work. The data, however, are not at all clear. Consider the following examples: Page 3 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (2)

These examples all provide seemingly prima facie evidence for the syntactic account. Example (2a) involves a syntactic mismatch: the ellipsis clause is in the active voice, whereas the antecedent clause is in the passive. Therefore the VP necessary on the syntactic account (looked into the problem) is not present in the context, hence the example’s infelicity. The VPs in examples (2b,c) are viable antecedents, on the other hand, but seem odd on the noted strict readings: exactly what Conditions A and C of the binding theory would respectively predict if syntactic structure exists at the ellipsis site. The example of antecedent-contained deletion (ACD) in (2d) is not only unacceptable, but has the same feel of a subjacency violation as its unelided counterpart (*John read everything which Bill believes the claim that he read).1 All of this is just what we would (p. 317) expect on a syntactic analysis. On the other hand, it seems hard to argue that the event representations that correspond to these VPs are not made available by the discourse for subsequent anaphoric reference. In (2c), for instance, a discourse representation for defend George should be made available by the antecedent clause; the particular syntactic form used to introduce the event (which here, crucially, involves a name rather than a reflexive pronoun) should not matter. This would all be convincing if it were not for the fact that there is equally compelling evidence for the anaphoric theory. Consider: (3)

Example (3a) features the same passive–active mismatch as (2a), but is acceptable. Similarly, (3b,c) should result in the same respective Condition A and C violations as (2b,c), but speakers judge them to be acceptable on the strict reading. Like (2d), (3d) is a case in which the syntactic account would predict a trace violation (consider its unelided counterpart *Which problem did you think John would solve because of the fact that Susan solved?), but again it is acceptable. The anaphoric view correctly predicts these

Page 4 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse cases to be acceptable for all of the same reasons that it gets the predictions for (2a–d) wrong. How about the deaccenting theory? We can also evaluate this approach easily enough, by comparing the felicity of ellipsis examples against their variants in which the VP is overt but deaccented.2 Interestingly, not only do the judgments in some cases diverge from those for their elided counterparts, they do so in different ways depending on the type of violation. On the one hand, the deaccented versions of both mismatch cases (2a) and (3a) seem acceptable: (4)

Whereas example (4a) might be less than a perfectly constructed discourse (more on that a bit later), it seems considerably more acceptable than its unelided counterpart. On the other hand, the binding theory cases seem bad with deaccented VPs on all sides: (p. 318)

(5)

Not only does Condition A block the strict reading where we would have expected it (5a), but apparently also in (5b) despite the fact that the effect is ameliorated in its elided counterpart (3b). Examples (5c,d) show the same pattern for Condition C. What is common to all of these cases is that the normal syntactic constraints governing overt material apply to deaccented VPs, in contrast to some of the elided versions.3,4 Since the foregoing data do not settle the question, we can also look to independent criteria for additional insight. For instance, the anaphoric approach predicts that the behavior of VP-ellipsis should pattern with other types of anaphora like pronominal reference. This appears to be the case. For instance, as noted by Lakoff (1968) and Jackendoff (1972), VP-ellipsis and pronouns may be cataphoric in similar circumstances, as shown in examples (6a–d). (6)

Page 5 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

Cataphoric VP-ellipsis is allowable when it is embedded as in sentence (6a), as it is for pronominal reference in sentence (6b). Cataphora is not allowable when the ellipsis is not embedded as in sentence (6c), on the other hand, as is the case for pronominal reference in sentence (6d). This pattern receives no independent explanation on a syntactic approach. It is also mysterious on a deaccenting approach: Whereas the anaphora and deaccenting analyses share a number of properties, cataphoric deaccenting of a VP is not permitted in contexts in which its meaning has not already been given (e.g., via previous evocation) in the discourse at the time of mention: (p. 319)

(7)

That is, assuming that going to the mall has not been previously discussed, cataphoric deaccenting is infelicitous in (7a); the fact that the VP meaning is not Given means that accent is required (7b). The version with VP-ellipsis (7c) is nonetheless acceptable under the same conditions. So whereas we normally think of the constraints on deaccenting as being less restrictive than for VP-ellipsis (see n. 2), here is a case in which the constraints on deaccenting are more restrictive. Again, this is expected if VP-ellipsis is anaphoric in the same way that pronouns are. A second characteristic of anaphora is that, while preferring local antecedents, it can access referents evoked from clauses other than the most immediate one. Such reference is not uncommon for VP-ellipsis; Hardt (1990) reports that 5 percent of the examples in the Brown corpus have an antecedent that is at least two sentences back in the discourse, as in example (8). (8)

Likewise, the subject pronoun he in the final sentence in passage (8) locates its referent from two sentences back. Page 6 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse A third property of anaphora is that it allows for split antecedents: in (9), the pronoun them successfully refers to Mary and Fred as a group, even though they are mentioned from distinct constituents. (9)

Analogously, the VP-ellipsis in (9) refers to a plural referent composed of meanings evoked by non-conjoined VPs. Fourth, under certain conditions an anaphoric pronoun can access a referent that is not coreferent with, but is nonetheless semantically derivable from, an antecedent expression: (10)

Here, the pronoun there successfully refers to France, which has a transparent semantic relationship to the meaning of the antecedent Frenchman. Similarly, Webber (1978) offers an example of a related sort involving VP-ellipsis: (11)

Whereas the antecedent clause does not contain a VP that denotes the intended referent —i.e., the meaning of dance with Irv—this meaning can be transparently derived from the (p. 320) meaning of the antecedent clause. Such behavior is thus consistent with the anaphoric account in principle, whereas it is more difficult to explain on a syntactic analysis. Finally, and more controversially, there is evidence that VP-ellipsis can refer to situationally evoked referents; examples (12a,b) are from Schachter (1977b): (12)

Obviously there is no syntactic VP available to serve as an antecedent in such examples. However, Hankamer (1978) disputed Schachter’s data, arguing that such cases are, in his Page 7 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse terms, either formulaic or conventionalized, occurring only as ILLOCUTIONALLY CHARGED EXPRESSIONS and not generally as declarative statements or informational questions. For instance, the elliptical expressions in (13a,b) are infelicitous, even though the contexts are the same as for Schachter’s examples: (13)

More recently, these data have been argued by Miller and Pullum (2014) to admit to a parsimonious explanation whereby VP-ellipsis with situationally evoked referents is subject to the same discourse conditions as cases with linguistic antecedents. Specifically, they argue that VP-ellipsis requires that the discourse context give rise to a highly salient set of alternatives that contains the meaning of the ellipsis clause (the nature of this set depends on whether the ellipsis is subject-focused or auxiliary-focused; see the discussion of Kertz 2013 below, and of Question-Under-Discussion models in section 13.3). The illusion of a constraint against VP-ellipsis with situationally evoked referents results from the fact that only in limited circumstances (e.g., in (12a,b), but not (13a,b)) does the discourse context raise the salience of the alternative set to the required level. All of these properties are consistent with an anaphoric theory of VP-ellipsis, rather than being indicative of a more restricted, local syntactic operation. Crucially, such behavior is not found with certain other forms of ellipsis, such as bare remnant ellipsis (e.g., gapping and stripping), which are far less sympathetic to non-local antecedents, cataphora, situationally evoked antecedents, and so forth. This is an important observation, since it might be tempting to draw conclusions about one form of ellipsis from the behavior of another. Different forms of ellipsis have different syntactic, semantic, and discoursal properties, and hence theorizing should not be guided by the assumption that a unified theory of ellipsis is possible. That is to say, there is no a priori reason why it could not be the case that VP-ellipsis is anaphoric and other types of ellipsis are not. As an ensemble, the aforementioned facts create a quandary. It seems clear that, as it stands, syntactic approaches undergenerate whereas anaphoric approaches overgenerate. Two types of approach to reconciling the facts therefore suggest themselves. First, (p. 321) proponents of syntactic accounts (and to some extent, deaccenting approaches that posit unpronounced syntactic structure) might say that the fact that we see any apparent effects of syntactic structure demonstrates that syntax must be relevant to licensing and recovery, since any sensitivity to syntax would seem out of the reach of anaphoric accounts. Proponents of syntactic accounts would therefore seek to explain why examples like (3a–d) are judged as grammatical even though they are not. On the other hand, proponents of the anaphoric account might argue that the paradox presented by (2a–d) and (3a–d) is fundamentally more difficult to resolve on a Page 8 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse syntactic theory than on an anaphoric one, since if examples like (3a–d) are ruled out by the grammar, then no semantic or pragmatic factors should be able to rescue them. Proponents of the anaphoric account would thus claim that all of the cases just reviewed are potentially acceptable, and look for independent causes for the unacceptability of (2a– d) and other problematic cases. This opposition gave rise to a lineage of research that bounced back and forth between the two approaches, in search of the key factors that determine VP-ellipsis felicity. Kehler (1994, 2000, 2002), concluding on the basis of examples of the sort in (6)–(12) that VPellipsis is anaphoric, took the route of attempting to explain the contradictory data with a theory that offered an independent reason why some examples are ruled out. Specifically, he argued that whereas anaphoric interpretation requires no syntactic reconstruction of an antecedent VP, there is a second, independent interpretation process that does under certain conditions. This process—the establishment of DISCOURSE COHERENCE— requires that constraints associated with one or more COHERENCE RELATIONS be established between the antecedent and ellipsis clauses, and part of this process is the identification of the arguments to which these constraints apply. The constraints associated with certain relations—for instance, CAUSE–EFFECT relations such as Violated Expectation (3a), Explanation (3b,d), and Denial of Preventer (3c)—only require access to the propositions denoted by the clauses. Since anaphoric resolution of the VPellipsis will result in a complete propositional meaning for the ellipsis clause, no syntactic construction of the VP is predicted to be necessary. On the other hand, the identification of the arguments to RESEMBLANCE relations, including the Parallel relation operative in (2a–d), is a more complex process that requires that subsentential arguments be retrieved from each clause and their correspondences identified. It is this requirement that triggers the recovery of syntactic structure, explaining the sensitivity we see to syntactic constraints in (2a–d). Although Kehler’s analysis potentially explained both the mismatch data and the binding theory violation facts, it was unsatisfying in several respects. For one, it is not obvious why syntactic reconstruction would be necessary for the establishment of Resemblance relations, as it is possible to have Parallel coherence without parallel syntax in discourses with no ellipsis. Second, as noted by Kehler (2000), there are cases which appear to have properties of more than one relation (all examples and judgments from the paper): (14)

Page 9 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse Examples (14a,b), which contain passivized and nominalized antecedents respectively, seem more marginal than their counterparts in (14c,d). Kehler conjectured that despite the fact that the connective in all four cases signals a Cause–Effect relation, a Parallel relation is also triggered in (14a,b) in light of existence of the corresponding agents and parallel tenses (hence the congruence with the adverbial too). On the other hand, (14c,d) lack these triggers for Parallel, and hence no reconstruction occurs. While this conjecture is possible in theory, the idea that multiple coherence relations would be operative in some cases is not entirely natural, and its use here is somewhat post hoc. So whereas the data suggest that coherence relations may be a relevant factor, the muddled middle ground presented by (14a–d) indicates that the coherence analysis might not make quite the right slice through the data. (p. 322)

A series of papers by Lyn Frazier, Charles Clifton, and colleagues took the opposite approach, particularly in advocating a syntactic theory that sought to explain mismatched cases by way of other means. To start, Frazier and Clifton (2006) set out to evaluate Kehler’s coherence theory experimentally, concluding that there was little to support its predictions.5 Their first experiment, for instance, used examples that paired cases like (14a), shown in (15a), with clear Parallel examples (15b): (15)

After hearing these sentences read aloud, participants were asked to say whether they “got” the sentence or did not “get it.” The results showed no benefit for examples in which the connective indicates the Explanation relation (15a) over those in which it indicates Parallel (15b); in fact, there was a small effect in the other direction. Noting that some of their causal stimuli (e.g., (15a)) may have been pragmatically odd, they ran a follow-up written acceptability study for which they modified their stimuli (e.g., changing because to even though in (15a)) and paired the mismatch cases with syntactically matched versions. Whereas the mismatched examples were rated significantly worse than the matched ones, the stimuli that employed Cause–Effect relations were again rated no better than the Parallel cases. Per the foregoing discussion, the results of these studies were not a surprise on Kehler’s analysis, as Kehler (2002) had already acknowledged the unacceptability of cases like (15a) (cf. (14a)) and attempted to explain why they were bad. These experiments therefore confirmed previously acknowledged intuitions about such examples. Importantly, Frazier and Clifton did not conduct experiments that paired examples like (15a,b) with those like (14c), and hence the central contrast between examples like (2a) and (3a) upon which Kehler’s theory was based was left unresolved. A second paper published around the same time (Arregui et al. 2006) aimed to explain mismatch facts with a different tack—one that maintains the fundamental assumptions of Page 10 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse a syntactic approach. Specifically, Arregui et al. posit the RECYCLING HYPOTHESIS, which states that although VP-ellipsis imposes a requirement for a suitable syntactic antecedent, the language processor has a repair strategy for structure that allows it to “recycle” whatever linguistic materials are at hand to create one when one does not already exist. The degree of (p. 323) acceptability of VP-ellipsis is then predicted to be proportional to how close the existing syntactic material in the discourse context is to the VP required to license the ellipsis, and hence the complexity of operations that need to be performed to fashion this VP from those materials. This characterization, according to Arregui et al., means that we should expect declining rates of acceptability across the four types of example in (16a,d), as they require increasingly greater amounts of repair. (16)

This prediction was confirmed in an acceptability study.6 A few provisos are in order, however. First, it is not completely clear that their results refute the predictions of an anaphoric theory. For one, as the authors themselves note, the difference between (16a) and (16b) does not concern whether a matching VP exists, but whether it appears in the expected position—an issue bearing on referent accessibility, and not syntactic mismatch.7 Second, it is not at all unnatural to think that form of reference would impact the relative level of accessibility of events for subsequent reference—for instance, whereas the adjective unapproachable in (16d) might activate the concept of approaching in the hearer’s mental model of the discourse, it might do so less strongly than when evoked by a verb as in (16a,b).8 On the assumption that successful anaphoric reference depends on a sufficient degree of activation of and/or attention to the referent in the hearer’s mental model of the discourse, such cases might therefore simply require additional contextual support to become accessible.9 Third, when making the case that a (p. 324) form is not anaphoric, it is important to have a control condition that uses a form that is uncontroversially anaphoric. Do it anaphora is a natural choice; consider the variants of (16a,d) in (17a,d): (17)

The question is whether we would find a similar reduction of acceptability in (17a,d) as was found for (16a,d). If we do, then one cannot attribute the effect to a condition on Page 11 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse ellipsis. Indeed, it would suggest instead that the linguistic form of the antecedent affects the accessibility of the event referent similarly in both cases, which would actually support the anaphoric view of VP-ellipsis. Resolution of the issue awaits further study. Finally, Arregui et al.’s study also did not attempt to resolve the conflicting evidence regarding mismatch: there is no difference in acceptability predicted between examples (2a,d) and their counterparts in (3a,d), which despite the results of Frazier and Clifton (2006), intuitions suggest still exists. In another paper, however, Grant et al. (2012) posit an analysis based on what they call NON-ACTUALITY IMPLICATURES (NAIs). They note that in many cases of mismatch offered in the literature, the antecedent clause does not assert that the event it describes actually happened, which in turn could be taken to implicate that it did not. For instance, the antecedent clause in (3a), repeated below as (18), might be taken to suggest that no one did look into the problem, an implicature that the ellipsis clause reinforces. (18)

Grant et al. (2012) state their hypothesis as follows: NAI hypothesis: The presence of an NAI will set up an alternative which implicitly focuses the antecedent clause, and sets up a potential QUD [AK: Question Under Discussion], thereby motivating and guiding the processing of a later clause which comments on the QUD, especially when the antecedent of an ellipsis needs to be repaired. Note that, as indicated by the final clause, Grant et al. build their analysis on top of the Recycling Hypothesis. As such, despite the semantico-pragmatic character of the process of providing answers to QUDs that arise from implicatures, their analysis is based on a syntactic, and not semantic, interpretation procedure. NAIs only serve the role of “motivating and guiding” the processor to repair an ungrammatical ellipsis.10 Grant et al. argue that their analysis is supported by the results of an acceptability study they conducted using stimuli like (19a,d): (p. 325)

(19)

As expected, they found that mean acceptability ratings for the mismatch cases were higher when an NAI was present (19b) than when it was not (19a). The pattern was reversed for the match cases: Examples like (19d) were judged worse than examples like Page 12 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (19c). Results of an online study that tracked eye movements while reading also provided evidence that cases involving NAIs are easier to process. Whereas the intuitions concerning such examples seem clear, it is important to distinguish the predictions of this analysis from others that cover the same set of data. For one, there is considerable overlap between the examples that support the coherence analysis and the NAI hypothesis, so we need to examine cases in which they make the opposite predictions. Example (19a), which is a Violated Expectation case, is predicted to be good on Kehler’s analysis, and it indeed seems marginal. But is this due to the mismatched ellipsis? Consider the variant in (20): (20)

This version, which the NAI hypothesis predicts to be as bad as (19a), seems much improved, suggesting that (19a) is awkward for reasons independent of the ellipsis mismatch.11 Second, we can ask what happens if we modify examples like (18) such that there is no longer an NAI evoked from the first clause: (21)

This change alters the coherence relation from one Cause–Effect relation (Violated Expectation) to another (Denial of Preventer), and hence the example is predicted to still be acceptable on Kehler’s analysis. It should be unacceptable on Grant et al.’s analysis, however, as it is crucial to their story that it is the antecedent clause (and not the ellipsis clause) that gives rise to the NAI.12 Third, we can ask what happens in passages in which the antecedent gives rise to an NAI, but which participates in a Parallel relation. In such cases Kehler would predict mismatches (p. 326) to be bad, but Grant et al. would presumably predict them to be good, since we should see the same effect of raising the salience of the antecedent as we did before. Consider the variant of (18) in (22): (22)

This version seems bad by my judgment, although confirmation will have to await experimental study.13 Finally, whereas many of the examples that motivate Kehler’s analysis involve NAIs, not all of them do. Examples include (23a,b): Page 13 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (23)

We have already discussed example (23a), which seems on par with other cases of acceptable mismatch but which does not have an antecedent that gives rise to an NAI. Example (23b) is likewise at least a marginally acceptable case of mismatch, featuring an active antecedent and passive ellipsis clause rather than the other way around. At this point we have examples of the two types of approach to reconciling the facts introduced earlier: Kehler’s analysis treats VP-ellipsis as anaphoric and then appeals to coherence establishment to explain the existence of syntactic constraints, whereas the approaches of Frazier, Clifton, and colleagues maintain a syntactic analysis and then appeal to external strategies to explain the acceptability of mismatch cases. Kertz (2008, 2013) offered a different analysis that took yet another slice through the data, appearing to capture many of the aforementioned facts in a more straightforward way. According to her account, the crucial factor that determines acceptability pertains neither to the type of coherence relation between the clauses nor the existence of NAIs, but is instead based on the information-structural properties of the ellipsis clause: specifically, whether the elided clause participates in a SUBJECT FOCUS or auxiliary focus construction. Examples like (2a), repeated in (24a), display subject focus: accent falls on the subject and not the auxiliary. On the other hand, examples like (18), repeated as (24b), feature auxiliary focus: accent falls on the auxiliary rather than the subject. (24)

For Kertz, the problem with (24a) does not bear directly on coherence relations, but is instead due to defective information structure. Specifically, Kertz points out that utterances participating in Parallel relations typically feature contrastive topics, and a defining feature (p. 327) of a contrastive topic is being the sentence topic. Unlike syntactically matched active-voice cases of VP-ellipsis, however, in (24a) the speaker has chosen to express the proposition denoted by the antecedent using a passive, which has the effect of demoting the logical subject (i.e., John) to a non-topical position in favor of another event participant. As a result, interpretation is then disrupted when, in the ellipsis clause, the processor has to recognize that it is in a contrastive topic structure and reanalyze John as a topic. There is no such issue in (24b), on the other hand, where the focus in the ellipsis clause falls on the auxiliary, and hence no similar information structure principle is violated. One of Kertz’s experiments was an acceptability study that employed tough alternations rather than a voice manipulation: Page 14 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (25)

Kertz found main effects of both mismatch and focus, as well as the predicted interaction whereby mismatched examples were judged to be significantly worse in the subject focus condition (25b) than in the auxiliary focus condition (25d). Further, two follow-up studies compared judgments on mismatch cases like (25d) against variants in which the ellipsis clauses contained unelided VPs (Venomous snakes are easy to identify, and most experienced hikers can identify them). These studies revealed that there is a penalty for mismatch even in the unelided versions, a crucial prediction of her theory that is unshared by previous ellipsis-specific proposals. Finally, she noted that her analysis makes a novel prediction concerning processing. Specifically, if defective topic structure is part of the reason for the unacceptability of cases with mismatch, such structures should not only disrupt processing at the ellipsis site, but also on the subject noun phrase in the ellipsis clause, since this is the point at which the defective topic structure can first be detected. A self-paced reading time study that used modified versions of (25a,d) as stimuli confirmed this prediction. The fact that mismatch disrupts processing even before the comprehender could know that an ellipsis will ensue is particularly striking evidence that the source of unacceptability in such examples is partially independent of ellipsis. Whereas further work is needed to confirm or refute the predictions of the information structure analysis, it does explain a number of the puzzles we have encountered thus far, without the need to invoke coherence relations, recycling processes, or NAIs. First, consider again the contrast between examples (14a) and (14c), repeated as (26a,b): (26)

Recall that whereas Kehler had based his analysis on cases like (26b), Frazier and Clifton (2006) used examples like (26a) in their experiments. The conflicting judgments are (p. 328) precisely what Kertz’s analysis predicts: (26a) is a case of subject focus (nuclear accent is placed on Bob), whereas (26b) is a case of auxiliary focus (nuclear accent is placed on had). Her analysis therefore makes the right slice through the data without having to posit multiple coherence relations; examples like (26a) simply feature a Cause–

Page 15 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse Effect coherence relation but subject focus information structure. Second, recall that Grant et al.’s study compared judgments on example pairs like (27a,b): (27)

Here we see that Kertz’s analysis predicts the difference in acceptability as well as Grant et al.’s does, since (27a) is subject focus whereas (27b) is auxiliary focus. Indeed, Kertz also captures the relatively acceptable status of (20) for the same reason.14 Finally, there is a range of other cases that are acceptable even though the antecedent clause does not give rise to an NAI. Two cases were pointed out already in (23a,b); both feature auxiliary focus. The same is true of the stimuli that Kertz used in her experiments; none of the antecedent clauses in (25a,d) yields an NAI, for instance, and as such the NAI hypothesis cannot explain the relative acceptability of (25d) compared to (25b). Kertz’s generalization therefore appears to capture the pattern found in the mismatch data much more robustly than either the coherence analysis or the NAI hypothesis. This notwithstanding, a question remains as to how to explain why mismatched ellipsis does seem to degrade acceptability to at least some degree in all context types (cf. nn. 8 and 9), which appears to be a robust finding across the experimental work just described. Another remaining question is how binding theory violations like (2b–c), in so far as they are robust (cf. Frazier and Clifton 2006), would be captured by these accounts. A range of related issues pertaining to binding relations will in fact play a prominent role in the set of questions addressed in section 13.3.

13.3 Strict and sloppy readings The previous section addressed the question of under what conditions an antecedent can license the felicitous use of VP-ellipsis. As mentioned in the introduction, a second and related question concerns the space of possible readings that a particular antecedent can (p. 329) give rise to, particularly with respect to so-called strict and sloppy ambiguities. Recall example (1), repeated as (28): (28)

Assuming that the pronoun his corefers with John, this sentence admits of two interpretations: the strict reading (Bill loves John’s mother) and the sloppy reading (Bill loves his own mother). The question that immediately arises is where this sloppy reading comes from, in light of the fact that the antecedent clause says nothing about Bill.

Page 16 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse A classic analysis posits that these readings arise from an ambiguity in the antecedent clause, specifically whereby the pronoun can have either a REFERENTIAL or BOUND interpretation (Keenan 1971; Sag 1976a; Williams 1977b, inter alia). These interpretations lead to the VP meanings in (29a,b) respectively: (29)

Both meanings lead to the same interpretation for the antecedent clause, so the ambiguity only reveals itself when there is a subsequent ellipsis. On a theory in which VPs are reconstructed (or deleted under identity) at a level of syntactic representation (surface syntactic or syntactic logical form), either of these meanings is available. Using (29a) will derive the strict reading for the ellipsis clause, whereas (29b) will derive the sloppy reading. A number of facts follow straightforwardly from the referential/bound ambiguity analysis (henceforth RBAA).15 First, it predicts that choice of linguistic form matters: a variant of example (28) in which John is rementioned with a name does not license the sloppy reading:16 (30)

This is predicted by the fact that there is no possibility of a bound interpretation for the second occurrence of John. Second, no sloppy interpretation is possible if a pronoun in the antecedent refers extrasententially. That is, if his in (28) refers to George, then the ellipsis clause can only mean that Bill loves George’s mother—the pronoun cannot rebind in the ellipsis clause. Again, this follows directly from the lack of a bound variable interpretation for the pronoun in the antecedent. Third, when the pronoun can bind to more than one possible antecedent, sloppy interpretations must exhibit a parallel binding pattern. First consider (31): (31)

If his in the antecedent refers to John, the only sloppy reading available for the elided clause is Bill said that Fred loves Bill’s mother; the interpretation Bill said that Fred loves Fred’s mother is out. The reverse is true if his refers to Fred in the antecedent clause. (p. 330)

Page 17 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse A fourth argument is more controversial. Specifically, the RBAA predicts that a licit binding configuration must hold between the pronoun and its antecedent, which generally requires that the antecedent c-command the pronoun. This predicts that example (32), which admits only of a referential relationship between the pronoun and its antecedent, gives rise only to a strict reading: (32)

Judgments on such cases vary, however. For instance, whereas Lasnik (1976) claims that an example similar to (32) only has a strict reading, Tomioka (1999) reports that his informants find the sloppy reading acceptable.17 Rooth (1992a) presents the similar sentence John’s coach thinks he has a chance and Bill’s coach does too as having a marginal status. Other facts seem more problematic for the RBAA. One is what Dalrymple et al. (1991) term CASCADED ELLIPSIS. Example (33) was provided by Dahl (1973), which he adapted from an example due to Schiebe (1973): (33)

An acceptable, and perhaps preferred, reading for (33) is: (34)

Example (33) contains two ellipses; the reading in (34) results from the second clause receiving a sloppy interpretation from the first and the third clause receiving a strict interpretation from the second. The RBAA, however, specifically predicts that reading (34) should not exist. Since the second clause will only have the sloppy VP meaning received from the first, the strict meaning that the third clause requires from the second is not present. A second problem is that strict readings are possible even when there is no possibility for a referential interpretation in the antecedent: (35)

Page 18 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse Both cases have a strict reading in addition to the (unproblematic) sloppy one; for instance, (35b) can mean the dean will buy supplies for the chair’s office or her own (see Heim 1998 for a discussion of similar cases). However, because the pronoun his can only be bound by the quantifier in the antecedent, there is no possibility for a referential interpretation. (p. 331)

A third problem is that sloppy readings are licensed within domains that extend further than the minimal clause that contains the ellipsis. Consider (36): (36)

(37)

Whereas him in (36) can be bound to the c-commanding John in the antecedent, the elided VP has no mechanism for rebinding the bound variable to the subject of the ellipsis clause in light of the intervening overt material. Rooth’s example (37)—which admits of the sloppy reading Sue heard I was bad-mouthing Sue—has the same property, and further illustrates the difference between constraints on deaccenting and ellipsis: Whereas the VP was bad-mouthing her is elidable, the remainder of VP that contains it—heard I was— is merely deaccentable, as it can be inferred from, but is not identical with, material in the antecedent clause. Such examples have inspired a variety of analyses that resolve ellipsis by way of parallelism constraints applied to expanded parallelism domains and discourse structures (Dalrymple et al. 1991; Prüst 1992; Rooth 1992a; Hobbs and Kehler 1997; Asher et al. 2001; Hardt and Romero 2004; Hardt et al. 2013, inter alia). A fourth problem is that there are clear cases in which sloppy readings are possible in the absence of a licit binding relationship between a pronoun and its antecedent: (38)

Example (38a) exhibits a sloppy reading despite the lack of c-command between him and John.18 Example (38b) is a case in which the required dependency applies across clause boundaries. Finally, example (38c) demonstrates that sloppy readings—in this case, I’ll kiss you even if you don’t want me to kiss you—can result from anaphoric dependencies at the event level.19 Whereas some requirement for parallelism is clearly at play in such examples, each has a sloppy reading despite the lack of a possible binding relationship. Page 19 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse Other problems result when the RBAA comes into contact with treatments of related phenomena. For instance, Reinhart (1983a) proposed what has come to be known as the Coreference Rule, which rules out referential interpretations for pronouns in clauses when a bound reading is possible and yields an equivalent interpretation. This condition (p. 332) eliminates the possibility of ‘accidental’ coreference in cases in which Condition B should apply (i.e., to disallow him coreferring with John in John likes him by virtue of a referential interpretation). It’s clear that the RBAA and Coreference Rule cannot both hold.20 Due in part to such issues, various approaches have used methods for identifying strict and sloppy readings without positing a referential/bound ambiguity for pronouns. Dalrymple et al. (1991), for example, propose a semantic analysis that resolves VP-ellipsis at the clause level. Roughly speaking, interpretation involves the solving of an equation in which the meaning of the antecedent clause as a whole is equated with the meaning of the ellipsis clause VP as applied to the meanings of the elements in the antecedent clause that are parallel to overt elements in the ellipsis clause. For example (28), this identity is captured by equation (39), which under suitable assumptions has (29a,b) as its two solutions for the meaning P of the elided VP. (39)

Because the method operates on purely semantic interpretations (at which binding relations and distinctions among referential forms are lost), however, it does not capture various facts we have discussed. For instance, it is insensitive to whether a binding relationship exists in generating sloppy readings (which might be a good thing or not, depending on one’s judgments for examples like (32)). More importantly, it will also generate sloppy readings based on coreferential expressions even when a pronoun is not used (cf. (30)) (Tancredi 1992; Kehler 1993a). Fox (2000), on the other hand, takes a different tack. Working within the assumptions of a syntactic approach, he posits a parallelism condition on the syntactic representations of elided VPs: (40)

As is no doubt apparent, options (40a) and (40b) respectively recreate the effects of the referential and bound options in the RBAA without locating the ambiguity in the antecedent clause.21 The problem, as Fox admits, is that the condition is purely Page 20 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse stipulative: it does not follow from other properties of the grammar. In addition, it doesn’t work for examples like (35b), where there is an ambiguity even though the pronoun does not have a referential value, and hence option (40a) cannot apply. A more radical way of handling the data was proposed by Hardt (1992a, 1992b, 1993, 1999). He posits an anaphoric theory, but one in which VP meanings are copied to the ellipsis clause with any pronoun representations contained in them left unbound. These pronouns are then reinterpreted in the ellipsis clause according to normal rules of pronoun interpretation. For example, in (28) the semantic representation for loves his mother would be copied to the ellipsis clause and then his re-resolved in its new context, in the same manner as if the elided VP appeared overtly. The fact that both John and Bill are salient in the discourse context yields the strict and sloppy ambiguity. Note that some care must be taken to restrict the possible readings on this approach—for instance, to disallow sloppy readings when a pronoun in the antecedent clause refers extrasententially, sloppy readings that do not maintain binding parallelism in examples like (31), and so forth. We will revisit Hardt’s approach in section 13.3.2. (p. 333)

These are but a few of many approaches that attempt to explain strict/sloppy ambiguities without appealing to a referential/bound ambiguity in pronouns. The problem with all of these accounts is that they each go beyond what could be seen to naturally fall out from independent properties of the grammar, i.e., all have additional machinery tailored specifically to the data at hand. This is an unsatisfactory situation. Now when it comes to building a theory from first principles, it may stand to reason that syntactic analyses are on the right track, in light of the fact that matters that bear on the linguistic form of the antecedent—e.g., choice of referring expression and binding relationships—are at least in part relevant to the question of when sloppy readings are available. It therefore seems all the more surprising that we find that strict/sloppy readings are far from limited to elliptical contexts (Dalrymple et al. 1991; Tancredi 1992; Kehler 1993a; Hobbs and Kehler 1997, inter alia): (41)

As there is no reason to think ellipsis is involved in any of these constructions, not only does the existence of strict/sloppy readings not specifically favor syntactic approaches to VP-ellipsis, but any approach that recovers such readings as a side effect of reconstructing (or resulting from the deletion of) syntactic material misses a significant generalization. The question then is why these other phenomena would display the same sensitivity to (presumably syntactically governed) binding relationships in the antecedent.

Page 21 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse One property common to these phenomena is that they are all sensitive to information structure, i.e., they all depend on their denotations being Given in the discourse. Indeed, the fact that these ambiguities are shared not only by event-referential forms (41a,b) but also overt, deaccented VPs (41c) and other focus-sensitive constructions (41d) points us directly toward deaccenting theories, as we expect the other forms to also be constrained by principles governing focus marking and accent placement. Rooth (1992a), for instance, proposes an analysis based on such principles. In his analysis, VP-ellipsis is subject to a constraint that requires that the ellipsis clause FOCUS MATCHES the antecedent clause whereby, roughly speaking, there exists a replacement for the focused elements in the elided clause that gives rise to the meaning of the antecedent. Consider how this would (p. 334) work for (28). Let us assume that, per the Coreference Rule, c-commanded pronouns in antecedent clauses can only be bound. Example (42a) represents the strict reading. (42)

Focus matching the ellipsis clause subject Bill against the antecedent subject John will give rise to the proposition that John loves John’s mother. Since this is the meaning of the antecedent clause, ellipsis is licensed. Example (42b), on the other hand, represents the sloppy reading. Focus matching will give rise to the meaning by which John loves his own mother. This also corresponds to the meaning of the antecedent clause, and so again ellipsis is licensed. As pointed out by Büring (2005: section 6.5), Rooth’s analysis handles at least the simple cases in an independently motivated way and without additional stipulations about binding and parallelism.22 However, the situation is about to become more complicated.

13.3.1 Missing readings examples Dahl (1974) famously introduced what have come to be known as MISSING READINGS puzzles in ellipsis. Consider example (43): (43)

Assuming that both pronouns are bound in the antecedent clause, a naive theory of strict and sloppy interpretations such as the RBAA—in which each pronoun can be interpreted strictly or sloppily—predicts the four readings shown in (44a,d). (44)

Page 22 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

The all-strict and all-sloppy readings in (44a,b) are clearly acceptable. Dahl’s judgments are that the mixed reading in (44c) “seems rather dubious” and that (44d) “is clearly not possible.” Since then, the literature has generally considered (44a–c) to all be acceptable, leaving the puzzle of why (44d) is not.23 (p. 335)

To explain this fact, a number of researchers have appealed to a locality condition on pronoun binding (Kehler 1993a; Fox 2000; Schlenker 2005, inter alia). Specifically, the ability to derive all four readings for (43) is dependent on the antecedent clause having a binding configuration whereby both pronouns are bound directly by the matrix subject John. Note, however, that his could also be bound by the embedded subject he. Let us stipulate, as Fox (2000: 115) does in his Rule H, that his can in fact only take this more local binder in such a circumstance: (45)

Now the missing reading cannot be derived, because the binding configuration in the antecedent that would be necessary—in which the second pronoun is bound directly to John—is no longer considered licit. With his bound locally, the combination that would have otherwise lead to the missing reading—where, using Fox’s terminology, the first and second pronouns satisfy referential (40a) and structural (40b) parallelism respectively— will instead derive the all-strict reading (redundantly with the scenario in which both pronouns satisfy referential parallelism), since the interpretation of the second pronoun will necessarily follow the interpretation of the first.24 In cases in which local binding is not possible between two pronouns, analyses of this sort predict that all four readings should be available: (46)

(47)

Page 23 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

In these cases, neither pronoun c-commands the other, and hence both can only be bound by John. As predicted, examples (46) and (47) admit of all four expected readings for most informants.25 (p. 336)

In light of examples (41a,d) discussed previously, we are led to ask whether other

phenomena also exhibit the missing readings pattern, including event anaphora, deaccenting, and focus-sensitive operators. Indeed they do (Kehler and Büring 2007): (48)

So the data again indicate that the constraint is coming from somewhere other than ellipsis per se. For instance, the fact that the deaccented version in (48c) and event anaphora in (48a) are missing reading (44d) suggests that perhaps the constraint is information-structural, as again all of these forms require their antecedent to be Given in the discourse.26 This yields an odd state of affairs, however, if we consider what is Given to be that information which is entailed by the context (Schwarzschild 1999). That is, we are then forced to the conclusion that the antecedent of (48c) renders as Given the properties said that John loves John’s wife (BJJ reading), said that oneself loves one own’s wife (BBB reading), and said that oneself loves John’s wife (BBJ reading), but not said that John loves one own’s wife (BJB reading), even though obviously all four are equally entailed by the context.27 Kehler and Büring (2007) attempt to explain these facts with a different type of information-structural analysis, particularly one that utilizes QUDs (Roberts 1998a; Büring 2003, inter alia). The proposal relies on the fact that parallel clauses such as those in (43) are plausibly analyzed as each providing partial answers to a (usually implicit) QUD. As a result, one would only expect an antecedent/ellipsis pair to be acceptable under a particular interpretation for the ellipsis if a suitable QUD can be inferred to which each clause provides a felicitous answer. It is here, according to the analysis, that things go wrong for reading (44d). Specifically, the QUD that is required to license the reading—Who1 thinks that John loves his1 wife?, where crucially, the possessive is bound by the wh-expression and not John—is one that is normally understood to be asking about people other than John himself. For instance, an answer of no one to this question can be judged true even if John (p. 337) thinks that he loves his own wife (as long as no one else does); and indeed, answering Well John does, of course! in such a circumstance seems snarky. Hence, according to this analysis, it is the meaning of the antecedent clause of (43), and not the ellipsis clause, that creates the problem for reading (44d), since this Page 24 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse meaning is presupposed to not be in the alternative set denoted by the QUD required to license the reading. Whatever the relative merits of this particular analysis, we see again how models of discourse coherence may play a role in determining the felicity of VPellipsis. We will return to the topic of QUDs momentarily. According to the analyses just discussed, the anaphoric dependencies that eliminate missing readings result from binding relationships. We have already seen examples (i.e., 38a–c) that suggest that that notion must be broadened, however. A different missing readings example that has been discussed in the literature, due to Gawron and Peters (1990), provides another case in which the space of readings appears to be dependent on the establishment of anaphoric dependencies, but not in a way that can be readily captured by binding relations: (49)

There are two VP-ellipses at play in (49). The first (the teacher did) has the first simple clause (John revised his paper) as its antecedent. The second (Bill did too) then has the entire previous sentence as its antecedent, including the first ellipsis site. We are interested in the set of possible readings that result when his corefers with John in the first conjunct. In theory, (49) could have as many as six readings:28 (50)

Gawron and Peters themselves claim that (49) has only three readings (49b, 49c, and 49d), which are those generated by their situation-theoretic analysis. Dalrymple et al. (1991), however, argue that there are five readings, and Kehler (1993b) concurs—all readings except (49f) are possible, although (49a) and (49e) are perhaps less immediately accessible. A straightforward application of Dalrymple et al.’s analysis generates all six readings, (p. 338) although they appeal to a mechanism-external constraint on relation formation to rule out (49f). Fox’s system appears to generate four readings: (49a–c) and (49e)—missing (49d), a reading that appears uncontroversially to exist. Kehler’s (1993b) Page 25 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse system, while being similar to Fox’s but cast within an event semantics instead of at LF, generates the five readings in (49a–e). The key difference between the accounts comes when the strict reading of the first ellipsis is recovered; unlike Fox’s system, in Kehler’s system the reconstructed pronoun will be anaphorically linked to its parallel pronoun in the antecedent clause despite there being no binding relationship, which ultimately allows reading (49d) to be derived when the second ellipsis is resolved.

13.3.2 Hardt’s puzzle Although approaches to resolving strict and sloppy readings vary in their details, one thing seems to be clear: there is some sort of parallelism constraint that governs the possibility of sloppy readings. We see evidence for this in the lack of sloppy readings when a pronoun in the antecedent clause refers extrasententially, the lack of readings with non-parallel binding in (31), and the missing readings cases. Whereas it may not be clear whether the requisite parallelism constraints are best stated in terms of syntax, semantics, or discourse structure, it would seem that constraints on parallelism are applied at some level. Hardt (1992a, 1992b, 1993, 1999) offered a collection of examples that were problematic for the early treatments of strict/sloppy ambiguities of Sag (1976a) and Williams (1977b). Most of these, including his (38b), can be captured by treatments involving more extended notions of parallelism (Dalrymple et al. 1991; Prüst 1992; Rooth 1992a; Hobbs and Kehler 1997; Asher et al. 2001; Hardt and Romero 2004; Hardt et al. 2013, inter alia). One of his examples, however, provides a more direct challenge for any analysis based on either syntactic or semantic parallelism: (51)

The elided clause in this example clearly admits of what would appear to be a sloppy interpretation, that is, I think Mrs Smith will pass John. Yet this reading arises without the existence of the type of syntactic or semantic parallelism between the clauses that all theories we have discussed require; in particular, the anaphoric relationship that is required in the ellipsis clause does not mirror the binding configuration between the pronoun and quantified noun phrase in the antecedent clause. Despite the fact that this example provides a clear adequacy criterion for any theory, substantive discussion of it is conspicuously absent in the literature.29 Hardt relied on such examples to argue that the interpretations of pronouns in VP representations that are copied to the ellipsis clause are not generated by virtue of parallelism constraints, but instead using the same discourse principles that apply to overt pronouns. In (51), for instance, the semantic representation of the VP pass him would be copied to the ellipsis clause representation with the pronoun unresolved, which (p. 339)

Page 26 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse would then be resolved to John via ordinary pronoun interpretation mechanisms on analogy with how we would expect the unelided version in (52) to be interpreted. (52)

Kehler and Shieber (1997) subsequently argued against this proposal, however, claiming that it can only be maintained if it is assumed that the unelided versions of reconstructed VPs are fully deaccentable. They considered minimal pairs such as (53): (53)

The reading where Mary asked out Bob at Bob’s party, while readily available in example (53a), is not available in its elided counterpart (53b). As Kehler and Shieber note, however, the intended interpretation (53a) appears to require accent within the VP, specifically on the pronoun. This is predicted on standard theories of focus marking and accent placement (e.g. Schwarzschild 1999); whereas both the verb meaning (ask X out) and the object NP meaning (Bob) represent Given information, the result of combining the two (ask Bob out) does not, and hence some part of the VP must bear accent. As such, Kehler and Shieber argued, a deaccenting analysis already gets these facts right without the need to posit a less constrained pronoun interpretation process on top of it. The real mystery is why the VP can be deaccented in the unelided version of (51) shown in (52), but not in (53a), in light of the fact that the reasoning we gave for why the VP in (53a) should carry accent should apply equally to (52) as well. To explain this, Kehler (2007, 2015) argued for an analysis that makes crucial reference to QUDs. Briefly, the idea is that phrases like in X’s case (to which one might add as for X, regarding X, etc.) are not ordinary prepositional phrases (cf. last night at Bob’s party in (53)), but instead introduce QUDs into the discourse that are created by substituting X for a parallel entity within the meaning of an anaphorically identified, contextually salient proposition. In the case of (51), for instance, the phrase in John’s case is anaphorically resolved to the antecedent clause (shown again in (54a)); establishing parallelism between John and the bound pronoun and taking pass to be the matrix of the QUD then yields the question in (54b). The remainder of the clause then provides an answer to this QUD. (54)

Page 27 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

The treatment of the ellipsis in (54c) is now straightforward: Despite appearances, it is not (54a) that serves as the antecedent of the VP-ellipsis, but instead the meaning of in John’s case given in (54b). As such, the example doesn’t involve rebinding of a pronoun at all;indeed with this antecedent the reading is an otherwise ordinary strict interpretation. The difference between this case and (53) is that last night at Bob’s party is not anaphoric—it is a normal adverbial phrase—and hence does not similarly introduce a QUD. The aforementioned analogous deaccenting facts follow directly. In so far as this analysis is on the right track, it provides more evidence that analyses of VP-ellipsis must account for higher-order coherence properties of the discourse. (p. 340)

13.4 Conclusions Despite the fact that it has remained a focus of attention in linguistics for over forty years, the analysis of VP-ellipsis is still rich and fertile ground, with many unresolved questions. We have seen examples that point specifically to syntactic, anaphoric, and deaccenting analyses, with the data supporting each often being problematic for the other two. Further work will obviously be required to sort it all out. For now, the data in this chapter is offered as a partial set of adequacy criteria for any theory, with which future work can hopefully engage. As is fitting for the topic of this chapter, one theme we have touched on repeatedly is the idea that constraints bearing on the establishment of discourse coherence may play a role beyond those imposed by the linguistic properties of the antecedent and ellipsis clauses themselves. Kehler (2000), which operated within a relational theory of discourse coherence, was an early example; more recently QUD analyses are emerging as a potential source of explanatory power. We have seen a varied set of works begin to converge in this regard. For instance, although Grant et al. (2012) posit a syntactic account of interpretation, they appeal to QUDs in their explanation for why their NonActuality Implicatures would play a role in processing. Kertz’s (2013) information structure theory is also closely related, in that the difference between subject focus and auxiliary focus constructions corresponds directly to the sets of focus alternatives that are commonly considered to be the meanings of wh- and polar QUDs respectively. The same is true for Miller and Pullum’s (2014) treatment of VP-ellipsis with situationally evoked antecedents, which builds on Kertz’s insights. In related work, Miller and Hemforth (2015) invoke QUDs in their treatment of VP-ellipsis with nominalized antecedents; according to their analysis, such cases are acceptable in just those cases in which the nominalization functions as a concealed question. In the realm of accounting for strict/sloppy readings, Kehler and Büring (2007) appealed to QUDs to explain Dahl’s Page 28 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse puzzle, and Kehler (2015) used them in his account of Hardt’s puzzle. Other work along these lines has recently emerged as well: see, e.g., Keshet (2013) and Elliott et al. (2014). Whereas it is the linguist’s temptation to ascribe the behavioral properties of a linguistic phenomenon to the syntax and semantics of the form itself, sometimes the picture only clarifies when one identifies the role of independent discourse interpretation processes with which discourse-sensitive phenomena like ellipsis would be expected to interact. Having said that, our understanding of QUD models of discourse coherence is still in a relative state of infancy; indeed the study of VP-ellipsis may be as useful for developing theories of QUDs as the other way around. Research on VP-ellipsis also provides an exemplar of how theoretical and experimental linguistics can profitably interact. Whereas experimental research on language is burgeoning, much of it bears on processing issues rather than the sorts of fundamental questions about the structure and meaning of language that have been the primary concern of theoretical linguistics. VP-ellipsis is rich in this regard precisely because approaches differ as to whether infelicity is predicted to be caused by the grammar or by processing, and hence it is among a small set of phenomena at the center of the grammar vs performance debate. As discussed earlier, Kertz’s self-paced reading (p. 341)

study—in showing that processing difficulty occurs at a place that her theory predicts, but before the comprehender could know that an ellipsis will ensue—illustrates how experimentation can add to the empirical base for theory building beyond what one can contribute from one’s armchair. Perhaps what is most fascinating about ellipsis is how a process that seems so conceptually simple can give rise to such complexity in the data. This is no doubt due in part to the fact that it interacts with so many other complex things—spread across the domains of syntax, semantics, anaphora, and information structure—and hence why the benefits of studying ellipsis go far beyond gaining an understanding of the phenomenon itself. Indeed, studying ellipsis is so exciting precisely because it gives us a window into big questions across such a wide range of language phenomena.

Notes: (1) Note that acceptable examples of ACD are among a variety of cases that demonstrate that, in principle, there is no constraint against elided VPs containing traces: ((i))

On a standard syntactic analysis, the quantified NP everything raises, leaving behind an antecedent VP containing a trace, thereby licensing the deletion (or reconstruction) of a parallel VP in the ellipsis site. Another type of example involves VPs containing traces left behind by topicalization; (1b) is from Merchant (this volume; see also Merchant 2008b): ((ii)) Page 29 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

Like (i), example (iia) is felicitous, in this case despite the fact that the antecedent does not contain a trace. The variant in (iib) is unacceptable, on the other hand, a fact predicted by an island violation on a syntactic theory. Both facts can be explained by a deaccenting theory in which unpronounced syntactic structure resides at the ellipsis site. On the other hand, whereas anaphoric analyses can be constructed that allow for cases like (iia), it is far from clear how such an approach could simultaneously rule out (iib). (2) Here I set aside the well-known fact that a wider range of VPs can be deaccented than can be elided; for instance (ib), unlike (ia), cannot be used to describe a situation in which Bill insulted Mary by doing something other than calling her a Republican (here and throughout I use smaller font to indicate deaccented material): ((i))

Deaccenting in (ia) is licensed because (under suitable political assumptions) the antecedent clause renders insult Sue as Given. The VP cannot be elided, however, as the hearer would have no way of distinguishing which of many deaccentable VPs (said something to Mary, offended Mary, etc.) was the one intended by the speaker. (3) One could posit an analysis that allows for referring expressions that are compatible with binding theory to occur as the objects of the elided VPs instead (e.g., blame him in (5a); see for instance the VEHICLE CHANGE proposal of Fiengo and May 1994). In this case, all of the examples would be predicted to be acceptable, again not capturing the contrast between (2b,c) and (3b,c). (4) It is worth noting that the deaccenting theory captures the difference between (2d) and (3d). Whereas the unelided version of (3d) is unacceptable if a trace is left in the ellipsis site, it is fine with a full VP—Which problem did you think John would solve because of the fact that Susan solved it?—and hence can be deleted under identity. Example (2d), on the other hand, is as unacceptable on the full VP variant as the tracecontaining variant (*John read everything which Bill believes the claim that he read it.). (5) For additional experimental evaluations of the coherence analysis, see Kim and Runner (2009, 2011) and SanPietro et al. (2012). (6) Although “unacceptable” in this task was defined as “violating the normal rules of everyday English”—a task description that runs the risk of participants bringing prescriptive biases to bear.

Page 30 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (7) It should be noted that while there is a syntactic match between the VP in the antecedent clause (approaching the lion) and the form required by the ellipsis clause (approach the lion), there is not a perfect morphological match. Such mismatches are typically unproblematic for successful VP-ellipsis; indeed even examples like (16a) lack a perfect match. (8) This issue is familiar from the debate on the status of so-called anaphoric islands for nominal reference. Whereas Postal (1969) appealed to a grammatical constraint against ‘word-internal’ reference (# Fritz is a cowboy. He says that they can be difficult to look after), Ward et al. (1991) point out that such cases are acceptable if a sufficient degree of semantic transparency exists between the antecedent expression and its word-internal counterpart (Do parental reactions affect their children?; see also example (10)). That is to say, the difference results from the fact that parental activates the concept of PARENT more strongly than cowboy does for COW. In the domain of event reference, Kehler and Ward (2007) cited the same factors in their treatment of do so anaphora with nominalized antecedents. They argue that the reason that examples like Four out of five female smokers do so in order to lose weight are acceptable but Most professors will do so even when no one is listening is not is because smokers activates the concept of smoking more strongly than professors activates the concept of professing. This type of sensitivity of an anaphor to the level of activation of its referent is just what one expects on an anaphoric theory. (9) However, Arregui et al. (2006) are quite right when they say: “A semantic account that could mirror our results would need to find the same level of relevant fine-grained distinctions in purely semantic representations. Such an account would also have to deal with some nontrivial issues. Since all conditions have event-denoting verb roots, why would these event properties be made more salient or available in an active sentence than a passive, and why would they be made more available by an inflected VP than by a gerund, by a gerund than by a nominalization, etc.” (p. 243). Whereas I am less pessimistic than Arregui et al. that a plausible account of this type could be constructed (see n. 8), the onus is indeed on anaphoric approaches to specify the principles that link linguistic form to level of accessibility in the discourse model, a topic that has received little attention to date. (10) What Grant et al. mean by implicatures “motivating and guiding” the language processor is admittedly not clear to me. (11) More on this in a moment. (12) Grant et al. (2012: 333) briefly address a similar case, saying that “the negation in the first clause may have the effect of introducing alternatives and thus behaving much like the cases of NAI we have discussed.” No reason is given for this conjecture, however, and indeed one has to ask why a clause that asserts that an event did not occur would

Page 31 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse give rise to a QUD that it already provides an answer to, whereas a clause that asserts that an event did occur would not. (13) To explain the unacceptability of (22), one might note that in this case the ellipsis clause does not ultimately answer the QUD that Grant et al. posit to be evoked by the NAI. However, it is hard to see how the mechanics of an explanation based on this fact could work, since the role of the NAI is to encourage the processor to repair the VP as a necessary prerequisite to determining congruence of the ellipsis clause with the QUD. (14) Grant et al. briefly address Kertz’s account (as described in Kertz 2008) based on the results of their Experiment 1b, which demonstrated that cases of VP-ellipsis that involve modals that give rise to NAIs (ib) are judged to be more acceptable than ones that do not (ia): ((i))

Grant et al. argue that this is a problem for Kertz, since “examples with the modal may arguably involve auxiliary focus” (2012: 335). Their judgment does not at all accord with mine, however; I find that (ia) requires subject focus to be felicitous (i.e., if someone made them it wasn’t the BABYSITTER who did). Example (ib), on the other hand, is more natural with auxiliary focus, at least in a context in which the babysitter was the person expected to make the cookies. (15) The RBAA is related to what Dalrymple et al. (1991) call “identity-of-relations” analyses, a term that they, following Dahl (1973), use more generally for any analysis in which the possible readings in the ellipsis clause result from a parallel ambiguity in the antecedent clause. Such approaches contrast with non-identity accounts, which posit no ambiguity in the antecedent but in which ambiguities result instead from the mechanics of the VP-ellipsis meaning recovery procedure itself. (16) Of course, the antecedent clause here violates Condition C of the binding theory, but that fact is orthogonal to the question at hand. (17) However, it may be that speaker judgments covary with those for ordinary quantificational binding, whereby the sloppy reading for (32), for example, is available to just those speakers for whom (17) admits of a bound variable interpretation: ((i))

In this case, the facts would support the RBAA even for those speakers that get a sloppy reading for (32).

Page 32 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse (18) Tomioka (1999) and Elbourne (2001) give alternative analyses of examples like (38a). (19) Schwarz (2000) provides a detailed discussion of such examples, and Charlow (2012) a more recent treatment. (20) That is unless we extend the ‘equivalent interpretation’ restriction to apply to subsequent clauses (in this case, the ellipsis clause) rather than just the one containing the binding relation, a possible but unattractive option (Büring 2005). (21) Kehler (1993b) proposed a similar system, except implemented in a version of event semantics that maintains referential dependencies between pronouns and their referents. Unlike Fox’s system, however, all such dependencies are represented, and thus the system derives sloppy readings even when there is no c-command relationship between the pronoun and its antecedent. (22) As Rooth notes, reliance on the focus-matching condition alone is insufficient in that it is insensitive to the forms of reference used in the antecedent clause. As such, the condition predicts (30), repeated below as (i), to be acceptable on a sloppy reading. ((i))

Rooth suggests that the principle be augmented with a second that requires syntactic identity. As we saw in section 13.2, however, such a condition would have to be reconciled with substantial data that points away from such a requirement. (23) An exception is Tancredi (1992), who offers a context which he claims allows for the fourth reading of a similar example. His judgments are at odds with mine, however. (24) There are a number of variations within this class of analyses that would take us too far afield to discuss here. See, for instance, Kehler (1993a), Fiengo and May (1994), Büring (2005), Schlenker (2005), Roelofsen (2011), and Drummund (2013), inter alia. (25) Kehler and Shieber (1997), however, note that example (i) only has two readings instead of the expected four: ((i))

Like (46) and (47), neither possessive pronoun in the antecedent can bind the other, but unlike those cases, the readings in which the two pronouns have disjoint reference are clearly out. Interestingly, this fact follows from the deaccenting theory, since the unelided, deaccented variant of (i) is also missing the mixed readings—each would require accent on at least one of the two pronouns in the ellipsis clause. It is admittedly

Page 33 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse not clear to me what the source of that constraint is, although it likely bears on the fact that the two pronouns occupy parallel positions in a coordinated noun phrase. (26) This leads us to ask how Rooth’s (1992a) focus-matching constraint discussed in section 13.3 fares on these more complex examples. Unfortunately, it will straightforwardly derive all four readings. (27) The missing reading becomes available in the unelided version, however, if the second pronoun is accented: …and Bill said that hej loves HISb wife. This version is also compatible with the all-sloppy reading, and other variants in which one or both pronouns are accented are also compatible with various readings that are acceptable with the fully deaccented version. Since the lack of the missing reading seems to result from the fact that the second pronoun must receive accent in the unelided version, coming to grips with this phenomenon may require that we come to understand why it is that bound pronouns would ever receive accent in the first place. Sauerland (1998, 2008), Jacobson (2000b), Mayr (2012), and Keshet (2013) offer proposals and discussion. (28) Each reading contains descriptions of four papers that were revised. The notation to the right of each example indicates the respective owners of these papers. (29) Fiengo and May (1994: 111, n. 14) are an exception, arguing that a similar example “preclude[s] a strict reading, and so appears to be subject to somewhat different constraints on anaphoric resolution,” and that therefore “such cases…are irrelevant to the matters of parallelism under discussion.” It is unclear to me why an infelicitous strict reading would necessarily license an otherwise unavailable sloppy reading (as opposed to simply being infelicitous on any interpretation), and no alternative account of the example is offered.

Andrew Kehler

Andrew Kehler is a Professor in the Department of Linguistics at the University of California, San Diego. His primary research foci are discourse interpretation and pragmatics, studied from the perspectives of theoretical linguistics, psycholinguistics, and computational linguistics. His publications include Coherence, reference, and the theory of grammar (2002) and numerous articles on topics such as ellipsis, discourse anaphora, and discourse coherence.

Page 34 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Discourse

Page 35 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics

Oxford Handbooks Online Ellipsis and Computational Linguistics   Daniel Hardt The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.14

Abstract and Keywords This chapter considers approaches to ellipsis within computational linguistics. It begins with the structure of the ellipsis site—a topic that has received little attention in computational linguistics. There are two prominent accounts of the recovery of ellipsis: that of Lappin and McCord (1990) and Dalrymple et al. (1991). The topic of licensing follows. This topic has not been directly addressed in the computational literature, but the chapter covers two related issues of direct computational interest: the identification and generation of ellipsis occurrences. This is followed by three additional topics: identifying the antecedent, dialogue, and text mining. Keywords: ellipsis, computational linguistics, unification, generation, dialogue

Daniel Hardt

14.1 Introduction COMPUTATIONAL linguistics differs from other approaches to linguistics in that it always keeps in view a computational implementation. This can provide a useful and different perspective on ellipsis, both because the computational perspective can bring different problems to the fore, and also because the emphasis on computational techniques can result in new ways to approach these problems. For example, in the approach of Dalrymple et al. (1991), the higher-order unification techniques used in automated theorem proving and related fields provided direct inspiration for a new and fruitful perspective on the resolution of ellipsis. Also, the requirements of a concrete application require consideration of issues that have otherwise been ignored, such as the identification of ellipsis occurrences and the identification of the antecedent. In what follows, I begin with the structure of the ellipsis site—a topic that has received little attention in computational linguistics. I turn then to the question of recoverability: Page 1 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics here there are two prominent accounts, that of Lappin and McCord (1990) and Dalrymple et al. (1991). The topic of licensing follows. This topic has not been directly addressed in the computational literature, but I consider two related issues of direct computational interest: the identification and generation of ellipsis occurrences. Finally, I consider three additional topics: identifying the antecedent, dialogue, and text mining.

14.2 Structure of the ellipsis site Many theorists (Fiengo and May 1994; Merchant 2001) have posited elaborate syntactic structure at the ellipsis site, while others have argued for a WYSIWYG approach (see Culicover and Jackendoff 2005 and references therein), where there is no implicit (p. 343) structure at the ellipsis site. This question has not received much direct attention in the computational linguistics literature. There is a general reluctance in computational approaches to posit implicit syntactic structure, and indeed theorists who argue against implicit structure sometimes argue that their position is more congenial to computational approaches. The account of Dalrymple et al. (1991), which will be treated in depth in section 14.3.2, is often named as a prominent representative of the no implicit structure view.1 However, Dalrymple et al. (1991) do not stake out a clear position here: in fact, they note: “The incorporation of an ellipsis analysis such as ours into a transformational framework is quite conceivable, merely requiring the ability to form abstractions over the syntactic objects representing semantic construals of sentences, that is, LF trees” (p. 414, n. 9). Thus it appears that Dalrymple et al. (1991) are agnostic on the question of implicit structure in ellipsis. One exception to this general agnosticism is Lappin and McCord (1990), who propose “an analysis of VP anaphora that involves applying rules of interpretation directly to Sstructure (parsed surface structure)” (p. 197). Thus Lappin and McCord argue that there is implicit syntactic structure, and they go on to present an algorithm for resolving VPellipsis that relies on the fact that the ellipsis site contains syntactic material. In particular, the VP-ellipsis resolution algorithm interacts with filters on pronoun resolution based on syntactic constraints on binding.

14.3 Recoverability There are two general computational approaches to recoverability: copying, represented primarily by Lappin and McCord (1990), and unification, represented primarily by Dalrymple et al. (1991).

14.3.1 Copying Lappin and McCord (1990) present “an algorithm for interpreting elliptical VP’s in antecedent-contained deletion structures, subdeletion constructions, and intersentential cases” (p. 197). They describe a complete implementation of the algorithm, which

Page 2 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics operates on the output of a Slot Grammar parser. The algorithm consists of three parts, A, B, and C, which we consider in turn. Part A identifies an elliptical-verb–antecedent pair . Since Lappin and McCord are describing a fully implemented system for VPE interpretation, they must address the VPE identification and antecedent location problems that are not addressed by most theoretical accounts, including computational accounts such as Dalrymple et al. (1991). The VPE “is identified by the presence of an auxiliary verb or infinitival ‘to’ without a (p. 344) realized verb complement” (p. 201). The antecedent A is defined as a nonelliptical verb, occurring in one of three configurations relevant to the VPE: 1. VPE is contained in a clausal complement to a clause containing A. 2. VPE is contained in a relative clause to a clause containing A. 3. VPE is contained in the right conjunct of a sentential conjunction S, and A is contained in the left conjunct of S. The first two conditions describe cases where the VPE clause is in the same sentence as the antecedent, while in the third condition they are in adjacent, conjoined sentences. Hardt (1992a) points out that these configurations are incomplete, giving numerous examples from the Brown Corpus that do not satisfy any of these conditions. For example, Hardt notes that in 5 percent of the examples in his data, at least one sentence intervenes between the antecedent and VPE. Furthermore, when the VPE and the antecedent are in the same sentence, there are many cases that are not covered by the first two conditions, such as Hardt’s example (21), repeated here as (1): (1)

Here the VPE is the matrix VP, while it is the antecedent use folklore that is contained in a relative clause. One unusual aspect of Lappin and McCord’s approach is that they offer a characterization of the possible VPE–antecedent configurations, instead of characterizing the impossible configurations. This is the approach taken, for example, in Hardt 1997a; in the theoretical literature a common view is that the impossible VPE–antecedent configurations are similar to the impossible pronoun–antecedent configurations. For example, Kim (2001) points out that VPE obeys the Backwards Anaphora Constraint (Langacker 1969)—it can precede, but not command its antecedent. Similar observations are made in Lobeck (1995) and Hardt (1999). Thus, while Lappin and McCord’s account is notable in being a fully implemented system, and therefore must confront the issue of antecedent location, it is too limited in the configurations that it permits. Furthermore, it does not address the issue of ambiguity: when more than one potential antecedent occurs, some means for selection is required.

Page 3 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics We turn now to the central part of Lappin and McCord’s algorithm, steps B and C. Here the missing material at the ellipsis site is filled in: first, the missing head V of the VPE is copied from the antecedent. Then, each argument for V is also copied from the antecedent, unless an argument is already present at the VPE site. The same is done for adjuncts. Since Lappin and McCord allow the possibility that an argument is present at the VPE site, it seems incorrect to describe this as VP-ellipsis, since in such cases what is elided is not a VP, as illustrated by Lappin and McCord’s (6a), shown here as (2): (2)

Lappin and McCord use the term subdeletion to describe such examples, which more typically are described as pseudogapping. There is an admirable simplicity to this approach: for each argument or adjunct of the antecedent, it is copied to the ellipsis site unless it (p. 345) is already present. However, there are known to be differences in the distribution of pseudogapping and VPE, and these differences are not accounted for here. This is particularly relevant for the characterization of VPE–antecedent configurations, which, as I noted above, are too restrictive for VPE. These configurations are also meant to apply to pseudogapping, and here the configurations are not restrictive enough. For example, Hardt (1997b) points out that backwards pseudogapping does not occur—the antecedent must precede the ellipsis site. Furthermore, while the VPE can be separated from the antecedent by one or more sentences, this is not possible for pseudogapping. The VPE algorithm can be combined with two other algorithms implemented by Lappin and McCord: a syntactic filter on pronominal anaphora and an anaphor binding algorithm. The result is an “Integrated System for Anaphora Resolution” (p. 207), which can be illustrated by Lappin and McCord’s examples (42) and (44), given here as (3) and (4): (3)

(4)

In (3), Lappin and McCord’s system ensures that him does not corefer with John in the first conjunct, and that him does not corefer with Bill in the second. In (4), the system ensures that herself corefers with the girl in the first conjunct, and that herself corefers with Mary in the second.

Page 4 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics In constructing this integrated system, Lappin and McCord seek to show that the ellipsis site contains syntactic material that is subject to ordinary syntactic constraints, such as the constraints on pronominal and anaphor binding. This became a controversial topic in subsequent literature: Kitagawa (1991) makes a somewhat similar argument for copying syntactic material from antecedent to ellipsis site, to which binding theory constraints apply. Fiengo and May (1994) also claim that ellipsis sites contain syntactic material to which constraints on binding can apply. However, Fiengo and May note many cases where such constraints do not apply to elided material in this way, and as a result, introduced some complex machinery to attempt to account for these observations, including the mechanism of “vehicle change.” The issue of the extent to which binding constraints apply to elided material remains controversial.

14.3.2 Unification The unification-based approach to ellipsis, originally introduced by Dalrymple et al. (1991), is probably the most influential computational contribution to the study of ellipsis.

14.3.2.1 Ellipsis: An abstract characterization Dalrymple et al. (1991) begin with what they call an “abstract statement of the ellipsis problem”, starting with (5). (5)

(p. 346)

Referring to (5), Dalrymple et al. (1991) give the following characterization of

ellipsis: The sentence is interpreted as meaning that Dan and George both like golf. The source clause, ‘Dan likes golf’, parallels the target ‘George does too’, with the subjects ‘Dan’ and ‘George’ being parallel elements, and the VP of the target sentence being vestigially represented by the target phrase ‘does too’. (p. 400) This establishes some important terminology for Dalrymple et al. (1991): • Source Clause: the antecedent clause for the ellipsis occurence • Target Clause: the clause containing the ellipsis • Parallel Elements: the source and target clauses contain one or more pairs of parallel elements. Parallel elements typically share the same grammatical role, but this is not necessary. Dalrymple et al. (1991) go on to describe “the problem of ellipsis interpretation” as follows: Given this abstract view of ellipsis, the problem of ellipsis interpretation is just to recover a property of (or relation over) the parallel element (respectively, elements) in the target that the missing or vestigial material stands proxy for. Of Page 5 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics course this property is not arbitrary. We know that the application of the property or relation to the parallel elements in the source constitutes the interpretation of the source clause. In example (1) above, we know that the property P being predicated of George in the second sentence is such that when it is predicated of Dan, it means that Dan likes golf. Dalrymple et al. (1991) state this with an equation, as follows: (6)

They go on to point out that a possible value for P is the property represented by the lambda term λx.like(x, golf). The general, abstract problem of ellipsis is given as follows: (7)

Once P is determined, the interpretation of the target clause is P(t1, t2,…, tn), where t1 through tn are the parallel elements in the target clause. This is the abstract characterization of the ellipsis problem offered by Dalrymple et al. (1991). Remarkably, it essentially constitutes their entire analysis, because, as they show, the equational statement completely determines a set of interpretations for target clauses.

14.3.2.2 Generating interpretations with unification In general, unification provides a technique for determining the values of variables in equations. In the case of ellipsis, as in the general equation given in (7), the value of the variable is a lambda term, in particular, a function from one or more individuals to a (p. 347) sentence meaning type (for example, a truth value). Dalrymple et al. (1991) refer to Huet (1975), which provides a method to enumerate solutions for such equations—that is, possible substitutions for P, which represents the ellipsis. Thus for (7) the solution for P is this: λx1, x2,…, xn.s[s1/x1, s2/x2,…, sn/xn] There are two steps to the construction of such a solution: first, each parallel element must be abstracted over, giving lambda-bound variables x1, x2,…, xn. Second, the body of the lambda term is the same as s, except that each parallel element is replaced with the corresponding lambda-bound variable. A complication arises when a parallel element occurs more than once in s. If there are multiple occurrences, at least one occurrence must be substituted with the lambda variable. The other occurrences may or may not be substituted. This issue of multiple

Page 6 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics occurrences is crucial, because it provides the basis for strict and sloppy readings in ellipsis. This is illustrated with (8): (8)

We assume that the interpretation of (8) is: (9)

As Dalrymple et al. point out (1991: 406), there are in principle four possible substitutions for P that solve the equation in (8): (10)

The first two solutions do not correspond to attested readings. To capture this, Dalrymple et al. (1991) introduce the notion of primary occurrences, and introduce the primary occurrence constraint which states that primary occurrences must be abstracted over. In the first two solutions for (8), the underlined occurrences of dan are primary occurrences that are not abstracted over, and thus they are ruled out. For Dalrymple et al. (1991), a primary occurrence is a term that is directly associated with a source parallel element. However, neither of the key terms, directly associated and parallel element, are given a formal definition, and they note that “Intuitively, the distinction between primary and secondary occurrences is clear:…At present, however, we have no way of making [this] precise…” (p. 412).2 I will return to this issue below. To produce the interpretation for the target clause, we select one of the two remaining solutions (3 or 4 in (10)), and apply the property P to the parallel element, George in the target clause, as follows: (p. 348)

(11)

Page 7 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics

Dalrymple et al. (1991: 413) go on to present the following algorithm to produce solutions under certain restrictions (“second-order matching”):3 Given an equation of the form: P(s1, s2,…, sn) = s0 Construct a term s from s0 by replacing zero or more instances of the si by xi. For each such s, construct a possible binding for P given by λx1…λxn.s. Clearly, there are at most 2c such unifiers… This algorithm provides the set of possible readings for ellipsis occurrences. Next we will see how it differs from previously proposed approaches.

14.3.2.3 Comparison with “Identity of Relations” accounts Dalrymple et al. (1991) contrast their account with what they term “Identity of Relations” accounts; the most prominent example at the time of their writing was undoubtedly that of Sag (1976a). In an important sense, Sag’s account actually involves the same algorithm as Dalrymple et al. (1991): first a lambda abstract is produced—for Sag, this is performed by the Derived Verb Phrase rule, which always produces a lambda abstract with a single lambda-bound variable, associated with the subject. Next, there is an optional Pronoun Rule, where indexed pronouns can be replaced with a lambda-bound variable. If we look back at example (8), it’s easy to see that Sag’s algorithm produces the same two solutions as the Dalrymple et al. (1991) algorithm above. There are, however, two key differences between Dalrymple et al.’s (1991) account and Sag’s: these have to do with generality and directionality. Generality. Sag imposes two requirements in the application of his algorithm: 1. There is a single parallel element: the Subject; 2. The variable to be solved for is the VP meaning. Dalrymple et al. (1991) do not impose these requirements; from their perspective, these conditions are typically (but not necessarily) met in the case of VP-ellipsis. By relaxing these requirements Dalrymple et al. (1991) can apply their algorithm to other forms of (p. 349) ellipsis, and also capture certain cases of VP-ellipsis where other parallel elements are required, such as Dalrymple et al.’s (1991: 420) example (30b), given here as (12): (12) Page 8 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics

The reading of interest is this: (13)

For Sag’s account, the sloppy reading is not possible—the only way to derive a sloppy reading is to replace a pronoun with a lambda-bound variable, but only when the pronoun corefers with the subject. Here there are two pronouns, her and him, which get sloppy readings. But neither of them corefer with the subject. Dalrymple et al. (1991) show how the desired reading can be derived here, with parallel elements other than the subject. The first clause receives the following representation: refuse(pwi(m, j), give(m, phone(j))) It is determined that there are three parallel elements: in the source clause, they are ‘the person who introduced Mary to John’, ‘Mary’, and ‘John’. The equation is therefore: P(pwi(m, j), m, j) = refuse(pwi(m, j), give(m, phone(j))) The desired sloppy reading is given by the following solution for P: λx.λy.λz.refuse(x, give(y, phone(z))) When applied to the parallel elements of the target clause, we get the desired reading: P(pwi(s, b), s, b) = refuse(pwi(s, b), give(s, phone(b))) Directionality. We arrive now at what I consider the most important contribution of the Dalrymple et al. (1991) approach: it is the first account in which the problem of ellipsis is approached in what I call a backward-looking fashion. Previous accounts, including that of Lappin and McCord discussed above, and the influential account of Sag (1976a), are what I call forward-looking accounts. What I mean by a forward-looking account is that the relevant interpretative decisions are made in the construction of the antecedent, which are then carried forward in providing the interpretation of the ellipsis. In particular, the Pronoun Rule in Sag’s account is applied in the construction of the antecedent, and this decides whether a given pronoun will receive a strict or sloppy interpretation under ellipsis. In the backward-looking approach of Dalrymple et al. (1991), the decision of strict or sloppy is not made in constructing the antecedent, but is only made as part of the interpretation of the ellipsis.

Page 9 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics Forwards vs backwards might seem like a distinction without a difference, or perhaps merely a matter of processing. And indeed, in many cases the predictions are the same, as we saw with example (8). However, Dalrymple et al. (1991) demonstrate convincingly that the forward-looking approach is too restrictive in certain key cases— these are cases where a given clause serves as source for more than one elliptical target clause. Here there is a clear difference in the prediction of the forward-looking vs backward-looking approach. (p. 350)

This is shown for two key examples in Dalrymple et al. (1991): the case of “Cascaded Ellipsis” (p. 423) and their example (55) (p. 432). In their discussion of these examples, Dalrymple et al. (1991) show that forward-looking accounts such as Sag’s are insufficiently flexible. Consider Dalrymple et al.’s (1991) example (41), given here as (14): (14)

This example is originally due to Dahl (1973), who points to the following reading, which I will term the mixed reading: (15)

Two other readings are possible, which I will term across the board strict (16) and across the board sloppy (17). (16)

(17)

The forward-looking approach of Sag only permits the across the board strict and sloppy readings. On the mixed reading, the pronoun he first gives rise to a sloppy reading and then a strict reading. For Sag’s approach this would not be possible. If the Pronoun Rule Page 10 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics applies in the derivation of the source clause, then we have the across the board sloppy reading. If the Pronoun Rule does not apply, we have the across the board strict reading. This argument applies not only to Sag’s approach, but indeed to any approach in which strict and sloppy readings are tied to a specific choice made in the derivation of the source clause. In my view this constitutes a compelling argument that readings in ellipsis are determined in a backward-looking fashion—a forward-looking account simply cannot provide the necessary flexibility for the “cascaded ellipsis” example. It should be noted that Dalrymple et al.’s (1991) account appears to overgenerate with respect to this example, permitting other mixed readings, such as: (18)

This reading is easily generated, if the first clause functions as the source clause for both ellipsis clauses (this is pointed out by Fiengo and May 1994). One (p. 351)

counterargument to this is that there is a requirement or strong preference that the second clause functions as the source for the third clause, because of a general preference for more recent antecedents. While no such restrictions are mentioned by Dalrymple et al. (1991), one might imagine that such preferences or constraints might arise independently from a model of discourse. I will say more below about how the Dalrymple et al. (1991) account is best integrated with a model of discourse.

14.3.2.4 Non-elliptical counterparts In characterizing the interpretive possibilities for ellipsis examples, Dalrymple et al. (1991) fail to address what might seem to be an obvious question: what are the interpretive possibilities for similar examples that do not involve ellipsis? At the time when Dalrymple et al. (1991) was published, they were not alone in this omission—this was essentially a problem with the entire literature on ellipsis up to this point. To my knowledge, it is Tancredi (1992) that first clearly brings this issue to the fore. Tancredi points to examples like the following: (19)

The italicized said he’s intelligent is deaccented, because, Tancredi argues, it is sufficiently salient in prior discourse. Tancredi points out that accounts such as that of Dalrymple et al. (1991) correctly characterize the two readings of the ellipsis example in (19), but do not apply to the apparently very similar case involving deaccenting, where one observes the exact same strict/sloppy ambiguity options for the embedded pronoun Page 11 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics he in both examples: it can be strict or sloppy. And this would appear to hold for many or perhaps all of the ellipsis examples considered by Dalrymple et al. (1991) and prior literature. Thus Tancredi argues that any account, like that of Dalrymple et al. (1991), that is specifically tailored to ellipsis occurrences, is missing a generalization. This suggests a revisioning of the Dalrymple et al. (1991) account: rather than an account of ellipsis occurrences, it could be thought of as a way of applying higher-order unification to determine interpretive possibilities for related clauses in discourse more generally. And this is perhaps the primary contribution of Dalrymple et al. (1991)—the argument for a backward-looking approach to the interpretation of parallel structures in discourse, within a unification-based framework. This perspective has been a fruitful one in subsequent literature, with works such as Prüst et al. (1994), Hobbs and Kehler (1997), Asher et al. (2001), and Hardt et al. (2013) all taking the general approach that the interpretive possibilities of a target clause can be characterized by applying a unification-like mechanism to determine common material. Possible readings, as well as preferences among them, can be seen as side effects of applying the unification mechanism. A discourse process of this sort has the additional benefit of providing a way to address the important issue of parallel elements. While this is left unspecified by Dalrymple et al. (1991), these discourse approaches all provide mechanisms for this, as a part of the general process of determining a more general semantic structure.

(p. 352)

14.4 Licensing

What factors must be present in the surrounding environment to permit an ellipsis occurrence? While this question has not been directly addressed in the computational literature, some closely related topics have been addressed. In this section I look at identifying ellipsis occurrences and the generation of ellipsis.

14.4.1 Identifying ellipsis occurrences For any computational application, the identification of ellipsis occurrences is a prerequisite for any further processing of ellipsis. Thus the study in Hardt (1997a) concerning identification of ellipsis antecedents includes a simple approach to the identification of ellipsis occurrences. This work applies to a corpus of WSJ texts from the Penn Treebank, and uses Treebank parse tree patterns to identify ellipsis occurrences. This approach was only partially successful: on a small sample Hardt (1997a) reports recall of 44 percent and precision of 53 percent. Nielsen (2003) improves on this, extending Hardt’s patterns with POS tags and lexical information to achieve recall of 69.6 percent and precision of 85.1 percent—a considerable improvement over Hardt’s results. Bos and Spenader (2011) report on an annotated corpus of the complete Wall Street Journal Corpus distributed with the Penn Treebank. This is a resource that can be used to study VP-ellipsis phenomena including VP-ellipsis resolution systems. Bos and Spenader Page 12 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics (2011) make some interesting observations about the frequency of various constructions. They note that comparative and equative constructions are very frequent, constituting 31 percent of the data, although such constructions are not prominent in theoretical discussions of VP-ellipsis. They point out that “the standard example of VPE in theoretical work consists of two sentences conjoined with and, where the second sentence is marked by the presupposition trigger too” (p. 487). But this structure is actually quite infrequent in the data, with only five cases. They also note that the strict–sloppy ambiguity which has received so much attention in the theoretical literature is virtually entirely absent: only nine examples include a bound pronoun in the source clause, and in all nine cases, the sloppy interpretation is strongly preferred according to Bos and Spenader (2011). Of course, as Bos and Spenader (2011) point out, the fact that some of these VPE types are infrequent doesn’t necessarily mean they are not worth studying. But surely an awareness of the actual distribution of VPE types is potentially valuable, and perhaps could provide theoreticians with some new perspectives. Fernández et al. (2004) report on a study of sluicing in the BNC, focusing on what they call “bare sluices”, i.e., sentences that consist of a single wh-word (or a phrase of the form which N). They use regular expressions to identify 5183 bare sluices in the approximately 10 million word dialogue transcripts of the BNC. Fernández et al. (2004) point out that two very common types of such bare sluices in dialogue have not received much attention in the theoretical literature, where focus has been on embedded sluices. Fernández et al. (2004) define the direct sluice as one “used in queries to request further elucidation of quantified parameters,” and the reprise sluice is “used to request clarification of reference of a (p. 353) constituent in a partially understood utterance.” Fernández et al. (2004) give (20a) and (20b) to illustrate direct and reprise sluices: (20)

Fernández et al. (2004) report that direct and reprise sluices constitute over 75 percent of all sluices in the BNC.4 Fernández et al. (2004) manually annotate the sluice category for 665 bare sluices. Using machine learning techniques, Fernández et al. (2004) produce a classifier with success rates of approximately 90 percent.

Page 13 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics

14.4.2 Generation Hardt and Rambow (2001) explore the generation of VPE, and “present conditions under which verb phrases are elided based on a corpus of positive and negative examples” (p. 1). Using data from the Penn Treebank, Hardt and Rambow (2001) draw on positive examples from previous work (Hardt 1997a), and for negative examples they identify cases where ellipsis had not occurred, although two VPs with the same meaning appeared within reasonably close promixity. A variety of syntactic and discourse factors were explored using machine learning techniques, and it was found that “the decision to elide VPs is statistically correlated with several factors, including distance between antecedent and candidate VPs by word or sentence, and the presence or absence of syntactic and discourse relations” (p. 8). They argue, “These findings provide a strong foundation on which to build algorithms for the generation of VPE.” They go on to examine where the decision to elide should occur in the generation process, and they argue that the decision should not be made in the text planning phase, but rather in the sentence planner or realizer phase. This is because the decision appears to depend on syntactic information that would not be available to the text planner.

14.5 Other topics 14.5.1 Antecedent location Hardt (1997a) reports on a study of 644 examples of VPE from the Penn Treebank, describing a system for determining the antecedent of ellipsis occurrences. The system applies to parsed sentences from the Penn Treebank, and the goal of the system is to determine the syntactic material that constitutes the antecedent for a given VPE example. (p. 354) The system determination is compared with the determination of a human coder, with different levels of strictness: 1. Head Overlap: either the head (i.e., the main verb) of the system choice is contained in the coder choice, or the head of the coder choice is contained in the system choice. 2. Head Match: the system choice and the coder choice have the same head. 3. Exact Match: the system choice and the coder choice are exactly the same. Hardt reports results of 94.8 percent, 84.4 percent, and 76.0 percent on these three measures, and notes that a baseline approach of always picking the most recent antecedent achieves much lower results of 75.0 percent, 61.5 percent, and 14.6 percent. The antecedent location system filters potential antecedents with a syntactic filter that eliminates potential VP antecedents that contain the VPE occurrence in a sentential complement, as in the following example: (21)

Page 14 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics

Here, the VP headed by “said” is ruled out as a potential antecedent for the VPE occurrence. Remaining candidates are then ordered according to the following preference features: 1. Recency 2. Clausal Relations 3. Parallelism 4. Quotation The Clausal Relation preference is illustrated by Hardt’s example (8), given here as (22): (22)

Here the correct antecedent, the VP headed by felt, is modified by the comparative clause containing the VPE. The VP headed by discuss is actually closer to the VPE, but the Clausal Relation preference causes the correct antecedent to be selected. Hardt suggests that there is a general preference for similar Parallel Elements, and indeed such a preference is either explicit or implicit in works such as that of Dalrymple et al. (1991), Asher (1993), Prüst et al. (1994), and Hobbs and Kehler (1997). A typical VPE configuration contains a subject and an auxiliary and sometimes other elements; Hardt suggests that there is a preference for similarity for each element’s parallel counterpart. Hardt only implements a preference for similar or identical auxiliaries; similarity of subjects is also suggested, but was not implemented. The Quotation preference is that a VPE within quotation prefers an antecedent that is also quoted. Overall, Hardt points out that the success rate of his system for VPE compares favorably with pronoun resolution systems. The preference factors have interesting relationships to discourse factors: the similarity of parallel elements is a basic feature of many leading theories of discourse relations and the existence of a syntactic clausal relation is one way to signal a discourse relation. Nielsen (2003) builds on this work, describing an end-to-end solution that identifies VPE occurrences, and locates the antecedent. (p. 355)

14.5.2 Dialogue

Dialogue systems have long been an important area for computational linguistics, and there has been an awareness that ellipsis occurs frequently in dialogues. Carberry (1989) examines intersentential sentence fragments in the context of task-related dialogues, arguing that discourse context and conversational goals play an important role in interpreting such fragments. Frederking (1993: 6) notes that “Natural language

Page 15 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics interfaces need to be able to interpret elliptical utterances, both because they tend to be quite frequent in dialogue and because they allow briefer and more natural interactions.” The emphasis on dialogue can provide a fresh perspective on the study of ellipsis, because it brings to light frequently occurring forms that haven’t received much attention in the theoretical literature, and these different forms can help to inform the theoretical debate in new ways. As was noted in section 14.4.1, Fernández et al. (2004) found that “bare sluices” like Clarifications are quite frequent in the BNC, and in an examination of the dialogue form of Clarification ellipsis, Ginzburg and Cooper (2004) raise fundamental issues about the nature of utterances and semantic representations.

14.5.3 Text mining Chae et al. (2014) describe a method for resolving ellipses in coordinated NPs, for the purpose of disambiguating Named Entity mentions in the biomedical domain, and they point out that Named Entity Recognition is important in many text-mining applications. An example of an elliptical coordinated NP is alpha- and beta-globin, where the intended interpretation is alpha-globin and beta-globin. Bergsma et al. (2011) also investigate the problem of elliptical coordinated NPs. They report on an approach in which systems are trained on the Penn Treebank, the multilingual Europarl texts, and Google N-grams. As Bergsma et al. (2011) point out, “…if an Internet search engine is given the phrase rocket attacks as a query, it should rank documents containing rocket and mortar attacks highly, even though rocket and attacks are not contiguous in the document” (p. 1346). These recent works are interesting in perhaps suggesting some new general directions for ellipsis research in computational linguistics. They take as their point of departure concrete practical challenges, such as the optimization of search queries and the improvement of parsing accuracy. Furthermore, in the case of Bergsma et al. (2011), they address the problem using machine learning techniques applied to large collections of data.

14.6 Conclusions The perspective of computational linguistics has provided significant insights in the investigation of ellipsis over the past several decades. The higher-order unification approach has brought important empirical material to the fore, and also has given rise to a backward-looking approach to ellipsis which I believe has been a very fruitful one. Furthermore, computational linguistics has brought focus on problems that have otherwise (p. 356) been given little attention, such as the identification of ellipsis occurrences and the identification of their antecedents. Most of the work described in this chapter focuses on VP-ellipsis, which is also the most well-studied form of ellipsis in the theoretical literature. As we have seen, VP-ellipsis is fairly infrequent in English, and indeed does not seems to exist in most of the world’s languages. On the other hand, sluicing is a very widespread form of ellipsis—as Page 16 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics Fernández et al. (2004) report, 33 percent of all wh-queries are direct or reprise sluices in the BNC. Thus it makes sense for computational linguistics researchers to focus more on sluicing than on VP-ellipsis; the analysis and processing of elliptical questions is a topic that has evident practical importance for NLP systems. It is not always clear what to conclude from distributional facts about ellipsis phenomena: for example, I find it surprising that the strict–sloppy ambiguity in VP-ellipsis is apparently absent from the entire million-word Penn Treebank, but of course that does not make it a less interesting object of study for theoretical linguistics. But surely such distributional analyses can be relevant for theoretical investigations. Consider the ongoing corpus development described by Anand and McCloskey (2015), where they are in the process of extracting and annotating 4100 sluicing examples from the Gigaword Corpus. Anand and McCloskey (2015) report that, while only 10 percent of the data has been annotated, they have “already encountered phenomena of real theoretical interest” (p. 7). For example, they describe a previously unknown phenomenon of modal mismatch, where the interpretation of the elided material requires a modal that is different from the corresponding modal in the antecedent, as in Anand and McCloskey’s (13), shown as (23): (23)

Here there is a shift from could to would. This kind of difference between antecedent and ellipsis has not been previously observed, and Anand and McCloskey (2015) argue that it presents a challenge for leading theories. It seems to me that this represents a promising direction for future work on ellipsis in computational linguistics: focusing on frequent types of ellipsis with a large data collection effort. Not only does this sort of work have the potential to provide new insights of theoretical relevance, but it provides the kind of empirical foundation necessary to develop systems to process ellipsis in ways that have real practical relevance.

Notes: (1) For example, Kennedy (2003) gives Dalrymple et al. (1991) as an example of an approach that claims that “elided constituents have no syntactic representation at all.” (2) See Gardent and Kohlhase (1996) for a proposal using Higher-Order Coloured Unification to provide a general theory for the primary occurrence constraint. (3) An anonymous reviewer points out that Stirling (2009) has shown that higher-order matching is decidable, so the restriction to second-order matching is no longer required. Page 17 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Computational Linguistics (4) In addition to Direct and Reprise, the categories of bare sluices include Clarification, Unclear, and Wh-Anaphor.

Daniel Hardt

Daniel Hardt is Associate Professor in the Department of Management, Society, and Communication at Copenhagen Business School, and is Visiting Research Associate at the University of California, Santa Cruz. He holds a PhD in Computer Science from the University of Pennsylvania. His research deals with both theoretical and computational linguistics, with a particular interest in ellipsis and other matters involving semantics, anaphora, and discourse. He has published articles on ellipsis in journals such as Linguistics and Philosophy, Journal of Semantics, and Computational Linguistics.

Page 18 of 18

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

Oxford Handbooks Online Ellipsis and Prosody   Susanne Winkler The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.15

Abstract and Keywords This chapter investigates the relation between ellipsis and prosody. Prosody is most frequently described as “the organizational structure of speech” (Beckman 1996). It has been defined in terms of three independent factors of the phonological representation: “intonation, phrasal rhythmic patterning and prosodic phrasing” (Selkirk 1995: 550). Prosody plays an important role in the interpretation of elliptical utterances and bridges the gap between what is overtly expressed and what is understood. There are three specific research areas where prosody has been claimed to be relevant: first, prosodyrelated licensing of ellipsis; second, prosody-related conditions of recoverability of deletion; third, prosodic effects with respect to the question of whether there is structure in the ellipsis site. In this area, research has focused on prosodic conditions on extraction and locality. This chapter is structured accordingly. It first summarizes the prosodic system of English and then reviews the research on how prosody bears on the central issues of the theory of ellipsis (licensing, recoverability, structure in the ellipsis site). The discussion shows that the interaction of prosody and the absence of sound contributes to the understanding of the theories of sound–meaning correspondence. Keywords: prosody, intonation, pitch accent, prosodic phrasing, prosodic licensing, focus, deaccentuation, parallelism, recoverability, contrast

Susanne Winkler

15.1 Introduction THE study of ellipsis and prosody explores the interrelatedness of the presence and the absence of structure and sound in relation to meaning. The term prosody is most frequently described as “the organizational structure of speech” (Beckman 1996: 21; Speer and Blodgett 2006). It has been defined in terms of three independent factors of the phonological representation: “intonation, phrasal rhythmic patterning and prosodic Page 1 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody phrasing” (Selkirk 1995: 550). Prosody, thus, also constitutes a cover term for intonation that more narrowly refers to the pitch movement (Beckman and Venditti 2011: 486). Prosody and intonation affect the interpretation of sentences (Beckman and Pierrehumbert 1986; Pierrehumbert and Hirschberg 1990; Féry 1994; Ladd 2008; Wagner and Watson 2010). The claim is that prosody plays an important role in the interpretation of elliptical utterances and bridges the gap between what is overtly expressed and what is understood, as is shown in (1) and (2). The omission of linguistic material is marked by an underscore, and pitch accents by capitals. (1)

(2)

This discussion of ellipsis and prosody is couched in the phonological model of intonation and prosodic structure developed by Pierrehumbert (1980), Beckman and Pierrehumbert (1986), Pierrehumbert and Hirschberg (1990), Hirschberg (2004), and others, as one strand of autosegmental phonology (cf. Goldsmith 1990; Ladd 2008). The general phonological system proposed there has been further developed in an analysis system by Beckman and Elam (1997), known as the Tones and Break Indices (ToBI) analysis system for English (cf. also Beckman et al. 2006; Brugos et al. 2006), and GToBi by Grice, Baumann, and Benzmüller (2005) for German and by Jun (2005) for other languages. The ToBI prosodic system implements two basic functions of prosody: (i) pitch accent assignment and (ii) prosodic phrasing. Pitch accents mark the most prominent segments in an utterance, which are often also perceived as focused. Prosodic phrases, usually delimited by prosodic boundaries (breaks or pauses), divide speech into sense units (cf. Selkirk 1984, 2000). Together they make up the prosodic contour, often also referred to as intonational contour. More precisely, an intonational contour consists of only two phonemic tones, H(igh) and L(ow). The prosody of a sentence containing a gap, such as example (1), is made up of several possible sequences of pitch accents (H*, L*, L+H*, L*+H, H+L*, H*+L). The target of the pitch accent, indicated by a * in ToBI, is associated with the stressed syllable of a lexical element to which it is assigned (according to certain rhythmical and focusstructural considerations) and functions as an anchoring point for the fundamental frequency (f0-)contour, measured in Hertz. Another reference point for the phonological Page 2 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody description is the phrase-final boundary tones, indicated by the percentage sign %, as in H% and L%. The phrase accent, indicated by the dash – (H-, L-), controls the pitch movement between the last pitch accent of the intermediate phrase and the beginning of the next intermediate phrase, or the end of the utterance. Thus, there are two levels of prosodic phrasing over which the intonational contour is defined. The larger intonational phrase (IPh) is marked by longer breaks and the smaller intermediate phrase (ip) by shorter breaks. The idea is that an IPh is composed of one or more phrase-like prosodic units that are smaller than the IPh and larger than the prosodic word. These intermediate phrases provide a structure for the intonational phrase, as in (3): (3)

A ToBI-style pitch extraction analysis of the verb phrase ellipsis (VPE) in example (1) is provided in Figure 15.1. Example (1), which can be completed as…but two swallows probably do make a summer, contains two cases of ellipsis in the second conjunct—a noun phrase ellipsis (NPE) in the subject position and a VPE. These two types of ellipses occur in a coordinate construction Figure 15.1 Pitch extraction analysis of the VPE in (1): Distribution of pitch accents and prosodic boundaries

where the end of the first conjunct, or antecedent clause, is prosodically marked by a pause and a

low boundary tone (L−L%) and the end of the second conjunct, or elliptical clause, is also marked by a low boundary tone. The overall prosodic structure thus consists of two IPhs which both contain two parallel pitch movements: a fall (H*+L) on the number head one in the antecedent clause and a fall (H*+L) on the number head two in the elliptical clause, and a fall on the most deeply embedded DP summer in the antecedent clause and a fall on the auxiliary do immediately preceding the ellipsis site in the elliptical clause. In both IPhs, the phrase accent L- controls the pitch movement of the last H* accent and the boundary tone. The accent assignment in VPE follows from information-structural considerations. In this particular instance, do contrasts with doesn’t in the first conjunct. Different distributions of pitch accents in VPE are discussed in section 15.2.3. The prosody of the German example in (2) bears some similarity to that of the English example in (1). The overall melody of the embedded dass(that)-clause (bracketed) consists Page 3 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody of two intonational phrases with two parallel pitch accents (H*) which are realized as falls on the modals will (want) and muss (have to). Note that, although the second conjunct looks like an instance of VPE at first glance, the prosodic deletion has targeted almost all of the structure including the subject and the main verb, and therefore constitutes a case of conjunction reduction in German (cf. Hudson 1976a). The relevant observation here is that the elliptical clauses in (1) and (2) contain a means of picking up the missing material from the antecedent clauses, that is, the missing material is interpreted in the elliptical sites at the level of semantic interpretation. At the same time, the interpretations are supported pragmatically and by the parallel prosody of the contrastively focused remaining constituents, referred to as remnants, in each case. (p. 359)

The central questions with respect to ellipsis and prosody are given in (4i–iii). (4)

(p. 360)

The answers to these questions will be discussed in the following sections.

Section 15.2 investigates the prosodic characteristics of the different types of ellipses and shows that the two main groups, the contrastive and the givenness-marking ellipses, are subject to different prosodic licensing conditions (location and type of pitch accent and prosodic phrasing). Section 15.3 discusses the prosodic evidence for recoverability of deletion. Section 15.4 investigates evidence for structure in the ellipsis site from the perspective of prosody. Section 15.5 concludes.

15.2 Licensing of ellipsis Research on prosody and ellipsis has mainly concentrated on the effects of the deaccentuation and deletion of given and redundant material and on the distribution of pitch accents realized on the remnant(s) and correlate(s). In the classic view set out by Chomsky (see Chomsky 1965, 1981, 1995b), the syntactic component of the grammar accounts for the matching of sound and meaning. More precisely, syntactic structures are interpreted at two different levels, the Phonological Form (PF) and the Logical Form (LF), which constitute interfaces with other systems, the articulatory–perceptual and the conceptual–intensional systems (Chomsky 1995b: 168). Under this conception, it is one of the most important issues to find an explanation of how it is possible for speakers to produce ellipses and for listeners to interpret them in the absence of form. One central claim in the literature is that the prosody of the elliptical utterance is relevant in answering this question (cf., e.g., Hartmann 2000; Merchant Page 4 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody 2001; Schwabe and Winkler 2003; Féry and Hartmann 2005; Repp 2009; Winkler 2005 and references quoted there). In the linguistic literature, there are two main observations with respect to the relevance of prosody in explaining the nature of ellipses: information-structurally given material is subject to prosodic reduction (givenness-marking hypothesis) and material that remains must be prosodically highlighted (contrastive remnant condition). These conditions play an essential role in the prosodic description of ellipsis and will be considered in more detail in 15.2.1.

15.2.1 Types of ellipses and their prosodic characteristics The prosodic and information-structural literature on ellipsis distinguishes at least two different kinds of ellipses, contrastive ellipsis as in (5) and givenness-marking ellipsis as in (6). The strikethrough marks the unpronounced material: (5)

(6)

Example (5) is an instance of gapping, which typically occurs in coordinate structures with parallel information structure; here the verb plays is deleted in the second conjunct and the (p. 361) remnants Anna and flute bear a contrastive pitch accent. Example (6) is an instance of VPE, where the redundant VP play the piano is information-structurally given and therefore prosodically deleted. Since here a complete VP-constituent is elided, this type of ellipsis has also been termed constituent ellipsis. Example (5), thus, is also called nonconstituent ellipsis (Ross 1967) since only the verbal head has been elided (cf. Chao 1988, and for a more recent discussion Reich 2011). The intonational contour of (5) is visually illustrated with a pitch extraction contour in Figure 15.2.

Page 5 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody The antecedent clause and the gapping clause are semantically parallel except for those points where contrastive accents signal their difference. Contrastive prosody is realized on the subject Anna and on the direct object flute in the second conjunct. They occur in a Figure 15.2 Pitch extraction analysis of gapping in contrastive relationship to (5): Parallel contrastive accents and prosodic their correlates Manny and boundaries piano in the first conjunct. The parallel interpretation allows the verbal head plays to be gapped. Kuno (1976, 1979) first observed that contrastive focus and deletion of redundant strings play a crucial role in the discourse appropriateness of gapping (see also Hartmann 2000; Johnson 2006, this volume). Compare the paradigm in (7) without gapping and that in (8) with gapping: (7)

(p. 362)

The unreduced (7a) can occur as an answer to the multiple wh-question Who

bought what? The subjects of the coordinate clauses Manny and Anna each receive contrastive rising pitch accents (L+H*) and the objects apples and bananas contrastive falling pitch accents (H*). Examples (7b) and (7c) show that these prosodic characteristics are not mandatory for the conjoined and nonelliptical clauses. Manny can occur as the subject of both conjoined clauses in (7b) and apples as the object in both clauses in (7c). There is obviously no requirement for contrast or newness on the subject and object in the second clause. This is different for the gapped clauses as shown in (8b,c): (8)

Page 6 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody The set in (8) corresponds to (7), but the verb bought is gapped from the second conjunct in each case. Gapping needs to be a matching response to a multiple wh-question Who bought what? Example (8a) fulfills this requirement and is a well-formed answer to this multiple focus question. In contrast, the examples in (8b) and (8c) are highly marked and perceived as ungrammatical. In (8b), the second coordinate clause does not contain a contrastive subject (Manny is the subject of both clauses) and in (8c), the second conjunct does not contain a contrastive object (apples is the object of both clauses). Question– answer congruence of this type will be used as a heuristic test for discourse-appropriate prosody throughout (cf. Klein and von Stutterheim 1987; Rochemont 2012). The ungrammaticality of (8b, c) results from the violation of two information-structural requirements on gapping which are expressed prosodically: the redundant verb in the second conjunct has the information-structural status of given, as defined in (9), and can therefore remain unpronounced. The remnants in gapping must be interpreted as contrastive with respect to their correlates, as defined in (10). This contrastive interpretation requirement has a prosodic reflex: the contrastive remnants must be prosodically highlighted and bear a strong contrastive pitch accent. The information-structural notion of givenness that has been central to the discussion of deletability and ellipsis (see, e.g., Halliday 1967: 206; Kuno 1975, 1976, 1982 for early observations) is informally defined as in (9) (see Krifka 2008; Féry and Ishihara 2016; Rochemont 2016): (9)

The notion of contrastive focus is defined by Krifka (2008: 247) as in (10): (10)

These information-structural notions can be understood as instructions to the PF component. The prosodic consequences are defined in the givenness-marking hypothesis (GMH) (p. 363) in (11) and the contrastive remnant condition (CRC) in (12) (cf. Hartmann 2000, 2003; Neijt 1979; Pesetsky 1982: 640ff.; Reich 2007, 2011; Sag 1976b; Winkler 2016). (11)

Page 7 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (12)

The GMH in (11) states that syntactic material is deaccented or deleted at PF if the material is either given or redundant. The CRC in (12) functions as an instruction to PF and requires that contrastive remnants must be prosodically highlighted and bear a strong contrastive pitch accent. The CRC is functionally dependent on the GMH. Gapping is an instance of contrastive ellipsis and subject to the CRC. The violation of the CRC leads to ungrammaticality as in (8b) and (8c). The pitch extraction contour of (8c) is provided in Figure 15.3. The pitch extraction contour of the gapping example in Figure 15.3 shows that the subjects of both conjuncts are contrasted (signaled by L+H*-pitch accents), but the objects are not. In particular, the DP apples in the second conjunct is given and therefore deaccented by virtue of being mentioned in the first conjunct (see (11)). The missing contrastive pitch accent on the second remnant in the gapping construction is perceived as highly marked if not ungrammatical. In this sense, the pairwise contrastive pitch accents prosodically license gapping. A case in point is provided in (13): (13)

Figure 15.3 Pitch extraction analysis of (8c): Violation of pairwise contrast in gapping

The phonetic contrast, which can be found in different varieties of English, is sufficient to license gapping in (13). Note, however, that the semantic requirement of the CRC that the remnants must be interpreted as contrastive foci and semantic alternatives is still violated, causing a humorous effect. (p. 364)

Next, consider the prosodic features of givenness-marking ellipsis. The question is: how can the ungrammatical example (8c), repeated in (14a), be saved? InformationPage 8 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody structurally, the verb bought and the object apples in (8c) have the information-structural status of given in both conjuncts, as in the response to the question Who bought apples? There is one straightforward way. We apply the givenness-marking hypothesis in (11) and delete the object, as in (14b). Note that (14b) remains ungrammatical. If the complete VP is deleted, as in (14c), the result is still highly marked. Obviously, there are further syntactic requirements on givenness-marking ellipses. As is well known from the syntactic literature, a functional head (here the inflected auxiliary did) and the affirmative particle, too, are needed in order to license VPE, as in (14d) (cf. Lobeck 1995; López and Winkler 2000; Aelbrecht and Harwood, this volume). (14)

The conclusion of this discussion is twofold: first, there are two main prosodic licensing conditions: the givenness-marking hypothesis in (11) and the contrastive remnant condition in (12). In the class of contrastive ellipsis, the contrastively focused elements are highlighted by omitting the given material. In the class of givenness-marking ellipsis, the givenness-marked complements of a functional head are prosodically reduced, as stated in (11). In addition to these prosodic licensing conditions on ellipsis, there are additional syntactic and semantic requirements that must be observed (cf. Chao 1988; Aelbrecht and Harwood, this volume; Johnson, this volume). The next two sections will discuss the prosodic licensing mechanisms of further instances of contrastive ellipsis (section 15.2.2) and givenness-marking ellipsis (15.2.3).

15.2.2 Prosodic licensing of contrastive ellipsis The term contrastive ellipsis applies to the set of constructions in (15): (15)

The set of contrastive ellipses in (15) have also been referred to as clausal ellipses in the syntactic literature. Clausal ellipses have been defined by van Craenenbroeck and Merchant (2013) as “a subspecies of ellipsis whereby an entire clause is missing, including the canonical subject position and the agreement domain, but often to the exclusion of one or more clause-internal constituents. Those constituents are usually argued to move to the (p. 365) left periphery of the clause prior to deletion” (p. 718). On Page 9 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody the basis of this syntactic account, it has been argued that the moved constituents are semantically interpreted as contrastive foci, as defined in (10). In addition, they adhere to the CRC in (12) in that they are prosodically highlighted and marked by a high pitch accent.1 This analysis seems straightforward for the contrastive focus requirement on the remnants in the elliptical constructions in (15), since here the remnants have been moved to the left periphery of the clause. In the gapping example in (15a), both remnants are assigned contrastive focus accents (cf. Johnson 2014, this volume, for references). In (15b) the complete conjoined clause is elided except for Anna and the focus particle too; this phenomenon is called stripping (cf. Hankamer and Sag 1976; Johnson, this volume; Wilder, this volume). Both remaining constituents are mandatorily assigned a pitch accent (cf. Konietzko and Winkler 2010) and must be interpreted contrastively (cf. Depiante 2000; López and Winkler 2000;Konietzko and Winkler 2010; Konietzko 2014, among others). Example (15c) is a fragment answer where the DP a new piano in the answer corresponds to the wh-expression in the question and where the head noun piano is prosodically highlighted and focused (cf. Merchant 2004a and Temmerman 2013 for a movement account and Culicover and Jackendoff 2005, this volume, for a non-movement account). The term contrastive ellipsis has also been applied to the set of constructions in (16), although they are not instances of clausal ellipsis: (16)

Example (16a) is known as pseudogapping (cf. Kuno 1975; Levin 1978, 1986). It is parallel to gapping, since the verb play is missing, but it also bears some features of VPE since the missing string occurs after an auxiliary. The complement of the missing verb, flute, is overt and contrastively focused in relation to its correlate, piano. Syntactic approaches to pseudogapping assume that the remnant has been moved out of the verb phrase to the left edge of vP (Lasnik 1999b), where the prosodic component assigns a contrastive pitch accent (cf. Gengel 2013).2 Example (16b) is an instance of comparative inversion ellipsis (cf. Merchant 2003b). A strong pitch accent is realized on the subject Manny, which occurs to the right of the auxiliary complex in the comparative clause (cf. Gergel et al. 2007). Culicover and Winkler (2008) propose that the contrastively focused subject undergoes a similar kind of movement to the object remnant in pseudogapping before the vP is deleted. Example (16c) is an instance of right-node raising (RNR). Here the predicate plays and the predicate tunes are assigned high pitch accents since they occur in a contrastive relationship (cf. Wilder 1997, this volume; Hartmann 2000; Selkirk 2002; Féry and Hartmann 2005; Ha 2008a, b, c). There are two characteristic features that the constructions in (15) and (16a,b) have in common. First, the syntactic analyses propose that the remnants are derived by A’-movement either to the edge of the clause (15) or the edge of vP (16a,b) before the clause or the verb phrase is deleted. Second, the syntax–prosody interface highlights the (p. 366)

Page 10 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody remnants with a pitch accent that marks their special discourse status. The remnants are interpreted contrastively with respect to their parallel antecedents. This feature gave rise to the observation that contrastive ellipses occur in coordinate constructions and are subject to the Parallelism Condition (PC) in (17), which has the prosodic dimension in (18). (17)

(18)

The condition in (18) provides a prosodic requirement on the discourse appropriateness of contrastive ellipsis. Prosodic and information-structural licensing conditions have explicitly been formulated with respect to gapping as in (15a) by Kuno (1976), Sag (1976b), Neijt (1979), Pesetsky (1982), Hartmann (2000, 2003), and Reich (2007, 2011). The violation of (17) and (18) leads to ungrammaticality if gapping occurs in non-parallel structures, as in (19a,b) (cf. Johnson 2006). In (19a) the gap occurs in a subordinate weil (because)-clause and not in a coordinated clause; in (19b) the gap occurs in a topicalized construction which changes the word order. (19)

The parallelism requirement also accounts for the prosodic and information-structural markedness of the so-called dangling remnant cases in (20) (cf. judgements by Schwarz 1999: 356; accents added): (20)

Page 11 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody While (20a) answers the question Who talked with whom about what?, (20b) doesn’t. In (20b) the first conjunct answers the question Who talked about what? but the gapped (p. 367) conjunct requires a wh-question with three operators. That is, the markedness of (20b) results from the violation of the PC in (17) and (18), which is supported by the fact that no uniform multiple wh-question can be found which is answered by both conjuncts. An attested German gapping example with mismatching parallelism is provided in (21): (21)

This example constitutes a case of contrastive ellipsis with mismatching semantic and prosodic parallelism because the remnants do not occur in a pairwise contrast. The first conjunct has a subject–aux–dative object–predicate structure. The elliptical conjunct, however, only mentions the subject–negation–adverb. Thus, the deletion involves the predicate plus the dative object. The parallelism condition requiring a pairwise contrast between the remnants and the antecedents is therefore violated. Due to this fact, this example is ambiguous in three ways: it can mean that she wasn’t always faithful to him (sloppy reading), that she wasn’t always faithful to her (external reference), or that she wasn’t always faithful to herself (strict reading). Pragmatically, the last two readings are probably marked. Thus, licensing of gapping requires a parallel syntactic, semantic, and prosodic structure and contrastive accents on the remnants. Mismatches show that prosodic licensing is essential for the discourse appropriateness of gapping constructions. Another type of contrastive ellipsis for which prosodic licensing conditions have been proposed is RNR as in (16c) (cf. Wilder, this volume). The phenomenon of RNR refers to a coordinate construction in which parts of the first conjunct are omitted, but are spelled out at the right periphery of the second conjunct. RNR is characterized by a specific prosodic pattern as in (22), where the capitalized verbs receive distinct contrastive pitch accents (cf. Selkirk 2002). The pitch accents are followed by clearly distinguishable intonational breaks, as shown in (22) and illustrated by the pitch extraction contour in Figure 15.4. (p. 368)

(22)

Page 12 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody In response to the context question Who did what? the subjects of both conjuncts are realized with rising accents (L+H*) and the verbs bought and broke each receive a contrastive pitch accent (H*) preceding a high boundary Figure 15.4 Pitch extraction contour of RNR in (22): tone, the latter of which Licensing pitch accents and boundary tones precedes the common target (cf. Wilder, this volume). Proceeding from left to right, the first conjunct corresponds to an IPh which ends in a high boundary tone (H%) after bought followed by a pause. The second conjunct, starting with the conjunction and, is also realized with a high boundary tone after the verb broke signaling continuation. The subjects John and Mary are interpreted as contrastive topics and the high pitch accents on the verbs bought and broke are interpreted as contrastive foci. The third IPh includes the target DP an expensive Chinese vase. Since it has not been mentioned in the context question, it is realized with a fall on vase which signals new information. If the context question mentions the target as in Did you hear the story about an expensive Chinese vase? the target is realized with L* accents and a longer deaccented target string, as is expected for given information. For RNR constructions like (22), there exist two different versions of prosodic licensing analyses. The first is a phonological reduction account by Hartmann (2000) and Féry and Hartmann (2005). They propose on the basis of the GMH in (11) that the missing target in the first conjunct an expensive Chinese vase is phonologically reduced by “radical deaccentuation” (Féry and Hartmann 2005: 69). The second is an E-feature account, provided by Ha (2008a, b). Ha proposes that RNR is licensed by a contrastively focused element which bears an E-feature as originally proposed by Merchant (2001) for VPE and sluicing. According to Ha, the E-feature in RNR “instructs PF to leave the RNRed element unpronounced” (Ha 2008a: 150). From the perspective of prosodic licensing, both accounts make the same predictions. The structural parallelism of the construction and the prosodic highlighting of the verbs license the phonological silence of the first of two redundant strings. Féry and Hartmann (2005) propose the licensing conditions for RNR and gapping in (23): (23)

Page 13 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody Féry and Hartmann observe that the prosodic phrasing in RNR generally seems to follow the same pattern: there is a high boundary tone preceding the covert target, which is followed by a pause. They propose that this specific prosodic RNR pattern at the end of the first conjunct “signals not only the incompleteness of the utterance at this stage, as well as the concomitant intention of the speaker to continue the sentence, but also the fact that some material is missing. In other words, part of the upstep could be motivated by the incompleteness of the utterance and part of it by the ellipsis” (Féry and Hartmann 2005: 102–3). Although Ha (2008a, b) challenges Féry and Hartmann’s approach and claims that an E-feature approach fares better with respect to empirical coverage, from a prosodic perspective the proposals are difficult to distinguish since they make the same empirical predictions. The typical prosody (p. 369) with the contrastive pitch accents on the verbs signals to the PF-component that the segmental and suprasegmental features of the target may remain unrealized. As for mismatches in RNR, there is an interesting case in which the typical RNR prosody seems to license an otherwise highly unpredicted instance of RNR, given in (24).3 (24)

In (24), the ellipsis occurs at the end of the first conjunct immediately following the pitchaccented noun swings and the preposition at realized with a rise. The second conjunct is subjectless (conjunction reduction); there is a pitch accent on the verb pushed followed by the target each other and it continues with high pitch accents on the particle out, middle, and on traffic. This example is an interesting case of RNR because the target is the reciprocal each other and not the prepositional phrase into the middle of oncoming traffic at the right edge of the clause. The interpretation is as follows Four or five girls took swings at each other and pushed each other out into the middle of oncoming traffic. This example calls the generalizations about RNR into question, particularly that the pivots must be final in all conjuncts (see Chaves 2014). Note that the seemingly parallel structure of the two intonational phrases disguises the fact that the example in (24) does not match the generalizations on RNR in the literature.

15.2.3 Prosodic licensing of givenness-marking ellipsis The theoretical precursor of the givenness-marking account of elliptical constructions (cf. (11) and (30)) is the so-called phonological reduction hypothesis, as in (25). (25)

Page 14 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody In the early PF-deletion accounts (cf. Tancredi 1992; Chomsky and Lasnik 1993; Chomsky 1995b), phonological deletion was applied to syntactically identical or redundant material. Chomsky and Lasnik (1993) discuss the so-predicate phrase construction in (26) and propose that deletion should be treated as an optional phonological rule which applies to deaccented material (i.e., the material which occurs in the square brackets and arguably lacks phonological accent).4 (26)

(p. 370)

The proponents of this theory assume that deaccentuation and deletion take place

at PF and follow from identity conditions that are stated at PF. Alternatively, it is assumed that the relevant constraints follow from focus considerations (cf. Rooth 1992; Tancredi 1992; W. Klein 1993; Hartmann 2000, 2003; Merchant 2001; Romero 2003; Winkler 2005; Repp 2009; Konietzko 2014). In particular, Tancredi (1992) claims that the constraints on VP-deletion can be reduced to those on deaccentuation, where deaccentuation is a process that takes place at PF. Tancredi (1992: 120) formulates this claim with respect to VPE more radically by proposing that “VP-ellipsis is no more than an extreme case of deaccenting where a VP ceases to be audible altogether.” Thus, Tancredi subsumes VPE under other cases of deaccentuation by proposing that VP deaccentuation shows restrictions similar to those active in VP deletion, as seen in the paradigm in (27) (cf. Tancredi 1992: 25–35). There are two arguments relevant for Tancredi’s claim: the first one is that the process of deaccentuation is triggered by parallel interpretation, as is deletion. This can be seen by the ambiguous paradigm in (27) (originally from Sag 1976b). Italics signal deaccentuation, strikethrough signals deletion: (27)

Following Sag’s basic insights, Tancredi observes that although the first conjunct in (27) is ambiguous between an NP-complement reading and a gerund reading, the resulting elliptical examples in (27b,d) are also only two-ways (and not four-ways) ambiguous. However, Tancredi’s particular claim is that similar conditions hold not only for ellipsis, but also for deaccentuation, as in (27a,c).

Page 15 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody Tancredi’s second argument is that the semantic interpretation of variables in the ellipsis site and in the deaccented string follow the same interpretative rules. He investigates strict and sloppy readings of pronouns in examples like (28). (28)

Tancredi observes that similar readings result in both reduction contexts. Examples (28a,b) are more extreme cases of the type of deaccentuation which can be witnessed in (28c). The difference between deletion and deaccentuation is that “identity is not required for deaccenting” (Tancredi 1992: 33), as seen in (28d). The discussion shows that the phonological reduction hypothesis is directly linked to the information-structural notion of givenness as provided in (9). Note, however, that within the e-GIVENness account (cf. Merchant 2001), Tancredi’s condition on deaccentuation which encodes a one-directional entailment relation has been strengthened by going both ways, from the antecedent to the elliptical clause and from the elliptical to the antecedent clause, as is formulated for VPE in (29). (p. 371)

(29)

The condition in (29) states that a VP can be deleted if it qualifies as e-GIVEN as defined in (30). (30)

E-GIVENness in (30) states that a VP can be deleted if it qualifies as e-GIVEN, where eGIVENness requires a mutual entailment relation to hold between the antecedent (A) and the elided VP (E). Implementing e-GIVENness in the grammar of ellipsis, Merchant (2001, 2004a, 2008b) proposes a syntactic E-feature on heads that licenses ellipsis. E-givenness applies straightforwardly to the complement of the functional heads in VPE, noun phrase ellipsis, and sluicing in (31a,b,c).

Page 16 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (31)

The syntactic literature on VPE has paid considerable attention to the functional elements that occur at the left edge of an elided VP and function as licensors of VPE. The most influential proposal, by Lobeck (1995), argues that VPE is licensed by agreement between the functional head of the clause and its VP-complement (cf. Zagona 1988b; Aelbrecht 2010; Aelbrecht and Harwood, this volume). From the information-structural perspective there is agreement that the deletion of the VP is licensed by e-GIVENness as spelled out by Merchant (2001) (see also Rooth 1992a, b, 1996; Schwarzschild 1999 for earlier proposals). There is much less work on the prosody of the remnants and the question of how they contribute to the licensing of the construction. Rooth (1992a, b) proposes that in instances where VPE occurs in parallel constructions, the deleted VP licenses contrastive focus on the subject, as in example (32a) with the corresponding LF in (32b). At LF, the scope of the focus feature F (MaryF) is marked by the focus interpretation operator ~ (informally referred to as squiggle operator) which ensures that congruence is satisfied with respect to the coindexed antecedent.5 (32)

Two alternative proposals have been made that consider VPE in non-parallel constructions. López and Winkler (2000) propose that these instances of VPE are licensed (p. 372)

by focus on the negative/affirmative features (polarity) situated in a functional category called Σ, which takes a VP as a complement, as in (33) and the attested example in (34):6 (33)

(34)

While in (33) and (34) the subject in the elliptical clause is pronominalized, it is deleted by the process of conjunction reduction (Hudson 1976a) in (35).

Page 17 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (35)

Gergel (2009) extended this proposal by suggesting that the aspectual feature [+tense] licenses VPE in modal–auxiliary combinations as in (36) (see also Aelbrecht 2010; Temmerman 2012; Aelbrecht and Harwood 2015): (36)

It seems that a unifying proposal is missing. Note, however, that the theory of focus and givenness marking in question–answer contexts would predict exactly the behavior observed in these examples. The prosodic requirements in VPE are not fixed by the CRC and the PC, but they follow from the theory of information structure and discourse appropriateness (cf. Kehler, this volume), as seen in (37). (37)

In (37B), the subjects of both conjuncts are contrastive topics, the negated modal in the elliptical clause is contrastively focused (could vs couldn’t), and perfective have and progressive been are deaccented. In (38B), it is shown that if the focus is realized on been, the elliptical clause is marked, since progressive been is obviously not a licensor of VPE. (p. 373)

(38)

The GMH also applies to sluicing in (39) and deletes the phonologically redundant material it exists following the wh-expression where. (39)

Page 18 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody According to Merchant (2001, 2008b), the E-feature on C licenses the deletion of the TP in (40) (see also Vicente, this volume). (40)

According to the GMH, the TP it exists is deleted at PF and the remnant of the elliptical construction is prosodically highlighted. The remaining overt wh-expression where bears a downstepped H* pitch accent immediately followed by an L- phrase accent and an L% boundary tone. Note that sluicing is traditionally analyzed as a givenness-marking ellipsis (cf. Merchant 2001). At the same time, sluicing is also an instance of contrastive ellipsis since the wh-constituent is moved to the left periphery and the CRC in (12) also holds. The remnant must carry a pitch accent which is interpreted as a contrastive focus (cf. Molnár 2006; Repp 2009, 2016; Molnár and Winkler 2010). This section has answered the question asked in (4i), namely how does prosody contribute to the licensing of ellipsis? The discussion has shown that prosody participates in the licensing process of ellipsis together with syntactic, semantic, and informationstructural constraints. In particular, it has been shown that there are two different types of constructions, where the types of accents fulfill a specific licensing function. The contrastive ellipses are licensed in parallel contexts by a contrastive accent on the remnants with respect to their correlates, as stated in the CRC (12). The givennessmarking ellipses are licensed by deaccentuation or deletion of a given complement of a functional head, as was first proposed in the phonological reduction hypothesis. This hypothesis has been incorporated into a more general information-structural requirement, the givenness-marking hypothesis in (11). The prosodic regularities of both types of ellipsis in discourse follow from question–answer congruence.

15.3 Prosody and recoverability of ellipsis site The general assumption is that the syntax–semantics interface is involved in recovering ellipsis (e.g., Reich 2011; van Craenenbroeck and Merchant 2013). This section will investigate whether the syntax–prosody interface takes any part in the reconstruction of the (p. 374) ellipsis site and how it contributes to the process of ellipsis recovery (cf. 4ii). In particular, prosody has been claimed to be relevant with respect to conditions of recoverability of deletion fed by the prosodic requirements of the contrastive remnant condition, the givenness-marking hypothesis, the parallelism condition, and the position of pitch accents and prosodic boundaries. The discussion concentrates on different types of mismatches that allow one to draw specific conclusions about the division of labor between the syntax–phonology and syntax–semantics interfaces. Where relevant, the discussion also refers to psycholinguistic research.7

Page 19 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

15.3.1 Phonetic mismatches At first glance, it seems straightforward that the representation of the elliptical site cannot simply be a surface identity of phonetic sounds as represented at PF. An intriguing phonological surface structure mismatch is provided by van Craenenbroeck and Merchant (2013: 710) in (41a), where the second of the homophonous words (right and write) is deleted in VPE (see also Arregui et al. 2006). The second verb pronounced [raɪts] cannot be recovered. A similar point can be made with respect to the German gapping example in (41b), which illustrates that the phonologically similar verbs mahlen (grind) and malen (paint) both pronounced as [maːlt] in the third person singular do not allow the recovery of the ellipsis in the second conjunct. (41)

However, it is too early to conclude that phonetic and phonological representations are completely excluded from the recovery process of ellipsis. Although the identity relation cannot be determined by the phonetic structure at PF alone, there are examples where the syntax–phonology and syntax–semantics interface seem to interact in recovering the ellipsis site. Consider the recovery processes of the VPE in (42), an attested example in English. (42)

The unmarked recovery of the VPE in (42) You disarm or we will uttered out of context is the non-causative reading You disarm or we disarm. However, the speaker is George W. Bush and the statement is addressed to Saddam Hussein. In the political context of the time, the unmarked interpretation cannot be the right one. Therefore, a lexical retrieval process which involves the activation of phonetic features (cf. Frauenfelder and Floccia 2006) sets in. The lexical element disarm not only has a non-causative reading but (p. 375) also a causative one, You disarm or we will disarm you. The interpretation of this utterance as a threat is the contextually adequate one. The observation in the processing literature is that lexical retrieval processes activate the phonetic features. This can also be observed in the gapping example (43) in German. (43)

Page 20 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

Example (43) shows that gapping in German can apply to the verbal head ging (left) alone, leaving the subject NP der andere (the other), the PP von dort (from there), and the particle of the verb weg behind. Here it seems that the phonetically identical verb ging in the second conjunct is gapped although weggehen (leave) is semantically the opposite of the verb gehen nach (go to) in the antecedent clause. The observation is that gapping can occur in the second conjunct in (43) despite the fact that the semantic meanings of the lexical entries (gehen vs weggehen) differ. The discussion shows that the identity relation between the deleted string and its antecedent is a matter of syntax and semantics and is not uniquely determined by phonetic and phonological representation. However, it seems that the lexical encoding of the identity relation interacts with the phonological component and the syntacticosemantic component in the antecedent retrieval. In particular, the homophonic representations seem to suggest that phonetic representations play a role in the search process for an appropriate interpretation of these examples.

15.3.2 Prosodic disambiguation The underlying argument regarding the prosodic disambiguation of syntactically ambiguous clauses is that prosody, and in particular prosodic phrasing and pitch accent assignment, can influence semantic interpretation (cf. Winkler 2015). Prosodic phrasing and the placement of pitch accents influence the interpretation of ellipsis.

15.3.2.1 Prosodic phrasing and disambiguation Féry (1994) observed that prosodic phrasing may disambiguate homophonous syntactic structures as in (44): (44)

Example (44) has two different prosodic representations resulting from the different instantiations of the argument frame of the verb schaukeln (swing) as in (45a,b): (45)

Page 21 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

The prosody of (45a) consists of one IPh, signaling a single clause in response to a question such as What’s wrong with Maria and Martin? The verb schaukeln is interpreted as a transitive verb Maria isn’t pushing Martin (on the swing). Example (45b) is an answer to a question such as Who does what? The prosodic realization contains a break after the verb and thereby signals the division of the clause into two IPhs, where the second one is an instance of stripping in German. The prosodic realization follows the parallelism condition, where the subjects Maria and Martin are realized with rising pitch accents which are interpreted as contrastive topics. The verb schaukelt (3rd pers. sing.) in the first conjunct is assigned a focus accent, as is the negative particle nicht (not) in the second conjunct. The parallel prosody supports the interpretation of the two IPhs as two coordinate clauses, where the intransitive verb schaukeln is reconstructed in the second conjunct: Maria is swinging, Martin isn’t swinging. (p. 376)

The relevant observation with respect to the prosodic disambiguation of (44) is that the transitive single IPh-reading in (45a) is the preferred interpretation. The stripping reading in (45b) is brought about by prosodic phrasing in two IPhs and parallel pitch accent assignment. The recovery of the ellipsis site in (45b) is dependent on the givenness-marking hypothesis and the contrastive remnant condition. In addition, it conforms to the prosodic requirement of the parallelism condition in (18). The underlying argument is that this particular structural ambiguity can be prosodically disambiguated (cf. Wiedmann and Winkler 2015; Remmele et al. forthcoming). The parallel prosodic realization of the two IPhs in (45b) supports the recovery of the ellipsis site.8

15.3.2.2 Pitch accents and disambiguation Elliptical constructions have been observed to cause ambiguous readings since Ross (1967) and Hankamer (1973a). Kuno (1976) observed that information structure and prosody provide hints as to how to recover the ellipsis site. Hankamer proposed a general No Ambiguity Condition based on the observation that the example in (46a) has the reading provided in (46b) and not the one in (46c). (46)

Note that (46a) can receive the reading that Hankamer marked as impossible (46c) if an appropriate discourse question that requires a particular prosodic realization precedes the utterance, as in (47): Page 22 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (47)

The multiple wh-question in (47a) requires the deaccentuation of the verb gave and the indirect object Sally and the pairwise prosodic highlighting of the subjects and the direct objects in (47b). Thereby, the reading that Hankamer excluded arises as the preferred one. (p. 377)

The type of ambiguity shown in (46a) is referred to as subject vs object ambiguity in the literature on ellipsis and has been widely observed with case-ambiguous DP-remnants. The reading in (46b) is generally referred to as conjunction reduction, the one in (46c) and (47) is an instance of gapping. Psycholinguistic studies have investigated the factors which influence the preferred reading in ambiguous examples, as in (48) (cf. Carlson 2001; Stolterfoht 2004; Hoeks et al. 2006; Stolterfoht et al. 2007; Hoeks et al. 2009; Frazier, this volume). (48)

The psycholinguistic studies show that the question of whether Manny is interpreted as the subject or the object of the second clause in (48) mainly depends on prosodic and contextual factors, as seen in the small dialogues in (49) (gapping) and (50) (conjunction reduction, or non-gapping). (49)

(50)

Subject vs object ambiguities also occur in stripping as in (51) (cf., e.g., Radford 1997: 492; Carlson 2002; Lechner 2004; Konietzko 2014). Note that the ambiguity of the comparative ellipsis in (51a,b) cannot be resolved by the accentuation of the remnant alone. The pronoun you must be stressed in both readings. The prosodic pattern of the antecedent clause is relevant. In the subject reading, the contrasting subjects are accented as in (51a), in the object reading, the contrasting objects as in (51b): (51)

Page 23 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

The ambiguous stripping examples in English are disambiguated by nominative case assignment in (52a) and accusative case marking in (52b): (52)

A similar type of subject vs object ambiguity has been observed with respect to sluicing in (53) and (54) by Merchant (2001: 23) (cf. also Frazier and Clifton 1998; Carlson et al. 2009). (53)

(p. 378)

(54)

Note that the disambiguation in these cases depends on the context question and the contrastive prosodic highlighting of the correlate in the antecedent clause. A different type of ambiguity has been observed with respect to the remaining subject in VPE. The subject of the second clause can function either as a matrix subject as in (55B) or an embedded subject as in (56B) (cf. Frazier and Clifton 2001; Frazier et al. 2007; Traxler and Frazier 2008: 315). (55)

(56)

Page 24 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody In the example (55B) the subject reading is brought about by the context question, which requires a pitch accent on the subject John in the matrix antecedent clause and on the parallel subject Mary in the elliptical clause. In (56) John is the matrix subject of both conjuncts. Mary is the subject of the subordinate clause. All the NPs are given, except Europe, which is highlighted. The general observation is that the distribution of the pitch accents in the antecedent clause influences the interpretation of the elliptical clause. The relevant prosodic requirements are influenced by the context questions and formalized by the GMH, CRC, and the prosodic requirement of the PC.

15.3.3 Active–passive mismatches The literature on active–passive mismatches, such as in (57) and (58), is extensive (cf. Dalrymple et al. 1991; Hardt 1993; Merchant 2008a, 2010, 2013a; Kim et al. 2011). (57)

(58)

Merchant’s empirical observations of active–passive mismatches constitute strong evidence for a syntactic identity account of ellipsis. Note that there is an alternative account by Kertz (2013) which suggests that information-structural and prosodic factors play a crucial role in the interpretation of these mismatches. Kertz (2013: 390) investigates the puzzle given in (59)–(61). Example (59) is an instance of VPE. Example (60) is identical, except that it contains an active–passive mismatch and is considerably marked. The puzzle arises in (61), (p. 379) which also contains an active–passive mismatch but does not seem to be degraded. Pitch accents are added following Kertz (2013: 398). (59)

(60)

(61)

Page 25 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody The example sentences (59)–(61) differ with respect to their information structure and intonation. This can best be seen by the question–answer test. Example (59) answers the question Who reported the incident? The subject remnants occur in a contrastive relationship in a parallel structure (contrastive topics, cf. Kertz). The deleted and eGIVEN VP in the second conjunct can be readily recovered by the antecedent VP. Example (60) also occurs in a parallel construction but there is no single context question which it could answer. The subject of the first conjunct the incident and the subject pedestrian of the second conjunct do not contrast because they do not form alternatives of the same set (cf. Rooth 1992a). This is a violation of the contrastive focus requirement in (10) and causes a severe mismatch. Example (61) constitutes a non-parallel construction involving a main clause which takes an embedded clause (cf. Reich and Reis 2013). The VPE occurs in the embedded clause. Example (61) answers the context question What was reported by the driver? and provides the additional information that the driver didn’t really need to report it. The subject of the subordinate clause is a given topic he that refers back to the thematic subject of the main clause the driver, and the focus accent is realized on the auxiliary need. The VPE in the embedded clause is easily recoverable because it finds its e-GIVEN antecedent in the matrix clause. The prosody supports this interpretation, as discussed in section 15.2.3.

15.3.4 Pitch accents and the interpretation of pronouns under ellipsis Prosodic research has focused on the relation between pitch accents and the interpretation of pronominal reference for almost five decades (Lakoff 1971; Akmajian 1973; Hirst 1981;Kameyama 1985, 1999; Grosz et al. 1995; Kehler 1997, 2000; Beaver 2004). In an early paper, Hirschberg and Ward (1991) investigate the relation of pitch accent assignment and semantic interpretation in VPE contexts. They propose that speakers use the stressed pronominal form to signal a marked interpretation, as originally observed by Lakoff (1971) for the well-known example John called Bill a Republican and then he/HE insulted him/HIM. They propose two tendencies: first, the sloppy interpretation is available where no c-command relation between anaphor and antecedent obtains; second, the presence or absence of a pitch accent on the anaphor affects whether the utterance is assigned a sloppy or strict interpretation (Hirschberg and Ward 1991: 112). More precisely, they propose that “pitch accent facilitates a strict interpretation when the unmarked case (i.e. the (p. 380) interpretation favored in the written condition) is sloppy, or a sloppy interpretation when the unmarked case is strict” (p. 117). Their observations tie in with earlier observations by Reinhart (1983b). She showed that for bound pronouns, such as the reflexive in (62), the referential reading is systematically absent under VPE. That is, in (62) only the reading in which John voted for himself and his lawyer voted for himself is available, as provided in the paraphrase. Accentuation of the reflexive does not change the reading. (62)

Page 26 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

However, this prediction does not hold for all cases, as (63) from Hardt (1993) shows: (63)

VPE in example (63) occurs in a subordinate clause and allows either a strict or a sloppy reading. Here the stressed pronoun condition by Hirschberg and Ward would predict that the referential reading because his lawyer didn’t [defend him] where him refers to John is not only available, but even preferred. Note, however, that the study by Hirschberg and Ward (1991) has shown that speaker intuitions on binding relations are not as robust as they may seem at this point. This has also been one of the major points of criticism by Hardt (1993) and others with respect to the data discussed above. Other restrictions seem to play a role, such as accent assignment, focus, and discourse structure (cf. Hardt and Romero 2004; Kehler, this volume). The general answer to the second main question in (4ii) is that prosody contributes to the process of ellipsis recovery by showing that the syntax–semantics and the syntax– phonology interfaces follow specific prosody-related mapping relations.

15.4 Prosody and structure in the ellipsis site Research on the question of whether there is structure in ellipsis sites usually targets syntactic and semantic arguments. This section investigates the third main question raised in (4iii) regarding whether there is evidence stemming from prosody suggesting that ellipsis sites contain syntactic representations that are unpronounced. Prosody provides indirect evidence for structure in the ellipsis site. By investigating the prosody and information structure of the remaining elements, the prosodic structure of the elided material can be inferred (cf. Hartmann 2000; Carlson 2001; Féry and Hartmann 2005; Winkler 2013, 2015). One core idea is the complementary distribution of the complete phonological reduction of those parts that are contained in the ellipsis site and the highlighting of the remnants, as discussed in section 15.2 with respect to the givennessmarking hypothesis, the contrastive (p. 381) remnant condition, and the prosodic requirement of the parallelism condition. However, complete deaccentuation of syntactic material is not an entirely convincing argument, since empty proforms are also characterized by phonological silence (cf. Lobeck 1995). Therefore, this section presents evidence from prosody-related research that concentrates on two main topics: first, on prosodic conditions that allow movement out of

Page 27 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody ellipsis sites, and second, on locality restrictions and the question of prosody-related repair.

15.4.1 Extraction from ellipsis site and prosody It has been widely observed in the literature that A’-extraction from an ellipsis site is possible (Sag 1976b; Williams 1977b; Fiengo and May 1994; Lasnik 2001; Merchant 2001, 2008b; Schuyler 2002; Takahashi and Fox 2005; see also Abels, this volume, for an overview). In most generative accounts, the remnants of contrastive ellipsis (i.e., gapping, stripping, pseudogapping, and sluicing) are assumed to be extracted from the ellipsis site (cf. section 15.2.2). The prosodic requirement on contrastive ellipsis is that it obeys the prosodic requirement of the PC and that the movement-derived remnant must be accented and contrastively interpreted in relation to its correlate. The underlying argument is that extraction from a VPE site is sensitive to prosodic and informationstructural constraints. The logic of the argument is as follows: if the ellipsis site contains unpronounced (completely deaccented) structure, it should be possible to extract those elements that are not deaccented because they are not given. It follows that these extracted elements must be accented and contrastively focused. This is exactly what we observe in contrastive ellipsis. In extraction from VPE the wh-expression is inherently focused and requires contrastive focus on the auxiliary. The most explicit focus-related proposal for extraction from VPE is provided by Schuyler (2002). She proposes the Contrast-Locality Condition as in (64): (64)

The contrast-locality condition in (64) accounts for the grammaticality of extraction out of VP in (65a) and the ungrammaticality of (65b): (65)

The contrast-locality condition in (64) also accounts for the examples in (66)–(70), where the extracted wh-phrases, which are considered focus operators, take another focused element in their scope. In (66) and (67) the licensing pitch accent of the VPE is realized on the functional head.

Page 28 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (p. 382)

(66)

(67)

The pitch extraction contour for (67) is provided in Figure 15.5. It shows that the negative auxiliary in the first conjunct and the positive polarity item in the second conjunct are both heavily accented. Alternative prosodic realizations are marked, because they violate (64), as seen in (68a, b): (68)

Figure 15.5 Pitch extraction contour of extraction from VPE

The contrast-locality condition for VPE extraction also applies to Johnson’s example in (69B1) (from Johnson 2001b: 456). (69B1) is a discourse-adequate answer to (69A), while (69B2) is not. (69)

The contrast-locality condition also applies to topicalization from VPE, as seen in (70B1) (from Merchant 2001). The parallelism of the construction is also expressed prosodically with rising topic accents on the topicalized elements and further contrasting accents on the (p. 383) auxiliaries. The violation of (64) leads to the ungrammatical (70B2) where the auxiliary in the second conjunct is not focused. (70) Page 29 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

Building on Schuyler’s insight, Merchant (2008a) proposes the MaxElide constraint, which states that under A’-extraction out of ellipsis the largest possible constituent must be elided (cf. also Lasnik 2001; Merchant 2001; Takahashi and Fox 2005). This observation seems to capture the contrast in (71) at first sight (cf. van Craenenbroeck and Merchant 2013: 706). (71)

However, a more principled explanation of the facts is still missing. It is unclear whether the notion of economy implicit in the MaxElide constraint can account for the optionality requirement which holds for ellipsis in general. For example, one would expect (71a) to be felicitous in the fully spelled version, as in (72). (72)

However, (72) is still marked because it does not fulfill the discourse appropriateness conditions in the undeleted version. An alternative explanation is that extraction from VPE only occurs in parallel structures and must therefore strictly follow the PC and the CRC, while sluicing, which occurs in an indirect question, doesn’t have to strictly follow these constraints (cf. Winkler 2013). Under this view, the ungrammaticality of (71a) follows from the fact that the focus requirements on sluicing differ from those on VPE extraction: while extraction from VP is subject to the PC and CRC, the wh-remnant in sluicing must itself be a pitch-accented contrastive focus which occurs in a set relation to the focused antecedent. That is, there are diverging focus and prosody requirements on the extraction from VPE and sluicing which are difficult (but not impossible) to fulfill at the same time. The diagnostic test of question–answer congruence makes the focus requirements for (71a) explicit. The relevant observation is that each of the elliptical constructions with wh-extraction must be an answer to an appropriate question, as seen in (73)–(75): wh-extraction from VPE answers an implicit or explicit parallel VP question like What did Ed do and what did Mary do? as in (73A). (73)

Page 30 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

Sluicing can only occur in an indirect question and must have an indefinite antecedent that is questioned. (p. 384)

(74)

Bringing together both information-structural conditions requires a complex parallel context question which induces a parallel focus structure and parallel prosody in the answer, as in (75): (75)

The prosodic structure of the wh-extraction from VPE in (75B) conforms optimally to the focus requirements of the question in (75A). Parallel contrastive prosody is realized on Ed and on Mary as well as on the complex noun phrase lecture on Balkan languages and which one. Thus, the perfectly grammatical (75B) raises some doubt about the validity of the grammatical principle MaxElide and provides an argument for the assumption that information-structurally controlled context and prosody are the factors that determine the appropriateness of extraction from ellipsis. There are two important conclusions of this discussion: First, the extraction from ellipsis examples show that there must be structure in the ellipsis site. Second, it shows that specific information-structural and prosodic requirements must be met for extraction from VPE and from sluicing.

15.4.2 Locality restrictions and prosody The observation that A’-extraction is possible out of contrastive ellipsis and VPE under certain syntactic and prosodic conditions constitutes a strong argument for movement. However, it must also be shown that A’-extraction is subject to the same locality restrictions as the fully spelled-out versions of the ellipsis clause. Research over the last three decades has shown that A’-movement out of an island under VPE results in ungrammaticality, as in (76) and (77) (cf. Haïk 1987: 511; Johnson 2001b: 457; Merchant 2008b: 143–4). (76)

Page 31 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

(77)

Kennedy (2003: 30) shows that similar observations hold for topicalization. An object DP cannot be topicalized out of a VPE in a relative clause, as seen by the ungrammaticality of (78). (p. 385)

(78)

To the best of my knowledge, there has not been an explicit discussion of the prosodic properties of these types of extractions. Also rating studies that manipulate the prosody and verify the grammaticality judgments reported here remain a desideratum. A visualization of the second conjunct of (78) in Figure 15.6 shows that the island violation cannot be prosodically repaired. Although the prosodic requirements on this extraction construction are met (parallel contrastive accents on topicalized phrases, accent on VPE licensing auxiliary does), it

Figure 15.6 Pitch extraction contour of topicalization from VPE in a relative clause

seems that prosody cannot save extraction out of the island configuration. Example (78) remains

highly marked. This is different in sluicing where the island is fully contained in the ellipsis site, as in the widely discussed island repair cases (see Abels, this volume, and references cited there). An example that shows that the wh-remnant and its antecedent are obligatorily focused: (79)

Page 32 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody

In (79a), the remnant who refers to the object antecedent a repeat offender who robbed a store. In (79b), the remnant which (one) refers to the object DP a store inside the relative clause. As relative clauses are islands to movement, the version of (79b) without ellipsis, given in (80), is ungrammatical. The extraction site is marked by underscore: (80)

As already observed by Ross (1969b), (79b) is considerably better than (80). Previous analyses of sluicing have established that antecedent selection in sluicing is determined by information structure (e.g., Romero 1998; Winkler 2013). A recent empirical investigation of sluicing in German showed that sluicing with relative clause (p. 386)

antecedents is as acceptable as main clause antecedents if parallel focus assignment is respected and if the relative clause is extraposed which is argued to add additional prosodic weight to the focused antecedent (cf. Konietzko et al. forthcoming). The general answer to the question (4iii) is that prosody provides indirect evidence for the structure in ellipsis site. There are two arguments: The first one is theoretical. The theoretical account of ellipsis as a phonological deletion process at the level of phonological form is prosody-related. The second is empirical. Close investigation of the main syntactic argument for structure in the ellipsis site, namely extraction from the ellipsis site, shows that there are prosodic requirements and discourse requirements that license extraction. Thus, there is prosodic evidence that supports the assumption that there is structure in the ellipsis site.

15.5 Conclusion This chapter has investigated ellipsis from the perspective of prosody. It has provided a summary of the prosodic system of English and reviewed the research on how prosody bears on the central issues of the theory of ellipsis. The chapter has focused on three prosody-related questions raised by the research on ellipsis: (i) How does prosody contribute to the licensing of ellipsis? (ii) How does prosody contribute to the recovery of the ellipsis site? (iii) What evidence can prosody contribute for the assumption that there is structure in the ellipsis site? With respect to the first question, it could be shown that most theories assume that the deletion of linguistic material is subject to formal syntactic, semantic, and prosodic conditions. There are in particular two prosodic constraints that interact with contextual features, the givenness-marking hypothesis and the contrastive remnant condition. In addition, there is a prosodic requirement on the parallelism Page 33 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody condition. As for the second question, the discussion has shown that prosodic features play a considerable role in the recovery of ellipsis. In particular, it has been shown that prosody figures prominently in the disambiguation process of different readings of elliptical constructions. With respect to the third question, the prosodic evidence that there is structure in the ellipsis site is indirect. However, it could be shown that A’extraction out of an ellipsis site is subject to prosodic constraints. Finally, the discussion has shown that the interaction of prosody and absence of sound contributes to the understanding of the theories of syntax–sound–meaning correspondence.

Acknowledgements This chapter has benefited from the valuable comments of the editors of this handbook, Jeroen van Craenenbroeck and Tanja Temmerman, and two anonymous reviewers. I am also grateful to Peter Culicover, Andreas Konietzko, Jason Merchant, Valéria Molnár, and Michael Rochemont for comments, fruitful discussions, and feedback. This material is based upon work supported by the German Research Foundation (Deutsche Forschungsgemeinschaft DFG) under the following grants: SFB 833, RTG 1808.

Notes: (1) For the syntactic literature on A’-movement, see e.g. Chomsky (2001, 2004). For the syntax-information structure and prosody interface of remnant movement in contrastive ellipsis, see Winkler (2005, 2015). (2) Although pseudogapping is not an instance of clausal ellipsis, it clearly shows the same behavior at the syntax–information structure–prosody interface. The remnant is moved out of the ellipsis site, which is the vP in this instance. (3) The prosody stems from a pilot production study with three English native speakers. (4) Note that (26) is not a particularly straightforward example to demonstrate the analogy between elliptical constructions and prosodic reduction in redundancy contexts, because the do-so construction functions as a predicative anaphor and occurs in complementary distribution with VPE (see López 1995). (5) The accent assignment in VPE follows from information-structural considerations. The main contrast can be realized on the remaining subject in the second conjunct. As pointed out by an anonymous reviewer, it would be ungrammatical to assign prominence only on did as in (i). ((i))

The subjects, John and Mary, occur in a contrastive topic relation, typically expressed by a prosodic rise. Page 34 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Ellipsis and Prosody (6) See King (1970), who first observed that contracted auxiliaries which typically don’t bear a pitch accent cannot license VPE as seen in (i): ((i))

(7) The discussion of prosody in relation to the different scopal interpretations in gapping is an interesting issue but goes beyond the scope of this chapter (cf. Johnson 2004a; Winkler 2005: ch. 4;Repp 2009; Temmerman 2012). (8) From a psycholinguistic perspective one could argue that there is a temporary ambiguity at the point where the verb schaukeln is parsed.

