114 6 49MB
English Pages 1297 Year 2017
T h e Ox f o r d H a n d b o o k o f
E RG AT I V I T Y
OXFOR D HANDBOOKS IN LINGUISTICS Recently Published
THE OXFORD HANDBOOK OF CORPUS PHONOLOGY Edited by Jacques Durand, Ulrike Gut, and Gjert Kristoffersen
THE OXFORD HANDBOOK OF LINGUISTIC FIELDWORK Edited by Nicholas Thieberger
THE OXFORD HANDBOOK OF DERIVATIONAL MORPHOLOGY Edited by Rochelle Lieber and Pavol Štekauer
THE OXFORD HANDBOOK OF HISTORICAL PHONOLOGY Edited by Patrick Honeybone and Joseph Salmons
THE OXFORD HANDBOOK OF LINGUISTIC ANALYSIS Second Edition Edited by Bernd Heine and Heiko Narrog
THE OXFORD HANDBOOK OF THE WORD Edited by John R. Taylor
THE OXFORD HANDBOOK OF INFLECTION Edited by Matthew Baerman
THE OXFORD HANDBOOK OF DEVELOPMENTAL LINGUISTICS Edited by Jeffrey Lidz, William Snyder, and Joe Pater
THE OXFORD HANDBOOK OF LEXICOGRAPHY Edited by Philip Durkin
THE OXFORD HANDBOOK OF NAMES AND NAMING Edited by Carole Hough
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 LANGUAGE AND LAW Edited by Peter M. Tiersma and Lawrence M. Solan
THE OXFORD HANDBOOK OF PRAGMATICS Edited by Yan Huang
For a complete list of Oxford Handbooks in Linguistics, please see pp 1263–1264
The Oxford Handbook of
ERGATIVITY Edited by
JESSICA COON, DIANE MASSAM, and
LISA DEMENA TRAVIS
1
3 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 Jessica Coon, Diane Massam, and Lisa deMena Travis 2017 © the chapters their several authors 2017 The moral rights of the authors have been asserted First Edition published in 2017 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 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2016944782 ISBN 978–0–19–873937–1 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.
Contents
Notes on Contributors
xi
1. Introduction Jessica Coon, Diane Massam, and Lisa demena Travis
1
PA RT I AC C OU N T I N G F OR E RG AT I V I T Y Representing Ergativity 2. Ergativity in Discourse and Grammar John W. Du Bois
23
3. Parameterizing Ergativity: An Inherent Case Approach Michelle Sheehan
59
4. Accusative and Ergative in Hindi Anoop Mahajan
86
The Nature of Ergative Case 5. On Inherent and Dependent Theories of Ergative Case Mark C. Baker and Jonathan David BOBALJIK
111
6. The Locus of Ergative Case Julie Anne Legate
135
7. Ergative need not Split: An Exploration into the TotalErg Hypothesis Itziar Laka
159
8. The Structural Source of Split Ergativity and Ergative Case in Georgian Léa Nash
175
vi Contents
PA RT I I C HA R AC T E R I ST IC S A N D E X T E N SION S Characteristics 9. Split Ergativity in Syntax and at Morphological Spellout Ellen Woolford
205
10. Split Ergativity is not about Ergativity Jessica Coon and Omer Preminger
226
11. Ergativity and Differential Case Marking Andrej Malchukov
253
12. Three-Way Systems do not Exist Gereon Müller and Daniela Thomas
279
13. Antipassive Maria Polinsky
308
14. Remarks on the Relation between Case-Alignment and Constituent Order Tarald Taraldsen
332
Extensions 15. Ergativity in Nominalization Artemis Alexiadou
355
16. Ergativity and Austronesian-Type Voice Systems Michael Yoshitaka Erlewine, Theodore Levin, and Coppe van Urk
373
17. On the Morphosyntactic Reflexes of Information Structure in the Ergative Patterning of Inuit Language Alana Johns and Ivona Kučerová 18. Ergative Constellations in the Structure of Speech Acts Martina Wiltschko
397 419
Contents vii
PA RT I I I A P P ROAC H E S TO E RG AT I V I T Y Diachronic 19. Grammaticalization of Ergative Case Marking William B. McGregor
447
20. Deconstructing Iranian Ergativity Geoffrey Haig
465
21. Intransitivity and the Development of Ergative Alignment Edith Aldridge
501
22. Developments into and Out of Ergativity: Indo-Aryan diachrony Miriam Butt and Ashwini Deo
530
23. Ergativity and Language Change in Austronesian Languages Ritsuko Kikusawa
553
24. Lexical Category and Alignment in Austronesian Daniel Kaufman
589
Acquisition 25. The Acquisition of Ergativity: An Overview Edith L. Bavin
631
26. The Role of Defaults in the Acquisition of Basque Ergative and Dative Morphology Jennifer Austin
646
27. A Comparative Study of the Acquisition of Nominative and Ergative Alignment in European and Mayan Languages Clifton Pye and Barbara Pfeiler
665
Experimental 28. Processing Ergativity: Behavioral and Electrophysiological Evidence 693 Adam Zawiszewski
viii Contents
29. Experimental Approaches to Ergative Languages Nicholas Longenbaugh and Maria Polinsky
709
PA RT I V C A SE ST U DI E S 30. Correlates of Ergativity in Mayan Judith Aissen
737
31. Ergative Case in Burushaski: A Dependent Case Analysis Mark C. Baker
759
32. Ergativity in Basque Ane Berro and Ricardo Etxepare
782
33. Hindi/Urdu and Related Languages Miriam Butt
807
34. Ergativity in Inuktitut Richard Compton
832
35. Ergativity in Nakh–Daghestanian Diana Forker
851
36. Ergativity in Neo-Aramaic Geoffrey Khan
873
37. Ergativity in Africa Christa König
900
38. Ergativity in Tibeto-Burman Shobhana Chelliah
924
39. The Ergative in Warlpiri: A Case Study Mary Laughren
948
40. Ergative–Absolutive Patterns in Tongan: An Overview Yuko Otsuka
989
41. Alignment across Tsimshianic Tyler Peterson
1007
Contents ix
42. What being a Syntactically Ergative Language means for Katukina-Kanamari Francesc Queixalós 43. Ergativity in Jê Languages Andrés pablo Salanova 44. Interaction of Ergativity and Information Structure in Jaminjung (Australia) Eva Schultze-Berndt
1035 1065
1089
45. Alignment and Orientation in Kartvelian (South Caucasian) Kevin Tuite
1114
References Author Index Language Index Subject Index
1139 1241 1245 1251
Notes on Contributors
Judith Aissen is Research Professor of Linguistics at UC Santa Cruz. The primary focus of her work since the 1970s has been on the morphology, syntax, and information structure of the Mayan languages, especially Tsotsil, Tz’utujil, and K’ichee’. She has always been particularly interested in the properties and analysis of Agent Focus constructions, and their position at the interface of these three modules. Edith Aldridge is Associate Professor in the Department of Linguistics at the University of Washington. Her work focuses on syntax, particularly diachronic change, of Austronesian languages, Chinese, and Japanese. Her work on ergativity has appeared in the journals Language and Linguistics, Compass, Lingua, Linguistics Vanguard, Sophia Linguistica, as well as the collected volumes Grammatical Change: Origins, Nature, Outcomes and Encyclopedia of Chinese Language and Linguistics. Artemis Alexiadou is Professor of English Linguistics at the Humboldt University of Berlin. She received her PhD in linguistics in 1994 from the University of Potsdam. Her research interests lie in theoretical and comparative syntax, morphology, and most importantly in the interface between syntax, morphology, the lexicon, and interpretation. Jennifer Austin is Associate Professor in the Department of Spanish and Portuguese Studies at Rutgers University, Newark. She received her PhD in linguistics from Cornell University with a minor in cognitive science. Her research interests include language acquisition, bilingualism, and the effects of language contact. She is a co-author of the book Bilingualism in the Spanish-speaking World (2015) and has published articles on the acquisition of Basque, English, and Spanish. Mark C. Baker is Distinguished Professor of Linguistics and Cognitive Science at Rutgers University. He received his PhD in linguistics in 1985 from MIT, and previously taught at McGill University. He specializes in the syntax and morphology of less-studied languages, seeking to bring together generative-style theories and data from fieldwork and typological comparison. He has written five research monographs, including Case: Its Principles and Its Parameters (2015). Edith L. Bavin received her PhD from the University of Buffalo and taught at the University of Oregon before moving to Australia, where she conducted fieldwork on Nilotic languages and on the acquisition of Warlpiri. She switched to experimental work focusing on both typical and atypical language development. She was editor
xii Notes on Contributors of the Journal of Child Language (2006–12), edited the Cambridge Handbook of Child Language (2009) and a second expanded edition (2015) together with Letitia Naigles, and co-edited with Sabine Stolle The Acquisition of Ergative Languages (2013). Now an honorary professor at La Trobe University, she is still conducting acquisition research. Ane Berro is a post-doc researcher at the University of the Basque Country (UPV/EHU) and the laboratory Structures Formelles du Langage (UMR 7023, CNRS/ Paris 8). She did her doctoral dissertation ‘Breaking Verbs: From Event Structure to Syntactic Categories in Basque’ under the supervision of Beatriz Fernández (UPV/ EHU) and Ricardo Etxepare (CNRS IKER UMR 5478), and, currently, she is working on aspect, categorization, and the relation between syntax and morphology. Jonathan David Bobaljik has held appointments at Harvard and McGill, and is currently Professor of Linguistics at the University of Connecticut. His theoretical research has focused primarily on morphology and syntax, and he has conducted descriptive fieldwork on the Itelmen language, Kamchatka, Russia. Publications include Universals in Comparative Morphology (2012). Miriam Butt is Professor for Theoretical and Computational Linguistics at the University of Kontanz. Her research interests include morphosyntax, historical linguistics, and computational linguistics. The bulk of her research is on South Asian languages, with a special emphasis on Urdu, though she has also worked on English and German. She has written and edited several books on syntax, semantics, and computational linguistics, including a textbook on theories of case. Shobhana Chelliah is a documentary linguist working primarily on the Tibeto- Burman languages of Manipur state in northeast India. Her interests lie in morphosyntactic issues such as case morphology, referent tracking, and affix ordering. Her publications include A Grammar of Meithei (1997) and the Handbook of Descriptive Linguistic Fieldwork (co-authored with Willem de Reuse (2011). With funding from the National Science Foundation, she is currently working with members of the Lamkang community (Tibeto-Burman) to develop a practical orthography and create an online dictionary for their language. She is also working towards the creation of a language archive for Tibeto-Burman languages. Chelliah served as the Program Director for the Documenting Endangered Languages program at the National Science Foundation (2012–15), and is currently a professor of linguistics at the University of North Texas. Richard Compton is Assistant Professor in the Department of Linguistics at the Université du Québec à Montréal in Canada. His works examines polysynthetic word- formation, nominal and verbal incorporation, lexical categories, modification, and agreement in Eastern Canadian Inuktitut. He has conducted fieldwork in the communities of Iqaluit and Baker Lake in Nunavut and is currently co-editing a new edition of a dictionary of the Kangiryuarmiut dialect of Western Canadian Inuit.
Notes on Contributors xiii Jessica Coon is Associate Professor of Linguistics at McGill University. She finished her PhD at MIT in 2010 and then spent one year as a postdoctoral fellow at Harvard University. Jessica has worked on topics including ergativity, split ergativity, verb-initial word order, and agreement, with a special focus on Mayan languages. Her book Aspects of Split Ergativity was published by Oxford University Press in 2013. Ashwini Deo received a Master’s in Sanskrit grammar and linguistics from Pune, India, followed by a PhD in linguistics from Stanford University (2006). She is an associate professor at Yale University. Her main research interest is in systematic semantic change phenomena—particularly in the ways in which functional morphemes like tense–aspect, negation, possession markers change over time in the ways that they do. Within semantics– pragmatics she also works on phenomena in the domains of aspect, temporal reference, lexical semantics of verbs, and genericity. Her empirical focus is on the Indo-Aryan languages. John W. Du Bois is Professor of Linguistics at the University of California, Santa Barbara. A specialist in functional linguistics, discourse, Mayan languages, and sociocultural linguistics, his work centers on the interaction between discourse and grammar. He has long been interested in the complex functional competitions that drive the emergence of grammar as a complex adaptive system, yielding the extraordinary typological diversity of argument structure constructions and syntactic alignments in the world’s languages. His publications include Competing Motivations (1985), The Discourse Basis of Ergativity (1987), Preferred Argument Structure (2003), Discourse and Grammar (2003), Motivating Competitions (2014), and Towards a Dialogic Syntax (2014). Michael Yoshitaka Erlewine is Assistant Professor at the National University of Singapore. He received his PhD in linguistics from MIT in 2014. His research interests are syntactic theory and the syntax–semantics interface. Ricardo Etxepare is a permanent researcher at the Centre National de la Recherche Scientifique, and director of the lab IKER (UMR5478), a research center devoted to the study of the Basque language and Basque texts in Bayonne, France. Diana Forker teaches general linguistics at the University of Bamberg. She completed her PhD at the Max Planck Institute for Evolutionary Anthropology. Her main interests are languages of the Caucasus, typology, and morphosyntax. She currently works on the documentation of the Nakh-Daghestanian language Sanzhi Dargwa and on a typological investigation of gender agreement. Among her publications are A Grammar of Hinuq (2013) and several articles on different aspects of Nakh-Daghestanian languages. Geoffrey Haig received his PhD in general linguistics from the University of Kiel in 1997. He is currently professor of linguistics in the Institute of Oriental Studies at the University of Bamberg. His research interests focus on the languages of the Middle East, in particular the syntactic features they have inherited, and those they have shared with their neighbors. He is also active in language documentation, and in corpus-based approaches to language typology and areal linguistics.
xiv Notes on Contributors Alana Johns is Professor of Linguistics at the University of Toronto. Her research concerns the morphosyntax of complex words, especially in the Inuit language. She has published on ergativity, noun incorporation, and verbal inflection. One of her main research interests are syntactic differences between closely related dialects, where small distinctions can lead to a wide range of effects. Another area of involvement is language maintenance, where linguistics can help in capacity development within communities who are engaged with these issues. She has worked with the Inuit of Nunatsiavut for many years. Daniel Kaufman specializes in historical, descriptive, and theoretical issues in Austronesian languages with a focus on the languages of the Philippines and Indonesia. He is co-founder and executive director of the Endangered Language Alliance, a non- profit organization dedicated to documenting and conserving the endangered languages of New York City’s immigrant communities and is also Assistant Professor at the Department of Linguistics and Communication Disorders at Queens College, CUNY. Geoffrey Khan is currently Regius Professor of Hebrew at the University of Cambridge. He was awarded his PhD by the University of London in 1984 for a thesis on extraposition in the Semitic languages. His published books include a series of grammars of Neo- Aramaic dialects, editions of medieval Judaeo-Arabic grammatical texts and medieval Arabic documents and an introduction to the Hebrew Bible. He is general editor of the Encyclopedia of Hebrew Language and Linguistics. Ritsuko Kikusawa is Associate Professor of the National Museum of Ethnology and the Graduate University for Advanced Studies in Japan. Her research interests include description of Fijian dialects and Betsimisaraka Malagasy, methodology of morphosyntactic comparison and reconstruction, and the linguistic prehistory of Oceania. Her recent publications include “The Austronesian language family” (The Routledge Handbook of Historical Linguistics, ed. Claire Bowern & Bethwyn Evans, 2015). Christa König is Apl. Professor at the Institute of African Linguistics, University of Frankfurt. Her research interests include verbal aspect and case systems. She has carried out extensive field research on the following languages: Maa (Kenya, Tanzania), Ik (Northeastern Uganda), !Xun (Namibia), and Akie (Tanzania). Her publications include ‘Marked nominative in Africa’ (Studies in Language 30(4): 705–782, 2006) and Case in Africa (Oxford University Press, 2008). Ivona Kučerová is Assistant Professor of Linguistics at McMaster University. She specializes in theoretical syntax and semantics, and their interface. Her works explores information structure and its morphosyntactic correlates, definiteness systems and their relation to aspect, the morphosyntax and morpho-semantics of case, agreement, and case splits, the syntax of null languages, and the syntax of copular clauses. She works mainly on Slavic, Germanic, and Romance languages.
Notes on Contributors xv Itziar Laka is Full Professor at the Department of Linguistics and Basque Studies and Director of The Bilingual Mind research group at the University of the Basque Country. Her research combines theoretical linguistics and experimental methods to inquire into the representation and processing of language, with a strong focus on syntax and bilingualism. Mary Laughren is an honorary senior research fellow in the School of Languages and Cultures at the University of Queensland, where she taught linguistics (1993–2009). Since 1975 she has studied the Warlpiri language and has been involved in the implementation of bilingual education programs in Warlpiri-speaking communities. One focus of her research is the interplay between lexical and syntactic organization. She is currently collaborating in the documentation of traditional Warlpiri songs, and since 2000 of another Australian language, Waanyi. Julie Anne Legate is Professor of Linguistics and the Chair of Undergraduate Studies in the Department of Linguistics at the University of Pennsylvania. She received her PhD from MIT in 2002. She is the author of Voice and v: Lessons from Acehnese, and is editor- in-chief of Natural Language & Linguistic Theory. Theodore Levin received his PhD in linguistics from MIT in 2015 and is a postdoctoral research associate in the Department of Linguistics at the University of Maryland. His primary research interests are in syntactic theory with specific interest in the interaction of case, agreement, and word order. Nicholas Longenbaugh is a PhD student in linguistics at MIT, with a BA in computer science and linguistics from Harvard University (2014). His work focuses on issues in syntax, with an emphasis on cross-linguistic variation and universal principles. He has done original research and fieldwork on Austronesian and Romance languages. Nicholas has also worked within the Tree-Adjoining Grammar framework, and has explored topics concerning formal complexity in language. Anoop Mahajan is Professor in the Department of Linguistics at UCLA. He received his PhD in linguistics from MIT in 1990 and has taught at UCLA since 1992. His research includes work in formal generative syntax with a special emphasis on how to account for typological variation across languages. He has published research on various topics in syntax that include word order and scrambling, agreement and case, ergativity, partial wh-movement, and relative clauses. Andrej Malchukov is a senior researcher at the St.-Petersburg Institute for Linguistic Research (Russian Academy of Sciences), currently affiliated as Visiting Professor to the University of Mainz. Apart from descriptive work on Siberian (in particular, Tungusic) languages, his main research interests lie in the domain of language typology. He published extensively on the issues of morphosyntactic typology; in particular, he edited The Oxford Handbook of Case (together with Andrew Spencer; 2009), Studies in Ditransitive Constructions: A Comparative Handbook (together with Bernard Comrie and Martin
xvi Notes on Contributors Haspelmath; 2010); Competing Motivations in Grammar and Cognition (together with Brian MacWhinney and Edith Moravcsik; OUP, 2014), and Valency Classes in the World’s Languages (together with Bernard Comrie; 2015). Diane Massam (PhD MIT 1985) is Professor in the Department of Linguistics at the University of Toronto. Her research areas are argument structure, case, predication, and word order, with a focus on the Niue language (Polynesian), and an interest in register variation in English. She has edited volumes on Austronesian syntax, ergativity, and the count–mass distinction, and was co-editor of Squibs for Linguistic Inquiry. William B. McGregor is Professor of Linguistics in the Department of Aesthetics and Communication, Aarhus University, Denmark. His primary research interest is in the languages of the Kimberley region of Western Australia; more recently he has begun work on a Khoisan language of Botswana. He has published widely on these languages, including descriptive grammars, and has a long-term interest in ergativity. Gereon Müller is Professor of general linguistics at Leipzig University, and head of the graduate program “Interaction of Grammatical Building Blocks” (IGRA). He got his Dr. phil. from Tübingen University in 1993, and his Dr. habil. from the same university in 1996, both with works on theoretical syntax. His main research interest is grammatical theory, with a special focus on syntax and morphology. Léa Nash PhD (1995), Paris 8 University, is Professor of Linguistics at that university. She has published many articles in theoretical syntax, especially on argument structure, case theory, and ergativity. Yuko Otsuka is Associate Professor in Linguistics at the University of Hawai‘i at Mānoa. Her research interests are theoretical syntax and Austronesian languages. She has worked extensively on Tongan (Polynesian) with special focus on case and ergativity. Tyler Peterson received his PhD from the University of British Columbia in 2010 and joined the University of Auckland in 2016. He has undertaken extensive fieldwork on the endangered indigenous language Gitksan (Tsimshianic, British Columbia), and has also worked with indigenous languages in the southwestern United States, the South Pacific, and the Brazilian Amazon. Barbara Pfeiler is Professor of Linguistics at the National Autonomous University of Mexico in Mérida, Yucatán. She has conducted fieldwork on the acquisition of Yukatek and Teenek since 1997. She has published articles on the acquisition of these languages as well as on the sociolinguistics and dialectology of Yukatek. She edited the volume Learning Indigenous Languages: Child Language Acquisition in Mesoamerica (2007). Maria Polinsky (PhD 1986) is Professor of Linguistics and Associate Director of the Language Science Center at the University of Maryland. Her main interests are in theoretical syntax, with an emphasis on cross-linguistic variation. She is also interested in the
Notes on Contributors xvii integration of experimental methodologies in linguistic research. She has done extensive work on ergative languages across several language families, namely, Austronesian, Kartvelian, Nakh-Dagestanian, and Paleo-Siberian. Omer Preminger is Assistant Professor of Linguistics at the University of Maryland (College Park). He graduated from MIT in 2011, and was a postdoctoral associate at MIT and at Harvard. Before joining UMD, he was a faculty member at Syracuse University. Omer has worked on issues of agreement and case in a variety of languages, including Basque, Hebrew, Kaqchikel, and Sakha. His publications include Agreement and Its Failures (2014). Clifton Pye is Associate Professor of Linguistics at the University of Kansas. His specialty is the documentation of the acquisition of Mayan languages with a special focus on the acquisition of K’iche’, Mam, Q’anjob’al, and Ch’ol. He has published numerous articles on the acquisition of phonology, morphology, syntax, and semantics in these languages. Francesc Queixalós as a member of the Centre National de la Recherche Scientifique and, periodically, of the Institut de Recherche pour le Développement (France), Francesc Queixalós has conducted in-depth research on two languages of lowland South America, Sikuani (Guahiban) spoken in the savanna area of the middle Orinoco, and Katukina-Kanamari (Katukinan), spoken in the rain forest south of the middle Amazon, while addressing several issues in the morphosyntax of Tupi- Guarani languages. He has taught linguistics in several Universities in France and South America. Andrés Pablo Salanova is Associate Professor of Linguistics at the University of Ottawa. He has worked with the Mẽbengokre since 1996, totaling over one year in the field and writing on several different aspects of their language. Salanova holds a BA in mathematics from Brown University, an MA in linguistics from Campinas (Brazil), and a PhD in linguistics from MIT. Eva Schultze-Berndt is Professor of Linguistics at the University of Manchester, UK. She received an MA from the University of Cologne, and a PhD from the University of Nijmegen. Her research interests include complex predication, overt classification of verbs/events, secondary predication, spatial language, parts of speech, information structure, language contact, and documentary linguistics. Michelle Sheehan is a senior lecturer at Anglia Ruskin University, specializing in comparative syntax. She has worked on null arguments, Control, word order variation, clausal–nominal parallels, and case/alignment. She is co-author of Parametric Variation: Null Subjects in Minimalist Theory (2010); The Philosophy of Universal Grammar (OUP, 2013); The Final-over-Final Constraint (forthcoming). Michelle is co- editor of Theoretical Approaches to Disharmonic Word Order (OUP, 2013) and of the journal IBERIA.
xviii Notes on Contributors Tarald Taraldsen (PhD University of Tromsø, 1983) has worked as Professor of Linguistics at the University of Tromsø since 1984. He was a senior researcher at the Center for Advanced Study in Theoretical Linguistics (2002–12). He has worked on the syntax of Scandinavian and Romance languages and, since 2008, also on the syntax and morphology of Bantu languages. Daniela Thomas gained her MA from Leipzig University in 2015, with a thesis on a harmonic grammar approach to scale effects in argument encoding that combines minimalist syntax with weighted constraints in post-syntactic morphology. Her bachelor thesis from 2013 tackles split ergativity in subordinate contexts from a minimalist perspective. Lisa deMena Travis completed her PhD in Linguistics at MIT in 1984, writing on parameters of word order variation. She is currently Professor in the Department of Linguistics at McGill University where she has been teaching since 1984. Her research focuses mainly on phrase structure, head movement, language typology, Austronesian languages (in particular, Malagasy and Tagalog), and the interface between syntax and phonology. Recent publications include Inner Aspect: The Articulation of VP (Springer, 2010). Kevin Tuite is Professor of Anthropology at the Université de Montréal. He directed the Caucasus Studies program at the Friedrich-Schiller-Universität Jena (2010–14). Since 1985, he has been researching the languages and cultures of the Caucasus, with a focus on Georgia. He is presently working on a grammar of the Svan language, and a study of the cult of St George in the Caucasus. Coppe van Urk received his PhD in linguistics from MIT in 2015 and is a lecturer at Queen Mary University of London. His research focuses on syntax, with specific interests in movement, case, agreement, and the structure of Dinka. Martina Wiltschko is Professor of Linguistics at the University of British Columbia (Vancouver). She is interested in the range and limits of variation in the domain of syntax and its interfaces (syntax–morphology; syntax–semantics; syntax–pragmatics). She has extensively published on several empirical domains pertaining to this question including her recent monograph The Structure of Universal Categories: Towards a Formal Typology. Ellen Woolford received a BA from Rice University and a PhD from Duke University, with a dissertation on Tok Pisin based on fieldwork in Papua New Guinea. She is currently a professor of linguistics at the University of Massachusetts. She has published on a range of topics in syntax, including locality constraints on wh movement and the behavior of passives in double-object constructions. Her recent work focuses on the theory and typology of case and agreement.
Notes on Contributors xix Adam Zawiszewski graduated in Romance philology (Adam Mickiewicz University), obtained a Master’s in Linguistics from the University of the Basque Country (UPV/ EHU) where he also defended his PhD. He completed a postdoctoral training at the Max-Planck Institute for Human Sciences and Cognition (Leipzig). In 2011 he was awarded a Juan de la Cierva Fellowship by the Spanish Ministry of Science and Innovation. Currently he is working as Assistant Professor at the University of the Basque Country (UPV/EHU).
Chapter 1
Introdu c t i on Jessica Coon, Diane Massam, and Lisa deMena Travis
1.1 Introduction 1.1.1 Introducing Ergativity This volume tackles the phenomenon known as ergativity. The term “ergativity” has been most commonly used to refer to systems with one or both of the following properties: (i) transitive subject (A arguments in (1)) pattern differently from intransitive subjects (S arguments) and from transitive objects (P arguments); and (ii) transitive objects and intransitive subjects pattern alike (see Figure 1.1). In such a system, schematized in Figure 1.1(a), the A argument is referred to as the “ergative” argument, and the S and P arguments are the “absolutive” arguments. This type of system contrasts with a “nominative–accusative” systems, shown in Figure 1.1(b). Just as there is more than one way to be “ergative,” it is important to point out that “ergativity” may refer to any characteristic which aligns arguments as in Figure 1.1(a)— this includes not only the more common morphological case marking and agreement, but also word order, discourse and information structure, or the extractability of arguments. A wide range of work across different traditions converges on the idea that “ergativity” is not a single unitary phenomenon, and is not realized in the same way across different languages. Dixon (1994: 219), for example suggests that “there is no necessary connection between ergative characteristics and any other linguistic feature,” and Johns (2000: 67) writes in a similar vein that there may be “little value in studying ergativity as a thing in itself.” In her recent survey of ergativity, Deal (2015b) suggests that “ergativity is not one but many phenomena.” Nonetheless, certain patterns and correlations emerge, suggesting that while there is certainly diversity, there is also some unity—perhaps motivating the existence of this additional volume on the topic. The general themes of unity and diversity in and among ergative systems are touched on in the chapters that follow.