Susanne Winkler

Susanne Winkler is Professor of English Linguistics at the University of Tübingen (Germany). She has a long-standing research interest in syntactic theory, information structure, and the syntax–prosody interface. She has written extensively on the information structure of elliptical constructions and focus constructions. She is the author of Ellipsis and focus in generative grammar (Mouton de Gruyter, 2005) and Focus and secondary predication (Mouton de Gruyter, 1997), and of papers in a variety of volumes and journals. She directs a DFG-funded research project on Focus and Extraction in Complex Constructions and Islands (SFB 833) and co-directs an interdisciplinary DFG-funded research training group (RTG 1808) on Ambiguity: Production and Preception.

Page 35 of 35

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

Oxford Handbooks Online Movement and Islands   Klaus Abels The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.17

Abstract and Keywords This chapter discusses two strands of research where the interaction of ellipsis with movement has been used to construct arguments pertaining to the implementation of movement in the grammar, the architecture of the grammar, and the nature of the grammar more broadly. Both lines of research discussed in this chapter take as their empirical starting point asymmetries in the behavior of moved constituents depending on whether movement originates within an ellipsis site or not. Movement from an ellipsis site is sometimes less restricted than movement from an overt phrase hypothesized to correspond to the silent structure at the ellipsis site and sometimes more restricted. The overall conclusions that can be drawn from the phenomena sampled here are that syntactic structure is present at the ellipsis site, that locality constraints are most likely not lifted within the ellipsis site, and that the identity condition on ellipsis is semantic/ pragmatic rather than syntactic. Keywords: islands, island repair, successive cyclicity, sluicing, contrast sluicing, multiple sluicing, punctuated paths, island evasion, case matching

Klaus Abels

16.1 Introduction THIS chapter discusses two strands of research where the interaction of ellipsis with movement has been used to construct arguments pertaining to the implementation of movement in the grammar, the architecture of the grammar, and the nature of the grammar more broadly. Such discussions stand in contrast to the simpler, more customary arguments where the possibility of extracting elements from an ellipsis site (or lack thereof) is used as a probe in the analysis of a particular elliptical construction (keeping assumptions about the analysis of movement and the architecture of the grammar essentially constant). For example, the possibility of (apparent) extraction from Page 1 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands a particular ellipsis site might be taken as evidence for the presence of articulated syntactic structure at the ellipsis site (see Aelbrecht 2010; Baltin 2012 for clear examples of this line of argumentation). Here we take the other perspective. We ask what ellipsis can teach us about the nature of movement and the grammar in general. Of course, the distinction is one of emphasis rather than one of principle, since analytical discussion informs theory and vice versa. Both lines of research discussed in this chapter take as their empirical starting point asymmetries in the behavior of moved constituents depending on whether movement originates within an ellipsis site or not. Movement from an ellipsis site is sometimes less restricted than movement from an overt phrase hypothesized to correspond to the silent structure at the ellipsis site and sometimes more restricted. The chapter is structured as follows. Section 16.2 introduces the phenomenon of WH-movement. WH-movement will be shown to be unbounded in some sense, though subject to certain constraints. We briefly review the notion of a syntactic island, that is, of a domain from which WH-movement is impossible. Then we introduce the idea that movement dependencies are created in a successive cyclic fashion; it will be contrasted with some important alternative views. Section 16.2 serves as background. It can be skipped by readers familiar with unbounded WH-movement, syntactic island phenomena, and the ways in which unbounded dependencies are modeled in different syntactic frameworks. Section 16.3 discusses the interaction of the elliptical phenomenon of sluicing with syntactic island constraints. Under certain assumptions about the analysis of ellipsis in general and sluicing in particular, sluicing ameliorates island effects (Ross 1969b). The section introduces some basic facts about sluicing, concentrating on case connectivity between antecedent and remnant and on Ross’s conjecture: the idea that (p. 390)

ellipsis repairs island violations. Subsection 16.3.1 explores the two canonical approaches to Ross’s conjecture, these are first the idea that ellipsis literally repairs island violations and second the idea that island violations cannot arise under ellipsis because there is no syntactic structure at the ellipsis site. Subsection 16.3.2 introduces two systematic counterexamples to Ross’s conjecture—contrast sluicing and multiple sluicing—and reviews approaches to sluicing that take these counterexamples at face value. Subsection 16.3.3 outlines the contour of a recent debate on the question of how damaging the prima facie counterexamples to Ross’s conjecture are. It is shown that the counterexamples to Ross’s conjecture tilt the balance in favor of approaches to ellipsis that assume the presence of unpronounced syntactic structure at the ellipsis site and against approaches that do not assume the presence of such structure. Among the approaches that do posit syntactic structure at the ellipsis site, a consensual answer to the question of whether the identity condition on ellipsis is syntactic or semantic (or both) has not been reached. Section 16.4 turns to VP ellipsis and related phenomena. VP ellipsis allows certain types of extraction from the ellipsis site but not others—at least not in an unrestricted way. The complex restrictions on movement from the ellipsis site have been used as indirect Page 2 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands arguments for a particular implementation of movement, cyclicity, and ellipsis. Facts and arguments in this area have not enjoyed the prominence of the island amelioration phenomenon. The section discusses some observations, sketches the arguments, and points out some open questions. Any conclusions must, at this point, remain extremely tentative.

16.2 Movement, islands, cyclicity This section serves as a brief reminder of some of the basic properties of the phenomenon that the generative literature calls movement and which is, in various contexts and theories, also referred to as “long distance dependency”, “unbounded dependency”, “displacement”, or “extraction”. A slightly more in-depth introduction to the phenomenon can be found in Abels (2016a). Example (1) illustrates movement of the interrogative pronoun ‘what’ to the front of the sentence: (1)

There is a dependency between the interrogative pronoun and the verb: It depends on the choice of verb whether the interrogative pronoun can appear at all—impossible with an intransitive verb like ‘sleep’—and, if it is possible, which interpretation is assigned to ‘what’—the thing created with ‘make’ and the thing destroyed with ‘burn’. Verbs that obligatorily take objects, like ‘devour’ in (2), allow their object position, exceptionally, to be empty when a suitable interrogative pronoun introduces the sentence, (2b), and, indeed, in such a case the object position must remain empty, (2c). In a nutshell, the interrogative pronoun acts semantically and for purposes of selection as the object of the verb in these examples. Gaps can occur in various positions in the clause, with correspondingly different interpretations, and different grammatical functions. (p. 391)

(2)

We can describe the relation between the interrogative pronoun and the object position in terms of fillers and gaps. In (1) ‘what’ is the filler and the object position of the verbs ‘make’ and ‘burn’ is the gap, as shown in (3). The gap is indicated by an underscore. (3)

Page 3 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

We have seen so far that the filler fulfills the selectional requirement associated with the gap and receives the thematic interpretation associated with the gap. In languages that show case morphology on interrogative pronouns, the form of the interrogative pronouns varies in the same way with the verb as the form of a non-interrogative phrase at the gap site would: (4) shows that the German verb unterstützen ‘support’ takes an accusative object while helfen ‘help’ takes a dative object. Questions about those being supported or helped must be introduced by interrogative pronouns in the appropriate form: accusative for unterstützen ‘support’ and dative for helfen ‘help’, (5). (4)

(5)

The syntactic category of a clause will here be abbreviated as CP (complementizer phrase). All CPs in (6) are enclosed by square brackets. Questions introduced by an interrogative pronoun or an interrogative phrase are called WH-questions. In such questions the WH-phrase occupies the specifier of CP in English. (6)

Furthermore, the complementizer position is occupied by an auxiliary in direct questions (6a), while the complementizer position remains empty in indirect WH-questions, (6b).

Page 4 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands The category of the remainder of the clause will be abbreviated here as TP (tense phrase). Thus, the very approximate structure of the indirect question in (6b) can be represented as in (7). The gap is represented in the structure by what. Strikethrough encodes the idea that the gap site is occupied by a silent copy of the filler. The dependency itself is represented by the arrow pointing from the silent copy to the filler. It represents the movement relation holding between the filler and the gap. Theories and analyses differ in how exactly the movement relation is represented and whether silent elements (copies or traces) occupy the site of the gap or not. Some of these issues will be discussed in subsection 16.2.2 and section 16.4. The notation with a struck-out copy and with (or usually without) an additional movement arrow to indicate the long-distance dependency will be used for convenience and perspicuity when nothing hinges on the choice. (p. 392)

(7)

16.2.1 Islands The examples up to this point have all contained a filler and a corresponding gap in the same clause. It is important to note, though, that filler–gap dependencies can be unbounded, i.e., there is no grammatically determined upper limit to the linear and structural distance between filler and gap, (8). In (8a) the linear distance between filler and gap is seven words, in (8b) it is ten words, in (8c) twelve, and in (8d) fifteen. The structural distance between filler and gap, as measured by how deeply embedded the gap is compared to the filler is one clause in (8a), two in (8b), three in (8c), and zero in (8d). In the latter, filler and gap are both in the main clause. (8)

Page 5 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

The observation that the distance between filler and gap can be unbounded does not entail by any means that it is unconstrained. Ross (1967) was the first to systematically catalogue constraints on movement. The discovery of the existence of systematic structural constraints on filler–gap dependencies is among the most important empirical discoveries of modern syntactic theory. Ross (1967) called a particular structural configuration an island if the gap cannot grammatically be situated inside of the configuration while the filler is outside. (For an excellent overview and introduction to the topic, see Szabolcsi 2006.) Among the island constraints operative in English, the following four are illustrated here, since they play a particular role in the discussion later on in the chapter: the Complex NP Constraint (going back to Ross 1967: 127), which bans movement of elements contained in a clause dominated by a noun phrase (DP) out of that noun phrase; (p. 393)

the Subject Condition (see Ross 1967: 243 for clausal and Chomsky 1973: 249–50 for DP subjects), which bans extraction from subjects; the Adjunct Condition (Huang 1982), which bans extraction from clausal adjuncts; and the Left-Branch Condition (Ross 1967: 207), which bans movement of the leftmost constituent of a DP containing other material. The constraints are illustrated below. In each case, the ungrammatical extraction is accompanied by relevant semantic and structural controls. The Complex NP Constraint is illustrated in (9a); the remaining examples in (9) serve as controls. The Subject Condition is illustrated in (10). In each case, the grammatical subject is enclosed in square brackets. The grammatical and ungrammatical examples differ in what constituent acts as the subject. The grammatical controls show that extraction from those constituents is possible in principle, the ungrammatical versions show that the sentence becomes degraded when the constituent in question acts as the subject. There are two examples of DP subjects, (10b) and (10d), and one of a clausal subject, (10f). The Adjunct Condition is illustrated in (11). The grammatical and ungrammatical examples are synonymous. In the grammatical cases, the extraction site is in the main clause; in the ungrammatical cases, in the adjunct clause. Finally, (12) illustrates the Left-Branch Condition. In the ungrammatical example, a left branch within a DP has been extracted. The grammatical example is synonymous; the violation is avoided by moving the entire containing DP under a strategy that Ross (1967) called ‘pied-piping’. (9)

Page 6 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

(10)

(11)

(12)

In the immediately preceding discussion, we saw that movement of WH-phrases in English is characterized simultaneously by its unbounded nature and its sensitivity to island configurations. The same simultaneous characterization also describes a substantial number of other constructions in English involving long-distance dependencies and gaps (see Chomsky 1977b). Even more significantly, this characterization describes long-distance dependencies in many other languages as well (Ross 1967). While there can be no doubt about the importance of island effects to linguistic theory, the question of the ultimate cause of island effects remains disputed. There are three different styles of answer to this question. One line of thinking holds that island effects reflect constraints on syntactic derivations or representations (Chomsky 1977, 1986; Huang 1982; Pesetsky 1987; Cinque 1990; Rizzi 1990; Manzini 1992; Starke 2001; Stepanov 2007; Müller 2011). Under such an approach, island violations cannot be (p. 394)

Page 7 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands derived and/or always lead to a deviant output. According to a second tradition, island violations are not syntactic violations proper. Instead, islands are viewed as derivable and syntactically well-formed but deviant either at the interface between syntax and a different system (interpretation for Lasnik and Saito 1992; Chomsky and Lasnik 1993, information structure for Erteschik-Shir and Lappin 1979; Van Valin and LaPolla 1997; Erteschik-Shir 1998, 2007; Goldberg 2013, and phonology for Chomsky 1973; Uriagereka 1999; Hornstein, Lasnik, and Uriagereka 2007; Abe and Hornstein 2012) or as deviant representations in a non-syntactic module (semantics in the account of weak islands in Szabolcsi and Zwarts 1992; Abrusán 2014). A third strand of thinking treats some island effects as the result of processor overload: island structures are not ungrammatical in any sense but lead to a breakdown of the parser due to their complexity (see Pritchett 1991; Kluender 1998; Sag, Hofmeister, and Snider 2007; Hofmeister and Sag 2010; Hofmeister, Staum Casasanto, and Sag 2013; Kluender and Gieselman 2013; among others). We will see in section 16.3 how the interaction of island effects with ellipsis has been used in arguments concerning the nature of island effects.

16.2.2 Successive cyclicity and punctuated paths As mentioned in section 16.2.1, the discovery of unbounded filler–gap dependencies and of island effects proved extremely important to the field of modern syntax. Once their existence is recognized, the question immediately arises how such unbounded dependencies are implemented in the grammar. We saw above that the filler has semantic properties usually associated with the gap site, (1), and that it is morphologically realized in the case form usually associated with the site of the gap, (5). Clearly, syntactic theory must explain how the grammatical properties usually associated with the gap site end up being expressed on the filler and why the filler must obey constraints associated with the gap. Two main metaphors have guided theorizing about the question of how filler–gap dependencies are mediated: The first metaphor is movement, the second metaphor is percolation. The movement metaphor says that, as part of the process that derives sentences with filler–gap dependencies, the filler moves from the position of the gap to the position of the filler, occupying each at distinct points of this derivation. In elaborating this metaphor, theorists have posited that unbounded movement is implemented as an arbitrarily long sequence of short, bounded steps. The idea of an arbitrarily long sequence of individually bounded movement steps has been exploited to explain some of the island effects mentioned above (see, for example, Chomsky 1973, 1977b, 1986), to explain why a filler–gap dependency may influence the shape of elements along the structural path of movement that are locally related neither to the position of the filler nor to the position of the gap (see, for example, McCloskey 1990), and to explain why the filler itself may have properties associated uniquely with certain positions along the path of movement (see, for example, the discussion of intermediate reconstruction and chain binding in Barss 1986; Fox 1999; Lebeaux 2009). With the ancillary assumption that (p. 395)

Page 8 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands movement targets the edge of VP and the edge of CP (Chomsky 1986, 2000, 2001), example (13) would roughly have the representation in (14). It is important to note how the filler leapfrogs through the structure, entering very local relations with some nodes along the path, those boxed gray in the diagram, but not with others, particularly the TP nodes in the structure in (14). (13)

(14)

Under the percolation metaphor, the filler never occupies the site of the gap. Instead, the information that there is a gap, what its category and interpretation are, and what restrictions are placed on it, percolates through the syntactic structure locally, from node to immediately dominating node. This percolation of information is depicted by the gray shading in (15). (p. 396)

Page 9 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands (15)

There are three main differences between the percolation conception and the movement conception of filler–gap dependencies. First, under the movement conception but not under the percolation conception, the filler occupies the position of the gap at some stage of the syntactic derivation. Second, under the movement conception but not under the percolation conception, the filler occupies positions between the position of the gap and that of the filler at some stage of the syntactic derivation. Third, under the percolation conception but not under the movement conception, the information that a particular constituent contains a gap, its nature, case, and interpretation are locally available at every node along the structural path between filler and gap. Abels (2003) coined the terms punctuated versus uniform paths for the movement conception and the percolation conception of movement, respectively. Section 16.4 discusses arguments from the interaction between movement and ellipsis that purport to demonstrate the superiority of a model with punctuated paths, that is, a model based on the movement metaphor.

Page 10 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands (p. 397)

16.3 Sluicing and islands

Ross (1969b) was the first generative paper to discuss sluicing. Sluicing is a type of elliptical WH-question, in which only the WH-phrase is pronounced and the rest of the question is elided. For an introduction to the phenomenon and leading analyses, see Vicente (this volume). (16a–c.i) provide typical examples of sluicing. The understood question in each case is spelled out in (16a–c.ii), but elided save for the WH-phrase in the sluice. (16)

I will use the following terminology: The entire elliptical question will be called the sluice; the pronounced WH-phrase will be called the remnant; the unpronounced remainder of the sluiced question will be called the ellipsis site. The clause providing the content for the sluice will be called the antecedent; finally, the phrase in the antecedent that intuitively corresponds to the remnant, if there is such a phrase, will be called the correlate. (17)

Furthermore, the following three types of sluices will be distinguished terminologically whenever this is necessary. In what I will call merger-type sluices following Chung, Ladusaw, and McCloskey (1995), the remnant has an overt, indefinite correlate whose identity is in question. Examples were provided in (16a.i) and (16b.i). In contrast sluices, the remnant has an overt, usually definite correlate. The remnant asks for alternatives to the correlate. An example is given in (18). Finally in the sprouting type of sluicing, the remnant has no overt correlate, (16c.i). (18)

Page 11 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands With terminology settled, we turn to three important facts discovered by Ross. First, Ross (1969b) argued that sluices are clauses as far as their syntactic category is concerned, a conclusion that is consensual in the literature (for a review of the arguments see Levin 1982; Vicente this volume). Beyond this, Ross (1969b) contains two claims about sluicing that, taken at face value, point to mutually incompatible conclusions about the nature of the ellipsis site. The first claim is given a general statement by Merchant (2001: ch. 3): The remnant must bear the case that its correlate bears. This observation is illustrated in (19) with German examples. Recall from (5) that the verb helfen ‘help’ takes its object in the dative case while unterstützen ‘support’ takes an object in the accusative case and that this case marking is (p. 398) maintained under movement. Now observe the behavior of the remnant WH-phrase when the correlate is either the object of helfen ‘help’ or unterstützen ‘support’. (19)

As can be seen, the case of the remnant is the same as the case of the correlate. I will refer to this phenomenon as the case-matching effect. Ross (1969b) argues that this behavior can be accounted for on the assumption that the sluice has the same syntactic structure as the synonymous full question but that the structure may remain unpronounced just in case it is structurally identical to the antecedent. Under these assumptions, no new mechanisms of case assignment or interpretation are needed. The only mechanism needed to derive sluicing is a deletion/ non-pronunciation operation under syntactic identity.1 We can represent this analysis of sluicing as deletion under syntactic identity schematically as follows, where grayed-out material in angle brackets indicates the analysis of the unpronounced syntactic structure present at the ellipsis site: (20)

Under a theory of ellipsis under syntactic identity, examples like those below motivate the claim that merger-type sluicing does not obey islands. The crucial behavior is illustrated here only for those island constraints that were introduced above in 16.2.1. Other island Page 12 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands effects are discussed in Ross (1969b), Chung, Ladusaw, and McCloskey (1995), and Merchant (2001). Sauerland (1995) concentrates on so-called weak islands. In (21)–(25), the example with sluicing is always given first. It is followed by the sentence which, under a syntactic identity analysis of sluicing, would be the syntactic representation of the sluice, the pre-sluice in the terminology of Dayal and Schwarzschild (2010). Example (25) offers a third example, whose function is to show that independent ellipsis of the noun phrase ‘a mug’ is unavailable. (21)

(p. 399)

(22)

(23)

(24)

(25) Page 13 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

The examples with sluicing are much more acceptable than their putative sources.2 The phenomenon that sluices are much more acceptable than the island-containing sources that an analysis of sluicing as deletion/non-pronunciation under syntactic identity would posit is known as island amelioration or island repair under sluicing. The phenomenon is surprising because, if the ellipsis site is indeed occupied by syntactic structure identical to the antecedent and if island effects are syntactic in nature, then we expect ellipsis to give rise to the same robust type of effect found in examples without ellipsis. Ross assumed that island repair characterizes all types of sluicing and, in fact, other elliptical processes as well. Consequently, his paper ends (Ross 1969b: 277, ex. 75) by reformulating his own earlier (1967) syntactic theory of islands as follows: “If a node is moved out of its island [footnote omitted, K.A.], an ungrammatical sentence will result. If the island-forming node does not appear in surface structure, violations of lesser severity will (in general) ensue.” The leap from merger-type sluices, which are discussed in detail in Ross (1969b), to all types of sluicing and to all ellipsis phenomena is not motivated in Ross’s paper. However, the generalized form of the claim—that is, that there is a phenomenon of (p. 400) island repair by ellipsis—has proved very influential in the literature. I will refer to the claim that island violations are repaired by ellipsis in general as Ross’s conjecture. I will now briefly introduce a cross-classification of theories of sluicing that structures the following discussion. Subsection 16.3.1 explores some consequences of assuming that Ross’s conjecture is essentially correct. Subsection 16.3.2 will question whether Ross’s conjecture is factually correct and explore consequences of the assumption that it is incorrect. Subsection 16.3.3 attempts to assess how serious the challenges to Ross’s conjecture really are. Theories of sluicing can be cross-classified by the answers they give to the following three questions: Is there contextually variable syntactic structure present at the ellipsis site? Is ellipsis conditional on a syntactic identity relation between the antecedent and the sluice? Is ellipsis fed by regular WH-movement (regular movement of the remnant)? Ross’s analysis of sluicing assumes that the answer to all three of these question is yes. The table in (26) shows where various approaches situate themselves.3 26

Page 14 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

4

The next subsection (subsection 16.3.1) explores the extreme positions marked out in this table: ‘No’ to all questions and ‘yes’ to all questions. Subsection 16.3.2 then turns to mixed positions. (p. 401)

16.3.1 Ross’s conjecture and its consequences The first question in the table in (26) distinguishes theories that assume that the ellipsis site is occupied by syntactic structure that varies in some systematic way with the structure of the antecedent (‘yes’ for question 1) from theories that assume that there is either no syntactic structure at all at the ellipsis site or that ellipsis targets some fixed, contextually invariable structure (‘no’ for question 1). Proponents of the latter type of approach must assume that the identity condition on ellipsis is non-syntactic. This is why one set of positions is marked as ‘incoherent’ in the table. I will not comment further on these. Furthermore, the question whether or not there is WH-movement at the ellipsis site is of minor importance in accounts that don’t Page 15 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands assume the presence of variable syntactic structure at the ellipsis site, and I will gloss over the distinction in the discussion to follow. Proponents of such theories usually cite the lack of island effects under sluicing (and Ross’s conjecture in general) as prima facie evidence for the absence of (variable) syntactic structure. The logic is simple: there is no island at the ellipsis site, hence no island effect. While lack of (contextually variable) structure provides an attractive solution to the island amelioration problem, it raises the question of how to account for the case-matching effect. Theories without (variable) structure at the ellipsis site require some form of transmission of case from the correlate to the remnant. Usually, syntactic theories allow for the full expression of morphological case only in a strictly circumscribed set of structural configurations. There must be a case licensor and case bearer and they must stand structurally in a case-licensing relation. This caselicensing relation has all the hallmarks of a well-behaved syntactic relation (Koster 1986; Neeleman and van de Koot 2002): It requires c-command, locality, and uniqueness of the licensor. However, in sluicing, the remnant does not stand in this type of relation to its putative case licensor—the remnant or the remnant’s case licensor in the antecedent. We are left with the question of whether case is a syntactic phenomenon, as all other evidence suggests, or a non-syntactic one, as case matching suggests under approaches to sluicing without syntactic structure at the ellipsis site.5 A further possibility formulable under some but not all theories of syntax would be the position that case is syntactic except (p. 402) in elliptical constructions. What this shows is that a definitive argument that there is no (contextually variable) syntactic structure at the ellipsis site would provide a strong argument against the assumption that case is (always) a syntactic phenomenon.6 It is difficult to overstate the importance of this issue for syntactic theory. Nevertheless, I will not pursue the issue further, since the evidence surveyed later in this chapter strongly suggests that there is contextually variable syntactic structure at the ellipsis site. Culicover and Jackendoff (2005) stress a further consequence of their approach to ellipsis, which admits no syntactic structure at the ellipsis site at all: the syntax-semantics mapping must be flexible and powerful enough to allow the sluice to be interpreted as a WH-question despite its incompleteness. We thus end up with two important conclusions. If it can be shown that there is no (variable) syntactic structure at the ellipsis site, then case cannot be (strictly) syntactic. And if there is no syntactic structure at the ellipsis site at all, then the syntax–semantics mapping must be quite powerful and flexible. While we have just sketched some consequences of assuming that there is no (variable) syntactic structure at the ellipsis site, historically the first analysis of sluicing was Ross’s.

Page 16 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands Ross answered all three questions in table (26) positively. We will now briefly survey the consequences this move has for the nature of the grammar and the nature of movement. The most immediate consequence of his analysis was pointed out by Ross himself. Syntactic processes (like the WH-movement rule or the ellipsis rule involved in sluicing) were seen as operations that map one phrase marker onto another with the additional assumptions that rules could be conditioned only by properties of the input phrase marker. At the time, phrase markers did not contain any indications that movement from one position to another along a particular path had taken place. In other words, the representational devices of copies or traces (14), movement arrows (7), or feature percolation along the path of movement (15), were unavailable. Without such representational devices, Ross’s analysis of sluicing entails that “ungrammaticality is a property not of merely deep or surface structures, or of pairs of trees which are related by rules, but rather of derivations” (Ross 1969b: 277). In the subsequent debate of the early 1970s, the researchers that espoused the view that ellipsis identity should be syntactically defined adopted one of two solutions to the island amelioration problem. Some researchers essentially endorsed Ross’s solution, which was to expand the types of permissible syntactic rules: that is, these researchers advocated a derivational solution to the problem. Thus, Lakoff (1970: 632–3) used island amelioration under sluicing as one among several arguments for the claim that constraints on syntactic rules should be allowed to access entire syntactic derivations rather than just adjacent derivational steps. Such rules were called global rules. By contrast, Chomsky (1972b: 71–3) suggested enriching the representational vocabulary of the theory. In essence, he proposed that when movement crosses an island, the island is marked with the diacritic #. When a (p. 403) phrase containing # is subsequently pronounced, the sentence is perceived as deviant. In the case of sluicing (and ellipsis more generally), the # diacritic marking an island violation remains unpronounced, therefore, the deviance disappears.7 The crucial idea here is that movement out of an island is not ungrammatical per se and neither is the # diacritic. Rather, a phrase carrying the # diacritic is perceived as deviant only if it is pronounced. The question of whether global rules were needed and desirable was a highly contentious and charged issue at the time. However, all modern theories of syntax have sufficient representational power (the copies, traces, movement arrows, feature percolation; see section 16.2.2) to make both global rules and the diacritic for ungrammaticality superfluous. Note that it does not follow that special diacritics for ungrammaticality are no longer used by syntacticians—they are (see for example Merchant 2008b; Bošković 2011). However, the fact of island amelioration under sluicing can be stated without a special diacritic and without powerful new rule types because of the enhanced representations employed by current syntactic theories. While modern theories have sufficient power to state Ross’s claim, that is, that islands are repaired by ellipsis, they do not automatically explain this fact. Indeed, Culicover and Jackendoff (2005) highlight just this point: What is it about the nature of movement and Page 17 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands islands that makes sentences ungrammatical when the island is pronounced, but acceptable when it is not pronounced?8 The question is addressed in the programmatic and speculative Hornstein, Lasnik, and Uriagereka (2007) and in Müller (2011). Both approaches rely heavily on the assumed derivational dynamics of syntactic structure building, albeit in very different ways. Hornstein, Lasnik, and Uriagereka (2007) assume, following Chomsky (1995a), that as far as the relation between the lexicon and the meaning interface is concerned, syntactic structures are purely hierarchical; they are partially ordered by dominance but not by precedence. The fact that words are ordered linearly, Chomsky suggests, is an imposition of the articulatory and perceptual interface (phonology) which can only realize words that are linearly ordered. Uriagereka (1999) then argued that what has to be ordered are the words, that is, the terminal elements of a syntactic structure. Following Kayne (1994), Uriagereka assumes that asymmetric c-command maps onto temporal precedence.9 Under this approach, since A in (27) asymmetrically c-commands all of B, C, and D, A precedes all of them. (27)

(p. 404)

A difficulty arises in a structure where both sisters are internally complex, (28a),

because there are no c-command relations (asymmetric or otherwise) between A, B, and C on the one hand and D, E, and F on the other.10 Assuming a bottom-up structurebuilding mechanism along the lines of Chomsky (1995a), Uriagereka (1999) suggests that the syntactic derivational engine overcomes this problem as follows: The internal structure of one of the phrases is removed and the phrase is turned into a single (huge) terminal before it is combined with its complex sister, (28b). By assumption, the internal order of elements in this new terminal is rigid and its internal parts inaccessible to (context-sensitive) syntactic operations. This move reduces (28a) to (27). A phrase that has thus lost its internal structure and flexibility is said to have been spelled out. Uriagereka compares the structure A-B-C in (28b) to a compound, as compounds, too, are made up of more than one lexical item but behave like syntactic terminals. (28)

Page 18 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

Uriagereka suggests that the proposed analysis of complex–complex junctures under which one of the complex phrases has to be linearized internally can derive island effects that hold at such junctures. In particular, following a line of work exemplified by Cattell (1976), Huang (1982), Kayne (1983), and Chomsky (1986), Uriagereka suggests that there is a unified explanation for the subject condition, the sentential subject constraint, and the adjunct condition. According to Uriagereka, that unifying property is that in each case there is a complex–complex juncture: subjects and adjuncts from which extraction is to take place are internally complex and they are sisters to phrases that are internally complex as well. According to Uriagereka, the syntactic engine must spell out subjects and adjuncts before combining them with their sisters, thus fixing the linear order of their component parts and removing their internal structure from the purview of contextsensitive syntactic operations. This derives the island effects just mentioned. Hornstein, Lasnik, and Uriagereka (2007) build on Urigereka’s theory just outlined. They observe that a phrase that is not pronounced doesn’t need to have its terminals linearly ordered—given that, by assumption, linear order is an imposition only needed to allow pronunciation. They suggest that material that ends up being elided need not be linearized at all. In particular, complex subjects and adjuncts within elided constituents need not be spelled out prior to becoming part of the structure. Consequently, their parts need not be removed from the purview of context-sensitive syntactic operations but can remain (p. 405) accessible to such operations. Thus, for those island effects that arise because of early spell-out of the island, island amelioration under ellipsis is expected. Müller (2011) relates island amelioration to a very different aspect of the dynamics of syntactic computation. He assumes a model of movement roughly along the lines of that depicted in (14). However, instead of skipping certain phrases (TPs in (14)) there is a copy of the moving item at the edge of each and every phrase. Special movementinducing features on the head through whose projection the moving element must pass are responsible for inducing these movements. These movement-inducing features are added to the relevant heads in the process of the derivation. They can be added to a head only while that head is still active, that is, while it still has other features that need to be Page 19 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands satisfied. In addition, simplifying somewhat, the movement-inducing features must be satisfied as soon as they are added to the head. These joint demands sometimes create a catch-22: movement from a given phrase can only be triggered when the head into whose projection the phrase is merged possesses the movement-inducing feature. The movement-inducing feature cannot be added to an inactive head and must be used as soon as it has been added. Therefore, a phrase that is merged and thereby satisfies the last active feature of a head can never be extracted from. Since movement-inducing features have to be used immediately, they have to be added to the relevant head after the phrase from which extraction is to proceed has been merged. But in the cases that interest us, once that phrase has been merged, the head is no longer active, hence the movement-inducing feature can no longer be added. In this way, Müller (2011) explains the subject condition, the sentential subject constraint, and the adjunct condition, briefly reviewed above in section 16.2. According to Müller (2011), ellipsis, like movement, proceeds on a phrase-by-phrase basis and is feature-driven. In order to elide a clause, it is not sufficient that the head of the clause bear the ellipsis feature. Instead, each and every phrase making up that clause must bear an ellipsis feature. Like movement-inducing features, ellipsis features are added to each head during the derivation. They make a head active. Since their effect is to elide the entire phrase, they are systematically the very last feature of a head that is satisfied. It follows from this that heads carrying the ellipsis feature never create the catch-22 for movement that heads without the ellipsis feature do. Therefore, the islands analyzed in terms of the catch-22 discussed above never arise within an ellipsis site. This, in outline, is Müller’s account of island amelioration under ellipsis. This section has discussed some prominent consequences of accepting Ross’s conjecture. On the assumption that there is no (contextually variable) syntactic structure at the ellipsis site, we are led to the conclusion that case is not a syntactic phenomenon and are likely to endorse a very powerful syntax–semantics mapping. On the completely contrary assumption that the ellipsis site is occupied by syntactic material identical to the antecedent, island amelioration becomes an important probe into the nature of island effects.

16.3.2 The island amelioration generalization The discussion in the preceding subsection has shown that Ross’s claim that islands are ameliorated by ellipsis, if true, has important theoretical consequences. It is therefore of paramount interest to ascertain whether the claim is true. The theories discussed in the previous subsection, whether they do or do not posit variable syntactic structure at the ellipsis site, assume that Ross’s conjecture is true and derive it: Everything else being equal, all forms of ellipsis should give rise to island amelioration.11 This subsection briefly introduces a number of elliptical phenomena, including certain types of sluicing, that do not show island repair. These phenomena are prima facie counterexamples to both types of analysis discussed above. I outline briefly (p. 406)

Page 20 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands how some of the approaches to sluicing from the table in (26) not discussed so far propose to deal with the emerging phenomenon of variable island repair (Merchant 2008b). Subsection 16.3.3, finally, assesses how damaging these facts really are for the two classical approaches discussed in the previous subsection. The claim that not all types of ellipsis repair islands has a long history. Thus, Reinhart (1991), Depiante (2000), and Fukaya (2007) contain examples like (29) that suggest that stripping is island-sensitive. (29)

Similarly, Merchant (2004a), Barross, Elliott, and Thoms (2014), and Griffiths and Lipták (2014) claim that contrastive fragments are island sensitive (though Culicover and Jackendoff 2005; Jacobson 2016b deny this, and Weir 2014 remains skeptical). A typical example is the following: (30)

VP ellipsis, too, shows restrictions on movement from the ellipsis site: (31)

While there might be interfering factors in example (31) that would make it ungrammatical quite apart from the island violation (see Schuyler 2002; Merchant 2008b), the relevant factors are controlled in the following example (Merchant 2008b: 143, ex. 39), which remains ungrammatical: (32)

Thus both stripping and VP ellipsis are prima facie counterexamples to Ross’s conjecture. Further evidence against Ross’s conjecture comes from sluicing itself. Above we based the discussion on merger-type sluices with a single WH-phrase as the remnant. While the island-insensitive behavior of such sluices constitutes a very stable cross(p. 407)

Page 21 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands linguistic fact, sluices with contrastive remnants and sluices with multiple WH-phrases as remnants show locality sensitivity. We look at multiple sluicing first. The following is a typical English example of multiple sluicing: (33)

In multiple sluicing, the two WH-phrases must originate in the same clause (see Bolinger 1978; Nishigauchi 1998; Richards 2001; Lasnik 2014 for discussion). Certainly, the original traces of the two WH-phrases must not be separated by an island, (34b), even though there is no (general) ban on multiple questions with WH-phrases from different clauses, as (34a) shows. (34)

English is not alone in showing the restriction that the remnants in multiple sluicing must come from the same clause. Indeed, for a language to obey this restriction is the crosslinguistic norm, whether we are looking at languages that usually front one WH-phrase (see Lasnik 2014 for English; Rodrigues, Nevins, and Vicente 2009 for Brazilian Portuguese and Spanish; E. Callegari, p.c., for Italian; the same also holds for German), at languages that generally do not front their WH-phrases (Takahashi 1994: 285–7; Nishigauchi 1998 for Japanese; Bhattacharya and Simpson 2012 for Bangla with some qualifications; V. Dayaal, p.c., for Hindi), or at languages that generally front all of their WH-phrases (Marušič and Žaucer 2013 for Slovenian; Adliene 2014 for Lithuanian).12 Surprisingly, though the multiple remnants must originate in the same clause, that clause may be contained in an island. This is illustrated with the Slovenian example (35). (35)

We return to this point after introducing the second type of island-sensitive sluicing. Page 22 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands According to Fukaya (2007), Merchant (2008b), Griffiths and Lipták (2014), and Winkler (2013), contrast sluicing is island-sensitive. A typical example is given in (36). The example is ungrammatical on the island-violating construal, where the speaker cannot remember the identity of a third person on whose recommendation the speaker tried a third restaurant. On this reading, the structure would have to be the islandviolating (36a). For some people the sentence is grammatical under a non-island-violating construal, which can be paraphrased as in (36b). The interpretations are clearly distinct, as (36a) is about three or more restaurants with three or more distinct recommenders, while (36b) is about a second recommender for the last restaurant mentioned. (p. 408)

(36)

We find a similar effect in the variant on (21a) given in (37). Some speakers report that this sentence is ungrammatical on any construal, others can get one of the readings paraphrased in (37a). Crucially, the reading paraphrased in (37b) is absent. This is the reading that we would expect if islands could freely be violated by ellipsis. The easiest way to distinguish the two readings is by paying attention to the fact that in (37a) there is only one person being hired who speaks (or needs to speak) two languages, while in (37b) there are two hires each of which speaks (or needs to speak) one language. The latter reading would result from island amelioration and it is absent. (37)

Examples of other types of islands are given in the literature cited. It seems safe to conclude that both multiple and contrast sluicing fail to conform to Ross’s conjecture. Multiple sluicing and contrast sluicing are therefore prima facie challenges to the two approaches discussed in the previous subsection. We will now see that approaches to sluicing that assume that there is contextually variable syntactic structure at the ellipsis site and combine this with a negative answer to one of the two remaining classifying questions from table (26) are potentially advantaged in accounting for these facts. Chung, Ladusaw, and McCloskey (1995) give an account of sluicing that posits syntactic identity between antecedent and sluice but that does not require WH-movement of the remnant—at least in merger-type sluicing. The account is built essentially around the Page 23 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands observation that merger-type sluices involve an indefinite correlate. Indefinites, unlike for example universal quantifiers, are known to be able to take very wide scope; in particular, the overt position of an indefinite and its scope position can cross an island boundary (Fodor and Sag 1982, though see in particular Winter 2001 for more nuanced discussion, and Szabolcsi 2010 for a recent overview). Thus, (38a) allows a reading where the indefinite (p. 409) “a particular paper” is referentially independent and is thus seen to scope over the negative quantifier “nobody.” The indefinite is inside of a relative clause and thus inside of a complex NP but can still scope over the negative quantifier. One prominent approach to this fact is to model indefinite scope after binding relations, which are also not island-sensitive, (38). Such an approach provides a possible implementation of Chung, Ladusaw, and McCloskey’s theory. (38)

Under such an approach wide-scope indefinites are bound by existential operators determining their scope. The indefinite correlate in merger-type sluicing is associated with just such an operator, potentially across an island. The WH-remnant in the sluice takes on the function of this operator in the sense that it binds the indefinite at the foot of the chain and is semantically restricted by it. Except for the operator at the root of the antecedent, the entire syntactic structure of the antecedent can be copied into the ellipsis site.13 The fact that the indefinite correlate exists and acts as a semantic restriction on the remnant is crucial. In contrast sluicing, binding of the correlate is either impossible (if it is definite) or the correlate cannot restrict the remnant semantically (if the correlate is a contrasting indefinite). Therefore instead of binding the correlate, a new branch has to be inserted in the structure. This new branch must be licensed through a chain rather than through binding. Island sensitivity arises from the assumption that the chain relation is island-sensitive while the binding relation is not.14 This accounts straightforwardly for the facts in (36) and (37). While Chung, Ladusaw, and McCloskey’s account is thus very successful in predicting the island insensitivity of merger-type sluicing on the one hand and the island sensitivity of contrast sluicing on the other hand, it is unsuccessful at predicting the behavior of multiple sluicing. If one WH-phrase can bind into an island, why shouldn’t the same be true when several remnants need to enter into a binding relation, as in (34b)? And if multiple binding could be shown to be island-sensitive, then why is multiple sluicing allowed into an island when all of the correlates are situated in the same clause, whether this clause is contained in an island or not?15 The final type of approach to be discussed in some detail is the island evasion approach. The island evasion approach is based on the idea that there is contextually variable structure at the ellipsis site, that the WH-phrase reaches its position through WH-movement, but that the (p. 410) recoverability condition on ellipsis is semantic rather Page 24 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands than syntactic. The idea was pioneered in Baker and Brame (1972) and developed in various ways in the subsequent literature (Merchant 2001; Fukaya 2007, 2012; Rodrigues, Nevins, and Vicente 2009; van Craenenbroeck 2010a; Barros 2014b; Barros, Elliott, and Thoms 2014; Abels 2016b). Under this approach, the ellipsis site does not have to contain a faithful syntactic copy of the antecedent but may contain a suitable paraphrase instead. Thus, example (21a) is analyzed as follows, (39): the ellipsis site cannot contain the structure of the overt question (39a) because it violates the complex NP constraint. Instead it might contain (39b) or (39c); both have an interpretation suitably matching that of the antecedent.16,17 (39)

Barros, Elliott, and Thoms (2014) call the strategy employed here island evasion as opposed to island repair, because the inaudible structure at the ellipsis site evades an island violation through paraphrase instead of repairing it. The strategy works in simple cases because short paraphrases like (39b) and (39c) often have meanings that are sufficiently close to the meaning of the structures isomorphic to the antecedent, (39a). The expectation of the theory is that the appearance of island repair should disappear in cases where there is no suitable paraphrase. Consider in this light example (36a). The strategies used above in (39) to get around the island fail here. (40a) violates the complex NP island and is therefore not a contender. (40b) and (40c) are grammatical, but cannot be construed as questions about the recommender for the third restaurant. (40)

Clearly, the account is built on the assumption that the space of allowable paraphrases is constrained. If there were no constraints, the following pre-sluice would have to be considered: (p. 411)

(41)

Page 25 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands With this as a candidate pre-sluice, (40) should be acceptable. In other words, proponents of the island evasion approach will have to supply a suitably restrictive theory of which paraphrases are and which ones are not contenders as pre-sluices; there cannot be a true theory of island evasion without a solution to this problem, which we might call the toomany-paraphrases problem. The literature offers no worked-out, explicit solution to the too-many-paraphrases problem, but proponents of the island evasion approach appear to assume in practice that paraphrases are either copulative structures or simplifications of the antecedent. On these assumptions, (37) can be analyzed in a way quite similar to (40). A pre-sluice identical to the full antecedent leads to an island violation and neither the copulative paraphrase (‘what other language it is’) nor the short paraphrase (‘what other language they (should) speak’) supports an interpretation corresponding to the intended reading. Island amelioration is thus viewed as an illusion under island evasion approaches. Consequently, proponents of such approaches are at pains to show that when no suitable paraphrase is available, the illusion of island amelioration breaks down. Recall from subsection 16.3.1 that sluicing appears to repair violations of Ross’s LeftBranch Condition, (25) repeated as (42a). Island repair approaches must posit (42b) as the pre-sluice because of parallelism. Island evasion approaches cannot posit (42b) as the pre-sluice, because it violates the Left-Branch Condition. Instead, island evasion approaches assume (42c) as the pre-sluice. (42)

Under evasion approaches, the sluice is thus really built on a predicative adjectival structure rather than the attributive adjectival structure in the antecedent. Prima facie evidence for this position comes from languages like German and Dutch, where there is a morphological distinction between attributive and predicative adjectives. Example (43a) illustrates that German attributive adjectives are always suffixed with a schwa and a further (though sometimes null) affix indicating number, gender, and case. Example (43b) shows that by contrast predicative adjectives are bare: they have neither the schwa suffix nor the agreement suffix. (43)

Page 26 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

In the German translation of example (42) the bare predicative form of the adjective is used, (44a), (see Merchant 2001, who also discusses parallel facts from Dutch), which strongly suggests that the pre-sluice can be the equivalent of (42c), as expected under the evasion approach. The attributive form of the adjective cannot be used, (44b), which strongly suggests that the pre-sluice cannot be the pre-sluice expected under island repair approaches. That is, it cannot be the equivalent of (42b). We are also not dealing with N-ellipsis, since, although N-ellipsis is independently available, it requires the presence of the article and the attributive form of the adjective, (44c).18 The island evasion approach offers an elegant explanation both of the acceptability of apparent left-branch extraction under sluicing and of the morphological observation. (p. 412)

(44)

We have seen above that the clausemateness condition on multiple sluicing poses a challenge for syntactic identity accounts and for accounts with no structure at the ellipsis site. Though there is currently no worked-out proposal about the syntax of multiple sluicing that could claim cross-linguistic validity (though Takahashi 1994;19 Rodrigues, Nevins, and Vicente 2009; Marušič and Žaucer 2013; Lasnik 2014 provide languageparticular solutions), the outline of such an account emerges quite clearly from these papers, and it involves island evasion. The idea in the papers cited is that the additional WH-phrase moves from the ellipsis site. The task then is to identify and characterize precisely the movement operating that lifts the additional WH-phrase from the ellipsis site and to attribute the clausemate condition to independently known constraints on this movement operation (see Abels and Dayal 2016). While this body of work makes it clear in principle what logical structure an account of the clausemateness condition of multiple sluicing will likely have, the logic of island evasion is invoked to deal with the island insensitivity of multiple sluicing. Under the island evasion approach, the pre-sluice for an Page 27 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands example like (35), repeated here as (45), can be assumed to be the equivalent of ‘who it kicked where’. This structure evades the complex NP island in the antecedent rather than repairing it. (p. 413) In the general case, violations of the clausemateness condition cannot be evaded by paraphrase. The island evasion approach is thus well positioned to develop an explanation for the otherwise paradoxical locality insensitivity of multiple sluicing combined with the clausemateness condition. (45)

It is difficult to see how either of the theories discussed in the previous subsection, 16.3.1, can be reconciled with the fact that multiple sluicing is island-insensitive while at the same time being subject to the clausemateness condition. If there is no syntactic structure at the ellipsis site, all sensible multiple WH-questions should in principle correspond to multiple sluices. This will explain island insensitivity but stumbles on the clausemateness condition, because there are many grammatical and sensible multiple questions where the WH-phrases are not clausemates. Crucially, these cannot generally be turned into multiple sluices, as (34) illustrated. Essentially the same problem arises for syntactic identity approaches. These approaches are built around the assumption that ellipsis repairs island violations. This assumption can be invoked to explain the island insensitivity of multiple sluicing. However, the assumption that syntactic locality constraints are repaired by ellipsis makes it difficult to see why the additional WH-phrase would have to originate in the same clause as the primary WH-phrase. Whatever syntactic locality constraint might underpin this fact under a syntactic identity account, the violation should be repaired by ellipsis (see Marušič and Žaucer 2013 on this point). While there is then substantial suggestive evidence for an island evasion approach, the case cannot be considered closed. There are two main reasons for this. First of all, there is no worked-out, explicit, and predictive solution to the too-many-paraphrases problem. Second, the island evasion approach does not by itself offer an approach to case connectivity. This is pointed out in Lasnik’s (2005) discussion of Merchant (2001). Lasnik notes that case connectivity between correlate and remnant is not predicted by the evasion approach, again, because of the too-many-paraphrases issue. The fact that syntactic identity accounts do explain case connectivity gives them an explanatory advantage over island evasion approaches.20 Researchers have since tried to solve the case connectivity and related problems in a number of different ways (see van Craenenbroeck 2010a; Chung 2013; Abels 2016b), but a consensus view has not yet emerged.

Page 28 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands In this subsection we have briefly looked at elliptical phenomena that fail to conform to Ross’s conjecture. The failure of island amelioration provides prima facie evidence against (p. 414) theories that posit no contextually variable syntactic structure at the ellipsis site and also against theories that impose strict syntactic parallelism between antecedent and ellipsis site and assume a grammatical mechanism of island repair. The facts favor theories that assume contextually variable syntactic structure at the ellipsis site but either do not impose strict syntactic parallelism between antecedent and ellipsis site (island evasion) or allow the island ameliorating examples to be derived without movement.21 While many aspects of these proposals remain to be worked out, the examples where Ross’s conjecture breaks down suggest that the ellipsis site contains (contextually variable) syntactic structure and that island violations cannot be repaired by ellipsis. It would follow from this that island amelioration under ellipsis is an illusion and should be explained neither by the absence of syntactic structure at the ellipsis site nor as a consequence of the nature of movement (as in Hornstein, Lasnik, and Uriagereka 2007 and Müller 2011). On the contrary, the cases where Ross’s conjecture fails have suggested to proponents of the island evasion approach that locality constraints are robustly attested within ellipsis sites. The facts discussed in this and the previous subsection also interact with the debate on whether island constraints are syntactic or arise from parsing considerations. In conjunction with the assumption of strict syntactic parallelism between antecedent and ellipsis site, Ross’s generalization might be construed as support for the idea that island constraints are not a grammatical phenomenon at all but instead arise from constraints on online processing. One could then assume that ellipsis resolution involves copying the potentially island-containing parse tree for the antecedent into the ellipsis site. Copying would sidestep the need to build a new parse tree online, thus avoiding the processing difficulty. It is hard to see, however, how such an account could be extended to the cases where ellipsis fails to ameliorate islands, especially since the relevant structures are often semantically perfectly well-formed, as (34) shows.

16.3.3 Ceteris paribus In the previous section, we have taken Ross’s conjecture to say that everything else being equal, all forms of ellipsis should give rise to island amelioration. We have further seen that stripping, contrastive fragments, VP ellipsis, multiple sluicing, and contrast sluicing are prima facie counterexamples to the generalization. Theoreticians defending a syntactic parallelism account have routinely invoked the idea that these counterexamples are only superficial and that not everything is equal in these cases (Fox and Lasnik 2003; Merchant 2004a; Temmerman 2013; Griffiths and Lipták 2014; Abe 2015). These authors suggest that the failure of Ross’s conjecture should not be explained as an island effect at the ellipsis site but in one of the three following ways instead: (i) the ungrammaticality arises (or is visible) along the movement path outside of the ellipsis site, (ii) ungrammaticality (p. 415) arises in the antecedent, or (iii) it is an effect of syntactic mismatch between antecedent and ellipsis site. Page 29 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands To understand what is intended by case (i), consider the contrast in island amelioration between sluicing, (46a), and VP-ellipsis, (46b), again. The crossed-out material represents material at the ellipsis site under a theory demanding syntactic identity. The overt material distinguishing sluicing from VP-ellipsis is italicized in (46b). Notice that the WH-phrase ‘what kind of language’ crosses the italicized material and the structure associated with it. In particular, ‘what kind of language’ crosses nodes 1 and 2, which form part of the WH-movement path. (46)

If the island violation arises or is made visible on node 1 or node 2 (or both), then the fact that these nodes are outside of the ellipsis site in (46b) but the corresponding nodes 1’ and 2’ in (46a) are inside of the ellipsis site can be used to explain why (46b) is ungrammatical while (46a) is grammatical. Fox and Lasnik (2003) use this idea in their explanation of the difference between VP ellipsis and sluicing. Merchant (2004a) and Temmerman (2013) use the same logic, coupled with a version of Chomsky’s #-mark, to make the island violation visible outside of the ellipsis site in fragments. Recall that Chomsky’s suggestion of using a diacritic feature representing the fact that islandviolating movement has occurred as an explicit part of the syntactic representation stems from a time when syntactic representations did not carry their derivational history on their sleeves (in the forms of copies of movement, traces of movement, movement arrows, a path of feature percolation,…). However, modern theories of syntax all make these representational devices available. As a consequence, it is always possible to determine by simple inspection of the output structure whether an island violation has taken place. The use of the diacritic features in modern theories therefore introduces a representational redundancy, which should be avoided if at all possible and which needs to be carefully motivated if it is to be assumed (see n. 6). Given these considerations, it is unclear whether the #-diacritic can carry any truly explanatory weight. To understand case (iii) it is useful to consider the following contrast. Example (47a) is well-formed while (47b) is deviant. The difference clearly has to do with the different antecedents, in particular the difference between ‘a certain’ and ‘any’. Indeed, ‘a certain colleague’ in (47a) takes scope above negation (‘There is a certain colleague that Bob didn’t consult’) but ‘any colleague’ in (47b) takes scope below negation (‘It is not the case that there is a colleague that Bob consulted’). (47)

Page 30 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands The difference can be represented syntactically with an existential operator taking scope below negation for (47b) and taking clausal scope for (47a): (p. 416)

(48)

If syntactic identity is sensitive to operators in the antecedent, then the contrast above in (47) may find a syntactic explanation. The WH-operator in the sluice takes clausal scope and c-commands negation. It is outside of the ellipsis site and there is no (other) operator binding the object of ‘consult’ within the ellipsis site. Under a syntactic identity account, we can take node α in (47a) as the antecedent for (47a). However, there is no clausal node excluding the existential operator in (47b). The antecedent provided by the structure with Page 31 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands the low existential operator cannot be used to resolve ellipsis under a syntactic identity account, since in this structure the object of ‘consult’ in the sluice would be bound twice: once by the existential below negation and a second time, vacuously, by the WH-operator. It is reasonable to assume that this structure can either not be derived or not be interpreted or both. The structure for the ellipsis site that can be derived and interpreted has no existential operator binding the object of ‘consult’ below negation. It is therefore not syntactically identical to the antecedent provided in (47b). Therefore, the sluice based on (47b) cannot meet the syntactic identity requirement and is correctly ruled out. This illustrates case (iii) above: the failure of syntactic parallelism.22 Notice that the diagrams in (48) associated the existential operator with the indefinite noun phrase through coindexing rather than through movement. This was intended to reflect the two claims we already encountered in our discussion of Chung, Ladusaw, and McCloskey (1995) above (see discussion around (38)), namely, that indefinites can take island-insensitive scope and that binding relations are not islandsensitive. Other operators such as WH-phrases, universal quantifiers, and comparative operators are often taken to be scoped by island-sensitive movement operations—though these movements can be covert and not have a reflex in word order. Case (ii) above, (p. 417)

where the ungrammaticality of an island-violating sluice is attributed to a violation in the antecedent, builds on the idea that certain types of scope are represented as movement. Specifically, consider the implication for the sluicing remnant if its correlate is an operator that is scoped by an island-sensitive movement operation. If the sluice itself obeys all island restrictions, then there is a structure for the antecedent with the corresponding syntactic scope for the correlate; both sluice and antecedent are wellformed and they match: no problem arises. However, if the remnant relates to a trace inside of an island in the ellipsis site, then the situation is quite different. By assumption, the antecedent is only grammatical if the correlate does not violate the island, but the syntactic identity constraint mandates just that. Thus, if the remnant relates to a trace inside of an island and the correlate is scoped in an island-sensitive way, then the two demands of island sensitivity and syntactic identity cannot be met simultaneously. Syntactic identity can only be achieved if the antecedent is ungrammatical. This is case (ii) above.23 Griffiths and Lipták (2014) and Abe (2015) account for the difference in island sensitivity between merger type and contrast sluicing exploiting this logic. Griffiths and Lipták (2014) argue that the difference between the island-insensitive merger type of sluicing and the island-sensitive contrast sluicing is part of the following larger pattern: Elliptical constructions in which the remnant contrasts with the correlate are generally island-sensitive and constructions where the remnant specifies or queries the identity of the correlate are generally island-insensitive. They argue convincingly that treating this generalization as a violation of type (i) is neither motivated nor insightful. To account for their generalization, Griffiths and Lipták assume that the indefinite correlates in merger-type sluicing and other non-contrastive structures are syntactically scoped by an island-insensitive mechanism (an idea we have already encountered in the discussion of Chung, Ladusaw, and McCloskey 1995 in 16.3.2), while the correlates in contrast sluicing are scoped by the island-sensitive focus movement operation. Under this view, Page 32 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands merger-type sluices are island-insensitive because the scoping mechanism in the antecedent is island-insensitive and because ellipsis repairs island violations. Contrast sluicing is island-sensitive because the correlate is scoped by an island-sensitive mechanism which means that either parallelism is violated (violation of type iii) or there is an unrepaired island violation in the antecedent (violation of type ii). Griffiths and Lipták’s attempt to uphold Ross’s conjecture is theoretically important. The analysis extends the scope of a syntactic identity account of sluicing to a number of cases where Ross’s conjecture fails and that we treated as prima facie counterarguments to a syntactic identity account. Despite its ingenuity, Griffiths and Lipták’s account is open to a number of criticisms which, on balance, appear to disfavor the syntactic identity approach. First, the account needs to make a distinction between (island-insensitive) merger-type sluicing and (island-sensitive) contrast sluicing. This distinction is related to the nature of the correlates: indefinites in merger-type sluicing and focused DPs in contrastive sluices. Indefinites are analyzed as having syntactically represented island-insensitive scope while foci are analyzed as being scoped syntactically with an island-sensitive mechanism. This analytical distinction between indefinites and foci seems to be driven by theoretical (p. 418)

expediency rather than by empirical necessity. Empirically, indefinites and foci behave in rather similar ways. As is well known, indefinites inside of islands can be referentially independent of c-commanding operators outside of islands (Fodor and Sag 1982). This behavior can—but by no means needs to (see Schwarzschild 2002)—be modeled in terms of island-insensitive syntactic scope of an existential operator. Equally important, though, is another aspect of the scopal behavior of indefinites: their distributive scope. The distributive scope of indefinites is island-sensitive (see Ruys 1992; Winter 2001) and tracks the locality of quantifier raising in many ways. The distributive scope of indefinites is a strong candidate for modeling in terms of a syntactically scoped operator. Foci show similarly mixed behavior. When we measure their scope in terms of the ability of a focus inside of an island to associate with a focus-sensitive operator, then foci come out as island-insensitive (Anderson 1972; Jackendoff 1972). On other measures, such as overt focus movement (Griffiths and Lipták 2014) and the ability of several foci from the same domain to associate with separate operators (Krifka 2006), foci come out as islandsensitive. Thus both foci and indefinites show a complex mix of island-sensitive and island-insensitive behavior. These empirical considerations show that Griffiths and Lipták’s choice to represent syntactically the island-insensitive aspects of the behavior of indefinites and the island-sensitive aspects of the behavior of foci is arbitrary at best. The similarities between indefinites and foci suggest a theoretical assimilation of the two phenomena rather than the dissimilation necessary for Griffiths and Lipták’s account. Indeed, Winter (2001) and Krifka (2006) explain the island-sensitive aspects of the behavior of foci and indefinites, respectively, by reducing these aspects to standard movement operations; and Schwarzschild (2002) and Krifka (2006) explain the islandinsensitive aspects of the behavior of foci and indefinites, respectively, by not representing them in the syntax at all. In light of this, Griffiths and Lipták’s analysis is not Page 33 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands only arbitrary, but would represent a step backwards in our understanding of the similar behavior of indefinites and foci. A second issue for Griffiths and Lipták’s (2014) analysis is raised by Abels and Thoms (2014) and Barros, Elliott, and Thoms (2014), who report that contrast sluicing loses its island sensitivity in certain cases in some languages. Concretely, Abels and Thoms and Barros, Elliott, and Thoms show that contrast sluicing loses its island sensitivity in languages that allow contrastive WH-phrases to bind resumptive pronouns inside of islands. They claim further that the acceptability of island amelioration in contrast sluicing correlates neatly with the island repair potential of a resumptive pronoun in a given language. All of this would follow, if the status of island-insensitive contrast sluicing does not depend on the properties of focus movement in the antecedent (as in Griffiths and Lipták’s approach) but depends instead on the availability of a grammatical presluice (see works cited for details). The pattern strongly suggests (a) that there is structure at the (p. 419) ellipsis site, (b) that the status of island-insensitive contrast sluicing is not tied to properties of the antecedent, (c) that extraction at the ellipsis site is island-sensitive, and (d) that the grammaticality of island-insensitive contrast sluicing depends on language-specific properties of the available pre-sluices, as expected under the island evasion approach. This argument turns Griffiths and Lipták’s defense of Ross’s conjecture into an argument against it. It weakens the syntactic identity approach but at the same time fails, in the absence of a solution to the too-many-paraphrases problem, to lend crucial support for the island evasion approach. Subsection 16.3.2 briefly introduced two kinds of sluicing that violate Ross’s conjecture: contrast sluicing and multiple sluicing. The existence of and constraints on these structures were taken as prima facie evidence against syntactic identity accounts with WH-movement and against accounts without (contextually variable) syntactic structure at the ellipsis site. We have taken these examples to provide prima facie evidence for island evasion approaches or for approaches with syntactic structure but without WH-movement. Given that to date there is no developed theory of multiple sluicing under any of the existing approaches to ellipsis, we could not pursue the issue of multiple sluicing further in this subsection (though see n. 14). This current subsection has briefly looked at the possibility of neutralizing contrast sluicing as a counterexample to Ross’s conjecture by invoking the implicit ceteris paribus clause. We first identified three general ceteris paribus strategies that have been used in the literature: (i) violations arise in the structure above the ellipsis site; (ii) violations arise in the antecedent; (iii) violations arise as a result of mismatch between antecedent and structure at the ellipsis site. We then saw that the most convincing rebuttal of the argument from contrast sluicing, namely Griffiths and Lipták (2014), relies on a combination of (ii) and (iii): the antecedent is either ill-formed (ii), being derived via island-violating focus movement, or it mismatches the antecedent (iii). We also saw that Griffiths and Lipták (2014) suffers both from conceptual and empirical weaknesses. This leaves largely unchanged the strength of the claim that syntactic identity accounts cannot give a proper analysis of contrast sluicing. The argument from contrast sluicing is a prima facie argument against Page 34 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands syntactic identity accounts. It cannot be turned into a compelling argument for island evasion approaches until the too-many-paraphrases problem has been solved. We can also evaluate to what extent the discussion from the present subsection impacts our evaluation of the prospect of a theory without (contextually variable) syntactic structure at the ellipsis site. Notice that Griffiths and Lipták’s account of the island sensitivity of contrast sluicing relies crucially on the presence of syntactic structure at the ellipsis site: Either antecedent and ellipsis site contain syntactically different material, in which case the syntactic identity condition is violated. Or antecedent and ellipsis site contain syntactically identical structure, and then the antecedent violates locality conditions. This logic cannot be reproduced in an account without (variable) syntactic structure at the ellipsis site. Nor can a directly semantic account rely on the notion that the elided material is semantically deviant as an explanation, since the intended interpretations of contrast sluicing are perfectly coherent. And nor can theories without syntactic structure at the ellipsis site avail themselves of Abels and Thoms (2014) and Barros, Elliott, and Thoms’ (2014) rebuttal of Griffiths and Lipták (2014) since this rebuttal, too, relies crucially on the nature of the syntactic representation at the ellipsis site. The counterexamples to Ross’s conjecture from subsection 16.3.2 remain as problematic as they were for accounts without (variable) syntactic structure at the ellipsis site. (p. 420)

16.3.4 Conclusion: Sluicing and islands In this section we have surveyed facts concerning and theoretical reactions to Ross’s conjecture, that is, the claim that ellipsis quite generally ameliorates island violations. In its simple, general form the generalization has elicited two types of response (subsection 16.3.1): (i) there is no (contextually variable) syntactic structure at the ellipsis site and (ii) there is a copy of the antecedent at the ellipsis site but island effects are masked or never incurred at the ellipsis site. Position (i) has important consequences for the theory of syntactic relations like case and for the syntax–semantic interface. Position (ii) has been taken to be revealing of the nature and derivational provenance of island constraints. The fact that Ross’s generalization has systematic counterexamples (subsection 16.3.2) poses problems for both of the above positions and lends prima facie support to theories that assume syntactic structure at the ellipsis site but do not assume that island violations are literally repaired by ellipsis. The latter type of theories undermine the conclusions drawn from the more traditional first two positions. Subsection 16.3.3 explored the severity of the problems caused by the counterexamples to Ross’s conjecture for the traditional two positions on ellipsis. While non-syntactic accounts of ellipsis have nothing to say about these cases, syntactic identity accounts have been defended by invoking the ceteris paribus clause. We have seen that these moves run into some difficulties, both conceptual and empirical. However, the issue remains open since proponents of the island evasion approach, which has the strongest prima facie support, have not conclusively answered the two main questions arising under the approach: how to account for case matching and how to solve the too-many-paraphrases problem. Page 35 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands As a consequence, none of the theoretical arguments from ellipsis for the nature of case, the flexibility of the syntax–semantics mapping, or pertaining to the nature of island effects holds water at the moment.

16.4 Punctuated paths Section 16.3 discussed a class of cases where movement is disallowed in non-elliptical structures and allowed in intuitively corresponding elliptical ones. This section turns to cases where movement is allowed in non-elliptical structures and selectively disallowed in elliptical structures. Such effects have been discussed particularly in relation to VP ellipsis and related constructions (Aelbrecht 2010; Abels 2012; Baltin 2012; Bošković 2014). A typical example of this type of interaction between movement type and ellipsis comes from Aelbrecht’s (2010) discussion of modal complement ellipsis in Dutch. Modal complement ellipsis is a type of ellipsis of the main verb, its dependents, and modifiers in the specific context where these appear as the complement of a modal verb. Modal complement ellipsis is illustrated in (49) (from Aelbrecht 2010: 116–17): (49)

Example (49a) features the active VP Sarah een cadeautje geven ‘give Sarah a present’ and (49b) the passive VP gewassen worden ‘be washed’. Sentences in the passive (p. 421)

voice, under many analyses, involve movement of the thematic object of the verb from object to subject position. Under those analyses, die broek ‘those pants’ in (49b) originates within the elided VP and moves out of it. Similarly, WH-question formation is usually taken to involve movement. But such movement cannot originate within the ellipsis site of modal complement ellipsis, (50a), while such movement is perfectly licit in non-elliptical contexts, (50b).24 (50)

Page 36 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands

The accounts for such asymmetries in Aelbrecht (2010), Abels (2012), Baltin (2012), and Bošković (2014) differ in their details but they share a number of common, crucial assumptions about the nature of syntax, ellipsis, and—importantly—movement. The first assumption common to the above accounts is that syntax is a derivational, bottom-up system, i.e., phrase markers are built along essentially minimalist lines (Chomsky 1995a; Epstein 1995) by merging lexical items bottom-up. Second, ellipsis is analyzed as involving regular syntactic structure at the ellipsis site. This structure is phonologically deleted and marked as syntactically inaccessible as a consequence of ellipsis. Third, movement from the ellipsis site requires access to the syntactic representation at the ellipsis site. It follows that when movement precedes ellipsis of the domain from which an item is to be moved, movement is possible, but when movement follows ellipsis of the domain from which an item is to be moved, movement is impossible, because the domain has been elided and is therefore syntactically inaccessible by the time movement happens. In other words, (p. 422) whether an item can or cannot be extracted from an ellipsis site depends on whether extraction happens before or after ellipsis. This is schematized in (51): (51)

The theory in Bošković (2014) for example would deal with the facts in (49) and (50a) as follows: movement to the subject position shown in (49) targets the specifier of TP while WH-movement illustrated in (50a) targets the structurally higher specifier of CP. The syntactic head triggering ellipsis is situated between those two positions (and identified Page 37 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands with C0 in Bošković 2014). Because of the bottom-up nature of syntax, movement of the subject to the specifier position of TP happens before ellipsis is triggered and is therefore licit, (51a). On the other hand, movement to the specifier of CP happens only after ellipsis is triggered (again because of the bottom-up nature of the syntactic computation) and is therefore illicit, (51b).25 The models in Aelbrecht (2010), Abels (2012), Baltin (2012), and Bošković (2014) all crucially rely on an implementation of movement in the leapfrogging way illustrated and discussed in connection with structure (14). Structure (14) claimed that WH-movement stops over at the edge of VP and at the edge of CP but not at the edge of TP. This is crucial. It is only because movement to the specifier of TP escapes the ellipsis domain before ellipsis can be triggered by the complementizer and because WH-movement to the specifier of CP happens after ellipsis is triggered that the distinction between the two movement types can be derived. If movement stopped over at the edge of every phrase along the path (as in Müller 2011, discussed in 16.3.1) both types of movement would target a position at the edge of TP and thus happen before ellipsis is triggered. The same reasoning applies to theories where movement is represented in terms of percolation of features, (15). Under percolation approaches, there is no obvious way of deriving the fact that movement to the specifier of TP is allowed from the ellipsis site, but movement to the specifier of CP is not. Extraction asymmetries from ellipsis sites can potentially provide an interesting argument about the punctuated or continuous nature of movement paths. Seductive though this argument may be, the case is far from closed. Thus, the simplified account of the Dutch modal complement ellipsis facts presented above predicts that any movement to a position outside of VP but below TP should be compatible with modal complement ellipsis while movements to a position above TP should not. As discussed at length in Aelbrecht (2010), the first of these predictions is wrong: scrambling of objects and adverbials out of the VP in Dutch is not licit under modal complement ellipsis. Worse still, even the second prediction is inaccurate, since modal complement ellipsis is possible in relatives with an abstract relative (p. 423) operator in contexts of so-called antecedentcontained deletion, (52a), it is possible in (certain) free relatives (see Abels 2012 for discussion), (52b), and in comparative clauses, (52c). (52)

Page 38 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands The difference between those movement types that do and those that do not allow extraction from the domain of modal complement ellipsis does not follow from any of the analyses mentioned above (see Aelbrecht 2010; Abels 2012 for discussion). Unfortunately, the other empirical cases are equally equivocal. Thus, Baltin (2012) claims that British English do is an ellipsis phenomenon whose behavior follows neatly from the logic outlined in (51). He claims that A-movement is and A¯-movement is not allowed from the ellipsis site. However, Abels (2012: 31–2) argues, largely on the basis of published examples, that the generalization is not true: A-movement does and WH-movement does not allow extraction from the ellipsis site in British English do. Some of the remaining A¯-movements pattern with A-movement, some with WH-movement. This leaves the overall picture muddled and the theory in need of re-evaluation. Other potential cases of ellipsis/movement-type interactions that could follow from the theory of punctuated paths involve verb phrase pronominalization in Scandinavian (Houser, Mikkelsen, and Toosarvandani 2007; Bentzen, Merchant, and Svenonius 2013), modal complement ellipsis in German, French, Italian, and Spanish (see Napoli 1985; Aelbrecht 2010; Dagnac 2010) and the high type of VP ellipsis in English discussed in Bošković (2014). In these cases, the database is not rich enough to reach firm conclusions about the nature of the effects. Although the theoretical question of the punctuated versus the continuous nature of movement paths is of great importance to syntactic theory, the empirical picture remains too fragmented for any firm conclusions at the moment. Moreover, the analyses presented in the literature, although exploiting the same logic, are parochial and contradict each other. It is to be hoped that ellipsis/movement-type interactions will be studied in more detail for more languages, at which point they might bear on such important questions as the derivational nature of syntax and the punctuated versus continuous nature of movement paths.

16.5 Conclusion This chapter has surveyed two types of interaction between movement and ellipsis. The first case study was devoted to cases where movement from an elliptical domain is less restricted than might be expected on the basis of comparable examples with fully (p. 424) pronounced structures. Such facts go by the name of island amelioration and have variously been taken as evidence for the phonological nature of island constraints, the derivational nature of syntax, the non-syntactic nature of case, and the flexibility of the syntax/semantics interface. It was shown that the most immediate empirical consequence of the two canonical theories of ellipsis, that is of theories that posit no (variable) syntactic structure at the ellipsis site and theories that demand syntactic identity between antecedent and ellipsis site, radically different though they are, is Ross’s conjecture. Ross’s conjecture says that, ceteris paribus, all cases of ellipsis give rise to island amelioration. On the surface, the conjecture is false. As a result, the various conclusions based on canonical theories of ellipsis are weakened. The most promising competing approach, the approach based on island evasion, cannot be considered a Page 39 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands successful theory until a principled account of case connectivity is presented and the toomany-paraphrases problem is solved. The second case study involved structures where movement from an ellipsis site is less liberal than movement from the corresponding pronounced structure. Some researchers have taken such structures to argue for the derivational nature of syntax and the hypothesis that movement paths are punctuated. At the moment, the facts do not line up neatly with the predictions of the various theories that try to make this argument and the theories, though building on the same logic, are mutually incompatible. For the time being, conclusions about the nature of syntactic computation and the representation of movement based on these facts must therefore be considered as highly tentative and preliminary.

Notes: (1) We review some of the reasons (see Merchant 2001: ch. 1; Vicente this volume, and for related arguments from other elliptical constructions Culicover and Jackendoff 2005: ch. 1.2) to doubt that this analysis is correct. Some such examples were already noted in Ross (1969b) but interpreted differently there. (2) The more modern literature, such as Levin (1982) and Chung, Ladusaw, and McCloskey (1995), usually treats sluicing examples like those in (21)–(25) as fully acceptable, while Ross (1969b) only claimed that the sluices were much improved compared to their putative source while still falling short of full acceptability. (3) The table is essentially adapted from the two-dimensional matrix in Merchant (this volume). Question 3 is added as a third dimension. The option of having a partially syntactic and partially semantic identity condition is suppressed here for simplicity. (4) For reasons given in Merchant (2001: ch. 4), this idea has never been seriously pursued for English, though an analysis in this spirit is pursued in Adams and Tomioka (2012) for Mandarin Chinese. (5) Pronominal forms in English fragments do not appear in the form that would be expected in the non-elliptical versions of the answers, as shown in (i). ((i))

This may be an instance where the case-matching generalization breaks down. However, analyses exist that treat pronominal forms as contextual allomorphs (see Parrott 2006, 2009) rather than different case forms. If such an analysis is correct, English fragments cease to be a counterexample to case matching. Evidence for this position comes from the somewhat erratic behavior of pronouns in coordinated structures (My brother and {me | Page 40 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands I} will be in town. Joe invited Fred and {me | I}.), a behavior that is not shared by pronouns in languages with bona fide case. (6) The debate about the distribution of preposition stranding under sluicing and its explanation (Merchant 2001; Sjepanović 2008; Rodrigues, Nevins, and Vicente 2009; van Craenenbroeck 2010a; Sato 2011b; Nykiel 2013; Abels 2016b) can also be viewed in this light. On the joint assumption that the ban on preposition stranding is an indicator of movement and that a remnant PP can only have moved if there is (contextually variable) syntactic structure at the ellipsis site, evidence that preposition stranding is disallowed under sluicing (or a well-defined subset of cases) will then count as evidence for the presence of syntactic structure at the ellipsis site. (7) The # diacritic had been introduced by Chomsky (1965: 138) as a boundary symbol together with a principle whereby # could not appear internally to a pronounced sentence; being a boundary symbol, it could only appear at the edge of a sentence. Chomsky (1972b) uses the diacritic to make the application of certain rules manipulating it obligatory: In the case at hand, the otherwise optional ellipsis rule becomes obligatory because otherwise # remains internally to a pronounced sentence. (8) Recall from above that Culicover and Jackendoff’s own analysis posits no syntactic structure at the ellipsis site and therefore explains island amelioration rather than just describing it. (9) A node in a syntactic structure asymmetrically c-commands all and only the nodes that are (irreflexively) dominated by its sister. (10) Another type of problem arises at the bottom of each complex branch, where two terminals will unavoidably meet. The case is irrelevant for what follows and will not be discussed here. (11) Hornstein, Lasnik, and Uriagereka (2007) and Müller (2011) derive, more precisely, that, everything else being equal, all islands that are ameliorated by one form of ellipsis should be ameliorated by all forms of ellipsis. (12) Abe (2015: ch. 6) discusses Japanese data and Lasnik (2014: 7) discusses Serbo Croatian data where multiple sluicing violates the same clause condition. As far as is known, the WH-phrases must still originate in positions fairly local to each other in both cases. (13) To account for case connectivity, Chung, Ladusaw, and McCloskey (1995) assume that case is transmitted via the binding chain to the remnant, despite the fact that binding relations usually do not transmit case. (14) Chung, Ladusaw, and McCloskey (1995) use the same type of account for the fact that sprouting is island-sensitive, but here an alternative explanation in terms of semantic/ pragmatic incoherence of the antecedent–sluice pair is available. See Romero (2000).

Page 41 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands (15) Using Chung, Ladusaw, and McCloskey’s model, Nishigauchi (1998) attempts to reduce the locality of Japanese multiple sluicing to the locality of so-called pair list interpretations of multiple questions. Such a reduction remains unsuccessful from a cross-linguistic perspective, because pair list readings have a wider distribution than multiple sluicing. This is shown by (34a), which has a pair list interpretation, yet (34b) is unacceptable. (16) See Romero (2000), Merchant (2001), AnderBois (2011), and Barros (2014b) for discussion of how to define what is a suitably matching interpretation. (17) I should note some limitations of the discussion of island evasion approaches here. For space reasons I concentrate exclusively on sluicing. It is not logically necessary that all types of ellipsis allow the same freedom for paraphrases at the ellipsis site. Detailed discussion of each case would lead us too far afield. The same is true, potentially, for variation across languages: not all languages necessarily allow the same range of paraphrases at the ellipsis site. Temmerman (2013) might be informative in this regard. It should also be noted that there are mixed approaches that combine a semantic identity condition with limited syntactic identity (Chung 2013). To keep the exposition manageable, I have to set such proposals aside (see Thoms 2015 for discussion). (18) For Italian the conclusion that apparent violations of the left-branch condition under sluicing are due to predicative pre-sluices can be based on the relation between possible adjective position in the NP and their ability to appear in predicative structures (Montali 2014). For English, the conclusion that apparent left-branch extractions are based on copulative pre-sluices can be supported by considering adjectives that are both gradable and restricted to attributive position (on the relevant reading) (Thoms, Barros, and Elliott 2013). (19) See also Park (2011) and Park and Kang (2007). (20) Some of Lasnik’s other arguments rely on scope and binding properties of complex WH-remnants. As discussed in Barros, Elliott, and Thoms (2014), these arguments are tenuous. (21) In the text we discussed Chung, Ladusaw, and McCloskey (1995) but we should also mention Abe (2015) as a member of this family of approaches. (22) The discussion here is intended to illustrate the logic of case (iii). It is not intended as an endorsement of the view that wide-scope indefinites have the syntactic representation in (47a). The issue will be taken up briefly in connection with Griffiths and Lipták (2014) just below. Given the difference in interpretation between the two structures in (48), the facts in (47) can be approached in semantic terms under a semantic identity account.

Page 42 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Movement and Islands (23) Fox and Lasnik’s account of the locality sensitivity of VP-ellipsis builds on a similar catch-22; although the details are quite different: either parallelism is violated (violation of type iii) or movement outside of the ellipsis site violates an island constraint (violation of type i). (24) Note that the status of (50a) is not attributable to a general incompatibility between modal complement ellipsis and WH-question formation, since WH-questions allow modal complement ellipsis just as long as the WH-movement does not originate inside of the elided VP (Aelbrecht 2010: 118–20). (25) For details of implementation, the interested reader is referred to the primary literature mentioned in this section.

Klaus Abels

Klaus Abels received his PhD from the University of Connecticut in 2003. He is a Reader in Linguistics at University College London and co-editor of the journal Syntax. His main interests relate to movement, constraints on movement, interactions of movement types, the formal modeling of movement, and the role of movement in deriving word order typology.

Page 43 of 43

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition

Oxford Handbooks Online Aphasia and Acquisition   Yosef Grodzinsky, Isabelle Deschamps, and Lewis P. Shapiro The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.19

Abstract and Keywords This chapter reports studies of parallelism in VP-ellipsis in two populations whose linguistic ability is lacking. First, a complex comprehension experiment was conducted with individuals with Broca’s aphasia, said to suffer from a specific syntactic deficit that some view as restricted to the deletion of traces of movement, while others argue is related to “complexity.” Though still at a preliminary stage, the experimental record suggests that parallelism, as evinced by the patients’ ability to reconstruct elided VPs, is relatively intact for these patients, indicating that their deficit is restricted. Second, a similar study was run with normally developing, 4–5-year-old children, whose linguistic abilities are not yet fully manifested. Nevertheless, these children use parallelism in an adult-like fashion, and correctly reconstruct elided material in comprehension. We also provide evidence that 3-year-olds use elided VPs in spontaneous language production. Both studies are used to argue for fine deficit analyses, and against generic, “complexity”-based accounts of language development and language loss. Keywords: aphasia, VP-ellipsis, acquisition, experiment, trace deletion, complexity

Yosef Grodzinsky, Isabelle Deschamps, and Lewis P. Shapiro

17.1 The content of this chapter THIS chapter is about verb phrase ellipsis (VPE) in speakers whose language faculty is incomplete—patients with a language deficit who have lost part of their linguistic ability subsequent to (mostly focal) brain damage, and developing children who have not yet reached the steady state of adult language. We show that despite certain syntactic limitations, the ability of both patients with Broca’s aphasia and children to reconstruct an elided VP is intact. While the present data set is of a relatively limited size, this result is nonetheless remarkable, we argue, and is important for our understanding of the

Page 1 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition nature of language loss subsequent to focal brain damage, as well as language acquisition. We briefly review some properties of VPE, move on to their psycholinguistic relevance, and then describe two pilot experiments—in adult aphasic patients and in children—that tested the comprehension of VPE. We establish that VPE (a) requires a comprehender to go beyond the information given on the surface; and (b) requires complex syntactic and semantic analysis. As such, it provides an excellent testing ground for claims regarding language-incomplete individuals. With that in mind, we consider the comprehension of complex linguistic material in patients with aphasia, and in developing children. We report data that indicate that adult patients with Broca’s aphasia and young children are quite capable with VPE, despite deficiencies elsewhere in syntax. We then consider these data in a broader context, and conclude by reflecting on the potential significance of the results of these experiments.

17.2 VPE The context of the present volume permits us to be relatively brief in discussing structural issues regarding VPE, and we restrict ourselves to just those aspects of the construction that are relevant to our endeavor. Our desire for brevity and focus leads us to be somewhat (p. 426) lenient in terms of precision and generality. The ingredients of the analysis of VPE are many, as are the associated questions and debates: What kind of operations are involved in the recovery of elided material? What is the type, size, and constituency of the elided piece? And what constrains the relation between the ellipsis clause E and the overt antecedent clause A? We concentrate on the last question, through an experimental exploration with language-impaired adults and developing children. In our experimental investigation, we started off by studying relatively simple constructions, that is, coordinate VPE, in which the VP of the antecedent clause A must be interpreted in the ellipsis clause E (1a) (Sag 1976a; Williams 1977b; Rooth 1992b; Hardt 1993; Fox 2000; Johnson 2001b; Merchant 2001; see van Craenenbroeck 2017 for a recent survey). Once the VP constituent in A (VPA) cannot be identified as such in E, the sentence becomes unacceptable. The piece sold a car in (1b) and told in (1c) are not constituents, and the resulting sentence is marked: (1)

Most (though likely not all) of the properties associated with the constructions in (1) carry over to VPE sentences with so and some version of auxiliary verb fronting: (2) Page 2 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition

1

Ellipsis does not only have syntactic reflexes. Meaning is also at issue: the content of the elided VPE must parallel that of VPA. In (1a), VPE is structurally identical to VPA, and yet it is clear that (1a) cannot be interpreted as (3). (3)

To account for these facts, assume that the form of the elided VP is governed by a constraint that requires parallelism between VPA and VPE (Fox 2000): (4)

If “identical” means “containing the same syntactic constituents”, then the acceptability of (1a), and the unacceptability of (1b,c) follows directly—VPE must consist of the same (p. 427) syntactic pieces as VP . The unacceptability of (3) follows if Parallelism requires A (roughly) semantic identity. Next, consider inflectional features (tense, person, number). These seem to be exempt from the Parallelism requirement, as the ellipsis clause needn’t parallel the antecedent clause. In (5), both tense and number are different in each of the coordinated clauses: (5)

Data like (5) suggest that the relevant features are left outside of the ellipsis, providing a first indication that the size of ellipsis may not amount to the complete size of VPA (see van Craenenbroeck 2017 for a recent review). Adjusting the size of ellipsis allows Parallelism to be maintained in its strict form (4). Still, a Parallelism constraint that forces content identity must allow for certain holes. That is, the VPE sometimes allows an interpretation that is different from that of the VPA. As is well known, anaphoric elements in object position of the antecedent clause bring about such situations (Williams 1977b; Bach and Partee 1980; Reinhart 1983a). Thus, the meaning of VPE in (6) is ambiguous between (6a) and (6b): (6)

Page 3 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition

Given the strict Parallelism in (4), how can the ambiguity in VPE arise in the (apparent) absence of ambiguity in VPA? Put differently, how can we reconcile between the need for a rigid Parallelism that would force the facts in (1)–(3), and the need for flexibility to account for (6)? One key idea (Reinhart 1983a; Grodzinsky and Reinhart 1993) allows us to maintain rigid Parallelism by playing with the interpretation of the antecedent pronoun in VPA. Assume that this pronoun is ambiguous: it may be assigned to Jim, but it may also be viewed as a variable, bound by a local antecedent t1. In this clause, the two meanings happen to converge. (7)

Now, consider the ellipsis clause. For each meaning, Parallelism guarantees an identity relation between VPA and VPE, and the ambiguity of the former (7) leads to these LFs of the latter: (8)

The difference in reading comes from the fact that while the pronoun in (8a) is assigned to Jim as before, in (8b) it is locally bound. The ambiguity follows, as the resulting LFs of the ellipsis sentence are roughly those in (9): (9)

The meaning in (9a), known as strict identity, allows only for one dog to have been walked—Jim’s; whereas the meaning of (9b), known as sloppy identity, allows two. The ambiguity is thereby obtained, and Parallelism is nonetheless observed. While this analysis is not problem-free (cf. Heim 1998, 2009; Fox 2000, among many others), it illustrates how Parallelism interacts with other assumptions to account for a range of VPE phenomena.2 (p. 428)

Lastly, there is the critical issue of recoverability of the elided part of VPE. Even on its broadest formulation, Parallelism only constrains the form and content of ellipsis; it says Page 4 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition nothing about the elision itself. But why and how can phonologically overt material be missing? Is the ellipsis site itself an empty category of some sort (Lobeck 1995 and especially van Craenenbroeck 2017, which propose that the conditions that license it are akin to those that license pro)? Do the conditions that allow for elision need to be specifically formulated for VPE, or are they, rather, general to all ellipses (cf. van Craenenbroeck and Merchant 2013)? Moreover, is VPE a result of PF deletion, or rather, of copying at an abstract syntactic or semantic level? At present, we remain agnostic on these issues, as the goals of this chapter are modest. Our experimental evidence from developing children and brain-damaged aphasic patients will mostly bear on developmental and neuropsychological issues. Still, we will argue that the aphasic deficit may shed some light on the nature of ellipsis licensing.

17.3 The psycholinguistic relevance of VPE: Broca’s aphasia Next, we turned to test the comprehension of patients with Broca’s aphasia (Walenski et al. 2012). Why can VPE be informative in this context? We can see two reasons: first, results from a comprehension experiment on VPE may bear on our view of the comprehension deficit in this syndrome, and as a consequence, on our understanding of the role of Broca’s region in language processing. Second, the pattern of impairment and sparing that would emerge might point to a neurological natural class that would bear on the theory of VPE.

17.3.1 Broca’s aphasia as a clinical entity Using patients with Broca’s aphasia requires some discussion regarding the status of this clinical entity. Some have questioned the validity of Broca’s aphasia as a stable and identifiable syndrome complex, presenting a pattern that is replicable across patients. The main argument behind the challenge of this 150-year-old syndrome has been that (p. 429) the observed inter-patient variability is too large for this collection of phenomena to count as unitary (Caramazza 1984, passim). Caramazza’s paper paved the way not only for an intense, protracted conceptual debate, but also for several quantitative analyses in which the inter-patient variability, mostly in the receptive domain, was measured. The result that seems to have settled this round of debate was based on a series of retrospective studies on a large data set (n=69, Drai and Grodzinsky 2006a, b): while there is considerable variation between patients who have been assigned this diagnosis, certain performance contrasts in the receptive domain remain very stable. The grouping of these patients, then, seems justified.

Page 5 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition

17.3.2 The comprehension deficit in Broca’s aphasia: TDH vs WM/ generic complexity Broca’s aphasia was traditionally thought of as a deficit in production (cf. Goodglass and Kaplan 1972 for the most frequently used diagnostic test). Later, this deficit was shown to extend to comprehension (Caramazza and Zurif 1976). Since then, its precise nature—and the subsequent role of Broca’s region in language processing—has been hotly debated. At present, there seem to be two dominant views of the comprehension deficit, and the role of Broca’s region: the movement view, embodied in the Trace Deletion Hypothesis (TDH) and its processing ramifications (e.g., Grodzinsky 1984, 1986, 2000; Shapiro et al. 1992; Zurif et al. 1993; Shapiro, Swinney, and Borsky 1998; Grodzinsky and Santi 2008) and the syntactic complexity view, that relates to Working Memory (WM, e.g., Stromswold et al. 1996; Caplan and Waters 1999; Friederici 2002). A brief description of the positions is followed by a discussion of the relevance of a test of VPE comprehension in aphasia. In brief, the movement account is based on the observation that the core comprehension deficit is restricted to structures that involve the displacement of a phrasal constituent. Patients are presented with binary-choice interpretive tasks (essentially ϑ-role assignment, evinced through the choice of the correct ϑ-order in a set-up with two scenarios, one of which makes the sentence true, and the other false via ϑ-reversal). Patients with Broca’s aphasia are typically at chance on object-gap relative clauses (whether these are on the subject to form a center-embedded sentence, or on the object, thereby right-branching); their performance, however, is above chance when the gap in the relative clause is in subject position (see Drai and Grodzinsky 2006a for a review). The idea behind the TDH is to set up the representational conditions that would derive this performance pattern—chance in displacement and above-chance otherwise. As the typical task involves proper ϑ-assignment, chance performance must stem from ϑ-confusion. This is obtained by the following: (10)

The consequence of the TDH is a limited role to Broca’s region in syntax—it only involves displacement.3 It has further been argued that the actual role of this region is the online linking of non-adjacent positions (Shapiro et al. 1998), most notably relating displaced constituents to extraction sites. The evidence comes from deficiencies in realtime tasks that critically rely on the linking of traces to their antecedent. Such deficiencies are observed for patients with Broca’s aphasia through a host of reaction (p. 430)

Page 6 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition time studies that so indicate. As we shall see, this type of task has also been used in testing VPE in aphasia (Poirier et al. 2009). Regarding comprehension performance on VPE structures, it is clear that a limited deficit is predicted, observed only when comprehension depends on the analysis of trace–antecedent relation in a manner that is not properly compensated for by the Direct ϑ-Link strategy. On the other side of the debate, it has been argued (e.g., Caplan and Waters 1999, and much subsequent literature) that Broca’s aphasia is a Working Memory (WM) failure. As WM for “language” is said to reside in Broca’s region (cf. Smith and Jonides 1999), its demise leads to problems with “syntactically complex” structures (the precise nature of complexity is rarely made explicit). The idea is that syntactically complex material taxes WM in an incremental manner, and hence its failure results in an impairment in speakers’ ability to process “complex sentences”. The absence of a formal complexity metric is justified empirically: based on reaction time data, certain constructions are deemed more complex than others. E.g., object-gap relative clauses take longer to process than their subject-gap analogues, and are hence viewed as more complex and more taxing on WM (and to increased activity in Broca’s region in neurologically unimpaired participants; see Makuuchi et al. 2013 for recent discussion). A relatively widespread comprehension failure subsequent to damage to Broca’s region follows. That is, any construction that is deemed as “syntactically complex” is likely to cause comprehension problems to patients that suffer from this brain disease. In particular, that VPE is a complex construction appears uncontroversial: any reasonable construal of the notion of syntactic complexity must include VPE, as this construction, at a minimum, is biclausal, involves copying/deletion of large chunks, and potentially LF movement. WM must be taxed: for Parallelism to be implemented, a comparison of the VP in A to the one in E is mandatory. At present, little evidence is available that pertains to this debate.4 Hence, a WM failure due to damage to Broca’s region should lead to serious comprehension problems (whose precise nature depends on the complexity metric that is assumed). (p. 431)

We can see that the two accounts contrast in prediction regarding VPE: the predicted TDH-based deficit is structured and restricted, whereas the predicted WM-based deficit is rather widespread.

17.3.3 Testing VPE with patients suffering from Broca’s aphasia We based ourselves on methodology developed in Grodzinsky (2005), and used a verification task with VPE sentences with diagnosed Broca’s aphasic patients. Focusing on Parallelism, we presented VPE structures along with situations whose interpretation forces the material in A to have its parallel in E. That is, we presented pairs to patients, where the denotation of the elided part sometimes matches

Page 7 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition the situation, and sometimes does not. The patient’s task was to assign a truth-value to the sentence. We now elaborate on this task. Verification is an experimental paradigm that aims to test the patient’s ability to fill in content from A into E: all scenarios make A true, and the only potential sentence/picture mismatch pertains to E, the ellipsis part. This set-up forces the patient to consult Parallelism in order to assign an interpretation to the sentence and determine its truthvalue. If sentences that accompany situations requiring improper reconstruction are deemed true, we can infer that Parallelism does not exist in the patient’s grammar (or at least cannot be put to use at the right time); otherwise, we can conclude that she possesses the relevant knowledge and can put it to use. To illustrate, consider the sentence in (11), and the accompanying scenarios (11a–d) in Table 17.1. If constraints on ellipsis are not known, or unavailable for use, then any interpretation of E that is consistent with an E whose elided part contains a man as either subject or object may be legitimate. This can be put to the test, in which the evaluation of VPE would provide the critical clue to the patient’s knowledge of Parallelism. If Parallelism is not known, and is hence violable, then the range of possible interpretations of VPE is extended. Table 17.1 Test conditions: VPE sentence + images Parallelism available? (11)

[A The woman kicked a tiger] and [E the man did [VP__]

Yes

No

too] Match

“A woman kicking a tiger, and a man kicking a tiger”

T

T

MisMatch 1

“A woman kicking a tiger, and a tiger kicking a woman”

F

T

MisMatch 2

“A woman kicking a tiger, and a woman kicking a man”

F

T

MisMatch 3

“A woman kicking a tiger, and a tiger kicking a man”

F

T

This is what was put to the test here. In (11), we present a sentence, and four accompanying scenarios (11a–d). Each scenario consists of two events, one that always makes A true, and a second one that makes E true (11a) and false (11b–d) in three different ways. A patient lacking Parallelism should accept all scenarios (11a–d) as true, because they all feature a man—the only constraining factor for VPE interpretation under this assumption. Conversely, a patient that masters Parallelism should only deem true Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition (11a). The difference between knowledge of Parallelism and lack thereof is testable, and presented in Table 17.1, which marks the possible responses of a patient with and without Parallelism: This part of our test consisted of Match cases (11a), as well as MM1 (11b), depicting role reversal in E, the ellipsis clause; MM2 (11c), corresponding to an incorrect object in E, and MM3 (11d), corresponding to an incorrect subject and object in E. Given the world created in the images, both yes and no responses are plausible. As controls, we used non-elided coordinate structures, whose primary meaning was identical to the elided ones. In the absence of ellipsis, the invocation of Parallelism is not necessary for interpretation, no error is attributable to a Parallelism failure, and therefore, all the relevant populations are expected to perform at ceiling (see Table 17.2). (p. 432)

Table 17.2 Control condition: Coordination + images (12)

[IP1 The tiger kicked a tiger] and [IP2 the man kicked a tiger]

Match

“A woman kicking a tiger, and a man kicking a

Expected response T

tiger” MisMatch 1

“A woman kicking a tiger, and a tiger kicking a woman”

F

MisMatch

“A woman kicking a tiger, and a woman kicking a

F

2

man”

MisMatch 3

“A woman kicking a tiger, and a tiger kicking a man”

F

17.3.4 Experiment As we discussed, our test required subjects to verify sentences against depicted visual scenarios—participants responded “yes”/“no” to pairs. In trying to adhere to principles of good design, we took extra measures to ensure that Parallelism must indeed be consulted, and that no tricks or heuristics would help the aphasic patient get around it. First, the design of scenarios was crucial: four scenarios were coupled with each experimental sentence—one scenario made the sentence true (the Match M) whereas the remaining three scenarios (MisMatches 1–3) represented three different ways of making the sentence false. Each scenario was shown separately, and thus every sentence was presented four times—once for each scenario. Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Second, as all our stimuli contained two conjoined clauses, they required two situations per scenario—one for each of the clauses A and E (11), or IP1 and IP2 in the control condition (12); the scene for A always made it true, as the mismatches were made false only by the scene that related to E. (p. 433)

Third, we tried to ensure that there is congruence between NPs in the sentences and the characters in images—every character in the pictures is mentioned, and every NP is depicted. This way, the presence or absence of mentioned referents cannot be used as a heuristic. Fourth, to avoid a violation of the uniqueness requirement imposed by definites, the object of the antecedent clause A was headed by an indefinite. Fifth, every character appeared in pictures as either actor or recipient of action. Sixth, animacy and plausibility were perfectly balanced (rotated across positions in the sentence). Seventh, we had one M and three MM conditions, which could lead to a “yes”/“no” imbalance. We therefore maintained a 2:3 ratio between tokens that required a correct “yes” (M) vs those requiring a “no” (MM) response, thereby preventing strategic responding. We had 18 Match and 27 (9*3) Mismatch tokens (=45 items per condition), for a total of 45*3=135 trials, as in Table 17.3, and the accompanying images in Figure 17.1.

Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Table 17.3 Experimental conditions Match a. COORDINATI ON:

MisMatch 1

MisMatch 2

MisMatch 3

18

9

9

9

18

9

9

9

18

9

9

9

The girl kicked a tiger and the boy kicked a tiger b. VPETOO: The girl kicked a tiger and the boy did too c. VPESO: The girl kicked a tiger and so did the boy

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition As control populations, we tested healthy and righthemisphere damaged patients, who were at- or near-ceiling in all conditions. Next, we started testing patients with a clinical diagnosis of Broca’s aphasia. At present, we have results for three patients, as finding candidates for Figure 17.1 Experimental stimuli testing is exceedingly difficult (Walenski et al. 2012). These patients’ clinical profiles can be found in Table 17.4.

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Table 17.4 Patients’ demographic and clinical profiles Subject

Gender

BDAEa severity level (1=severe, 5=mild)

Years post stroke

Age at test (years)

Educationa l level

009

M

2

5

45

College + 1

019

F

2

14

59

High School

101

M

2

2

60

PhD

Note: a = Boston diagnostic aphasia examination

Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition All three subjects exhibited the familiar TDH-based comprehension pattern: in a sentence-to-picture matching task with “reversible” sentences, they performed better on actives than on passives, and better on subject than on object relative clauses. Once clinical and TDH-based diagnoses were established, we proceeded to the VPE test. Subjects were tested in several sessions, with a short training session, and many breaks during testing. The experiment was conducted with an ethics approval and informed consent in keeping with the ethical principles of testing that apply at McGill University, as well as San Diego State University, where the subjects were tested. Healthy education- and age-matched control participants performed near or at-ceiling on all conditions. Table 17.5 details how our patients performed on the VPE test. (p. 434)

Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Table 17.5 Number of accurate responses per condition for each patient on the VPE test Control condition (number correct) Subj ect

M (/ 18)

MM 1 (/ 9)

MM 2 (/ 9)

MM 3 (/ 9)

009

18

8

0

9

019

101

12

17

8

7

5

8

7

8

VPE (number correct)

M (/ 18)

MM 1 (/ 9)

MM 2 (/ 9)

MM 3 (/ 9)

TOO

18

9

0

8

SO

17

9

0

9

TOO

16

6

4

7

SO

16

3

2

7

TOO

18

8

8

7

SO

17

8

6

6

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Overall, Patient 009 performed well (a total of 105/135 [78 percent] correct); Patient 019 performed slightly worse (a total of 93/135 [69 percent] correct); and Patient 101 performed best (a total of 118/135 [87 percent] correct). Patient 101 leaves little room for discussion, being near- or at-ceiling on all conditions. There can be little doubt that he knows and respects Parallelism. The other patients require a more detailed examination to reveal the source of their errors. (p. 435)

Patient 009 was near- or at-ceiling on M, MM1, and MM3 across all three conditions, and erred systematically on MM2 across the board—he was at 0 correct on both ellipsis and overt coordination. That is, he accepted as true every MM2 trial, which is judged False by healthy speakers, because it features a recipient of action that does not correspond to the object of VPA (hence the covert object of VPE did not refer to it). What is crucial in this case is the fact that he judged these sentence–image pairs as True, in both control and two test conditions. This highly systematic error, whose source is unknown, generalizes over both VPE and coordination, hence seems unrelated to VPE. Viewed against his performance on the control, then, we can safely assert that 009 demonstrated knowledge of Parallelism, and an additional yet unrelated deficit. Patient 019 presents a more complex picture. On the Match trials, she was not as apt as the previous patients, but nevertheless performed above chance. This relatively weak performance (across both VPE and COORD conditions) provides the backdrop against which her performance on the MisMatch trials must be evaluated. She was at chance on MM1 and MM2, and above chance on MM3. These results are not easy to analyze quantitatively or interpret, and will be left in their raw form. Performance on the COORD condition (expected to serve as a baseline) is rather low; performance level on the test conditions must be evaluated accordingly. Two aspects of the result provide hints regarding the preservation of Parallelism: first, on the Match trials of both +VPE conditions, the patient scores are well above chance; second, performance levels are almost identical for both +VPE conditions. Thus, from a perspective that focuses on Parallelism, the performance of this small group of patients provides no evidence for failure in both knowledge and execution. Moreover, the virtually identical performance scores of all three patients across the two VPE types (…did too,…so did) strongly suggests that the problem does not arise from a lexical deficit (e.g., misunderstanding of the word too). Together with the fact that the main sources of error were independent of VPE (as they occurred as frequently in the control condition), this cross-condition identity also lends credibility to our interpretation, according to which a Parallelism failure is not part of the deficit in Broca’s aphasia.

17.3.5 Patients’ relative success in analyzing VPE and the complexity/ WM view We conducted this experimental study with the hope of shedding new light on the comprehension deficit in Broca’s aphasia, and of gleaning new neurological insights regarding the theory of VPE. We can now clearly assert that the patients were able to use Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Parallelism for the comprehension of VPE. This assertion helps to rule out the (rather vague) view that Broca’s aphasic patients fail to comprehend complex sentences—VPE can hardly be viewed as a simple construction. Regarding a WM deficit: an overt VP in clause A and its elided copy in clause E are large non-adjacent chunks; VPA cannot be reconstructed online in VPE with WM. Our results thus exclude an account of Broca’s aphasia as a generic deficit in analyzing syntactically complex stimuli, or as a plain WM impairment.5 (p. 436)

17.3.6 Patients’ relative success in analyzing VPE and the movement (TDH) view What about a movement, TDH-based account? VPE contains a relation between the antecedent VP and the ellipsis site (VPA and VPE); we might therefore wonder whether this dependency relation is affected by the deficit in Broca’s aphasia. Standard approaches to VPE view the relation between overt and elided material as being akin to pronominal binding, albeit a silent one. Schwarz (2000) and Elbourne (2008) argue against a movement analysis on the basis of highly complex data.6 If the deficit in Broca’s aphasia is restricted to traces, this approach predicts that patients’ comprehension ability of VPE as tested here is preserved. Prima facie, then, VPE stays outside the scope of the TDH, hence its representation is expected to be intact in Broca’s aphasia, as our experiment indeed found.7

17.3.7 VPE in reaction time experiments So far, we have looked at experiments in which comprehension error rates are the dependent measure. We now turn to the time domain. Here, we find two reaction time studies on VPE with Broca’s aphasic patients (Poirier et al. 2009; Walenski et al. 2012). These present an interesting new puzzle. Consider first some results about priming-at-adistance—a paradigm in which participants are tested in a priming paradigm, however the target is not adjacent to the prime, but rather appears further downstream. Neurologically intact populations do show remarkable priming-at-a-distance that is based on trace–antecedent relations. A displaced antecedent can prime not only adjacently, but also at the position of their trace: (13)

Nicol and Swinney (1989), and many other authors since, have used the Cross-Modal Lexical Priming technique to determine whether listeners fill gaps online. They presented subjects with spoken sentences such as (13). At the positions marked by [1] and [2], visual (p. 437) probes that were semantically related to the correct antecedent (boy), as well as unrelated control probes, were presented. As always with this task, subjects were required to make lexical decisions to the visual probes. It was discovered that there was Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition significant priming for probes related to the correct antecedent at the gap ([2]), but not at the pre-gap ([1]) probe site. The same group later tested patients with Broca’s aphasia, who failed to prime in this way (e.g., Zurif et al. 1993), in keeping with the predictions of the TDH. With this picture in mind, Shapiro et al. (2003) turned to VPE. They found that in neurologically intact populations, the copy at the ellipsis site induces priming to a noun related to the elided object. Participants are given a Cross-Modal Priming task with sentences as in (14): (14)

Healthy adults reacted faster to a related word when it is presented at point [2], but not when presented at points [1] or [3]. This pattern seems to be due to the fact that this word was primed by the copy of the elided object, as priming was observed only at the ellipsis site (point [2]). Against this background, Poirier et al. (2009) tested patients with Broca’s aphasia on a similar task: Broca’s patients evinced no priming of the object from the antecedent VP at the elided position in the ellipsis clause, though in a similar study (Walenski et al. 2012), Broca’s patients evinced delayed priming (that is, priming at a probe position downstream from the elided position). The conflict between these results and ours is puzzling: previously, data from CrossModal Priming aligned with comprehension data. That is, patients failed to prime exactly in constructions where their comprehension was impaired. In VPE, we find evidence for the intactness of Parallelism in comprehension in the face of priming deficiencies. At present, we have no account of this performance split, but we note that the patients with Broca’s aphasia do prime for the elided material, though it is later in the time course of the sentence relative to unimpaired participants. Thus, while the results we obtained are highly structured, and help to exclude certain accounts, a full explanation of the experimental findings is not yet available. Experimentation with ellipsis that would go beyond VPE—gapping, sluicing, N’-ellipsis— would hopefully help to shed light on the nature of the deficit in Broca’s aphasia.

17.4 The psycholinguistic relevance of VPE: Acquisition We now move to discuss another population with incomplete linguistic abilities— developing children, whose abilities in VPE bear on yet another central issue—the “innateness debate”. If children only imitate input, as many have argued (cf. Gennari and Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition MacDonald 2006 for a recent statement), and if “input” only means overt phonological strings, then there is little room for ellipsis in children’s language, for it requires the reconstruction of missing material in the ellipsis clause (E) from the antecedent clause (p. 438) (A)—an operation for which no source of evidence is available on this approach. To produce and properly analyze VPE, children must go “beyond the information given,” especially given the paucity of ellipsis constructions in the input. To see how directly children’s abilities on ellipsis, and on VPE in particular, bear on this issue, consider first a recent exchange on “innateness” in the context of children’s comprehension of relative scope of logical operators, and specifically on whether they are able to represent “inverse scope” of quantifiers. At issue are scopally ambiguous sentences—ones that contain more than one Scope-Bearing Element as in (15). Representations of different orderings of the bolded words not and some produce distinct meanings—(16a) entails that no child was found by the detective, whereas (16b) does not: (15)

(16)

At issue is whether children who are confronted with (15) are able to represent not only surface scope representations (16a), but also inverse scope relations (16b). Going “beyond the information given” (Bruner 1957) amounts to analyzing (15) as (16b). Phonological material is identical in both cases, but (16b) requires an abstract (LF) reordering operation. Details aside, Musolino and Lidz (2004), as well as Conroy, Lidz, and Musolino (2009), have claimed that children (up to 5:9 years of age) are only able to represent surface scope (16a).8 Gennari and MacDonald (2006) argue that this result provides empirical support for the claim that language development is entirely dependent on input, rather than determined by innate knowledge. Specifically, they claim that distributional patterns observed in adult speech production influence both adult and child comprehension. Given that inverse scope is rarely produced by adults, children cannot be expected to possess it early on. Children’s performance on VPE may shed new light on whether syntax and semantics acquisition is a purely input-driven process, or is, rather, privy to prior, innate, linguistic knowledge. While relatively infrequent, VPE comprehenders must invariably go beyond the information provided by surface form—they must complete material missing in E, and guarantee that it parallels A. No surface-based strategy, allegedly used by children in the “inverse scope” case discussed above, can rescue the child. She must reconstruct elided Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition material in a highly constrained fashion. Parallelism, a crucially necessary principle, cannot be attested in the input, and the computation is not easy to perform if one’s parsing system is still budding. Moreover, if abilities depend on exposure, as Gennari and McDonald (2006) would have it, then VPE should be very difficult to master, as it is rather infrequent in use. Gennari and MacDonald, then, would expect constraints on ellipsis to be absent from the child’s syntactic arsenal.9 A radical innatist, by contrast, would expect ellipsis to be known at an early age. At hand is, therefore, a prediction that distinguishes the two approaches. And yet, the few studies of VPE in children (most notably, Thornton and Wexler 1999; cf. also Göksun et al. 2011 for some experimental work, and Santos 2006 for work on Portuguese) have not directly tested their ability to invoke Parallelism. These studies have by and large focused on the interpretation of pronouns in VPE (in the context of the so-called “condition B debate” of the 1990s; Chien and Wexler 1990; Grodzinsky and Reinhart 1993). Little experimental work has been done on other aspects of VPE (with the exception of Santos 2009a; see subsection 17.4.2). (p. 439)

We therefore set about providing experimental evidence from VPE that would bear on the debate.

17.4.1 Spontaneous production of VPE: Parallelism is used in early childhood Our first step was to investigate young children’s abilities to check that VPA=PVPE (see Grodzinsky 2005). To anticipate our findings, we found that their comprehension performance—for reconstruction under Parallelism—was near-perfect at a surprisingly early age. This remarkable finding also emerged in a search of the CHILDES database, which indicated that children use VP-ellipsis in free speech very early on, even before they turn 3. There were not many such uses (20 instances out of 3199 utterances for the 3-year-olds, and 11 out of 1254 for the 4-year-olds), but VPE was nonetheless attested in the spontaneous speech of several children (see Table 17.6 for examples). Table 17.6 Examples of spontaneous production of VPE by children Name of child

CHILDES file name

Example

Abe

Kuczaj_Abe102

CHI: I don’t know. FAT: you didn’t do anything? CHI: we did too. FAT: tell me. CHI: we did a lot of stuff.

Ben and Deb

Garvey_Bendeb

DEB: I will be the policemen okay?

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition BEN: no I wanta [:want to] I wanta [:want to]. DEB: I will too okay? DEB: okay? DEB: okay. Nina

Suppes_Nina46

CHI: what kind of bus does he have? MOT: he has a Volkswagen bus. CHI: Dennis does too. MOT: that’s right Gary and Dennis have the same car. CHI: uhhuh [=yes].

It is difficult to see how any of these utterances can be analyzed as anything other than VPE. Their meaning, moreover, cannot be reconstructed in the absence of Parallelism: Abe (p. 440) responds to a full negative question by eliding the VP and excluding its negation. To be sure, he then elaborates on the father’s request, and reiterates the VPE as a full sentence. Deb asks for consent to her being the police(wo)man, and Nina uses VPE by way of confirming that Dennis has a VW bus. And these, recall, are 3-year-olds, said to be unable to carry out operations that involve unattested stimuli and use inverse scope because it is too taxing on their parser. These data add to evidence by Santos (2009a), in which examples of child productions in Portuguese are given, that are claimed to be unambiguous instances of ellipsis.

17.4.2 Children comprehend VPE very early on: Experimental evidence Our success in finding evidence that children are capable of constructing VPE in production led us to a comprehension study, in the hope that it would bear directly on the “innateness/input dependence” debate. As 3-year-old children have difficulties in sitting still and concentrating, we turned to 4-year-olds. Focusing on Parallelism, and basing ourselves on methodology developed in Grodzinsky (2005), we planned to give children a verification task with VPE sentences, presented along with situations whose interpretation forces the material in A to have its parallel in E. The method was to present pairs to children, where the denotation of the elided part sometimes matches the situation, and sometimes does not. The child’s task would be to assign a truth-value to the sentence. We now elaborate on this task, which has also been used by Santos (2009a), who found that by and large, children interpret ellipsis in an adult-like fashion. As this experimental paradigm aims to test the child’s ability to fill in content from A into E, all scenarios make A true, and the only potential sentence/picture mismatch pertains to E, the ellipsis part. This set-up forces the child to consult Parallelism in order to assign an Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition interpretation to the sentence and determine its truth-value. If sentences that accompany situations requiring improper reconstruction are deemed true, we can infer that Parallelism does not exist in the child’s grammar (or at least cannot be put to use at the right time); otherwise, we can conclude that she possesses the relevant knowledge and can put it to use. As an example, consider the sentence in (17), and the accompanying scenarios (17a–d) in Table 17.7. If constraints on ellipsis are not known, or unavailable for use, then any VP may be (p. 441) a legitimate completion for E. If so, then she should accept all the situations (17a–d) as true (this deficit would become apparent when their errors on VPE are compared to the control coordination condition); conversely, if the child knows and uses constraints on ellipsis, she should only deem (17a) as true. The difference between these situations is depicted in Table 17.7, which marks the possible responses of the child with and without Parallelism. Table 17.7 Test Condition: VPE + images Parallelism available? (17) [A The detective found a child] and [E the policeman did [VP__]

Yes

NO

too] a.

“A detective finding a child, and a policeman finding a child”

T

T

b.

“A detective finding a child, and a policeman finding a detective”

F

T

c.

“A detective finding a child, and a detective finding a policeman”

F

T

d.

“A detective finding a child, and a child finding a policeman”

F

T

Note that all the situations in Table 17.7 are plausible, and the child’s task is to check the sentence against them, with constraints on ellipsis being the sole factor determining truth conditions. To make sure that sentence length, or number of clauses, are not an intervening factor that increases difficulty, we used non-elided coordinate structures with the same meaning as controls, as detailed in Table 17.8. Table 17.8 Control condition: Coordination + images (18) [IP1 The detective found a child] and [IP2 the policeman found a child] a.

“A detective finding a child, and a policeman finding a child”

T

Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition b.

“A detective finding a child, and a policeman finding a detective”

F

c.

“A detective finding a child, and a detective finding a policeman”

F

d.

“A detective finding a child, and a child finding a policeman”

F

Our test, then, required subjects to verify sentences against depicted visual scenarios— participants responded “yes”/“no” to pairs. In trying to adhere to principles of good design, we took extra measures to ensure that Parallelism must indeed be consulted, and that no tricks or heuristics would help the child get around it. We ran this experiment at the Centre de La Petite Enfance de McGill (with an ethics approval and informed parental consent in accordance with the McGill Research Ethics Board). All participants were right-handed with normal hearing. They range in age from 4:6 to 5:6 (5 boys, mean=59 months). Our participants were seven children. They were tested by one of us (I.D.) one child at the time, in a secluded area within their classroom. Sentences were read aloud, and the pictures shown concomitantly. The children responded to each pair verbally, and their responses were recorded immediately. The relatively large number of test items (n=45), and the young age of the children, forced 4–5 testing sessions per child. Prior to each testing session, the instructions and four training sentences were presented to each child. Each session lasted no more than 30 minutes. Results (number of correct responses) are presented in Table 17.9.

Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition Table 17.9 Experimental results Control condition Chil d

1

2

3

4

5

6

7

VPE

M (/ 6)

MM 1 (/ 3)

MM 2 (/ 3)

MM 3 (/ 3)

6

3

3

3

6

6

6

6

6

6

3

3

3

3

3

3

3

3

3

3

3

3

3

3

3

3

2

3

M (/ 6)

MM 1 (/ 3)

MM 2 (/ 3)

MM 3 (/ 3)

TOO

6

3

3

3

SO

6

3

3

3

TOO

6

3

2

3

SO

6

3

3

3

TOO

6

3

3

2

SO

6

3

3

2

TOO

6

3

3

1

SO

6

3

3

2

TOO

6

3

3

3

SO

5

3

3

3

TOO

6

3

3

2

SO

6

3

3

3

TOO

5

3

3

2

SO

5

3

3

2

Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition As can be easily seen, there were hardly any errors on any of the conditions. In both control and test conditions, children correctly verified the Match pictures, and rejected the MisMatches. We can safely conclude, it seems, that in every instance where the analysis of conjoined clauses was not an obstacle, Parallelism was used for ellipsis reconstruction. In conclusion, then, children invoke Parallelism in VPE in both production and comprehension. VPE is an infrequently used sentence type, and the task forced children to use a linguistic constraint to which they have no direct evidence in the input. Our results seem to provide clear evidence for the use of a highly abstract, unobserved, linguistic principle—Parallelism—in the processing of VPE. The significance of this result is further underscored by the relative rarity of VPE in the child’s ambient language. Early mastery of VPE speaks clearly against Genarri and MacDonald’s position, and weakens the force of the “isomorphism” claim, that can no longer cover VPE, even though VPE would be a natural candidate for such a requirement. (p. 442)

17.5 Where are we? What to do next? We have seen relatively good comprehension performance in our pilot study of Broca’s aphasia. We have also seen that very young children use and analyze VPE properly, despite the fact that such use requires highly abstract principles that are never attested in the data. Theoretically, the evidence from aphasia speaks against a generic “complexity” deficit in these patients, and suggests that VPE can be used as a research tool with patients, which would hopefully explore more refined structural issues, aimed at providing evidence that is relevant to theory construction. With children, we have provided yet another demonstration that highly complex knowledge is attained in the absence of evidence. Experimentally, (p. 443) it is clear that our perspective must first be broadened to other types of ellipsis—gapping, sluicing, N’-ellipsis. In children, we must look at how abstract principles like Parallelism interact with children’s presumed deficiencies with inverse scope. In aphasia, we must gain deeper understanding of the discrepancy between on- and offline performances. In the meantime, the preliminary hints at hand suggest that ellipsis and its kin are likely to provide important clues on language development and breakdown.

Acknowledgements We gratefully thank the reviewers, as well as the editors of this volume, whose comments helped to improve this manuscript. This work was supported by ELSC, and by a grant from the Israel Science Foundation (ISF) to Y.G.

Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition

Notes: (1) Inverted constructions like (2) are distributionally more restrictive than (i.a), e.g., they disallow negation: ((i))

A multitude of reasons—syntactic and semantic (perhaps even phonological)—may conspire to lead to this restrictiveness. Yet these issues are obviously beyond the scope of the present study. (2) Among the drawbacks of this analysis is that it has the consequence that a possessive pronoun in object position is always ambiguous (between the bound variable and referential readings), or else the copying of two different LF representations into E would not be possible (cf. Fox 2000; Heim 2009 for discussion). (3) No empirical test has been conducted to examine whether deletion extends to other empty categories and elements that are represented on the surface (Grodzinsky 2011). (4) We are aware of one attempt to study VPE in aphasia: Vasić, Avrutin, and Ruigendijk (2006b) used what they claimed are Dutch VPE constructions (although see van Craenenbroeck 2017: 20–2 for recent discussion of whether or not Dutch has VPE). Vasić et al. tested patients with Broca’s aphasia in a picture selection task. They were interested in the interpretation of pronouns, hence all their sentences contained anaphora, with distractors as described in (i). The relevant foil is the “related distracter” (i.c), which forces the possessive pronoun in the antecedent clause to have a local antecedent, but the possessive pronoun in the ellipsis clause to have a deictic antecedent. For this meaning to match the sentence, Parallelism must be violated. ((i))

These scenarios may have led to infelicity: a silent pronoun (the copy in the elided VP) refers deictically, even though emphatic stress (cum pointing) is necessary for deixis to work in such contexts. (NB. controls’ error rates were ~10 percent, a non-negligible level). This work, at any rate, did not test Parallelism. (5) Other pieces of empirical evidence point to a similar conclusion regarding a WM deficit (cf. Grodzinsky and Santi 2008). Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition (6) Grodzinsky (2005) argues, based on preliminary data from an experiment with patients, that VP-fronting at LF is mandatory in VPE, as VPE is impaired in Broca’s aphasia. He argued that if the TDH is the right deficit analysis, then mandatory VPfronting in VPE would predict the pattern he observed. However, the data presented above, which come from a study with much improved design, seem to point to the intactness of VPE in Broca’s aphasia, and hence no conclusions regarding VP-fronting follow. More refined experiments are needed to bear on this issue. (7) Tanja Temmerman suggests that if we add a trace to a VP-ellipsis construction, the deficit would manifest. She proposes to test cases of remnant movement: ((i))

((ii))

(8) Gualmini and colleagues (2004) argue against this position, on empirical grounds. (9) Gennari and MacDonald’s position would thus be unable to explain how Parallelism (or any other unattested constraint) is ever attained by humans.

Yosef Grodzinsky

Yosef Grodzinsky is Professor of Neurolinguistics at the Edmond and Lily Safra Center for Brain Research, the Hebrew University of Jerusalem, and a Visiting Professor at the Institute of Neuroscience and Medicine (INM-1), Forschungszentrum Jülich. His research has focused on the neural basis, acquisition, and processing of syntactic and semantic knowledge. At present, his work explores the neural bases of overt and covert negation. Previously, Grodzinsky was a Professor and Tier-I Canada Research Chair of Neurolinguistics at McGill University, a Professor of Psychology at Tel Aviv University, and a Research Professor of Neurology at the Boston University School of Medicine. He is the recipient of several awards, and his research has been funded by government agencies in the US, Canada, Israel, and Germany. Isabelle Deschamps

Isabelle Deschamps is a postdoctoral research scholar in the Department of Rehabilitation at Laval University. Previously she received her bachelor’s degree in Psychology and Linguistics as well as her PhD in Communication Sciences and Disorders from McGill University. Her research interests focus on issues pertaining to phonological processes during speech perception and production. In addition, her

Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Aphasia and Acquisition research aims to understand the relationship between phonological processes and other cognitive functions such as verbal working memory. Lewis P. Shapiro

Lewis P. Shapiro is an Emeritus Professor at San Diego State University. Research interests include charting the moment-by-moment unfolding of language and cognitive processing in neurologically healthy adults and those with brain damage; brain–language relations through lesion analyses and brain imaging; and the efficacy and neurological implications of treatment for adults with language disorders. Dr Shapiro’s work has been funded continuously through the US National Institutes of Health since 1988.

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies

Oxford Handbooks Online Parsing Strategies   Masaya Yoshida The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.20

Abstract and Keywords In order to successfully comprehend sentences involving ellipsis, the online sentenceprocessing mechanism must be able to identify the ellipsis site, find its antecedent, and recover the content of the ellipsis site by referring to the antecedent. The psycholinguistic study of ellipsis aims to reveal aspects of the mechanism of online sentence processing by studying how ellipsis constructions are processed online. In this chapter, the following questions are asked. What structure does the parser build in the ellipsis site? How does the parser build the structure of ellipsis? When does the parser recognize ellipsis? Where does the parser find the antecedent of ellipsis? By answering these questions, this chapter tries to reveal what parsing strategies are employed in the online processing of ellipsis constructions. Keywords: sentence processing, parsing strategies, copying, antecedent search, ambiguity resolution

Masaya Yoshida

18.1 Introduction THIS chapter summarizes what experimental studies on ellipsis can tell us about the mechanism of online sentence processing.1 Grammatical studies on ellipsis have revealed many properties of ellipsis constructions (see Chapter 1 and Chapter 2). Among others, the following three properties are prominent. First, when ellipsis is licensed, the ellipsis site normally has a salient linguistic antecedent (see Hankamer and Sag 1976 for detailed discussion). Second, related to the first property, the ellipsis site and its antecedent site must stand in a certain parallelism relation. In other words, the ellipsis site is not licensed when the ellipsis site does not have a parallel antecedent (see Merchant 2001; Chung 2013 for detailed summary and

Page 1 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies discussion on the parallelism requirement on ellipsis). Third, the ellipsis site is often licensed in a specific syntactic configuration (Lobeck 1995, among many others). Thus, in order to successfully comprehend sentences involving ellipsis, the online sentence-processing mechanism must be able to identify the ellipsis site, find its antecedent, and recover the content of the ellipsis site by referring to the antecedent. The psycholinguistic study of ellipsis aims to reveal aspects of the mechanism of online sentence processing by studying how ellipsis constructions are processed online.2 The major questions in the psycholinguistic study of ellipsis can be formulated in the following ways. What structure does the parser build in the ellipsis site? How does the parser build the structure of ellipsis? When does the parser recognize ellipsis? Where does the parser find the antecedent (p. 445) of ellipsis? In what follows, I discuss these four questions one by one and try to provide an overview of what the study of ellipsis has been revealing about the mechanism of online sentence processing.3

18.2 What structure does the parser build in the ellipsis site, and how? The content of the ellipsis site must be recovered from the antecedent. Therefore, during online processing, the parser must find the antecedent and recover the content of the ellipsis site from the antecedent. Thus, it is possible to infer the structure in the ellipsis site by looking at the structure of the antecedent. In other words, the degree to which the parser is sensitive to the structural details of the antecedent can indicate how extensive the unpronounced structure is. Making use of the Cross-Modal Priming (CMP) paradigm, Shapiro and colleagues (Shapiro and Hestvik 1995; Shapiro et al. 2003; Poirier et al. 2010) have investigated whether the content of the antecedent is “reactivated” at the ellipsis site. Their study is motivated by CMP studies in other domains. In the study of anaphora and empty categories, it has been argued that the content of the antecedent of a pronoun or of a gap linked to a wh-phrase is reactivated once the pronoun or the gap is processed (McElree and Bever 1989; Nicol et al. 1994; Nicol and Swinney 2003, among others).4 The interpretation of the pronoun or the gap depends on its antecedent, and to achieve the interpretation of the sentence, the processor needs to find the antecedent and link the pronoun or the gap to the antecedent. This process of finding the antecedent and linking the dependent element to the antecedent leads to the reactivation of the antecedent. Adopting similar logic, Shapiro and colleagues asked whether the content of the antecedent for the ellipsis site is reactivated. If the reactivation of the antecedent is observed, this means that the parser accesses the information of the antecedent in memory and tries to link the ellipsis site and the antecedent online. Shapiro et al. (2003) investigated the processing of VP-ellipsis (VPE). In their experiment, while sentences like (1) were audibly presented a word was visually presented. The participants performed a lexical decision task while listening to the sentence. Shapiro et al. found faster lexical decision times when the word semantically related to the object NP in the first clause Page 2 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies (e.g., neck for a tie) was presented at the VPE site (at the point of did). This faster lexical decision time was not observed when the word was presented in the earlier position in the sentence, prior to the VPE site. Furthermore, they did not find a priming effect of the subject the mailman. (1)

Their result suggests that when the parser recognizes VPE, it searches for the appropriate antecedent, the VP in the first clause, and links the VPE site to the antecedent, resulting in the reactivation of the antecedent VP.5 Therefore, reactivation of the object was observed while the reactivation of the subject, which is outside of VP, was not observed.6 This study suggests that the parser accesses the antecedent of an ellipsis, and also that the parser is sensitive to the content of the antecedent site, recovering only a VP as the antecedent of VPE to the exclusion of its containing clause. However, studies of this type do not tell us how the structure of ellipsis is built. (p. 446)

In a series of studies on the processing of VPE and sluicing, Frazier and Clifton argue that the ellipsis site in these constructions is associated with syntactic structure and that the parser employs a copying mechanism to build the structure of ellipsis (Frazier and Clifton 2000, 2005, 2006). Essentially, what they argue is that the structure of the antecedent clause is literally copied into the ellipsis site. As evidence for this proposition, Frazier and Clifton (2005) show that the distance between the ellipsis site and the antecedent impacts the processing of the ellipsis site.7 In one of the experiments in Frazier and Clifton (2005), they tested the following paradigm of sluicing: (2)

In a self-paced moving-window reading study, they found that when the position of the verb (studied) associated with the wh-phrase is farther from the ellipsis site, the processing speed of the elided (sluiced) clause is slower. Assuming that either just one of the conjuncts or the whole coordinated structure can serve as the antecedent of ellipsis (see Merchant 2001: 223–6 for related discussion), Frazier and Clifton argue that when the parser cannot find the antecedent in the nearest conjunct, it considers the whole coordinated structure as the antecedent.8 As a result, in (2b) the bigger size of the antecedent (i.e., the whole (p. 447) coordinated structure) is taken as the antecedent by the parser, which gives rise to the reading slowdown. In a study of gapping, Carlson, Dickey, and Kennedy (2005) also argue that the size of the antecedent impacts the processing of the ellipsis site.9 They found that in gapping sentences like (3a), the NP in the second conjunct (Sam) was preferably interpreted as Page 3 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies the object (…insulted Sam). However, in sentences like (3b), where the PPs are fronted in both of the conjuncts, this strong bias for the object interpretation of the NP in the second conjunct (Sabrina) was absent. (3)

They interpret this result as indicating that the structural complexity of the antecedent clause impacts the processing of the ellipsis site. They argue that in (3a), the VPcoordination structure is available, and thus, it is easy to interpret the NP Sam as the object of the verb insult. However, in (3b), due to the fronted PP, which is adjoined to the clausal projection, only the bigger clausal coordination structure is available. Thus, the subject interpretation of Sabrina is more accessible and the strong object interpretation bias is not seen. This result, like Clifton and Frazier’s results, indicates that the structure of the antecedent affects the processing of the ellipsis site. Even though these studies show that the size of the antecedent matters in terms of the processing of the ellipsis site, they do not show whether and to what extent the parser is sensitive to the structural details of the antecedent. If the parser employs a copying mechanism, it is expected that the detailed structure of the antecedent will be copied into the ellipsis site, and thus that the parser will be sensitive to the structural details of the antecedent site while processing the ellipsis site. Dickey and Bunger’s (2011) study compared the processing of merger-type sluicing (henceforth just sluicing: Chung, Ladusaw, and McCloskey 1995), in which the antecedent clause contains an indefinite NP as the antecedent of the wh-phrase, with the processing of sprouting-type sluicing (henceforth sprouting: Chung, Ladusaw, and McCloskey 1995), in which an explicit antecedent of the wh-phrase is missing. This research suggests that the structural parallelism of the antecedent and the ellipsis site affects the processing of the ellipsis site. They argue that in sluicing, the antecedent clause and the ellipsis site are structurally parallel because the antecedent clause contains an object NP, but in sprouting, they are not, because the projection of the object NP is missing in the antecedent clause. They tested the following paradigm in a self-paced moving-window experiment. (4)

In this paradigm, they compared the sluiced sentences and the non-elliptical sentences. They found that when the first clause (the clause before but) involves an implicit object (typed quickly), the reading of the region including the wh-phrase (what exactly/what she typed) was slower both in sluiced and non-elliptical conditions. Based on this finding they conclude that structural parallelism impacts the processing of an ellipsis (p. 448)

Page 4 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies site (similar results are reported in VPE contexts by Tanenhaus and Carlson 1990; Arregui et al. 2006; see also Kim et al. 2011 on the discussion of voice mismatches in VPE). Furthermore, they emphasize that the effect of structural parallelism is not specific to ellipsis. Given the similar result between the sluiced and non-elliptical conditions, they argue that sluicing and non-elliptical wh-interrogatives involve the same syntactic structure and operations. Importantly, for our purpose, Dickey and Bunger’s study shows that the parser is sensitive to the structural details of the antecedent clause when the ellipsis site is processed. Their result supports the claim that the parser builds the structure of the antecedent site inside the ellipsis site, as argued for by Frazier and Clifton. The slower reading time of the sprouting wh-phrase is expected if the parser copies the structure of the antecedent into the ellipsis site. In the sprouting structure, the object NP position is missing in the antecedent clause. Thus, the parser needs to reanalyze the copied structure and add an object NP node, because the wh-phrase needs a position where it can receive its thematic role and form an interpretable wh-gap chain. In sluicing, because the object NP position is available in the antecedent clause, such reanalysis need not take place. Dickey and Bunger argue that this reanalysis process leads to the slower reading time of the wh-phrase in the sprouting structure. Extending these previous findings, Yoshida and colleagues asked whether the parser is sensitive to further structural details of the antecedent site when the ellipsis site is processed (Yoshida et al. 2013a; Yoshida et al. 2013b). First, Yoshida, Lee, and Dickey (2013b) asked if the parser is sensitive to structural constraints such as islands (Ross 1967) in the processing of ellipsis. They contend that if the effect of an island is seen during the processing of an ellipsis site, this would strongly support the position that the parser builds a sufficiently detailed syntactic structure in the ellipsis site to support island constraints. It has long been known that while sluicing is insensitive to islands, sprouting is sensitive to them (Chung, Ladusaw, and McCloskey 1995). For example, in (5), when an indefinite NP (something) that serves as the antecedent of the wh-phrase is inside an island (clausal adjunct island), the example is acceptable. However, when the antecedent of the wh-phrase is an implicit object (i.e., in sprouting) and this implicit object is within an island, the example is not acceptable. (5)

Roughly put, Chung, Ladusaw, and McCloskey’s argument is as follows. In (5) the antecedent involves an island structure. Under LF-copying theories, the whole structure of the antecedent is copied into the ellipsis site at LF and the wh-phrase and the implicit object inside the copied structure form a dependency. As a result, this dependency must go across the adjunct island boundary in (5), resulting in the unacceptability due to an

Page 5 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies island violation. On the other hand, sluicing has different means to form a dependency that is not affected by islands. Taking advantage of this island sensitivity of sprouting, Yoshida, Lee, and Dickey investigated whether the parser will build an island-violating structure inside an ellipsis site. They tested the sentences in (6) in a word-by-word self-paced moving paradigm. (p. 449)

(6)

In (6a), the indefinite NP (something) and the implicit object are in a non-island complement clause. In (6b) they are embedded in an adjunct island (because-clause). In all of the conditions, the wh-phrase, what, is followed by a full clause. That is, none of the conditions involves ellipsis. Importantly, however, at the point of what precisely, each sentence is ambiguous between sluicing and non-sluicing. Faced with the ambiguity of the wh-phrase, does the parser choose one structure at the point of the wh-phrase, or does it delay structure building until decisive bottom-up information becomes available to disambiguate the structure? This can be tested by looking at the processing of the wh-phrases. A wh-phrase in a sprouting structure is known to be read slower than one in a sluicing structure (Dickey and Bunger 2011). Thus, if the parser chose the ellipsis structure at the point of the wh-phrase in (6a), the wh-phrase should be read slower when the first clause involves an implicit object than when the first clause involves the indefinite NP something: the former leads to a sprouting structure and the latter leads to a sluicing structure. Dickey and Bunger found that the wh-phrase is indeed read slower in the implicit object condition than in the indefinite NP condition. The parser, therefore, picks the ellipsis structure at the point of the wh-phrase. Importantly, at the same time they found that this contrast in slowdown of the wh-phrase disappears in (6b). They argue that the lack of reading time slowdown in the sprouting version of (6b) means that the parser did not attempt to construct the structure of sprouting. This follows from the island sensitivity of sprouting plus the well-known observation that the parser does not attempt to form a wh-gap dependency across an island in online sentence processing (Stowe 1986; Traxler and Pickering 1996; Philips 2006). The difference between (6a) and (6b) is whether the first clause contains an adjunct island or not. Building a wh-gap chain in the implicit argument version of (6b), unlike in the overt argument version, would result in an island-violating sprouting structure. To avoid the island-violating structure, the parser did not attempt to build a sprouting structure in (6b). Thus the slowdown of the wh-phrase associated with the sprouting structure was not observed. This result suggests that the processing of the ellipsis site is sensitive to the detailed syntactic structure of the antecedent. Furthermore, this result is predicted by the copying approach whereby the parser copies the structure of the antecedent into the ellipsis site Page 6 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies and builds silent syntactic structure within the ellipsis site. The parser avoids positing the structure of sprouting-type sluicing because the antecedent clause contains an adjunct island. This indicates that the parser refers to the antecedent clause and evaluates its structure. In terms of the copying approach, this means that the parser decides whether to copy the structure of the antecedent into the ellipsis site or not. If the result of the copying would lead to an island violation, the parser does not do so. Adopting similar logic, Yoshida, Dickey, and Sturt (2013a) asked whether the hierarchical structure of the antecedent clause is recovered in the ellipsis site. They examined whether the effect of the c-command relation underlying anaphora resolution (Chomsky 1981) is seen during the processing of an ellipsis site. Using a word-by-word self-paced moving window paradigm, they examined sentences like (7). (p. 450)

(7)

In (7) the gender of the subject NP (grandfather/grandmother) is manipulated. Furthermore, in (7a) the wh-phrase is an NP, but in (7b), the wh-phrase is a PP. Yoshida, Dickey, and Sturt argue that at the point of the wh-phrase (which story about himself), (7a) is locally ambiguous between sluicing and a non-elliptical wh-question. But (7b) is not ambiguous, because the wh-phrase is a PP, and there is no PP in the antecedent clause that can serve as the antecedent for the wh-phrase. Thus, it is possible that the parser predicts that (7a) is sluicing at the point of the wh-phrase, but in (7b), the parser does not consider sluicing as a possible structure. They further argue that if the parser picks the sluicing structure and if the structure of the antecedent is copied into the ellipsis site, the parser will try to link the reflexive to the potential antecedent. It has been reported that the parser tries to link a reflexive to its antecedent as soon as the reflexive is encountered (e.g., Sturt 2003). Furthermore, the parser will try to link the reflexive to a grammatically sanctioned antecedent, one which c-commands it and is located in the local domain of the reflexive (Sturt 2003). In (7a), the subject NP in the first clause is the only possible antecedent for the reflexive if the parser is building the structure of sluicing. This is because the structure of the antecedent clause is copied and reconstructed into the ellipsis site, and in the reconstructed structure, the subject NP ccommands the thematic position of the wh-phrase, so the reflexive is c-commanded by the antecedent NP. On the other hand, in (7b), a sluicing structure is not built and thus the subject NP in the first clause is not a candidate for the antecedent. Yoshida, Dickey, and Sturt found that the reflexive in (7a) is read significantly slower when the subject NP in the first clause and the reflexive mismatch in gender, the Gender Mismatch Effect (GMME: Sturt 2003; Kazanina et al. 2007). The GMME is not found in (7b). They interpret Page 7 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies this result as indicating that the parser does indeed copy the structure of the antecedent into the ellipsis site and tries to link the reflexive to the antecedent. The GMME reflects the failure of linking the reflexive to the antecedent due to the gender mismatch. In this section, I have reviewed several studies on the processing of ellipsis. All of these studies suggest the following: First, the parser accesses the information of the antecedent when the ellipsis site is recognized. Second, the parser copies the structure of the antecedent into the ellipsis site. In other words, the parser is using the strategy of copying to rapidly resolve the ellipsis site. These are the answers to the first two questions (what structure the parser builds and how the parser builds the structure). (p. 451)

18.3 When does the parser recognize ellipsis? The discussion so far has revealed that the parser accesses and copies the structure of the antecedent site into the ellipsis when the ellipsis site is recognized and processed. Now we turn to the next question, namely when does the parser recognize ellipsis? Looking at environments where ellipsis is licensed, it is clear that these environments are compatible with both elliptical and non-elliptical structures. In other words, in many cases, the potential position of ellipsis shows local ambiguity. An example is an embedded wh-question. A fragment of a sentence like (8) is locally ambiguous between ellipsis, i.e. sluicing, and a non-elliptical wh-question. (8)

(8) can be continued as a non-elliptical wh-construction like ‘I don’t know what he likes’, or it can be sluicing, in which no overt material follows. The same holds for VPE and other types of ellipsis constructions. This means that the processing of an ellipsis construction has an aspect of ambiguity resolution. Given this ambiguity, there are roughly two possible strategies that the parser can employ. One is what we can call the delay strategy. In the delay strategy, the parser does not decide immediately whether the wh-phrase is part of a sluicing structure or a non-elliptical wh-question structure. In other words, the delay strategy dictates that the parser waits on its decision until decisive bottom-up information becomes available. For example, if the wh-phrase in (8) is followed by a subject NP, it is clear that the wh-phrase is part of a non-elliptical wh-question structure. The other strategy can be called the incremental strategy. In the incremental strategy, the parser decides whether the wh-phrase is the part of a sluicing structure or a non-elliptical structure immediately upon encountering the wh-phrase, without waiting for the later information. The work by Yoshida and colleagues that I discussed previously directly investigates this point. Recall the sentences that they tested in (6) and (7). These sentences do not involve Page 8 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies ellipsis. However, at the point of the wh-phrase these sentences are ambiguous between sluicing and a non-elliptical wh-question. They found that in such environments effects of islands and effects of c-command and binding occur at the point of the wh-phrase, but not later. This suggests that the parser decides that these sentences involve clausal ellipsis upon encountering the wh-phrase in an ambiguous context without waiting for the later information. Furthermore, Yoshida, Dickey, and Sturt’s study showed that the effect of binding and c-command is not found in an unambiguous context where the wh-phrase is a PP and is not compatible with a sluicing structure. These studies suggest that the parser decides that the ambiguous wh-phrase is part of a sluicing context only when sluicing is possible. In other words, the parser recognizes the ellipsis site whenever the ellipsis is grammatically possible and does not wait for bottom-up information later in the sentence.10 Kaan and colleagues studied gapping constructions in an EEG experiment (Kaan, Wijnen, and Swaab 2004; Kaan et al. 2013) and reached a similar conclusion. Kaan, Wijnen, and Swaab (2004) asked whether semantic incompatibility between the gapped verb and the object NP is detected immediately when the object NP is processed. They tested the following pair of sentences: (p. 452)

(9)

The verb took is semantically compatible with the NP the hammer; however, the verb sanded is not. Thus, if the parser inserts the elided verb immediately when the NP-NP sequence (Bill the hammer) is processed, the semantic incompatibility of the gapped verb and the object NP should be detected, and the signal of such incompatibility, the N400 effect, should be seen. They indeed found the N400 effect, and thus they conclude that the gap is recognized immediately when the NP-NP sequence is processed. In line with these studies, Yoshida, Carlson, and Dickey (2013) asked when the ellipsis site in gapping is recognized. They point out that, like examples of sluicing, the environment where gapping is licensed can be compatible with a non-elliptical structure. For instance, the example in (10) involves the sequence of an NP the lawyer and a PP near the telephone pole. (10)

This NP-PP sequence can be analyzed as gapping in which the verb stood is elided. At the same time, this NP-PP sequence can also be analyzed as an NP that is modified by an adjunct PP. Yoshida, Carlson, and Dickey asked whether the parser decides that this NPPP sequence has the structure of gapping or the structure of an NP with a PP modifier.

Page 9 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies Using a word-by-word self-paced moving-window paradigm, they examined the following sentences: (11)

In (11a) the two clauses are connected by a coordination connective and, but in (11b), they are connected by a subordination connective while. It is well known that gapping is not possible in a subordinated clause (Johnson 2006 among many others). Thus, the NPPP sequence can potentially be analyzed as gapping in (11a), but not in (11b). After the NP-PP sequence, a VP follows. When the verb stood is processed, it is clear that the NPPP sequence must be an NP with a PP modifier. Yoshida, Carlson, and Dickey argue that if the parser employs the delay strategy, it can wait until the verb stood is encountered. On the other hand, if the parser employs the incremental strategy, it decides whether the NPPP sequence has the structure of gapping or the structure of an NP with a PP modifier at the point of the NP-PP sequence. If the parser chooses the NP structure, the presence of the verb at the later (p. 453) point is expected as the NP in that position is analyzed as the subject. The verb stood, therefore, should not surprise the parser. On the other hand, if the parser chooses the gapping structure, then the presence of the verb stood is not expected. This is because a coordinated or subordinated clause in sentences like (11) cannot have two verbs. As a result, the verb stood should surprise the parser if the gapping structure is chosen. However, the subordination condition (11b) does not allow a gapping structure, and so in (11b) the NP structure is the only possible structure. As a consequence, in (11b), the verb stood should not surprise the parser. Yoshida, Carlson, and Dickey found that the verb stood in (11a) was read significantly more slowly than in (11b). They thus conclude that the parser employs the incremental strategy and picks the gapping (i.e., ellipsis) structure immediately when the NP-PP sequence is encountered. In summary, these ambiguity studies all suggest that the parser prefers an ellipsis structure when the grammar allows one. In the sluicing studies, when sluicing is grammatically possible, the parser chooses the sluicing structure at the point of the wh-phrase, and in the gapping studies, when gapping is grammatically possible (in a coordination context, not in a subordination context), the parser chooses the gapping structure. Together, these results suggest that the parser employs the incremental strategy for ellipsis resolution.

Page 10 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies

18.4 Where does the parser find the antecedent of ellipsis? For the parser to successfully process sentences involving ellipsis, it has to search for an antecedent for the ellipsis and decide which constituent is the antecedent of the ellipsis. In simpler examples of ellipsis, finding the antecedent looks like a trivial matter. For example, in the sluicing example in (12), the only possible antecedent for the elided clause ([IP ø]) is the whole IP in the first clause. Thus, in such a case, there is an unambiguous antecedent, and determining which constituent is the antecedent of the ellipsis should be easy. (12)

A somewhat more complex example such as (13) reveals that the parser can need to decide which of multiple IPs is the antecedent of the ellipsis site. (13)

In principle, both of the IPs can be an appropriate antecedent for the elided clause, and thus (13) is compatible with both of the interpretations in (13a) and (13b). Such ambiguity in the interpretation of the ellipsis site can be found in other ellipsis constructions as well. This ambiguity arises clearly because both the long interpretation and the short interpretation are allowed for ellipsis (see Merchant 2001; Lasnik 2014 for related discussion). The problem of antecedent search exhibited by examples like (12) and (13) still does not look like a big problem, as the search space is bounded by the already-processed context. The situation becomes more complicated, however, in the context of backward sluicing (Giannakidou and Merchant 1998) in which the ellipsis site precedes the antecedent clause, as in (14). (p. 454)

(14)

In the processing of (14), at the point of the connective but it becomes clear that the first clause involves ellipsis of the clause following the wh-phrase. Once the parser recognizes the ellipsis site, it has to search for the antecedent of this IP. In an example like (14), it is the clause that immediately follows the connective. However, a potential antecedent for the elided IP can be embedded, and at the point of the connective but it is impossible for Page 11 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies the parser to decide where the antecedent IP is located. For example, in (15) there are multiple IPs that can serve as the antecedent for the elided IP, and the parser needs to decide which IP is the antecedent. (15)

Facing this situation, there are at least two possible antecedent search strategies. The first strategy is the delay strategy. In many sluicing cases, the wh-phrase has an antecedent, normally an indefinite NP. For example, in (15), the indefinite NP a book is the antecedent for the wh-phrase, which book. If the antecedent clause of sluicing normally contains an antecedent for the wh-phrase, then an indefinite NP can be a reliable cue to tell the parser that the IP that contains the indefinite NP is most likely the antecedent clause for the ellipsis clause. Thus the parser can rely on the presence of an indefinite phrase in identifying an antecedent clause. In this case, it waits for the indefinite NP to come into the input, and starts copying the materials in the antecedent IP into the ellipsis site only after recognizing the indefinite phrase. The delay strategy guarantees safe and accurate processing because it relies on the decisive bottom-up input in deciding which IP is the antecedent IP. The delay strategy, on the other hand, sacrifices rapid processing of ellipsis sentences. The second strategy is the incremental strategy. Looking at examples like (14) and (15), it is clear that immediately after but, an IP follows. Thus it is possible that once the parser recognizes the ellipsis site by means of the connective but, it starts copying the materials that follow but, without any delay. Using this strategy, the parser can rapidly process ellipsis sentences, but at the same time, it is at risk of making a wrong decision, i.e., it is always possible that the IP following but is not an appropriate antecedent for the ellipsis site. So far, there have been two studies on the processing of backward sluicing (Gullifer 2004; Yoshida et al. 2012a). Both of these studies support the incremental strategy. First, Gullifer (2004) asked whether backward sluicing and non-elliptical wh-questions are processed similarly or not. He compared the following type of sentences in a phraseby-phrase self-paced moving-window paradigm: (p. 455)

(16)

Page 12 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies

Gullifer reports that the reading time of the region fitted something to his boat in (16a) was slower than that of the region fitted to his boat in (16b).11 This result, Gullifer argues, suggests that the processing of backward sluicing is more difficult than the processing of non-elliptical wh-questions. He contends that this difference is attributed to the difference in how these two constructions are processed. In backward sluicing, unlike in non-elliptical wh-questions, the parser has to bind the indefinite NP to the wh-phrase and copy the antecedent IP into the ellipsis site. On the other hand, in the processing of nonelliptical wh-questions, these processes are not involved. Gullifer suggests that these extra steps are the source of the processing difficulty of backward sluicing sentences. Additionally, Gullifer argues that the parser copies the structure of the antecedent IP as soon as it can identify the complete IP. Thus he basically argues for the incremental strategy. Even though Gullifer’s study is compatible with the incremental strategy, his result is also compatible with the delay strategy illustrated above. It is possible that the parser starts copying the materials in the antecedent IP only after the indefinite NP is processed. This point, however, cannot be tested in Gullifer’s experiment as the indefinite NP and other material are in the same critical region of interest.12 Yoshida et al. (2012a) approached the problem of antecedent search from a slightly different angle. First, Yoshida et al. point out that both backward sluicing and nonelliptical wh-questions involve a long-distance dependency between the wh-phrase and the verb. For example, in (16), the wh-phrase what in both of the constructions must be linked to the verb fitted because the wh-phrase receives its thematic role from the verb. Furthermore, in the sentence-processing literature, it is well known that the parser tries to actively link the wh-phrase to the closest verb (Pickering and Barry 1991; Traxler and Pickering 1996; Phillips 2006, among others). Thus, if the verb closest to the wh-phrase is semantically incongruent with the wh-phrase, the verb gives rise to a reading time slowdown. Given that backward sluicing is a kind of wh-question, the processing of backward sluicing should involve a dependency formation process between the wh-phrase and the verb. Yoshida et al. argue that if the parser employs the delay strategy, this wh-verb dependency formation process should also be delayed. In the delay strategy, therefore, the semantic incongruence between the wh-phrase and the verb that is closest to the wh-phrase should not matter. On the other hand, if the parser employs the incremental strategy, the wh-verb dependency formation process should be launched immediately after the wh-phrase is encountered. Therefore, the semantic incongruence between the wh-phrase and the verb should surprise the parser, and cause a reading time slowdown at the incongruent verb. To test these two possibilities, (p. 456) Yoshida et al. examined the following type of sentences in a word-by-word self-paced reading paradigm:

Page 13 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies (17)

(17a) is the backward sluicing condition. (17b) is the non-elliptical wh-question condition. In both conditions, the type of wh-phrase is manipulated. Which book is not compatible with the verb notify, but which author is. In (17), if the parser employs the incremental strategy, it tries to link the wh-phrase to the verb. Thus, the verb should be read significantly slower when the verb and the wh-phrase are semantically incompatible than when they are semantically compatible. On the other hand, if the delay strategy is employed, such a reading time difference is not expected. Yoshida et al. found that the verb, when semantically incongruous, is read significantly slower both in the backward sluicing condition and in the wh-question condition. This result supports the incremental strategy. Furthermore, the observation that the parser is sensitive to the lexical content of the verb supports the copying strategy of ellipsis resolution. These two backward sluicing studies suggest the answer to the fourth question (Where does the parser find the antecedent of ellipsis?): the parser tries to find the antecedent IP immediately after the connective but, so the parser locates the antecedent IP in the closest grammatically possible position. Moreover, these findings indicate that the parser employs an incremental copying strategy in resolving ellipsis during online sentence processing, beginning the copying process before all the material to be copied is encountered.

18.5 Conclusions In this chapter, I have tried to survey some recent psycholinguistic studies on ellipsis. I have tried to show what the psycholinguistic study of ellipsis can tell us about the mechanism of online sentence processing. The studies I have discussed in this chapter have revealed the following properties of the mechanism of online sentence processing (which we can call parsing strategies). First, the parser builds the structure of ellipsis sites by copying the structure of the antecedent site. The copying strategy properly captures the parser’s sensitivity to the structural (and lexical) details of the antecedent during the processing of the ellipsis site. Second, the parser recognizes the ellipsis site immediately when it recognizes a structure in which ellipsis can be licensed. Thus, the parser processes the ellipsis site incrementally without delay. Finally, we have seen that the parser looks for the antecedent in the position closest to the ellipsis site. The backward sluicing studies suggest that the parser decides that the clause that is closest to the ellipsis site is the antecedent of the ellipsis site without referring to the information in the later portion of the sentence. In general, these findings support the claim that the parsing strategies that the human parser employs achieve incremental, rapid, and Page 14 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies grammatically detailed structure building. Thus, the study of online processing of ellipsis constructions points to a conclusion that is in line (p. 457) with those reached in the sentence-processing studies in other domains. For example, processing studies of cataphoric constructions have shown that the parser tries to link a pronoun to its antecedent, and the process of linking the pronoun to the antecedent obeys grammatical constraints such as Binding Conditions (Kazanina et al. 2007; Yoshida et al. 2014). Similarly, studies of filler–gap dependencies have revealed that the parser tries to link wh-fillers to the closest gap site as long as the gap is not embedded inside an island (Stowe 1986; Traxler and Pickering 1996; Phillips 2006, among others). The study of the online processing of ellipsis can provide further support for the position that the parser builds grammatically detailed structure rapidly and accurately (see Phillips 2006; Phillips and Lewis 2013 for a detailed discussion on this point). Finally, I would like to discuss what remains in the study of ellipsis processing. The study of ellipsis processing has been developed mostly based on research of clausal ellipses, such as sluicing (see Harris 2016; Harris and Carlson 2016, which investigate a different type of clausal ellipsis construction—the let alone construction). Therefore, there are not many studies on other types of ellipsis constructions. For example, the processing of NPellipsis has not been investigated. One of the open questions is whether the same mechanism is responsible for the online processing of clausal ellipsis constructions and other ellipsis constructions. Some studies on sluicing (Yoshida, Dickey, and Sturt 2013a; Yoshida, Lee, and Dickey 2013b) have suggested that ellipsis is preferred when ellipsis and non-ellipsis are both possible. Does the same hold true in other ellipsis constructions? Why does the parser prefer to pursue ellipsis when other structures are possible? Pursuing these questions can shed new light on the study of ellipsis processing.

Acknowledgements I am most grateful to Michael Frazier for his valuable discussion of and comments on an earlier version of this work. This work has been supported in part by NSF grant BCS-1323245 awarded to Masaya Yoshida. All the remaining errors are, of course, my own.

Notes: (1) This chapter is crucially NOT about what the psycholinguistic study of ellipsis can tell us about grammatical analyses of ellipsis. Given this purpose of this chapter, I will try to avoid going into detailed discussion on issues of the syntax and semantics of ellipsis throughout this chapter, and, as such, citations on these issues will be limited to representative works. It is important to note that, many times, the interpretation of experimental results depends on the theory of ellipsis adopted. However, given the limited space, I will not be

Page 15 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies able to discuss the details of alternative interpretations of each experimental study that I will touch on. (2) Phillips and Parker (2014) provide an extensive survey of the psycholinguistic study of ellipsis. What this chapter covers overlaps with some of Phillips and Parker (2014). This chapter, however, focuses more on issues of how ellipsis constructions are processed. (3) As far as I can tell, the vast majority of psycholinguistic studies of ellipsis have focused on the first two questions, and the latter two questions have not been asked until recently. (4) Note that McKoon et al. (1994) argue against the position that anaphors show reactivation of the antecedents (see also Nicol, Fodor, and Swinney 1994 for related discussion). (5) A recent structural priming study also supports this point (Xiang et al. 2014; but see Cai et al. 2013 who report that VPE in Chinese does not show structural priming effects and thus conclude that VPE is not associated with syntactic structures). (6) Snider and Runner (2014) report similar results using a visual world eye-tracking paradigm (for details, see Phillips and Parker 2014 and Snider and Runner 2014). (7) Frazier and Clifton (2000) suggest that the process of copying is cost-free in the sense that the size or complexity of the antecedent does not impact the processing time of the ellipsis site. Martin and McElree, on the other hand, argue that such a result points to a content-addressable memory retrieval mechanism (Martin and McElree 2008, 2009, 2011), which is not affected by the syntactic structure of the retrieved materials. However, as I will discuss, the existence of antecedent complexity effects is controversial as some studies do show such an antecedent complexity effect (e.g., Frazier and Clifton 2005 in their study on sluicing). Note that recent cue-based memory retrieval studies have shown that the parser utilizes structural cues as well as non-structural cues (Dillon et al. 2013; Kush et al. 2015; see also the discussion in Jäger et al. 2015). Thus, an antecedent complexity effect can be predicted by cue-based memory retrieval theories as well as by copying theories. I will avoid going into the details of the debate, but see Frazier and Clifton (2005), Martin and McElree (2008, 2009, 2011), and Phillips and Parker (2014). (8) Frazier and Clifton (2005) specifically assume that wh-movement out of a conjunct is possible when a wh-phrase is moved out of the nearest conjunct, but not when it is moved out of the more distant conjunct, based on the observation in (i). ((i))

Page 16 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Parsing Strategies They further assume that the nearest conjunct can be determined in terms of the surface position of the conjunct. Thus in the case of sluicing in (2), the second conjunct is the nearest conjunct. (9) There is a long-standing debate on whether gapping involves ellipsis or not. I leave this issue open (see Carlson et al. 2005 and Johnson 2009, this volume for detailed discussion). (10) In Yoshida and colleagues’ studies, it is not clear why the ellipsis structure is preferred over a non-ellipsis structure. This point requires further investigation. (11) Note, Gullifer’s (2004) experimental paradigm is a little bit more complicated, and here I am simplifying his study so that I can highlight its main finding. For the details and the full paradigm of this experiment, see Gullifer (2004). (12) Note that Gullifer’s main interest is whether the processing of backward sluicing is similar to that of non-elliptical wh-questions. Thus, the problem of antecedent search is not the main point for him.

Masaya Yoshida

Masaya Yoshida is an Associate Professor in the Department of Linguistics at Northwestern University. His research interests are online sentence processing and syntax with special focus on the syntax and processing of ellipsis. He has published articles in a number of linguistics and psycholinguistics journals such as Linguistic Inquiry, Natural Language and Linguistic Theory, and Journal of Language, Cognition, and Neuroscience.

Page 17 of 17

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

Oxford Handbooks Online Codeswitching   Kay González-Vilbazo and Sergio E. Ramos The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.21

Abstract and Keywords While research on codeswitching and on ellipsis has flourished in the last two decades, ellipsis in codeswitching has not received as much attention. In this chapter, we argue that using ellipsis to study codeswitching is insightful on two fronts: first, codeswitching data offer a new proving ground against which to test extant theories of ellipsis. Second, ellipsis data ought to bear on theories of codeswitching, which have for the most part remained silent on it. The chapter first offers an overview of the literature on codeswitching. Then, we cover the work on ellipsis in codeswitching done so far. Our claim is that codeswitching points largely towards a structural condition on identity in ellipsis. Conversely, we find that the ellipsis data from codeswitching can be accounted for within a constraint-free approach to codeswitching, without having to further complicate the system. Keywords: codeswitching, ellipsis, identity, sluicing, VP-ellipsis, bilinguals

Kay Gonzaález-Vilbazo and Sergio E. Ramos

19.1 Introduction MOST people in the world are bilingual (Grosjean 1982; Romaine 1995). Very often, when two individuals of a bilingual1 speech community speak to each other, they switch from one language to another effortlessly back and forth. This phenomenon is known as codeswitching or codemixing (Muysken 1995; Edwards 2004). Broadly speaking, codeswitching refers to the simultaneous use of two or more languages by fluent bilinguals within a discourse (Edwards 2004). A Taiwanese/Spanish example of codeswitching can be found in (1):2 (1)

Page 1 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

Codeswitching is to be differentiated from other types of language contact phenomena, such as borrowing, the integration (morphological, syntactic, and phonological) of loanwords into a recipient language (see Poplack 1980 and Romaine 1995 for further discussion). Further, codeswitching can be categorized as inter-sentential, i.e. involving switches between clauses, or intra-sentential, i.e. switches within a single clause. Competent codeswitchers have the ability to judge the acceptability of any codeswitched utterance, in a manner similar to how monolingual speakers can discriminate between constructions that are part of their I-language and those that are not. Put in another way, codeswitching is not random, but reflects the linguistic competence of codeswitchers. This in turn has motivated much theoretical work on codeswitching, for as Woolford (1983) noted, linguistic theory should account for codeswitching as just another expression of the faculty of language, (p. 459) while at the same time codeswitching can provide a rich new source of data against which to test existing theories. Indeed, several models have been proposed to explain the constraints on codeswitching and account for much data, particularly on intra-sentential codeswitching, given how linguistic theory is mostly concerned with the grammatical relations within a clause. One phenomenon that has received much attention in linguistic theory is ellipsis. To this day, however, very little work has been done on elliptical constructions under codeswitching. This is unfortunate on two counts. First, because codeswitching can and should inform our theories on ellipsis, just in the way Woolford argued. Second, because ellipsis data may contribute to a better understanding of the ways in which codeswitching is possible, even at the syntactic level prior to any phonology. In this chapter we review the work on ellipsis under codeswitching to date to show that (i) ellipsis under codeswitching exhibits several of the connectivity effects well known in the ellipsis literature since Ross (1969b), which we argue is further evidence not only for a deletion theory of ellipsis but for a hybrid identity requirement (see Chung 2013), and (ii) these data can naturally be accommodated in a constraint-free approach to codeswitching, without the need to posit codeswitching-specific constraints. The chapter is organized as follows: we first provide a historical overview of the literature on codeswitching, focusing on some of the most important theories proposed to date. We then examine the studies that have been pursued on ellipsis under codeswitching, and we illustrate how a constraint-free approach can account for the data. Finally, we conclude with a brief discussion on possible future directions of research. Before we review the particular theories, we believe a note on the methodology of codeswitching research is warranted. Many methodological challenges are unique or exacerbated when studying codeswitching. Some of these challenges include what to count as codeswitching, how to select participants (i.e. what constitutes a bilingual Page 2 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching codeswitcher), the types of tasks to perform, how to design the stimuli, whether participants are in what is known as the bilingual mode, the type of data to collect among others (see Grosjean 1998; MacSwan 1999, 2012, 2014; Poplack 2001; and MacSwan and McAlister 2010 for relevant discussion on these issues, and González-Vilbazo et al. 2013 for potential solutions to some of these challenges). We take the position that while the issues are real, they are not insuperable, and the potential rewards from studying codeswitching outweigh the difficulties.

19.2 Theories of codeswitching The study of codeswitching has progressed from a primary focus on social and pragmatic aspects to a much more comprehensive view, where theoretical attempts to understand its underlying mechanisms inch closer to, and can borrow from, more general considerations on linguistic theory. This was highly implausible fifty years ago. In fact, early discussion about codeswitching suggested no restrictions on where a switch might occur. As Pfaff (1979) noted, the idea that the alternation between languages was not rule-governed, suggested by researchers such as Labov (1971) and Lance (1975), endured well into the 1970s. It was nevertheless during this decade that important fieldwork on codeswitching (Timm 1975; Gumperz 1976; Wentz and McClure 1976; McClure 1977, among others) (p. 460) demonstrated that codeswitching was not random but rather constrained, both at the sociolinguistic level and also at the points at which a switch could occur within an utterance. An example of the extra-syntactic significance of codeswitching can be seen in Gumperz (1976), which examined how in some Spanish/ English communities, Spanish had become the language for affective expression, while English was used to communicate in a more objective or impersonal manner. And although works like these were mostly focused on the sociolinguistic role of codeswitching, as MacSwan (2012) observes, they nevertheless made important descriptive observations about some types of switches which bilinguals rejected. These restrictions were language-pair-specific, e.g. Timm (1975) reports that Spanish/English switches between a personal pronoun and a finite verb are strongly restricted (2). (2)

3

Page 3 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching Similarly, Wentz and McClure (1976) and McClure (1977) investigated the use of codeswitching by children in a Spanish/English bilingual community and observed that codeswitching could function as an index of identity and could allow bilinguals to use different languages for different situations or to mark different discourse styles. These authors were also the first ones, to our knowledge, to note that codeswitched elliptical constructions could also be part of the grammatical inventory of bilinguals, as in some of the answers reported in the following dialogues: (3)

(4)

Note that while in (3) consultants could respond in English in two different ways to the question in Spanish posed by speaker A, Wentz and McClure reported that in (4) bilingual children responded in three different ways: I am, I’m hungry and I do. But adult bilinguals consulted later only considered two of those responses acceptable, namely I am and I’m hungry. The response I have, suggested to adults by the researchers, was not accepted by adults and not produced by any of the children. We return to these findings (p. 461)

in more detail in the following section. What is important to note at this point, is that constraints on the possible responses for these elliptical constructions under codeswitching exist, dispelling the idea that codeswitching is random. Subsequent accounts have tried to capture the codeswitching data in a more systematic and generalizable way, suggesting rules not as mere descriptions of the data but rather as being fundamental parts of a more general theory of codeswitching (cf. MacSwan 2012 for discussion). Some of these theories were based on word order and were very much specific to codeswitching. Poplack (1980), for instance, proposed that codeswitches must respect the word order of both languages, as well as a general restriction against codeswitching below the word level, i.e. no intra-word codeswitching. In this view, codeswitching seems to be understood as a linear phenomenon as in (5) (where > represents a sequencing operator and / a switch between languages): (5)

Page 4 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

While this theory set out a more ambitious goal, that of explaining codeswitching in more general terms rather than just describing the data, several counterexamples in the literature exist, where neither constraint is violated and the utterance is nevertheless unacceptable. Consider the Spanish/English examples in (6): (6)

Both sentences in (6) only involve a switch between Aux and V, and the word order of both languages is respected, and yet they are both rejected, contra Poplack’s constraints. Further, Poplack’s constraints are specific to codeswitching and thus involve essentially a third grammar, i.e. rules that are not part of any of the grammars of the languages involved in the codeswitch. After Poplack’s seminal work, newer approaches incorporated hierarchical relations into their descriptive framework. These theories rely on concepts that at the time were considered to be part of UG, such as government or functional head selection. Di Sciullo, Muysken, and Singh (1986) proposed an anti-government constraint for intra-sentential codeswitching, so that complement DPs must be in the same language as their governors, be they verbs or prepositions. Conversely, Belazi, Rubin, and Toribio (1994) appeal to feature checking theory to propose that functional heads need to check, among other features, a language feature against their complements. These approaches greatly contributed to the progress of codeswitching research. Despite their theoretical appeal, however, (p. 462) these theories made the wrong predictions. Two naturalistic examples from a German/Spanish codeswitching corpus are given in (7) and (8) (González-Vilbazo 2005): (7)

(8)

Page 5 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

The German/Spanish example in (7) would violate Di Sciullo et al.’s anti-government constraint, since the verb and its complement are in different languages, and yet the utterance is acceptable. On the other hand, a theory like Belazi et al.’s would have difficulty accounting for an example such as (8). This sentence contains a switch between a functional head (Aux) and its complement VP, and would thus be predicted by Belazi et al. to be ungrammatical, contrary to fact. Another line of approach to codeswitching proposes an asymmetry between the languages involved. These theories (Joshi 1985, Klavans 1985, Myers-Scotton 1993, Myers-Scotton and Jake 2001, and subsequent work) rely on the idea that a codeswitched utterance reflects an asymmetry between a matrix language ML, from which the template for the utterance is derived, and an embedded language EL, which only inserts lexical items into the utterance in the ML structure. The model known as the Matrix Language Framework, MLF (see Myers-Scotton 1993 and Jake, Myers-Scotton, and Gross 2002) is perhaps the best known of these approaches. The MLF model is purported to be a model of competence, and it aims to account specifically for intra-sentential codeswitching, seen as the only instance of true codeswitching. This constraint-based model aims to find the conditions limiting when codeswitching can occur. While there have been different variants of the theory (see Myers-Scotton and Jake 2017 for one of the latest iterations), it has been a constant that these conditions can be reduced to a typology of morphemes: content morphemes and system morphemes; and within system morphemes, a subtype known as outsider late system morphemes (case-marking, verb agreement) determine the ML and the grammatical properties of the clause.4 The elements from the EL would be inserted in the structure provided by the ML. There is much discussion of the model and some empirical evidence against it in the literature (for an extended discussion on the MLF, see Jake, Myers-Scotton, and Gross 2002, 2005; MacSwan 2005a, 2005b; and González-Vilbazo and López 2011). It nonetheless remains difficult to conceive of terms such as ML or EL islands—typically adjunct phrases well-formed by EL grammar—as anything but codeswitching-specific. Indeed, one key question that constraint-based proposals raise is whether these constraints can be derived from more basic grammatical principles. (p. 463) We should favor simpler theories; ideally, already established theories for monolingual data would be extended to also cover the facts of codeswitching. If that is not possible, codeswitching data force us to revise current theories in order to account both for monolingual and bilingual data. Most recent generative codeswitching theories have aimed at explaining codeswitching without appealing to any third grammar factor.5 For instance, MacSwan (1999, 2012, 2014) and González-Vilbazo and López (2011, 2012) made two such proposals with no codeswitching-specific rules. Following the minimalist framework as laid out in Chomsky (1995, 2001, et seq.), these authors argue that only the

Page 6 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching interface conditions have to be met by the derivation of a codeswitched utterance in order for it to be grammatical. It is worth emphasizing, especially when studying bilingualism, that the term ‘language’ is not a primitive of linguistic theory, but a social, political, and/or cultural construct. Each speaker has individual competence. To the extent that competences overlap, this overlap is often referred to as a ‘language’. Given this, what does it mean to be bilingual, i.e., to speak two (or more) languages? What makes the difference between a monolingual and a bilingual competence? A possible answer would be to say that a bilingual has two grammars. This is still too vague. Do we mean syntax, morphology, semantics, and/or phonology? Rules, operations, lexica? Whatever is a universal property of human languages cannot make the difference between mono- and bilinguals. We have to look for the difference in those areas where we find variation across speakers. Current advances in generative linguistic theory take it that there is only one set of universal syntactic, morphological, and semantic operations, invariant across languages. Consequently, these operations/rules cannot constitute the difference, as both monolingual and bilingual speakers will necessarily have this set of operations as part of their linguistic system. There is, however, consensus among phonologists that phonological systems (e.g., constraint rankings) vary across languages. This would entail that bilinguals would have at least two phonological systems, whereas monolinguals only have one. Finally, the lexicon is of course another source of variation across languages. It is not surprising that most theories of codeswitching assume that bilinguals have two lexica. But this is not a logical necessity. It is possible to assume that bilinguals only have one lexicon (regardless of whether one assumes a DM-style lexicon or not) with lexical items (vocabulary items) from both ‘languages’. To the linguistic system it would make no difference whether the lexical items merged into a derivation are from what we call language X or Y. Summarizing, it looks like bilinguals have two phonological systems and possibly two lexica. Everything else is similar to the monolingual linguistic system.6 If we take these ideas on bilingualism seriously, one could say there are very few truly monolingual speakers in the world, as most speakers will know at least two linguistic varieties (either registers, dialects, or languages). If we stick to the idea that a speaker’s competence is individual, then there is no difference in principle between language, dialect, or register for the linguistic system. This is not to say that we cannot distinguish those concepts and they have certainly proven to be fruitful in sociolinguistics, but they are extralinguistic concepts. For the linguistic system there is no difference between these concepts. And everybody is competent in different linguistic varieties (language, dialects, registers). That means that almost everybody has two phonologies (to the extent that the dialects or registers they speak have different phonologies) and they certainly have two or more lexica (or an enlarged single lexicon). If this is true, these speakers are bilinguals. Thus, true monolingualism would represent the marked, not the unmarked, case. (p. 464)

Page 7 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching Based on these considerations and the research mentioned above, we take it that codeswitching is a natural instantiation of the faculty of language. That is, it is binary and hierarchically structured, and it undergoes the same universal linguistic operations (Merge, Agree) as monolingual derivations. In the generation of a codeswitched sentence, lexical items are drawn from two or more lexica,7 then merged into the derivation, and finally, the resulting structure is transferred to the interfaces, as represented in (9). (9)

This is just the same as for a monolingual structure; the derivation will crash if the conditions of the interfaces are not met, be it a monolingual or a bilingual structure. (p. 465)

19.3 What codeswitching tells us about ellipsis Ellipsis is a widely studied phenomenon in linguistics, as this volume can attest. Broadly, ellipsis refers to structures in which a part of a sentence appears to be missing. (10)

Page 8 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

Ellipsis can apply to clausal structures as in sluicing (10a), which is the deletion of a full interrogative clause with the exception of the wh-phrase; it can apply to verbal phrases (10b) as well as inside the nominal domain (10c); and it can also result in just one constituent conveying propositional information, such as in fragment answers (10d). A thorough review of the literature on ellipsis is well beyond the scope of this chapter, but we refer readers to Merchant (2011), van Craenenbroeck and Merchant (2013), and Merchant (this volume) for recent overviews of some of the questions raised and tools used for the study of ellipsis. Chung, Ladusaw, and McCloskey (2011) distinguish three different approaches for the study of ellipsis: a pragmatic approach, a copying approach, and a deletion approach. The pragmatic approach would propose that the ellipsis lacks any unpronounced structure and its interpretation depends only on pragmatic inference; the copying approach posits that the ellipsis site is empty at the time of the derivation but its content can be recovered by means of copying an already constructed structure at the level of interpretation. Finally, in the deletion approach, at some point of the derivation the ellipsis has full syntactic structure, which is then deleted or made inaccessible to PF under certain conditions. Currently, the copying and deletion approaches are favored. In a copy theory of ellipsis, or non-deletion theory, the structure of the ellipsis site amounts to a null element (or more than one) at the level of syntax, which can then either be given a fuller structure so it can be interpreted (as in Chung, Ladusaw, and McCloskey 1995), or resolved post-syntactically by an interpretation algorithm, similar to anaphora (Hardt 1993). In deletion approaches, the ellipsis contains full syntactic structure, as a corresponding non-elided sentence would, and the structure is then subject to a deletion operation, either at Spell-Out or at PF. Representative examples of deletion theories are Ross (1969b), Sag (1976a), and Merchant (2001), among others. Note that, both for approaches positing a deletion mechanism for ellipsis and for those adopting a copy system, one of the main questions is what type of relation of identity between antecedent and ellipsis must hold for the unpronounced material to be (p. 466) recoverable. Put another way, the question is whether the relation between antecedent and ellipsis need only be a semantic one (as proposed in Merchant 2001), a syntactic one (as in Saab 2009, or Thoms 2014) or whether a hybrid theory requiring semantic as well as other types of identity is necessary (as in Chung 2005, 2013; Merchant 2013d; González-Vilbazo and Ramos 2014). Codeswitching can be illuminating when answering this question. To our knowledge, there are only a few studies that look at ellipsis in codeswitching: Wentz and McClure (1976), Nee (2012), González-Vilbazo and Ramos (2014), and Merchant (2015b). We can say that all studies so far have provided evidence against both the pragmatic approach and the LF-copying approach and support for a deletion approach to ellipsis. Furthermore, within the deletion approach, some of the most recent literature provides evidence that the identity between the ellipsis and its Page 9 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching antecedent must be morphosyntactic as well as semantic. Below, we discuss each study and how it relates to different approaches to ellipsis. Wentz and McClure (1976) first reported elliptical constructions under codeswitching in a Spanish/English bilingual community. The data stem from interviews with bilingual children, and their responses were later presented to adult bilinguals for verification. Some of their findings are shown below: (11)

(12)

In (11) we see that the question ‘who’s hungry?’ in Spanish is formed with the equivalent of have and not with a copular verb. I am was accepted both by children and adults. It seems to be the elliptical result of the English sentence I am hungry, both of which (I am/ I’m hungry) were deemed acceptable by children and adults. I do, only accepted by children, is unlikely to be the result of an unelided English counterpart, since I do be hungry or I do hunger would be a heavily marked construction. The answer I have was not produced by any of the interviewees and was deemed unacceptable by adults. Similarly, in (12), children offered two very similar Spanish answers to the English question, the only difference being the case of the DP. Adults only accepted the answer bearing overt accusative case. From these judgments, Wentz and McClure concluded that children may take longer to develop the ability to codeswitch, thus explaining the difference between children and adult answers, and thus they deemed adult judgments to be more reliable. Furthermore, the authors concluded that the fragment answer must obey the grammar of its own language, regardless of the language of the question. Unfortunately, they provided no further analysis beyond this generalization. The authors further suggested that an account such as Morgan’s (1973) might be on the right track. Morgan (1973) proposes that the missing part of a clause in a fragment must be present at one point in the derivation, after which it is deleted. It is worth noting, however, that the bilingual children interviewed seemed to more readily (p. 467) provide a wider range of possible answers, whereas the adult bilinguals tended to narrow down the array of acceptable answers when compared to the bilingual children interviewed.

Page 10 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching For over thirty years, no further investigation of ellipsis in codeswitching was carried out. It was only in 2014 that González-Vilbazo and Ramos reconsidered the topic following a suggestion by Jason Merchant. González-Vilbazo and Ramos (2014) conducted an experiment to test connectivity effects on German/Spanish codeswitching, specifically case mismatch resolution. In order to test these connectivity effects they used codeswitched sluicing data. Both Spanish and German show case effects in sluicing: (13)

(14)

In some instances, a Spanish verb (13) assigns a different case to its object than the equivalent German verb does (14). Take, for instance, the verb threaten in these languages. German drohen assigns dative to its complement while Spanish amenazar assigns accusative. Now, consider a codeswitched utterance with a Spanish antecedent clause and a full non-elided embedded clause, as in (15)–(16). In this case, a German wh-phrase will bear the case that the verb in the embedded clause assigns, dative for German (15) and accusative for Spanish (16), regardless of the verb of the antecedent clause. (15)

(16)

Page 11 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching If the sentence is sluiced, however, the wh-phrase can only bear accusative, which in this instance is the case the Spanish antecedent verb assigns: (17)

This result strongly suggests that the verb elided must be Spanish amenazar. This is puzzling, as (15) shows that the non-sluiced embedded clause can in principle contain the German verb drohen. Why then is it not possible to just delete in (15)? This would result in a sentence like (17) with the wh-phrase bearing dative case: (p. 468)

(18)

If a semantics-only theory of ellipsis (as in Merchant 2001) were correct, a sentence like (18) should be grammatical, as the antecedent and the ellipsis are semantically identical. However, (18) is ungrammatical. González-Vilbazo and Ramos argue that a semantics-only theory of ellipsis is insufficient to capture the data, for (15) and (16) are semantically equivalent in terms of truth conditions. Instead, the authors claim that alongside a semantic identity condition, at least some morphosyntactic identity has to hold. More precisely, for these data, at least the case assigner of the wh-phrase has to be identical to the case assigner of its correlate in the antecedent clause. This analysis points in the same direction as other hybrid theories such as Chung’s (2005) ban on eliding words new to the derivation, Chung’s (2013) case and argument structure constraints, or Merchant’s (2013d) structure constraint. In Chung (2005), besides the semantic conditions, it is proposed that words which have no overt correlate in the antecedent clause cannot be left inside the ellipsis site. This proposal aims to explain, among other things, cases of sprouting: instances of sluicing where the wh-remnant has no overt correlate in the antecedent clause. In English, for instance, P-stranding, otherwise allowed (19), is prohibited in these constructions (20). (19)

(20)

Page 12 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching According to Chung (2005), the ungrammaticality of (20b) is explained by the fact that the preposition of does not have an overt correlate in the antecedent, and is thus unable to stay inside the ellipsis. Chung (2013) and Merchant (2013d), for their part, argue that no argument structure differences are possible between the antecedent and the ellipsis site. For Chung (2013), semantic constraints do not allow the ellipsis to have a different argument structure than its antecedent clause. As for case, if the remnant of the sluice is a DP, its licensing head within the ellipsis must be identical to the licensing head of its correlate in the antecedent clause. Evidence for these restrictions is provided by some intransitive verbs in Chamorro that have derived transitive counterparts. Consider (21): (21)

(p. 469)

The verb ekgu’ is an intransitive predicate which can introduce an internal

argument as a PP, though this argument may be omitted. Its counterpart ekgu’i is semantically similar yet it introduces its internal argument as a DP. Now, in (21a) we have an unsluiced example of an embedded question with the transitive ekgu’i preceded by an antecedent with intransitive ekgu’. (21b) shows how a sluice in this situation results in ungrammaticality, even though a semantic condition should be satisfied. Chung concludes that the ungrammaticality must lie in the argument structure difference between the two clauses. In Merchant (2013d), this structural constraint accounts for the difference in grammaticality that voice mismatches produce in VP-ellipsis and sluicing. In VP-ellipsis (22a), unlike in sluicing (22b), voice mismatches are allowed. (22)

The argument here is that in VP-ellipsis the ellipsis site is smaller than in sluicing. In VPellipsis (22a), a VP, the complement of the head determining voice, is elided. In sluicing, a full clause, including the voice head, is elided. Because in sluicing the voice head is elided, no mismatch with respect to voice between antecedent and ellipsis is possible. In a similar vein, Nee (2012) conducted research on Spanish/Zapotec codeswitching in Teotitlán del Valle, México. This language pair is interesting for the study of sluicing Page 13 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching because some Zapotec transitive verbs (23) have a Spanish equivalent that selects a PP as its complement (24). (23)

(24)

In (23) the Zapotec verb gunien selects a DP for its complement whereas the corresponding Spanish verb habló selects a PP as its complement (24). Nee takes advantage of this difference between Zapotec and Spanish in constructing codeswitched sluicing utterances. (25)

(26)

(27)

Nee analyzes example (25) by adopting Chung’s (2005) lexical requirement. According to her, the embedded clause contains the Spanish verb habló and its PP complement tu cun. Notice that the Spanish verb hablar, which is in the antecedent, selects for a PP complement, which in this case is satisfied by the Zapotec PP cun tu.8 The wh-phrase must then move up to Spec CP along with its preposition. The TP is then elided. In this case, Merchant’s semantic condition, Chung’s lexical requirement, and González-Vilbazo and Ramos’s morphosyntactic condition are all observed. On the other (p. 470)

Page 14 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching hand, Nee takes (26) to be ungrammatical because neither Zapotec nor Spanish allows for P-stranding. In (27), the antecedent clause contains the Zapotec transitive verb, and Nee argues, assuming Chung’s lexical requirement, the wh-phrase must be bare. This is so because the verb inside the ellipsis must be the Zapotec gunien, and thus selects a DP. This derivation would also satisfy all semantic, lexical, and morphosyntactic conditions mentioned above. As we can see from these examples, Nee also advocates for a deletion approach to sluicing. Moreover, she embraces the view that identity between antecedent and ellipsis goes beyond semantics, and must also include some sort of lexical identity. Merchant’s recent paper on codeswitching and ellipsis (2015b) looks into verb phrase ellipsis in Greek/English codeswitching. This paper is consistent with the studies described above and finds further evidence not just for structural identity, but also evidence that in some instances the elided structure could not be pronounced. The data in this study are Greek/English codeswitching dialogues uttered by balanced bilingual children and contain a Greek antecedent and a verb phrase ellipsis response in English. An example is shown below: (28)

In (28), the elided VP in the response appears to be in English, for two reasons. First, the corresponding unelided sentence would be fully grammatical (29a). Second, a Greek VP at this point produces an ungrammatical result (29b,c). (p. 471)

(29)

In (29b), Merchant argues, the Greek verb píra cannot agree with the subject, since the auxiliary did effectively blocks agreement. The Greek pern (29c), for its part, is a bare stem form that cannot remain unbound in Greek. However, if (29a) were the underlying representation for B’s reply in (28), this result would contradict González-Vilbazo and Page 15 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching Ramos’s condition as well as Chung’s (2005) lexical requirement and Chung’s (2013) argument structure and case condition. Chung (2005) would expect the ellipsis to be in Greek just as the antecedent, whereas Chung (2013) and González-Vilbazo and Ramos (2014) would expect at least part of the morphosyntax of the ellipsis to match the Greek antecedent. Other Greek/English data from Merchant (2015b) suggest that (29a) cannot be the elided structure in the ellipsis in (28). (30)

The example in (30) is similar to one of the answers given in the English/Spanish example in (11) reported by Wentz and McClure, although one adult speaker later rejected it. The unmarked way of expressing hunger in Greek is via an intransitive verb (30a), unlike in English where an adjective is used (I am hungry). In this case, if the underlying representation of the response in (30b) were either *I do hunger or *I do (be) hungry, neither of these options would correspond to the common way of responding to this question in English: I am hungry. In fact, it seems to correspond to a verb phrase ellipsis response (I do) which suggests a non-copula verb is inside the ellipsis. Similar to the previous example, the codeswitched non-elided counterpart to (30b) would also be unacceptable. (31)

Merchant offers a detailed analysis within the Distributed Morphology framework in which the verbal complex cannot enter into a local relation with T, and as a consequence the verb is not inflected. This makes the root unpronounceable, as Greek lacks infinitives. Again, these data suggest that a semantic condition alone is not sufficient and a structural constraint is necessary, echoing González-Vilbazo and Ramos’s proposal. In another example, the impossibility of an English VP inside the ellipsis with a Greek antecedent becomes clearer. (32)

Page 16 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

Let us focus on the instance of ellipsis in (32d). As Merchant notes, in Greek the existential predicate is expressed via an impersonal éxo ‘have’, which takes an accusative (p. 472)

object. This can be seen in the remark about the temperature in (32c). Just like in the previous examples, the elliptical response in (32d) is acceptable but it is not initially clear what the ellipsis must contain. Crucially, no unelided counterpart to this response, codeswitched or not, is grammatical.9 (33)

As seen earlier in (30) and (31), the ellipsis cannot be in English if it has a Greek antecedent. This explains (33a–d). However, it seems it cannot be in Greek either as (33e) shows. Similar to (31), Merchant proposes the following structure for the attested ellipsis in (32d): (34)

Page 17 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

This ellipsis contains the roots √EX and √ÐROSJA that correspond to the Greek antecedent. That is the only way to satisfy the structural condition on ellipsis, i.e., to have at least the same relevant morphosyntactic features in the antecedent and the ellipsis. (p. 473)

However, these roots are unpronounceable, as mentioned earlier. Merchant shows that the Greek/English codeswitching data offer more evidence for an irreducible syntactic condition on ellipsis. He further shows that this condition is so strong that it must hold even if the structure within the ellipsis would be otherwise unrealizable. This would thus be a new instance of elliptical repair,10 i.e., a situation in which the English answer to the Greek question as given in (32d) is only possible with an elliptical construction. In this section, we have seen that codeswitching can unveil some of the inner mechanisms of ellipsis. It has been known since at least Chung (2005) that a semanticsonly condition for ellipsis still faces challenges (specifically for sluicing). However, as shown by Nee (2012), González-Vilbazo and Ramos (2014), and Merchant (2014b), codeswitching offers some of the strongest evidence yet that a theory of ellipsis requires some morphosyntactic condition as well, beyond whatever semantic conditions must hold. Further, codeswitching (Merchant 2014b) also offers evidence of elliptical repair in previously unknown contexts.

19.4 What ellipsis tells us about codeswitching One goal of this chapter is to show how ellipsis informs our understanding of codeswitching. Indeed, the research done so far already suggests surprising insights. First, it is possible to identify the language of the material within the ellipsis. This is shown by the case of the German wh-remnant in the data from González-Vilbazo and Ramos (2014), as well as the fact that ellipsis is what seems to cause the ungrammaticality of (17), repeated as (37b). Page 18 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching (35)

(36)

(37)

(p. 474)

Given the remnant’s case, we can infer that the verb in the ellipsis has to be

Spanish; as we have seen, the Spanish verb amenazar ‘threaten’ assigns accusative, thus wen, whereas German drohen assigns dative (wem). Hence the ellipsis is in Spanish. The generalization that the ellipsis site has to be in the same language as in the antecedent is corroborated by the data from Merchant (2014), as seen in the previously discussed examples (32) and (33). Further, what these results also indicate is that codeswitching can occur inside the ellipsis site. In all the data from González-Vilbazo and Ramos (2014) and some of the data in Nee (2012), some element, the remnant, has been moved out of the ellipsis site prior to deletion. That in turn means that within the elided structure in (37), we find a Spanish verb amenazar and the lower copy of the German wh-phrase. This is a new observation. Finally, ellipsis provides further evidence that surface realization of structures plays a role in codeswitching. Take the structure in (34). This structure contains a switch between English it doesn’t and Greek exi drosula. Although the switch is perfectly grammatical, it is only so if the Greek verb phrase is not realized overtly. It appears that the codeswitched utterance is structurally fine but the surface realization of the VP is unacceptable for PF reasons. This problem is solved by eliding the Greek verb phrase and thus not realizing the codeswitch overtly. Page 19 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching Before the recent studies on ellipsis in codeswitching, it could have been speculated that codeswitching would prove to be fundamentally different from monolingual cases in ellipsis. For instance, in a theory of codeswitching as a surface-oriented phenomenon, the lack of overt surface realization in ellipsis could have entailed the impossibility of ellipsis in a codeswitching context. The results reviewed in the preceding section show that this is not the case, and ellipsis can occur in the context of codeswitching and codeswitching can occur within ellipsis. Ellipsis is another phenomenon that shows that bilingual linguistic structures are in principle no different from monolingual structures. Eventually it comes down to features and operations on those features. Whether those features (or feature bundles) are extralinguistically assumed to belong to one language or another is irrelevant to the linguistic system in general and to the derivation in particular.

19.5 Outlook As we have seen in this chapter, codeswitching has opened new avenues of research on ellipsis. At the same time, the extent to which ellipsis may be constrained in codeswitching remains unknown. More research is obviously necessary. Possible topics for further research are, among others, fragment answers and ellipsis in the nominal domain. Currently, an ongoing project (Ramos 2014) suggests that fragment answers in English/ Spanish codeswitching may be subject to the same constraints found in González-Vilbazo and Ramos (2014) for German/Spanish sluicing. Specifically, the morphosyntactic condition has to be observed with respect to the antecedent clause. In this ongoing study, constructions which would require a PP in one language but not in the other are investigated. (38)

(39)

The answer in (38) can be easily explained following the literature reviewed above. In this instance, the verb in the question, English marry, does not select for a PP, and the subsequent Spanish answer cannot be headed by a preposition. This is in line with previous findings, where the structure of the antecedent must match that of the ellipsis: Page 20 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching the underlying structure left unpronounced in the answer must contain the English verb marry and not the Spanish verb casarse, which does select for a PP complement. In (39), conversely, a new puzzle arises. The question contains a Spanish verb that selects for a PP. The possible codeswitched answers given, however, are an English bare DP or an English PP. The fact that a bare DP can be an acceptable response to the question in (39A) could suggest that the Spanish verb casarse inside the ellipsis can select for an English bare DP. This, however, contradicts what we know of the Spanish verb casarse, i.e., that it necessarily selects for a PP complement. The puzzling alternative would be that the preposition has been stranded inside the ellipsis. This in turn also runs counter to what we know about Spanish prepositions, namely, that they cannot be stranded.11 Other future research involves the resolution of gender in NP-ellipsis in codeswitching. In German/Spanish codeswitching, the following examples are attested: (40)

(41)

In (40), the elided noun appears to be masculine, which can be inferred from the remaining Spanish determiner. The German antecedent is also masculine, unlike the Spanish equivalent noun mesa, which is feminine. Consequently, the elided noun would have to be (p. 476) German Tisch, not Spanish mesa. On the other hand, the German neuter Mädchen ‘girl’ does not trigger the same effect (41): in the ellipsis, the remaining Spanish determiner bears feminine gender, suggesting that the elided noun is Spanish niña, and not German Mädchen. The differences between these two cases might reside in the fact that Tisch/mesa is inanimate and therefore only has grammatical gender, whereas Mädchen is animate, and in German its grammatical gender (neuter) differs from its biological gender (feminine). These preliminary results are examples of the many questions that remain to be studied.

Acknowledgements We want to thank Jason Merchant, Luis López and the UIC Bilingualism Research Lab for valuable feedback on this chapter. Our special gratitude to the editors and two anonymous reviewers for their comments and suggestions.

Page 21 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching

Notes: (1) While we will use the terms bilingualism and bilinguals throughout just for ease of exposition, note that what is asserted includes multilingualism and multilinguals as well. (2) As per the conventions in the literature, we will mark codeswitches with italics. (3) For a more complete review on the question of codeswitches of pronouns, see González-Vilbazo and Koronkiewicz (2016). (4) In the MLF, system morphemes are further divided into two types: early system morphemes and late system morphemes. “[Early system morphemes] are divided from late system morphemes because, along with content morphemes, they are conceptually activated to express a part of speakers’ meanings that they wish to communicate” (Myers-Scotton and Jake 2017). On the other hand, “late SMs [system morphemes] build hierarchical grammatical structure rather than enlarging or restricting the core meanings of referents” (Myers-Scotton and Jake 2017). (5) More recently, several theories of this sort have looked at different aspects of codeswitching, such as: codeswitching and language acquisition (Cantone 2007; Eichler 2011; Eichler and Müller 2012; Sunny Park-Johnson 2012), clausal structure (Mahootian 1993; González-Vilbazo and Struckmeier 2008; González-Vilbazo and López 2011, 2012; den Dikken and Bando Rao 2014), the DP (González-Vilbazo 2005; Liceras et al. 2008; Herring et al. 2010), pronouns (van Gelderen and MacSwan 2008; González-Vilbazo and Koronkowiecz 2016), wh-dependencies (Woolford 1985; Toribio and González-Vilbazo 2014), and classifiers (Bartlett and González-Vilbazo 2010, 2013). (6) Of course, this does not deny typological differences in grammars in languages (e.g., V2/non-V2). Rather, it states that the grammatical components are the same for monolingual and bilingual systems. In this view, the differences would be encoded in features or feature combinations in the lexicon. For further discussion, cf. GonzálezVilbazo and López (2012). (7) There is a continuous debate about the number of lexica in bilingual linguistic systems. Den Dikken (2011) argues for one lexicon for bilinguals, while the vast majority of the literature argue for two (Woolford 1983; MacSwan 2005; Bartlett and GonzálezVilbazo 2013). (8) Nee reports that in Valley Zapotec the prepositions are optionally inverted in interrogative contexts; it is only in sluicing contexts that P-inversion is mandatory. (9) The editors point us to the well-known observation that ellipsis has been argued to repair otherwise ungrammatical sentences (cf. Richards 2001; Lasnik 2001, 2010; van Craenenbroeck and den Dikken 2006; Merchant 2008a). Indeed, Merchant also considers this, though operating “at the morphological level” (Merchant 2015b).

Page 22 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Codeswitching (10) An editor wonders if any insight could be gained from studying elliptical island repair on codeswitching. This is very possible, although it would be necessary first to look into island effects in codeswitching without ellipsis. To our knowledge, no such study exists. Looking into codeswitching between two languages, which differ with respect to their island effects, we would expect to gain insights into the elements that play a role in licensing those effects. (11) An anonymous reviewer suggests the answer ‘my sister’ in both (38) and (39) could be the result of an underlying cleft or cleft-like structure, similar to cases investigated by Rodrigues et al. (2009) and van Craenenbroeck (2010a). This is certainly a possibility, and future research could control for this possibility by using multiple remnants or nonexhaustive modification.

Kay González-Vilbazo

Kay González-Vilbazo is Associate Professor at the University of Illinois at Chicago and Co-Director of the UIC Bilingualism Research Lab. His research focuses on linguistictheoretical aspects of bilingual phenomena, mostly the grammar of codeswitching. Current projects of his include gender agreement and concord, phase theory, wh-dependencies, ellipsis, pro-drop, case theory, the structure of PF, the phonology of codeswitching, and the theory of the bilingual lexicon. Sergio E. Ramos

Sergio E. Ramos is a graduate student at the University of Illinois at Chicago. His main research interests lie at the intersection of bilingualism and linguistic theory. These include codeswitching, identity, and second language acquisition as well as topics of linguistic theory such as nominal ellipsis, sluicing, case, and Romance linguistics more broadly.

Page 23 of 23

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Oxford Handbooks Online Sluicing and Its Subtypes   Luis Vicente The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.22

Abstract and Keywords This chapter discusses the phenomenon of sluicing, a type of clausal ellipsis whereby an entire interrogative clause is missing save for a wh-phrase. Sluicing consists of a number of subtypes, each one exhibiting its own properties and restrictions. Given that the properties and restrictions of any given subtype tend to be the properties and restrictions of some independently available strategy of wh-question formation, the most parsimonious analysis is the one advanced by Ross (1969) and Merchant (2001): sluicing is a purely PF-side operation, i.e. non-pronunciation of a regular wh-question, without affecting its syntactic or semantic properties. At the same time, we need to contend with the fact that this seems to be only a tendency (albeit a remarkably strong one), rather than an exceptionless generalization: there exist cases of sluicing that appear to resist assimilation to a pure deletion approach. Determining whether they are genuine exceptions (and if so, exactly which factors allow them to be exceptional) is arguably one of the main issues that an eventual comprehensive, insightful theory of the taxonomy of sluicing must resolve. Keywords: sluicing, pseudosluicing, isomorphic sluicing, non-isomorphic sluicing, multiple sluicing, clefting, island repair, wh-in-situ, wh-movement, focus movement

Luis Vicente

20.1 Introduction THE approach to sluicing that I discuss in this chapter is the one originally advanced by Ross (1969b) and later on revived and explored extensively by Merchant in a series of publications (see especially Merchant 1998, 1999, 2001, 2002, 2004a, 2008b, 2010, 2013d, and Giannakidou and Merchant 1998, plus Merchant 2006a, this volume, and Chapters 1 and 2 of this volume for overviews of data and arguments).1 Abstracting away from framework-related details, the core of Ross’s and Merchant’s proposals is the same: Page 1 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes sluicing is a PF-side operation, namely, deletion (non-pronunciation) of a syntactically complete wh-interrogative to the exclusion of the wh-expression, under identity with some suitable antecedent.2,3 (p. 480) Somewhat unimaginatively, I will refer to this line of analysis as the Ross–Merchant approach. By way of illustration, suppose that sluicing applies to (1a) and produces the surface string (1b), where [__] marks the position of the sluicing site. What Ross and Merchant are telling us is that the best analysis for (1b) is going to be one along the lines of (2), where the light grey font represents lexical material that fails to be phonetically realized.4 (1)

(2)

More precisely, both Ross and Merchant define deletion as constituent deletion (i.e., a specific node in the tree, and every node dominated by it, fails to receive a phonetic realization), which requires the remnant to evacuate the sluicing site prior to deletion, lest it also gets deleted. This requirement is easy to satisfy (in the general case) in languages with overt wh-fronting, given that the regular mechanism of wh-question formation independently takes the remnant to a position external to the sluicing site. More problematic, however, are sluices in wh-in situ languages, illustrated in (3) with Japanese (but see Kirchner 2006, Gribanova 2013c, Gribanova and Manetta 2016, and many of the chapters in Merchant and Simpson 2012 for comparable examples in other languages). How come sluicing is possible in the absence of overt wh-movement? (3)

Page 2 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

A different version of the same problem arises in wh-fronting languages when we combine sluicing and islands. Ross (1969b: 276ff.) already noted that (4b) is a grammatical sluice, even though the unsluiced question (4a) incurs a strong island violation. How come the sluice is grammatical if the putative movement step required to derive it is illicit? (p. 481)

(4)

There are, in principle, three different ways in which (3b), (4b), and comparable examples can be brought into the deletion fold. First, one could simply deny that constituent deletion is a necessity; instead, one could define a process of non-constituent deletion that elides everything inside IP except for the in situ remnant (Kimura 2010, Abe 2015, Ott and Struckmeier 2015, and in a different context, An 2007), as illustrated in (5)–(6). On the assumption that sluicing bleeds A-bar movement in wh-fronting languages (cf. Kimura 2010; Abe 2015), the lack of island effects follows directly from the fact that no movement is taking place. I will not discuss this solution any more in this chapter, given the accumulated evidence that regular A-bar movement is an integral part of sluicing in all wh-fronting languages and at least some wh-in situ languages (cf. Agüero-Bautista 2007, Abels 2011, Barros et al. 2014, plus Chapters 16 and 32 of this volume). (5)

(6)

Page 3 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes Alternatively, one could preserve the constituent deletion requirement and assume instead that sluicing invariably licenses movement of the remnant to an IP-external position (as illustrated in (7)–(8)), even if such movement is otherwise illicit in unsluiced questions—see Takahashi (1993, 1994), and Takahashi and Lin (2012) for an application of this idea to wh-in situ languages, plus Merchant (2001, 2004a) and Almeida and Yoshida (2007), among others, for an application to sluicing out of islands in wh-fronting languages. This approach faces the problem of not being fully generalizable: both across and within languages, the observed distribution of grammatical and ungrammatical sluices is inconsistent with an analysis where sluicing licenses exceptional, islandinsensitive movements on a regular basis (Rodrigues et al. 2009; Abels 2011; Gribanova 2013c; Barros et al. 2014; and references). However, we will see in §§20.3 and 20.4 some environments where an appeal to a weak version of this approach (p. 482) (i.e., sluicing exceptionally licenses only a small subset of otherwise illicit movements) might be necessary. (7)

(8)

Note that the two approaches above presuppose a high degree of syntactic isomorphism between the sluice and the antecedent, i.e., the sluice is invariably a question built on the same predicate type as the antecedent. Our third approach, which constitutes the core of this chapter, rejects this assumption: so long as the standard semantic identity conditions are satisfied, sluices may stem from a different predicate (more precisely, various types of clefts and copular clauses),5 and any required movements need not be wh-movement in the strict sense (e.g., sluicing can be derived by scrambling or focus movement; see van Craenenbroeck and Liptak 2006, 2013, Toosarvandani 2008, Ince 2009, 2012, and Chapter 32 of this volume). For example, Kizu (1997a) and Merchant (1998), among others, propose that (3b) stems from a copular clause exhibiting simultaneous subject drop and copula drop (both independently possible in Japanese) as illustrated in (9); similarly, Merchant (2001) and Barros et al. (2014), among others, propose that (4b) stems from deletion of a cleft, as in (10). Figure 20.1 partially illustrates the range of underlying structures that sluices have been argued to stem from across languages (although, due to space restrictions, I will not be able to go through this taxonomy in detail). (9)

Page 4 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

(10)

This line of attack is interesting in that, unlike the other two approaches, it commits us to a very clear and specific prediction—i.e., that across languages the taxonomy of sluices should parallel the taxonomy of unsluiced questions. It follows from this that any given sluice should have the same properties as the corresponding unsluiced Figure 20.1 A partial taxonomy of sluicing, based on question. This prediction the underlying syntax of the sluice has been explored extensively, with notable results, in a great variety of languages and sluicing types over the last fifteen years, starting with the seminal work of Shimoyama (1995), Nishiyama et al. (1996), Kizu (1997a), and Merchant (1998) on Japanese, (p. 483) and following with Merchant (2001, 2004a), Hiraiwa and Ishihara (2002, 2012), van Craenenbroeck (2004, 2010a), van Craenenbroeck and Lipták (2006, 2013), Potsdam (2007), Toosarvandani (2008), Vicente (2008), Ince (2009, 2012), Rodrigues et al. (2009), Barros (2012a, 2014b), Hoyt and Teodorescu (2012), Paul and Potsdam (2012), Gribanova (2013c), Barros et al. (2014), Barros and Vicente (2016), and Gribanova and Manetta (2016), among many others—see also the bibliography in Merchant and Simpson (2012) for an extensive list of related publications. As Merchant (2004a: 669) aptly points out: These parallels in distribution are immediately and straightforwardly accounted for by the theory of sluicing discussed above, since the grammatical constraints […] will be operative uniformly in both elliptical and non-elliptical structures. The goal of this chapter is to present a small subset of the data and arguments discussed in this body of literature (§§20.2 and 20.3), as well as some patterns that seem to put limits on the generalizability of a pure deletion analysis as described in Merchant’s quote above, i.e., grammatical sluices for which no grammatical unsluiced question seems to exist (§20.4). Due to space restrictions, I am going to deliberately (and unfortunately) Page 5 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes ignore a number of important issues, such as the different discourse functions of sluices (Ginzburg and Sag 2000; Schlangen 2003; Fernández 2006), the status of sluices without a linguistic (p. 484) antecedent (Ginzburg 1992; Ginzburg and Sag 2000; Stainton 2006b; Merchant 2010; Weir 2014), correlate–remnant congruence restrictions (Chung et al. 1995, 2011; Romero 1998; Dayal and Schwarzschild 2010; Barros 2012b; Winkler 2013; Messick et al. 2015; Jacobson 2016b), voice and argument structure mismatches (Chung et al. 1995, 2011; Chung 2005; Barros 2014b; Barros and Vicente 2016; Merchant this volume), the conditions on, and causes of, deletion (cf. Rooth 1992a; Tancredi 1992; Johnson 2001a; Merchant 2001; Hartman 2007; AnderBois 2011, 2014; Chung 2013; Barros 2014b), certain minor variants of sluicing like sprouting (Albert 1993; Chung et al. 1995, 2011; Romero 1998; Chung 2005; Barros and Vicente 2016), swiping (Merchant 2002; Culicover and Jackendoff 2005; Beecher 2008; Hartman and Ai 2009; van Craenenbroeck 2010b; Larson 2014; Radford and Iwakasi 2015), and sluices derived from relative clauses (Szczegielniak 2004; van Craenenbroeck and Lipták 2006, 2013), modal existential constructions (Šimík 2011: §20.5.5), or amalgams (Kluck 2011). Readers interested in these issues are referred instead to the works just cited and the references therein.

20.2 Many roads lead to a sluice 20.2.1 Non-isomorphic sluicing and island repair effects The idea that sluices can also stem from deletion of non-isomorphic predicates (specifically, clefts and copular clauses) is hardly a new one: Erteschik-Shir (1973: 170), Pollmann (1975: 286), and Rosen (1976) proposed this much in order to account for Ross’s observation that sluicing is able to circumvent island effects. Their claim, elaborated in more detail by Merchant (2001: §20.5.3), Rodrigues et al. (2009), and especially Abels (2011) and Barros et al. (2014), is that repair effects are only a superficial illusion: whenever sluicing appears to void an island violation, it is because the sluice in question stems from deletion of a non-isomorphic clause that doesn’t incur the pertinent violation (an evasive source, in the terminology of Barros et al. 2014). This line of attack makes a very clear prediction: because repair effects are necessarily linked to the availability of an evasive source, we should expect island effects to reappear whenever the evasive source can be independently blocked. Given that locality effects are discussed in more detail in Chapter 16 of this volume, I offer only a brief illustration of this pattern here: in many languages, extraction of an attributive adjective out of its containing DP is ungrammatical (a violation of Ross’s 1967 Left-Branch Condition), but the corresponding sluice is grammatical. Call these left-branch extraction (LBE) sluices, illustrated here with English. (11)

Page 6 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Merchant (2001: §20.5.1.1) proposes a literal repair analysis of these sluices, but Barros et al. (2014) observe that they exhibit various properties that support an evasive analysis instead. First, the repair effect is suspended for adjectives with non-intersective readings (12a), and (p. 485) adjectives ambiguous between an intersective and a non-intersective reading are invariably disambiguated in favor of the intersective reading (12b). (12)

Notably, non-intersective readings of adjectives are also disallowed in the predicate position of copular clauses. (13)

This parallelism suggests that (11b) doesn’t stem from deletion of (11a) but rather from deletion of a predicative copular clause, which doesn’t incur an LBC violation in the first place. (14)

Barros et al. reinforce this analysis by pointing out that, in languages where adjectives inflect differently depending on whether they are used predicatively or attributively, LBE sluices require the predicative inflection. I illustrate this below with German, where attributive adjectives bear the appropriate case morphology and both predicative adjectives and remnants of LBE sluices are bare.6 This pattern follows directly from an evasive analysis, but a literal repair analysis would predict sluicing remnants to bear the attributive inflection. Page 7 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes (15)

To complete the argument, note that languages that allow extraction of attributive adjectives in unsluiced questions (and therefore do not require an evasive source) exhibit the opposite pattern, i.e., they allow LBE sluices with intersective adjectives7 and attributive inflection. I illustrate this pattern here with Serbo-Croatian (data from Boban Arsenijević, p.c.). This asymmetry between language types is predicted to arise only in an analysis where the repair effect is an illusion.8 (p. 486)

(16)

I refer the reader to Lasnik (2001), Merchant (2001: §5.1), Fox and Lasnik (2003), Lasnik and Park (2003), van Craenenbroeck and den Dikken (2006), Agüero-Bautsta (2007), Nakao and Yoshida (2007), Merchant (2008b), Nakao (2009), Rodrigues et al. (2009), Abels (2011), Barros (2012a), Fukaya (2012), Nakamura (2012), Yoshida et al. (2012b), Cantor (2013),Marušič and Žaucer (2013), Rottman and Yoshida (2013), Barros (2014a), Barros et al. (2014), Griffiths and Lipták (2014), and especially Chapter 16 of this volume for additional discussion of the evasive approach to island repair, as well as of its limitations. See also §§20.4.1.1 and 20.4.1.2 for a brief illustration of two environments where literal island repair might be necessary.

20.2.2 Non-isomorphic sluicing and wh- in situ languages Beyond providing a means to evade island violations, cleft- and copula-based sluices have also been successfully applied to the analysis of sluicing in wh-in situ languages. Consider Japanese, which I already mentioned in §20.1. (p. 487)

(17)

Page 8 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Shimoyama (1995), Nishiyama et al. (1996), Kizu (1997a), Merchant (1998), Fukaya (2007, 2012), Hiraiwa and Ishihara (2012), and Nakamura (2012), among others, have developed a number of arguments in favor of a non-isomorphic cleft source (but see Iseda 2007 and Hasegawa 2008, as well as the discussion in §20.2.4 here). For example, Shimoyama notes that these sluices support a copular verb, which would be surprising if they didn’t stem from an underlying cleft. (18)

Additionally, Kizu (1997a) and Merchant (1998) note, following earlier literature, that cleft pivots tend to resist being overtly case-marked, each case morpheme triggering a specific degree of deviance (and some degree of cross-speaker variation). Compare, for example, the different acceptability of nominative -ga and accusative -o: (19)

As expected under a cleft-based analysis, sluices exhibit the same case-marking pattern: (20)

Page 9 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes This line of analysis extends easily to other wh-in situ languages. I refer the reader to Kirchner (2006) and Tomioka (2012) for Mandarin Chinese, and Gribanova (2013c) for Uzbek. (p. 488)

20.2.3 Non-isomorphic sluicing as a default strategy As is the case with other Austronesian languages, wh-extraction of non-subjects in Malagasy requires a pseudocleft (21). Potsdam (2007) and Paul and Potsdam (2012) argue that this restriction remains active under ellipsis, so sluices like (22) must stem from a comparable non-isomorphic, pseudocleft source. (21)

(22)

Potsdam (2007) and Paul and Potsdam (2012) support this analysis by pointing out that sluices exhibit certain restrictions that are otherwise specific to pseudoclefts; as should be obvious, such parallelisms follow without stipulation if sluices are derived from a pseudocleft base. First, there are certain constituents (e.g., accusative-marked arguments) that can neither be questioned through the pseudocleft strategy (23a) nor sluiced (23b). (23)

Page 10 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes Second, pseudocleft pivots (qua fronted predicates) can be directly followed by a variety of elements, such as the modifiers daholo ‘all’ and foana ‘always’ (24). Potsdam’s and Paul and Potsdam’s analysis of this pattern relies on the well-supported assumption that daholo, faona, and similar modifiers are contained in the constituent that undergoes fronting in pseudoclefts (PredP in their terminology), which allows them to escape deletion of IP1. As expected, the same range of elements can also follow sluicing remnants (25). (p. 489) (24)

(25)

20.2.4 The balance between isomorphic and non-isomorphic sluicing Given the pervasivity of cleft and copular sluicing, there might be a certain temptation at this juncture to jump to the conclusion that all cases of sluicing can be reduced to cleft or copular sluicing (i.e., there would be no proper isomorphic sluicing). This is a conjecture that can be safely rejected. Merchant (2001: 120–7) already provides ten independent arguments against a reduction along these lines. In the interest of brevity, I reproduce only two of them here: first, he notes that adverbial wh-phrases (e.g., how, why, when…) are allowed in wh-questions (26a), but not in clefts (26b). The fact that sluices with adverbial wh-remnants are grammatical (26c) suggests that they do not stem from an underlying cleft. (26)

Page 11 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Second, regular wh-questions allow else-modification (27a), but clefts do not (27b). As above, sluicing remnants pattern with regular wh-questions in allowing else-modification (27c). (27)

Importantly, Merchant notes that these arguments only contraindicate a general reduction of sluicing to cleft sluicing, but leave open the possibility that some sluices stem, in fact, from a cleft (or a copular clause); van Craenenbroeck (2010a) and Barros (2014b) make (p. 490) the same point.9 This much suggests, as already mentioned above, that sluicing is a syntactically heterogeneous construction both cross- and intralinguistically. Merchant’s arguments can be replicated in other languages—see, for example, Rodrigues et al. (2009) and Saab (2015) on Spanish, or Lipták (2013a) on Hungarian. Especially interesting in this respect is the status of Farsi (Toosarvandani 2008 and Chapter 32 of this volume) and Turkish (Ince 2009, 2012). Both are wh-in situ languages, and in both of them the battery of tests discussed above contraindicate deriving sluicing from a nonisomorphic source, at least in the general case. Toosarvandani and Ince conclude from this combination of factors that Farsi and Turkish sluicing is of the isomorphic kind. The difference with English-style isomorphic sluicing is that the movement of the remnant out of the sluicing site is not a case of wh-movement proper; rather, it is a case of focus movement, which is independently attested and productive in both languages. Takita (2009) develops a comparable analysis for the specific case of Japanese sluices embedded under control predicates (28). (28)

Page 12 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes Takita notes that these predicates cannot take cleft or copular complements (29), which contraindicates deriving sluicing from a non-isomorphic source; this conclusion is reinforced by the fact that the relevant sluices disallow the optional presence of a copular verb (30), unlike the cases discussed in §20.2.2. From this, Takita concludes that these sluices stem from an isomorphic source, with extraction of the remnant out of the sluicing site being a case of scrambling.10 (29)

(30)

These results suggest that non-isomorphic sources are not the only means to circumvent the limitations of wh-movement—one can also resort to an isomorphic source (p. 491)

coupled with a different type of movement. This conclusion, however, is hardly new: Merchant (2001: 15) has briefly hinted at a similar analysis of Turkish and Hindi, writing that “to the extent that sluicing structures pattern with those found in overt whmovement languages like English, [they] are employing a scrambling-type movement to create the input structures for deletion, and not using ‘true’ wh-movement to specCP.”

20.3 Multiple sluicing 20.3.1 Genuine multiple sluicing 20.3.1.1 Multiple focus/wh--fronting Given the discussion so far, it is unsurprising that multiple wh-fronting languages allow sluices with multiple remnants. I illustrate this possibility in (31) with Serbo-Croatian, but comparable paradigms can be constructed in other Balkan/Slavic languages (cf. Richards 2001 and Grebenyova 2006b, among others).11 (31) Page 13 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Given that there are different subtypes of multiple wh-fronting languages, each one with a different cluster of properties (Rudin 1988 et seq.), the Ross–Merchant approach to sluicing predicts that multiple sluices in each of these languages will inherit the properties of the corresponding overt multiple questions. In support of this prediction, Merchant (2001: 10) notes that Bulgarian multiple sluices (32a) respect Superiority, just as their unsluiced counterparts do (32b). (p. 492)

(32)

The reverse effect also obtains, i.e., Superiority violations under multiple sluicing are possible only in languages that also allow them in unsluiced multiple questions. I illustrate this pattern here with Russian (data from Grebenyova 2009).12 (33)

Grebenyova (2006a: ch. 3) further points out that Russian unsluiced multiple wh-questions allow pair-list, but not single-pair readings. The examples below demonstrate that, as expected, this restriction carries over to multiple sluices. Note that the manipulation of the subject in the antecedent clause (každyj ‘everyone’ vs ktoto ‘someone’) is done deliberately to induce the pair-list and single-pair reading, respectively. (34)

Page 14 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

In contrast to Russian, Serbo-Croatian allows single-pair readings in multiple wh-questions (Bošković 2003; Grebenyova 2006a). As expected, this enables multiple sluices with single-pair readings.13 (p. 493)

(35)

For a more extensive and detailed discussion of the correlations between multiple wh-fronting and genuine multiple sluicing, I refer the interested reader to Grebenyova (2006a, 2009), van Craenenbroeck and Lipták (2013), and references therein.

20.3.1.2 Multiple clefting Some wh- in situ languages also exhibit sluices with multiple remnants. Here, I will concentrate on Japanese, for which Takahashi (1994) and Kuwabara (1996), Hiraiwa and Ishihara (2012) and Takahashi and Lin (2012) have noted examples like (36). (36)

Page 15 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

The standard analysis of such examples relies on the fact that multiple clefting is independently possible in Japanese (37)—see Kuwabara (1996), Hiraiwa and Ishihara (2012), and references therein.14 (p. 494)

(37)

Hiraiwa and Ishihara (2012) discuss a number of parallelisms between multiple clefts and multiple sluices that suggest that the latter ought to be derived from the former. For example, Japanese DPs can appear without case markers under certain circumstances, but multiple clefting requires case markers to be retained (38a). In the same way, remnants of multiple sluicing necessarily each appear with their corresponding case markers (38b). (38)

Similarly, pivots of multiple clefts generally need to be clausemates except if they are wh-expressions—see especially Ishihara (2012) and references for an analysis of this pattern. As expected, remnants of multiple sluicing pattern with wh-pivots of multiple clefts in not being subject to the clausemate restriction.15 (39)

Page 16 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

As above, I refer the reader to Kuwabara (1996) and Hiraiwa and Ishihara (2012) for a more detailed discussion of these examples. What is relevant for the purposes of this (p. 495)

chapter is the fact that languages with non-isomorphic sluicing allow multiple sluicing remnants, so long as the corresponding unsluiced structure (in the case of Japanese, clefts with multiple pivots) is independently available.

20.3.2 Fake multiple sluicing 20.3.2.1 Single sluicing plus Heavy NP Shift Contrary to what one might expect from a Ross–Merchant analysis, multiple-sluicing-like configurations are also allowed in languages that do not otherwise allow multiple fronting (whether qua wh- or focus-driven movement) or multiple clefting. The like qualifier is important here: it turns out, upon closer inspection, that what these languages exhibit is a case of single sluicing supplemented with some additional process that produces the illusion of genuine multiple sluicing, as in the English examples in (40).16 (40)

Note that the placement of the [__] diacritic in between the two remnants is not accidental. Both Merchant (2001) and Richards (2001) originally proposed that sluicing licenses exceptional multiple overt wh-fronting in English (and, by extension in other nonmultiple wh-fronting languages). Against this background, Lasnik (2014) proposes that Page 17 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes only the first (p. 496) remnant undergoes regular focus/wh-fronting; the second one escapes the sluicing site through rightward Heavy NP Shift (41). (41)

Among Lasnik’s arguments in favor of (41) is the fact that the second remnant doesn’t allow P-stranding (42). This is a surprising restriction, given that P-stranding under sluicing is otherwise possible in English (cf. Merchant 2001); however, it follows directly from (41), given that P-stranding under HNPS is likewise impossible (42b)/(42c). (42)

Additionally, Lasnik also notes that multiple sluicing becomes impossible if the two remnants are not clausemates (43a); other languages I have checked (i.e., Farsi, Maziar Toosarvandani, p.c., and Turkish, as described in Ince 2009) behave in the same way.17 This restriction follows directly from Lasnik’s proposal, as the second remnant would have to undergo HNPS across a finite clause boundary in violation of Ross’s (1967) Right Roof Constraint (illustrated in (43b) for English). (43)

Note that, under Lasnik’s analysis, this clausemate restriction is predicted not to hold in languages where multiple sluicing can be derived through regular multiple wh-/focus fronting (see §20.3.1.1), at least to the extent that fronting of non-clausemate multiple wh-phrases is possible in non-elliptical contexts. The paradigm in (44) illustrates this effect for Serbo-Croatian. Lasnik points out that not all speakers he consulted accept (44a). Importantly, though, those same speakers didn’t accept (44b) either, which provides additional support for the hypothesis that there is a direct connection between the acceptability of the sluiced and unsluiced forms of these questions. (44)

Page 18 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

20.3.2.2 Null coordination of simple sluices Merchant (2001: 111) argues that the Turkish multiple sluice in (45a) is likely to be two separate simple focus sluices joined by a null coordinator. In support of this idea, he points out that the two remnants need to be separated by either a strong pause or an actual overt coordinator (45b). (p. 497)

(45)

Gribanova (2013c) proposes the same analysis for the Uzbek example in (46), which appears to instantiate multiple pseudosluicing. The evidence she offers is the same Merchant does for (45a), i.e., that “when larger, more weighty conjuncts are coordinated […] a large pause (47a) or an overt coordinator (47b) becomes necessary.” (46)

(47)

Page 19 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Obviously, if Merchant’s and Gribanova’s analyses are correct, then (45a) and (46) instantiate multiple sluicing only superficially, not in any analytically meaningful sense of the term. (p. 498)

20.4 Limits of a pure deletion analysis As should be obvious at this stage, the unambiguous prediction of the Ross–Merchant analysis is that sluices should invariably exhibit the same syntactic properties as the corresponding unsluiced question. This is admittedly a very strong prediction, so it is unsurprising that it runs into difficulties every now and then. The goal of this last section is to present a few of these difficult cases. The question of whether there is a way to subsume them under a pure deletion analysis (or whether they require extending the deletion analysis in specific ways) is one that runs beyond the limits of this chapter.

20.4.1 Repair effects without an evasive source 20.4.1.1 Parasitic gaps in the remnant Yoshida et al. (2015) argue at length that the remnant-internal gap in (48), notated [PG2__] is a parasitic gap licensed by a real gap [RG2__] contained inside the sluicing site (and thus not directly visible), paralleling the licensing of the parasitic gap in the first conjunct [PG1__] by its corresponding real gap [RG1__]. Note that standard licensing conditions on parasitic gaps prevent [PG1__] from being licensed by [RG1__]. (48)

Note, however, that the unsluiced counterpart of (48) is ungrammatical, a fact that Yoshida et al. are well aware of (49). Moreover, the source of the ungrammaticality of (49)

Page 20 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes is easy to pinpoint—namely, the illicit movement of the wh-phrase containing the parasitic gap across the wh-island boundary created by which book. (49)

The acceptability of (48) and similar examples suggests that the theory of sluicing must be supplemented with some island repair capabilities (as proposed by Ross 1969b and Merchant 2001, but contrary to the evasion approach discussed in §20.2.1). This conclusion, if valid, raises the difficult question of why some island effects require an evasive source to be “repaired,” whereas others allow literal repair. (p. 499)

20.4.1.2 P-stranding effects in non-P-stranding languages

Merchant (2001) argues that P-stranding effects under sluicing are observed only in those languages that allow P-stranding in unsluiced questions. A P-stranding language like English allows prepositions heading the remnant to be freely dropped (50), whereas a non-P-stranding language like German does not (51). Previous literature has referred to this as the P-stranding generalization (PSG), but here I will use Merchant01’s generalization instead.18 (50)

(51)

Some counterexamples to Merchant’s generalization have been reported, but the relevant question here is whether they are genuine counterexamples or simply instances of evasive sluices (in the sense discussed in §20.2.1), where a cleft/copular source creates a superficial P-stranding illusion. Rodrigues et al. (2009) argue at length that the second option is the correct one for Romance languages, focusing on Spanish and Brazilian Portuguese. Their argument is based on the fact that P-stranding sluices in these languages exhibit a number of properties typically associated with clefts. To give a single example (from Spanish), they observe that wh-phrases in non-cleft questions can be modified by más ‘else’, but those in cleft questions cannot. (52) Page 21 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

Under sluicing, the presence vs absence of a preposition correlates with the possibility vs impossibility (respectively) of más modification. This parallelism, among others, leads Rodrigues et al. to conclude that P-stranding effects under sluicing stem from an evasive source, rather than being direct counterexamples to Merchant’s generalization. (53)

Rodrigues et al. conclude their study by noting that “the strongest implication is that all languages that appear to violate this generalization should be reducible to a (p. 500)

pseudosluicing [i.e., evasive, LV] analysis.” A number of later studies have challenged this conjecture. For example, Sag and Nykiel (2011) and Nykiel (2013) argue that, in Polish Pstranding sluices, the remnant must have the case assigned by the missing preposition, rather than the characteristic instrumental of cleft pivots.19 (54)

Similarly, Sato (2011b) argues that P-stranding effects in Indonesian are not reducible to a cleft source, on the grounds that they also obtain in other types of ellipsis (e.g., pseudogapping), where a cleft source is clearly not at play. (55)

See also Stjepanović (2008, 2012) and Leung (2014) for a comparable line of argumentation in Serbo-Croatian and Emirati Arabic, respectively. Sato takes these data as an indication that the nature of the ban on P-stranding in unsluiced questions varies Page 22 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes across languages. He hypothesizes in Polish, Indonesian, and other languages, this ban can be characterized as a surface restriction that ellipsis can void, whereas in the languages that Rodrigues et al. examine, it is a “deeper” constraint that also holds under ellipsis. The nature of the ban, then, dictates whether P-stranding effects under sluicing require an evasive source or not. This approach, if correct, implies that literal repair effects under sluicing are possible (see also the conclusion of §20.4.1.1), contrary to the conjecture outlined in §20.2.1.

20.4.1.3 German Philipp (2015) observes that the fronted wh-phrase in the unsluiced question (56a) can contain an anaphor (sich) bound by the subject; note that this anaphor cannot be replaced with a coindexed pronoun (ihm), as this would induce a Condition B violation. The prediction of a Ross–Merchant analysis is that this anaphor/pronoun asymmetry should (p. 501) be preserved under sluicing, but as (56b) shows, it is not—in fact, Philipp reports that some of the speakers she tested had a slight preference for the pronoun over the anaphor (see also Ginzburg and Sag 2000: 297 for a similar pattern in English fragments). (56)

It is tempting to analyze the pronoun version of (56b) as stemming from deletion of a cleft, given that a pronoun is grammatical in this environment. (57)

However, this is not a general solution, given that the grammaticality of the pronoun persists even in sluices where a cleft source is not possible—e.g., sluices with a prepositional remnant (58a) or with a morphologically non-nominative remnant (58b). These two types of remnants block a cleft source because both are illicit cleft pivots. (58)

Page 23 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

The conclusion we can extract from this paradigm, then, is that sluicing interacts with Binding Theory in ways that we do not yet understand.

20.4.2 Morphological case parallelism effects Van Craenenbroeck (2008, 2012b) observes that the availability of cleft-based sluices in a variety of languages is subject to a morphological case identity condition. Consider the following examples: the ungrammaticality of a dative remnant wem in (59a) is unsurprising, given that it requires an illicit stranding of the dative-assigning preposition mit ‘with’ inside the sluicing site. More surprising is the fact that a nominative remnant wer is also (p. 502) ungrammatical, which suggests that, unlike in Spanish (see §20.4.1.2), the P-stranding violation cannot be circumvented through an underlying cleft. Note that the corresponding unsluiced cleft is grammatical (59b), so the ungrammaticality of a nominative remnant in (59a) is purely an ellipsis effect. (59)

One cannot attribute the ungrammaticality of a cleft source in (59a) to a language-wide ban on non-isomorphic sluices. For one, we already saw in §20.2.1 that German allows copula-based sluices as a means to circumvent LBC violations. Moreover, van Craenenbroeck also points out that P-stranding effects (and, by extension, the possibility of a cleft source) become licit under nominative case syncretism—consider the grammaticality of (60), where both etwas ‘something’ and was ‘what’ are nominativeaccusative syncretic. Van Craenenbroeck further shows that this pattern also holds in a number of languages beyond German (e.g., Greek, Czech, Hungarian…). (60) Page 24 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

On the basis of these data, van Craenenbroeck proposes that sluicing is subject to a morphological case parallelism requirement between the remnant and its correlate. Note that this requirement is a very superficial one, as it has to allow for abstract Case mismatches if case syncretism obtains, as in (60); Sag and Nykiel (2011) and Barros (2014b) reach the same conclusion. However, a requirement along these lines is too strong, as there exist multiple examples of licit sluices with non-identical case morphology (see Vicente 2014a for a non-exhaustive list). We already saw in §20.2.1 that German LBE sluices are grammatical despite the fact that the correlate is case-marked and the remnant is not. One might want to dismiss these sluices on the grounds that the lack of case morphology on the remnant satisfies the case identity requirement trivially, but this line of reasoning would not apply to languages that display licit sluices where the remnant and its correlate carry distinct case morphemes. For example, Sakamoto (2014) points out that, in Mongolian, remnants of embedded sluicing carry accusative case morphology, rather than the case morphology of their correlates.20 (p. 503)

(61)

Accounting for the observed distribution of case (non)-identity effects probably requires integrating a more sophisticated analysis of morphological case assignment and case syncretism (e.g., Caha 2009, among others) into our current analyses of sluicing. As far as I know, this integration has not been attempted yet.

20.5 Conclusions and outlook We can summarize this chapter as follows: sluicing consists of a number of subtypes, each one exhibiting a number of properties and restrictions. Given that the properties and restrictions of any given subtype tend to be the properties and restrictions of some independently available strategy of wh-question formation, the most parsimonious analysis is the one advanced by Ross and Merchant: sluicing is a purely PF-side operation, i.e., non-pronunciation of a regular wh-question, without affecting its syntactic or semantic properties.21 At the same time, we need to contend with the fact that it seems to be only a tendency (albeit a remarkably strong one), rather than an exceptionless generalization: we have seen in §20.4 that there exist cases of sluicing that appear to resist assimilation to a pure deletion approach. Determining whether they are genuine Page 25 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes exceptions (and if so, which factors exactly allow them to be exceptional) is arguably one of the main issues that an eventual comprehensive, insightful theory of the taxonomy of sluicing must resolve.

Acknowledgements I want to thank the editors of this volume, Matt Barros, Sandra Chung, Masaya Yoshida, and especially Jason Merchant for very insightful comments on a previous draft. All errors and shortcomings remain my responsibility.

Notes: (1) Throughout this chapter, I use the following terminology:

(2) Throughout this chapter I assume a general P&P/Minimalism framework (as do a majority of the works I cite), but readers should keep in mind that this is not a necessity. The essentials of a deletion analysis can be ported to other frameworks, so long as those frameworks allow us to talk about phonetically null constituents in a meaningful way (cf. Jacobson 2016b: §2 for a nuanced discussion of this issue). (3) There are a number of approaches to sluicing that propose a much less articulate, or even non-existent, syntax for the sluicing site. For example, Chung et al. (1995, 2011) propose that sluicing sites consist only of an atomic proform that inherits the LF of the antecedent (this approach effectively amounts to treating sluices as deep anaphors, contrary to Hankamer and Sag 1976); more radically, Ginzburg and Sag (2000), Culicover and Jackendoff (2005), Sag and Nykiel (2011), Barker (2013), and Jacobson (2016b) have argued that sluicing sites, qua independent syntactic constituents, don’t exist; this line of attack requires accepting the existence of rules that can map a bare wh-phrase to the semantics of a full wh-question. This whole chapter can be viewed as an extended argument against such non-deletion approaches: if the syntactic properties of any given sluice follow from the kind of syntactic structure that underlies the sluicing site, then analyses that deny the existence of such structure are inadequate, at least in the general case. See also Merchant (2001, 2004a, 2006a), Agüero-Bautista (2007), Abels (2011), Barros et al. (2014b), and references therein for more detailed arguments, and Vicente (2014b) for a summary of these arguments. See also the discussion of French sluicing in Chapter 30 of this volume for some problematic patterns. (4) This particular tree comes from Merchant (2001), but the exact category that the remnant moves to depends largely on the granularity of the functional structure one wishes to assume (as does the category that undergoes deletion). For example, Hartman Page 26 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes (2007), who assumes that Rizzi’s (1997) decomposition of the Italian CP layer carries over to English, proposes that the remnant moves to the specifier of ForceP after an intermediate stop in the specifier of FocusP, with deletion targeting a lower TopicP. See also Radford and Iwakasi (2015) for a further exploration of this line of analysis in English, as well as van Craenenbroeck (2004, 2010b) for Dutch. (5) There is a tendency in the literature to refer to sluices derived from clefts and copular clauses as pseudosluices, which is a perversion of the original meaning of the term, as popularized by Merchant (1998). Merchant used pseudosluicing to refer to a class of Japanese sluices that do not involve deletion of IP or a comparable clausal constituent, but rather simultaneous subject drop and copula drop, which under certain circumstances can produce the same surface result as genuine sluicing (hence the pseudo- qualifier). Here I am going to use pseudosluicing in the sense originally intended by Merchant and use the more accurate labels cleft sluicing and copular sluicing to cover the subtypes of sluicing other people use pseudosluicing for. (6) Merchant (2001: 173) contests this claim. He provides (i), with the indicated judgment, and the comment that “sluicing is fairly degraded, with or without inflection.” I have been unable to replicate this judgment in my own research: all the speakers I have consulted agree with the judgments that Barros et al. indicate. ((i))

(7) This generalization is somewhat of an idealization, in the sense that some adjectives in these languages lose their non-intersective reading upon extraction, regardless of whether sluicing happens. What is important for this particular argument is whether the availability of the non-intersective reading of a certain adjective under LBE sluicing correlates with the availability of this reading under unsluiced left-branch extraction. As far as I have been able to determine, this is indeed the case. (8) As an additional argument in favor of this view of LBC violation repair, note that many Slavic languages disallow multiple LBE. Example (i) illustrates this for Russian. This restriction persists under sluicing (ii), which would be surprising if sluicing could actually repair LBE violations (all data from Grebenyova 2006b). ((i))

((ii))

Page 27 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

(9) Although van Craenenbroeck and Barros make slightly different proposals about the availability of cleft sources: van Craenenbroeck defines an analysis where cleft sources are available only if using the corresponding non-cleft source would have produced an ungrammatical result; on the other hand, Barros proposes that both cleft and non-cleft sources are equally available in principle. (10) He treats the relevant movement as scrambling, and assumes that scrambling is itself a subtype of wh-movement (pace Takahashi 1993, 1994). It is unclear to what extent this assimilation is justifiable. (11) Here it is appropriate to note the work of Ortega-Santos et al. (2014) and Yoshida et al. (2014b) on English ellipses with two remnants, out of which only one is a wh-item. ((i))

((ii))

Ortega-Santos et al.’s and Yoshida et al.’s proposal is that such examples involve a combination of sluicing and stripping, the latter involving focus movement rather than wh-movement. In this sense, they are not cases of multiple sluicing; however, given the close parallelisms between sluicing and stripping (Merchant 2004a), this might be the closest that English gets to genuine multiple sluicing. (12) Note that this is only a one-way generalization. For example, Grebenyova points out that Russian Superiority-violating multiple sluices require the order of the remnants to parallel the order of the correlates. She attributes this restriction to a combination of Fox’s (2000) scope parallelism constraint and the fact that scope in Russian is defined on surface structures (Ionin 2001). Similarly, Stjepanović (2003) observes that SerboCroatian, which allows Superiority-violating unsluiced questions, bans Superiorityviolating multiple sluices, given that scrambling of an object across the subject in the antecedent blocks sluicing altogether.

Page 28 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes (13) It is not clear to me to what extent this correlation can be preserved in analyses where sluices do not have a regular syntax. Bošković (1999, 2003), Grebenyova (2006a), and others have argued at length that the availability of a single-pair reading is contingent on wh-fronting being triggered by a [FOCUS] feature, rather than a [WH-] feature. If this conjecture is cross-linguistically consistent (see Bošković 1999, 2003 for discussion), then the contrast between Russian-type languages and Serbo-Croatian-type languages can be construed as an additional argument in favor of sluices having a regular syntax. (14) Specifically, Hiraiwa and Ishihara make a distinction between clefts and pseudoclefts, which exhibit a number of asymmetries (e.g., possibility of multiple pivots, possibility of case dropping, possibility of nominative–genitive conversion, etc.). Given that sluices in general, and multiple sluices in particular, exhibit the same range of properties as clefts, Hiraiwa and Ishihara propose that (multiple) sluices stem exclusively from clefts, rather than pseudoclefts. (15) Although Takahashi (1994: §4.2) makes the opposite claim, citing (i) as evidence. At present, I do not know why judgments differ in this way. ((i))

(16) Takahashi (1994: 284) marks (i) as ungrammatical, which leads him to argue that English lacks multiple sluicing altogether (Takahashi and Lin 2012 makes the same claim). While (i) is indeed ungrammatical, it only shows that the distribution of English multiple sluices is not totally free; for one, Nishigauchi (1998) and Merchant (2001) point out that licit multiple sluices need to have a pair-list reading, which is not available in (i). ((i))

(17) Norwegian is a special case, as some of the speakers I polled do not report a clausemate restriction (i). Other than this, all the speakers I surveyed reported Lasniklike judgments in accepting multiple sluicing with clausemate remnants and rejecting Pstranding on the second remnant. I have currently no explanation for this pattern. ((i))

Page 29 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes

(18) Merchant (2001) himself uses the more opaque term Form-Identity Generalization II, where Form-Identity Generalization I is the observation that remnants of sluicing bear the same case morphology as their correlates (see §20.4.2). (19) Nykiel’s article reverses Szczegielniak’s (2008) previous claims that Polish patterns together with Spanish and Brazilian Portuguese. Note, however, that Szczegielniak introduces an unnecessary confound by focusing exclusively on the preposition z ‘with’, which assigns the same instrumental case that cleft pivots receive. Independently of this issue, I have attempted to replicate Nykiel’s reported judgments in an informal setting, but the two Polish speakers I polled rejected P-stranding sluices except in cases where both the remnant and the correlate carried instrumental case morphology. (20) Given that Mongolian embedded subjects are typically accusative, the pattern in (61) suggests that Mongolian sluices invariably stem from a cleft or copular clause (i.e., I don’t know who it is). Notably, embedded subjects can be nominative, under certain circumstances (Klein et al. 2012 and references), so one would expect (61) also to be grammatical with a nominative remnant. As of this writing, I have not been able to check if this is true. (21) In and of itself, this is a surprising result. Sluicing in the languages that are discussed here is a case of surface anaphora, in the sense of Hankamer and Sag (1976). But are there languages with deep anaphoric sluicing—that is, languages where the sluicing site is an unstructured proform? These would be languages that would behave exactly as predicted by Chung et al. (1995), i.e., they would not exhibit any locality or connectivity effects contingent on there being an internal syntax to the sluicing site, they would regularly allow sluicing with non-linguistic antecedents, etc. At present, I do not know of any languages where sluicing behaves deep-anaphorically in the general case.

Luis Vicente

Luis Vicente was a lecturer and researcher in the Department of Linguistics at the University of Potsdam, Germany. He earned a BA in English Philology from the Universidad de Deusto, Spain; a PhD in Linguistics from Leiden University, the Netherlands; and after a short lecture engagement at the University of Amsterdam, he was a postdoctoral researcher at the University of California Santa Cruz, supported by a prestigious scholarship from the Basque Government. He successfully obtained his Habilitation in July 2016 at the University of Potsdam. He was a prolific researcher who published a large number of important and influential papers on various phenomena, mainly focusing on the interaction between syntax and semantics. He passed away on 6 February 2018 at the age of 38 in Freiburg im

Page 30 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Sluicing and Its Subtypes Breisgau, Germany. His absence is a great loss to academia in general and linguistics in particular. He is sorely missed by us all.

Page 31 of 31

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

Oxford Handbooks Online Predicate Ellipsis   Lobke Aelbrecht and William Harwood The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.23

Abstract and Keywords This chapter discusses the phenomenon of predicate ellipsis, which covers several types of ellipsis targeting the clausal predicate. The best-known type is VP-ellipsis, but there is also pseudogapping, and two lesser-known types, namely Modal Complement Ellipsis and British English do. For each type the basic properties and main research questions are discussed. Extra attention is given to the question of what is actually elided, and it turns out that many instances of predicate ellipsis delete more than just the verb phrase. Although this phenomenon is best studied in English, VP-ellipsis also occurs in other languages, where it often appears as so-called V-stranding VP-ellipsis: the main verb undergoes movement when it is finite, allowing it to move out of the ellipsis site and survive ellipsis. These languages will also be discussed. Keywords: VP-ellipsis, pseudogapping, predicate, verb phrase, V-stranding VP-ellipsis, modal complement ellipsis, British English do

Lobke Aelbrecht and William Harwood

21.1 What is predicate ellipsis? AS the name leads one to suspect, predicate ellipsis is a type of ellipsis that leaves the main predicate of the clause unpronounced, most often together with one or more of its internal arguments or (low) adjuncts. Unlike clausal ellipsis, predicate ellipsis typically does not affect the canonical subject position and the finite verbal element. A prototypical example of predicate ellipsis is VP-ellipsis as it occurs in English: (1)

Page 1 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis What is missing in the second clause in (1) is the verbal predicate and its direct object, i.e. [eaten pumpkins]. The external argument which appears as the subject his wife, the (high) adverb surely, and the finite auxiliary has are unaffected by ellipsis. This phenomenon, especially in English, is probably the most widely discussed elliptical construction of all, and has a great amount of literature devoted to it. Some important publications on this topic include Hankamer and Sag (1976), Sag (1976a), Williams (1977c), Zagona (1982), Hardt (1993), Fiengo and May (1994), Lobeck (1995), Fox (2000), Johnson (2001b), Goldberg (2005), and van Craenenbroeck (2017), but this list is by no means exhaustive. The second section of this chapter discusses the distinctive properties and various analyses of VP-ellipsis (henceforth VPE) in more detail. VPE is not the only kind of ellipsis that targets the predicate and its internal arguments. A second type of predicate ellipsis is pseudogapping (first identified by Stump 1977; Levin 1978, 1986). This phenomenon elides the main verbal predicate, parallel to VP-ellipsis, but leaves one argument or (low) adjunct unaffected. As we saw with VPE, the finite auxiliary and the surface subject are not elided either. An example is given in (2), in which the main verbal predicate buy is elided, but the direct object lilies remains unaffected (just like the subject others and the finite auxiliary did). (2)

This type of ellipsis is called pseudogapping because it is reminiscent of another elliptical phenomenon, gapping, exemplified in (3). (p. 505)

(3)

The difference between gapping and pseudogapping is that pseudogapping does not include the inflectional domain—in this case the finite auxiliary—in its ellipsis site, whereas gapping does (for more on gapping, see Chapter 23 of this handbook). Despite the name, most accounts of pseudogapping consider this phenomenon to be a subtype of VP-ellipsis. In the mainstream generative analysis, a VP-internal remnant with contrastive focus is assumed to have moved out of the ellipsis site prior to the ellipsis. Section 21.3 zooms in on pseudogapping and discusses various analyses that have been proposed, which differ mostly in terms of the type of movement that extracts the remnant out of the ellipsis site (see Jayaseelan 1990; Johnson 1996; Lasnik 1999a, 1999b, 2001; Kennedy and Merchant 2000a; Takahashi 2003, 2004; Gengel 2007; Merchant 2008a; and Aelbrecht 2010), although Hardt (1993) and Lobeck (1995), for instance, advocate a different approach. Since these two elliptical constructions occur in Standard English, they have received a high degree of attention in the ellipsis literature. However, recently two other kinds of predicate ellipsis have been brought to the fore. The first type is called British English do Page 2 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis (henceforth BE do). The example of BE do in (4) shows that this type of predicate ellipsis looks identical to VPE, except for the addition of a non-finite form of the verb do (see Chalcraft 2006, Haddican 2007, Aelbrecht 2010, Thoms 2011b, and Baltin 2012). (4)

The second elliptical phenomenon that was recently added to the spectrum of predicate ellipsis is called Modal Complement Ellipsis (or MCE), first discussed in Busquets and Denis (2001), Depiante (2001), Cyrino and Matos (2002), and Dagnac (2010) for French, Italian, and Spanish, and Aelbrecht (2010) for Dutch. As the examples in (5) show, MCE is reminiscent of VPE as it occurs in English, but is attested in languages which have been claimed not to display VPE, as illustrated in (6) (see Lobeck 1995). (5)

(6)

As will be discussed in section 21.5, MCE also elides the main predicate of the clause, namely the infinitival complement of the modal, but differs from VPE in that it only occurs with root modals, not with aspectual auxiliaries. First, however, we focus on VPE as it occurs in English. Section 21.2 presents the main properties and analyses of VPE. Section 21.3 then turns to pseudogapping, and section 21.4 extends the VPE data range beyond English. The properties and analyses for both MCE and BE do are presented in section 21.5, and section 21.6 concludes. (p. 506)

Page 3 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

21.2 English VP-ellipsis The ellipsis literature has been dominated by research on VP-ellipsis in English for many years, and although this chapter does not limit itself to dealing exclusively with the English variant, most of what is known about predicate ellipsis starts with English VPE (see Zagona 1982, 1988a, 1988b; Lobeck 1987, 1995; Hardt 1993, 1999; Potsdam 1997b; Johnson 2001b, among others). First, in section 21.2.1, we look at the licensing of VPE in English. Section 21.2.2 discusses recoverability in VPE, and then the main approaches to VPE are discussed in section 21.2.3, revolving around the question of whether or not there is syntactic structure in the ellipsis site. Finally, section 21.2.4 investigates the size of the ellipsis site in VPE.

21.2.1 Licensing of English VPE VPE elides the main verbal predicate and its internal arguments, but leaves the finite verbal element and the subject intact. This implies that modals and finite auxiliaries survive the ellipsis, as illustrated in (7).1 The examples also show that both direct and indirect objects are missing in the ellipsis clause, as well as low adjuncts. (7)

In cases where the ellipsis clause does not contain an auxiliary, dummy do is introduced. The lexical verb itself cannot survive the ellipsis: (8)

Hence, these examples illustrate the generalization that VPE is found after tensed auxiliaries and not after lexical verbs. (p. 507)

No other verbal elements or aspectual verbs selecting non-finite clauses allow for

VPE, as the following sentences from Johnson (2001b: 440) illustrate: (9)

Page 4 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis One way of capturing this empirical generalization is to say that it is the finite auxiliary in T° that licenses VPE. This ties in with one of the conditions on ellipsis, namely syntactic licensing. As was first noted by Bresnan (1976: 17), Sag (1976a), Williams (1977c), and Zagona (1982, 1988a, 1988b) and further explored by Martin (1992, 1996), Lobeck (1993, 1995), and Johnson (2001b), VPE is only allowed in clauses with an overt T° head; or as Johnson (2001b: 439) puts it: “the ellipsis site must be in construction with, or perhaps governed by, a member of ‘Aux.’” The instantiations of these “Aux” elements are argued to be modals, the auxiliaries have, be, and do, and the infinitival marker to. However, the issue of what licenses VPE in English is not as straightforward as this. A first point of concern has to do with VPE in non-finite clauses, with infinitival to. Not all instances of to allow for VPE (see Lobeck 1987, 1992, 1995, 1999; Zagona 1988a, b; Johnson 2001b): when the infinitival clause is a complement, as in (10), VPE is licensed, but not when it is an adjunct, as in (11) (examples from Lobeck 1995). (10)

(11)

Furthermore, when an infinitival clause occurs in subject position, VPE is disallowed, as in (12), as observed by Zwicky (1982), and also in certain wh-clauses VPE is restricted; see (13).2 We refer the interested reader to Zagona (1982, 1988a, b), Zwicky (1982), Lobeck (1995), and more recently, Thoms (2011b) for further discussion. (12)

(13)

Page 5 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis Moreover, whilst control sentences involving a to-infinitive permit VPE, this is not the case for subject-to-object raising constructions that favour stative VP complements (see Martin 2001): (p. 508)

(14)

Note, however, that VPE following infinitival to in other types of subject-to-object raising constructions is permissible (Miller p.c.): (15)

Another problem for the claim that T° licenses VPE is the fact that negation also sometimes independently licenses VPE (example from Potsdam 1997b, see also Johnson 2001b): (16)

The exact contexts in which negation can license VPE are still unclear. A last remark concerning VPE licensing is that Gergel (2009) has claimed that VPE is degraded when licensed by epistemic must. This is illustrated in (17a), whereas (17b) shows that the ungrammaticality disappears when the modal is followed by a non-finite auxiliary. (17)

3

However, this restriction does not appear to be absolute, as Miller (p.c.) notes the following counterexamples: (18)

Page 6 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

Other epistemic modals, e.g. should or might, allow for VPE with the relevant epistemic reading: (19)

Recapitulating, English VPE requires the presence of a finite auxiliary, a modal, or the infinitival marker to (the latter only in certain syntactic environments). Sometimes, however, negation can license VPE, and VPE licensed by epistemic must is sometimes degraded. It is clear, therefore, that not all problems have been solved when it comes to (p. 509) VPE-licensing, nor do we have a thorough formal understanding of what ellipsis licensing actually is. In the next section we discuss recoverability in VPE constructions.

21.2.2 Recoverability Whether or not VPE is syntactically licensed is not the only factor determining whether or not it is licit in a given linguistic context. Recoverability also plays an important role. That is, an elided constituent must have a salient antecedent elsewhere in the linguistic discourse so that its meaning is recoverable by the hearer. In the following example, for instance, the elided constituent in the second clause can only be interpreted as punch Rocky since this phrase appears as a salient antecedent in the preceding clause. (20)

Much debate has revolved around what can serve as an antecedent for ellipsis. Importantly, Hankamer and Sag (1976) claim that VPE cannot be used exophorically. That is, it must always have a linguistic antecedent. Its meaning cannot be recovered via inference from the non-linguistic context. Consider, for instance, the following minimal pair: (21)

The important point here is that (21a) contains VPE whereas (21b) does not. The fact that (21a) is infelicitous, while (21b) is not, illustrates that VPE is not recoverable from the

Page 7 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis non-linguistic context (though see Schachter 1977b, Hankamer 1978, and in particular Miller and Pullum 2014 for more detailed discussion and counterexamples). A further issue is to what extent the elided constituent must be identical to its antecedent in order for it to be recoverable. Mismatches between the two can be informative as to the nature of the recoverability requirement: if formal mismatches that don’t affect the (truth-conditional) semantics are allowed, then this suggests that the recoverability requirement is semantic in nature. On the other hand, if even the slightest formal change leads to ellipsis no longer being an option, it seems justified to conclude that an ellipsis site has to be structurally identical to its antecedent in order for the application of ellipsis to be recoverable. The literature on VPE abounds in cases involving ellipsis–antecedent mismatches, but the theoretical conclusions that can be drawn from them invariably turn out to be more subtle and complex than they appear to be at first glance. Due to space limitations we only discuss one such case here, but see van Craenenbroeck (2017) for an overview of other mismatches. The case we focus on is that of VPE with a nominal antecedent, as illustrated in (22) (from Hardt 1993). (22)

(p. 510)

In this example, VPE (indicated by the underscore) does not take another VP as

its antecedent, but an NP, in particular the noun laugher in the first clause. At first sight, then, this seems like an open-and-shut case in favour of a semantic recoverability requirement over a syntactic one: while the VP laugh and the NP laugher can arguably mean (roughly) the same thing, they are clearly not structurally identical. Johnson (2001b), however, uses these data to make the exact opposite claim. The key to understanding this example, he argues, is the fact that laugher is a deverbal noun. If, as proposed by Fu et al. (2001), such nouns feature in their underlying representation an actual VP, then there is in fact a level of representation at which the ellipsis site and its antecedent are structurally identical, and examples like (22) cannot be used as arguments against an approach in terms of structural identity. More recently, however, Miller and Hemforth (2014a) have taken issue with Johnson’s analysis. They point out that it wrongly predicts (a) that all deverbal nouns should be able to serve as VPEantecedents, and (b) that underived (and hence non-deverbal) nouns should never be able to do so. Miller and Hemforth propose instead that VPE can only take a nominal antecedent when that NP expresses an implicit polar question. Since it is hard to see how this could be implemented in terms of a structural identity requirement, they contend that examples like (22) can indeed be used as an argument against such an approach. The next section addresses the issue of whether the elided constituent in VPE contains syntactic structure or not.

Page 8 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

21.2.3 Deleted syntactic structure? As discussed elsewhere in this handbook, there are several ways of analysing VPE. Roughly, there is a distinction between structural and non-structural accounts, which provide opposite answers to the question of whether there is syntax in the ellipsis site. Non-structural accounts assume no covert syntax in the VP-ellipsis site: under these approaches, the syntax of an elliptical sentence lines up with its phonology, and the computational burden is placed on the syntax–semantics interface in order to derive the correct interpretation (see Culicover and Jackendoff 2005, but also accounts in the HeadDriven Phrase Structure Grammar and Construction Grammar frameworks: Pollard and Sag 1994; Ginzburg and Sag 2000; Sag et al. 2003; Schlangen 2003; Stainton 2006a; Jacobson 2008; Kim and Sells 2008). The other end of the spectrum has the syntax lined up with the semantics of the elliptical clause, while ellipsis affects the phonology. This is known as the structural approach.4 According to this approach, which was first proposed by Chomsky (1955, 1965) and Ross (1969b), the syntax of an elliptical sentence does not really differ from the syntax of its non-elliptical counterpart. Ellipsis is an operation that marks part of the syntactic structure for non-pronunciation at PF, the phonological interface. It is entirely plausible that a structural approach suits certain types of ellipsis, while a non-structural approach suits other types, but we show here that VPE favours a structural analysis. A first argument has to do with agreement facts. Consider the following data: (p. 511)

(23)

In these examples the finite auxiliary agrees in number with the associate that is left unpronounced. This indicates that the associate should be present in the underlying syntax, otherwise these facts are less straightforwardly accounted for. Another argument in favour of the structural approach involves extraction out of the ellipsis site. Although there are certain restrictions to be considered (see Schuyler 2002 for extensive discussion), VPE allows for elements to be extracted from the ellipsis site. This is only possible if the ellipsis site contains enough structure to host the movement trace or copy. Some examples are given in (24), showing subject extraction out of passive and unaccusative verb phrases, subject raising out of VPE, and also wh-object extraction (see also Fox 2000 for covert movement out of the ellipsis site).5 (24)

Page 9 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

In section 21.4 we will see that head movement out of the ellipsis site is possible too: in languages with V-stranding VPE the lexical verb moves to T° from its base position. This section has provided a quick overview of the debate as to whether VPE involves hidden syntactic structure or not. Next, we have a closer look at the size of the ellipsis site, before turning to other types of predicate ellipsis.

21.2.4 The size of the ellipsis site In the introduction, predicate ellipsis was presented as a kind of ellipsis that targets the predicate of the clause, and the name VP-ellipsis suggests that it is the VP(-predicate) that is targeted. However, VPE does not necessarily involve a verbal predicate. Prepositional, nominal, or adjectival predicate phrases can be elided too, as (25) shows: (p. 512)

(25)

This suggests that VPE uniformly targets all types of predicates in finite clauses. But what exactly is to be considered part of the predicate is less straightforward. So far, we have seen that VPE elides more than just the (verbal, nominal, adjectival, or prepositional) predicate; it can also elide verbal complements and VP-adjuncts. What is not elided are the subject and the finite auxiliary in T°. However, more elements can be present in the clause between T° and VP. Within the Minimalist framework the VP-layer is standardly assumed to be dominated by a functional projection vP, which hosts the base position of the predicate’s external argument. Most analyses take VPE to include at least this vP-layer in the ellipsis site as well (see Johnson 2001b, 2004; Merchant 2001, 2008a, 2013d). One piece of evidence is found in there-existential sentences: the associate of there in unergative or transitive existential surfaces in spec,vP, but under VPE this associate is obligatorily elided, as shown in (26).6 (26) Page 10 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

Non-finite auxiliaries also help us identify the size of VPE. Consider the sentence in (27a), with the largest possible sequence of auxiliaries in English: (27)

Akmajian and Wasow (1975) and Sag (1976a), and more recently Thoms (2012), Bošković (2014), Sailor (2014), and Aelbrecht and Harwood (2015), observed that when VPE is applied to such a sequence, the following pattern emerges (from Sag 1976a: 31): (28)

This illustrates that perfect have cannot be elided (see (28a,b)), whilst been is optionally elided (see (28b,c)), and being is obligatorily included in the ellipsis site (see (28c,d)). We (p. 513) briefly review some accounts of this data pattern (see also Lasnik 1995b, Potsdam 1997b, and Nunes and Zocca 2009 for discussion of the behaviour of auxiliary verbs under VPE).7 The non-ellipsis of have is rather straightforward under most approaches: given that have always precedes forms of the auxiliary be, the former is taken to be higher in the structure than the latter, specifically in a position that is external to the ellipsis site. Consequently, have is never elided. Most accounts also assume, one way or another, that the position of being is always included in the ellipsis site.8 Bošković (2014) and Sailor (2014), following Akmajian and Wasow (1975), Iwakura (1977), Akmajian, Steele, and Wasow (1979), Lobeck (1987), Bošković (2004a), and Thoms (2011b), claim that being does not raise for inflectional purposes and instead has its inflection lowered onto it in its base position of Voice°, where it is subsequently elided (cf. (29a)). Aelbrecht and Harwood (2015), on the other hand, argue that being raises into a progressive aspectual projection in order to receive

Page 11 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis its non-finite -ing inflection, but that the ellipsis site includes this landing site: under their account VPE targets the progressive aspectual layer above VoiceP (cf. (29b)): (29)

The optional ellipsis of been has also received two possible accounts. Thoms (2012), Sailor (2014), and Aelbrecht and Harwood (2015) argue that been can optionally move out of the ellipsis site, thus optionally surviving VPE: (30)

On the other hand, instead of optional raising and a fixed ellipsis site, Bošković (2014), as well as Akmajian, Steele, and Wasow (1979), argues that the size of the ellipsis site can fluctuate: it does not contain been, but can be optionally extended to include the perfect aspectual projection, which Bošković (2014) assumes been has risen into for inflectional purposes:9 (p. 514)

(31)

Whatever analysis one adopts, bringing non-finite auxiliaries into the picture leads to the conclusion that the VP ellipsis site must be larger than simply VP or vP, extending as far as VoiceP, ProgP, or even PerfP. In addition, Sailor (2014) claims that the size of VPE in English differs depending on whether the ellipsis clause is coordinated with or subordinated to its antecedent. It should be clear by now that the last has not been said on this topic. This section has discussed the licensing and recoverability requirements of VPE in English, the question of whether the elided constituent involves syntactic structure or not, and the size of the ellipsis site in English VPE. The next section focuses on a different kind of predicate ellipsis that is standardly considered a subtype of VPE, but displays some notable differences: pseudogapping.

Page 12 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

21.3 Pseudogapping Pseudogapping is a kind of ellipsis that was first named by Stump (1977), who argued that it was similar to gapping (see also Levin 1978, 1986). Both kinds of ellipsis seem to target non-constituents (or discontinuous strings) of a clause. (32)

Because of this discontinuity, pseudogapping has been viewed as a subtype of VPE involving movement: in pseudogapping a contrastive phrase has moved out of the VP prior to the occurrence of VPE, hence surviving the ellipsis (see Jayaseelan 1990).10 This is illustrated in (33). (33)

Many have adopted this view and the main differences between the proposals are the motivation for the movement out of the ellipsis site and the direction of this movement (see Johnson 1996; Lasnik 1999a, 1999b, 2001; Kennedy and Merchant 2000a; Takahashi 2003, 2004; Merchant 2008a; Aelbrecht 2010; Gengel 2013).11 If this is on the right track, pseudogapping strengthens the argument in favour of syntactic structure inside the (p. 515) ellipsis site in VPE, as it instantiates another case of extraction out of that ellipsis site (see section 21.2.3). First, we discuss some basic properties of pseudogapping, especially those in which this phenomenon differs from VPE. Then we come back to the question of the movement operation responsible for extracting the remnant phrase out of the verb phrase.

21.3.1 Properties of pseudogapping Pseudogapping seems to be licensed by the same mechanisms that license VPE, namely either the modal verb, dummy do, or finite auxiliary verb; cf. (34). As (35) illustrates, finite main verbs or other aspectual verbs do not license it, parallel to VPE. (34)

Page 13 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis (35)

However, the distribution of pseudogapping is more restricted than that of VPE. First of all, pseudogapping gives rise to variable judgements depending on the sentence and the speaker. While all speakers accept it in comparatives and other subordinate structures, they are less widely accepted (Levin 1986: 99; Lasnik 1999a) and far less common in coordinations: in Miller’s (2014) corpus study of pseudogapping, coordinated pseudogaps constitute only 3.3 percent of all pseudogapping examples. Second, the infinitival marker to does not license pseudogapping, unlike VPE (cf. (36)). (36)

Thirdly, unlike VPE, pseudogapping generally does not allow for backwards ellipsis (examples from Levin 1980: 99):12 (37)

A fourth asymmetry traditionally believed to hold between VPE and pseudogapping is that the latter does not allow for voice mismatches, while the former does (cf. Merchant 2008a, 2013d; (p. 516) but also Sag 1976a; Dalrymple et al. 1991; Hardt 1993; Fiengo and May 1994; Kehler 2000, 2002; Johnson 2001b; Arregui et al. 2006; and Frazier and Clifton 2006 for more examples and discussion): (38)

Page 14 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis These data are brought forward by Merchant (2008a, 2013d) as supporting evidence for the claim that pseudogapping elides a larger part of the structure than VPE. He assumes ellipsis to be subject to a strict syntactic identity condition, which states that the morphosyntactic specification of the ellipsis site matches that of the antecedent. He explains the contrast between VPE and pseudogapping by arguing that VPE elides vP, thus not including the Voice head in the ellipsis site, while pseudogapping targets VoiceP. Because in the latter case, the Voice head is part of the ellipsis site, the voice specification in the ellipsis site has to match that of the antecedent.13 However, Miller (2014) has recently shown Merchant’s generalization to be incorrect, with attested examples from corpora, illustrating that voice mismatches are just as acceptable in pseudogapping as they are in VPE:14 (39)

As a result, the voice mismatch data no longer seem to bear on the debate surrounding the size of the ellipsis site involved in VPE and pseudogapping, and if anything, only highlight the syntactic similarity between the two phenomena. (p. 517)

As was pointed out above, most accounts argue that pseudogapping is a subtype

of VPE. In the mainstream generative analysis, a VP-internal remnant with contrastive focus is assumed to have moved out of the ellipsis site prior to the ellipsis. What the analyses disagree on mostly is what kind of movement is responsible for the evacuation of the remnant. The different approaches are presented in the next section.

21.3.2 The nature of the movement operation Pseudogapping is traditionally analysed as movement of an element out of the verb phrase plus VP-ellipsis (Jayaseelan 1990; Lasnik 1999a, 1999b, 2001; Merchant 2008a, 2013b; Gengel 2013). Crucially, this movement does not seem to take place in nonellipsis, or is at least disguised by further movement of the verb. In this section, we give an overview of the main proposals. For extensive discussion of the arguments for and against A- and A’-movement in pseudogapping we refer the reader to Johnson (2001b) and Gengel (2013). As a first proposal, Jayaseelan (1990) claims that the remnant is moved out of the ellipsis site by Heavy NP Shift (HNPS): it adjoins to the right of the clause, as in (40). (40)

Page 15 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis The examples below, however, show that the remnant is not necessarily a heavy NP (or PP): it can also be a pronoun, which cannot undergo HNPS (cf. (41a,b)). (41)

Furthermore, Takahashi (2003, 2004) argues against an approach with only HNPS on the basis of pseudogapping with multiple remnants. The sentence in (42a) (= Jayaseelan 1990: 66 (10)) cannot be derived by rightward movement of both remnants because multiple applications of HNPS are disallowed in non-elliptical contexts (cf. (42b) = Jayaseelan 1990: 66 (9); see also Gengel 2013, and see Shiobara 2010 and Park and Kim 2016 for alternative views on multiple HNPS). (42)

Moreover, Lasnik (1999a, 1999b) observes that unlike pseudogapping, HNPS is impossible with the first object of a double object construction: (43)

Thus, HNPS cannot fully account for moving the pseudogapping remnant out of the ellipsis site. Another proposal, advocated by Lasnik (1999a, 1999b, and subsequent work), (p. 518) argues that the remnant undergoes Object Shift (OS) to the left of the verb phrase (Vanden Wyngaerd 1989; Chomsky 1991; Koizumi 1995), parallel to Scandinavian OS. This movement is triggered by an optional feature that attracts the pseudogapping remnant when present. Lasnik additionally takes the main verb in English to move to a position preceding the object in non-ellipsis. Under pseudogapping such verb movement is bled and the verb is deleted with the rest of the VP. However, the movement in pseudogapping behaves differently from Scandinavian OS. Firstly, the latter does not target prepositional objects (see (44)), unlike pseudogapping in both English and Scandinavian (Gengel 2013). (44)

Page 16 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis (45)

Secondly, Scandinavian OS is blocked when an auxiliary is present in the sentence, i.e., it only occurs when the main verb moves out of the verb phrase (Holmberg’s Generalization, cf. Holmberg 1986). Given that in pseudogapping an auxiliary is crucially present, the two phenomena do not pattern together (see also Gengel 2013). In short, object shift does not seem to be responsible for the movement of a pseudogapping remnant either. Takahashi (2004) therefore proposes a hybrid account and argues that pseudogapping is derived by either HNPS or OS, depending on the context. Another alternative is put forward by Jayaseelan (1990, 2001), who observes that pseudogapping requires the moved element to bear contrastive focus. Gengel (2013) builds directly on this idea and argues that Focus movement is at stake in pseudogapping, not OS or HNPS. Gengel assumes that there is a clause-internal focus projection above VP that attracts the pseudogapping remnant to its specifier (following Jayaseelan 2001 for Malayalam and cleft constructions in English). For arguments in favour of such a focus position, see Belletti and Shlonsky (1995) for Italian and Hebrew, and É. Kiss (1998) for Hungarian. Summing up, there is much debate as to which operation is responsible for moving the pseudogapping remnant. Jayaseelan (1990) argues that Heavy NP Shift (HNPS) is involved, while Lasnik (1999a) claims the remnant to have undergone Object Shift (OS). Takahashi (2004) takes pseudogapping to be derived by either of these two operations, and finally, Gengel (2013) takes focus movement to be at stake (see also Jayaseelan 2001).15 The next section turns back to VPE, but extends the data set to languages other than English. (p. 519)

21.4 VPE cross-linguistically English has dominated the literature on ellipsis, and particularly VPE. Compared to other well-documented Western European languages, such as German, French, Italian, Spanish, and Dutch, English is the odd one out, as none of these other languages display VPE. However, that is not to say that VPE is restricted to English. It has also been argued to exist in Danish (Houser et al. 2007), Persian (Toosarvandani 2009), Finnish (Holmberg 2001), Galicean (Rouveret 2012), Hebrew (Sherman (Ussishkin) 1998; Doron 1999), Hungarian (Bartos 2000b), Irish Gaelic (McCloskey 1991a, 2004), Japanese (Otani and Whitman 1991), Korean (Otani and Whitman 1991; Kim 1997, though see Park 1997 for an opposing view), Libyan Arabic (Algryani 2012), Ndenduele (Goldberg 2005), Polish (Sczcegielniak 2006), Brazilian and European Portuguese (Raposo 1986; Martins 1994; Page 17 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis Cyrino and Matos 2002, 2005; Santos 2009b), Russian (Gribanova 2009b), Samoan (Sailor 2009), Scottish Gaelic (Sailor 2009), Serbo-Croatian (Stjepanović 1997), Swahili (Ngonyani 1996), Taiwanese (Sailor and Kuo 2010), and Welsh (Rouveret 2012). We focus in this section on what is known as V-stranding VPE languages such as (European and Brazilian) Portuguese, Hebrew, and Irish.16 An interesting property of these languages is that VPE only appears to target the material following the lexical verb, namely its internal arguments, and does not include the lexical verb itself: (46)

(47)

(48)

A complication is that some of the relevant languages also exhibit object drop. Portuguese, for instance, allows for a null object in (49a) (Raposo 1986). One could therefore argue that Portuguese does not in fact have VPE and that what looks like VPE in (48) is a null object construction. However, closer scrutiny reveals that in Portuguese, specific instances of ellipsis must be analysed as VPE. For instance, the typical null object cases only involve (p. 520) direct objects, while VPE elides all verbal complements; see (49b) (Raposo 1986; examples from Cyrino and Matos 2002). (49)

Page 18 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

Data such as these show that proper identification is a crucial step in VPE research: one needs to distinguish cases of object and topic drop from actual VPE. In this vein, Goldberg (2005) has argued that Japanese and Korean do not exhibit genuine V-stranding VPE, and that such cases can be ascribed to object and topic drop. Similar claims have been made for Russian (see Bailyn 2011; Erteshik-Shir, Ibn-Bari, and Taube 2011, 2013; and Gribanova 2013a, 2013b for discussion), though the matter is far from closed. Nevertheless, if the Hebrew, Irish, and Portuguese examples in (46), (47), (48), and (49b) are genuine cases of VPE, how does the lexical verb survive the ellipsis in such cases? Unlike English, these languages display main verb movement: when the main verb is finite, it raises to the T°-head to receive its inflection. Earlier we have seen that both Aand A’-movement are allowed out of the VP ellipsis site. The data in (46), (47), (48), and (49b) illustrate that the same holds for head movement: when the verb undergoes V-to-T movement, it raises out of the ellipsis site and so survives predicate ellipsis. Therefore these languages are called V-stranding VPE languages (see also Goldberg 2005).17 It is worth noting, however, that the analysis presented here is a purely transformational approach. Many other frameworks would not necessarily claim that languages such as Portuguese, Irish, and Hebrew actually exhibit VPE, though it remains to be seen exactly how those frameworks would tackle such data. The fact that the main verb can survive the ellipsis in these languages brings along a new identity restriction. Goldberg (2005: 171) calls this the Verbal Identity Requirement: (50)

This is illustrated in the Irish example in (51), from Goldberg (2005: 168). The main verb thuig has moved out of the ellipsis site, and is therefore recoverable, but still VPE is disallowed, because the verb needs to be identical to the verb in the antecedent.18 (p. 521)

(51)

Page 19 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

No such requirement applies to phrasal movement out of a verbal ellipsis site, as the following examples with subject and object extraction illustrate: the subject in (52a), Summer, differs from the subject in the antecedent, Clara; and the same holds for the topicalized objects in (52b). (52)

This contrast in identity requirements between phrasal movement and head movement out of the ellipsis site poses an interesting puzzle that is the subject of ongoing debate in ellipsis research. It should be noted, however, that counterexamples to the VIR have been noted in the literature. Gribanova (2013b: 119), for instance, notes the following data from Russian in which the stranded lexical verb in the ellipsis clause does not match the lexical verb in the antecedent: (53)

It seems therefore that the VIR is not an entirely robust generalization for all V-stranding VPE languages.19 A closely related phenomenon to V-stranding VPE is that of little v-stranding VPE, as discussed in Toosarvandani (2009) for languages such as Persian (see also Chapter 35 of this handbook). This language is argued to exhibit complex predicates consisting of a light verb in v°, and a nominal, adjectival, or prepositional element in its complement: (54)

Toosarvandani (2009) argues that VPE exists in Persian, except it does not affect any verbal elements within the clause, only the nominal, adjectival, or prepositional element

Page 20 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis of the complex predicate, and its internal arguments.20 The light verb is stranded by such ellipsis (example from Toosarvandani 2009: 61): (p. 522)

(55)

Unlike the main verb in V-stranding VPE, however, v is not stranded by raising out of the ellipsis site. Using the distribution of the adverb dobāre (again) as a diagnostic for the position of the light verb, Toosarvandani (2009) argues that it does not undergo v-to-T movement, and instead remains in its base position. For the light verb to survive VPE this would imply that in Persian, only the complement of v° is elided. vP itself is not the target of VPE. This shows that VPE in Persian targets a much smaller constituent than in English, and, more generally, that the size of VPE can differ cross-linguistically. This section expanded the VPE data set to languages other than English. In section 21.5, we deal with instances of predicate ellipsis that have only recently been getting more attention, namely Modal Complement Ellipsis and British English do.

21.5 Other predicate ellipses: Modal Complement Ellipsis and British English do 21.5.1 Modal Complement Ellipsis The previous section has shown that VPE is not restricted to English; it also occurs in Hebrew, Irish, Swahili, Portuguese, and several other languages. It is remarkable, then, that VPE is unattested in many languages related to these, such as German and Dutch for the Germanic languages and French, Italian, and Spanish in the Romance family. The question of why VPE has such a limited distribution is one of the most prominent and long-standing issues surrounding this construction. However, these languages display a type of predicate ellipsis following modal verbs. The phenomenon was first discussed in Busquets and Denis (2001), Depiante (2001), Cyrino and Matos (2002), and Dagnac (2010) for French, Italian, and Spanish, and Aelbrecht (2010) for Dutch, under the names of Modal Ellipsis (or ME) and Modal Complement Ellipsis (or MCE). The properties of M(C)E differ slightly between these languages, and this section focuses on the Dutch variant. In Dutch MCE, the infinitival complement of root modals can go unpronounced, as in (56): (56) Page 21 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

Unlike VPE, MCE only occurs with modals, not with aspectual auxiliaries (see (57a)), and even epistemic modals do not allow for the ellipsis (cf. (57b)). (p. 523)

(57)

When it comes to the ellipsis site, the examples in (58) show that MCE targets not only the main infinitive and its complement, but also the aspectual (cf. (58a)) and voice (cf. (58b)) auxiliaries: (58)

Page 22 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis Curiously, MCE seems to allow for voice mismatches, even though it elides the Voice head (as well as the aspectuals dominating VoiceP; see Aelbrecht 2010): (59)

(p. 524)

This presents a further problem for Merchant’s (2008a, 2013d) analysis of voice

mismatches in VPE and pseudogapping (see section 21.3.1), since it cannot be claimed that voice mismatches are permitted only if the voice head is situated external to the ellipsis site. What is remarkable about MCE is that extraction out of the ellipsis site is allowed, but restricted in a way that extraction out of VPE is not (see Aelbrecht 2010; Dagnac 2010). It turns out that only subjects—including derived subjects originating in object position— are able to move out of the to-be-elided constituent, not objects. This is illustrated in (60) for object extraction, with (60b) indicating that the ungrammaticality is due to ellipsis: the non-elliptical counterpart of this example is fine.21 (60)

The example in (61) displays subject extraction, here with the derived subject of an unaccusative verb (but see also (58b) for a derived subject of a passive).22 (61) Page 23 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

These data are relevant in light of the debate around the structure of the ellipsis site. The ban on object extraction at first hints at a proform analysis, without a full syntactic structure, but that would leave the subject extraction facts unexplained. Aelbrecht (2010) develops a theory of ellipsis licensing which captures this contrast by proposing that the timing of the ellipsis is crucial: only elements that have moved out of the ellipsis site before the licensing head is merged can escape ellipsis (see Aelbrecht 2010 for the details). As it turns out, MCE is not the only type of predicate ellipsis to display this property of limited extraction. British English do does as well, as the next section shows. (p. 525)

21.5.2 British English do British English do (henceforth BE do) is a phenomenon reminiscent of English VPE, and it is discussed by Chalcraft (2006), Haddican (2007), Aelbrecht (2010), Thoms (2011b), and Baltin (2012). In the examples in (62), the verb phrase run the race is elided and there is a finite auxiliary present. Hence, in the VPE sentence in (62b) there is no need to insert a dummy do. In British English on the other hand, an additional non-finite do can appear, as in (62a). (62)

This do can only occur when the verb phrase is elided: (63)

Parallel to what we have seen for MCE, BE do yields mixed results when the extraction test for internal structure of the ellipsis site is applied to it (see Aelbrecht 2010, Baltin 2012): object extraction is banned (unlike in the VPE counterparts, as in (64)). Moreover, pseudogapping is disallowed with BE do (see (65a)). Subjects, on the other hand, can be extracted from the ellipsis site, as in (66). (64)

(65)

Page 24 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis

(66)

Aelbrecht (2010) applies the analysis proposed for MCE to BE do to capture this contrast. Hence, these recently emerged types of predicate ellipsis shed some new light on issues that have been at the centre of the VPE literature for years, such as the question of internal structure and the value of the extraction test.

21.6 Concluding remarks Predicate ellipsis is a cover term for different types of ellipsis targeting the predicate part of a clause. The types presented here are VP-ellipsis (the most widely discussed elliptical phenomenon), pseudogapping, verb-stranding VP-ellipsis, modal complement ellipsis, and British English do. We have seen that these types display differences in behaviour and distribution, and it has become apparent that the question of what exactly is elided is not always easy to answer.

Notes: (1) Section 21.2.4 deals with non-finite auxiliaries. (2) We indicate the ellipsis site with an underscore here, for clarity’s sake. (3) The asterisk here indicates that an epistemic reading is unacceptable. A deontic interpretation is fine. (4) Here we only present the two opposite ends of the spectrum. Some structural accounts exist in which there is only a limited amount of hidden syntactic structure involved (see e.g. Hardt 1993). (5) Jacobson (1992), Hardt (1993), and Lobeck (1995) argue that these cases do not involve extraction at all, but are in fact pseudogaps with a fronted remnant. Under their account, VPE does not allow for extraction and therefore does not necessarily contain syntactic structure. However, as Johnson (2001b) points out, pseudogapping is more restricted than the extraction cases with VPE: not all phrases that can move out of a VPellipsis site can act as pseudogapping remnants. See Miller (2014), however, for critical discussion of Johnson (2001b). (6) Johnson (2004b) provides another argument in favour of this claim involving different interpretations of the adverb again. Merchant (2008a, 2013d) adds a third, showing that argument structure alternations are not allowed between the VPE site and its antecedent. Page 25 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis (7) The claim that perfect have is never elided under VPE is arguably too strong. Although the early references do not acknowledge it, subsequent literature on VPE has recognized that perfect have can be elided under certain circumstances and in some dialects (see discussion in Aelbrecht and Harwood 2015). In particular, Sailor (2014: 33–4) shows that in certain ellipsis–antecedent configurations, ellipsis of perfect have becomes possible, and is even preferred by speakers of particular varieties of English: ((i)) If you’d been there, he’d have got you one too, I know he would. (BNC) The opposite can be said about being: some authors have claimed that it can be stranded in certain dialects (see also Chapter 38 in this handbook). ((ii)) %Remember, always be respectful and courteous, even if the officer isn’t being. (8) See Baker, Johnson, and Roberts (1989), however, for an alternative account in which the ellipsis of being is ascribed to a ban on ellipsis following V+ing forms in general. See Aelbrecht and Harwood (2015) for critical discussion of this approach. (9) Earlier work by Akmajian and Wasow (1975) represents a third option: they proposed optional Aux-deletion operations, which could occur separately from VPE. See Sag (1976a), however, for arguments against this two-operation approach to VPE size variability. (10) Toosarvandani (2009) and Potter (2014) have recently attempted to also analyse gapping as VPE. See Chapter 23 of this handbook for more discussion about gapping. (11) See Miller (1990, 2014), Hardt (1993), Lobeck (1995), and Thoms (2011b), however, for an alternative view. (12) Miller (2014), however, notes the following counterexample: ((i))

(13) Recall, however, the discussion in the previous section: since being is always elided, several authors have claimed that VPE targets at least as much as VoiceP. These approaches are incompatible with Merchant’s explanation for the voice mismatch contrast. (14) An additional complication is that voice mismatches in VPE are themselves not always acceptable (Kehler 2002; Kertz 2013; examples from Sailor 2014: 4): ((i))

Page 26 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis On the basis of these data Sailor (2014) claims that there are two types of VPE. When the ellipsis clause is coordinated with its antecedent, VPE includes VoiceP, and voice mismatches are disallowed. However, when the ellipsis clause is subordinate to its antecedent clause, VPE targets a constituent smaller than VoiceP, and voice mismatches are permitted. (15) See also Johnson (2001b) for the hypothesis that pseudogapping involves a movement operation that resembles Dutch scrambling, and Miller (2014) for general critical discussion of the movement analysis of pseudogapping. (16) Though there are differences between European and Brazilian Portuguese when it comes to their V-stranding VPE properties; see e.g. Cyrino and Matos (2002) for discussion. (17) Other languages that are only Aux-stranding, like English, are Taiwanese and Mainland Scandinavian. See Sailor (2009, 2012b) for general discussion. (18) The principle in (50) requires the two verbs to be identical not just in their root, but also in their derivational morphology. See Goldberg (2005: 169–71) for further illustrations of this. (19) See Lipták (2012), Rouveret (2012), Schoorlemmer and Temmerman (2012), and Gribanova (2013b) for more discussion. (20) Toosarvandani (2009) argues for the existence of VPE in Persian on the basis of identity requirements, extraction, and the lack of pragmatic control. (21) Note that Modal Ellipsis in Romance languages differs in several respects from Dutch (Germanic) MCE. According to Busquets and Denis (2001), Dagnac (2010), and Authier (2011), extraction is much less limited in Romance ME than in Germanic MCE (more than just subjects can be extracted), and voice mismatches are illicit in French (and not fully investigated in other Romance languages). This potentially has consequences for the size of the elided constituent in M(C)E and the licensing of the ellipsis operation. Whether MCE and ME constitute the same type of ellipsis remains an open question. (22) See Aelbrecht (2010) for detailed discussion as well as evidence that deontic modals behave like raising verbs in Dutch and, consequently, that the subject in these cases is extracted from the ellipsis site.

Lobke Aelbrecht

Lobke Aelbrecht obtained her PhD at the Catholic University of Brussels in 2009 with a thesis entitled ‘You have the right to remain silent: The syntactic licensing of ellipsis’. Her main research interests are ellipsis, VP topicalization and VP pronominalization, and the Dutch adpositional domain. In 2010, she published the monograph The syntactic licensing of ellipsis (John Benjamins).

Page 27 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Predicate Ellipsis William Harwood

William Harwood obtained his PhD at Ghent University in 2013 with a thesis entitled ‘Being progressive is just a phase: Dividing the functional hierarchy’. His main research interests include ellipsis, phase theory, idiomatic expressions, auxiliary verbs, aspect, verb movement, VP fronting, existential constructions, and relative clauses.

Page 28 of 28

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Oxford Handbooks Online Gapping and Stripping   Kyle Johnson The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.24

Abstract and Keywords The chapter looks at the main attempts to model two constructions: gapping and stripping. It reviews the ways in which these constructions are similar, exploring the thesis that they both stem from the same underlying process. One of those shared properties is their restriction to coordinations—a restriction that is not found elsewhere in the menagerie of ellipsis constructions. Methods of tying this restriction to the similarly gapping/stripping-specific constraint against embedding the ellipsis or its antecedent are examined. The chapter focuses on the shortcomings of previous attempts to explain these conditions, and their counter-examples, and points to new directions of analysis that seem promising. Keywords: gapping, stripping, ellipsis, coordination, movement, small conjuncts, large conjuncts

Kyle Johnson

23.1 Introduction THIS chapter looks at two constructions which are sometimes characterized as involving ellipsis. One is known as gapping, a name Ross (1970) suggested. And the other is variously known as stripping or Bare Argument Ellipsis.1 An example of gapping is (1), and stripping is illustrated by (2). (1)

(2)

Page 1 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping The second conjunct of (1) is understood to have the same verb that the first conjunct has: likes. Similarly, in the second conjunct of (2), there is understood to be both a verb and a subject, each identical to those found in the first conjunct: Jones likes. These constructions are sometimes alleged to have the same source. They each involve material in a conjunct that is allowed to go unexpressed when there is identical material in the preceding conjunct. Gapping and stripping are typically described as being found only in conjunctions or disjunctions. Cases like (1) and (2) contrast with examples such as (3) and (4). (3)

(4)

(p. 563)

To the extent that this is correct, gapping and stripping fit the description of a

process that has been hypothesized to give conjunctions and disjunctions their meaning. This process is sometimes called “Conjunction Reduction,” which allows the mappings in (5). (5)

Hankamer (1979) is the first attempt to reduce gapping and stripping to Conjunction Reduction. One fact such an account might be able to capture is the putative generalization in Hernández (2007) that for a language to have gapping it must use the same morpheme for coordinating DPs and clauses. English, for instance, uses and for both clausal coordination and DP coordination, as do all the other Germanic languages, the Romance languages, Hindi, Persian, and Turkish. Yoruba, Wolof, Chinese, Korean, and Japanese, however, have one morpheme that coordinates DPs and another that coordinates clauses. The first group of languages all have gapping and the second does not.2 If Conjunction Page 2 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping Reduction cannot change the morphological form of the coordination morpheme, then this typology would be explained. An important component of Hernández’s survey is her assumption that gaps only arise in the second conjunct; she assumes that what Ross (1970) calls “Backwards Gapping” doesn’t exist. Ross argued that examples like the Japanese (6) arise from gapping the verb from the first conjunct. (6)

Ross suggested that whether a language has “Forward Gapping,” where the gap shows up in the second conjunct, or backwards gapping, correlates with whether the language has Verb–Object word order or Object–Verb word order.3 Hankamer (1971, 1979) and Maling (1972) argue, however, that examples like (6) are not gapping, but have a different source: Right-Node Raising. Right-Node Raising is a process that causes material at the right edge of an initial conjunct to be expressed in a conjunct that follows.4 Because languages whose VPs end with the verb will allow Right-Node Raising to apply to their verbs, Right-Node Raising and backwards gapping will produce the same surface strings. Ross’s observation that only verb-final languages have backwards gapping can be explained if all such cases are in fact Right-Node Raising. Hernández adopts this now standard view and for this reason doesn’t classify Japanese as a language that has gapping.5 (p. 564)

Reducing gapping to Conjunct Reduction predicts that it will be found only in

coordinations and disjunctions, but there are examples which resemble gapping and stripping in other environments. For instance, comparatives include cases like (7). (7)

And there are dialogues whose independent sentences, like those in (8), appear to arise from gapping and stripping. (8)

Page 3 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping For the most comprehensive investigation of gapping in the context of comparatives, see Lechner (1998, 2004). For a recent sketch of how gapping might arise in question/answer pairs, see Reich (2007), Boone (2014), and Weir (2014). Those analyses of gapping and stripping that tie them directly to coordinations would require these cases to have a different source. Gapping and stripping also have superficial resemblances to certain other ellipsis processes. Gapping, for instance, looks similar to “pseudogapping,” which (9) illustrates. (9)

Pseudogapping, given that name by Stump (1977), and investigated more extensively with corpora in Levin (1986), in turn resembles “VP-ellipsis,” exemplified by (10). (10)

The salient difference between pseudogapping and gapping is the presence of the word that stands in the highest position of what I’ll call the “verbal series.” In (9), that word is does; in non-tensed sentences, that word is a modal or the infinitival marker to. (11)

Gapping examples, by contrast, do not have this word. One strategy we’ll see is to make gapping the name we give to examples of pseudogapping in which the highest word in the verbal series is also elided. Stripping also looks very much like various forms of clausal ellipses. The short answers to questions like that in (12), for instance, can have the superficial appearance of stripping. (12)

See Merchant (2004a) for an account of (12) that involves ellipsis. There are also discourses like those in (13) which have a superficial similarity to (12). (p. 565)

(13)

Page 4 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Questions themselves can, in the right contexts, involve an ellipsis that appears like stripping. Ross (1969b) called these cases “sluicing”; examples are in (14).6 (14)

How to situate gapping and stripping among the panoply of ellipsis types remains open, as does the question of whether they belong together as two cases of the same process. In sections 23.2 and 23.3, we examine some of the problems that remain in understanding how to classify gapping and stripping.

23.2 Stripping One early analysis of stripping, in Ross (1967), is that it involves a rightward movement rule. This putative rule would move the string made up of and and the second conjunct to the right in the way indicated in (15). See Hudson (1976a, b) for an extended defense of this approach. (15)

Reinhart’s (1991) proposal is similar; it suggests that Quantifier Raising (QR) can create an unspoken, but semantically interpreted, representation by moving a nominal to the left of and, as illustrated in (16). (16)

Page 5 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping Under both views, the idea is to generate a representation that would make these sentences semantically equivalent to (17). (p. 566)

(17)

But the meaning of stripping sentences and sentences with conjoined nominals is not exactly the same, suggesting that they shouldn’t be equated. For instance, because the subject argument of (17a) and the object argument of (17b) are plurals, they can support semantic material that depends on a plurality. (18a,b) are therefore grammatical. (18)

This isn’t true of the stripping versions, however. (19)

Stripping has two “singular” conjoined clauses rather than one “plural” clause, like these accounts claim. There are other more syntactic problems with an approach that gives strips an underlying form like (17)—Depiante (2000) has a good list—but this is reason enough to abandon it.7 The other approach to stripping gives the second conjunct of a stripping construction its meaning through anaphora. This view is perhaps first defended in Ross (1969b), and given more extensive support in Hankamer and Sag (1976), Fiengo and May (1994), Heim and Kratzer (1998), Depiante (2000), Kolokonte (2008), and Algryani (2012). It assumes that there is a silent anaphor in the second conjunct whose meaning is recovered from the material in the first conjunct. If we represent that anaphor with ‘Δ’, the hypothesis is that stripping involves representations like those in (20). (20)

Page 6 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

This correctly captures the fact that stripping can involve the conjunction of singular sentences. If (20) is correct, the anaphor involved is one that Hankamer and Sag (1976) dubbed a “surface anaphor.” Surface anaphors get a denotation from their antecedent instead of referring independently to their antecedent’s referent. Thus, for instance, in (21), ‘Δ’ is understood to have a denotation that, like its antecedent, includes an indefinite. (21)

(p. 567)

As a consequence, the event of seafood eating that Smith and Jones participate in

doesn’t have to be the same one in (21), and the seafood they eat can, thankfully, be different. One way of deriving the fact that the anaphor in stripping is a surface anaphor is to let it be an ellipsis. Ellipses are uniformly surface anaphors, and this is typically derived by taking ellipses to be unspoken syntactic phrases. The antecedence conditions on ellipsis require that the elided phrase have a denotation that “follows” from the denotation of its antecedent. In the limiting case, the denotations of antecedent and ellipsis are the same. It is this view of stripping that dominates, and is the one we shall explore more fully here. If stripping removes a portion of the second conjunct whose meaning follows from that of an antecedent, then the words it elides should combine to have a meaning. This poses a problem for those examples in which the words that have elided don’t seem to form a constituent, as in (20b). If the string ‘Smith ate yesterday’ which has elided in (20b) is to form a meaningful constituent, we should expect the spoken form of (20b) to be (22). (22)

But the word order in (22) is generally judged ungrammatical, though the meaning it has is precisely the one needed in (20b). There are cases of stripping and, as we shall see, gapping too, in which the meaningful constituent that is alleged to have elided is difficult to find among the corresponding ellipsis-less sentences of English. This is a cloud that hangs over the view that strips are kinds of surface anaphors. If they are anaphors, then the material that has elided must form a semantically contentful Page 7 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping phrase, and in many cases the phrase cannot be found in unelided sentences. This occasionally encourages the idea that what has elided is merely a collection of words, and that how those words compose semantically is irrelevant. On this view, what would have elided in (20b) are the words Smith, ate and bread independently. Williams (1997) proposes a system like this and Ackema and Szendrői (2002) employ it to analyze cases of determiner gapping that we will see in (49). There are two problems that lie in taking up an approach along these lines, however. First, it needs to be prevented from running amok. In the majority of cases of stripping, the set of words that elide is very nearly just those that can compose into meaningful phrases. From (23), for instance, only (23a) is a possible strip. (23)

This follows if only strings that form meaningful phrases can strip. A second problem is that there are cases which suggest that it’s not just the words that make up a strip and its antecedent that matter, but also how they combine semantically. For instance, (24) is four ways ambiguous. (24)

The words ‘read the riot act’ have an idiomatic meaning, roughly equivalent to ‘chastise’, as well as their compositional meaning: to read that famous 1714 act of the British Parliament to a riotous assembly in order to dispel it. Both meanings are available to each of the conjuncts in (24), and they can be independently chosen. But this is not the case for the stripped version of (24).8 (p. 568)

(25)

Whatever meaning is given to the words ‘read the riot act’ in the first conjunct must be applied to those elided words in the second conjunct. Strips seem to depend on how the elided words are put together. We should restrict the groups of words that can strip to just those that make phrases. A virtue of taking strips to be ellipsis is that it provides a ready-made analysis of the cases where the strings that have elided do not seem to make a meaningful phrase. Because elided phrases are just those phrases unspoken, it is possible for material within them to Page 8 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping move out. This provides a way of characterizing the strings that have elided in stripping constructions that is consistent with normal views of constituency. The sentence in (23a), for instance, could arise from moving the object out of the elided constituent, as indicated in (26). (26)

This correctly predicts that the elided material in a strip should not contain an island from which the remnant moves. The ungrammaticality of the stripping examples in (27) matches the ungrammaticality of the island violations in (28). (The sentence-final adverbs in the first conjuncts of (27) must be understood as belonging to the root clause, otherwise we cannot be sure that the elided phrase will contain the island.) (27)

(28)

Island effects are diagnostic of movement, and their presence in stripping implicates movement in forming the strings that elide. (p. 569)

There are different kinds of movement operations, however, and they are sensitive to slightly different locality conditions. If the movement of the remnant out of the phrase which elides in stripping is like the movement of the interrogative pronouns which form the questions in (28), then we should expect the phrases that can elide under stripping to be just those that this kind of movement can escape. We should expect, for example, the strings traversed by movement in (29) to be strippable. (29)

Page 9 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

But that doesn’t seem to be the case; (30a,b) are degraded. (In these examples, in a letter must be understood to be part of the root clause, otherwise the stripped material would include only the embedded clause.) (30)

If movement of the remnant is what is responsible for producing the constituents that stripping can elide, then the type of movement operation involved must be different than that used to form constituent questions. To identify what movement operations are implicated, one might try to match the constraints on these strings with the constraints that identify different kinds of movement operations. Indeed, there is a rough match between the strings that can strip and those that can elide in pseudogapping (= (31)). (31)

A popular account of pseudogapping (see Jayaseelan 1990, Lasnik 1999a, 2003, and Aelbrecht 2010) is that it is simply VP ellipsis, from which the phrase that doesn’t elide has moved. Takahashi (2004) argues that the operations responsible for moving the remnant out of a pseudogap are “Heavy NP Shift” and “Object Shift.” These two operations obey Ross’s islands and also cannot move things out of (most) clauses, so they have the same properties we’ve seen are necessary for stripping. This also correctly captures the fact that pseudogapping cannot strand a preposition or move a predicate, as (32) shows. (p. 570)

(32)

Page 10 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Neither Heavy NP Shift nor Object Shift can strand prepositions or move predicates. Interestingly, the grammaticality judgments for stripping a preposition without its complement, as Kuno (1976: 301, n. 3) first noted, are not clear-cut. There are speakers who find (33) grammatical. (33)

Depiante (2000) argues that the cross-linguistic grammaticality of examples in which the preposition is included in the strip, leaving its complement behind, correlate with the ability of the language to strand a preposition with movement. She finds that English examples like (33) contrast with Spanish examples like (34). (34)

Unlike English, Spanish doesn’t allow preposition stranding under any kind of movement. (35)

Similarly, I believe stripping has the ability to strand predicates too. I and my consultants find examples like (36) grammatical. (36)

Page 11 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

If stripping has the ability to strand prepositions and predicates, and pseudogapping doesn’t, then a movement account of the strings that can elide must involve more movement kinds than just Heavy NP Shift and Object Shift: the movement operations that allegedly feed pseudogapping. One popular view is that stripping employs a movement operation that is tied to the information structure of the construction. Depiante (2000), Kolokonte (2008), and Algryani (2012), among others, argue that the movement involved is to a position reserved for focus-marked phrases. This responds to the observation that the material left over from the strip is focus-marked, and the fact that in many languages, focused phrases are moved leftwards to either the edge of the local clause or the edge of the local VP. (See, e.g., Rizzi 1997, 2004 and Belletti 2004.) While this movement doesn’t occur overtly in English, we might speculate that it is nonetheless available. Adopting the view that the focus position is at the left edge of VP, this would give to examples like (37), representations like (38). (37)

(38)

Page 12 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Brown has moved to the Specifier of a Focus Phrase in (38), leaving a variable ‘(=t)’ in the position it moves from. Stripping would presumably elide the boxed VP in the second conjunct. Recall that ellipsis is allowed just when the denotation of the elided constituent matches, or follows from, the denotation of an antecedent constituent. This requirement is not trivially satisfied in (38): the constituent that elides has a trace of the moved phrase Brown in the same place that the antecedent has the DP Jones. One way of addressing the issue is to recognize that Jones in (37) is focused too, and allow it to covertly move, forming the representation in (39).9 (p. 572)

(39)

Page 13 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

It is known from other kinds of ellipses that phrases with variables in them can be in the right matching relationship for one to serve as the antecedent to the other even if those variables are bound by different things. When an elided phrase, and its antecedent, have different bound variables in them, we have a situation that Ross (1967) called “sloppy identity.” Another solution would be to use an antecedence condition on ellipsis that allows focused phrases to behave like bound variables and partake in sloppy identity without moving. That would allow the focus-marked Jones in (38) to “match” the trace of Brown. Merchant (2001) offers an account of this type. If movement to a focus phrase can make strippable strings, we should understand why this normally non-overt movement becomes possible in ellipsis contexts.10 And if it is to give us a handle on the locality conditions—like the inability of strips to include part of an embedded finite clause—then we should be able to discover that this locality condition holds of covert focus movement too.