2 Jessica Coon, Diane Massam, Lisa Travis (b) NOMINATIVE–ACCUSATIVE
(a) ERGATIVE–ABSOLUTIVE
transitive:
intransitive:
A
P
A
P
ERG
ABS
NOM
ACC
S ABS
S NOM
Figure 1.1 Alignment patterns
Before discussing the organization and content of the volume, a few disclaimers are in order. First, in this introduction, we do not attempt to provide a comprehensive survey of the wide range of existing work on ergativity. We refer readers to works such as Johns (2000), Aldridge (2008a), and Deal (2015) for overviews in the generative tradition; Dixon (1979, 1994), McGregor (2009), and Comrie (2013a, 2013b) for more typologically oriented overviews; and Coon and Adar (2013) for an annotated bibliography of work on this topic. Rather, we aim to highlight some of the topics that have emerged over the years which we feel reappear throughout this volume, and point to some common themes. Second, due to the broad reach of the term “ergative,” ergativity may be viewed from a number of different perspectives. We have done our best to include chapters representing a range of different theoretical and methodological traditions, though, as with any volume, imbalances reflecting the orientations of the editors nonetheless exist. Finally, many of the chapters contained here would fit naturally into more than one of the book’s four parts, and the reader should interpret these divisions as loose guidelines rather than a strict packaging.
1.1.2 Themes and Organization The volume is organized into four main parts. Part I, on accounting for ergativity, focuses on factors which distinguish ergative from non-ergative systems and how these may be parameterized and formalized in the grammar. In Part II, common as well as less-common characteristics and manifestations of ergative systems are discussed. Topics here include alignment splits, antipassive constructions, and word order correlations, as well as nominalization, voice systems, and connections to speech acts and information structure. Part III focuses on approaches which draw on data from a diverse range of methodologies; these chapters focus on ergativity through the lens of diachronic, experimental, and acquisition research. Finally, Part IV turns to case studies—in-depth looks at ergativity and ergative phenomena in particular languages or language families.
Introduction 3 Throughout the four parts of this volume, several themes emerge. One such theme is the impressive diversity of languages which exhibit ergativity—languages from nearly every continent and an impressive number of language families are represented—as well as the wide range of phenomena that have been associated with the label “ergative.” In addition to diversity in the geographic and empirical landscape, the contributions to this volume also reflect the range of different analyses, views, and theoretical approaches of how to interpret these facts. Relatedly, it becomes clear that ergativity is not as fully isolated a phenomenon as it is sometimes made out to be. Some characteristics that have been argued to hold of ergativity do not hold in a uniform way, such as the existence of split systems, antipassives, or extraction restrictions, or continue to resist full explanation, such as correlations between word order and ergativity. As has been frequently noted, not only do we find non-ergative patterns throughout languages traditionally labeled “ergative,” we also find ergative patterns in a number of language and domains normally considered “nominative–accusative.” To pick just a couple of illustrative examples, this volume includes a number of chapters on split ergativity (see especially sections 1.2.2 and 1.3.1) that demonstrate not only the complexity of defining ergative splits and differentiating them from other types of differential argument marking systems, but also disagreement about how they should be formally represented: as simply morphophonological rather than syntactic; connected to something specific to the syntax of ergative languages; driven by competing functional discourse pressures; or even that they might just be the natural fallout of other structural properties, and hence not be a hallmark of ergative languages at all. In addition, the antipassive, once commonly thought to be exclusive to ergative languages, is argued to be found across other types of languages as well (see sections 1.3.1 and 1.5). The derivational origins and limits of ergativity are similarly unclear, being possibly based on information structure, or perhaps related to voice systems and nominalizations, and with a possible reach to other domains such as speech act structure (sections 1.2.1 and 1.4). Despite the diversity reflected here, a number of points of commonality or areas of agreement emerge. A look at the contributions in Part IV drives home the point that simply labeling a language as “ergative” or “accusative” is not enough. Many contributions here highlight the importance of careful, holistic investigations into individual languages. Just as a given language must be examined carefully, it may be examined from more than one angle. In this volume we see the benefits of increasing the diversity of approaches to the study of ergativity (see section 1.4), as well as an increase in cross-collaboration in various disciplines—through studies of acquisition (Bavin; Austin; Pye and Pfeiler), experimental work (Longenbaugh and Polinsky; Zawiszewski), diachronic analyses (McGregor; Haig; Aldridge; Butt and Deo; Kikusawa; Kaufman), or through discourse and speech act structure (Wiltschko; Du Bois). The implications discussed in the chapters in this volume are similarly far- reaching, with consequences for the representation of case and agreement systems more generally, for argument structure, and the role of constraints in the grammar, to name just a few.
4 Jessica Coon, Diane Massam, Lisa Travis Another striking point of commonality across many of the formal accounts of ergativity presented here is that ergativity or ergative assignment occurs low in the structure. Whether this is formalized as a low, in situ licensing of ergative subjects (see Legate; Sheehan; Laka; Müller and Thomas; Woolford; Aldridge), or as the result of ergative being assigned configurationally to the higher of two nominals in some domain, perhaps by virtue of the ergative argument remaining low (Baker and Bobaljik; Baker; Nash; Coon and Preminger), the relevance of the structural height of subjects is discussed throughout many of the chapters in this volume. However it is formalized, the proposed “lowness” of ergative subjects may in turn have consequences elsewhere in the grammar, for example in word order (Taraldsen), or for connections between nominalization and ergativity (Alexiadou; Kaufman). We do not explore further the many threads of research in this volume, but will let the contributions—which we go on to introduce briefly in turn—speak for themselves. Far from being the last word on ergativity, we anticipate that this volume will serve to spark further interest and study of this topic, which we hope to have demonstrated has implications for linguists working in any discipline or subfield of linguistics.
1.2 Part I: Accounting for Ergativity The chapters in Part I share a common goal: to understand and model how ergativity arises either in a specific language, or cross-linguistically. Part I has two sections: in the first of these, on representing ergativity, Du Bois, Sheehan, and Mahajan each discuss the “parameterization” of ergativity. Du Bois’ chapter focuses on functional motivations for ergativity, and the competing pressures which might result in ergative or accusative grammatical systems. Sheehan seeks to capture not only differences between ergative and non-ergative languages, but also differences among what she identifies as subtypes of ergative languages, with a parameter hierarchy. Mahajan narrows in to discuss differences in how “absolutive” arguments are represented, with a focus on Hindi. Though they have different scopes and approaches, the chapters share a theme found throughout the volume and in other work, namely that languages may manifest ergativity in different ways and perhaps to different degrees. The chapters in the second section, on the nature of the ergative case, tackle a specific question in the formal representation of ergativity: What is special about ergative arguments? Two main approaches are presented, labeled in Baker and Bobaljik’s contribution as the “Inherent Case Theory” (ICT) and the “Dependent Case Theory” (DCT). In the former, ergative case is assigned to an external argument in its base position (e.g. specifier of vP). In the DCT approach, ergative case is assigned configurationally; it is not tied to a specific functional head, but rather is assigned to the higher of two nominals in some specified domain. Baker and Bobaljik introduce both options, presenting evidence in favor of DCT. In her contribution, Legate presents an overview of the behavior of ergative marking in a wide range of different languages.
Introduction 5 Despite the range of variation, she argues that what they share in common is a low source or ergative case, registering concerns for a DCT approach. Both Laka and Nash tackle the question of ergative case assignment and splits in specific languages— Basque and Georgian, respectively—coming down on different sides of the debate. Laka argues in favor of a consistently low locus of ergative case in Basque, providing a detailed analysis of the verb behar (‘need’). Nash, on the other hand, argues that the non-ergative alignment in Georgian arises when the subject is outside of the vP domain, which she formalizes in terms of dependent case. We summarize each of the volume’s chapters in more detail in the following sections.
1.2.1 Representing Ergativity In Chapter 2, Du Bois discusses the relationship between ergativity and an “ergative discourse profile.” Du Bois argues that the presence of ergative alignment in discourse— specifically, an ergative alignment in terms of which roles arguments play when they are introduced and tracked throughout utterances—reveals motivations for grammaticalized patterns of ergativity cross-linguistically. A conflicting universal discourse– pragmatic pressure, namely for “topicality,” gives rise to accusativity. In this chapter, the ergative discourse profile is examined through the lens of typology, language acquisition, and language change. Additional functional factors which contribute to the grammaticalization of ergativity including verb semantics, aspects, and inherited morphosyntax, are also discussed. Chapter 3 by Sheehan outlines a parameter hierarchy to capture variation in alignment systems. The first parameter determines the presence of ergative vs. non- ergative alignment based on whether or not v0 is able to assign inherent ergative case. Further micro-parameters within the ergative setting determine (i) the full distribution of ergative case (i.e. whether there are splits or active alignment); (ii) the presence or absence of extraction restrictions on ergative subjects; and (iii) the source of absolutive case in transitive contexts. Sheehan’s contribution allows for variation within ergative systems, while still restricting the range of possible alignment systems. She discusses how the rankings between parameters connect to the need to create convergent derivations. Mahajan (Chapter 4) tackles the mechanism by which direct objects are licensed in Hindi. Through the examination of the syntax of perfective and imperfective prenominal relative clauses, Mahajan argues that morphologically bare (“absolutive”) direct objects in Hindi are licensed by T. Specifically, Mahajan proposes that the restrictions on which arguments can be relativized in prenominal relatives provide evidence for how case licensing works in participial clauses; this in turn offers a window into licensing mechanisms in ergative constructions. These results contrast with recent work (e.g. Legate 2008 and others) which has argued that transitive objects in Hindi are licensed low by v0. Differential Object Marking is also discussed, and argued to not be a substitute for structural case licensing.
6 Jessica Coon, Diane Massam, Lisa Travis
1.2.2 The Nature of the Ergative Case Baker and Bobaljik (Chapter 5) discuss two approaches to how the ergative case is assigned in the grammar, labeled Inherent Case Theory (ICT) and Dependent Case Theory (DCT). In the former, ergative case is an inherent case assigned to the subject by v0, while in the latter, ergative is a dependent case assigned configurationally to the higher of two arguments in some local domain. Baker and Bobaljik discuss the predictions of the two accounts and argue in favor of DCT through an examination of languages such as Shipibo, Kalaallisut, and Chukchi. As evidence against the ICT, they present constructions in which non-agents bear ergative case, and in which agents fail to receive ergative case. They also discuss the absence of active patterns of morphological case marking, argued to be predicted on the ICT. Legate argues in Chapter 6 that while the ergative case is not determined by a single factor cross-linguistically, ergative-assignment is governed by a consistent constellation of factors which share the property of occurring low in the clause, centered around vP. The factors Legate identifies include: theta-position and theta-role of the subject, the presence of a complement, the presence of a DP object, the theta-role of the object, the case of the object, the presence of object agreement, the lexical predicate, the light verb, and the aspectual head which selects vP. A wide range of languages are discussed, including two for which ergative initially seems to have a higher locus (TP or CP): Kurmanji Kurdish and Yukulta. Legate concludes that even here, ergative is assigned low and that high ergative languages may not exist. Laka examines ergativity in Basque in Chapter 7. In particular, she discusses what she calls the “TotalErg” hypothesis: the hypothesis that ergative is an inherent case, and that ergativity does not actually split. Apparent splits, under this account, are epiphenomenal, resulting from different structures rather than from different case-assignment properties of functional heads. She examines the Basque predicate behar (‘need’), which shows a split in the assignment of case to subjects and has been recently argued to provide evidence in favor of structural assignment of ergative by T (Rezac, Albizu, and Etxepare 2014). Laka argues instead that predicates like behar are nominals, and not raising modals. She concludes that there is no raising-to-ergative in Basque, and that ergative case is uniformly assigned by transitive v0. Nash’s contribution (Chapter 8), on the structural source of split ergativity and ergative case in Georgian, examines split ergativity in Georgian in order to understand the difference between ergative and nominative behavior within a single language, with cross-linguistic implications. Specifically, Nash argues that nominative alignment arises when the transitive subject is case-licensed in a position outside of vP. In an ergative system, on the other hand, the transitive subject is licensed inside of vP. Nash formalizes this variation in terms of a configurational approach to case assignment, in which the low subject is local enough to the object to receive dependent ergative case. Nash attributes the differences in licensing patterns to the presence or absence of an aspectual category, “Event,” which has both semantic and syntactic consequences.
Introduction 7
1.3 Part II: Characteristics and Extensions Part II has two sections. The first section includes chapters that examine some of the key grammatical characteristics that are commonly considered to be correlated with ergativity, and the second includes chapters that extend our notion of ergativity in one of two ways. Some chapters extend our view of ergativity empirically, by looking at phenomena that are usually considered to lie outside ergativity but which have been argued to be related to it, while other chapters extend the usual discussion of ergativity theoretically, by tying ergativity to theories of speech acts and information structure.
1.3.1 Characteristics Languages with ergative systems are said to exhibit properties and constructions that are characteristic of ergativity. In this first section of Part II, some of these characteristics are explored. One common claim about languages with ergative case systems is that they are never uniformly ergative—rather, they always exhibit other case patterns as well. This property is referred to as split ergativity, in which the ergative pattern is lost in certain contexts, often in non-perfective aspects or in contexts with “highly ranked” (e.g. first-and second- person) subjects. Several chapters in this section explore split ergativity, as we now outline. In Chapter 9, Woolford focuses on types of split ergative languages, providing an overview of conflicting definitions in the literature. She argues that a consistent definition is important in evaluating claims about whether all ergative languages exhibit splits. She discusses familiar triggering factors such as person and aspect (e.g. Marathi, Chol) and lesser-known triggers such as stage or individual predicates (Nepali) and social conventions (Folopa, Mongsen Ao). She includes languages where ergativity depends on object properties (e.g. Niuean), and she also examines languages with “active” alignments, arguing that while some are split (Choctaw), others are fully ergative (Laz), since all verbs that can license ergative case do so in all contexts. In their contribution, Coon and Preminger (Chapter 10) examine both aspect-based splits (as in Basque and Samoan) and person-based splits (as in Sakha and Dyirbal), arguing that split ergativity is epiphenomenal, and that it is not in fact limited to ergative languages. They consider that case splits are due to structural factors, with the non- ergative pattern arising as the result of a bifurcation of the clause, so that the clause becomes intransitive, hence straightforwardly not ergative. They argue that bifurcation can be the result of non-perfective (i.e. locative) syntax, or of first-and second-person licensing requirements, thus accounting for the universal directionality of the splits. They conjecture that because all subjects pattern similarly in nominative languages, such splits are not as apparent as they are in ergative languages.
8 Jessica Coon, Diane Massam, Lisa Travis Malchukov (Chapter 11) takes a functional–typological approach to splits, examining Differential Object Marking (DOM), which is widely attested, and differential subject marking (DSM), which is less discussed, and is found mainly in ergative languages. He shows that while DOM can be uniformly explained via markedness, DSM cannot be so explained, as many patterns mark subjects that are higher on the hierarchy (e.g. Hindi). He explores two views about case: indexing and distinguishability. Though these are sometimes taken to be in conflict, he argues that both are needed, as together they can account for the varying patterns of DSM and other case patterns. He presents an OT (Optimality Theory) analysis, showing that two unranked constraints, DIFF and INDEX, can converge or not, allowing for the existence of different patterns. Müller and Thomas (Chapter 12) discuss three-way systems, arguing that such systems do not exist syntactically, but diverge from two-way systems through scale-driven optimization operations at the syntax–morphology interface. They argue that such languages are actually either ergative or accusative, with case markers that disappear in certain contexts because of morphological processes. Through examination of a range of languages such as Kham, Djapu, Nez Perce, Upriver Halkomelem, and Dyribal, they propose adding a transitive scale to the standard definiteness, animacy, and person scales, which are also usually active in these systems. This allows the successful Principles and Parameters approach to syntactic case assignment to remain just as it is for nominative, ergative, and for three-way systems. Ergative languages are often said to exhibit a particular construction, sometimes considered to be the ergative version of the passive construction, known as the antipassive. Polinsky, in her contribution (Chapter 13), examines this construction across languages (e.g. Chukchi, Diyari, Labrador Inuit, Warlpiri), summarizing its properties and key approaches. She defines it as a construction where the logical object of a predicate is not realized as a direct object but as a non-core argument or is left unexpressed. She demonstrates various realizations including some less typical, such as (pseudo) noun incorporation and bi-absolutive constructions. She argues that there are interpretative effects, but that none is a defining factor across all antipassives cross-linguistically. She shows that antipassive and passive constructions are not mutually exclusive, and that antipassive is not limited to ergative languages, though it is more noticeable in such languages. Another proposed characteristic property of ergativity is word order; in particular, it has been claimed that SVO order and ergativity do not coincide. Taraldsen (Chapter 14) examines this generalization, demonstrating that SVO can be derived in a multitude of ways, as can ergative case marking. He questions whether the generalization holds of all these possible derivations, and argues that we would expect counterexamples, hopefully within certain types of well-defined languages. He examines tripartite and neutral languages with ergative agreement patterns but he finds that no conclusion can be drawn, due to lack of data. The chapter also examines key proposals about ergativity, pointing out necessary modifications in order to account for the word order restriction. The chapter richly illustrates the complexity involved in developing detailed analyses of broad generalizations.
Introduction 9
1.3.2 Extensions In the second section of Part II, our familiar view of ergativity as a sentential argument indexing system is extended to allow for consideration of the role of ergativity in other domains such as nominalizations, voice systems, information structure, and speech act theory. The authors of the four chapters in this section take different positions on how and whether these extensions can be posited. Alexiadou argues that ergativity is linked to nominalization, and Erlewine, Levin, and van Urk argue against a currently widespread view that Austronesian-style voice systems are an expression of ergativity. Johns and Kučerová argue that ergative patterns stem from structural properties of information structure, while Wiltschko argues that at the thematic level, ergative patterns are basic, and that they can be extended into speech act theory. In the following paragraphs, we outline each of these chapters. It has been noted that ergativity is related to nominality, both because nominalizations often exhibit ergative case, and because verbs in ergative languages seem to exhibit fewer verbal properties than verbs in nominative systems. Alexiadou explores these issues in Chapter 15, noting that many authors have attributed ergativity to the presence of a defective v or Voice head, which yields a more nominal clause. She observes that ergativity only arises in a subset of nominalizations in languages that have more than one nominalization pattern, and that these are cases that contain an n head. She argues that n-based nominalizations allow only one structural case and do not contain an external argument. She also includes a discussion on the nature of unergative subjects. In her chapter, ergativity is related to characteristics that enable it to extend to other construction types. In Chapter 16, Erlewine, Levin, and van Urk examine recent extensions of ergativity to Austronesian voice system languages. They begin by reviewing and critiquing ergative analyses of voice systems as in Tagalog, Malagasy, and Atayalic languages, and they bring in new data from Balinese and (non-Austronesian) Dinka. These languages have similar voice systems to the other languages, but they do not exhibit ergativity, thus they demonstrate the necessary dissociation of these two phenomena. They argue that there must be mechanisms other than ergativity that yield the behavior of Austronesian-style voice systems. Their chapter thus suggests that there are limits on extensions of ergativity to explain other grammatical phenomena. Johns and Kučerová show in Chapter 17 that there is variability in the presence of object agreement in the ergative–antipassive alternation in the Inuit language. They argue that this is related to information structure, and, given this, the case and agreement patterns fall out from familiar principles. They propose that absolutive object “agreement” is in fact cliticization, and that such cliticization is tied to the fact that absolutive objects are always “aboutness” topics. Such topics must be at the edge of a phase in order to be assigned a referential address, and this affects the locality relations of the arguments, yielding an ergative pattern. They also touch on dialect variation across the Inuit languages. Their chapter thus raises a new perspective on the nature of case splits.
10 Jessica Coon, Diane Massam, Lisa Travis Wiltschko argues in her programmatic contribution (Chapter 18) that ergativity extends beyond the familiar argument structure domain into the domain of speech act structure. She argues for the existence of this domain and shows that, as with argument structure and case structure, ergativity is a possible and indeed, expected pattern at this level. She overviews proposals about speech act structure, touching on assertions, imperatives, and presentatives, and on particles such as eh, and German jo. She argues that speech act structure consists of a grounding layer and a response system layer, and that ergative constellations can be detected in each of these.
1.4 Part III: Approaches to Ergativity The chapters in Part III show how diachronic, acquisition, and experimental work can probe data points and theoretical questions in ways that can both complement and support the work reported on in other parts of this volume—and is divided into three sections, accordingly. Recurring themes in these chapters involve issues such as the amount of variation that is found in the instantiation of ergativity and the possibility of reanalysis and/or grammaticalization of structure. There is the basic question of how closely-related languages can come to have very different grammatical systems, thereby raising further questions concerning how languages change, what the influence of language contact is, and what parts of language are susceptible to reanalysis. Diachronic work takes these puzzles as the starting point. But questions of language change and reanalysis lead to questions about acquisition. Acquisitionists explore what might be subject to reanalysis, what might be a default setting for a parameter, what triggers are salient, and what structures are learnable. Experimental work outside the domain of acquisition looks at related areas where similar questions are investigated, such as what systems are more easily processed, and what elements in the linguistic string aid intelligibility. Specifically, in the context of ergativity, we can ask whether ergative systems are stable or are prone to reanalysis; how one arrives at an ergative system; what the paths out of such a system are; and whether there is any evidence that an ergative system is either more or less complex than a nominative system.
1.4.1 Diachronic A striking characteristic of ergativity is how differently it may present itself from language to language. This is particularly noticeable where microvariation appears within language families. In this section, six chapters tackle the problem of variation by investigating paths of change. Several different types of focus are evident in these chapters— variation vs. commonality and description vs. theory. As more and more details about variation are uncovered (see also Part IV, the case studies, which we discuss shortly), the
Introduction 11 puzzle of what we mean by ergativity and what a theoretical account for ergativity might be becomes more complex. Some of these chapters stress the fact that the paths to ergativity are more varied than previously thought (McGregor, Haig), others try to reduce the number of possible paths (e.g. Aldridge). Without getting down to the details of specific mechanisms, a larger question can be raised as to whether change to an ergative system can be tied to a shift of one language-particular characteristic, or whether a general flavor of ergativity is constructed by coinciding but logically independent changes. It is no surprise that these issues appear in Parts I and II of this volume as well when discussing what the parameters of ergativity are and how the particular characteristics of ergativity are accounted for. McGregor (Chapter 19) traces the creation and loss of ergative case morphemes cross-linguistically, arguing that the range of sources for this case marker is wider than what is often assumed. He outlines and evaluates various proposals in the literature, critiquing the viability of lexical sources, but giving multiple examples where markers of other cases, indexical items, and directional elements have been reanalyzed to produce ergative case morphemes. He also discusses instances where ergative case markers themselves are reanalyzed as other case markers or grammatical categories. In the final section of the chapter he discusses the role of language contact in the development of ergative case markers and ergative systems. Haig focuses on ergativity in Iranian languages in Chapter 20. He introduces three case studies, Kurdish, Balochi, and Taleshi, to illustrate the extent of the micro variation of ergativity within Iranian languages. He focuses on the path of the emergence of these systems, supporting the claim that this micro-variation stems from independent changes in interrelated subsystems such as case, agreement, and pronominal clitic systems. These findings result in raising doubts for any proposal that ergativity is best represented by a monolithic alignment parameter. Aldridge (Chapter 21) takes a different tack from the previous two chapters, emphasizing what characteristics paths to ergativity might share. Couched in a generative syntax framework, she explores data from several languages and language families, e.g. Indo-Aryan, Iranian, Inuit, and Austronesian, arguing that ergativity is derived from nominal structures. She examines two cases (Indo-Aryan and Inuit) where it has been claimed that the ergative structures came from earlier passives but advances an argument that even in these cases, the ergative was originally a possessive, supporting the hypothesis of a nominal base. The next three chapters focus on two language families that have been introduced in Aldridge’s chapter, but offer different viewpoints. Butt and Deo (Chapter 22) take a close look at four stages in the rise and fall of ergativity from Early Old Indo-Aryan to New Indo-Aryan, starting with the development of ergativity from participial constructions. Within New Indo-Aryan, they describe three major innovative patterns. In Hindi and Nepali, ergative case marking is strengthened with new morphology, in Bengali and Oriya, both ergative case and ergative pattern agreement is lost, and in Marathi and Gujarati, ergative agreement remains in spite of complications, such as surface syncretism of morphology and differential subject case marking.
12 Jessica Coon, Diane Massam, Lisa Travis Kikusawa (Chapter 23) uses the Comparative Method to reconstruct the direction of change in various Austronesian languages and to explain the typological diversity found within this language family. More specifically, she outlines three paths of change: (i) the shift from a morphologically marked ergative system to a fixed word order voice system (Ibaloy, Pendau); (ii) the development of an accusative clitic system (Tongan, Samoan); and (iii) the development of a system of lexically marked NPs that can be analyzed as ergative (Tongan) or accusative (Maori). In Chapter 24, Kaufman focuses on Western Malayo-Polynesian languages. He starts by comparing three theoretical accounts of Tagalog, posing problems for two of them— an ergative analysis and a case agreement analysis. He argues rather that Tagalog is a symmetric language where predicates are nominal rather than verbal. In order to support this account, he compares the structure of Tagalog to that of another Western Malayo-Polynesian language, Mamuju, a canonically ergative language. Kaufman shows how ergative structures found in Mamuju, as distinct from the structures of Tagalog, are developed through the reappearance of verbal predicates.
1.4.2 Acquisition It is difficult to talk about language change without invoking questions about language acquisition. The next three chapters raise many of the relevant questions for the acquisition of an ergative language, the answers for which have an impact on how language change should be viewed. As is pointed out, ergative languages are often split systems creating complex input for the language learner, which makes this a particularly interesting and informative field of study. The problems addressed include issues of methodology, the status default cases, and the use of acquisition data to support theoretical claims. Bavin (Chapter 25) outlines various issues that arise in the study of the development of an ergative system in child language. By summarizing studies from the literature representing a range of languages and language families, she highlights several possible confounds in the input data that could create problems for acquisition. These include split systems, multiple uses for the same case marker, and the contribution of pragmatic function to the choice of construction. She also discusses potential hurdles such studies face, such as drawbacks in using naturalistic data and potential ambiguities in the acquisition data. In spite of this, the cross-linguistic data show similar results of timely successful acquisition with very little overgeneralization of ergative case marking. Austin (Chapter 26) presents data from previous studies on the acquisition of the verbal morphology and case in Basque. She shows that children resort to a default morphological system where forms that encode fewer features substitute for more complex forms, for example absolutive case is produced rather than ergative or dative. Austin argues that this repair strategy is not surprising given a Distributive Morphology analysis of Basque morphology. In a morphological system where morphemes compete to
Introduction 13 realize a set of features, the notion of “best fit” will ensure that a less marked morpheme will appear in instances where the more complex form has not yet been acquired. Pye and Pfeiler (Chapter 27) use acquisition data to probe the status of person marking in Mayan languages by comparing the acquisition of nominative person markers in French (clitics) and Spanish (agreement) with the acquisition of both absolutive and ergative person marking in four Mayan languages: Wastek, Yukatek, Ch’ol, and K’iche’. The acquisition of the ergative person marking in all four Mayan languages followed neither the French nor the Spanish pattern; while the acquisition of the absolutive person marking produced mixed results. There were insufficient Ch’ol data, but the acquisition patterns of Yukatek absolutive personal marking were similar to those of Spanish agreement. Those of Wastek and K’iche’, however, differed from both those of French clitics and Spanish agreement markers. They argue that the grammaticalization of person markers as determined by their specific combination of clitic and affix properties predicts children’s production of the person markers more accurately than their categorical status as absolutive or ergative.