Page 14 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping This is presently the most popular view of stripping. It is a form of ellipsis that results when a movement operation removes a focused item from the ellipsis. Many of the considerations that lead to this conclusion show up again in gapping. Indeed, certain of the properties that seem diagnostic of stripping—that it is confined to certain kinds of coordinations, for instance—are found in gapping too, fueling the hypothesis that they are the same phenomenon. (p. 573)

23.3 Gapping Like stripping, gapping is a surface anaphor, as can be seen from the fact that two books can be involved in (40). (40)

This example also demonstrates that the verb gapping elides can be part of a longer string. Just as with stripping, it is necessary to find a way of characterizing the strings that can gap, and there is a rough match with the strings that can strip. Neijt (1979) showed, for instance, that island effects are obeyed. (41)

As with stripping and pseudogapping, then, we can hypothesize that the second phrase left behind by gapping has moved from the constituent that elides. Neijt also discovered that, as in stripping, the constraints on gapping are stricter than the islands which govern wh-movement would predict. For instance, the remnant of a gap cannot have moved out of an embedded finite clause, though, as we’ve seen, that is possible for wh-phrases.11 (42)

Most of the literature takes gapping to be like pseudogapping in not permitting preposition stranding. Lasnik and Saito (1991), Jayaseelan (1990), and Abe and Hoshi (1997) characterize examples like (43) as ungrammatical. (43) Page 15 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Steedman (1990), by contrast, provides examples of preposition stranding in English that seem perfectly grammatical, such as (44). (44)

(p. 574)

Vanden Wyngaerd (2009) argues that the availability of preposition stranding in

gapping tracks a language’s ability to strand prepositions under movement. I will assume that gapping in English does permit preposition stranding, but it is worth noting that this requires more investigation. Predicates are also capable of being stranded by gapping. (45)

If the remnants of gapping can include predicates and separate prepositions from their complements, then the class of remnants for gapping matches that for stripping. Like the remnants in stripping, the remnants of gaps are also contrastively focus-marked. Kuno (1976) argued that the remnants of a gap must present discourse-new information, and Hartmann (1998) and Winkler (2003, 2005, 2006) show that the remnants of a gap must be in a contrastive focus relationship to matching material in the antecedent. If the explanation for this fact in stripping is that the remnant focus moves to a position outside the ellipsis, then we should be tempted to give the same explanation for gapping. An important question to be settled is what are the focus structures of gapping and stripping and to what extent they can be wedded to a theory of what remnants are possible. This is especially well discussed in Susanne Winkler’s work. There is an interesting argument in Yoshida (2005) for the more general conclusion that the remnants of a gap have moved from the elided phrase. His argument is based on the generalization that the contents of phrases that have moved cannot themselves be moved. For instance, if (46b) is derived from (46a) by rightward movement of the PP, then the contrast in (47) indicates that the PP becomes impenetrable for movement in its derived position. (46)

Page 16 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (47)

Using this as a diagnostic for movement, Yoshida argues that the remnant of a gap has moved. His centerpiece example is (48). (48)

In (48a), which topic has moved in across-the-board fashion out of a PP that resides in its unmoved position. (“Across-the-board movement” describes a scenario in which one term moves from two parallel positions in conjoined or disjoined phrases. See Ross 1967.) The contrast with (48b) suggests that when the verb gaps, the PP that remains has become an island. If this PP moved, then this will follow from whatever is responsible for the contrast in (47). As in the case of stripping, then, there is evidence that the strings which can gap correspond to those that are phrases from which something has moved. There are, (p. 575)

however, a class of cases discovered by McCawley (1993) which seem to be clear counterexamples. These are cases where a determiner seems to gap leaving the rest of its nominal behind.12 (49)

If the gapped predicate is restored in these examples, they either become ungrammatical, or acquire a different meaning. (50)

Page 17 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping The grammatical (50a,b) and (50e) differ semantically from the corresponding examples in (49) in having bare plurals as the subject of the second conjuncts. In the matching (49) examples, the subject of the second conjunct is understood to have a determiner: too many in (49a), for instance. The absence of prenominal material in (50c) and (50d) is what makes these cases ungrammatical. McCawley’s conclusion is that gapping has removed the determiner from the subject of the second conjunct in (49). While that might be a way of characterizing the examples in (49a–c), which are synonymous with the examples in (51), it doesn’t extend to (49d–e) which aren’t synonymous with (52). (51)

(52)

Instead (49d) and (49e) have the meanings indicated in (53). (53)

(p. 576)

The negation part of the meaning of no and few falls outside the scope of the

disjunction. McCawley noted a similar fact about his (54). (54)

(54) is not a disjunction of denials (‘Not enough linguists study Russian or not enough literary scholars study French or…’) but a denial of disjunctions (‘It’s not the case that enough linguists study Russian or enough literary scholars study French…’). An account of these cases should derive the fact that the negation part of these determiners must be outside the coordination. Johnson (2000a,b) suggests that what is happening in (49d) and (49e) involves a more complicated syntax for the determiners no and few. If we adopt the view in Bech (1983) that the determiner no is the exponent of two morphemes, one of which has the meaning of not and the other the meaning of any, then we can see (49d) as involving a surprising

Page 18 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping way of spelling out that exponent. Suppose, for instance, that we assume no to arise when the not and any parts of the meaning are adjacent at some point of the derivation. (55)

If (49d) is parsed as in (56), then (55) causes the ψ in the first subject to be pronounced as no, but leaves the ψ in the second subject silent. (56)

A parallel syntax–semantics can be put together for (49e); few could be seen as the exponent of a morpheme that means not and another that means many. Because this account lives on the complex two-part syntax of the determiners, it predicts that the only prenominal material that should be able to “gap” in this way are those terms which have this two-part syntax. McCawley discovered that neither prenominal adjectives nor the determiner a is able to gap. (57)

Many other determiners, however, can. (p. 577)

(58)

Page 19 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping The list of terms that can gap roughly match the class of items that trigger Quantifier Raising (QR)—the rule that fixes the scope of quantificational determiners. In some analyses of Quantifier Raising, the determiner involved does indeed have two syntactic parts: one that makes the DP involved a variable and the other that creates a binder for that variable. One possibility, then, is that the complex syntax of determiners quite generally is responsible for their ability to appear to gap. The implementation of this idea won’t be straightforward. It’s not clear, for instance, how to get it to remove the possessive pronoun in (49c), or how to reconcile the ability of indefinite DPs headed by a to scope in the way that QR predicts but not be able to gap. A fact that may follow from an account along these lines is that a determiner may gap only if it is in a position consistent with it being at the left edge of the coordination. McCawley points out, for instance, that a determiner associated with an object cannot be gapped. (59)

But if that object is at the left edge of the coordination, then its determiner can gap. (60)

If the relationship between the two parts of these complex determiners is part of the syntax of QR, then one way of viewing this fact is that it follows from conditions on across-the-board movement. Williams (1977a) shows that across-the-board movement is subject to a condition that requires the moved item to come from parallel positions in the coordinations. For instance, an item cannot move across-the-board from a subject position in one conjunct and an object position in the other conjunct; examples like (61) are ungrammatical. (61)

If the relationship between the part of the complex determiner that is outside the coordination (the ϕ in (55)) and the parts of the determiner within the coordination (the Page 20 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping ψs in (55)) is subject to the same condition, then the parts of the determiner inside the coordination will have to be in parallel positions. Because the two parts of the determiner must be adjacent to get mapped onto their exponent, the DP in the left conjunct will have to be at (p. 578) the left edge of that conjunct. This will force the DP in the right conjunct to also be at the left edge of its conjunct. On this view, the determiners have not gapped in (49). Instead, the coordinations allow for a syntax that reveals the otherwise hidden complex relationship between the morphology of a determiner and its syntactic/semantic components. If that is the case, then we must understand why the coordinations in (50) without gapping do not allow this revelation: what is it about gapping that allows a determiner to be unexpressed? Levine and Kubota (2013) argue that gapping is not required, but instead that a kind of parallelism which gapping is a special instance of is the relevant requirement. They offer the examples of successful determiner gapping without any obvious gap in (62). (62)

Similarly, Centeno (2011) finds that wh-determiners in Spanish can also gap without an accompanying predicate gapping. (63)

Note that in both of these examples, the tenses of the conjoined clauses are the same, and at least for the English cases, this looks like a requirement. So whereas (62) and (64) are grammatical, cases like (62) where the tenses are different are worse. (64)

(65)

Page 21 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping A feature of gapping that Hankamer (1979) discovered is that the tenses must be the same. In (66), he noted, the conjuncts are either both present or both past.13 (66)

If, as is standardly assumed, the highest functional head in a clause is the one that carries the tense morpheme, then we might see this sameness in tense as a sign that gapping has applied (p. 579) to just this tense morpheme, leaving the rest of the clause behind. Perhaps, then, the gapped determiners in (62) and (63) are accompanied by a gapped portion of the verb: its tense.14 If this method of analyzing gapped determiners is successful, then our working hypothesis can be that a movement operation is responsible for forming the strings that gap. Those same movement operations also form the strings that strip. Gapping and stripping are then cases of remnant ellipsis, just like pseudogapping, but with a wider inventory of movement rules. If the conclusion that gapping cannot elide material in the first coordinate is correct, then this also matches the inability of a pseudogap to be in this position: (67)

These similarities encourage the view that gapping, and stripping too, are kinds of remnant ellipsis constructions. A salient difference between gapping and pseudogapping is that the highest term in the verbal sequence elides in gapping but not pseudogapping. This is related to another property that distinguishes gapping and stripping from pseudogapping. If gapping or stripping occurs in the second of two conjuncts, then it must elide the highest term of the verbal sequence in that conjunct. Examples like (68) are ungrammatical. (68)

Their pseudogapping and VP ellipsis correlates, by contrast, are fine: (69)

Page 22 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping Hankamer (1979: 20), who discovered the constraint, calls it “downward bounding.” I’ll formulate it with (70). (70)

Downward Bounding is violated in (68) because thought is not included in the ellipsis. The Downward Bounding Constraint has inspired one class of analyses of gapping that sometimes goes by the name of the “small conjuncts” account. (p. 580)

23.4 Accounts 23.4.1 Small conjuncts account The small conjuncts account gets started with Siegel’s (1987) observation that some cases of gapping can be captured with no ellipsis at all. In these cases, just the highest item in the verbal sequence is missing from the second conjunct.15 (71)

Siegel argues that (71) should be given a representation that, in today’s syntax, would look like (72).16 (72)

Page 23 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

This syntax assumes that the Specifier of TP requires a subject argument to move into it. For the account illustrated in (72) to work, there must be a constraint on this movement that would prevent the subject of the second conjunct from moving into the Specifier of TP, for otherwise we would expect the ungrammatical ‘Jones can’t Smith make natto or eat it’ to arise as well. The movement of the subject in (72) appears to violate John Ross’s Coordinate Structure Constraint, which prohibits movement out of a single conjunct. But Lin (2001) argues that (72) satisfies the Coordinate Structure Constraint, once it is formulated properly. She (p. 581) follows Ruys (1992), who argues that the Coordinate Structure Constraint is a condition that holds of binders rather than movement. We can formulate it with (73). (73)

If movement of Smith in (72) leaves a variable bound by Smith, this condition is violated. Lin argues that in contexts like (72), the movement of the subject is semantically vacuous, leaving no bound variable in the first conjunct. This can be seen by considering examples like (74). (74)

Page 24 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping If the first conjunct of (74) occurs on its own, it is ambiguous. (75)

The two readings of (75) depend on the relative scope of the subject and can’t. When Bob or Mary scopes outside of can’t, the meaning can be paraphrased by (76a). When it scopes within can’t, the resulting meaning can be paraphrased by (76b). (76)

Lin’s consultants report that (74) only permits a reading in which the first conjunct has the meaning paraphrased by (76b). That would follow if the movement of the subject in (76) is forced to be semantically vacuous because of the Coordinate Structure Constraint in (73). The syntax in (72) seems like a viable analysis of (71), then. Siegel (1987) argues on its behalf from the observation that hasn’t and can’t in (77) scopes over the conjunction. (77)

This is just what the constituency in (72) predicts. There are, however, apparent counterexamples to this prediction studied by Hulsey (2008), that involve disjunctions. One of these is (78). For the Red Sox to make the playoffs… (78)

This sentence is ambiguous. It has a reading in which must scopes over the disjunctions in line with the meanings found in (77). On this interpretation, (78) describes a requirement that can be (p. 582) met in one of two ways: what must happen for the Red Sox to make the playoffs is either that they beat the Yankees or that the Angels lose to the Mariners. This interpretation arises if must scopes over the disjunction. But (78) can also get an interpretation in which the disjunction scopes over must. It can be synonymous with The Sox must beat the Yankees or the Angels must lose to the Mariners. On this interpretation, (78) is expressing ignorance about which of two requirements hold; this Page 25 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping meaning allows the continuation: but I don’t know which. Hulsey explains this apparent counterexample by making appeal to an analysis of disjunctions that uses an “alternative semantics.” On this analysis of disjunctions (see Alonso-Ovalle 2006 and references therein), the semantic scope of disjunctions can be wider than their syntax alone would allow. Crediting the disjunct-wide reading of (78) to this analysis of disjunctions explains why we don’t find a similar wide-scope reading for conjunctions, whose scope is constrained more rigidly by the syntax.17 If Hulsey is right, the generalization that emerges is that the modal and negation which appears to be in the first coordinate is in fact outside the coordination. This supports Siegel’s proposal that these sentences have the structure in (72). Putting (72) together with the version of the Coordinate Structure Constraint in (73) predicts that when the second conjunct contains something that the moved subject of the first conjunct can bind, it will be able to do so. That is correct, as (79) shows.18 (79)

In this context, the Coordinate Structure Constraint is satisfied when (79) has the representation in (80) (no father binds a variable in both conjuncts: its trace in the first, and the pronoun his in the second). (80)

Page 26 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping And, indeed, (79) does have an interpretation in which his is a variable bound by no father. (That this is possible in gapping contexts is discovered by Oehrle 1987 and McCawley 1993 and part of Lin’s 2001 discussion.) (p. 583)

This class of cases, then, could be handled without ellipsis of any kind. The material that is shared by both conjuncts is simply outside the coordination; the appearance of ellipsis is achieved by letting the material that has conjoined be sufficiently small. If this is the only way gaps of this kind can be formed, then we also have an understanding of why they only arise in coordinations and why they obey Downward Bounding. An example like (81), which illustrates a violation of Downward Bounding, cannot be formed from a low coordination. (81)

If a way of spreading this treatment to all other gaps can be found, then we could derive these attributes for all gaps. Consider first how the low conjuncts account might be applied to a case, like (82), in which the highest term in the verbal sequence is the main verb. (82)

One approach to this case involves the hypothesis that main verbs move a short distance in English. Imagine, for instance, that the verbal root combines with the external θ-role assigner, ‘v’, and they jointly move to a higher head position, X0. This would allow a small conjuncts analysis of (82) like that in (83), where the complex made of the verbal root and v have, in each conjunct, moved in across-the-board fashion into X. (83)

Page 27 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

With the addition of verb movement and the presence of XP, the cases of gapping that Siegel described would have the slightly different analysis shown in (84). (p. 584)

(84)

Page 28 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

The coordinates are large enough in (84) that movement of the main verb does not remove them from the coordination. As before, the subject of the first conjunct moves into the Specifier of TP. But we must now also conjecture that the subject of the right conjunct also moves, for otherwise it would remain lower than the main verb of that conjunct. There is no satisfactory account for this. (But see the discussion of (117) in section 23.4.2.) This technique for producing a gap out of the main verb, in addition to an object which has shifted, preserves the account of Downward Bounding, as well as ensuring that gaps only occur in coordinations. We are left with the task of extending this account to more complex gaps, where more than just the highest term in the verbal sequence, plus possibly a shifted object, gaps. There have been two approaches: ellipsis and rampant across-the-board movement.

23.4.1.1 Ellipsis Adding ellipsis to a small conjuncts analysis generates many of the cases we’ve seen. In Coppock (2001), this ellipsis is just garden variety pseudogapping; in Lin (2002), it is a purpose-built ellipsis process. In either case, the remnant moves out of the phrase to be elided in the way we’ve reviewed in the previous sections.

Page 29 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping To see how this works, let’s walk through (85), which cannot be produced with just low coordination and movement of the main verb and/or object. (85)

We’ll start the derivation by Object Shifting the objects in each of the coordinated XPs. (I’ve called the phrase into whose Specifier objects shift: µP.) (p. 585)

(86)

The vPs in this representation are close enough a match for ellipsis to apply, forming (87). (87)

Page 30 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

The main verb will do its movement in the left conjunct, forming (88). (p. 586)

(88)

Page 31 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

This is the correct outcome. Because the highest term in the verbal sequence still gaps by way of a low coordination on this view, it preserves the effects we’ve seen credited to low coordinations: its restriction to coordinations and its compliance with Downward Bounding. It’s not that Downward Bounding is a constraint on gapping on this view, but rather that Downward Bounding names the environment where pseudogapping and small conjuncts come together to form the outputs we call gapping. When Downward Bounding isn’t honored, examples such as (89) arise, where the combination of small conjuncts and pseudogapping is obvious. (89)

This is the proposal in Coppock (2001).19 Note that if movement of the verb in the second conjunct could happen, we would expect to get (90).20 Page 32 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (p. 587)

(90)

In many of the world’s languages, it is possible to elide a VP from which the verb has moved (see Goldberg 2005) and (90) is just where we’d expect it to occur in English. This is a problem that comes with the small conjuncts analysis. I don’t know that this problem has been observed or addressed, so it is an open problem. There are other problems as well. The first is, simply, that it predicts that gaps which involve more than one term will not arise in contexts where pseudogapping isn’t licensed. This doesn’t appear to be correct. For instance, neither VP-ellipsis nor pseudogapping can occur after the auxiliary verb have in Icelandic, and yet gapping can occur in a sentence in which have is the highest term in the verbal sequence.21 (91)

Similarly, whereas VP-ellipsis and pseudogapping can occur in English to VPs and APs when they are embedded under modals, finite auxiliaries, and other similar terms (see (92)), they cannot when VPs or APs are complements to lexical verbs (compare (93)). (92)

(93)

Page 33 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Gapping, on the other hand, can apply to verbs and adjectives that are embedded as complements to lexical verbs, as (94) illustrates. (p. 588)

(94)

If pseudogapping is the source of a gap that includes more than just the highest term in the verbal sequence, this isn’t what we should find. Another problem for the pseudogapping-based account is one that Siegel (1987) notes. We have seen that when the highest term in the verbal sequence is all that gaps, the resulting semantic interpretation puts that term outside the scope of the coordination. The example in (95) has just the interpretation in which can’t scopes outside the disjunction. (95)

But when the highest term in the verbal sequence gaps along with the main verb, as in (96), the sentence can have a reading in which that term doesn’t scope out of the disjunction. (96)

Like (95), (96) has a reading in which can’t scopes over the conjunction; it can be synonymous with (96). But it also has an interpretation in which can’t is in each conjunct; a reading synonymous with Ward can’t eat caviar and his guest can’t eat dried beans. Neijt concludes from this contrast that gapping is something different than a mere low

Page 34 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping coordination. If she is right, (95) is low coordination and (96) isn’t—it is an independent process instead: gapping. A third problem is that the antecedence conditions on pseudogapping and gapping don’t seem to be the same. The antecedence conditions on gapping are, as Hernández (2007) points out, stricter than those for pseudogapping. Like VP-ellipsis, pseudogapping is able to have a deverbal noun as its antecedent, as in (97). (97)

See Fu et al. (2001) for an analysis. Gapping, by contrast, cannot; as (98) demonstrates.22 (98)

And a final problem is that there is a condition like Downward Bounding that holds of the antecedent for the gap in the left conjunct. That condition requires that the (p. 589)

antecedent for the gap not be embedded. It is responsible for the ungrammaticality of (99). (99)

(99) is only grammatical if the disjunction is interpreted as the complement to require; it cannot have an interpretation in which the disjunction is the one indicated by the brackets. Pseudogapping is possible in this context, as we can see from (100). (100)

If gapping is just a low coordination in tandem with pseudogapping, then (99) should be able to get the representation in (101). (I’ve suppressed here the irrelevant XP, µP, and the movements they incur.) (101)

Page 35 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

An ellipsis-based account of gapping doesn’t explain what Hankamer (1979), who first discovered this constraint, called the Upward Bounding Constraint: (p. 590)

(102)

Page 36 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping 23.4.1.2 Across-the-board movement In an attempt to solve these problems with the ellipsis-based account, Johnson (2009) proposes that gapping only involves across-the-board movement. Following proposals in Kayne (1994), Johnson suggests that the syntax of English can involve short movement of a verbal projection. A sentence like (103), for instance, can have a representation like (104). (103)

(104)

We can imagine that this is the phrasal version of movement of the single verb into X. Here, I’ll assume that English allows either head movement of the V to X, or phrasal movement of vP to XP, to bring the verb into its surface position.23 If this movement is available, then it can happen in across-the-board fashion in contexts where there are low coordinations, and this will produce gaps of verbal projections, not just verbs. For instance, the gap in (105) can be derived by across-theboard moving give a book as shown in (106). (p. 591)

Page 37 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (105)

(106)

The complete set of relevant cases aren’t given in Johnson 2009, however, and it is not transparent how the right word orders are always achieved—a problem that has been pointedly raised by Vicente (2010a) and Boone (2014). For instance, to form the gap in (107) requires gapping give and to me. (107)

That can be done in one of two ways on Johnson’s account. Either the verb and the PP independently move across-the-board, as in (108), or they move across-the-board as a phrase, out of which the contrasting objects have moved, as in the derivation in (109). (108)

(109)

In neither case is the word order in (107) achieved, and it is not transparent how to derive that word order. This is the same problem we saw in (37) for stripping, where the movement operations necessary to form the string that elides in the second conjunct cannot be mirrored by overt movement operations in the first conjunct’s antecedent to the ellipsis. On an ellipsis-based account, this can be solved by letting the movement operations in the first conjunct happen covertly. But that’s not a possible solution in the across-the-board account, since it is the overt form of the elided phrase that is spoken in the left conjunct on that account. The across-the-board account does, however, provide a way to capture the fact that a modal can be interpreted inside the coordination when gapping involves both the modal and the main verb. In that scenario, the phrase that moves across-the-board could be made to include the modal. If the standard view that modals surface in T is correct, then the easiest way of getting this to work would be to assume that modals have a lower underlying position from which they move. If that lower position is inside the phrase that Page 38 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping moves across-the-board, then representations like (110) could produce the desired interpretation. (p. 592)

(110)

The surface form could be produced from (110) by moving can’t from the phrase it heads (i.e., MP) into T. If movement of MP and movement of can’t into T are semantically vacuous, this syntax will achieve the desired reading, one in which can’t is interpreted in both conjuncts.24 The across-the-board account also derives Upward Bounding. Because the movement of vP to XP is by definition clause-bound, it will not be able to produce violations of Upward Bounding. It also doesn’t make the erroneous prediction that gapping of this type will only be possible in languages, and syntactic contexts, that have pseudogapping. Instead,

Page 39 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping it makes the more difficult to confirm prediction that it only happens in languages that have predicate movement of the sort that (104) illustrates. Both executions of the small conjuncts analysis harbor problems that have not yet been overcome. The across-the-board based analysis has its most serious difficulties in explaining the word orders that arise, and the pseudogapping-based analysis has its greatest difficulties in explaining why pseudogapping qua gapping becomes available in (p. 593) environments where it is not otherwise available. In addition to these problems there are various more particular problems of execution which we will review in section 23.4.2. But let’s highlight first the reasons for maintaining a small conjuncts analysis. It explains why gapping is found in coordinations, and not other sentential connectives. It also explains Downward Bounding, as we’ve seen, and if the across-the-board account is adopted, Upward Bounding as well. And, finally, it explains why the movement operations that seem to be responsible for forming the strings that can gap are just Heavy NP Shift, Object Shift, and the putative short focus movement, normally covert in English. If gapping can be fed by just these movement operations, and not by movement operations that can span longer distances, then we will explain why gapping and stripping cannot remove part of an embedded finite clause. How does the small conjuncts analysis restrict the class of movement operations that can form gaps and strips? The small conjuncts analysis traps these operations into structures where T remains outside coordinated phrases that contain a vP, as in (111). (111)

If the constituent that gaps or strips is vP, then the only material that can remain behind will have to reside somewhere in XP but outside vP. The only established movement operations in English that target this region are Heavy NP Shift, Object Shift, and presumably Focus Movement. The small conjuncts analysis therefore restricts the class of

Page 40 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping movement rules available to form gapped phrases by forcing those movement rules to be ones whose landing site is within the small conjunct. This predicts that the kinds of gaps a language has should vary with the inventory of short movement rules it possesses. Languages with short scrambling, for instance, should be able to use this rule to form a gapped constituent. We should also see differences cross-linguistically in what a gap can include that will track what material can be put between vP and XP. This might be the reason that German, as Repp (2006a, 2009) shows, can have gaps in which the sentential negator is interpreted in just the first conjunct. (112)

This could reflect the fact that sentential negators in German are low adverbs and can therefore be positioned within the first conjunct while still allowing the vP of that conjunct (p. 594) to serve as the antecedent for the gap. See Repp (2009) for a different approach, and for a discussion of the factors that influence the availability of this reading.25

23.4.2 Large conjuncts Let’s now consider an account that allows gapping and stripping to occur when two full clauses are coordinated: a “large conjuncts” account. An across-the-board account is incompatible with this possibility, so the gap or strip will have to be created by ellipsis. Because Heavy NP Shift and Object Shift do not move their phrases far enough out of the VP to escape an elided clause, it relies on Focus Movement that targets the left edge of the clause. A stripping example like (37), repeated here, will have the structure in (113). (37)

(113)

Page 41 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Ellipsis silences the boxed TP. A gapping example like (114) will get a representation like (115). (p. 595)

(114)

(115)

Page 42 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Note that both remnants of the gap are in a contrastive focus relationship to the parallel arguments in the first conjunct, and this is captured by moving both of them to the Specifier of Focus Phrase. Because there can be an indefinite number of remnants to a gap, we will have to allow there to be an indefinite number of focus positions to be moved into.26 These are the analyses advocated for by Depiante (2000) and Boone (2014). Something must ensure that the remnants of the gap are ordered in a particular way. It isn’t possible to move the remnants out of the gapped TP in (115) so that the object precedes the subject: Some ate beans and rice others is ungrammatical. A parallel requirement is needed for the small conjuncts account as well, since the remnants of the gapped vP on that account must be similarly ordered (witness the contrast in (116)). (116)

The generalization seems to be that the linear order of the gap’s remnants must be the same as the linear order of the phrases they are contrasted with in the antecedent. That is what paradigms like (117) imply.27 (p. 596)

Page 43 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (117)

As far as I know, this effect is unexplained. Allowing for a large conjuncts analysis gives an immediate remedy for several of the afflictions of the small conjuncts account. It permits an explanation, for instance, of Siegel’s paradigm concerning the scope of modals. Recall that when only the modal gaps, as in (77b), it necessarily scopes outside the conjunction. But when both modal and verb gap, as in (96), it can scope either outside or inside the conjunction. (77b)

(96)

If both a large conjuncts and small conjuncts syntax for gapping are available, then the ambiguity of (96) emerges. The wide-scope reading for can’t arises when it is outside the conjunction and the small conjuncts syntax is used. The narrow-scope reading for can’t arises when his guest and dried beans have focus-moved out of a TP and the large conjuncts syntax is used. The unambiguity of (77b) can be ensured if the large conjuncts syntax can be blocked in this configuration. That might be achieved by disallowing focus movement of VPs, which is what would be required if (77b) were to get a large conjuncts analysis. Allowing gapping and stripping to come by way of large conjuncts also removes the problem that gapping is possible in places where pseudogapping isn’t. Recall that there are languages which have gapping and stripping, but not pseudogapping (e.g., Icelandic) and there are contexts in English where gapping is possible but pseudogapping isn’t (e.g., complements to causative or perception verbs). If large conjuncts, and an accompanying TP ellipsis, can produce gaps, then this could be the source of the gaps in these situations. Of course, this requires that there be independent justification for the TP ellipsis in these languages and environments that the large conjuncts account requires. A candidate for this independent evidence is fragment answers, which we will look at in section 23.5. This is the ellipsis needed to produce fragment answers, of the kind that (118) illustrates. (118) Page 44 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Merchant (2004a) makes a convincing case that the question–answer versions of these conversations derive from an ellipsis which removes everything but Bread from Bread, Jones ate. (See Weir 2014, 2015 for an extension of Merchant’s analysis to cases like these.) It’s conceivable, then, that the gaps and strips could be achieved with this ellipsis process applying to clause-sized coordinates. Boone (2014) is an extended argument for this approach. If languages like Icelandic only have gapping in clause-sized conjuncts, we expect that gaps of the form in (96) will unambiguously have a narrow-scope reading for can’t in (p. 597) languages like Icelandic.28 As noted above, Repp (2006a) reports that German, which also fails to have pseudogapping, has facts inconsistent with the small conjuncts analysis when negation and gapping are combined. Finally, allowing a large conjuncts analysis alongside a small conjuncts analysis gives an account for the fact that the subject in a clause with gapping can be either accusative or nominative in English. (119)

On a large conjuncts account, the subject of the clause with gapping will have moved into the nominative case-marked position and, from there, into its focus position. On a small conjuncts account, the conjunct with the gap will be too small to contain Specifier of TP, where nominatives appear in English, and we can expect the default accusative case form instead. When this is put together with the account for Siegel’s contrast between (77b) and (96) sketched above, we make the prediction that (77b) should only allow the accusative form, as this example only has a small conjuncts syntax. I don’t believe that prediction is confirmed, as there is only the slightest difference in grammaticality between the nominative and accusative examples of (120). (120)

It also predicts that using a nominative or accusative subject should disambiguate (96). A nominative subject in the second conjunct should require a large conjuncts analysis, forcing can’t to be within each conjunct, and an accusative subject in the second conjunct should require a small conjuncts analysis, forcing can’t to scope outside the conjunction. I

Page 45 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping haven’t examined this prediction with enough thoroughness, but it is in only partial accord with my judgments about the cases in (121). (121)

As expected, the accusative subject in (121b), which is only compatible with a small conjuncts syntax, forces can’t to scope outside the conjunction. But the nominative subject in (121a) allows for both readings, suggesting that it is compatible with either a small conjuncts or large conjuncts syntax. The grammaticality of (120b) and the widescope reading for can’t in (121a) would both follow if English subjects can be nominative when they remain in their underlying position. It isn’t trivial to discover whether this is the case (p. 598) since the scenarios where subjects in their low position can be pronouns —the only terms that show Case morphology—are rare. These are some of the advantages to letting gapping and stripping occur with clausesized conjuncts. There are also certain kinds of examples that are admitted by widening the set of coordinations that host gapping to clause-sized ones. We should, for instance, expect gapping to be possible in coordinated CPs. This does not generally seem to be the case, however. Gapping is blocked in coordinated complement CPs (= (122a) and (122b)), adjunct CPs (= (122c) and (122d)), and relative clause CPs (= (122e)).29 (122)

But gapping does seem possible in coordinated CPs if they are interrogatives:30 (123)

Page 46 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Sluicing is a clausal ellipsis that is found only in questions, and so it might be the source of examples like (123). If sluicing is responsible for these constructions, and not gapping, then there are no examples of coordinated CPs that support gapping. That is an unsolved problem for the view that gapping can occur in clause-sized conjuncts. Allowing gapping and stripping to occur in clause-sized conjuncts brings real benefits. But we lose some of the features that drive a small conjuncts account. Because a small conjuncts account claims that the highest head in the verbal sequence isn’t elided, but is instead simply outside the coordination, it derives the fact that gapping and stripping are found only in coordinations and it derives Downward Bounding. The across-the-board variant of the small conjuncts account also derives Upward Bounding. If gapping and stripping aren’t restricted to just small conjuncts, then there is no explanation for these three properties. (p. 599)

23.5 Future directions While the small conjuncts analysis predicts that gapping and stripping show up in coordinations but not adjuncts, it doesn’t explain why fragment answers and similar discourses, as well as comparatives, should have superficially similar-looking ellipses. If these cases are to be brought into the analysis, then a force that uniquely singles out these environments should be found. The only attempt I have found in the literature to do this that doesn’t use the small conjuncts analysis points to the antecedent conditions on ellipsis instead. Perhaps the most salient of these accounts is in Andrew Kehler’s work, which I will briefly sketch here. Kehler argues that the antecedence conditions on ellipsis track relations that hold discourses together, and that they are responsible for segregating one kind of ellipsis from another. In Kehler (2000) (and see also Kehler 2002), he adopts a typology of discourse relations from Hobbs (1990) and applies them to the antecedence conditions on ellipsis. He singles out two relations in particular: “cause–effect” and “parallel.” The “cause–effect” relation is invoked to connect the two sentences in (124a), and the parallel relation is invoked to connect the sentences in (124b). (124)

Kehler (2000) suggests that the antecedence conditions on VP-ellipsis are tied to either one of these relations. I’ll frame that hypothesis informally with (125).

Page 47 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (125)

Cases which satisfy (125) are in (126). (126)

Gapping, by contrast, is subject to a narrower condition, one that forces the gapped material and its antecedent to be in a parallel relation. I’ll frame this condition with (127). (127)

This captures the contrast in (128). (128)

(p. 600)

If the notion “parallel” can be formalized correctly, it might manage to restrict the

range of connectives that gapping can exist in to just those involving and and or. This approach to the question of what derives the environments where gapping can occur is wholly semantic, and Kehler argues that is supported by the fact that gapping has an effect on the meanings of the coordinations that host it. To see this, consider the range of meanings that (129) can have. (129)

In each of these cases, the conjunction can have a strictly Boolean interpretation: the sentences report merely that both propositions are true. But it is also possible to understand these sentences as communicating more: that there is a temporal or causal Page 48 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping connection between the propositions each conjunct conveys. In (129c), for instance, the second conjunct could be construed as the result of the first. Kehler reports that Levin and Prince (1986) discovered that this second reading disappears when gapping applies. The sentences in (130) have only the strictly Boolean interpretation. (130)

This follows from (127), if the parallel relation can be defined so that it precludes all but the Boolean construal. For Kehler’s system to explain why gapping is restricted to coordinations an adequate definition of the parallel relation is required, but an explanation for why gapping is only capable of using this relation is needed too. Kehler suggests that the cause–result relation —and all the other similar discourse relations—cannot be used when syntactic reconstruction of the anaphor is at play and that this is why only the parallel relation can be employed in gapping. He assumes, along the lines described above, that gapping necessarily involves ellipsis of a constituent from which the remnants have moved and that the only method of achieving this is for the gap to have syntactic structure that is reconstructed from the antecedent. VP-ellipsis, by contrast, he argues can arise without syntactic reconstruction, and is therefore able to make use of more discourse relations in finding its antecedent. Note, however, that if pseudogapping is similarly an instance of ellipsis of a constituent from which the remnant has moved, this would wrongly predict that pseudogapping is restricted to coordinations in the same way that gapping is. (131)

An explanation for the putative connection between the restriction to parallel discourse relations and gapping remains wanting. (p. 601)

A definition of “parallel” that might do the job needed comes from Craige Roberts’ notion of “Question Under Discussion.” Roberts (2012) sketches a model of discourse cohesion that relates each sentence of a discourse to an implicit goal that the contributors to that discourse share. Those goals, she suggests, can be modeled as the answer to an implicit question: the “Question Under Discussion.” Reich (2007) suggests that the antecedents to gapping constructions come from these implicit Questions Under Discussion. Perhaps we can take Kehler’s “parallel” relation to be the sentences that are related to the Question Under Discussion in the way that Reich suggests. See Weir (2014) for an elaboration of Reich’s idea that solves some of its problems, like those in AnderBois (2010). Page 49 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping Consider how this model will apply to the environments we have seen gapping licensed in. When the Question Under Discussion is overt, we have the special case of gapping being licensed in answer contexts: (132)

A declarative sentence can sometimes also be used to invoke a Question Under Discussion, thereby also providing the antecedent to a gap. Thus, for instance, (133a) could invoke the Question Under Discussion in (133b), licensing the gap in (133c). (133)

And (134a) could invoke the Question Under Discussion in (134b), thereby licensing the strip in (134c). (134)

Coordinations could be understood as having a first conjunct which invokes a Question Under Discussion that, in turn, licenses gapping in the second conjunct. To formalize Kehler’s parallel relation with the Question Under Discussion model, and to use it to explain gapping’s restriction to coordinations, requires that only root assertions be able to contribute to the Question Under Discussion.31 Imagine that coherent discourses consist of assertions offered by participants which constitute partial answers to the Question Under Discussion. Under normal circumstances, these assertions are the root, independent, sentences uttered by the participants. A rough description of the condition on gapping could then be (135). (135)

Page 50 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping Because coordinated clauses in a strictly Boolean coordination act like root assertions, they could each be understood as independently providing a partial answer to the Question Under Discussion. The first conjunct, then, would offer information about what the Question Under Discussion is by providing a partial answer to it, and the second conjunct provides another partial answer to that Question Under Discussion and is thereby able to house a gap. This is roughly the direction that Reich (2007) pursues. (p. 602)

(135) also comes close to deriving Upward and Downward Bounding. It requires that the clause with the gap and the clause containing the antecedent are parallel and this, as Toosarvandani (2016) has argued, could capture many of the cases Downward and Upward Bounding are designed for. If the typical means of making a contribution to a discourse is with the meaning of a root clause, then only root clauses will be able to be gapped. But, as noted, conjuncts of a coordination are straightforward exceptions to that, as their semantics allows them to be understood as independent contributions to the discourse when they get a strictly Boolean interpretation.32 This will effectively block violations of Downward Bounding like those in (136). (136)

The first conjunct can license the Question Under Discussion who ate what?, but the second conjunct can only obliquely be construed as a partial answer to that question. In a parallel way, (135) can explain why violations of Upward Bounding like (137) are blocked. (137)

A Question Under Discussion that the first disjunct licenses is who thought that who ate what?, to which the second disjunct cannot be construed as a partial answer. There are no other Questions Under Discussion that the first coordinates of (136) or (137) can license that the second coordinates can be partial answers to, and (135) will therefore correctly block them. An advantage of deriving Downward Bounding in this way is that it offers a method for relaxing it, and there are reports of cases where it does seem to be relaxed. Weir (2014), for instance, reports that the clausal complements to certain verbs allow relaxation of Downward Bounding when the complementizer is unexpressed. (138)

Page 51 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (p. 603)

And Wurmbrand (2017) reports a similar effect for stripping:

(139)

Weir and Wurmbrand propose that the obligatory absence of the complementizer is connected to the purely syntactic conditions that determine which constituents can be elided. As we’ve seen, the presence of a complementizer blocks gapping in cases where embedded CPs are conjoined, and this could be connected to these cases. (138) and (139) show that when this syntactic constraint is satisfied, Downward Bounding is sometimes relaxed. Similarly, Farudi (2013) shows that certain cases of gapping in Farsi also allow violations of Downward Bounding. (140)

The relaxation of Downward Bounding here could be credited to the ability of the embedded clauses in these examples to constitute a partial answer to the Question Under Discussion. Perhaps with these verbs, the propositions they embed can be understood as proffered by the speaker, thereby satisfying (135). When the embedding verb presupposes the truth of the proposition they embed, however, it would not allow those propositions to be part of what the speaker is asserting. This predicts correctly that Downward Bounding holds in the factives of (141). (141)

See Weir 2014 for a worked-out account along these lines. Kush (2017) argues that deriving Downward and Upward Bounding from something other than the small conjuncts analysis is warranted from a cross-linguistic perspective as well. He shows that Hindi-Urdu obeys Downward and Upward Bounding, and yet doesn’t have many of the other properties predicted by the small conjuncts analysis. For instance, Hindi-Urdu doesn’t permit material that appears to reside in the first conjunct to scope over the second conjunct, as we’ve seen is possible in English gapping constructions. The subject of the first conjunct in (142), for instance, cannot bind the pronoun in the second conjunct. Page 52 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (142)

He suggests that languages vary with respect to whether gapping elides a full clause or a smaller VP-like constituent, but that this variation doesn’t track whatever it is that derives Downward and Upward Bounding. An approach that credits the antecedence conditions on gapping to something like (135), and derives Upward and Downward Bounding from it, would allow for such a typology.33 (p. 604)

For (135) to successfully constrain gapping and stripping would require a formalization of its key parts, including how it is that a gap or strip recovers the elided material from the Question Under Discussion. As we’ve just seen, it offers a method for constraining gapping and stripping to just coordinations and certain cross-discourse constructions, and it also offers a way of deriving Upward and Downward Bounding that relaxes its effects in ways that might be required. Interestingly, though, it doesn’t block simultaneous violations of Upward and Downward Bounding, like those in (143). (143)

A Question Under Discussion that the first conjunct of (143) can invoke is (144). (144)

The second conjunct of (143) is a partial answer to this question and should, therefore, satisfy (135). Moreover, though an account based on (135) would do a better job of allowing gapping and stripping in constructions other than coordinations—like question– answer exchanges—it wouldn’t seem to permit gapping and stripping in comparatives, which (7) seems to indicate is necessary. Nonetheless, a more careful investigation of the discourse and information structure of stripping and gapping seems likely to reduce the number of open mysteries these interesting constructions harbor. I recommend this direction to those embarking on a gapping and stripping career.

Page 53 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping

Notes: (1) The term “stripping” might have been first coined by Hankamer (1971), though he uses it only to refer to the cases that would today be called “fragment answers” (see section 23.5). (2) See Paul (1999) for discussion of a limited form of gapping in Chinese. (3) Koutsoudas (1971) supports this correlation and seeks to build it into gapping. (4) See Hartmann (1998), Wilder (1999, 2008), Phillips (2003), Bachrach and Katzir (2009), Sabbagh (2007), and Ha (2008b) for discussions of Right-Node Raising. (5) But see Abe and Hoshi (1997) for a recent reinvocation of Ross’s typology. (6) See Hoji and Fukaya (1999) for the relationship between sluicing and stripping in Japanese. (7) See Merchant (2004a) for a similar semantically based argument against accounts like these. (8) The word too is present in (25), but not (24). Without too, (25) is either ungrammatical or requires an intonation (stress on and). If either this intonation, or the word too is added to (24), then (24) also loses the meanings in which ‘read the riot act’ has a different denotation in each conjunct. I interpret this to mean that stripping requires either intonation or something equivalent to too to signal that the meanings of the deaccented portions of the two conjuncts are the same. That amounts to the same problem for the semantically independent theory of ellipsis that I am describing in the text. (9) The way this hypothesis is usually expressed is by positing a focus position at the left edge of the clause, rather than the left edge of a VP. (See Depiante 2000, for instance.) But there’s no reason for this, so far as I can see, and assuming a lower position enables a “low conjunct” analysis of stripping—see section 23.3. (10) See Weir (2014) for a solution to the similar problem in fragment answers that could be applied here. (11) She notes, however, that there are exceptions that arise when the embedded clause has a subject that corefers with the subject of the higher clause. See Moltmann (1992) for discussion. (12) (49a–c) are from McCawley (1993). Determiner gapping has not been discovered in stripping, and this has no explanation if gapping and stripping have the same source. (13) See also Chao (1987). (14) For accounts of determiner gapping, see also Lin (1999, 2002), Ackema and Szendrői (2002), Citko (2007), Arregi and Centeno (2005), and Centeno (2011). Page 54 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping (15) A reviewer points out that similar examples can be found in Dutch: ((i))

See Heycock and Kroch (1994), Büring and Hartmann (1998), and Johnson (2002). (16) Siegel did not have the “derived subjects hypothesis” at her disposal, but conjectured that the right conjunct in (72) is what is called a small clause in contemporary syntax. (17) It also explains (i), where the disjunction cannot include the negated modal. ((i))

Negation blocks the scope-widening effect of an alternative-based semantics for disjunction. (18) Although, as Jeroen van Craenenbroeck notes, this is inconsistent with the account I offered for the ungrammaticality of (59). (19) David Pesetsky was the first I know of to make this proposal, but his suggestion never made it to print: it was part of a seminar he gave at MIT in 1997. (20) To the extent that (90) is grammatical, it has an interpretation paraphrased by ‘Some put books on a table and others put magazines on it’, not the interpretation that would be derived by gapping: ‘Some put books on a table and others put magazines on a table’. (21) My thanks to Johannes Gisli Johansson for the examples. (22) Hernandez’s example illustrating this point is *Sal is a forger of passports and Holly paintings. This example, however, also violates Upward Bounding, which we will encounter next. Indeed, my own example might too, depending on the exact parse of deverbal nominals and what derives Upward Bounding. (23) I’ve remained cryptic about the identity of X, but one could adopt the view in Zwart’s (1993, 1994) work on Dutch, or the similar ideas in Bowers (2002), that it is ‘Pred’. (24) A problem with letting a modal move from an underlying position lower than T is that it allows for derivations in which it moves across-the-board out of conjoined MPs. This would be another way of achieving a gap of the modal alone, and this could wrongly allow it to be interpreted in both conjuncts of a gapping construction. Letting modals move threatens to unravel Siegel’s explanation for the fact that a modal has just wide scope in

Page 55 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping these cases. In addition, as Jeroen van Craenenbroeck notes, these derivations violate the Freezing Principle. (25) Repp suggests that a similar interpretation is available in the English (i). ((i))

I and most of my consultants find this sentence ungrammatical. It can be marginally improved by adding rather after but. Consultants who find this sentence grammatical do report that it has the meaning ascribed to it by Repp. I haven’t found an example with and or or that has a similar effect. See Repp (2006b) for some discussion. (26) Jackendoff (1971) suggests that gapping can have no more than two remnants, but I believe the facts suggest a steady decline in grammaticality as the number of remnants increase, with no sharp cut-off point. (27) Hudson (1976b) discusses gapping in inversion structures like (117a). (28) Similarly, it predicts that examples like No father should make okonomiyaki or his daughter eat it won’t allow the bound variable reading for his in Icelandic and similar languages. (29) The fact that a complementizer in coordinated CPs blocks gapping is reported in Hartmann (1998). (30) In all the examples of gapping we’ve seen so far, the remnants are in a contrastive focus relation to parallel phrases in the antecedent. If that holds for (123), then the interrogative determiners, which, in the second conjunct should be contrastively related to something in the first conjunct. See Romero (1998) for an account. For a discussion of gapping in question contexts, and an exploration of some of their constraints, see Pesetsky (1982). (31) Jeroen van Craenenbroeck points out that this does not lead to the expectation that gapping can be inside embedded contexts, as in ‘John regrets that no one asked whether Sally likes beans or Bill rutabagas’. (32) Disjunctions work similarly, though in a weaker way. Each disjunct can be linked to a Question Under Disjunction in the way necessary to license gapping, but the disjuncts are not offered as independently partial answers of the QUD, but instead potential partial answers. (33) Erschler (2016) also reports a number of languages in which violations of Downward Bounding are found. However, in the ellipses he is studying many of the other attributes of gapping are not found—they are found in adjunct clauses, Upward Bounding is

Page 56 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Gapping and Stripping violated, and backwards gapping also seems possible. I tentatively conclude that his cases have a different source than gapping does.

Kyle Johnson

Kyle Johnson is Professor in the Linguistics Department at the University of Massachusetts Amherst. His research focuses on the connection between syntax and semantics.

Page 57 of 57

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

Oxford Handbooks Online Fragments   Alison Hall The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.25

Abstract and Keywords This chapter discusses two kinds of fragments: phrases such as DPs and PPs that occur apparently unembedded in a sentence, without any overt antecedent, and fragment answers to questions. It presents non-sententialist (Stainton 2006, Progovac 2006) and sententialist approaches (focusing especially on Merchant’s 2004 movement-thendeletion account), then considers the main evidence for each approach, and how the opposing approach has tried to address this evidence. Arguments discussed in favour of non-sententialism are anticonnectivity effects, differences in the behaviour of fragment answers and their overtly sentential counterparts with regard to presupposition inheritance, and questions about the psychological plausibility of silent syntax; arguments in favour of sententialism are various kinds of connectivity effects—case, island effects, binding, and requirements for complementizers and prepositions in fragment answers. Keywords: fragment answers, antecedentless fragments, sententialist approach, non-sententialist approach, connectivity, case, island effects, presupposition inheritance, preposition stranding, complementizers

Alison Hall

24.1 Introduction FRAGMENTS are utterances that appear to be smaller than a sentence, including short answers to questions, such as (1B), and various kinds of utterances occurring without an overt antecedent, such as the apparently isolated DPs and PPs in (2)–(4): (1)

(2)

Page 1 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

(3)

(4)

Such fragments are frequently used to express a fully propositional meaning: as has been argued by Stainton (e.g. 2006b), this is strongly suggested by the fact that utterances of fragments can be used to make assertions, are evaluable as true or false, and do not merely communicate implicatures. The debate about fragments used in this way is whether what you see is basically all the linguistic structure you get—that is, whether they are base-generated words or subsentential clauses—or whether they have the syntax of full sentences, part of which is subject to ellipsis.1 Sections 24.2 and 24.3 present the non-sententialist and sententialist approaches to fragments. Sections 24.4 and 24.5 present the main evidence used in the literature to support each approach, and discuss recent attempts to meet the challenges posed by these arguments. Arguments for the latter approach, which assumes silent sentential syntax, are based largely on a set of connectivity effects, particularly case marking; arguments for the former include ‘anti-connectivity’ effects, and differences between fragments and their overtly sentential counterparts. Section 24.6 concludes.

(p. 606)

24.2 Non-sententialist accounts Stainton (2006b) concentrates on utterances of phrases such as NPs, DPs, APs, and PPs occurring without an overt antecedent, and argues that appearances reflect reality and they are syntactically non-sentential: for instance, example (3), ‘From Italy’, as uttered while holding up a bottle of wine, has the structure of just the overt prepositional phrase. This is then integrated in the central conceptual system with information from other perception modules, inference, and memory. This integration, proposes Stainton (2006b), occurs through function–argument application: the speaker utters either (i) a word or phrase whose content is a propositional function, and context provides the argument to this function, or (ii) a word or phrase whose content is an argument to some propositional function, and context provides the function. Applying the propositional function to the argument results in the proposition expressed. The first of these possibilities is illustrated by (3), ‘From Italy’. The central system will receive from the visual module a concept α of the bottle of wine being pointed at (i.e. the argument); the input from the language faculty is a property concept (the propositional function), and the two inputs are concatenated in language of thought/mentalese to give the proposition α IS FROM ITALY

Page 2 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments (Stainton 2006b: 156–7).2 The second possibility is illustrated by Stainton’s example given in (5): (5)

Given the context, the propositional function in (6a) would be salient, and the determiner phrase uttered in (5) would provide the argument to this function, giving the proposition (6b) (where β is a concept of the speaker of (5)): (6)

Stainton remains neutral on the question of whether fragment answers contain silent syntax, though one of his arguments—that antecedentless fragments are non-sentential appeals to the leeway for different interpretations that results from their lack of overt antecedent—suggests that he might accept a sententialist account of those varieties of ellipsis where a determinate content is recoverable from an overt antecedent. Progovac et al. (2006c) apply a non-sententialist analysis much more broadly than the fragments Stainton focuses on, to cover not just fragment answers, but also ‘small-clause’ structures, such as (7):3 (7)

(p. 607)

The core proposal, presented in Progovac (2006) and building on earlier work by

Barton (e.g. 1990), is that non-sententials are generated without merger of (at least) TP, and then, as on Stainton’s account of antecedentless fragments, are combined with salient language of thought material. Among the predictions of this analysis is the appearance of non-sententials with tenseless/non-finite verb forms, as in (7), and, in the kinds of fragments that are the focus of this chapter, default case on subject pronouns when these occur as apparently unembedded fragments, as in (8). This latter prediction relies on Longobardi’s (1994) argument that a DP layer on top of NP is required in an argument position only when needed to check structural nominative case in a TP (Progovac 2013). The idea, also argued for by Cardinaletti (1994) and Progovac (1998), is that pronouns may be generated in the N position, and moved to D in order to check Case features. Where TP is absent, structural nominative case is not required, so subject pronouns would not move to D to check structural case, and will surface in default case (for example, accusative in

Page 3 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments English and French; nominative in German and Spanish). This feature systematically appears in fragments: (8)

Case plays an important role in arguments about whether sententialism or nonsententialism is correct, and I return to it in section 24.5. In the next section, I present the influential sententialist analysis of Merchant (2004a).

24.3 The sententialist approach On Merchant’s (2004a) account, fragments are generated by movement of the fragment into a left-peripheral position Functional Projection, FP, followed by ellipsis of the clause that the fragment has moved out of. B’s answer in (9) would have the structure in (10): (9)

(10)

What licenses the ellipsis is a feature that Merchant calls the [E] feature, which is located in the projection to which the fragment has moved, and tells the phonological component not to parse its complement. What licenses the presence of [E], in turn, is the presence of the appropriate kind of similar content in the prior discourse: this is, in fragment answers, the antecedent question. This analysis can immediately be seen to improve on previous accounts of fragment answers, such as Morgan (1973), which involved only deletion but not movement, and therefore required that various constituents and non-constituents surrounding the (p. 608) fragment be deleted. The problem with such a proposal is in giving a principled account of what kinds of constituent and non-constituent strings can be deleted that does not overgenerate by allowing for impossible deletions.4 Merchant’s move-then-delete analysis constrains the material that can be deleted and that which can appear as the fragment: the deleted material must be a TP, and the fragment must be a constituent that can undergo this kind of movement. As will be seen in section 24.5, this analysis also elegantly accounts for a range of connectivity effects, capturing several parallels between fragment answers and fronted constituents.

Page 4 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments When it comes to fragments without overt antecedents, it appears that there is no linguistic content in the prior discourse to license [E], but Stanley (2000) argues that they should be analysed similarly to fragment answers, VP-ellipsis, and so on: as encoding the syntax of a full sentence, part of which has undergone ellipsis. These cases do not have an overt antecedent, but Stanley downplays this difference, claiming that the felicitous use of these utterances depends on something in the non-linguistic context making salient a linguistic expression to serve as the antecedent for ellipsis. For example, they would often be infelicitous without a preceding ostensive stimulus (such as a demonstration) to draw attention to some object or person. They therefore are not discourse-initial in any relevant sense, claims Stanley, because the prior context necessary for their acceptability makes linguistic antecedents salient to serve as licensers for ellipsis. Consider an example where two people are talking, when across the room, a woman comes in. One of the interlocutors glances at the woman, then back to his partner with a quizzical look. The latter utters ‘John’s mom.’ The idea is that the implicit question ‘Who is that woman?’ is made salient by the questioning look, so can serve as the antecedent. Implicit question and overt answer, then, work similarly to an overt question–answer pair: The elliptical clause would be ‘That is John’s mom’ or ‘That woman is John’s mom’; ‘John’s mom’ would be fronted, followed by deletion of the rest of the clause. There are various kinds of data that sententialists appeal to that make a compelling case for silent syntax in many fragments—both of the answer and the antecedentless variety. Among this data is the case connectivity displayed by fragment answers, as illustrated in the German example (11). Here, the determiner in the fragment answer must have the same case as it would in the full-sentence answer: (11)

As (12) and (13) show, antecedentless fragments in German are also required to have the same case marking that they would have if embedded in a full sentence. (p. 609)

(12)

Page 5 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

(13)

In section 24.5, I discuss in more detail how case and other connectivity effects are used as evidence for sententialism, and also assess non-sententialist attempts to account for them. In the next section, I review arguments that have been given in favour of nonsententialism, and how these could be dealt with on a sententialist account.

24.4 Arguments for non-sententialism, and sententialist replies In this section, I present arguments that non-sententialists have given against sententialism, or varieties of sententialism, and discuss sententialist responses. These arguments concern whether silent syntax is psychologically plausible, particularly in antecedentless fragments (section 24.4.1); ‘anti-connectivity’ effects (24.4.2), and the fact that fragment answers and their overtly sentential counterparts behave differently with regard to whether they inherit presuppositions conveyed by the question (24.4.3).

24.4.1 The psychological plausibility of silent syntax Non-sententialists have tended to cite concerns about the psychological plausibility of silent syntax as an argument for their approach. Stainton (2006b) raises the question of whether it is plausible that elided linguistic material is what combines with the fragment to produce a propositional interpretation in the overtly antecedentless cases such as ‘John’s mom’, discussed in section 24.3, and argues instead for a non-sententialist account on which the fragment combines with a salient mentalese property or argument (see also Bezuidenhout 2013 for similar considerations). Stainton’s argument is that, since the mentalese representation is prior to (and makes available) the natural-language representation ‘That woman is’, the more parsimonious explanation is that the process of recovering the proposition expressed is not mediated by a fully sentential linguistic representation. Merchant (2004a: 274–5) and van Craenenbroeck (2012b) take up the question of the nature of the linguistic material that can be made salient in this way: while it might not be plausible that natural-language representations as specific as those posited by Stanley (2000) are recovered, there are certain phrases, they suggest, that are so general that they (p. 610) are available in almost any context. These phrases are ‘do it’ and ‘that/this is’. The words in these phrases are inherently underdetermining of the Page 6 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments concepts they express: pronouns and demonstratives require reference assignment, and a more specific sense of the verb ‘do’ than the very general encoded meaning would often need to be inferred. This captures the fact that there is some leeway for hearers to recover different propositions expressed within a range that is compatible with the speaker’s communicative intention. A question about this so-called ‘limited ellipsis’ strategy, however, is whether it is too limited to capture all the data. The kinds of elements involved simply do not appear to account for many cases—consider two of Stainton’s examples in (14)–(15): (14)

(15)

It does not seem that the propositions expressed using (14) and (15) would be anything that could count as the content of ‘To Segovia it is/do it’ or ‘Both hands do it!/It is both hands!’. Instead, it looks like the elided phrases must be, for (14), ‘Take me’, and for (15), ‘Use’. Merchant (2010) does mention a different possible analysis: that certain fragments are abbreviations of ‘scripts’: we memorize dialogues for conventional situations such as ordering things (e.g. ‘A coffee, please’ in (2)) or getting into a taxi. It is the script, whose linguistic form is memorized, that supplies the antecedent for the elided clause. However, (15) is not such a conventionalized phrase, so it is far less plausible that there is a script for it. Stainton (2006b: 140–1) gives more such examples, which seem to require greatly increasing the inventory of phrases available in ‘limited ellipsis’, thus entailing that any discourse-initial fragment is massively linguistically ambiguous. Alternatively, it might be that the sententialist would treat as non-elliptical any isolated fragment that cannot be accounted for by limited ellipsis in the original version Merchant (2004a) suggests. Merchant (2006b: 87–8) lists a range of fragment types that he does not consider elliptical, extending beyond those (labels, etc.) on which the consensus is that they do not express propositions; I return to these briefly in section 24.4.3. It is unclear, then, how far the limited ellipsis analysis can go in accounting for antecedentless fragments; however, there is still a problem for the non-sententialist argument that limited ellipsis was intended to address: considerations about whether the recovery of silent syntax is less parsimonious and psychologically plausible than Stainton’s mentalese/language of thought only become relevant if non-sententialism can successfully address the problem of evidence for silent syntactic structure in fragments, such as the case connectivity briefly presented at the end of section 24.3, and discussed further in section 24.5. Another objection based on psychological plausibility is raised by Casielles (2006), who points out that children produce fragment answers before they acquire the movement Page 7 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments that is essential to Merchant’s account: for example, answers such as ‘Me’ in (8) are produced well before children are able to reliably produce wh-questions, at around 3 years 6 months, and subject cleft structures, at around 4 years old (see Guasti 2002). This is a concern for sententialist accounts of fragments if movement turns out to be inevitable, though would not affect non-movement accounts such as that currently being developed by Abe (2016) and Abe and Tancredi (2013). (p. 611)

24.4.2 Anti-connectivity effects

Connectivity effects, introduced briefly at the end of section 24.3, provide evidence for sententialism; on the other hand, there are also certain anti-connectivity effects that do not, at least initially, appear to be explained on Merchant’s move-then-delete account. Example (8), repeated here, demonstrates one such effect: the pronoun in (8B) appears in accusative case, whereas nominative would be expected if the source were the full sentence, as in (16): (8)

(16)

(17)

Merchant claims (2004a: 703) that the ellipsis account is supported by the fact that the fragment answer must bear the same case as it does when in the hanging topic leftdislocation structure in (17), suggesting the latter is the sentential source for (8B).5 This proposal faces the question of where this source comes from; as Progovac (2013: 605) notes, this no longer appears to be a movement-then-deletion analysis. A possible answer is that the source is the biclausal structure (18): (18)

This analysis is based on Ott’s (2014) argument that, in contrastive left-dislocation, the left-dislocated XP is a remnant of clausal ellipsis.6 However, a problem that remains to be solved for this account is that it appears to wrongly predict the possibility of the fragment answer in German, (19B), derived from (20): (19)

Page 8 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

(20)

To account for examples such as (8B), ‘Me’, Progovac uses the idea of default case, although if fragments with non-default case are also to be treated as non-sentential, (p. 612)

an explanation is needed for why accusative case is obligatory in the answer to (19A); I return to this question in section 24.5.1. It should also be noted that the idea of default case is controversial in the Minimalist Program, with Merchant (2004b), for one, rejecting it, though others (e.g. Schütze 2001a) arguing for it, so the success of Progovac’s nonsententialist analysis here will hinge on the outcome of that debate. Casielles (2006) also highlights anti-connectivity in bare infinitive answers, such as (21), and negative answers, such as (22): (21)

(22)

Casielles claims that the non-elliptical source of B’s answer in (21) looks like it would have to be ‘John play baseball’; that for (22B) would be ‘I talked not to John’, both of which are impossible. There are at least the following responses open to the sententialist, though. First, it could be that the sentential source for (21) is ‘Play baseball, John did’, as such movement of tenseless VPs is allowed in English. An alternative suggestion (den Dikken et al. 2000) is that the underlying full forms of these fragments could be pseudocleft or it-cleft constructions, as in ‘What John did is play baseball’ and ‘It is not to John that I talked’. Van Craenenbroeck (2010a) suggests ways of accounting for when the Page 9 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments pre-deletion movement required by Merchant’s account results in simply fronting the pronounced fragment, as in (1), and when it results in it-clefting or pseudoclefting.

24.4.3 The problem of presupposition inheritance A recent argument against silent syntax is based on the differences between fragment answers on the one hand and full-sentence and VP-ellipsis answers on the other. Jacobson (2016b) points out that when the question contains a presupposition, fragment answers and their corresponding full-sentence and VP-ellipsis answers have different properties: (23)

The fragment answer, ‘Jill’, commits the speaker to the view that Jill is a maths professor, but the full answer does not; it seems to communicate that the speaker thinks Jill is not a maths professor. Importantly, the VP-ellipsis answer also behaves differently from the fragment answer. This difference becomes more obvious when considering the following continuations: (p. 613)

(24)

Ellipsis accounts such as Merchant’s do not predict this behaviour, claims Jacobson, who takes it as evidence that ellipsis is not involved in deriving fragment answers. A problem for this conclusion, though, is that sluicing behaves the same way as fragment answers (Chung et al. 1995): (25)

In (25), ‘who’ means ‘which maths professor’. Because sluicing is generally analysed as movement plus ellipsis, the parallels between fragment answers and sluicing seem to provide evidence for sententialism (though Ginzburg and Sag 2000 propose a nonsententialist account of sluicing).

Page 10 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments Jacobson’s alternative proposal is that fragment answers compose directly with the meaning of the antecedent question. This is based on the idea that question–answer pairs are a basic syntactic category of our grammar, Qu-Ans. The question is taken to be a function (or partial function) from objects/entities to propositions, and only the fragment answer—not the full answer—can combine to give the Qu-Ans proposition. For example, the question in (23), ‘Which maths professor left the party at midnight?’, denotes a partial function defined for maths professors: (26)

This can then combine with 〚Jill〛 to give a Qu-Ans proposition only if Jill is a maths professor. This accounts for why the fragment answer commits the speaker to the presupposition while the full answer does not.7 A question that Jacobson herself raises for the Qu-Ans account, though, is whether it extends to fragment answers to implicit questions, such as those created by indefinites and by focus, illustrated in (27) and (28): (27)

(28)

These are exactly the same kind of fragments that appear as answers to explicit questions: for example, in a language with morphological case, they would appear obligatorily with the same case marking as in the full-sentence answer. Antecedentless fragments will pose the (p. 614) same problem for Jacobson’s account—after all, they are often answering an implicit question that is salient in the context, and they display case connectivity. To enable the sententialist analysis to account for the different kinds of fragments— fragment answers to explicit and implicit questions, and antecedentless fragments—while avoiding the problem of presupposition inheritance illustrated by Jacobson’s examples, Weir (2014) adds to the account a semantic licensing condition on fragments based on Roberts’ (2012) Question Under Discussion (QUD). Drawing on Merchant’s (2004a) eGIVENness condition and an earlier QUD account by Reich (2007), Weir proposes a condition that he dubs QUD-GIVENness, where ‘a clause can be elided as long as the union of its focus value mutually entails the union of some unresolved question’ (2014: 117). Returning to example (23), the QUD is 〚Which maths professor left the party at Page 11 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments midnight?〛 = {Jill left the party at midnight, Bill left the party at midnight,…}. Mutual entailment between the union of the QUD and the union of the focus value requires that Jill is a maths professor. Antecedentless fragments where there appears to be no existing QUD in the discourse, and fragment answers to implicit questions, can be dealt with straightforwardly on this account by allowing the QUD to be implicit, and accommodated by the hearer. Consider (29), from Schlangen (2003). B’s answer seems to express (30): (29)

(30)

(31)

Merchant (2006b: 87–8) claims that this example does not involve ellipsis. The question appears to provide the antecedent in (31), which cannot combine with (29B) to produce (30). The QUD account, however, allows accommodation in the context of an alternative QUD, which could be ‘What does seaweed contain that makes it so good for you?’ The QUD is a semantic object, and has to be inferred in context, but what is inferred, on the QUD accounts mentioned here, is not just the semantics of the QUD but also the corresponding sentential syntax. Evidence for this silent syntax is presented in the next section.

24.5 Connectivity effects as evidence for sententialism, and non-sententialist responses Evidence for silent syntax in ellipsis comes from a range of connectivity effects, which are discussed in detail in Merchant (2004a, this volume). One that applies to both fragment answers and many antecedentless fragments is case marking, discussed in section 24.5.1. (p. 615) In 24.5.2–24.5.5 I introduce some of the other main connectivity effects demonstrated by fragment answers: island effects, obligatoriness of complementizers, binding, and requirements for prepositions. I also discuss how non-sententialists have attempted to respond to these arguments from connectivity effects.

Page 12 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

24.5.1 Case connectivity The strongest evidence for sentential structure in the class of fragments as a whole comes from the fact that, in languages with morphological case, the case that appears on a fragment is the same case that the fragment would have if embedded in a full sentence. For example, to order a coffee in German, you can utter the apparent subsentence (32): (32)

Accusative case on the determiner is obligatory, as it would be if the fragment were embedded in a full overt sentence, as in (33): (33)

This fact could be easily explained on the assumption that there is elided syntax in the fragment in (32): it seems, then, that the fragment actually has the syntactic structure of the full sentence in (33). The presence of the phonologically unrealized expression gerne haben (‘would like’), whose direct object is required to have accusative case, would explain the otherwise puzzling requirement for accusative case on the fragment. Merchant (2010) presents evidence of such effects in a wide variety of languages, which points to the conclusion that all apparently isolated determiner/noun phrases might turn out to display some such effect, even if they are largely invisible in English. Some examples involving fragment answers are, from German (34) and Greek (35): (34)

Page 13 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

(35)

Unless case marking can be shown to be explained by interpretive effects, its presence in fragments strongly suggests that the fragment encodes the syntax of a full (p. 616)

sentence. Stainton (2006b: 109) considers the issue of obligatory case marking in antecedentless fragments. He discusses the German coffee-ordering example, where accusative is obligatory, and suggests that case marking plays a semantic role in fragments. Here, accusative case indicates that the denotation of the determiner phrase is the patient. Merchant (2010) rejects this explanation on the grounds that structural cases including nominative and accusative cannot be assigned a consistent semantics and their function is purely syntactic. The idea of case having a semantic role is more plausible for inherent cases, such as instrumental case, which several languages have (Slavic languages, Basque, Finnish, Turkish, and others). Instrumental would be obligatory on, for instance, the Russian version of ‘Both hands!’ (Merchant 2010: section 3.2.3). As Progovac (2013: 606–7) notes, an object that has the theta role of goal/recipient is typically marked with dative case in Serbian and other languages, meaning that dative in such languages could be interpretable, as could certain genitives in some languages (although Merchant 2010: section 5 rules this out for others). Progovac suggests that if a language has some Page 14 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments inherent cases, then even a structural case such as accusative becomes informative, even without a consistent semantics, therefore its presence in fragments could be explained on these grounds. In Serbian, for example, which has dative case linked to objects, and instrumental case, accusative case on an argument would reduce ambiguity and help narrow down the possibilities for interpretation, by indicating that the argument ‘is not an agent, not a goal/recipient, not an instrument’ (2013: 607). While this account remains to be developed in detail, one obstacle that it faces is that, if, for example, a language where dative indicates goal/recipient can employ another way of expressing that something is the goal/recipient, then the lack of dative case marking does not rule out this interpretation. An alternative suggestion is that structural cases encode not a semantics in the sense at issue above, but constraints on interpretation—that is, on the conceptual structure to be built. According to Kempson et al. (this volume), accusative case marking on an expression, such as the German coffee example, introduces the constraint that the immediate combination of that expression with another must result in a predicate. It is semantic expressions, rather than syntactic ones, that are combining, so this is a nonsententialist account. The case marker indicates how the fragment should fit into the proposition expressed, so provides a clue to how to enrich the fragmentary logical form that is encoded. The idea seems to be that using the nominative (which acts as the default case in German) could be misleading, depending on what constraint is proposed for nominative case, or would at (p. 617) least be less informative in examples where the accusative could be used, which would predict the infelicity of nominative in an example such as this.8 However, this account, like Progovac’s, has yet to be fleshed out and does not currently appear to have an answer to why, for example, the nominative should not be able to trigger the construction of a different proposition, such as It is a coffee that I want. So far, then, case connectivity provides evidence for the silent syntactic structure assumed by sententialists. Other connectivity effects have been offered in support of the sententialist account of fragment answers, and those which I introduce briefly in the following subsections have been used specifically to support Merchant’s idea that the fragment has been fronted before elision of the TP.

24.5.2 Islandhood At least some fragment answers seem to be subject to island effects, as shown in (36) and (37): (36)

(37) Page 15 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

The sentential sources of B’s answers, if there is movement before deletion, would have to be (38) and (39), the ungrammaticality of which would explain the ungrammaticality of the fragments derived from them: (38)

(39)

One issue that this analysis raises is that island effects might not be expected to appear in fragments at all, since sluicing does not show the same effects, as can be seen from a sluicing example similar to (36): (40)

Because Merchant analyses fragments and sluicing as TP ellipsis, this difference in island sensitivity is unexpected at first sight, but this is where the Functional Projection mentioned in section 24.3 comes in. In sluicing, the remnant moves to CP, then TP is elided; in (p. 618) fragments, there is an additional step of movement to the FP, which leaves a non-elided trace in the CP which accounts for the island sensitivity. Griffiths and Lipták (2014) criticize the ad hoc nature of this extra step of movement: it is posited purely to create a trace that will explain the island sensitivity of fragments. While it is ad hoc for English, however, Temmerman (2013) presents evidence that motivates the extra movement in Dutch, as Griffiths and Lipták appear to acknowledge, despite noting some variability in judgments about the Dutch data (see Griffiths and Lipták 2014: 221–2, n. 28 for discussion). Griffiths and Lipták (2014: 205–6) also show that in several kinds of example, including the following, island effects appear to be voided: (41)

(42)

Page 16 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

Their conclusion is that only contrastive fragments are island-sensitive; non-contrastive ones (where the antecedent contains an indefinite correlate of the fragment answer) are island-insensitive. This mirrors the pattern identified in sluicing by Merchant (2008b): contrastive sluicing, illustrated in (43), from Barros et al. (2015), is island-sensitive; noncontrastive sluicing, as in (44), is not: (43)

(44)

As with the parallels between sluicing and fragment answers mentioned in section 24.4.3, on presupposition inheritance, this looks to provide more evidence for treating fragment answers as requiring, like sluicing, a movement-then-deletion analysis. To capture the difference between the contrastive and non-contrastive fragment answers, Barros et al. (2015) propose that what is happening in these non-contrastive cases is not that ellipsis voids island violations, but that what is elided is not isomorphic to the antecedent; instead, it is a cleft, for example: (45)

The island effects discussed here do not provide unambiguous evidence for sentential structure, as there is a lack of consensus about the empirical facts. Casielles (2006), Culicover and Jackendoff (2005), and Stainton (2006b) all note variation in judgments of acceptability depending on the salience of the intended contrast in examples such as (36)–(37). For data (p. 619) that tends to elicit more uniform judgments, though, we can look to similar locality effects. Merchant (2004a: 698–700) discusses predicate answers such as (46)–(47): (46)

(47) Page 17 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

As the object is easily recoverable from the context, it is not clear how the nonsententialist would explain why the object cannot be omitted. This follows from the movement-then-deletion analysis: in English, only the whole VP, e.g. ‘totaled it’, not the verb alone, can move to the left periphery.

24.5.3 Complementizer distribution If the move-then-delete account of fragment answers is correct, then complementizers should be required in English fragment answers, as in (48). This is because the full-form equivalents of these answers only allow fronting with an overt complementizer, illustrated in (49): (48)

(49)

Answers of the kind in (48B) are significantly more acceptable with ‘that’ than without, as confirmed in a judgment task run by Merchant et al. (2013). Their conclusion is that, since ‘complementizers do not obviously add any lexical semantic meaning to the answers, it is difficult to see how a syntactic theory without inaudible syntax for fragment answers could explain these results’ (2013: 27). In response, Jacobson (2016b) and Progovac (2013) point out that the correlation between the degradedness of the fragment answer and that of the full sentence in (48) and (49) does not mean that one caused the other; a third factor could be the cause of both namely, they suggest, the fact that null complementizers are licensed only under adjacency with the verb, as shown by (48)–(49) along with data such as the following: (50)

(51)

The adjacency explanation would work for Jacobson 2016b, which has the fragment answer composing directly with the question; verb and complementizer are not adjacent Page 18 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments in the construction that results, so the complementizer must be overt. On the kind of analysis (p. 620) developed in Progovac (2006), however, where there is no verb ‘deny’ in (48B), and no natural-language complementizer—overt or null—it cannot be adjacency or non-adjacency of them to each other that determines whether an overt complementizer is required; another factor must be responsible. Culicover (2013a) gives an account of where English zero-relatives such as (52) are possible that hints at how a non-sententialist, non-Qu-Ans account of complementizer distribution in fragment answers might go: (52)

Building on work by Jaeger and Tily (2011) which finds that the more predictable ‘that’ is in a given context, the more likely it is to be omitted, Culicover argues that, where zero relatives are unacceptable, this reflects complexity of processing—how easy it is to identify the relative clause as a relative clause without the relative ‘that’—and such examples are syntactically well-formed. Applying this idea to complementizers in fragment answers, a Progovac-style non-sententialist analysis could try to argue that ‘that’, while not adding lexical semantic meaning, does provide an interpretive clue which could be expected to ease processing: it indicates that what follows should be interpreted as the direct answer, as opposed to the various other possibilities for interpretation available when the complementizer is absent.

24.5.4 Binding in fragment answers Another connectivity effect cited by Merchant is binding connectivity. Two of his examples are: (53)

(54)

The fragments in (53B) and (54B) are not possible as answers to the antecedent questions. On Merchant’s account, the sources for these fragments would be (53B’) and

Page 19 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments (54B’), which violate principles B and C of binding theory respectively, and this is seen as another strong argument that there is sentential syntax in the fragments.9 (p. 621)

24.5.5 Prepositions in fragment answers Languages that do not allow preposition stranding (P-stranding) in full sentences require a preposition to appear in corresponding fragment answers. In English questions, Pstranding is possible, as in (55A), and the fragment answer (55B) does not require the preposition, whereas in German, which forbids P-stranding, the preposition moves together with the object in question formation, and in the corresponding fragment answers, the preposition is obligatory, as illustrated by (56): (55)

(56)

This is easily explained on Merchant’s account of fragment answers, where the derivation of the fragment has involved movement then deletion of the phrase that has been moved out of it. In English, the object can be fronted without its preposition, as in (57), but in German, only the entire PP can be fronted, as illustrated by (58): (57)

(58)

What remains after deletion of the TP, then, must be the fronted PP in German and other non-P-stranding languages. This requirement for prepositions in fragment answers is a major challenge for nonsententialist accounts, at least in languages such as German (56) and Spanish (59) where the preposition does not appear to contribute anything to the interpretation: (59) Page 20 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

In Serbian, in contrast, as discussed by Progovac, there is interaction between prepositions and determiners that affects interpretation. Here is one of her examples (2013: 608): (p. 622)

(60)

When not governed by a preposition, accusative in Serbian is associated with the theme/ patient role and whole-object affectedness; the preposition is needed to override this interpretation. Progovac does not suggest an account for those non-P-stranding languages where the preposition in fragment answers does not have an obvious effect on interpretation, such as German and Spanish, but Jacobson (2016a) addresses the P-stranding facts, raising several questions that remain to be answered before it can be concluded that they are evidence for silent syntax. These include, first, apparent counterexamples to the Pstranding generalization, where languages that forbid P-stranding seem to allow preposition omission in sluicing. In the Spanish answer in (59), con (‘with’) is required, but it can be omitted in sluicing, as illustrated in (61) (Casielles 2006: 137); Brazilian Portuguese and Polish also allow this: (61)

Furthermore, it is not known what determines whether or not a language allows Pstranding, and, as Progovac et al. (2006a: 342) also point out, only if the explanation must Page 21 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments invoke the full-sentence projection and not just the PP would prepositions in fragment answers constitute evidence against the non-sententialist approach. While the crosslinguistic picture for P-stranding is complex, though, there are analyses of examples such as (61) that maintain the P-stranding generalization by proposing that they involve an underlying cleft (Rodrigues et al. 2009; van Craenenbroeck 2010a; Vicente 2008). On this analysis, then, the structure of (61) without the preposition would be (62): (62)

While Jacobson (2016a: 4–6) raises some more questions for this account, the P-stranding facts pose a major challenge that non-sententialists have so far made little progress with, and are most naturally understood as strong support for the movement-and-deletion account of fragments.

24.6 Conclusion Sententialist accounts of fragments (Merchant 2004a, 2006b, 2010; Abe 2016; Barros et al. 2015; van Craenenbroeck 2010a; Ott 2014; Temmerman 2013; Weir 2014) and the nonsententialism (p. 623) of Progovac et al. (2006c; Progovac 2013), Barton (1990, 2006), Casielles (2006), Stainton (2006b) each successfully account for certain features of fragments, but face difficulties with other features. Merchant’s movement-then-deletion analysis, followed by many others, deals elegantly with many examples of case connectivity in both fragment answers and antecedentless fragments, and with binding connectivity, island violations, and complementizer and preposition requirements in fragment answers, though there are arguments which may instead motivate a nondeletion analysis (Casielles 2006; Abe and Tancredi 2013). Additional connectivity effects that have recently been cited as support for sententialism, though not yet widely discussed, are gender marking and attributive versus predicative marking on fragments (Martí 2015). While those who are sceptical about silent syntax have addressed the connectivity data (Progovac 2013; Jacobson 2016a,b) and begun to suggest some routes for non-sententialists to explore, the facts so far strongly support the idea that fragments require a sententialist analysis.

Acknowledgements Thanks to the reviewers and the editors for their comments on earlier versions, and to Patrick Elliott, Matthew Reeve, and Ye Tian for discussion.

Page 22 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

Notes: (1) I set aside, then, those uses of fragments that are generally agreed not to express propositions, including labels or titles (Jane Eyre), and items on lists (e.g. a shopping list). (2) Small caps indicate mental representations (as opposed to natural-language representations). (3) Structures containing both an overt subject and overt predicate, such as (7), will not be discussed here, as the main issue about fragments is how a fully propositional meaning is recovered from structures that look like they only contain DP/NP/PP, etc. (4) Though see Abe and Tancredi (2013) for a recent attempt to develop an approach that allows non-constituent deletion. (5) For Merchant, the island sensitivity of fragments, to be discussed in section 24.5, also supports this analysis. (6) Ott (2014: 284, n. 26) suggests that HTLD could be analysed along the lines he suggests for CLD. He notes the problem that in examples such as (18), the fronted element does not show connectivity into the host clause (271, n. 5) but does not pursue an account of HTLD in that paper. (7) For some suggestions about how this account deals with the connectivity effects that are a problem for any non-sententialist account, see Jacobson (2016a,b, and this volume). (8) This suggestion, that it is infelicity rather than ungrammaticality strictly speaking, is also made by Merchant (2006b: 89). (9) A recent attempt, yet to be assessed, to account for the binding facts without appeal to silent syntax was made by Jacobson (2016b: 362–3), who applies to fragment answers the direct compositionality account of binding detailed in Jacobson (1999, 2000a).

Alison Hall

Alison Hall is a Lecturer in English Language at De Montfort University, Leicester, UK. She received her PhD in Pragmatics from University College London and has been a postdoctoral researcher at UCL and at Institut Jean Nicod, Paris. She has published on linguistic underdeterminacy, lexical pragmatics, and the debate between contextualism, indexicalism, and semantic minimalism.

Page 23 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Fragments

Page 24 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Oxford Handbooks Online Nominal Ellipsis   Andrés Saab The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.26

Abstract and Keywords This chapter centers on nominal ellipsis phenomena from a broad perspective. First, several diagnostics are provided in order to make a basic distinction between empty nouns and true instances of nominal ellipses. One set of such diagnostics is related to uniformity considerations; i.e. the parallelism between elliptical and non-elliptical nominal phrases with respect to thematic assignment, matching effects, and extraction. Another set has to do with the specific conditions that distinguish empty nouns from nominal ellipses with respect to the need for an antecedent, identity effects, productivity, and lexical meaning. Second, the chapter also focuses on the recoverability conditions that regulate the distribution of nominal anaphora. It is shown why lexical identity is unavoidable in nominal ellipses. Moreover, some putative instances of pragmatically controlled nominal ellipses are, in reality, conceived of as empty nouns requiring contextual salience. Finally, the chapter addresses the problem of licensing nominal ellipses. On the one hand, it is argued that there is no licensing by inflection of any sort. Instead, the morphological effects we observe are epiphenomena resulting from the way in which morphology resolves different stranded affix scenarios. On the other hand, a typology of nominal ellipses is proposed based on the different sizes that the elided constituents might have, depending on some selectional properties of functional heads. Keywords: nominal ellipsis, empty nouns, identity, licensing, anaphora

Andrés Saab

22.1 Introduction NOMINAL ELLIPSIS is the somewhat vague label applied to different types of anaphoric phenomena involving a gap within the internal structure of the nominal phrase. It is used in the literature for a set of arguably disparate phenomena within and across languages.

Page 1 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis Just to give an idea of the many constructions to which the term is (correctly or incorrectly) applied, let us consider Japanese. In this language, the term nominal ellipsis (and relatives) is used to refer to: (i) nominal gaps with at least one genitive remnant (see (1)), (ii) nominal gaps filled with the pronominal -no ‘one’ (see (2)), (iii) nominal gaps of place and time nouns modified by ku-inflected adjectives (see (3)), and (iv) radical gaps in argument position (see (4)):1,2 (1)

(2)

(3)

(p. 527)

(4)

Page 2 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

As the English glosses show, some of these constructions can be translated by a nominal gap or a similar nominal anaphor as in (1) and (2), respectively, but in other cases, the gap must be overtly expressed with the corresponding noun (3) or pronoun (4). In other languages, like Spanish and Romance in general, both (1) and (2), for instance, have a silent counterpart: (5)

Like English, Spanish lacks temporal/spatial elliptical nouns, but like many languages (including English and Japanese) it makes productive use of human ones (the so-called people-deletion or human null construction: Pullum 1975, Kester 1996a, 1996b, Giannakidou and Stavrou 1999, Panagiotidis 2002, Kornfeld and Saab 2005, among many (p. 528)

others). (6)

Finally, Spanish licenses radical gaps in a productive way for subject arguments, but regarding objects, the phenomenon is only attested with indefinite objects (with the exception at least of Andean dialects; see Suñer and Yépez 1988): (7)

Page 3 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Thus, in language after language we observe the ubiquity of nominal ellipsis phenomena. In this chapter, I will adopt the hypothesis—very widespread in the generative tradition— that the term ellipsis refers to a syntactic mechanism that generates some gaps on the basis of salient linguistic information. Put differently, the term ellipsis only applies to what Hankamer and Sag (1976) call surface anaphora. In contradistinction, Hankamer and Sag argued that other anaphora are simply (silent) base-generated proforms, namely, deep anaphora. The general conditions that account for the distribution of these two types of anaphor are discussed at length in Hankamer and Sag’s original paper and much subsequent work since then. This chapter is framed in this tradition. Thus, the set of facts to be discussed in what follows raises at least two basic questions: (Q1) To what extent does the term ellipsis adequately describe the entire set of phenomena illustrated in (1)–(7)? (Q2) Under what general conditions are the nominal gaps in (1)–(7) allowed in natural languages? Given the deep/surface assumption, and the analytical tools it provides, only a subset of the examples discussed so far will be derived as instances of ellipsis. Other cases illustrated in (1)–(7) will be conceived of as empty nouns or deep anaphora; i.e., basegenerated nominals (or different projections of a nominal), whose meaning and syntactic distribution respond to different conditions from those attested for nominal ellipsis. In section 22.2, I will provide several tests to distinguish both types of anaphora with the hope of making a descriptive contribution in an area that is still poorly understood. This will answer (Q1) at least partially. In section 22.3, I will (also partially) answer (Q2) by addressing the problem (p. 529) of the identity condition in nominal ellipses. The so-called licensing problem, also connected to (Q2), will be explored in section 22.4, where I will argue that there is no morphological licensing of any sort; the morphological reflexes attested in nominal ellipsis environments are epiphenomena arising from the interaction of various morphological, syntactic, and semantic factors. Before entering into the varieties of nominal ellipsis, let me clarify some aspects of the DP structure I will assume. Following the tradition initiated by Abney (1987), I adopt the minimal structure of DPs illustrated in (8), in which D-features are encoded in an independent projection dominating the nominal root. Features pertaining to Number are universally encoded in an independent functional head Num above the nP (Ritter 1991 and much subsequent work). As for the nP domain, I assume that it minimally consists of

Page 4 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis a lexical Root, √, and a category-defining head, n, and that both heads are combined via head movement in the syntax (see Embick and Marantz 2008).3 (8)

I take adjectival modifiers to be phrasal adjuncts (or specifiers) that attach to the nP or above and AP/PP complements of the noun (e.g., la destrucción italiana vs la destrucción de Italia, ‘the Italian destruction’ vs ‘Italy’s destruction’) to be selected by the Root (= √). Gender features, when present in a language, are encoded inside the nP (Saab 2004, 2008, 2010a).

22.2 Empty nouns vs nominal ellipsis: some diagnostics Hankamer and Sag (1976) distinguish two basic types of anaphora, namely, deep and surface anaphora. Briefly summarized, with the term deep anaphora Hankamer and Sag refer to a base-generated (c)overt proform, whose basic recovery conditions boil down to those of a (free) pronoun. Surface anaphora instead are elliptical structures derived by transformation (PF deletion in their terms) and, consequently, their recoverability conditions reduce to the theory of identity in ellipsis, whatever the right theory of identity is (syntactic or semantic, in broad terms). Since then, surface anaphora are conceived of as invisible/inaudible full-fledged structures, the result of the operation that we call ellipsis. In turn, deep anaphora are conceived of as lexical proforms made available by the Universal Feature Inventory of Universal Grammar and the combinatory system that produces syntactic objects; i.e., they are not derived by any particular transformation of the computational system. The particular abstract form of these two types of entities is sometimes obscured by their surface form. In the nominal domain, this is especially clear in languages like Spanish which make productive use of both empty noun constructions (ENs) and NP-ellipsis (NPE). (p. 530)

(9)

Page 5 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis The three expressions in (9) might be ambiguous in the right contexts. Consider, for instance, the following sentences: (10)

The sentence in (10a) can mean either that smart dogs and silly dogs cannot be distinguished or that smart dogs and silly people cannot. Likewise, the sentence in (10b) can be true either in a scenario in which the dogs that live in front and the dogs living next door are noisy, or in a situation in which the dogs that live in front and the people living next door are. This ambiguity is straightforwardly derived under the hypothesis that we are dealing with different types of nominal gaps in each of the DPs in the second conjunct of the sentences in (10). Thus, the human reading arises because of the presence of an empty noun in the second conjunct of both sentences in (10) (see Panagiotidis 2002, 2003a,b, and Schütze 2001b and Corver and van Koppen 2011 for related analyses). I will assume that an empty noun is, strictly speaking, a functional nominal category which might encode some syntactic–semantic features such as [+/− human], [+/− female], [+/− count], and so on. It is the same category that provides nominal status to a given bare Root (see Embick and Marantz 2008 and (8)). In turn, the readings under which we are always talking about dogs are derived by NPE. As already mentioned, NPE entails deletion/non-pronunciation of a full-fledged nP including the n itself and minimally the √P. The two configurations are illustrated in (11), where I cross out the elliptical constituents following standard conventions: (11)

Applied to the ambiguity of the sentences in (10), we get the following underlying structures (omitting some important details): (p. 531)

Page 6 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis (12)

The two (rough) representations in (12) adequately describe the ambiguity in (10). The next question is what other types of predictions arise by virtue of the two configurations just provided. An immediate set of predictions can be grouped under what I will call the uniformity assumption, which can be traced back to Ross’s (1969b) seminal work on sluicing (see Saab 2008, 2010b, and references therein): (13)

This assumption allows us to distinguish between ENs and NPE in at least three related domains, namely, (i) thematic assignment (22.2.1.1), (ii) extraction (22.2.1.2), and (iii) case and other matching effects (22.2.1.3). Another set of diagnostics emerges precisely from the particular conditions contemplated in the ceteris paribus proviso. Some of these conditions are quite general, like those that regulate the distribution of antecedents and remnants, while others are language-particular and have to be determined in a case-bycase fashion. In section 22.2.2, some of these diagnostics are addressed with special reference to the human EN construction and NPE in Spanish.

22.2.1 Uniformity tests 22.2.1.1 Thematic assignment A crucial difference between the two configurations in (11) is that ENs are not theta-role assigners. Elided NPs, in turn, conserve the thematic properties of their non-elliptical counterparts. This is a general property that distinguishes pronouns or deep anaphora from ellipsis phenomena. In the case at hand, it is predicted that the internal theta-roles of the Root (or lower N, depending on different assumptions on the DP geometry) cannot be assigned in EN configurations. As shown by Panagiotidis (2002, 2003a,b), if wellknown instances of one-anaphora in English were analyzed as the surface realization of an EN configuration, then the old observation by Lakoff (1970) on the asymmetry between nominal adjuncts and complements in the licensing of one would be immediately Page 7 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis derived, given that ENs lack theta-grids. In effect, whereas nominal adjuncts can modify one, internal complements cannot: (p. 532)

(14)

In (15)–(17), there are some additional examples taken from Corver and van Koppen (2011): (15)

(16)

(17)

By virtue of these facts, Panagiotidis (2002, 2003a,b) proposes the following EN analysis, according to which one realizes an empty noun incapable of assigning theta-roles:4 (18)

Page 8 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

It is worth mentioning that neither this analysis nor the evidence that seems to support it is uncontroversial. Empirically, examples in which one replaces deverbal nouns (19a), picture-like nouns (19b), and human relational nouns like queen, supporter, and (p. 533)

student (19c–e) are attested in corpora (see Payne et al. 2013 for the source of each example): (19)

The fact that some replacements are more frequent than others is attributed by Payne et al. to the existence of more successful competitors (NumP-ellipsis, for instance, like Rome’s destruction and Carthage’s; see section 22.4.3.2). The conclusion for the distinction made in this chapter would be that theta-role assignment would not be a reliable test, against what is generally assumed. In effect, if there is no argument structure projected in nominals and complements are indeed modifiers (see also Llombart-Huesca 2010), then there would be no basis for the distinction between empty Page 9 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis nouns and ellipsis. I think, however, that such a conclusion would be misleading both empirically and conceptually. One way of handling the facts under the EN analysis is to reinterpret Payne et al.’s idea that frequency is accounted for in terms of blocking: more than to a frequency effect we can attribute the difference between NumP-ellipsis with genitive remnants and one-replacement precisely to the distinction between ellipsis and empty nouns. Thus, in my opinion, the sharp contrast between *the destruction of Rome and the one of Carthage and Rome’s destruction and Carthage’s is not because of some blocking effect, but because the second case is a true instance of ellipsis. However, I agree with Payne et al. that the difference between complements and adjuncts might dissolve in favor of the second when it comes to human relational nouns like student or supporters. On the empirical side, more research is needed to know whether other noun complements are really grammatical or not under one-replacement (e.g., (17)). Alternatively, the paradigm in (19) could also be taken as evidence in favor of a surface anaphor analysis for one (see, for instance, Hankamer and Sag 1976). This idea is extensively discussed in Llombart-Huesca (2002), where one is taken as the surface reflex of a support strategy for the Number head in cases in which the NP is indeed deleted by ellipsis. According to Llombart-Huesca, who does not consider data like (19), the basic pattern in (14)–(17) shows that NP complements are part of the ellipsis site and, as a consequence, not (p. 534) visible on the surface.5 On this account, the asymmetry between complements and adjuncts would follow from the size of the ellipsis site. At any rate, both the NPE and the EN analyses should account for variation in judgments with respect to the availability of one to take thematic complements (i.e., the paradigm in (19)). The diagnostic we are discussing here raises no controversy when it comes to some nominal gaps in Spanish. As shown in Kornfeld and Saab (2004) and Saab (2008), this diagnostic gives positive results when applied to the following Spanish examples: (20)

Page 10 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

The facts in (20) clearly favor an analysis in terms of NPE. If the nominal gaps in (20) were analyzed as containing an EN, we would expect these examples to be ungrammatical given that ENs are not theta-role assigners. In contradistinction, a nominal phrase that is elided retains its thematic dependencies after ellipsis. Thus, the (again rough) representation in (21a), but not the one in (21b), accounts for the occurrence of internal arguments of the Root: (21)

Note, however, that the analysis in (21a) is inconsistent with the assumptions made regarding the size of the ellipsis site and the geometry of DPs (cf. (8)). Given that this type of NPE in Spanish is ellipsis of nP (see 11b), we predict that thematic PPs should be part of the ellipsis site, as argued by Llombart-Huesca (2002) for English one (see above). (p. 535) A possible way to overcome this problem is by assuming that the internal complement of destruction in the second conjunct has to vacate the nP domain to some higher position (see Ticio 2003, Saab 2008, and also Eguren 2010 for another approach): (22)

Page 11 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis On this approach, this kind of Spanish NPE parallels the behavior of pseudogapping in English (i.e., VP-ellipsis + movement of an internal complement; see Chapter 21). The same analysis has also been defended for English NumP-ellipsis by Yoshida et al. (2012b) (e.g., John’s books of poems and Peter’s of theater). In both languages, the evidence goes against any assimilation of these constructions to gapping.6 Space limitations prevent us from entering into the nature of such evidence, although I will present an argument in favor of the analysis in (22) for some putative instances of NPE in English in section 22.2.1.2.

22.2.1.2 Extraction tests A well-known test for surface anaphora is extraction from elliptical sites (see Depiante 2000 and this volume, for extensive discussion and references). In short, if a constituent can establish a chain dependency with a position within a putative nominal gap, then we can safely conclude that such a gap has internal structure; i.e., the gap is indeed derived by ellipsis. In the nominal domain, however, there is no easy way to construct this type of example because extraction out of DPs requires manipulating too many variables (e.g., the relative position of the DP, its definiteness/specificity, among other poorly understood variables). However, Lipták and Saab (2010) provide the following example from Hungarian, a language that makes productive use of NPE: (23)

Merchant (2014a) also gives the following example from Greek, another NPE language: (24)

(p. 536)

The same result is obtained in Spanish NPE:

(25)

Page 12 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Yet, English presents a more interesting case related to subextraction facts. In principle, extraction is ungrammatical with numerals or quantifiers as remnants (sometimes taken as remnants of true NPEs). Consider the English counterparts to (25) (all from Gary Thoms, p.c.): (26)

On the basis of this paradigm, one would be tempted to conclude that English lacks NPE in these environments. However, as observed by Jason Merchant (p.c.), the examples become grammatical whenever the preposition is left stranded in the DP domain. As an illustration, consider the following counterpart to (26a): (27)

On an NPE account, these facts are accounted for along the lines proposed for Spanish in the previous section (see (22)); i.e., the PP complement vacates the ellipsis site and then the wh-constituent is subextracted from the PP: (28) Page 13 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Thus, the contrast observed in (26a) and (27) would constitute evidence in favor of a pseudogapping approach, at least in this case. However, the interaction between NPE and extraction is an area almost unexplored and whatever conclusion one draws from such facts should be taken as provisory at this point.7 (p. 537)

22.2.1.3 Matching effects As illustrated in (9), human EN constructions in languages like Spanish can contain, among other modifiers, de-PPs or adjunct CPs as modifiers. NPE gaps, however, subcategorize for the same categories as their non-elliptical counterparts. Take for instance psych nouns like amor ‘love’ which subcategorize for PPs headed by the preposition por ‘by’, or deverbal nouns like insistencia ‘insistence’ which inherits PP complements headed by en ‘in’ from the Root: (29)

(30)

Case markers are also inherited in nominal gaps. Thus, the deverbal noun entrega ‘delivery’ takes a dative complement headed by the dative marker a ‘to’. (31)

Page 14 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis As with the other diagnostics connected to the Uniformity Assumption, the test has to be checked language by language, taking into account the particular lexical and non-lexical properties of the languages under consideration. In principle, and as expected, the test gives positive results in clear cases of NumP-ellipsis in English. (p. 538)

(32)

22.2.2 More diagnostics 22.2.2.1 Antecedents Recall that a sentence like (10a), repeated in (33), is ambiguous between a human EN construction and an NPE one. (33)

Interestingly, reversing the order of the conjuncts eliminates the ambiguity: (34)

This new sentence can only mean that foolish people and smart dogs cannot be distinguished. This follows from the basic division made between ENs and NPE. As observed by Hankamer and Sag (1976), one of the basic properties distinguishing deep and surface anaphora is the mandatory requirement of a linguistic antecedent for surface anaphora. Given that backward anaphora is not allowed in coordinated structures (at least in NPE), the contrast between (33) and (34) is derived as a matter of linguistic Page 15 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis antecedence. Put differently, the NPE reading in (34) is blocked because there is no linguistic antecedent in the required configuration.

22.2.2.2 Identity effects Given that NPE requires a linguistic antecedent, ellipsis sites are formally linked to them. Such a formal dependency has particular manifestations across languages. In the case of NPE in Spanish and other languages, the gender specification must be identical between the antecedent and the nominal gap (for Spanish see Leonetti 1999, Depiante and Masullo 2001, Ticio 2003, Kornfeld and Saab 2004, Saab 2004, 2008, 2010a, and Eguren 2010; for Brazilian Portuguese see Zocca 2003, Nunes and Zocca 2009, and Bobaljik and Zocca 2010; for Greek see Giannakidou and Stavrou 1999 and Merchant 2014a). Although nouns differ as to how natural gender is morphologically represented, the ban on nominal ellipsis under gender mismatches remains constant with some subtle differences among speakers: (p. 539)

(35)

Number does allow for mismatches under ellipsis: (36)

Gender and number asymmetries have been accounted for in at least two ways in the literature. On one approach, the difference boils down to the lexical vs syntactic nature of gender and number, respectively. The general idea is that gender is a lexical property of Page 16 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis nouns and, as such, it has to respect lexical identity under ellipsis. This suggestion has been made by Giannakidou and Stavrou (1999), Depiante and Masullo (2001), and Kornfeld and Saab (2004). This type of analysis has been criticized for a number of reasons by Saab (2004, 2008, 2010a), who proposes that NPE in Spanish and other languages is ellipsis of the nP, to the exclusion of other extended projections of the nominal domain such as NumP and DP (see also Ticio 2003 and, more recently, Merchant 2014a). Given that gender is specified on the nP level and number on the head of NumP (see (8)), the asymmetry follows from the licensing condition on ellipsis. See the following representation, where the [E] feature on Num is just a convention to indicate the head licensing the elliptical phrase (see Merchant 2001 and section 22.4.2): (37)

As mentioned, this analysis captures the gender–number asymmetry on the basis of the ellipsis site size. Yet, it still remains to be seen whether identity in nominal ellipsis must be captured in semantic or lexico-syntactic terms, but this will be the topic of section 22.3.1. For our purposes here, what matters is that identity effects of this type are a property of NPE, but not of ENs, which, as already noted, are subject to the general recoverability conditions on pronouns. When used as a free expression, the natural gender specification (p. 540) on ENs would be consistent only if the female or male presupposition encoded in the empty noun meets its conditions. Therefore, whenever you say las tontas de al lado ‘the fools (female people) living next door’ in an out-of-the-blue context, the expression would be felicitous only under the condition that the people living next door are indeed female. Note that this difference between NPE and human EN predicts that the ambiguity observed in (33) would vanish whenever gender features between antecedent and elliptical gap differ. This is borne out: (38)

Page 17 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis Here, the second conjunct can only have a human EN reading; the identity constraint on gender discussed above blocks an NPE reading.

22.2.2.3 Lexical restrictions and productivity As noticed by Giannakidou and Stavrou (1999) and others, ENs, but not NPE, are lexically restricted.8 On the one hand, they express some meanings but not others. As we have already seen, ENs in Spanish and other languages are commonly used to express a human entity. The types of meanings that ENs might express are, however, a lexical matter. Thus, there is no other obvious reason, beyond lexical specification, why Japanese has empty nouns of time and space (see (3)), but Spanish does not. The meanings that empty nouns might encode correspond to general concepts such as HUMAN, SPACE, TIME, FACT, MANNER, etc. On the other hand, ENs are especially productive with certain types of modifiers but not others. Thus, as discussed in detail by Panagiotidis (2002), poor and rich are especially good EN modifiers in out-of-the-blue cases like the poor and the rich, under generic readings. Such effects could lead us to conclude that these are not manifestations of a particular empty noun syntax, but just nominal uses of adjectives. Given that there are indeed certain lexical and superficial similarities between these two phenomena, it is important to know whether there are arguments for keeping them apart. As shown by Panagiotidis (2002: 56), adjectives like poor maintain their adjectival properties in EN environments; thus, they license degree modification but not plural morphology (e.g., the very poor vs *the poors). This follows if in the relevant cases poor is just an adjective modifying an empty noun and not a nominalized adjective: (39)

There are other morphological particularities that are relevant when it comes to making this distinction between ENs and nominalizations clearer. The diagnostics, once again, depend on language-internal properties. The use of diminutive forms in Dutch, for instance, is a good diagnostic to distinguish pure EN constructions from nominalizations, because adjectives in this language do not tolerate diminutive suffixes. (p. 541)

(40)

Moreover, Dutch human ENs present an irregular form of the plural; instead of using the [‑s] ending that characterizes most nouns with final schwa, they show the plural form [en]. (41) Page 18 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Compare with true nominalizations which can bear diminutive morphemes and show the regular plural ending: (42)

However, it is not always easy to keep empty noun constructions and nominalizations apart. In Spanish, for instance, there are good reasons to think that adjectives like rich, poor, and others can participate in both types of underlying configurations. Thus, a nominal phrase such as los ricos (lit. ‘the rich’) can be ambiguous between an empty noun structure, where the adjective modifies the EN (43a), or a nominalization structure, where ricos occupies the head noun position (43b): (43)

As shown in (44), the second conjunct is ambiguous between a human EN construction according to which we are talking about people in general and an ellipsis reading according (p. 542) to which we are talking about rich people. This second reading follows if ricos is analyzed as a noun like in (43b) and, as such, can be a legitimate antecedent for another occurrence of the same noun in the elliptical gap. Of course, an EN reading is also available in this particular context: (44)

Page 19 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis We can also disambiguate in the opposite direction. As the following examples from Kester (1996a) show, degree adjectival modification disambiguates in favor of an EN analysis: (45)

As expected, NPE cannot apply in this case, because rich, when acting as an adjective, is not a legitimate nominal antecedent. Consider (46): (46)

Here, the reading according to which people from the neighboring country are very rich is just not available. Dropping the intensifier muy ‘very’ makes the elliptical reading fully available again, showing that Spanish distinguishes adjectives that modify ENs (e.g., los muy ricos, lit. ‘the very rich’) from deadjectival nouns. Only the latter can be the target of NPE. In summary, besides superficial similarities, modified empty nouns and nominalizations are different phenomena. Once these possible confusion factors are set aside, we can safely conclude that ENs, as a phenomenon distinct from nominalization, are subject to two basic lexical restrictions: (i) they encode general concepts (TIME, MANNER, HUMAN, etc.), and (ii) they can be subject to particular morphological processes (irregular plurals, for instance). None of these properties applies to NPE, a systematic and productive process of deletion/non-pronunciation quite unrestricted semantically. (p. 543)

22.2.3 Summary I have provided several tests to distinguish NPE from ENs. As shown in section 22.2.1, some diagnostics follow from uniformity considerations (theta-role assignment, extraction, and matching effects) and, as shown in section 22.2.2, others follow from the general conditions that license either NPE or ENs (the need for an antecedent, identity effects, productivity, and so on). Unfortunately, there is some degree of sloppiness in the Page 20 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis literature when it comes to making this basic division. Even though I am aware of the controversial status of some of the diagnostics discussed in this section, I still think that the distinction between NPE and ENs is robust and should be taken seriously when discussing these phenomena within and across languages. In the next section, I discuss the nature of the recoverability conditions that regulate NPE and ENs.

22.3 Recoverability conditions on nominal ellipses In this section, I briefly discuss the problem of recoverability for empty anaphora in general. Building on the basic distinction made in the previous section, I will focus on the identity condition for NPE (22.3.1) first, and then on some particular cases of pragmatic ENs in Spanish (22.3.2).

22.3.1 Identity in NPE As is well known, the proper nature of the identity condition on surface anaphora is a matter of controversy. Broadly speaking, the debate centers on whether identity should be formulated in purely semantic terms (Merchant 2001), in purely lexico-syntactic terms (Chomsky 1965 and much subsequent work in transformational grammar), or in mixed ones (Chung 2006, 2013 for a recent approach). Curiously, NPE has not been the focus of such a debate, even though it constitutes an ideal scenario to evaluate competing theories on identity in ellipsis. An important exception is Giannakidou and Stavrou (1999) who propose the following condition: (47)

Importantly, for Giannakidou and Stavrou recoverability is semantic and not lexicosyntactic. This is so, they argue, because case and number mismatches, as in Spanish, are attested in Greek NPE:9 (p. 544)

(48)

Page 21 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

According to (47), the descriptive content must be given and asserted in the previous discourse and cannot be entailed; otherwise, the following example should be grammatical with the intended meaning that Andreas bought three books, given that buying three dictionaries entails buying three books: (49)

One may wonder in what sense we can say that an antecedent A and an ellipsis site E are semantically identical when it comes to evaluating NPs for ellipsis. Assuming that in the general case NPs denote properties, we can understand the condition in (47) as making reference to equivalence of properties. In this respect, like other semantic approaches to ellipsis (e.g., Merchant 2001), the theory predicts that lexical mismatches should be legitimate in NPE whenever the properties denoted by the antecedent and the elided phrase are identical. Empirical evidence demonstrates that this formulation of the identity condition is too weak and that some sort of lexical identity is needed. The main argument comes from synonymy relations among pairs of words. Consider, for instance, the synonymous nouns doctor ‘doctor’ and médico ‘medic’ in Spanish as they occur in the following DPs: (50)

As it stands, (47) predicts that médico would be a legitimate antecedent for elision of doctor and vice versa, given that, for instance, John met Peter’s medic entails that John Page 22 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis met Peter’s doctor and vice versa. The prediction cannot be (dis)confirmed in this particular case because both nouns are masculine. Yet, it is easy to find pairs of synonymous nouns that differ in gender specification. Take, for instance, the masculine noun casamiento ‘marriage’ and the feminine one boda ‘wedding’, which are synonymous in (p. 545) Spanish under the ceremony and party meanings.10 Ellipsis under strict lexical identity is of course allowed: (51)

Lexical mismatches are however strongly ungrammatical in both directions: (52)

It is worth noticing that the ungrammaticality of (52) cannot be triggered only by gender. Unlike the mismatches discussed in (35), gender specification in (52) is arbitrary (i.e., not related to natural gender or sex). Therefore, no semantic explanation for this particular gender mismatch seems to be available here. English NumP-ellipsis behaves similarly in this respect. As shown by Merchant (this volume), pairs of synonyms like nuptials/wedding differ arbitrarily in number specification: i.e., nuptials is a plurale tantum noun that triggers plural agreement with the verb in spite of being semantically singular. (53)

(54) Page 23 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Verbal agreement mismatches, however, are not allowed under NumP-ellipsis: (55)

(56)

The fact that an arbitrary property of nouns, such as being a plurale tantum, must be preserved under ellipsis gives support to the hypothesis that semantic identity alone is (p. 546)

not enough to identify the ellipsis site. In summary, we are led to conclude that NPE requires some type of lexical/formal identity condition, more along the lines of the lexical–syntactic approach proposed by Merchant (2013d) (see also Saab 2008, 2014, and Chung 2006) than in the semantic theory defended by Merchant (2001) (and much subsequent work) or in even more radical semantic/pragmatic views (Culicover and Jackendoff 2005 and this volume, or Dalrymple et al. 1991). Notably, this type of evidence also goes against laxer syntactic approaches, like the recent proposal by Thoms (2015), according to which syntactic identity does not require making reference to the content encoded in the syntactic nodes but only to the syntactic distribution of variables.

22.3.2 Recoverability and ENs I have claimed that ENs are particular instances of deep anaphora. According to Hankamer and Sag (1976), recoverability conditions for deep anaphora are similar to those applying to pronouns and indexicals in general. In the nominal domain, standard cases of deep anaphora are indeed pronouns of different sorts like she, we, I, this, that, and so on. In fact, the semantic and syntactic literature on pronouns agrees in the postulation of an empty noun in the underlying structure of pronouns. Whether all pronouns contain an empty noun is something that we cannot resolve here (see Elbourne 2013 and references therein). The intricacies of the semantics and syntax of pronouns in general are beyond the limits of the present chapter. Instead, I would like to make a brief remark on some putative instances of pragmatically controlled NPEs in Romance and beyond. Consider the following examples: (57)

Page 24 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Some clarifications are in order. First, I have avoided using cases of human EN constructions, given that we have already provided several tests in favor of their nonelliptical nature in section 22.2. Second, note that the remnants of each example are modifiers and not internal arguments of some putative elliptical noun. Third, this entire set of examples can be translated into English by one or by just a demonstrative. (p. 547)

The examples in (57) seem to pattern like putative cases of English VP-ellipsis

controlled pragmatically (e.g., Shall we? as an invitation to dance; from Merchant 2004a) and other surface anaphora (e.g., sluicing). In view of this, some researchers have proposed abandoning the requirement of a linguistic antecedent (see section 22.2.2.1) as a reliable test for distinguishing surface and deep anaphora (see Chao 1988 and Lobeck 1995, among many others) and claim that surface anaphora can also be controlled pragmatically. However, Merchant (2004a) has provided evidence for analyzing putative examples of English VP-ellipsis controlled pragmatically as instances of silent deep anaphora. I think that a similar approach is generalizable to (57): they are empty nouns pragmatically controlled. I propose then that the sentences in (57) contain ENs minimally specified for some features like gender. Thus, the DP in (57a) can be analyzed along the following lines: (58)

Page 25 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

It seems that one crucial property that these empty nouns encode is that the entity pointed to must be a discrete entity, something that can be indeed signaled and individuated depending on properties of the context and an act of demonstration. In other words, the difference between cases like (57) and, say, human ENs is that the former are pure demonstratives (in Kaplan’s 1989 sense) that require an associated demonstration to be complete. As is well known, the notion of associated demonstration is vague as it does not necessarily imply a stereotypical act of pointing; some salient properties of the context seem to be enough in order to license a correct use of a demonstrative. Again, more research is needed in this almost unexplored area with the detail it deserves. In any case, we can reasonably conclude that pragmatically controlled ENs should not be confused with NPE.

22.4 Licensing nominal ellipses 22.4.1 Some background A crucial question for any theory of ellipsis is to what extent languages can vary in the elliptical constructions they allow. Although not always explicit, the idea is that natural languages do not parameterize the identity condition. We assume that anaphoric recoverability in general is not subject to different language-particular conditions. In other words, recoverability conditions are universal (see Chomsky 1965). The obvious next question is (p. 548) then why some languages have productive NPE or VPE while others do not. The answer involves what we might call the licensing component of the theory of ellipsis. Two prominent formal theories in the literature on NPE are Saito and Murasugi’s (1990a) Specifier-Head agreement approach (see also Lobeck 1990) and Lobeck’s (1995) proper government approach. According to the first view, a given head licenses the ellipsis of its complement whenever such a head has an agreeing specifier. The following pattern is taken as evidence in favor of this line of analysis:

Page 26 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis (59)

Note that only in (59a) is the specifier of the DP filled and, as predicted, it allows for NPE. The appeal of this proposal is that it seems to generalize to other types of ellipsis— concretely, to VP-ellipsis and sluicing. For instance, it is well-known that only some C heads license sluicing, in particular, those that are in a Specifier-Head configuration: (60)

Similar patterns are attested in VP-ellipsis contexts. Schematically, the theory can be illustrated for each type of ellipsis as follows: (61)

Beyond its initial appeal, Saito and Murasugi’s idea that Specifier-Head agreement is what accounts for some types of elliptical phenomena across languages has some shortcomings. Its main weakness is that its empirical coverage is too restricted: languages show many other forms of ellipsis than those that would depend on such an abstract configuration. Moreover, there are Specifier-Head agreement configurations that do not license ellipsis (e.g., sluicing is not generally allowed in relative clauses). These reasons might be behind Lobeck’s (1995) reformulation of the theory of formal licensing. First, for Lobeck, ellipses (i.e., NPE, VPE, and sluicing) are not derived by deletion. On her account, ellipsis sites are instances of the so-called little pro that must be licensed and identified through government (see Rizzi 1990). Here is Lobeck’s particular implementation: (p. 549) Page 27 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis (62)

The crucial notion is strong agreement. Lobeck proposes that a given head X0 is specified for strong agreement if and only if the head or the phrases and heads with which X0 agrees morphologically realize agreement in a productive number of cases (Lobeck 1995: 51). Note that this approach explains why the configurations in (61) license ellipsis, but also extends to cases where the X0 itself is inherently specified for certain formal features. In English, for instance, [+plural] is taken by Lobeck as a strong feature, which would derive why certain demonstratives and quantifiers license NPE without the need for Specifier-Head agreement: (63)

This contrasts with the or a we saw in (59b,c) where the plural feature is not realized on the relevant D head. Other strong features in English are [+possessive] and [+ partitive]. I refer the reader to Lobeck for detailed analyses of particular paradigms. At any rate, the problem with this government-based approach is not only that the conceptual apparatus is not consistent with current approaches to formal syntax, but also that it makes the empirically suspicious claim that the particular morphological requirements some NPE gaps show are an indisputable indication of the role that morphology plays in licensing ellipsis. So the fact that Germanic languages like Dutch or German license NPE ellipsis under the condition of having inflectional adjectives (see also Kester 1996a,b) or that Romance languages use designated inflected determiners and not uninflected ones in NPE environments is taken as a demonstration that inflectional morphology is needed in order to license ellipsis. As we will see in section 22.4.3, these facts seem to be independent of the licensing problem per se. Other current approaches to the licensing problem in ellipsis are also controversial. Theories based on some putative contrast condition on remnants (Giannakidou and Stavrou 1999 or Eguren 2010) are falsified by the mere existence of non-contrastive remnants (see Saab 2008 and Saab and Lipták 2016). Theories based on the quantificational nature of adjectival remnants (Sleeman 1993, 1996) fail because of the existence of non-quantificational remnants (such as simple thematic PP remnants). More morphologically oriented approaches, which attribute a crucial role to word marker projections (Bernstein 1993) or gender/classifier ones (Alexiadou and Gengel 2012) are not only too weak but also empirically incorrect at least for Romance. Attributing licensing to a gender or classifier phrase in Spanish, as Alexiadou and Gengel do, overgenerates gender mismatches under nominal ellipsis (see 22.2.2.2 and Saab 2010a Page 28 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis for more discussion). Finally, the proposal to assimilate licensing to D-linked functional heads has a flavor of circularity in part because of the vagueness of the notion of D-linked functional heads (López 2000). Of course, I do not wish to deny the important contribution of the aforementioned works. The fact that not every complement/modifier can be a legitimate remnant under NPE is an important discovery of the past decades. There are two aspects to be distinguished: (i) on the one hand, some remnants must show some sort of inflectional (p. 550) morphology, (ii) on the other hand, remnants must bear some particular semantic import. Government approaches (e.g., Lobeck 1995, Kester 1996a,b) focus on the first aspect of the problem. Semantic accounts (e.g., Giannakidou and Stavrou 1999 or Eguren 2010) center on the second one. I claim here that neither of these aspects forms part of the theory of ellipsis licensing. As we will see in the subsection that follows, morphological effects in NPE are epiphenomena arising from the way in which morphology resolves some stranded affix filter configurations. As for the second aspect, the issue is still poorly understood, but the restrictive character of remnants seems to be a necessary condition for ellipsis to apply, as already observed in Hernanz and Brucart (1987). For Sleeman (1993), this would follow from the need to look for an antecedent. So, in el auto rojo y el verde ‘the red car and the green one’, the restrictive nature of the color adjective in the second conjunct makes the linguistic antecedent salient. This does not happen with non-restrictive modifiers which cannot license ellipsis. Thus, prenominal adjectives like pobre ‘poor’ in el pobre hombre ‘lit. the poor man’ cannot be legitimate remnants for a gap like el pobre ‘the poor’ which can only be restrictive when understood as an instance of NPE (i.e., Había varios hombres en la fiesta: uno pobre, uno rico… ‘There were several men in the party: a poor one, a rich one…’); otherwise, it is interpreted as a human empty noun. This kind of semantic-discursive effect on remnants led some grammarians to exaggerate the role of contrast and focus in NPE (Giannakidou and Stavrou 1999 and Eguren 2010, among others). This position is criticized in Saab (2008) and Alexiadou and Gengel (2012), among others, on empirical grounds. At any rate, whatever the ultimate explanation for the distribution of remnants, it is important to have the morphological and the semantic factors separated. In the rest of this section, I address some morphological effects in nominal ellipses and try to show that they are epiphenomena derived from the way in which morphology and syntax interact in NPE contexts.

22.4.2 Licensing as selection Licensing can be implemented in a purely mechanistic way. Call this (maybe skeptical) way to think the E-feature approach, which is pursued in Merchant (2001) and subsequent work. It simply states that some phrases are eligible to be elliptical in a given language for the syntactic occurrence of a specific ellipsis feature, [E]. Thus, English, but not Spanish, has VP-ellipsis because the functional node T can be optionally specified for taking elliptical complements. This mechanistic way of thinking has a positive consequence, namely, that ellipsis is a matter of phrasal selection. In the nominal domain, it predicts different sorts of nominal ellipses depending on the loci of the E selection Page 29 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis feature—plus, of course, other conditions on possible remnants, on the one hand, and legitimate morphological outputs, on the other: (64)

The explicative force of such a view depends on the empirical justification of this formal feature and also on the grammatical correlations that it implies within and across (p. 551)

languages.

22.4.3 Government effects in nominal ellipses as epiphenomena We must wonder then how the selection theory of ellipsis licensing may account for the morphological effects attested in different nominal ellipses across languages. Space limitations prevent us from doing justice to the empirical richness of this aspect of nominal ellipsis, but we will advance some lines of research.

22.4.3.1 Selection by Num: nP ellipses Let’s start with nP ellipses (see also (37)): (65)

Depending on some properties of the morphological make-up of DPs, we predict different government effects. There is, indeed, a clear systematic pattern according to whether (i) the language at hand has nP ellipsis or not, and (ii) the language is agglutinative or Page 30 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis inflectional. For agglutinative languages with productive nP ellipsis, it is common to observe that number and case morphemes get stranded whenever nP ellipsis applies. Consider the following examples from Hungarian taken from Saab and Lipták (2016): (66)

(67)

(68)

(69)

What is interesting about this type of example is that Hungarian adjectives do not inflect in number or case in non-elliptical contexts; such an inflection is mandatory when ellipsis applies, as in (66)–(69). As shown in detail in Saab and Lipták (2016), strictly speaking, these are not agreement markers (pace Kester 1996a) but the stranded case and number affixes that are attached to the adjacent adjectival remnant under specific conditions. This is not a particularity of Hungarian grammar but rather is attested in other agglutinative languages like Turkish (Saab 2008), Quechua (Weber 1983), and Persian (Ghaniabadi 2010). We conclude then that the special adjectival inflection that NPE remnants show is the result of nP ellipsis, which leaves the number and case morphemes stranded. (p. 552)

Inflectional languages with productive nP ellipsis, like most Romance languages, resolve this stranded affix scenario by deleting the stranded number affix via number identity with an agreeing morpheme. This would explain the observation that determiners Page 31 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis accompanying nominal gaps must minimally inflect in number, as illustrated in the following Spanish example from Kornfeld and Saab (2004): (70)

Number agreement on cuáles ‘which.pl’, then, licenses the deletion of the conflictive plural feature specified on the Num head: (71)

If this additional deletion operation did not apply, then a stranded affix filter violation would arise at PF. This is exactly what happens in the case in which uninflected qué fills the D head position: (72)

Crucially, the ungrammaticality qué triggers seems to be enough to reject a NumP-ellipsis analysis for these ellipses, at least under our assumptions (see Eguren 2010 for another account). If the Num head were part of the ellipsis site, then it would be deleted under identity with the antecedent NumP.11 Under the NPE analysis, then, government effects are illusory, the surface reflex of a stranded affix filter configuration. (p. 553)

Page 32 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis 22.4.3.2 Selection by D: NumP ellipses Suppose now that the licensor is D and not Num. The predictions about the syntactic and morphological correlations we should expect are quite different from those for nP ellipsis, because now there is no need for “rescuing” stranded number affixes, so no government effects would show up. The syntactic/LF correlate of this is that, as a counterpart, number mismatches should not be allowed. (73)

The best-known example of NumP-ellipsis is English (for a first approach in terms of ellipsis, see Jackendoff 1971, 1977, and Lobeck 1995 for NumP-ellipsis analysis in terms of the theory of pro): (74)

As already mentioned, the morphological correlate of selecting NumP as elliptical is absence of government effects for the category of number. Put differently, this type of nominal gap does not require number-inflected adjectives or determiners, as number is part of what is being deleted. In turn, the syntactic/LF correlate of NumP-ellipsis is that number mismatches should not be allowed. In simple cases, like John’s book/s and Peter’s, the elliptical gap is interpreted as singular or plural depending on the number information encoded in the antecedent. Interestingly, adding grammatical information— through, for instance, verbal agreement—makes number mismatches acceptable for the (p. 554) speakers I consulted, with a preference for the identical cases for some of them ((75a) and (75c)):12,13 (75)

Page 33 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis The % symbol ranges from speakers that found the sentences perfect to those that found a subtle difference between the identical sentences and the non-identical ones. The fact that we find subtle variation across speakers contrasts the behavior of gender with that of number features in nominal ellipses. In effect, recall that gender concord does not improve the basic cases we saw in section 22.2.2.2 (e.g., *el tío de Juan y la de María ‘the.M uncle of Juan and the.FEM of María’). The question is why number specification in verbal agreement improves the number mismatches but gender concord does not. I will not try to formulate an answer here as the matter was not discussed in the previous literature. Instead, I will provide a robust piece of evidence in favor of distinguishing NumP and nP ellipses. Assuming, as is standard, that numerals are specifiers (or modifiers) of the Num head, a clear prediction arises. Concretely, numerals should be part of the ellipsis site whenever NumP-ellipsis applies. According to the informants I have consulted, this is correct: the example in (76) is strongly deviant followed by the assertion that Mary has ten books of poems.14 Since the NumP antecedent contains the numeral three, the ellipsis site must be also modeled as containing an identical numeral. Evidently, this is inconsistent with the follow-up assertion of Mary having ten books.15 (p. 555)

(76)

Interestingly, similar examples are perfectly coherent in Spanish NPE: (77)

There is then evidence for distinguishing these two ellipses, beyond the nature of number and gender mismatches, which, needless to say, deserves careful examination.

22.4.3.3 Selection by n: RootP ellipses As shown by Saito and Murasugi (1990a), some types of Japanese NP-ellipses seem to parallel what we call in the previous section NumP-ellipsis (see the Japanese examples in (1), repeated in (78), and Saito et al. 2008 for a recent approach), where ellipsis is licensed by the sole presence of genitive remnants. As we have seen in the introduction, the form of the genitive is homophonic with the no anaphor (see (2) repeated in (79)):

Page 34 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis (78)

(79)

An important difference between both types of phenomena is that anaphoric no cannot be used to replace eventive nouns: (80)

I will assume here that the anaphoric no construction is amenable to an EN analysis. In turn, (78) has all the properties of a surface anaphora (NumP-ellipsis, strictly speaking). Interestingly, in dialects that morphologically distinguish the genitive and the pronominal form, such as the Nagasaki dialect of Japanese, co-occurrence of a genitive remnant (‑n) plus the putative proform to is attested. Consider (81). In (81a), the particular -n form of the genitive marker is shown, whereas (81b) illustrates the particular form of the anaphoric empty noun to. (p. 556)

(81)

In (82), the genitive marker and to co-occur: (82) Page 35 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

At first sight, one could be tempted to adopt an EN analysis for (82). However, Maeda and Takahashi (2013) present evidence in favor of an analysis in terms of ellipsis. One such piece of evidence comes from extraction of internal arguments in multiple genitive constructions: (83)

As mentioned in section 22.2.1.2, extraction from an ellipsis site can be considered as robust evidence in favor of ellipsis. As proposed by Maeda and Takahashi, the derivation for the elliptical gap in (83c) should minimally contain the trace of the internal argument: (84)

As the reader can check, this analysis resembles the pseudogapping analysis already discussed for Spanish NP-ellipsis and English NumP-ellipsis (see section 22.2.1.1). As in those ellipses, here the internal argument of the noun is also extracted to some peripheral position in the DP structure, which is an indication of the surface anaphor behavior of this particular construction. (p. 557)

As also noticed by Maeda and Takahashi, the form to cannot co-occur with overt

nouns: (85)

Page 36 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis Moreover, to occurs below classifiers like satu, which by assumption occupy a designated functional head, a Class(ifier)P projection (see Tang 1990 and much subsequent work): (86)

The solution proposed by Maeda and Takahashi is that to is the surface form of the n head, which contains the [E] feature licensing ellipsis. The elliptical DP in (86b) is represented as follows: (87)

Thus, to is seen as the nominal counterpart of English do in VP-ellipsis. Just like English do, Nagasaki to does not have any semantic import; it just supports a stranded head in elliptical contexts. This analysis accounts for all the properties seen above; i.e., the extraction facts, the impossibility of co-occurrence with overt nouns, and its relative position with respect to classifiers. A similar analysis could be extended to some dialects of Dutch. Consider the case of Afrikaans and Frisian, both languages that have been argued to use a pronominalization

Page 37 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis strategy. According to this analysis, Afrikaans, among other strategies, makes use of the proform een ‘one’ to fill some nominal gaps: (p. 558)

(88)

However, as noticed by Corver and van Koppen (2011), Afrikaans een can combine with internal arguments: (89)

(90)

Frisian behaves similarly. Among other ways to produce nominal gaps in the language, Corver and van Koppen observe that the form -en attached to adjectival remnants can be followed by a nominal gap optionally realized by a zero morpheme or the form ien ‘one’. (91)

Page 38 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis As shown in (92) and (93) both strategies can take internal arguments of the missing noun as remnants: (92)

(93)

According to Corver and van Koppen, the Afrikaans and Frisian facts are an indication that, as opposed to English one (although see the discussion in section (p. 559)

22.2.1.1), some empty nouns like een and ien can inherit the argument structure properties of the antecedent noun (Corver and van Koppen 2011: 397, n. 24). This stipulation is needed in a theory like Corver and van Koppen’s and related ones, according to which nominal ellipsis sites are empty proforms.16 An alternative way to account for this particular paradigm could assume that these variants of Dutch make use of RootP-ellipsis, which would bring the analysis of these facts in line with the other data discussed in this subsection. So the underlying representation for a case like (92) and (93) in Frisian, for instance, would have a full-fledged RootP in the nominal gap position with all its theta assigners abstractly represented in the syntax. Again, as in other cases explored in the chapter, the internal arguments of the elided RootPs would vacate this projection in the relevant cases (i.e., when they are visible as in (92) and (93)). Evidently, this view would have to assume that the occurrence of a putative proform like een or ien is the morphological reflex of the surviving n head that selects the elided RootP, as in Nagasaki Japanese.17 Yet such an account would still have to explain the doubling effect we observe in Frisian, among other potential problems. Hence, the final form of the analysis would depend on the results of other NPE diagnostics in these Dutch dialects.

Page 39 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

22.4.4 Summary The E-selection approach to the licensing problem predicts different sorts of nominal ellipses depending on the size of the elliptical site, which is selected by the [E] feature. It seems that there is good evidence to postulate the following elliptical sizes: nP ellipses, NumP ellipses, and RootP ellipses, among other options (e.g., Class(ifier)P-ellipsis). In turn, this approach predicts different morphological effects depending on the size of the elliptical site and internal properties of each language (inflectional or agglutinative, for instance). Thus, we have seen that there are at least three morphological strategies that give us what in other accounts are considered as government effects, namely: (i) morpheme dislocation (Hungarian), (ii) morpheme deletion (Spanish), and (iii) support strategies (Nagasaki Japanese). If correct, this approach allows us to dispense with (p. 560) the morphological licensing component of the theory of nominal ellipses proposed in previous analyses.

22.5 Conclusion In this chapter, I have presented several diagnostics to distinguish empty noun constructions from different sorts of nominal ellipses. Ultimately, nominal gaps within and across languages are derived under one or the other analysis and their subtypes. Thus, first of all, we have seen that empty nouns are lexically restricted to express some general concepts and are sensitive to particular morphological processes. Moreover, their structure is that of a pronoun (i.e., a set of nominal functional rootless projections) as is also their semantic distribution. Second, nominal ellipses are derived from full-fledged nominals via an operation that gives us their silent final forms (i.e., deletion or nonpronunciation). Such an ellipsis operation is minimally subject to a lexical identity condition, although other pragmatic and semantic accommodations may be at work. Third, phrasal selection by an [E] feature gives us different forms of nominal ellipses, depending on the locus of such a feature: nP ellipsis, NumP ellipsis, RootP ellipsis and so on. This account predicts different illicit morphological outputs which are resolved by different strategies across languages, like morpheme dislocation, morpheme deletion, and morpheme support. Thus, the so-called morphological licensing in NPE (Lobeck 1995; Kester 1996a,b) is illusory. The survey of nominal ellipsis phenomena we have explored here is, of course, incomplete.18 I hope, however, that the diagnostics presented and discussed throughout this (p. 561) chapter make some contribution to a better understanding of nominal ellipsis and its subtypes in future research.

Acknowledgements I would like to thank Tanja Temmerman and Jeroen van Craenenbroeck for inviting me to write this chapter and for their endless patience and support at every stage of this long process. I also thank Marjo van Koppen and an anonymous reviewer for comments, Page 40 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis criticisms, and corrections that considerably improved the final result. My gratitude extends to Dave Embick, Patrick David Elliott, James Griffiths, Jason Merchant, and Gary Thoms, who helped me a lot with English judgments and references. Finally, many thanks to Verónica Ferry for proofreading the chapter. The usual disclaimers apply.

Notes: (1) I use the names that each of these constructions receives in the literature, without any theoretical commitment. (2) Through this chapter, I follow the Leipzig Glossing Rules (https://www.eva.mpg.de/ lingua/resources/glossing-rules.php). For the sake of simplicity, in most cases I use Rule 4 for expressing one-to-many correspondences and optional Rule 4C, which avoids unnecessary segmentation in the original examples. I also use Rule 6 for non-overt elements, especially in the Spanish examples. In the general case, I avoid modification of glosses from other sources. Abbreviations: ACC = accusative; ADJ = adjective; ATR = attributive marker; CLASS = classifier; CL = clitic; COMP = complementizer; DAT = dative; DET = determiner; DIM = diminutive; F = feminine; FUT = future; GEN = genitive; IMP = imperative; INF = infinitive; LOC = locative; M = masculine; N = neuter, NOM = nominative; PASS = passive; PART = particle; PL = plural; PRS = present; PST = past; SBJV = subjunctive; SG = singular; TOP = topic. (3) When irrelevant, I will just refer to the complex formed by n and RootP as NP. (4) Additional evidence for the nominal status of one is that: (i) it can be the complement of a determiner, (ii) it can receive adjectival modification, and (iii) it can occur in the plural form. ((i))

((ii))

((iii))

Yet Kayne (2015) has criticized the nominal status of one on empirical grounds. According to Kayne, in its anaphoric use, one is not a noun, but a complex determiner. The link with

Page 41 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis its antecedent is, nevertheless, mediated by a silent noun, so anaphoric one is still a deep anaphor on Kayne’s account. (5) An alternative analysis for one-replacement is proposed in Harley (2005), where one is seen as the phonological realization of the n head specified as [+identity] and [+count]. Despite important differences with Llombart-Huesca’s analysis, one also represents a surface anaphor underlying structure under Harley’s account. (6) The reader is referred to the aforementioned works for extensive discussion. (7) NumP-ellipsis in English genitive constructions seems to be less uncontroversial, since in these cases extraction of internal arguments gives grammatical results (e.g., Rome’s destruction and [DP Carthage’s [NumP Num [NP destruction t ]] ]). (8) Giannakidou and Stavrou do not propose an empty noun analysis. Instead, they refer to a substantivization process to account for some of the constructions that I analyze as involving an underlying empty noun. Other constructions explored by Giannakidou and Stavrou also do not seem to instantiate EN constructions. Concretely, what they call the abstract construction of Greek, which is translated by the so-called neuter article lo ‘it’ in Spanish (e.g., lo desconocido, ‘the:N unknown’), does not have the properties of EN constructions (see Kornfeld and Saab 2005). Space limitations prevent us from providing an overview of the properties of this construction. (9) Yet this argument does not hold if number (and also case) mismatches are the result of the fact that ellipsis only deletes the NP layer, to the exclusion of NumP and K(ase)P, as discussed in section 22.2.2.2. (10) Note that wedding and marriage are not synonymous in English. (11) Of course, another crucial argument against the NumP-ellipsis analysis is the asymmetry between gender and number with respect to identity effects already discussed in section 22.2.2.2 (see the contrast between (35) and (36)). (12) Arnold Zwicky contrasts the following examples in his blog: ((i)) I accept the first argument, but reject the other two ___. [understood arguments] ((ii)) I accept the first two arguments, but reject the third ___. [understood argument] ((iii)) That was your dream. Kim’s ___ were all nightmares. [understood dreams] ((iv)) Those were your dreams. Kim’s ___ was a nightmare. [understood dream] (http://arnoldzwicky.wordpress.com/2009/12/08/nominal-ellipsis) According to Zwicky, (i) and (ii) are fully grammatical, but (iii) and (iv) require extra processing work, even when verbal agreement provides the relevant information for the elided number feature inside the DP. It is not clear to me whether the particular grammatical status that Zwicky assigns to (iii) and (iv) is in accord with the judgments that my informants provided for (75). At any rate, it is interesting that he finds (i) and (ii)

Page 42 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis fully grammatical. If confirmed, this contrast would follow from the size of each gap, NP in (i)–(ii) and NumP in (iii)–(iv). (13) Thanks to Dave Embick, James Griffiths, Jason Merchant, and Gary Thoms for judgments and comments. (14) I follow Yoshida et al.’s (2012b) hypothesis that the remnant moves to the right of the ellipsis site in English, but the point is orthogonal to this particular argument. For Spanish NPE, I will continue assuming that remnants move to the left, as in Ticio (2003) and Saab (2008) (see the discussion surrounding (22) in section 22.2.1.1). (15) Thanks to Patrick David Elliott for this particular example. Other informants share exactly the same intuition. (16) Indeed, this problem applies to Lobeck’s (1995) general theory of ellipsis, according to which all types of ellipses (i.e., NPE, VPE, and sluicing) have underlying pros. (17) As discussed in section 22.2.1.1, Llombart-Huesca (2002) also defends an analysis along these lines for English one. (18) For instance, I have remained silent about the nature of null arguments in languages like Spanish or Japanese. As we have seen in the introductory section (cf. examples in (4) and (7)) both languages make use of null arguments in subject and object position. However, it is well known that the distribution of such arguments differs between these two languages. An important observation is that while null arguments in Japanese allow for sloppy readings, Spanish null subjects do not. This observation is due originally to Oku (1998), who provided the following minimal pair. ((i))

((ii))

Page 43 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Nominal Ellipsis

Thus, while the null subject in the Japanese example in (ib) is ambiguous between a strict and a sloppy reading, according to which either John thinks that Mary’s proposal will be accepted or his (=John’s) own proposal will, the null subject in (iib) only admits the strict reading, according to which the empty subject can only refer to María’s proposal and not to Juan’s. A prominent line of analysis, mainly represented by Oku (1998), Saito (2007), and Takahashi (2014), among others, claims that the difference between Spanish and Japanese must be done on the basis of the surface/deep anaphora distinction. Thus, while Japanese null arguments would derive from a radical DP-ellipsis analysis, Spanish null subjects would be base-generated proforms (i.e., they are deep anaphora). Given that deep anaphora do not allow for sloppy readings in the general case, the contrast in (i) and (ii) would be accounted for (although see Tomioka 2003 for another analysis of sloppy readings in Japanese). If this is on the right track, then the distinction between nominal ellipses and empty nouns would generalize to cover also the entire DP structure.

Andrés Saab

Andrés Saab studied Literature and Linguistics at the Universidad de Buenos Aires and at the Universidad Nacional del Comahue (Argentina). In 2009, he defended his doctoral dissertation on the theory of ellipsis. Currently, he is Associate Researcher at the Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET) and Associate Professor at the University of Buenos Aires. His main research areas are ellipsis, copy theory of movement, null subjects and, more broadly, the syntax– interface connection.

Page 44 of 44

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Oxford Handbooks Online Comparative Deletion   Winfried Lechner The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.27

Abstract and Keywords This chapter presents a selective overview of three classes of ellipsis phenomena that manifest themselves in the comparative construction: Comparative Deletion (CD), Comparative Ellipsis (CE), and phrasal comparatives (PCs). A survey of some central empirical generalizations and their theoretical interpretation in the extant literature consolidates three findings. First, while CD displays all the characteristics of a syntactic ellipsis operation, its exact nature still remains elusive. Although some core properties of CD fall out from modeling CD in analogy to the matching analysis for the relative clause, the competing raising account is better equipped to tackle others, such as the identity condition on ellipsis and opacity. Second, CE proves less recalcitrant in that most of its core manifestations are reducible to independently attested ellipsis operations, specifically those found in coordinate structures. Finally, as for PCs, the evidence available at the moment is best compatible with a hybrid approach that treats PCs as base-generated constructions in some languages, but derives them by syntactic ellipsis in others. Keywords: comparatives, Comparative Deletion, phrasal comparatives, gapping, stripping, ATB movement, coordination vs subordination

Winfried Lechner

25.1 Introduction THE surface appearance of the comparatives in (1) is shaped by six deletion operations— gapping, VP-ellipsis, pseudogapping, stripping, across-the-board (ATB) movement and Right-Node Raising (RNR)—which are also attested in other syntactic environments, typically coordinate structures (unpronounced parts marked by angled brackets): (1) Page 1 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

All of the deletion processes in (1) are, modulo the presence of do, optional. The examples in (2) document another type of ellipsis found in comparatives, traditionally referred to as COMPARATIVE DELETION (CD; Bresnan 1973, 1975), which renders unpronounced the gradable property (2a) or the common noun minimally including the gradable property (2b,c) inside the clause following than. Unlike the ellipsis processes in (1), CD is construction-specific and obligatory:1 (2)

(p. 625)

The present chapter surveys the main characteristics of the deletion mechanisms

illustrated in (1) and (2), and explores analytical options that have been pursued in attempts to account for the principles underlying these phenomena. Central to the discussions will be three questions which over the last forty years have time and again served as a practical vantage point for studying ellipsis phenomena (see recent surveys in Merchant, this volume, and Aelbrecht 2016): (3)

As already indicated by the grouping in (1) and (2), in comparatives, these questions materialize in two partially overlapping guises. On the one hand, the triad (3) helps to clarify whether degree constructions afford new insights into the principles governing ellipsis. For instance, if (1a)–(1c) are actually the result of forward verb deletion and movement, two processes that are known to be restricted to coordinate structures, how come that comparatives, which are usually held to establish a subordination relation, emulate the behavior of coordinations? On the other hand, it is also possible to use (3) in an attempt to gain a better understanding of the comparative construction itself. Here, in particular, two topics have attracted a significant amount of attention in the literature, both of which will be addressed below: the nature of Comparative Deletion (see (2)), and the mechanisms responsible for the formation of PHRASAL COMPARATIVES (PCs), illustrated by (1b) and Page 2 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (1c). Superficially, PCs are simply comparatives in which the particle than combines with a single, typically nominal constituent. Since the early 1970s, it has been recognized, though, that there are two competing analytical strategies to capture this observation. PCs can be parsed either as reduced clausal comparatives embedding hidden syntactic structure, or base-generated PPs that are introduced by than (Hankamer 1973b). At the time, cross-linguistic variation had been thought to be related to typological universals. Interestingly, recent studies have revealed that PCs do not form a homogeneous class in that PCs react to routine tests for the presence of silent syntactic structure in some languages, including English, but not in others, among them Japanese and Hindi (Bhatt and Takahashi 2011; Bochnak 2013). The debate leading up to this conclusion, which also has important consequences for the semantic treatment of degree constructions more generally, will be traced in some detail in the sections to follow. In this context, a concomitant question to consider will be whether those instances of PCs for which the diagnostics indicate the presence of hidden structure are best uniformly treated as the outcome of standardly sanctioned ellipsis operations, or whether there are designated kinds of deletion processes only operative in comparatives. The chapter is organized as follows. Subsequent to some preliminaries on the syntax and semantics of comparatives in section 25.2, section 25.3 turns to CD, comparing three competing analyses in terms of ellipsis and movement which mainly diverge in the amount of hidden structure they allocate inside the CD-site. Section 25.4 expands on the different kinds of ellipsis in comparatives. PCs will be treated in section 25.5. As both the phenomenology of and the literature about comparatives is extraordinarily rich, the presentation will by necessity be selective and limited in depth. Complementary surveys of comparatives which also treat ellipsis phenomena are Klein (1991), Corver (2006), Pancheva (2012), Morzycki (2015), Lechner (forthcoming), and Lechner and Corver (2017).

(p. 626)

25.2 The syntax and semantics of comparatives DEGREE CONSTRUCTIONS, which include comparatives, equatives, superlatives, and enough/too-constructions, form a class of structures which express an ordering between two (sets of) degrees. Comparatives induce an asymmetric ordering between the degree introduced by the main clause and the degree specified by the DEGREE COMPLEMENT (henceforth than-XP). For instance, (2a), repeated below as (4a), expresses the proposition that the degree of the ship’s length exceeds the degree of the train’s length, while the ATTRIBUTIVE COMPARATIVE (4b) and THE AMOUNT COMPARATIVE (4c) assert that the quality and quantity of ships built by Korea exceeds the quality and quantity, respectively, of ships owned by Greece: (4)

Page 3 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Following a widely adhered-to terminological convention, the second degree of the comparative relation is also referred to as STANDARD OF COMPARISON, whose left edge is demarcated by the STANDARD MARKER (than). While the standard in (4) is EXPLICIT, it can also be left IMPLICIT, as in John is tall, but Sam is taller. The degrees themselves are supplied by a GRADABLE PROPERTY, typically represented by a gradable adjective denotation, as in (4a) (long) and (4b) (good), or by the DEGREE HEAD much/many, as in (4c). Degree adjectives have been analyzed in two ways, as measure functions from individuals to degrees (logical type ; Bartsch and Vennemann 1972; Kennedy 1999), or as relations between degrees and individuals (type ; Cresswell 1976; von Stechow 1984, among others). Here, I will adopt the latter strategy, without any discernable consequences for the argumentation. On this conception, long denotes the relation which maps each degree to the individuals which are long to that degree. As mentioned above, the gradable property is removed from the comparative complement by CD; for details see section 25.3. Degree clauses share various properties with relative clauses. Just as relative clauses denote derived predicates of individuals, clausal than-XPs can be analyzed as derived predicates of degrees, or, equivalently, as sets of degrees. In both instances, set formation is the semantic reflex of empty operator movement to a clause-initial position (von Stechow 1991; Heim and Kratzer 1998). In the case of comparatives, movement is usually interpreted as λ-abstraction over the degree variable inside the gradable property which has been silenced by CD. The LF of the degree complement for our sample comparative (4a) accordingly looks as in (5a), and denotes, as seen in (5b), the set of degrees to which the train is long (von Stechow 1984; Rullmann 1995; Heim 2000, inter alia):2 (p. 627)

(5)

Syntactic evidence for OP-movement comes from the observation that comparatives respond to barriers imposed by islands, among them the Complex Noun Phrase Constraint in (6a) and the Coordinate Structure Constraint (CSC) in (6b) (Ross 1967; Chomsky 1977b): (6)

Page 4 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

The comparative relation is expressed by an abstract degree head MORE which projects a DEGREE PHRASE (DegP; Abney 1987; Corver 1990) and selects for a degree complement and the gradable property. A more fine-grained representation of (4a) is given in (7). (7)

On a popular analysis, which models degree heads in analogy to quantificational determiners in the individual domain (type ), MORE denotes a twoplace second-order relation between degrees (type ). One of various possible implementations is given in (8) (Gawron 1995; Heim 2000, 2006; Bhatt and Pancheva 2004): (8)

The DegP of (7) can then be expanded into the tree in (9), with AP occupying specDegP (Izvorski 1995; Lechner 1997, 2004) and the degree complement in complement position: (9)

In (9), the unit consisting of MORE and the than-XP (henceforth also DegGQ) denotes a generalized degree quantifier, that is a second-order property of degrees. Similar to generalized quantifiers in the complement position of individual predicates, this DegGQ is not type-compatible with its sister node and therefore needs to undergo Quantifier Raising (QR), as spelled out in (10a) (Heim 2000).3 The moved DegGQ in (10a) binds a degree variable which serves as the inner argument of the matrix AP. Together with the (p. 628) lexical semantics of MORE and the degree complement, the semantic computation yields (10b), which demands that the set of degrees to which the train is long forms a proper subset of the set of degrees to which the ship is long (or, equivalently, that the maximal degree of the ship’s length exceeds the maximal degree of the train’s length; for maximality analyses see von Stechow 1984; Rullmann 1995). (10) Page 5 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

A noteworthy property of this account is that comparatives parallel propositions with universally quantified objects both in their logical syntax and meaning. Discussion of empirical predictions generated by a corollary of the assumption that DegGQs undergo QR will be taken up presently. The factorization in (9) is one of the three main approaches towards the constituency of gradable predicates that have been pursued in the literature, all of which are associated with a distinct set of consequences. On the ‘classical’ view, sketched in (11) (Chomsky 1965; Selkirk 1970; Bresnan 1973; Heim 2000), which is directly compatible with the semantics above, the DegGQ is not a daughter of DegP, but is situated in the specifier of AP. If the DegP is organized as in (11), QR targets the specifier of AP, instead of the sister of AP. The differences between (9) and (11) are minor, but will be seen to render the phrase structure (11) inconsistent with a particular perspective on CD (to anticipate, the raising analysis of CD; see section 25.3). (11)

Finally, the DegP has also been assigned the template in (12), in which MORE and the AP form a constituent to the exclusion of the degree complement. The than-XP joins the derivation as a DegP adjunct, resulting in (12), or is merged as the external argument of the degree head (von Stechow 1984; Abney 1987; Larson 1988b; Corver 1990, 1997; Rullmann 1995; Kennedy 1999): (12)

Page 6 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion Due to this structural modification, either the lexical entry of MORE has to be adjusted, or the order in which the components are assembled needs to be changed. The former option can be operationalized by adopting the meaning rule (13a), which while retaining the subset (p. 629) condition of the generalized quantifier analysis does not require scoping. Alternatively, it is also possible to switch to a semantics that treats the than-XP as a nominalized degree term, as in (13b) (von Stechow 1984 and Rullmann 1995; for a third option in terms of measure functions see Kennedy 1999). The typing in (12) reflects (13a): (13)

But the phrase structure in (12) can also be made consistent with the generalized quantifier analysis in (8) by abandoning the standard phrase structure axiom that all syntactic trees grow strictly monotonically. Specifically, a variant of (12) becomes interpretable if Bhatt and Pancheva’s (2004) hypothesis is adopted that only the degree head MORE moves at LF and that the degree complement is counter-cyclically inserted in the scope position of MORE by Late Merge (Lebeaux 1990). On this conception, the derivation of (7) starts with the base-generated structure in (14a). Next, MORE raises covertly, leaving a degree trace that serves as the inner argument of the AP-denotation (14b). Finally, the than-XP is Late Merged as a sister node of MORE (14c). (14)

The resulting LF-representation (14c) for all intents and purposes parallels (10a), except that the derivation does not create a movement copy in the base position of the than-XP. This has, as Bhatt and Pancheva point out, various desirable consequences. Among others, the Late Merge hypothesis improves on the standard QR-account in that it offers an explanation for systematic correlations between the scope of MORE and the size of ellipsis that the than-XP is able to embed, captured by the ELLIPSIS‑SCOPE GENERALIZATION (15) (adapted from Bhatt and Pancheva 2004: (59); Williams 1974: ch. 4; Gawron 1995; Fox 2002; Lechner 2004: 199): (15)

Empirically, (15) manifests itself, among others, in the observation that in elliptical comparatives, possible coreference patterns reflect the height of attachment of the than-XP, which in turn regulates possible choices for the ellipsis antecedent. To illustrate, Page 7 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (16) admits coreference between Mary and the pronoun her in case the missing VP is construed with broad ellipsis, paraphrased in (16b), whereas the narrow ellipsis reading (16a) incurs a Principle C violation:4 (p. 630)

(16)

The paradigm is accounted for as follows. The narrow ellipsis reading in (16a) requires the degree complement to be merged in the embedded clause (16c), triggering a disjoint reference effect between the name and the c-commanding coindexed pronoun (underlined). By contrast, construing MORE with wide scope, as in (16b/d), makes it possible to Late Merge the than-XP within the matrix clause, above her, which removes the name from the c-command domain of the pronoun. Essentially, what (16) demonstrates is that matrix scope for MORE does not entail that the degree complement is syntactically represented in a position next to the degree adjective, as would be expected in the montonic derivation (17b): (17)

The behavior of degree complements mimics in this respect that of extraposed relative clauses (18b), which are equally able to escape the verdict of Principle C by Late Merge (Fox and Nissenbaum 1999; Fox 2000, 2002, inter alia): (18)

Two remarks regarding the Ellipsis‑Scope Generalization (15) are in order. First, on the assumptions above, the effects of (15) cannot be derived from the denotation of MORE in (13a) or (13b). This supplies an argument for the generalized quantifier analysis of the degree head (8).

Page 8 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion Second, not all researchers agree that the best account of the Ellipsis‑Scope Generalization (15) resides with the Late Merge theory. Alrenga et al. (2012), for example, capture (15) by designing a semantics that admits multiple possible attachment sites for the comparative complement. To recapitulate, although syntactically comparatives fit at least three distinct syntactic templates, the generalized degree quantifier analysis limits the analytical options: while the structures (9) and (11) can be embedded both within a conservative QR analysis and the Late Merge account, (12) is compatible with the latter only. Further consequences of this finding will be discussed in section 25.3. Moreover, Late Merge and, concomitantly, the second-order degree predicate analysis of the comparative relation were seen to receive support from their ability to handle systematic correlations among scope, Principle C, and (p. 631) ellipsis (but see Alrenga et al. 2012). Similar interactions between ellipsis and comparatives will be taken up again in section 25.4, following a survey of empirical and theoretical challenges posed by Comparative Deletion (CD) in the next section.

25.3 Comparative Deletion It is possible to distinguish among at least three families of approaches towards CD, which diverge mainly in whether they endow the silent gradable property with syntactic structure or not, and the mechanism which identifies the elliptical gradable property. More specifically, CD has been analyzed as (i) the result of syntactic ellipsis in combination with movement of a degree operator (which, on some accounts, itself consists of a deleted constituent); (ii) a designated type of movement operation (‘head raising’), and (iii) an instance of semantic ellipsis, with the CD-site being syntactically inert. The present section provides a synopsis and outlines how the syntactic and semantic assumptions collected in section 25.2 align with these competing perspectives on CD.

25.3.1 The movement and deletion (matching) analysis of CD Proponents of the movement and deletion analysis (Lees 1961; Hankamer 1973b; Chomsky 1977b; Kennedy 2002) agree that CD combines fronting of a constituent inside the comparative complement with the instruction to forgo pronunciation of the degree predicate, possibly together with other nodes. What exactly moves or is deleted depends on the particulars of the theory, though. To exemplify, the classical empty operator analysis (Chomsky 1977b; von Stechow 1984; Heim 1985; Izvorski 1995; Rullmann 1995, among others) postulates movement of a null operator to specCP. From there, the operator binds a degree variable within the gradable predicate, which is deleted by a construction-specific ellipsis operation. In a recent incarnation, developed in Kennedy (2002), displacement targets the whole DegP-complex instead of just the null operator, followed by ellipsis of the higher occurrence of DegP. Kennedy’s analysis also diverges from the classical account in that the variable is not located within the degree predicate, Page 9 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion but is syntactically represented by the lower copy of the DegP. On this view, the effects of which are exposed in (19), CD is a manifestation of matching the unit [DegP Deg° long] in specCP with the external head of the degree clause [MORE long]:5 (19)

Modeling CD in terms of DegP-movement has two immediate consequences. First, it is no longer necessary to stipulate a designated ellipsis operation affecting the lowest DegP-occurrence, because deletion follows from the general principles regulating the distribution of audible copies. Second, the semantic part of the analysis, to be expanded on momentarily, includes a device which operationalizes the matching relation between the higher DegP and the external head. (p. 632)

Adjusted to current assumptions according to which gradable adjectives denote relations between degrees and individuals, the abstract degree head Deg° of Kennedy (2002) can be defined as in (20a).6 (20b) repeats from above the non-scoping version of MORE’ and (21) supplies relevant parts of the semantic calculation for (19). The LF to be compositionally interpreted is (21a): (20)

(21)

As shown by (21b), the lower DegP translates as a variable of type which is bound by the λ-operator created by DegP-movement to specCP. Semantically, abstraction over T generates a predicate of ‘passivized’ (i.e. instead of ) gradable adjective meanings. Once this derived predicate is combined with the DegPdenotation, given in (21c), the derivation produces, as desired, the predicate of degrees Page 10 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion to which the train is long (21d). The output of (21d) can finally be picked up by the comparative head MORE’ (20b). A noteworthy feature of this set-up is that the identity condition between the internal head in specCP and the external head is defined syntactically, yet the content of CD is restored in semantics, by β-converting the degree predicate inside the fronted DegP into the CD-site. Moreover, as DegP movement to specCP arguably leaves a copy, the gradable property inside the than-XP contains silent structure in specCP, as well as in its base position. As observed by Kennedy (2000: n. 2), the derivation in (19) is reminiscent of the matching analysis of relative clauses (Hulsey and Sauerland 2006; Bhatt 2015) in that a category which is internal to the relative/degree clause is raised and elided under identity with an external head. Syntactically, the proposal is based on the constituency for the DegP in (12), which, as exposed by (22a), guarantees that the internal DegP ([Deg° AP], (p. 633) underlined) actually finds a constituent which supplies a matching external head ([MORE AP], double underlining): (22)

None of the alternative syntactic templates for DegP satisfy this requirement. Both in (22b) (which models the DegP after (9)) and (22c) (which follows (11)), the internal DegP lacks a corresponding external head that excludes the than-XP, suggesting that the matching analysis is compatible only with the phrase structure in (12). While the observation above does not have any detrimental effects in itself, it relates to another, more problematic trait of the analysis: the matching account is, at least in its present form, inconsistent with the Late Merge hypothesis. This follows from the fact that the two theories impose two conflicting sets of requirements on the constituency of MORE and the AP. On the one hand, the Late Merge account mandates that the than-XP be inserted as the sister of MORE, instead of as a sister to the unit [MORE AP]. That is, MORE has to form a unit with the than-XP to the exclusion of the AP. By contrast, the matching analysis makes just the opposite kind of demand since the local context must supply an external head—for matching with the internal DegP—which needs to contain both MORE and the AP, but not the than-XP. Put differently, counter-cyclical Merge inevitably locates the than-XP in an environment that fails to provide an external head.7 Note, moreover, that the problem generalizes into two directions. First, it is not restricted to the particular template in (12) but extends to the alternative two phrase structures. Neither (9) nor (11) is able to reconcile the assumption that the than-XP is Late Merged Page 11 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion as a sister of MORE with the need of the internal DegP for an external matching [MORE AP] unit. Interestingly, while this finding poses a substantial challenge for the matching analysis, a similar (yet not identical) shortcoming will also be seen to impinge upon its main competitor to be addressed in the next subsection. Second, it is hard to see how the matching analysis derives scoping of the comparative relation MORE in the first place, given that (i) MORE and the than-XP do not form a constituent, which prohibits them from QRing as a unit, and (ii) movement of MORE in isolation bleeds Late Merge. Naturally, it would of course be possible to scope only the than‑XP, but there is no discernable strategy for transporting MORE together with the than-XP into their scope positions. This second generalization of the problem makes the matching analysis incompatible with the standard degree quantifier approach. An implementation that avoids this complication can be found in Alrenga et al. (2012), though. (p. 634)

25.3.2 The raising analysis of CD Similarities between relative clause and comparative are also exploited by the second strand of analyses, which construe comparatives in analogy to the head-raising analysis of relative clauses (Vergnaud 1974; Kayne 1994; Bhatt 2015; on head raising in comparatives see Donati 1997). In Lechner (2004), it is proposed that the gradable property is attracted by a feature on the matrix degree head, landing in its specifier position, where the feature is morphologically expressed by comparative morphology under spec–head agreement: (23)

Just like in the head-raising account of relative clauses, only the higher copy is spelled out. This can be made to follow from the assumption that the lexical specification of the lower, semantically inert degree head contains an ellipsis feature [E] (Merchant 2001), which instructs the grammar to forgo pronunciation of specDegP.8 Moreover, semantic considerations demand that both the higher and the lower copy of the gradable property are interpreted, resulting in an instance of movement without chain formation.9 Since the relation between the comparative AP and the CD-site is defined in terms of movement, the identity condition on CD is the same as the identity requirement operative in other instances of dislocation. Thus there is no need to grant special dispensation to the relation between the external head [MORE AP] and the internal head [Deg° AP] from the laws defining syntactic or semantic identity, as was necessary in the matching analysis (see also n. 5). In general, the raising analysis does not make it necessary to stipulate a construction-specific, obligatory deletion process for comparatives; rather, the core properties of CD fall out from the universal principles regulating movement. It is obvious that for head raising to succeed, the internal structure of gradable APs must be like (9), because only (9) ensures that the pronounced, morphologically marked AP cPage 12 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion commands the than-XP and at the same time resides in the specifier of the degree head. Once head raising has applied, the quantificational degree determiner MORE in (8) together with the than-XP undergoes QR, providing the requisite logical syntax for interpretation. The LF-representation and corresponding interpretation were already introduced by (10). Independent support for the phrase structure in (9) comes from an observation due to Bresnan (1973), who notes that in attributive phrasal comparatives, the size of CD systematically correlates with the position of the degree complement. Only (24a) admits a sensible reading, while (24b) entails that Sally is a man, resulting in a category mistake. This indicates that the CD-site comprises of the AP in (24a), but includes the NP in (24b): (p. 635)

(24)

A successful analysis of the contrast above hinges on finding a way to secure that the size of CD is functionally determined by syntactic properties such that the CD-site corresponds to the sister node of the than-XP as expressed by Gawron’s (1995: 343) ‘Hypothesis A’: (25)

On the raising analysis, Hypothesis A follows directly from the c-command condition on movement. By contrast, the matching account, which posits the structure in (22a), repeated below, either has to stipulate that identity in comparatives ignores the difference between ‘MORE AP’ and ‘Deg° AP’, or needs to assume that the matching operation reaches into the external head, targeting the complement of Deg°: (22)

If the latter option is chosen, matching is no longer defined in terms of c-command. But the raising analysis comes at a cost, too. Similar to what was observed with the matching account, an independent yet inherent feature of the head-raising mechanism renders the analysis incompatible with the Late Merge hypothesis. The raising account of CD presupposes that the than-XP is merged early, as a sister of MORE, so as to allow movement out of the than-XP into specDegP. But this is clearly inconsistent with the assumption that the than-XP joins the derivation only at a later stage, once MORE has reached its scope position. Thus, unlike in the matching analysis, the complication is

Page 13 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion created by conflicting criteria on the sequencing of derivational steps, and not by a conflict of constituency.10 One possible response to this challenge consists in assuming that than-XPs are merged cyclically after all, but fail to create movement copies due to independent factors which are operative in comparatives, but not in relative clauses. These factors might in turn be reduced to logical properties that distinguish comparatives from other clause types. More precisely, Bhatt and Pancheva (2004: 39–40) point out that LF-configurations in which the than-XP is merged cyclically lead to logical contradictions. Together with a principle that prohibits certain, formally definable, representations with trivial truth conditions (Ltriviality; Gajewski 2002), this would be sufficient to guarantee that QR of the unit (p. 636) [MORE than-XP] leaves a degree variable, instead of a copy. Moreover, as contradictions are limited to comparatives, Late Merge of relative clauses would, as desired, still be possible. While this looks like a promising move for rendering the raising analysis of CD consistent with the Ellipsis Scope Generalization, a more careful investigation of the problem has to await another occasion.

25.3.3 Semantic theories of CD and diagnostics for structure On semantic theories of CD, the unpronounced gradable property is either λ-converted into the degree clause (Kennedy 1999, 2002; see (21)) or treated as an instance of discourse anaphora (Lerner and Pinkal 1995; see also Klein 1980; Larson 1988b; Heim 2006). By nature, these accounts require a syntax in which the external head of the comparative is already present at the point where CD is interpreted. In principle, this configuration can be provided both by a raising syntax or a matching derivation. While the specific mechanism implicated in restoring the CD-site is not of direct relevance for our present concerns, there is an important property discriminating between the semantic analysis and its other competitors. Both the deletion and the raising analysis have in common that the CD-site contains hidden syntactic structure that is predicted to react to standard diagnostics for the presence of silent nodes. By contrast, semantic analyses (Lerner and Pinkal 1995; Kennedy 1999) generate the expectation that the CD-site is syntactically inert. Various tests document that CD does indeed embed unpronounced structure, in compliance with the syntactic approach (Lechner 2004). To begin with, names inside the CD-site induce Principle C violations (26a), indicating that the degree complement contains a silent copy of the gradable AP and its complement. CD behaves in this respect just like VP-ellipsis (26b). Moreover, CD also parallels VP-ellipsis in that clausal embedding of the name ameliorates the disjoint reference effect, as shown by (26c) and (26d) (vehicle change; see Fiengo and May 1994; Safir 1999, among others): (26)

Page 14 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Second, the CD-site can host reflexives and reciprocals, which are commonly assigned sloppy readings.11 For theories which express the conditions on anaphor licensing structurally, this entails that the abstract representation of the degree complement has to include unpronounced occurrences of the bound variables: (27)

Next, extraction from the CD-site is subject to the CSC, a condition which is usually held to be verified at the syntactic level of LF (Fox 2000). The contrast in (28) receives a natural (p. 637) explanation under the syntactic account, but is difficult to reconcile with the tenets of semantic theories, which lack the means to express conditions on LFrepresentations. (28)

Finally, the fact that the silenced AP in (28a) can contain a trace whose binder is located outside the elliptical node provides further, independent confirmation that the CD-site contains hidden structure. Thus, it seems very likely that CD is an operation that either consists in movement and deletion or head raising. To recapitulate, while the ellipsis process that renders unpronounced the degree predicate inside the comparative displays all the signatures of a syntactic operation, the exact nature of CD is still somewhat elusive. On the one hand, it was seen that some of the core properties of CD are successfully captured by the matching analysis. On the other hand, the raising account avoids complications regarding the identity condition on the CD-site and its antecedent, and offers at least the prospect of an explanation for Late Merge effects. After this brief, necessarily selective survey of theories of the obligatory process of CD, the discussion to follow elaborates on optional ellipsis operations in comparatives. For more information on CD, see Pancheva (2012), Morzycki (2015), and Lechner and Corver (2017).

Page 15 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

25.4 Ellipsis in comparatives—Comparative Ellipsis It has been observed at least since Hankamer (1973b) that various kinds of reduction processes that are attested in coordinate structures can also be found in comparatives. (29) provides a representative sample from coordinate contexts, and (1), repeated from above, examples of arguably the same phenomena in comparatives (nothing bears on the particular parses): (29)

(1)

Historically, three questions have been central to the study of ellipsis in comparatives. First, can all manifestations of ellipsis in comparatives be subsumed under independently attested ellipsis operations, or are there processes that are only operative in degree constructions (Comparative Ellipsis; Bresnan 1973; Pinkham 1982)? Second, are all seemingly elliptical comparatives actually elliptical? More specifically, do all phrasal comparatives (PC), exemplified by (1b) and (1c), embed unpronounced structure, or are there also base-generated PCs? Third, it has been asked how phenomena that are generally confined to coordinate structures are licensed in the subordinate environments provided by comparatives. The present section reports findings relevant to the first and third domain, while the dispute about PCs will be taken up in section 25.5. (p. 638)

Before proceeding, a brief methodological note is in order. It is important to keep in mind that degree complements are (unless the remnant is a measure phrase) elliptical as a result of CD. When lining up comparatives and coordinations, the CD-site should therefore, all things being equal, be phonologically filled in the corresponding coordinate structures. Thus, the analogous coordinate structure of (30a) is (30b), and not (30c). (30) Page 16 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Throughout, I will assume that the right-hand-side bracketed unit in (30a) corresponds to the second conjunct in a coordination, with than usurping the role of a syntactic coordinator.

25.4.1 Ellipsis in comparative and coordinate structures While it is uncontroversial that the reduction processes in (1) have the appearance of standard ellipses, demonstrating that the comparatives in (1) and their coordinate analogues (29) actually share a common derivational history turns out to be a less trivial task. In what follows, I will present reasons to believe that this strong hypothesis is correct at least for gapping, ATB movement, and RNR (Pinkham 1982; Hendriks 1995; Moltmann 1992; Lechner 2004).

25.4.1.1 Gapping Gapping is a deletion operation exclusively found in coordinate structures ((31a) vs (31b)) which targets a proper subset of verbs in non-initial conjuncts ((31c); Johnson, this volume, for an overview). (31)

Six signature characteristics which are known to define the behavior of coordinate gapping are also attested in comparatives (Lechner 2004): (i) Gapping operates progressively only, in coordinations as well as in comparatives: (p. 639)

(32)

(33)

(ii) Gapping cannot affect infinitives to the exclusion of finite verbs. This is shown for coordinate structures in (34a) and adverbial, subject, and object comparatives in (34b–d). Page 17 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (For expository convenience, the discussion will focus on a single exponent of the paradigm from now on.) (34)

(iii) Just like gapping is prohibited from targeting finite, embedded clauses in coordinate structures (35a), it cannot do so in comparatives (35b), either: (35)

(iv) In both constructions, the results are strongly degraded if the gap includes finite sentence boundaries: (36)

(v) In V2-languages such as German or Dutch, gapping must not operate across overt complementizers (Hendriks 1995): (37)

Page 18 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (vi) In V2-languages, it is only marginally possible to gap clause-final, finite auxiliaries to the exclusion of the main predicate (38). The results improve drastically if the finite verb undergoes V2 (39): (p. 640)

(38)

(39)

25.4.1.2 Right-Node Raising Unlike gapping, RNR removes material from non-final constituents and targets strings that contiguously extend from the right edge of the first conjunct (40), largely ignoring locality conditions such as the Right Roof Constraint ((41b); Hartmann 1998; Sabbagh 2007). RNR treats comparatives and coordinations alike (see Lechner 2004 for further details): (40)

(41)

Page 19 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

In OV-languages, RNR overrides the prohibition on auxiliary ellipsis seen in (38). Thus, finite auxiliaries can be RNRed (42), but not gapped (38). (42)

(43)

Finally, comparatives, just like coordinations, tolerate RNR in certain noncoordinate contexts. The status of RNR in marginally acceptable subject relative clauses ((44a)/(45a); Hudson 1976) parallels that of subject comparatives ((44b)/(45b); Lechner 2004). In (45b), RNR generates a PC. (p. 641)

(44)

Page 20 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

(45)

(p. 642)

These findings corroborate the hypothesis that comparatives can be modulated by

RNR.

25.4.1.3 Across-the-board movement Direct evidence for ATB movement in comparatives comes from examples like (46a), in which relativization reaches both into the matrix clause and the than-XP, suggesting an analysis that assigns to (46a) a representation similar to that of the coordinate structure in (46b):12 (46)

(47) adds a semantic argument for the existence of ATB movement in comparatives. The missing indefinites in the coordination (47a) and the comparative (47b), respectively, are interpreted as variables bound by the existential subject. Given that there is no ‘small conjunct’ analysis for comparatives, which would equally be able to derive the intended reading (see Hirsch 2016 for recent implementation), it is hard to escape the inference that the subject has been removed by ATB movement: Page 21 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (47)

ATB is also implicated in the curious exception to the ban on auxiliary gapping in (39). The puzzle disappears once it is recognized that the verb (haben) is not gapped but has reached its surface position by ATB verb-second movement (ATB-V2), as in (48) (Lechner 2004: 3.4): (48)

That the deletion process at work in comparatives targets a genuine coordinate structure becomes apparent by inspecting the effects of the Coordinate Structure Constraint (CSC) in (p. 643) paradigm (49). (49a) illustrates a legitimate combination of two ATB movements, viz. V2 and subject raising. No conflict arises, because both operations presuppose a coordinate parse. But whenever one of the two movements skips a conjunct, the results drastically degrade. In (49b/c), ATB subject raising to specTP establishes a coordinate structure, such that asymmetric V2 is bound to incur a violation of the CSC (traces that do not abide by the CSC are marked by ☛). Conversely, (49d/e) are blocked by the CSC due to illegitimate asymmetric subject movement out of a coordination established by ATB-V2: (49)

Page 22 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

The effects visible in (49) are replicated for ATB-V2 and object extraction in (50): (50)

Finally, what (51) demonstrates is that the coordinate parse for comparatives is, unless forced by independent factors such as ATB movement, optional. In the absence of other ATB dependencies, the than-XP can be parsed into a position subordinate to the (p. 644)

Page 23 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion matrix clause, which enables syntactic operations to target the two clauses individually, accounting for the availability of asymmetric V2-movement of the finite auxiliary hat (see Lechner 2001, 2004, 2015 for details): (51)

In sum, the standard diagnostic for coordination (CSC) yields solid evidence that comparatives which include ATB movement establish a coordinate structure. This is expected if the symmetric deletion processes observed in (47b), (48), (49a), and (50a) actually consist in ATB movement.

25.4.1.4 Other ellipsis operations in comparatives Additional ellipsis operations that have been observed to target comparatives include MODAL COMPLEMENT ELLIPSIS ((52b); Aelbrecht 2010), and NULL COMPLEMENT ANAPHORA (NCA; (53)). Just like NCA in coordinate structures ((54); Jacobson 1992c), ellipsis is possible with control, but not with raising predicates: (52)

(53)

(54)

In German, NCA co-occurs with the expletive es, a requirement that carries over to comparatives: (55)

Page 24 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Interestingly, comparatives permit SENTENTIAL ARGUMENT ELLIPSIS also with verbs like think and seem, which usually do not license NCA (Kennedy and Merchant 2000b; Pancheva 2012, attributing (56) to Irene Heim): (56)

(57)

(p. 645)

The effect, yet to be accounted for, appears to be restricted to stative predicates:

(58)

An additional puzzle is raised by German sentential argument ellipsis in (59). The contrast in (59) demonstrates that the deletion may consist of a discontinuous string made up of a finite auxiliary (wurde) and the complement clause, provided that the stranded participle is embedded in a passive clause. (The active version (59b) is wellformed with a finite auxiliary): (59)

Page 25 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion To summarize, the observations collected in the last four subsections provide strong support for the claim that comparatives can be targeted by the same kinds of deletion operations—specifically gapping, RNR, and ATB movement—that are operative in coordinate structures. Sentential argument ellipsis, which is attested in comparatives only, is a to date poorly understood exception to this generalization. Note, incidentally, that in all the examples examined up to now, ellipsis and ATB movement do not discriminate between PCs and elliptical comparatives with more than one remnant. This parallelism will become relevant again in the discussion of different approaches towards PCs in section 25.5.

25.4.2 Coordination vs subordination Even though the question why comparatives are able to emulate the behavior of coordinations is largely unexplored, there are a few speculations in the literature. In principle, three factors have been identified to influence the availability of a coordinate parse for comparatives: linearization, non-containment, and morphosyntactic properties of the standard marker (than in English). The contrast between (60a/b) and (60c/d) illustrates that gapping and ATB movement in comparatives is contingent upon extraposition of the than-XP, suggesting that the degree (p. 646) complement must, just like regular non-initial conjuncts, be properly linearized to the right of the matrix clause. Notably, (60c) can only be blocked by linearization conditions on the whole conjunct since (60c) satisfies the requirement that the antecedent precede the gap: (60)

Proper linearization can be effected in various ways: by extraposition of the than-XP (Pinkham 1982; Lechner 2004); by Late Merge of the than-XP subsequent to covert rightward movement of MORE (which is another strategy to model extraposition); by overt QR of the DegGQ; or by using enriched phrase structures, such as multidimensional trees (Moltmann 1992) or multidominance (Sabbagh 2007; Gračanin‑Yüksek 2007). Second, elliptical comparatives in which the than-XP contains hidden structure are constellations of antecedent contained deletion (ACD; Wold 1995). The than-XP accordingly has to attach to a node c-commanding the ellipsis site at the point in the derivation where ellipsis is resolved. Importantly, while containment is related to linearization, the two criteria do not define the same class of expressions. In the structure (61a), the degree clause is properly linearized, but still contained inside the (VP)-node

Page 26 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion which embeds the antecedent. Hence, ACD-resolution requires re-bracketing, for instance as in (61b): (61)

As with linearization, the proper factorization can be achieved by movement of the than-XP or the degree quantifier. Which of these options is best suited to account for the coordinate behavior of comparatives is not known at the moment. A third factor apart from linearization and containment implicated in coordination formation is the lexical inventory of standard markers provided by a language. Greek, Polish, and Serbo-Croatian, for instance, distinguish between two versions of than (see (70)), only one of which licenses ellipsis. Conversely, Hankamer (1973b) suggested that English than is ambiguous between a coordinating complementizer and a preposition. A modern implementation of this line of thought is Alrenga et al. (2012). In sum, the fact that comparatives replicate the restrictions on deletion operations typically attested in coordinate structures makes it seem likely that the processes at work are the same. Naturally, it was not possible in the confines of the presentation to demonstrate that the analysis is complete. Further evidence in support of the claim that all deletion in comparatives is reducible to the union of the operations exclusively targeting coordinations (gapping, ATB movement, RNR) and more liberal deletion processes also attested in subordinate contexts like VP-ellipsis or pseudogapping (Kennedy and Merchant 2000a) is collected in Hendriks (1995) and Lechner (2004).13 (p. 647)

25.5 Phrasal comparatives From the early 1970s on, there has been an intense debate about the proper treatment of PCs, illustrated in (62). (62)

Scholars have noticed that PCs meet certain, but not all criteria typically associated with elliptical constructions (Hankamer 1973b), which led to the formation of two groups of approaches: the DIRECT ANALYSIS (DA), which analyzes PCs as base-generated PPs headed by a prepositional version of than, and the REDUCTION ANALYSIS (RA; terminology by Bhatt and Takahashi 2011), adherents of which hold that PCs embed hidden syntactic structure.14 In general, the standard heuristics for detecting the effects Page 27 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion of RA has been that PC formation by ellipsis affords phenomenologically richer structures than by DA. Over the last fifteen years, a consensus has emerged that the choice between DA and RA is not universal, but subject to typological variation, such that some languages employ DA, while others opt for RA. More recent results indicate that both strategies might be employed even within a single language. This final section traces synoptically the most important steps of this still ongoing controversy, which not only affects the analysis of comparatives and ellipsis, but also has broader repercussions for the theory of the lexicon, language typology, and learnability, among others. The remainder of this section falls into three parts. Reporting findings from classical studies of the 1970s and 1980s, as well as two more recent studies, section 25.5.1 lists in a synoptic form arguments in support for DA, accompanied by a brief critical assessment. Next, section 25.5.2 presents the compositional semantics for PCs, turning from there to evidence in support of RA in 25.5.3. Before proceeding, a final note on terminology: The DP following than will be referred to as the REMNANT, while the category which takes up the same grammatical function as the remnant in the main clause is the CORRELATE. In (62a), for example, Bill serves as the remnant and Ann is the correlate.

25.5.1 Arguments for DA Arguments for DA, and thereby for the claim that at least some PCs are base-generated and lack hidden syntactic structure, include observations about the morphology of the standard marker or the remnant, the syntax of PCs, and interpretive properties exclusively found with PCs. In the remainder of this section, I will list arguments in support of DA culled from the literature. (i) English PCs with subject accusative remnants (63a) lack a clausal source (63b), and have therefore been argued to be base-generated (Hankamer 1973b; Napoli 1983a). But accusative subjects are also found in elliptical conjunctions (63c) and fragment answers (63d); Merchant 2004a), suggesting that the contrast (63a) vs (63b) does not reflect the clausal vs phrasal distinction, but is an artifact of whatever mechanism is responsible for the distribution of default case (Lechner 2004; Merchant 2008b). (p. 648)

(63)

(ii) Extraction and preposition stranding is possible from PCs only ((64a); Hankamer 1973b). At first sight, the best analysis of (64a) seems to reside with the assumption that

Page 28 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion the remnant is a base-generated prepositional complement which has been separated from its head under preposition stranding: (64)

There is another conceivable explanation of the phenomenon, though, which treats (64a) as an instance of island repair under ellipsis (Merchant 2004a, 2008b). One possible implementation of this idea involves the two assumptions that degree complements are inherently islands for extraction, and that the ellipsis in (64a), but not in (64b), is large enough to include all offending intermediate traces. As a result, the representation of (64a), relevant parts of which are given in (65a), does not contain any illegitimate syntactic objects, while (64b), where ellipsis affects a node lower than TP (65b), embeds at least one non-locally bound trace: (65)

On this conception, (64a) could be elliptical, after all. Naturally, various challenges remain, most prominently the observation that ellipsis does not amend island violations in comparatives (66a) in the same way it does under sluicing (66b): (66)

However, given that as to date, there is no satisfactory account for the contrast in (66), the island repair analysis of (65) cannot be excluded as a plausible analytical alternative to DA. (iii) Reflexives can be externally licensed in PCs, but not in clausal degree complements, suggesting that PCs lack a clause boundary, as predicted by DA (Hankamer 1973b; Brame 1983): (67)

Page 29 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion Again, an alternative interpretation of the data exists, though, which is equally compatible with RA. It is also possible to treat (67a) as a small clause isomorphic to (68a). The small-clause analysis has the added benefit that it provides a reason for why the subject remnant in (63a) surfaces with accusative case: it is a small-clause subject, just like the embedded subject of (67b): (p. 649)

(68)

That clausal degree complements do not always have to include functional heads but also come as small clauses was already seen in (24), repeated below as (69). In the intended readings, the remnant is followed by a single adjectival (69a) or nominal (69b) projection made up by the CD-site: (69)

(iv) On the DA, the PC remnant is a prepositional complement. Since prepositions generally select a single, nominal complement, one is led to expect that genuinely phrasal comparatives admit only a single remnant (single remnant restriction) and resist prepositional remnants. Although these predictions are not confirmed for English, which generally treats PCs as reduced clauses (see section 25.5.3 for justification), they accurately characterize languages such as Greek, Polish, Russian, and Serbo-Croatian, where base-generated PCs and clausal comparatives are morphologically distinguished by the shape of the standard marker (Pancheva 2006; Merchant 2009). To exemplify, Greek PCs formed with the standard marker apo display all properties typically associated with the DA, while the particle ap’oti introduces elliptical, clausal comparatives. This division of labor is reflected by the fact that only apo-PCs abide by the single remnant restriction (70) and the prohibition on prepositional remnants ((71); Merchant 2009: 139, (21)–(22)): (70)

Page 30 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (71)

Furthermore, as will be seen in some detail in 25.5.3, whenever a language makes available a clausal and a prepositional strategy to form PCs, RA and DA properties cluster together. (p. 650)

(v) In some languages, among them Greek and Polish, the formation of PCs is limited to certain grammatical functions. As observed by Pancheva (2006), in those Slavic languages that possess a designated standard marker for base-generated PCs, the comparative DP must not surface as a subject. The same restriction is operative in attributive comparatives in German (Lechner 1997, 2016):15 (72)

In Slavic (as well as in German), the judgments are gradient and subject to speaker variation, but clearly distinguish between PCs that fall under the DA and those which are derived by RA. To recapitulate, even though morphosyntactic and syntactic criteria (i)–(iii) provide some suggestive initial evidence for the existence of base-generated PCs, none of the classical arguments for DA survives exposure to closer scrutiny; in all three cases, alternative ellipsis analyses for the phenomena were seen to exist, partially with broad empirical coverage. By contrast, criteria (iv) and (v), which entered the debate more recently, provide solid evidence for the hypothesis that at least in some languages, some PCs are base-generated, as posited by DA.

25.5.2 Interpreting PCs PCs can be interpreted compositionally by defining a homomorphic lexical entry for the comparative degree head that applies to the remnant meaning first, and then combines with the gradable property and the subject denotation, in any order (Heim 1985). A version of this three-place version MORE3, adopted from Bhatt and Takahashi (2011), is given in (73). On this view, first explicitly argued for in Hankamer (1973b) and recently Page 31 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion defended in Bhatt and Takahashi (2011, henceforth BT), more is lexically ambiguous between (73) (p. 651) and the two-place interpretation MORE2 ((8), repeated below), which generates clausal comparatives:16 (73)

(8)

BT defend the conjecture that the choice between MORE2 and MORE3 is subject to crosslinguistic variation, in that some languages, among them English and German, treat PCs as reduced clauses formed with the help of MORE2, while others parse them as basegenerated PPs introduced by MORE3 (Hindi-Urdu). Still others, such as Greek (Merchant 2009) and Polish (Pancheva 2006) employ both options. (74) provides a sample derivation for predicative PCs based on the individual comparison degree head MORE3, which proceeds without changes in the surface constituency: (74)

By contrast, the overt representations of nominal attributive PCs such as (75a) are not compositionally interpretable, given the meaning rule (73), and need to be manipulated by two covert movement operations (Bhatt and Takahashi 2007; Kennedy 2007). First, the correlate Sue moves (① in (75b) and the binder index is attached to the sister node of the moved category (Heim and Kratzer 1998). Then, the degree quantifier [DEGQP MORE3 than Ann] raises (②) and lands in-between Sue and the binder index of Sue, resulting in a configuration of PARASITIC SCOPE (Barker 2007; Sternefeld 1998; Nissenbaum 1998; Beck and Sauerland 2000). As detailed by (75c), Parasitic Scope supplies DegQP with a two-place relation between degrees and individuals. (75)

Page 32 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Note that, somewhat surprisingly, PCs in which the remnant consists of a measure phrase instead of an individual term are not handled by MORE3 but by the clausal version MORE2, at least if measure phrases are assumed to denote sets of degrees (Schwarzschild 2005). (p. 652)

(76)

From this it follows that languages which only have access to MORE3 should not sanction measure phrase remnants, a prediction which still deserves further study. Turning to a first set of empirically falsifiable consequences of the DA, the specific implementation above entails the five claims in (77): (77)

Corollary (77a) predicts that base-generated PCs should not display Principle C effects of the sort seen in (26a). One is accordingly led to expect that the Greek PC (78a) should contrast with its ill-formed clausal variant (78b). This prognosis is not confirmed, though: (78)

Page 33 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

However, as pointed out by a reviewer, the coreference pattern (78a) is still amenable to DA, given that at LF, the pronominal remnant (auton/‘heACC’) c-commands the predicate containing the name (Gianni), inducing a Principle C violation. Second, BT demonstrate that the second consequence of the DA, according to which the than-XP does not embed unpronounced structure, is empirically supported by the absence of disjoint reference effects in PCs in Hindi and Japanese, among others. Discussion of details will be postponed to section 25.5.3. Third, as Heim (1985) pointed out, covert movement of the remnant and the correlate should be subject to syntactic island constraints. That this is correct can be inferred from the observation that the degree head FEW in (79a) is unable to scope out of a relative clause. Movement of FEW together with the remnant, which is required to generate the interpretable LF-representation (79b), incurs a strong violation of the Complex NP Constraint:17 (79)

(80a), also discussed by Heim, demonstrates that correlate raising is equally bounded by locality. As revealed by the pertaining LF (80b), the correlate (the clarinet) must not bind a trace across a complex NP barrier: (p. 653)

(80)

While supplying a prima facie argument in support of DA, it should not go unnoticed that the paradigms in (79) and (80) also find a natural explanation under the ellipsis analysis (Lechner 2004). This is so because gapping in (79) and (80) removes a finite CP, in violation of constraint (iv) of section 25.4.4.1 (see (36)). The pertinent underlying representations are given in (81). (81)

Page 34 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

Fourth, BT observe that the DA requires quantificational remnants to leave their surface position, because MORE3 selects for an individual term as its first argument. More precisely, the remnant needs to QR and is therefore obligatorily assigned wide scope with respect to the correlate. (82) depicts the relevant relations schematically (for examples see BT, pp. 603–4): (82)

BT show that effects of this requirement are visible in languages that unambiguously treat PCs as base-generated, among them Japanese and Hindi-Urdu. A further prediction, which has not been attended to in the literature, is that in these languages, NPs that resist wide scope such as bare indefinites, should not be able to function as remnants. Thus, the structure corresponding to (83) should be ill-formed in Japanese and HindiUrdu.18 (83)

Finally, the derivation of attributive PCs should reflect properties generally characteristic of Parasitic Scope constellations. Positive evidence to this effect is discussed in Lechner 2016. The information accumulated so far lends plausibility to the idea that PCs also have a base-generated parse. Further substantiation for this claim will be presented in the (p. 654) following section, which summarizes the results of RA, and traces current and recent developments which lead to a partial synthesis incorporating aspects of both approaches.

25.5.3 Arguments for RA Proponents of RA have recruited the following arguments in support of the claim that the degree complement of PCs is endowed with abstract syntactic representation. (i) The ellipsis analysis offers a natural explanation for the fact that in languages such as German, the case of the remnant matches that of the correlate (Heim 1985): (84)

Page 35 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

This is unexpected on DA in its present form, because the representations of DA do not supply the remnant with a case assigner. While it is possible to amend this shortcoming— for example, by relegating case assignment to LF after movement of the two DPs in configurations akin to (85)—RA clearly offers a more concise explanation for case matching. (85)

(ii) English PCs obligatorily extrapose (Pinkham 1982: 108; but see also Bhatt and Takahashi 2007). The DA lacks the means to express this linearization and/or containment condition, which excludes (86b): (86)

(iii) If the degree clause contains hidden structure, one is led to expect that names embedded in the remnant should not be able to corefer with pronouns that c-command the correlate, as expressed by generalization (87) (Lechner 2004; improved examples and (87) from BT): (87)

(88) documents that this prediction is confirmed for English PCs. As the LFs in (89) reveal, the pronoun c-commands the name inside the degree complement in (88a), but not in (88b): (88)

Page 36 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (89)

By contrast, the DA assigns to both sentences representations in which the name is free. Hence, the disjoint reference effect in (88a) remains unaccounted for: (p. 655)

(90)

BT observe that not all languages react to Principle C alike. In Hindi-Urdu, for one, structures isomorphic to (88a) admit coreference (Bhatt and Takahashi 2011: (35)): (91)

They take this to signal that English assigns PCs a clausal analysis and employs MORE2, while Hindi-Urdu parametrically has access to MORE3 only. This conjecture is, as BT show, supported by the systematic clustering of properties such as scope (see (82)) and the single remnant condition, illustrated for Hindi-Urdu in (92): (92)

(iv) The ellipsis analysis leads one to expect that PCs display the same characteristics as partially reduced comparatives, i.e., constructions in which the than-XP either contains more than one remnant or a remnant that is not nominal. As shown by Lechner (2004; see also section 25.3.1) this prediction is borne out for reduction languages such as German and English. (v) A well-defined subclass of PCs admit tense mismatches between the matrix clause and the reconstructed degree complement (McCawley 1998; Pinkham 1982; Lechner 2004, 2016): (93)

Page 37 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion

The existence of atemporal readings is not expected on DA, because MORE3 λ-converts an exact copy of the matrix predicate, including the temporal specification, into the than-XP. Specifically, DA generates the LF for (93) in (94a), which is mapped into the meaning (94b): (94)

(p. 656)

Evidently, the mechanics of DA do not provide the means to derive mismatches in

temporal specification, indicating that PCs that admit atemporal readings are derived by ellipsis.19 Taken together with the findings from 25.5.1, where it was concluded that arguments for base generation on closer scrutiny turn out to be equally compatible with RA, the five generalizations listed in the present section provide additional support for the existence of hidden structure in the representation of PCs. Moreover, there is also clear evidence for the competing base-generation approach. In light of these two conflicting sets of facts, a consensus has emerged that the model currently best suited to handle the full range of phenomena is synthetic, incorporating both RA and DA, subject to typological variation and possibly other parameters (Kennedy 2009).

25.6 Conclusion Unpronounced constituents in comparatives fall into three discrete classes, none of which, it was argued, requires designated syntactic or semantic mechanisms: (i) the silent degree variable created by empty operator movement; (ii) material affected by Comparative Deletion, which targets degree predicates (including the common noun in attributive comparatives) and consists, I have suggested, in a version of the raising operative known from NP-movement; and (iii) non-construction-specific ellipsis operations. The present chapter has set out to demonstrate that attempts at bringing this idealization closer to reality have been successful in many respects. Naturally, a number of questions regarding ellipsis in comparatives remain under debate at the point of writing, among them (i) how to rein in the zoo of possible meanings for the degree head MORE (with important consequences for delimiting the class of PCs that are not elliptical); (ii) the inner architecture of degree predicates (affecting the analysis of how Page 38 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion much ellipsis is tolerated in amount and degree comparatives, respectively; Lechner 2016); (iii) and numerous issues related to the scope of the degree relation (Beck 2011).20

Notes: (1) Two notable exceptions are MEASURE PHRASE COMPARATIVES (see (i)), which are non-elliptical on all extant accounts, and COMPARATIVE SUBDELETION ((ii); Corver 2006; Lechner and Corver 2017); subdeletion will be ignored throughout. ((i))

((ii))

(2) Than will be assumed to be vacuous throughout; see Alrenga et al. (2012) for an analysis in which than has a semantic contribution. The empty operator will be ignored for the moment. (3) Concerns that QR affects an intermediate projection (Deg’) can be defused by adopting Bhatt and Pancheva’s (2004) Late Merge hypothesis or Alrenga et al. (2012); see discussion of (14). Just like covert movement of nominal quantifiers, QR in (10a) leaves a movement copy (not represented). Again, see the discussion of Late Merge surrounding (14). (4) The full paradigm also involves a licit reading on which MORE and the degree clause take matrix scope but the ellipsis is identified narrowly, by the embedded VP works hard ((69c) in Bhatt and Pancheva 2004). This documents the well-known, yet orthogonal, phenomenon that VP-ellipsis can reach into embedded clauses in search for their antecedents (Fiengo and May 1994). (5) On why the two occurrences of DegP in the matrix clause and fronted position of the than-XP do not need to be strictly identical (MORE long vs Deg° long), see Kennedy 2002: 590. For an analysis that observes strict identity, see Alrenga et al. (2012). (6) For the original analysis, on which adjectives denote measure functions, see Kennedy (2002: 572–3). (7) The problem does not in arise in head-external relative clauses, because the node that is moved prior to Late Merge is also the head of the relative. (8) It has been suggested that there is an additional functional layer above DegP, which hosts measure phrases and more in synthetic comparatives (5cm more distant). On this view, the [E]-feature is hosted by the head of this projection, and ellipsis affects the complement, instead of the specifier of DegP. Crucially, adopting the more fine-grained Page 39 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion structure ensures uniformity with Merchant (2001) and subsequent work, according to which [E]-features always affect the complement domain. (9) Multiple LF-copies are also attested in other domains. On the raising analysis of relative clauses, the higher copy can but, crucially, does not have to be ignored by the computation. The same rationale is also underlying the explanation of disjoint reference effects with A’-movement and QR. For instance, (i.a) is assigned the LF in (ii), in which both copies are interpreted (see Fox 2000; Sauerland 1998 for details): ((i))

((ii))

(10) Note incidentally that the same complication shows up in relative clauses, yet with diametrically opposite effects. Extraposed, i.e. Late Merged, relative clauses actually resist a raising construal, as witnessed by the absence of reconstruction effects, among others (see Bhatt 2015 for an overview). (11) For exceptional strict readings see Hestvik (1995), Kennedy and Lidz (2001), and McKillen (2016). (12) A reviewer points out that the well-formedness of (i) suggests that (46a) does not implicate ATB-movement, but is a parasitic gap (Hendriks 1995). While this is a viable analytical option, (i) is also compatible with the view that comparative coordination is optional and that the degree clause in (i) is subordinated. See discussion of German V2 in (49) and (50) for evidence in support of this view. ((i))

(13) On ellipsis in comparatives see also Napoli (1983a); Moltmann (1992); Hendriks (1995); Pinkham (1982); Lin (2009), among many others. Complications for Lechner (2004) are noted in Osborne (2009). (14) On the debate surrounding DA vs RA see also Hankamer (1971, 1973b), Bresnan (1975), Brame (1983), Pinkham (1982), Napoli (1983a), Heim (1985), McCawley (1998), Kennedy (1999, 2009), Xiang (2003), Lechner (2004, 2016), Pancheva (2006, 2010), Beck et al. (2009), Lin (2009), Merchant (2009), and Sudo (2014), among others.

Page 40 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion (15) More generally, the restriction demands that in attributive degree comparatives, the correlate c‑commands the comparative DP. For unknown reasons, the effect disappears with numerical/amount DPs, the German equivalent of More conductors supported the composers than the directors is impeccable. See Lechner (2016) for discussion. (16) Kennedy (2009: (40)) observes that MORE3 can be defined in terms MORE2, but not vice versa, creating the expectation that no language uses MORE3 only; this seems, as Kennedy notes, to be correct. (17) (66) showed that remnant movement in PCs differs from remnant movement under sluicing in that—for yet undisclosed reasons—only the latter grants amnesties to island violations. (18) A reviewer notes that the example is acceptable in Hindi, but that bare nouns in that language behave like kind names. A more complete investigation of the issue has to await another occasion. (19) Interestingly, the availability of atemporal readings is syntactically restricted in similar ways as attributive nominal comparatives in German or base‑generated PCs in Slavic: in atemporal PCs, the correlate has to c‑command the comparative DP. See Lechner 2016 for discussion. (20) To close with a puzzle observed by Larson (1988b), universals in subject position can scope over matrix clause negation in PCs but not in clausal comparatives: ((i))

((ii))

Winfried Lechner

Winfried Lechner is Associate Professor for German Linguistics and Theoretical Linguistics at the National and Kapodistrian University of Athens, Greece. His main academic interests are located in the areas of syntax, semantics, and the interaction between these two components. He has been working on the logical syntax of scope and reconstruction, reflexivization, comparatives, ellipsis, the cross-linguistic Page 41 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Comparative Deletion typology of same/different, additive and scalar focus particles, Duke of York opacity, and the architecture of the grammar.

Page 42 of 42

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

Oxford Handbooks Online Null Complement Anaphora   Marcela Depiante The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.28

Abstract and Keywords In this chapter, the different positions regarding NCA in the literature are discussed and the analysis proposed in Hankamer and Sag (1976) is favored. NCA is taken to be a type of deep anaphor as opposed to a surface anaphor, in Hankamer and Sag’s proposed typology. It behaves like a deep anaphor in that it can take pragmatic antecedents, missing antecedents, and does not seem to require strict syntactic parallelism with its antecedent. In addition, NCA does not allow overt or covert extraction out of it, which provides evidence in favor of the view that NCA is to be represented as an element with no internal syntactic structure. It is also shown that the types of predicates that select NCA seem to be lexically determined, since no reliable pattern has yet been found to be able to predict which predicates select it. Keywords: Null Complement Anaphora (NCA), deep anaphor, surface anaphor, typology of anaphora, extraction, phonetically null element

Marcela Depiante

26.1 Characteristics of Null Complement Anaphora 26.1.1 The basic data The phenomenon of Null Complement Anaphora (henceforth NCA) refers to the null complements of certain verbs as in sentences (1) and (2). (1)

Page 1 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

(2)

The possibility of having null complements is only available to a certain group of verbs as we see from the ungrammaticality in (3):1 (p. 658)

(3)

26.1.2 The questions In this chapter we will look at the behavior of these null elements. What are their properties? Which predicates allow them and which ones do not and why? What syntactic category/categories do they belong to, and what type/s of semantic interpretation can they take? In addition, we will discuss the theoretical questions that they raise. The main question that arises with the null elements to be analyzed in this chapter is how the grammar can account for them given that they are not pronounced but nonetheless interpreted. What is their structure, how are they represented, how are they interpreted? Are they the same type of null element as other elliptical constructions, like VP-ellipsis? When asking how the syntax accounts for the non-pronounced strings that are the focus of this chapter, there are a number of logical possibilities available as answers. I list these logical possibilities below: a. Nothing in the syntax is present in the position of the null complement. In other words, what you see is what you get.2 b. The null element is the result of a deletion operation (such as PF-deletion) and the full syntactic structure is available. c. The null element is not the result of a deletion operation. Instead, a null pro-form with no internal syntactic structure is present in the syntax. Page 2 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora d. The null element is not the result of a deletion operation. Instead, a null pro-form with some/full internal syntactic structure is present in the syntax. e. Some other combination of the options mentioned above. Some version of the possibilities given above has been put forth in different proposals by various authors in the literature since the 1970s. We will present and discuss these proposals in the next section of this chapter.

26.2 Previous studies The null complements illustrated in (1) and (2) have been referred to in the literature by various names. Shopen (1972, 1973) calls this feature definite constituent ellipsis, Hankamer and Sag (1976), Grimshaw (1979), and Napoli (1983b) call it Null Complement Anaphora. Fillmore (1986) calls it Definite Null Complement and Saeboe (1993, 1996) uses the term definite ellipsis. Huddleston and Pullum (2002) use the label lexical verb complement ellipsis. In this section, we present each one of the different proposals. (p. 659)

26.2.1 Shopen (1972, 1973)

Shopen (1972, 1973) called examples like the ones in (1) and (2) definite constituent ellipsis. He notes that some verbs allow definite constituent ellipsis and others do not. Whether a verb allows it is a lexical issue, i.e., that information must be encoded in the lexical entry of the verb. So, according to Shopen, the fact that we can say Henry explained but not Henry expected has to do with the lexical properties of the verbs explain and expect. Shopen notes that definite constituent ellipsis can be used either anaphorically or deictically. If used anaphorically, the null argument has an antecedent within the linguistic context as in (4). (4)

When used deictically—as in (5)—there is no linguistic antecedent present in the discourse. The speaker “relies on knowledge shared with the hearer or the obviousness of the identity of the referent within the perceivable physical context for the utterance” (Shopen 1972: 151). (5)

Page 3 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora In addition, Shopen (1972, 1973) claims that the grammar of sentences such as (4) is just what you see. In other words, he does not propose any syntactic structure in the position of the null complement. The null complement is not a consequence of a deletion transformation.

26.2.2 Hankamer and Sag (1976) 26.2.2.1 Deep vs surface anaphora Hankamer and Sag (1976) put forward a mixed theory of anaphora. They propose that there are two types of anaphora, independently of whether they are null or not. They call them deep and surface anaphora. In their system, deep anaphors are not derived via an operation (or transformationally, in their terms) but are present in underlying representations, while surface anaphors are derived via an operation (or transformationally). Null Complement Anaphora is—in their system—a type of deep anaphor. According to Hankamer and Sag (1976: 421–5), deep and surface anaphors differ in the following ways: (6)

Given their typology, Hankamer and Sag (1976) classify different anaphoric constructions as either deep anaphora or surface anaphora, as summarized below: (p. 660)

(7)

What Hankamer and Sag (1976) call deep anaphors are anaphors that can get their interpretation either from a linguistic or a non-linguistic antecedent. Independently of one’s theoretical position and perspective on the details of Hankamer and Sag’s proposal, the gist behind Hankamer and Sag’s typology is still empirically accurate in the sense that all languages have anaphors that need linguistic antecedents and anaphors that can get their interpretation from the non-linguistic context. In that sense, any analysis of NCA has to take this fact into account, because NCA is an anaphor that can get its interpretation both from the discourse and from the non-linguistic context.3 In the following section, we will concentrate on how NCA behaves with respect to Hankamer and Sag’s (1976) three main tests for distinguishing between deep and surface Page 4 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora anaphors: whether an anaphor can have a pragmatic antecedent or not, whether it can host a missing antecedent, and whether it needs to have an antecedent that is syntactically parallel or not.

26.2.2.2 Hankamer and Sag’s (1976) tests (i) Pragmatic antecedents Hankamer and Sag propose that anaphors split into two groups depending on whether they can take a non-linguistic antecedent or not. NCA and the Do It pro-form can take a pragmatic antecedent while VP-ellipsis cannot,4 as illustrated in (8): (8)

(ii) Missing antecedents Hankamer and Sag (1976) argue that only surface anaphors exhibit the missing antecedent phenomenon previously discussed in Grinder and Postal (1971) and Bresnan (1971). Deep anaphors cannot contain an antecedent for a pronoun that follows it, while surface (p. 661) anaphors can. The data in (9) show that NCA and the Do It pro-form pattern together with respect to this test. On the other hand, VP-ellipsis behaves differently, in that it can contain the antecedent for a following pronoun and therefore patterns as a surface anaphor according to the typology proposed by Hankamer and Sag (1976).5 (9)

The idea here is that a VP-ellipsis site contains enough syntactic structure to contain an antecedent for the following pronoun while the pro-form Do It or an NCA-site do not.

(iii) (Non)-strict syntactic parallelism with the antecedents Hankamer and Sag’s (1976) third test for distinguishing between deep and surface anaphora is to see whether the anaphor in question demands strict syntactic parallelism with its antecedent. Deep anaphors do not require such strict syntactic parallelism with their antecedent, as shown with the pro-form Do It and NCA in examples (10b) and (10c). Page 5 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora In contrast, VP-ellipsis must have strict syntactic parallelism with its antecedent, as we see in (10a). (10)

26.2.3 Grimshaw (1979) Within the context of her discussion of categorical and semantic selection, Grimshaw (1979: 288–97) discusses the phenomenon of NCA. She proposes that NCA has no syntactic representation at all. In terms of how to get the interpretation for NCA, Grimshaw proposes the Null Complement Rule, which is a discourse rule that applies after sentence grammar (along the lines of the VP Deletion Rule in Williams 1977b): (11)

6

into the complement position of the verb.

Grimshaw limits the rule to copying “sentential expressions” only, which means that a predicate alone or a term alone cannot be copied. This is shown in the unacceptable examples in (12b) and (13b): (p. 662)

(12)

(13)

Equipped with the Null Complement Rule, Grimshaw can account for the fact that an NCA can have a concealed question as an antecedent. In (14) Grimshaw proposes that the concealed question ‘what time it was’ is copied onto the NCA, not the NP ‘the time’. (14)

Page 6 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

In addition, she observes that whatever restrictions are imposed on the semantic type of the complement of these verbs will be imposed on the null complements of these verbs. So, for example, the verb ‘know’ can take a complement that is semantically a question as in ‘I know who left’ or a proposition as in ‘I know that John is telling lies again’. When this verb selects an NCA, this NCA will be able to be interpreted either as a question or as a proposition, just as if the complement had been overtly expressed, as shown in the contrasts in (15) and (16): (15)

(16)

With respect to which verbs can take NCA, Grimshaw (1979: 291) observes that “it is very unlikely that the difference between the predicates allowing null complements and predicates which require filled complements is predictable.” So, in her system, the subcategorization frame of the verb will need to include that information.

26.2.4 Napoli (1983b) Napoli proposes that examples such as (17) and (18) should be analyzed as intransitive uses of the verb. (17)

(18)

She argues that the representations in (19) with deletion or with a base-generated null pro-form are not the right analyses for NCA. (p. 663)

(19)

Page 7 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

Napoli argues that the representations in (19) are not possible representations for NCA. What you see in (17) and (18) is what you get. According to Napoli, there is nothing present in the syntax which will later be deleted, nor is there an empty element present at the base. One of the arguments that she gives against a deletion account and a basegenerated account is that these elements do not seem to obey the Backwards Anaphora Constraint (BAC), if they were either the product of ellipsis (of the VP-ellipsis type) or base-generated elements (as is the case with pronouns). Napoli (1983b) presents examples such as the one in (20a) in favor of her position. However, Hankamer and Sag (1976: 412) had already shown that NCA does in fact obey the Backward Anaphora Constraint as illustrated in their examples given in (20b). (20)

There seems to be disagreement with respect to the possibility of BAC with NCA and that would be something interesting to explore in future research. To summarize, Napoli takes verbs that select NCA to be the manifestation of intransitive uses of those verbs and proposes that the interpretation is done in the pragmatics. To analyze NCA as an intransitive use of a predicate is incompatible with the fact that NCA is not interpreted in a generic way as are null complements of verbs like eat when used in an intransitive way. Such an analysis cannot account for the anaphoric nature of NCA. This had already been pointed out by Hankamer and Sag (1976: 412, n. 21) who show that an example such as the following: I play cards and shoot dice, and my wife doesn’t approve, can only mean that the speaker’s wife disapproves of the speaker playing cards and shooting dice, not of anything in general, which a view such as Napoli’s would predict to be possible.

26.2.5 Fillmore (1986) Fillmore (1986) distinguishes between what he calls Indefinite Null Complement, as in (21) and Definite Null Complement, as in (22). According to Fillmore, Definite Null Complements must be “retrieved from something given in the context” (p. 96) while in Indefinite (p. 664) Null Complements “the referent’s identity is unknown or a matter of indifference” and does not require a context as we can see from the contrast between (21) and (22).

Page 8 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora (21)

(22)

Fillmore also claims that the predicates that can take NCA cannot be semantically derived since synonymous predicates do not all take NCA. However, he notes that there appear to be “some commonalities across word meanings within particular semantic domains in the semantic roles of omissible arguments” (p. 104). These will be discussed in more detail in later sections.

26.2.6 Depiante (2000, 2001) Depiante (2000, 2001) revisits the issue of NCA in English and looks at NCA in Spanish and Italian as well. She concludes that in all three languages there is evidence that NCA is a deep anaphor and that the distinction between deep and surface anaphora is empirically accurate and should be maintained. In addition to the tests put forth by Hankamer and Sag (1976), she takes the following two facts as evidence for NCA having no internal syntactic structure: 1. NCA does not allow overt extraction (overt wh-movement and topicalization). 2. NCA does not allow inverse scope readings. These two properties will be discussed in more detail in the following two subsections.

26.2.6.1 NCA does not allow extraction If in fact NCA—like other deep anaphors—has no internal structure at all, as proposed by Hankamer and Sag (1976), then we would expect there to be no extraction possible out of it. Extraction out of ellipsis sites has been shown in the literature to be a diagnostic for internal structure of that element (see Haïk 1987, Fiengo and May 1994, Johnson 2001b, Houser 2010, and others). One of the ways to show that VP-ellipsis, for example, has internal syntactic structure is by showing that extraction out of it is allowed, as shown by the following examples: (23)

Page 9 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora The reasoning goes like this: if a null string disallows extraction, then that is evidence that there is no internal structure in that element. However, other authors such as Aelbrecht (2010) claim that this implication is not valid. We will discuss all this in detail in section 26.4.1. Consider the following examples of lack of extraction in Do it anaphora from Johnson (2001b) and taken as evidence for lack of internal structure in the anaphor Do it. (p. 665)

(24)

Depiante (2000) shows that NCA does not allow overt wh-movement nor topicalization out of it, as shown in (25a) and (25b) respectively. This is taken as evidence for the lack of internal structure in NCA. (25)

Similar arguments have been made by others in the literature. For example, Merchant (2013a: 538) gives the following minimal pair: a sentence with VP-ellipsis where extraction is possible and a sentence with NCA where extraction is not possible, in order to show that extraction can be used as a diagnostic for the existence of internal structure inside a null element: (26)

Haynie (2010) also uses the lack of extraction to argue for the lack of internal structure in NCA—this will be discussed in section 26.2.9.

26.2.6.2 NCA does not allow inverse scope readings Inverse scope readings are not allowed with NCA, as opposed to their availability with VPellipsis constructions (see Hirschbühler 1982; Fiengo and May 1994; Fox 1995; among others). Sentence (27) illustrates a standard example of a sentence containing quantifiers and no ellipsis that allows two readings, as a consequence of the scope of each quantifier

Page 10 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora phrase: one reading in which ‘some doctor’ has scope over ‘every patient’ and another reading in which ‘every patient’ has scope over ‘some doctor’. (27)

In the examples in (28), we can see the contrast between VP-ellipsis and NCA. Sentence (28a) is the example with VP-ellipsis, and in this case two readings are possible: one in which ‘some nurse’ has scope over ‘every patient’ and another in which ‘every patient’ has scope over ‘some nurse’. In contrast, sentence (28b) contains NCA, and only one reading is possible: the one in which ‘some nurse’ has scope over ‘every patient’. (p. 666)

(28)

In this respect, NCA behaves like other deep anaphors such as Do it in not allowing inverse scope readings, as shown in example (29), where an inverse scope reading is not possible. (29)

26.2.7 Huddleston and Pullum (2002) Huddleston and Pullum (2002: 1527) refer to NCA as lexical verb complement ellipsis and compare it to VP-ellipsis, which they call auxiliary-stranding ellipsis. They observe three differences between lexical verb complement ellipsis and auxiliary-stranding ellipsis, as well as a list of verbs that can license lexical verb complement ellipsis. The first difference that they observe is that lexical verb complement ellipsis is not compatible with relatives, while auxiliary-stranding ellipsis is, as we see in the contrast in (30). (30)

Page 11 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

The second difference that they observe is that in lexical verb complement ellipsis, the full complement must be null, while in auxiliary-stranding ellipsis it doesn’t have to be, as we can see from the contrast in (31). (31)

Finally, the third difference that they observe is that lexical complement ellipsis allows changes of voice while, according to them, auxiliary-stranding ellipsis does not—as illustrated in (32): (p. 667)

(32)

Huddleston and Pullum (2002) is descriptive in nature and no theoretical analysis is proposed.

26.2.8 Cinque (2004) Cinque (2004: 159) discusses NCA when it occurs with Italian restructuring verbs and proposes that in the syntax of Italian, an NCA with a restructuring verb is really the deletion of the Italian overt pro-form far lo (‘do it’). A sentence such as (33a) in Italian would be represented as (33b), according to Cinque (2004). (33)

Under this view, the null element has some internal structure. Cinque’s analysis can account for the lack of object extraction out of the null site, which is also not possible in Italian.

Page 12 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora One could conceive of a proposal that extended Cinque’s (2004) analysis of NCA with restructuring verbs in Italian to English7 where NCA could perhaps be conceived as the product of deletion of the pro-form Do it, which is suggested by Cinque (2004: 181). (p. 668)

26.2.9 Haynie (2010) Haynie (2010) also argues that NCA is a null element without any internal structure. She does a thorough study of NCA in English where she explores the properties of a large sample of predicates that license NCA. She uses Hankamer and Sag’s (1976) tests and demonstrates that NCA behaves like a deep anaphor. She also assumes that the lack of extraction out of an NCA-site indicates that NCA lacks internal structure. Below is a minimal pair where she shows the difference between the possibility of extraction from a VP-ellipsis site as opposed to an NCA-site: (34)

Haynie (2010) provides a thorough critique of Depiante’s semantic account of NCA which attempts to answer the question of the interpretation of these null complements. We will review her critique and her exact proposal in section 26.4.2 where we will address the question of what syntactic category NCA belongs to and what its exact semantics should look like.

26.2.10 Summary and discussion of the different analyses of NCA in the literature Sections 26.2.1–26.2.9 have presented the different proposals regarding NCA that have been put forth in the literature. As we have seen, some authors propose that there is nothing in the syntax that is present in the position where the NCA appears (Shopen 1972, 1973; Grimshaw 1979; Napoli 1983b). Grimshaw proposes an interpretative rule that can account (p. 669) for its interpretation. Shopen does so as well. Napoli, on the other hand, takes NCA verbs to be intransitive uses of those verbs and proposes that the interpretation is done in the pragmatics. We have seen that another set of authors (Hankamer and Sag 1976, Depiante 2000, 2001, and Haynie 2010) propose that NCA is an element with no internal syntactic structure that behaves like other deep anaphors with respect to the tests proposed in Hankamer and Sag (1976). Importantly, NCA behaves like other deep anaphors in that it does not allow extraction out of it, as opposed to what happens with surface anaphors. Under

Page 13 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora these analyses, NCA is not the product of a deletion operation, regardless of how we want that operation to be understood/formulated. Finally, there is another possible analysis which takes a mixed/hybrid view of NCA. In this view, NCA has some internal syntactic structure and does involve some deletion. That is the position taken by Cinque (2004) for Italian and one that could perhaps be extended to English. Under this view, the complement of the verb would be generated as Do it, and later be deleted. This analysis is appealing in the sense that Do it is also a deep anaphor with respect to Hankamer and Sag’s (1976) tests, and doesn’t tolerate extraction. Furthermore, it would predict that all predicates that take NCA also take Do it. However, that prediction is not borne out, because NCA doesn’t just have to be interpreted as a property but also can be interpreted as a proposition. So, such a proposal would not be able to account for sentences where the NCA is interpreted as a proposition, such as in (2) and to be discussed in section 26.4.2.

26.3 Which predicates allow NCA in English? Most authors agree that whether a predicate can take NCA or not is a lexical issue and cannot be derived from semantic facts. Fillmore (1986) presents pairs of nearsynonymous verbs that don’t both license NCA as evidence against a semantic criterion delineating the class of predicates that take NCA; see (35): (35)

In (36), I have listed the English predicates that have been thought to license NCA in the literature. (p. 670)

(p. 671)

(p. 672)

(p. 673)

(p. 674)

(36)

Page 14 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

Page 15 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

Page 16 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

As we see in the table in (36), there is attested variation among speakers. This type of intra-linguistic variation is expected, since it seems that it is a lexical property of particular predicates that determines whether a predicate can take NCA or not. We also (p. 675)

expect that there would be cross-linguistic variation in the predicates that take NCA. That seems to be the case if we look at work on NCA in Spanish and Italian for example (see Depiante 2000, 2001). Authors agree that it seems difficult to find a natural class of verbs that take NCA. Fillmore (1986) argues that the class of verbs that take NCA (in his terms Definite Null Complement) cannot be derived from a common semantics. Nevertheless, he says that verbs that take NCA seem to be “in a great many cases, verbs having to do with causing, inducing or allowing someone to perform an action” (Fillmore 1986: 105). See the examples in (37). (37)

Fillmore (1986) proposes that the right generalization must involve something more than just causation or enablement since we don’t get NCA with verbs like cause, get, or have (which have more general meanings than the verbs that take NCA) but he says that a “social act of some sort” (Fillmore 1986: 105) is involved. He also says that most instances of aspectual complementation also allow NCA, which is true for start, stop, continue, finish, resume, stay, and begin. Somewhat related to Fillmore’s observations is Saeboe’s (1993, 1996) observation that “quite many of the verbs where zero argument anaphora is possible describe reactions and generate presuppositions as to the Page 17 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora stimulus” (Saeboe 1993: 39) such as in the verbs agree or refuse or in predicates such as to be delighted. Saeboe (1996) tries to distinguish the predicates that take NCA (which he calls definite ellipsis) from the predicates that select implicit complements (which he calls indefinite ellipsis) such as eat by virtue of the availability of a presupposition. He proposes that the predicates that take NCA have presuppositions involving their complements while the predicates that have implicit arguments do not. So, for verbs like agree or refuse, Saeboe says that when we say Sue agrees, that is represented as Sue agrees that p and that will be interpreted by way of the presupposition that someone else thinks that p. For Saeboe, an NCA (or a zero anaphor, as he calls it) is not really an anaphor but is instead an epiphenomenon of presuppositions. However, not all verbs that have presuppositions involving their complements will select NCA. In fact, Saeboe ends up admitting that the availability of NCA for a particular predicate cannot be predicted. Another attempt at trying to characterize the predicates that take NCA is the one in Depiante (2000, 2001), where it is proposed that the predicates that select NCA are the same predicates that do not take the overt predicate/propositional pro-forms it/so. She thus claims that NCA is in complementary distribution with it/so. In order to account for that, Depiante proposes an extension of Chomsky’s (1981) Avoid Pronoun Principle. In later work, Haynie (2010) shows that this characterization is not empirically accurate since there are numerous predicates which allow both NCA and it. (p. 676)

In agreement with other authors, Haynie (2010) also admits that the class of

predicates that select NCA cannot be defined either semantically or syntactically. Semantically, it cannot be defined because, following Fillmore, predicates with similar meanings do not all take NCA, as shown earlier. Syntactically, they cannot be defined either, because verbs with identical syntactic complementation behave in different ways with respect to whether they license NCA or not. Haynie illustrates this with the contrast in (38). (38)

In example (38), Haynie shows that bother selects an NCA but endeavor does not, even though they both select the same type of syntactic complement.

Page 18 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

26.4 How is NCA represented in the syntax? How is NCA interpreted? As we have seen, there are different proposals in the literature as to what the syntax of NCA looks like. Some authors believe that nothing is present in the syntax and that an interpretative rule in the semantics or in the pragmatics provides the right interpretation. Other authors believe that NCA is represented in the syntax by a phonetically empty proform with no internal syntactic structure; in other words, that NCA is the null counterpart of it/so as in (39). (39)

As we have seen, the basis for arguing for the lack of internal structure comes from the unavailability of extraction out of an NCA-site. In the following section, we discuss the validity of this diagnostic.

26.4.1 Is lack of extraction a good diagnostic for lack of internal structure? As mentioned in previous sections, the literature on ellipsis has used the possibility of extraction out of null elements as a diagnostic for internal structure inside that element. In other words, it has assumed the following implication: (40)

The reverse implication, given in (41), has been assumed by some authors but challenged by others. (p. 677)

(41)

One such challenge comes from Aelbrecht (2010) in her work on Dutch Modal Complement Ellipsis (MCE) and British English do. She shows that these types of ellipsis allow some extractions out of them but not all. They allow A-movement (passive and subject raising) but do not allow A-bar-extraction. She claims that “the lack of extraction does not automatically mean that the ellipsis site is a null pro-form without internal structure, although it can be” (Aelbrecht 2010: 9–10). The gist of her account is that Dutch MCE and British English do exhibit internal syntactic structure and that the reason why A-movement is allowed and A-bar-movement is not has to do with the timing of the ellipsis operation. So, these types of anaphors in Page 19 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora Dutch and British English are in fact, according to her, the product of a deletion operation. According to this analysis, the actual deletion operation will happen after Amovement but before A-bar-movement. The fact that Dutch MCE and British English do behave the way they do appears to call into question the implication in (41). However, it doesn’t follow from this that (41) does not work for NCA in English. As we have seen, overt extraction out of English NCA is not possible and neither is covert extraction (recall the inverse scope facts in section 26.2.6.2). So, given these data, NCA in English cannot be equated to a surface anaphor, just because Dutch MCE or British English do behave differently. The point is that they behave differently and they are a different type of anaphor. The locus of contention should not be (41) as a diagnostic but instead, the discussion should be focused on the empirical adequacy of the typology of anaphors proposed or whether a typology is necessary at all. Along these lines, some authors such as van Craenenbroeck and Merchant (2013: 710) have suggested that “it could well be that even null complement anaphora involves abstract syntax, but that the timing of the ellipsis process in this specific construction precludes any movement operation from targeting material inside the ellipsis site.” However, the fact that some null anaphoric elements in some languages allow some extractions but not others seems to indicate that there still is a difference between different types of anaphora. We either need to expand the typology of anaphors (see Thompson 2014) or propose that all null anaphors are the product of deletion. Then, the difference between deep and surface anaphors will be lost as a theoretical primitive and would be reduced to the timing of the deletion operation. Under this view, the grammar must stipulate the exact timing of each deletion. If it turns out that this is how the grammar works, then we would lose the empirically based distinction inherent in the typology of deep and surface anaphora. On the other hand, it might be theoretically appealing to say that all null anaphors are generated by just one deletion operation and that the fact that the deletion operation applies at different points in the derivation is what is perceived as the distinction between deep and surface anaphora. Future research in this area will be able to tell us more about these alternative positions. (p. 678)

26.4.2 On the semantics of NCA

In terms of the semantics, one clear proposal is put forth by Depiante (2000, 2001), who claims an NCA-site contains a free variable that is specified for three semantic types (property, proposition, or question), as summarized in (42) and illustrated in (43)–(45) respectively. (42)

Page 20 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora

(43)

(44)

(45)

However, Depiante (2000, 2001) claims that NCA is never specified for the semantic type e, an individual, as we can see from the ungrammaticality of (46). (46)

Upon examination of a larger set of NCA-selecting predicates, Haynie (2010) discovers that NCA can in fact be interpreted as an individual (type e) in certain circumstances. In this way, Haynie (2010) challenges the proposal put forth in Depiante (2000, 2001). The interesting discovery by Haynie (2010) is that PPs which contain a semantically vacuous preposition and an individual-denoting DP can be replaced by NCA as in example (47). So, in this case, NCA is interpreted as an individual. (47)

Based upon examples such as (47), Haynie argues that Depiante’s “semantic” account of the nature of NCA is not viable, since it doesn’t account for the fact that NCA can have an individual interpretation in certain cases when it replaces a PP with a semantically vacuous preposition.

Page 21 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora However, she recognizes that NCA can never have an individual interpretation given by a DP. So, she proposes what she calls a “syntactic” account of NCA to account for the new (p. 679) set of data. In her account, the grammar does not need to specify the semantic types that NCA will be interpreted as; it only needs to specify the types of syntactic elements that it can be. The semantic types follow from the interpretation of the syntactic types. According to Haynie (2010), NCA can be of the syntactic category CP, VP, or PP, but crucially not DP or AP. She proposes that the semantic interpretation of NCA will be determined by its antecedent and the predicate that selects it. Here is her exact proposal, which we believe to be empirically accurate: (48)

26.5 Summary of the properties of NCA To summarize, here is a list of the properties that characterize NCA in English: 1. It is licensed by a particular group of predicates, determined lexically—no natural class has been found yet. 2. It can be interpreted anaphorically or exophorically: that is, it can have a nonlinguistic antecedent. 3. It behaves like a deep anaphor according to Hankamer and Sag’s (1976) tests: it allows pragmatic antecedents, disallows missing antecedents and doesn’t need strict syntactic parallelism with its antecedent—as exemplified by changes in voice. 4. It does not allow extraction: no overt wh-extraction, no topicalization. 5. It does not allow inverse scope readings. 6. It must be completely null; no subpart of the NCA can be overt. 7. Syntactically, it can replace a CP, VP, or PP but not a DP or an AP. 8. Semantically, it can be interpreted as a property, a proposition, or a question as well as an individual—when it replaces a PP with a vacuous preposition.

26.6 NCA cross-linguistically NCA is common cross-linguistically. The question arises as to whether it receives the same analysis in other languages as its English counterpart or not. Because of space limitations, I will not discuss here all of the cross-linguistic work available in the Page 22 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora literature, but will refer the reader to some of the work that is available. Research on NCA in Spanish (Zubizarreta 1982; Bosque 1984; Brucart 1987, 1999; Brucart and MacDonald 2013; Bošković 1994; Depiante 2000, 2001) and Italian (Cinque 2004, 2006; Cecchetto and Percus 2006) seems to point in the direction that NCA in these languages is also a deep anaphor (according to (p. 680) the results of the tests presented in the literature on English) or some version of it (see Cinque 2004, 2006). However, other authors like Dagnac (2010) seem to suggest that at least the null complements of modals in Spanish and Italian (as well as French) seem to exhibit some of the characteristics of surface anaphors. Work on Portuguese (Cyrino and Matos 2006) seems to point to the idea that in Portuguese, the equivalents of Spanish NCA do not behave like deep anaphors. Research on Romance thus seems to suggest that there might be a split between the way the null complements of modals behave versus the null complements of other types of predicates that seem to select NCA. The former appear to exhibit some of the characteristics of surface anaphors while the latter do not. Further research with larger sets of predicates and more data is needed to elucidate what the exact nature of these complements is and in what ways the type of predicate determines the type of null complement it takes. Cross-linguistic research is fundamental to further our understanding of the nature of NCA and its interactions with other elements of the grammar, as well as to see whether a typology of anaphors is necessary and whether there is a universal typology or not.

26.7 Conclusions and directions for future research In this chapter, we looked at NCA in detail and discussed its nature and behavior. The empirical evidence reviewed points in the direction of NCA in English being a deep anaphor in Hankamer and Sag’s (1976) sense, and thus seems to further support their typology of anaphora. The syntactic and semantic properties of NCA were presented. Syntactically, NCA appears to be a syntactic unit without any internal structure. The diagnostic of lack of extraction out of null elements as evidence for lack of internal structure was used for English NCA. Challenges to the use of this diagnostic were discussed. An alternative view that has been hinted at—but not been completely explored—in the literature by van Craenenbroeck and Merchant (2013) is that NCA might perhaps have abstract syntax and that the lack of extraction is a consequence of the timing of the deletion operation à la Aelbrecht (2010) for Dutch MCE and British English do. Advantages and disadvantages of such a view were discussed. Further research along these lines should be interesting and elucidating. One of the questions that still remains to be answered is how to define the class of predicates that take NCA. It seems to be lexically determined and it exhibits intraPage 23 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora linguistic and cross-linguistic variation. A corpus study combined with work on the lexical semantics of the predicates that license NCA is a direction in which future research might go. In this chapter, we have been able to see that NCA is clearly a phenomenon of the interface. The lexicon, syntax, semantics, and pragmatics all conspire to license it, restrict it, and provide an interpretation for it. We think future research on NCA will benefit from more research focused on the interfaces of the different components of the grammar.

Notes: (1) These null complements differ from the intransitive use of verbs such as eat; see (i). ((i))

The null complement of eat in (i) is not interpreted anaphorically as the null complements in (1) and (2) are. In addition, the null complement of eat in (i) has a very specific interpretation (i.e., John ate something edible, not anything at all). We will not be discussing these types of null complements here. (2) This is the approach taken in Culicover and Jackendoff (2005) for ellipsis. (3) See also Sag and Hankamer (1984) on further differences in interpretation between surface and deep anaphors. (4) There has been a lot of discussion in the literature as to whether it is really true that surface anaphors can never take pragmatic antecedents. Schachter (1977b) challenges Hankamer and Sag (1976) and argues that VP-ellipsis in English can take pragmatic antecedents. However, Hankamer (1978) challenges these counterarguments in a very convincing way. See also Pullum (2000) and Miller and Pullum (2014) who revisit this issue, and also Thompson (2014) for discussion. (5) There is discussion in the literature regarding whether the Missing Antecedent Test is reliable. Hankamer and Sag (1976: 413) recognize that the judgments are “admittedly delicate.” See Thompson’s (2014: 16) discussion of some of the problems that this test has. Bresnan (1971: 595) and Johnson (2001b: 456 n. 30), among others, discuss further limitations of this test. See also Merchant’s (2013a: 541) arguments against using the missing antecedent test as a diagnostic for ellipsis. (6) In other words, sentential expressions composed of a predicate and its arguments, with or without any associated sentential operators. (7) Interestingly, we independently know that Do it cannot have a stative predicate as its antecedent. The same holds for Italian ‘far lo’ (thanks to Guglielmo Cinque for the Italian Page 24 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora judgments) and Spanish ‘hacerlo’, as shown in (i). Given this, we would be able to make an interesting prediction, which is that stative predicates could not be antecedents to NCA. ((i))

It appears that the prediction is borne out in Italian as we see from the examples in (ii): ((ii))

The data for English in (iii) and Spanish in (iv) seems a bit mixed. More data would need to be explored in order to test this interesting prediction accurately. ((iii))

((iv))

Marcela Depiante

Marcela Depiante received her PhD from the University of Connecticut in 2000. She is an Associate Professor in the Department of Languages at the University of Wisconsin, Eau Claire. Her research interests include comparative syntax and Page 25 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Null Complement Anaphora morphology, in particular the syntax of ellipsis, as well as the grammar of Heritage Spanish Speakers and Spanish L1 attriters in the US.

Page 26 of 26

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

Oxford Handbooks Online Conjunction Reduction and Right-Node Raising   Christopher Wilder The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.29

Abstract and Keywords This chapter reviews Conjunction Reduction (CR) phenomena—forward CR (stripping, Phrase Cluster Coordination), and backward CR (Right-Node Raising), and discusses theoretical concerns they raise and approaches to explaining them (ellipsis, across-theboard movement, flexible constituency, multiple dominance/MD). It covers the main empirical generalizations, illustrating mostly with English data but also some from German; and touches on important less well-known phenomena (non-coordinate RNR, Right-Node Wrapping). Striking right–left asymmetries indicate that different mechanisms lie behind forward and backward CR. Similarities between regular ellipsis types and forward CR, and evidence that non-deleted units inside conjuncts targeted by forward CR move to the conjunct periphery, are argued to support the move-and-elide approach. Difficulties afflicting both ellipsis and ATB-movement approaches to RNR are highlighted, and the MD approach is presented as a promising alternative. Keywords: coordination ellipsis, Conjunction Reduction, Right-Node Raising, move-and-elide, ATB movement, multiple dominance

Chris Wilder THIS chapter surveys ellipsis or sharing phenomena peculiar to coordination structures, known as Conjunction Reduction and Right-Node Raising, and reviews some leading issues in their theoretical treatment, including their relation to non-coordinate ellipsis phenomena such as verb phrase ellipsis (VPE). While focusing on analyses from mainstream generative grammar, the chapter seeks to highlight empirical patterns and generalizations. No attempt is made at wide cross-linguistic coverage; the data are mostly English, though some German examples are discussed as well.

Page 1 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.1 Introduction The term Conjunction Reduction (CR), which goes back to generative grammar of the 1960s, is nowadays associated with the idea that conjuncts in a coordinate structure can be reduced by ellipsis, in the sense of omission of repeated material. Examples like (1), where material in one conjunct is repeated in another, have paraphrases in which the relevant words (underlined) are missing from one conjunct, as shown in (2). (1)

(2)

CR phenomena fall into two classes according to directionality. Forward CR can be modelled as ellipsis in the non-initial conjuncts of a coordination licensed by pronounced material inside the initial conjunct. In an analysis of (2a) that takes the conjuncts to be sentential, the subject and verb in the second conjunct are elided, under (some notion of) identity with corresponding pronounced material in the first conjunct (p. 682)

(underlined).1 (3)

In backward CR, commonly known as Right-Node Raising (RNR), material is missing in non-final conjuncts, so the pattern can be modelled as ellipsis under identity with pronounced material in the final conjunct.2 (4)

An ellipsis approach cashes out the omission of words from a larger expression in terms of the non-pronunciation, or deletion, of items that are present inside a conjunct at an underlying syntactic level (indicated by strikeouts in (3)–(4)). However, there is no consensus on whether any, and if so which, of the data in question involve an ellipsis mechanism in this sense. I use sharing (in coordination) to describe the phenomenon without prejudging a particular approach. In (2), the underlined elements are shared, while the remainder Page 2 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (excluding the coordinators) is non-shared material, belonging solely to one conjunct. What is uncontroversial is that the underlined material in (2) is semantically shared by the conjuncts. In (2a), the non-shared NP-PP sequences are semantically arguments of (tokens of) the same shared predicate offer, along with the shared subject they. In (2b), any losses incurred likewise figures in the meaning of both conjuncts, supplying an argument to the non-shared predicates. Contemporary approaches diverge widely on whether and in which cases shared expressions are structurally (syntactically) shared, in the sense of being structurally contained in both conjuncts, as in CR analyses like (3)–(4). This variety reflects different assumptions concerning constituent structure, the size of conjuncts and the treatment of movement phenomena (see 27.2). In (2) and many other cases associated with CR, the shared material (underlined) is peripheral to the coordination, being located at the left (2a) or right (2b) periphery of the non-shared material of the conjuncts. This partly explains why these phenomena are susceptible to an alternative claim concerning the structural location of the shared items —namely, that they are outside the coordination (5). (5)

Such approaches—including across-the-board movement (ATB) analyses (see 27.2.6)—I call direct analyses, in contrast with CR analyses which locate the shared (p. 683)

material inside the coordination in surface structure. Direct approaches are challenged by cases where shared items are not peripheral but sandwiched among non-shared material in one of the conjuncts, as in gapping (6). Viewed as forward CR, gapping involves medial ellipsis of the verb, auxiliaries, and possibly other elements.3 (6)

CR patterns are mostly found in structures with and/or/but, hence the term coordination ellipsis. They normally do not occur in clauses linked by subordinators such as because, while, or unless (7). This sets them apart from types which are not coordination ellipses— sluicing (8a), verb phrase ellipsis (VPE) (8b), and noun phrase ellipsis (NPE) (8c). (7)

Page 3 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

(8)

However, “coordination ellipsis” may be a misnomer—both forward CR and RNR patterns also turn up in configurations not normally viewed as coordinations. Most notably, both figure in comparative constructions (more-than, as-as, etc.); cf. Lechner (2004). In addition, examples similar to (7c) with the RNR pattern may, unlike (7a,b), be judged acceptable to various degrees. Non-coordinate RNR is taken up in 27.4.5. The divide between forward and backward CR is deep (Neijt 1979: ch. 2); the directionality difference correlates with distinct clusters of properties, reviewed in 27.3 and 27.4. While backward CR (RNR) is typically considered a unitary phenomenon, individual forward patterns are often treated under different labels, including gapping, stripping (see 27.2.4), and what I term Phrase Cluster Coordination (PCC), illustrated by (2a) (see 27.2.5). Gapping and stripping are treated in depth in Chapter 23. The present chapter largely ignores medial gapping, but briefly discusses stripping and PCC—firstly, to illustrate the rationale behind CR approaches to coordinations that resist direct analysis; and secondly, to demonstrate the affinity of forward CR to non-coordinate ellipses such as VPE. The chapter is structured as follows. Section 27.2 introduces assumptions and data that figure in the debate about direct vs CR approaches to sharing. Sections 27.3 and 27.4 look at forward and backward sharing in more depth, highlighting their different empirical profiles, and what appear to be promising theoretical accounts. 27.3 focuses on a CR (p. 684) approach to forward patterns that combines conjunct-internal movement and ellipsis. 27.4 reviews arguments for and against direct (ATB movement) and ellipsis analyses of RNR, and outlines the case for an alternative approach using Multiple Dominance. Section 27.5 concludes.

27.2 Coordination syntax and CR This section begins with the early conception of CR and why it was abandoned, and then sketches the subsequent standard view on coordination syntax, which allows direct analysis of phrasal coordinations. Specific classes of data that resist direct analysis within that frame have motivated more recent CR proposals (27.2.3–27.2.5), which compete with

Page 4 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising alternative direct analyses utilizing ATB movement, or flexible syntactic constituency (27.2.6–27.2.7). Other kinds of motivation for CR are pointed out in 27.2.8.

27.2.1 Conjunction Reduction vs phrasal coordination Many an ordinary phrasal coordination is at first blush semantically equivalent to a sentential coordination, as indicated in (9) by the parentheses. Matching the paraphrase relation is a syntactic equivalence which can be stated as (10), adapted from Chomsky (1957: 36). (9)

(10)

This characterization prefigures both the early idea that subsentential coordinations are generally derived from sentential coordinations, and two issues (discussed in 27.2.3 and 27.2.5) behind later CR proposals, namely whether the non-shared strings (X and Y) constitute single constituents of the same type. Early generative work pursued the idea that phrasal coordinations result from a syntactic rule applying to underlying coordinated sentences. (10) can be reformulated as a “Conjunction Reduction” transformation (11), which takes a coordination of sentences as input, and delivers a reduced structure, a single sentence containing coordinated substrings (cf. Gleitman 1965, Ross 1967, Dougherty 1970/1971, Koutsoudas 1971, Hankamer 1979, Hudson 1976b, and others). (11)

The approach was consistent with the framework of Chomsky (1965): the deep structure of a sentence determines its meaning, transformations derive the surface structure. It captures the paraphrase relation between phrasal and corresponding sentential coordinations, and the intuition that the meaning of and and or is constant across phrasal and sentential coordinations. It further suggests that their meanings can be equated with logical conjunction and disjunction of classical propositional logic. (p. 685)

Page 5 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising However, (11) posits massive restructuring of the input. Two sentences are reduced to one, and X and Y, separate in the input, are assumed to form a constituent [X Conj Y] in the output. Such derivations are incompatible with subsequent developments (e.g. the idea that syntactic operations are structure-preserving) and have long been rejected in mainstream generative grammar. The approach was also abandoned on other grounds. Underlying the rule in (11) is the idea that the paraphrase relation between phrasal and sentential coordinations is due to their having the same structure (at the relevant level). Yet by no means do all phrasal coordinations have sentence coordination paraphrases. One class concerns and-coordinations of singular nominals that function as the argument of a collective predicate such as alike in (12) (Lakoff and Peters 1967). Since the predicates in question select for a semantically plural argument, the corresponding sentential coordination is simply ill-formed. (12)

The paraphrase also fails where the phrasal coordination is interpreted in the scope of a higher negation, attitude predicate, or quantificational expression. The examples in (13) express something different than their full sentential equivalents in (14); cf. also (15). (13)

(14)

With the breakdown in the paraphrase relation, there is no motivation from interpretation for assuming sentential conjuncts. These cases lead naturally to the assumption that phrases of all types (NP, AP, VP, PP, etc.) can form coordinated constituents embedded directly within the larger sentence, where relative semantic scope is reflected in structure. (15)

Page 6 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

To the extent that phrasal conjuncts do not denote propositions, the semantics of and and or cannot be limited to the logical conjunction/disjunction of propositional logic. A compositional approach requires a different semantics for coordinators that link phrasal (p. 686) conjuncts, such as the “generalised conjunction” of Partee and Rooth (1983). The denotations of (13a,b) can then be captured straightforwardly on the basis of a phrasal coordination in the scope of a single shared operator. At the same time, the paraphrase relation in cases like (9) falls out as a result of direct semantic interpretation of the phrasal coordination. In addition, a different semantics is needed for the and of (12) (“group-forming and”). See Zamparelli (2011) for an overview. The upshot is that direct phrasal coordination, in which the conjuncts contain exclusively non-shared material, can capture not only (13)–(15), but also (9), with no need of CR to capture semantic sharing effects reflected in sentential paraphrases. Such an approach, now more or less universally adopted, underpins a common view on the core of coordination syntax. Nevertheless, a version of the old CR idea is retained by ellipsis analyses of sharing in coordination for cases in which the core approach breaks down. The structural reduction of (11) is replaced by a conception in which the output retains the structure of the (not necessarily sentential) input, and the reduction itself affects only the pronunciation of words contained in the input. The output in (16) symbolizes the combination of forward and backward CR sketched in (3)–(4). (16)

27.2.2 Coordination syntax: Constituency, substitutability, and coordination of likes This section reviews core assumptions concerning coordination syntax that are widely held.4 Firstly, a coordination comprises a sequence of two or more adjacent expressions (conjuncts), which are interrupted only by the coordinator which links them, and which are independent of one another in the sense that no conjunct is part of another in the

Page 7 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising coordination. Secondly, the coordination itself and its individual conjuncts are constituents. (17)

Descriptively, possibilities seem to be further constrained by a condition like (18): (18)

For example, (18) rules out an analysis like (19a) on the basis of the illformedness of *he a tiny house, while permitting the correct analysis (19b).5 (p. 687)

(19)

Examples like (20), where each of the italicized expressions satisfies (17b) and (18), are nevertheless ill-formed. Such cases motivate an additional condition that the conjuncts of a coordination be syntactically and/or semantically similar (cf. (10)), which has come to be known as the Law of Coordination of Likes (LCL) following Williams (1981). (20)

(21)

What exactly “same type” means, whether for example, traditional syntactic categories, or semantic types, is subject to debate (and of course depends on what types are defined by one’s theory). There is a widespread practice in the syntax literature of assuming that “same type” means “same syntactic category.”6 The classic phrase structure for coordination is a special configuration type (22), according to which for X = any category, a coordination is a constituent of type X that immediately dominates two or more (signalled by the Kleene star) constituents of the same type X (the conjuncts), plus a coordinator.

Page 8 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (22)

(22) encodes both the LCL and, by identifying the type of the conjuncts with the type of coordination, substitutability. It also straightforwardly accounts for the possibility for coordinations to be nested. In frameworks that assume phrase structure uniformly conforms to X’-theory, the multiple headed (22) is anomalous, and endocentric alternatives have been proposed that take the coordinator to head the coordination. In one version (23), & forms a phrase with the second conjunct that is right-adjoined to the first conjunct (Munn 1993). In another (24), the conjuncts are specifier and complement to & (Larson 1990: 556; Kayne 1994: ch. 6; Johannesen 1998).7 (p. 688)

(23)

(24)

It is uncontroversial that phrases of any category can be coordinated, from root clauses down to the smallest individual phrases, whether these be functional categories (CP, TP, DP) or lexical. The broader consensus breaks over whether (in X’-theoretic terms) nonmaximal constituents can form conjuncts. While the coordination of heads, for example, is often assumed under (22), the endocentric approaches do not permit the conjuncts to be subphrasal.8 Cases of apparent coordination of heads, e.g. as in (25a), have been argued to involve phrasal coordination with structural sharing of the direct object (backward ellipsis in Kayne 1994: ch. 6).9 Apparent coordination of sub-word units as in (26a) is problematic for either approach, and has been argued to involve backward ellipsis (Booij 1985; Wilder 1997; Chaves 2008). (25)

(26)

Page 9 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising In sum, the consensus picture has it that coordinations whose non-shared strings form adjacent constituents of the same type (perhaps further restricted to phrasal constituents) do not involve structural sharing. That picture is disturbed by three types of cases (in addition to (26)), each of which has been argued to motivate a CR-type analysis. One involves conjuncts that appear not to correspond to a well-formed single constituent, thus threatening (17b) (see 27.2.5). Another concerns apparently non-adjacent conjuncts (see 27.2.4). A third involves Unlike Category Coordinations (UCCs), i.e. what appear to be phrasal conjuncts of different categories in violation of the LCL.

27.2.3 Unlike Category Coordination Examples like (27a) (Sag et al. 1985), where the non-shared elements are predicates of different categories, illustrate the UCC phenomenon. (27)

The direct analysis (27b) requires a modification of LCL or of assumptions concerning categories (the latter pursued by Sag et al. 1985). (28) illustrates another less well-known UCC type, involving focus or focus particles (Grosu 1985, 1987). Related cases involve wh-questions (What and when does he drink?). (p. 689)

(28)

Most UCCs conform to substitutability.10 Exploiting this property, a CR approach can offer an account that satisfies LCL while allowing leeway on the category of non-shared constituents within the conjuncts. If α can combine with constituents of different categories X and Y such that both [Z α X ] and [Z α Y ] are independently well-formed, and if α can be structurally shared, then Z-coordination with sharing of α can yield the effect of a UCC. This is outlined in (29), where α = the verb (plus other material). (29)

On (29a), see Beavers and Sag (2004), Chaves (2006) and for critical discussion, Levine (2011).11 On (29c), see 27.4.4.

Page 10 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.2.4 Discontinuous coordination and stripping The phenomenon illustrated by (30) poses a related challenge to the core assumptions. (30)

Taking the first conjunct to be a phrase left-adjacent to and (PP, VP, or TP in (31)) conflicts not only with the LCL but also with the intuition that the first conjunct is the DP his phone. (31)

Some have taken this to indicate that the conjuncts of a coordination need not always be adjacent. Munn (1993: 15) suggests an approach in terms of rightward (p. 690)

movement of the second conjunct of a DP-coordination (and his keys forms a phrasal unit &P according to (23)), paralleling extraposition of a postnominal modifier (see also Johannesen 1998: 215ff.). (32)

However, an indication that (30) does not involve DP-coordination is that the pattern fails with collective predicates like combine (cf. Merchant 2003a).12 (33)

Forward CR analyses attribute discontinuous coordinations to the data set derived by stripping, “a rule that deletes everything in a clause under identity with corresponding parts of a preceding clause, except for one constituent (and sometimes a clause-initial adverb or negative)” (Hankamer and Sag 1976: 409).13 Under this approach, the nonshared DPs of (31) are contained in larger, adjacent conjuncts conforming with the LCL, and (33b) is straightforwardly accounted for.14 (34)

Considerations about relative scope of the coordinator and shared negation or quantifiers (cf. 27.2.1) also apply in determining the height of the coordination in stripping. The

Page 11 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising negation naturally scopes over the coordination in (35a), indicating that the coordination is at VP (or vP) level. (35)

Interpreting the negation inside both conjuncts, expected if the coordination is at clause level, is virtually impossible in (35). The possibility for larger conjuncts is evidenced by (36), where the negation is interpreted inside both conjuncts, in the scope of or.15 (36)

Stripping is discussed further in 27.3; see also Chapter 23.

(p. 691)

27.2.5 Non-constituent conjunct strings The string of the first conjunct in (37a) does not correspond to a constituent (Chomsky 1957: 35); the same applies to (37b) and (2b). (37)

The examples in (37) instantiate what has come to be known as Right-Node Raising. Standard assumptions about constituent structure strongly favour a structural sharing approach, such as backward ellipsis (38). While (37a) is a sentential coordination, the plural agreement and collective predicate in (37b) suggest a DP coordination with structural sharing of the N child only. (38)

Two forward patterns are found where the non-shared material in the second conjunct does not form a single constituent (in a traditional structure). One is gapping (see (6)). In the other, (39), Phrase Cluster Coordination (PCC) (cf. also (2a)), the non-shared material is a sequence of postverbal phrases, often analysed (following Sag 1980) as VPcoordination with forward ellipsis of the shared verb. (39)

Page 12 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

The non-constituent status of the non-shared material evident in RNR and PCC (and medial gapping) has provided perhaps the strongest impetus behind recent explorations of CR analyses in the form of ellipsis (see 27.3 and 27.4.3) and/or Multiple Dominance (see 27.4.4).

27.2.6 Conjunction Reduction and movement Movement delivers an alternative way of reconciling apparent non-constituent conjuncts in the RNR, gapping, and PCC patterns with the core assumptions of 27.2.2. This perspective was opened up by Ross’s (1967) discovery of the Coordinate Structure Constraint (CSC) and the across-the-board (ATB) movement phenomenon (see de Vries 2017 for recent discussion). (40)

16

Examples (41) and (42) illustrate the ban on leftward movement of a conjunct and of a constituent out of a conjunct (Ross 1967: 160). (43) exemplifies the possibility for leftward ATB extraction, showing that the constituent must be extracted from all conjuncts.17 (p. 692)

(41)

(42)

(43)

Page 13 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Ross proposed to treat RNR as ATB movement.18 The missing object in non-final conjuncts, giving the impression of a non-constituent string, is (in more recent terms) analysed as a movement trace; and a corresponding trace is posited for the final conjunct. The shared object is located (after movement) outside of the coordination (44). Like leftward ATB movement, RNR applies in ATB fashion in a coordination of more than two conjuncts (45) (Ross 1967: 176–7). (44)

(45)

The relative merits of movement and backwards CR approaches to the RNR phenomenon are discussed in 27.4. The interdependence of assumptions on movement and coordination is illustrated by an early problem with the direct analysis of VP-coordination and its interaction with passivization (i.e. the movement assumed to derive the subject of a passive clause) (Dougherty 1970, 1971). Under the direct analysis (46a), passive movement violates the CSC. A CR-style analysis avoids the CSC problem (46b), but runs into the interpretation problem sketched in 27.2.1. The adoption of the VP-internal subject hypothesis allows a reinterpretation in terms of ATB movement out of small conjuncts (46c) (Burton and Grimshaw 1992; McNally 1992). (46)

ATB movement out of a low coordination has also been proposed to account for PCCs like (39) (Larson 1988: 345); and also for medial gapping (Johnson 2004b, 2009), as sketched (p. 693) in (47). Johnson’s analysis assumes that the CSC does not govern movement of the subject. See Chapter 23. (47)

Page 14 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.2.7 Alternative: Flexible constituency The case for structural sharing approaches to non-constituent conjuncts seen in RNR, PCC, etc. is founded on the assumption that the pronounced non-shared material in the relevant conjunct does not form a constituent on its own; that there is, for instance, no well-formed syntactic constituent containing Mary praised in (48) that does not also contain a direct object as sister to the verb, unpronounced in that position, whether as a movement trace or target of ellipsis. (48)

An alternative hypothesis is that such conjoined strings do in fact correspond to single constituents, albeit under a non-standard view of what counts as a constituent. Such an approach is pursued in categorial frameworks, for instance Combinatory Categorial Grammar (CCG) (see Steedman 2000, 2007; Steedman and Baldridge 2011).19 CCG syntax is based on categories defined as functions or arguments, with the possibility for the latter to be shifted to higher functional types. Complex units are formed by combinatory operations such as function application or function composition, directionally specified; cf. the sample derivations in (49)–(51) (derived categories in bold).20 While basic categories combine in ways consistent with standard constituency as in (49), typeshifting makes alternative derivations available that generate non-standard constituents as in (50) (based on Steedman 2000: 39ff.). (49)

(50)

Non-standard constituents such as [Mary praised] in (50) are then available to be coordinated and subsequently combined with an object NP (51), which forms the basis of semantic composition, to give a direct account of the rightward ATB movement effect. (p. 694)

(51)

Page 15 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

A similar approach is taken to PCC. A double object VP can be derived as [[ V NP ] NP ] by forward function application, or the two NPs can be typeshifted so as to be able to combine by function composition to give a constituent that combines with the ditransitive verb as [ V [ NP NP ]]. The latter permits a direct analysis of a PCC (Steedman 2000: 46). For an account of medial gapping, see Steedman 2000: 183ff. In short, the syntax of all coordination is given a direct analysis, involving neither movement traces nor ellipsis. A basic feature of CCG and related approaches is that “everything that can coordinate…is a constituent under the generalized definition of that notion that is afforded by categorial grammars” (Steedman 2000: 198). The same flexible notion of constituency is claimed to capture prosodic (intonation) units (Steedman 2000: ch. 5). On the other hand, the CCG notion of constituency is not constrained by traditional notions of dominance and ccommand which underlie mainstream approaches. Flexible constituency approaches are not considered further here.21

27.2.8 Other kinds of motivation for CR Motivations for CR are not limited to considerations pertaining to adjacency and constituenthood of conjuncts and the LCL, but extend for instance to assumptions about how meaning relates to structure, and about how agreement works. Two cases which appear at first blush consistent with a direct analysis illustrate the point.22 McCawley’s (1987) example (53) appears, like (52), to fit a pattern [DP < X Conj Y > α ] in which X and Y corresponds to a coordination of APs modifying a single head noun. Yet in (p. 695) the case of (53), interpretation (two kinds of knowledge) and number agreement (a plural verb agrees with an and-coordination of non-count singular DPs but not with a single such DP) indicate larger conjuncts, two DPs, each with its own head noun, reduced by backward CR. (52)

(53)

Page 16 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Sentences with corrective but provide another example. In (54) with and, negation is naturally interpreted as taking scope over the coordination, as reflected in the structure under the direct analysis (55a). With but, however, the negation’s scope in (54) is restricted to the first conjunct, suggesting that this case must involve larger conjuncts— which in turn suggests forward CR (55b) (Vicente 2010b; Toosarvandani 2013a). (54)

(55)

27.3 Forward CR If CR is a species of ellipsis, one expects it to share basic properties with canonical (noncoordinate) ellipsis types (verb phrase ellipsis, sluicing, noun phrase ellipsis). This section begins by considering the relation between forward CR (stripping, PCC) and canonical ellipsis, and then discusses the ‘move-and-elide’ approach.

27.3.1 Forward CR and non-coordinate ellipsis The canonical non-coordinate ellipsis types involve omission of a single constituent. In the context of a local head of a specific type, shown as X in the configuration (56), YP can be elided (Lobeck 1995; Merchant 2001, 2004a). (56)

Approximately, X stands for an interrogative wh-complementizer (sluicing), certain auxiliaries (VPE), or certain determiners, including numerals and possessors, but not the or every (NPE). The ellipsis site is a “surface anaphor” (Hankamer and Sag 1976), one formed during and interacting with the syntactic derivation (see van Craenenbroeck and Merchant 2013 for an (p. 696) overview). According to the phonological deletion approach championed by Merchant and others, the ellipsis site contains a fully fledged syntactic structure, whose subparts show syntactic interactions with elements external to the ellipsis, including movement out of the ellipsis site (57). The ellipsis itself arises through suppression of the pronunciation of the constituent in question.23 (57) Page 17 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

The forward CR approach to stripping and PCC sketched in 27.2 invokes conjuncts of variable size containing a focused phrase (the pronounced remnant), with the remainder, i.e. shared material, elided by a specific type of ellipsis rule, applying in ATB fashion in non-initial conjuncts (Forward Deletion / FWD in Wilder 1997). FWD involves ellipsis of phrases or individual words in the second conjunct, under identity with parallel material in the initial conjunct, leaving solely the focused element(s) as remnants (cf. also Hartmann 2000). The idea that forward CR is of a kind with ellipsis types such as VPE24 finds support in patterns of non-identity of forms known as “vehicle change” effects (Fiengo and May 1994). As documented in the literature on VPE and sluicing, an ellipsis target may contain forms that differ from corresponding forms in the antecedent, as long as the semantic identity condition (Merchant’s 2001 e-GIVENness) is satisfied. The VP elided in (58a) (italicized) contains a pronoun (as required by Binding Theory) where the antecedent VP contains a name. The antecedent and elided VPs of (58b) contain bound (sloppy) pronouns which take different forms, as required by conditions on pronominal agreement. (58)

The same effects can be observed with stripping and PCC (59)—cf. inform himj about {hisj /*Johnj’s} results, and put the dog out of {its/*their} misery. (59)

Verbal agreement may also diverge in stripping (60a), an effect replicated in VPE (60b) (Fiengo and May 1994: 103). (60)

Another vehicle change effect concerns indefinites (any-some, no-a, etc.) in polarity-switching VPE (61) (Johnson 2001: 468–9; Merchant 2013b). This effect is not detectable in English stripping (perhaps because the polarity switch that would trigger it depends on the presence of an emphatic finite auxiliary in that language), but it can be observed in German. In (62), the polarity switch is marked by the contrast between (p. 697)

Page 18 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising negation (in niemanden ‘nobody’) and the particle schon (which signals the emphatic affirmative). (61)

(62)

Furthermore, like canonical ellipsis, stripping and PCC show interaction with movement; both ellipsis sites (63) and remnants (64) may contain the trace of an item extracted (ATB) out of the coordination. (63)

(64)

In sum, forward CR shares much of the empirical profile of canonical ellipses.

27.3.2 Move-and-elide In the FWD conception, elliptical conjuncts have the same syntax as corresponding nonelliptical expressions, involving no special operations except for the elision of given material. Unlike canonical ellipsis types, what is targeted by FWD evidently need not be a single constituent (e.g. (63a) involves non-adjacent gaps). However, not just any given Page 19 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising material may be elided (cf. *He bought the book and READ the book); conversely, not just any part of the second conjunct can surface as a remnant in an elliptical non-initial (p. 698) conjunct, and difficulties arise in accounting for the attested patterns under the FWD approach.25 An alternative move-and-elide approach, sketched in (65), is proposed for stripping in Merchant (2003a), building on Sag (1980), Depiante (2000), and others. (65)

The key idea is that ZP (the focused remnant) is moved to the periphery of the conjunct (XP). What is elided is a single constituent (YP in (65)), containing the trace of ZP.26 (66)

Stripping is thereby aligned with a number of other constructions analysed in terms of move-and-elide, including sluicing (Merchant 2001), pseudogapping (Jayaseelan 1990; Lasnik 1999a), and fragment answers (Merchant 2004a). The movement of ZP correlates with A'-movement to the clausal periphery. As discussed in 27.2.4, the conjuncts in which stripping applies are in many cases smaller than a full clause (a point emphasized by Toosarvandani 2013a). The account thus needs to be formulated in terms of A'-movement of the focused remnant to the conjunct periphery, whatever its category (CP, TP, vP, etc.), even if that movement pattern does not surface in non-coordinate environments (67b). (67)

The move-and-elide approach has the twin advantages of predicting what can be a remnant (namely, any constituent within the conjunct which can undergo A'-movement to its periphery), and what can be elided (namely, the constituent YP). A further conceptual attraction is that it brings ellipsis in stripping into line with non-coordinate ellipses; deletion targets a single constituent in each case. The prediction that the class of stripping remnants should correspond to all and only (focusable) units that are licit targets for A'-movement is borne out widely. For example, a remnant may not correspond to a phrase contained in an island (68b); but the island itself (p. 699) may front, giving (68c).27 Nor may the remnant be a unit that resists A'Page 20 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising movement, such as one headed by a finite auxiliary (69b), (70b), as opposed to a full CP (compare (69a) and (70a), which allow a direct, non-elliptical coordination parse). Nor may it be the object of a preposition in a language, such as German (71b), that does not have preposition stranding. (68)

(69)

(70)

(71)

Cases like *He has bought the book and READ the book also fall into place. The verb cannot raise alone (remnants arise by phrasal A'-movement), and the VP minus its object cannot be an A'-moved remnant either, in English at least (cf. *Read he has the book). The hypothesis that stripping remnants have moved is strongly supported by such patterns.28 Move-and-elide has further been argued to apply in PCCs (Sailor and Thoms 2014). Similar evidence motivates movement of remnants in medial gapping (Neijt 1979; see also Chapter 23). It seems worth exploring whether the UCC types (27) and (28) (see 27.2.3) can be captured as a kind of stripping under move-and-elide. That approach may also shed new light on UCCs like (72) (see n. 10). (72)

Page 21 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising For whatever reason, English does not tolerate a that-clause in the complement of a preposition (*You can depend on that my assistant will be on time)—but the clause may surface if it (p. 700) has moved (That my assistant will be on time, you can depend on). A move-and-elide derivation would generate (72) with the that-clause as a stripping remnant (73): (73)

The movement analysis of stripping and PCC remnants raises issues requiring further study (see also n. 24). For example, ATB extraction is possible from the ellipsis site (cf. (63)); (74) shows the move-and-elide structure. Focus movement of the remnant to the conjunct periphery might be expected to create an intervention configuration, i.e. turn the conjunct into an island for extraction out of the coordination. (74)

Also, remnant movement creates a classic “freezing” configuration (see Corver 2014) in which the remnant itself becomes an island. Yet extraction (ATB) from a remnant is apparently possible; cf. (64). Whether the move-and-elide approach can extend to other forward CR cases is an open question. Stripping is not confined to clausal or verbal coordinations; (75) illustrates its occurrence in coordinations of nominals, a phenomenon on which the literature is largely silent (but see Neijt 1979: 29). (75)

Another case involves the sharing of the initial part of a compound word in the pattern (76) (Booij 1985; Chaves 2008). If coordination below the word level is rejected, a direct analysis is not possible; yet at the same time, given that parts of a word are not moveable elements, movement and ellipsis within larger conjuncts is also ruled out.29 (76)

Page 22 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.4 RNR According to Ross’s (1967) proposal, RNR is similar to leftward ATB movement. Although the shared element α is introduced inside the conjuncts, it surfaces outside the coordination. A CR approach (77a) thus competes with an ATB movement analysis (77b), according to which α has undergone extraction from each of the conjuncts. (77)

RNR displays properties that differ sharply from those of forward CR cases analysed in terms of ellipsis, and also from leftward movement, including leftward ATB movement. These special RNR properties pose puzzles for both ATB movement and ellipsis approaches (see 27.4.2 and 27.4.3). 27.4.4 discusses a third approach, according to which shared material in the RNR pattern arises through Multiple Dominance (MD). Two related and understudied phenomena, non-coordinate RNR and the mysterious RightNode Wrap configuration, are addressed in 27.4.5 and 27.4.6. (p. 701)

27.4.1 RNR as rightward ATB movement The rightward movement approach is what Abels (2004) terms an ex situ analysis; it makes the assumption in (78). (78)

The movement analysis is natural in a model which assumes rightward and leftward movement operations, the CSC, and ATB movement. It directly captures the ATB nature of RNR (cf. (44)–(45)); if α moves out of the coordination, it is subject to the CSC. Another potential advantage is the prediction that RNRed quantifier expressions can scope over the coordination. Sabbagh (2007) cites contrasts such as that in (79), where (79a), unlike (79b), allows the distributive (‘different ways’) reading (80a) expected if every patient takes scope over or. (79)

(80)

Page 23 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

Under the movement approach, the wide scope of the RN follows as a direct consequence of its derived position, from which it takes scope over the coordination. A narrow-scope interpretation of an RN may be explained via the possibility for reconstruction, similar to what is found with leftward ATB movement (Höhle 1991). See also 27.4.4.1.

27.4.2 Special RNR properties Various traits of RNR are unexpected from the viewpoint of the ATB movement approach. These relate to (i) general linear properties of the RNR configuration, (ii) possible targets of RNR, and (iii) RNR–ellipsis interaction.

27.4.2.1 Linear properties: String-vacuous nature, Right-Edge Restriction (p. 702)

A hallmark of movement, rightward and leftward, is that it typically results in a change in the order of constituents. For leftward ATB movement, that includes reordering the shared item α with respect to material X outside of the coordination (81). RNR never reorders α with respect to elements external to the coordination; cf. (82). If RNR is rightward ATB movement, it is string-vacuous—unlike its leftward counterpart, and unlike non-ATB rightward displacements (Extraposition, Heavy NP Shift). (81)

(82)

(83)

30

The ex situ assumption (78) may be compatible with (83), but the absence of cases like (82b) means that the order facts are also compatible with the in situ assumption that the RN is inside (has not moved out of) the final conjunct (Abels 2004).

Page 24 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Unlike leftward ATB movement (and forward CR), RNR is subject to an edge restriction on the placement of RN gaps (Wilder 1999).31 (84)

The RER is illustrated by the contrast in (85) (cf. Oehrle 1991). The object of invite in (85a) can appear at the right edge of its conjunct following the PP, i.e. it can undergo Heavy NP Shift (cf. I invited into my house all the winners). The first object in a double object construction resists Heavy NP Shift (*I gave a present all the winners); hence the gap in the first conjunct of (85b) cannot satisfy (84). (85)

The RN itself also shows a right-edge effect. Under the movement analysis, the contrast in (86) suggests that the rightmost position in the final conjunct must be a licit position for α if it is to undergo RNR.32 (86)

(p. 703)

Properties (83) and (84) govern the RNR pattern across languages. Comparison of

VO and OV languages is instructive; (83) and (84) interact with the word order of the language to produce different choices of RN, as illustrated for German in (87). Verb-final contexts permit RNR of V or O+V, but RNR of the object alone is possible only if it is in final position.33 (87)

Page 25 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising An account within the ATB movement approach is developed by Sabbagh (2007). His theory aims to derive both the RER and string vacuity of RNR from the way movement interacts with linearization (see his work for details).

27.4.2.2 RNR targets Inspection of possible RNs reveals a range of choices that are unexpected if RNR is movement. Units which can be RNs but undergo neither leftward movement (ATB or otherwise) nor simple rightward shift include: TP, stranding complementizers (88a) (Bresnan 1974); T’/VP (88b,c), with do-support not triggered, unlike in leftward movement; nominals, stranding predeterminers like every (88d) or adjectives (cf. (37b) and (53)); and parts of words (88e) (cf. also (26)), (Booij 1985; Wilder 1997; Chaves 2008). (88)

RNR can also target the complement of P, stranding P—a pattern unexpected in English, since rightward DP-shift cannot strand P otherwise; and more so in languages like Irish (McCloskey 1986) or German (89b) which otherwise generally disallow P-stranding. (89)

The fact that RNR can target an element contained in an island inside one or both conjuncts (Wexler and Culicover 1980) also calls into question the idea that RNR involves movement out of the coordination, even if that element is mobile otherwise. (p. 704)

(90)

(91)

Page 26 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising In (92)–(93), the RNRed string does not correspond to a single constituent. Such cases have been taken to indicate that RNR can target a sequence of constituents at the right edge of the conjuncts. While (92a) involves multiple moveable elements, (92b) and (93) certainly do not. (92)

(93)

In (93), RNR targets the verbs of the main clause and a relative clause, which form a clause-final sequence due to the OV nature of the language (Ha 2008: 13 and Kubota 2015: 4 give similar examples from Korean and Japanese respectively). Such data contribute to the case for considering an in situ approach.34

27.4.2.3 Interaction of RNR with leftward movement In (93) (following standard analyses of German V2), the shared verbs are followed by the trace of a non-shared auxiliary hat ‘has’ inside each conjunct, cf. (94). (94)

This illustrates the fact that only overt material counts for the RER. Distinct (nonshared) null elements such as the traces in (94) or (95) do not cause RER violations. (p. 705)

(95)

To the extent that silent elements can be said to figure in a linear string, a better formulation of (84) would be that ‘no RNR gap may be followed by a pronounced nonshared element in its conjunct’. The fact that only overt elements count is a further indicator that linearization principles are what underlie the restriction.

Page 27 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Non-shared material can be extracted from a position inside the shared string, such as the two wh-phrases in (96), or the auxiliary belonging to the first conjunct (97). (96)

(97)

Such patterns are tricky to accommodate if RNR is treated as ATB movement of a single constituent (Grosz 2015).35 Under an ellipsis analysis, (96)–(97) are relevant for the question of what counts as ‘identical with the antecedent’, since there is a mismatch between the RN and the gap with respect to null elements contained within them. (98)

27.4.2.4 Interaction of RNR with ellipsis Another asymmetry between leftward ATB movement and RNR is pointed out by Abels (2004). In contrast with leftward movement, an RNRed element cannot survive independent ellipsis of the VP containing its original position. In the configuration (99a), the VP of the second conjunct can be elided (100b), while the example corresponding to (99b) is impossible (constructing test examples with P-stranding eliminates a potential pseudogapping source). (99)

(100)

(101)

Page 28 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising If the RN in (101a) is outside the coordination, ellipsis ought to be possible just as it is in (100). That this is not so suggests that in fact the RN remains inside its conjunct.36 (p. 706)

27.4.3 RNR as ellipsis The facts reviewed above are taken by many to indicate that (78) is mistaken, and that an in situ analysis of RNR is more likely to be correct. One possible in situ approach is an ellipsis analysis (77a), whereby each conjunct contains a (syntactically independent) token of α, with all but the final token undergoing phonological deletion, under identity with the pronounced token. As such, RNR shows a directionality opposite to that of both forward CR and of canonical ellipsis. There are other significant differences between RNR and canonical ellipsis phenomena (NPE, VPE, sluicing). One concerns the grammatical type of the gap, which in each of NPE, VPE, and sluicing is a specific category (NP, VP, TP). RNR is promiscuous—there is quite possibly no category type that cannot undergo RNR, once independent factors such as prosody are controlled for.37 Another concerns identity. Forward CR shows form identity mismatches largely in line with those found in canonical ellipsis; RNR does not (see 27.4.3.1). In addition, certain types of RN are at odds with the assumption of two independent syntactic tokens (see 27.4.3.2 and 27.4.3.3). Focus is seen by some to play a crucial role in RNR. Certainly, RNR involves semantic contrast between the non-shared parts of the conjuncts, which typically leads to material preceding the gap and the RN being focused. Hartmann (2000) proposes that focus conditions the phonological deletion of the RNRed material in non-final conjuncts. Ha (2008) suggests that the RNR is syntactically conditioned ellipsis, and that the context for RNR is identified by a syntactically encoded contrastive focus which immediately precedes the gap in each non-final conjunct. Chaves (2014) presents arguments against taking (contrastive) focus to play a defining grammatical role in RNR. (102) illustrates that RNR gaps need not be directly preceded by contrastively focused elements. (102)

27.4.3.1 Form identity RNR generally does not tolerate form mismatches (vehicle change) of the type observed in forward CR and VPE etc., discussed in 27.3.1. It appears that the form of the RN must satisfy form requirements imposed by both conjuncts (Wilder 1997) (exceptions are discussed below). (p. 707)

Page 29 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Firstly, an R-expression contained in the RN cannot corefer with a pronoun ccommanding the RNR gap in the first conjunct (Levine 1985): (103)

Secondly, while an RN can contain bound (sloppy) pronouns (Höhle 1991; Jacobson 1999), mismatch of forms leads to ill-formedness (Chaves 2014), unlike in VPE and forward CR. (104)

Thirdly, mismatching verb agreement forms cause ill-formedness, though morphosyntactic feature mismatches are tolerated in the case of syncretism, i.e. if the form is compatible with requirements in both conjuncts (Eisenberg 1973; Wilder 1997). (105)

Similar examples involving other verb forms in English, and case forms in German and other languages are discussed in Pullum and Zwicky (1986) and Chaves (2014). These facts indicate that, if RNR is a kind of ellipsis, then it is subject to a different kind of identity requirement than forward ellipsis. Wilder (1997) suggests that RNR involves deletion at a level (in terms of the Minimalist model, following Spell-Out and operations of inflectional morphology) at which grammatical categories are no longer relevant. Chaves proposes an operation of ‘backward peripheral deletion’ which ‘imposes morph form identity conditions’ (Chaves 2014: 868). A puzzling counterexample to the identity generalization is seen in (106) (Kayne 1994: 146). (106)

The pattern resembles the some-any vehicle change found in VPE, yet the shared constituent is a DP, not a regular ellipsis target in English. Chaves (2014: 875–6) suggests that the RN in (106a) is not the whole DP but the bare plural books about linguistics

Page 30 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising following any. However, natural-sounding instances of the pattern can be found which lack a suitable RN following any (examples from the Web). (107)

27.4.3.2 A total of, same, different Ellipsis accounts break down in the face of examples like (108a), involving relational adjectives like same, different (Jackendoff 1977), and what Chaves (2014) calls the additive reading of (108b). (p. 708)

(108)

(109)

In the reading in which John’s tune was different from Mary’s tune (Carlson’s 1987 internal reading of different), (108a) is not equivalent to its undeleted counterpart (109a). Likewise, (109b) lacks the prominent reading of (108b) according to which the sum of the money borrowed and stolen was $3000. An analogous difference was described in 27.4.1 with respect to quantifier scope. The source of the problem these examples pose is that the RN takes scope over the remainder of the coordination. This in and of itself does not force the conclusion that the RN is syntactically outside the coordination (Abels 2004). But ellipsis approaches suppose that each conjunct contains a syntactically independent token of the RNRed phrase; an RNR example is syntactically the same as its counterpart without deletion. It is unclear what syntactic or semantic mechanism could be responsible for the wide scope of the RN, and why such a mechanism should not be available in the absence of the deletion. These cases, in short, are highly problematic for in situ deletion theories such as Wilder (1997) or Hartmann (2000b).

27.4.3.3 Summative agreement The phenomenon of summative agreement (Yatabe 2003), illustrated in (110), was observed in English by Postal (1998: 173) and in German by Schwabe and Heusinger

Page 31 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (2001). The RNRed string includes a plural verb which agrees with non-shared singular subjects that are not themselves coordinated but each in a separate larger conjunct. (110)

The pattern appears to exist in some but not all languages/varieties and for some but not all speakers (Grosz 2015). It clearly presents a problem for a deletion account, in which each subject would be in a syntactic agreement relation with a separate token of the plural verb. Postal (1998), Yatabe (2003), and Chaves (2014) conclude that the pattern is to be handled (p. 709) via an ex situ analysis of the RNRed string in which some special mechanism provides a plurality for the verb agreement. Other phenomena involving a plural form in the RN related to singular non-shared elements, illustrated by (111), cause similar problems for in situ deletion accounts: (111)

Relative clauses with split antecedents (Perlmutter and Ross 1970), insofar as it is correct to suppose that they instantiate RNR, pose a similar problem. (112)

27.4.3.4 Backwards ellipsis vs RNR Canonical ellipses (sluicing, VPE, NPE) usually follow their antecedents, but backward ellipsis is also possible. That raises the question of whether some apparent RNR gaps (specifically, RNR of TP, VP, or NP) might have an ellipsis source in addition to or instead of an RNR source. The configuration of interest is (113), where YP = TP, VP, or NP, the gap is final in its conjunct, and the content of the gap is identical with overt final YP in the second conjunct. (113)

Barros and Vicente (2011) propose that some cases of (113) are derived by backward ellipsis, discussing YP=VP in English. Chaves (2014) makes a similar claim, discussing examples where YP= NP and YP=TP as well. Sluicing, NPE, and VPE are all licensed by specific choices of X in the left-hand context (114). Page 32 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (114)

Certain choices of X only occur where YP undergoes ellipsis, i.e. do not permit a nonelliptical YP. These can be deployed to probe for genuine backward ellipsis, and to distinguish ellipsis from RNR gaps. Given the form identity condition on RNR (see 27.4.3.1), an RNR gap should behave like overt YP in this respect. In English, the inversion of a preposition and wh-word (swiping) is an instance of X that only occurs in sluicing; swiping does not occur when the TP is not elided (Merchant 2002).38 (115)

(p. 710)

Gračanin-Yüksek (2007) analyses wh&wh constructions (What and when did he

eat?, cf. (29c)) as a form of RNR. Chaves (2014) suggests similar examples are derived by ellipsis (backwards sluicing) rather than RNR. However, as Gračanin-Yüksek notes (pp. 156–7), swiping is degraded in this construction, implying that it does not have a sluicing source. (116)

The same contrast is seen in cases involving larger conjuncts; the deviance of (117b) suggests that (117a) does not have a sluicing source: (117)

English NPE is licensed in the complement of prenominal possessors (118), but in the case of possessor pronouns, a special ellipsis form (mine, yours, theirs) replaces the regular my, your, their. (118)

This alternation directly distinguishes RNR gaps from NPE. (119a) (from the Web) illustrates a possessed NP in the RNR configuration where the gap follows the regular pronoun. The ellipsis form is degraded (119b). Page 33 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (119)

German allows NPE in the complement of quantifiers like ein ‘one/a’, kein ‘no’, and pronominal possessors (mein ‘my’, sein ‘his’, etc.). These also have a special ellipsis form: the regular form ein (or mein, kein, etc.), as in ein Wagen ‘one car’ (MASC.SG.NOM) or ein Buch ‘one book’ (NEUT.SG.NOM/ACC), is replaced by einer (MASC.SG.NOM) or eins (NEUT.SG.NOM/ACC) if the NP is elided (120) (Lobeck 1995: ch. 4). A contrast similar to (119) shows up in German RNR (121); again, the ellipsis form is degraded. (120)

(121)

The licensing contexts for English VPE and RNR of VP differ in four cases. Firstly, VPE cannot target the complement of progressive being (passive or copular); cf. (122) (p. 711)

(Akmajian and Wasow 1975; Sag 1980; Harwood 2014). RNR of a passive VP or other predicate, stranding being, is perfectly possible (123): (122)

(123)

Secondly, RNR can take out non-finite perfect have. VPE normally cannot (Sag 1980; Harwood 2014). (124)

Thirdly, RNR (unlike VPE) can take out a finite lexical verb without triggering do support:

Page 34 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (125)

Fourthly, RNR (unlike VPE) cannot trigger do-support except in emphatic do contexts. Do-support in affirmative declaratives is restricted to cases of emphatic do, with do focused (John DID leave, but *John did LEAVE / *JOHN did leave)—except in VPE contexts, where do-support is mandatory even if do is not focused (Who left? – JOHN did). Where VP is RNRed, do-support only occurs when focused (emphatic do, (126c)).39 (126)

The first three contrasts indicate that RNR can elide English VPs which VPE cannot, whereas the contrast in (126) suggests that VPE is unable to create an RNR-like configuration where the VP gap in the first conjunct depends on the VP in the second conjunct. This matches what (116) and (117) suggest for sluicing, and (119) and (121) for NPE.40 (p. 712)

On the other hand, Bošković (2004b), Barros and Vicente (2011), and others have

documented that apparent RNR of English VPs allows identity mismatches that are unexpected from the viewpoint of the form identity condition on RNR discussed in 27.4.3.1, and that mirror those permitted in VPE. This is illustrated by the elision of the bare form write licensed by the participle written in both examples in (127). (127)

Barros and Vicente argue that this pattern and the way it interacts with phenomena discussed in 27.4.3.2 and 27.4.3.3 show that RNR is not a unitary phenomenon (they propose a hybrid ellipsis/MD approach).41 In sum, although there is evidence to suggest that the configuration (113) generally only arises via RNR (and not regular ellipsis), there are still unresolved issues in this domain, notably relating to facts like (127a).

27.4.4 RNR as Multiple Dominance

Page 35 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising 27.4.4.1 Sharing as MD Multiple Dominance has been explored as an alternative in situ approach to RNR (McCawley 1982; Wilder 1999; Gračanin-Yüksek 2007; Kluck and de Vries 2013, and others). It differs from ellipsis/deletion accounts in that there is assumed to be only one token of the shared constituent in the structure. Thus instead of two copies of the shared element α with one deleted, there is a single element α, dominated by two different mother nodes, one in each conjunct. (128)

This approach is claimed to enable an account of in situ effects while avoiding some (at least) of the drawbacks of ellipsis approaches. If it is on the right track, then RNR is a phenomenon very different in nature from regular ellipsis, one that does not involve a relation between an anaphor (or rather, cataphor) and an independent antecedent. Rather, RNR is more akin to movement (even if it is not movement), involving a single item (p. 713) occupying multiple positions in a structure. There have also been proposals to treat movement in terms of MD (see de Vries 2009 for a review), an idea extended to ATB movement by Citko (2005) (see also Wilder 2008). The placement of α in the string needs to be accounted for by independent linearization assumptions, such that a gap appears in the string in α’s position in non-initial conjuncts. Wilder (1999, 2008), Bachrach and Katzir (2009), and others explore ways of deriving linear properties of RNR, including the RER, from a general linearization algorithm applying to structures with or without MD. More generally, extending ideas of Kayne (1994), the idea is pursued that the extra structural possibilities admitted by MD are severely constrained by a linearizability requirement. By invoking a basic option of structural combination (‘parallel merge’, to use Citko’s term), the MD approach fits well with the facts reviewed above showing that just about any type of constituent can be shared in the RNR pattern (modulo linear constraints). Even subword deletion can be captured in terms of MD, assuming complex words have internal structure (Wilder 2008). By not invoking movement, the MD approach avoids questions facing the ATB movement approach about RNs that do not otherwise move.

Page 36 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising The MD approach forms a reasonable basis for understanding the form identity effects reviewed in 27.4.3.1 (cf. also Citko 2005 on similar effects in leftward ATB movement). Mismatches such as those concerning any ((106)–(107)) remain unexplained, though. The MD structure also offers a basis for accounting for the interpretation of shared quantificational elements within the scope of the coordinator (the ‘different ducks’ reading of (129a)). (129)

While there is only one token of the indefinite a duck in the structure, it occupies two separate positions in the scope of the coordinator (one in each conjunct). Depending on one’s theory of indefinites, a duck may undergo local Quantifier Raising within each conjunct or be interpreted in situ as a variable; regardless, it will be interpreted in both conjuncts. At the same time, the MD approach does not suffer the defect of the ellipsis approach outlined above with respect to (130a), the wide-scope (‘same song’) reading of the shared object in (130b), or the quantifier in (79). (130)

Assuming MD for sharing in RNR does not in principle exclude that an RNRed element may undergo rightward ATB movement. Equally, it is not incompatible with the possibility that an RN may undergo covert ATB movement (e.g. ATB Quantifier Raising).42 Grosz (2015) presents an argument concerning the summative agreement pattern (27.4.3.3) favouring MD. While agreement in (131) cannot be handled as a regular syntactic relation under a deletion approach or an ex situ treatment, this is possible within an MD account. This is because in the structure (132), the shared auxiliary, by virtue of being dominated by both TP1 and TP2, enters an independent structural relation (represented in (132) by coindexing) with each of the non-shared singular subjects. (p. 714)

(131)

(132)

Page 37 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

The subjects in (132) do not form a single unit (a DP coordination). It is thus correctly predicted that group readings of the DPs are impossible (133) (Grosz reports that such examples are uniformly rejected by speakers who accept summative agreement in the pattern (131)). (133)

27.4.4.2 Non-bulk sharing The MD approach permits sharing of a sequence of more than one constituent. The facts noted in 27.4.2.2 and 27.4.2.3 concerning multiple constituent sequences and the interaction of RNR with leftward movement are both captured in terms of multiple constituent sharing. In (128), what is shared is a single complex unit, viz. [DP the book]. There is one node in each conjunct which is a mother of the shared unit. In (134) (cf. (92b)), two units are shared, A and N, each of which has a mother in each conjunct—AP1 and AP2, NP1 and NP2 (Wilder 2008). (134)

Gračanin-Yüksek (2007) distinguishes bulk sharing from non-bulk sharing. In (128), a single shared constituent, formed from smaller shared units, is shared as a ‘bulk’ unit. In (134), two smaller units (A and N) are shared separately (‘non-bulk’). (p. 715)

Page 38 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising RNR cases like (135) have non-shared constituents (which book and which film) extracted from a position within the shared string, posing tricky questions for an ATB movement account (cf. 27.4.2.3). (135)

In an MD approach, RNRed strings containing distinct null elements involve non-bulk sharing—one trace belongs to one conjunct and the other trace belongs to the other. The shared items of the RNRed string are shared individually, a sequence of MDed constituents. Any constituent XP1 that dominates a non-shared trace within a conjunct is itself non-shared (i.e. distinct from its counterpart XP2 in the other conjunct by virtue of dominating an element not dominated by XP2), even where the constituent in question (e.g. the TPs and VPs in (136)) dominates no pronounced non-shared elements. (136)

The same account is applied to wh&wh questions like (137) by Gračanin-Yüksek (2007) (see also Citko and Gračanin-Yüksek 2013). In (137b), the positions of the individually shared elements (did, he, eat) in the first conjunct are indicated with underscores. (137)

The non-bulk sharing approach resolves the issue of RNRed strings containing distinct (non-shared) null elements by denying that the RNRed string corresponds to a single constituent.43 Any such RNRed string will be analysed as non-bulk sharing, including examples involving A-movement (138a). For the same reason, though, this approach is unable to derive (138b) as ATB VP-fronting; the structure will not contain a single shared VP to input ATB fronting. (138)

Page 39 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.4.5 Non-coordinate RNR The RNR configuration (139), with a gap at the right edge of a constituent A corresponding to a string within its sister B, is also found where A and B do not form a classic coordination. One case involves comparatives (140a), where B is the than-clause. Various authors argue that comparatives can be structured as coordinate constructions (e.g. Lechner 2001, 2004; Phillips 2003; Osborne 2009). Comparatives also license forward CR; cf. (140b). (139)

(140)

RNR also turns up in environments which are non-coordinate in the sense that they bar forward CR. Some involve arguments of predicates of comparison (141a,b) or motion (141c), or subordinators like whereas or without (141d,e), while in other cases, material is shared by a subject and its predicate (141f). Many such examples have a marginal flavour. (141)

That such examples belong to RNR is supported by several characteristics. Subword deletion as in (141b) is only otherwise found in coordinate RNR. The possibility for Pstranding precludes an analysis of (141e) as parasitic gap construction licensed by rightward shift of all the members across the without phrase (Williams 1990; Postal 1994). Like coordinate RNR, these dependencies are not constrained by islands; cf. (141a) and many examples in Postal (1994). In German, non-coordinate RNR in subordinate clauses Page 40 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising can yield the sharing of verbs in the predicate and a relative clause in the subject (142a), a possibility bled in main clauses by V2 (Wilder 1997; Lechner 2001: 705). (142)

(p. 717)

A general characterization of the environments permitting non-coordinate RNR is

lacking, as is a systematic study of other factors impacting on its occurrence.

27.4.6 Right-Node Wrapping The VP coordinations in (143) involve a shared object that surfaces in a non-final position in the final conjunct (Wilder 1999, 2008; Whitman 2009), a pattern which Whitman calls Right-Node Wrap. RNW shows a characteristic intonation pattern: the shared element is deaccented (cf. the pronoun in (143c)) and the non-shared material that follows it is accented. (143)

Object-sharing in the RNW pattern is attested for German by Hartmann and Schmitt (2013) and for French by Mouret and Abeillé (2011); it is also found in Norwegian (144). (144)

Page 41 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising The phenomenon itself and its implications for theories of RNR have only recently begun to be explored. One reason for treating RNW as part of RNR is that non-initial conjuncts show the RER (145). (145)

The non-final placement of α in the final conjunct makes it difficult to pursue an ATB movement analysis—if α has raised out of the coordination, then the non-shared phrase following it presumably has, too, in violation of the CSC. Another RNW pattern is found in German nominals, illustrated by (146b) (from the Web). Coordinated possession nominals with shared possession can take the form (146a), whereby the possessor pronoun (mein) has the bare form used in RNR but not NPE (cf. 27.4.3.4). (146)

Sabbagh (2012) notices that a quantifier phrase in the RNW configuration does not scope over the coordination, unlike a regular RNRed QP. Unlike (147a), (147b) cannot describe a prediction that some will be fired and others promoted. (p. 718)

(147)

Taking wide scope to be a reflex of overt movement of the RN out of the coordination (Sabbagh 2007), the narrow scope of the RN in (147b) indicates an in situ analysis for that case (as the word order independently suggests). However, the wide-scope reading of everyone also correlates with accenting; unstressed everyone, whether non-final or final (147c), does not have the distributive (wide-scope) reading.

Page 42 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising

27.5 Conclusion Viewed as a whole, CR is characterized by striking right–left asymmetries, suggesting that different mechanisms lie behind forward and backward gaps in coordination ellipsis. Forward CR shows affinities with canonical ellipsis with respect to vehicle change etc. This, together with indications that non-elided remnants in conjuncts affected by forward CR have undergone conjunct-internal movement, renders the move-and-elide approach particularly promising, though important questions remain. If move-and-elide is correct, then forward sharing arises from an ellipsis mechanism like that involved in VPE, NPE, and sluicing. RNR is different. An ellipsis approach falters in the face of identity patterns (absence of vehicle change) and wide-scope effects found with quantifiers, relational adjectives, etc. The major alternative has been Ross’s original rightward ATB movement proposal; however, doubt is cast on that approach by a range of considerations, including the RNR ability of sharing otherwise unmoveable elements. The third possibility, an in situ MD approach, appears to offer ways out of the dilemma, though much remains to be explored in this domain. If the MD hypothesis is correct, then RNR gaps are due to a basically different mechanism than forward CR and regular ellipsis.