1.4.3 Experimental Experimental work on ergative languages is relatively new but clearly very important to our understanding of ergativity as a typological language category and as a theoretical construct. The existence of ergative languages raises questions concerning markedness, ease of acquisition, and ease of processing. Whatever the answers to these questions may be, it is clear that no universal pronouncements about language can be made without including data from ergative languages. Both chapters in this section give overviews of experimental research on ergative languages. They discuss the methodologies used, the importance of the work, and some ideas for future research. Zawiszewski (Chapter 28) presents an overview of current experimental studies on ergativity using a variety of methodologies (self-paced reading, ERP, fMRI, grammaticality judgments) on a variety of languages (Basque, Hindi, Avar). After an introduction of the different experimental methods, he summarizes the studies and shows how they can be used to further probe the results from earlier experiments on nominative–accusative languages investigating issues such as the distinction of syntactic vs. semantic processing, subject–object asymmetries, and the effect of L1 and L2 acquisition. He concludes with a discussion of the overall results and directions to be explored further. Chapter 29 is a review by Longenbaugh and Polinsky of recent experimental work testing ergative-specific questions involving alignment, long-distance relations, and agreement in a range of languages including Hindi, Basque, Niuean, and Avar. They stress the importance of doing experimental work on ergative languages to resolve some confounds that are found in the existing literature on accusative languages. More specifically, the alignment of grammatical case with grammatical function can be teased apart in ergative languages. Their chapter ends with a suggestion that further experimentation can probe the heterogeneous nature of ergative languages.
14 Jessica Coon, Diane Massam, Lisa Travis
1.5 Part IV: Case Studies This section of the book contains sixteen case studies of a range of languages from a range of language families. While the approaches of the authors and the scope of the studies vary considerably, the common threads that have appeared throughout the book reappear here. On one hand, ergative languages vary from one another to such an extent that one might suspect that they have no unifying feature, yet they also evidence enough recurring features to confirm their membership in the class. There are SOV, VOS, VSO, and SVO exemplars. Both morphological and syntactic ergativity are explored, and a variety of types of ergative systems are outlined in families such as Nakh–Daghestanian, Tibeto-Burman, and Kartvelian, as well as some apparently emergent systems in African languages. Several languages that are discussed exhibit both ergative case and ergative agreement systems, leading to insights about the relation between case and agreement, and to the relation between ergativity and features such as animacy, gender, number, and person, as well as tense and aspect. Most of the languages explored have split systems—some sensitive to aspect, some to person, while some are Split-S systems, in which different types of intransitive verbs are marked differently for case. The relation between ergativity and other constructions is also explored, focusing on constructions such as the antipassive and control structures, relative clauses, coordination, and non- finite sentences and nominalizations. Aissen (Chapter 30) examines ergative characteristics of Mayan languages, with a focus on constraints on extraction. Like some of the other languages discussed in this volume, some (but not all) Mayan languages restrict the extraction of ergative subjects, which Aissen calls the Ergative Extraction Constraint (EEC). In this contribution, Aissen reviews the empirical facts and discusses two main approaches to the EEC in the recent literature on Mayan languages: (i) a Case-based approach, in which restrictions are attributed to abstract Case assignment configurations; and (ii) a morphosyntactic approach which attributes extraction asymmetries to special morphology, in particular the “Agent Focus” morphology used when transitive subjects are extracted. Through a detailed look at a range of constructions, Baker (Chapter 31) provides a dependent case analysis of ergative case in Burushaski, a language of Northern Pakistan. To understand the distribution of ergative marking, Baker investigates three environments in which the canonical ergative pattern of the language disappears: (i) verbs with two absolutive arguments; (ii) verbs with an ergative argument and a dative argument; and (iii) future-tense clauses which permit absolutive transitive subjects. Baker argues that the syntax of each of these constructions is more complex than surface appearances show, lending support to the proposal that ergative case is assigned only when one NP (the ergative) c-commands another NP in the same local domain. Berro and Etxepare (Chapter 32) on ergativity in Basque, explore an ergative system that is manifested by both case and agreement morphology. They provide a thorough and detailed overview of the case and agreement systems in Basque and of their
Introduction 15 interaction across both the nominal and verbal inflectional systems. They also demonstrate how ergative marking interacts with other systems such as number, person, and tense. They present a cross-dialectal study of the marking of ergative case on subjects of unergative predicates, which has been referred to as Split-S system, while critiquing some of the claims that have been made about this system, such as the positing of implicit objects and light verb structures. They discuss claims that have been made that ergative is an inherent case linked to causation, by considering a range of construction types, including nominal and adjectival predicates, perception verb complements and raising verbs. In their closing section they discuss the notion of “marked case” in relation to case marking in Basque. Butt (Chapter 33) gives an overview of ergativity in Hindi/Urdu but crucially sets the Hindi/Urdu facts against a background of other South Asian languages such as Nepali, Gujarati, Marathi, and Bengali. She highlights the range of variation and also details the different roles that differential case marking plays in each of the languages. Butt argues that the variation in the behavior of case and agreement in these languages, and the variation that role of differential case marking in agreement patterns, makes a tight link between case and agreement difficult to maintain. Compton (Chapter 34) focuses on how ergativity is realized both in morphological case marking as well as in the rich agreement system of the language. After reviewing basic characteristics of Inuktitut, Compton discusses the various approaches to ergative and absolutive case assignment in the literature. Finally, he turns to antipassive constructions and their relationship to Differential Object Marking and aspect. Forker (Chapter 35) surveys the range of ergative alignment patterns found in the Nakh–Daghestanian (or East Caucasian) language family, concluding that the main correlates of ergativity in these languages are morphological. In particular, Forker discusses the system of gender and person agreement on verbs and the morphological case marking found on nominals. Biabsolutive constructions—in which both A and P arguments are marked absolutive—are reviewed, as well as valence-changing operations (causative, antipassive). Forker also provides an in-depth discussion of control constructions, noting that there is a general tendency for syntactic accusativity in this domain. Kahn (Chapter 36) focuses on ergativity in Neo-Aramaic. He organizes and presents a complex set of patterns of ergativity in modern spoken form of Aramaic (Neo-Aramaic) split into four subgroups: Western, Turoyo, Northeastern, and Mandaic. Khan discusses the nature of split ergativity evidenced in the patterns of verbal suffixes across a number of dialects, which are described in detail. He argues that the influence of Iranian languages on Eastern Aramaic explains both why Neo-Aramaic differs from other Semitic languages in its expression of ergativity and the non-canonical type of ergativity that it displays. While African languages are generally left out of any discussion of ergativity, König (Chapter 37), describes ergative patterns that appear in the West Nilotic family of Nilo- Saharan, in particular Shilluk. She points out four features particular to ergativity in African languages—marked nominative, no-case before the verb, OVA word order, and its relationship to pragmatically marked word order. She argues that ergative case
16 Jessica Coon, Diane Massam, Lisa Travis developed in these languages through reanalysis of either determiners, genitive markers, or prepositions. Chelliah (Chapter 38) surveys morphological case marking in several representative languages of the Tibeto-Burman family: Dolakha Newar, Chintang, Tibetan, Meitei, and Burmese, using these to demonstrate four possible alignment patterns for core arguments. The first two languages exemplify a typical ergative alignment pattern. In Tibetan, there is a general pattern of ergativity but one in which transitivity factors influence whether the transitive subject receives ergative marking. Burmese shows an accusative alignment in which information structure (topicality, contrastiveness, and theticity) influences when “subject marking” occurs. Finally, Chelliah discusses Meitei, demonstrating that it falls somewhere between Tibetan and Burmese insofar as both transitivity and information structure considerations affect the marking of core arguments. Laughren (Chapter 39) focuses on the ergative in Warlpiri, and examines what has been claimed to be a morphologically ergative case system in a syntactically nominative– accusative language. Laughren begins with an overview of ergativity in Australian languages, then focuses on Warlpiri, which has certain verbs which take ergative subjects in finite clauses and other verbs which take unmarked or absolutive subjects. This chapter examines the distribution of the ergative morpheme, including on body parts and circumstantial adjuncts, and the functions of the ergative DPs in both finite and non-finite clauses, with a focus on the relation between subject marking and instrument marking. Otsuka (Chapter 40) demonstrates that Tongan has an ergative pattern in both morphology and syntax, but that this pattern is not consistent throughout the language, as nominal morphology is split between clitic pronouns and other nominals. There are three syntactic manifestations of ergativity in the language: relativization strategies, coordinate reduction strategies, and anaphoric antecedence patterns. Interestingly, these cannot be accounted for in a unified manner, and Otsuka argues that the first two are in fact PF phenomena. She claims that this necessitates a view of ergativity as a construction-specific rather than a language-specific phenomenon. Peterson (Chapter 41) demonstrates that the Tsimshianic languages have fully ergative agreement systems. Although there are splits, conditioned by clause type and person hierarchy, all sides of the splits exhibit ergativity. He describes the agreement patterns across the family, including a discussion of connectives, which are determiner- like particles that appear to contribute a further split. The more conservative languages are purely ergative, while other branches also exhibit transitive, contrastive and neutral alignments. He considers all to be expansions of ergativity, since A and S are never grouped together. Queixalós (Chapter 42) presents a detailed examination of alignment, and grammatical relations more generally, in the Amazonian language Katukina-Kanamari (KatKan)—a language which Queixalós describes as “remarkably suited for raising pivotal issues on grammatical relations.” KatKan is shown to have two patterns of bivalent clauses: ergative and accusative. The latter, Queixalós shows, is more highly restricted in its distribution, and is found with unindividuated patient arguments. Queixalós’
Introduction 17 contribution includes a thorough survey of the empirical facts surrounding the two types of construction, as well as more general discussion of the interactions among grammatical roles, argument structure, and alignment. Salanova (Chapter 43) describes the distribution of ergative case marking in Jê languages in general, and Mẽbengokre more specifically. In these languages, the link between the ergative constructions and nominalization is clear, where the subject DP is marked with a postposition when it occurs with the nominal/adjectival form of the verb. Further, he shows that the use of the nominalized structure is pervasive, appearing not only in embedded contexts, but in independent clauses as well depending on other considerations including aspect and the presence of post-verbal modifiers. Schultze-Berndt (Chapter 44) tackles the problem of a system where ergative case- marking appears to be optional, alternating with zero-marking, and, less frequently, ablative case. She describes several factors that influence the choice, factors which include animacy, verb class, tense/aspect, and information structure. Schultze-Berndt shows that factors that categorically determine morphological marking in other languages show up only as tendencies in Jaminjung, connecting it to differential case- marking of subjects. Finally, Tuite (Chapter 45) traces the history of linguistic accounts of Georgian ergativity from the seventeenth century. This history is followed by a detailed description of the different case and agreement patterns found in Georgian, as well as in Laz, Mingrelian, two members of the Zan branch of the family, and in Svan, an outlier. Tuite further outlines the role of verb classes in determining these patterns. Once the present variation has been established, an overview is given of case, agreement, verb classes, and morphosyntax of Proto-Kartvelian.
Acknowledgments In concluding this introduction, the editors would like to acknowledge the valuable help of several people. We would like to thank Crystal Chen, Patrick Murphy, and Rebecca Tollan for editorial assistance, Justin Royer for help with proofing, and Hisako Noguchi for work on the index. We would also like to thank the editorial team at Oxford University Press, in particular Vicki Sunter and Julia Steer, who has provided invaluable help and advice throughout the process of producing this volume. Finally, we would like to thank all the authors for their work not only on their own contributions, but also in the reviewing process.
PA RT I
AC C OU N T I N G F OR E RG AT I V I T Y
Representing Ergativity
Chapter 2
Ergativit y in di s c ou rse and gramma r John W. Du Bois
Every language provides its users with systematic ways of organizing the core arguments of the clause, establishing a more or less stable and consistent framework for the foundations of its grammar. Remarkably, languages differ even in this most basic level of structural organization. Yet certain configurations of arguments tend to recur, emerging again and again in the grammars of the world’s languages. For syntactic alignment what proves pivotal is how the grammar treats the sole argument of a one-place predicate (S), aligning it with one or the other argument of a two-place predicate, with respect to case- marking, agreement, word order, extraction, and so on. Some languages treat the S like the object of a two-place predicate (O), yielding ergative alignment (S=O vs. A), while others treat S like the subject of a two-place predicate (A), yielding accusative alignment (S=A vs. O). Still other languages are sensitive to the semantic variability inherent in the population of one-place predications, aligning the more agent-like subset of S (Sa ) with A, and the more patient-like subset of S (So) with O, giving active alignment (Sa=A vs. So=O). Yet languages are not simply ergative, accusative, or active. There is great diversity, as well as convergence, across the world’s languages with respect to the various systematizations of basic grammatical relations (Silverstein 1976; Comrie 1978, 2013a, 2013b, 2013c; Dixon 1979, 1994; Mithun 1991; Malchukov 2005; Bickel & Nichols 2009; Bickel 2010; Siewierska 2013). This diversity can penetrate into the grammar of a single language, where a mix of distinct alignments is often found in different parts of the same grammar (Silverstein 1976; Comrie 1978; Mithun 1991; Coon 2010a, 2013a; Malchukov 2014; see also Coon and Preminger, Chapter 10 in this volume). Accounting for diversity and convergence within and across languages remains a compelling yet elusive task for linguistics (Evans & Levinson 2009). Argument structure configurations (Goldberg 1995) represent the very foundations of the clause—what may be considered the “basic body plan” (Mayr 2001) of the utterance. No theory of grammar can be considered explanatory without contributing to an understanding of how such a diversity of basic
24 John W. Du Bois plans could have emerged in the world’s languages. Yet the challenge of accounting for structural variability at a foundational level has proved baffling, such that many leading linguists have postponed the day of reckoning with ergativity. Fillmore considered his principles of subject selection universal, “given certain qualifications for the interpretation of ergative systems” (1977: 61). Dowty acknowledged that “argument selection in ergative languages” (1991: 581) was relevant to his proto-role model, but invoked an “inverse” model of ergativity which “means in effect treating the transitive ‘Patient’ as a grammatical subject and the transitive ‘Agent’ as analogous to an object” (1991: 582). Ergative languages are said to employ the same proto-agent and proto-patient roles as in accusative languages, but “merely reverse the syntactic association” with subject and object (1991: 582). Often it seems that ergativity is taken up only after commitments to basic theoretical assumptions are set (Ackerman & Moore 2001: 1, fn. 1). But ergativity is unlikely to reveal its secrets to those who approach it superficially or too late, whether with afterthoughts or mere mirror-image models. This chapter explores the connection between the well-known ergative pattern in grammar and a pattern in discourse that is isomorphic to it, with the goal of providing a functional explanation for ergativity. The ergative discourse pattern holds the key to the grammaticization of ergativity, perhaps—or is just a piece of the larger puzzle. The specific approach presented here is known informally as discourse and grammar (Givón 1979; Hopper & Thompson 1980; Thompson & Mulac 1991; Du Bois 2003b, 2014), which seeks to understand grammar in light of discourse, and discourse in light of grammar. Patterns of grammatical form are linked to communicative functions on the evidence of naturally occurring language use, in order to shed light on why grammars are the way they are. Crucially, “grammars” is plural, inviting attention to typological diversity. Not only must the broad alignment types of ergative, active, and accusative be distinguished, but also such cross-cutting typological features as head-marking vs. dependent-marking, optional vs. consistent, aspect-based vs. person-based splits, and so on (DeLancey 1981; Nichols 1986; Garrett 1990; Bickel & Nichols 2009; Malchukov 2014; Nichols & Bickel 2013a; van de Velde 2014). The discourse-and-grammar approach accords equal importance to discourse and to grammar. In this chapter, however, I will devote more space to the discourse side of the equation. This is feasible because ergative grammar is well documented in the literature (including this volume); it is necessary because ergative discourse is not. That said, a key task will be to bring together the facts of ergative grammar and ergative discourse, and to clarify the connection between them. If ergativity is seen as a problem, it’s one that is not going away any time soon. That may be a good thing for linguists, who have a lot to gain by taking up the challenge of explaining ergativity. But one group for whom ergativity has never been a problem is the speakers of ergative languages. Ergative speakers do just fine, learning their language with equal ease (Ochs 1982; Narasimhan, Budwig, & Murty 2005; Bavin & Stoll 2013; Brown et al. 2013; Pye et al. 2013), and using it to perform the full range of functions that every language serves (de León 1999, 2000). Linguists seeking explanations might take a cue from the language users, and treat ergative
Ergativity in discourse and grammar 25 grammar as a system that works. The idea is to see ergativity as the solution, and ask what the problem is. This is not to suggest that there will be easy answers, nor that every aspect of ergative grammar will be transparently motivated or directly functional. Nonetheless, how a grammar works for its users is one of the most productive questions a linguist can ask. The view that grammars solve problems in new and creative ways accords well with the approach advocated by Evans and Levinson (2009), who, while arguing for massive cross-linguistic diversity, also maintain that languages tend toward “evolutionarily stable strategies,” representing “recurrent solutions across time and space.” These strategies result from myriad interactions between communicative, cognitive, and processing constraints which reshape existing structures through use. A major achievement of functionalist linguistics has been to map out, under the rubric of grammaticalization, the complex temporal sub-processes by which grammar emerges as frequently used patterns sediment into conventionalized patterns (Bybee 2000; Givón 2009). (Evans & Levinson 2009: 444–445)
From the discourse-and-grammar perspective, the enterprise begins in discourse with the search for “frequently used patterns”; it continues by showing how the “recurrent solutions” resolve universal functional needs; and it ends, if successful, by elucidating the emergence of the “conventionalized patterns” known as grammar—including the grammar of ergativity and its competitors. One seeming paradox is that grammar is already present in discourse from the start. Discourse is never found without its grammatical clothing; but by the same token, grammar is never realized except in discourse. A basic task for discourse-and-grammar research, then, is to tease discourse and grammar apart. With ingenuity and a little typology, the problem is solvable, as will become clear. This then sets the stage for investigating the connection between the respective ergative patterns in discourse and in grammar. I begin (2.1) with a look at a stretch of discourse in an ergative language, identifying a recurrent pattern which is isomorphic with the ergative–absolutive pattern of grammar. The next section (2.2) documents this pattern quantitatively as the ergative discourse profile, based on evidence from a typologically diverse array of ergative, active, and accusative languages. I propose that the ergative discourse profile is shaped by a set of soft constraints known as Preferred Argument Structure (Du Bois 1987b, 2006; Du Bois, Kumpf, & Ashby 2003). The next section (2.3) explores whether the ergative discourse profile represents a discourse universal, examining evidence from child language, typology, genre, and diachrony. In the following section (2.4) I take up the functional explanation for ergativity, having introduced the analysis of competing motivations (Du Bois 1985, 2014; MacWhinney, Malchukov, & Moravcsik 2014; Malchukov 2014) that favor either ergative or accusative alignment. Next (2.5) I respond to some common objections to the discourse explanation for ergativity. Finally, I present some directions for future research (2.6) and conclusions (2.7).
26 John W. Du Bois
2.1 Ergativity in Discourse To understand ergativity, it is important to look at how discourse connects to grammar and to meaning. As Dixon says The most important task for future work on ‘why some languages are ergative in a certain way and others are not’ is to investigate the semantic and discourse-pragmatic makeup of each of a sample of languages, and study the way in which this determines (or partly determines) its grammatical profile. (1994: 219–220)
If discourse has the power to affect grammar, it is because “discourse, clause structure, and verb semantics are all intimately interwoven” (Foley & Van Valin 1984: 373). But to move beyond broad generalities about the interdependence of form and function, it is necessary to tease apart these three forces, if only to show how they come together again to shape the fundamental grammar of the clause. To make good on the promise that “recurrent patterns” shape grammar (Evans & Levinson 2009), it is necessary to do the empirical work to document the specific “discourse profiles” (Du Bois 2003a: 40–44) that are linked to the grammatical constructions of interest, and their functional niches. The relevant work on discourse profiles focuses on “discourse inside the clause” (Du Bois 2003a: 13; 2003b: 83), seeking to identify the distinctive discourse correlates of the clause, its arguments, and other aspects of structure. While researchers sometimes speak broadly of the discourse profile of a language, akin to whole-language typology (Nichols & Bickel 2013b), it is often more useful to target the discourse profile at a more specific level—on a par with a specific argument structure construction, for example. Thus one can ask about the discourse profile of the intransitive, transitive, or ditransitive clause; or the agentless vs. agentive passive, and so on. One challenge in working with discourse is its evident variability, born of freedom. Seemingly, speakers exercise the absolute liberty to construct their utterances as they wish, within the broad limits circumscribed by the rules of grammar. The result appears, to some, as unpredictable variability. Yet a closer look reveals recurrent regularities in discourse, including some which are not strictly required by any grammatical rule. Understanding ergativity depends on sorting out how argument structure constructions balance the multiple demands of information processing, anaphoric reference, topic continuity, event semantics, and other factors, including the inherited morphosyntax of the language at hand. To make these matters more concrete, it will be useful to examine a sample of discourse from an ergative language. The following narrative is in Sakapultek, an ergative, head-marking, verb-initial language of the Mayan family, spoken in highland Guatemala (Du Bois 1981). The narrative was produced as a telling of a short film, the Pear Film (Chafe 1980; Du Bois 1980). To highlight the grammatical elements that matter for the discourse profile, the following conventions are used: underscore represents a
Ergativity in discourse and grammar 27 referent mention marked by a reduced referential form (pronoun, agreement, or zero);1 boldface indicates a lexical noun phrase; italics in the free translation indicates a verb or preposition (head of its phrase). The distribution of lexical vs. reduced forms is also indicated schematically, with capital letters (A S O) indicating a lexical noun in the designated grammatical role, while small letters represent a reduced form (a s o). A clause- by-clause analysis of this narrative’s ergative discourse profile is published elsewhere (Du Bois 2006); here I present just the first 13 lines. (1)
Pear Story (Sakapultek)
1
… (H) Ee x-Ø-inw-il-anh, foc cp-3.abs-1.erg-see-tv … (H) What I saw was,
oaV
2
.. x- Ø- aq’an jun achenh, cp-3.abs-ascend one man a man climbed up,
VS
3
.. ch-u’ chee’, at-top tree .. in a tree,
PX
4
… (H) x-Ø-a-r: … ch’up- o’ niky’aj péera-s. cp-3.abs-lat-3.erg pick-dep some pear- pl … (H) he went and … picked some pears.
aVO
5
… Tik’ara’ Ø-Ø-qaaj-uu: l, then cp-3.abs-descend-hither … Then he came dow: n,
sV
6
… Ø-Ø-r- su’ r- iij juu: n, cp-3.abs-3.erg-wipe 3.erg-back one … he wiped the surface of one,
aVO
7
… (H) ii despwee: s, and then … (H) and the: n,
8
… (H) x-Ø-r-ya’2 qaj p l chikech, cp-3.abs-3.erg-put down at the basket … (H) he put it in the basket,
..
oaV PX
1 Note that this annotation focuses on referent mentions and how they are expressed. Thus a referent expressed with a lexical noun phrase plus a cross-referencing affix in the same clause is treated as one mention, not two (Du Bois 1987b: 813). Here it always the heavier form (noun phrase) that is marked (with boldface). 2 The underlying r-is devoiced in this phonological environment, coalescing with preceding voiceless fricative x-and effectively disappearing in the surface form, yielding xya’ ‘he gives it.’
28 John W. Du Bois 9
Ø-Ø-r- alsa- aj p l r:-… m komo ber gabaacha. oaV PX cp-3.abs-3.erg-remove-tv at the 3.erg um like see apron he removed it from his: … um like apron.
10
… Despwees tik’ara’, then then … Then,
11
… Ø- Ø- pee jun aj-laab’, cp-3.abs-come one dim-boy … a little boy came,
VS
12
.. ch-ij bisikleeta, at-back bicycle .. on a bicycle,
PX
13
… (H) xaq x-Ø-a-r"- k’am- a’ jun chkech peera. just cp-3.abs-lat-3.erg-take-dep one basket pear … (H) he just came and took a basket of pears.
aVO
Viewed in grammatical terms, the data exhibit the hallmarks of ergative alignment in the grammar of verbal agreement (pronominal clitics). Focusing on third person singular referents, transitive subjects (A) are cross-referenced with r-‘3rd person singular ergative’ (lines 4, 6, 8, 13). In contrast, intransitive subjects (S) are unmarked, i.e. cross- referenced with Ø-‘3rd person singular absolutive’ (lines 2, 5, 11). Transitive objects (O) receive the same treatment as S (lines 4, 6, 8, 13). The ergative–absolutive pattern holds throughout the agreement paradigm (Du Bois 1981, 1987a: 210; 1987b: 809–810), as in virtually all Mayan languages (Larsen & Norman 1979; England 1983; Robertson 1983; Kaufman & Norman 1984; Law 2009). Viewed in functional terms, the data illustrate some common patterns in the realization of basic discourse functions. For example, a new human referent is introduced using a full lexical noun phrase in the S role (lines 2, 11), and is subsequently tracked through the discourse with reduced forms (e.g. zero). The tracking of the most topical referents occurs most often in reduced A’s (lines 4, 6, 8, 13), but also in a reduced S (line 5). Inanimate referents are introduced here with a lexical mention in O role (lines 4, 6, 13), and tracked in subsequent discourse using a reduced O (lines 8, 9). Locative prepositional phrases also serve for the introduction of lexical and New inanimates, which may occur as adjuncts in separate intonation units (lines 3, 12), or in more tightly integrated verbalizations within the same intonation unit as the verb (lines 8, 9). The point of this exercise is to show what discourse looks like when viewed through the lens of grammar. To generalize from this tiny sample, lexical mentions occur mostly in absolutive argument positions (S and O roles), expressing New information. A’s are mostly reduced forms, and express given information. In this verb-initial language, the preferred order of overt lexical nouns is VO and VS (in the four-way typology of Dryer 1997, 2013a, 2013b), which can be generalized as V-Lex word order: verb followed by a
Ergativity in discourse and grammar 29 lexical noun.3 The attentive reader will have noticed that the discourse distribution of lexical arguments (and of new information) corresponds to the absolutive category in the grammar of ergative languages, while topically continuous elements are found in what would be the subject in accusative languages. The latter reflects, perhaps, the common wisdom that subject is a grammaticization of agent and topic (Givón 1983a; Comrie 1988). The ergative and accusative discourse patterns coexist in one and the same stretch of narrative, in a language whose grammar is basically ergative. The tension between these two discourse patterns will prove fruitful for understanding the role of competing motivations (Du Bois 1985, 2014) in the discourse motivation of ergativity (1987a; Du Bois 1987b, 2006); see section 2.4. But we are getting ahead of ourselves. First we must ask: Do these observations characterize the discourse of one speaker, or do they represent something broader? This calls for a quantitative perspective, which is addressed in terms of discourse profiles.
2.2 The Ergative Discourse Profile The first systematic study of information structure to distinguish A, S, and O was by Du Bois (Du Bois 1981, 1985, 1987a, 1987b, 2003a, 2003b, 2006; Du Bois, Kumpf, & Ashby 2003). Previous studies had commonly been framed in terms of subjects, documenting contrasts in information structure that naturally seemed to favor accusative languages. But as long as S and A were collapsed within the all-too-familiar subject category, the differences between them remained effectively hidden. Adopting the grammatical terms A, S, O (Dixon 1979) or P (Comrie 1978), originally designed for typologically neutral comparisons encompassing ergative and accusative languages, made it possible to recognize a new kind of pattern in discourse. Early evidence from Sakapultek, illustrated in example (1), revealed a skewed distribution of lexical arguments across A, S, and O. A similar skewing was discovered for New information. Specifically, most lexical mentions occur in absolutive argument positions (S or O), but are avoided in the ergative (A) slot, which is mostly restricted to reduced forms (pronoun, agreement, zero). Correspondingly, most new mentions occur in S or O, with few occurring in A. Many more languages have since been investigated; see Table 2.1 for a selection. In most cases the findings reported tend to fit the generalizations of Preferred Argument Structure. This holds true whether the language is ergative, active, or accusative. Of course, not all scholars agree on the interpretation of the findings; for a discussion of controversial issues, see 2.5. It is useful to distinguish between discourse profiles, understood as observable patterns of linguistic behavior in discourse, and Preferred Argument Structure, understood as the functional constraints that govern them. The ergative discourse profile
3
The implications of V-Lex as a kind of ergative–absolutive word order (Dryer 2013b) are developed below in “Typology” and “Diachrony” (2.3).
30 John W. Du Bois Table 2.1 Selected studies of Preferred Argument Structure Language
Family
Region
Type
Genre
Studies
Sakapultek
Mayan
Guatemala
ergative
narrative (PF)
(Du Bois 1987b)
Mam
Mayan
Guatemala
ergative
narrative
(England & Martin 2003)
Tektiteko
Mayan
Guatemala
ergative
narrative
(England & Martin 2003)
Mochó
Mayan
Mexico
ergative
narrative
(England & Martin 2003; Martin 2003)
Q’anjob’al
Mayan
Guatemala
ergative
narrative
(England & Martin 2003)
Itzaj
Mayan
Guatemala
ergative
narrative
(Hofling 2003)
Tsotsil
Mayan
Mexico
ergative
narrative
(Martínez Álvarez 2012)
Hieroglyphic Maya
Mayan
Guatemala, Mexico
ergative
dynastic history
(Mora-Marín 2004)
Inuktitut
Eskimo-Aleut Canada
ergative
child
(Allen & Schröder 2003)
Hindi
Indo-Iranian
India
ergative
child
(Narasimhan et al. 2005)
Nepali
Indo-Iranian
Nepal
ergative
narrative
(Genetti & Crain 2003)
Hawrami
Indo-Iranian
Iran
ergative
narrative (PF)
(Mahand & Naghshbandi 2014)
Gooniyandi
Australian
Australia
ergative
narrative
(McGregor 1998)
Roviana
Austronesian
Solomon Is.
ergative
monologue
(Corston-Oliver 2003)
Chamorro
Austronesian
Guam
ergative?
narrative
(Scancarelli 1985)
Acehnese
Austronesian
Indonesia
active
narrative
(Durie 1988, 1994, 2003)
Chol
Mayan
Guatemala
active
narrative
(Vázquez Álvarez & Zavala Maldonado 2013)
Yagua
Yaguan
Peru
active
narrative
(Thomas Payne 1993)
Mapudungun
Araucanian
Chile
hierarchical
narrative
(Arnold 2003)
Tohono O’odham
Uto-Aztecan
Arizona
narrative
(Doris L. Payne 1987)
Ergativity in discourse and grammar 31
Language
Family
Region
Type
Genre
Studies
Jarawara
Arauan
Amazon
accusative
narrative
(Dixon 1994: 209)
Japanese
Japonic
Japan
accusative
conversation
(Matsumoto 1997, 2000)
Korean
Koreanic
Korea
accusative
child
(Clancy 2003)
To’aba’ita
Austronesian
Solomon Is.
accusative
narrative
(Lichtenberk 1996)
Finnish
Uralic
Finland
accusative
conversation
(Helasvuo 2003)
Hebrew
Semitic
Israel
accusative
narrative
(Sutherland-Smith 1996)
English
Indo- European
United States
accusative
narrative (PF)
(Kumagai 2001, 2006)
English, diachronic
Indo- European
England
accusative
drama
(Shibasaki 2006)
French
Indo- European
France
accusative
conversation
(Ashby & Bentivoglio 1993)
French, diachronic
Indo- European
France
accusative
epic poetry
(Ashby & Bentivoglio 2003)
Spanish
Indo- European
Venezuela
accusative
conversation
(Ashby & Bentivoglio 1993)
Spanish, diachronic
Indo- European
Spain
accusative
epic poetry
(Ashby & Bentivoglio 2003)
Portuguese
Indo- European
Brazil
accusative
narrative
(Dutra 1987)
Portuguese
Indo- European
Brazil
accusative
various
(Brito 1996, 1998; Everett 2009)
multiple
various
various
various
various
(Haig & Schnell 2016)
Notes The languages are arranged by alignment type, then by family. The designation as ergative or accusative is not meant to dichotomize, as its necessarily glosses over details of split ergativity, etc. “PF” indicates Pear Film narratives (Chafe 1980).
can be observed in a corpus as a skewed distribution of new and lexical noun tokens across the argument slots of the clause. Preferred Argument Structure represents the constraints or generalizations which—as a first approximation—appear to govern the skewed distribution of utterance tokens. Whether the four constraints should be considered mere generalizations about linguistic practices, or whether there are deeper cognitive motivations behind them, is a question that remains open to alternative theoretical
32 John W. Du Bois interpretations (see section 2.5). Though closely related, the two perspectives are useful to distinguish, at least for the moment. Specifically, Preferred Argument Structure comprises four soft constraints, which collectively influence the discourse distribution of grammatical expressions and pragmatic statuses. In the grammatical dimension, the Quantity constraint limits the number of lexical arguments in the clause core to a maximum of one. The Role constraint specifies where in the clause the single lexical argument may appear, excluding it from the A role. Paralleling this pair of constraints is a second pair in the pragmatic domain. Here the Quantity constraint limits the number of new information arguments in the clause core to a maximum of one. The Role constraint specifies where in the clause this new argument will appear, again excluding it from the A role. In contrast to the generalizations of Preferred Argument Structure, the ergative discourse profile represents the empirically observable pattern of recurrent linguistic behavior, characterized by a statistical skewing in the distribution of lexical argument and new information tokens in a population of utterances. Broadly speaking it is the difference between rules and consequences—or generalizations and facts on the ground. Table 2.2 summarizes the relation between the four constraints of Preferred Argument Structure and the corresponding consequences observable in the ergative discourse profile. To be more precise, the pragmatic constraints should be described as applying, not to new information, but to low accessibility referents (Ariel 1990, 2001), i.e. to referents whose cognitive status motivates a verbalization with a relatively informative (high surprisal) form. For present purposes, the looser formulation in terms of new information is adequate. But the more precise theoretical framing has important implications for research design, including the need for a continuous scale of accessibility/surprisal (Ariel 2001) and, correspondingly, a continuous variable for the size of linguistic forms. For future research, such terms of analysis are to be preferred. Preferred Argument Structure is neither a syntactic structure nor a discourse structure, but a preference in discourse for the use of certain configurations of grammar. All four constraints are soft constraints (Givón 1979; Bresnan, Dingare, & Manning 2001), which can be violated without producing ungrammaticality. Yet in spontaneous language use, speakers tend to follow them. Together they influence the shape of discourse productions, yielding the distribution of argument structure constructions recognizable as the ergative discourse profile. Figure 2.14 presents findings from several languages regarding lexical argument quantity, that is, the frequency of clauses with zero, one, or two lexical arguments. (For Figures 2.1–2.4, the languages are sorted in the same sequence as Table 2.1, with ergative languages presented to the left, and accusative languages to the right.) While there is considerable variability, the key finding here is that clauses with two lexical arguments are consistently rare across languages of different typologies, regions, and genealogies.
4
Figures 2.1–2.4 are based on selected studies in Table 2.1 (see also t ables 2.1–2.4 of Du Bois 2003b: 62–69).
Ergativity in discourse and grammar 33 Table 2.2 Preferred Argument Structure and the ergative discourse profile Preferred Argument Structure
Ergative discourse profile
Grammar
Avoid more than 1 lexical core argument
lexical arguments ≤ 1
Pragmatics
Avoid more than 1 New core argument
New arguments ≤ 1
Grammar
Avoid lexical A
free use of lexical S & O
Pragmatics
Avoid new A
free use of New S & O
Constraint
Domain
Quantity
Role
100 0 1 2
90 80 70 60 50 40 30 20 10
w
ish gl En
re eb
se
H
ne
m ’o O
no
Ja
da
an iy To
ho
oo n G
pa
di
di in H
ut In
uk
tit
aj Itz
al b’ jo ’an
Q
ko
hó oc M
ite
am M
kt Te
Sa
ka
pu
lte
k
0
Figure 2.1 Lexical argument quantity: frequency of clauses with 0, 1, or 2 lexical core arguments
Noun phrases are not produced in a functional vacuum, of course, but are tied to cognitive-pragmatic processes of information management. The use of a lexical noun is linked to, but far from equivalent to, the presenting of new information. Thus the Quantity constraint on lexical core arguments is paralleled by a similar constraint on new core arguments. Figure 2.2 presents findings across several languages regarding New argument quantity, that is, the frequency of clauses containing zero, one, or two new arguments. Some language samples show no instances at all of clauses containing two new core arguments (Sakapultek, Roviana), while others show very few (Inuktitut, English).
34 John W. Du Bois 100 90 80 70 60 0 1 2
50 40 30 20 10 0
Sakapultek
Inuktitut
Roviana
English
Figure 2.2 New argument quantity: Frequency of clauses with 0, 1, or 2 new core arguments
While the Quantity constraint allows speakers to freely deploy One Lexical Argument per clause core, not all syntactic positions in the clause are equally good candidates for realizing this mention. This is where the Role constraint comes in, constraining the use of lexical arguments in A role, while freely allowing their use in S or O roles. Figure 2.3 summarizes findings from a number of languages regarding where lexical arguments are realized within the clause, showing how lexical core arguments are distributed across A, S, and O. While there is considerable variation in some dimensions (e.g. the relative preference for S or O), again what matters most is what is specifically constrained: the A role, which shows relatively low frequencies of lexical mentions as a recurrent pattern across many of a typologically diverse array of languages. Nonetheless, variation here is substantial, raising interesting typological questions (see below). For one-place predicates, it is easy to satisfy the Quantity constraint, since the S role is unconstrained, freely allowing the introduction of a new referent. Indeed, this may be one reason language users select a one-place predicate over a transitive alternative— for its pragmatic, rather than semantic, affordances. But for two-place predicates, the Quantity constraint on new information is more directly consequential. Thus the Role constraint establishes an opposition within the transitive clause between A and O, constraining new arguments in A, while freely allowing them in O. Figure 2.4 presents the distribution of new core argument realizations across A, S, and O. A close examination of the frequencies in Figures 2.1–2.4 shows considerable cross- linguistic similarities in some dimensions, along with substantial variability in others.
Ergativity in discourse and grammar 35 100 90 80 70
A S O
60 50 40 30 20 10
Sa
ka pu lte k M a Te m kt ite k M o oc Q hó ’an jo b’a l Itz In aj uk tit ut H in d N i ep al M Rov i ap ia To ud na ho un no gu O n ’od Ja am pa ne s Ko e re H an eb Po re rtu w gu En g es l i e ( sh Br az i Fr l) en Sp ch O anis ld h O fren ld Sp ch an ish
0
Figure 2.3 Lexical argument role: Distribution of lexical arguments across A, S, and O
This is why it is important to evaluate the discourse evidence in light of a theoretical framework, such as Preferred Argument Structure, which offers specific hypotheses about which aspects of argument realization are constrained, and which are not. One misunderstanding that often arises in the literature involves an attempt to compare raw frequencies between two languages (one of which is usually Sakapultek). But whether frequencies match or not is beside the point (and no one language of the many surveyed has a privileged position). What matters instead is the testing of specific, theoretically motivated hypotheses about constraints on syntactic positions in argument structure constructions, e.g. with respect to information structure. Another common temptation is to attribute the difference in frequencies found in two studies to the structural attributes of the languages in question. This may prove to be correct in some cases, and is certainly a question worth asking. But there are other candidates that should be considered as well in accounting for variance, especially genre differences. As the field of Preferred Argument Structure studies develops further, these issues are likely to become more visible, and we can anticipate new findings that tease apart the subtle variables involved. Now that so many typologically diverse languages have been studied, new questions arise about universality and variability of Preferred Argument Structure. Especially interesting is the potential for two-way interactions between the grammar of a language and its discourse profiles, occasioned by the never-ending cycle of grammaticization and language use. Are there different Preferred Argument Structures for
36 John W. Du Bois 90 80 70 60 50 A S O
40 30 20 10
Sa
ka pu lte k M a Te m kt ite ko M oc Q hó ’an jo b’a l Itz In aj uk tit u Ro t vi M an ap a ud un gu n Fi nn ish H eb re w En gl ish Fr en ch Sp a O nish ld Fr en O ch ld Sp an ish
0
Figure 2.4 New argument role: Distribution of new arguments across A, S, and O
different languages, and does this correlate with the ergative–active–absolutive typological contrast (Durie 1988, 2003)? The way is open for a broader theoretical framing of questions about why these recurrent patterns arise across such a broad typological array of languages, yet vary within seemingly well-defined limits. There is much that remains to be discovered.
2.3 A Discourse Universal? Is the ergative discourse profile universal? To answer this question, I begin with the child, and go on to examine the evidence from typology and diachrony.
2.3.1 Child Language As might be expected, children who are exposed to ergative languages exhibit the ergative discourse profile. This has been shown for Tzeltal (Mayan, Brown 1998, 2008; de León 1999), Inuktitut (Eskimo-Aleut, Allen 2000; Allen & Schröder 2003), and Hindi
Ergativity in discourse and grammar 37 (Indo-Iranian, Narasimhan 2013; Narasimhan et al. 2005), and similar observations have been made for Samoan child language (Austronesian, Duranti, & Ochs 1989; Ochs 1982). But if the question is about universals, what is more telling is what happens in the discourse of children learning accusative languages. Clancy’s extensive studies of Korean child language show that even when exposed to an accusative grammar, children produce the ergative discourse profile (Clancy 1993, 1995, 1996, 1997, 2003). For Spanish, Bentivoglio argues for late acquisition of the ergative discourse profile (1998), while other studies of Spanish and French show a more complex picture (Khorounjaia & Tolchinsky 2004; Salas 2010). Older English-speaking children with autism show the ergative discourse profile (Weber 2003), as do adults with aphasia (Kohn & Cragnolino 2003). How deep does the ergative discourse profile go? For Goldin-Meadow this leads her to ask what the child’s discourse can reveal about the mind, and she goes to great lengths to find out. Seeking out “people who have had no exposure to any conventional language whatsoever” (2003: 493), Goldin-Meadow works with profoundly deaf children who have had no access to either spoken language or sign language. What these children do is invent their own gesture systems, with revealing implications for cognition: The thoughts of these individuals cannot possibly have been shaped by language. As a result, whatever categories they express reveal thoughts that do not depend on language—thought before language. (Goldin-Meadow 2003: 493)
One of Goldin-Meadow’s most compelling discoveries is that in the invented gesture systems, discourse tends to “pattern like ergative languages: intransitive actors and patients are treated alike (produced), whereas transitive actors are treated differently (omitted)” (2003: 502). She observes: We have found the ergative pattern to be robust in communication situations. Deaf children of hearing parents who are inventing their own gesture systems tend to organize their gesture sentences around an ergative pattern. Equally striking, we found that when asked to describe a series of action vignettes using their hands rather than words, English-speaking adults invented an ergative structure identical to the one developed by the deaf children, rather than the accusative pattern found in their spoken language. These findings suggest that the ergative pattern may reflect a robust solution to the problem of communicating information from one mind to another, be it an adult or a child mind. (Goldin-Meadow 2003: 516)
While Goldin-Meadow finds evidence to support the absolutive distribution of lexical arguments, she does not find evidence for the role of new information (2003: 516; Schulman, Mylander, & Goldin-Meadow 2001). Pointing to eye-tracking research by Griffin and Bock (2000), Goldin-Meadow suggests that “focusing on patients may be a default bias found in both processing and acquisition tasks” (2003: 517). While their proposed “patient focus” motivation must be balanced against other
38 John W. Du Bois factors—competing motivations—deriving from pragmatics, semantics, etc., Goldin- Meadow and associates have made a compelling case for the presence of an ergative discourse profile at the earliest stages of communication. This begins even before exposure to language, arguing for a deep-rooted cognitive basis for “thinking ergative” (Goldin-Meadow, Yalabik, & Gershkoff-Stowe 2000; Goldin-Meadow 2003).
2.3.2 Typology From a typological perspective, the first question is whether the ergative discourse profile is restricted to languages with ergative grammar, or rather represents a property of the discourse of all languages, independent of syntactic alignment. If it is restricted to ergative languages, it could be a consequence of ergative grammar. But if it is a discourse universal, this is no longer a tenable position, and it becomes a potential factor in the functional motivation for ergativity (2.4). As already noted, studies of Preferred Argument Structure have been carried out for a number of languages (2.2, Table 2.1, and Figures 2.1–2.4). While a truly random typological sample is not possible, the available languages appear to be reasonably diverse in typology, region, and genealogy, covering all major alignment types. Based on these languages, the preponderance of evidence supports the view that most languages, whether ergative, active, or accusative, tend to display an ergative–absolutive patterning in the discourse distribution of lexical (or “heavy”) nouns, and of new (or low-accessible) information, in the core argument positions of the clause. To be sure, the interpretation of the discourse patterns, and even their existence, remains controversial for some scholars; I address these issues in 2.5. I believe that when the evidence from discourse, typology, child language, and diachrony are all taken into account, the picture that emerges is of an ergative discourse profile pervasive across languages, independent of the typology of syntactic alignment. That said, one of the most interesting questions raised by cross-linguistic evidence like that in Figures 2.1–2.4 is whether ergative, accusative, and other language types may have their own distinctive variants of Preferred Argument Structure, maintaining some universal aspects while also fine-tuning to the grammar at hand, through cyclic interactions between grammar and use.
2.3.3 Diachrony If the ergative discourse profile represents a universal, found in modern languages regardless of their type, the uniformitarian principle (Hock 1991)5 predicts that older
5
“The general processes and principles which can be noticed in observable history are applicable in all stages of language history” (Hock 1991: 630).
Ergativity in discourse and grammar 39 stages should share this feature as well. While few languages can provide direct textual evidence for older stages, some do. Diachronic studies conducted so far, though few, have consistently found the ergative discourse profile in both earlier and later stages of the same language. Among accusative languages, diachronic studies of Old Spanish, Old French, and their modern descendants show that “despite a gap of seven centuries … the medieval and modern forms of French and Spanish are remarkably similar in their manifestations of Preferred Argument Structure” (Ashby & Bentivoglio 2003: 73; see also Bentivoglio 1994). In a study of English spanning six periods from Old English to Modern English, Shibasaki concludes that, despite higher lexical densities6 in older stages, each period largely conforms to the constraints of Preferred Argument Structure (Shibasaki 2006). Among ergative languages, the Mayan family offers especially rich historical implications, given the available comparative and philological evidence. With 30 modern descendants, almost all of the languages have remained consistently ergative over four millennia (Norman & Campbell 1978; England 1983, 1991, 1990; Robertson 1983, 1992; Kaufman & Norman 1984; Law 2009, 2014).7 In the domain of discourse, the ergative discourse profile has been documented in detailed case studies covering four of the six major branches of Mayan, extending across a wide expanse of the Mayan territory: Mamean (Mam and Tektiteko, England & Martin 2003); Q’anjob’alan (Q’anjob’al and Mochó, England & Martin 2003); K’ichean (Sakapulteko, Du Bois 1987b, 2006); and Yucatecan (Itzaj, Hofling 2003).8 While at least one nineteenth century scholar claimed to reconstruct the text of a proto-Indo-European myth, such feats were ultimately deemed beyond the reach of the comparative method. But a discourse profile is not a text. Its wide distribution across the Mayan family makes the ergative discourse profile a plausible candidate for reconstruction to older stages. Support comes from early textual evidence in yet a fifth branch of Mayan (Ch’olan-Tzeltalan). Mora-Marín shows that Maya hieroglyphic inscriptions, written in an early Ch’olan language, display the ergative discourse profile (Mora-Marín 2004). While areal diffusion is a theoretical possibility, the family-wide distribution and textual evidence taken together point to a 6
It bears noting that in several language families, older texts show higher lexical densities than modern texts (Ashby & Bentivoglio 2003), while still conforming generally to Preferred Argument Structure. This may reflect differences in genre of the older texts (epic poetry for Old French and Old Spanish, dynastic history for hieroglypic Maya), for which the modern languages lack a common counterpart. 7 The main exception is Chol, which has innovated an agentive system (Law, Robertson, & Houston 2006; Coon 2013; Vázquez Álvarez & Zavala Maldonado 2013). A few Mayan languages show some split ergativity, but this is a late and partial development (England 1983; Law 2009, 2014), which coexists with ergative morphosyntactic alignment. Yet one line is never crossed: No Mayan language has ever been accusative, from Proto-Mayan till today. 8 Discourse evidence from a fifth branch is provided for Chol (Vázquez Álvarez & Zavala Maldonado 2013), which seems to have a discourse profile similar to that identified by Durie for Acehnese (Durie 1988, 1994, 2003), with A=Sa opposed to So=O in both grammar and discourse profile. Whether Durie’s conclusion, that Acehnese is compatible with an (extended) Preferred Argument Structure account, can be applied as well to the Chol data is beyond the scope of this chapter.
40 John W. Du Bois more likely conclusion: that the ergative discourse profile was present in early stages of Mayan, even Proto-Mayan. This conclusion is also consistent with the uniformitarian principle, given the typological evidence for universality. The wealth of evidence for the Mayan family offers unique opportunities for tracing the history of ergativity in grammar and discourse. The implications are worth dwelling on here, as they are relevant not only for discourse-and-grammar reconstruction, but also for tracking the development of ergativity down through the daughter languages. One point of intersection between studies of comparative Mayan and the ergative discourse profile comes in the work of Nora England and Laura Martin (England 1983, 1991; England & Martin 2003). Drawing on their own research on the grammar and discourse of four languages from two branches of Mayan, they find the Given A constraint in both Mamean and K’ichean, and consider how discourse tendencies may become grammatical rules: It may be the case that, in K’ichean languages in particular, a grammatical restriction against indefinite subject NPs exists or is developing. This would presumably be a grammaticalization of the discourse constraint noted by Du Bois (1987b), that agents (ergators) are typically not used to convey new information. His analysis and analyses on other languages by England and Martin (1989) [=(England & Martin 2003)] show that, in texts from five different Mayan languages, lexical new mentions in agent (transitive subject) role typically occupy about three percent of the total lexical new mentions. Therefore, it is very rare to encounter an indefinite transitive subject noun, for discourse reasons. K’ichean languages appear to be creating a syntactic rule that reflects the same constraint. (England 1991: 484)
Thus the ergative discourse profile, present in Mamean and K’ichean languages and presumably in their common ancestor, is undergoing incipient grammaticization in K’ichean but not Mamean. This means the K’ichean change cannot be considered deterministic. Nevertheless, in moving from soft constraints to hard constraints, it traverses a path from ergative discourse to ergative grammar.
2.3.4 Interim Conclusions Is Preferred Argument Structure universal? The evidence so far paints a complex picture (2.5), and a full answer must await further inquiry (2.6). But what seems clear is that something like the ergative discourse profile represents a “recurrent pattern,” appearing again and again in the discourse of children just learning their language—or exposed to no language at all—and remaining constant in languages widely separated by space, time, genealogy, and typology. Whether the complexity and variation evident in the research record can be subsumed under a single set of generalizations remains to be seen. For now, it seems productive to explore the view that the ergative discourse profile,
Ergativity in discourse and grammar 41 and the Preferred Argument Structure constraints that motivate it, reflect enduring properties of language in use, with consequences for cognitive processing, acquisition, typology, diachrony, and grammaticization. Having introduced several parts of the problem, it is time to see how they fit together, and to assess what it would take to provide a functional explanation for ergativity.
2.4 Explaining Ergativity Preferred Argument Structure is claimed to be universal, and to motivate the grammaticization of ergativity. So far the functional explanation for ergativity seems on track—until we stop to think about accusative languages. If the ergative discourse profile is present as a motivating force in all languages, why aren’t all languages ergative? “A system-external functional force, once appealed to, cannot simply be turned off at will” (Du Bois 1985: 353). To explain ergativity, one more piece of the puzzle must be introduced: competing motivations (Du Bois 1985, 2014; MacWhinney et al. 2014; Malchukov 2014). In the arena of discourse, the two top competitors are powerful and pervasive, even universal: topic continuity and new information. The grammaticizing power of topicality is widely acknowledged. Subject is said to be a grammaticization of topic, or topic-cum-agent (Givón 1983a; Comrie 1988). But what about languages that don’t have subjects, or at least don’t grammaticize them in the central way that accusative languages do? Speakers of ergative languages have been claimed to differ from speakers of accusative languages in having “different conceptions of prototypical agenthood” and “different basic topicalizations” (Plank 1979: 28). Curiously, these psychological claims were made in the absence of any experimental or corpus evidence. To put such speculations to the test, a corpus-based study of topic continuity was devised (Du Bois 1987b: 842–843), which showed that in the ergative language Sakapultek, topical referents (appearing in two successive clauses) overwhelmingly favor S=A continuity over S=O continuity, 80 percent to 20 percent. The idea that ergative speakers think differently about topicality is debunked. But this leaves us with a quandary. Admitting that ergative speakers track topics just like accusative speakers do, it is still only accusative languages that grammaticize topic in an all-encompassing syntactic and morphological subject. If speakers of all languages track topic continuity in the S=A groove that motivates accusative alignment, why aren’t all languages accusative? The typological question comes full circle. Both questions receive the same answer: competing motivations. Ergativity and accusativity are both motivated, each with its own dedicated motivation operating at all times in all languages. But only one motivation can prevail at a time, in organizing the basic structure of a grammar (or part of a grammar) in a given language. The discourse profiles that drive this eternal competition were hinted at already, in the discussion of the discourse excerpt in (1) (2.1). Within a single short stretch of discourse, two
42 John W. Du Bois recurrent patterns are observed to coexist: first, introduction of new referents is managed in O and S (absolutive); and second, topic continuity is managed in A and S (subject). Quantitative cross-linguistic evidence shows the two discourse profiles are widely distributed, found together in diverse genres and across languages of all types. As usual, S is Janus-faced: Variable in discourse, it becomes pivotal in grammar. Having seen some key pieces of the ergative puzzle, while acknowledging the competition from accusativity, we can now ask how it all fits together—to articulate the discourse motivation for ergativity, at least, if not a complete explanation for ergativity. The Quantity and Role constraints of Preferred Argument Structure set broad limits on the information structure of the clause. In effect they define a gross template for any argument structure construction, including intransitive, transitive, and ditransitive clauses. In the arena of discourse, utterance tokens realizing such constructions undergo selection to satisfy the constraints, producing the “facts on the ground” of the distribution of argument structure configurations in the utterance population. The ergative discourse profile represents a generalization of the statistical distribution of recurrent patterns, as candidates for grammaticization in the grammar of a given language. At the same time, the accusative discourse profile, lurking in the very same utterance population, presents its own patterns as alternative attractors for grammaticization. In general, the distribution of New information (reflecting cognitive processing demands) motivates a discourse pattern isomorphic to ergative–absolutive grammar. The distribution of topic continuity motivates a discourse pattern isomorphic to nominative–accusative grammar. Both functional pressures are present in the discourse of all languages, but at any given point only one can grammaticize, determining the syntactic alignment of a specific argument structure construction. If just one syntactic alignment is to prevail, the competition must be resolved. And this is just what grammaticization does: It resolves competitions, converting functional motivation into normative motivation (Du Bois 2014: 280). What emerges is a grammar that may seem arbitrary in its specific forms and normative rules, but that works for its users, serving as a unified framework for communicative practice and cognitive affordance. Consider the Mayan case, which shows what can happen when the ergative discourse profile interacts with word order, setting up the conditions for the emergence of ergativity. First, in line with the ergative discourse profile, the single lexical argument (reflecting the Quantity constraint) typically occurs in either O for transitives, or S for intransitives (reflecting the Role constraint). Second, for Mayan languages going back to Proto-Mayan, the dominant word order (Dryer 2013a) is verb-initial. Taken together, these two factors place both lexical arguments to the right of the verb, producing a structurally consistent V-Lex{s/o} or V-Absolutive word order (an ergative order in the sense of Dryer 2013b). This constitutes the gross structure of verb and noun, maintained consistently in the discourse profiles of most if not all Mayan languages from Proto-Mayan to now. But Mayan languages also have a fine structure of pronominal clitics, implemented in head-marking of both ergative and absolutive on the verb. How does the fine structure of agreement morphology interact with the gross structure of word order syntax?