Notes: (1) Throughout the chapter, angled brackets mark the coordination, i.e., the constituent comprising the conjuncts plus the coordinator. (2) The pattern in (4) was attributed by Ross (1967) to a rightward across-the-board extraction rule which he called Conjunction Reduction (see 27.2.6); it has been known as Right-Node Raising (RNR) at least since Postal (1974). The term RNR is commonly applied to the phenomenon itself without implying a particular analysis of it. (3) Another case of medial sharing is discussed in 27.4.6 (Right-Node Wrap). (4) To facilitate the presentation, the remainder of the chapter assumes (except where otherwise stated) a standard X'-theoretic analysis of non-conjoined expressions, in terms of a CP-TP(-vP)-VP clause structure, and a DP-NP structure for nominals. (5) Counterexamples to substitutability arise where an external item depends on the presence of the coordination—either, respectively, and plural agreement in (i)–(iii): ((i))

((ii))

Page 43 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising ((iii))

(6) For early discussion, see Schachter (1977a). Whitman (2004) has insightful discussion of semantic and pragmatic factors influencing judgements on coordinations of different types of phrases. (7) The LCL is not an inbuilt feature of (23) or (24); the type equality of XP and YP needs to be ensured independently. See Progovac (2003) and Borsley (2005) for critical discussion of endocentric approaches. (8) Examples like (i) suggest coordination of an intermediate projection (C’) (likewise for conjoined finite verb-initial constituents with shared initial specifier phrase in Germanic V2 languages); but under alternative assumptions (e.g., multiple heads in the C-domain, following Rizzi 1997), they may be analysed as phrasal coordination. ((i))

(9) For a defence of head coordination, see Abeillé (2006a). (10) On (i) (from Sag et al. 1985), which contradicts substitutability as well as the LCL, see 27.3.2. ((i))

(11) Constructions where a predicative UCC is isolated from the verb, such as fronting (i), raise questions for (29a). Beavers and Sag (2004) consider the analysis (ii) involving larger conjuncts with backward CR, but Levine (2011) points out problems. For instance, (iii) indicates a smaller coordination, since both cannot modify conjoined root clauses. ((i))

((ii))

((iii))

Page 44 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising Chaves (2006) discusses related adjunct UCCs (Both tired and in a foul mood, Bob packed his gear). (12) Neijt (1979: 62–6) gives further arguments against ‘conjunct extraposition’. (13) The construction is sometimes termed Bare Argument Ellipsis, yet it is also found with predicates (He seemed angry yesterday, and rather drunk). A different approach to stripping, according to which the correlate in the first conjunct raises covertly to attach to the second conjunct, is argued for by Reinhart (1991); see also Fiengo and May (1994: 111–12). (14) Prinzhorn and Schmitt (2010) discuss German data problematic for this account. (15) On the interaction of either and ellipsis, see Den Dikken (2006) and Hofmeister (2010). (16) The ATB exception to the CSC applies only to movement of an element contained within a conjunct. Conjuncts themselves appear to be immobile (Grosu 1973); granted, that is, that there is no conjunct extraposition. (17) Cases of (apparent) coordination that do not obey CSC, such as What did he turn round and say to you?, already noted by Ross (1967) and later discussed by Goldsmith (1985), Lakoff (1986), Postal (1998), and others, are not considered here. (18) Ross’s (1967: 174) rule (“Conjunction Reduction”) was responsible for both RNR and leftward ATB movement. (19) Proposals for direct coordination accounts of PCC in terms of flexible constituency also include Pesetsky’s (1995) “layers and cascades” and Phillips’ (1996, 2003) constituency-altering derivational merge approach, which also extends to RNR. (20) X/Y signals a function that combines with Y to its right to give X; while X\Y is one that combines with Y to its left to give X, where X and Y can both be function types. (S\NP)/NP is the category for a finite verb in an SVO language. (21) The CCG approach outlined by Steedman suggests a basic right–left symmetry for ATB constructions. RNR is claimed to be island-constrained like leftward movement (Steedman 2000: 17), contra Wexler and Culicover (1980) and many since (see 27.4.2). (22) A case which does not motivate a CR approach is the phenomenon of respectively readings in coordination, whereby (i) is paraphrased by (ii). ((i))

((ii))

Page 45 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising These readings have been described as fundamentally problematic for the syntax of coordination (Postal 1998: 135). In the logic of early Transformational Grammar, the paraphrases suggest a clausal coordination source, and syntactic solutions that assign sentences like (i) (in the relevant reading) a biclausal syntax have been suggested, e.g. by Goodall (1987). However, the fact that such readings also arise with non-conjoined plurals, as in The first two boys met the last two girls (respectively), indicates that their source is not to be found in coordination syntax (via some kind of CR), but in semantics (e.g., a covert distributive operator). See Gawron and Kehler (2004) and Chaves (2012). (23) In Merchant’s (2001) proposal, phonological deletion is triggered by an “E-feature” introduced by X in (56), which simultaneously imposes the requirement for the elided YP to stand in a suitable semantic relation (“e-GIVENNESS”) with an antecedent. (24) A central (and hitherto unanswered) question facing attempts to assimilate forward CR to regular ellipsis is why with forward CR, unlike the canonical ellipsis types, the antecedent–ellipsis relation is coordination-bound. See Johnson (2009) for relevant discussion with regard to gapping. (25) Attempts at characterizing the possibilities are Hankamer’s (1979) Major Constituent Condition on remnants, combined with a Head Condition prohibiting deletion of material c-commanded by an overt head (Wilder 1997). (26) The idea can be implemented in the Minimalist framework by assuming an abstract head X which attracts ZP, and carries the E-feature which simultaneously imposes the e-GIVENNESS requirement on YP and triggers its phonological deletion. If the LCL is interpreted in terms of syntactic categories, this kind of solution would require the initial conjunct also to be an XP, even though the correlate to the focused remnant typically remains in situ. ((i))

(27) Sluicing can in certain cases remedy island violations (cf. Merchant 2001), and the question arises of why the effect is found with some kinds of ellipsis but not others. See Griffiths and Liptak (2014) for discussion. (28) A case that deserves closer scrutiny is Reinhart’s (1991) examples involving a remnant extracted from a subject, as in (i): ((i))

((ii))

Page 46 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising (29) This pattern appears to be restricted to coordinations, unlike its mirror image RNR counterpart (cf. Chaves 2008: 271); compare ?distinguishing neurolinguistics from psycholinguistics and *distinguishing half-brothers from half-sisters. A similar pattern in Japanese is discussed under the rubric of Left-Node Raising by Yatabe (2001). (30) Kluck and de Vries (2013) argue that RNR can feed non-string-vacuous rightward processes such as extraposition of nominal modifiers. (31) Postal (1998: 178 and 195, n. 18) presents examples as grammatical that do not conform to RER. His judgements are contested by Levine (2001: 163) and others. (32) Cases where the shared element surfaces in a non-final position in the final conjunct are discussed in 27.4.6. (33) This suggests that backwards V-gapping, famously restricted to SOV languages (SO&SOV vs *SO&SVO; Ross 1970), is RNR. (34) Sabbagh (2007) offers an account for why (91) should be possible, i.e., why RNR does not respect islands. His approach does not, however, afford insight into other surprising RNR targets, including sublexical units and multiple constituent sequences. (35) A similar problem arises in leftward ATB cases like (i), with a single fronted constituent apparently containing the traces of the non-shared subjects John and Mary. See 27.4.4.2. ((i))

(36) This argument is not addressed in Sabbagh (2007). The failure of RNR to interact with sluicing points in the same direction: ((i))

((ii))

(37) Postal (1998: 106–7) and others have proposed that certain types of expression cannot undergo RNR, whether for syntactic or prosodic reasons. Chaves (2014: sections 2.1–2.2) gives some counterarguments. (38) Giannakidou and Merchant (1998) analysed examples like (i) as “reverse sluicing” (i.e., backward ellipsis). However, if does not license regular (forward) sluicing (ii), and (i) is better analysed as RNR. Page 47 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising ((i))

((ii))

The following exemplifies backward sluicing, evidenced by wiping (Merchant 2002: 294): ((iii))

(39) In a context like (i), (126b) improves considerably, presumably because the VP gap is parsed as an ellipsis whose antecedent is provided by the preceding question, rather than the following conjunct. ((i))

A contextual antecedent can also render (117b), (119b), and (121b) acceptable. (40) However, natural-sounding examples with but can be found that appear to involve backward ellipsis: ((i))

((ii))

((iii))

(41) Responding to Barros and Vicente, Larson (2012) argues that there are data paradoxes that a hybrid approach cannot resolve. Larson (2013) further claims that no approach which attributes internal structure to RNR gaps is able to avoid paradoxes, and that a better theory is one that simply denies that there is any structure in RNR gaps. (42) Bošković and Franks (2000) argue that covert ATB movement does not exist. However, their arguments relate to examples which contain two pronounced copies of the relevant expression inside the conjuncts. Under an account in which both RNR and ATB Page 48 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Conjunction Reduction and Right-Node Raising movement involve MD (Citko 2005), covert ATB movement would target a shared constituent inside the conjuncts, i.e. one that surfaces in the RNR pattern. The possibility for a wide scope of an RNRed QNP pointed out by Sabbagh (2007) is consistent with that possibility, as is the possibility for a multiple wh-question such as (ii). ((i))

((ii))

(43) An ATB movement approach to RNR could in principle avoid the non-shared trace problem by denying that the shared string forms a single constituent, and assuming that the examples involve movement of multiple constituents. This is suggested by Sabbagh (2007: 395–7) for cases like (134). The same would apply in the leftward movement case (138b).

Christopher Wilder

Chris Wilder is Professor of English Linguistics at the Norwegian University of Science and Technology in Trondheim, where he has been employed since 2005. Prior to that he spent fifteen years as a researcher and lecturer in Germany and the USA. His research interests include English and German syntax, comparative syntax in general, and ellipsis and constituent-sharing phenomena in particular.

Page 49 of 49

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

Oxford Handbooks Online Hungarian   Anikó Lipták The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.30

Abstract and Keywords This chapter reviews ellipsis in Hungarian, providing examples of the majority of ellipsis types that are discussed in this book. The discussion starts with nominal ellipsis and shows that, unlike ordinary noun phrases, possessed noun phrases do not show evidence for ellipsis, V-stranding ellipsis, and pseudogapping) and includes a section on preverbstranding ellipsis and the analytical challenges it represents. Clausal ellipsis phenomena (sluicing, fragments, stripping) are treated with reference to single remnant ellipsis and multiple remnant ellipsis as well, the latter including a description of gapping. Separate sections are dedicated to right-node raising, null complement anaphora, and ellipsis in comparative clauses. Analytical and theory-specific details about the structure of Hungarian are introduced along the way. Keywords: Hungarian, left periphery, movement, A-bar extraction, ellipsis licensing

Anikó Lipták

31.1 Nominal ellipsis AS in many other languages, the head noun or the head noun together with one or more modifiers can be unpronounced to the exclusion of a modifier, numeral, or (quantificational) determiner in Hungarian. Ellipsis is strictly only found in nonpossessive noun phrases, while in possessed noun phrases, a non-elliptical anaphoric strategy is used. The next two subsections give details of both strategies.

31.1.1 Nominal ellipsis in non-possessed nominals In non-possessed nominals, the missing nominals can be understood with reference to an entity in the linguistic or the extra-linguistic context. In the following examples, the part of the noun phrase that is understood to be missing is indicated by __ .1 Page 1 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (1)

As can be observed, when the noun is missing, the overt number morphemes and case morphemes that normally get spelled out on the noun only appear on the linearly (p. 816)

last remnant preceding the missing noun (in case they contain a harmonic vowel, they harmonize with the last remnant, too), as was noted in Bánréti (1992, 2007), Kenesei et al. (1998), Laczkó (2007), and Saab and Lipták (2016). The linearly last remnant can also be a clausal modifier: for example, an adjectival participial clause, as shown in the following example, (2b): (2)

Missing nouns or nominal constituents have been analyzed as involving a silent pronominal pro in Bánréti 1992, 2007 (see also Laczkó 2007) and as ellipsis in Dékány 2015 and Saab and Lipták 2016.2 Here I follow the latter kind of proposals and assume that the missing nominal is the result of ellipsis. Adopting a structure like (3) for (unpossessed) nominal constituents (see, e.g., Cinque 2005), in which adjectives are adjoined to the noun phrase, and the plural marker originates in the NumP projection, the deleted category in Hungarian nominal ellipsis corresponds to an NP. The missing NP may be unmodified or modified (see (4) and (5) for examples of both types). (3) Page 2 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

Evidence that the elided category cannot be as big as a numeral phrase (NumP) comes from the observation that numerals and number morphology must always survive the ellipsis (cf. (4) and (5)). (4)

(5)

There are some information-structural criteria that NP ellipsis usually complies with. Preferentially, NP ellipsis contains adjectival remnants that are new and not given in the discourse—in the sense of not having been mentioned yet. Pronouncing given adjectival modifiers gives a slightly awkward, redundant utterance, but they do not count as ungrammatical. In the case of numerals as remnants, givenness is fully tolerated and gives rise to no sense of redundancy. (p. 817)

(6)

It is also important to mention that non-identity between remnants of NP ellipsis and their correlates does not trigger any syntactic marking of contrast via contrastive focusing. Consider the following examples where the adjectival remnant is non-identical to another adjective in the antecedent clause: (7)

Page 3 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian In (7), the elliptical noun phrase is in postverbal position and is intonationally unmarked―there is no contrastive stress on the remnant. Both properties are earmarks of constituents that do not distribute as contrastive focus or contrastive topic expressions (with contrast on their adjective/numeral). That being said, elliptical noun phrases can contain contrastively focused remnants, and―in line with the rules of Hungarian―appear in the preverbal focus position as contrastively focused phrases (cf. (8)). (8)

This shows that contrastive focus on the remnant is optional, and not a necessary property of elliptical nouns, and thus cannot be considered the licensing factor for noun phrase ellispis in general (contra Corver and van Koppen 2009; Eguren 2010). (p. 818)

31.1.2 Anaphoric possessed nominals In contrast to non-possessed noun phrases, possessed noun phrases do not allow for nominal ellipsis. This equally holds for possessives with dative (9a) and with nominative possessors (9b) (see Szabolcsi 1994 for basic differences between the two). The lack of elliptical possessives stems from the fact that anaphoric possessives make exclusive use of a pronominal strategy that is earmarked by the use of the -é suffix (Bartos 2000b; Laczkó 2007; Dékány 2011, 2015), illustrated in (9c). (9)

Page 4 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian The precise analysis of the -é suffix is a point of contention in the literature (Bartos 2000b equates it with a functional head that selects the noun, Laczkó 2007 with the proform, and Bartos 2001b, Dékány 2015 with genitive case). What all analyses agree on is that anaphoric noun phrases involve a proform. As can be seen in (10a), the anaphoric noun phrase is obligatorily adorned with the possessor agreement morpheme (such agreement morphemes are present with pronominal possessors in Hungarian) and the number morpheme indicating plurality of possession: -tek spells out agreement with a 2PL possessor and -i indicates plural possession. Importantly, as (10b) shows, -é can never co-occur with the possessedness morpheme -(j)a/(j)e, neither can it co-occur with the possessed noun (10c). (10)

Anaphoric possessed noun phrases furthermore cannot contain any overt numeral or adjectival modifier; see the next example as illustration. In Dékány 2015, the latter property (p. 819) is explained with reference to the fact that anaphoric possessed noun phrases contain a proform and pronominals cannot be modified in Hungarian. The ungrammaticality of numeral and adjectival remnants in possessed nominals sharply contrasts with non-possessed nominals, where such remnants are allowed (see (4) and (5) again). This difference provides a strong argument to the effect that the missing element in anaphoric possessed noun phrases is a proform, while in non-possessed noun phrases it corresponds to ellipsis of a nominal projection (see Dékány 2015). (11)

Concerning the “size” of the anaphoric pronoun in possessives, Dékány argues that the anaphoric proform replaces a piece of structure that is bigger than an NP. To understand why, consider the structure of possessives as in (12) (Dékány 2015; also É. Kiss 2002) where there are two functional projections dedicated to marking the possessive relation: the possessor agreement projection (Poss2P) and the possessedness projection (PossP). Page 5 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian These two projections flank the NumP that hosts the plural possession marker -i. Standardly, the possessor is taken to be generated in Spec,PossP and the possessed noun as head of the NP: (12)

From the fact that in anaphoric possessives, the noun and the possessedness morphemes are never overt (cf. (10b,c)), Dékány concludes that the proform must minimally correspond to the Poss’ node, which subsumes the NP projection and the Poss0 head. To finish off the discussion of anaphoric possessives, consider the following example, which at first sight seems to contradict the claim that ellipsis is impossible in possessed noun phrases. The interpretation of the missing noun is preferred to be that of a possessed nominal. (13)

There are, however, two strong indications that the possessed interpretation is only pragmatically controlled for in cases like this and that we are dealing with an unpossessed nominal undergoing ellipsis here. One indication is provided by the nominal morphology found in the elliptical nominal: the endings are characteristic of non-possessed noun phrases. In possessed noun phrases, (p. 820) the plurality of the possession is spelled out by the invariable -i morpheme, cf. kabát-a-i (coat-POSS-PL), while in the elliptical új-ak (new-PL.ACC) the plural marker is the ordinary -k morpheme (together with an epenthetic vowel) that is found on nonpossessed nouns. As the ungrammatical forms furthermore illustrate in (14), there is no other variant that is acceptable (as noted in Kenesei et al. 1998). (14)

Page 6 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian The other argument against a possessed NP analysis of these data comes from the observation that the possessor can never be overtly present in the elliptical nominal, either in dative or nominative case (15). (15)

These two observations jointly confirm that the elliptical noun phrase in (13) is not a possessed NP, but an unpossessed one, and the possessed reading of the missing nominal must be derived pragmatically.

31.2 Predicate ellipsis The main predicate of the clause can be missing in Hungarian in two configurations: what can be referred to as auxiliary(AUX)-stranding VP-ellipsis and V-stranding VP-ellipsis, following a distinction made by Goldberg (2005). This section reviews these two types, together with a further reduced variant of V-stranding, the so-called preverb-stranding pattern. This section closes with a short discussion of pseudogapping. Before turning to the specifics of predicate ellipsis, it is important to introduce key assumptions about the clause structure of Hungarian that are going to be made use of in the discussion. In most syntactic accounts, Hungarian clauses are taken to contain an inflectional layer, termed TP in the following (comprising tense, agreement, and mood specifications/ projections not distinguished here any further), and a predicate layer, termed vP below. The predicate layer also comprises various subprojections, most notably an aspectual projection (AspP/PredP) and the core lexical predicate, the VP. Lexical verbs in Hungarian often combine with so-called verbal modifiers with aspectual/predicative meaning, which are phrasal constituents (Koopman and Szabolcsi 2000; den Dikken 2004; Surányi 2009a) and comprise preverbal particles (or preverbs, PVs for short), incorporated nominals, and PPs of distinct types. (p. 821)

(16)

Page 7 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian Verbal modifiers are syntactically independent of the verbal head (for this reason they will be spelled as separate words in this chapter, contrary to rules of Hungarian orthography). Following Piñón (1995), Olsvay (2000, 2004), and Surányi (2009a), I take verbal modifiers to originate from an AspP/PredP-internal position and to move to specTP in overt syntax (in order to satisfy an EPP property, note that subjects do not raise to specTP). Finite verbs raise to the tense head (Brody 1990; Kenesei 1998; Surányi 2009a), and this results in the obligatory adjacency between verbal modifiers and the verbal head that characterizes all clauses without focal material. The structure in (17), indicating the position of the verb and the most frequent verbal modifier, the preverbal particle, will be adopted in the discussion in 31.2.1–31.2.3. (17)

31.2.1 AUX-stranding predicate ellipsis AUX-stranding predicate ellipsis data have been described in Bánréti (1992) and Bartos (2000a) (see also Gyuris 2001). This is a type of ellipsis that removes a predicate and strands a finite auxiliary or a so-called semi-lexical verb. Hungarian has two frequently used auxiliaries, the habitual auxiliary szokott HABIT and the future auxiliary fog FUT (Kenesei 2001). Semi-lexical verbs are verbs like akar ‘want’, szeretne ‘would like’, or modals like kell ‘need’ or lehet ‘may’. AUX-stranding VP-ellipis can occur with all these items, in matrix or embedded contexts (including ACD contexts and comparative clauses as well): (18)

In (18a) the tenses of the two clauses are not identical, which provides evidence that the elided category is smaller than a tense phrase. This conclusion dovetails with accounts that place finite auxiliaries in T0, such as Kenesei (1998) and Surányi (2009a). In line with these accounts, AUX-stranding predicate ellipsis is ellipsis of a vP constituent.

Page 8 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian While the examples in (18) contain overt subjects (nominal and dative), other types of phrases can also appear A-bar extracted out of the site of AUX-stranding predicate ellipsis and occupy positions such as that of topic, contrastive topic, or focus/question word phrase before the auxiliary (19a,b,c). (p. 822)

(19)

The constituents that line up before the auxiliaries in these examples move to specific positions in the left periphery of Hungarian. Te in (19a) is a topic, kivel is a question word, Marival in (19b) is a contrastive topic, and EGY BICIKLIT in (19c) is a contrastive focus constituent. The order of these items reflects the usual order of left peripheral elements in the language. Hungarian places focus, topic, and quantificational material in various ordered positions in the left periphery (giving rise to the often quoted discourseconfigurationality of the language; É. Kiss 1995). The articulated left periphery, illustrated in (20), houses topics (contrastive and non-contrastive, in TopPs), quantifiers (in DistPs as in Szabolcsi 1997 or adjoined to FocP and TP as in Surányi 2002 and É. Kiss 2010), and contrastive focus and wh-phrases (in a unique FocP, cf. É. Kiss 1978, Brody 1995, and Szabolcsi 1997).3 (20)

That we are dealing with A-bar extraction out of predicate ellipsis in (19) and not (English-type) pseudogapping is evidenced by at least three observations. First, as (19c) shows, the second remnant need not be contrastive with respect to a correlate in the antecedent (it is the same item egy biciklit ‘a bike’), unlike in English-type pseudogapping. Second, long-distance extraction is possible in examples of this type: see (21), where the missing predicate corresponds to akar, hogy fölvegyünk, meaning ‘want

Page 9 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian that we hire’. Such long-distance dependencies cannot be established in English pseudogapping (Johnson 2001b). (21)

Last but not least, examples like (19) can support a sloppy reading (cf. the interpretation of 19a), while pseudogapping does not support such a reading (see Johnson 2001b and section 31.2.4. for illustration). (p. 823)

31.2.2 V-stranding predicate ellipsis Predicate ellipsis in Hungarian can also exhibit a pattern of V-stranding (Bánréti 1992) similar to the one found in Finnish (Holmberg 2001, 2016), Irish (McCloskey 1991a), or Hebrew (Doron 1999), among other languages. There are two syntactic contexts in which V-stranding can rear its head in Hungarian: one is polarity contexts, such as yes/no questions and answers or confirmations to declaratives (cf. 22), the other is a context with non-emphatic polarity, typically involving an is-phrase (also/too) before the stranded verb (cf. (23); Surányi 2009a,b). Interestingly, there is extensive microvariation concerning these two types of V-stranding: one dialect of Hungarian only allows for V-stranding in polarity contexts (call it variant A), another for both types (call it variant B).4 (22)

(23)

Page 10 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian A quick comparison with patterns of pro-drop shows that the missing material in these examples can only be due to ellipsis of an entire verb phrase, and not to individual argument ellipsis. As example (24) shows, 3PL objects (animate and non-animate alike) cannot be dropped in Hungarian. (24)

Based on this consideration as well as others, Lipták (2013b) argues that facts like (22B/23) involve predicate ellipsis: V-stranding VP-ellipsis strands the finite verb in T and elides the vP, including all arguments contained in there. (p. 824)

Lipták (2013b) furthermore suggests that the microvariation concerning the availability of the two patterns identified above is due to variation in the licensing of V-stranding ellipsis in the two variants of Hungarian. While in variant B, V-stranding ellipsis can be licensed by finite tense, just like in AUX-stranding predicate ellipsis that strands an auxiliary, in variant A V-stranding ellipsis is licensed by emphatic polarity only. The precise mechanism of licensing by emphatic polarity can be successfully modelled in the theory of ellipsis licensing in Aelbrecht (2010), where the licensor corresponds to a syntactic head that must c-command and establish an Agree relation with the head whose complement is elided. In polarity contexts, the licensor is a polarity head, Pol0 (similar to Laka’s 1990 Σ) that selects the TP, and whose lexical content amounts to stress on the (preverb+)verb combination. In elliptical configurations, the T head hosts an ellipsisspecific feature [E] that brings about the non-pronunciation of the complement category. Via an agreement process between Pol and T, ellipsis is licensed. In variant B, V-stranding ellipsis is licensed in a local relation by T. (25)

Page 11 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian Further evidence for the long-distance Agree mechanism between an ellipsis licensor and the head that triggers ellipsis, can be provided from V-stranding phenomena in sentences containing verbal complexes. When a finite verb is followed by a series of infinitival complements, V-stranding ellipsis can strand any finite or non-finite verbal projection and yield exactly the same meaning. The following example illustrates the various acceptable patterns in B1, B2, and B3.5 (p. 825)

(26)

In examples like (26B1) or (26B2), the licensing category (Pol0) is non-adjacent to the ellipsis site, as a number of overt verbal projections (akarni hívni in B1 and akarni in B2) intervene (and note that these verbal projections appear in the base-generated order). Such non-adjacency provides a strong argument for a long-distance approach to ellipsis licensing such as that of Aelbrecht (2010).

31.2.3 Preverb-stranding ellipsis Hungarian also exhibits a stranding-type ellipsis that does not strand the entire verb, but rather only the preverbal particle that combines with the verb. With reference to the frequently occurring instance of stranded preverbal particles, this type of ellipsis can be referred to as preverb-stranding ellipsis. Preverb stranding is allowed in both variant A and variant B of Hungarian and is strictly confined to polarity contexts. This type of ellipsis can occur in positive answers to polar questions and in affirmations to declaratives. It cannot occur in non-polarity contexts like (23) above; cf. (28). (27)

Page 12 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

(28)

The most straightforward account of preverb stranding would have it that it is structurally identical to V-stranding, except that the verbal head fails to raise out of the ellipsis site, cf. (29) (as v-to-T-movement is bled by ellipsis, as proposed in Lipták 2012; see also van Craenenbroeck and Lipták 2008 for other bleeding effects of ellipsis on verb movement). (p. 826)

(29)

There are, however, indications that (29) and the account offered in terms of bleeding are unlikely to be on the right track. A closer look at the data reveals that the syntactic distribution of V-stranding and that of preverb stranding is not fully identical. One crucial difference between the two is that while verb stranding can be used as a positive response to a negative yes/no question, preverb stranding cannot:6 (30)

One way of explaining this contrast is to say that preverb stranding deletes more structure than just the vP projection, and in fact even more than the entire TP: it deletes the entire PolP that is standardly generated in answers and affirmations to polar questions. With the assumption that preverb stranding elides a full clause including the

Page 13 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian polarity specification, the ungrammaticality of (30B2) follows as a failure of identity between the elided constituent and its antecedent. To give a sketch of how this could potentially work, consider the following preliminary analysis. First, assume that PolP is the locus of both negative and positive polarity specifications in the clause and contains interpretable [Neg] and [Aff] features in negative and positive clauses respectively (see independent evidence for these features in Lipták 2013b). Assume furthermore that preverbs move to a position outside PolP in preverb stranding, a position that will be referred to as FP below.7 The complement of FP is affected (p. 827) by deletion in preverb stranding, as shown in (31). Importantly, this contrasts with ellipsis in V-stranding which affects a lower category, vP only. (31)

Using (31), we can explain the pattern in (30). V-stranding with a negative antecedent (cf. 30B1) is grammatical, because V-stranding involves vP-deletion and the elided vP in (32B) (marked by < >) is strictly identical to the vP in the antecedent (32A). (32)

The preverb-stranding answer in (30B2 / 33B), however, is ill-formed since in this case a larger category is elided. Most importantly, the elided category contains the PolP which is featurally non-identical to its antecedent: it contains an affirmative feature while the antecedent has a negative one. (33)

If this account is on the right track, the Hungarian facts in this section demonstrate that stranding-type ellipsis does not always elide vP predicates, but also higher projections in the clause. If this is correct, preverb stranding should be classified as a case of clausal ellipsis.

31.2.4 Pseudogapping As was mentioned in section 31.2.1, AUX-stranding predicate ellipsis can with A-bar extraction, resulting in sentences that resemble but in fact do English-type pseudogapping (see examples (19a–c)). There are, however, closely match the syntactic profile of English-type pseudogapping. These

occur together not instantiate examples that feature an

Page 14 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian auxiliary followed by a focal remnant that is contrastive with respect to a preverbal constituent. The postverbal and the preverbal constituents form an ordered pair whose thematic relation is the reverse of that in the antecedent (small caps stand for emphatic stress): (34)

Note that the post-auxiliary remnant necessarily corresponds to a given constituent that ‘switches’ its argument position with respect to the antecedent. The pre-auxiliary constituent is normally a topic or an also-phrase. This shows that the ordered pair of constituents in the elliptical clause do not form a multiple focus construction (as such constructions always involve the preverbal constituent in focus position, see sections 31.3.2 and 31.3.3). At the (p. 828) same time, examples like (34) express pair-wise focus in the sense that the elliptical clause and its antecedent necessarily differ only in the order of the pairing relation between participants. This kind of focus will be referred to as reversing focus below.8 A clear indication that we are dealing with pseudogapping in (34) comes from (i) the fact that the final remnant is contrastive in the way defined above (see also n. 8), (ii) the fact that long-distance extraction akin to (21) is impossible in these cases. Last but not least, the unavailability of sloppy identity readings in these constructions further supports the suspicion that we are dealing with an instance of English-type pseudogapping here. As Johnson (2001b) points out (attributing the observation to Chris Kennedy), pseudogapping does not license sloppy identity, while VP ellipsis does: (35)

Exactly the same difference is observable in Hungarian. Consider first the case of VP ellipsis in (36), which allows for both sloppy and strict readings: (36)

Page 15 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian Pseudogapping on the other hand is incompatible with a sloppy reading, as the following example shows―note that the sloppy reading is unavailable even though this is the reading favored by the context, the strict reading being unlikely: (37)

Concerning the derivation of Hungarian pseudogapping, only speculations can be given at this point. Clearly, reversing focus never moves to the preverbal focus position (it is not an instance of exhaustive focus), and its position in the postverbal domain is rather free: it can occur in various positions, showing a slight preference for the clause-final (p. 829)

one, as the following non-elliptical versions of (34) show. (38)

The elliptical variant in (34) can be derived assuming that the focal remnant undergoes short A-bar movement to end up right before the infinitive. In that position, the infinitival VP undergoes ellipsis as in (39). This account is compatible with analyses of pseudogapping in the literature that postulate a (short) A-bar movement process to the left (Jayaseelan 2001; Gengel 2013, among others). (39)

Finally, it can be noted that the same type of reversing focus can also survive ellipsis of a predicative constituent out of which V-movement has taken place, i.e. pseudogapping can also take place when accompanied by V-stranding: (40)

Page 16 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

31.3 Clausal ellipsis Hungarian, like many other languages, exhibits various instances of clausal ellipsis, such as sluicing, stripping, and fragments. All these types involve TP-ellipsis and as such are characterized (and can be identified) by strong tense-identity effects. The tense of the elliptical clause is always identical to that of the antecedent clause in clausal ellipsis. Observe this in the following instance of stripping. (41)

(p. 830)

The following three subsections review single-remnant and multi-remnant clausal

ellipsis (including gapping) in Hungarian.

31.3.1 Single-remnant clausal ellipsis and the theory of ellipsis licensing Remnants of clausal ellipsis move to the structurally rich left periphery (see (20)). As a direct result of Hungarian’s rich left periphery, clausal ellipsis exists with various remnant constituents (Bánréti 1992, 2007), such as is-phrases in stripping (cf. (41)), whand focus constituents in sluicing and fragments (cf. (42)), and focus constituents in fragments (cf. (43)). (42)

Page 17 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (43)

Sluicing and fragments with single remnants share identical derivations, including the position targeted by the remnants in them (Lipták 2011; Griffiths and Lipták 2014)— which is not surprising given that wh-constituents and preverbal focus occupy the same structural position, FocP, in non-elliptical sentences as well. The structural similarity extends to embedded contexts as well. In sluicing contexts (where an ellipsis remnant corresponds to an indefinite in the antecedent clause), TP-ellipsis can have a wh- as well as a focus remnant, as example (44) shows (Horvath 2005; van Craenenbroeck and Lipták 2006)—this example can also be treated as a case of a genuine embedded fragment.9 (p. 831)

(44)

This is in line with the fact that wh-questions and focus constructions make use of the same syntax and occupy the same position in the left periphery. They are in complementary distribution in main clauses and trigger the separation of the preverbal particle from the verb in exactly the same way: (45)

In fact, TP-ellipsis is perfectly well-formed in embedded contexts (just as it is in fragments) with fronted emphatic operator constituents such as universal quantifiers (whose position is just above FocP):

Page 18 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (46)

(45) and (46) have important repercussions for the theory of ellipsis as they show that wh-movement is not a distinctive trait of sluicing. Rather, ellipsis licensing is sensitive to the type of feature the wh-phrase checks in overt syntax, which in Hungarian happens to be an operator feature. Based on data like (45) and (46) in Hungarian and other languages, van Craenenbroeck and Lipták (2006) suggest that sluicing tracks wh-syntax in this sense across languages: the feature content of wh-elements in non-elliptical questions determines what kind of remnants can escape TP-ellipsis in sluicing, referred to as the wh/sluicing correlation:10 (p. 832)

(47)

The syntactic features that the [E]-feature checks in a certain language are identical to the strong features a wh-phrase checks in a non-elliptical constituent question in that language. The reach of (47) interestingly extends to the domain of relative clauses as well in Hungarian: relative pronouns, which arguably have operator features just like question words, allow for ellipsis of their complement in free relatives and pronominally headed relatives (Lipták 2015)—contradicting the received opinion that TP-ellipsis stranding a relative operator should be impossible (van Riemsdijk 1978, Lobeck 1995, and Merchant 2001). (48)

Page 19 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian Evidence for clausal ellipsis in (48) is culled from various domains in Lipták 2015, including the observation that the annyi-amennyi pronominal+relative pronoun complex occurs only in syntactic positions where one would expect a relative clause, and the fact that (48) exhibits characteristic properties of antecedent-contained sluicing (Yoshida 2010), such as tense and modality mismatches between the elliptical and the antecedent clause. (49)

Sluicing in this type of relative clause furthermore has a characteristic (crosslinguistically rare) prosodic profile that is unlike the prosodic profile of non-elliptical relatives.11 (p. 833)

31.3.2 Multiple-remnant clausal ellipsis and ellipsis repair Just like many languages, Hungarian can also form clausal ellipsis stranding multiple remnants. This is not surprising in the domain of wh-syntax, since it has been known since É. Kiss (1987, 1993) that Hungarian allows for multiple wh-movement to the left periphery. In fact, there is evidence from interpretation that multiple sluicing can only be formed via the derivational route of ordinary multiple wh-movement and no other. As Grebenyova (2007, 2012), and following her, van Craenenbroeck and Lipták (2013b) observe, multiple wh-sluicing only allows for interpretations that are also available for multiple wh-movement as well. Hungarian multiple wh-movement only supports a pair-list reading (É. Kiss 1993), cf. (50), and multiple wh-sluices are only possible in contexts that support such a reading, too. Thus they are ruled out in single-pair contexts such as (50), but are well-formed in contexts such as (52) where a distributive reading is established. (50)

(51)

Page 20 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

(52)

In the domain of focus, however, the derivational options of multiple focus under ellipsis do not run parallel to the choices of overt focus movement in Hungarian: multiple focus remnants are allowed notwithstanding the fact that multiple focus fronting in nonelliptical contexts does not exist. In other words, two adjacent foci are only possible when followed by ellipsis: (p. 834)

(53)

One possible way of thinking about this apparent breakdown between ellipsis and nonelliptical syntax is to consider it to be a PF-repair effect of ellipsis, as proposed in van Craenenbroeck and Lipták (2013b). The proposal is based on the assumption that multiple focus movement in Hungarian triggers a PF-violation, namely that non-final focus constituents are not adjacent to the verb (recall from n. 3 that there is obligatory adjacency between the lexical verb and the fronted focused constituent). (54)

Page 21 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian If the observed focus-verb adjacency is a phonological requirement, it is possible to attribute the repair effect of ellipsis to the fact that it removes the verb in clausal ellipsis and thus eliminates the confounding factor altogether: if there is no verb, it does not have to be adjacent to anything and multiple focus movement is allowed as in (55):12 (55)

Finally it must be mentioned that multi-remnant TP-ellipsis can also strand constituents with distinct discourse functions. It can exhibit a combination of topics, quantifiers, and focus constituents, in the order allowed in non-elliptical sentences, i.e., topic < quantifiers < focus. For illustration, consider the following contrastive fragments in (56B1) and (56B2). B1 involves a quantifier followed by a focus, B2 has a contrastive topic followed by a focus. (56)

31.3.3 Multiple remnants in gapping As Bánréti (1992, 2007) has pointed out, gapping-like elliptical constructions have two basic types in Hungarian: cases in which the final remnant is a contrastively focused constituent and cases in which the final remnant is not focal, but instead the elliptical clause (and its antecedent) corresponds to an information-structurally unmarked sentence. Starting with the former type, consider (57), which illustrates gapping with two remnants, a contrastive topic (én) and a preverbal focus (ma). (57)

Page 22 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

As the reader can ascertain, this kind of example resembles the cases in (56) in the sense that it features left peripheral constituents arranged in the expected order. Accordingly, the ellipsis here corresponds to ordinary TP-ellipsis, and thus does not in fact instantiate English-type gapping. One strong reason to think so is that unlike English gapping, the elliptical clause here can be embedded with respect to its antecedent—cf. the following example, adapted from Bánréti (2007) (see Farudi 2013 for similar facts in Farsi.) (58)

A distinct subtype of gapping with focal remnants is illustrated in (59). This contains two focus remnants, corresponding to the two members of a complex multiple focus construction (as defined in Krifka 1991, corresponding to focus on an ordered pair). As reflected by word order in the antecedent clause (and the preference for parallelism in structure), both members correspond to focused entities: the first has a preverbal correlate and the second a postverbal one. Note that this kind of gapping cannot contain the elliptical clause embedded with respect to its antecedent, thus more closely resembling gapping in other languages (59b): (59)

Page 23 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian The derivation of these constructions is arguably the same as that presented for multiple focus remnants in section 31.3.2. Multiple focus remnants are allowed in the left periphery in these contexts due to elliptical repair: two focal remnants normally cannot both front as they would violate the focus–verb adjacency rule. Under ellipsis, this rule is vacuously satisfied and both constituents can undergo fronting: (p. 836)

(60)

The least understood pattern of gapping is gapping with remnants that are non-focal in nature, which are anteceded by an information-structurally unmarked clause. The following examples, adapted from Bánréti 2007, illustrate this pattern. As Bánréti notes, the remnants normally correspond to arguments rather than adjuncts, and the number of remnants can be higher than two. (61)

Since in this construction type the antecedent is an information-structurally unmarked clause, and the remnants do not have any specific discourse functions, one would be inclined to think that the missing finite verb is elided in situ as a result of non-constituent deletion. (62)

Given, however, that Hungarian does not show evidence for such in-situ deletion in other contexts, the account in (62) is unlikely. (p. 837)

31.4 Right-node raising Right-node raising (RNR) in Hungarian is an attested phenomenon (Bánréti 1992, 2001, 2007), as the following examples (adapted from Bánréti 2001 and Surányi 2009b respectively) show. The shared pivot is enclosed in brackets in these examples. Page 24 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (63)

(64)

As (63) shows, RNR need not observe syntactic constituency, which supports the conclusion that this example of RNR is not derived by movement. Importantly, Hungarian only allows for RNR in which the pivot corresponds to non-focal material. Focal pivots that represent contrastive focus or the answer to a wh-question cannot be formed; consider (65) and (66): (65)

(66)

The fact that focal pivots are disallowed in Hungarian supports the generalization by Valmala (2013) according to which RNR with focal and RNR with non-focal pivots have distinct information-structural, prosodic, and syntactic properties (see also Hirsch and Wagner 2015). If, as Valmala claims, focal-pivot RNR is derived via rightward movement of the focal chunk in languages like English, where (65) and (66) are grammatical, the impossibility of this kind of RNR in Hungarian can be understood with reference to the fact that Hungarian lacks rightward movement processes for the expression of focus. (p. 838)

Page 25 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

31.5 Ellipsis in comparative clauses Hungarian comparative clauses may contain unpronounced elements. The most important distinction between comparative clauses in Hungarian and those in English in this domain is that a degree phrase does not have to be eliminated in comparative clauses via comparative deletion in Hungarian: such a phrase may be covert (67a), or overt (67b) (Kenesei 1992). When it is overt, the nominal or adjectival degree expression always contains an overt operator (a relative wh-operator), as shown in (67b). (67)

When overt, the degree expression must appear in the left periphery of the comparative clause, it cannot be left in situ (cf. (68)). This follows straightforwardly as Hungarian does not allow in situ relative pronouns, either. (68)

Bacskai-Atkari (2010, 2014) claims that the non-existence of covert degree operators in the left periphery in Hungarian and the lack of an obligatory process of comparative deletion are interrelated. She argues that comparative deletion only affects degree expressions in the left periphery, featuring covert operators, and thus it never applies to Hungarian as in this language the degree operator is overt. She also argues that cases where the entire degree constituent is missing (cf. (67a)) are derived by ordinary ellipsis processes that are operative in Hungarian clauses in general, such as predicate ellipsis, sluicing, or gapping (cf. (69) for the latter two cases). In these examples the degree phrase stays in situ in Bacskai-Atkari’s analysis, and gets eliminated due to it being inside an ellipsis site. The obligatoriness of ellipsis is supported by the observation that in cases where the degree phrase is covert, a given verb cannot be pronounced, as can be observed in (69a).13 (p. 839)

(69) Page 26 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

31.6 Null complement anaphora Hungarian null complement anaphora (NCA) are difficult to detect for the reason that many verbs that are known to allow NCA complements in other languages take complements in Hungarian that can be missing as a result of pro-drop. Verbs with nominative or accusative nominal complements or finite clausal complements associated with nominative and accusative sentential pronouns are such: their complements, which are headed or spelled out by definite pronouns can undergo the regular process of subject and object pro-drop (see Kenesei 1994). Null complement anaphora can be evidenced among verbs that normally select prepositional complements, however. These are verbs that take nominal complements marked with oblique case, or finite clause complements associated with oblique sentential pronouns. These types of complements cannot undergo pro-drop in the language, so the lack of an otherwise obligatory complement is telling in these cases. Verbs where such complements can be missing, and which therefore allow for NCA are e.g. vállalkozik ‘volunteer’ (with a sublative PP/finite clause complement), ajánlkozik ‘offer’ (with a sublative PP/finite clause complement), jelentkezik ‘sign up’ (with a sublative PP/finite clause complement), csodálkozik ‘wonder’ (with a superessive PP/finite clause complement), emlékezik ‘remember’ (with a sublative PP/finite clause complement), or egyetért ‘agree’ (with an instrumental PP complement). The following two examples illustrate the use of these verbs without overt complements. (70)

(71)

Page 27 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

Note that extraction cannot be used as a test for the null complement analysis in these cases, as extraction from the complements of these verbs is disallowed anyway. Extraction (p. 840) can be used as a test for predicates selecting infinitival complements, verbs such as akar ‘want’, mer ‘dare’, szeretne ‘like’, hajlandó ‘be willing’, etc. on the other hand. Interestingly, this test reveals that the complements of this class of verbs do allow for extraction, which in turn shows that these predicates do not take null anaphors as complements. (72)

31.7 Conclusions This chapter has illustrated the major types of elliptical constructions in Hungarian that have been identified and analyzed in some detail in previous works. It was shown that Hungarian has a wealth of elliptical constructions. Many of these constructions are derived by leftward movement of a syntactic constituent and ellipsis of (a subpart of) the domain out of which movement has taken place. The existence of sluicing inside relative clauses and preverb stranding furthermore points to the important role prosody plays in the formation of elliptical clauses.

Acknowledgements I thank the editors and three anonymous reviewers for their spot-on remarks on the previous version of this chapter. All mistakes and discrepancies are my own responsibility. This research was supported by NWO (Netherlands Organization for Scientific Research) in the form of the Vrije Competitie grant ‘Ellipsis licensing beyond syntax’.

Notes: (1) Non-standard abbreviations in this chapter are the following: AFF: affirmative particle; COMP: complementizer; COND: conditional, DE: reversing response particle; FUT: future auxiliary; HABIT: habitual auxiliary; NOM: nominalizer; PART: participial form; POT: Page 28 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian potential modality; PRT: discourse particle; PV: preverbal particle; REL: morpheme marking relative pronouns; VALO: copulative element linking arguments/adjuncts to derived nouns. Past tense and object definiteness agreement on the verb are not glossed throughout. Preverbal particles and verbs are spelled in two words, contrary to the Hungarian orthographic tradition. (2) Under either type of analysis, these facts have to be differentiated from nominalized adjectives (see Giannakidou and Stavrou 1999; Panagiotidis 2003b) such as gazdag-ok ‘rich-pl’ ‘the rich’, beteg-ek ‘sick-pl’ ‘the sick’, fiatal-ok ‘young-pl’ ‘the young’ for at least two reasons: (i) nominalized adjectives, unlike missing nouns like in (1) or (2), can be used without any (linguistic or non-linguistic) antecedent; (ii) nominalized adjectives have a generic [+human] interpretation, while the missing nouns in (1) and (2) are not restricted in this way. (3) In one theory of focus placement (Brody 1990), specFocP harbors a single focused or wh-moved constituent of the clause. The verbal head raises up to the Foc head stranding its verbal modifier and creating obligatory adjacency between the raising head and the focused item. The verb does not raise any further than Foc. In other words, there is no head movement to Dist or Top in Hungarian. (4) The precise geographical spread of this interspeaker variation is unknown. In a smallscale survey involving thirteen informants (dating back to 2012), the B speakers appeared to be from the Budapest area (see Lipták 2013b for further details). Note that dialectal variation along the same lines is also reported to exist between Capeverdean and Portuguese as described in Costa et al. (2012): Capeverdean can only resort to Vstranding in polarity contexts, while Portuguese can do so both in polarity and nonpolarity contexts. See Costa et al. (2012) for the facts and an account of the variation. (5) The descibed optionality in the size of V-stranding ellipsis in verbal complexes is only available under the so-called ‘straight order’ of verbal complexes (Koopman and Szabolcsi 2000; É. Kiss 2002). See an explanation of why it is missing in the ‘inverted’ orders in Lipták (2013b). Note also that the preverb appears left-adjacent to the finite auxiliary in these examples due to an ellipsis-independent process of what is referred to as preverb-climbing in the literature: a preverb belonging to an infinitive obligatorily appears before the finite auxiliary or semi-lexical verb that selects the infinitive. ((i))

Page 29 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (6) De is a sentential answer particle encoding the reverse function that indicates a relative switch of polarity (from negative to positive) with respect to an antecedent. See Farkas (2009) for this element in Hungarian. (7) What motivates the placement of the preverb to the clause-initial FP projection is far from clear. Since the preverb is not a morphosyntactic category that is related to either polarity or affirmation/answerhood, it is unlikely that this movement is driven by a morphosyntactic feature. Note that the movement is sensitive to prosodic structure, specifically to accenting, as preverb–verb combinations have pitch accent on the preverb. See Dvořák (2007) for the role that accenting plays in a similar ellipsis process in Slovenian. FP should be thought of as a position that hosts the sentential answer particles igen ‘yes’ and nem ‘no’ as well (possibly a polarity focus position of some sort). Independent justification for this comes from the similar distribution of igen and preverb-stranding in some contexts: e.g. neither a stranded preverb nor a single igen ‘yes’ can be used as a positive answer to an alternative question. As (i.B1) shows, V-stranding can be used in this context. ((i))

(8) The necessity of the inverse pairing relation is further indicated by the fact that examples like (i), where the post-auxiliary remnant does not repeat an already mentioned participant are judged as degraded by speakers. Some speakers report that they only accept such non-inverse examples if they understand the relation between the participants in the elliptical clause (i.e., me and Tamás), to be similar to the relation between the participants in the antecedent (i.e., me and Balázs). ((i))

(9) Note that the embedding verb tud ‘know’ used in (44) cannot introduce embedded fragments in other languages such as English, where only parenthetical embedded fragments are allowed. These can only feature verbs like think, expect, or hope (Morgan 1973; Temmerman 2013): ((i)) Page 30 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian

(10) The [E]-feature in (47) should be understood as the specific notation introduced by Merchant (2001: 55–61, 2004a: 670–3). Merchant argues that ellipsis in sluicing should be implemented by means of a syntactic [E]-feature, which resides on C0 and has all the relevant properties that distinguish elliptical structures from their non-elliptical counterparts, summarized in (i): ((i))

Van Craenenbroeck and Lipták (2006) argue that the syntax of [E] can vary across languages. (11) The relative pronouns in examples like (48) and (49) must carry stress (in line with Sprouse 2006; Sáez 2011). The latter is not possible for relative pronouns in non-elliptical clauses; compare the non-elliptical (i) and the elliptical (ii), where ▾ indicates lexical accent and 0 lack thereof. ((i))

((ii))

The stress pattern in (ii) is independently attested in Hungarian in environments not featuring clausal ellipsis: it is also found in relative clauses that contain no other overt material but the relative pronoun due to pro-drop and copula drop. This is possible in sentences that relativize an AP or NP predicate and where both the subject and the 3SG present tense copula are zero, such as (iii). ((iii))

Page 31 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Hungarian (12) See van Craenenbroeck and Lipták (2013) for the specific technical implementation: multiple focus movement does take place in narrow syntax but it has to spell out the lower copy of Beának in (54b). Note also that the movement of the verb from T0 to Foc0 (see n. 3) is bled by clausal ellipsis taking place in (54), similarly to (42), (43), (44). See Horvath (2005) and van Craenenbroeck and Lipták (2008) for details on bleeding. (13) Non-given verbs and verbs expressing a tense distinction with respect to the antecedent can survive the ellipsis. ((i))

Anikó Lipták

Anikó Lipták is an Assistant Professor at the Leiden University Centre for Linguistics (LUCL), Leiden University. Her main field of research is comparative syntax and Hungarian, and she has published extensively on elliptical phenomena. She is currently researching issues concerning ellipsis identity, the interaction between ellipsis and morphology, and between ellipsis and intonation.

Page 32 of 32

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

Oxford Handbooks Online Varieties of English   Gary Thoms The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics, Morphology and Syntax Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.013.31

Abstract and Keywords This chapter provides an overview of dialectal variation in English with respect to ellipsis phenomena. The discussion focuses on VP-ellipsis, describing dialectal variation in the range of licensors and in the retention of additional non-finite auxiliaries (e.g. John has been fired and Mary has (been) too). The range of variation attested, drawing upon data from British and North American varieties, provides an instructive test for theories of licensing, pushing in the direction of a theory which pays closer attention to the role of prosody in licensing ellipsis. There is also a discussion of British do, an ellipsis construction which seems to show an interaction between ellipsis licensing and extraction. Keywords: ellipsis, varieties of English, dialectal variation, VP-ellipsis, auxiliaries, British do, licensing, extraction, tag questions

Gary Thoms

38.1 Introduction IN this chapter I describe how certain ellipsis phenomena differ across varieties of English. The discussion here is mainly focused on VP-ellipsis1 because, insofar as I know, there is little to no description in the literature of dialectal variation in the domain of clausal ellipsis, nominal ellipsis, or other such elliptical constructions.

38.2 The licensing of VP-ellipsis To begin with, let us consider dialectal variation in which verbs may occur as the sole surviving verbal head in the context of VP-ellipsis. The retained verb is typically described as the “licensor” of VP-ellipsis (Lobeck 1995; Johnson 2001b), since it is clearly required for ellipsis to work,2 and the standard description of English VP-ellipsis is that it is only Page 1 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English possible if at least one auxiliary verb is retained as a licensor. This auxiliary may be be, have, a modal, and if none of these options is available, the dummy verb do must appear. (1)

Dialect data is useful for making this description more precise because there is substantial dialectal variation in English in the domain of auxiliary verbs. In what follows I describe a few different sources of variation and how they interact with VP-ellipsis licensing, and I discuss how this allows us to make our generalizations about licensors more precise. (p. 1021)

38.2.1 VP-ellipsis and auxiliary-like verbs First let us consider variation in whether or not a given verb shows the morphosyntactic behaviour of an auxiliary verb. One case of interest is ‘main verb’ have used to express possession: this may survive VP-ellipsis in at least British and Australian varieties of English, whereas this is impossible in all3 standard English varieties in the USA (thus requiring do-support). (I henceforth refer to these varieties as US English, USE.) (2)

The difference between the dialects shown by (2) is perhaps unsurprising, since it is known that have retains the last vestiges of auxiliary verb syntax in the British and Australian ones but behaves fully like a lexical verb in the American ones. The auxiliarylike status of have in these dialects is indicated by cliticization to the subject, raising to C in questions and raising past negation. These tests do not all pattern together uniformly though, as many standard British varieties (i.e. from southeast England) only allow (3a), in others inversion is possible in questions but not with negation, and in some varieties (northern England) raising past negation in declaratives is either impossible or restricted only to cases where the possessum is indefinite (dialects in the southwest of Scotland). Broadly, the tendency in the UK seems to be that have tends to retain more raising properties the further north you go, and the dialects which allow raising past negation also allow raising to C (but not vice versa). All these dialects contrast with North American dialects, in which all of (3) are unacceptable (just like with (2)). (3)

Page 2 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

The correlation between auxiliary-like behaviour and the ability to survive ellipsis can also be observed with other uses of have. Consumption have (as in have steak) shows a (p. 1022) much more lexical verb-like syntax in most British varieties, failing all the raising tests, and it also fails to license ellipsis (much like possessive have for American dialects). The same pattern shows up when we consider other main verb uses of have, such as its causative use. (4)

(5)

The same correlation between raising and ellipsis even seems to be attested across different instantiations of possessive have in single dialects. For many speakers of have-raising dialects, raising is much better with present tense forms of have; for instance, in my own dialect the examples in (6) are degraded, even though (3a)–(3b) are wholly acceptable.4 Quite why this would be is not clear, but what’s of interest is that these judgments are again tracked by ellipsis judgments, in that possessive have is not a good VP-ellipsis licensor in its past form. (6)

(7)

This indicates that it is not some idiosyncratic property of British dialects that they allow all uses of have to be conflated with auxiliary have; rather, it is the syntax of the verb which determines whether or not a given have may survive ellipsis, and it seems there is a correlation between whether a verb displays auxiliary-like properties and whether it can survive ellipsis.

Page 3 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English Interestingly, it is not the case that auxiliary-like verbs always survive as ellipsis remnants. First, consider the need passive, a non-standard passive construction which is available in various East Coast American dialects5 and in all Scottish English dialects (see Edelstein 2014 and references cited therein). In this construction, a passive VP is embedded under the verb need, which functions as a passive auxiliary of sorts; however need does not show auxiliary-like properties like verb-raising to T or precluding do-support. (9) shows that need does not license ellipsis (Thoms and Walkden 2015), thus indicating that not all auxiliary-like verbal elements license ellipsis. (p. 1023)

(8)

(9)

The same thing is seen with habitual be in African American English (AAE).6 In this construction, an uninflected form of be encodes a habitual interpretation of the predicate it takes as its complement (see e.g. Green 2000, 2002). In terms of our diagnostics, habitual be does not behave like an auxiliary verb, in that it does not raise and it triggers do-support in various raising contexts (Green 2000). Like passive need, habitual be is also unable to license ellipsis (Green 2002).7 (10)

(11)

(12)

Page 4 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

These facts allow us to narrow our VP-ellipsis generalization further: only those auxiliaries which can occur in T can serve as licensors.

38.2.2 Null verbs Another aspect of VP-ellipsis which emerges from looking at dialect data is that the surviving auxiliary-like element must be overt. For instance, finite non-past instances of auxiliary and copular be may optionally be null in African American English (AAE), as (13a) shows. Conner (2014) observes (following up on observations in Labov 1969) that this optionality disappears when the auxiliary is the licensor of VP-ellipsis. (p. 1024)

(13)

One further relevant observation to add is that VP-ellipsis is possible with the null auxiliary construction in AAE when an uncontracted not is present at the ellipsis site. (14)

8

One might speculate that the same effect is seen in infinitives in standard English. In a discussion of VP-ellipsis in infinitives Johnson (2001b) notes that while VP-ellipsis is impossible in all of the environments in (15) (data from Zwicky 1982 and Zagona 1988a,b), in all cases VP-ellipsis is ameliorated by inserting not to the left of the non-finite auxiliary to. I add here the observation that the same effect is not seen when negation follows to. (15)

(16)

Page 5 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (17)

Zwicky (1982) and Zagona (1988a,b) argue that the problem with (15) is that non-finite to is a clitic which must attach to a host to its left, and they submit that this cliticization process is impossible when to is separated from its potential host by an island boundary. Assuming this is correct, the effect of adding negation in (16) and the absence of any such effect in (17) would then follow if it were the case that a preceding not may provide a host for encliticization of to. In both (14) and (16), then, licensing becomes possible when some additional prosodic constraint on the licensing element is satisfied. Similar effects can be seen in English varieties in some parts of Scotland. Smith (2000) reports that Buckie English optionally allows for null forms of finite non-past do and auxiliary have when the inflection would be zero (e.g. first person), so long as they cooccur (p. 1025) with negation, as shown in (18).9 The elicited data in (19) shows that VPellipsis is unacceptable when the null auxiliary is the licensor. (18)

(19)

One point to note regarding the Buckie case is that the negation which occurs with the null auxiliary is the contracted form -na (-nae in western Scottish varieties), which is similar but not equivalent to StE -n’t (see Smith 2000, Weir 2007, and Thoms et al. 2013 for discussion). This contracted negation forms a phonological word with the preceding subject, which is typically pronominal, so it is likely that ellipsis in this context is not possible for the same reason that contracted auxiliaries may not occur adjacent to ellipsis sites or gap sites. (20)

Page 6 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (21)

Going further, it could be the case that the overtness requirement for ellipsis licensors is due to some general prosodic property of structures of this kind too, whereby auxiliary verbs at the edge of a gap must be stressed and hence cannot be null or reduced. Interestingly, there is another corner of Scots dialects which indicates that this might not be adequate either. Thoms et al. (2016) note that in certain Scots dialects from the central belt of Scotland, contraction is possible in constructions like the following, which are typically used in contexts where the speaker has discovered something. (22)

Speakers who accept and produce examples like these nevertheless reject (20)–(21) just like other English speakers. What these examples suggest is that a fully prosodic account of auxiliary contraction is not viable, and that there must be some irreducible syntactic component to the contraction restriction. A complete analysis of these facts should prove instructive when it comes to assessing the extent to which restrictions on ellipsis licensing can be reduced to prosodic constraints. (p. 1026)

38.3 VP-ellipsis and non-finite auxiliaries 38.3.1 Optionality in ellipsis of non-finite auxiliaries The preceding discussion of ellipsis licensors focuses on simple cases where there is just one auxiliary. However, an interesting property of English VP-ellipsis is that there is often optionality in the size of the ellipsis site in clauses with multiple verbal heads, with some heads having the option of either being included or excluded in the ellipsis site. This is most widely available with the non-finite form of auxiliary be. In (23)–(24), we see that non-finite be may optionally be included in the ellipsis site when the auxiliary immediately above is a finite modal or finite have. In (25), we see that the same optionality is attested when be occurs under non-finite have in clauses with three auxiliaries. (23)

(24)

Page 7 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (25)

The same pattern is observed with be used as the copula, even though this is often analysed as being the ‘main verb’ of copular clauses. (26)

However this optionality is not always observed with all auxiliaries, and much discussion of the nature of VP-ellipsis licensing has focused on cases where this optionality fails (e.g. Akmajian and Wasow 1975; Sag 1976a; Thoms 2011b; Sailor 2012a, 2014; Aelbrecht and Harwood 2015). One such case is the failure of VP-ellipsis licensing when the auxiliary adjacent to the ellipsis site is in the -ing form (Sag 1976a; Johnson 2001b); I’ll call this the -ing constraint. (27)

We see the opposite kind of failure of optionality with non-finite forms of have. As noted by Sailor (2012a, 2014), Harwood (2013), and Aelbrecht and Harwood (2015), even though non-finite be may be optionally elided when it follows a modal auxiliary, the same does not hold with non-finite have: for many speakers, ellipsis of non-finite have is completely impossible. (p. 1027)

(28)

These patterns seem to show that it is not just the feature specification of T which is relevant to whether or not ellipsis is licensed, but rather non-finite auxiliaries have a role to play as well.

Page 8 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English These facts are relevant in the present context because there is substantial dialectal variation with respect to the ability of non-finite auxiliaries to be retained in ellipsis contexts. We turn to this next.

38.3.2 Non-finite auxiliaries and raising An important fact about VP-ellipsis licensing in English is that, for the most part, a verbal element’s ability to function as a licensor of ellipsis in its finite form tracks its ability to survive ellipsis as an additional non-finite auxiliary in multiple auxiliary constructions of the kind seen in section 38.3.1. Thus the patterns of successful and failed VP-ellipsis licensing demonstrated in section 38.2 can be recreated with their non-finite counterparts. The dialectal variation seen with possessive have can be replicated for cases where it’s non-finite, while non-finite versions of the other main verb haves still fail to survive ellipsis. (29)

(30)

Habitual be and the need of the need-passive fail to be retained in VP-ellipsis contexts even when there’s a licensor in T: (31)

(32)

The generalization, then, is that only verbs which show the morphosyntactic properties of auxiliary verbs may survive ellipsis, be it the licensor or a non-finite hangeron. (p. 1028)

Page 9 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

38.3.3 Have-deletion Let’s go back now to the have and -ing restrictions mentioned at the outset of section 38.3.1. First, consider have: while it is reported by Sailor (2012a) and Harwood (2013) that have-deletion is unacceptable, the data in (28) are somewhat controversial, since many speakers accept such examples readily. Examples where have is deleted are given by authors from various different dialect regions: see Lasnik (1995a) for USE, Thoms (2011b) for a Scottish variety,10 and Aelbrecht and Harwood (2015) report that a significant proportion of their northern England-based informants accept these examples. It seems likely, then, that this is a matter of interspeaker variation, rather than regional variation. Harwood (2013: 119) (picking up on suggestions in Kayne 1997) offers an intriguing speculation regarding this variability. He proposes that in general have cannot be deleted by VP-ellipsis and that when have does go missing this is due to some independent rule which allows modals to embed a participial phrase with a perfect interpretation but without the perfect head have; in other words, there is an independent means for removing have for these speakers. As Harwood notes, such rules seem to be available for some Scandinavian languages, and his speculation is lent further plausibility by the fact that have-deletion is possible in a number of environments in some Scottish dialects. For instance, Macafee (1980) has noted naturally occurring examples like (33) in Scottish varieties, and fieldwork in the SCOSYA project has found that such examples are rated as acceptable quite widely in Scotland (see also the discussion of Smith 2000 in 38.2.2). (33)

What seems to be crucial for such structures is the presence of some other non-verbal element in the area which would normally host have (or indeed the phonologically reduced enclitic schwa which it is often realized as), so it is viable that this is a phonologically conditioned deletion rule which may also apply adjacent to an ellipsis site. This kind of analysis also makes it easier to understand the variability of the data, since a phonological rule for deleting have (or indeed its highly reduced schwa form) seems more likely to be acquired variably by speakers with broadly similar grammars than a syntactic rule which allows for ellipsis of different sizes of constituent in a limited set of syntactic contexts (i.e. those where have is dominated by a modal). Indeed it is relevant that very few other ellipsis-related rules which are normally analysed syntactically are as subject to ‘in-the-room variation’11 across (p. 1029) varieties as this one; for instance, I have never encountered any such variability in accepting modals as licensors of VP-ellipsis, nor have I encountered variability with British speakers with respect to judgments like (2) for ellipsis licensing with possessive have. This kind of argument is somewhat indirect, and unlikely to prove convincing in the absence of further supporting evidence, but nevertheless it shows another way in which microcomparative evidence may be brought to bear on the analysis of ellipsis and other such grammatical phenomena. A full Page 10 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English assessment of Harwood’s have-deletion analysis would require us to give greater consideration to the other factors which condition its availability. For instance, have-deletion is typically only acceptable when the preceding modal auxiliary is might or should, and examples with must are typically rejected even by the most liberal speakers.

38.3.4 The anti -ing constraint Now consider the anti -ing constraint, which was identified by Sailor (2012a) and Harwood (2013) as a strong constraint which is not subject to the same kind of variation which we see with have. It turns out that there are in fact very many English speakers who allow VP-ellipsis in the presence of non-finite being, as the following examples from Google searches indicate: (34)

12 13 14

,

,

Although NAmE speakers consistently reject these kinds of examples,15 I have found that they are readily accepted by British English speakers from all around the country, although not all speakers accept them. In general this seems to only occur when there is no progressive auxiliary in the antecedent, and it may be relevant that many speakers reject (p. 1030) ellipsis of -ing forms when there is no progressive aspect in the antecedent, as Lasnik (1995a) reports that examples like the following are unacceptable:16 (35)

It is possible, then, that the pressure to retain being to avoid violation of the constraint exemplified by (35) may thus force being to survive ellipsis exceptionally in cases like (34). This option also seems to be conditioned by prosody to some extent, as those speakers who do accept examples like (34) acknowledge that this requires stressing the preceding finite auxiliary. Clearly, the anti -ing condition is not a hard constraint on English VP-ellipsis. But if examples like (34) are exceptional, in that they are wholly unacceptable for many speakers, they should prove useful in identifying the general rule for retaining additional auxiliaries, and perhaps for VP-ellipsis licensing generally.

Page 11 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

38.3.5 Additional auxiliaries in tag questions and retorts Most of the variation discussed so far has concerned differences between British and North American dialects, with USE and Canadian English patterning together against the British varieties. However there is at least one case that I know of where the judgments of Canadian English speakers pattern with British dialects and contrast with those of USE speakers, in Sailor’s (2014) discussion of the retention of additional auxiliaries in instances of ‘high VPE’. The starting point for this is the observation that voice mismatches under VP-ellipsis are typically only acceptable when the elided VP is in a subordinate clause, for instance an adverbial; in most other configurations, such as coordinate structures, emphatic retorts, or tag questions, voice mismatches are impossible. (36)

(37)

Sailor shows that the latter class of VP-ellipsis environments behave differently from subordinate clause structures with respect to a number of diagnostics, and he explains these asymmetries in terms of the size of the antecedent for ellipsis: with subordinate clauses the antecedent may be a small chunk of the lower part of the inflectional domain, one that excludes VoiceP, whereas with the other environments the antecedent must be a higher constituent which includes VoiceP. (p. 1031)

Sailor dubs the environments in (37) ‘high VPE environments’ and the point of relevance here is that he describes intriguing dialectal variation with respect to the retention of additional auxiliaries in high VPE contexts. In outline, the dialectal difference is that in USE, VP-ellipsis in clauses with multiple auxiliaries strongly favours retention of all additional auxiliaries, whereas in Canadian, British, and Australian Englishes (clubbed together as ‘Commonwealth English’, CE), ellipsis of as many auxiliaries as possible is favoured. This manifests itself in some quite sharp differences in judgments when it comes to clauses with several auxiliaries. First, consider cases where been can be elided: USE speakers find retaining all auxiliaries (other than being) acceptable, whereas for CE speakers this is highly marked or worse; rather, the preference is to delete at least been. Sailor notes that this difference is minimal in the case of coordination, although for some CE speakers there is still a markedness to retaining been in this case. Page 12 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (38)

(39)

(40)

(41)

The following examples show the other side of this variation: in the same high VPE environments, eliding a much larger chunk of the clause, including non-finite have, is typically acceptable for CE speakers but reported as unacceptable for USE speakers. Note that for coordination Sailor uses an idiom which requires a perfect auxiliary in order to ensure the ellipsis site contains have, although as before the contrasts here are still quite slight. (42)

(43)

(44)

(45)

The challenge for making sense of this striking observation is to pin down other dialectal properties which unite CE varieties as distinct from USE and which might therefore be used to explain the variation we see. (p. 1032)

Page 13 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

38.3.6 Interim summary What we’ve seen in this section is that there is variation across dialects with respect to the optionality of retaining additional non-finite auxiliaries in VP-ellipsis contexts. Some of these conditions recall those discussed in the previous section on licensors and keyed to the morphosyntax of auxiliary verbs. But others, such as the –ing restriction, seem distinct and more likely to be the result of interface conditions (at least in part), and so it’s not yet clear whether a unified account is viable. If anything, a unified account is likely to lead us in the direction of reassessing whether the more syntactic conditions discussed in section 38.2 can be reduced to interface conditions, with the right model of the syntax–phonology interface, as reducing the dialectal variation seen with -ing to syntactic conditions seems unviable.

38.4 British do ‘British do’ is the phenomenon whereby a superfluous non-finite form of do occurs at the right edge of a cluster of auxiliaries in VP-ellipsis contexts bearing the morphological ending that would normally occur on the finite verb (Haddican 2007; Baltin 2006, 2012; Aelbrecht 2010; Thoms 2011a).17 The superfluous do typically occurs after modals and have, but it may also be retained after progressive be in the same circumstances where being is retained in examples like (34) (with similar variability in judgments18). It cannot occur alongside passive be.19 (46)

(47)

British do has attracted a reasonable amount of attention in the literature in recent years, and one of the major issues is whether or not it is best analysed as an elliptical structure or a verbal proform. This discussion has mainly focused on some curious extraction asymmetries between British do environments and their more standard VP-ellipsis counterparts. (p. 1033)

British do is analysed as a proform akin to do so by Haddican (2007), and it is easy to see why this is appealing: on the surface, the only difference between the two is the so, and these putative proforms may not co-occur with the VPs which they seem to be replacing. (48)

Page 14 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

Haddican strengthens the proform analysis on the basis of intriguing facts from extraction. Building on Baltin (2004), Haddican reports that various A′-extractions— wh-extraction, QR of objects for scope inversion with subjects, comparative extraction, topicalization, and pseudogapping—are all incompatible with British do; in this respect it patterns with do so rather than VP-ellipsis, which is compatible with all of these extractions. (49)

(50)

(51)

(52)

(53)

Noting that do so is also incompatible with passivization, Haddican therefore arrives at the generalization that both British do and do so disallow extraction, and he proposes that this follows from an analysis where they are both verbal proforms with no internal content. The argument from extraction for the proform analysis has since been contested by Baltin (2007, 2012), Aelbrecht (2010), Thoms (2011a), and Abels (2012), who point out that (p. 1034) it is not quite right that all extractions from British do are impossible. First, Baltin notes that A-extraction is possible with unaccusative and raising predicates, Page 15 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English seeming to pattern against do so which forbids these.20 Second, Thoms shows that while QR for scope inversion of subjects and objects is not possible, QR to scope an object over negation is as acceptable with do as it is with VP-ellipsis, but impossible with do so.21 Third, Abels (2012) reports that relativization and topicalization are both possible with British do for many speakers, the latter somewhat marginally. No such leeway is found with do so. (54)

(55)

(56)

(57)

(58)

Finally,22 Baltin (2012) notes a set of facts which seem to point to the conclusion that it might not even be right to say that do so does not involve ellipsis, undermining a key premise of Haddican’s argument. Baltin notes that while only VP-ellipsis (of the three constructions) allows scope inversion of subjects and objects when this would be derived by QR (i.e. where the object is a strong quantifier), scope inversion is possible when the object is an indefinite which would be able to scope out of the putative ellipsis site by insitu scoping mechanisms; to this, we can add the observation that inverse scope is also possible with British do in these configurations as well. Crucially, this is not possible in cases with do it. (The following examples are adapted versions of Baltin’s, which he attributes to an anonymous reviewer.) (p. 1035)

(59) Page 16 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

This difference between do it and the other constructions indicates that the former is the only true verbal proform, with the other constructions being derived by ellipsis, as it is hard to see why do so and do it would differ in their ability to take VPs containing specific indefinites as their antecedents. This should be a welcome conclusion, as we know from Hankamer and Sag (1976) that do it behaves differently from do so and other elliptical constructions in other respects; that is, it seems to be deep anaphor, since it does not require a linguistic antecedent (see also Bentzen et al. 2013). Accepting this conclusion, the desiderata for an ellipsis analysis of do so is to account for the fact that it systematically blocks A′-extraction but not A-extraction, and that it requires a non-stative volitional antecedent. It seems likely that the correct analysis will be one where the so part of do so is some kind of focus-related operator which occupies a peripheral position in the vP and acts as an A′-intervener, but the matter warrants more attention than I can grant it here. The challenge with British do, then, is to work out an analysis which will derive the extraction restrictions, the ellipsis-like licensing conditions, and of course the presence of the superfluous do just in ellipsis contexts, and it needs to relate to some minor parameter which differs between British dialects and others. No existing account is (p. 1036)

without its problems. Baltin (2007, 2012) and Aelbrecht (2010) fail to predict that some A′-extractions are possible, and they both resort to stipulation to derive the fact that do occurs only in ellipsis contexts: for both, do is a realization of v which is stipulated as only being possible in the context of ellipsis. Thoms (2011a) fares better on the latter point: he also assumes that the do is a spell-out of v, but he states a rule for this realization which generalizes to standard do-support, and the Britishness is somewhat loosely tied to the general British propensity for retaining additional auxiliaries. Thoms does not offer a fully worked-out account of the extraction pattern, although he does succeed in pinning down the concrete generalization for when extraction is possible. Thoms begins with an apparent contradiction with respect to QR: while (50) seems to show that British do prohibits reconstruction, (56) shows that QR can sometimes escape this environment. He argues that this contradiction goes away if the effect in (50) is not due to the inability of the object to QR, but rather the subject’s inability to reconstruct back into its base position in the vP. This would follow since it is known independently that scope inversion of the subject and the object requires subject reconstruction as well as QR (Hornstein 1995b; Johnson and Tomioka 1998; Fox 2000; Nevins and Anand 2003). Thoms shows that this analysis is supported by the fact that in cases where subject reconstruction would invert the scope of the subject and some other non-argument, the scope inversion is possible with VP-ellipsis and even with do so, but it is substantially degraded with British do.

Page 17 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (60)

Thoms thus argues that in those cases where extraction from British do fails, what is going wrong is reconstruction of the extracted constituent, and he argues that this accounts for the distribution of extraction possibilities: for instance, wh-movement from British do sites is strictly prohibited since wh-movement shows obligatory reconstruction (Chomsky 1995b), whereas relativization and topicalization do not show obligatory reconstruction effects in many situations and may be analysed as being derived as null operator constructions (Lasnik and Stowell 1991). Abels (2012) shows that this account makes the right predictions when we come to consider relativization a bit more closely: when we look at structures which would require obligatory reconstruction, British do becomes much worse than it is with regular relativization. This is demonstrated for amount relatives, free relatives, and relatives headed by idioms, all of which are known to require reconstruction of the relativized category (Bianchi 2004). (61)

23

(62)

(63)

Since the operator movement analysis is not readily available for these relative types, they are derived by raising and hence require reconstruction, in violation of the restriction against reconstructing into British do. (p. 1037)

As for comparatives, subsequent work has since revealed a number of exceptions to the claim that comparative extraction from do is prohibited:24 (64) Page 18 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

It seems the main difference between these examples and Haddican’s (52b) is that in the former the subjects of the comparative are pronouns, whereas in (52b) it is a focused full DP. Haddican discusses a number of other facts about British do which indicate that its distribution is sensitive to prosodic properties of its host; for instance, it is impossible when the finite auxiliary is contracted. (65)

The generalization seems to be that the finite verb must be sufficiently strong in its prosodic form to host the clitic do, so one possible analysis for (52b) is that placing strong accent on the embedded subject for focus reasons precludes giving strong enough accent to the finite auxiliary in order for it to host do (see also n. 22). Putting this to one side, the acceptability of (64) requires an explanation. One possibility, explored in more detail in Thoms and Sailor (2017), is to unify comparatives with topicalization and matching relatives and treat them all as involving operator movement (Chomsky 1977a; see Kennedy 2000 for an operator movement analysis of comparatives). Thoms’ reconstruction-based generalization seems to stand up well, then. This is still far from a satisfying result, though, since it is hard to see what would actually explain a generalization of this kind. This is the heart of the matter with British do: on the one hand, all that we seem to have is a superfluous little do which is the reflex of some prosodic (p. 1038) property of British dialects to retain additional non-finite auxiliaries, and yet there seems to be a hard syntactic restriction at play, one which crucially implicates an LF-phenomenon which is not expected to interact with prosody normally.

38.5 Conclusion In this chapter I have reviewed several cases of dialectal variation with respect to VPellipsis phenomena in English and shown that they allow us to refine generalizations about the licensing conditions on ellipsis and other such matters. The description here is only scratching the surface, and is based on a very limited understanding of the extant variation, but hopefully the recent upsurge in microcomparative syntax research will continue apace and bring with it further advances in our understanding of ellipsis and the language faculty more generally.

Page 19 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

Notes: (1) Throughout I use the term ‘VP-ellipsis’ following tradition, with no strong commitment to whether or not each case involves deletion of a node identifiable as a VP. Miller and Pullum (2014) argue that Sag’s (1976a) term ‘post-auxiliary ellipsis’ is more accurate, but it is not clear that that would apply to all the cases discussed here. (2) For different theories of ellipsis licensing, see Lobeck (1995), Merchant (2001), Aelbrecht (2009), and Thoms (2011b), among many others. (3) A note on the contextual restriction of quantification: throughout this chapter I will often make claims about “all dialects” of some kind or another having one characteristic or another. Obviously this is overstating things, as there is still a lot that is not known about just how much some of the phenomena in question vary within dialect communities: for instance, I know little about whether or not have-raising is possible in the southern American English dialects which share a number of relic features with the British dialects which they are closely related to historically. Thus throughout, the set of dialects in question should be understood as those which I have been able to access data on directly or indirectly. (4) Preliminary results in fieldwork for the Scots Syntax Atlas indicate that this is a general pattern, in that in all eighty-three locations sampled at the time of writing the present tense forms of have are rated higher than the past tense forms. (5) For details see the Yale Grammatical Diversity Project’s page on this phenomenon: , accessed on 17 June 2018. (6) Thanks to Tracy Conner for bringing these facts to my attention. (7) A similar pattern holds for the aspect marker BIN, as discussed in Green (2002). (8) This example is taken from the song ‘Errbody’ by Yo Gotti. Tracy Conner (p.c.) has confirmed for me that it is grammatical in AAE. (9) These are based upon naturally occurring examples. Further examples of these have been found in data from other communities in the northeast of Scotland as well. (10) Sailor introduces a number of controls for ensuring that what is elided in these cases really is a larger constituent rather than some smaller VP, as he claims that ambiguities muddy the waters somewhat. My own judgment for a Scottish variety is that have-ellipsis is still possible when these controls are put in place. (11) This term is due to Jim Wood. Wood identifies transitive control promise as an example of a grammatical phenomenon in English which is subject to unpredictable inthe-room variation. (12) , accessed on 19 June 2018. Page 20 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (13) , accessed on 14 March 2015. (14) , accessed on 14 March 2015. (15) One point of potential relevance is that examples of VP-preposing which strand auxiliary being are reported as only marginal by Kyle Johnson in his survey article (Johnson 2001b), even though he finds basic VP-ellipsis examples like (27) unacceptable. ((i))

Given that the licensing conditions on VP-fronting track those on VP-ellipsis consistently otherwise, this could be taken as a further indication that some NAmE speakers may marginally allow licensing with retention of being in the right discourse conditions. (16) Again this judgment seems to be subject to some variation, with some speakers accepting other examples with similar configurations. Bronwyn Bjorkman (p.c.) reports that the following example (attributed to David Pesetsky) is relatively acceptable for some speakers: ((i))

And while examples like (i) are not always perfect for all speakers, Rooryck and Schoorlemmer have noted that there is even less objection to examples like (ii), where the antecedent is an infinitive. Rooryck and Schoorlemmer argue that the important difference between the good cases and the bad cases is that the bad ones involve mismatches in Mood specification, but this would leave examples like (ii) unaccounted for. ((ii))

(17) Following others, I generalize this do to all British dialects, as I have not identified any specific syntactic differences between dialects with respect to the properties described below. It is possible that do is used more commonly in southeast dialects, but this needs to be investigated more thoroughly. Note also that ‘British dialects’ is (innocently) taken to include all dialects from the British Isles, including those in Ireland, and as with a lot of British dialect phenomena it is also found in Australian English dialects.

Page 21 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (18) My impression is that retention of doing is particularly common in southeastern dialects, although Craig Sailor (p.c.) reports that it is also common in the speech of Yorkshire-born Top Gear presenter Jeremy Clarkson. (19) Thoms and Sailor (2017) point out that this is not some incompatibility with passivization as such, since British do occurs in other passive-like structures such as the Scottish need-passive discussed in section 38.2.1 and in get-passives: ((i))

(20) Thoms notes (citing Hallman 2004 and p.c. from Carson Schütze) that it is not quite right to say that do so cannot appear with raising predicates, as the effect in examples like (55b) is to be attributed to a general restriction which requires that do so must appear with non-stative volitional predicates. When this is controlled for, do so may take an unaccusative as its antecedent. ((i))

Similar comments may apply with passivization, as Bruening (2015) notes that many speakers seem to accept passives with do so as well, citing numerous examples like (ii) from Google searches (most of which are perfectly acceptable to me). ((ii))

This may provide a further argument for providing do so with an ellipsis analysis, as suggested below. (21) Baltin (2012) provides the following example as evidence against QR of objects out of British do sites and over negation. ((i))

The contrast with VP-ellipsis is far from clear for me though, and it seems likely that the degradation of this example might be independent of scope, since couldn’t do is somewhat marked in general (for reasons that are not clear).

Page 22 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English (22) Thoms also notes a set of contrasts from Condition B (building on Kennedy 2003), where there is an apparent Condition B effect with VP-ellipsis and British do but not with do so. ((i))

These contrasts are subtle and subject to some variation, but the majority of speakers I have consulted share them. However, it strikes me as plausible that the contrast here is not due to some fundamental difference in the structure of the putative ellipsis sites, but rather to differences in how the three structures may realize focus on the subject: it is much more natural to put emphatic stress on the subject in the context of do so than with British do, and with VP-ellipsis focusing the subject seems to require emphatically focusing the following negation as well, whereas this isn’t the case with do and do so. There may also be interactions between focus and the height of attachment of the reason adjunct which bring antecedent containment and the problems that that poses into the picture as well. And as with many of the other phenomena discussed here, the variability of the data is highly relevant and so one might hope that a full account of these contrasts will provide the right kind of wriggle room to allow for such variation. (23) While Abels marks this as only somewhat degraded, I and my informants find it completely impossible. (24) The do in (64a) is allowed not just for British speakers, but for some North Americans as well. Tim Stowell (p.c. to Craig Sailor) reports that he finds such examples acceptable, and has found agreement among other older Canadian English speakers, although for the most part USE speakers reject (64a). It would be intriguing to explore whether this is related to the only other Canadian English/USE dialect split discussed in this chapter, namely the variation with respect to retorts and tags discussed in section 38.3.5.

Gary Thoms

Gary Thoms is Assistant Professor of Linguistics at New York University. His main research interests include ellipsis, reconstruction, dialectal and intraspeaker variation in English, Celtic syntax, polarity phenomena, predicate fronting, and the language of poetry.

Page 23 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

Varieties of English

Page 24 of 24

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). Subscriber: University of Glasgow; date: 13 February 2019

References

References   The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019 DOI: 10.1093/oxfordhb/9780198712398.004.0001

(p. 1039)

References

Aarons, Debra (1996). Topic and topicalization in American Sign Language, Stellenbosch Papers in Linguistics 30: 65–106. Abbott, Barbara (1976). Right node raising as a test for constituency, Linguistic Inquiry 7(4): 639–42. Abdulkarim, Lamya and Tom Roeper (1997). Economy of representation: Ellipsis and NP reconstructions, in Antonella Sorace, Caroline Heycock, and Richard Shillcock (eds), Pro­ ceedings of GALA 97 Conference on Language Acquisition. Edinburgh: University of Edin­ burgh. Abe, Jun (2014). Antecedent-contained deletion in Japanese: Support for the VP-ellipsis analysis. Ms. Tohoku Gakuin University. Abe, Jun (2015). The in-situ approach to sluicing. Amsterdam and Philadelphia: John Ben­ jamins. Abe, Jun (2016). Make short answers shorter: Support for the in-situ approach, Syntax 19(3): 223–55. Abe, Jun and Norbert Hornstein (2012). ‘Lasnik-effects’ and string-vacuous ATB move­ ment, in Myriam Uribe-Etxebarria and Vidal Valmala (eds), Ways of structure building. Oxford: Oxford University Press, 169–205. Abe, Jun and Hiroto Hoshi (1997). Gapping and P-stranding, Journal of East Asian Linguis­ tics 6: 101–36. Abe, Jun and Chizuru Nakao (2012). String-vacuity under Japanese right node raising, in Matthew A. Tucker, Anie Thompson, Oliver Northrup, and Ryan Bennett (eds), Proceed­ ings of FAJL 5. MIT Working Papers in Linguistics 64. Cambridge, MA: MITPWL, 1–14. Abe, Jun and Christopher Tancredi (2013). Non-constituent deaccenting and deletion: A phase-based approach. Ms. Tohoku Gakuin University and Keio University.

Page 1 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Abeillé, Anne (2005). Les syntagmes conjoints et leurs fonctions syntaxiques, Langages 160: 42–67. Abeillé, Anne (2006a). In defense of lexical coordination, in Olivier Bonami and Patricia Cabredo Hofherr (eds), Empirical issues in formal syntax and semantics. Paris: CNRS, 7– 36. Abeillé, Anne (2006b). Stripping constructions in French. Paper presented at the Coordi­ nation and Ellipsis Conference. Université Paris 7, 26 June. Abeillé, Anne, Gabriela Bîlbîie, and François Mouret (2014). A Romance perspective on gapping constructions, in Hans Boas and Francisco Gonzálvez Garcia (eds), Romance per­ spectives on Construction Grammar. Amsterdam and Philadelphia: John Benjamins, 227– 67. Abeillé, Anne, Berthold Crysmann, and Aoi Shiraishi (2015). Syntactic mismatches in French peripheral ellipsis. Presentation at CSSP (Colloque de Syntaxe et Sémantique à Paris), Université Paris 7. Abeillé, Anne and Danièle Godard (2000). Varieties of ‘esse’ in Romance languages, in Dan Flickinger and Andreas Kathol (eds), Proceedings of the 7th International HPSG Con­ ference. Stanford: CSLI Publications, 2–22. Abeillé, Anne and Danièle Godard (2003). The syntactic flexibility of French degree ad­ verbs, in Stefan Müller (ed.), Proceedings of the 10th International HPSG Conference. Stanford: CSLI Publications, 26–46. Abels, Klaus (2003). Successive cyclicity, anti-locality, and adposition stranding. PhD the­ sis, University of Connecticut at Storrs. Abels, Klaus (2004). Right Node Raising: Ellipsis or across the board movement? in Keir Moulton and Matthew Wolf (eds), NELS 34: Proceedings of the 34th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, 45–60. Abels, Klaus (2009). Don’t repair that island! It ain’t broke. Paper presented at the University of the Basque Country, Vitoria/Gasteiz. (p. 1040)

Abels, Klaus (2011). Don’t fix that island! It ain’t broke. Talk at the Islands in Contempo­ rary Syntactic Theory workshop, University of the Basque Country. Abels, Klaus (2012). Phases: An essay on cyclicity in syntax. Berlin: Mouton de Gruyter. Abels, Klaus (2016a). Displacement in syntax, in Mark Aronoff (ed.), Oxford research en­ cyclopedia of linguistics. New York: Oxford University Press. DOI: 10.1093/acrefore/ 9780199384655.013.313.

Page 2 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Abels, Klaus (2016b). On the interaction of P-stranding and sluicing in Bulgarian. Ms. UCL. Abels, Klaus and Veneeta Dayal (2016). On the syntax of multiple sluicing. Talk presented at the 47th annual meeting of North East Linguistic Society (NELS 47), UMass, Amherst, MA. Abels, Klaus and Gary Thoms (2014). Resumption and island repair under sluicing. Talk presented at the LAGB (Linguistics Association of Great Britain) Annual Meeting, Oxford. Abney, Steven (1987). The English noun phrase in its sentential aspect. PhD thesis, Mass­ achusetts Institute of Technology. Abrusán, Márta (2014). Weak island semantics. Oxford: Oxford University Press. Ackema, Peter and Kriszta Szendrői (2002). Determiner sharing as an instance of depen­ dent ellipsis, Journal of Comparative Germanic Linguistics 5(1): 3–34. Adams, Perng Wang and Satoshi Tomioka (2012). Sluicing in Mandarin Chinese: An in­ stance of pseudo-sluicing, in Jason Merchant and Andrew Simpson (eds), Sluicing: Crosslinguistic perspectives. Oxford: Oxford University Press, 219–47. Adliene, Egle (2014). Multiple sluicing in Lithuanian. Master’s thesis, UCL. Aelbrecht, Lobke (2007a). A movement account of Dutch gapping. Talk presented at Lin­ guistics in the Netherlands Day, Utrecht University, 3 February. Aelbrecht, Lobke (2007b). A movement account of Dutch gapping. Ms. Catholic Universi­ ty of Brussels. Aelbrecht, Lobke (2009). You have the right to remain silent: The syntactic licensing of el­ lipsis. PhD thesis, Catholic University of Brussels. Aelbrecht, Lobke (2010). The syntactic licensing of ellipsis. Amsterdam: John Benjamins. Aelbrecht, Lobke (2012). Modal complement ellipsis: VP ellipsis in Dutch?, in Peter Acke­ ma, Rhona Alcorn, Caroline Heycock, Dany Jaspers, Jeroen van Craenenbroeck, and Gui­ do Vanden Wyngaerd (eds), Comparative Germanic syntax: The state of the art. Amster­ dam and Philadelphia: John Benjamins, 1–34. Aelbrecht, Lobke (2016). Ellipsis, in Artemis Alexiadou and Tibor Kiss (eds), Syntax: Theo­ ry and analysis. An international handbook. Berlin: Mouton de Gruyter, 562–95. Aelbrecht, Lobke and Will Harwood (2015). To be or not be elided: VP ellipsis revisited, Lingua 153: 66–97. Agüero-Bautista, Calixto (2007). Diagnosing cyclicity in sluicing, Linguistic Inquiry 38: 413–44.

Page 3 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Ahmed-Chamanga, Mohamed (2010). Introduction à la grammaire structurale du co­ morien. Volume 1: Le shiNgazidja. Paris: KomEdit. Akmajian, Adrian (1973). The role of focus in the interpretation of anaphoric expressions, in Stephen R. Anderson and Paul Kiparsky (eds), A Festschrift for Morris Halle. New York: Holt, 215–25. Akmajian, Adrian (1984). Sentence types and the form–function fit. Natural Language and Linguistic Theory 2: 1–23. Akmajian, Adrian, Susan Steele, and Thomas Wasow (1979). The category AUX in univer­ sal grammar, Linguistic Inquiry 10(1): 1–64. Akmajian, Adrian and Thomas Wasow (1975). The constituent structure of VP and Aux and the position of the verb be, Linguistic Analysis 1: 205–45. Ala-Sippola, Sanna (2012). Agentin ilmaiseminen suomalaisella viittomakielellä tuotetussa asiatekstissa [Expressing agent in a formal FinSL narrative]. MA thesis in General Lin­ guistics, Department of Modern Languages, University of Helsinki. Albert, Chris (1993). Sluicing and weak islands. Ms. UC, Santa Cruz. Alcántara, Manuel and Nuria Bertomeu (2005). Ellipsis in spontaneous spoken language, in Jennifer Spenander and Petra Hendriks (eds), Proceedings of the cross-mod­ (p. 1041)

ular approaches to ellipsis. Edinburgh: Heriot-Watt University, 1–10. Alexiadou, Artemis and Kirsten Gengel (2008). NP-ellipsis without focus movement/pro­ jection: The role of classifiers. Talk presented at the Workshop on Interface-Based Ap­ proaches to Information Structure, University College London. Alexiadou, Artemis and Kirsten Gengel (2012). NP-ellipsis without focus movement/pro­ jections: The role of classifiers, in Ivona Kučerová and Ad Neeleman (eds), Contrast and positions in information structure. Cambridge: Cambridge University Press, 177–205. Algryani, Ali (2012). The syntax of ellipsis in Libyan Arabic: A generative analysis of sluic­ ing, VP ellipsis, stripping and negative contrast. PhD thesis, Newcastle University. Algueró, Anna, Anna Gavarró, Maria Teresa Guasti, Laurice Tuller, Philippe Prévost, Adri­ ana Belletti, Luca Cilibrasi, Hélène Delage, and Mirta Vernice (2012). The acquisition of partitive clitics in Romance five-year-olds, IBERIA: An International Journal of Theoretical Linguistics 3(2): 1–19. Almeida, Diogo A. de A. and Masaya Yoshida (2007). A problem for the preposition strand­ ing generalization, Linguistic Inquiry 38(2): 349–62. Alonso-Ovalle, Luis (2006). Disjunction in alternative semantics. PhD thesis, University of Massachusetts.

Page 4 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Alrenga, Peter, Chris Kennedy, and Jason Merchant (2012). A new standard of compari­ son, in Nathan Arnett and Ryan Bennet (eds), Proceedings of the 30th West Coast Confer­ ence on Formal Linguistics. Somerville, MA: Cascadilla Press, 32–42. Amaral, Luiz, Marcus Maia, Andrew Nevins, and Tom Roeper (2018). Recursion across do­ mains. Cambridge: Cambridge University Press. Amaral, Luiz and Tom Roeper (2014). Multiple grammars and second language represen­ tation, Second Language Research 30(1): 3–36. Amaral, Patricia, Craige Roberts, and E. Allyn Smith (2007). Review of The Logic of Con­ ventional Implicatures, Linguistics and Philosophy 30: 707–49. An, Duk-Ho (2007). Syntax at the PF interface: Prosodic mapping, linear order, and dele­ tion. PhD thesis, University of Connecticut at Storrs. Anand, Pranav and James McCloskey (2015). Annotating the implicit content of sluices, in Adam Meyers, Ines Rehbein, and Heike Zinsmeister (eds), Proceedings of the 9th Linguis­ tic Annotation Workshop (LAW-IX). Red Hook, NY: Curran Associates, 178–87. AnderBois, Scott (2010). Sluicing as anaphora to issues, in Nan Li and David Lutz (eds), SALT 20: Proceedings of the 20th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 451–70. AnderBois, Scott (2011). Issues and alternatives. PhD thesis, UC Santa Cruz. AnderBois, Scott (2012a). Focus and uninformativity in Yucatec Maya questions, Natural Language Semantics 20(4): 349–90. AnderBois, Scott (2012b). Indefiniteness and the typology of implicit arguments, in Nathan Arnett and Ryan Bennett (eds), Proceedings of the 30th West Coast Conference on Formal Linguistics (WCCFL 30). Somerville, MA: Cascadilla Press, 43–53. AnderBois, Scott (2014). The semantics of sluicing: Beyond truth-conditions, Language 90(4): 887–926. AnderBois, Scott, Adrian Brasoveanu, and Robert Henderson (2015). At-issue proposals and appositive impositions in discourse, Journal of Semantics 32(1): 93–138. Anderson, Stephen R. (1972). How to get ‘even’, Language 48: 893–905. Ariel, Mira (1990). Accessing Noun Phrase antecedents. London and New York: Rout­ ledge. Arnold, Doug and Robert D. Borsley (2008). Non-restrictive relative clauses, ellipsis and anaphora, in Stefan Müller (ed.), Proceedings of the 15th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 325–45.

Page 5 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Arnold, Doug and Robert D. Borsley (2010). Auxiliary-stranding relative clauses, in Stefan Müller (ed.), Proceedings of the 15th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 47–67. (p. 1042)

Arnold, Doug and Andrew Spencer (2015). A constructional analysis for the skeptical, in Stefan Müller (ed.), Proceedings of the 22nd International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 41–60. Arnon, Inbal, Bruno Estigarribia, Philip Hofmeister, Florian Jaeger, Jeanette Pettibone, Ivan A. Sag, and Neal Snider (2005). Rethinking superiority effects: A processing model. Poster presented at the CUNY Sentence Processing Conference, University of Arizona. Arregi, Karlos (2010). Ellipsis in split questions, Natural Language and Linguistic Theory 28(3): 539–92. Arregi, Karlos and Naiara Centeno (2005). Determiner sharing and cyclicity in wh-move­ ment, in Randall Gess and Edward J. Rubin (eds), Theoretical and experimental approach­ es to Romance Linguistics. Amsterdam and Philadelphia: John Benjamins, 1–19. Arregui, Ana, Jr., Charles Clifton, Lyn Frazier, and Keir Moulton (2006). Processing elided verb phrases with flawed antecedents: The recycling hypothesis, Journal of Memory and Language 55: 232–46. Asher, Nicholas (1993). Reference to abstract objects in English. Dordrecht: Kluwer Acad­ emic Publishers. Asher, Nicholas, Daniel Hardt, and Joan Busquets (2001). Discourse parallelism, ellipsis, and ambiguity, Journal of Semantics 18(1): 1–25. Asher, Nicholas and Alex Lascarides (2003). Logics of conversation. Cambridge: Cam­ bridge University Press. Ashton, Ethel O. (1944). Swahili grammar, including intonation. London: Longman. Authier, Jean-Marc (2011). A movement analysis of French modal ellipsis, Probus 23: 175– 216. Authier, Jean-Marc (2012). Ellipsis as movement and silence: Evidence from French, Uni­ versity of Pennsylvania Working Papers in Linguistics 17(1): 1–9. Authier, Jean-Marc (2013). Phase-edge features and the syntax of polarity particles, Lin­ guistic Inquiry 44(3): 345–89. Babby, Leonard H. (1987). Case, prequantifiers, and discontinuous agreement in Russian, Natural Language and Linguistic Theory 5(1): 91–138. Bach, Emmon and Barbara H. Partee (1980). Anaphora and semantic structure, in Jody Kreiman and Almerindo E. Ojeda (eds), Papers from the Parasesion on Pronouns and Anaphora. Chicago: University of Chicago, 1–28. Page 6 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bachrach, Asaf and Roni Katzir (2009). Right-node raising and delayed spellout, in Klean­ thes K. Grohmann (ed.), InterPhases: Phase-theoretic investigations of linguistic inter­ faces. Oxford: Oxford University Press, 283–316. Bacskai-Atkari, Julia (2010). Parametric variation and comparative deletion, The Even Yearbook 9: 1–21. Bacskai-Atkari, Julia (2014). The syntax of comparative constructions: Operators, ellipsis phenomena and functional left peripheries. PhD dissertation, Universität Potsdam. Bahan, Benjamin, Judy Kegl, Robert G. Lee, Dawn MacLaughlin, and Carol Neidle (2000). The licensing of null arguments in American Sign Language, Linguistic Inquiry 31(1): 1– 27. Bailyn, John Frederick (1995). A configurational approach to Russian ‘free’ word order. PhD thesis, Cornell University. Bailyn, John Frederick (2004). Generalized inversion, Natural Language and Linguistic Theory 22(1): 1–49. Bailyn, John Frederick (2011). Kak tebe lingvistika? Nenavižu! How (not) to analyze Russ­ ian verb-stranding constructions. Paper presented at Formal Approaches to Slavic Lin­ guistics 20. MIT, Cambridge, MA. Bailyn, John Frederick (2012). The syntax of Russian. Cambridge: Cambridge University Press. Bailyn, John Frederick (2014). The case of Q, in Olga Arnaudova, Wayles Browne, Maria Luisa Rivero, and Danijela Stojanović (eds), Formal Approaches to Slavic Linguis­ (p. 1043)

tics 12: The Ottawa Meeting. Ann Arbor, MI: Michigan Slavic Publications, 1–36. Bailyn, John Frederick (2017). Against a VP ellipsis account of Russian verb-stranding constructions, in Alexander Vovin (ed.), Studies in Japanese and Korean linguistics and beyond. Boston: Brill, 93–109. Bailyn, John Frederick and Tanya Scott (2007). Asymmetries in Russian multiple sluicing: A purely syntactic account. Abstract for a talk at FASL (Formal Approaches to Slavic Lin­ guistics) 15. Baker, C. L. (1984). Two observations on British English do, Linguistic Inquiry 15(1): 155– 7. Baker, C. L. and Michael Brame (1972). ‘Global rules’: A rejoinder, Language 48(1): 51– 75. Baker, Mark, Kyle Johnson, and Ian Roberts (1989). Passive arguments raised, Linguistic Inquiry 20(2): 219–51.

Page 7 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bakker, Dik (1968). Samentrekking in Nederlandse syntactische groepen. PhD thesis, Lei­ den University. Baltin, Mark (1993). Negation and clause structure. Ms. New York University. Baltin, Mark (2004). The position of adverbials. Ms. New York University. Baltin, Mark (2006). The nonunity of VP-preposing, Language 82(4): 734–66. Baltin, Mark (2007). Deletion versus pro-forms: A false dichotomy? Ms. New York Univer­ sity. Baltin, Mark (2012). Deletion versus pro-forms: An overly simple dichotomy? Natural Lan­ guage and Linguistic Theory 30: 381–423. Banfield, Ann (1981). Stylistic deletion in coordinate structures, Linguistic Analysis 7: 1– 32. Bánréti, Zoltán (1992). A mellérendelés, in Ferenc Kiefer (ed.), Strukturális magyar nyelv­ tan. Mondattan. Budapest: Akadémiai Kiadó, 715–96. Bánréti, Zoltán (2001). Multiple lexical selection and parallelism in Hungarian VP-ellipsis, Acta Linguistica Hungarica 48: 25–58. Bánréti, Zoltán (2007). A mellérendelés és az ellipszis nyelvtana a magyarban. Budapest: Tinta Könyvki-adó. Barbiers, Sjef (2000). The right-periphery in SOV-languages, in Peter Svenonius (ed.), The derivation of VO and OV. Amsterdam and Philadelphia: John Benjamins, 181–219. Barker, Chris (1998). Partitives, double genitives and anti-uniqueness, Natural Language and Linguistic Theory 16(4): 679–717. Barker, Chris (2007). Parasitic scope. Linguistics and Philosophy 30. 3: 407–44. Barker, Chris (2013). Scopability and sluicing, Linguistics and Philosophy 36: 187–223. Barros, Matthew (2012a). Sluiced fragment answers: Another puzzle involving islands and ellipsis, Snippets 25: 5–6. Barros, Matthew (2012b). Else-modification as a diagnosis for pseudosluicing. Ms. Rut­ gers University. Barros, Matthew (2013). Harmonic sluicing: Which remnant/correlate pairs work and why, in SALT 23: Proceedings of the 23rd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 295–315. Barros, Matthew (2014a). A non-repair approach to island sensitivity in contrastive TP el­ lipsis, in Proceedings from the 48th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 61–75. Page 8 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Barros, Matthew (2014b). Sluicing and identity in ellipsis. PhD thesis, Rutgers University, New Brunswick, NJ. Barros, Matthew and Jeroen van Craenenbroeck (2013). Tag questions and ellipsis. Paper presented at Deutsche Gesellschaft für Sprachwissenschaft, Theme session: Parentheses and ellipsis: Crosslinguistic and theoretical perspectives. Barros, Matthew, Patrick Elliott, and Gary Thoms (2014). There is no island repair. Ms., Rutgers University, University College London, and University of Edinburgh. Barros, Matthew, Patrick Elliott, and Gary Thoms (2015). More variation in island repair: The clausal/non-clausal island distinction, in Proceedings of the 49th annual meeting of the Chicago Linguistic Society, Chicago: Chicago Linguistic Society, ch. 2. Barros, Matthew and Luis Vicente (2011). Right node raising requires both ellip­ sis and multidomination, University of Pennsylvania Working Papers in Linguistics 17(1): 1–9. (p. 1044)

Barros, Matthew and Luis Vicente (2016). A remnant condition for ellipsis, in Kyeong-min Kim, Pocholo Umbal, Trevor Block, Queenie Chan, Tanie Cheng, Kelli Finney, Mara Katz, Sophie Nickel-Thompson, and Lisa Shorten (eds), Proceedings of the 33rd West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, 57– 66. Barss, Andrew (1986). Chains and anaphoric dependence: On reconstruction and its im­ plications. PhD thesis, Massachusetts Institute of Technology. Bartlett, Laura and Kay González-Vilbazo (2010). Classifiers in Spanish-Taiwanese codeswitching, in Proceedings of the 46th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 1–15. Bartlett, Laura and Kay González-Vilbazo (2013). The structure of the Taiwanese DP in Taiwanese–Spanish bilinguals: Evidence from code-switching, Journal of East Asian Lin­ guistics 22(1): 65–99. Barton, Ellen (1990). Nonsentential constituents: A theory of grammatical structure and pragmatic interpretation. Amsterdam and Philadelphia: John Benjamins. Barton, Ellen (2006). Toward a nonsentential analysis in generative grammar, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsen­ tentials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 11–32. Barton, Ellen and Ljiljana Progovac (2006). Nonsententials in Minimalism, in Reinaldo El­ ugardo and Robert J. Stainton (eds), Ellipsis and nonsentential speech. Amsterdam and Philadelphia: Springer, 71–93.

Page 9 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bartos, Huba (2000a). Az inflexiós jelenségek szintaktikai háttere, in Ferenc Kiefer (ed.), Strukturális magyar nyelvtan 3: Morfológia. Budapest: Akadémiai Kiadó, 653–761. Bartos, Huba (2000b). VP-ellipsis and verbal inflection in Hungarian, Acta Linguistica Hungarica 47: 3–23. Bartos, Huba (2001a). A mutató névmási módosítók a magyarban: Egyeztetés vagy osz­ tozás? in Marianne Barkó-Nagy, Zoltán Bánréti, and Katalin É. Kiss (eds), Újabb tanul­ mányok a strukturális magyar nyelvtan és a nyelvtörtenet köréből. Budapest: Osiris, 19– 41. Bartos, Huba (2001b). Sound–form non-insertion and the direction of ellipsis, Acta Lin­ guistica Hungarica 48(3): 3–24. Bartsch, Renate and Theo Vennemann (1972). Semantic structures. Frankfurt am Main: Athenaeum. Barwise, John and John Etchemendy (1987). The liar. New York: Oxford University Press. Barwise, John and John Perry (1983). Situations and attitudes. Cambridge, MA: MIT Press. Bassong, Paul Roger (2014). Information structure and the Basa’a left peripheral. PhD thesis, University of Yaounde I. Beaver, David I. (2004). The optimization of discourse anaphora, Linguistics and Philoso­ phy 27(1): 3–56. Beaver, David I. and Brady C. Clark (2008). Sense and sensitivity: How focus determines meaning. Malden, MA: Wiley-Blackwell. Beavers, John and Ivan A. Sag (2004). Coordinate ellipsis and apparent non-constituent coordination, in Stefan Müller (ed.), Proceedings of the 11th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 48–69. Bech, Gunnar (1983). Studien über das deutsche Verbum infinitum. Tübingen: M. Niemey­ er. Beck, Sigrid (2011). Comparison constructions, in Claudia Maienborn, Klaus von Heusinger, and Paul Portner (eds), Semantics: An international handbook of natural lan­ guage meaning. Berlin and New York: Mouton de Gruyter, 1341–89. Beck, Sigrid, Svetlana Krasikova, Daniel Fleischer, Remus Gergel, Christiane Savelsberg, John Vanderelst, and Elisabeth Villalta (2009). Crosslinguistic variation in comparison constructions, Linguistic Variation Yearbook 9: 1–66. Beck, Sigrid and Uli Sauerland (2000). Cumulation is needed: A reply to Winter (2000), Natural Language Semantics 8(4): 349–71. Page 10 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Beckman, Mary E. (1996). The parsing of prosody, Language and Cognitive Processes 11: 17–67. (p. 1045)

Beckman, Mary E. and Gayle E. Elam (1997). Guidelines for ToBI labeling, Version 3. Ms. Ohio State University. Beckman, Mary E., Julia Hirschberg, and Stefanie Shattuck-Hufnagel (2006). The original ToBI system and the evolution of the ToBI framework, in Sun-Ah Jun (ed.), Prosodic typol­ ogy: The phonology of intonation and phrasing. Oxford: Oxford University Press, 9–54. Beckman, Mary E. and Janet B. Pierrehumbert (1986). Intonational structure in Japanese and English, Phonology Yearbook 3: 255–309. Beckman, Mary E. and Jennifer J. Venditti (2011). Intonation, in John Goldsmith, Jason Riggle, and Alan C. L. Yu (eds), Handbook of Phonological Theory. Malden, MA: Blackwell Publishing, 485–532. Beecher, Henry (2008). Pragmatic inference in the interpretation of sluiced prepositional phrases, San Diego Linguistics Papers 3: 2–10. Belazi, Hedi M., Edward J. Rubin, and Almeida Jacqueline Toribio (1994). Code switching and X-bar theory: The functional head constraint, Linguistic Inquiry 25(2): 221–38. Belletti, Adriana (2004). Aspects of the low IP area, in Luigi Rizzi (ed.), The Structure of CP and IP. Oxford: Oxford University Press, 16–51. Belletti, Adriana and Ur Shlonsky (1995). The order of verbal complements: A compara­ tive study, Natural Language and Linguistic Theory 13(3): 489–526. Bennett, Ryan, Emily Elfner, and James McCloskey (2015). Prosody, focus, and ellipsis in Irish. Ms. Yale University, University of British Columbia, and University of California, Santa Cruz. Bennis, Hans (1977). Het kwantitatieve er in komparatiefconstructies, Spektator 6: 384–7. Bennis, Hans (1978). Comparative deletion is subdeletion. Master’s thesis, University of Amsterdam. Bennis, Hans (1986). Gaps and dummies. Dordrecht: Foris Publications. Bentzen, Kristine, Jason Merchant, and Peter Svenonius (2013). Deep properties of sur­ face pronouns: Pronominal predicate anaphors in Norwegian and German, Journal of Comparative Germanic Linguistics 16(2): 97–125. Bergsma, Shane, David Yarowsky, and Kenneth Church (2011). Using large monolingual and bilingual corpora to improve coordination disambiguation, in Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Tech­ nologies, Volume 1. Stroudsburg, PA: Association for Computational Linguistics, 1346–55. Page 11 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bernolet, Sarah, Robert Hartsuiker, and Simona Collina (2016). The persistence of syntac­ tic priming revisited, Journal of Memory and Language 91: 99–116. Bernstein, Judy (1993). The syntactic role of word markers in null nominal constructions, Probus 5: 5–38. Bertomeu, Nuria and Valia Kordoni (2005). Integrating pragmatic information in gram­ mar: An analysis of intersentential ellipsis, in Stefan Müller (ed.), Proceedings of the 12th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 46–69. den Besten, Hans (1978). On the presence and absence of wh-elements in Dutch compar­ atives, Linguistic Inquiry 9(4): 641–71. de Beuzeville, Louise, Trevor Johnston, and Adam Schembri (2009). The use of space with indicating verbs in Auslan: A corpus-based investigation, Sign Language and Linguistics 12: 53–82. Bever, Thomas G. (1970). The cognitive basis for linguistic structure, in John R. Hayes (ed.), Cognition and the development of language. New York: John Wiley, 279–362. Bezuidenhout, Anne (2013). Structuring silence versus the structure of silance, in Lau­ rence Goldstein (ed.), Brevity. Oxford: Oxford University Press, 36–52. Bhat, D. N. Shankar (2000). The indefinite-interrogative puzzle, Linguistic Typology 4: 365–400. Bhatt, Rajesh (2015). Relative clauses and correlatives, in Artemis Alexiadou and Tibor Kiss (eds), Syntax: Theory and Analysis. An International Handbook. Berlin: Mouton de Gruyter, 708–49. Bhatt, Rajesh and Roumyana Pancheva (2004). Late merger of degree clauses, Linguistic Inquiry 35(1): 1–46. Bhatt, Rajesh and Shoichi Takahashi (2007). Direct comparisons: Resurrecting the direct analysis of phrasal comparatives, in Masayuki Gibson and Tova Friedman (eds), SALT 17: Proceedings of the 17th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 19–36. (p. 1046)

Bhatt, Rajesh and Shoichi Takahashi (2011). Reduced and unreduced phrasal compara­ tives, Natural Language and Linguistic Theory 29(3): 581–620. Bhattacharya, Tanmoy and Andrew Simpson (2012). Sluicing in Indo-Aryan: An investiga­ tion of Bangla and Hindi, in Jason Merchant and Andrew Simpson (eds), Sluicing: Crosslinguistic perspectives. Oxford: Oxford University Press, 183–218. Bianchi, Valentina (2004). Resumptive relatives and LF chains, in Luigi Rizzi (ed.), The Structure of CP and IP. Oxford: Oxford University Press, 76–114. Page 12 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Biezma, Maria (2014). Multiple focus strategies in pro-drop languages: Evidence from el­ lipsis in Spanish, Syntax 17: 91–131. Bîlbîie, Gabriela (2011). Grammaire des constructions elliptiques: Une étude comparative des phrases sans verbe en roumain et en français. PhD thesis, Université Paris Diderot. Bîlbîie, Gabriela (2013). A quantitative study on right node raising in the Penn treebank. Presentation at the International Congress of Linguists, Geneva. Bjorkman, Bronwyn (2011). BE-ing default: The morphosyntax of auxiliaries. PhD thesis, MIT. Blackburn, Patrick and Wilfried Meyer-Viol (1994). Linguistics, logic and finite trees, Log­ ic Journal of the Interest Group of Pure and Applied Logics 2(1): 3–29. Blom, A. (1975). Against conjunction reduction, in A. Kraak (ed.), Linguistics in the Netherlands 1972–1973. Amsterdam: Van Gorcum, 128–40. Blom, Elma, Daniela Polišenská, and Fred Weerman (2008). Articles, adjectives and age of onset: The acquisition of Dutch grammatical gender, Second Language Research 24(3): 297–331. Boas, H. C. (ed.) (2010). Contrastive studies in construction grammar, vol. 10. Amster­ dam: John Benjamins. Bobaljik, Jonathan and Cynthia Zocca (2010). Gender markedness: The anatomy of a counter-example, Morphology 21: 1–26. Bochnak, Michael (2013). Cross-linguistic variation in the semantics of comparatives. PhD dissertation, University of Chicago. Bock, Kathryn and Zenzi M. Griffin (2000). The persistence of structural priming: Tran­ sient activation or implicit learning? Journal of Experimental Psychology 129(2): 177–92. Bolinger, Dwight (1978). Asking more than one thing at a time, in Henry Hiz (ed.), Ques­ tions. Dordrecht: D. Reidel, 107–50. Bonami, Olivier and Danièle Godard (2008). On the syntax of direct quotation in French, in Stefan Müller (ed.), Proceedings of the 15th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 358–77. de Bont, A. P. (1962). Dialekt van Kempenland. Meer in het bijzonder d’Oerse taol. Deel 1. Klank- en vormleer en enige syntactische bijzonderheden. Assen: Van Gorcum. Booij, Geert (1985). Coordination reduction in complex words: A case for prosodic phonol­ ogy, in Harry van der Hulst and N. Smith (eds), Advances in Nonlinear Phonology. Dor­ drecht: Foris Publications, 143–60.

Page 13 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Boone, Enrico (2014). The syntax and licensing of gapping and fragments. PhD thesis, Leiden University. Borsley, Robert D. (2005). Against ConjP, Lingua 115: 461–82. Bos, Johan and Jennifer Spenader (2011). An annotated corpus for the analysis of VP ellip­ sis, Language Resources and Evaluation 45(4): 463–94. Bošković, Željko (1994). D-structure, Theta-Criterion and movement into theta-positions, Linguistic Analysis 24: 247–86. Bošković, Željko (1997). Fronting wh-phrases in Serbo-Croatian, in Martina Lindseth and Steven Franks (eds), Formal Approaches to Slavic Linguistics: The Indiana meeting. Ann Arbor, MI: Michigan Slavic Publications, 86–107. Bošković, Željko (1998). Wh-movement and wh-phrases in Slavic. Position paper presented at the Comparative Slavic Morphosyntax Workshop, Spencer, IN. (p. 1047)

Bošković, Željko (1999). On multiple feature checking: Multiple wh-fronting and multiple head movement, in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, 159–87. Bošković, Željko (2000). Sometimes in [Spec, CP], sometimes in situ, in Roger Martin, David Michaels, and Juan Uriagereka (eds), Step by step. Cambridge, MA: MIT Press, 53– 156. Bošković, Željko (2002). On multiple wh-fronting, Linguistic Inquiry 33(3): 351–83. Bošković, Željko (2003). On the interpretation of multiple questions, Linguistic Variation Yearbook 1: 1–15. Bošković, Željko (2004a). Be careful where you float your quantifiers, Natural Language and Linguistic Theory 22(4): 681–742. Bošković, Željko (2004b). Two notes on Right Node Raising, University of Connecticut Working Papers in Linguistics 12: 13–24. Bošković, Željko (2008). What will you have, DP or NP? in Emily Elfner and Martin Walkow (eds), NELS 37: Proceedings of the 37th annual meeting of the North East Lin­ guistic Society. Amherst, MA: GLSA, 101–14. Bošković, Željko (2011). Rescue by PF deletion, traces as (non)interveners, and the thattrace effect, Linguistic Inquiry 42(1): 1–44. Bošković, Željko (2013a). Adjectival escapades, in Steven Franks, Markus Dickinson, George Fowler, Melissa Witcombe, and Ksenia Zanon (eds), Proceedings of Formal Ap­ proaches to Slavic Linguistics 21. Ann Arbor, MI: Michigan Slavic Publications, 1–25.

Page 14 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bošković, Željko (2013b). Traces do not head islands: What can deletion rescue? in Yoichi Miyamoto, Daiko Takahashi, Hideki Maki, Masao Ochi, Koji Sugisaki, and Asako Uchibori (eds), Deep insights, broad perspectives: Essays in honor of Mamuro Saito. Tokyo: Kaitakusha, 56–93. Bošković, Željko (2014). Now I’m a phase, now I’m not a phase: On the variability of phas­ es with extraction and ellipsis, Linguistic Inquiry 45(1): 27–89. Bošković, Željko (2016). On pronouns, clitic doubling and argument ellipsis: Argument el­ lipsis as predicate ellipsis. Ms. University of Connecticut, Storrs. Bošković, Željko (2017). On clitic doubling and argument ellipsis, in Aaron Kaplan, Abby Kaplan, Miranda K. McCarvel, and Edward J. Rubin (eds), Proceedings of the 34th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, 97–106. Bošković, Željko and Steven Franks (2000). Across-the-board movement and LF, Syntax 3(2): 107–28. Bošković, Željko and Daiko Takahashi (1998). Scrambling and last resort, Linguistic In­ quiry 29(3): 347–66. Bosque, Ignacio (1984). Negación y ellipsis, Estudios de Lingüística 2: 171–99. Bouchard, Denis (2002). Adjectives, number and interfaces: Why languages vary. Oxford: Elsevier Science. Bouton, L. F. (1970). Antecedent contained pro-forms, in Papers from the Sixth Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 154–67. Bowers, John (2002). Transitivity, Linguistic Inquiry 33(2): 183–234. Brame, Michael (1983). Ungrammatical notes 4: Smarter than me, Linguistic Analysis 12(3): 323–8. Branco, António and Francisco Costa (2006). Noun ellipsis without empty categories, in Stefan Müller (ed.), Proceedings of the 13th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 81–101. Bresnan, Joan (1971). A note on the notion ‘identity of sense anaphora’, Linguistic Inquiry 2(4): 589–97. Bresnan, Joan (1973). Syntax of the comparative clause construction in English, Linguis­ tic Inquiry 4(3): 275–343. Bresnan, Joan (1974). The position of certain clause-particles in phrase structure, Linguis­ tic Inquiry 5(4): 614–19.

Page 15 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Bresnan, Joan (1975). Comparative deletion and constraints on transformations, Linguistic Analysis 1(1): 25–74. (p. 1048)

Bresnan, Joan (1976). On the form and functioning of transformations, Linguistic Inquiry 7(1): 3–40. Bresnan, Joan and Sam A. Mchombo (1995). The lexical integrity principle: Evidence from Bantu, Natural Language and Linguistic Theory 13(2): 181–254. Brody, Mihály (1990). Remarks on the order of elements in the Hungarian focus field, in István Kenesei (ed.), Approaches to Hungarian 3. Szeged: JATE Press, 95–122. Brody, Mihály (1995). Focus and checking theory, in István Kenesei (ed.), Levels and structures. Approaches to Hungarian 5. Szeged: JATE Press, 29–44. Broekhuis, Hans (2013). Syntax of Dutch: Adjectives and adjective phrases. Amsterdam: Amsterdam University Press. Broekhuis, Hans and Norbert Corver (2015). Syntax of Dutch: Verbs and verb phrases, vol. II. Amsterdam: Amsterdam University Press. Brown, Roger (1958). How shall a thing be called? Psychological Review 65(1): 14–21. Brown-Schmidt, Sarah, Christine Gunlogson, and Michael K. Tanenhaus (2008). Ad­ dressees distinguish shared from private information when interpreting questions during interactive conversation, Cognition 107(3): 1122–36. Brucart, José María (1987). La elisión sintáctica en español. Bellaterra: Publicacions de la Universitat Autonoma de Barcelona. Brucart, José María (1999). La ellipsis, in Ignacio Bosque and Violeta Demonte (eds), Gramática descriptiva de la lengua española. Madrid: Espasa Calpe, 1787–866. Brucart, José María and Jonathan E. MacDonald (2013). Empty categories and ellipsis, in José Ignacio Hualde, Antxon Olarrea, and Erin O’Rourke (eds), The handbook of Hispanic linguistics. Malden, MA: John Wiley & Sons. Bruening, Benjamin (2015). Passive do so. Ms. University of Delaware. Brugos, Alenja, Stefanie Shattuck-Hufnagel, and Nanette Veilleux (2006). Transcribing prosodic structure of spoken utterances with ToBI. Tutorial through MIT OpenCourse­ Ware. Bruner, Jerome Seymour (1957). On perceptual readiness, Psychological Review 64: 123– 52. Buell, Leston (2005). Issues in Zulu verbal morphosyntax. PhD thesis, UCLA.

Page 16 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Büring, Daniel (2003). On D-trees, beans, and B-accents, Linguistics and Philosophy 26: 511–45. Büring, Daniel (2005). Binding theory. Cambridge: Cambridge University Press. Büring, Daniel and Katharina Hartmann (1998). Asymmetrische Koordination, Linguistis­ che Berichte 174: 172–201. Burton, Strang and Jane Grimshaw (1992). Coordination and VP-internal subjects, Lin­ guistic Inquiry 23(2): 305–12. Busquets, Joan and Pascal Denis (2001). L’ellipse modale en français: Le cas de pouvoir et devoir, Cahiers de Grammaire 26: 55–74. Bybee, Joan, Revere Perkins, and William Pagliuca (1994). The evolution of grammar: Tense, aspect, and modality in the languages of the world. Chicago: University of Chicago Press. Byrant, Doreen (2006). Koordinationsellipsen im Spracherwerb: Die Verarbitung poten­ tieller Gapping-Strukturen. Berlin: Walter de Gruyter. Cabredo Hofherr, Patricia (2006). Pronouns, determiners and N-ellipsis in Spanish, French, and German, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS 36: Proceedings of the 36th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 167–78. Caha, Pavel (2009). The nanosyntax of case. PhD thesis, University of Tromsø. Caha, Pavel (2011). Case in adpositional phrases. Ms. CASTL, Tromsø. Cai, Zhenguang G., Martin J. Pickering, and Patrick Sturt (2013). Processing verb-phrase ellipsis in Mandarin Chinese: Evidence against the syntactic account, Language and Cog­ nitive Processes 28: 810–28. Camacho, José (2003). The structure of coordination: Conjunction and agreement phenomena in Spanish and other languages. Dordrecht: Kluwer Academic Publishers. (p. 1049)

Cann, Ronnie (2011). Towards an account of the English auxiliary system: Building inter­ pretations incrementally, in Ruth Kempson, Eleni Gregoromichelaki, and Christine Howes (eds), Dynamics of lexical interfaces. Chicago: CSLI Press, 279–317. Cann, Ronnie, Ruth Kempson, and Lutz Marten (2005). The dynamics of language. Ox­ ford: Elsevier. Cann, Ronnie, Ruth Kempson, and Matthew Purver (2007). Context and well-formedness: The dynamics of ellipsis, Research on Language and Computation 5(3): 333–58. Cantone, Katja Francesca (2007). Code-switching in bilingual children. Berlin: Springer.

Page 17 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Cantor, Sara (2013). Ungrammatical double-island sluicing as a diagnostic of left-branch positioning. Master’s thesis, University of California at Santa Cruz. Caplan, David and Gloria S. Waters (1999). Verbal working memory and sentence compre­ hension, Behavioral and Brain Sciences 22: 77–126. Caponigro, Ivano and Kathryn Davidson (2011). Ask and tell as well: Clausal question-an­ swer pairs in ASL, Natural Language Semantics 19: 323–71. Caramazza, Alfonso (1984). The logic of neuropsychological research and the problem of patient classification in aphasia, Brain and Language 21: 9–20. Caramazza, Alfonso and Edgar Zurif (1976). Dissociations of algorithmic and heuristic processes in sentence comprehension: Evidence from aphasia, Brain and Language 3: 572–82. Carberry, Sandra (1989). A pragmatics-based approach to ellipsis resolution, Computa­ tional Linguistics 15(2): 75–96. Cardinaletti, Anna (1994). On the internal structure of pronominal DPs, The Linguistic Re­ view 11(3–4): 195–220. Carlson, Greg (1987). Same and different: Some consequences for syntax and semantics, Linguistics and Philosophy 10(4): 531–66. Carlson, Katy (2001). The effects of parallelism and prosody on the processing of gapping structures, Language and Speech 44: 1–26. Carlson, Katy (2002). Parallelism and prosody in the processing of ellipsis sentences. New York: Routledge. Carlson, Katy, Michael Walsh Dickey, Lyn Frazier, and Charles Clifton Jr (2009a). Informa­ tion structure expectations in sentence comprehension, Quarterly Journal of Experimen­ tal Psychology 62: 114–39. Carlson, Katy, Michael Walsh Dickey, and Christopher Kennedy (2005). Structural econo­ my in the processing and representation of gapping sentences, Syntax 8: 208–28. Carlson, Katy, Lyn Frazier, and Charles Clifton, Jr. (2009b). How prosody constrains com­ prehension: A limited effect of prosodic packaging, Lingua 119(7): 1066–82. Carroll, John M., Thomas G. Bever, and Chava R. Pollack (1981). The non-uniqueness of linguistic intuitions, Language 57(2): 368–83. Casielles, Eugenia (2006). Big questions, small answers, in Ljiljana Progovac, Kate Pae­ sani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisci­ plinary perspectives. Amsterdam and Philadelphia: John Benjamins, 117–45.

Page 18 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Cassimjee, Farida and Charles W. Kisseberth (1998). Optimal Domains Theory and Bantu tonology, in Larry M. Hyman and Charles W. Kisseberth (eds), Theoretical aspects of Ban­ tu tone. Stanford: CSLI, 33–132. Cassimjee, Farida and Charles W. Kisseberth (in preparation). The Shingazidja lexicon ex­ emplified. Ms. Columbia University and Tel Aviv University. Cattell, Ray (1976). Constraints on movement rules, Language 52: 18–50. Cecchetto, Carlo, Alessandra Cecchetto, Carlo Geraci, Mirko Santoro, and Sandro Zucchi (2015). The syntax of predicate ellipsis in Italian Sign Language (LIS), Lingua 166: 214– 35. Cecchetto, Carlo and Orin Percus (2006). When we do that and when we don’t: A contrastive analysis of VP-ellipsis and VP-anaphora, in Mara Frascarelli (ed.), Phases of interpretation. Berlin and New York: Mouton de Gruyter, 71–105. (p. 1050)

Centeno, Naiara (2011). Gapping and determiner sharing in Spanish. PhD thesis, Univer­ sity of the Basque Country, Vitoria-Gasteiz. Chae, Jeongmin, Younghee Jung, Taemin Lee, Soonyoung Jung, Chan Huh, Gilhan Kim, Hyeoncheoi Kim, and Heungbum Oh (2014). Identifying non-elliptical entity mentions in a coordinated NP with ellipses, Journal of Biomedical Informatics 47: 139–52. Chafe, Wallace L. (1976). Givenness, contrastiveness, definiteness, subjects, topics, and point of view, in Charles N. Li (ed.), Subject and topic. New York: Academic Press, 25–55. Chalcraft, Faye (2006). Do-doubling in West Yorkshire English. Paper presented at the workshop Syntactic Doubling in European Dialects. Chao, Wynn (1988). On ellipsis. New York and London: Garland Press. Charlow, Simon (2008). Free and bound pro-verbs: A unified treatment of anaphora, in To­ va Friedman and Satoshi Ito (eds), SALT 8: Proceedings of the Eighth Annual Conference on Semantics and Linguistic Theory. Ithaca, NY: CLC Publications, 194–211. Charlow, Simon (2012). Cross-categorial donkeys, in Maria Aloni, Vadim Kimmelman, Floris Roelofsen, Galit W. Sassoon, Katrin Schulz, and Matthijs Westera (eds), Logic, lan­ guage, and meaning. Berlin: Springer, 261–70. Charlow, Simon (2014). On the semantics of exceptional scope. PhD thesis, NYU. Chaves, Rui P. (2006). Coordination of unlikes without inlike categories, in Stefan Müller (ed.), Proceedings of the 13th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 102–22. Chaves, Rui P. (2007). Coordinate structures: constraint-based syntax-semantics. PhD the­ sis, University of Lisbon. Page 19 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Chaves, Rui P. (2008). Linearization-based word-part ellipsis, Linguistics and Philosophy 31: 261–307. Chaves, Rui P. (2009). A linearization-based approach to gapping, in James Rogers (ed.), Proceedings of FG-MoL 2005 The 10th Conference on Formal Grammar and the 9th Meet­ ing on Mathematics of Language. Stanford: CSLI Publications, 205–18. Chaves, Rui P. (2012). Conjunction, cumulation and respectively readings, Journal of Lin­ guistics 48: 297–344. Chaves, Rui P. (2014). On the disunity of right-node phenomena: Extraposition, ellipsis, and deletion, Language 90: 834–86. Chien, Yu-Chin and Kenneth Wexler (1990). Children’s knowledge of locality conditions as evidence for the modularity of syntax and pragmatics, Language Acquisition 1: 225–95. Chomsky, Noam (1955). The logical structure of linguistic theory. Ms. Harvard University and MIT. (Revised 1956 version published in part by New York: Plenum, 1975; Chicago: University of Chicago Press, 1985). Chomsky, Noam (1957). Syntactic structures. The Hague: Mouton & Co. Chomsky, Noam (1964). Current issues in linguistic theory. The Hague: Mouton & Co. Chomsky, Noam (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, Noam (1971). Deep structure, surface structure and semantic interpretaion, in Danny Steinberg and Leon Jacobovits (eds), Semantics. London: Cambridge University Press, 183–216. Chomsky, Noam (1972a). Some empirical issues in the theory of transformational gram­ mar, in Stanley Peters (ed.), The goals of linguistic theory. Englewood Cliffs, NJ: PrenticeHall, 63–130. Chomsky, Noam (1972b). Studies on semantics in generative grammar. The Hague: Mou­ ton. Chomsky, Noam (1973). Conditions on transformations, in Stephen Anderson and Paul Kiparsky (eds), Festschrift for Morris Halle. London: Rinehart and Winston, 232–86. Chomsky, Noam (1977a). Conditions on transformations, in Essays on form and interpre­ tation. New York: Elsevier North-Holland, 81–162. Chomsky, Noam (1977b). On wh-movement, in P. Culicover, T. Wasow, and A. Ak­ majian (eds), Formal syntax. New York: Academic Press, 71–132. (p. 1051)

Chomsky, Noam (1981). Lectures on government and binding. Dordrecht: Foris Publica­ tions.

Page 20 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Chomsky, Noam (1982). Some concepts and consequences of the theory of government and binding. Cambridge, MA: MIT Press. Chomsky, Noam (1986). Barriers. Cambridge, MA: MIT Press. Chomsky, Noam (1991). Some notes on economy of derivation and representation, in Robert Freidin (ed.), Principles and parameters in comparative grammar. Cambridge, MA: MIT Press, 417–54. Chomsky, Noam (1995a). Bare phrase structure, in Gert Webelhuth (ed.), Government binding theory and the minimalist program. Oxford: Oxford University Press, 383–439. Chomsky, Noam (1995b). The minimalist program. Cambridge, MA: MIT Press. Chomsky, Noam (2000). Minimalist inquiries: The framework, in Roger Martin, David Michaels, and Juan Uriagereka (eds), Step by step: Essays on minimalist syntax in honor of Howard Lasnik. Cambridge, MA: MIT Press, 89–156. Chomsky, Noam (2001). Derivation by phase, in Michael Kenstowicz (ed.), Ken Hale: A Life in Linguistics. Cambridge, MA: MIT Press, 1–52. Chomsky, Noam (2004). Beyond explanatory adequacy, in Adriana Belletti (ed.), Struc­ tures and beyond: The cartography of syntactic structures, vol. 3. Oxford: Oxford Univer­ sity Press, 104–31. Chomsky, Noam (2005). Three factors in language design, Linguistic Inquiry 36(1): 1–22. Chomsky, Noam (2008). On phases, in Robert Freidin, Carlos Otero, and Maria Luisa Zu­ bizarreta (eds), Foundational issues in linguistic theory. Cambridge, MA: MIT Press, 133– 66. Chomsky, Noam (2013). Problems of projection, Lingua 130: 33–49. Chomsky, Noam (2014). Minimal recursion: Exploring the prospects, in Tom Roeper and Margaret Speas (eds), Recursion: Complexity in cognition. Berlin: Springer, 1–15. Chomsky, Noam and Howard Lasnik (1993). The theory of Principles and Parameters, in Arnim von Stechow, W. Sternefeld and T. Vennemann (eds), Syntax: An international hand­ book of contemporary research. Berlin: Walter de Gruyter, 506–69. Chung, Sandra (2006). Sluicing and the lexicon: The point of no return, in Rebecca T. Cover and Yuni Kim (eds), Proceedings of the annual meeting of the Berkeley Linguistics Society, vol. 31. Berkeley, CA: Berkeley Linguistics Society, 73–91. Chung, Sandra (2013). Syntactic identity in sluicing: How much and why, Linguistic In­ quiry 44(1): 1–44. Chung, Sandra, William A. Ladusaw, and James McCloskey (1995). Sluicing and Logical Form, Natural Language Semantics 3(3): 239–82. Page 21 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Chung, Sandra, William A. Ladusaw and James McCloskey (2011). Sluicing(:) Between structure and inference, in Rodrígo Gutiérrez-Bravo, Line Mikkelsen, and Eric Potsdam (eds), Representing language: Essays in honor of Judith Aissen. Santa Cruz: Linguistics Research Center, 31–50. Ciardelli, Ivano (2009). Inquisitive semantics and intermediate logics. Master’s thesis, University of Amsterdam. Ciardelli, Ivano, Jeroen Groenendijk, and Floris Roelofsen (2011). Attention! Might in In­ quisitive Semantics, in Ed Cormany, Satoshi Ito, and David Lutz (eds), SALT 19: Proceed­ ings of the 19th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publica­ tions, 91–108. Ciardelli, Ivano, Jeroen Groenendijk, and Floris Roelofsen (2013). Inquisitive semantics: A new notion of meaning, Language and Linguistics Compass 7(9): 459–76. Cinque, Guglielmo (1990). Types of A′-dependencies. Cambridge, MA: MIT Press. Cinque, Guglielmo (2004). ‘Restructuring’ and functional structure, in Adriana Belletti (ed.), Structures and beyond: The cartography of syntactic structures, vol. 3. Oxford: Ox­ ford University Press, 132–91. Cinque, Guglielmo (2005). Deriving Greenberg’s universal 20 and its exceptions, Linguis­ tic Inquiry 36(3): 315–32. Cinque, Guglielmo (2006). Restructuring and functional heads: The cartography of syn­ tactic structures, vol. 4. New York: Oxford University Press. (p. 1052)

Citko, Barbara (2005). On the nature of merge: External merge, internal merge,

and parallel merge, Linguistic Inquiry 36(4): 475–96. Citko, Barbara (2007). Determiner sharing from a crosslinguistic perspective, in Pierre Pi­ ca (ed.), Linguistic variation yearbook 2006. Amsterdam and Philadelphia: John Ben­ jamins, 73–96. Citko, Barbara (2011). Symmetry in syntax. Cambridge: Cambridge University Press. Citko, Barbara (2015). To gap or to Right Node Raise. Presentation at Formal Approaches to Slavic Linguistics 24, New York. Citko, Barbara and Martina Gračanin-Yuksek (2013). Towards a new typology of coordi­ nated wh-questions, Journal of Linguistics 49(1): 1–32. Clark, Herbert (1996). Using language. Cambridge: Cambridge University Press. Clark, Herbert and Jean Fox Tree (2002). Using uh and um in spontaneous speaking, Cog­ nition 84: 73–111.

Page 22 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Clifton, Charles, Jr (2014). The (non)effect of discourse relations on comprehending verb phrase ellipsis. Paper presented at Architectures and Mechanisms of Language Process­ ing. Clifton, Charles, Jr and Lyn Frazier (2012). Discourse integration guided by Question Un­ der Discussion, Cognitive Psychology 65: 352–79. Clifton, Charles, Jr and Lyn Frazier (2016). Focus in corrective exchanges: Effects of pitch accent and syntactic form, Language and Speech 59(4): 544–61. Clifton, Charles, Jr and Lyn Frazier (in progress). Direct and indirect answers to ques­ tions. Ms. University of Massachusetts at Amherst. Colbert, Stephen (2007). I am America (and so can you!). New York: Grand Central Pub­ lishing. Cole, Peter, Gabriella Hermon, and Norhaida Aman (to appear). Clefted questions in Malay, in David Gil and James Collins (eds), Malay/Indonesian linguistics. London: Curzon Press. Collins, James N. (2013). The discourse potential of narrow scope indefinites in Samoan. Ms. Stanford. Collins, James N., Daria Popova, Ivan A. Sag, and Thomas Wasow (2014). Sluicing and the inquisitive potential of appositives. . Conner, Tracy (2014). Heads must be overt: Overtness and ellipsis licensing, in Marlies Kluck and Mark de Vries (eds), Parenthesis and ellipsis. Berlin: Mouton de Gruyter, 75– 108. Conner, Tracy (2015). Ellipsis licensing: Testing overtness. Ms. University of Massachu­ setts at Amherst. Conroy, Anastasia, Jeffrey Lidz, and Julien Musolino (2009). The Fleeting Isomorphism Ef­ fect, Language Acquisition 16: 106–17. Cooper, Robin (2005). Austinian truth, attitudes and type theory, Research on Language and Computation 3(4): 333–62. Cooper, Robin (2012). Type theory and semantics in flux, in Ruth Kempson, Nicholas Ash­ er, and Tim Fernando (eds), Handbook of the philosophy of science. Amsterdam: Elsevier, 271–323. Cooper, Robin and Jonathan Ginzburg (2015). Type theory with records for natural lan­ guage semantics, in Chris Fox and Shalom Lappin (eds), Handbook of Contemporary Se­ mantic Theory. Oxford: Blackwell, 375–407.

Page 23 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Coppen, Peter-Arno (1991). Specifying the Noun Phrase. PhD thesis, Catholic University Nijmegen. Coppock, Elizabeth (2001). Gapping: In defense of deletion, in Mary Andronis, Christoper Ball, Heidi Elston, and Sylvain Neuvel (eds), Proceedings of the 37th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 133–47. Coquand, Thierry, Randy Pollack, and Makoto Takeyama (2003). A logical framework with dependent types, Fundamenta Informaticae 20: 1–21. Corbett, Greville (1978). Universals in the syntax of cardinal numbers, Lingua 46: 355–68. Corblin, Francis, Jean-Marie Marandin, and Petra Sleeman (2003). Nounless determiners, in Francis Corblin and Henriette de Swart (eds), Handbook of French semantics. Stan­ ford: CSLI Publications, 21–38. Cormack, A. (1984). VP anaphora: Variables and scope, in F. Landman and F. Veltman (eds), Varieties of formal semantics. Dordrecht: Foris Publications, 81–102. Cornish, Francis (1999). Anaphora, discourse, and understanding: Evidence from French and English. Oxford: Oxford University Press. (p. 1053)

Corver, Norbert (1990). The syntax of left branch extractions. PhD thesis, Katholieke Uni­ versiteit Brabant. Corver, Norbert (1993). A note on subcomparatives, Linguistic Inquiry 24(4): 773–81. Corver, Norbert (1997). Much-support as a last resort, Linguistic Inquiry 28(1): 119–64. Corver, Norbert (2006). Comparative deletion and subdeletion, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax, vol. 1. Malden, MA and Oxford: Blackwell, 582–637. Corver, Norbert (2014). Freezing effects. Ms. University of Utrecht. Corver, Norbert and Marjo van Koppen (2009). Let’s focus on NP-ellpsis, Groninger Ar­ beiten zur Germanistischen Linguistik 48: 3–26. Corver, Norbert and Marjo van Koppen (2010). Ellipsis in Dutch possessive noun phrases: A microcomparative approach, Journal of Comparative Germanic Linguistics 13(2): 99– 140. Corver, Norbert and Marjo van Koppen (2011). NP-ellipsis with adjectival remnants: A mi­ cro-comparative perspective, Natural Language and Linguistic Theory 29: 371–421. Corver, Norbert and Marjo van Koppen (2015). Pronominalization and variation in Dutch demonstrative expressions. Ms. Utrecht University. Forthcoming in Tonjes Veenstra (ed.), Demonstratives. Amsterdam and Philadelphia: John Benjamins.

Page 24 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Costa, João, Ana Maria Martins, and Fernanda Pratas (2012). VP ellipsis: New evidence from Capeverdean, in Irene Franco, Sara Lusini, and Andrés Saab (eds), Romance lan­ guages and linguistic theory 2010. Amsterdam and Philadelphia: John Benjamins, 155–76. van de Craats, Ineke, Norbert Corver, and Roeland van Hout (2000). Conservation of grammatical knowledge: On the acquisition of possessive noun phrases by Turkish and Moroccan learners of Dutch, Linguistics 38(2): 221–314. van Craenenbroeck, Jeroen (2004). Ellipsis in Dutch dialects. PhD thesis, Leiden Universi­ ty. van Craenenbroeck, Jeroen (2008). Ellipsis and accommodation: The (morphological) case of sluicing. Paper presented at MIT. van Craenenbroeck, Jeroen (2010a). Invisible last resort: A note on clefts as the underly­ ing source for sluicing, Lingua 120(7): 1714–26. van Craenenbroeck, Jeroen (2010b). The syntax of ellipsis: Evidence from Dutch dialects. New York: Oxford University Press. van Craenenbroeck, Jeroen (2012a). How do you sluice when there is more than one CP? in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Ox­ ford: Oxford University Press, 40–67. van Craenenbroeck, Jeroen (2012b). Ellipsis, identity and accommodation. Ms. KU Leu­ ven. van Craenenbroeck, Jeroen (2017). VP-ellipsis, in Martin Everaert and Henk van Riems­ dijk (eds), The Wiley Blackwell companion to syntax, 2nd edition. Hoboken: Wiley-Black­ well, 1–35. van Craenenbroeck, Jeroen and Marcel den Dikken (2006). Ellipsis and EPP repair, Lin­ guistic Inquiry 37(4): 653–64. van Craenenbroeck, Jeroen and Anikó Lipták (2005). Ellipsis in Hungarian and the typolo­ gy of sluicing, in Kiyong Choi and Changguk Kim (eds), Ellipsis in minimalism: Proceed­ ings of the 7th Seoul International Conference on Generative Grammar. Seoul: Hankook, 103–33. van Craenenbroeck, Jeroen and Anikó Lipták (2006). The crosslinguistic syntax of sluic­ ing: Evidence from Hungarian relatives, Syntax 9(3): 248–74. van Craenenbroeck, Jeroen and Anikó Lipták (2008). On the interaction between verb movement and ellipsis: New evidence from Hungarian, in Charles B. Chang and Hannah J. Haynie (eds), Proceedings of the 26th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 138–46.

Page 25 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References van Craenenbroeck, Jeroen and Anikó Lipták (2013). What sluicing can do, what it can’t and in which language: On the cross-linguistic syntax of ellipsis, in Lisa L.-S. Cheng and Norbert Corver (eds), Diagnosing syntax. New York: Oxford University Press, 502–36. van Craenenbroeck, Jeroen and Jason Merchant (2013). Ellipsis phenomena, in Marcel den Dikken (ed.), The Cambridge handbook of generative syntax. Cambridge: Cambridge University Press, 701–45. (p. 1054)

Cremers, Crit (2014). Syntax of Dutch: Ellipsis and grammar. Ms. Leiden University. Cresswell, Max J. (1976). The semantics of degree, in Barbara Partee (ed.), Montague grammar. New York: Academic Press, 261–92. Croft, William (2001). Radical Construction Grammar: Syntactic theory in typological per­ spective. New York: Oxford University Press. Cruse, Alan (1977). The pragmatics of lexical specificity, Journal of Linguistics 13: 153–64. Crysmann, Berthold (2008). An asymmetric theory of peripheral sharing in HPSG, in Ger­ ald Penn (ed.), Proceedings of FG Vienna: The 8th Conference on Formal Grammar. Stan­ ford: CSLI Publications, 45–64. Culicover, Peter (2013a). English (zero-)relatives and the competence-performance dis­ tinction, International Review of Pragmatics 5: 253–70. Culicover, Peter (2013b). Grammar and complexity. Oxford: Oxford University Press. Culicover, Peter and Ray Jackendoff (2005). Simpler Syntax. Oxford: Oxford University Press. Culicover, Peter and Ray Jackendoff (2012). A domain-general cognitive relation and how language expresses it, Language 88: 305–40. Culicover, Peter and Paul Postal (eds) (2001). Parasitic gaps. Cambridge, MA: MIT Press. Culicover, Peter and Susanne Winkler (2008). English focus inversion, Journal of Linguis­ tics 44(3): 625–58. Cutler, Anne and Jerry A. Fodor (1979). Semantic focus and sentence comprehension, Cognition 7: 49–59. Cyrino, Sonia (2004). On the existence of Null Complement Anaphora in Brazilian Por­ tuguese, Revista Letras 63: 97–117. Cyrino, Sonia and Gabriela Matos (2002). VP-ellipsis in European and Brazilian Por­ tuguese: A comparative analysis, Journal of Portuguese Linguistics 1: 177–95. Cyrino, Sonia and Gabriela Matos (2005). Local licensers and recovering in VP ellipsis, Journal of Portuguese Linguistics 4: 79–112. Page 26 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Cyrino, Sonia and Gabriela Matos (2006). Null Complement Anaphora in Romance: Deep or surface anaphora? in Jenny Doetjes and Paz González (eds), Romance Languages and Linguistic Theory 2004. Amsterdam and Philadelphia: John Benjamins, 95–120. Dadan, Marcin (2016). Preposition omission in sluicing: Teasing apart LF-copying and PFdeletion, in Christopher Hammerly and Brandon Prickett (eds), NELS 46: Proceedings of the 46th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 251– 60. Dagnac, Anne (2010). Modal ellipsis in French, Spanish and Italian: Evidence for a TPdeletion analysis, in Karlos Arregi, Zsuzsanna Fagyal, Silvina Montrul, and Mireille Trem­ blay (eds), Romance linguistics 2008: Interactions in Romance. Amsterdam and Philadel­ phia: John Benjamins, 157–70. Dagnac, Anne (2016). Gapping needs vP-coordination: An argument from French NPI li­ censing, The Linguistic Review 33(4): 503–30. Dahl, Östen (1973). On so-called sloppy identity, Synthese 26: 81–112. Dahl, Östen (1974). How to open a sentence: Abstraction in natural language, Logical Grammar Reports 12. University of Götenburg. Dalrymple, Mary (1991). Against reconstruction in ellipsis. Unpublished ms. Xerox-PARC and CSLI. Dalrymple, Mary, Stuart M. Shieber, and Fernando C. N. Pereira (1991). Ellipsis and high­ er-order unification, Linguistics and Philosophy 14(4): 399–452. Davidson, Kathryn (2013). ‘And’ and ‘or’: General use coordination in ASL, Semantics and Pragmatics 6: 1–44. Davies, Mark (2008). The Corpus of Contemporary American English: 450 million words, 1990–present. Available online at . Dayal, Veneeta and Roger Schwarzschild (2010). Definite inner antecedents and wh-correlates in sluices, Rutgers Working Papers in Linguistics 3: 92–114. (p. 1055)

Dékány, Éva (2011). A profile of the Hungarian DP: The interaction of lexicalization, agreement and linearization with the functional sequence. PhD thesis, University of Tromsø. Dékány, Éva (2015). The syntax of anaphoric possessives in Hungarian, Natural Language and Linguistic Theory 33(4): 1121–68. Depiante, Marcela (2000). The syntax of deep and surface anaphora: A study of null com­ plement anaphora and stripping/bare argument ellipsis. PhD thesis, University of Con­ necticut at Storrs.