Ergativity in discourse and grammar 43 As noted earlier, different parts of the cross-referencing paradigm behave differently. First and second person referents are given information, and thus are regularly marked by overt (non-zero) pronouns/clitics, in both ergative and absolutive. For 3rd person, referents in A are typically given, and thus reduced, and potentially cliticized. But S and O are often new, hence lexical, and so are not likely to be reduced or cliticized. The absolutive mention, being expressed overtly in the clause with a lexical rather than pronominal mention, is not itself cliticizable. The absence of a clitic may then be interpreted as absolutive zero agreement. Once the pronoun-to-agreement shift takes places (presumably in pre-Proto-Mayan), the features that define the Mayan ergative complex (V-Lex word order, head-marking, absolutive zero, and the ergative discourse profile) create a kind of V-Lex lock-in. This proves to be an evolutionary stable strategy, resistant to change over four millennia. Along with the gross structure of information, there is also the fine structure of inflectional detail, semantic nuance, and the specificity of pragmatic interpretation. The gross structure defined by Preferred Argument Structure sets broad constraints, but leaves speakers and languages plenty of room to maneuver at the level of fine structure. This is where the precise details of morphosyntactic analysis and historical development become critical, as processes of utterance production, interpretation, analogy, reanalysis, and grammaticization deploy and reconfigure the fine points of grammar to serve the exacting needs of language users. The grammarian’s, semanticist’s, and historical linguist’s attention to detail comes into its own here, analyzing the fine structure to elucidate the precise accounting of grammar and meaning, which is indispensable to the functionality of language. But the gross structure matters too. In the end there is no need to choose between them. Gross and fine work together in all of human action, and language is no exception. Languages often show an apparent harmony between gross structure (e.g. information structure, topicality, word order) and fine structure (e.g. inflection, agreement). But what if a discord or disruption arises between gross structure and fine—how is it to be bridged? Must it be? To locate the critical arena where the crisis comes to a head, look to word order in use. This is where discourse-and-grammar research may contribute to resolving certain mysteries about the grammaticization of syntactic alignment. Preferred Argument Structure motivates the gross configuration of lexical nouns, e.g. the ergative discourse profile. If these nominal elements are arrayed on the same side of the verb as in most Mayan languages, perhaps motivated by a formal and/or functional analogy, the result is a recurrent word order configuration (e.g. V-Lex{s/o} or Lex{s/o}-V) that constitutes a potential model for ergativity. If another historical development (say a phonological merger or analogical leveling) collapses key distinctions between elements in the agreement system, this may undermine the fine structure of morphological agreement. Yet what remains is the gross structure. Now the ergative distribution of gross structure becomes a potential model upon which to rebuild a fine structure, this time along new lines, perhaps ergative. Whether such a trajectory can be confirmed or disconfirmed in a language family with a long written history remains an open question. But the only way to find out is to follow the patterning of language in use. This
44 John W. Du Bois means documenting the discourse profiles that define the gross structure of the relevant utterances. It may be rare to find suitable conditions for inquiry into both gross and fine structure at the required level of detail, but it is well worth seeking out. The analysis of the Mayan ergative complex in relation to the ergative discourse profile provides an important case study illustrating certain general principles of functional explanation, insofar as it combines universal generalizations about gross structure of the ergative discourse profile with locally specific generalizations about fine structure of the inherited morphosyntax of a particular language family. Extending this approach to other language families, each with its own unique history, one should expect three things. First, the layer of gross structure produced by principles like the ergative discourse profile has a discourse-based coherence of its own, and is likely to remain relatively stable; yet word order developments can introduce fundamental changes even here. Second, the layer of fine structure (inflection, agreement, and so on) linked to the language-specific inherited morphosyntax has its own logic of continuity and change, which operates in part independently of that of gross structure. Third, interactions between gross structure and fine structure may trigger a dynamic of change that disrupts the ecology of grammar, setting in motion events that lead to restructuring the system of grammar. The story of the interaction among gross and fine must be discovered anew in the history each new family. Yet even here, analogy, reanalysis, and adaptive selection (based on cognitively motivated information processing constraints, for example) offer general principles for a theory of grammaticization. The roster of motivations with the power to shape ergative and other syntactic alignments is not exhausted by the two considered here. A more complete account will have to incorporate interactions with competing and converging factors such as event structure, causal chains, verb semantics, tense and aspect, voice, the constructional repertoire, evidentiality, analogy, and more (DeLancey 1990; Croft 1998; see also the chapters in this volume). Many of these involve morphosyntactic, semantic, and pragmatic fine structure, which is essential to complete the functional explanation of ergativity. A critical task for future research is show how multiple layers of fine structure interact with layers of gross structure to shape the grammaticization of ergative and other argument structure configurations.
2.5 Objections and Refutations The claim of a discourse basis of ergativity (Du Bois 1987b) has generated a certain amount of controversy, which I address in this section. Some objections question the existence of the ergative discourse profile, or try to explain it away, while others accept it but doubt it supports a functional explanation for ergativity. While some studies raise useful points that warrant attention in future research, others reveal a misunderstanding of the nature of discourse-based explanation. Common conceptual errors are essentialism, reductionism, and epiphenomenalism, all of which involving idealization of the
Ergativity in discourse and grammar 45 facts of language use. In the following I will try to distinguish the useful critiques from the dead ends; and, for the latter, suggest an alternative approach. A recurrent theme is that generalizations about language must be grounded in linguistic realism rather than idealization.
2.5.1 Diversity One of the earliest and most productive critiques was developed in a series of discourse- and- grammar studies by Mark Durie (1988, 1994, 2003). Preferred Argument Structure studies had adopted the A–S–O grammatical categories (Dixon 1979) for their neutrality with respect to typological diversity in syntactic alignment, to avoid imposing alien categories on ergative languages. But Durie argued they were doing just that to active languages. He rejected the view of S as a universal category for linguistic analysis, arguing that it obscured important diversity in both discourse and grammar, in active languages like Acehnese (Austronesian, Indonesia). Here, intransitive subjects are internally differentiated, both in grammar and in discourse profile, between Sa and So. A better analysis of Acehnese discourse could be achieved by respecting the alignment typology evident in Acehnese grammar, which distinguishes Actor (Sa=A) and Undergoer (So=O). By investigating categories relevant to the grammar at hand, Durie was able to show that each discourse profile was both internally consistent and distinct from its counterpart. Subsequent studies in ergative, accusative, and especially active languages have confirmed and extended these findings. For example, Chol, a Mayan language once characterized as “split-ergative” but now recognized as active, makes a similar distinction between Actor and Undergoer in both grammar and discourse profiles (Vázquez Álvarez & Zavala Maldonado 2013). These studies challenge the assumption that S is uniform in language use, and provide a useful corrective to the A–S–O schema. More generally, they remind us that the categories of the language being investigated are a potent guide to new discourse profiles waiting to be discovered. This raises the question of whether there are different Preferred Argument Structures for active, ergative, and accusative languages, or even finer-grained distinctions; and, if so, whether they can be interpreted as principled variations on a single unifying theme. Durie nevertheless saw value in framing his analysis in terms of Preferred Argument Structure, once the necessary adjustments were made to accommodate the diversity of grammar and discourse profiles of the language being analyzed. Durie recognizes a key point that befuddles some critics, regarding the role played in discourse-functional explanation by the discourse profile (Du Bois 2003a: 40–44), what I have previously called “recurrent clusterings in parole” (Du Bois 1985: 357). Discourse profiles are generalizations about recurrent patterns of linguistic behavior, including recurrent co- occurrences of pragmatic, semantic, and grammatical features. As statements about language use, they are not to be confused with the grammatical categories they motivate, which have a normative status in the linguistic system. Durie nicely points up a
46 John W. Du Bois common misconception about discourse-functional motivation of grammatical categories in a response to Herring (1989) which remains relevant today: Herring (1989), misunderstanding the concept of motivation, regards the mismatch in Sakapulteko as a logical flaw in Du Bois’s argument. A further distortion in Herring’s argument derives from her demand that the grouping of S and [O] should involve functional specialization in terms of cognitive or semantic factors. In Du Bois’s account, the cognitive/semantic motivations do not themselves directly impact upon coding choices, but only proximately, through the “recurrent clusterings in parole” (Du Bois 1985: 357) which they produce. This is of course a key difference between the Du Bois account of ergativity and more directly cognitively or semantically based approaches. (Durie 2003: 192, fn. 193)
Discourse profiles are general statements about the “facts on the ground” of language use. They arise as a result of multiple factors, including factors such as cognitive processing, salience, and so on. As system-external motivations they interact with system- internal factors to shape the emergence of grammar. But the process is not deterministic. While discourse profiles influence the grammaticization of linguistic categories, the profile is not the category.
2.5.2 Essentialism Less productive are critiques which, while bringing no empirical research to bear on the question at hand, translate originally statistical observations into the language of categorical statements, the better to draw logical deductions—or contradictions. Needless to say, the meaning of a discourse profile may be severely compromised if its empirical variability is disregarded. When soft constraints are restated as hard constraints, the result is unlikely to be faithful to the original. Such a confusion mars the critique of Harris and Campbell (1995), who paint a portrait of the discourse basis of ergativity (Du Bois 1987b) that is almost unrecognizable to someone familiar with the theory. They confidently appeal to the “spirit” of Preferred Argument Structure constraints, which they feel are “blatantly violated” by the passive origin of ergativity (1995: 253). They enthusiastically assert that transitive “A [is] prohibited almost totally from introducing new information” (1995: 253), then quietly delete the “almost” as they substitute a categorical rule for the original soft constraint. They perceive “an unpleasant circularity in Du Bois’ picture” (1995: 254), based on their questionable reframing of Nora England’s findings of incipient grammaticization of certain Preferred Argument Structure constraints into grammatical constraints on the ergative in K’ichean (England 1991; England & Martin 2003). Their story of the passive origins of ergativity begins with a proto-language agented passive, represented as containing two proper nouns (Mary was.hit by Jane) (1995: 252). This contrasts with the attested
Ergativity in discourse and grammar 47 examples they themselves cite from Old Persian to illustrate the passive to ergative change, which contain a pronoun rather than a lexical noun in the agentive by-phrase (1995: 244, 255). Though somewhat murky, the reasoning behind their argument seems to depend on the following assumptions and inferences:
1. To get from accusative to ergative alignment, a well-known diachronic pathway is via the reanalysis of an intransitive passive construction as a transitive active construction. 2. To express a two-place predication fully explicitly, the passive sentence cannot be agentless, but must include the agent. 3. Thus the agented passive construction is used, with the agent expressed in an oblique role, e.g. a prepositional phrase adjunct (by-phrase). 4. Obliques (prepositional phrase adjuncts) often contain new information and lexical nouns (by Preferred Argument Structure). 5. Therefore the oblique agent must have been new and lexical. 6. The two-participant predication, being a passive, is intransitive. 7. Therefore the subject of the two-participant predication (the semantic Patient) must be an S. 8. Intransitive S arguments are often new and lexical (by Preferred Argument Structure). 9. Therefore the S role argument must have been new and lexical. 10. By reanalysis, the by-phrase (originally an optional adjunct) is reinterpreted as an obligatory core argument (transitive subject A), and the S becomes O. 11. The formerly rare agented passive intransitive construction undergoes a massive change in frequency to become the new normal for transitive constructions, but speakers make no changes in their use of pronouns vs. nouns, or given vs. new information. 12. Now there are two new and lexical nouns in the core arguments of the clause (violating Preferred Argument Structure). 13. Now there is a new and lexical A (violating Preferred Argument Structure).
The problems with this account are several, arising on multiple levels. First, the logical flaws. The reasoning depends on converting a statistical tendency to a categorical rule. This must be done twice (in deriving 5 from 4, and 9 from 8), in order to generate the desired contradiction. Second, language users are presented as being sufficiently creative to introduce major changes to the structure of their grammar, reanalyzing an optional oblique as an obligatory core argument (step 11), but they were apparently too timid to begin using a pronoun in place of a lexical noun—during the decades or centuries it must have taken for the frequency shift and ergative reanalysis to be completed. Ignored is the fact that there are no constraints against using pronouns or given information: Not only are the Preferred Argument Structure constraints always soft constraints, the limits are always upper limits, not lower (Du Bois 1987b: 834; 2003b: 73).
48 John W. Du Bois Thus there has never been any minimum requirement to fill a syntactic slot, whether argument or adjunct, with either new information or a lexical noun. One begins to understand why it was necessary for Harris and Campbell to appeal to the “spirit” of Preferred Argument Structure in making their argument, instead of employing the actual constraints. The result is unfortunately fairly typical of essentialism, which, faced with statistical evidence of diversity in argument realization, substitutes a categorical stereotype, and then uses the stereotype to work out the logic of its reasoning. But surely this is antithetical to the “spirit” of Preferred Argument Structure, if it must have one. Is there an alternative? Actually, very little needs to be changed in the above account to make it clear that not only is Preferred Argument Structure compatible with the grammaticization pathway in question, it actually facilitates it. Moreover, the only changes needed involve replacing the rigidly idealized conjectures of Harris and Campbell with common-sense observations on how speakers talk—as confirmed in corpus-based studies of naturally occurring language use. Preferred Argument Structure allows speakers the flexibility to use a pronoun in any slot, whether argument or adjunct, and speakers routinely do just that (Du Bois 1987a, 2003b). Moreover, as Ariel et al. (2015) have shown, speakers use pronouns when innovating a new argument slot, precisely because pronouns fly under the radar of the Quantity constraint. This leaves speakers free, when extending the use of a structure like the passive, to choose a pronoun in the by-phrase. In fact this is precisely what the textual evidence from Old Persian shows: (2) ima tya manā kartam Parθavaiy “This is what was done by me in Parthia” (Darius the Great) (John R. Payne 1980: 151) (cited in Harris & Campbell 1995: 255) The genitive pronoun (manā ‘me.gen’) expresses the agent in a by-phrase, initially an optional adjunct. But as this structure comes to be routinely used to express two- place predications, it undergoes reanalysis (step 11) as an ergative core argument. The new transitive structure easily satisfies the Quantity and Role constraints of Preferred Argument Structure.
2.5.3 Reductionism One attraction of essentialism is that it feeds reductionism. By replacing a complex empirical generalization with a handy rule of thumb, a theoretical economy is achieved. If the rule is not only categorical but universal, the conclusion may be drawn that empirical investigation of the language at hand is unnecessary. But when statistical generalizations are replaced with categorical statements, the likely result is a false economy. Such a chain of essentialist substitutions is found in the reductionist proposal of Haspelmath (2006). Where I had pointed out the functional motivation linking the avoidance of lexical nouns with avoidance of new information (Du Bois 1987b: 829–830), Haspelmath wants to go further, suggesting that “since new information is mostly coded
Ergativity in discourse and grammar 49 by lexical NPs, … the [lexical] tendencies … could be reducible to the [new information] tendencies.” Moreover: [T] he ‘quantity’ tendencies … [may] follow straightforwardly from the ‘role’ tendencies. (Haspelmath 2006: 910) [I]t appears that the strong tendency to avoid clauses with two new/lexical core arguments can be reduced to the strong avoidance of new/lexical As … . [W]e simply like to talk about human beings and their actions, so animates tend to be topical. … Since the A argument is almost invariably animate, it follows that it is typically topical and hence given. (Haspelmath 2006: 911, emphasis added)
Haspelmath concludes that “the whole notion of preferred argument structure may be reducible to other, well-established tendencies and generalizations” (2006: 911, emphasis added); (see also Dahl 2000: 50; Everett 2009; Haig & Schnell 2016). This style of argument slides easily from most to all, from it tends to it follows, and draws inferences accordingly. While it overlooks the fact that the Quantity constraint probably has a better independent motivation as a cognitive limitation on information processing resources, Haspelmath’s proposal to dispense with the Quantity constraint, and indeed all of Preferred Argument Structure, makes a certain sense—if we accept a series of inferences about language use, each apparently well motivated on its own:
1. 2. 3. 4.
Humans are interesting to humans. Therefore, humans are topical. Therefore, humans are given. Therefore, humans are expressed by reduced forms (pronouns or zeroes).
Leaving aside the inferential gaps, this is still not enough to make viable predictions about argument realizations in syntactic argument slots. If you add the further assumption that humans are agents, and that the subject role expresses agent-cum-topic, you can, seemingly, derive the inference that subjects are expressed by pronouns. In contrast, direct objects are said to encode inanimates. Inanimates are less interesting, therefore not topically continuous, therefore not accessible, therefore new, therefore lexical. But do these conclusions actually follow? Goldberg evaluates a similar reductionist proposal and rejects it: “[T]he Given A constraint does not follow directly from the prevalence of animate topics. … [T]he Given A constraint is not simply epiphenomenal” (Goldberg 2004: 431). The problem with the essentialist–reductionist line of reasoning is that the conclusions follow only if each of a long chain of assumptions is valid—specifically, only if each generalization is categorical. But none of them are. And because each step in the chain represents at best a statistical tendency, the inferential failure compounds with each step. Flawed logic aside, the most serious problem is empirical: the observable facts of discourse do not confirm the logical deductions about what discourse “should” look like. The fact is, only some
50 John W. Du Bois humans are topical, and therefore pronominalizable, and this makes a critical difference for grammaticization pathways, e.g. in the role of pronouns in the emergence of innovative argument structure constructions (Ariel 2000; Ariel et al. 2015). Is there another way? A recent corpus-based study of Hebrew datives (Ariel et al. 2015) offers a relevant model, even if the construction involved is different. Ariel et al. compare datives (syntactic arguments) with adjuncts headed by the preposition bishvil ‘for.’ Both introduce mostly humans, and both express the same thematic role (roughly, benefactive). Yet they part company in their discourse profiles: Only 5 percent of the datives are lexical, but as many as 23 percent of the bishvil adjuncts are. The difference cannot be attributed to animacy, but only to the difference in syntactic status: argument vs. adjunct. This accords with the predictions of constructional Preferred Argument Structure (Ariel et al. 2015: 270–272; Du Bois & Lester in progress), which freely allows lexical mentions in adjuncts, but not in core argument slots other than S/O. More generally, pronouns sidestep any problems with the One Lexical Argument constraint, and thus are exploited as pioneers in an incipient grammaticization introducing an additional argument position into the clause, allowing a second human participant (in addition to the agent) to be expressed in a benefactive-like role. What are the implications for ergativity? While the alignment types and constructions in question differ, interesting parallels can be drawn nonetheless. In both cases, speakers modify an existing argument structure construction, adding a new argument role (ergative or dative). And the strategy they adopt to accomplish it without violating Preferred Argument Structure constraints is essentially the same: cognitive containment (Ariel et al. 2015: 270–272; Du Bois & Lester in progress). The safe strategy is to use a pronoun in the innovative argument slot, to avoid violating the One Lexical Argument constraint—or, to put it in cognitive–functional terms, to avoid overloading limited processing resources.
2.5.4 Historical Accident Claims of functional motivations have implications for language change, but are sometimes at odds with the findings of historical linguists (Cristofaro 2014). For example, Næss (2015) points to a series of seemingly random historical changes in rejecting, not only the competing motivations analysis of ergativity (citing Du Bois 1985; Du Bois 1987b), but any functional explanation for the ergative structure of Äiwoo (Oceanic, Solomon Islands). After detailing the complicated historical processes that produced the ergative verb phrase, he states: What the [ergative] VP structure of Äiwoo demonstrates is … that any linguistic system at any point in time is the outcome of a number of interacting factors which do not add up to either a universal structural template or a set of functionally-based competing motivations. (Næss 2015: 102)
Ergativity in discourse and grammar 51 This statement about grammar as “the outcome of … interacting factors” is fine up to the conclusion, which does not follow. The assumption that the claimed VP merits its own “set of functionally-based competing motivations” seems premature, in the absence of the relevant discourse profile information. Be that as it may, it is surely unwise to conclude that the randomness of historical changes precludes an adaptive account of the historical evolution of linguistic structure. From the perspective of linguistic as well as biological evolution (Beckner et al. 2009; Lane 2015: 172–204), there is no principled contradiction between the fact that a system is subject to historical processes, which may randomly introduce partial arbitrariness, and the applicability of the evolutionary processes of adaptive selection, which yield partial motivation. The origin of ergativity seems particularly prone to invite frustration and despair, leading some historical linguists to go so far as to question whether there is any functional motivation for ergativity: The absence of a clear case of extension creating ergativity argues against a clear functional motivation unique to the ergative pattern. With the exception of the passive > past/perfective ergative, it appears that ergativity is a historical accident that has come up again and again in many parts of the world. (Gildea 2004)
To conclude that “ergativity is a historical accident,” while acknowledging that it is a recurrent pattern arising independently in languages around the world, seems a contradiction. Anything in language can be made to look like a historical accident—even the grammar of accusative languages. But such a stance appears valid only if one restricts the inquiry to tracing etymological sources and describing the arbitrary signs that result. What looks like a “historical accident” may well turn out to involve adaptive selection, given an evolutionary account. What is the alternative? The evolutionary development of any aspect of language can be seen as the result of many small, locally motivated actions, taken by speakers who lack an overarching view of the system (Keller 1994). But in this respect linguistic evolution is no different from biological evolution (Dediu et al. 2013; Lane 2015; Mayr 2001). Evolutionists don’t give up on adaptive explanation just because random mutations introduce one accident after another. On the contrary: Such accidents (mutations) provide the necessary variability (Bybee 2007) for selection to act on. On one interpretation, functional constraints play a role in grammaticization by acting as selective processes that winnow the variability of naturally occurring discourse. The forms and constructions that survive and reproduce in the longer term will be those that satisfy cognitively motivated constraints, like Quantity. In general, constraints on the selection of functionally viable linguistic structures can act over time to constrain the possible outcomes of grammaticization processes, leading to functionally motivated structures (Hopper & Traugott 2003; Traugott & Trousdale 2014).
52 John W. Du Bois
2.5.5 Epiphenomenalism Epiphenomenalism may seem an arcane philosophy, originating as it does in nineteenth- century mind–body dualism (Walter 2009: 1137). But despite its abandonment in most modern sciences, in linguistics (and in sociobiology, Searle 2013) the epiphenomenon is invoked surprisingly often (Hopper 1987; Jaeger & Snider 2008; Malchukov 2008). This includes in claims about ergativity (Everett 2009; Haig & Schnell 2016). In practice, labeling an empirical observation as epiphenomenal typically prefigures a move to dismiss it as inconsequential. “An epiphenomenon is a secondary symptom, a mere “afterglow” of real phenomena” (Walter 2009: 1137). While the supposed epiphenomenon is acknowledged to have a real cause, it is claimed to have no further causal consequences in the world (Walter 2009: 1137). By setting up a disconnect between mind and materiality—or langue and parole—epiphenomenalism inherits the problems of dualism (Searle 2013). Not surprisingly, attempts to partition facts into real phenomena and epiphenomena tend to suffer from a lack of consensus about criteria for deciding which is which: One scholar’s epiphenomenon is another’s phenomenon. But the real problem lies in the idealization that is introduced into otherwise empirical questions: a dualism, not of mind and body, but of grammar and use. Appeals to epiphenomenality often arise in response to claims about language use as an influence on grammar. An empirically observable pattern in use is said to be epiphenomenal—in effect, not a phenomenon, but only illusory—to the extent that it can be explained away as caused by other factors—in a word, reduced (Haspelmath 2006). But this neglects the first question that should be asked: Does the observable pattern in language use have downstream causal consequences? Specifically, the epiphenomenal charge has been made regarding discourse patterns identified by Du Bois (1987b) as consequential for the grammaticization of ergativity (Everett 2009; Haig & Schnell 2016). In a study otherwise notable for its careful analysis and impressive multilanguage database,9 Haig and Schnell, speaking of the Given A Constraint, maintain that: [T]he apparently marked behavior of the A role, another cornerstone of the ergativity claims, … is an epiphenomenal by-product of two well-documented and robust tendencies: the pervasive tendency for transitive subjects to be [+hum], and the pervasive tendency for all subjects (S or A) to be topical, hence given information. (2016: 612)
Their claim to distinguish some patterns as epiphenomenal (while others, presumably, are not) leads them to conclude, somewhat surprisingly, that ergativity itself is epiphenomenal: 9 The massive study by Haig and Schnell (2016), encountered online as this chapter was going to press, makes a valiant effort to raise the bar statistically, and merits a more complete assessment than can be given here. Nevertheless, it introduces conceptual problems of its own, touched on here. Note that they cite different numbers of subjects and objects for transitive clauses (e.g. for English they report 1,111 transitive objects but only 422 transitive subjects; see their appendix 2). The gap reflects their omission of 1st and 2nd persons, making it difficult to compare with the findings reported here.
Ergativity in discourse and grammar 53 In Iranian for example, a shift from accusative to ergative alignment (restricted to past tenses) … was a particular, and highly contingent, combination of … changes that conspired to yield ergative alignment … . These and similar diachronic developments speak of a more contingent approach to ergativity, according to which ergativity arises as an epiphenomenal and construction-specific constellation, through the combination of essentially independent morphological and phonological processes. (Haig & Schnell 2016: 614–615)
Here the idea of epiphenomenality begins to collapse on itself. Against claims of functional explanation, the epiphenomenalist proposes an absence of explanation: “ergativity is a historical accident” (Gildea 2004) or “epiphenomenal” (Haig & Schnell 2016: 615). True, a grammatical pattern may seem accidental, when viewed exclusively in terms of its etymological source materials. But some historical accidents are accidents waiting to happen. As a typologically recurrent pattern, ergative syntactic alignment must be recognized as a powerful attractor state, that is, an evolutionary stable strategy. Accusativity too is a potent attractor. But there is no contradiction here. Both recurrent patterns are attested worldwide, and each has its motivations. Indeed, the two motivations coexist within the discourse of every language. The eternal tension between ergative and accusative motivations, evident in split ergativity, is best understood in terms of the theory of competing motivations (Du Bois 1985, 2014; MacWhinney et al. 2014). But to dismiss a linguistic pattern—even ergativity—as epiphenomenal is to cut off inquiry prematurely. It would be better to drop the epiphenomenal stance altogether, and take up the very real challenges of building theory and method capable of accounting for recurrent discourse patterns and their downstream consequences for grammar. In the study of ergativity, the facts on the ground of discourse hold much promise for understanding split ergativity (Silverstein 1976; Comrie 1978; Dixon 1979, 1987; DeLancey 1981; Durie 1988, 2003; Malchukov 2005, 2014; Law et al. 2006; Mahand & Naghshbandi 2014; see also Laka, Nash, Coon & Preminger, and Woolford Chapters 7, 8, 9, and 10, this volume).
2.5.6 Interim Conclusions I have considered various objections to Preferred Argument Structure and the ergative discourse profile. The most productive tend to come from researchers who combine in-depth first-person research on the grammar in question with detailed and sensitive empirical investigations into the discourse distribution of grammatical elements (Durie 1988, 1994, 2003; Vázquez Álvarez & Zavala Maldonado 2013). Others were found wanting: mired in essentialism, reductionism, and epiphenomenalism. What is the problem? Not generalization per se, which is indispensable for understanding and explaining grammar. Rather, the problem lies in idealization, cutting the system of language off from the reality of its use. Idealization begins with a misplaced essentialism that reifies categories, obscuring the variability that characterizes
54 John W. Du Bois populations of naturally occurring utterances. It continues (sometimes) with a reductionism that creates an illusion of economy, without testing to see if the reduced principles can in fact reconstitute (predict) the facts of the world it claims to have reduced. On the other hand, sometimes scholars dwell on a maze of intricate historical details—no lack of empirical facts here—but when the time comes for an explanation they come up empty, claiming a historical accident. Finally, the apotheosis of idealization is reached in epiphenomenalism, which dismisses certain facts as not phenomena, banishing them to the black hole of epiphenomena—from which no downstream causal consequences can escape. What is the alternative? First, linguists must commit to linguistic realism. The facts on the ground of discourse are not to be dismissed, lest their consequentiality in the world of language be overlooked. Nor are they to be shunted aside as epiphenomena. Rather, language use is firmly located in the world. Here it coexists and interacts with the practices, norms, and knowledge of language, even if all these have somewhat different ways of being in the world. Linguistic realism urges documenting the empirical generalizations about language use that define its capacity to shape grammar: the discourse profile. Second, it is equally important to follow through on the theoretical end. Discourse profiles have downstream causal consequences. Identifying the consequences serves at the very least to verify that a meaningful discourse profile has been identified. The combination of linguistic realism and theoretical generalization is critical for explaining ergativity, and for all questions of the interinfluence of discourse and grammar. While some objections have proved lacking, even so they serve to elicit clarification of issues left murky in previous formulations, and point to gaps in our knowledge that call for further research. Important issues have been broached regarding the relation of language use to grammar, and of linguistic realism to functional explanation. Certainly some aspects of the discourse approach to ergativity are likely to remain controversial, until resolved by further research. It remains for new collaborations among researchers, bringing together corpus-based methodologies, multifactorial statistical techniques, grammar, typology, competing motivations, and functional theory, to advance our understanding of the outstanding questions.
2.6 Directions for Future Research What new possibilities does discourse research bring to the study of ergativity? Whether the topic is pragmatics, syntax, semantics, constructions, splits, morphology, prosody, cognition, typology, diachrony, or grammaticization, all can benefit from incorporating a focus on language in use. The study of discourse integrates well with a wide range of research questions, bringing new perspectives on how a targeted set of grammatical resources serves the communicative goals of its users. For ergativity the biggest payoff is likely to come from “discourse inside the clause” (Du Bois 2003a: 13; 2003b: 83), an approach that seeks to map out the distinctive functional
Ergativity in discourse and grammar 55 correlates of the various structural components of the clause. A useful research strategy is to follow the trail of difference, in both discourse and grammar. Differences in linguistic structure are shaped by the multiplicity of functional needs, modulated by complex interactions among competing motivations, and driven by the dynamics of grammaticization. Interesting structural differences arise between contrasting elements in split grammar, variation, typology, child language, diachrony, and grammaticization, all potentially linked to distinctive functions. The world’s languages offer a rich set of natural experiments (Evans & Levinson 2009), where each case represents a potentially novel way of linking contrasts in form to contrasts in function. Integrating a discourse methodology into cross-disciplinary research can bring a key piece of the puzzle, helping to trace out the similarities and contrasts that manifest in such fundamental differences as, for example, ergativity vs. accusativity. For discourse to become an integral and valued component of such research, it must identify the specific discourse profiles that are relevant to the problem at hand. One ongoing challenge is to tease apart grammar and discourse: grammar with its seemingly static structure, and discourse with its free plasticity. The challenge is compounded, yet all the more important, because grammar and discourse are so intimately intertwined. For untangling the multiplicity of factors that impinge on the discourse distribution of referential forms across syntactic roles, a much-needed development is the application of newer multivariate statistical techniques (Du Bois & Lester in progress). The increasing popularity of soft constraints (Bresnan et al. 2001) as a way of describing morphosyntactic and functional variation makes the development of statistical models all the more urgent for corpus-based ergativity research. Such techniques are needed to address questions that have been raised about functional correlates of variation in ergative marking, for example in exploring Duranti’s observations about how the so-called “optional ergative” is exploited in discourse to index social power and access to agency (Duranti & Ochs 1990; Duranti 2004). Similarly, Dixon, observing that discourse profiles differ in detail from one ergative language to the next, asks how this may reflect differences in cognitive style between their respective cultures (Dixon 1994: 219–220). Yet perhaps the most productive questions, offering a combination of the most effective methodological purchase and the greatest theoretical rewards, will be those that explore the diversity of constructions that characterize the grammar of ergative and other languages. The constructional approach is particularly well positioned to shed light on issues of split ergativity, with large theoretical implications. The exploding interest in the grammar of constructions provides a useful framing for new investigations of the discourse profiles of a wide variety of constructions, each potentially revealing some new aspect of the structural and functional diversity of language. Constructions combining aspects of argument structure, valence, reference, person, voice, tense, aspect and so on are promising topics for new research on syntactic alignment that incorporates a discourse-and-grammar dimension. For the targeted argument structure construction, it will be important to map out its information structure, as revealed in its distinctive discourse profile, operating within the relevant functional niche (Du Bois 2014).
56 John W. Du Bois A further step is to work out how such construction-level discourse profiles interact with the semantics of the verbal repertoire; inherited morphosyntax; functional strategies for utterance processing; and other factors that combine to motivate the grammaticization of ergative and other alignments. A particularly incisive version of this problem concerns the three-way interaction between split ergativity, split function, and split discourse, where each new language’s “natural experiment” differentiates contrasting elements within the scope of a single language. Insights gained from split ergativity can extend even to languages that do not themselves overtly exhibit these splits, insofar as they provide clues to concealed patterns and discourse profiles that reveal the incipient fault lines of potential change. For linguists who are prepared to use the combined tools of discourse and grammar to investigate the diversity of natural experiments in the world’s languages, the potential for discovery is open-ended. A valuable research strategy is to (1) document a construction, or better, a contrasting set of grammatical constructions; (2) identify their respective discourse profiles; (3) clarify the connection of each to its corresponding functional niche; and (4) explore the ramifications for grammaticization. Such a research agenda is well positioned to expand and refine our understanding of the dynamic processes that motivate the grammaticization of the structures of ergative languages in all their diversity, with broad implications for understanding all forms of grammar.
2.7 Conclusions Amidst increasing interest in the complexity, diversity, and heterogeneity of grammar in all its forms and functions (Beckner et al. 2009; Page 2010; Evans 2013), the challenge of coming to terms with ergativity takes on special significance. Ergativity has long revealed an uneasy fit with the conventional categories of standard theories of grammar, including the supposedly fundamental categories of subject and object, or even A, S, and O. Ergativity calls into question the universality of such preconceived categories; yet neither can its own indigenous categories of ergative and absolutive be set up as universal structures in their stead. Ergativity and its variations and competitors challenge the linguist’s desire for easy answers. One way to engage with this complexity is by attending to the variability, and the recurrence, of patterns of language use. For example, the ergative discourse profile has been documented in a typologically diverse array of languages around the world: ergative, active, and accusative. To be sure, it does not stand alone: competing discourse forces of topicality strongly motivate ergativity’s main competitor, the accusative type. Yet whether expressed in grammar or only in discourse, the robustness of the ergative discourse profile remains, attested in its continuity across grammatical typology, historical change, child language, and genre. Its presence, sometimes variable and often submerged, is nevertheless revealed in small ways, shaping the child’s earliest productions and the occasional grammaticized construction even in accusative languages. While it
Ergativity in discourse and grammar 57 cannot constitute a complete account of ergativity on its own, neither can any explanatory account of ergativity be complete, realistically, without incorporating the “facts on the ground” of the ergative discourse profile, with all its consequences. As a counter to reductionism, essentialism, and idealization, the empirical trend of modern linguistics invites attention to the explanatory power of corpus-based evidence. Typologically aware research on discourse profiles documents the complexity and diversity of vast populations of utterances—but also the consistency that gives them power to shape the adaptive evolution of grammars. Surely the functional, structural, and historical basis of ergativity is more complex than is envisioned in any one current analysis. A full explanation of ergativity and its variants and alternatives will require the corpus-based identification of multiple competing and converging motivations, and their integration into a larger explanatory framework. Relevant forces include the distribution of given and new information across clausal arrays of argument slots, the semantics of force dynamics in the clause, the lexicon of verbs and argument structure constructions, the learning and reanalysis of inherited morphosyntax, recurrent pathways of change and grammaticization, and more. All impact the dynamics of the discourse profile, and all come together to shape the emergence of the system of grammar. Ergativity is too important to be left to the specialists of ergative languages. It is a problem for all linguistics, and a useful one at that. Ergativity invites linguists to investigate the most fundamental structures of grammar, not only in ergative languages but in every language; and to revisit questions of function and structure, of universality and diversity, that were once thought to be settled. As the field of linguistics turns more and more to evidence-based analysis, traditional methods of elicitation are increasingly supplemented by the empirical tools of corpus-based and experimental methods. There is much work to be done to document the discourse profiles that shed light, not only on the syntactic alignments that broadly define ergative, accusative, and active languages, but also on the details of the rich constructional repertoires of more specialized argument structure constructions, such as passives, antipassives, perfectives, and nominalizations, that may serve as bridges for diachronic realignment from one structural type to another. Ergativity has the potential to disrupt conventional thinking and existing explanations for grammar. Let the disruption begin.
Acknowledgments I wish to thank Jessica Coon and Shobhana Chelliah for comments on an early draft that have greatly improved this chapter. For helpful input on specific points I thank Matthew Dryer and Danny Law. For conversations that have deepened my thinking on this and other questions of linguistic explanation, I am profoundly indebted to Mira Ariel.
58 John W. Du Bois
Abbreviations The following symbols and abbreviations are used in glosses: CP, completive aspect; DEP, dependent; DIM, diminutive; ICP, incompletive aspect; FOC, focus; LAT, lative; PL, plural; TV, transitive verb; 1.ERG, ergative 1st person singular; 3.ERG, ergative 3rd person singular; 3.ABS, absolutive 3rd person singular. In addition to the standard A–S–O symbols introduced in the first paragraph, the following are used in the syntactic schemas: X, oblique/adjunct NP; V, verb; P, preposition. Transcription symbols (Du Bois 2014b; Du Bois et al. 1993) include: ; [semi-colon] speaker label; … pause; .. micro-pause; : [colon] prosodic lengthening; (H) in-breath; , [comma] continuing intonation; . [period] final intonation. Intonation units are indicated by a carriage return (one IU per line). Sakapultek data are presented in a practical orthography (Du Bois 2006: 198), in accordance with standards of the Academy of Mayan Languages, with the following special values: x voiceless laminopalatal fricative; j voiceless velar fricative; q voiceless uvular stop; tz voiceless apicoalveolar affricate; nh voiced velar nasal stop (engma). Apostrophe (’) following a consonant represents a glottalized consonant; following a vowel it represents a glottal stop. Double vowels represent phonemic length.
Chapter 3
Parameteri z i ng ergativit y: An i nh e re nt case approac h Michelle Sheehan
3.1 Introduction: Theta-Roles and Inherent Case Many recent (and not so recent) approaches argue that ergative is an inherent case associated with the specifier of little v (see Levin 1983; Mahajan 1989; Woolford 1997, 2006; Massam 1998, 2006; Aldridge 2004; Anand and Nevins 2006; Legate 2006, 2012a; among others):1 (1)
vP DPERG
v' v
...
GB versions of this proposal took inherent case to be assigned at D-structure, as opposed to structural case, which was assigned at S-structure. In Minimalist terms, inherent case can be thought of as a K-projection dominating DP, which is s-selected by a class of thematic heads, or simply as case-valuation coupled to theta-role assignment (Woolford 2006). The inherent-case approach to ergativity is attractive because 1 In the following discussion, I use the term ‘accusative’ to denote languages which lack anything which could be classified as ergative case and ‘non-accusative’ to denote those languages which use ergative case in some way. This is intended to avoid the problem of referring to a rather heterogeneous class of languages as ‘ergative’ (complete with scare quotes).
60 Michelle Sheehan (i) inherent (theta-related) cases appear to be independently needed in order to model the case/agreement properties of accusative languages, making ‘ergativity’ (i.e. non- accusativity) less exotic, and (ii) if ergative is an inherent case, this immediately explains Marantz’s (2000 [1991]) much discussed generalization that non-thematic subjects do not bear ergative, as many others have noted (Woolford 2006; Legate 2012a). Essentially, the inherent approach to ergative case makes four distinct kinds of predictions not made by structural or dependent case approaches: A. Ergative will only occur on (a subset of) arguments externally merged in spec vP. B. The presence of ergative may be independent of transitivity, so we might find ergative subjects without absolutive objects. C. There will be no derived/non-thematic ergative subjects (no ergative expletives, raising to ergative or ergative subjects of passives, ditransitives, or otherwise). D. Ergative case will not be lost in contexts where structural case is not available (no change of case under ECM, no loss of ergative under raising). While C and D can only be evaluated via in-depth language-specific consideration of raising, passives, applicatives, and ECM in non-accusative languages (to the extent that they exist—see Rezac et al. 2014; Berro and Extepare, Chapter 32 and Laka, Chapter 7, this volume, on Basque; Baker and Bobaljik, Chapter 5, this volume on Burushaski), A and B should be much easier to evaluate on a broad cross-linguistic basis as they concern the surface distribution of ergative case.2 Nonetheless, to my knowledge, no systematic cross-linguistic survey of the distribution of ergative case has been given in favour of the inherent-case proposal. The main aim of this chapter, then, is to fill this gap, in the context of a broader parametric account of basic alignment. The obvious challenge in relation to A is that the inherent-case proposal only has predictive power inasmuch as there is an independent theory of theta-roles, distinguishing those arguments introduced by little v from those introduced by other (lower) thematic heads such as Appl and V. While there has been rich cross-linguistic research on argument structure, it is still much debated exactly how many thematic distinctions need to be made syntactically and even how best these thematic distinctions should be described: by distinct theta-positions (Baker 1988, 1997) or by the combination of proto-roles/features (Dowty 1990; Reinhart 2003), possibly accumulated via movement (Ramchand 2011). Nonetheless, certain proposals regarding theta-roles have become widely accepted. There is a general (empirically grounded) consensus, for example, that certain (proto) theta-roles are subject-like (agent, causer, initiator) while others are more object-like (theme, patient, undergoer, result), or lie somewhere in between (experiencer, recipient, process) (Baker 1997, 2009; Platzack 2009, 2011; Ramchand 2011;
2 Of course, the very real possibility exists that ergative has a different inherent/structural/dependent status in different languages. The null hypothesis, though, should be that it has the same status cross-linguistically.
Parameterizing ergativity 61 but see also Bowers 2013).3 Here, I will adopt the conservative position, stemming from Baker (1988), that there are distinct theta-roles, which are configurationally determined.4 As Folli & Harley (2007) point out, there is good reason to take agents and (animate/ inanimate) causers to be introduced by distinct little vs: vdo and vcause respectively. The evidence for distinguishing these two theta-roles comes from the fact that some predicates require agent subjects, and cannot take causers (Oehrle 1976; Hale & Keyser 1993; Folli & Harley 2007, among others). For example, as Oehrle (1976) showed, prepositional datives require agent subjects but the double object construction takes a causer subject, at least with verbs of transfer of possession (see also Pesetsky 1995): (2) a. My relationship with him gave/brought me a daughter. b. *My relationship with him gave/brought a daughter to me. The fact that the same lexical verb occurs in both examples shows, moreover, that this is not a lexical but a syntactic effect. A similar contrast holds of certain transitive verbs in English: those derived from unaccusative change of state verbs allow causer subjects, unlike unergative activity verbs like read, which take only agents (Hale & Keyser 1993): (3) a. *The homework assignment read several books. (Intended. caused the reading of several books) b. The snowfall closed several roads. It has also been observed that in many languages anticausatives permit causers but not agents to be overtly expressed as PPs (headed by from in English) (Alexiadou & Schäfer 2006: 41): (4) a. The window broke from the pressure /from the explosion /from Will’s banging b. *The window broke from John The opposite pattern is observed with the by-phrase in passives in some languages, as Alexiadou & Schäfer also note. Finally, certain ‘causative’ constructions are actually ‘agentive’, requiring an agent and not a causer subject (see Folli & Harley 2007 on Romance ‘causatives’): (5) a. The fact that it was hot in the room made/?let/*had Mary take off her jacket. b. The teacher made/let/had Mary take off her jacket.
3 Assuming that the goal theta-role in prepositional datives is distinct from the recipient theta-role in double object constructions, with recipients being externally merged higher than themes, unlike goals (see Pesetsky 1995; Harley 2002). 4 For our purposes here, it is not important whether theta-roles can be acquired only by external merge or by either external or internal merge (Hornstein 1999), but see Sheehan (2014a) for arguments in favour of the second possibility.
62 Michelle Sheehan The fact that both agents and causers can, nonetheless, be introduced as additional arguments in ‘causative’ constructions and (to varying degrees in different languages) appear as a by-phrase in the passive, provides strong evidence that they are both ‘external arguments’ introduced by little v. A first prediction of the inherent approach to ergative case is that agents and/or causers can appear as ergatives. In many accusative languages, there is also evidence that the subjects of some intransitive predicates are introduced by little v. As Burzio (1986) showed, intransitive verbs divide into those which have a thematic object or theme, externally merged as the complement of V (so-called unaccusatives) and those whose single argument is a thematic agent, now usually taken to be externally merged in spec vP (unergatives), following Koopman & Sportiche (1991).5 A further prediction of the inherent case approach is therefore that, among intransitive subjects, only the subjects of unergatives should be able to surface with ergative case.6 In line with Alexiadou & Schäfer (2006), I take instrument subjects to reduce to either agents or causers. Differences in binding possibilities strongly suggest that subject instruments are externally merged in a higher position than PP instruments: (6) a. *Mary hit himi on the foot with Johni’s baseball bat b. Johni’s baseball bat hit himi on the foot (when it fell off the shelf). It does not seem, however, that there is a dedicated theta-position for instruments high in the clause, but rather that subject instruments behave like either agents or causers. The main evidence for this comes from the fact that not all instruments can surface as subjects (examples adapted from Alexiadou & Schäfer 2006): (7) a. The doctor cured the patient with a scalpel/chamomile tea. b. ??The scalpel cured the patient. c. Chamomile tea cured the patient. (8) a. The crane picked up the crate. b. *The fork picked up the potato. Alexiadou & Schäfer (2006: 42) claim that instruments make good subjects when they can be conceived of as ‘acting on their own without being (permanently) controlled by 5 There is some disagreement with respect to the correct label or semantic characterization of the argument of unergative subjects, partly due to the fact that there is non-trivial cross-linguistic (lexical) variation regarding the unaccusative/unergative divide (see Rosen 1984; Sorace 2000). While these issues are of course interesting and relevant, we abstract away from them here for reasons of space (see Pesetsky 1995 for relevant discussion). 6 It has sometimes been claimed, in fact, that the inherent-case approach predicts that transitive and intransitive little v should pattern alike in this respect. As we shall see, though, this does not necessarily follow if the distribution of inherent ergative is subject to more intricate parameterization (see also Legate 2012a).
Parameterizing ergativity 63 a human agent’, as is the case with chamomile tea and cranes (in German, Greek, and Dutch as well as English). This means that there is no independent theta-role ‘instrument’ introduced by little v, subject instruments are either agents or causers, depending on their semantics. In non-accusative languages, then, we expect to find ergative instruments to the extent that they are semantically permitted, but there might be quite subtle restrictions on their use. Deciding which other kinds of arguments are introduced by little v is more difficult. The subjects of verbs of perception, though often labelled experiencers, actually pattern with agents/causers and unlike other experiencers in certain respects: they can form so- called agentive nominalizations (hearer, feeler, sooth-seer) for example, and can appear in by-phrases in the passive. Note also that verbs of perception also give rise to causative- like constructions in English and other languages (see Guasti 1993), again making them syntactically similar to causers/agents as ‘external arguments’. There are other experiencers too, which pattern like this (lover, cheese-hater, ghost-fearer). Broadly speaking, these are the subjects of stative psych-predicates, which, in many accusative languages, surface as subject-experiencer predicates, taking a nominative subject (Grimshaw 1990). Eventive psych-predicates, on the other hand, often surface as object-experiencer predicates, taking a dative or accusative experiencer (Belletti & Rizzi 1988; Pesetsky 1995; Landau 2009). The fear/frighten contrast illustrates this difference in English: (9) a. I fear ghosts. b. *I am fearing (some) ghosts (right now). (10) a. Ghosts frighten me. b. Some ghosts are frightening me (right now). Building on Platzack (2011) and Ramchand (2011), I take the subjects of (stative) psych- predicates/verbs of perception to be holders rather than experiencers, introduced as external arguments by vhold. The experiencers in eventive object-experiencer constructions, I take to bear the true experiencer theta-role, introduced by a lower thematic head (Appl) (in the spirit of Belletti & Rizzi 1988; Pesetsky 1995; Landau 2010). This means that we might expect to find a split in some non-accusative languages between stative psych-predicates which take an ergative holder and eventive psych-predicates which take a dative or absolutive experiencer. As we shall see, this is exactly what we find in many non-accusative systems. Additionally, the second argument in object-experiencer predicates is persuasively argued by Pesetsky (1995) to be either (i) an internal theme/ target (with the unaccusative piacere class) or (ii) an external causer (with the causative preoccupare class). A further prediction, then, is that the non-experiencer argument of an eventive psych-predicate will also be able to surface with ergative case, again something which is born out in many cases. Following Pylkkänen (2008) among others, I take recipients and benefactives to be introduced by low/high Appl heads respectively. Finally, themes/patients and goals, I assume to be externally merged low down inside VP. This gives the following
64 Michelle Sheehan range of thematic positions in the clause (not all of which can co-occur—see, again, Pesetsky 1995): (11) [vP agent/causer/holder vagent/vcauser/vholder [ApplP benefactive Applbenefactive [VP V [ApplP recipient Applrecipient theme ([PP P goal])]]]]7 The prediction for ergative languages is therefore that only holders, agents, and causers (including instruments and the non-experiencer arguments of eventive psych- predicates) should be able to bear ergative case. In the remainder of this chapter I show that this prediction holds for a range of non-accusative languages, but that languages vary as to exactly which subset of little vs assigns ergative case. In section 3.2 I consider the differing distribution of ergative case in Basque, Hindi, Tsez, Lezgian, Trumai, Cavineña, and Chamorro. I argue that the best way to account for this variation is via a series of parameters arranged in transitive dependencies in the general way proposed by Roberts (2012). The section further considers two additional dependent parameters active only in languages with transitive- sensitive ergative case, determining the presence of syntactic ergativity (understood in the narrow sense) and the source of absolutive case (T or v). Section 3.3 briefly discusses the resultant parameter hierarchy and its theoretical status as well as raising some questions for future research. Finally, section 3.4 concludes.
3.2 The (Parameterized) Distribution of Ergative Case This section considers variation in the distribution of ergative case across languages. It does not consider, for reasons of space, the kind of split-ergativity that Sheehan (2015) calls ‘variable alignment’, whereby the same predicate in the same language displays different case/agreement properties depending on syntactic context (root/embedded and/or tense/aspect properties of the clause, person/animacy of the subject). This kind of variability, I assume, can be attributed to independent facts about the languages in question which serve to obscure basic alignment in certain contexts, rather than to variable parameter settings (see Laka 2006a, Chapter 7, this volume; Coon 2013a; Coon & Preminger 2012, Chapter 10, this volume; see also Woolford, Chapter 9, this volume). As Hindi, Basque, and Tsez show aspect-sensitive variability, Chamorro mood-based
7
Note that the theme appears as the complement of V where no recipient is present. Where a recipient is present, I assume that the theme is the complement of Appl, for the reasons discussed in Pylkkänen (2008). Likewise, where a PP goal is present, the theme is the specifier of P. For this reason, theta-roles are configurationally determined in a relative rather than an absolute sense. Thanks to Lisa Travis for asking me to clarify this point.
Parameterizing ergativity 65 variability, and Yidiɲ personal sensitive variability, I focus on the distribution of ergative case in ergative contexts here, to control for this effect.8
3.2.1 Ergative Unergatives: Basque and Hindi In Western Basque (henceforth Basque), it has been claimed that the subjects of unergatives surface with ergative case, while the surface subjects of unaccusatives surface with absolutive (Laka 2006b).9, 10 (12)
a. Txalupa hondora-tu boat.def.abs sink.perf ‘The boat sank.’
da. is
b. Klara-k ondo eskia-tzen Klara-erg well ski.impf ‘Klara skis well.’
du. has
[Basque] (Laka 2006b: 376)
(Laka 2006b: 379)
The fact that the Basque verb hondura ‘sink’ occurs with an absolutive subject in (12a) follows if the surface subject is base-generated as the complement of V (i.e. it is unaccusative). The surface subject of the (unergative) verb eskia ‘ski’, on the other hand, surfaces as ergative because it base generated in spec vP. This structural difference is also illustrated by the different auxiliaries selected by the two kinds of predicates in Basque (‘be’ vs. ‘have’). While many unergative Basque verbs are N-do compounds and hence might be considered transitive under some definitions (Bobaljik 1993a; Laka 1993b), Laka (2006b), and Preminger (2012) show that this is not true of all unergative verbs, as is obvious from (12b). It therefore seems to be the case that in Basque intransitive little v also assigns ergative case to its specifier.
8
An anonymous reviewer asks about the status of tripartite systems in this approach. Sheehan (2014b) shows one option regarding how such systems might be accommodated in the hierarchy, based on the idea that absolutive on transitive objections in some languages is underlying equivalent to accusative Case (Legate 2006, 2012a). This would mean that whether a low ABS non-accusative language is ergative or tripartite is purely a matter for the morphology. I leave a discussion of this complication to one side here for reasons of space (see also Müller and Thomas, Chapter 12, this volume for arguments that genuine tripartite systems do not exist). 9 The following languages have been argued to behave similarly in this respect: Georgian, Kartvelian (Harris 1982); Chol, Mayan (Coon 2013a); Lakhota, Siouan; Caddo, Caddoan (Mithun 1991b), Hindi (Bhatt 2003). We consider Hindi shortly. 10 Guaraní also displays a split regarding the behaviour of intransitive verbs. While Mithun (1991b) characterizes this as an active–stative split, Velázquez-Castillo (2002) suggests that matters are more complex than this. Like the other split-S systems discussed here, Guaraní is not syntactically ergative (Velázquez-Castillo 2002), but it is not totally clear at present how to fit this language into the parameter hierarchy presented here. I therefore leave this as a matter for future research.
66 Michelle Sheehan In fact, closer examination of Basque suggests that all arguments introduced by little v in transitive and intransitive contexts seem to bear ergative case. Consider first agents and animate/inanimate causers:11 (13) Maddi-k sagarr-a jan du. Maddi-erg apple-det eat 3sabs.aux.3serg ‘Maddi ate the apple.’
[Basque] (Oyharçabal 1992: 313)
(14) (Nik) zuri lan egin arazi dizut. 1.erg you.dat work do caus aux.2dat.1erg ‘I made you work.’ (Oyharçabal 1992: 332 fn.) (15) Haize-a-k ate-a ireki du. wind-det-erg door-det open 3sabs.aux.3serg ‘The wind opened the door.’ Even instruments can surface with ergative case in Basque, as noted by Woolford (2006). This follows if, as discussed, subject instruments are actually causers/agents: (16) Giltz-a-k ate-a ireki zuen. key-det-erg door- det open 3sabs.aux. pst.3serg ‘The key opened the door.’ (Woolford 2006: 124, citing unpublished work by Juan Uriagereka) Turning to experiencers, we find that the latter surface either with dative, absolutive, or ergative case, depending on predicate type (Etxepare 2003). These three options seem to be equivalent to the three kinds of psych-predicates identified by Pesetsky (1995) and Landau (2010), building on Belletti and Rizzi (1988). Thus, unaccusative object experiencer verbs (the piacere class) surface with a dative experiencer and an absolutive theme/target (see Etxepare 2003; Rezac 2008): (17) Ni-ri zure oinetako-a-k gustatzen zaizkit. I-dat your shoes-det-pl like 3plabs.aux.1sdat ‘I like your shoes.’ (Woolford 2006: 115, citing Austin and Lopez 1995: 12) Other verbs in this class include interesatu ‘to interest’ and dolutu ‘to repent’ (Etxepare 2003: 39). Causative object experiencer psych-predicates follow an erg–abs pattern, whether or not the causer is animate (and hence potentially agentive) or inanimate:
11
Thanks to Maia Duguine for help with the Basque data.
Parameterizing ergativity 67 (18)
Mikel-ek ni Michael-erg I.abs ‘Michael angered me.’
haserretu nau. angry-perf 1sabs.aux.3serg (Woolford 2006: 124, citing Manandise 1988: 118)
(19)
Berri-ek (ni) haserretu naute. new-det.pl.erg I anger 1sabs.aux.3plerg ‘The news angered me.’
Verbs in this class, which often alternate with an intransitive form, include kezkatu ‘to worry, to become worried’, gogoratu ‘to remember’, zoratu ‘to madden’ (Etxepare 2003: 41). The ergative here is as expected if the non-experiencer argument is a causer rather than a theme, as discussed. Moreover, the absolutive case on the experiencer is plausibly equivalent to the accusative case received by experiencers in the equivalent accusative structures (the preoccupare class) (but see Landau 2009 for complications). There is a class of experiencers, though, including the subjects of verbs of perception, which surface with ergative case (including miretsi ‘to admire’, gutxietsi ‘to despise’, desiratu ‘to desire’—Etxepare 2003: 41) (20) Ni-k asko ikusi ditut I-erg many.abs seen 3plabs.aux.1serg ‘I saw many.’ (21)
Jon-ek liburu hauek nahi/ezagutzen Jon-erg book these want/know ‘John wants/knows these books.’