Page 27 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Depiante, Marcela (2001). On null complement anaphora in Spanish and Italian, Probus 13(2): 193–221. Depiante, Marcela and Jorge Hankamer (2005). Non-constituent ellipsis. Talk presented at the Identity in Ellipsis workshop, University of California, Berkeley. Depiante, Marcela and Pascual Masullo (2001). Género y número en la elipsis nominal: Consecuencias para la hipótesis lexicalista. Talk given at the I Encuentro de gramática Generativa, National University of Comahue. Desmets, Marianne (2001). Les typages de la phrase en HPSG: Le cas des phrases en ‘omme’. PhD thesis, University of Paris 10. Desmets, Marianne (2010). Ellipses dans les constructions comparatives en ‘comme’, Linx 58: 47–74. Dickey, Michael Walsh and Ann C. Bunger (2011). Comprehension of elided structure: Evi­ dence from sluicing, Language and Cognitive Processes 26(1): 63–78. Dik, Simon (1986). Coordination: Its implications for the theory of general linguistics. Am­ sterdam: North-Holland. den Dikken, Marcel (2004). Agreement and ‘clause union’, in Katalin É. Kiss and Henk van Riemsdijk (eds), Verb clusters: A study of Hungarian, German and Dutch. Amsterdam and Philadelphia: John Benjamins, 445–98. den Dikken, Marcel (2006). Either-float and the syntax of co-or-dination, Natural Lan­ guage and Linguistic Theory 24(3): 689–749. den Dikken, Marcel (2011). The distributed morphology of codeswitching. Talk given at the UIC Bilingualism Forum, University of Illinois at Chicago. den Dikken, Marcel and Shoba Bando-Rao (2014). On v as pivot in codeswitches, and the nature of the ban on word-internal switches, in Jeff MacSwan (ed.), Grammatical theory and bilingual codeswitching. Cambridge, MA: MIT Press, 161–84. den Dikken, Marcel, André Meinunger, and Chris Wilder (2000). Pseudoclefts and ellipsis, Studia Linguistica 54: 41–89. Dillon, Brian, Alan Mishler, Shayne Sloggett, and Colin Phillips (2013). Contrasting intru­ sion profiles for agreement and anaphora: Experimental and modeling evidence, Journal of Memory and Language 69: 85–103. Dixon, Robert M. W. (1994). Ergativity. Cambridge: Cambridge University Press. Dixon, Robert M. W. (2010). Basic linguistic theory, vol. 1: Methodology. Oxford: Oxford University Press. Dobrovie-Sorin, Carmen (1993). The syntax of Romanian. The Hague: Mouton. Page 28 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Donati, Caterina (1997). Comparative clauses as free relatives: A raising analysis, Probus 9(2): 145–66. Doron, Edit (1999). V-movement and VP ellipsis, in Shalom Lappin and Elabbas Ben­ mamoun (eds), Studies in ellipsis and gapping. New York: Oxford University Press, 124– 40. Dougherty, Ray C. (1970). A grammar of coordinate conjoined structures, I, Language 46(4): 850–98. Dougherty, Ray C. (1971). A grammar of coordinate conjoined structures, II, Language 47(1): 298–339. Dowty, David (1988). Type raising, functional composition, and non-constituent conjunc­ tion, in Richard Oehrle, Emmon Bach, and Deirdre Wheeler (eds), Categorial grammars and natural language structures. Dordrecht: D. Reidel, 153–97. Dowty, David, Robert Wall, and Stanley Peters (1981). Introduction to Montague grammar. Dordrecht: D. Reidel. (p. 1056)

Drai, Dan and Yosef Grodzinsky (2006a). A new empirical angle on the variability

debate: Quantitative neurosyntactic analyses of a large data set from Broca’s aphasia, Brain and Language 76: 117–28. Drai, Dan and Yosef Grodzinsky (2006b). The variability debate: More statistics, more lin­ guistics, Brain and Language 76: 157–70. Drummond, Alex (2013). Dahl’s paradigm: In defense of the crossover analysis. Poster presented at NELS 44: the 44th annual meeting of the North East Linguistic Society. Uni­ versity of Connecticut. Drummond, Alex, Norbert Hornstein, and Howard Lasnik (2010). A puzzle about P-strand­ ing and a possible solution, Linguistic Inquiry 41(4): 689–92. Dryer, Matthew S. (2006). Descriptive theories, explanatory theories, and Basic Linguistic Theory, in Felix K. Ameka, Alan Dench, and Nicholas Evans (eds), Catching language: Is­ sues in grammar writing. Berlin: Mouton de Gruyter, 207–34. Duffield, Nigel and Ayumi Matsuo (2001). A comparative study of ellipsis and anaphora in L2 acquisition, in Anna H.-J. Do, Laura Domínguez, and Aimee Johansen (eds), Proceed­ ings of the 25th Boston University Conference on Child Language Development. Somerville, MA: Cascadilla Press, 238–49. Dvořák, Bostjan (2007). Slovenian clitic pronouns and what is so special about them, Slovene Linguistic Studies 6: 209–33. É. Kiss, Katalin (1987). Configurationality in Hungarian. Budapest: Akadémiai Kiadó.

Page 29 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References É. Kiss, Katalin (1993). Wh-movement and specificity, Natural Language and Linguistic Theory 11: 83–120. É. Kiss, Katalin (ed.) (1995). Discourse configurational languages. Oxford: Oxford Univer­ sity Press. É. Kiss, Katalin (1998). Identificational focus and information focus, Language 74: 245– 73. É. Kiss, Katalin (2002). The syntax of Hungarian. Cambridge: Cambridge University Press. É. Kiss, Katalin (2008). Free word order, (non)configurationality, and phases, Linguistic Inquiry 39(3): 441–75. É. Kiss, Katalin (2010). An adjunction analysis of quantifiers and adverbials in the Hun­ garian sentence, Lingua 120(3): 506–26. Edelstein, Elspeth (2014). This syntax needs studied, in Lawrence Horn and Rafaella Zanuttini (eds), Microsyntactic variation in North American English. Oxford: Oxford Uni­ versity Press, 242–68. Edwards, John (2004). Foundations of bilingualism, in Tej K. Bhatia and William C. Ritchie (eds), Handbook of bilingualism. Malden, MA: Blackwell, 7–31. Egg, Markus and Katrin Erk (2002). A compositional account of VP-ellipsis, in Frank van Eynde, Lars Hellan, and Dorothee Beerman (eds), Proceedings of the 8th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, 162–79. Eguren, Luis (2010). Contrastive focus and nominal ellipsis in Spanish, Lingua 120(2): 435–57. Eichler, Nadine (2011). Code-Switching bei bilingual aufwachsenden Kindern: Eine Analyse der gemischtsprachlichen Nominalphrasen unter besonderer Berücksichtigung des Genus. Tübingen: Narr Verlag. Eichler, Nadine and Natascha Müller (2012). The derivation of mixed DPs: Mixing of func­ tional categories in bilingual children and in second language learners, in Marzena Wa­ torek, Sandra Benazzo, and Maya Hickmann (eds), Comparative perspectives on language acquisition: A tribute to Clive Perdue. Bristol: Multilingual Matters, 263–81. Eisenberg, Peter (1973). A note on ‘identity of constituents’, Linguistic Inquiry 4(3): 417– 20. Elbourne, Paul (2001). E-type anaphora as NP deletion, Natural Language Semantics 9(3): 241–88. Elbourne, Paul (2005). Situations and individuals. Cambridge, MA: MIT Press. Page 30 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Elbourne, Paul (2008). Ellipsis sites as definite descriptions, Linguistic Inquiry 39(2): 191– 220. Elbourne, Paul (2013). Definite descriptions. Oxford: Oxford University Press. Elliot, Patrick, Andreea Nicolae, and Yasutada Sudo (2014). VP ellipsis without parallel binding: A QUD approach. Poster presented at SALT (Semantics and Linguistic Theory Conference) 24. Embick, David and Alec Marantz (2008). Architecture and blocking, Linguistic Inquiry 39(1): 1–53. Emonds, Joseph E. (1978). The verbal complex V’-V in French, Linguistic Inquiry 9: 151–75. (p. 1057)

Emonds, Joseph E. (1985). A unified theory of syntactic categories. Dordrecht: Foris Pub­ lications. Engberg-Pedersen, Elisabeth (1993). Space in Danish Sign Language: The space and mor­ phosyntax of the use of space in a visual language. Hamburg: Signum Verlag. Engberg-Pedersen, Elisabeth (2002). Grammatical relations in Danish Sign Language, in Anneli Pajunen (ed.), Mimesis, sign, and the evolution of language. Turku: University of Turku Press, 5–40. Engdahl, Elisabet (1983). Parasitic gaps, Linguistics and Philosophy 6(1): 5–34. Engdahl, Elisabet (1986). Constituent questions. Dordrecht: D. Reidel. Engel, Ulrich (1994). Syntax der deutschen Gegenwartssprache. Berlin: Erich Schmidt. Epstein, Samuel (1995). A derivational approach to syntactic relations. Unpublished man­ uscript, Harvard University. Eroms, Hans-Werner (2000). Syntax der deutschen Sprache. Berlin: Walter de Gruyter. Erschler, David (2016). On embedded gapping. Talk presented at IATL 32: The 32nd an­ nual meeting of the Israel Association for Theoretical Linguistics. Erteschik, Naomi (1973). On the nature of island constraints. PhD dissertation, MIT. Erteschik-Shir, Nomi (1998). The syntax-focus interface, in Peter W. Culicover and Louise McNally (eds), The limits of syntax. San Diego: Academic Press, 211–40. Erteschik-Shir, Nomi (2007). The syntax-discourse interface. Oxford: Oxford University Press. Erteschik-Shir, Nomi, Lena Ibn-Bari, and Sharon Taube (2011). Object ellipsis as topic drop. Ms. Ben Gurion University of the Negev. Page 31 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Erteschik-Shir, Nomi, Lena Ibn-Bari, and Sharon Taube (2013). Missing objects as topic drop, Lingua 136: 145–69. Erteschik-Shir, Nomi and Shalom Lappin (1979). Dominance and the functional explana­ tion of island phenomena, Theoretical Linguistics 6: 41–86. Eshghi, Arash, Christine Howes, Eleni Gregoromichelaki, Julian Hough, and Matthew Purver (2015). Feedback in conversation as incremental semantic update, in Proceedings of the 11th International Conference on Computational Semantics. Stroudsburg, PA: Asso­ ciation for Computational Linguistics, 261–71. Evans, Frederic (1988). Binding into anaphoric verb phrases, in Joyce Powers and Ken­ neth de Jong (eds), Proceedings of the 5th Annual Eastern States Conference on Linguis­ tics. Ohio: ESCOL Publications Committee, Linguistics Department, Ohio State Universi­ ty, 122–9. Evans, N. and S. C. Levinson (2009). The myth of language universals: Language diversity and its importance for cognitive science, Behavioral and Brain Sciences 32(5), 429–48. Fanselow, Gisbert and Stefan Frisch (2006). Effects of processing difficulty on judgments of acceptability, in Gisbert Fanselow, Caroline Féry, Ralf Vogel, and Matthias Schlesewsky (eds), Gradience in grammar. Oxford: Oxford University Press, 291–316. Farkas, Donka (2009). Polarity particles in Hungarian, in Marcel den Dikken and Robert Vago (eds), Approaches to Hungarian 11. Amsterdam and Philadelphia: John Benjamins, 95–118. Farkas, Donka and Floris Roelofsen (forthcoming). Polar initiatives and polarity particle responses in an inquisitive discourse model. Language. Accepted subject to minor revi­ sions. Farudi, Annahita (2013). Gapping in Farsi: A cross-linguistic investigation. PhD thesis, University of Massachusetts at Amherst. Fernández, Raquel (2006). Non-sentential utterances in dialogue: Classification, resolu­ tion and use. PhD thesis, King’s College. Fernández, Raquel and Jonathan Ginzburg (2002). Non-sentential utterances: A corpus study, Traitement automatique des languages 43(2): 13–42. Fernández, Raquel, Jonathan Ginzburg, and Shalom Lappin (2004). Classifying ellipsis in dialogue: A machine learning approach, in COLING’04: Proceedings of the 20th Interna­ tional Conference on Computational Linguistics. Stroudsburg, PA: Association for Compu­ tational Linguistics, 240–6. Fernández, Raquel, Jonathan Ginzburg, and Shalom Lappin (2007). Classifying non-sen­ tential utterances in dialogue: A machine learning approach, Computational Linguistics 33(3): 397–427. Page 32 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Ferrara, Lindsay and Trevor Johnston (2014). Elaborating who’s what: A study of constructed action and clause structure in Auslan (Australian Sign Language), Australian Journal of Linguistics 34: 193–215. (p. 1058)

Féry, Caroline (1994). Prosodische und tonale Faktoren bei der Disambiguierung syntak­ tischer Strukturen, in Karl H. Ramers, Heinz Vater, and Henning Wode (eds), Universale phonologische Strukturen und Prozesse. Tübingen: Niemeyer, 97–114. Féry, Caroline and Katharina Hartmann (2005). The focus and prosodic structure of Ger­ man Right Node Raising and gapping, The Linguistic Review 22(1): 69–116. Féry, Caroline and Shinichiro Ishihara (eds) (2016). The Oxford Handbook of Information Structure. Oxford and New York: Oxford University Press. Fiengo, Robert (1980). Surface structure. Cambridge, MA: Harvard University Press. Fiengo, Robert and Robert May (1994). Indices and identity. Cambridge, MA: MIT Press. Fillmore, Charles (1969). Types of lexical information, in Ferenc Kiefer (ed.), Studies in syntax and semantics. Dordrecht: D. Reidel, 109–37. Fillmore, Charles (1986). Pragmatically controlled zero anaphora, in Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguis­ tics Society, 95–107. Fillmore, Charles, Paul Kay, and Mary Catherine O’Connor (1988). Regularity and id­ iomaticity in grammatical constructions: The case of ‘let alone’, Language 64(3): 501–38. Filtopoulos, Rebecca (2015). Dialogical analysis of theatre: A corpus study of non-senten­ tial utterances in Waiting for Godot and Endgame. Master’s thesis, Université Paris Diderot. von Fintel, Kai (1999). NPI licensing, Strawson entailment and context dependency, Jour­ nal of Semantics 16(2): 97–148. Fischer, Klaus (1997). German-English verb valency: A contrastive analysis. Tübingen: Narr Verlag. Fitzpatrick, Justin M. (2006). Deletion through movement, Natural Language and Linguis­ tic Theory 24(2): 399–431. Fletcher, Charles (1994). Levels of representation in memory for discourse, in Morton Ann Gernsbacher (ed.), Handbook of psycholinguistics. San Diego: Academic Press, 589– 607. Fodor, Janet Dean and Ivan Sag (1982). Referential and quantificational indefinites, Lin­ guistics and Philosophy 5: 355–98.

Page 33 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Fodor, Jerry A., Thomas G. Bever, and Merrill F. Garrett (1974). The psychobiology of lan­ guage. New York: McGraw-Hill Publishers. Foley, Claire, Zelmira Núñez del Prado, Isabella Barbier, and Barbara Lust (2003). Knowl­ edge of variable binding in VP-ellipsis: Language acquisition research and theory con­ verge, Syntax 6(1): 52–83. Folli, Raffaella, Heidi Harley, and Simin Karimi (2005). Determinants of event type in Per­ sian complex predicates, Lingua 115: 1365–1401. Fortin, Catherine (2007a). Some (not all) nonsententials are only a phase, Lingua 117(1): 67–94. Fortin, Catherine (2007b). Indonesian sluicing and verb phrase ellipsis: Description and explanation in the Minimalist framework. PhD thesis, University of Michigan. Fortin, Catherine (2011). We need LF-copying: A few good reasons why, in Mary Byram Washburn, Sarah Ouwayda, Chuoying Ouyang, Bin Yin, Canan Ipek, Lisa Marston, and Aaron Walker (eds), Proceedings of the 28th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 87–95. Fortin, Catherine (2013). Indonesian sluicing is LF copying. Ms. Carleton College. Fortin, Catherine (2014). The distribution of [E]: A view from Indonesian verb phrase el­ lipsis, in Joseph Sabbagh (ed.), Proceedings of the 20th Meeting of the Austronesian For­ mal Linguistics Association (AFLA 20). London: University of Western Ontario. Fox, Danny (1995). Economy and scope, Natural Language Semantics 3(3): 283–341. Fox, Danny (1999). Reconstruction, binding theory, and the interpretation of chains, Lin­ guistic Inquiry 30(2): 157–96. (p. 1059)

Fox, Danny (2000). Economy and semantic interpretation. Cambridge, MA: MIT

Press. Fox, Danny (2002). Antecedent-contained deletion and the copy theory of movement, Lin­ guistic Inquiry 33(1): 63–96. Fox, Danny and Howard Lasnik (2003). Successive-cyclic movement and island repair: The difference between sluicing and VP-ellipsis, Linguistic Inquiry 34(1): 143–54. Fox, Danny, and John Nissenbaum (1999). Extraposition and scope: A case for overt QR. Paper presented at WCCFL 18: The 18th West Coast Conference on Formal Linguistics. Fraser, Bruce (1970). Vice versa, Linguistic Inquiry 1(2): 277–8. Frauenfelder, Uli and Caroline Floccia (2006). Das Erkennen gesprochener Wörter, in An­ gela D. Friederici (ed.), Sprachrezeption. Göttingen: Hofgrefe, 1–48.

Page 34 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Frazier, Lyn (2008). Processing ellipsis: A processing solution to the undergeneration problem? in Charles B. Chang and Hannah J. Haynie (eds), Proceedings of the 26th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 21–32. Frazier, Lyn (2015). Two interpretive systems for natural language? Journal of Psycholin­ guistic Research 44(1): 7–25. Frazier, Lyn and Charles Clifton, Jr (1998). Comprehension of sluiced sentences, Lan­ guage and Cognitive Processes 13: 499–520. Frazier, Lyn and Charles Clifton, Jr (2000). On bound-variable interpretations: The LF-on­ ly hypothesis, Journal of Psycholinguistic Research 29: 125–39. Frazier, Lyn and Charles Clifton, Jr (2001). Parsing coordinates and ellipsis: Copy a, Syn­ tax 4(1): 1–22. Frazier, Lyn and Charles Clifton, Jr (2005). The syntax-discourse divide: Processing ellip­ sis, Syntax 8(2): 121–74. Frazier, Lyn and Charles Clifton Jr (2006). Ellipsis and discourse coherence, Linguistics and Philosophy 35: 315–46. Frazier, Lyn and Charles Clifton Jr (2011). Dynamic interpretations: Finding an an­ tecedent for VPE, in Jesse A. Harris and Margaret Grant (eds), University of Massachu­ setts occasional papers in linguistics: Processing linguistic structure. Amherst, MA: GLSA, 21–35. Frazier, Lyn and Charles Clifton Jr (2015). Without his shirt off he saved the child from al­ most drowning: Intepreting uncertain input, Language, Cognition and Neuroscience 30(6): 635–47. Frazier, Lyn, Charles Clifton Jr, and Katy Carlson (2007). Focus and VP ellipsis, Language and Speech 50: 1–21. Frazier, Lyn, Lori Taft, Tom Roeper, Charles Clifton Jr, and Kate Erlich (1984). Parallel structure: A source of facilitation in sentence comprehension, Memory and Cognition 12: 421–30. Frederking, Robert E. (1993). Understanding dialogue ellipsis. Pittsburgh: Center for Ma­ chine Translation, Carnegie Mellon University. Fried, Mirjam (2009). Construction grammar as a tool for diachronic analysis, Construc­ tions and Frames 1(2): 262–91. Friederici, Angela D. (2002). Towards a neural basis of auditory sentence processing, Trends in Cognitive Sciences 6: 78–84. Friedman, Lynn (1976). The manifestation of subject, object and topic in American Sign Language, in Charles N. Li (ed.), Subject and topic. New York: Academic Press, 125–48. Page 35 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Fu, Jingqi, Tom Roeper, and Hagit Borer (2001). The VP within process nominals: Evi­ dence from adverbs and the VP anaphor do so, Natural Language and Linguistic Theory 19(3): 549–82. Fukaya, Teruhiko (2003). Island (in)sensitivity in Japanese sluicing and stripping, in Gina Garding and Mimu Tsujimura (eds), Proceedings from the 22th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 179–92. Fukaya, Teruhiko (2007). Sluicing and stripping in Japanese and some implications. PhD thesis, USC. Fukaya, Teruhiko (2012). Island-sensitivity in Japanese sluicing and some implications, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Ox­ ford: Oxford University Press, 123–63. Fukaya, Teruhiko and Hajime Hoji (1999). Stripping and sluicing in Japanese and some implications, in S. Bird, A. Carnie, J. Haugen, and P. Norquest (eds), Proceedings of the 18th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 145–58. (p. 1060)

Fukui, Naoki (1986). A theory of category projection and its application. PhD thesis, Massachusetts Institute of Technology. Fukui, Naoki and Margaret Speas (1986). Specifiers and projection, in Naoki Fukui, Tova R. Rapoport, and Elizabeth Sagey (eds), Papers in theoretical linguistics: MIT working pa­ pers in linguistics 8. Cambridge, MA: MIT Working Papers in Linguistics, 128–72. Funakoshi, Kenshi (2012). On headless XP-movement/ellipsis, Linguistic Inquiry 43: 519– 62. Funakoshi, Kenshi (2014). Syntactic head movement and its consequences. PhD thesis, University of Maryland, College Park. Funakoshi, Kenshi (2016). Verb-stranding verb phrase ellipsis in Japanese, Journal of East Asian Linguistics 25: 113–42. Gagnon, Michaël (2013a). Anaphors and the missing link. PhD thesis, University of Mary­ land, College Park. Gagnon, Michaël (2013b). Part and parcel of eliding partitives, SALT 23: Proceedings of the 23rd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 316– 35. Gajewski, Jon (2002). L-analyticity and natural language. Ms. MIT. Gärdenfors, Peter (2014). The geometry of meaning: Semantics based on conceptual spaces. Cambridge, MA: MIT Press.

Page 36 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Gardent, Claire (1991). Gapping and VP ellipsis in a unification-based grammar. PhD the­ sis, University of Edinburgh. Gardent, Claire and Michael Kohlhase (1996). Higher-order coloured unification and nat­ ural language semantics, in Proceedings of the 34th Annual Meeting on Association of Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 1– 9. Gardent, Claire and Michael Kohlhase (1997). Computing parallelism in discourse, IJCAI’97: Proceedings of the Fifteenth International Joint Conference on Artifical Intelligence. San Francisco: Morgan Kaufmann Publishers, 1016–21. Gardent, Claire and Michael Kohlhase (1998). Higher-order colored unification: A linguis­ tic application, Technique et Science Informatique 18: 181–209. Gargett, Andrew, Eleni Gregoromichelaki, Chris Howes, and Yo Sato (2008). Dialoguegrammar correspondence in Dynamic Syntax, in Jonathan Ginzburg, Pat Healey, and Yo Sato (eds), Proceedings of the 12th Workshop on the Semantics and Pragmatics of Dia­ logue (SEMDIAL), 37–44. Gargett, Andrew, Eleni Gregoromichelaki, Ruth Kempson, Matthew Purver, and Yo Sato (2009). Grammar resources for modeling dialogue dynamically, Cognitive Neurodynamics 3(4): 347–63. Gawron, Jean Mark (1995). Comparatives, superlatives, and resolution, Linguistics and Philosophy 18(4): 333–80. Gawron, Jean Mark and Andrew Kehler (2004). The semantics of respective readings, conjunction, and filler-gap dependencies, Linguistics and Philosophy 27(2): 169–207. Gawron, Jean Mark and Stanley Peters (1990). Anaphora and quantification in situation semantics. Stanford University: Center for the Study of Language and Information. van Gelderen, Elly and Jeff MacSwan (2008). Interface conditions and code-switching: Pronouns, lexical DPs, and checking theory, Lingua 118(6): 765–76. Gengel, Kirsten (2007). Focus and ellipsis: A generative analysis of pseudogapping and other elliptical structures. PhD thesis, University of Stuttgart. Gengel, Kirsten (2013). Pseudogapping and ellipsis. Oxford: Oxford University Press. Gennari, Silvia P. and Maryellen C. MacDonald (2006). Acquisition of negation and quan­ tification: Insights from adult production and comprehension, Language Acquisition 13(2): 125–68. Gergel, Remus (2004). Short-distance reanalysis of Middle English modals: Evidence from ellipsis, Studia Linguistica 58(2): 53–87.

Page 37 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Gergel, Remus (2006). Interpretable features in vP-ellipsis: On the licensing head, in S. Blaho, E. Schoorlemmer, and L. Vicente (eds), Proceedings of ConSOLE XIV. Leiden: SOLE, 165–88. (p. 1061)

Gergel, Remus (2009). Modality and ellipsis: Diachronic and synchronic evidence. Berlin and New York: Mouton de Gruyter. Gergel, Remus, Kirsten Gengel, and Susanne Winkler (2007). Ellipsis and inversion: A fea­ ture-based account, in Kerstin Schwabe and Susanne Winkler (eds), On information struc­ ture, meaning and form. Amsterdam and Philadelphia: John Benjamins, 301–22. Ghaniabadi, Saeed (2010). The empty noun construction in Persian. PhD thesis, Universi­ ty of Manitoba. Giannakidou, Anastasia and Jason Merchant (1998). Reverse sluicing in English and Greek, The Linguistic Review 15: 233–56. Giannakidou, Anastasia and Melita Stavrou (1999). Nominalization and ellipsis in the Greek DP, The Linguistic Review 16: 295–331. Gibson, Edward, Pauline Jacobson, Peter Graff, Kyle Mahowald, Evelina Fedorenko, and Steven T. Piantadosi (2015). A pragmatic account of complexity in definite antecedentcontained-deletion relative clauses, Journal of Semantics 32(4): 579–618. Ginzburg, Jonathan (1992). Questions, queries and facts: A semantics and pragmatics for interrogatives. PhD thesis, Stanford University, Stanford. Ginzburg, Jonathan (1994). An update semantics for dialogue, in Harry Bunt (ed.), Pro­ ceedings of the 1st International Workshop on Computational Linguistics. Tilburg: ITK. Ginzburg, Jonathan (1995). Resolving questions, I, Linguistics and Philosophy 18(5): 459– 527. Ginzburg, Jonathan (1999). Semantically-based ellipsis resolution with syntactic presup­ positions, in Harry Bunt and Reinhard Muskens (eds), Computing meaning. Springer, 255–79. Ginzburg, Jonathan (2012). The interactive stance: Meaning for conversation. Oxford: Ox­ ford University Press. Ginzburg, Jonathan and Robin Cooper (2004). Clarification, ellipsis, and the nature of contextual updates in dialogue, Linguistics and Philosophy 27(3): 297–365. Ginzburg, Jonathan and Robin Cooper (2014). Quotation via dialogical interaction, Journal of Logic, Language, and Information 23(3): 287–311. Ginzburg, Jonathan, Robin Cooper, and Tim Fernando (2014a). Propositions, questions, and adjectives: A rich type-theoretic approach, in Proceedings of the EACL 2014 Work­ Page 38 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References shop on Type Theory and Natural Language Semantics (TTNLS). Gothenburg: Association for Computational Linguistics, 89–96. Ginzburg, Jonathan, Robin Cooper, Julian Hough, and David Schlangen (2017). Incremen­ tality and clarification/sluicing potential, in Proceedings of Sinn und Bedeutung 21. Edin­ burgh. Ginzburg, Jonathan and Raquel Fernández (2010). Computational models of dialogue, in Alex Clark, Chris Fox, and Shalom Lappin (eds), Handbook of computational linguistics and natural language processing. Oxford: Blackwell, 429–81. Ginzburg, Jonathan, Raquel Fernández, and David Schlangen (2014b). Disfluencies as in­ tra-utterance dialogue moves, Semantics and Pragmatics 7(9): 1–64. Ginzburg, Jonathan and Dimitra Kolliakou (2009). Answers without questions: The emer­ gence of fragments in child language, Journal of Linguistics 45: 641–73. Ginzburg, Jonathan and Massimo Poesio (2016). Grammar is a system that characterizes talk in interaction, Frontiers in Psychology 7: 1–22. Ginzburg, Jonathan and Ivan Sag (2000). Interrogative investigations. Stanford: CSLI Publications. Gleitman, Lila (1965). Coordinating conjunctions in English, Language 41: 260–93. Göksun, Tilbe, Tom Roeper, Kathy Hirsh-Pasek, and Roberta M. Golinkoff (2011). From noun phrase ellipsis to verb phrase ellipsis: The acquisition path from context to abstract reconstruction, in Jassie A. Harris and Margaret Grant (eds), Processing linguistic struc­ ture. Amherst, MA: GLSA, 53–74. Goldberg, Adele E. (1995). Constructions: A Construction Grammar approach to argument structure. Chicago: University of Chicago Press. (p. 1062)

Goldberg, Adele E. (2002). Surface generalizations: An alternative to derivations, Cogni­ tive Linguistics 4: 327–56. Goldberg, Adele E. (2013). Backgrounded constituents cannot be ‘extracted’, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, 221–38. Goldberg, Lotus Madelyn (2005). Verb stranding VP-ellipsis: A cross-linguistic study. PhD thesis, McGill University. Goldsmith, John (1985). A principled exception to the Coordinate Structure Constraint, in William H. Eilfort, Paul D. Kroeber, and Karen L. Peterson (eds), Proceedings of the 21st Annual Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 133–43. Goldsmith, John (1990). Autosegmental and metrical phonology. Oxford: Blackwell. Page 39 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References González-Vilbazo, Kay (2005). Die Syntax des Code-Switching. PhD thesis, University of Cologne. González-Vilbazo, Kay, Laura Bartlett, Sarah Downey, Shane Ebert, Jeanne Heil, Brad Hoot, Bryan Koronkiewicz, and Sergio Ramos (2013). Methodological considerations in code-switching research, Studies in Hispanic and Lusophone Linguistics 6(1): 119–38. González-Vilbazo, Kay and Bryan Koronkiewicz (2016). Tú y yo can codeswitch, nosotros cannot: Pronouns in Spanish-English codeswitching, in Rosa E. Guzzardo Tamargo, Catherine M. Mazak, and M. Carmen Parafita Couto (eds), Spanish-English codeswitching in the Caribbean and the US. Amsterdam and Philadelphia: John Benjamins, 237–60. González-Vilbazo, Kay and Luis López (2011). Some properties of light verbs in codeswitching, Lingua 121(5): 832–50. González-Vilbazo, Kay and Luis López (2012). Little v and parametric variation, Natural Language and Linguistic Theory 30(1): 33–77. González-Vilbazo, Kay and Sergio Ramos (2014). A morphosyntactic condition on sluicing: Evidence from Spanish/German code-switching. Ms. University of Illinois at Chicago. González-Vilbazo, Kay and Volker Struckmeier (2008). Asymmetrien im code-switching: Eine DM-Lösung zur Partizipselektion, in Eva-Maria Remberger and Guido Menshing (eds), Romanistische Sprachwissenschaft–minimalistisch. Tübingen: Narr Verlag, 83–102. Goodall, Grant (1987). Parallel structures in syntax. Cambridge: Cambridge University Press. Goodglass, Harold and Edith Kaplan (1972). The assessment of aphasia and related disor­ ders. Philadelphia: Lea & Febiger. Gračanin-Yüksek, Martina (2007). About sharing. PhD thesis, MIT. Grant, Margaret (2012). The parsing and interpretation of comparatives: More than meets the eye. PhD thesis, University of Massachusetts at Amherst. Grant, Margaret, Charles Clifton Jr, and Lyn Frazier (2012). The role of non-actuality im­ plicatures in processing elided constituents, Journal of Memory and Language 66(1): 326– 43. Grebenyova, Lydia (2004). Interpretation of Slavic multiple wh-questions, in Proceedings of the Annual Workshop on Formal Approaches to Slavic Linguistics (FASL) 12. Grebenyova, Lydia (2006a). Multiple interrogatives: Syntax, semantics, and learnability. PhD thesis, University of Maryland, College Park. Grebenyova, Lydia (2006b). Multiple left branch extraction under sluicing, in Proceedings of the 41st Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguis­ tic Society. Page 40 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Grebenyova, Lydia (2006c). Sluicing puzzles in Russian, in Hana Filip, Steven Franks, James Lavine, and Mila Tasseva-Kurktchieva (eds), Formal approaches to Slavic linguis­ tics: The Princeton meeting. Ann Arbor: Michigan Slavic Publications, 156–70. Grebenyova, Lydia (2007). Sluicing in Slavic, Journal of Slavic Linguistics 15(1): 49–80. Grebenyova, Lydia (2009). Sluicing and multiple wh-fronting, in Nguyen Chi Duy Khuong and Richa Samar Sinha (eds), Proceedings of GLOW in Asia 5. New Delhi: Central Insti­ tute of Indian Languages, 219–42. Grebenyova, Lydia (2012). Syntax, semantics and acquisition of multiple interrog­ atives. Amsterdam and Philadelphia: John Benjamins. (p. 1063)

Green, Lisa (2000). Aspectual BE-type constructions and coercion in African American English, Natural Language Semantics 8(1): 1–25. Green, Lisa (2002). African American English: A linguistic introduction. Cambridge: Cam­ bridge University Press. Green, Lisa (2011). Language and the African American child. Cambridge: Cambridge University Press. Gregoromichelaki, Eleni (2006). Conditionals: A Dynamic Syntax account. PhD thesis, King’s College London. Gregoromichelaki, Eleni (2012). Review of J. Ginzburg (2012), The interactive stance, Fo­ lia Linguistica 47: 293–316. Gregoromichelaki, Eleni (2013). Clitic left dislocation and clitic doubling: A dynamic per­ spective on left-right asymmetries in Greek, in Gert Webelhuth, Manfred Sailer, and Heike Walker (eds), Rightward movement in a comparative perspective. Amsterdam and Philadelphia: John Benjamins, 321–68. Gregoromichelaki, Eleni (2016). The case of non-sentential speech. Ms. University of Os­ nabrueck. Gregoromichelaki, Eleni (2017). Quotation in dialogue, in Paul Saka and Michael Johnson (eds), Semantic and pragmatic aspects of quotation. Berlin: Springer, 195–255. Gregoromichelaki, Eleni, Ronnie Cann, and Ruth Kempson (2013). On coordination in dia­ logue: Subsentential talk and its implications, in Laurence Goldstein (ed.), On brevity. Ox­ ford: Oxford University Press, 53–73. Gregoromichelaki, Eleni, Ruth Kempson, Matthew Purver, Greg J. Mills, Ronnie Cann, Wil­ fried Meyer-Viol, and Pat G. T. Healey (2011). Incrementality and intention-recognition in utterance processing, Dialogue and Discourse 2(1): 199–233.

Page 41 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Gregoromichelaki, Eleni, Yo Sato, Ruth Kempson, Andrew Gargett, and Christine Howes (2009). Dialogue modelling and the remit of core grammar, in Harry Bunt, Volha Petukho­ va, and Sander Wubben (eds), Proceedings of the Eighth International Conference on Computational Semantics (IWCS-8). Tilburg, 128–39. Gregory, Howard and Shalom Lappin (1999). Antecedent contained ellipsis in HPSG, in Gert Webelhuth, Jean-Pierre Koenig, and Andreas Kathol (eds), Lexical and constructional aspects of linguistic explanation. Stanford: CSLI, 331–56. Gribanova, Vera (2008). Exposing the Russian verbal complex via evidence from prefixa­ tion and verb phrase ellipsis. Paper presented at the 17th Annual Workshop on Formal Approaches to Slavic Linguistics, Yale University. Gribanova, Vera (2009a). Ellipsis and the syntax of verbs in Russian. Talk given at NYU. Gribanova, Vera (2009b). Prefixation, verb phrase ellipsis and the structure of the Russ­ ian verbal complex. Ms. UCSC. Gribanova, Vera (2009c). Structural adjacency and the typology of interrogative interpre­ tations, Linguistic Inquiry 40(1): 133–54. Gribanova, Vera (2010). On diagnosing ellipsis and argument drop: The view from Russ­ ian. Talk given at MIT. Gribanova, Vera (2013a). A new argument for verb-stranding verb phrase ellipsis, Linguis­ tic Inquiry 44(1): 145–57. Gribanova, Vera (2013b). Verb-stranding verb phrase ellipsis and the structure of the Russian verbal complex, Natural Language and Linguistic Theory 31(1): 91–136. Gribanova, Vera (2013c). Copular clauses, clefts, and putative sluicing in Uzbek, Lan­ guage 89: 830–82. Gribanova, Vera and Emily Manetta (2016). Ellipsis in wh-in-situ languages: Deriving ap­ parent sluicing in Hindi-Urdu and Uzbek, Linguistic Inquiry 47(4): 631–68. Grice, Martine, Stefan Baumann, and Ralf Benzmüller (2005). German intonation in au­ tosegmental-metrical phonology, in Sun-Ah Jun (ed.), Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press, 55–83. Grice, Paul (1975). Logic and conversation, in Peter Cole and Jerry Morgan (eds), Syntax and Semantics, vol. 3. New York: Academic Press, 22–40. (p. 1064)

Griffiths, James (2015). On appositives. PhD thesis, University of Groningen.

Griffiths, James and Anikó Lipták (2014). Contrast and island sensitivity in clausal ellipsis, Syntax 17: 189–234.

Page 42 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Grimshaw, Jane (1979). Complement selection and the lexicon, Linguistic Inquiry 10: 279– 326. Grimshaw, Jane (1991). Extended projection. Unpublished manuscript, Brandeis Universi­ ty. Grinder, John and Paul M. Postal (1971). Missing antecedents, Linguistic Inquiry 2(3): 269–312. Grodzinsky, Yosef (1984). Language deficits and linguistic theory. PhD thesis, Brandeis University. Grodzinsky, Yosef (1986). Language deficits and the theory of syntax, Brain and Language 27: 135–59. Grodzinsky, Yosef (2000). The neurology of syntax, Behavioral and Brain Sciences 23: 1– 71. Grodzinsky, Yosef (2005). Ellipsis in deficient language users: Parallels (and divergences) in parallelism. Plenary talk, Generative Approaches to Language Acquisition (GALA), Siena. Grodzinsky, Yosef (2011). Two puzzles in experimental syntax and semantics, in Proceed­ ings of the 47th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Lin­ guistic Society, 41–64. Grodzinsky, Yosef and Tanya Reinhart (1993). The innateness of binding and coreference, Linguistic Inquiry 24(1): 69–102. Grodzinsky, Yosef and Andrea Santi (2008). The battle for Broca’s region, Trends in Cog­ nitive Sciences 12(12): 474–80. Groenendijk, Jeroen (2007). Inquisitive Semantics: Two possibilities for disjunction, in Pe­ ter Bosch, David Gabelaia, and Jérôme Lang (eds), Logic, language, and computation. Heidelberg: Springer, 80–94. Groenendijk, Jeroen (2011). Erotetic languages and the inquisitive hierarchy, in J. van der Does and C. Dutilh Novaes (eds), This is not a Festschrift: A Festschrift for Martin Stokhof (online). Groenendijk, Jeroen and Floris Roelofsen (2009). Inquisitive semantics and pragmatics, in Jesús M. Larrazabal and Larraitz Zubeldia (eds), SPR-09: Proceedings of the ILCLI Inter­ national Workshop on Semantics, Pragmatics and Rhetoric. San Sebastián: Universidad del País Vasco. Groenendijk, Jeroen and Martin Stokhof (1984). The semantics of questions and the prag­ matics of answers. PhD thesis, University of Amsterdam.

Page 43 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Grootveld, Marjan Jeannette (1984). Parsing coordination generatively. PhD thesis, Lei­ den University. Grosjean, François (1982). Life with two languages: An introduction to bilingualism. Cam­ bridge, MA: Harvard University Press. Grosjean, François (1998). Studying bilinguals: Methodological and conceptual issues, Bilingualism: Language and Cognition 1(2): 131–49. Groß, Thomas (1999). Theoretical foundations of dependency syntax. Munich: Ludicium. Groß, Thomas and Timothy Osborne (2009). Toward a practical DG theory of discontinu­ ities, SKY Journal of Linguistics 22: 43–90. Groß, Thomas and Timothy Osborne (2013). Katena und Konstruktion: Ein Vorschlag zu einer dependen-ziellen Konstruktionsgrammatik, Zeitschrift für Sprachwissenschaft 32(1): 41–73. Grosu, Alexander (1973). On the nonunitary nature of the Coordinate Structure Con­ straint, Linguistic Inquiry 4(1): 88–92. Grosu, Alexander (1985). Subcategorization and parallelism, Theoretical Linguistics 12: 231–9. Grosu, Alexander (1987). On acceptable violations of parallelism constraints, in René Dir­ ven and Vilém Fried (eds), Functionalism in linguistics. Amsterdam and Philadelphia: John Benjamins, 425–57. Grosu, Alexander (1994). Three studies in locality and case. London: Routledge. Grosz, Barbara, Aravind Joshi, and Scott Weinstein (1995). Centering: A framework for modeling the local coherence of discourse, Computational Linguistics 21: 203–26. Grosz, Patrick (2015). Movement and agreement in Right-Node Raising constructions, Syntax 18: 1–38. Grover, Claire, Chris Brew, Suresh Manandhar, Marc Moens, and Andreas Schöter (1995). Priority union and generalization in discourse grammar, Edinburgh Working Papers in Cognitive Science 10: 157–96. Gualmini, Andrea, Sarah Hulsey, Valentine Hacquard, and Danny Fox (2004). The question-answer requirement for scope assignment, Natural Language Semantics 16: 205–37. (p. 1065)

Guasti, Maria Teresa (2002). Language acquisition: The growth of grammar. Cambridge, MA: MIT Press. Guida, Flore (2013). Les phrases sans verbes. Ms. Université Paris-Diderot.

Page 44 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Gullifer, Jason (2004). Processing reverse sluicing: A contrast with processing filler-gap dependences, in Keir Moulton and Matthew Wolf (eds), NELS 34: Proceedings of the 34th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 1–20. Gumperz, John (1976). The sociolinguistic significance of conversational code-switching. Papers on language and context, Working paper 46, University of California, Berkeley, 1– 46. Gundel, Jeanette, Nancy Hedberg, and Zacharski (1993). Cognitive status and the form of referring expressions in discourse, Language 69(2): 274–307. Guthrie, Malcolm (1967). Comparative Bantu: An introduction to the comparative linguis­ tics and prehistory of the Bantu languages. Farnborough: Gregg Press. Gyuris, Béata (2001). Semantic licensing of VP-ellipsis, Acta Linguistica Hungarica 48(1– 3): 59–78. Ha, Seungwan (2008a). Contrastive focus: Licensor for Right Node Raising, in Emily Elfn­ er and Martin Walkow (eds), NELS 37: Proceedings of the 37th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 247–60. Ha, Seungwan (2008b). Ellipsis, Right Node Raising, and across-the-board constructions. PhD thesis, Boston University. Ha, Seungwan (2008c). On ellipsis features and Right Node Raising, in Proceedings of ConSOLE XV. Leiden: SOLE, 67–90. Hackl, Martin, Jorie Koster-Hale, and Jason Varvoutis (2012). Quantification and ACD: Ev­ idence from real time sentence processing, Journal of Semantics 29: 145–206. Haddican, Bill (2007). On egin: do-support and VP focus in Central and Western Basque, Natural Language and Linguistic Theory 25(4): 735–64. Haddican, Bill (2008). The structural deficiency of verbal pro-forms, Linguistic Inquiry 38(3): 539–47. Haeseryn, W., K. Romijn, G. Geerts, J. de Rooij, and M. C. van den Toorn (eds) (1997). Al­ gemene Nederlandse Spraakkunst. Groningen / Deurne: Martinus Nijhoff / Wolters Plan­ tyn. Haida, Andreas (2008). The indefiniteness and focusing of wh-words. PhD thesis, Hum­ boldt-Universität zu Berlin. Haïk, Isabelle (1984). Indirect binding, Linguistic Inquiry 15(2): 185–224. Haïk, Isabelle (1987). Bound VPs that need to be, Linguistics and Philosophy 10(4): 503– 30.

Page 45 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Halliday, Michael (1967). Notes on transitivity and theme in English, part 2, Journal of Linguistics 3: 199–244. Hallman, Peter (2004). Symmetry in structure building, Syntax 7(1): 79–100. Halpert, Claire (2012). Argument licensing and agreement in Zulu. PhD thesis, MIT. Halpert, Claire (2015). Argument licensing and agreement. Oxford: Oxford University Press. Hamblin, Charles (1973). Questions in Montague grammar, Foundations of Language 10(1): 41–53. Hankamer, Jorge (1971). Constraints on deletion in syntax. PhD thesis, Yale University. Hankamer, Jorge (1973a). Unacceptable ambiguity, Linguistic Inquiry 4(1): 17–68. Hankamer, Jorge (1973b). Why there are two than’s in English, in Claudia Corum, T. Cedric Smith-Stark, and Ann Weiser (eds), Proceedings of the 8th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 179–91. Hankamer, Jorge (1978). On the nontransformational derivation of some null VP anaphors, Linguistic Inquiry 9(1): 66–74. Hankamer, Jorge (1979). Deletion in coordinate structures. New York: Garland Publishing. Hankamer, Jorge and Ivan Sag (1976). Deep and surface anaphora, Linguistic Inquiry 7(3): 391–428. Hardt, Daniel (1990). A corpus-based survey of VP ellipsis. Ms. University of Pennsylva­ nia. Hardt, Daniel (1992a). An algorithm for VP ellipsis, in Proceedings of the 30th annual meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 9–14. Hardt, Daniel (1992b). VP ellipsis and contextual interpretation, in COLING-92: Proceedings of the 15th International Conference on Computational Linguistics. Strouds­ burg, PA: Association for Computational Linguistics, 303–9. (p. 1066)

Hardt, Daniel (1992c). VP ellipsis and semantic identity, in Chris Barker and David Dowty (eds), SALT 2: Proceedings of the 2nd Semantics and Linguistic Theory Conference. Itha­ ca, NY: CLC Publications, 145–62. Hardt, Daniel (1993). Verb phrase ellipsis: Form, meaning and processing. PhD thesis, University of Pennsylvania.

Page 46 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hardt, Daniel (1994). Sense and reference in dynamic semantics, in P. J. E. Dekker and M. J. B. Stockhof (eds) Proceedings of the 9th Amsterdam Colloquium. Amsterdam: ILLC, 333–48. Hardt, Daniel (1997a). An empirical approach to VP ellipsis, Computational Linguistics 23(4): 525–41. Hardt, Daniel (1997b). VPE as a proform: Some consequences for binding. Paper present­ ed at the 2nd Colloque de Syntaxe et Sémantique à Paris (CSSP). Hardt, Daniel (1999). Dynamic interpretation of verb phrase ellipsis, Linguistics and Phi­ losophy 22(2): 185–219. Hardt, Daniel and Owen Rambow (2001). Generation of VP ellipsis: A corpus-based ap­ proach, in Proceedings of the 39th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 290–7. Hardt, Daniel and Maribel Romero (2004). Ellipsis and the structure of discourse, Journal of Semantics 21(4): 375–414. Hardt, Daniel, Nicholas Asher, and Julie Hunter (2013). VP ellipsis without indices, in SALT 23: Proceedings of the 23rd Semantics and Linguistic Theory Conference, Ithaca, NY: CLC Publications, 239–56. Harley, Heidi (2005). One-replacement, unaccusativity, acategorial roots, and bare phrase structure, in Slava Gorbachov and Andrew Nevins (eds), Harvard Working Papers in Lin­ guistics, vol. 11. Cambridge, MA: Harvard University, 59–78. Harley, Heidi (2009). Compounding in Distributed Morphology, in Rochelle Lieber and Pavel Stekauer (eds), Oxford Handbook of Compounding. Oxford: Oxford University Press, 129–44. Harris, Jesse (2016). Processing ‘let alone’ coordination in silent reading, Lingua 169: 70– 94. Harris, Jesse and Katy Carlson (2016). Keep it local (and final): Remnant preferences for ‘let alone’ ellipsis, Quarterly Journal of Experimental Psychology 69(7): 1278–1301. Hartman, Jeremy (2007). Focus, deletion, and identity: Investigations of ellipsis in Eng­ lish. BA thesis, Harvard University. Hartman, Jeremy and Ruizi Ressi Ai (2009). A focus account of swiping, in Kleanthes K. Grohmann and Phoevos Panagiotidis (eds), Selected papers from the 2006 Cyprus Syn­ taxfest. Cambridge: Cambridge Scholars Publishing, 91–122. Hartmann, Katharina (1998). Right Node Raising and gapping. PhD thesis, Goethe-Uni­ versität zu Frankfurt am Main, Frankfurt.

Page 47 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hartmann, Katharina (2000). Right Node Raising and gapping: Interface conditions on prosodic deletion. Amsterdam and Philadelphia: John Benjamins. Hartmann, Katharina (2003). Background matching in Right Node Raising constructions, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, 121–51. Hartmann, Katharina (2009). Right node raising and gapping: Interface conditions on prosodic deletion. Amsterdam: John Benjamins. Hartmann, Katharina and Viola Schmitt (2013). Violations of the Right Edge Constraint in Right Node Raising, Snippets 27: 8–10. Harves, Stephanie (2002). Genitive of negation and the syntax of scope, in Marjo van Kop­ pen, Erica Thrift, Erik-Jan van der Torre, and Malte Zimmerman (eds), Proceedings of ConSOLE IX. Leiden: SOLE, 96–110. Harwood, Will (2013). Being progressive is just a phase: Dividing the functional hierarchy. PhD thesis, Ghent University. (p. 1067)

Harwood, Will (2014). Being progressive is just a phase: Celebrating the uniqueness of progressive aspect under a phase-based analysis, Natural Language and Linguistic Theo­ ry 33: 523–73. Hasegawa, Nobuko (1984/5). On the so-called ‘empty pronouns’ in Japanese, The Linguis­ tic Review 4: 289–341. Hasegawa, Nobuko (2008). Wh-movement in Japanese: Matrix sluicing is different from embedded sluicing, in Cedric Boeckx and Suleyman Ulutas (eds), MIT Working Papers in Linguistics 55: Proceedings of the 4th Workshop on Formal Altaic Linguistics (WAFL-4). Cambridge, MA: MITWPL, 63–74. Haspelmath, Martin (1997). Indefinite Pronouns. Oxford: Oxford University Press. Haspelmath, Martin (2007a). Coordination, in Timothy Shopen (ed.), Language typology and syntactic description: Complex constructions, vol. 2. Cambridge: Cambridge Universi­ ty Press, 1–51. Haspelmath, Martin (2007b). Pre-established categories don’t exist: Consequences for language description and typology, Linguistic Typology 11: 119–32. Haspelmath, Martin (2008). Parametric versus functional explanations of syntactic uni­ versals, in T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam: John Ben­ jamins Publishing Company, 75–107. Haspelmath, Martin (2010). Comparative concepts and descriptive categories in cross-lin­ guistic studies, Language 86: 663–87.

Page 48 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hausser, Roland and Dietmar Zaefferer (1979). Questions and answers in a context de­ pendent Montague grammar, in Franz Guenther and Max Schmidt (eds), Formal seman­ tics and pragmatics for natural languages. Dordrecht: D. Reidel, 339–58. Hawkins, John A. (2004). Efficiency and complexity in grammars. Oxford: Oxford Univer­ sity Press. Hawkins, John A. (2014). Cross-linguistic variation and efficiency. Oxford: Oxford Univer­ sity Press. Hayashi, Shintaro (2015). Head movement in an agglutinative SOV language. PhD thesis, Yokohama National University. Haynie, Hannah J. (2010). What is Null Complement Anaphora? A syntactic account can explain. Ms. University of California at Berkeley. van der Heijden, Emmeken (1999). Tussen nevenschikking en onderschikking: een onder­ zoek naar verschillende vormen van verbinding in het Nederlands. PhD thesis, Radboud University Nijmegen. van der Heijden, Emmeken and Maarten Klein (1995). Rekenkundige voegwoorden: De ‘logica’ van samentrekkingen, De Nieuwe Taalgids 88: 22–38. Heim, Irene (1982). The semantics of definite and indefinite noun phrases. PhD thesis, University of Massachusetts, Amherst. Heim, Irene (1985). Notes on comparatives and related matters. Unpublished manuscript, University of Texas-Austin. Heim, Irene (1998). Anaphora and semantic interpretation: A reinterpretation of Reinhart’s approach, in Orin Percus and Uli Sauerland (eds), The interpretive tract. Cam­ bridge, MA: MIT Working Papers in Linguistics, 205–46. Heim, Irene (2000). Degree operators and scope, in Brendan Jackson and Tanya Matthews (eds), SALT 10: Proceedings of the 10th Semantics and Linguistic Theory Con­ ference. Ithaca, NY: CLC Publications, 40–64. Heim, Irene (2006). Remarks on comparative clauses as generalized quantifiers. Ms. MIT. Heim, Irene (2009). Forks on the road to Rule I, in Muhammad Abdurrahman, Anisa Schardi, and Martin Walkow (eds), NELS 38: Proceedings of the 38th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 256–84. Heim, Irene and Angelika Kratzer (1998). Semantics in generative grammar. Malden, MA: Blackwell. Heine, Lena (2011). Non-coordination-based ellipsis from a Construction Grammar per­ spective: The case of the coffee construction, Cognitive Linguistics 22(1): 55–80. Page 49 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hendriks, Petra (1991). The coordination-like structure of comparatives, in Frank Drijkoningen and Ans van Kemenade (eds), Linguistics in the Netherlands 1991. Amster­ dam and Philadelphia: John Benjamins, 41–50. (p. 1068)

Hendriks, Petra (1995). Comparatives and categorial grammar. PhD thesis, University of Groningen. Hendriks, Petra (2004). Coherence relations, ellipsis and contrastive topics, Journal of Se­ mantics 21: 375–414. Hendriks, Petra and Jennifer Spenader (2005). Why be silent? Some functions of ellipsis in natural language, in Jennifer Spenader and Petra Hendriks (eds), Proceedings of the Cross-modular approaches to ellipsis (ESSLLI), 29–38. Hennig, Mathilde (ed.) (2013). Die Ellipse: Neue Perspektive auf ein altes Phänomen. Berlin: De Gruyter. Heringer, Hans (1996). Deutsche Syntax dependentiell. Tübingen: Stauffenberg. Hernández, Ana Carrera (2007). Gapping as a syntactic dependency, Lingua 117(12): 2106–33. Hernanz, María Luisa and José María Brucart (1987). La sintaxis I. Barcelona: Crítica. Hestvik, Arild (1995). Reflexives and ellipsis. Natural Langauge Semantics 3(2): 211–37. Herring, Jon, M. Carmen Parafita Couto, Moro Quintanilla, and Mónica Moro Quintanilla (2010). ‘When I went to Canada, I saw the madre’: Evaluating two theories’ predictions about codeswitching between determiners and nouns using Spanish-English and WelshEnglish bilingual corpora, International Journal of Bilingual Education and Bilingualism 13(5): 553–73. Heycock, Caroline and Anthony Kroch (1994). Verb movement and coordination in a dy­ namic theory of licensing, The Linguistic Review 11(3–4): 257–84. Hilbert, David and Paul Bernays (1939). Grundlagen der Mathematik II. Berlin: Julius Springer. Hinds, John (1973). On the status of the VP node in Japanese, Language Research 9: 44– 57. Hiraiwa, Ken and Shinichiro Ishihara (2002). Missing links: Clefts, sluicing and ‘no da’ construction in Japanese, in Tania Ionin, Heejeong Ko, and Andrew Nevins (eds), The MIT Working Papers in Linguistics 43: Proceedings of the 2nd HUMIT Student Conference in Language Research (HUMIT 2001). Cambridge, MA: MITWPL, 35–54. Hiraiwa, Ken and Shinichiro Ishihara (2012). Syntactic metamorphosis: Clefts, sluicings, and in-situ focus in Japanese, Syntax 15(2): 142–80. Page 50 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hirsch, Aron (2016). A case for conjunction reduction. Ms. MIT. Hirsch, Aron and Michael Wagner (2015). Right Node Raising, scope, and plurality, in Thomas Brochhagen, Floris Roelofsen and Nadine Theiler (eds), Proceedings of the 20th Amsterdam Colloquium. Amsterdam: ILLC, 187–96. Hirschberg, Julia (2004). Pragmatics and intonation, in Laurence R. Horn and Gregory Ward (eds), The Handbook of Pragmatics. Malden, MA: Blackwell Publishing, 515–37. Hirschberg, Julia and Gregory Ward (1991). Accent and bound anaphora, Cognitive Lin­ guistics 2(2): 101–21. Hirschbuhler, Paul (1982). VP deletion and across-the-board quantifier scope, in James Pustejovsky and Peter Sells (eds), NELS 12: Proceedings of the 12th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 132–9. Hirst, Graeme (1981). Discourse-oriented anaphora resolution in natural language under­ standing: A review, American Journal of Computational Linguistics 7(2): 85–98. Hobbs, Jerry (1990). Literature and cognition. Technical report, CLSI Lecture Notes 21. Hobbs, Jerry and Andrew Kehler (1997). A theory of parallelism and the case of VP ellip­ sis, in Philip R. Cohen and Wolfgang Wahlster (eds), Proceedings of the 35th annual meet­ ing of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 394–401. Hoeks, John C. J., Gisela Redeker, and Petra Hendriks (2009). Fill the gap! Combining pragmatic and prosodic information to make gapping easy, Journal of Psycholinguistic Re­ search 38: 221–35. Hoeks, John C. J., Gisela Redeker, and Louisa J. Zijlstra (2006). The predominance of nonstructural factors in the processing of gapping sentences, in Ron Sun (ed.), Proceedings of the 28th Annual Conference of the Cognitive Science Society. Philadelphia: Taylor & Francis, 1511–16. Hoeksema, Jack (1983). Negative polarity and the comparative, Natural Lan­ guage and Linguistic Theory 1(3): 403–34. (p. 1069)

Hoeksema, Jack (2006). Pseudogapping: Its syntactic analysis and cumulative effects on its acceptability, Research on Language and Computation 4(4): 335–52. Hofmann, Ronald T. (1976). Past tense replacement and the modal system, in James D. McCawley (ed.), Notes from the linguistic underground. New York: Academic Press, 85– 100. Hofmeister, Philip (2010). A linearization accout of either…or constructions, Natural Lan­ guage and Linguistic Theory 28(2): 275–314.

Page 51 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hofmeister, Philip and Ivan A. Sag (2010). Cognitive constraints and island effects, Lan­ guage 86(2): 366–415. Hofmeister, Philip, Laura Staum Casasanto, and Ivan A. Sag (2013). Islands in the gram­ mar? Standards of evidence, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, 42–63. Hofmeister, Philip, T. Florian Jaeger, Ivan A. Sag, Inbal Arnon, and Neal Snider (2007). Lo­ cality and accessibility in wh-questions, in Sam Featherston and Wolfgang Sternefeld (eds), Roots: Linguistics in search of its evidential base. Berlin: Mouton de Gruyter, 185– 206. Hogue, Alan (2014). VP ellipsis exhibits structural priming. Poster presented at the 88th Annual Meeting of the Linguistic Society of America. Minneapolis, MN. Höhle, Tilman N. (1991). On reconstruction and coordination, in Hubert Haider and Klaus Netter (eds), Representation and derivation in the theory of grammar, vol. 22. Dordrecht: Kluwer Academic Publishers, 139–98. Hoji, Hajime (1987). Japanese clefts and reconstruction/chain binding effects. Talk pre­ sented at WCCFL VI: Sixth West Coast Conference on Formal Linguistics held at Universi­ ty of Arizona. Hoji, Hajime (1995a). Demonstrative binding and Principle B, in Jill Beckman (ed.), NELS 25: Proceedings of the annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 255–72. Hoji, Hajime (1995b). Theories of anaphora and aspects of Japanese syntax. Ms. USC. Hoji, Hajime (1997a). Sloppy identity and formal dependency, in Brian Agbanyani and Sze-Wing Tang (eds), Proceedings from the 15th West Coast Conference on Formal Lin­ guistics. Somerville, MA: Cascadilla Press, 209–29. Hoji, Hajime (1997b). Sloppy identity and principle B, in Hans Bennis, Pierre Pica, and Jo­ han Rooryck (eds), Atomism and binding. Dordrecht: Foris Publications, 205–35. Hoji, Hajime (1998). Null object and sloppy identity in Japanese, Linguistic Inquiry 29(1): 127–52. Hoji, Hajime (2003). Surface and deep anaphora, sloppy identity, and experiments in syn­ tax, in Andrew Barss (ed.), Anaphora: A reference guide. Malden, MA: Oxford University Press, 172–236. Hoji, Hajime (2009). On Saito et al. 2008 (and Saito and Murasugi 1990). Class handout at USC. Hoji, Hajime (2013). Gengo kagaku o mezasite [Aiming at linguistic science]: Issues on anaphora in Japanese. Shiga: Oshumi Shoten. Page 52 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Hoji, Hajime (2015). Language faculty science. Cambridge: Cambridge University Press. Hoji, Hajime (2016). Issues on anaphora in Japanese. Shiga: Oshumi Shoten. Hoji, Hajime, Satoshi Kinsui, Yukinori Takubo, and Aymui Ueyama (2000). Demonstra­ tives, bound variables, and reconstruction effects, in Proceedings of the Nanzan GLOW. Nagoya: Nanzan University, 141–58. Holmberg, Anders (1986). Word order and syntactic features. PhD thesis, Stockholm. Holmberg, Anders (2001). The syntax of yes and no in Finnish, Studia Linguistica 55: 141– 74. Holmberg, Anders (2016). The syntax of yes and no. Oxford: Oxford University Press. Hopper, P. and Elizabeth Closs Traugott (2003). Grammaticalization. Cambridge: Cam­ bridge University Press. Hornstein, Norbert (1995a). Logical Form: From GB to Minimalism. Cambridge, MA: Basil Blackwell. Hornstein, Norbert (1995b). Putting truth into Universal Grammar, Linguistics and Philos­ ophy 18(4): 381–400. Hornstein, Norbert, Howard Lasnik, and Juan Uriagereka (2007). The dynamics of islands: Speculations on the locality of movement, Linguistic Analysis 33: 149–75. (p. 1070)

Horvath, Julia (2005). Is focus movement driven by stress? Approaches to Hungarian 9: 131–58. Hough, Julian (2011). Incremental semantics driven natural language generation with self-repairing capability, in Proceedings of the Student Research Workshop associated with RANLP 2011. Hissar, Bulgaria, 79–84. Hough, Julian (2015). Modelling incremental self-repair processing in dialogue, PhD the­ sis, Queen Mary University of London. Hough, Julian and Matthew Purver (2012). Processing self-repairs in an incremental typetheoretic dialogue system, in Proceedings of the 16th SemDial Workshop on the Seman­ tics and Pragmatics of Dialogue (SeineDial). France, 136–44. Houser, Michael (2010). The syntax and semantics of do so anaphora. PhD thesis, Univer­ sity of California at Berkeley. Houser, Michael, Line Mikkelsen, and Maziar Toosarvandani (2007). Verb phrase pronom­ inalization in Danish: Deep or surface anaphora?, in Erin Brainbridge and Brian Ag­ banyani (eds), Proceedings of the 34th Western Conference on Linguistics. Fresno, CA: Department of Linguistics, 183–95.

Page 53 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Howes, Christine, Matthew Purver, Pat G. T. Healey, Greg J. Mills, and Eleni Gre­ goromichelaki (2011). On incrementality in dialogue: Evidence from compound contribu­ tions, Dialogue and Discourse 2(1): 279–311. Hoyt, Frederick and Alexandra Teodorescu (2012). How many types of sluicing, and why: Single and multiple sluicing in Romanian, English, and Japanese, in Jason Merchant and Andrews Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, 83–103. Huang, C.-T. James (1982). Logical relations in Chinese and the theory of grammar. PhD thesis, Massachusetts Institute of Technology. Huang, C.-T. James (1984). On the distribution and reference of empty pronouns, Linguis­ tic Inquiry 15: 531–74. Huang, C.-T. James (1988a). Comments on Hasegawa’s paper, in Tawa Wako and Mine­ haru Nakayama (eds), Proceedings of Japanese Syntax Workshop Issues on Empty Cate­ gories. New London: Connecticut College, 77–93. Huang, C.-T. James (1988b). Wo pao de kuai and Chinese phrase structure, Language 64: 274–311. Huang, C.-T. James (1991). Remarks on the status of the null object, in Robert Freidin (ed.), Principles and parameters in comparative grammar. Cambridge, MA: MIT Press, 56–76. Hubert, Anja (2009). The acquisition of ellipsis and the structure of the noun phrase. Master’s thesis, University of Potsdam. Huddleston, Rodney (1976). Some theoretical issues in the description of the English verb, Lingua 40: 331–83. Huddleston, Rodney and Geoffrey K. Pullum (2002). The Cambridge grammar of the Eng­ lish language. Cambridge: Cambridge University Press. Hudson, Richard A. (1976a). Arguments for a non-transformational grammar. Chicago: University of Chicago Press. Hudson, Richard A. (1976b). Conjunction reduction, gapping and right-node raising, Lan­ guage 52(3): 535–62. Hudson, Richard A. (2003). Gerunds without phrase structure, Natural Language and Lin­ guistic Theory 21: 579–615. Hudson, Richard A. (2007). Language networks: The new Word Grammar. Oxford: Oxford University Press. Hudson, Richard A. (2010). An introduction to Word Grammar. Cambridge: Cambridge University Press. Page 54 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Huet, Gérard P. (1975). A unification algorithm for typed λ-calculus, Theoretical Comput­ er Science 1(1): 27–57. Hulsey, Sarah (2008). Focus sensitive coordination. PhD thesis, MIT. Hulsey, Sarah and Uli Sauerland (2006). Sorting out relative clauses: A reply to Bhatt, Natural Langauge Semantics 14(1): 111–37. (p. 1071)

Hurford, J. R. (2012). The origins of grammar: Language in the light of evolution. Volume II: The origins of grammar. Oxford: Oxford University Press. Hyman, Larry M. and Francis X. Katamba (1993). The argument in Luganda: Syntax or pragmatics?, in Sam Mchombo (ed.), Theoretical aspects of Bantu grammar. Stanford: CSLI, 209–56. Ince, Atakan (2009). Dimensions of ellipsis: Investigations in Turkish. PhD thesis, Univer­ sity of Maryland, College Park. Ince, Atakan (2012). Sluicing in Turkish, in Jason Merchant and Andrew Simpson (eds), Sluicing: Crosslinguistic perspectives. Oxford: Oxford University Press, 248–69. Inoue, Kazuko (1976). Henkei bunpoo to nihongo [Transformational grammar and Japan­ ese]. Tokyo: Taishukan. Inoue, Kazuko (1978). Nihongo-no bunpoo kisoku [Grammatical rules in Japanese]. Tokyo: Taishukan. Ionin, Tania (2001). The one girl who was kissed by every boy: Scope, scrambling, and discourse function in Russian, in Marjo van Koppen, Joanna Sio, and Mark de Vos (eds), Proceedings of ConSOLE X. Leiden: SOLE, 79–94. Ionin, Tania and Ora Matushansky (2013). More than one comparative in more than one Slavic language: An experimental investigation, in Proceedings of the 24th Meeting of Formal Approaches to Slavic Linguistics (FASL 21). Ann Arbor: Michigan Slavic Publica­ tions. Ionin, Tania, Ora Matushansky, and Eddy G. Ruys (2011). Parts of speech: Towards a uni­ fied semantics for partitives, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS 36: Proceedings of the 36th annual meeting of the North East Linguistic So­ ciety. Amherst, MA: GLSA, 357–71. Iseda, Takako (2007). On focused NPs with nominative case particle -ga in Japanese, in M. Kelepir and B. Öztürk (eds), MIT Working Papers in Linguistics 54: Proceedings of the 2nd Workshop on Altaic Formal Linguistics (WAFL 2). Cambridge, MA: MITWPL, 185–97. Ishihara, Shinichiro (2012). The clausemate condition in Japanese multiple clefts, in Matt Tucker, Anie Thompson, Olver Northrup, and Ryan Bennett (eds), Proceedings of FAJL 5: Formal Approaches to Japanese Linguistics. Cambridge, MA: MITWPL, 75–88. Page 55 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Iwakura, Kunihiro (1977). The auxiliary system in English, Linguistic Analysis 3: 101–36. Izvorski, Roumyana (1995). A solution to the subcomparative paradox, in José Camacho, Lina Choueiri, and Maki Watanabe (eds), Proceedings from the 14th West Coast Confer­ ence on Formal Linguistics. Stanford: CSLI, 203–19. Jackendoff, Ray (1971). Gapping and related rules, Linguistic Inquiry 2(1): 21–35. Jackendoff, Ray (1972). Semantic interpretation in Generative Grammar. Cambridge, MA: MIT Press. Jackendoff, Ray (1977). X′ syntax. Cambridge, MA: MIT Press. Jackendoff, Ray (1990). Semantic structures. Cambridge, MA: MIT Press. Jackendoff, Ray (2002). Foundations of language. Oxford: Oxford University Press. Jacobson, Pauline (1992a). Antecedent contained deletion in a variable-free semantics, in Chris Barker and David Dowty (eds), SALT 2: Proceedings of the 2nd Semantics and Lin­ guistic Theory Conference. Ithaca, NY: CLC Publications, 193–213. Jacobson, Pauline (1992b). Flexible categorial grammars: Questions and prospects, in Robert D. Levine (ed.), Formal grammar: Theory and implementation. Oxford: Oxford Uni­ versity Press, 129–67. Jacobson, Pauline (1992c). Raising without movement, in Richard K. Larson, Sabine Iatri­ dou, Utpal Lahiri, and James Higginbotham (eds.), Control and Grammar. Dordrecht: Kluwer Academic Publishers, 149–94. Jacobson, Pauline (1994). Binding connectivity in copular sentence, in Mandy Harvey and Lynn Santelmann (eds), SALT 4: Proceedings of the 4th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 145–60. Jacobson, Pauline (1999). Towards a variable-free semantics, Linguistics and Phi­ losophy 22(2): 117–84. (p. 1072)

Jacobson, Pauline (2000a). Paycheck pronouns, Bach-Peters sentences, and variable-free semantics, Natural Language Semantics 8(2): 77–155. Jacobson, Pauline (2000b). Paychecks, stress, and variable free semantics, in T. Matthews and B. Jackson (eds), SALT 10: Proceedings of the 10th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 65–82. Jacobson, Pauline (2003). Binding without pronouns (and pronouns without binding), in G.-J. Kruiff and Richard Oehrle (eds), Binding and resource sensitivity. Dordrecht: Kluwer Academic Publishers, 57–96.

Page 56 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Jacobson, Pauline (2008). Direct compositionality and variable-free semantics: The case of Antecedent Contained Deletion, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cam­ bridge University Press, 30–68. Jacobson, Pauline (2014). Compositional semantics: An introduction to the syntax-seman­ tics interface. Oxford: Oxford University Press. Jacobson, Pauline (2016a). Online appendix to: The short answer: Implications for direct compositionality and vice versa, Language 92: s1–s10. Jacobson, Pauline (2016b). The short answer: Implications for direct compositionality and vice versa, Language 92: 331–74. Jäger, Lena A., Lena Benz, Jens Roeser, Brian Dillon, and Shravan Vasishth (2015). Teas­ ing apart retrieval and encoding interference in the processing of anaphors, Frontiers in Psychology 6, . Jaeger, T. Florian and Harry Tily (2011). On language ‘utility’: Processing complexity and communicative efficiency, Wiley Interdisciplinary Reviews Cognitive Science 2: 323–35. Jake, Janice L. and Carol Myers-Scotton (2009). Which language? Participation potentials across lexical categories in codeswitching, in Ludmila Isurin, Donald Winford, and Kees de Bot (eds), Multidisciplinary approaches to code-switching. Amsterdam and Philadel­ phia: John Benjamins, 207–42. Jake, Janice L., Carol Myers-Scotton, and Steven Gross (2002). Making a minimalist ap­ proach to codeswitching work: Adding the matrix language, Bilingualism: Language and Cognition 5(1): 69–91. Jake, Janice L., Carol Myers-Scotton, and Steven Gross (2005). A response to MacSwan (2005): Keeping the matrix language, Bilingualism: Language and Cognition 8(3): 271–6. James, William (1890). The principles of psychology, vol. I. New York: Henry Holt and Co. Jantunen, Tommi (2007). The equative sentence in Finnish Sign Language, Sign Language and Linguistics 10: 113–43. Jantunen, Tommi (2008). Fixed and free: Order of the verbal predicate and its core argu­ ments in declarative transitive clauses in Finnish Sign Language, SKY Journal of Linguis­ tics 21: 83–123. Jantunen, Tommi (2009). Tavu ja lause: Tutkimuksia kahden sekventiaalisen perusyksikön olemuksesta suomalaisessa viittomakielessä [Syllable and sentence: Studies on the na­ ture of two sequential basic units in FinSL]. Jyväskylä: University of Jyväskylä. Jantunen, Tommi (2010). Suomalaisen viittomakielen pääsanaluokat [The main parts of speech in FinSL], in Tommi Jantunen (ed.), Näkökulmia viittomaan ja viittomistoon [Per­ spectives on sign and lexicon]. Jyväskylä: University of Jyväskylä, 57–78. Page 57 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Jantunen, Tommi (2013). Ellipsis in Finnish Sign Language, Nordic Journal of Linguistics 36: 303–32. Jantunen, Tommi (2015). Rinnastuksen prosodiaa suomalaisessa viittomakielessä [The prosody of clausal coordination in FinSL], in Mona Lehtinen and Unto K. Laine (eds), XXIX Fonetiikan päivät, Espoo 20. 21.3.2015, Julkaisut—Papers [Proceedings of the XXIX Phonetics Symposium]. Helsinki: Aalto University, 15–21. Jantunen, Tommi (2016). Clausal coordination in Finnish Sign Language, Studies in Lan­ guage 40(1): 204–34. Jayaseelan, Karattuparambil A. (1990). Incomplete VP deletion and gapping, Linguistic Analysis 20(1–2): 64–81. Jayaseelan, Karattuparambil A. (2001). IP-internal topic and focus phrases, Stu­ dia Linguistica 55(1): 39–75. (p. 1073)

Jensen, Britta and Rosalind Thornton (2008). Fragments of child grammar, in M. João Fre­ itas and Anna Gavarró Alguéro (eds), Language acquisition and development: Papers from GALA 2007. Cambridge: Cambridge Scholars Publishing, 303–12. Johannessen, Janne Bondi (1998). Coordination. Oxford: Oxford University Press. Johnson, Kyle (1996). When verb phrases go missing, GLOT 2(5): 3–9. Johnson, Kyle (2000a). Few dogs eat Whiskers or cats Alpo, in Kiyomi Kusumoto and Elis­ abeth Villalta (eds), University of Massachusetts Occasional Papers 23. Amherst, MA: UMOP, 47–60. Johnson, Kyle (2000b). Gapping determiners, in Kerstin Schwabe and Nina Zhang (eds), Ellipsis in conjunction. Tübingen: Niemeyer, 95–115. Johnson, Kyle (2001a). Sluicing and constraints on quantifier scope, GLOT International 5(6): 217–21. Johnson, Kyle (2001b). What VP ellipsis can do, and what it can’t, but not why, in Mark Baltin and Chris Collins (eds), The handbook of contemporary syntactic theory. Oxford: Blackwell Publishers, 439–79. Johnson, Kyle (2002). Restoring exotic coordinations to normalcy, Linguistic Inquiry 33(1): 97–156. Johnson, Kyle (2004a). In search of the English middle field. Ms. University of Massachu­ setts, Amherst. Johnson, Kyle (2004b). How to be quiet, in Nikki Adams, Adam Cooper, Fey Parrill, and Thomas Wier (eds), Proceedings from the 40th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 1–20. Page 58 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Johnson, Kyle (2006). Gapping, in M. Evereart, H. C. van Reimsdijk, R. Goedemans, and B. Holledbrandse (eds), The Blackwell companion to syntax, vol. II. Malden, MA and Oxford: Blackwell, 407–35. Johnson, Kyle (2008a). Introduction, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, 1–14. Johnson, Kyle (ed.) (2008b). Topics in ellipsis. Cambridge: Cambridge University Press. Johnson, Kyle (2009). Gapping is not (VP-)ellipsis, Linguistic Inquiry 40(2): 289–328. Johnson, Kyle (2014). Gapping. Ms. University of Massachusetts at Amherst. Johnson, Kyle and Satoshi Tomioka (1998). Lowering and mid-size clauses, in Graham Katz, Shin-Sook Kim, and Winhart Haike (eds), Proceedings of the 1997 Tübingen work­ shop on reconstruction. Tübingen, Germany: Sprachteoretische Grundlagen für die Com­ puter Linguistik, 185–206. Johnson-Laird, Philip N. (1983). Mental models. Cambridge, MA: Harvard University Press. Johnston, Trevor and Adam Schembri (1999). On defining lexeme in a signed language, Sign Language and Linguistics 2: 115–85. Johnston, Trevor and Adam Schembri (2007). Australian Sign Language: An introduction to sign language linguistics. Cambridge: Cambridge University Press. Joshi, Aravind (1985). Processing of sentences with intrasentential code switching, in David R. Dowty, Lauri Karttunen, and Arnold Zwicky (eds), Natural language parsing: Psy­ chological, computational, and theoretical perspectives. Cambridge: Cambridge Universi­ ty Press, 190–205. Jun, Sun-Ah (2005). Korean intonational phonology and prosodic transcription, in Sun-Ah Jun (ed.), Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press, 201–29. Kaan, Edith, Carlie Overfelt, Do Tromp, and Frank Wijnen (2013). Processing gapped verbs, Journal of Psycholinguistic Research 42: 307–38. Kaan, Edith, Frank Wijnen, and Tamara Y. Swaab (2004). Gapping: Electrophysiological evidence for immediate processing of ‘missing’ verbs in sentence comprehension, Brain and Language 89(3): 584–92. Kadowaki, Makoto (2005). N’-ellipsis reconsidered, in Yukio Otsu (ed.), Proceed­ ings of the 6th Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, 191–215. (p. 1074)

Kallestinova, Elena (2007). Aspects of word order in Russian. PhD thesis, University of Iowa. Page 59 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kameyama, Megumi (1985). Zero anaphora: The case of Japanese. PhD thesis, Stanford University. Kameyama, Megumi (1999). Stressed and unstressed pronouns: Complementary prefer­ ences, in Peter Bosch and Rob van der Sandt (eds), Focus: Linguistic, cognitive and com­ putational perspectives. Cambridge: Cambridge University Press, 306–621. Kamio, Akio (1983). Meisiku-no koozoo [Noun phrase structure], in Kazuko Inoue (ed.), Nihongo-no kihon koozoo [Basic structures in Japanese]. Tokyo: Sanseido, 77–126. Kaplan, David (1989). Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals, in Joseph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan. Oxford: Oxford University Press, 481–564. Karimi, Hossein, Kumiko Fukukura, Fernanda Ferreira, and Martin J. Pickering (2014). The effect of noun phrase length on the form of referring expressions, Memory and Cogni­ tion 42: 993–1009. Karimi, Simin (2005). A minimalist approach to scrambling: Evidence from Persian. Berlin: Mouton de Gruyter. Katz, Jerrold J. and Paul Postal (1964). An integrated theory of linguistic descriptions. Cambridge, MA: MIT Press. Kay, Paul (1989). Contextual operators: respective, respectively, and vice versa, in Kira Hall, Michael Meacham, and Richard Shapiro (eds), Proceedings of the 15th annual meet­ ing of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 181–92. Kay, Paul (2002). English subjectless tagged sentences, Language 78(3): 453–81. Kayne, Richard S. (1983). Connectedness, Linguistic Inquiry 14(2): 223–49. Kayne, Richard S. (1984). Connectedness and binary branching. Dordrecht: Foris Publica­ tions. Kayne, Richard S. (1994). The antisymmetry of syntax. Cambridge, MA: MIT Press. Kayne, Richard S. (1997). The English complementizer of, The Journal of Comparative Germanic Linguistics 1(1): 43–54. Kayne, Richard S. (2015). One and ones as complex determiners. Ms. New York Universi­ ty. Kazanina, Nina, Ellen Lau, Moti Lieberman, Masaya Yoshida, and Colin Phillips (2007). The effect of syntactic constraints on the processing of backwards anaphora, Journal of Memory and Language 56: 384–409. Kazenin, Konstantin (2002). On coordination of wh-phrases in Russian. Ms. University of Tuebingen. Page 60 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kazenin, Konstantin (2006). Polarity in Russian and typology of predicate ellipsis. Ms. Moscow State University. Kazenin, Konstantin (2007). O nekotoryx ograničenijax na ellipsis v russkom jazyke [On some restrictions on ellipsis in Russian], Voprosy jazykoznanija 2: 92–107. Kazenin, Konstantin (2010). Russian gapping: Against ATB, in Gerhild Zybatow, Philip Dudchuk, Serge Minor, and Ekaterina Pshehotskaya (eds), Formal studies in Slavic lin­ guistics. Bern: Peter Lang, 85–100. Keenan, Edward (1971). Names, quantifiers, and a solution to the sloppy identity prob­ lem, Papers in Linguistics 4(2): 211–32. Kehler, Andrew (1993a). A discourse copying algorithm for ellipsis and anaphora resolu­ tion, in Proceedings of the Sixth European Chapter of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 203–12. Kehler, Andrew (1993b). The effect of establishing coherence in ellipsis and anaphora res­ olution, in Proceedings of the 31st Conference of the Association for Computational Lin­ guistics. Stroudsburg, PA: Association for Computational Linguistics, 62–9. Kehler, Andrew (1994). Common topics and coherent situations: Interpreting ellipsis in the context of discourse inference. In Proceedings of the 32nd Conference of the Associa­ tion for Computational Linguistics. Stroudsburg, PA: Association for Computational Lin­ guistics, 50–7. Kehler, Andrew (1997). Current theories of centering for pronoun interpretation: A critical evaluation, Computational Linguistics 23: 467–75. (p. 1075)

Kehler, Andrew (2000). Coherence and the resolution of ellipsis, Linguistics and Philoso­ phy 23(6): 533–75. Kehler, Andrew (2002). Coherence, reference, and the theory of grammar. Stanford: CSLI Publications. Kehler, Andrew (2007). Contrastive topics and illusory sloppy interpretation in VP-ellipsis. Paper presented at the Annual Meeting of the Linguistic Society of America. Kehler, Andrew (2015). On QUD-based licensing of strict and sloppy ambiguities, in Sarah D’Antonio, Mary Moroney, and Carol-Rose Little (eds), Semantics and linguistic theory, vol. 25. Washington DC: LSA and CLC Publications, 512–32. Kehler, Andrew and Daniel Büring (2007). Be bound or be disjoint!, in NELS 38: Proceed­ ings of the 38th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 487. Kehler, Andrew and Stuart Shieber (1997). Anaphoric dependencies in ellipsis, Computa­ tional Linguistics 23(3): 457–66. Page 61 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kehler, Andrew and Gregory Ward (2007). Event reference and morphological transparen­ cy, in Proceedings of the 35th Western Conference on Linguistics (WECOL-07). Depart­ ment of Linguistics, California State University, Fresno, 115–27. Kempson, Ruth, Ronnie Cann, Arash Eshghi, Eleni Gregoromichelaki, and Matthew Purv­ er (2015). Ellipsis, in Shalom Lappin and Chris Fox (eds), Handbook of contemporary se­ mantic theory. Hoboken: Wiley-Blackwell, 114–40. Kempson, Ruth, Andrew Gargett, and Eleni Gregoromichelaki (2007). Clarification re­ quests: An incremental account, in Ron Artstein and Laure Vieu (eds), Decalog 2007: Pro­ ceedings of the 11th Workshop on the Semantics and Pragmatics of Dialogue (SemDial 11). Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay (1999). VP ellipsis: Toward a dy­ namic, structural account, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, 227–89. Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay (2001). Dynamic Syntax: The flow of language understanding. Hoboken: Wiley-Blackwell. Kendon, Adam (2004). Gesture: Visible action as utterance. Cambridge: Cambridge Uni­ versity Press. Kenesei, István (1992). On Hungarian complementizers, in István Kenesei and Csaba Pléh (eds), Approaches to Hungarian 4. Szeged: JATE Press, 37–50. Kenesei, István (1994). Subordinate clauses, in Ferenc Kiefer and Katalin É. Kiss (eds), Syntax and Semantics 27. New York: Academic Press, 275–354. Kenesei, István (1998). Adjuncts and arguments in VP focus in Hungarian, Acta Linguisti­ ca Hungarica 45: 61–88. Kenesei, István (2001). Criteria for auxiliaries in Hungarian, in István Kenesei (ed.), Argu­ ment structure in Hungarian. Budapest: Akadémiai Kiadó, 73–106. Kenesei, István, Anna Fenyvesi, and Robert M. Vago (1998). Hungarian. Oxford: Rout­ ledge. Kennedy, Christopher (1997). Antecedent contained deletion and the syntax of quantifica­ tion, Linguistic Inquiry 28(4): 662–88. Kennedy, Christopher (1999). Projecting the adjective: The syntax and semantics of grad­ ability and comparison. New York: Garland Press. Kennedy, Christopher (2000). Comparative (sub)deletion and ranked, violable constraints in syntax, in Masako Hirotani, Andries Coetzee, Nancy Hall, and Ji-yung Kim (eds), NELS 30: Proceedings of the 30th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 389–414. Page 62 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kennedy, Christopher (2002). Comparative deletion and optimality in syntax, Natural Lan­ guage and Linguistic Theory 20(3): 553–621. Kennedy, Christopher (2003). Ellipsis and syntactic representation, in Kerstin Schwabe and Susanne Winkler (eds), The Interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, 29–53. Kennedy, Christopher (2007). Standards of comparison. Paper presented at the Colloque de Syntaxe et Sémantique de Paris, 6 October. (p. 1076)

Kennedy, Christopher (2008). Argument contained ellipsis, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, 95–131. Kennedy, Christopher (2009). Modes of comparison, in Malcolm Elliott, James Kirby, Os­ amu Sawada, Eleni Staraki, and Suwon Yoon (eds), Proceedings of the 43rd annual meet­ ing of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 141–65. Kennedy, Christopher and Jeffrey Lidz (2001). A (covert) long distance anaphor in Eng­ lish, in Karine Megerdoomian and Leora Anne Bar-el (eds), Proceedings of the 20th West Coast Conference in Formal Linguistics. Somerville, MA: Cascadilla Press, 318–31. Kennedy, Christopher and Jason Merchant (2000a). Attributive comparative deletion, Nat­ ural Language and Linguistic Theory 18(1): 89–146. Kennedy, Christopher and Jason Merchant (2000b). The case of the missing CP and the secret case, in Sandra Chung, James McCloskey, and Nathan Sanders (eds), Jorge Han­ kamer Webfest. UCSC. . Kentner, Gerrit, Caroline Féry, and Kai Alter (2008). Prosody in speech production and perception: The case of Right Node Raising in English, in Anita Steube (ed.), The dis­ course potential of underspecified structures: Event structures and information struc­ tures. Berlin: Mouton de Gruyter, 207–23. Kerstens, Johan (1980). Over samentrekking, De Nieuwe Taalgids 73: 375–94. Kerstens, Johan (1981). Bestaat gapping eigenlijk wel?, Spektator 11: 61–79. Kertz, Laura (2008). Focus structure and acceptability in verb phrase ellipsis, in Natasha Abner and Jason Bishop (eds), Proceedings of the 27th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 283–91. Kertz, Laura (2010). Ellipsis reconsidered. PhD thesis, University of California, San Diego. Kertz, Laura (2013). Verb phrase ellipsis: The view from information structure, Language 89(3): 390–428. Keshet, Ezra (2013). Sloppy identity unbound, in Todd Snider (ed.), Semantics and lin­ guistic theory 23. Ithaca: CLC Publications, 412–31. Page 63 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kester, Ellen-Petra (1996a). Adjectival inflection and the licensing of empty categories in DP, Journal of Linguistics 32: 57–78. Kester, Ellen-Petra (1996b). The nature of adjectival inflection. Utrecht: LEd. Khamisi, Abdu M. (1988). A typology of gaps in Kiswahili sentences, Kiswahili 55: 120–33. Kiguchi, Hirohisa and Rosalind Thornton (2004). Binding principles and ACD construc­ tions in child grammars, Syntax 7(3): 234–71. Kim, Christina, Gregory M. Kobele, Jeffrey Runner, and John T. Hale (2011). The accept­ ability cline in VP ellipsis, Syntax 14: 318–54. Kim, Christina and Jeffrey Runner (2009). Strict identity, coherence, and parallelism in VP ellipsis, in Ed Cormany, Satoshi Ito, and David Lutz (eds), Proceedings of the 19th Seman­ tics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 275–87. Kim, Christina and Jeffrey Runner (2011). Discourse structure and syntacic parallelism in VP ellipsis, UMass Occasional Papers in Linguistics 38: 75–102. Kim, Jong-Bok (2001). Interactions between constructions and constraints in VP ellipsis and VP fronting, in Charles Chang et al. (eds), Proceedings of the 27th Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 107–24. Kim, Jong-Bok (2006). Similarities and differences between English VP ellipsis and VP fronting: An HPSG analysis, Studies in Generative Grammar 13(3): 429–59. Kim, Jong-Bok (2015). Syntactic and semantic identity in Korean sluicing: A direct inter­ pretation approach, Lingua 66: 260–93. Kim, Jong-Bok and Peter Sells (2008). English syntax: An introduction. Stanford: CSLI. Kim, Jeong-Seok (1997). Syntactic focus movement and ellipsis: A minimalist approach. PhD thesis, University of Connecticut. Kim, Soowon (1999). Sloppy/strict identity, empty objects, and NP ellipsis, Jour­ nal of East Asian Linguistics 8(4): 255–84. (p. 1077)

Kimmelman, Vadim (2014). Information structure in Russian Sign Language and Sign Language of the Netherlands. PhD thesis, University of Amsterdam. Kimura, Hiroko (2010). A wh-in-situ strategy for sluicing, English Linguistics 27: 43–59. King, Harold V. (1970). On blocking rules for contraction in English, Linguistic Inquiry 1(1): 134–6. King, Paul John (1989). A logical formalism for Head Driven Phrase Structure Grammar. PhD thesis, University of Manchester.

Page 64 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References King, Paul John (1996). An expanded logical formalism for Head Driven Phrase Structure Grammar. Technical report, Working Papers of the SFB 340. Kinsui, Satoshi (1995). Nihongo-no iwayuru N’ sakuzyo-ni tuite [On so-called N’-deletion in Japanese], in Yasuaki Abe (ed.), The Reports for the Third Nanzan University Interna­ tional Symposium on Japanese Education and Linguistics. Nagoya: Nanzan University, 153–76. Kirchner, Jesse (2006). Sluicing and quasisluicing in Mandarin. Ms. UCSC. Kitagawa, Chisato and Claudia Ross (1982). Prenominal modification in Chinese and Japanese, Linguistic Analysis 9: 19–53. Kitagawa, Yoshihisa (1986). Subjects in Japanese and English. PhD thesis, University of Massachusetts, Amherst. Kitagawa, Yoshihisa (1991). Copying identity, Natural Language and Linguistic Theory 9(3): 497–536. Kitahara, Hisatsugu (1999). Eliminating * as a feature (of traces), in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, 77–93. Kizu, Mika (1997a). A note on sluicing in wh-in-situ languages, MIT Working Papers in Linguistics 36: 143–60. Kizu, Mika (1997b). Sluicing in wh-in-situ languages, in Kora Singer, Randall Eggert, and Gregory Anderson (eds), Proceedings of the Thirty-Third Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistics Society, 231–44. Klavans, Judith (1985). The syntax of code-switching: Spanish and English, in Larry D. King and Catherine A. Maley (eds), Selected papers from the XIIIth Linguistic Symposium on Romance Language. Amsterdam and Philadelphia: John Benjamins, 213–31. Klein, Ewan (1980). A semantics for positive and comparative adjectives, Linguistics and Philosophy 4(1): 1–45. Klein, Ewan (1991). Comparatives, in Arnim von Stechow and Dieter Wunderlich (eds), Semantik/Semantics: Ein internationales Handbuch der zeitgenössischen Forschung/An International Handbook of Contemporary Research. Berlin and New York: Mouton de Gruyter, 673–91. Klein, Elaine C. (1993). Toward second language acquistion: A study of Null Prep. Dor­ drecht: Kluwer Academic Publishers. Klein, Udo, Dolgor Guntsetseg, and Klaus von Heusinger (2012). Case in conflict: Embed­ ded subjects in Mongolian, in Monique Lamers and Peter de Swart (eds), Case, word or­ der, and prominence. Dordrecht: Springer, 43–64.

Page 65 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Klein, Wolfgang (1985). Ellipse, Fokusgliederung und thematischer Stand, in Reinhard Meyer-Hermann and Hannes Rieser (eds), Ellipsen und fragmentarische Ausdrücke. Tübingen: Niemeyer, 1–24. Klein, Wolfgang (1993). Ellipse, in Joachim Jacobs, Arnim von Stechow, Wolfgang Sterne­ feld, and Theo Vennemann (eds), Syntax: An international handbook of contemporary re­ search. Berlin and New York: Mouton de Gruyter, 763–99. Klein, Wolfgang and Christiane von Stutterheim (1987). Quaestio und referentielle Bewe­ gung in Erzählungen, Linguistische Berichte 109: 163–83. Vander Klok, Jozina (2016). Diagnosing VP-ellipsis in Javanese: Evidence for a non-move­ ment account, in Henrison Hsieh (ed.), AFLA 22: Proceedings of the 22nd meeting of the Austronesian Formal Linguistics Association. Canberra: Australian National University, 202–19. Kluck, Marlies (2009). Good neighbors or far friends. Matching and proximity ef­ fects in Dutch Right Node Raising, Groninger Arbeiten zur Germanistischen Linguistik 48: 115–58. (p. 1078)

Kluck, Marlies (2011). Sentence amalgamation. PhD thesis, University of Groningen. Kluck, Marlies (2014). Merchant’s wrinkle: The ban on Dutch bare R-pronoun ellipsis remnants. Paper presented at the West Coast Conference on Formal Linguistics 32 (WC­ CFL32). Kluck, Marlies and Mark de Vries (2013). Cumulative rightward processes, in Gert Webel­ huth, Manfred Sailer, and Heike Walker (eds), Rightward movement in a comparative per­ spective. Amsterdam and Philadelphia: John Benjamins, 281–317. Kluender, Robert (1998). On the distinction between strong and weak islands: A process­ ing perspective, in Peter W. Culicover and Louise McNally (eds), Syntax and semantics 29: The limits of syntax. New York: Academic Press, 241–79. Kluender, Robert and Simone Gieselman (2013). What’s negative about negative islands? A re-evaluation of extraction from weak island contexts, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge Universi­ ty Press, 186–207. Kobele, Gregory M. (2006). Generating copies: An investigation into structural identity in language and grammar. PhD thesis, University of California, Los Angeles. Kobele, Gregory M. (2012a). Eliding the derivation: A minimalist formalization of ellipsis, in Stefan Müller (ed.), Proceedings of the 19th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, 409–26. Kobele, Gregory M. (2012b). Ellipsis: Computation of, WIREs Cognitive Science 3(3): 411– 18. Page 66 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kobele, Gregory M. (2015). LF-copying without LF, Lingua 166: 236–59. Koeneman, Olaf, Sergio Baauw, and Frank Wijnen (1998). Reconstruction in VP-ellipsis: Reflexive vs. nonreflexive predicates. Poster presented at the 11th Annual Conference on Human Sentence Processing. Koizumi, Masatoshi (1995). Phrase structure in Minimalist syntax. PhD thesis, Massachu­ setts Institute of Technology, Cambridge, MA. Kolokonte, Marina (2008). Bare Argument Ellipsis and information structure. PhD thesis, Newcastle University. Konietzko, Andreas (2014). The syntax and information structure of Bare Argument Ellip­ sis in English and German: Experimental and theoretical evidence. PhD thesis, EberhardKarls-Universität Tübingen. Konietzko, Andreas, Janina Radó, and Susanne Winkler (forthcoming). Focus constraints on relative clause antecedents in sluicing, in Sam Featherston, Robin Hörnig, Sophia von Wietersheim, and Susanne Winkler (eds), Information structure and semantic processing. Berlin and New York: Mouton de Gruyter. Konietzko, Andreas and Susanne Winkler (2010). Contrastive ellipsis: Mapping between syntax and information structure, Lingua 120: 1436–57. Koopman, Hilda Judith and Dominique Sportiche (1991). The position of subjects, Lingua 85: 211–58. Koopman, Hilda Judith and Anna Szabolcsi (2000). Verbal complexes. Cambridge, MA: MIT Press. Kornfeld, Laura and Andrés Saab (2004). Nominal ellipsis and morphological structure in Spanish, in Reineke Bok-Benneman, Bart Hollebrandse, Brigitte Kampers-Manhe, and Pe­ tra Sleeman (eds), Romance languages and linguistic theory 2002. Amsterdam and Philadelphia: John Benjamins, 183–98. Kornfeld, Laura and Andrés Saab (2005). Hacía una tipología de las anáforas nominales en español. Talk given at the III Encuentro de Gramática Generativa. Koster, Jan (1975). Dutch as an SOV language, Linguistic Analysis 1: 111–36. Koster, Jan (1986). Domains and dynasties. Dordrecht: Foris Publications. Koulidobrova, Elena (2012). When the quiet surfaces: ‘Transfer’ of argument omission in the speech of ASL-English bilinguals. PhD thesis, University of Connecticut. Koutsoudas, Andreas (1971). Gapping, conjunction reduction, and coordinate deletion, Foundations of Language 7(3): 337–86. (p. 1079)

Kraak, Albert (1967). Presuppositions and the analysis of adverbs. Ms. MIT.

Page 67 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Krahmer, Emiel and Reinhard Muskens (1995). Negation and disjunction in discourse rep­ resentation theory, Journal of Semantics 12: 357–76. Kranendonk, Huib (2011). Quantificational constructions in the Dutch nominal domain: Facets of Dutch microvariation. PhD thesis, Utrecht University. Kratzer, Angelika (1996). Severing the external argument from its verb, in Johan Rooryck and Laurie Zaring (eds), Phrase structure and the lexicon. Dordrecht: Kluwer Academic Publishers, 109–37. Kratzer, Angelika and Junko Shimoyama (2002). Indeterminate pronouns: The view from Japanese, in Yukio Otsu (ed.), Proceedings of the Tokyo conference on psycholinguistics, vol. 3. Tokyo: Hituzi Syobo, 1–25. Krifka, Manfred (1991). A compositional semantics for multiple focus constructions, in Steven Moore and Adam Zachary Wyner (eds), Proceedings of the First Semantics and Linguistic Theory Conference. Ithaca, NY: Cornell University, 127–58. Krifka, Manfred (2006). Association with focus phrases, in V. Molnár and Susanne Win­ kler (eds), The architecture of focus. Studies in Generative Grammar 82. Berlin and New York: Mouton de Gruyter, 105–35. Krifka, Manfred (2008). Basic notions of information structure, Acta Linguistica Hungari­ ca 55(3–4): 243–376. Kubota, Yusuke (2015). Nonconstituent coordination in Japanese as constituent coordina­ tion: An analysis in Hybrid Type-Logical Grammar, Linguistic Inquiry 46: 1–42. Kubota, Yusuke and Robert Levine (2012). Gapping as like-category coordination, in De­ nis Bechet and Alexander Dikovsky (eds), Logical aspects of computational linguistics. Lecture notes in computer science 7351. Berlin: Springer, 135–50. Kubota, Yusuke and Robert Levine (2016). Gapping as hypothetical reasoning, Natural Language and Linguistic Theory 34: 107–56. Kubota, Yusuke and Robert Levine (2017). Pseudogapping as pseudo-VP-ellipsis, Linguis­ tic Inquiry 48(2): 213–57. Kučerová, I. and A. Neeleman (eds) (2012). Contrasts and positions in information struc­ ture. Cambridge: Cambridge University Press. Kuno, Susumu (1975). Conditions for verb phrase deletion, Foundations of Language 13: 161–75. Kuno, Susumu (1976). Gapping: A functional analysis, Linguistic Inquiry 7(2): 300–18. Kuno, Susumu (1978a). Danwa-no bunpoo. [Discourse grammar]. Tokyo: Taishukan.

Page 68 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Kuno, Susumu (1978b). Japanese: A characteristic OV language, in Winfred Lehmann (ed.), Syntactic typology. Austin: University of Texas Press, 57–138. Kuno, Susumu (1979). Newness of information and order of deletion, Cahiers Charles V: Recherches de l’Institut d’Anglais Charles 1: 211–21. Kuno, Susumu (1980). Discourse deletion, in Susumu Kuno (ed.), Harvard Studies in Syn­ tax and Semantics 3. Cambridge, MA: Harvard University, 1–144. Kuno, Susumu (1982). Principles of discourse deletion: Case studies from English, Russ­ ian, and Japanese, Journal of Semantics 1: 61–93. Kuno, Susumu (1986). Anaphora in Japanese, in Sige-Yuki Kuroda (ed.), Working Papers from the First SDF Workshop in Japanese Syntax. La Jolla: USCD, 11–70. Kuno, Susumu and Kiyoko Masunaga (1986). Questions with wh-phrases in islands, Uni­ versity of Massachusetts Occasional Papers in Linguistics 11: 139–65. Kunze, Jürgen (1975). Abhängigkeitsgrammatik. Studia Grammatica XII. Berlin: Akademie-Verlag. Kuroda, Sige-Yuki (1965). Generative grammatical studies in the Japanese language. PhD thesis, MIT. Kuroda, Sige-Yuki (1988). Whether we agree or not: A comparative syntax of English and Japanese, Linguisticae Investigationes 12: 1–47. Kuroda, Sige-Yuki (1992). Japanese syntax and semantics. Dordrecht: Kluwer Academic Publishers. Kush, Dave (2017). Notes on gapping in Hindi-Urdu: Conjunct size and focus-parallelism, Linguistic Analysis 40(3–4): 255–96. Kush, Dave, Jeffrey Lidz, and Colin Phillips (2015). Relation-sensitive retrieval: Evidence from bound variable pronouns, Journal of Memory and Language 82: 18–40. (p. 1080)

Kuwabara, Kazuki (1996). Multiple wh-phrases in elliptical clauses and some aspects of clefts with multiple foci, in Masatoshi Koizumi, Masayuki Oishi, and Uli Sauerland (eds), MIT Working Papers in Linguistics 29: Formal Approaches to Japanese Linguistics 2. Cambridge, MA: MITWPL, 97–116. Kuwabara, Kazuki (1997). On the properties of truncated clauses in Japanese, in Kazuko Inoue (ed.), Researching and verifying an advanced theory of human language. Chiba: Kanda University of International Studies, 61–84. Labov, William (1969). Contraction, deletion and the inherent variability of the English copula, Language 45(4): 715–62.

Page 69 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Labov, William (1971). The notion of system in Creole studies, in Dell Hymes (ed.), Pidginization and creolization of language. Cambridge: Cambridge University Press, 447– 72. Laczkó, Tibor (2007). On elliptical noun phrases in Hungarian, in Miriam Butt and Tracy Holloway King (eds), Proceedings of the LFG’07 Conference. Stanford: CSLI, 323–42. Ladd, D. Robert (2008). Intonational phonology. Cambridge: Cambridge University Press. Ladewig, Silva (2014). Creating multimodal utterances: The linear integration of gestures into speech, in Cornelia Müller, Alan Cienki, Ellen Fricke, Silva Ladewig, David McNeill, and Jana Bressem (eds), Body—Language—Communication. Berlin: Mouton de Gruyter, 1662–77. Laka, Itziar (1990). Negation in syntax: On the nature of functional categories and their projections. PhD thesis, Massachusetts Institute of Technology. Lakoff, George (1968). Pronouns and reference, in James D. McCawley (ed.), Syntax and semantics 7: Notes from the linguistic underground. New York: Academic Press, 275–335. Lakoff, George (1970). Global rules, Language 46: 627–39. Lakoff, George (1972). The arbitrary basis of transformational grammar, Language 48(1): 76–87. Lakoff, George (1974). Syntactic amalgams, in Michael Galy, Robert Fox, and Anthony Bruck (eds), Papers from the 10th regional meeting of the Chicago Linguistic Society. Chicago: University of Chicago, 321–44. Lakoff, George (1986). Frame semantic control of the Coordinate Structure Constraint, in Anne M. Farley, Peter T. Farley, and Karl-Erik McCullough (eds), Proceedings of the 22nd annual meeting of the Chicago Linguistic Society 22. Chicago: Chicago Linguistic Society, 152–67. Lakoff, George and Stanley Peters (1967). Phrasal conjunction and symmetric predicates, in David A. Reibel and Sandford A. Schane (eds), Modern studies in English: Readings in transformational grammar. Englewood Cliffs, NJ: Prentice Hall, 113–42. Lakoff, Robin T. (1971). Ifs, ands, and buts about conjunction, in Charles J. Fillmore and D. Terence Langendoen (eds), Studies in linguistic semantics. New York: Holt, Rinehart and Winston, 115–49. Lambrecht, Knud (1990). ‘What, me worry?’—‘Mad magazine sentences’ revisited, in Pro­ ceedings of the Annual Meeting of the Berkeley Linguistics Society, vol. 16. Berkeley: Berkeley Linguistics Society, 215–28. Lambrecht, Knud (1994). Information structure and sentence form: Topic, focus, and the mental representations of discourse referents. Cambridge: Cambridge University Press. Page 70 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lance, David (1975). Spanish-English code-switching, in Eduardo Hernández-Chávez, An­ drew D. Cohen, and Anthony F. Beltramo (eds), El lenguaje de los chicanos. Arlington, VA: Center for Applied Linguistics, 138–53. Landau, Idan (2013). Control in generative grammar: A research companion. Cambridge: Cambridge University Press. Langacker, Ronald W. (1969). On pronominalization and the chain of command, in David A. Reibel and Sanford A. Schane (eds), Modern Studies in English: Readings in Transfor­ mational Grammar. Englewood Cliffs, NJ: Prentice Hall, 160–86. Lappin, Shalom (1992). The syntactic basis of ellipsis resolution, in Steve Berman and Ar­ ild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart: Universität Stuttgart. Lappin, Shalom (1996). The interpretation of ellipsis, in Shalom Lappin (ed.), The handbook of contemporary semantic theory. Oxford: Blackwell, 145–75. (p. 1081)

Lappin, Shalom (1999). An HPSG account of antecedent-contained ellipsis, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Ox­ ford: Oxford University Press, 68–97. Lappin, Shalom (2005). A sequenced model of anaphora and ellipsis resolution, in António Branco, Tony McEnery, and Ruslan Mitkov (eds), Anaphora processing: Linguistic, cogni­ tive, and computational modelling. Amsterdam and Philadelphia: John Benjamins, 3–16. Lappin, Shalom and Michael McCord (1990). Anaphora resolution in slot grammar, Com­ putational Linguistics 16(4): 197–212. Larson, Bradley (2012). A dilemma with accounts of Right Node Raising, Linguistic In­ quiry 43: 143–50. Larson, Bradley (2013). The syntax of non-syntactic dependencies. PhD thesis, University of Maryland. Larson, Bradley (2014). Unelided basic remnants in Germanic ellipsis or: ÜBRIGE argu­ ments, University of Pennsylvania Working Papers in Linguistics 20(1): article 20. Larson, Richard (1988a). On the double object construction, Linguistic Inquiry 19(3): 335– 92. Larson, Richard (1988b). Scope and comparatives, Linguistics and Philosophy 11(1): 1– 26. Larson, Richard (1990). Double objects revisited: Reply to Jackendoff, Linguistic Inquiry 21(4): 589–632.

Page 71 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Larson, Richard and Hiroko Yamakido (2003). A new form of nominal ellipsis in Japanese, in Patricia M. Clancy (ed.), Japanese/Korean Linguistics 11. Stanford: CSLI Publications, 485–98. Larsson, Staffan (2002). Issue-based dialogue management. PhD thesis, Gothenburg Uni­ versity. Larsson, Staffan (2015). Formal semantics for perceptual classification, Journal of Logic and Computation 25(2): 335–69. Lasnik, Howard (1976). Remarks on coreference, Linguistic Analysis 2: 1–22. Lasnik, Howard (1995a). Case and expletives revisited: On greed and other human fail­ ings, Linguistic Inquiry 26(4): 615–34. Lasnik, Howard (1995b). Verbal morphology: Syntactic structures meets the minimalist program, in Hector Campos and Paula Kempchinsky (eds), Evolution and revolution in lin­ guistic theory. Washington, DC: Georgetown University Press, 251–75. Lasnik, Howard (1997). A gap in an ellipsis paradigm: Some theoretical implications, Lin­ guistic Analysis 27: 166–85. Lasnik, Howard (1999a). A note on pseudogapping, in Howard Lasnik (ed.), Minimalist analysis. Oxford: Blackwell Publishers, 151–74. Lasnik, Howard (1999b). Pseudogapping puzzles, in Shlom Lappin and Elabbas Ben­ mamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, 141–74. Lasnik, Howard (2001). When can you save a structure by destroying it?, in Minjoo Kim and Uri Strauss (eds), NELS 31: Proceedings of the 31st annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 301–20. Lasnik, Howard (2003). On feature strength: Three Minimalist approaches to overt move­ ment, Minimalist Investigations in Linguistic Theory 30: 197–217. Lasnik, Howard (2005). Review of The syntax of silence, by Jason Merchant, Language 81(1): 259–65. Lasnik, Howard (2010). On ellipsis: Is material that is phonetically absent but semantical­ ly present present or absent syntactically? in H. Götzche (ed.), Memory, mind and lan­ guage. Newcastle upon Tyne: Cambridge Scholars Publishing, 221–42. Lasnik, Howard (2011). Ellipsis and island violation repair. Handout for LING 819, Uni­ versity of Maryland. Lasnik, Howard (2014). Multiple sluicing in English? Syntax 17: 1–20.

Page 72 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lasnik, Howard and Myung-Kwan Park (2003). The EPP and the Subject Condition under sluicing, Linguistic Inquiry 34(4): 649–60. Lasnik, Howard and Myung-Kwan Park (2013). Locality and MaxElide in extrac­ tion out of elided VP, in Yoichi Miyamoto, Daiko Takahashi, Hideki Maki, Masao Ochi, Koji Sugisaki, and Asako Uchibori (eds), Deep insights, broad perspectives: Essays in honor of Mamuro Saito. Tokyo: Kaitakusha, 235–56. (p. 1082)

Lasnik, Howard and Mamoru Saito (1984). On the nature of proper government, Linguis­ tic Inquiry 15(2): 235–89. Lasnik, Howard and Mamoru Saito (1991). On the subject of infinitives, in Lise M. Dobrin, Lynn Nichols, and Rosa M. Rodriguez (eds), Proceedings of the 27th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 324–43. Lasnik, Howard and Mamoru Saito (1992). Move α. Cambridge, MA: MIT Press. Lasnik, Howard and Tim Stowell (1991). Weakest crossover, Linguistic Inquiry 22(4): 687– 720. Lebeaux, David (1990). Relative clauses, licensing, and the nature of the derivation, in Juli Carter, Dechaine Rose-Marie, Bill Philip, and Tim Sherer (eds), NELS 20: Proceedings of the 20th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, 318–32. Lebeaux, David (2009). Where does the Binding Theory apply? Cambridge, MA: MIT Press. Lechner, Winfried (1997). Some remarks on phrasal comparatives. Paper presented at the International Syntax Roundtable, Ypps/Danube, 6–9 July 1997. Lechner, Winfried (1998). Comparative deletion. PhD thesis, University of Massachusetts at Amherst. Lechner, Winfried (2001). Reduced and phrasal comparatives, Natural Language and Lin­ guistic Theory 19(4): 683–735. Lechner, Winfried (2004). Ellipsis in comparatives. Berlin and New York: Mouton de Gruyter. Lechner, Winfried (2015). Zwei bemerkungen zur Verbbewegung. Paper presented at the University of Hamburg. Lechner, Winfried (2016). On the typology of phrasal comparatives. Paper presented at the University of Vienna. Lechner, Winfried (forthcoming). Clausal vs. phrasal comparatives, in Daniel Gutzmann, Lisa Matthewson, Cécile Meier, Hotze Rullmann, and Thomas Ede Zimmermann (eds), The Blackwell Companion to Semantics. Malden: Blackwell. Page 73 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lechner, Winfried and Norbert Corver (2017). Comparative deletion and comparative subdeletion, in Martin Everaert and Henk van Riemsdijk (eds), The Wiley Blackwell com­ panion to syntax. 2nd edition. Hoboken: Wiley-Blackwell, 865–922. Lee, Yong-hun (2012). A unified approach to VP-ellipsis and VP-anaphora, in Stefan Müller (ed.), Proceedings of the 19th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, 271–90. Lees, Robert (1961). The English comparative construction, Word 17: 171–85. Leonetti, Manuel (1999). El artículo, in Ignacio Bosque and Violeta Demonte (eds), Gramática descriptiva de la lengua española. Madrid: Espasa Calpe, 787–890. Lerner, Jan and Manfred Pinkal (1995). Comparative ellipsis and variable binding, in Mandy Simons and Tim Galloway (eds), Semantics and linguistic theory. Ithaca, NY: Cor­ nell University Press, 222–36. Leung, Tommi (2014). The preposition stranding generalization and conditions on sluic­ ing: Evidence from Emirati Arabic, Linguistic Inquiry 45: 332–40. Levelt, Willem J. (1983). Monitoring and self-repair in speech, Cognition 14(4): 41–104. Levin, Lori (1982). Sluicing: A lexical interpretation procedure, in Joan Bresnan (ed.), The mental representation of grammatical relations. Cambridge, MA: MIT Press, 590–654. Levin, Nancy S. (1978). Some identity-of-sense deletions puzzle me, in D. Farkas, W. Ja­ cobsen, and K. Todrys (eds), Papers from the 14th regional meeting of the Chicago Lin­ guistic Society. Chicago: Chicago Linguistic Society, 229–40. Levin, Nancy S. (1980). Main verb ellipsis in spoken English, in Arnold M. Zwicky, Clitics and ellipsis. Columbus: Ohio State University, 65–165. Levin, Nancy S. (1986). Main-verb ellipsis in spoken English. New York: Garland Publishing. (p. 1083)

Levin, Nancy S. and Ellen F. Prince (1986). Gapping and clausal implicature, Papers in Linguistics 19: 351–64. Levine, Robert D. (1985). Right Node (non)-Raising, Linguistic Inquiry 16(3): 492–7. Levine, Robert D. (2001). The extraction riddle: Just what are we missing? Journal of Lin­ guistics 37: 145–74. Levine, Robert D. (2011). Linearization and its discontents, in Stefan Müller (ed.), Pro­ ceedings of the 18th International Conference on Head-Driven Phrase Structure Gram­ mar. Stanford: CSLI, 126–46. Levine, Robert D. and Yusuke Kubota (2013). Gapping as hypothetical reasoning. Ms. Ohio State University. Page 74 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lewis, David (1979). Score keeping in a language game, in Rainer Bauerle, Urs Egli, and Arnim von Stechow (eds), Semantics from different points of view. Dordrecht: Springer, 172–87. Lewis, Richard L., Shravan Vasishth, and Julie A. Van Dyke (2006). Computational princi­ ples of working memory in sentence comprehension, Trends in Cognitive Sciences 10: 44– 54. Liceras, Juana M., Raquel Fernández Fuertes, Susana Perales, Rocío Pérez-Tattam, and Kenton T. Spradlin (2008). Gender and gender agreement in bilingual native and non-na­ tive grammars: A view from child and adult functional-lexical mixings, Lingua 118(6): 827–51. Liddell, Scott K. (1980). American Sign Language syntax. The Hague: Mouton. Liddell, Scott K. (2003). Grammar, gesture, and meaning in American Sign Language. Cambridge: Cambridge University Press. Lin, Jo-Wang (2009). Chinese comparatives and their implicational parameters, Natural Language Semantics 17(1): 1–27. Lin, Vivian (1999). Determiner sharing, MIT Working Papers in Linguistics 33: 241–77. Lin, Vivian (2001). A way to undo A movement, in Karine Megerdoomian and Leora Anne Bar-el (eds), Proceedings of the 20th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 358–71. Lin, Vivian (2002). Coordination and sharing at the interfaces. PhD thesis, Massachusetts Institute of Technology. Lindenbergh, Charlotte, Angeliek van Hout, and Bart Hollebrandse (2015). Extending el­ lipsis research: The acquisition of sluicing in Dutch. Poster presented at the 39th Boston University Child Language Development Conference (BUCLD 39). Lipták, Anikó (2011). A fragmentumok mondattana a magyarban, Általános nyelvészeti tanulmányok 23: 317–49. Lipták, Anikó (2012). Verb-stranding ellipsis and verbal identity: The role of polarity fo­ cus, Linguistics in the Netherlands 29: 82–96. Lipták, Anikó (2013a). A note on overt case marking as evidence against pseudo-sluicing. Unpublished manuscript, University of Leiden. Lipták, Anikó (2013b). The syntax of positive polarity in Hungarian: Evidence from ellip­ sis, Lingua 128: 72–94. Lipták, Anikó (2015). Relative pronouns as sluicing remnants, in Éva Dékány, Katalin É. Kiss, and Balázs Surányi (eds), Approaches to Hungarian 14. Amsterdam and Philadel­ phia: John Benjamins, 187–207. Page 75 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lipták, Anikó and Andrés Saab (2010). Nominal ellipsis and the stranded affix filter. Pa­ per presented at Nijmegen University. Lipták, Anikó and Andrés Saab (2014). No N-raising out of NPs in Spanish: Ellipsis as a diagnostic of head movement, Natural Language and Linguistic Theory 32: 1247–71. Llombart-Huesca, Amàlia (2002). Anaphoric one and NP-ellipsis, Studia Linguistica 56(1): 59–89. Llombart-Huesca, Amàlia (2010). La elipsis nominal y los sintagmas introducidos por de, Cuadernos de la ALFAL 1: 39–53. Lobeck, Anne (1987). Syntactic constraints on ellipsis. PhD thesis, University of Washing­ ton, Seattle. Lobeck, Anne (1990). Functional heads as proper governors, in Juli Carter (ed.), NELS 20: Proceedings of the 20th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 348–62. (p. 1084)

Lobeck, Anne (1991). The phrase structure of ellipsis, in Susan Rothstein (ed.), Perspec­ tives on phrase structure. San Diego: Academic Press, 81–103. Lobeck, Anne (1992). Licensing and identification of ellipted categories in English, in Steve Berman and Arild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart. Lobeck, Anne (1993). Strong agreement and identification: Evidence from ellipsis in Eng­ lish, Linguistics 31: 777–811. Lobeck, Anne (1995). Ellipsis: Functional heads, licensing, and identification. New York: Oxford University Press. Lobeck, Anne (1999). VP ellipsis and the Minimalist Program: Some speculations and pro­ posals, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, 98–123. Lobeck, Anne (2006). Ellipsis in DP, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax. Malden, MA and Oxford: Blackwell, 145–73. Loewen, Gena (2011). The syntactic structure of noun phrases in Indonesian. Master’s thesis, University of Manitoba. Longobardi, Giuseppe (1994). Reference and proper names, Linguistic Inquiry 25(4): 609– 66. Lopes, Ruth E. Vasconcellos and Ana Lúcia Santos (2014). VP-ellipsis comprehension in European and Brazilian Portuguese, in João Costa, Alexandra Fiés, M. João Freitas, Maria

Page 76 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Lobo, and Ana Lúcia Santos (eds), New directions in the acquisition of Romance lan­ guages. Newcastle upon Tyne: Cambridge Scholars Publishing, 181–202. López, Luis (1995). Polarity and predicate anaphora. PhD thesis, Cornell University. López, Luis (2000). Ellipsis and discourse linking, Lingua 110: 183–213. López, Luis and Susanne Winkler (2000). Focus and topic in VP-anaphora constructions, Linguistics 38(4): 623–64. Ludlow, Peter (2005). A note on alleged cases of nonsentential assertion, in Reinaldo Elu­ gardo and Robert J. Stainton (eds), Ellipsis and nonsentential speech. Dordrecht: Kluwer Academic Publishers, 95–108. Lukasczyk, Ulrika (2008). Sanottua, ajateltua ja tehtyä: Referointi kolmessa suomalaisella viittomakielellä tuotetussa fiktiivisessä kertomuksessa [Referencing in three fictive FinSL narratives]. Master’s thesis, University of Jyväskylä. Macafee, Caroline (1980). Characteristics of non-standard grammar in Scotland. Ms. Uni­ versity of Aberdeen. Machobane, Malillo (2003). Variation in Bantu DP-structure: Evidence from Sesotho, Malawian Journal of Linguistics 3: 85–104. MacSwan, Jeff (1999). A Minimalist approach to intrasentential code-switching. New York: Garland Publishing. MacSwan, Jeff (2000). The architecture of the bilingual language faculty: Evidence from intrasentential code switching, Bilingualism: Language and Cognition 3(1): 37–54. MacSwan, Jeff (2005a). Codeswitching and generative grammar: A critique of the MLF model and some remarks on ‘modified minimalism’, Bilingualism: Language and Cogni­ tion 8(1): 1–22. MacSwan, Jeff (2005b). Remarks on Jake, Myers-Scotton and Gross’s response: There is no ‘matrix language’, Bilingualism: Language and Cognition 8(3): 277–84. MacSwan, Jeff (2012). Code-switching and grammatical theory, in Tej K. Bhatia and William C. Ritchie (eds), The handbook of bilingualism and multilingualism. Hoboken: Wi­ ley-Blackwell, 323–50. MacSwan, Jeff (2014). Grammatical theory and bilingual codeswitching. Cambridge: Cam­ bridge University Press. MacSwan, Jeff and Kara T. McAlister (2010). Naturalistic and elicited data in grammatical studies of codeswitching, Studies in Hispanic and Lusophone Linguistics 3(2): 521–32. MacWhinney, Brian (2000). The CHILDES project: Tools for analyzing talk. Mah­ wah, NJ: Lawrence Erlbaum Associates. (p. 1085)

Page 77 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Maeda, Masako and Daiko Takahashi (2013). NP-ellipsis in the Naghasaki dialect of Japanese. Paper presented at the 23rd Japanese/Korean Linguistics Conference, MIT. Mahootian, Shahrzad (1993). A null theory of code-switching. PhD thesis, Northwestern University. Makuuchi, Michiru, Yosef Grodzinsky, Katrin Amunts, Andrea Santi, and Angela D. Friederici (2013). Processing noncanonical sentences in Broca’s region: Reflections of movement distance and type, Cerebral Cortex 23: 694–702. Maling, Joan M. (1972). On ‘gapping and the order of constituents’, Linguistic Inquiry 3(1): 101–8. Malouf, Robert (1998). Mixed categories in the hierarchical lexicon. PhD thesis, Stanford University. Manning, Christopher and Ivan Sag (1999). Argument structure, valence, and binding, Nordic Journal of Linguistics 21: 107–44. Manus, Sophie and Cédric Patin (2011). The prosodic structure of elliptical constructions in Bantu. Paper presented at the 4th International Conference on Bantu Languages (B4ntu). Manzini, Maria Rita (1992). Locality, vol. 19. Cambridge, MA: MIT Press. Marandin, Jean-Marie (1996). Pas d’identité sans identité: L’analyse des groupes nom­ inaux en Det + A, in Bernard Fradin and Jean-Marie Marandin (eds), Mot et grammaire. Paris: Didier Erudition, 129–64. Marchena, Oscar Garcia (2015). Phrases averbales et fragments de l’espagnol oral. PhD thesis, Université Paris Diderot. Martí, Luisa (2015). Grammar vs. pragmatics: Carving nature at the joints, Mind & Lan­ guage 30: 437–73. Martin, Andrea E. and Brian McElree (2008). A content-addressable pointer mechanism underlies comprehension of verb-phrase ellipsis, Journal of Memory and Language 58: 879–906. Martin, Andrea E. and Brian McElree (2009). Memory operations that support language comprehension: Evidence from verb-phrase ellipsis, Journal of Experimental Psychology: Learning Memory and Cognition 35: 1231–9. Martin, Andrea E. and Brian McElree (2011). Direct-access retrieval during sentence comprehension: Evidence from sluicing, Journal of Memory and Language 64: 327–43. Martin, Roger (1992). On the distribution and case features of PRO. Ms. University of Connecticut. Page 78 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Martin, Roger (1996). A Minimalist theory of PRO and control. PhD thesis, University of Connecticut. Martin, Roger (2001). Null case and the distribution of PRO, Linguistic Inquiry 32(1): 141–66. Martins, Ana-Maria (1994). Enclisis, VP-deletion and the nature of sigma, Probus 6(2/3): 173–206. Marušič, Franc and Rok Žaucer (2013). A note on sluicing and island repair, in Steven Franks (ed.), Annual Workshop on Formal Approaches to Slavic Linguistics: The Third In­ diana Meeting 2012. Ann Arbor: Michigan Slavic Publications, 176–89. Matos, Gabriela (1992). Construções de elipse do predicado em Português: SV nulo e de­ spojamento. PhD thesis, University of Lisbon. Matsuo, Ayumi (2007). Differing interpretations of empty categories in English and Japan­ ese VP ellipsis contexts, Language Acquisition 14(1): 3–29. Matthews, Peter (1981). Syntax. Cambridge: Cambridge University Press. May, Robert (1985). Logical Form: Its structure and derivation. Cambridge, MA: MIT Press. Mayr, Clemens (2012). Focusing bound pronouns, Natural Language Semantics 20(3): 299–348. McCawley, James D. (1970). On the applicability of vice versa, Linguistic Inquiry 1(2): 278–80. McCawley, James D. (1982). Parentheticals and discontinuous constituent structure, Lin­ guistic Inquiry 13(1): 91–106. McCawley, James D. (1987). Some additional evidence for discontinuity, in Geoffrey Huck and Almerindo E. Ojeda (eds), Syntax and semantics, vol. 20: Discontinuous constituency. Orlando: Academic Press, 185–200. McCawley, James D. (1993). Gapping with shared operators, in David A. Peterson (ed.), Proceedings of the annual meeting of the Berkeley Linguistics Society. Berkeley: Berke­ ley Linguistics Society, 245–53. McCawley, James D. (1998). The syntactic phenomena of English. Chicago: Uni­ versity of Chicago. (p. 1086)

McCloskey, James (1986). Right node raising and preposition stranding, Linguistic Inquiry 17: 183–6.

Page 79 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References McCloskey, James (1990). Resumptive pronouns, A’-binding, and levels of representation in Irish, in Randall Hendrick (ed.), The syntax of the Modern Celtic languages. San Diego: Academic Press, 199–248. McCloskey, James (1991a). Clause structure, ellipsis, and proper government in Irish, Lin­ gua 85: 259–302. McCloskey, James (1991b). There, it, and agreement, Linguistic Inquiry 22: 563–7. McCloskey, James (2004). A note on predicates and heads in Irish clausal syntax, in An­ drew Carnie, Heidi Harley, and Sheila Ann Dooley (eds), Verb first: On the syntax of verbinitial languages. Amsterdam and Philadelphia: John Benjamins, 155–74. McCloskey, James (2007). A language at the edge: Irish and the theory of grammar. Paper presented at the University of North Carolina, Chapel Hill. McCloskey, James (2009). Subjects, ellipsis, and the EPP in a VSO language. Paper pre­ sented at USC. McClure, Erica. F. (1977). Aspects of code-switching in the discourse of bilingual Mexi­ can-American children. Technical Report 044. Urbana: Center for the Study of Reading, Technical Report. McConnell-Ginet, Sally (1973). Comparative constructions in English: A syntactic and se­ mantic analysis. PhD thesis, University of Rochester. McDowell, Joyce P. (1987). Assertion and modality. PhD thesis, University of Southern California. McElree, Brian and Thomas G. Bever (1989). The psychological reality of linguistically de­ fined gaps, Journal of Psycholinguistic Research 18: 21–35. McKillen, Alanah (2016). On the interpretation of reflexive pronouns. PhD dissertation, McGill University, Montreal. McKoon, Gail, Roger Ratcliff, and Gregory Ward (1994). Testing theories of language pro­ cessing: An empirical investigation of the on-line lexical decision task, Journal of Experi­ mental Psychology: Learning Memory and Cognition 20: 1219–28. McNally, Louise (1992). VP-coordination and the VP-internal subject hypothesis, Linguis­ tic Inquiry 23: 336–41. McNeill, David (1992). Hand and mind: What gestures reveal about thought. Chicago: University of Chicago Press. McShane, Marjorie (2005). A theory of ellipsis. New York: Oxford University Press. Megerdoomian, Karin (2002). Beyond words and phrases: A unified theory of predicate composition. PhD thesis, University of Southern California. Page 80 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Mehler, Jacques (1963). Some effects of grammatical transformations on the recall of English sentences, Journal of Verbal Learning and Verbal Behavior 2: 346–51. Mel’čuk, Igor (1988). Dependency syntax: Theory and practice. Albany, NY: State Univer­ sity of New York Press. Mel’čuk, Igor (2003). Levels of dependency description: Concepts and problems, in Vil­ mos Ágel, Ludwig M. Eichinger, Hans-Werner Eroms, Peter Hellwig, Hans Jürgen Heringer, and Henning Lobin (eds), Dependency and valency: An international handbook of contemporary research. Berlin: Walter de Gruyter, 188–229. Melnik, Nurit (2007). Extending partial pro-drop in modern Hebrew: A comprehensive analysis, in Stefan Müller (ed.), Proceedings of the 14th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 173–93. Merchant, Jason (1998). ‘Pseudosluicing’: Elliptical clefts in Japanese and English, ZAS Working Papers in Linguistics 10: 88–112. Merchant, Jason (1999). The syntax of silence: Sluicing, islands, and identity in ellipsis. PhD thesis, University of California, Santa Cruz. Merchant, Jason (2000). Antecedent-contained deletion in negative polarity items, Syntax 3(2): 144–50. Merchant, Jason (2001). The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford: Oxford University Press. (p. 1087)

Merchant, Jason (2002). Swiping in Germanic, in C. Jan-Wouter Zwart and Werner Abra­ ham (eds), Studies in comparative Germanic syntax. Amsterdam: John Benjamins, 295– 321. Merchant, Jason (2003a). Remarks on stripping. Ms. University of Chicago. Merchant, Jason (2003b). Subject-auxiliary inversion in comparatives and PF output con­ straints, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and in­ terpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, 55–77. Merchant, Jason (2003c). Section on stripping excised from submitted version of ‘Frag­ ments and ellipsis’, Linguistics and Philosophy 27(6): 661–738. Merchant, Jason (2004a). Fragments and ellipsis, Linguistics and Philosophy 27(6): 661– 738. Merchant, Jason (2004b). Resumptivity and non-movement, Studies in Greek Linguistics 24: 471–81. Merchant, Jason (2005). Revisiting syntactic identity conditions. Paper presented at the Workshop on Ellipsis, University of California, Berkeley. Page 81 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Merchant, Jason (2006a). Sluicing, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax. Maldon, MA and Oxford: Blackwell, 269–89. Merchant, Jason (2006b). Small structures: A sententialist perspective, in Ljiljana Progov­ ac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 73–91. Merchant, Jason (2006c). Why no(t)?, Style 20(1–2): 20–3. Merchant, Jason (2008a). An asymmetry in voice mismatches in VP-ellipsis and pseudo­ gapping, Linguistic Inquiry 39(1): 169–79. Merchant, Jason (2008b). Variable island repair under ellipsis, in Kyle Johnson (ed.), Top­ ics in ellipsis. Cambridge: Cambridge University Press, 132–53. Merchant, Jason (2009). Phrasal and clausal comparatives in Greek and the abstractness of syntax, Journal of Greek Linguistics 9: 134–64. Merchant, Jason (2010). Three types of ellipsis, in François Recanati, Isidora Stojanovic, and Neftali Villanueva (eds), Context-dependence, perspective and relativity. Berlin and New York: Mouton de Gruyter, 141–94. Merchant, Jason (2011). Ellipsis: A survey of analytical approaches. Ms. University of Chicago. Merchant, Jason (2013a). Diagnosing ellipsis, in Lisa Lai-Shen Cheng and Norbert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, 537–42. Merchant, Jason (2013b). Polarity items under ellipsis, in Lisa Lai-Shen Cheng and Nor­ bert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, 441–62. Merchant, Jason (2013c). Revisiting conditions on predicate anaphora, in Philip Hofmeis­ ter and Elisabeth Norcliffe (eds), The core and the periphery: Data-driven perspectives on syntax inspired by Ivan A. Sag. Stanford: CSLI Publications, 1–4. Merchant, Jason (2013d). Voice and ellipsis, Linguistic Inquiry 44(1): 77–108. Merchant, Jason (2016). An embarrassment of riches? Cutting up the elliptical pie, Theo­ retical Linguistics 42(3–4): 297–304. Merchant, Jason (2014a). Gender mismatches under nominal ellipsis, Lingua 151: 9–32. Merchant, Jason (2014b). Greek-English bilingual code-switching under ellipsis. Paper presented at the 4th Annual Midwest Greek Linguistics Workshop, 4 April, University of Chicago. Merchant, Jason (2015a). How much context is enough? Two cases of span-conditioned stem allomorphy, Linguistic Inquiry 46: 273–304.

Page 82 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Merchant, Jason (2015b). On ineffable predicates: Bilingual Greek-English code-switching under ellipsis, Lingua 166: 199–213. Merchant, Jason and Andrew Simpson (eds) (2012). Sluicing: Cross-linguistic perspectives. Oxford Studies in Theoretical Linguistics 38. Oxford: Oxford University Press. Merchant, Jason, Lyn Frazier, Charles Jr. Clifton, and Thomas Weskott (2013). Fragment answers to questions: A case of inaudible syntax, in Laurence Goldstein (ed.), Brevity. Oxford: Oxford University Press, 21–35. (p. 1088)

Messick, Troy and Gary Thoms (2016). Ellipsis, economy, and the (non)uniformity of traces, Linguistic Inquiry 47: 306–32. Messick, Troy, Andrés Saab, and Luis Vicente (2015). Deep properties of a surface anaphor: Contextual restrictions of sluicing sites. Ms. University of Connecticut, Universi­ ty of Buenos Aires, and University of Potsdam. Miller, Philip (1990). Pseudogapping and do so substitution, in Michaela Ziolkowski, Manuela Noske, and Karen Deaton (eds), CLS 26: Papers from the 26th Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 293–305. Miller, Philip (1997a). Auxiliary verbs in Old and Middle French. A diachronic study of substitutive ‘fair’ and a comparison with the Modern English auxiliaries, in Ans van Ke­ menade and Nigel Vincent (eds), Parameters of morphosyntactic change. Cambridge: Cambridge University Press, 119–33. Miller, Philip (1997b). Les morphèmes zéro à l’épreuve du rasoir d’Occam, in Jean Chu­ quet and Marc Fryd (eds), Travaux linguistiques du Cerlico. Rennes: Presses Universi­ taires de Rennes, 13–42. Miller, Philip (2011a). The choice between verbal anaphors in discourse, in Ilse Hen­ drickx, Sobha Lalitha Devi, António Branco, and Ruslan Mitkov (eds), Discourse anaphora and anaphor resolution colloquium. Heidelberg: Springer, 82–95. Miller, Philip (2011b). The use of the English verbal anaphor ‘do so’ in discourse: Descrip­ tive analysis and theoretical consequences. Paper presented at Typology and Language Universals Workshop, University of Paris 7. Miller, Philip (2014). A corpus study of pseudo-gapping and its theoretical consequences, in Christopher Piñón (ed.), Empirical issues in syntax and semantics 10. Paris: CSSP, 73– 90. Miller, Philip and Barbara Hemforth. (2014a). Verb phrase ellipsis with nominal an­ tecedents. Ms. Université Paris, Diderot. Miller, Philip and Barbara Hemforth (2014b). VP ellipsis beyond syntactic identity: The case of nominal antecedents. Linguistic Evidence. DOI: 10.13140/2.1.4713.2488. Page 83 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Miller, Philip and Geoffrey K. Pullum (2014). Exophoric VP-ellipsis, in Philip Hofmeister and Elisabeth Norcliffe (eds), The core and the periphery: Data-driven perspectives on syntax inspired by Ivan A. Sag. Stanford: CSLI Publications, 5–32. Miller, Philip and Ivan A. Sag (1997). French clitic movement without clitics or move­ ment, Natural Language and Linguistic Theory 15(3): 573–639. Milner, Jean-Claude (1990). Some remarks on Principle C, in Anna Maria di Sciullio and Anne Rochette (eds), Binding in Romance. Ottawa: Canadian Linguistic Association, 41– 67. Miyagawa, Shigeru (1997). Against optional scrambling, Linguistic Inquiry 28(1): 1–25. Miyagawa, Shigeru (2006). On the ‘undoing’ property of scrambling: A response to Bošković, Linguistic Inquiry 37(4): 607–24. Miyagawa, Shigeru (2010). Why Agree? Why Move? Unifying agreement-based and dis­ course-configurational languages. Cambridge, MA: MIT Press. Miyamoto, Yoichi (2016). Meisikunai-no syooryaku [Ellipsis within noun phrases], in Keiko Murasugi, Mamoru Saito, Yoichi Miyamoto, and Kensuke Takita (eds), Nihongo bunpoo handobukku [A handbook of Japanese grammar]. Tokyo: Kaitakusha, 265–98. Mohamed-Soyir, Kassim (2014). Le nom en shingazidja (G44a): Morphologie, phonologie, sémantique et syntaxe. PhD thesis, Université Paris Diderot. Mohammad, Jan and Simin Karimi (1992). ‘Light’ verbs are taking over: Complex verbs in Persian, in Joel Ashmore Nevis and Vida Samiian (eds), Proceedings of the Western Con­ ference on Linguistics (WECOL 5). Fresno, CA: Department of Linguistics, 195–215. Moline, Estelle (2013). ‘Comme’ et ‘comment’ en français contemporain: Spécificités et recouvrements, in Antoine Gautier and Thomas Verjans (eds), ‘Comme’, ‘comment’, ‘com­ bien’: Concurrences et complémentarité. Paris: L’Harmattan, 57–96. Molnár, Valeria (2006). On different kinds of contrast, in Valeria Molnár and Su­ sanne Winkler (eds), The architecture of focus. Berlin and New York: Mouton de Gruyter, 197–234. (p. 1089)

Molnár, Valeria and Susanne Winkler (2010). Edges and gaps: Contrast at the interfaces, Lingua 120: 1392–415. Moltmann, Friederike (1992). Coordination and comparatives. PhD thesis, MIT. Montague, Richard (1970). English as a formal language, in Bruno Visentine (ed.), Lin­ guaggi nella società e nella tecnica. Milan: Edizioni di Comunità, 189–224. Montague, Richard (1974). The proper treatment of quantification in ordinary English, in Richmond H. Thomason (ed.), Formal philosophy: Selected papers of Richard Montague. New Haven: Yale University Press, 247–70. Page 84 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Montali, Emanuele (2014). Sluicing and adjectives. Term paper, UCL. Morgan, Jerry (1973). Sentence fragments and the notion ‘sentence’, in Brai Kachru, Robert Lees, Yakov Malkiel, Angelina Pietrangeli, and Sol Saporta (eds), Issues in linguis­ tics: Papers in honor of Henry and Renée Kahan. Urbana: University of Illinois Press, 719–51. Morris, Amanda (2008). Polarity ellipsis and negative stripping. Master’s thesis, UCSC. Morzycki, Marcin (2015). Modification. Key topics in semantics and pragmatics. Cam­ bridge: Cambridge University Press. Moulton, Keir (2007). Small antecedents: Syntax or pragmatics, in Emily Elfner and Mar­ tin Walkow (eds), NELS 37: Proceedings of the 37th annual meeting of the North East Linguistics Society. Amherst, MA: GLSA, 45–58. Moulton, Keir (2013). Small antecedents. Ms. Simon Fraser University. Mouret, François (2006). A phrase structure approach to argument cluster coordination, in Stefan Müller (ed.), Proceedings of the 13th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, 247–67. Mouret, François (2007a). Grammaire des constructions coordonnées: Coordinations sim­ ples et coordinations à redoublement en français contemporain. PhD thesis, Université Paris Diderot. Mouret, François (2007b). Syntaxe et sémantique des construction en ‘ni’, Faits de langues 28: 196–205. Mouret, François (2008). Les coordinations de terms dissemblables sont-elles elliptiques? in Jacques Durand, Benoît Habert, and Bernard Laks (eds), Actes du Congrés Mondial de Linguistique française. Paris: CMLF Online Publications. Mouret, Francois and Anne Abeille (2011). On the rule of Right Node Raising in French. Handout at the Colloque international ‘Topics in the Typology of Elliptical Constructions’. University of Chicago Center, Paris, June. Muadz, H. (1991). Coordinate structures: A planar representation. PhD thesis, University of Arizona, Tucson. Mukai, Emi (2003). On verbless conjunction in Japanese, in Makoto Kadowaki and Shige­ to Kawahara (eds), NELS 33: Proceedings of the 33rd Annual Meeting of the North East Linguistic Society. Amherst, MA: GLSA, 205–24. Müller, Claude (1997). ‘De’ partitif et la négation, in Danielle Forget, Paul Hirschbühler, France Martineau, and Maria-Luisa Rivero (eds), Negation and polarity. Amsterdam and Philadelphia: John Benjamins, 251–70.

Page 85 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Müller, Gereon (1998). Incomplete category fronting: A derivational approach to remnant movement in German. Dordrecht: Kluwer Academic Publishers. Müller, Gereon (2011). Constraints on displacement: A phase-based approach. Amster­ dam and Philadelphia: John Benjamins. Müller, Stefan (2015a). Grammatical theory: From transformational grammar to con­ straint-based approaches. Berlin: Language Science Press. Müller, Stefan (2015b). HPSG—A synopsis, in Tibor Kiss and Artemis Alexiadou (eds), Syntax—Theory and analysis: An international handbook. Berlin: Mouton de Gruyter, 937– 73. Munn, Alan Boag (1993). Topics in the syntax and semantics of coordinate structures. PhD thesis, University of Maryland. Murasugi, Keiko (1991). Noun phrases in Japanese and English: A study in syn­ tax, acquisition and learnability. PhD thesis, University of Connecticut. (p. 1090)

Murasugi, Keiko (2005). Sakuzyo [Deletion], in Masatake Arimoto and Keiko Murasugi (eds), Sobukaku to sakuzyo [Binding and deletion]. Tokyo: Kenkyusha, 130–229. Murphy, Gregory (1985). Processes of understanding anaphora, Journal of Memory and Language 24: 290–303. Musolino, Julien and Jeffrey Lidz (2004). The scope of isomorphism: Turning adults into children, Language Acquisition 11: 277–91. Muysken, Pieter (1995). Code-switching and grammatical theory, in Lesley Milroy and Pieter Muysken (eds), One speaker, two languages. Cambridge: Cambridge University Press, 177–98. Muysken, Pieter and Henk van Riemsdijk (1986). Projecting features and featuring pro­ jections, in Pieter Muysken and Henk van Riemsdijk (eds), Features and projections. Dor­ drecht: Foris Publications, 1–30. Myers-Scotton, Carol (1993). Dueling languages: Grammatical structure in codeswitching. Oxford: Clarendon Press. Myers-Scotton, Carol (2001). The matrix language frame model: Developments and re­ sponses, in Rodolfo Jacobson (ed.), Codeswitching worldwide II. Berlin and New York: Mouton de Gruyter, 23–58. Myers-Scotton, Carol and Janice L. Jake (2001). Explaining aspects of codeswitching and their implications, in Janet L. Nicol (ed.), One mind, two languages: Bilingual language processing. Malden: Blackwell, 84–116.

Page 86 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Myers-Scotton, Carol and Janice L. Jake (2017). Revisiting the 4-M model: Codeswitching and morpheme election at the abstract level, International Journal of Bilingualism 21(3): 340–66. Nagata, Hiroshi (1989). Effect of repetition on grammaticality judgments under objective and subjective self-awareness conditions, Journal of Psycholinguistic Research 18(3): 255– 69. Nakamura, Masanori (2012). Case morphology and island repair, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, 104–22. Nakao, Chizuru (2009). Island repair and non-repair by PF strategies. PhD thesis, Univer­ sity of Maryland. Nakao, Chizuru and Masaya Yoshida (2005). Japanese sluicing as a specificational pseu­ do-cleft, in Yukio Otsu (ed.), Proceedings of the Sixth Tokyo Conference on Psycholinguis­ tics. Tokyo: Hituzi Syobo, 243–68. Nakao, Chizuru and Masaya Yoshida (2007). ‘Not-so-propositional’ islands and their impli­ cations for swiping, in Erin Bainbridge and Brian Agbanyani (eds), Proceedings of the Western Conference on Linguistics (WECOL 2006). Fresno, CA: University of California Press, 323–33. Napoli, Donna Jo (1982). Initial material deletion in English, Glossa 16: 85–111. Napoli, Donna Jo (1983a). Comparative ellipsis: A phrase structure analysis, Linguistic In­ quiry 14(4): 675–94. Napoli, Donna Jo (1983b). Missing complement sentences in English: A base analysis of Null Complement Anaphora, Linguistic Analysis 12: 1–28. Napoli, Donna Jo (1985). Verb phrase deletion in English: A base-generated analysis, Jour­ nal of Linguistics 21: 281–319. Nee, Julia (2012). Spanish/Zapotec code-switching and the theory of ellipsis. BA thesis, University of Chicago. Neeleman, Ad and Hans van de Koot (2002). The configurational matrix, Linguistic In­ quiry 33(4): 529–74. Neeleman, Ad, Elena Titov, Hans van de Koot, and Reiko Vermeulen (2009). A syntactic typology of topic, focus, and contrast, in Jeroen van Craenenbroeck (ed.), Alternatives to cartography. Berlin: Mouton de Gruyter, 15–52. Neijt, Anneke (1979). Gapping: A contribution to sentence grammar. Dordrecht: Foris Publications.