(Rezac et al. 2014: 6)
ditu. 3plabs.aux.3serg
These appear to be equivalent to subject-experiencer (temere class) verbs in accusative languages. The case patterns observed in Basque are therefore exactly as expected if all little vs assign ERG to their specifiers and thematic structure in Basque follows expected universal patterns.12 Hindi shares many of these properties with Basque even to the extent that unergative but not unaccusative predicates can surface with ergative subjects. In Hindi, though, few intransitive verbs actually require ergative subjects, the verb ‘to bathe’ being a rare exception in this respect: (22)
Raam*(-ne) nahaayaa Ram-erg bathe.perf ‘Ram bathed.’
(Mohanan 1994a: 71)
12 It may turn out, of course, that the inherent case approach to Basque fails for reasons C–D (as Rezac et al. 2014 argue). If this is the case, the implication is merely that the approach to non-accusative alignment here is incomplete as there are other kinds of ergative languages in which ergative is not an inherent case. I leave this as a matter for ongoing research. See also Laka, Chapter 7, this volume for a defense of the inherent ergative approach to Basque.
68 Michelle Sheehan The verbs which behave in this way appear to have a reflexive meaning. In most cases, intransitive verbs take ergative subjects only optionally. More precisely, a subset of the class of verbs which satisfy independent diagnostics for unergativity can surface with either an absolutive or ergative subject (Bhatt 2003; Butt, Chapter 33, this volume; Surtani, Jha, & Paul 2011, Surtani & Paul 2012).13 This includes jhool ‘swing’, dauR ‘run’, kood ‘jump’ naac ‘dance’, hans ‘laugh’ tair ‘swim’, gaa ‘sing’, kʰel ‘play’ and chillaa ‘shout/ scream’ (Mahajan 1990; Mohanan 1994a; Surtani, Jha, & Paul 2011). As Mohanan (1994a) shows, however, the choice between an absolutive and ergative subject with these verbs is not purely optional but correlates with a semantic difference: ergative subjects imply volition in intransitive contexts: (23) raam-ko acaanak šer dikʰaa. vah/us-ne cillaayaa [Hindi] Ram-dat suddenly lion.abs appear.perf he/he.obl-erg scream.perf ‘Ram suddenly saw a lion. He screamed.’ (Mohanan 1994a: 71) (24) Us-ne/*vah jaan buujʰkar cillaayaa He.obl-erg/he.abs deliberately shout.perf ‘He shouted deliberately.’
(Mohanan 1994a: 72)
Verbs which can be independently diagnosed as unaccusative, however, never take ergative subjects, even where volition is implied (gir ‘fall’ soo ‘sleep’, jaa ‘go’, phail ‘spread’): (25) Raam(*-ne) giraa Ram-erg fall.perf ‘Raam fell hard.’ The implication seems to be that unergative verbs in Hindi can, but need not, take ergative subjects; where they do, the subject is interpreted as volitional. These facts are consistent with ergative being an inherent case but they are not immediately explained by such an account. What is required is some further parameter differentiating intransitive little v in Basque from intransitive little v in Hindi.14 Because of these facts, Mohanan (1994a) proposes to associate ergative case directly with the semantic feature [volition] across the board. The problem with this idea,
13
These diagnostics include (i) the possibility of a cognate object, (ii) participation in impersonal passives, (iii) non-participation in participial relatives, (iv) compound verb selection of le ‘take’, de ‘give’, daal ‘did’ and not jaa ‘go’, and (v) inability to appear without genitive marking in non-finite clauses. 14 Another complication in Hindi is that a small number of transitive verbs (verbs taking absolutive/ ko-marked objects) fail to assign ergative case (bolnaa ‘speak’ and laanaa ‘bring’) or do so only optionally (samajʰnaa ‘understand’ and jannaa ‘give birth to’) (see Mohanan 1994a). These appear to be idiosyncratic lexical gaps.
Parameterizing ergativity 69 though, is that ergative case assignment with transitive predicates is insensitive to [volition]. In fact, the distribution of ergative case across transitive predicates in Hindi follows very closely the Basque distribution. Non-volitional animate and inanimate causers receive ergative just as volitional agents do. This is true of instruments as well as the causers in object-experiencer constructions:15 (26) Havaa-ne patte bikʰer diye tʰe Wind-erg leaves scatter give.perf be.pst ‘The wind scattered the leaves.’
(Mohanan 1994a: 75)
(27) [Mina-ke cillaa-ne]-ne sab-ko Daraa diyaa Mina-gen scream-inf/ger-erg all-dom scare give.perf Mina’s screaming scared everyone. (28) nayii khabaroN-ne Sita-ko dukhii kar diyaa new news-erg Sita-dom sad do give.perf The new news saddened Sita. (29) caabhii-ne taalaa khol-aa key-erg lock open.tr-perf The key opened the door. Finally, holders of states also receive ergative with both verbs of perception and subject-experiencer verbs: (30) tuṣaar-ne caand dekʰaa Tushar-erg moon see/look.at.perf ‘Tushar saw the moon.’ (31)
tuṣaar-ne vah kahaanii yaad kii Tushar-erg that story memory do.perf ‘Tushar remembered that story.’
(Mohanan 1994a: 141)
(Mohanan 1994a: 141)
Eventive object-experiencers versions of these verbs surface with dative subjects: (32) tuṣaar-ko caand dikʰaa Tushar-dat moon appear.perf ‘Tushar saw the moon.’ (lit. The moon appeared to Tushar.)
15
Thanks to Rajesh Bhatt for help with the Hindi data.
(Mohanan 1994a: 141)
70 Michelle Sheehan (33)
tuṣaar-ko vah kahaanii yaad aayii Tushar-dat that story memory come.perf ‘Tushar remembered that story.’ (Mohanan 1994a: 141) (lit. The memory of that story came to Tushar.)
The correct characterization of the distribution of ergative case appears to be that it surfaces wherever v is transitive and additionally in intransitive volitional contexts (a non- natural class). The pattern observed in Basque falls out straightforwardly from a parametric approach along the lines proposed by Roberts (2012). Assuming that the basic alignment parameter concerns whether transitive v assigns a theta-related case, an alignment of the Basque kind arises where ergative is generalized to all little vs via ‘input generalization’, an acquisition strategy. The Hindi system arises where instead of generalizing the transitive system to all intransitives, the system is simply extended to a subset of all possible contexts (i.e. volitional intransitive vs). The non-natural class of ergative DPs arises as a result of this extension procedure. This can be represented via the following parameter hierarchy: (34) Parameter hierarchy (first version, to be extended and revised) P1: Does transitive v assign a theta-related case (ERG) in a language L? N - accusative
Y– non-accusative P2: is this generalised to all vs in L? Y – Basque
N – P3: is this extended to intransitive [volition] vs in L? Y – Hindi
N-…
The rather idiosyncratic Hindi system therefore reduces to an extension of the basic non-accusative system. In the following sections, we see that further parameterization is also required in order to account for the distribution of ergative case in other contexts.
3.2.2 Instigators Only: Tsez and Lezgian The distribution of ergative case in Tsez and Lezgian suggests that a further kind of parameterization must be added to (34). In these languages, transitive agents and animate/inanimate causers receive ergative case, just as in Basque and Hindi (Comrie 2004; Polinsky 2015. on Tsez; Haspelmath 1993 on Lezgian):16 16 Tsez also marks accidental causation via the poss-essive case, though this is optional (Comrie 2004: 118; Polinsky 2015: 144–146) and does not seem to occur on inanimate causers (Bernard Comrie and Maria Polinsky, p.c.).
Parameterizing ergativity 71 (35)
žek’-ā bišwa r-ac’-xo man-erg food.abs.iv iv-eat-pres ‘The man is eating the food.’
[Tsez] (Comrie 2004: 115)
(36) žek’-ā is b-exu-r-si Man-erg bull.abs.III III-die-caus-pst.wit ‘The man killed the bull.’
(Comrie 2004: 116)
(37) C’i-d-ä ʕаˤɣur y-iku-r-si. fire-erg mill.abs.II II-burn.intr-caus-pst.wit ‘Fire burnt the mill.’
(Polinsky 2015: 138)
[Tsez]
(38) Nes-ä ža kaɣat kid-be-q t’et’r-er-si. dem.i-erg dem letter.abs.ii girl-os-poss.ess read-caus-pst.wit ‘He made the girl read that letter.’ (Polinsky 2015: 160) An apparent difference between the languages concerns instruments. In Tsez, instruments can also surface with ergative case, as expected: (39) Yiła rek-ä ħišimuku r-aˤɣi-x. dem key-erg lock.abs.iv iv-open-pres ‘This key opens the lock.’
[Tsez] (Polinsky 2015: 139)
While this was also previously true of Lezgian instruments, Haspelmath (1993: 84) claims that ergative instruments are ‘never used in the modern language’, raising some potential issues for the collapsing of instrument subjects with agents/ causers. This is where the similarities with Hindi and Basque end, however. The subjects of intransitive verbs always receive absolutive case in Tsez, regardless of the unaccusative/ unergative distinction, which is independently diagnosable in the language (Polinsky 2015: 121–123): (40) is b-exu-s bull.abs.III III-die-pst.wit ‘The bull died.’ (41) Ecru žek’u qoqoƛi-s old man.abs laugh-pst.wit ‘The old man laughed.’
[Tsez]
(Comrie 2004: 115)
While this is also generally the case in Lezgian, there is a class of verbs derived from N- do compounds which take ergative subjects despite appearing intransitive:
72 Michelle Sheehan (42) ada k’walax-zawa she.erg work-impf ‘She was working.’
[Lezgian]17 (Haspelmath 1993: 284)
Based on a range of facts, Haspelmath proposes that such examples involve incorporation of the nominal into the (light) verb do, giving rise to absolutive absorption in an otherwise transitive structure.18 In other cases, though, Lezgian makes no distinction between unaccusative and unergative verbs, with both taking absolutive subjects (e.g. change of state verbs such as kisun ‘fall silent’, kusun ‘fall asleep’, ifin ‘become hot’, q̃un ‘become cold’, q’ežin ‘become wet’ and the verb of motion juzun ‘move’, which are presumably unaccusative, as well as verbs which are presumably unergative: q̃uǧun ‘play’, qʰürün ‘laugh’, q̃eq̃ün ‘walk around’—Haspelmath 1993: 271): (43) a. stxa k’wal.i-z xta-na brother.abs house-dat return-aor ‘The brother came back home.’
[Lezgian] (Haspelmath 1993: 5)
b. q’if wiči-n t’ekwen galaj.pataqʰ kat-na. mouse.abs self-gen hole toward run-aor ‘The mouse ran towards its hole.’ (Haspelmath 1993: 223) In both Lezgian and Tsez, then, intransitive v does not assign ergative, assuming that in examples like (42), little v is formally transitive.19 Another difference between Tsez/Lezgian vs. Basque/Hindi is that holders never seem to surface with ergative case in the former. The subjects of verbs of perception and psych-predicates surface with dative in Lezgian and either lative or absolutive in Tsez: (44) Zamira.di-z Diana aku-na. Zamira-dat Diana see-aor ‘Zamira saw Diana.’ (lit. Diana was visible to Zamia.)
17
[Lezgian] (Haspelmath 1993: 270)
Further examples include the verbs meaning ‘howl’ and ‘dance’ (Haspelmath 1993: 284). This recalls Hale and Keyser’s (1993) analysis of unergative verbs as well as Bobaljik’s (1993a) account of Basque. 19 In Lezgian, verbs taking an oblique complement can also surface with either an absolutive or ergative subject. As is the case with Warlpiri (Legate 2012a) this appears to depend on the thematic status of the subject. Agents appear to surface with ergative (q̃arǧi šun ‘curse’, ewerun ‘call’, ikramun ‘bow to’, taʔsirun ‘influence’, hürmetun ‘respect’) (Haspelmath 1993: 284), whereas non-agents are absolutive alatun ‘falls off, passes, exceeds’, eläč’un ‘crosses’, agaq’un ‘reaches’ ac’un ‘becomes full of ’, gaw ‘is near’ (Haspelmath 1993: 272–277). An apparent outlier is raxun ‘talks to’, which takes an absolutive subject, despite being apparently agentive. 18
Parameterizing ergativity 73 (45) Elu-r mašina c’aq’ b-et-äsi yoł. [Tsez] 1pl-lat car.abs.iii very iii-want-res.ptcp aux.pres ‘We badly need a car.’ (lit.: car is wanted to us) (Polinsky 2015: 156) (46) [Yedu kid] [meži-z ɣwˤay-q] y-uƛ’-xo. dem girl.abs.ii 2pl-gen2 dog-os-poss.ess ii-fear-pres ‘This girl is afraid of your dog.’ (lit.: fears on your dog)
(Polinsky 2015: 157)
Polinsky (2015) further shows that these verbs can be causativized in Tsez, giving rise to two different patterns. The first pattern is as expected: an external causer is added: (47) Eni-y-ä debe-q yedu čorpa b-et-ir-xo. mother-os-erg 2sg-poss.ess dem.ni soup.abs.IV IV-like-caus-prs ‘The mother is making/will make you like this soup.’ (Polinsky 2015: 165) In the second pattern, however, causativization serves merely to alter the case and theta- role of the subject, making it ergative and agentive: (48) Madin-ä [gagali-s maħ] b-iy-r-si. Madina-erg flower-gen smell.abs.III III-know-caus-pst.wit ‘Madina smelled flowers.’ [ERG-agent, ABS-stimulus] (Polinsky 2015: 163) Based on other diagnostics such as binding and Control, it appears that the like-type verbs are unaccusative and equivalent to the piacere class in Italian (see Polinsky 2015: 163–166), meaning that the smell-type verbs are presumably subject-experiencer verbs. If this is the case, then holders in these languages are not ergative. Finally, Tsez also has what appear to be object-experiencer verbs of the preoccupare class which take an ergative causer (ambiguous between direct and indirect causation) and an absolutive experiencer: (49) Meži-z ɣwˤay-ä kid y-uƛ’-er-xo. 2pl-gen dog-erg girl.abs.ii ii-fear-caus-pres ‘Your dog frightens the girl.’
[Tsez] (Polinsky 2015: 168)
In this way, both Tsez and Lezgian limit ergative to (animate and inanimate) causers and agents, with dative case surfacing on holders.20 In order to capture the basic 20
A further difference, not encoded in the hierarchy is that Lezgian but not Tsez can have ergative subjects in the absence of absolutive case with a small class of intransitives (derived from N-do compounds) and with bivalent verbs taking oblique complements. It seems that this difference is due to an independent parametric difference between the languages regarding what counts for transitivity, a discussion of which would take us too far afield here.
74 Michelle Sheehan behaviour of these languages a further kind of parameter must be added to the proposed hierarchy: (50) (Parameter hierarchy (second version, to be extended and revised)
P1: Does transitive v assign a theta-related case (ERG) in a language L? N - accusative
Y – non-accusative P2: is this generalised to all vs in L? Y – Basque
N – P3: is this extended to intransitive [volition] vs in L? Y – Hindi
N – P4: is this restricted to [instigator] vs in L?
Y – Tsez, Lezgian
N…
The pattern in Tsez and Lezgian suggests that in addition to generalizing and extending the distribution of ergative case it must also be possible to restrict it to a subset of transitive vs. As discussed in section 3.3, the parameters in (50) are not intended to be pre-specified by Universal Grammar, nor is their format intended to be fixed. Rather, the hierarchy, it is proposed, emerges as the result of acquisition based on the acquisition strategies of feature economy, input generalization and analogy (see Roberts 2012).
3.2.3 Default Transitive Ergativity: Trumai and Chamorro The languages discussed up to now are all morphologically rather than syntactically ergative in that they do not have syntactic operations sensitive to the transitive/intransitive subject distinction. It is well known, however, that many non-accusative languages do display a form of syntactic ergativity in that they prohibit (straightforward) A-bar extraction of ergative DPs.21 Following the general approach in Aldridge (2004, 2008a, 2008b) and Coon et al. (2014), I assume that this property results from the presence of a movement-triggering EPP feature on ergative-assigning little v.22 21 An anonymous reviewer asks about the other apparent instances of syntactic ergativity such as topic chaining in Dyirbal. These apparent instances of syntactic ergativity remain controversial (see Legate 2012a). The ban on A-bar extraction is a much more robust effect, attested in many unrelated languages, and I therefore limit the discussion to this narrower definition of syntactic ergativity here (see also Aldridge 2008a). The typologically and genetically diverse languages displaying this restriction include some Mayan languages (Assmann et al. 2012; Aissen, Chapter 30, this volume; Campana 1992; Coon et al. 2014; ), some Austronesian languages (Tongan, Otsuka 2006; Otsuka, Chapter 40, this volume; Tagalog, Seediq, Aldridge 2004; 2012b; Chamorro, Chung 1982), some Eskimo–Aleut languages (Manning 1996) as well as Australian and Brazilian languages: Dyirbal (Dixon 1979, 1994), Trumai (Guirardello-Damian 2010), Karitiana (L. Storto, p.c.), and Katukina (Queixalós 2012; Queixalós, Chapter 42, this volume). 22 As an anonymous reviewer reminds me, the idea that syntactic ergativity can be attributed to movement of the absolutive DP past the ergative DP was first proposed by Bittner and Hale (1996a).
Parameterizing ergativity 75 (51)
vP DPABS
v' DPERG
v' vEPP
VP V
t
This feature attracts the absolutive argument to the external specifier of v and serves to trap the ergative DP inside the vP phase.23 Implementations of this basic idea differ in Aldridge and Coon et al.’s work. I assume that it is an effect of anti-locality (see also Erlewine 2016). This follows if only the outermost specifer of vP can avoid the phase impenetrability condition (PIC). In this way the absolutive DP is always available for further extraction and lower adjuncts/locatives can raise to a further external spec vP, avoiding the PIC, past both the ergative and absolutive DPs. The ergative DP is, however, doomed to be trapped inside vP as it can never raise to the external specifier of vP, due to anti-locality (one cannot raise from the internal specifier of a given head to its external specifier).24 In parametric terms then, syntactic ergativity of this kind arises where a dependent parameter associates an EPP feature with the ergative property, in the following way (see also Sheehan 2014b): (52) Parameter hierarchy (third version, to be extended and revised)
P1: Does transitive v assign a theta-related case (ERG) in a language L? N - accusative
Y – non-accusative P2: is this generalised to all vs in L?
Y – Basque
N – P3: is this extended to intransitive [volition] vs in L?
Y – Hindi Y – Tsez, Lezgian
N – P4: is this restricted to [instigator] vs in L? N – P5: does verg bear an EPP in L?
N – Warlpiri, Cavineña(?)
Y – Chamorro, Trumai
A direct consequence of (52) is that P5 (the syntactic ergativity parameter) is only active in languages which answer Y to P1 and N to P2–P4. In other words, a prediction is that syntactic ergativity will be possible only in languages which have default (transitive) 23 Note that this movement must be covert in VSO languages (Aldridge 2004). In many languages, it nonetheless triggers a Diesing-type effect on the absolutive DP. I leave a full exploration of the relationship between syntactic ergativity, object interpretation, and word order to one side here. 24 One assumption that needs to be made is that this movement does not lead to ‘tucking in’ (Richards 1997).
76 Michelle Sheehan ergative alignment rather than generalized (Basque), extended (Hindi), or restricted (Tsesz, Lezgian) ergative alignment. This is a one-way implication, though, and there can be languages which are only morphologically ergative, such as Warlpiri, but which have default ergative alignment (ergative on transitive causers, agents, and holders—Legate 2012a).25 Chamorro is an example of a syntactically ergative language displaying default non- accusative alignment, whereby ergativity is tied very closely to transitivity. In the realis mood, the verb in Chamorro displays ergative agreement with transitive subjects only, but this agreement is lost where such subjects are A-bar extracted, and the infix -um-is added instead (Chung 1982). This kind of morphological compliance strategy is common in syntactically ergative languages (see Dixon 1994 on antipassives; Coon et al. 2014, Erlewine 2016, on Mayan Agent focus): (53) a. Ha-fa’gasi si Juan i kareta. erg.3s-wash unm Juan the car ‘Juan washed the car.’ b. Hayi f-um-a’gasi i kareta? who um-wash the car ‘Who washed the car?’
[Chamorro]
(Chung 1982: 49)
Extraction of objects and intransitive subjects, however, can proceed straightforwardly:26 (54) Hayi na famalao’an man-ma’pus ? who l women pl-leave ‘Which women left?’
(Chung 1982: 46)
(55) Hafa ha-fahan si Maria gi tenda? what erg.3s-buy unm Maria loc store ‘What did Maria buy at the store?’
(Chung 1982: 51)
In addition to agents, (animate/inanimate) causers trigger ergative agreement, including the subjects of object experiencer verbs, as do holders (chat-‘hate’ patterns
25 An anonymous reviewer points out that some Mayan languages (Ixil and Chuj) are syntactically ergative in this sense but require unaccusative subjects to be co-indexed by set A (ergative) marking on the verb in durative/progressive aspect (see Assmann et al. 2015). Crucially, these apparently ergative intransitive subjects can be A-bar extracted unlike their transitive counterparts. While a full discussion of these facts is beyond the scope of the present discussion, Coon (2013a) and Coon and Preminger (Chapter 10, this volume) argue convincingly that aspect-sensitive splits of this kind in Mayan result from embedded nominalizations, so that the set A marking is actually genitive, rather than ergative, case (the two are often homophonous in Mayan). For this reason, these examples are only a superficial counterexample to the prediction. 26 Where obliques are extracted, the clause must be nominalized, however (Chung 1982: 51). It is not clear why this should be the case or to what extent this holds in other syntactically ergative languages.
Parameterizing ergativity 77 like ya- ‘like’, ga’o- ‘prefer’, gusto- ‘like’, ga’ña- ‘prefer’ in this respect—Chung 1982, 1998, p.c.): (56) Hu-punu’ i lalu’ ni niuis. erg.1s-kill the fly obl newspaper ‘I killed the fly with the newspaper.’
[Chamorro] (Chung 1982: 51)
(57) a. Ha-istotba ham [na malagu’ i lahi-nmami ni kareta]. erg.3s-disturb us comp want the son-our obl car ‘That our son wants the car disturbs us.’ b. Ha-istototba yu’ si Juan. erg.3s-disturb.impf me unm Juan ‘Juan is disturbing me.’ (58) Ha-chatli’i’ yu’ atyu na taotao. erg.3s-hate me that lk man ‘That man hates me.’
(Chung 1982: 54–55)
(S. Chung, p.c.)
In causative object-experiencer constructions, experiencers are absolutive (i.e. they do not trigger agreement as in (57)). Where they take an oblique complement, psych- predicates also take an absolutive subject: (59) Ma’a’niao yu’ ni ga’lagu. fear I obl dog ‘I am afraid of the dog.’
[Chamorro] (Chung 1982: 51)
The subjects of intransitive predicates never trigger ergative agreement, regardless of the thematic status of the subject (in realis mood). In all cases, then, it seems that v inflects for ergative agreement exactly where it is transitive, due to its positive setting of P1 (and negative setting of P2–P4).27 Trumai is another syntactically ergative language which displays a strong connection between transitivity and ergative case (Guirardello-Damian 2003, 2010). In Trumai where an absolutive argument (subject or object) is relativized, the verb is modified by the relativizer ke: (60) Ha hu’tsa chï_in [axos-a-tl]i [Øi esa-t’ ke] [Trumai] I see foc/tense child-ev-dat dance-nzr.past rel ‘I saw the boy who danced.’ (Guirardello-Damian 2010: 218)
27
No data is available regarding the behaviour of instrument subjects (S. Chung, p.c.).
78 Michelle Sheehan Where a (transitive) ergative subject is relativized, however, the verb is modified by chïk:28 (61) Ha hu’tsa ka_in [axos-a-tl]i [ha aton mud husa-t’ chï-k Øi] I see foc/tense child-ev-dat 1 pet neck tie-nzr.past rel ‘I saw the boy who tied my pet.’ (Guirardello-Damian 2010: 219) As in Chamorro, agents and animate/ inanimate causers alike take ergative case (Guirardello-Damian 2003), including the subjects of causative object experiencer verbs (Guirardello-Damian 2010: 221): (62) [sud yi]-k [pike xop yi] mahan. wind yi -erg house mouth yi close ‘The wind closed the door.’
[Trumai] (Guirardello-Damian 2003: 201)
(63) [martelu yi]-k [atlat] mapa. hammer yi-erg pan break ‘The hammer broke the saucepan (by falling on it).’ (Guirardello-Damian 2003: 201) (64) hai-ts Yakairu-ø sa ka 1-erg Yakairu dance caus ‘I made Yakairu dance.’
(Guirardello-Damian 2003: 210)
It is even possible to have two ergatives following the causativization of a transitive verb: (65) Hai-ts chï_in Atawa-k atlat-ø mapa ka 1-erg foc/tense Atawak-erg pan-abs break caus ‘I made Atawak break the pan.’ (Guirardello-Damian 2003: 210) Somewhat surprisingly, instruments are reported never to be ergative: (66) chavi letsi [ pike xop yi] mahan. [Trumai] key instr house mouth yi close Lit. ‘pro closed the door with a key.’ (Guirardello-Damian 2003: 201) An apparent difference between Chamorro and Trumai is that in the latter, holders are never ergative, but rather surface with absolutive case (with verbs of perception like see, hear, smell, feel as well as subject experiencer verbs: like, think, believe, forget, remember):
28
Note, though, that this is true also where recipients are relativized (Guirardello-Damian 2010: 219).
Parameterizing ergativity 79 (67) axos-ø hu’tsa de kasoro-tl child-abs see already puppy-dat ‘The child saw the puppy.’
[Trumai] (Guirardello-Damian 2003: 204)
A crucial point here, though, is that the theme/target complement in such cases is always dative, meaning that the verbs in question are not formally transitive. Note that this situation is different from that displayed in Tsez/Lezgian where it is the holder which receives dative/lative while the theme/target is absolutive. In Trumai, unlike in Warlpiri and Lezgian oblique arguments can never count for transitivity. This is true even where the subject is an agent: if the complement of V is oblique, then the subject remains absolutive. This is the case with the heterogeneous class of verbs of routine events translating variously eat, drink, cook, roast, kiss, hunt, fish, which take absolutive subjects and dative complements. In fact, as Guirardello-Damian notes, many abs– dat verbs are direct synonyms of erg–abs verbs: kapan/chuda ‘make/produce’, disi/ fa ‘kill/beat’ tako/make ‘bite’, tuxa’tsi/dama ‘pull’, padi/fatlod ‘wait’. As long as dative objects never count for transitivity in Trumai, then it shares with Chamorro the default ergative alignment whereby ergative case is simply tied to transitivty. Intransitive verbs, whether unergative or unaccusative always take absolutive subjects (Guirardello- Damian 2003: 196).29
3.2.4 Parameterizing the Source of Absolutive Case A final point of variation between non-accusative languages concerns the source of absolutive case (Aldridge 2004, 2008a; Legate 2006, 2012a; Coon et al. 2014). Whereas in some languages it appears that absolutive case has a mixed source, coming from v in transitive and T/Asp in intransitive contexts (mixed ABS), in other languages it appears to come from T/Asp across the board, leading to what has been called a ‘high ABS’ system. The evidence for this comes from the distribution of absolutive in non-finite contexts, where T/Asp loses it structural Case-assigning capabilities. It has been observed that at least in some accusative languages, non-finite T/Asp fails to assign nominative case and, for this reason, the only possible subject of such clauses is PRO (possibly derived via movement, at least in some cases—Hornstein 1999; Sheehan 2014a). In some non-accusative languages, we see that the absolutive on transitive objects is retained in non-finite contexts, suggesting it does not come from T/Asp, but from v. In other cases, we find that absolutive case in not straightforwardly available in non-finite clauses at all, so that the transitive object must be licensed in some special way, if transitive control is possible. In these high ABS languages, then, it seems that absolutive always comes from T/Asp. 29
As noted, an independent parameter is needed to govern whether obliques count for transitivity.