Page 87 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Neijt, Anneke (1981a). Gapping and remnants: Sentence grammar aspects of gapping, Linguistic Analysis 8: 69–93. (p. 1091)

Neijt, Anneke (1981b). Gapping bestaat, Spektator 11: 80–4. Nerbonne, John, Masayo Iida, and William Ladusaw (1989). Running on empty: Null heads in Head-Driven Grammar, in Jane Fee and Katherine Hunt (eds), Proceedings of the Eighth West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, 276– 88. Nerbonne, John, Masayo Iida, and William Ladusaw (1990). Semantics of common Noun Phrase anaphora, in Aaron L. Halpern (ed.), The Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, 379–94. Nerbonne, John and Tony Mullen (2000). Null-headed nominals in German and English, in Frank van Eynde, Ineke Schuurman, and Ness Schelkens (eds), Proceedings of Computa­ tional Linguistics in the Netherlands: Selected Papers from the Ninth CLIN Meeting. Am­ sterdam and Philadelphia: John Benjamins, 143–64. Nevins, Andrew and Pranav Anand (2003). Some AGREEment matters, in M. Tsujimura (ed.), Proceedings of the West Coast Conference on Formal Linguistics 22. Somerville, MA: Cascadilla Press, 370–83. Ngonyani, Deo (1996). VP ellipsis in Ndenduele and Swahili applicatives, in Edward Gar­ rett and Felicia Lee (eds), Syntax at sunset. UCLA Working Papers in Syntax and Seman­ tics 1. Los Angeles: University of California Los Angeles, 109–28. Ngonyani, Deo (1998). V-to-I movement in Kiswahili, Afrikanistische Arbeitspapiere 55: 129–44. Ngonyani, Deo and Peter Githinji (2006). The asymmetric nature of Bantu applicative con­ structions, Lingua 116: 31–63. Nicol, Janet, Janet Dean Fodor, and David Swinney (1994). Using cross-modal lexical deci­ sion tasks to investigate sentence processing, Journal of Experimental Psychology: Learn­ ing Memory and Cognition 20: 1228–38. Nicol, Janet and David Swinney (1989). The role of structure in coreference assignment during sentence comprehension, Journal of Psycholinguistic Research 18(1): 5–19. Nicol, Janet and David Swinney (2003). The psycholinguistics of anaphora, in Andrew Barss (ed.), Anaphora: A reference guide. Oxford: Oxford University Press, 72–104. Nicolae, Alexandru (2012). P-stranding in a language without P-stranding? The case of sluicing in Romanian. Paper presented at the 45th annual meeting of the Societas Lin­ guistica Europaea, University of Stockholm.

Page 88 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Nielsen, Leif Arda (2003). A corpus-based study of verb phrase ellipsis identification and resolution. Ms. King’s College London. . Nishigauchi, Taisuke (1990). Quantification in the theory of grammar, vol. 37. Dordrecht: Kluwer Academic Publishers. Nishigauchi, Taisuke (1998). ‘Multiple sluicing’ in Japanese and the functional nature of wh-phrases, Journal of East Asian Linguistics 7(2): 121–52. Nishigauchi, Taisuke (1999). Ronri koozoo to bunpoo riron: Nitieigo no WH gensyoo [Log­ ical structure and the theory of grammar: WH phenomena in Japanese and English]. Tokyo: Kurioso. Nishigauchi, Taisuke and Tomohiro Fujii (2006). Short answers: ellipsis, connectivity, and island repair. Ms. Kobe Shoin Graduate School and University of Maryland. Nishiyama, Kunio, John Whitman, and Eun-Young Yi (1996). Syntactic movement of overt wh-phrases in Japanese and Korean, in Noriko Akatsuka, Shoichi Iwasaki, and Susan Strauss (eds), Japanese and Korean linguistics. Stanford: CSLI Publications, 337–51. Nissenbaum, Jon (1998). Derived predicates and the interpretation of parasitic gaps, in Kimary Shahin, Susan Blake, and Eun-Sook Kim (eds), Proceedings of the 17th West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, 247–95. Nunes, Jairo (2004). Linearization of chains and sideward movement. Linguistic Inquiry Monographs. Cambridge, MA: MIT Press. (p. 1092)

Nunes, Jairo and Cynthia Zocca (2009). Lack of morphological identity and ellip­

sis resolution in Brazilian Portuguese, in Jairo Nunes (ed.), Minimalist essays on Brazilian Portuguese syntax. Amsterdam and Philadelphia: John Benjamins, 215–36. Nykiel, Joanna (2013). Clefts and preposition omission under sluicing, Lingua 123: 74– 117. Nykiel, Joanna (2015). Constraints on ellipsis alternation: A view from the history of Eng­ lish, Language Variation and Change 27: 227–54. Nykiel, Joanna (2017). Preposition stranding and ellipsis alternation, English Language and Linguistics 21(1): 27–45. Nykiel, Joanna and Ivan A. Sag (2008). Sluicing and stranding. Ms. University of Silesia and Stanford University. Obdeijn, Margreet (2005). Discourse anaphora in children and the role of ‘er’ [there] in Dutch. BA thesis, Universiteit van Utrecht.

Page 89 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Oehrle, Richard T. (1987). Boolean properties in the analysis of gapping, in Geoffrey J. Huck and Almerindo E. Ojeda (eds), Syntax and semantics: Discontinuous constituency. San Diego, CA: Academic Press, 203–40. Oehrle, Richard T. (1991). Categorial frameworks, coordination, and extraction, in Aaron L. Halpern (ed.), Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford, CA: Center for the Study of Language and Information, 411–26. O’Grady, William (1998). The syntax of idioms, Natural Language and Linguistic Theory 16: 279–312. Oirsouw, Robert R. van (1987). The syntax of coordination. New York: Croom Helm. Okrent, Anrika (2002). A modality-free notion of gesture and how it can help us with the morpheme vs. gesture question in sign language linguistics (or at least give us some cri­ teria to work with), in Richard P. Meier, Kearsy Cornier, and David Quinto-Pozos (eds), Modality and structure in signed and spoken languages. Cambridge: Cambridge Universi­ ty Press, 175–98. Oku, Satoshi (1998). A theory of selection and reconstruction in the Minimalist perspec­ tive. PhD thesis, University of Connecticut. Olsvay, Csaba (2000). Formális jegyek egyeztetése a magyar nemsemleges mondatokban, in László Büky and Márta Maleczki (eds), A mai magyar nyelv leírásának újabb módszerei. Szeged: Szegedi Tudományegyetem, 119–51. Olsvay, Csaba (2004). The Hungarian verbal complex: An alternative approach, in Katalin É. Kiss and Henk van Riemsdijk (eds), Verb clusters: A study of Hungarian, German and Dutch. Amsterdam and Philadelphia: John Benjamins, 290–333. Ortega-Santos, Iván, Masaya Yoshida, and Chizuru Nakao (2014). On ellipsis structures involving a wh-remnant and a non-wh-remnant simultaneously, Lingua 138: 55–85. Osborne, Timothy (2005). Beyond the constituent: A Dependency Grammar analysis of chains, Folia Linguistica 39(3–4): 251–97. Osborne, Timothy (2006). Shared material and grammar: A dependency grammar theory of nongapping coordination, Zeitschrift für Sprachwissenschaft 25: 39–93. Osborne, Timothy (2008). Major constituents and two dependency grammar constraints on sharing coordination, Linguistics 46: 1109–65. Osborne, Timothy (2009). Comparative coordination vs. comparative subordination, Nat­ ural Language and Linguistic Theory 27: 427–54. Osborne, Timothy (2012). Edge features, catenae, and dependency-based Minimalism, Linguistic Analysis 34(3–4): 321–66.

Page 90 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Osborne, Timothy (2014). Dependency grammar, in Andrew Carnie, Yosuke Sato, and Daniel Saddiqi (eds), The Routledge handbook of syntax. London: Routledge, 604–26. Osborne, Timothy and Thomas Groß (2012). Constructions are catenae: Construction Grammar meets Dependency Grammar, Cognitive Linguistics 23(1): 165–216. Osborne, Timothy, Michael Putnam, and Thomas Groß (2011). Bare phrase structure, la­ bel-less trees, and specifier-less syntax: Is Minimalism becoming a dependency grammar? The Linguistic Review 28: 315–64. Osborne, Timothy, Michael Putnam, and Thomas Groß (2012). Catenae: Introduc­ ing a novel unit of syntactic analysis, Syntax 15(4): 354–96. (p. 1093)

Oshima, Shin (1979). Conditions on rules: Anaphora in Japanese, in George Bedell, Eichi Kobayashi, and Masatake Muraki (eds), Explorations in linguistics: Papers in honor of Kazuko Inoue. Tokyo: Kenkyusha, 423–48. Otani, Kazuyo and John Whitman (1991). V-raising and VP-ellipsis, Linguistic Inquiry 22(2): 345–58. Ott, Dennis (2014). An ellipsis approach to contrastive left-dislocation, Linguistic Inquiry 45: 269–303. Ott, Dennis and Volker Struckmeier (2015). Deletion of non-constituents in clausal ellip­ sis: Remnants in the middle field. Ms. Humboldt University of Berlin and University of Cologne. Ott, Dennis and Mark de Vries (2016). Right-dislocation as deletion, Natural Language and Linguistic Theory 34(2): 641–90. Paape, Dario (2015). A reactivation advantage for sluicing antecedents in German. Poster presented at the 28th Annual Conference on Human Sentence Processing. University of Southern California. Padden, Carol (1990). The relation between space and grammar in ASL verb morphology, in Christopher Lucas (ed.), Sign language research: Theoretical issues. Washington DC: Gallaudet Press, 118–32. Paesani, Kate (2006). Extending the nonsentential analysis: The case of special registers, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Ben­ jamins, 147–82. Panagiotidis, Phoevos (2002). Pronouns, clitics and empty nouns. Amsterdam and Philadelphia: John Benjamins. Panagiotidis, Phoevos (2003a). One, empty nouns and θ-assignment, Linguistic Inquiry 34(2): 281–92. Page 91 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Panagiotidis, Phoevos (2003b). Empty nouns, Natural Language and Linguistic Theory 12(2): 381–432. Pancheva, Roumyana (2006). Phrasal and clausal comparatives in Slavic, in James Lavine, Steven Franks, Mila Tasseva-Kurktchieva, and Hana Filip (eds), Formal Approaches to Slavic Linguistics 14. Ann Arbor: Michigan Slavic Publications, 236–57. Pancheva, Roumyana (2010). More students attended FASL than CONSOLE, in Wayles Browne, Adam Cooper, Alison Fisher, Esra Kesici, Nikola Predolac, and Draga Zec (eds), Formal Approaches to Slavic Linguistics 18: The Cornell Meeting, 2009. Ann Arbor: Michigan Slavic Publications, 383–400. Pancheva, Roumyana (2012). Syntax and semantics of comparatives and superlatives. Class handouts, EGG 2012 summer school, 30 July–3 August. Pantcheva, Marina (2009). First phase syntax of Persian complex predicates: Argument structure and telicity, Journal of South East Asian Linguistics 2: 53–72. Park, Myung-Kwan (1997). The syntax of VP ellipsis in Korean, Language Research 33(4): 629–48. Park, Myung-Kwan (2011). The syntax-semantics interface of pair-list interpretation in wh-constructions and their sluicing counterparts, Studies in Generative Grammar 21(1): 311–31. Park, Myung-Kwan and Jung-Min Kang (2007). Multiple sluicing in English, Studies in Generative Grammar 17(1): 419–34. Park, Myung-Kwan and Jaejun Kim (2016). English Heavy NP Shift as clause-internal frag­ menting, The Journal of Studies in Language 31(4): 863–85. Park-Johnson, Sung (2008). The acquisition of wh-questions by Korean-English bilingual children: The role of crosslinguistic influence. Master’s thesis, Purdue University. Park-Johnson, Sunny (2012). The effect of subject person in auxiliary movement acquisi­ tion by Korean-English bilingual children, in LSA Annual Meeting Extended Abstracts, vol. 3, 37:1–5. Parrott, Jeffrey K. (2006). Distributed morphological mechanisms of pronoun-case varia­ tion, University of Pennsylvania Working Papers in Linguistics 12(2): 173–87. Parrott, Jeffrey K. (2009). Danish vestigial case and the acquisition of vocabulary in Dis­ tributed Morphology, Biolinguistics 3(2–3): 270–304. Parsons, Terence (1990). Events in the semantics of English: A study in subatom­ ic semantics. Current Studies in Linguistics 19. Cambridge, MA: MIT Press. (p. 1094)

Page 92 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Partee, Barbara Hall and Mats Rooth (1983). Generalized conjunction and type ambiguity, in R. Bäuerle, C. Schwarze, and Arnim von Stechow (eds), Meaning, use and interpreta­ tion of language. Berlin: de Gruyter, 362–83. Patin, Cédric (2007). La tonologie du shingazidja, langue bantu (G44a) de la Grande Co­ more: Nature, formalisation, interfaces. PhD thesis, Université Paris 3. Patin, Cédric (2010). The prosody of Shingazidja relatives, ZAS Working Papers in Lin­ guistics 53: 187–210. Paul, Hermann (1889). Principles of the history of language. New York: MacMillan & Co. Paul, Ileana and Eric Potsdam (2012). Sluicing without wh-movement in Malagasy, in Ja­ son Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Amster­ dam and Philadelphia: John Benjamins, 164–82. Paul, Waltraud (1999). Verb gapping in Chinese: A case of verb raising, Lingua 107: 207– 26. Payne, John, Geoffrey K. Pullum, Barbara C. Scholz, and Eva Berlage (2013). Anaphoric one and its implications, Language 89(4): 794–829. Penn, Gerald (2000). The algebraic structure of attributed type signatures. PhD thesis, Carnegie Mellon University. Pereltsvaig, Asya (2006). Small nominals, Natural Language and Linguistic Theory 24(2): 433–500. Pereltsvaig, Asya (2007). On the universality of DP: A view from Russian, Studia Linguisti­ ca 61(1): 59–94. Perez-Leroux, Ana, Mihaela Pirvulescu, and Yves Roberge (2008). Children’s interpreta­ tion of null objects under the scope of negation, in Susie Jones (ed.), Proceedings of the 2008 annual conference of the Canadian Linguistic Association. Vancouver: University of British Columbia, 1–10. Perez-Leroux, Ana, Mihaela Pirvulescu, and Yves Roberge (2018). Direct objects and lan­ guage acquisition. Cambridge: Cambridge University Press. Perlmutter, David M. and John R. Ross (1970). Relative clauses with split antecedents, Linguistic Inquiry 1(3): 350. Pesetsky, David (1982). Paths and categories. PhD thesis, Massachusetts Institute of Tech­ nology. Pesetsky, David (1987). Wh-in-situ: Movement and unselective binding, in Eric J. Reuland and Alice G. B. ter Meulen (eds), The representation of (in)definiteness. Cambridge, MA: MIT Press, 98–129. Page 93 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Pesetsky, David (1995). Zero syntax. Cambridge, MA: MIT Press. Phillips, Colin and Dan Parker (2014). The psycholinguistics of ellipsis, Lingua 151: 78–95. van Peteghem, Marleen (2009). Sur le subordonnant comparatif dans les langues ro­ manes, Langages 174: 99–112. Pfaff, Carol W. (1979). Constraints on language mixing: Intrasentential code-switching and borrowing in Spanish/English, Language 55: 291–318. Philip, William (2011). Acquiring knowledge of universal quantification, in Jill de Villiers and Tom Roeper (eds), Handbook of generative approaches to language acquisition. Ams­ terdam: Springer, 351–94. Philipp, Mareike (2015). German: Anaphor binding in sluiced sentences. Ms. University of Potsdam. Philippova, Tatiana (2014). P-omission under sluicing, [P clitic] and the nature of P-strand­ ing, in Martin Kohlberger, Kate Bellamy, and Eleanor Dutton (eds), Proceedings of ConSOLE XXII. Leiden: SOLE, 133–55. Phillips, Colin (1996). Order and structure. PhD thesis, Massachusetts Institute of Tech­ nology, Cambridge, MA. Phillips, Colin (2003). Linear order and constituency, Linguistic Inquiry 34(1): 37–90. Phillips, Colin (2006). The real-time status of island phenomena, Language 82: 795–823. Phillips, Colin and Shevaun Lewis (2013). Derivational order in syntax: Evidence and ar­ chitectural consequences, Studies in Linguistics 6: 11–47. Piñón, Christopher (1995). Around the progressive in Hungarian, in István Kenesei (ed.), Approaches to Hungarian 5. Szeged: JATE Press, 153–89. Piantadosi, Steven T., Harry Tily, and Edward Gibson (2012). The communicative function of ambiguity in language, Cognition 122: 280–91. Pickering, Martin J. and Guy D. Barry (1991). Sentence processing without empty categories, Language and Cognitive Processes 8: 229–59. (p. 1095)

Pickering, Martin J. and Simon Garrod (2013). An integrated theory of language produc­ tion and comprehension, Behavioral and Brain Sciences 36(4): 329–47. Pierrehumbert, Janet (1980). The phonology and phonetics of English intonation. PhD the­ sis, MIT. Pierrehumbert, Janet and Julia Hirschberg (1990). The meaning of intonational contours in the interpretation of discourse, in Cohenm Philip R., Jerry L. Morgan, and Martha E. Pollack (eds), Intentions in communication. Cambridge, MA: MIT Press, 271–312. Page 94 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Pieters, Lody (2011). Lees maar! Er staat (niet!) wat er staat. Over de beschrijving van sa­ mentrekking en andere onvolledige taaluitingen in Nederlandse grammatica’s na 1900. PhD thesis, Radboud University Nijmegen. Pinkham, Jessie E. (1982). The formation of comparative clauses in English and French. Bloomington, IN: Indiana University Linguistics Club. Plewnia, Albrecht (2003). Sätze, denen nichts fehlt: Eine dependenzgrammatische Unter­ suchung elliptischer Konstruktionen. Germanistische Linguistik Monographien 11. Hildesheim et al.: Olms. Plewnia, Albrecht (2013). Moglichkeiten und Wahrscheinlichkeiten: Was man in der De­ pendenzgrammatik mit syntaktischen Leerstellen tun kann. In Mathilde Hennig (eds), Die Ellipsis: Neue Perspectiven auf ein altes Phänomen. Berlin: Walter de Gruyter, 227–51. Poesio, Massimo and Hannes Rieser (2010). Completions, coordination, and alignment in dialogue, Dialogue and Discourse 1: 1–89. Poirier, Josée, Lewis Shapiro, Tracy Love, and Yosef Grodzinsky (2009). The on-line pro­ cessing of verb phrase ellipsis in aphasia, Journal of Psycholinguistic Research 38: 237– 53. Poirier, Josée, Katie Wolfinger, Lisa Spellman, and Lewis P. Shapiro (2010). The real-time processing of sluiced sentences, Journal of Psycholinguistic Research 39: 411–27. Pollard, Carl Jesse and Ivan A. Sag (1987). Information-based syntax and semantics. San­ ford, CA: Center for the Study of Language and Information. Pollard, Carl Jesse and Ivan A. Sag (1994). Head-driven phrase structure grammar. Stan­ ford, CA: Center for the Study of Language and Information. Pollmann, Thijs (1975). Een regel die subject en copula deleert? Spektator 5: 282–92. Pollock, Jean-Yves (1989). Verb movement, UG and the structure of IP, Linguistic Inquiry 20(3): 365–424. Poplack, Shana (1980). Sometimes I’ll start a sentence in Spanish and termino en Es­ pañol, in Jon Amastae and Lucia Elfas-Olivares (eds), Spanish in the United States: Soci­ olinguistic aspects. Cambridge: Cambridge University Press, 230–63. Poplack, Shana (2001). Code-switching (linguistic), in Neil J. Smelser and Paul B. Baltes (eds), International encyclopedia of the social and behavioral sciences. Amsterdam: Else­ vier, 2062–5. Poplack, Shana, Lauren Zentz, and Nathalie Dion (2012). Phrase final prepositions in Quebec French: An empirical study of contact, codeswitching, and resistance to conver­ gence, Bilingualism: Language and Cognition 15(2): 203–25.

Page 95 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Postal, Paul M. (1969). Anaphoric islands, in Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds), Papers from the 5th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 205–39. Postal, Paul M. (1974). On raising. Cambridge, MA: MIT Press. Postal, Paul M. (1994). Parasitic and pseudoparasitic gaps, Linguistic Inquiry 25(1): 63– 118. Postal, Paul M. (1998). Three investigations of extraction. Cambridge, MA: MIT Press. Potsdam, Eric (1997a). English verbal morphology and VP ellipsis, in Kiyomi Kusumoto (ed.), NELS 27: Proceedings of the 27th annual meeting of the North East Linguistic Soci­ ety. Amherst, MA: Graduate Student Linguistic Association, 353–68. Potsdam, Eric (1997b). NegP and subjunctive complements in English, Linguistic Inquiry 28(3): 533–41. (p. 1096)

Potsdam, Eric (2007). Malagasy sluicing and its consequences for the identity require­ ment on ellipsis, Natural Language and Linguistic Theory 25(3): 577–613. Potsdam, Eric (2012). A direct analysis of Malagasy phrasal comparatives, in Lauren Eby Clemens, Gregory Scontras, and Maria Polinsky (eds), Proceedings of the 18th annual meeting of the Austronesian Formal Linguistics Society (AFLA 18). Cambridge, MA: Aus­ tronesian Formal Linguistics Association, 140–55. Potter, David (2014). A heterogeneous approach to gapping, in Robert E. SantanaLaBarge (ed.), Proceedings of the 31st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 344–50. Potts, Christopher (2005). The logic of conventional implicatures. Oxford: Oxford Univer­ sity Press. Potts, Christopher, Ash Asudeh, Seth Cable, Yurie Hara, Eric McCready, Luis AlonsoOvalle, Rajesh Bhatt, Christopher Davis, Angelika Kratzer, Tom Roeper, and Martin Walkow (2009). Expressives and identity conditions, Linguistic Inquiry 40: 356–66. Prinzhorn, Martin and Viola Schmitt (2010). Discontinuous DP-coordination in German, Linguistic Variation Yearbook 10: 161–200. Pritchett, Bradley (1991). Subjacency in a principle-based parser, in Robert C. Berwick, Steven P. Abney, and Carol Tenny (eds), Principle-based parsing: Computation and psy­ cholinguistics. Dordrecht: Kluwer Academic Publishers, 301–45. Progovac, Ljiljana (1998). Detereminer phrase in a language without determiners, Journal of Linguistics 34: 165–79.

Page 96 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Progovac, Ljiljana (2003). Structure for coordination, in Lisa L.-S. Cheng and Rint Sybesma (eds), The second Glot International state-of-the-article book. Berlin: Mouton de Gruyter, 241–88. Progovac, Ljiljana (2006). The syntax of nonsententials: Small clauses and phrases at the root, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 33–72. Progovac, Ljiljana (2013). Non-sentential vs. ellipsis approaches: Review and extensions, Language and Linguistics Compass 7: 597–617. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (2006a). Epilogue: Wherefrom and whereto?, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 323–53. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (2006b). Introduc­ tion, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 1–10. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds) (2006c). The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins. Prüst, Hub (1992). On discourse structuring, VP anaphora, and gapping. PhD thesis, Uni­ versity of Amsterdam. Prüst, Hub (1993). Gapping and VP anaphora. PhD thesis, University of Amsterdam. Prüst, Hub, Remko Scha, and Martin van den Berg (1994). A discourse perspective on verb phrase anaphora, Linguistics and Philosophy 17(3): 261–327. Pullum, Geoffrey K. (1975). People deletion in English, Working Papers in Linguistics 14: 95–101. Pullum, Geoffrey K. (1997). The morpholexical nature of English to-contraction, Language 73(1): 79–102. Pullum, Geoffrey K. (2000). Hankamer was!, in Sandra Chung, James McCloskey, and Nathan Sanders (eds), Jorge Hankamer Webfest. Santa Cruz: UCSC. Pullum, Geoffrey K. and Arnold Zwicky (1986). Phonological resolution of syntactic fea­ ture conflict, Language 62: 751–73. Pulman, Stephen G. (1997). Higher order unification and the interpretation of fo­ cus, Linguistics and Philosophy 20(1): 73–115. (p. 1097)

Page 97 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Purver, Matthew (2006). Clarie: Handling clarification requests in a dialogue system, Re­ search on Language and Computation 4(2): 259–88. Purver, Matthew, Ronnie Cann, and Ruth Kempson (2006). Grammars as parsers: Meeting the dialogue challenge, Research on Language and Computation 4(2–3): 289–326. Purver, Matthew, Arash Eshghi, and Julian Hough (2011). Incremental semantic construc­ tion in a dialogue system, in Johan Bos and Stephen Pulman (eds), Proceedings of the 9th International Conference on Computational Semantics. Oxford: Association for Computa­ tional Linguistics, 365–9. Purver, Matthew, Jonathan Ginzburg, and Patrick Healey (2001). On the means for clarifi­ cation in dialogue, in Jan van Kuppevelt and Ronnie Smith (eds), Current and new direc­ tions in discourse and dialogue. Dordrecht: Kluwer Academic Publishers, 235–56. Purver, Matthew, Eleni Gregoromichelaki, Wilfried Meyer-Viol, and Ronnie Cann (2010). Splitting the ‘I’s and crossing the ‘You’s: Context, speech acts and grammar, in Pavel Lup­ kowski and Matthew Purver (eds), Aspects of Semantics and Pragmatics of Dialogue: SemDial 2010, 14th Workshop on the Semantics and Pragmatics of Dialogue. Poznan: Pol­ ish Society for Cognitive Science, 43–50. Purver, Matthew and Ruth Kempson (2004). Incremental context-based generation for di­ alogue, in Anja Belz, Roger Evans, and Paul Piwek (eds), Proceedings of the 3rd Interna­ tional Conference on Natural Language Generation (INLG04). Brockenhurst: Springer, 151–60. Quer, Josep and Joanna Rossello (2013). On sloppy readings, ellipsis and pronouns: Miss­ ing arguments in Catalan Sign Language (LSC) and other argument-drop languages, in Victoria Camacho-Taboada, Ángel L. Jiménez-Fernández, Javier Martin-Gonzalez, and Mariano Reyes-Tejedor (eds), Information structure and agreement. Amsterdam and Philadelphia: John Benjamins, 337–70. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik (1985). A compre­ hensive grammar of the English language. London: Longman. Radford, Andrew (1997). Syntactic theory and the structure of English. Cambridge: Cam­ bridge University Press. Radford, Andrew and Eiichi Iwasaki (2015). On swiping in English, Natural Language and Linguistic Theory 33(2): 703–44. Ramos, Sergio. (2014). A look at fragment answers under code-switching. Talk given at the UIC Talks in Linguistics. University of Illinois at Chicago. Rankin, Tom (2014). Variational learning in L2: The transfer of L1 syntax and parsing strategies in the interpretation of wh-questions by L1 German learners of L2 English, Lin­ guistic Approaches to Bilingualism 4(4): 423–61. Page 98 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Rankin, Tom (2016). Quotative inversion and particles. Ms. University of Vienna. Ranta, Aarne (1994). Type theoretical grammar. Oxford: Oxford University Press. Raposo, Eduardo (1986). On the null object in European Portuguese, in Osvaldo Jaeggli and Carmen Silva- Corvalán (eds), Studies in Romance linguistics. Dordrecht: Foris Publi­ cations, 373–90. Rapp, Irene and Arnim von Stechow (1999). Fast ‘almost’ and the visibility parameter for functional adverbs, Journal of Semantics 16(2): 149–204. Rasekhi, Vahideh (2014). Missing object in Persian. Ms. Stony Brook University. Reeve, Matthew (2014). If there’s anything cleft ellipsis resembles, it’s (pseudo)gapping, in Robert E. Santana-LaBarge (ed.), Proceedings of the 31st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 351–60. Reich, Ingo (2007). Toward a uniform analysis of short answers and gapping, in Kerstin Schwabe and Susanne Winkler (eds), On information structure, meaning and form. Ams­ terdam and Philadelphia: John Benjamins, 467–84. Reich, Ingo (2011). Ellipsis, in Claudia Maienborn, Klaus von Heusinger, and Paul Portner (eds), Semantics: An international handbook of natural language meaning. New York: Mouton de Gruyter, 1849–74. Reich, Ingo and Marga Reis (2013). Koordination und Subordination, in Jörg Meibauer, Markus Steinbach, and Hans Altmann (eds), Satztypen des Deutschen. Berlin (p. 1098)

and New York: Mouton de Gruyter, 536–69. Reinhart, Tanya (1983a). Anaphora and semantic interpretation. Chicago: University of Chicago Press. Reinhart, Tanya (1983b). Coreference and bound anaphora: A restatement of the anapho­ ra questions, Linguistics and Philosophy 6: 47–88. Reinhart, Tanya (1991). Elliptic conjunctions—non-quantificational LF, in Asa Kasher (ed.), The Chomskyan Turn. Cambridge, MA: Blackwell Publishers, 360–84. Remmele, Bettina, Sophia Schopper, Robin Hörnig, and Susanne Winkler (forthcoming). (Non)strategic production planning and ambiguity: Experimental evidence, in Matthias Bauer and Angelika Zirker (eds), Strategies of ambiguity. London: Routledge. Renato, Lacerda (2016). Contrastive topicalisation in child English. Ms. University of Con­ necticut. Repp, Sophie (2006a). Interpreting ellipsis: The changeable presence of the negation in gapping. PhD thesis, Humboldt-Universität zu Berlin.

Page 99 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Repp, Sophie (2006b). ¬(a&b) Gapping, negation and speech act operators, Research on Language and Computation 4: 397–423. Repp, Sophie (2009). Negation in gapping. Oxford Studies in Theoretical Linguistics. Ox­ ford: Oxford University Press. Repp, Sophie (2016). Contrast, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, 270–89. Richards, Norvin (1997). What moves where when in which language? Doctoral disserta­ tion, Massachusetts Institute of Technology, Cambridge, MA. Richards, Norvin (2001). Movement in language: Interactions and architectures. Oxford: Oxford University Press. van Riemsdijk, Henk (1978). A case study in syntactic markedness. Dordrecht: Foris Pub­ lications. Rieser, Hannes and David Schlangen (2011). Special issue on incremental processing in dialogue, Dialogue and Discourse 2. Rijkhoek, Paulien (1998). On degree phrases and result clauses. PhD thesis, University of Groningen (GroDiL 27). Rissanen, Terhi (1985). Viittomakielen perusrakenne [The basic structure of sign lan­ guage]. Helsinki: University of Helsinki. Rissanen, Terhi (1998). The categories of nominals and verbals and their morpholoy in Finnish Sign Language. Master’s thesis, University of Turku. Ritter, Elizabeth (1991). Two functional categories in noun phrases: Evidence from Mod­ ern Hebrew, in Susan Rothstein (ed.), Syntax and semantics 25: Perspectives on phrase structure: Heads and licensing. San Diego and London: Academic Press, 37–62. Rivero, María-Luisa (1990). The location of nonactive voice in Albanian and Modern Greek, Linguistic Inquiry 21: 135–46. Rizzi, Luigi (1990). Relativized minimality. Cambridge, MA: MIT Press. Rizzi, Luigi (1997). The fine structure of the left periphery, in Liliane Haegeman (ed.), Ele­ ments of grammar. Dordrecht: Kluwer Academic Publishers, 281–337. Rizzi, Luigi (2004). Locality and left periphery, in Adriana Belletti (ed.), Structures and be­ yond: The cartography of syntactic structures, vol. 3. Oxford: Oxford University Press, 223–51. Roberts, Craige (1998a). Information structure in discourse: Towards an integrated for­ mal theory of pragmatics, in James H. Yoon and Andreas Kathol (eds), OSU working pa­ pers in linguistics 49: Papers in semantics (1996). Ohio: Ohio State University, 91–136. Page 100 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Roberts, Craige (2012). Information structure in discourse: Towards an integrated formal theory of pragmatics, Semantics and Pragmatics 5(6): 1–69, revision of 1998 version of paper originally circulated in 1996. Roberts, Ian (1998b). Have/be raising, move F, and procrastinate, Linguistic Inquiry 29(1): 113–25. Roberts, Leah, Ayumi Matsuo, and Nigel Duffield (2013). Processing VP-ellipsis and VP-anaphora with structurally parallel and non-parallel antecedents: An eye-tracking study, Language and Cognitive Processes 28(1–2): 29–47. (p. 1099)

Rochemont, Michael S. (2012). Question/answer congruence and focus phrase. Ms. Uni­ versity of Tuebingen and University of British Columbia. Rochemont, Michael S. (2016). Givenness, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, 41–63. Rodgers, Richard and Lorenz Hart (1937). Where or when. Van Nuys, CA: Alfred Music Publishing. Rodrigues, Cilene, Andrew Nevins, and Luis Vicente (2009). Cleaving the interactions be­ tween sluicing and preposition stranding, in Danièle Torck and Leo Wetzels (eds), Ro­ mance languages and linguistic theory 2006. Amsterdam: John Benjamins, 175–98. Roelofsen, Floris (2011). Free variable economy, Linguistic Inquiry 42(4): 682–97. Roelofsen, Floris (2013). Algebraic foundations for the semantic treatment of inquisitive content, Synthese 190(1): 79–102. Roeper, Tom (1982). The role of universals in the acquisition of gerunds, in Eric Wanner and Lila R. Gleitman (eds), Language acquisition: The state of the art. Cambridge: Cam­ bridge University Press, 267–87. Roeper, Tom (1999). Universal bilingualism, Bilingualism: Language and Cognition 2(3): 169–86. Roeper, Tom (2003). Multiple grammars, feature attraction, pied-piping, and the question: Is AGR inside TP?, in Natascha Müller (ed.), Vulnerable domains in multilingualism. Ams­ terdam and Philadelphia: John Benjamins, 335–60. Roeper, Tom (2014). Strict interface principles and the acquisition engine: From unla­ beled to labeled and minimal modular contact, Language Sciences 46: 115–32. Roeper, Tom (2016). Multiple grammars and the logic of learnability in second language acquisition, Frontiers in Psychology 7(14). . Roeper, Tom, Barbara Zurer Pearson, and Margaret Grace (2011). Quantifier spreading is not distributive, in Nick Danis, Kate Mesh, and Hyunsuk Sung (eds), Proceedings of the Page 101 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References 35th annual Boston University Conference on Language Development (BUCLD 35). Somerville, MA: Cascadilla Press, 526–39. Roeper, Tom and Harry N. Seymour (1994). The place of linguistic theory in the theory of language acquisition and language impairment, in Yonata Levi (ed.), Other children, other languages: Issues in the theory of language acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates, 305–30. Roeper, Tom and Muffy E. A. Siegel (1978). A lexical transformation for verbal com­ pounds, Linguistic Inquiry 9(2): 199–260. Roeper, Tom, Uri Strauss, and Barbara Zurer Pearson (2006). The acquisition path of the determiner quantifier ‘every’: Two kinds of spreading, in Tanya Heizmann (ed.), Current issues in first language acquisition. Charleston: Booksurge Publishing, 97–129. Romaine, Suzanne (1995). Bilingualism. Malden, MA: Blackwell. Rombi, Marie-Françoise and Pierre Alexandre (1982). Les parlers comoriens, caractéris­ tiques différentielles, position par rapport au swahili, in Marie-Françoise Rombi (ed.), Études sur le bantu oriental. Paris: SELAF, 17–39. Romero, Maribel (1998). Focus and reconstruction effects in wh-phrases. PhD thesis, Uni­ versity of Massachusetts, Amherst. Romero, Maribel (2000). Antecedentless sluiced wh-phrases and islands, in Kerstin Sch­ wabe and Ning Zhang (eds), Ellipsis in conjunction. Tübingen: Niemeyer, 195–220. Romero, Maribel (2003). Correlate restriction and definiteness effect in ellipsis, in Ker­ stin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omit­ ted structures. Amsterdam and Philadelphia: John Benjamins, 263–300. Ronat, Mitsou (1977). Une contrainte sur l’effacement du nom, in Mitsou Ronat (ed.), Langue: Théorie generative étendue. Paris: Collection Savoir, 153–69. Rooryck, Johan (2001). Evidentiality, part I, GLOT International 5: 125–33. Rooth, Mats (1981). A comparison of three theories of verb phrase ellipsis, in Wynn Chao and Dierdre Wheeler (eds), University of Massachusetts Occasional Papers in Linguistics. Amherst, MA: GLSA, 212–44. (p. 1100)

Rooth, Mats (1992a). Ellipsis redundancy and reduction redundancy, in Steve Berman and Arild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart. Heidelberg: IBM. Rooth, Mats (1992b). A theory of focus interpretation, Natural Language Semantics 1(1): 117–21. Rooth, Mats (1996). Focus, in Shalom Lappin (ed.), The handbook of contemporary se­ mantic theory. Oxford: Blackwell Publishers, 271–97. Page 102 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Rooth, Mats (2016). Alternative semantics, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, 19–40. Rosen, Carol (1976). Guess what about?, in Alan Ford, John Reighard and Rajendra Singh (eds), NELS 6: Proceedings of the sixth annual meeting of the North East Linguistic Soci­ ety. Amherst, MA: GLSA, 205–11. Ross, John Robert (1967). Constraints on variables in syntax. PhD thesis, Massachusetts Institute of Technology. Ross, John Robert (1969a). Auxiliaries as main verbs, in W. Todd (ed.), Studies in philo­ sophical linguistics, Series One. Evanston, IL: Great Expectations Press, 77–102. Ross, John Robert (1969b). Guess who?, in Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds), Proceedings of the 5th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 252–86. Ross, John Robert (1970). Gapping and the order of constituents, in Manfred Bierwisch and Karl E. Heidolph (eds), Progress in linguistics. The Hague: Mouton, 249–59. Ross, John Robert (1986). Infinite syntax! Norwood, NJ: Ablex. Rottman, Isaac and Masaya Yoshida (2013). Sluicing, idioms, and island repair, Linguistic Inquiry 44: 651–68. Roussarie, Laurent and Pascal Amsili (2007). Interpréter les pronoms phrastiques, Mod­ èles Linguistiques 56: 83–112. Rouveret, Alain (2012). VP ellipsis, phases and the syntax of morphology, Natural Lan­ guage and Linguistic Theory 30: 897–963. Rowlett, Paul (2007). The syntax of French. Cambridge: Cambridge University Press. Rowling, Joanne K. (1999). Harry Potter and the prisoner of Azkaban. London: Blooms­ bury Publishing. Rowling, Joanne K. (2000). Harry Potter and the goblet of fire. London: Bloomsbury Pub­ lishing. Rudin, Catherine (1988). On multiple questions and multiple wh-fronting, Natural Lan­ guage and Linguistic Theory 6: 445–501. Rullmann, Hotze (1995). Maximality in the semantics of wh-constructions. PhD thesis, University of Massachusetts at Amherst. Runic, Jelena (2014). A new look at clitics, clitic doubling, and argument ellipsis: Evi­ dence from Slavic. PhD thesis, University of Connecticut. Ruwet, N. (1972). Théorie syntaxique et syntaxe du français. Paris: Editions du Seuil. Page 103 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Ruys, E. G. (1992). The scope of indefinites. PhD thesis, Utrecht Univeristy. Saab, Andrés (2004). El dominio de la elipsis nominal en español. Master’s thesis, Univer­ sity of Comahue. Saab, Andrés (2008). Hacia una teoría de la identidad parcial en la elipsis. PhD thesis, University of Buenos Aires. Saab, Andrés (2010a). (Im)possible deletion in the Spanish DP, IBERIA: An International Joumal of Theoretical Linguistics 2(2): 45–83. Saab, Andrés (2010b). Silent interactions. Spanish TP-ellipsis and the theory of island re­ pair, Probus 22: 73–116. Saab, Andrés (2014). Ineffable narratives in Spanish: Another case of overgeneration by e-GIVENness, Probus 28(2): 367–89. Saab, Andrés (2015). A note on someone (else): An island repair solution and its competi­ tors, Linguistic Inquiry 46(3): 553–68. Saab, Andrés and Anikó Lipták (2016). Movement and deletion after syntax: El­ lipsis by inflection reconsidered, Studia Linguistica 70(1): 66–108. (p. 1101)

Saab, Andrés and Pablo Zdrojewski (2012). Anti-repair effects under ellipsis, in Irene Franco, Sara Lusini, and Andrés Saab (eds), Romance languages and linguistic theory 2010: Selected papers from ‘Going Romance’ Leiden 2010. Amsterdam and Philadelphia: John Benjamins, 177–202. Sabbagh, Joseph (2007). Ordering and linearizing rightward movement, Natural Lan­ guage and Linguistic Theory 25(2): 349–401. Sabbagh, Joseph (2008). Right node raising and extraction in Tagalog, Linguistic Inquiry 39(3): 502–11. Sabbagh, Joseph (2012). Right Node Raising. Ms. University of Texas. Sacks, Harvey (1992). Lectures on conversation. Malden, MA: Blackwell. Saeboe, Kjell Johan (1993). Anaphoric presuppositions and zero anaphora. Ms. University of Oslo. Saeboe, Kjell Johan (1996). Anaphoric presuppositions and zero anaphora, Linguistics and Philosophy 19(1): 187–209. Sáez, Luis (2011). Peninsular Spanish prenominal possessives in ellipsis contexts: A phase-based account, in Julia Herschensohn (ed.), Romance linguistics 2010: Selected pa­ pers from the 40th Linguistic Symposium on Romance Languages (LSRL). Amsterdam and Philadelphia: John Benjamins, 155–76.

Page 104 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Safir, Ken (1999). Vehicle change and reconstruction in a’-chains, Linguistic Inquiry 30(4): 587–620. Sag, Ivan A. (1976a). Deletion and logical form. PhD thesis, MIT. Sag, Ivan A. (1976b). A logical theory of verb phrase deletion, in Salikoko S. Mufwene, Carol A. Walker, and Sanford B. Steever (eds), Papers from the 12th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 533–42. Sag, Ivan A. (1977). Deletion and logical form. Bloomington, IN: Indiana University Lin­ guistic Club Publications. Sag, Ivan A. (1980). Deletion and logical form. New York: Garland Publishing. Sag, Ivan A. (2012). Sign-based Construction Grammar: An informal synopsis, in Hans C. Boas and Ivan A. Sag (eds), Sign-based Construction Grammar. Stanford: CSLI Publica­ tions, 69–102. Sag, Ivan A., Hans C. Boas, and Paul Kay (2012). Introducing sign-based Construction Grammar, in Hans C. Boas and Ivan A. Sag (eds), Sign-based Construction Grammar. Stanford: CSLI Publications, 1–29. Sag, Ivan A. and Janet Dean Fodor (1994). Extraction without traces, in Paul Aranovich, William Byrne, Susanne Preuss, and Martha Senturia (eds), Proceedings of the 13th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 365–84. Sag, Ivan A., Gerald Gazdar, Thomas Wasow, and Steven Weisler (1985). Coordination and how to distinguish categories, Natural Language and Linguistic Theory 3(2): 117–72. Sag, Ivan A. and Jorge Hankamer (1984). Toward a theory of anaphoric processing, Lin­ guistics and Philosophy 7: 325–45. Sag, Ivan A., Philip Hofmeister, and Neal Snider (2007). Processing complexity in subja­ cency violations: The Complex Noun Phrase Constraint. Ms. Stanford. Sag, Ivan A. and Joanna Nykiel (2011). Remarks on sluicing, in Stefan Müller (ed.), Pro­ ceedings of the 18th International Conference on Head-Driven Phrase Structure Gram­ mar. Stanford: CSLI, 188–208. Sag, Ivan A., Thomas Wasow, and Emily Bender (2003). Syntactic theory: A formal intro­ duction. Stanford: CSLI Publications. Sailor, Craig (2009). Tagged for deletion: A typological approach to VP ellipsis in tag questions. Master’s thesis, UCLA. Sailor, Craig (2012a). Inflection at the interfaces. Ms. UCLA. Sailor, Craig (2012b). Tag questions and the typology of VP ellipsis. Ms. UCLA.

Page 105 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Sailor, Craig (2014). The variables of VP-ellipsis. PhD thesis, UCLA. Sailor, Craig and Grace Kuo (2010). Taiwanese VP ellipsis and the progressive prohibi­ tion. Paper presented at the 22nd North American Conference on Chinese Linguistics. Sailor, Craig and Gary Thoms (2014). On the non-existence of non-constituent co­ ordination and nonconstituent ellipsis, in Robert E. Sandana-LaBarge (ed.), Proceedings of the 31st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Pro­ ceedings Project, 361–70. (p. 1102)

Saito, Mamoru (1985). Some asymmetries in Japanese and their theoretical implications. PhD thesis, Massachusetts Institute of Technology. Saito, Mamoru (1987). Three notes on syntactic movement in Japanese, in Issues in Japan­ ese linguistics. Dordrecht: Foris Publications, 301–50. Saito, Mamoru (1989). Scrambling as semantically vacuous A′-movement, in Mark R. Baltin and Anthony S. Kroch (eds), Alternative conceptions of phrase structure. Chicago: University of Chicago, 182–200. Saito, Mamoru (1992). Long-distance scrambling in Japanese, Journal of East Asian Lin­ guistics 1(1): 69–118. Saito, Mamoru (2004). Ellipsis and pronominal reference in Japanese clefts, Nanzan Lin­ guistics 1: 21–50. Saito, Mamoru (2007). Notes on East Asian argument ellipsis, Language Research 43: 203–27. Saito, Mamoru (2014). Japanese syntax in a comparative perspective. New York: Oxford University Press. Saito, Mamoru and Keiko Murasugi (1990a). N′-deletion in Japanese, in Javier Ormazabal and Carol Tenny (eds), University of Connecticut working papers in linguistics 3. Con­ necticut: University of Connecticut, 87–107. Saito, Mamoru and Keiko Murasugi (1990b). N′-deletion in Japanese: A preliminary study, in Hajime Hoji (ed.), Japanese and Korean linguistics 1. Stanford: CSLI Publications, 285– 301. Saito, Mamoru, Jonah T.-H. Lin, and Keiko Murasugi (2008). N′-ellipsis and the structure of noun phrases in Chinese and Japanese, Journal of East Asian Linguistics 17(3): 247–71. Sakamoto, Yuta (2014). Absence of case-matching effects in Mongolian sluicing. Ms. Uni­ versity of Connecticut. Sakamoto, Yuta (forthcoming). Japanese null arguments as mixed anaphora: Evidence for the LF-copy analysis, in Proceedings of the 11th Workshop on Altaic Formal Linguistics (WAFL11). MIT Working Papers in Linguistics. Cambridge, MA: MITWPL. Page 106 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Samiian, Vida (1983). Origins of phrasal categories in Persian: An X-bar analysis. PhD the­ sis, UCLA. San Pietro, Steve, Ming Xiang, and Jason Merchant (2012). Accounting for voice mis­ match in ellipsis, in Nathan Arnett and Ryan Bennett (eds), Proceedings of the 30th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 303–12. Sandler, Wendy and Diane Lillo-Martin (2006). Sign language and linguistic universals. Cambridge: Cambridge University Press. Santos, Ana Lúcia (2006). Minimal answers: Ellipsis syntax and discourse in the acquisi­ tion of European Portuguese. PhD thesis, University of Lisbon. Santos, Ana Lúcia (2009a). Early VP ellipsis: Production and comprehension evidence, in Acrísio Pires and Jason Rothman (eds), Minimalist inquiries into child and adult language acquisition: Case studies across Portuguese. Berlin and New York: Mouton de Gruyter, 155–76. Santos, Ana Lúcia (2009b). Minimal answers: Ellipsis syntax and discourse in the acquisi­ tion of European Portuguese. Amsterdam and Philadelphia: John Benjamins. Sato, Yosuke (2007). P-stranding generalization and Bahasa Indonesian: A myth?, Snip­ pets 16: 17–18. Sato, Yosuke (2008). Sluicing in Bahasa Indonesia, P-stranding, and interface repair, in Sarah Berson, Alex Bratkievich, Daniel Bruhn, Amy Campbell, Ramon Escamilla, Allegra Giovine, Lindsey Newbold, Marilola Perez, Marta Piqueras-Brunet, and Russell Rhomieux (eds), Proceedings of the annual meeting of the Berkeley Linguistics Society, vol. 34. Berkeley: Berkeley Linguistics Society, 269–80. Sato, Yosuke (2010). Minimalist interfaces: Evidence from Indonesian and Javanese. Ams­ terdam and Philadelphia: John Benjamins. Sato, Yo (2011a). Local ambiguity, search strategies and parsing in Dynamic Syn­ tax, in Ruth Kempson, Eleni Gregoromichelaki, and Christine Howes (eds), The dynamics of lexical interfaces. Stanford: CSLI Publications, 201–28. (p. 1103)

Sato, Yosuke (2011b). P-stranding under sluicing and repair by ellipsis: Why is Indonesian (not) special? Journal of East Asian Linguistics 20: 339–82. Sato, Yosuke (2013). Fragments, ellipsis and PF repair: New evidence from Indonesian, Canadian Journal of Linguistics 58: 497–505. Sato, Yosuke (2014). Cyclic Spell-Out and modal complement ellipsis in Javanese, Linguis­ tic Analysis 38(3–4): 183–206. Sato, Yosuke (2015). Argument ellipsis in Javanese and voice agreement, Studia Linguisti­ ca 69: 58–85. Page 107 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Sauerland, Uli (1995). Sluicing and islands. Ms. MIT. Sauerland, Uli (1996). Guess how? Proceedings of the 5th conference of the Student Or­ ganization of Linguistics in Europe (ConSoLE 5). Leiden: SOLE, 297–311. Sauerland, Uli (1998). The meaning of chains. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Sauerland, Uli (2008). The silent content of bound variable pronouns, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, 183–209. Savolainen, Leena (2006). Interrogatives and negatives in Finnish Sign Language, in Ul­ rike Zeshan (ed.), Interrogative and negative constructions in sign languages. Nijmegen: Ishara Press, 284–302. Schachter, Paul (1977a). Constraints on coordination, Language 53: 86–103. Schachter, Paul (1977b). Does she or doesn’t she? Linguistic Inquiry 8(4): 763–6. Schachter, Paul (1978). English propredicates, Linguistic Analysis 4: 187–224. Schank, Roger and Robert Abelson (1977). Scripts, plans, goals, and understanding: An inquiry into human knowledge structures. Hillsdale, NJ: Lawrence Erlbaum Associates. Schegloff, E. A. (2007). Sequence organization in interaction: A primer in conversation analysis I. Cambridge: Cambridge University Press. Schiebe, Traugott (1973). Zum Problem der grammatisch relevanten Identität, in Ferenc Kiefer and Nicholas Ruwet (eds), Generative grammar in Europe. Dordrecht: D. Reidel, 482–527. Schlangen, David (2003). A coherence-based approach to the interpretation of non-sen­ tential utterances in dialogue. PhD thesis, University of Edinburgh. Schlenker, Philippe (2005). Non-redundancy: Towards a semantic reinterpretation of Binding Theory, Natural Language Semantics 13(1): 1–92. Schmeh, Katharina, Peter W. Culicover, Jutta Hartmann, and Susanne Winkler (2015). Dis­ course function ambiguity of fragments: A linguistic puzzle, in Susanne Winkler (ed.), Am­ biguity. Berlin: Mouton de Gruyter, 368–400. Schoorlemmer, Erik and Tanja Temmerman (2012). Head movement as a PF phenomenon: Evidence from identity under ellipsis, in Jaehoon Choi, E. Alan Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz, and Alex Trueman (eds), Proceedings of the 29th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 232–40. Schütze, Carson T. (2001a). On the nature of default case, Syntax 4(3): 205–38.

Page 108 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Schütze, Carson T. (2001b). Semantically empty lexical heads as last resorts, in Norbert Corver and Henk van Riemsdijk (eds), Semi-lexical categories: The function of content words and the content of function words. Berlin: Mouton de Gruyter, 127–87. Schuyler, Tamara (2002). Wh-movement out of the site of VP ellipsis. Master’s thesis, UCSC. Schwabe, Kerstin and Klaus von Heusinger (2001). On shared indefinite expression in co­ ordinative structures, Journal of Semantics 18: 243–69. Schwabe, Kerstin and Susanne Winkler (eds) (2003). The interfaces: Deriving and inter­ preting omitted structures. Amsterdam and Philadelphia: John Benjamins. Schwarz, Bernhard (1999). On the syntax of either…or, Natural Language and Linguistic Theory 17(2): 339–70. Schwarz, Bernhard (2000). Topics in ellipsis. PhD thesis, University of Massachusetts, Amherst, MA. Schwarzschild, Roger (1999). GIVENness, Avoid F and other constraints on the placement of focus, Natural Language Semantics 7(2): 141–77. (p. 1104)

Schwarzschild, Roger (2002). Singleton indefinites, Journal of Semantics 19(3): 289–314. Schwarzschild, Roger (2005). Measure phrases as modifiers of adjectives, Recherches lin­ guistiques de Vincennes 34: 207–28. di Sciullio, Anna Maria, Pieter Muysken, and Rajendra Singh (1986). Government and codeswitching, Journal of Linguistics 22: 1–24. Scott, Tatiana V. (2012). Whoever doesn’t HOP must be superior: The Russian left-periph­ ery and the emergence of superiority. PhD thesis, Stony Brook University. Sczcegielniak, Adam (2006). VP ellipsis and topicalization, in L. Bateman and C. Ussery (eds), NELS 35: Proceedings of the 35th annual meeting of the North East Linguistic So­ ciety. Amherst, MA: GLSA, 603–15. Selkirk, Elisabeth O. (1995). Sentence prosody: Intonation, stress and phrasing, in J. A. Goldsmith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell Publish­ ers, 550–69. Selkirk, Elisabeth O. (1970). On the determiner systems of noun phrases and adjective phrases. Ms. MIT. Selkirk, Elisabeth O. (1984). Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth O. (2000). Focus types and tones in English. Paper presented at the North American Phonology Conference 1. Page 109 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Selkirk, Elisabeth O. (2002). Contrastive FOCUS vs. presentational focus: Prosodic evi­ dence from Right Node Raising in English, in Bernard Bel and Isabelle Marlien (eds), Speech Prosody 2002: Proceedings of the First International Speech Prosody Conference. Aix-en-Provence: Université de Provence, Laboratoire Parole et Langage, 643–6. Sener, Serkan and Daiko Takahashi (2010). Ellipsis of argument in Japanese and Turkish, Nanzan Linguistics 6: 79–99. Seymour, Harry N., Tom Roeper, and Jill de Villiers (2003). DELV: Diagnostic evaluation of language variation-norm referenced. London: Pearson. Shabani, Mansour (2013a). Erteqâye gereye râst dar zabâne fârsi [Right Node Raising in the Persian language], Adabe pazhohi 24: 149–70. Shabani, Mansour (2013b). Pâreye jomlehâ dar zabâne fârsi [Fragments in the Persian language], Pazhoheshhâye zabâni 4: 81–100. Shabani, Mansour (2015). Erteqâye gereye râst: Sâxtâriye nâhamgen dar zabâne fârsi [Right Node Raising: A nonhomogeneous construction in Persian], Jostârhâye zabâni 6: 151–72. Shafiei, Nazila (2015). Ellipsis in Persian complex predicates: VVPE or something else. Paper presented at the annual conference of the Canadian Linguistic Association. Shannon, Claude E. (1993). Claude E. Shannon: Collected papers. Hoboken, NJ: John Wi­ ley & Sons. Shapiro, Lewis P., Betty Gordon, Nancy Hack, and Jennifer Killackey (1993). Verb argu­ ment structure processing in complex sentences in Broca’s and Wernicke’s aphasia, Brain and Language 45: 423–47. Shapiro, Lewis P. and Arild Hestvik (1995). On-line comprehension of VP-ellipsis: Syntac­ tic reconstruction and semantic influence, Journal of Psycholinguistic Research 24: 517– 32. Shapiro, Lewis P., Arild Hestvik, Lesli Lesan, and A. Rachel Garcia (2003). Charting the time-course of sentence processing: Evidence for an initial and independent structural analysis, Journal of Memory and Language 49: 1–19. Shapiro, Lewis P., David Swinney, and Susan Borsky (1998). Online examination of lan­ guage performance in normal and neurologically impaired adults, American Journal of Speech-Language Pathology 7: 49–60. Sherman (Ussishkin), Adam (1998). VP ellipsis and subject positions in Modern Hebrew, in Adam Zachary Wyne (ed.), Proceedings of the 13th annual meeting of the Israel Associ­ ation of Theoretical Linguistics. Jerusalem: Akademon, 211–29.

Page 110 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Shimoyama, Junko (1995). On ‘sluicing’ in Japanese. Ms. University of Massachu­ setts at Amherst. (p. 1105)

Shinohara, Michie (2004). Nihongo-no sakuzyo gensyoo-nituite [On the deletion phenome­ na in Japanese]. BA thesis, Nanzan University. Shiobara, Kayono (2010). Derivational linearization at the syntax–prosody interface. Tokyo: Hituzi Syobo. Shopen, Timothy (1972). A generative theory of ellipsis: a consideration of the linguistic use of silence. PhD thesis, UCLA. Shopen, Timothy (1973). Ellipsis as grammatical indeterminacy, Foundations of Language 10: 65–77. Shriberg, Elizabeth (1994). Preliminaries to a theory of speech disfluencies. PhD thesis, University of California at Berkeley. Siegel, Muffy E. A. (1987). Compositionality, case, and the scope of auxiliaries, Linguistics and Philosophy 10(1): 53–76. Šimík, Radek (2011). Modal existential wh-constructions. PhD thesis, University of Groningen. Simons, Mandy (2005). Semantics and pragmatics in the interpretation of ‘or’, in Effi Georgala and Jonathan Howell (eds), SALT 15: Proceedings of the 15th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 205–22. Simons, Mandy, Judith Tonhauser, David Beaver, and Craige Roberts (2011). What projects and why, in SALT 20: Proceedings of the 20th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, 309–27. Sleeman, Petra (1993). Noun ellipsis in French, Probus 5: 271–95. Sleeman, Petra (1996). Licensing empty nouns in French. Amsterdam: Holland Institute of Generative Linguistics. Sleeman, Petra and Aafke Hulk (2013). L1 acquisition of noun ellipsis in French and in Dutch: Consequences for linguistic theory, in Sergio Baauw, Frank Drijkoningen, Luisa Meroni, and Manuela Pinto (eds), Romance languages and linguistic theory. Amsterdam and Philadelphia: John Benjamins, 249–66. Smith, Edward E. and John Jonides (1999). Storage and executive processes in the frontal lobes, Science 283: 1657–61. Smith, Jennifer (2000). ‘You Ø na hear o’ that kind o’ things’: Negative do in Buckie Scots, English World-Wide 21(2): 231–59.

Page 111 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Sneddon, James Neil, Alexander Adelaar, Djenar Dwi N., and Michael C. Ewing (2010). In­ donesian: A comprehensive grammar. London: Routledge. Snider, Neal and Jeffrey Runner (2011). Structural parallelism aids ellipsis and anaphor resolution: Evidence from eye movements to semantic and phonological neighbors. Talk presented at the Linguistic Society of America Annual Meeting. Snider, Neal and Jeffrey Runner (2014). Semantic and phonological neighbors are re­ trieved in verb phrase ellipsis: Evidence for syntactic structure. Ms. University of Rochester. Soh, Hooi Ling and Hiroki Nomoto (2015). Degree achievements, telicity, and the verbal prefix meN- in Malay, Journal of Linguistics 51: 147–83. Speer, Shari and Allison Blodgett (2006). Prosody, in Matthew J. Traxler and Morton Ann Gernsbacher (eds), Handbook of psycholinguistics. Amsterdam: Elsevier, 505–37. Sportiche, Dominique (2008). Inward bound: Splitting the wh-paradigm and French rela­ tive qui. Ms. UCLA. Sprouse, Jon (2006). The Accent Projection Principle: Why the hell not? University of Pennsylvania Working Papers in Linguistics 12: 349–59. Sprouse, Jon and Norbert Hornstein (eds) (2013). Experimental syntax and island effects. Cambridge: Cambridge University Press. Stabler, Edward P. (1997). Derivational minimalism, in Christian Retoré (ed.), Logical as­ pects of computational linguistics. Dordrecht: Springer, 68–95. Stabler, Edward P. (2013). Two models of minimalist, incremental syntactic analysis, Top­ ics in Cognitive Science 5(3): 611–33. Stainton, Robert J. (2006a). Neither fragments nor ellipsis, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, 93–116. (p. 1106)

Stainton, Robert J. (2006b). Words and thoughts: Subsentences, ellipsis, and the philoso­ phy of language. Oxford: Oxford University Press. Stanley, Jason (2000). Context and logical form, Linguistics and Philosophy 23: 391–434. Starke, Michal (2001). Move dissolves into Merge: A theory of locality. PhD thesis, Uni­ versity of Geneva. von Stechow, Arnim (1984). Comparing semantic theories of comparison, Journal of Se­ mantics 3(1): 1–77.

Page 112 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References von Stechow, Arnim (1991). Syntax und Semantik, in Arnim von Stechow and Dieter Wun­ derlich (eds), Semantik: Ein Handbuch der zeitgenoessischen Forschung/Semantics: An international handbook of contemporary research. Berlin and New York: de Gruyter, 90– 148. von Stechow, Arnim (1996). The different readings of wieder ‘again’: A structural account, Journal of Semantics 13: 87–138. Steedman, Mark (1987). Combinatory grammars and parasitic gaps, Natural Language and Linguistic Theory 5(3): 403–39. Steedman, Mark (1990). Gapping as constituent coordination, Linguistics and Philosophy 13: 207–64. Steedman, Mark (2000). The syntactic process. Cambridge, MA: MIT Press. Steedman, Mark (2007). On ‘the computation’, in Gillian Ramchand and Charles Reiss (eds), The Oxford handbook of linguistic interfaces. Oxford: Oxford University Press, 575– 611. Steedman, Mark and Jason Baldridge (2011). Combinatory Categorial Grammar, in Robert D. Borsley and Kersti Börjars (eds), Non-transformational syntax: Formal and explicit models of grammar. Oxford: Blackwell, 181–224. Stepanov, Arthur (1998). On wh-fronting in Russian, in Pius N. Tamanji and Kiyomi Kusumoto (eds), NELS 28: Proceedings of the 28th annual meeting of the North East Lin­ guistic Society. Amherst, MA: Graduate Linguistic Student Association, 453–68. Stepanov, Arthur (2007). The end of CED? Minimalism and extraction domains, Syntax 10(1): 80–126. Stepanov, Arthur and Penka Stateva (2009). When QR disobeys superiority, Linguistic In­ quiry 40(1): 176–85. Sternefeld, Wolfgang (1998). Reciprocity and cumulative predication, Natural Language Semantics 6(3): 303–37. Stirling, Colin (2009). Decidability of higher-order matching, Logical Methods in Comput­ er Science 5(3): 1–52. Stjepanović, Sandra (1997). VP ellipsis in a verb raising language and implications for the condition on formal identity of verbs, in Jeong-Seok Kim, Satoshi Oku, and Sandra St­ jepanović (eds), Is the logic clear? Papers in honor of Howard Lasnik. Connecticut: Uni­ versity of Connecticut, 287–306. Stjepanović, Sandra (1998). Short-distance movement of wh-phrases in Serbo-Croatian matrix clauses. Paper presented at the Comparative Slavic Morphosyntax Workshop.

Page 113 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Stjepanović, Sandra (1999). Multiple sluicing and superiority in Serbo-Croatian, in Pius N. Tamanji, Masako Hirotani, and Nancy Hall (eds), NELS 29: Proceedings of the 29th annu­ al meeting of the North East Linguistic Society. Amherst, MA: GLSA, 145–60. Stjepanović, Sandra (2003). Multiple wh-fronting in Serbo-Croation matrix question and the matrix sluicing construction, in Cedric Boeckx and Kleanthes K. Grohmann (eds), Mul­ tiple wh-fronting. Amsterdam and Philadelphia: John Benjamins, 255–84. Stjepanović, Sandra (2008). P-stranding under sluicing in a non-P-stranding language? Linguistic Inquiry 39(1): 179–90. Stjepanović, Sandra (2012). Two cases of violation repair under sluicing, in Jason Mer­ chant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, 68–82. Stolterfoht, Britta (2004). Processing word order variations and ellipses: The in­ terplay of syntax and information structure during sentence comprehension. Leipzig: Max Planck Institute for Human, Cognitive, and Brain Sciences. (p. 1107)

Stolterfoht, Britta, Angela D. Friederici, Kai Alter, and Anita Steube (2007). Processing fo­ cus structure and implicit prosody during reading: Differential ERP effects, Cognition 104(5): 65–90. Stowe, Laurie A. (1986). Parsing wh-constructions: Evidence for on-line gap location, Lan­ guage and Cognitive Processes 3: 227–45. Stromswold, Karin, David Caplan, Nathaniel Alpert, and Scott Rauch (1996). Localization of syntactic comprehension by positron emission tomography, Brain and Language 52: 452–73. Stump, Gregory T. (1977). Pseudogapping. Ms. Ohio State University. Sturt, Patrick (2003). The time-course of the application of binding constraints in refer­ ence resolution, Journal of Memory and Language 48: 542–62. Sudo, Yasutada (2014). Hidden nominal structure in Japanese clausal comparatives, Jour­ nal of East Asian Linguistics 24(1): 1–51. Sugisaki, Koji (2005). ‘One’ issue in acquisition, in Yukio Otsu (ed.), Proceedings of the sixth Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, 345–60. Sugisaki, Koji (2007a). NP ellipsis and parameters: An acquisitional perspective, in Aly­ ona Belikova, Luisa Meroni, and Mari Umeda (eds), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition North America (GALANA). Somerville, MA: Cascadilla Press, 416–25. Sugisaki, Koji (2007b). The syntax of swiping: The view from child language. Ms. Univer­ sity of Connecticut. Page 114 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Sugisaki, Koji (forthcoming). Argument ellipsis in child Japanese revisited, in Proceedings of the 10th Workshop on Altaic Formal Linguistics (WAFL 10). Cambridge, MA: MITWPL. Suñer, Margarita and María Yépez (1988). Null definite objects in Quiteño, Linguistic In­ quiry 19(3): 511–19. Suomalaisen viittomakielen perussanakirja [The basic dictionary of FinSL] (1998). Helsin­ ki: Kuurojen Liitto ry ja Kotimaisten kielten keskus. Surányi, Balázs (2002). Multiple operator movements in Hungarian. PhD thesis, Utrecht University. Surányi, Balázs (2009a). ‘Incorporated’ locative adverbials in Hungarian, in Katalin É. Kiss (ed.), Adverbs and adverbial adjuncts at the interfaces. Berlin and New York: Mouton de Gruyter, 39–74. Surányi, Balázs (2009b). Verbal particles inside and outside vP, Acta Linguistica Hungari­ ca 56(2–3): 201–49. de Swart, Henriette (2001). Négation et coordination: La conjonction ‘ni’, in Reineke BokBenneman, Bob de Jonge, Brigitte Kampers-Manhe, and Arie Molendijk (eds), Adverbial Modification. Amsterdam: Rodopi, 109–24. Syrett, Kristen (2016). Acquisition of comparatives and degree constructions, in Jeffrey Lidz, William Snyder, and Joe Pater (eds), The Oxford handbook of developmental linguis­ tics. Oxford: Oxford University Press, 463–97. Syrett, Kristen and Jeffrey Lidz (2009). QR in child grammar: Evidence from antecedentcontained deletion, Language Acquisition 16(2): 67–81. Szabolcsi, Anna (1992). Combinatory grammar and projection from the lexicon, in Ivan A. Sag and Anna Szabolcsi (eds), Lexical matters. Stanford, CA: Center for the Study of Lan­ guage and Information, 241–68. Szabolcsi, Anna (1994). The noun phrase, in F. Kiefer and K. É. Kiss (eds), Syntax and se­ mantics 27: The syntactic structure of Hungarian. San Diego: Academic Press, 179–245. Szabolcsi, Anna (1997). Ways of scope taking. Dordrecht: Kluwer Academic Publishers. Szabolcsi, Anna (2006). Strong vs. weak islands, in Martin Everaert and Henk van Riems­ dijk (eds), The Blackwell companion to syntax. Malden, MA and Oxford: Blackwell, 479– 531. Szabolcsi, Anna (2010). Quantification. Cambridge: Cambridge University Press. Szabolcsi, Anna and Frans Zwarts (1992). Weak islands and an algebraic semantics for scope taking, Natural Language Semantics 1(3): 235–84. Szczegielniak, Adam (2004). Relativization and ellipsis. PhD thesis, Harvard University. Page 115 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Szczegielniak, Adam (2006). All sluiced up, but no alleviation in sight. Ms. Boston Col­ lege. (p. 1108)

Szczegielniak, Adam (2005). Relativization that you did. Cambridge, MA: MIT

Press. Szczegielniak, Adam (2008). Islands in sluicing in Polish, in Natasha Abner and Jane Bish­ op (eds), Proceedings from the 27th West Coast Conference on Formal Linguistics. Somerville, CA: Cascadilla Proceedings Project, 404–12. Takahashi, Daiko (1993). Movement of wh-phrases in Japanese, Natural Language and Linguistic Theory 11(4): 655–78. Takahashi, Daiko (1994). Sluicing in Japanese, Journal of East Asian Linguistics 3(3): 265– 300. Takahashi, Daiko (2006). Apparent parasitic gaps and null arguments in Japanese, Journal of East Asian Linguistics 15(1): 1–35. Takahashi, Daiko (2008a). Noun phrase ellipsis, in Shigeru Miyagawa and Mamoru Saito (eds), The Oxford handbook of Japanese linguistics. Oxford: Oxford University Press, 394– 422. Takahashi, Daiko (2008b). Quantificational null objects and argument ellipsis, Linguistic Inquiry 39(2): 307–26. Takahashi, Daiko (2014). Argument ellipsis, anti-agreement, and scrambling, in Mamoru Saito (ed.), Japanese syntax in comparative perspective. Oxford: Oxford University Press, 88–116. Takahashi, Daiko (2016). Koo syooryaku [Argument ellipsis], in Keiko Murasugi, Mamoru Saito, Yoichi Miyamoto, and Kensuke Takita (eds), Nihongo bunpoo Handobukku [A hand­ book of Japanese grammar]. Tokyo: Kaitakusha, 228–64. Takahashi, Daiko and Sichao Lin (2012). Two notes on multiple sluicing in Chinese and Japanese, Nanzan Linguistics 8: 129–45. Takahashi, Shoichi (2003). Pseudogapping: The view from Scandinavian languages. Paper presented at the Comparative Germanic Syntax Workshop 18. Takahashi, Shoichi (2004). Pseudogapping and cyclic linearization, in Keir Moulton and Matthew Wolf (eds), NELS 34: Proceedings of the 34th meeting of the North East Linguis­ tic Society. Amherst, MA: Graduate Linguistic Student Association, 571–85. Takahashi, Shoichi and Danny Fox (2005). MaxElide and the re-binding problem, in Effi Georgala and Jonathan Howell (eds), Proceedings of Semantics and Linguistic Theory (SALT XV). Ithaca, NY: CLC Publications, 223–40.

Page 116 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Takita, Kensuke (2007). NP-internal honorification and N′-deletion in Japanese, Nanzan Linguistics 1(2): 45–67. Takita, Kensuke (2009). ‘Genuine’ sluicing in Japanese, in Ryan Bochnak, Nassira Nicola, Peet Klecha, Jasmine Urban, Alice Lemieux, and Christina Weaver (eds), Proceedings of the 45th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 577–92. Takita, Kensuke (2011). An argument for argument ellipsis from -sika NPIs, in Suzi Lima, Kevin Mullin, and Brian Smith (eds), NELS 39: Proceedings of the 39th annual meeting of the North East Linguistic Society. Amherst: GLSA, 771–84. Takita, Kensuke and Nobu Goto (2013). On (im)possible N′-deletion within PPs, Nanzan Linguistics 9: 215–31. Tanaka, Hidekazu (2011). Voice mismatch and syntactic identity, Linguistic Inquiry 42: 470–90. Tancredi, Christopher (1992). Deletion, deaccenting and presupposition. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Tanenhaus, Michael K. and Greg N. Carlson (1990). Comprehension of deep and surface verb phrase anaphors, Language and Cognitive Processes 5(4): 257–80. Tang, C.-C. Jane (1990). Chinese phrase structure and the extended X′-theory. PhD thesis, Cornell University. Tao, Hongyin and Charles F. Meyer (2006). Gapped coordinations in English: Form, us­ age, and implications for linguistic theory, Corpus Linguistics and Linguistic Theory 2(2): 129–63. Temmerman, Tanja (2012). Multidominance, ellipsis, and quantifier scope. PhD thesis, Leiden University. Temmerman, Tanja (2013). The syntax of Dutch embedded fragment answers: On the PFtheory of islands and the wh/sluicing-correlation, Natural Language and Linguistic Theory 31: 235–385. (p. 1109)

Tesnière, Lucien (1959). Éléments de syntaxe structurale. Paris: Klincksieck.

Tesnière, Lucien (2015). Elements of structural syntax. Amsterdam and Philadelphia: John Benjamins. Testelets, Yakov (2011). Ellipsis v russkom jazyke: Teoretičeskij i opisatelnyj podxody. Pa­ per presented at the conference Tipologija morfosintaksičeskix parametrov. Thompson, Andrea (2014). Beyond deep and surface: Explorations in the typology of anaphora. PhD thesis, UCSC. Page 117 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Thoms, Gary (2011a). From economy to locality: Do-support as head movement. Ms. Strathclyde University. Thoms, Gary (2011b). ‘Verb floating’ and VP-ellipsis: Towards a movement account of el­ lipsis licensing, in Jeroen van Craenenbroeck (ed.), Linguistic variation yearbook 2010. Amsterdam: John Benjamins, 252–97. Thoms, Gary (2012). Towards a movement theory of ellipsis licensing. Paper presented at UiL-OTS. Thoms, Gary (2014). MaxElide and clause structure in Scottish Gaelic, Linguistic Inquiry 45: 158–68. Thoms, Gary (2015). Syntactic identity, parallelism and accommodated antecedents, Lin­ gua 166: 172–98. Thoms, Gary (2016). Short answers in Scottish Gaelic and their theoretical implications, Natural Language and Linguistic Theory 34: 351–91. Thoms, Gary, David Adger, Caroline Heycock, and Jennifer Smith (2013). ‘Remarks on negation in varieties of Scots’. Paper presented at the ReCoS Workshop on English Di­ alects. Thoms, Gary, David Adger, Caroline Heycock, and Jennifer Smith (2016). On some surpris­ ing contractions in Scots dialects. Paper presented at LAGB (Linguistic Association of Great Britain). Thoms, Gary and Craig Sailor (2017). When silence gets in the way: Extration from do-el­ lipsis in British dialects. Paper presented at the LSA. Thoms, Gary and George Walkden (2015). vP-fronting with and without remnant move­ ment. Ms. University of Glasgow and University of Manchester. Thornton, Rosalind (2011). Minimal answers: Ellipsis, syntax and discourse in the acquisi­ tion of European Portuguese, Language 87(1): 229–34. Thornton, Rosalind and Kenneth Wexler (1999). Principle B, VP ellipsis, and interpreta­ tion in child grammar. Cambridge, MA: MIT Press. Tian, Ye and Jonathan Ginzburg (2017). No, I am: What are you saying ‘no’ to. Ms. Univer­ sité Paris-Diderot. Ticio, Emma (2003). On the structure of DPs. PhD thesis, University of Connecticut at Storrs. Timm, Lenora A. (1975). Spanish-English code-switching: El porqué and how-not-to, Ro­ mance Philology 28: 473–82.

Page 118 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Titov, Elena (2012). Encoding focus and contrast in Russian, in Ad Neeleman and Reiko Vermeulen (eds), The syntax of topic, focus and contrast: An interface-based approach. Berlin: Mouton de Gruyter, 119–55. Todorovic, Neda (2014). VP-ellipsis and all its phases: The role of aspect in VP-ellipsis in Serbian, in Cassandra Chapman, Olena Kit, and Ivona Kučerová (eds), Proceedings of For­ mal Approaches to Slavic Linguistics 22. Ann Arbor: Michigan Slavic Publications. Tomioka, Satoshi (1999). A sloppy identity puzzle, Natural Language Semantics 7(2): 217– 41. Tomioka, Satoshi (2003). The semantics of Japanese null pronouns and its crosslinguistic implications, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, 321–39. Tomioka, Satoshi (2008). A step-by-step guide to ellipsis resolution, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, 210–28. Tomioka, Satoshi (2012). Sluicing in Mandarin Chinese, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, 219–47. Toosarvandani, Maziar (2006). v-stranding VPE: Ellipsis in Farsi complex predicates, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS 36: Proceedings of the 36th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 639–52. Toosarvandani, Maziar (2008). Wh-movement and the syntax of sluicing, Journal of Linguistics 44: 677–722. (p. 1110)

Toosarvandani, Maziar (2009). Ellipsis in Farsi complex predicates, Syntax 12(1): 60–92. Toosarvandani, Maziar (2013a). Corrective but coordinates clauses not always but some­ times, Natural Language and Linguistic Theory 31: 827–63. Toosarvandani, Maziar (2013b). Gapping is VP-ellipsis: A reply to Johnson. Ms. MIT. Toosarvandani, Maziar (2016). Embedding the antecedent in gapping: low coordination and the role of parallelism, Linguistic Inquiry 47: 381–90. Toribio, Almeida Jacqueline and Kay González-Vilbazo (2014). Operator movement in Eng­ lish-Spanish and German-Spanish codeswitching, in Jeff MacSwan (ed.), Grammatical the­ ory and bilingual codeswitching. Cambridge, MA: MIT Press, 87–118. Traugott, Elizabeth (2014). Towards a constructional framework for research on language change, Cognitive Linguistic Studies 1(1): 3–21. Traxler, Matthew J. and Lyn Frazier (2008). The role of pragmatic principles in resolving attachment ambiguities: Evidence from eye movements, Memory and Cognition 36(2): 314–28. Page 119 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Traxler, Matthew J. and Martin J. Pickering (1996). Plausibility and the processing of un­ bounded dependencies: An eye-tracking study, Journal of Memory and Language 35(3): 454–75. Ueno, Yoshio (2015). An automodular view of ellipsis. Tokyo: Waseda University Press. Ueyama, Aymui (1998). Two types of dependency. PhD thesis, University of Southern Cali­ fornia. Ueyama, Aymui (2003). Two types of scrambling constructions in Japanese, in Andrew Barss (ed.), Anaphora: A reference guide. Malden, MA: Oxford University Press, 23–71. Uriagereka, Juan (1999). Multiple spell-out, in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, 251–82. Vallduví, Enric (2016). Information structure, in Maria Aloni and Paul Dekker (eds), The Cambridge handbook of formal semantics. Cambridge: Cambridge University Press, 728– 55. Valmala, Vidal (2013). On Right Node Raising in Catalan and Spanish, Catalan Journal of Linguistics 12: 219–51. Valois, Daniel, Phaedra Royle, Ann Sutton, and Evelyne Bourdua-Roy (2009). L’ellipse du nom en français: Le rôle des données de l’acquisition pour la théorie linguistique, Canadi­ an Journal of Linguistics 54(2): 339–66. Van Valin Jr, Robert D. (1986). An empty category as the subject of a tensed S in English, Linguistic Inquiry 13: 557–61. Van Valin Jr, Robert D. and Randy J. LaPolla (1997). Syntax: Structure, meaning and func­ tion. Cambridge: Cambridge University Press. Vanden Wyngaerd, Guido (1989). Object shift as an A-movement rule, MIT Working Pa­ pers in Linguistics 11: 256–71. Vanden Wyngaerd, Guido (2009). Gapping constituents. Ms. HUB. Vasić, Nada, Sergey Avrutin, and Esther Ruigendijk (2006a). The economy hierarchy of referential dependencies in child language: Evidence from Dutch VP-ellipsis, in Kamil Ud Deen, Jun Nomura, Barbara Schulz, and Bonnie D. Schwartz (eds), University of Connecti­ cut Occasional Papers in Linguistics 4: Proceedings of the Inaugural Conference on Gen­ erative Approaches to Language Acquisition—North America (GALANA). Storrs: Universi­ ty of Connecticut Linguistics Club, 413–24. Vasić, Nada, Sergey Avrutin, and Esther Ruigendijk (2006b). Interpretation of pronouns in VP-ellipsis constructions in Dutch Broca’s and Wernicke’s aphasia, Brain and Language 96: 191–206. Vergnaud, Jean-Roger (1974). French relative clauses. PhD dissertation, MIT. Page 120 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Vicente, Luis (2006). Negative short replies in Spanish, in Jeroen van de Weijer and Bet­ telou Los (eds), Linguistics in the Netherlands 2006. Amsterdam and Philadelphia: John Benjamins, 199–211. Vicente, Luis (2008). Syntactic isomorphism and non-isomorphism under ellipsis. Ms. UCSC. Vicente, Luis (2010a). A note on the movement analysis of gapping, Linguistic Inquiry 41(3): 509–17. Vicente, Luis (2010b). On the syntax of adversative coordination, Natural Lan­ guage and Linguistic Theory 28(2): 381–415. (p. 1111)

Vicente, Luis (2014a). Morphological case mismatches under sluicing. Ms. University of Potsdam. Vicente, Luis (2014b). Sluicing and its subtypes. Ms. University of Potsdam. Vicente, Luis (2015). Morphological case mismatches under sluicing. Snippets 29: 16–17. de Villiers, Jill, Peter A. de Villiers, and Tom Roeper (2011). Wh-questions: Moving beyond the first phase, Lingua 121(3): 352–66. Vlachos, Christos (2011). The mechanics of sluicing, Linguistic Analysis 37: 273–314. de Vries, Gertrud (1992). On coordination and ellipsis. PhD thesis, Tilburg University. de Vries, Mark (2005). Ellipsis in nevenschikking: Voorwaarts deleren maar achterwaarts delen, Tabu 34: 13–46. de Vries, Mark (2009). On multidominance and linearization, Biolinguistics 3: 344–403. de Vries, Mark (2013). Multidominance and locality, Lingua 134: 149–69. de Vries, Mark (2017). Across-the-board phenomena, in Martin Everaert and Henk van Riemsdijk (eds), The Wiley Blackwell companion to syntax. 2nd edition. Hoboken: WileyBlackwell, 1–31. Wagner, Michael and Duane Watson (2010). Experimental and theoretical advanes in prosody: A review, Language and Cognitive Processes 25(7–9): 905–45. Walenski, Matthew, Tracy Love, Michelle Ferrill, Isabelle Deschamps, Peter Pieperhoff, Katrin Amunts, Yosef Grodzinsky, and Lewis P. Shapiro (2012). Structure–function corre­ spondences in Broca’s aphasia: Evidence from MRI and comprehension of verb phrase el­ lipsis constructions. Paper presented at the Clinical Aphasiology Conference. Walker, Edward, Phillip Gough, and R. Wall (1968). Grammatical relations and the search of sentences in immediate memory. Paper presented at the Midwestern Psychological As­ sociation. Page 121 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Walters, Minette (1996). The dark room. London: Pan Books. Wang, Chyan-An (2006). Sluicing and resumption, in Emily Elfner and Martin Walkow (eds), NELS 37: Proceedings of the 37th annual meeting of the North East Linguistic So­ ciety. Amherst: GLSA, 239–52. Ward, Gregory and Julia Hirschberg (1985). Implicating uncertainty: The pragmatics of fall-rise intonation, Language 61: 747–76. Ward, Gregory, Richard Sproat, and Gail McKoon (1991). A pragmatic account of so-called anaphoric islands, Language 67: 439–74. Warner, Anthony (1985). The structure of English auxiliaries: A phrase structure grammar. Bloomington, IN: Indiana University Linguistics Club. Warner, Anthony (1986). Ellipsis conditions and the status of the English copula, York Pa­ pers in Linguistics 12: 153–72. Wasow, Thomas (1972). Anaphoric relations in English. Doctoral dissertation, MIT, Cam­ bridge, MA. Wasow, Thomas (1979). Anaphora in generative grammar. Ghent: E. Story-Scientia. Wasow, Thomas and Tom Roeper (1972). On the subject of gerunds, Foundations of Lan­ guage 8(1): 44–62. Webber, Bonnie (1979). A formal approach to discourse anaphora (Outstanding Disserta­ tions in Linguistics). New York: Garland. Weber, David (1983). Relativization and nominalized clauses in Huallaga Quechua. Berke­ ley: University of California Press. Weber, Heinz J. (1992). Dependenzgrammatik: Ein Arbeitsbuch. Tübingen: Gunter Narr Verlag. van Weelden, Lisanne (2009). Discourse integration in children: An eye-tracking study to children’s interpretation of bare cardinals. Master’s thesis, Utrecht Univeristy. Weerman, Fred (1989). The V2 conspiracy: A synchronic and diachronic analysis of verbal positions in Germanic language. Dordrecht: Foris Publications. Weir, Andrew (2007). Scots na: Clitic or affix? Ms. University of Massachusetts at Amherst. Weir, Andrew (2014). Fragments and clausal ellipsis. PhD thesis, University of Massachusetts at Amherst. (p. 1112)

Page 122 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Weir, Andrew (2015). Fragment answers and exceptional movement under ellipsis: A PFmovement account, in Thuy Bui and Deniz Özyildiz (eds), NELS 45: Proceedings of the 45th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, 175–88. Wentz, J. and E. McClure (1976). Ellipsis in bilingual discourse. In Papers from the 12th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 656–65. Werner, Edwin (2011). The ellipsis of ‘ellipsis’: A reanalysis of ‘elided’ noun phrase struc­ tures in German. Master’s thesis, Utrecht Univeristy. Wesche, B. (1995). Symmetric coordination. Tübingen: Niemeyer. Wescoat, Michael (1989). Sloppy readings with embedded antecedents. MS. Stanford Uni­ versity. Wexler, Kenneth and Peter Culicover (1980). Formal principles of language acquisition. Cambridge, MA: MIT Press. Whitman, Neal (2004). Semantics and pragmatics of English verbal dependent coordina­ tion, Language 80(3): 403–34. Whitman, Neal (2009). Right-node wrapping: Multimodal categorial grammar and the ‘Friends in Low Places’ coordination, in Erhard Hinrichs and John Nerbonne (eds), Theory and evidence in semantics. Stanford: CSLI, 235–56. Wiedmann, Natalie and Susanne Winkler (2015). The role of prosody in children’s pro­ cessing of ambiguous sentences, in Susanne Winkler (ed.), Ambiguity. Berlin: Mouton de Gruyter, 185–98. Wijnen, Frank and Edith Kaan (2006). Dynamics of semantic processing: The interpreta­ tion of bare quantifiers, Language and Cognitive Processes 21(6): 684–720. Wijnen, Frank, Tom Roeper, and Hiske van der Meulen (2003). Discourse binding: Does it begin with nominal ellipsis?, in Jacqueline van Kampen and Sergio Baauw (eds), Proceed­ ings of GALA 2003 on language acquisition. Utrecht: Utrecht University, 505–16. Wilder, Chris (1994). Coordination, ATB, and ellipsis, Groninger Arbeiten zur Germanistis­ chen Linguistik 37: 291–331. Wilder, Chris (1997). Some properties of ellipsis in coordination, in Artemis Alexiadou and T. Hall (eds), Studies on Universal Grammar and typological variation. Amsterdam: John Benjamins, 59–107. Wilder, Chris (1999). Right Node Raising and the LCA, in Sonya Bird, Andrew Carnie, Ja­ son D. Haugen, and Peter Norquest (eds), Proceedings of the 18th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 586–98.

Page 123 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Wilder, Chris (2008). Shared constituents and linearization, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, 229–58. Williams, Edwin (1974). Rule ordering in syntax. PhD thesis, Massachusetts Institute of Technology. Williams, Edwin (1977a). Across-the-board application of rules, Linguistic Inquiry 8(4): 419–23. Williams, Edwin (1977b). Discourse and logical form, Linguistic Inquiry 8: 101–39. Williams, Edwin (1977c). On deep and surface anaphora, Linguistic Inquiry 8: 692–6. Williams, Edwin (1981). Transformationless grammar, Linguistic Inquiry 12(4): 645–53. Williams, Edwin (1990). The ATB-theory of parasitic gaps, The Linguistic Review 6: 265– 79. Williams, Edwin (1997). Blocking and anaphora, Linguistic Inquiry 28(4): 577–628. Winkler, Susanne (2003). Ellipsis at the interfaces: An information-structural investiga­ tion of sentence-bound and discourse-bound ellipsis in English. PhD thesis, University of Tübingen. Winkler, Susanne (2005). Ellipsis and focus in generative grammar. Berlin: Mouton de Gruyter. Winkler, Susanne (2006). Ellipsis and the division of labor at the interfaces, in Valeria Molnár and Susanne Winkler (eds), The architecture of focus. Berlin and New York: Mou­ ton de Gruyter, 1–28. Winkler, Susanne (2013). Syntactic diagnostics for extraction of focus from ellipsis site, in Lisa L.-S. Cheng and Norbert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, 463–84. Winkler, Susanne (2015). Exploring ambiguitiy and the ambiguity model from a transdis­ ciplinary perspective, in Susanne Winkler (ed.), Ambiguity. Berlin: Mouton de Gruyter, 1– 28. Winkler, Susanne (2016). Ellipsis and information structure, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, 359–82. (p. 1113)

Winkler, Susanne and Kerstin Schwabe (2003). Exploring the interfaces from the perspec­ tive of omitted structures, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Ben­ jamins, 1–26.

Page 124 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Winter, Yoad (2001). Flexibility principles in Boolean semantics. Cambridge, MA: MIT Press. Winterson, Jeanette (1987). The passion. Westminster: Penguin Books. Wittgenstein, Ludwig (1953). Philosophical investigations. Oxford: Basil Blackwell. Wold, Dag (1995). Antecedent-contained deletion in comparative constructions. Ms. MIT. Wong, Kwong-Cheong and Jonathan Ginzburg (2013). Investigating non-sentential utter­ ances in a spoken Chinese corpus, in Proceedings of PACLING 2013, Tokyo. Woolford, Ellen (1983). Bilingual code-switching and syntactic theory, Linguistic Inquiry 14(3): 520–36. Woolford, Ellen (1985). On the application of wh-movement and inversion in code-switch­ ing sentences, Revue Québécoise de Linguistique 14(1): 77–86. Wulf, Alyssa, Paul Dudis, Robert Bailey, and Ceil Lucas (2002). Variable subject presence in ASL narratives, Sign Language Studies 3: 54–76. Wurmbrand, Susi (2017). Stripping and topless complements, Linguistic Inquiry 48: 341– 66. Xiang, Ming (2003). A phrasal analysis of Chinese comparatives, in Amy Franklin, Jon Cih­ lar, David Kaiser, and Irene Kimbara (eds), Proceedings of the 39th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 739–54. Xiang, Ming, Jason Grove, and Jason Merchant (2014). Ellipsis sites induce structural priming effects. Ms. University of Chicago. Yamauchi, Nobuyuki (2006). Some properties of vice versa: A corpus-based approach, Journal of Culture and Information Science 1: 9–15. Yang, Charles D. (2002). Knowledge and learning in natural language. Oxford: Oxford University Press. Yatabe, Shuichi (2001). The syntax and semantics of left-node raising in Japanese, in Daniel P. Flickinger and Andreas Kathol (eds), Proceedings of the 7th International Con­ ference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 325–44. Yatabe, Shuichi (2003). A linearization-based theory of summative agreement in peripher­ al-node raising constructions, in Jong-Bok Kim and Stephen Wechsler (eds), Proceedings of the 9th International Conference on Head-Driven Phrase Structure Grammar. Stan­ ford: CSLI Publications, 391–413. Yatabe, Shuichi (2012). Comparison of the ellipsis-based theory of non-constituent coordi­ nation with its alternatives, in Stefan Müller (ed.), Proceedings of the 19th International

Page 125 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, 454–74. Yoshida, Masaya (2005). The rightward movement analysis of gapping in NP and its struc­ tural implications, in John Aldrete, Chung-Hye Han, and Alexei Kochetov (eds), Proceed­ ings of the West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 388–96. Yoshida, Masaya (2010). ‘Antecedent-contained’ sluicing, Linguistic Inquiry 41(1): 348–56. Yoshida, Masaya, Lauren Ackerman, Rebekah Ward, and Morgan Purrier (2012a). The processing of backward sluicing. Paper presented at the 25th Annual CUNY Conference on Human Sentence Processing. Yoshida, Masaya, Katy Carlson, and Michael Walsh Dickey (2013). Incremental parsing, gapping, and connectives. Paper presented at the 26th Annual CUNY Conference on Hu­ man Sentence Processing. Columbus, SC. Yoshida, Masaya, Michael Walsh Dickey, and Patrick Sturt (2013a). Predictive processing of syntactic structure: Sluicing and ellipsis in real-time sentence processing, Language and Cognitive Processes 28(3): 272–302. Yoshida, Masaya, Jiyeon Lee, and Michael Walsh Dickey (2013b). The island (in)sensitivity of sluicing and sprouting, in Jon Sprouse and Norbert Hornstein (eds), Experimental syn­ tax and island effects. Cambridge: Cambridge University Press, 360–76. Yoshida, Masaya, Tim Hunter, and Michael Frazier (2015). Parasitic gaps li­ censed by elided syntactic structure, Natural Language and Linguistic Theory 33: 1439– (p. 1114)

71. Yoshida, Masaya, Nina Kazanina, Leticia Pablos, and Patrick Sturt (2014a). On the origin of islands, Language, Cognition and Neuroscience 29: 761–70. Yoshida, Masaya, Chizuru Nakao, and Iván Ortega-Santos (2014b). The syntax of why-stripping, Natural Language and Linguistic Theory 33: 323–70. Yoshida, Masaya, Honglei Wang, and David Potter (2012b). Remarks on ‘gapping’ in DP, Linguistic Inquiry 43(3): 475–94. Zagona, Karen (1982). Government and proper government of verbal projections. PhD thesis, University of Washington. Zagona, Karen (1988a). Proper government of antecedentless VPs in English and Span­ ish, Natural Language and Linguistic Theory 6: 95–128. Zagona, Karen (1988b). Verb phrase syntax: A parametric study of English and Spanish. Dordrecht: Kluwer Academic Publishers.

Page 126 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

References Zamparelli, Roberto (2011). Coordination, in Klaus von Heusinger, Claudia Maienborn, and Paul Portner (eds), Semantics: An international handbook of natural language mean­ ing. Berlin: Mouton de Gruyter, 1713–41. Zipf, George K. (1949). Human behavior and the principle of least effort. Cambridge: Ad­ dison Wesley. Zocca, Cynthia (2003). O que não está lá: Um estudo sobre morfologia flexional em elipses. Master’s thesis, Universidad Estatal de Campinas. Zoerner, Cyril Edward, III (1995). Coordination: The syntax of &P. PhD thesis, University of California at Irvine. van Zonneveld, Ron (1994). Kleine syntaxis van het Nederlands. Dordrecht: ICG Publica­ tions. Zribi-Hertz, Anne (1986). Relations anaphoriques en français: Esquisse d’une grammaire générative raisonnée de la féflexivité et de l’ellipse structurale. PhD thesis, Université Paris 8. Zubizarreta, María-Luisa (1982). On the relationship of the lexion to syntax. PhD thesis, MIT. Zurif, Edgar B., David Swinney, Penny Prather, Julie Solomon and Camille Bushell (1993). An on-line analysis of syntactic processing in Broca’s and Wernicke’s aphasia, Brain and Language 45: 448–64. Zwaan, Rolf A. and Gabriel A. Radvansky (1998). Situation models in language compre­ hension and memory, Psychological Bulletin 123(2): 162–85. Zwart, C. Jan-Wouter (1993). Dutch syntax: A Minimalist approach. PhD thesis, University of Groningen. Zwart, C. Jan-Wouter (1994). Dutch is head-initial, The Linguistic Review 11(3–4): 377– 406. Zwart, C. Jan-Wouter (2014). The syntax of Dutch. Cambridge: Cambridge University Press. Zwarts, Frans (1978). Extractie uit prepositionele woordgroepen in het Nederlands, in Ans van Berkel, Willem Blok, Gerard Brummel, and Theo A. J. M. Janssen (eds), Proeven van Neerlandistiek. Groningen: Nederlands Instituut, 303–99. Zwicky, Arnold (1982). Stranded to and phonological phrasing, Linguistics 20(1/2): 3–58.

Page 127 of 127

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index

Index   The Oxford Handbook of Ellipsis Edited by Jeroen van Craenenbroeck and Tanja Temmerman Print Publication Date: Dec 2018 Subject: Linguistics Online Publication Date: Jan 2019

(p. 1115)

Index

acceptable ungrammaticality 266 acquisition 3, 5, 10, 76–7, 212, 228, 277–84, 287, 289, 291–4, 296–301, 303–4, 306, 313, 425, 438, 463, 813 acquisition path 278–80, 282–4, 291, 293, 296, 299, 313 across-the-board movement 574, 577, 584, 590, 642, 683, 882–3, 885, 939–40; see also ATB agentive 130, 785 agreement 9, 25, 31, 35, 39, 48–9, 60, 171, 276, 297, 302, 304, 306, 364, 371, 411, 462, 471, 510, 531, 545, 548–9, 552–4, 634, 676, 687, 691, 694–6, 707–9, 714, 748, 762, 799–800, 813, 815, 818– 20, 824, 891, 901, 903, 906, 911, 925, 941, 947, 952, 983, 986, 989, 991, 1008, 1037 align 167, 170, 174, 178, 180–1, 185, 187, 631 alternative semantics 582 ambiguity 85, 101, 116, 207, 209–10, 223, 248, 268, 270, 293, 329, 332–3, 344, 351–2, 356, 376–8, 427–8, 449, 451, 453, 530–1, 538, 540, 596, 616, 767, 778, 858, 908, 950–1 anaphor deep anaphor 7, 129, 137, 196, 259, 287–8, 528–9, 531–2, 546–7, 561, 659–60, 664, 668–9, 679–80, 855–6, 866, 893, 899, 950, 1035 discourse anaphora 178, 636 do it anaphor 176, 324, 660–1, 665 do so anaphor 323, 660 one anaphor 179, 531 sentential it anaphor 660 surface anaphor 126, 129, 259, 503, 528–9, 533–5, 538, 543, 547, 555–6, 566–7, 573, 659–61, 664, 677, 695, 856, 866, 899, 951 VP anaphor 171, 175, 260, 264, 287, 298, 343, 786, 808–9 antecedent ACD 14–15, 65, 126, 134–5, 137–41, 259, 267, 272, 316, 646, 787, 821, 985; see also an­ tecedent-contained deletion antecedent-contained deletion 15, 65, 101, 122, 137, 208, 259, 280, 292–3, 312, 316, 343, 423, 646, 787, 932, 985; see also ACD antecedentless 606–10, 613–14, 616, 623; see also non-linguistic antecedent; implicit corre­ late; pragmatic antecedent; pragmatic control linguistic antecedent 53, 95–6, 129–30, 151–2, 173, 179, 196–7, 202–3, 206, 210, 241, 259, 320, 444, 503, 509, 538, 547, 550, 608, 659–60, 727, 866–7, 951, 966, 1002, 1013, 1035 Page 1 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index missing antecedent 62, 175, 659–61, 679, 787 nominal antecedent 94, 97, 509–10, 542 non-linguistic antecedent 179, 660, 679, 727, 951, 1002; see also antecedentless; implicit correlate; pragmatic antecedent; pragmatic control pragmatic antecedent 659–60, 679, 855–6; see also antecedentless; non-linguistic an­ tecedent; implicit correlate; pragmatic control split antecedent 92–3, 319, 709 anti-connectivity 605, 609, 611–12 anti-government constraint 461–2 aphasia 5, 11, 425, 428–31, 433–7, 442–3 appositives 240, 247, 252 argument cluster conjunction 194–5 argument deletion 298 argument drop 11–12, 19, 280, 296–8, 312, 979–80, 1013 argument ellipsis 20, 27, 76, 78, 83, 86, 90, 97–8, 163, 165, 168, 298, 562, 645, 690, 727, 765, 769–72, 776, 780, 823, 854, 873, 886, 889–91, 893–4, 973, 988, 1014 Argument Realization Principle 100 argument-adjunct asymmetry 895 assertion 65, 105, 111, 120, 166, 269–71, 275, 312, 435, 554 ATB 82, 624, 637–8, 642–6, 683–4, 691–2, 694, 696–7, 700–3, 705, 713, 715, 717–18, 897–8, 1012; see also across-the-board movement augment 153, 900–2, 904, 919 auxiliary 14, 43, 59, 61, 73–5, 79, 93, 98–101, 127, 130, 137, 152, 156–7, 176–7, 193, 199, 221, 268, 289, 314, 320, 326–8, 340, 343, 354, 358–9, 364–5, 372, 379, 381–3, 385, 391, 426, 504–8, 511–13, 515, 518, 525, 579, 587, 640, 642, 644–5, 666–7, 697, 699, 704–5, 714, 722, 729, 777, 785–6, 788–9, 815, 820–1, 824, 827–8, 845, 847, 905, 937, 947–8, 952, 979–80, 985, 1006, 1009, 1020–32, 1037 Avoid Pronoun Principle 675 BAC 663, 1002; see also Backward Anaphora Constraint Backward Anaphora Constraint 344, 663, 992, 1001; see also BAC backward conjunction reduction 682–3, 686, 689, 695 backward ellipsis 688, 691, 709, 711, 1001 bare nominal 749, 888–9 bilingualism 458, 460, 463–4, 466–7, 476 binary branching 142, 163 (p. 1116) binding 25, 38, 44, 87–8, 96, 107, 131, 136, 139–40, 178–9, 183, 185, 211, 216, 224, 230, 263, 290–2, 296, 298, 316, 318, 321, 328–35, 337–8, 343, 345, 348, 380, 395, 409, 413, 416–17, 436, 451, 457, 501, 615, 620, 623, 696, 739, 743, 809, 814, 868, 891, 944, 947, 996, 1016 borrowing 458 bound pronouns 336, 380 bound variable 66, 243, 290–2, 329–31, 347–9, 428, 572, 581, 597, 636, 853–4, 867–9, 880, 888–9 British English do 15, 423, 505, 522, 524–5, 677, 680, 1032–7 Broca 11, 425, 428–31, 433, 435–7, 442 c-command 73–4, 131, 258, 263, 273, 330–2, 379, 401, 404, 450–1, 579, 630, 635, 654, 694, 804, 824, 868, 870, 873, 875, 888–9, 892, 897 cascaded ellipsis 350 Case Condition 44, 471

Page 2 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index case marking 15, 127–8, 183–4, 201, 203, 216, 224, 276, 377, 397, 462, 487, 605, 609, 613–14, 616, 843, 940, 963, 1015 case matching 25, 29, 48, 53, 128, 398, 401, 420, 654, 843, 857, 963, 965, 970, 972 case-marked cleft 869–70, 876–8, 880, 898 cataphora 227, 318, 320 Categorial Grammar 3, 88–9, 122, 134, 693; see also CG catena 8, 146–7, 153–61, 795 CG 109, 122–3, 126, 134–5, 138, 140–1; see also Categorial Grammar clarification 83–4, 86, 103, 105–6, 110, 115, 117–18, 207, 209, 225, 229, 352–3, 355, 977 clausal ellipsis 5, 63, 311, 365, 451, 457, 504, 598, 611, 758, 776, 778, 785, 790, 827, 829–30, 832–4, 916, 920, 922–3, 988, 1004–6, 1009, 1017, 1020 cleft 25, 29, 35, 50, 64–5, 251, 475, 482–4, 486–7, 489–90, 493–5, 499–502, 518, 610, 612, 618, 622, 759, 793–4, 858–9, 861, 866, 869–70, 873–4, 876–8, 880, 898, 919, 966, 970, 975 codeswitching 5, 11, 458–64, 466–7, 469–70, 473–5 coherence 8, 10, 117–18, 269–70, 321–2, 325–8, 337, 340, 782 coindexing 300, 417, 714, 796, 810, 814 communicative motivation 188 comparative attributive comparative 650, 656 case-marked comparative 874–6, 887, 889, 892–3, 896, 899 clausal comparative 625, 649, 651, 656, 935, 984–5, 1015–17 comparative deletion 6, 13, 255, 625, 631, 656, 721, 749–50, 755, 764, 805, 807, 838, 841, 935, 984, 988, 1015, 1017 comparative ellipsis 26, 32, 178, 201, 377, 638, 785, 920, 922, 933 comparative subdeletion 749–51, 805, 922 direct analysis 683–4, 689, 692, 694–5, 700, 851–2, 985 matching analysis 632–5, 637 non-case-marked comparative 874, 876, 899 non-elliptical comparative 874, 893 phrasal comparative 635, 638, 649, 749–50, 754–5, 757, 922, 935, 984–5, 1015–17 complete subtree 156, 158–60 complex predicate 521, 948, 951, 953, 961 concealed cleft 873 Condition C 64, 66, 318, 329, 868, 944, 946; see also Principle C conjunction 6, 13, 20, 59, 109, 128, 194–5, 226, 237, 255, 257, 294, 344, 359, 368–9, 372, 377, 414, 563, 566, 581, 588, 596–7, 600, 660, 681–2, 684–6, 691–2, 694, 721, 730–1, 733–4, 737, 754, 764–5, 773, 779, 781–2, 803–4, 841, 849, 853, 863, 912, 936, 940, 992, 1004, 1012 conjunction reduction 6, 13, 20, 359, 369, 372, 377, 563, 660, 681–2, 684, 691–2, 721, 730–1, 733–4, 737, 754, 764–5, 773, 841, 853, 863, 936; see also backward conjunction reduction; CR; forward conjunction reduction connectivity 9, 15, 25, 31, 35–8, 77, 86–9, 112–13, 128, 131, 183, 185, 200–3, 390, 409, 413, 424, 459, 467, 503, 605, 608–15, 617, 620, 623, 742–3, 843–845, 928–9 Constituent Raising 869 constraint 8, 10–11, 19, 26, 29, 31, 36–8, 55–7, 62, 76–7, 79–80, 86, 89–91, 93–4, 96, 100–1, 103, 108, 113, 116, 131, 140, 153, 163, 170, 179, 183, 188, 199, 202–4, 208–10, 213, 216, 219, 223–4, 227, 229–32, 251–2, 254, 258, 265–6, 274, 282, 289, 306, 311–12, 314–16, 318–21, 323, 326, 331, 333, 336, 338–40, 343–5, 347, 351, 370, 373, 381, 383, 386, 389, 392–4, 398–9, 402, 404–5, 410– 14, 417, 419–20, 424, 426–7, 431, 438, 440–1, 448, 457, 459, 461–3, 468–9, 471, 474, 483, 492, Page 3 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index 496, 500, 540, 569, 573, 579–82, 586, 590, 598, 603, 616, 627, 640, 642, 652–3, 663, 691, 713, 727, 731, 752, 773–4, 853, 857, 859, 861, 877, 880, 933, 939, 943–6, 957–8, 963–5, 971, 973, 975, 979, 992, 997, 1001, 1018, 1024–6, 1029–30 constraint-based 76, 462 constraint-free 11, 459 constructed action 768–9, 773–6, 782–3 Construction Grammar 3, 77, 162, 169, 510 context-dependence 12, 125, 206, 213 contrast contrast sluicing 28, 390, 408–9, 414, 417–19, 872 contrastive ellipsis 360, 363–7, 373, 381, 384 contrastive focus 82, 361–3, 365, 371, 373, 379, 381, 383, 505, 517–18, 574, 595, 598, 706, 817–18, 822, 837, 964–5, 1008–9 contrastive remnant condition 360, 363–4, 374, 376, 386 contrastive stress 170, 172, 180, 182, 187, 189, 193, 817 contrastive remnant 360, 362–4, 374, 376, 386, 407, 549, 786, 964–5 contrastive topic 195, 268, 326–7, 368, 371–2, 376, 379, 817, 821–2, 834–5, 919, 1002–3 conversational dialogue 211, 232 coordinate ellipses 683, 698, 785, 798, 804–5, 814 (p. 1117) coordination 13, 139, 143–4, 167, 193, 204–5, 212–13, 228–9, 232, 256, 277, 294, 432–3, 435, 441, 447, 452–3, 483, 497, 563, 576–7, 581–4, 586, 588–9, 591, 598, 602, 638, 642–6, 681– 97, 699–702, 704, 706, 708, 714, 716–18, 726, 733, 750, 752–4, 768, 773, 779, 781–2, 797–804, 850, 853, 857, 863, 913, 916, 920–1, 934, 936, 938, 940–1, 946, 961, 980, 984, 991, 997–8, 1031 coordination ellipsis 683, 718 copula drop 11–12, 19, 482, 832 copular clause 482–3, 485, 489, 502, 860, 973 copy 9, 23, 38, 133, 181, 183, 192, 203–4, 218–19, 230, 279, 292, 392, 405, 410, 420, 431, 435, 437, 449–50, 455, 465, 474, 511, 627, 629, 631–2, 634, 636, 655, 661, 788, 795, 834, 882, 884, 1021–2, 1027 copy theory of movement 133, 882 coreference 41, 331–2, 334, 345, 349, 629, 652, 654–5, 707, 868, 876, 929 Coreference Rule 331–2, 334 correlate 27–8, 32, 36, 81, 113, 247, 255, 290, 359–61, 365, 378, 381, 397–8, 401, 408–9, 413, 417, 468, 479, 484, 500, 502, 553, 570, 618, 647, 650–4, 656, 690, 698, 788, 790, 792–3, 795–6, 800, 807, 822, 835, 842–3, 848, 857, 866, 869–72, 874, 883–4, 925–6, 956–8, 963–5, 970, 973, 999, 1004 CR 13, 84, 681–4, 686, 688–92, 694–7, 700–2, 706–7, 716, 718, 853–4, 857, 863, 869–70; see also backward conjunction reduction; conjunction reduction; forward conjunction reduction cross-modal priming 437, 445 cumulative agreement 947, 983 D-linked 63–4, 66, 246, 254, 549, 792 deaccentuation 40, 125, 315, 317–18, 333, 335–6, 339, 351, 360, 363, 366, 368–70, 372–3, 377, 381, 568, 717 default case 607, 612, 616, 648, 755, 966 definite article 747, 761–2 definite constituent ellipsis 658–9 definite correlate 397, 957 definite ellipsis 658, 675, 812 Page 4 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index definite null complement 658, 663–4, 675 definiteness 280, 299, 313, 535, 815, 919–20 degree 3, 9, 33, 55, 174, 256, 271, 291, 302, 322–3, 328, 445, 482, 487, 505, 540, 542, 625–38, 642, 645–6, 648–52, 654–6, 766, 781, 785, 791–2, 794, 796, 808, 838, 1019 Degree Phrase 33, 302, 791, 838 degree quantifier 627, 630, 633, 646, 651 deontic 508, 524 Dependency Grammar 3, 795; see also DG DG 143, 146–7, 149, 152–4, 159–61; see also Dependency Grammar DGB 104, 108, 111–13, 118–20, 970; see also Dialogue Gameboard Dialogue Gameboard 104, 970; see also DGB direct compositionality 122, 125, 140–1, 620 direct interpretation approach 963, 965–8, 970, 987 discourse 2–3, 5, 8, 10, 19, 38, 40, 77, 81, 83, 88, 90–7, 101, 117, 123, 125–7, 129–30, 132–3, 140, 170, 178, 183, 188, 193, 196, 225, 234, 236, 238, 240–1, 247, 250, 252, 262, 268–70, 272, 275, 284, 290, 294, 297, 302, 307, 309, 311–12, 315–20, 323–4, 331, 333, 336–40, 351, 353–5, 361–2, 366–7, 372–3, 376, 380, 382–3, 386, 458, 460, 483, 509, 543–4, 574, 599–602, 604, 607–8, 610, 614, 636, 659–61, 728, 756, 769–71, 774, 776, 780–4, 800, 814–15, 817, 822, 834, 836, 877, 880, 886–9, 931, 966, 970, 973, 1013, 1029 discourse deletion 877, 880 discourse drop 1013 discourse orientation 771, 784 discourse relations 170, 353–4, 599–601, 800, 814 disjunction 109, 202, 234–7, 239, 246, 252, 576, 582, 588–9, 602, 685 Distributed Morphology 471 do-support 70, 703, 711, 886, 1021–3, 1036 Downward Bounding Constraint 579 Dynamic Semantics 104, 238, 241, 252 E-feature 37, 279, 289, 303, 311, 368, 371, 373, 550, 696, 698, 941, 955 e-GIVENness 243–5, 370–1, 614 Economy 86, 188, 228, 257, 274, 290, 313, 383 ellipsis resolution 19, 75, 77, 83, 97, 103, 105, 343, 352, 414, 453, 456, 880 embedded language 462 embedded question 469, 737 emphatic do 711 empty noun 529–30, 532, 540–1, 546, 550, 556, 560 entailment 94, 243–7, 249–51, 370–1, 614, 953 epistemic 245, 271, 508, 522, 788, 790, 845–7 EPP 821; see also Extended Projection Principle equation 169, 332, 346–9 Evaluation Metric 312 exophoric ellipsis 96 Extended Projection Principle 163; see also EPP extraction 32, 51–2, 134, 254, 262, 293, 295, 358–9, 361, 363, 367, 381–6, 389–90, 393, 404–5, 412, 419, 422–3, 430, 484, 486, 488, 490, 511, 515, 521, 524–5, 531, 535–7, 543, 556–7, 636, 643, 648, 664–5, 667–9, 682, 692, 700, 787, 800, 808–9, 822, 827–8, 839–41, 845, 855, 857, 859, 963– 4, 981, 1012, 1033–7 extraposition 82, 645–6, 690–1, 702, 848, 861 Page 5 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index feature checking 461 focus FEC 104–5, 112–21; see also Focus Establishing Constituent Focus Establishing Constituent 105; see also FEC focus fronting 496, 833, 954–7 focus intonation 277–8 focus movement 417–19, 482–3, 490–1, 518, 572, 593–4, 596, 700, 833–4, 850, 862, 968 (p. 1118) focus projection 260, 518, 945, 954, 958, 993, 996 focus-verb adjacency 834, 836 forward conjunction reduction 681, 683, 689–90, 695–7, 700–2, 706–7, 716, 718; see also back­ ward conjunction reduction; conjunction reduction; CR fragments case-marked fragment 7, 866, 899 contrastive fragment answers 618, 965 declarative fragments 96, 112, 114 non-case-marked fragment 7, 866, 899 full reconstruction 277, 283, 295, 311 function composition 134–5, 693–4 g rule 136–8, 140–1 gapping 6, 13, 15, 20, 27, 32–3, 35, 42, 54–5, 80–1, 86, 90, 98, 140, 143–5, 147–9, 151–3, 155–6, 160, 165, 171–3, 176–8, 181–2, 189–90, 192–5, 198, 202–4, 256–7, 277, 320, 360–8, 374–5, 377, 381, 437, 443, 447, 452–3, 505, 514, 535, 562–5, 567, 572–5, 578–80, 583, 586–604, 624, 637–40, 642, 645–6, 653, 660, 683, 691–2, 694, 696, 699, 703, 721–8, 746, 753–4, 764–5, 776–9, 781, 783, 797–800, 803–7, 830, 835–6, 838, 841, 848–9, 857, 862–3, 865, 867, 882–5, 900, 903, 916, 918–22, 933–43, 961, 981–3, 988, 991, 1002, 1009–13 gender 136, 224, 263, 265, 273, 276, 303–4, 411, 434, 450, 475–6, 529, 538–40, 544–5, 547, 549, 553–5, 623, 747, 762, 813, 989 gender mismatch effect 263, 450 general inference 278, 280–3, 294, 309 gesture 7, 196, 765–6, 782–4 givenness 97, 243–5, 360, 362–4, 369–4, 380, 386, 614, 817 government 29, 73, 78, 151, 294, 315, 461–2, 548–51, 553, 559, 573, 693, 703, 875, 998, 1037 grounding 77, 105–6 Head-Complement Rule 98 Head-Driven Phrase Structure Grammar 3, 75–6, 510; see also HPSG Heavy NP shift 495–6, 517–18, 569–70, 593–4, 702 Higher Operator Phrase 994; see also HOP higher-order matching 348 higher-order unification 83, 89, 105, 222, 342, 351, 355 HOP 994–5; see also Higher Operator Phrase HPSG 75–8, 80–3, 86, 88, 91, 93, 97–107, 209; see also Head-Driven Phrase Structure Grammar identity condition 37, 40, 58, 60, 62–8, 74, 133, 244, 390, 400–1, 410, 419, 468, 501, 516, 529, 543–4, 546–7, 560, 632, 634, 637, 696, 709, 712, 958 implicit arguments 41, 247, 675 implicit correlate 32; see also antecedentless; non-linguistic antecedent; pragmatic antecedent; pragmatic control implicit questions 601, 613–14 implicit structure 343 Page 6 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index in-situ analysis 879–80 incrementality 11, 77, 92, 97, 114, 206–7, 212–13, 215, 218–19, 224–7, 229, 430, 451–6 indefinite ellipsis 675, 812 indefinite null complement 663–4 Indirect Licensing 23, 184–5, 187 inference 85, 93, 103, 106, 164, 207, 224, 250, 266, 274, 278, 280–4, 286, 294, 296, 309–13, 465, 509, 606, 642 infinitival to 507–8 infinitive 30, 61, 343, 506–8, 515, 522–3, 526, 564, 612, 723–4, 729–30, 737, 742, 788–9, 794, 824, 829, 840, 907, 923–4, 931, 1000, 1030 inflectional mismatches 947 innateness 277, 313, 437–8, 440 inquisitive semantics 3, 233–40, 243–7, 249, 251–2 interaction 10–11, 77–8, 83–4, 90, 103–7, 109–10, 112, 115, 205–6, 212, 214, 219, 225, 228–9, 232, 242, 254, 259, 261, 269, 273, 297, 311, 315, 327, 386, 389, 394, 396, 420, 423, 529, 537, 621, 690, 692, 697, 701, 704–5, 714, 989 interactive coordination 213 interrogative 23, 25, 49, 59, 94, 96, 113, 145–6, 157, 160, 170–1, 233–4, 239, 242, 245, 249, 251, 262–3, 267, 390–1, 465, 470, 479, 569, 598, 695, 723, 768, 779, 792–3, 841, 930, 955, 958, 962–3, 967–9, 973–5, 978, 993 intonation 112, 128, 189, 193, 277–8, 357, 369, 379, 568, 694, 717, 728, 734 intonational contour 358, 361 intonational phrase 358 inverse scope 438, 440, 443, 664–6, 677, 679, 1035 invisible syntactic structure 163–4, 182; see also unpronounced syntactic structure islands Adjunct Island 262, 448–9, 744, 756, 860, 963, 1018 Complex NP Constraint 208, 393, 398, 410, 652, 939, 943–4, 957–8, 1018 Coordinate Structure Constraint 580–2, 627, 642, 691, 727, 939, 943; see also CSC CSC 627, 636, 642–4, 691–3, 701, 717, 1012; see also Coordinate Structure Constraint island constraint 10, 26, 163, 202, 208, 219, 230, 232, 254, 258, 389, 393, 398, 414, 417, 420, 424, 448, 652, 752, 946, 963–4, 971, 973, 1018 island effect 5, 10, 27, 31–2, 34, 37, 389, 394–5, 398–9, 401, 404–5, 414, 420, 473, 481, 484, 498, 569, 573, 615, 617–18, 844, 859–60, 884–5, 965, 1018 island evasion 409–14, 419–20, 424 island repair 54–6, 58, 74, 385, 399–400, 406, 410–12, 414, 418, 473, 484, 486, 498, 648, 880 island violation 51, 53–6, 89–90, 262, 274, 316, 385, 390, 394, 400, 403, 406, 410–11, 413–15, 417, 420, 448–9, 481, 484, 486, 568, 618, 623, 648, 652, 699, 842, 845, 859, 884, 958, 965, 968, 971 LBE 484–6, 502; see also Left-Branch Condition (p. 1119) Left-Branch Condition 393–4, 399, 411–12, 845, 859; see also LBE PF islands 56, 859 Subject Condition 393, 399, 404–5, 880 weak islands 394, 398, 739 isomorphic reconstruction 300 Kase 68–9, 72 KoS 78, 83, 94, 104, 106, 111, 117, 120 lambda abstraction 135, 166–7, 170, 290, 346–9 Page 7 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index Late Merge 627, 629–30, 633, 635–7, 646 Law of Coordination of Likes 687; see also LCL LCL 687–90, 694, 698; see also Law of Coordination of Likes learnability 281, 294, 300, 303, 309, 647 LEE 101, 149, 152, 160, 270, 322, 448–9, 457; see also left-edge ellipsis left periphery 81, 365, 373, 619, 743, 756, 797, 799, 822, 830–1, 833, 836, 838, 850, 862, 944, 993–4, 996 left-edge ellipsis 147, 149, 152; see also LEE lexical verb complement ellipsis 658, 666–7 LF Pied-Piping 877 LF-copy 23, 38, 47, 52–4, 58, 74, 448, 466 LF-extraction 293 LF-projection 302, 313 licensing 3–5, 7–8, 10, 22–3, 25, 37–8, 47, 68–9, 72–4, 78, 97–8, 118, 128, 153, 160, 165, 171, 184– 8, 206–10, 213–14, 223–4, 227–30, 232, 237, 241, 244, 253, 269, 279, 287–8, 302–4, 306, 315–16, 321, 342, 359–60, 364, 366–9, 371, 373, 381, 385–6, 401, 428, 468, 473, 498, 506–9, 514, 524, 529, 531, 539, 548–1, 557, 559–60, 601, 614, 625, 636, 711, 747–8, 774, 790, 800, 803, 813–14, 818, 824–5, 830–31, 840, 845, 847, 857, 988, 991, 993, 1003, 1009, 1020–1, 1024–7, 1029–30, 1035, 1038 lift 134–5, 137, 139, 291 linearization 645–6, 654, 703, 705, 713 Logical Form 46, 123, 141, 208, 239, 290, 315, 329, 360, 616 low coordination 256, 583–4, 586, 588–9, 692, 940–1 main assertion principle 271 matrix language 462 Matrix Language Framework 462 MCE 505–6, 522–5, 677, 680, 787–8, 790, 846–7; see also Modal Complement Ellipsis MD 701, 712–15, 718; see also multidominance Meaning-Text Theory 145 Merge 210, 217–18, 238–9, 245, 247–8, 251, 397–9, 407–9, 417–18, 447, 464, 607, 627, 629–30, 633, 635–7, 646, 693, 713, 925 mime 7, 765–6, 769, 782–4 Minimalist framework 463, 512, 698 mismatch 41–2, 68, 74, 92–3, 129, 133, 258–60, 263, 265–9, 272–4, 316–17, 321–8, 356, 374, 378– 9, 415, 419, 431–5, 440, 450, 467, 469, 516, 545, 705, 707, 736–7, 967, 980 mixed reading 335, 350 modal 14, 67, 70, 100, 199, 214, 216, 272–3, 328, 356, 372, 420–3, 436–7, 445, 484, 505–6, 508, 512, 515, 522, 524–5, 564, 582, 591–2, 596, 677, 730, 777, 785–9, 808, 843, 846–7, 952, 1012, 1020, 1026–9 Modal Complement Ellipsis 420–3, 505, 522, 525, 677, 730, 787, 846; see also MCE modal ellipsis 522, 524, 787, 808 move-and-elide 695, 697–700, 718 movement-then-deletion 611, 618, 623 multidominance 646, 684, 691, 712, 946–8, 983; see also MD multiple cleft 483, 493–5 multiple wh-movement 833, 994, 1000 N’-deletion 865–6, 894–9 N-deletion 900, 910, 933 Page 8 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index NCA 13, 79, 98, 279–80, 300–1, 312, 644, 657, 660–9, 675–80, 755–7, 787–8, 808, 839, 853–6, 864, 923; see also Null Complement Anaphora negation 67, 72, 111, 147–9, 197, 236–7, 240–1, 246–8, 252, 325, 367, 415–16, 426, 440, 508, 576, 582, 597, 656, 685, 690, 695, 697, 728, 744, 786, 789, 802, 842, 847, 902, 940, 972, 1005, 1011– 12, 1021, 1024–5, 1034–5 negative polarity item 129, 803, 892 nominal ellipsis 5, 12–13, 15, 20, 31, 78, 86, 526, 528–9, 538–9, 549, 551, 559–61, 721, 745–8, 758, 760, 764, 814–16, 818, 841, 1020 nominalization 324, 340, 541–2 non-actuality implicature 340 non-constituent coordination 13, 798, 853, 857, 984 non-constituent deletion 481, 836, 879 non-movement account 365 non-sentential utterances 76–8, 80, 96, 168; see also NSUs NP-ellipsis 20, 69, 158, 302, 371, 457, 465, 475, 529–30, 548, 550–3, 556, 560, 785, 817, 988–9; see also NPE NPE 13, 158, 280, 285, 289, 301–2, 305–6, 312–13, 358, 529–31, 534–40, 542–4, 546–50, 552–5, 559–60, 683, 695, 706, 709–11, 717–18, 747–8, 761, 763, 810–14, 986, 988–91; see also NP-ellip­ sis NSUs 76, 80–1, 84, 86, 88, 96, 111; see also non-sentential utterances null argument 26, 62, 560–1, 659, 809, 854, 865, 874, 886–9, 891–4, 899, 936, 949–51, 953 Null Complement Anaphora 6, 25, 79, 127, 149, 269, 279, 301, 657–61, 677, 721, 755–6, 764, 787, 839, 841, 900, 923, 936, 986; see also NCA Null Complement Rule 661–2 null elements 24, 38, 144, 658, 676, 680, 705, 715, 885, 887 null object construction 519, 885–9 null subject 561, 853, 890, 898 (p. 1120) number 3, 12, 15–16, 26, 31, 40, 48, 61, 71, 76, 83–4, 89, 98, 105, 111, 113, 128–9, 143, 146, 149–50, 152, 154, 165, 173–4, 199, 225, 232, 234, 247, 250, 261, 265, 267, 276, 282, 302, 308, 318, 327, 329, 335, 349, 358, 394, 406, 411, 413, 417–18, 421, 427, 434, 441, 464, 479, 483, 487, 494, 499–500, 502–3, 511, 529, 533, 539, 543, 545, 549, 551–5, 595, 604, 656, 658, 695, 698, 722, 728, 737, 751–3, 762, 797, 813, 816, 818, 825, 836, 858, 879–80, 901, 947, 966, 980, 989, 1010, 1021, 1028, 1031, 1037 object drop 71, 519, 900, 905 object shift 518, 569–70, 593–4 Occam’s Razor 86, 162, 164, 191, 196, 268, 274 orphan 23, 90, 185 overt correlate 247, 397, 468, 790, 842, 925–6, 999 Parallel Architecture 162, 186 parallelism 19, 21, 29, 39–40, 81, 84–5, 88, 103, 105, 112, 115–16, 167, 172, 175, 180, 213, 227–8, 254–5, 257, 259, 267, 281, 294, 296, 300, 331–5, 338–9, 354, 366–8, 374, 376, 381–2, 386, 411, 414, 416–17, 426–8, 430–2, 435, 437–44, 447–8, 485, 492, 499, 501–2, 578, 645, 659, 661, 679, 732, 738, 773, 778–9, 781, 790, 835, 964–5, 968, 978, 995–7, 1002 parameter identification 110, 112, 115–16 parasitic gap 25–7, 51–2, 498, 642, 716 parasitic scope 651–3 parataxis 294 parsing 5, 11, 105, 212–14, 218, 220–1, 224–6, 228–9, 266, 274, 280, 294, 355, 414, 438, 456 Page 9 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index partitivity 280, 302, 305–8, 313, 549, 793, 811–13 passive 42–3, 62–3, 129, 151, 195, 199, 259, 266–7, 282, 316–17, 324–7, 378–9, 421, 511–12, 524, 526, 645, 677, 692, 711, 731, 767, 785, 789, 798, 842, 845, 890, 901, 1022–3, 1027, 1032 PCC 683, 691, 693–7, 700; see also Phrase Cluster Coordination perfect 28–9, 59, 130–1, 199, 282–3, 309, 312, 323, 439, 512–13, 554, 711, 786, 792, 937, 1028, 1030–1 PF-deletion 76, 82, 243, 251, 369, 428, 529, 658, 962–3, 965–70, 974–5, 979, 987 phrasal ellipsis 988 Phrase Cluster Coordination 683, 691; see also PCC Pied-Piping 258, 260, 293, 393, 725, 877 pitch accent 255–6, 358, 360–3, 365–6, 368–9, 372–3, 375–6, 378–9, 381, 826, 956 pitch extraction contour 361, 363, 367, 382, 385 pointer 37, 188, 191–2, 196, 203, 214–15 polarity 15, 41–2, 93, 129, 174, 199, 211, 253, 273, 372, 382, 697, 722, 728, 760, 800–3, 823–6, 849–50, 892, 979, 988, 1000–3, 1006, 1009 polarity ellipsis 15, 800–2, 979, 988, 1000–3, 1006, 1009 possessive 87, 291–2, 302–3, 335–6, 428, 430, 549, 577, 760–4, 772, 809–10, 815, 819, 895, 901– 2, 989–90, 1022, 1027, 1029 possessor 710, 717, 746, 760, 762, 818–20, 844, 857, 960 possessor doubling 760, 762 poverty of the stimulus 313 pragmatic control 521; see also antecedentless; implicit correlate; non-linguistic antecedent; pragmatic antecedent pragmatics 62, 77, 97, 101, 206, 238, 277–8, 308, 663, 669, 676, 680 predicate ellipsis 5, 13–15, 76, 504–6, 511, 514, 520, 522, 524–5, 721, 730, 764, 777, 786, 788, 790, 820–4, 827, 838, 847, 900, 931, 933, 962, 979, 988, 992, 1013 preposition omission 7, 622, 861–3, 974, 976, 978–9 preposition stranding 7, 15, 25, 28–9, 34–7, 49–50, 53, 57, 62, 88, 133–4, 260, 265, 293, 311, 402, 468, 470, 496, 499–500, 502, 570, 573–4, 621–2, 648, 699, 703, 705, 738, 741, 743, 792–3, 795, 841, 843, 857–62, 929–30, 974–5, 978–9, 987 Preposition Stranding Generalization 28, 34, 36, 53, 133, 499, 622, 792, 795, 841, 843, 974; see also PSG presentational focus 1007–8 presupposition inheritance 612, 614, 618 preverb 15, 820, 824–7, 840 preverb stranding 15, 820, 825–7, 840 preverbal particle 815, 821, 825, 831 Primary Linguistic Data 312 priming 25, 192, 255, 259, 263–5, 436–7, 445–6 Principle C 629–30, 636, 652, 655, 929; see also Condition C Principles and Parameters 163 pro 9, 12, 22, 32–3, 35, 38–9, 70, 75, 79, 100, 173–4, 186, 210, 255, 278–9, 283–4, 287, 294–302, 310, 312–13, 370, 428, 482, 548–9, 553, 630, 658, 660–1, 667, 675–7, 679, 723, 746–8, 765, 769, 785–7, 800–1, 808, 812, 816, 823, 832, 839, 853–4, 868–71, 877–8, 891, 929, 986, 1011, 1014 pro-form 9, 75, 137–8, 145, 210, 315, 381, 479, 503, 524, 528–9, 556–7, 559, 561, 658, 660–1, 663, 667, 675–7, 679, 746–8, 785–7, 800–1, 808, 812, 818–19, 855–6, 864, 894, 986, 989, 1008, 1033, 1035

Page 10 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index processing 5, 11, 77, 93, 97, 106, 114, 130, 132, 163, 192, 206, 210, 213–14, 216, 218–19, 221–32, 240, 253–5, 257–8, 261–5, 268–9, 272, 274, 315, 324, 327, 340–1, 350, 352, 356, 375, 414, 428–9, 442, 444–51, 454–7, 554, 620, 965, 969–70, 978–9, 987 progressive 40, 289, 372, 512–13, 711, 842, 847, 904, 947, 1029–30, 1032 Projection Principle 163 pronominal possessor 760 pronoun 25, 30, 34, 40–2, 59, 66, 87–8, 91, 94, 101–2, 126, 129, 131–2, 135–7, 140, 146, 148, 152, 178–80, 183, 185, 191, 201, 211, 216, 218–20, 223, 227, 230–1, 241, 258, 264, 266, 283–4, 287–8, 292, 305, 314–15, 317–19, 329–36, 338–40, 343–4, 348–52, 354, 370, 377, 379–80, 390–1, 401, 418, 427–8, 430–1, 439, 445, 457, 460, 463, 500–1, 517, 527, 529, 531, 539, 546, 560, 569, 577, 582, 598, 603, 607, 610–11, 629–30, 654, 659–61, 663, 675, 696, 707, 710, 717, 735, 737–8, 741– 3, 746, 748, 750, 757, 759, 761–4, 766, 769, 772, 785, 787, 794, 801, 806–8, 810–11, 814–15, 819, 832, (p. 1121) 838–9, 851–4, 857, 868, 877, 885, 891–2, 901, 907, 929, 986, 989–90, 1001, 1016, 1037 pronoun resolution 343, 354 Pronoun Rule 348–50 proposition 42, 80, 91, 94–6, 104–7, 109, 111, 113, 115, 121, 123, 127–8, 132, 136, 138, 189, 191, 194, 206, 232–5, 247, 314, 321, 327, 334, 339, 446, 575–6, 600, 603, 605–6, 609–10, 613, 616–17, 626, 628, 662, 669, 678–9, 685, 854, 966 prosodic boundaries 358–9, 361, 374 prosodic contour 358 prosodic licensing 360, 364, 367–9 prosodic phrasing 357–8, 360, 368, 375–6 prosody 3, 256, 357–2, 365–6, 368–9, 371, 373–6, 379–81, 383–6, 706, 768, 840, 914, 919, 921, 923, 933, 1030, 1038 pseudocleft 35, 185, 483, 488, 494, 612, 874 pseudogapping 14–15, 20, 35, 42, 86, 101, 122, 126, 134, 139–41, 147–9, 153, 155–7, 160, 171, 176–8, 193, 344–5, 365, 381, 500, 504–6, 511, 514–18, 524–5, 535, 537, 556, 564, 569–70, 573, 579, 584, 586–9, 592, 596–7, 600, 624, 637, 646, 698, 705, 721, 730, 764, 786, 820, 822–3, 827–9, 841, 847–8, 857, 861–3, 865, 886, 916, 931–3, 949, 979, 1033 PSG 499, 792–3, 797, 841, 843, 857–60, 863, 974–5, 977, 979; see also Preposition Stranding Generalization psycholinguistics 3, 11, 104, 191–2, 203, 253–4, 258, 265, 269, 273, 374, 376–7, 425, 444–5, 456, 700, 716, 986 punctuated paths 394, 396, 423 QR 65, 138, 293–4, 565, 577, 627–30, 634–5, 646, 653, 809, 1033–6; see also Quantifier Raising Qu-Ans 127–8, 133–4, 613, 620 quantifier 65, 69, 85, 134, 138, 211, 259, 291–3, 306, 331, 409, 418, 565, 577, 627, 629–30, 633, 646, 651, 665, 701, 708, 713, 718, 746, 749, 751–2, 787, 806–7, 809, 834, 853, 902, 985, 1035 Quantifier Raising 65, 138, 259, 292–3, 418, 565, 577, 627, 713, 985; see also QR QUD 8, 11, 16, 77, 88–9, 95, 97, 104–5, 109–21, 238, 252, 270, 272–3, 280, 311–12, 324–6, 336–7, 339–40, 601–2, 614; see also Question Under Discussion Question Under Discussion 16, 77, 94, 97, 270, 272, 274, 311–12, 320, 324, 601–4, 614, 929; see also QUD reaction time 430, 436 record 77, 102, 105–11, 113, 209, 213–14, 219–20, 247, 755, 970 record types 77, 106–9

Page 11 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index recoverability 3–4, 7–10, 39, 45–7, 58–60, 74, 78, 91–2, 96, 147, 149–52, 160, 165, 188–9, 201, 206–7, 210, 214, 216, 220, 228, 232, 251, 253, 265, 273, 279, 290, 312, 315, 342–3, 359–60, 374, 410, 428, 506, 509–10, 514, 529, 539, 543, 546–7 Recycling Hypothesis 324 Reduction Analysis 755, 985 relational adjective 947 relative clause 32, 91, 135, 138, 199, 208, 230–1, 247, 272, 344, 384–6, 409, 429, 598, 620, 634, 652, 704, 716, 757, 794, 832, 842, 845, 856–7, 860, 884, 957 repair 37, 51, 53–6, 58, 74, 77, 103, 106–7, 112, 117, 228–9, 240, 254, 262, 265, 267–8, 274, 280, 282–3, 322–4, 326, 381, 385, 399–400, 406, 410–12, 414, 418, 472–3, 484–6, 498, 500, 648, 793, 795, 833–4, 836, 859, 864, 880, 913–14, 957, 965, 978, 987 replacives 254–6, 270 reprise 83–6, 103, 115–16, 118, 225, 352–3, 356, 977 restructuring 667, 685 resumptive pronoun 418 Right-Node Raising non-coordinate RNR 683, 701, 716–17 RER 702–3, 705, 713, 717; see also Right-Edge Condition; Right-Edge Restriction Right-Edge Condition 914; see also RER; Right-Edge Restriction Right-Edge Restriction 702; see also RER; Right-Edge Condition Right-Node Raising 6, 13, 76, 81, 365, 563, 624, 640, 681–2, 691, 721, 733, 764, 837, 841, 852, 865, 881–3, 885, 899–900, 904, 912, 933–4, 946–8, 983, 988, 1011, 1015, 1017–18; see also non-coordinate RNR; RNR Right-Node Wrap 717–18 RNR 13, 81–2, 86, 98, 365, 367–9, 624, 637–8, 640–2, 645–6, 682–4, 691–4, 700–13, 715–18, 798–9, 837, 852–3, 857, 863, 912–15, 983–4, 1019; see also Right-Node Raising root 111, 145, 156, 215–16, 220, 226, 291, 301, 323, 409, 471, 473, 506, 520, 522, 529–31, 534, 537, 568–9, 583, 601–2, 688–9, 790, 792, 794–5, 846–7, 904, 906–7, 909, 916, 919, 927, 946, 962– 3 Saxon genitive 760–1, 763 scope 13, 16, 76, 174, 202, 215, 234, 236–7, 239–40, 245, 248, 250, 293–4, 371, 374, 381, 408–9, 413, 415–18, 426, 436, 438, 440, 443, 465, 492, 576–7, 581–2, 588, 592, 596–7, 603, 629–30, 633, 635–6, 651–3, 655–6, 664–6, 677, 679, 685–6, 690, 695, 701, 708, 713–14, 718, 768, 773, 786–7, 809, 853, 940, 964–5, 1011–12, 1033–6 scrambling 216, 422, 482, 490–2, 518, 593, 862, 865, 891, 954, 996 script 96, 610, 836 self-paced reading 263, 266, 271, 327, 341 semantic identity 40–1, 43, 58, 63, 65–8, 74, 125, 133, 265–6, 268–9, 274, 280, 315, 400, 410, 416, 427, 468, 482, 546, 634, 696 sentence processing 11, 262, 274, 444–5, 449, 456 short answer 30, 80, 87–9, 103, 112, 119–20, 128, 223–4, 564, 605, 865–6, 876–81, 898 short do replies 758–9 sign language 6–7, 13, 765, 769, 776–7, 780–2 Simpler Syntax Hypothesis 162; see also SSH (p. 1122) sloppy identity 40–1, 178, 194, 207, 223, 264, 280, 284, 290–1, 295, 298–9, 313, 329–36, 338, 340, 347, 349–50, 352, 367, 370, 379–80, 428, 430, 560–1, 572, 636, 778, 823, 828, 854–6, 865–73, 875–6, 881, 886–94, 896–9, 950, 1004, 1014; see also sloppy-like reading; strict identity; strict-sloppy identity Page 12 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index sloppy-like reading 871, 888–9, 897–8; see also sloppy identity; strict identity; strict-sloppy identi­ ty Slot Grammar 343 sluicing backward sluicing 454–6, 709 bare sluice 352–3, 355, 791–7 case-marked sluicing 871–4, 899 contrast sluicing 28, 390, 408–9, 414, 417–19, 872 embedded sluicing 502, 874 isomorphic sluicing 484, 486, 488–90, 495 matrix sluicing 874, 996 multiple coordinated sluicing 992, 997 multiple sluicing 15, 54–8, 74, 155, 157–8, 390, 407–9, 412–14, 419, 491–6, 498, 833, 842, 871, 967–9, 976, 992, 996, 1000 non-isomorphic sluicing 484, 486, 488–9, 495 pronominal sluicing 874 pseudosluicing 35, 482, 497, 500 spading 15, 25, 41, 741, 758–9, 791 sprouting 15, 42, 62–3, 68, 74, 171, 238–9, 244–5, 247–51, 262, 397, 409, 447–9, 468, 484, 741–2, 790, 795, 925, 965, 971, 978, 988, 999–1000, 1006 swiping 15, 293, 484, 709–10, 741–2, 791, 927, 974 split questions 972, 975 spontaneous production 439 SSH 162–5, 174, 177; see also Simpler Syntax Hypothesis stress 166, 170, 172, 180, 182, 187, 189, 193–4, 252, 402, 431, 568, 722, 817, 824, 827, 832, 903, 928, 933, 986, 1035 strict identity 207, 263–4, 292, 295, 316–18, 329–31, 334–5, 338, 340, 350, 367, 379, 560–1, 636, 828, 854–5, 868, 876, 890, 892–3, 896, 950; see also sloppy identity; sloppy-like reading; strictsloppy identity Strict Identity Hypothesis 290, 299, 313 strict interfaces 277–8, 280, 282–3, 290–1, 294, 312 strict-sloppy identity 41, 333–4, 340, 352, 356, 950; see also sloppy identity; sloppy-like reading; strict identity string deletion 882–5, 899 string-vacuous movement 884 stripping BAE 163, 165, 168, 170–3, 178, 183–4, 186; see also Bare Argument Ellipsis Bare Argument Ellipsis 20, 27, 76, 83, 90, 98, 163, 165, 168, 562, 690, 727, 973; see also BAE case-marked stripping 867–71, 873–4, 896, 899 multiple stripping 867 non-case-marked stripping 867–70, 873, 899 subjacency 316, 865–7, 870–2, 875–82, 884–5, 898–9 subject drop 482, 999 subordination 452–3, 625, 645, 768, 779, 991 subsentential 126, 209, 224, 321, 605, 684 subtract 169–70, 173, 178, 187 successive cyclicity 394 summative agreement 708, 714 Page 13 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index Superiority 396, 492, 968, 993–7 symmetric entailment 244–7, 250–1 syntactic complexity 429–0 syntactic identity 40, 42–4, 58, 62–5, 67–8, 74, 82, 92, 96, 163, 173, 175, 265, 334, 378, 398–400, 408, 410, 412–13, 415–20, 424, 516, 546, 967 syntactic matching 258, 265, 267–9, 274 syntactic mismatch 92–3, 258–60, 268–9, 274, 316, 323, 415 tag questions 65, 152, 1030 TDH 429–31, 433, 436–7; see also Trace-Deletion Hypothesis Theta Criterion 163 topic 2–5, 10, 26, 33, 47, 81, 84, 195, 218, 249, 265, 268, 277, 295, 297, 306, 324, 326–7, 337, 340, 342, 345, 352, 356, 368, 371–2, 376, 379, 381–2, 392, 467, 474, 504, 514, 520, 526, 539, 574, 611, 625, 756, 758, 768, 774–5, 817, 821–2, 827, 834–5, 866, 885, 919, 933, 940, 955, 1002–3 topicalization 25–6, 30, 184–5, 277, 284, 316, 366, 382, 384–5, 521, 664–5, 679, 788, 790, 1002, 1009, 1033–4, 1036–7 TP ellipsis 596, 617, 808, 829–32, 834–5, 850, 862–3, 993, 1002–3, 1006 Trace-Deletion Hypothesis 429; see also TDH transfer 82, 143 trigger 31, 37, 88, 91–2, 215, 352, 476, 577, 617, 697, 711, 755, 806, 817, 831, 901, 906, 912–13, 941 TTR 77, 106–7, 109, 111, 209; see also Type Theory with Records Type Theory with Records 77, 102, 106, 209; see also TTR UCC 688–9, 699; see also Unlike Category Coordination unification 83, 89, 105–6, 116, 218, 222, 342–3, 345–7, 351, 355 Unlike Category Coordination 688; see also Law of Coordination of Likes; UCC unpronounced syntactic structure 23, 41, 76, 78, 86, 88–9, 91, 201, 203, 315–16, 321, 390, 398; see also invisible syntactic structure Upward Bounding Constraint 590 variable 23, 66, 107, 114, 127, 131, 135–6, 139–40, 221, 230–1, 239, 262, 267, 278, 284, 290–2, 296, 298, 329–31, 346–9, 400–2, 405–6, 408–9, 414, 419–20, 424, 427–8, 515, 571, 577, 581–3, 597, 626–7, 631–2, 636, 656, 678–9, 696, 713, 795, 853–4, 867–9, 880, 885, 888–9, 996 variable-free semantics 131, 135–6 vehicle change 27, 37, 41, 64, 66, 68, 74, 129, 174, 202–3, 345, 636, 696–7, 707, 718, 946 (p. 1123) verbal phrase ellipsis 777–8 voice 41–4, 62–3, 68, 74, 93–4, 129, 151, 194–6, 259, 267, 282, 316, 327, 421, 448, 469, 472, 484, 513, 515–16, 523–4, 667, 679, 789–90, 798, 842, 845, 859, 967, 970, 980, 1030 voice alternations 845 voice mismatch 41–2, 62–3, 68, 74, 93–4, 129, 195, 267, 282, 448, 469, 515–16, 523–4, 789–90, 967, 980, 1030 VP-ellipsis auxiliary-stranding ellipsis 666–7 Derived Verb Phrase Rule 348 verb-stranding VP-ellipsis 15, 26, 71–3, 520, 820, 900, 905, 932–3, 979; see also VVPE Verbal Identity Requirement 520; see also VIR VIR 520–1; see also Verbal Identity Requirement VP Deletion 173, 370, 660–1 VVPE 14–15, 905, 907–9, 979–80; see also verb-stranding VP-ellipsis wh-fronting 54–5, 480–1, 491–3, 495–6, 740, 967–8, 992–8 Page 14 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021

Index wh-in-situ 860, 865, 954, 993 wh-movement 28, 36, 50, 52–4, 56–7, 66, 90, 280, 287–9, 294–5, 446, 480, 482–3, 490–1, 573, 664–5, 740, 792, 814, 833, 843, 859–60, 929–30, 954, 961, 968, 974, 993–5, 1000, 1036 wh-phrase 9, 12, 20, 23, 26, 28, 30, 32, 34–6, 38, 47–9, 51–4, 56–8, 63–4, 66, 68, 70, 85–6, 98, 105, 113, 171–2, 239–40, 243, 247, 252, 381, 445–51, 453–6, 465, 467–8, 470, 474, 479, 489, 496, 498– 500, 573, 705, 725, 728, 739, 790, 792–6, 822, 831–2, 841–3, 857, 859, 865, 871, 877, 924–5, 927, 929, 953–7, 967–9, 975, 977, 993–9 wh-question 47, 50, 85, 94, 103, 113, 120, 223, 237–8, 287, 367, 377, 450–1, 454–6, 479–80, 489, 492–3, 503, 610, 714, 768, 779, 787, 792, 831, 837, 842, 858, 860, 876, 972, 993, 995–7 wh-remnant 242–3, 383, 385, 468, 473, 489, 842, 874, 879, 927, 956–7, 992, 994–5, 999 WM 429–31, 435–6; see also working memory Word Grammar 145 working memory 11, 429–30; see also WM

Page 15 of 15

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice). date: 03 April 2021