80 Michelle Sheehan Chamorro appears to be a high ABS language in these terms. The evidence for this is that the infix -um-, which surfaces where ergative subjects are extracted, also surfaces where a transitive predicate appears in a control context (Chung 1982: 49, fn. 5): (68) Malagu’ gui’ b-um-isita want he um-visit ‘He wants to visit Rita.’
si Rita. unm Rita
[Chamorro]
Plausibly, -um-serves to license the absolutive DP in (68), as is the case with the “crazy antipassive” in high ABS Mayan languages (Coon et al. 2014). In Tagalog, on the other hand, ABS is retained on the transitive object in non-finite contexts because it is a mixed ABS language (see Aldridge 2004, 2013b): (69) Nagba-balak ang babae-ng [PRO intr.prog-plan abs woman-lk ‘The woman is planning to help the man.’
tulung-an ang lalaki] [Tagalog] help-appl abs man (Aldridge 2013: 2)
There is some evidence that Trumai is also high ABS in these terms. Verbs like padi ‘wait’ can take reduced clausal complements which appear to give rise to ECM, whereby what would be the absolutive argument of the embedded clause surfaces as an enclitic on the matrix verb: (70) hai-ts chï_in [Kumaru-k tïchï] padi-n 1-erg foc/tense Kumaru-erg scarify wait-3abs ‘I waited for Kumaru to scarify her.’ (71) hai-ts [huma] padi-n 1-erg take.bath wait-3abs ‘I waited for her to take a bath.’
[Trumai]
(Guirardello-Damian 2010: 220–1)
The fact that this process applies uniformly to absolutive subjects and objects in Trumai, whereas ergative case is retained (as in (70)), suggests that both get case from T in finite contexts. In non-finite contexts, T fails to assign absolutive case, but ergative, from v, is still available, as expected. High ABS languages appear to be a proper subset of syntactically ergative languages and given standard assumptions about intervention, there is a principled reason why this should be the case. In order for a lower DP1 to receive structural Case from a higher head where another DP2 intervenes, DP1 must first move past DP2. This is essentially the movement proposed to hold in syntactically ergative languages: v bears an EPP feature and attracts DP1 past DP2. No intervention obtains for this movement because DP2 occupies the specifier position of vP and so is not c-commanded by v. To ensure that DP1 receives Case from this higher head (T), though, it must also be the case that it has not received Case from v, prior to movement. In this way, the parameter distinguishing mixed ABS and
Parameterizing ergativity 81 high ABS languages is that which determines whether v loses its ability to assign a structural Case. In a mixed ABS language, v retains this ability (which is the default option for transitive vs—Burzio 1986). In a high ABS language, on the other hand, v loses the ability to assign a structural case (i.e. it bears no ɸ-features) and so DP1 receives Case from T, after movement has occurred.30 This final dependent parameter is thus active only in syntactically ergative languages for principled reasons, giving the following parameter hierarchy: (72)
Parameter hierarchy (fourth version)31 P1: Does transitive v assign a theta-related case (ERG) in a language L? N–accusative
Y – non-accusative P2: is this generalised to all vs in L?
Y – Basque
N – P3: is this extended to intransitive [volition] vs in L?
Y – Hindi
N – P4: is this restricted to [instigator] vs in L?
Y – Tsez, Lezgian
N – P5: does verg bear an EPP in L?
N – Warlpiri, Niuean(?) Y – P6: are verg’s -features suppressed in L? N – West Greenlandic, Tagalog, Cavineña(?) Yidi , Kuikuro, Tongan
Y – Chamorro, Mam, Q’anjobal, Dyirbal, Seediq, Trumai
3.3 The Status of the Hierarchy The parameter hierarchy in (72) serves to model micro-parametric variation among non-accusative languages. What is given, I assume, are the functional categories themselves, formal features such as EPP and ɸ (leading to structural Case valuation) and a requirement for nominal licensing. The other properties of v are open to parameterization: whether v is overt/covert; whether it assigns a theta-related Case (always, never, 30 This raises the question what happens to T’s ɸ-features in mixed ABS languages’ transitive clauses, where both arguments get Case from v. Either T must simply lack ɸ-features in this context or we must assume, following Preminger (2011a) that unvalued ɸ-features fail to crash the derivation. 31 With extra languages tentatively added for purposes of illustration: see Dixon (2010) on Yidiɲ; Massam (1998, 2006) on Niuean; Otsuka (2006) on Tongan; Guillaume (2008) on Cavineña; Aldridge (2004) on Seediq and Tagalog; Franchetto (2010) on Kuikuro; Coon et al. (2014) on Mam and Q’anjobal. It is not actually possible to say, as of yet, whether Cavineña is syntactically ergative or not, though there is suggestive evidence that it is (Guillaume 2008, p.c.). I have not been able to ascertain whether Niuean patterns with Warlpiri or Tsez/Lezgian, but it is reported not to be syntactically ergative (Levin and Massam 1985).
82 Michelle Sheehan sometimes); whether it assigns a structural Case; and whether it bears an EPP feature. The upper end of the hierarchy remains fairly descriptive, and I have no deep explanation for the fact that P1 refers to transitive rather than intransitive v, though this is clearly empirically justified, given that all non-accusative languages seem to share this property. It is possible that transitive v is the most salient instantiation of the category v and as such has a privileged status for acquisition. The format of P3–P4 is intended to be open rather than fixed by UG, the idea being that the child acquiring a non-accusative system can extend or restrict ergative case to any coherent class of vs, with Hindi and Tsez/Lezgian being just two possibilities (see Roberts 2012). While this is a fairly powerful model, it appears to be empirically necessary. It is an empirical question, though, to what extent all potential extensions/restrictions of ergative case are attested. I have found no language, for example, in which ergative is limited to agents, to the exclusion of causers, though there is at least one language which may limit ergative to animate DPs (Nepali).32 The model can therefore be seen as a working hypothesis. P5–P6 are substantively different from P2–P4. First of all, the dependencies between P2–P4 are negative, so that they do not determine cumulative properties of a system but rather mutually exclusive properties. A language either generalizes, extends, or restricts ergative case, but it cannot, by hypothesis, do more than one of these things. P5–P6, however are different. Syntactically ergative high ABS languages are a subset of syntactically ergative languages. It is only positive dependencies of this kind which are truly dependent and hierarchical in this sense. P2–P4 are non-cumulative and so could be reordered without altering potential outputs, but the same is not true of P5–P6. An anonymous reviewer points out that it is possible to rephrase P2–P4 so that they too are cumulative in this sense, giving the following alternative parameter hierarchy: (73)
Parameter hierarchy (alternative fifth version) P1: Does transitive v assign a theta-related case (ERG) in a language L? N – accusative
Y – non-accusative P4’: is this generalised to all transitive vs in L?
N (some) – Tsez, Lezgian N – P5’: does verg bear an EPP in L?
Y – P3’: is this extended to intransitive vs in L? Y – P2’: is this generalised to all intransitive vs?
N (some) – Hindi N – Warlpiri, Niuean(?) Y – P6’: are verg’s -features suppressed? N – West Greenlandic, Tagalog, Cavineña(?) Yidi , Kuikuro, Tongan
Y – Basque
Y – Chamorro, Mam, Q’anjobal, Dyirbal, Seediq, Trumai
In (73), there is only one negative dependency: that between extension of ergative to intransitive vs and the association of an EPP feature. Again, there is a principled 32
Thanks to Joe Perry for providing me with and discussing the Nepali data.
Parameterizing ergativity 83 reason why these two grammatical properties would be incompatible: a language which extended ERG to intransitive contexts would have no argument to satisfy an EPP feature in such contexts. With this exception, though, the parameters in (73) all involve positive dependencies. This has the advantage that, moving down the hierarchy, the output grammars stand in superset relations. The contexts in which ergative surfaces in Basque are a proper superset of the contexts where ergative surfaces in Hindi, which are a proper superset of the contexts where ergative surfaces in Tsez. Similarly, little v in Chamorro has all the properties of little v in West Greenlandic plus suppressed ɸ-features (the marked option) and little v in West Greenlandic has all the properties of little v in Warlpiri plus an additional EPP feature. In this way, assuming that the hierarchy models acquisition, this is a process of selecting grammars of ever increasing complexity and size, providing a potential solution to the subset problem identified by Wexler & Manzini (1987) and Manzini & Wexler (1987). Essentially, as they point out, given the negligible role played by negative evidence in language acquisition, children face a superset trap, whereby if they posit a grammar consistent with the data they observe, but not restrictive enough, they may never be able to posit a grammar which is a subset of that initial hypothesis. The kind of parameter hierarchy in (73) addresses this problem head on by proposing that children start off by positing smaller grammars and only extend them in the face of positive evidence. Reordering the hierarchy in this way thus seems attractive, although it appears to involve some redundancy concerning the actual parameters required. A remaining question concerns the relative ordering between parameters. Is there any deeper rationale for the positive dependencies between parameters in (73)? The answer appears to be that these dependencies are due to the need for convergence. We have already provided a potential explanation for the fact that syntactic ergativity is compatible only with transitive-sensitive ergative alignment. A similar account emerges for the dependency between P1, P5/P5’ and P6/P6’. In a language in which v fails to assign ergative and/or lacks an EPP feature, there will be no way for both arguments to receive Case if v loses its ability to assign a structural Case (i.e. loses its ɸ-features). In order for both DPs to get case in such as context: (i) the higher DP must get a non-structural case; and (ii) the lower DP must scramble past the higher DP so that it is in a position to receive a higher structural case (from T) without defective intervention. It is only in such contexts, then, that Burzio’s generalization can be violated. All of this is implicit in the parameter hierarchies in (72)/(73).
3.4 Conclusions This chapter has developed a parameter hierarchy for non-accusative alignment based on the hypothesis that ergative is an inherent Case. It has been shown that while there are minimal differences in the distribution of ergative case across languages, there are also many similarities all of which seem to be broadly in line with the predictions of
84 Michelle Sheehan the inherent case account, based on what is known about theta-roles in accusative languages. Of course, as mentioned in the introduction, the inherent case approach makes four different kinds of predictions and this chapter has focused mainly on the first of these (A): A. Ergative will only occur on (a subset of) arguments externally merged in spec vP. B. The presence of ergative may be independent of transitivity, so we might find ergative subjects without absolutive objects. C. There will be no derived/non-thematic ergative subjects (no ergative expletives, raising to ergative or ergative subjects of passives, ditransitives, or otherwise). D. Ergative case will not be lost in contexts where structural case is not available (no change of case under ECM, no loss of ergative under raising). In relation to B, it has been shown that Western Basque and Hindi both have ergative with unergative intransitives, but the possibility of ergative with oblique objects seems to be rarer (though it is observed in Warlpiri and Lezgian). The discussion of the source of absolutive case in Trumai and Chamorro touched on D, though, there is clearly much more to be said. Finally, C has not been discussed at all. In a sense, then, showing that ergative case occurs only on a (a subset of) arguments externally merged in spec vP is just the first step towards arguing that ergative is an inherent case. While the facts have been shown to be broadly compatible with the inherent case approach, the results are by no means conclusive (see especially Rezac et al. 2014 on Basque). Given the parameter hierarchy approach, though there are added advantages to the inherent case approach, not least because it enables us to conceive of accusative/ergative variation in terms of variation of the properties of a single class of functional heads (little v), broadly in line with the Borer–Chomsky conjecture. Parameter hierarchies, in these terms, are the pathways used by the child to acquire the properties of a class of functional heads, aided by the kinds of dependencies and acquisition strategies discussed. The basic case/alignment properties of a language are thus effectively encoded on little v in systematically defined ways.
Acknowledgments Many thanks to the European Research Council for funding this research under the auspices of the project Rethinking Comparative Syntax (ReCoS). Thanks also to the other members of that project: András Bárány, Tim Bazalgette, Theresa Biberauer, Alison Biggs, Georg Höhn, Anders Holmberg, Ian Roberts, and Jenneke van der Wal for providing critical feedback at various stages. Different parts of this work were presented at West Coast conference on Formal Linguistics (Arizona), the Syntax of the World’s Languages (Dubrovnik), GLOW Biolinguistics Workshop (Lund), Formal Ways of Analyzing Variation (Reykjavik), Towards a Theory of Syntactic Variation (Bilbao), International Congress of Linguists (Geneva), Societas Linguistica Europeae (Split), Workshop on Building Blocks (Leipzig), and at What Happened to Principles and Parameters? (Arezzo). Thanks to the audiences at those venues
Parameterizing ergativity 85 as well as at the University of Cambridge for providing thought-provoking questions and critiques, especially Joe Perry, Katya Pertsova, Gereon Müller, Edith Aldridge, Amy Rose Deal, Adam Ledgeway, and Bob Freidin. Finally, special thanks to an anonymous reviewer and to Lisa Travis for providing detailed comments on an earlier draft of this chapter which have (I hope) led to a great deal of improvements. As ever, I take full responsibility for the way I have interpreted and used comments and suggestions.
Abbreviations ABS, absolutive; AOR, aorist; APPL, applicative; AUX, auxiliary; CAUS, causative; COMP, complementizer; DAT, dative case; DEF, definite; DEM, demonstrative; DET, determiner; DOM, differential object marker; EPP, movement trigger; ERG, ergative; EV, evidential; FOC, focus marker; GEN, genitive; GER, gerund; I noun, class I agreement; II noun, class II agreement; III noun, class III agreement; impf, imperfect; inf, infinitive; INSTR, instrumental; INTR, intransitive; iv noun, class iv agreement; LAT, lative case; LK, linker; LOC, locative; NZR, nominalizer; OBL, oblique case; OS, oblique stem; PERF, perfect; PL, plural; POSS.ESS possessive case; PRES, present; PROG, progressive; PST, past; PTCP, participle; REL, relative marker; RES, resultative; S, singular; TENSE, tense marker; TR, transitive; UM, ‘um’ morpheme; UNM, unmarked; WIT, witnessed; YI, ‘yi’ morpheme.
Chapter 4
Ac cusative and e rg at i v e in Hindi Anoop Mahajan
4.1 Introduction This chapter deals with a specific issue of case licensing in ergative languages. The issue concerns the licensing of the case on the direct object (absolutive) argument in ergative constructions. This issue is of importance, since over the last few years the case of the direct object (DO) argument in ergative constructions has been argued to be licensed by a variety of conditions that include: case licensing by T; case licensing by an accusative licensing head, which could be either a little v or a lower agreement (AGR) head; or by some mechanism of case competition.1 The discussion is often complicated by the well-known fact that the overt case realization of absolutive is often morphologically null. Furthermore, in a language like Hindi, the DO can have a differential object marking that is often equated with the overt manifestation of accusative case (as in Mohanan 1994a, for example) in opposition to the null case on a non-differentially case marked object which is then sometimes labeled as nominative (Mohanan 1994a) or as accusative (Legate 2008). My objective in this chapter is to argue, employing data that has not been previously invoked within this domain, that the Hindi DO in ergative constructions does not have accusative case. This casts doubt over the universal validity of absolutive-as-accusative type theories, and also on Hindi specific proposals where it has been argued that the DOs in Hindi ergative constructions bear accusative case. Furthermore, it will once again highlight the question concerning why, at least in certain environments, the absence of accusative assignment is a prerequisite for the licensing of ergative subjects and perhaps other oblique subjects as well. The discussion in this chapter will
1
I use the terms ‘case’ and ‘inherent case’ in this chapter to distinguish between ‘structural case’ and ‘inherent case.’
Accusative and ergative in Hindi 87 also emphasize that focusing on case licensing as structural argument licensing in the original Government and Binding sense is a more fruitful strategy to resolve disputes over what case the arguments in ergative constructions bear, rather than looking at the morphology on the arguments—which that can often be confusing.2 To set the background for this discussion, consider (1) in Hindi:3 (1) Kabiir-ne vah laal gaaṛii jaldii-se beč-ii thii Kabir(masc)-erg that red car(fem) quick-with sell-perf.fem.sg be.pst.fem.sg ‘Kabir had sold that red car quickly.’ The central problem to which we seek an answer concerns the case borne by the DO in (1). The case of the subject in (1) is identifiable as the ergative case by its postpositional case ending –ne. However, the DO has no overt morphological case ending, therefore it is not a priori clear whether this DP has a case and if it does, what is the nature of that case. I am not aware of any proposal that asserts that the object DP in (1) does not have a case, and indeed if one assumes the validity of the traditional GB case filter, the DO in (1) must surely have a case in order to be visible.4 The usual practice in traditional literature on ergativity (for example in Comrie 1978 and in Dixon 1994) is to label the DO as having an absolutive case. However, the label ‘absolutive’ does not tell us how this absolutive case is assigned or licensed, and therefore the use of this label often obscures the formulation of the structural licensing conditions under which the object nominal is (case) licensed. My aim here is to suggest that the DO in (1) is case licensed by T, and within the GB-minimalist type theories, the case that the DP bears in (1) should be labeled as the nominative case. The alternative that has been proposed (for Hindi) is that the DO in (1) is case licensed by a little v and should be labeled as accusative (Bhatt 2005; Legate 2008; also related general proposals in Murasugi 1991, Bobaljik 1993b, and others). I will argue that this alternative view is not correct. It should however be made clear that the case labels themselves are not so important. The GB-minimalist theories are clear on this. The real issue concerns the nature of case licensing of the object, and 2
In this respect (realization of the morphological case), I am in general agreement with distributive morphology based approaches such as that of Legate (2008). 3 The glossing I provide for the examples includes the features that are relevant to the discussion as well as for clarity. Not all morphological features are always identified in the glosses. For example, I gloss gender agreement for participle verbs and do not usually gloss the number agreement, even though it can be morphologically realized in many cases. 4 However, the possibility that the DO in (1) may be caseless should perhaps be evaluated in view of proposals concerning the case of DOs in antipassive constructions, where it has been suggested by some (including Aldridge 2012b and Mahajan 2012) that DO nominals may be caseless and perhaps be a special case of incorporation (or pseudo-incorporation). I have tried to minimize that possibility in (1) by making the DO non-adjacent to the verb and by including a demonstrative and a modifier. Furthermore, the DO in (1) is clearly referential and it can be shown that its scope properties are different from that of the DOs in antipassive constructions (for an outline of the relevant properties of DOs in antipassives, see Polinsky 2005; for relevant issues concerning Hindi noun incorporation, see Mohanan 1995 and Dayal 2011).
88 Anoop Mahajan indirectly of the subject, particularly the identification of the relevant case licensing heads and the environments of case licensing.
4.2 A Selective Overview of Proposals about DO Case Licensing in Ergative Constructions 4.2.1 DO Has ACC Case in Ergative Constructions Within the GB tradition, this is perhaps the earliest proposal, represented by Massam (1985) and Levin and Massam (1985), and followed up in Bobaljik (1993b).5 Taking Bobaljik (1993b) to be representative of this tradition, the general idea is that the sources of structural case for subjects and objects in ergative–absolutive (ERG–ABS) languages and nominative–accusative (NOM–ACC) languages are parallel. Specifically, ABS=ACC and ERG=NOM (Bobaljik 1993b: 46). Even more specifically, Bobaljik proposes that NOM and ERG are both assigned by a structurally higher head, while ACC and ABS are assigned by a structurally lower head. He labels those heads as AGR1 (the higher head that takes TP as its complement) and AGR2 (the lower head that takes VP as its complement). The crucial way in which ERG–ABS languages differ from NOM– ACC languages in his system concerns the case that the sole argument of an intransitive clause is assigned. Bobaljik makes his proposal in the form of a parameter he calls the ‘Obligatory Case Parameter’, which essentially says that in NOM–ACC languages, this sole argument has NOM (=ERG) case, while in ERG–ABS languages, this argument has ACC (=ABS) case. Given the existence of split ergative languages, it is clear that this type of parameter cannot be a parameter distinguishing languages, though one can develop an implementation of this approach to include split ergativity (which I will not attempt here). I should note that I will follow Bobaljik and others in an important way in that I will assume that NOM is the label of the case licensed by a (higher) T head while ACC is the label of the case that is licensed by a (lower) v head under well-defined locality conditions. An implementation of this would be to use the locality involved in the AGREE relation of the minimalist tradition and I will assume that for this chapter. It would perhaps be wise to develop a different terminology and relabel NOM as case1 and ACC as case2 or something similar so that we can get away from the confusion caused by the use of the traditional terminology in this domain. However, no one, including me at this point, is making this move, and therefore I want to make sure that we are talking about licensing relations as opposed to morphological forms, which may sometimes yield useful clues, but can also add unnecessary confusion. 5
For Hindi, Bhatt (2005: 759–760) also makes the assumption that the absolutive DO has ACC. He also assumes that Hindi differentially case marked objects have ACC.
Accusative and ergative in Hindi 89
4.2.2 DO Has Nominative Case in Ergative Constructions This view is advocated and developed in Bittner (1987), Mahajan (1990), Bok-Bennema (1991), and Murasugi (1992) among others. The general idea in most of these approaches is that ergative licensing is handled by some head other than T (though see Bittner and Hale 1996a for a somewhat different perspective), leaving T to license NOM on the DO in ergative languages. Whether ergative is licensed as an inherent case or a structural case is a separate issue and I will not be directly concerned with that here (see among others, Marantz 1991, Bittner and Hale 1996a, Woolford 1997). My focus in this chapter is more on the nature of DO licensing than that of subject (SUB) licensing.
4.2.3 DO Has Sometimes NOM and Sometimes ACC in Ergative Constructions This view has been developed in Legate (2008) who argues that there are two typologically distinct groups of languages. In one group, the absolutive argument is NOM, while in the second group, the absolutive argument is ACC. Legate makes a specific proposal about Hindi, arguing that Hindi is a type of language in which absolutive objects have ACC. This approach agrees with the assumption made concerning the case of DO in Hindi ergative constructions in Bhatt (2005: 759–760). I will argue in this chapter that Hindi DOs in ergative constructions do not have ACC. I will use evidence from a variety of prenominal relative clauses that are based on the same verbal form as the Hindi ergative constructions to substantiate my argument. If my argument is on the right track, it will cast doubt over Legate’s proposal about Hindi, though I do not evaluate the Hindi external data that Legate provides for her general proposal.6
4.2.4 The Source of Ergative Case Once again, there are differing views on the licensing of the ergative case. Marantz (1991) and Bittner and Hale (1996a) propose in different ways that the ergative case is structural. Bobaljik’s (1993b) proposal tying ERG to the higher AGR head also makes the ergative case look like a structural case, though in ways different from Bittner and Hale, and Marantz. However, much of the recent work on ergativity has developed the idea that the ergative case is inherently assigned by little v. This view is represented in Mahajan (1990, 2000, 2012), Woolford (1997), Anand and Nevins (2006), and in numerous other recent papers. In this chapter, I will assume that ergative in Hindi is assigned inherently by the little v that heads the complement of a perfective Asp head 6
Another type of variation is discussed in Massam (1996) who argues that in Niuean the absolutive case patterns unlike both NOM and ACC.
90 Anoop Mahajan (more details in section 4.3). While this aspect of ergativity is not the focus of this chapter, there appears to be a connection between the unavailability of accusative case and the appearance of ergative case in Hindi. This raises the issue of whether the ergative case is some deviant form of the accusative case (Marantz 1991; Mahajan 2000). The unavailability of ACC in Hindi extends beyond the ergative subject construction to include dative subject constructions and a variety of Hindi passives (see Mahajan 2000 for relevant details). Therefore, the proper correlation in Hindi is between the unavailability of ACC and the appearance of an oblique case on the subject. I do not pursue this matter in this chapter.
4.3 Some Basic Hindi Ergativity Facts and the Basic Issues 4.3.1 Null Case Objects in Hindi Ergative Constructions As is well known, Hindi is a split ergative language. Ergative case appears on the subject of transitive perfective participle verbs as in (2) (on the connection between ergative and perfective in Hindi, see, among others, Porizka 1967, 1968, 1969; Kachru and Pandharipande 1978; Mohanan 1994a; Davison 2004b). The subjects of non-perfective participle transitive clauses must not have an ergative ending; and indeed in (3), a transitive imperfective construction, and in (4), a transitive future construction, the subjects are unmarked and cannot have an ergative ending.7 (2) mε˜-ne vah akhbaarẽ jəldii-se beç-ĩĩ thĩĩ I-erg(masc) those newspapers (fem) quickly sell-perf.fem.pl be.pres.fem.pl ‘I sold those newspapers quickly.’ (3) mε˜ (*-ne) vah akhbaarẽ jəldii-se I(masc) those newspapers (fem) quickly ‘I (habitually) read those newspapers quickly.’
paṛh-taa hũ read-imperf.masc.sg be.pres.IP.sg
(4) mε˜ (*-ne) vah akhbaarẽ jəldii-se paṛhũ-gaa I(masc) those newspapers (fem) quickly read.IP-fut.masc.sg ‘I will read those newspapers quickly.’ A lot of the later discussion in this chapter will focus on the verb forms in (2) and (3). At this point, what we need to note about (2)–(4) is that: (i) the morphological shape 7
Examples (3) and (4) are ungrammatical with a subject –ne ending even if the agreement is changed to object agreement.
Accusative and ergative in Hindi 91 of the DO in all of these sentences is identical; (ii) the subject in (2) has an ergative case postposition,8 while the subjects in (3) and (4) must be unmarked, and are therefore labeled as nominative by most Hindi linguists; and (iii) the perfective participle and the copula in (2) agree with the object, whereas we get subject agreement in (3) and (4).9 The Hindi verbal cluster (verb and the copula/auxiliary) always agrees with the same argument, though not exactly in the same features. Given that the DO in (2)–(4) looks exactly the same, there are various logical possibilities. These include: (i) the DOs in (2)–(4) are all NOM, since they are morphologically bare and are morphologically identical, and share the same bare case form as that of the subject in (3) and (4). The fact that the DO agrees with the verb in (2) but not in (3) and (4) is not relevant. This is the view held by Mohanan (1994a). She treats (3) and (4) as double nominative constructions.10 (ii) the DOs in (2)–(4) are all ACC, since they are all objects and licensed in the same way, arguably by the same case licensing head. Once again, the fact that the DO agrees with the verb in (2) but not in (3) and (4) is not relevant. This is the view held by Bhatt (2005) and Legate (2008). (iii) the DO in (2) is NOM and is case licensed by T, while the DOs in (3) and (4) are ACC and are case licensed by a distinct head, presumably a little v. Agreement relations in (2) vs. (3) and (4) mirror case licensing. This view is proposed by Mahajan (1990) and is consistent with Chomsky’s (1993) idea about the relationship between case and agreement feature checking. (iv) the DO in (2) is ACC while it is NOM in (3) and (4). This possibility has not been explicitly suggested by anyone and I will not discuss it further. A variant of this idea is followed by Mohanan (1994a) in the context of differentially case marked (DOM) objects, which I will discuss later. If the evidence I present and discuss in this chapter is on the right track, the possibilities and proposals in (i) and (ii) are wrong. That is, the proposals that treat all unmarked objects alike, whether as ACC or as NOM, are inadequate in dealing with the data that I will discuss. I argue that the DO in ergative constructions in Hindi, whether unmarked or marked, is never licensed by an ACC licensing head. This essentially leaves us with the proposal in (iii) that the DOs in (2) vs. (3) and (4) are licensed by different heads and should therefore be labeled differently: NOM in (2) vs. ACC in (3) and (4). A sketch of the proposal concerning case licensing that I defend is outlined in the structural configurations that represent perfective (Figure 4.1, corresponding to
8
Hindi case postpositions are clitics (see Mahajan 1990 and Mohanan 1994a for some discussion). Object agreement in ergative constructions in Hindi does not include person features. The subject agreement in non-ergative constructions (3) and (4) shows person agreement, though in (3) the person agreement shows up only on the auxiliary. In the present discussion, this fact will not be relevant. 10 Mohanan’s (1994a) discussion is within the LFG framework where the case linking conditions are quite distinct from the case licensing conditions of the GB-minimalist tradition. Therefore, a direct comparison of her proposal and its possible counterpart in the GB-minimalist tradition is difficult. However, Mohanan’s work does represent the tradition of taking morphological shape of the case endings (including null endings) seriously for linking/licensing purposes. 9
92 Anoop Mahajan sentence (2)) and imperfective (Figure 4.2, corresponding to sentence (3)) transitive clauses with a finite tense. For the sake of clarity, these configurations are depicted with pre-movement structures that only show case relations and not the final word order of Hindi.11
selection … TFIN
[AspP
Aspperf
[vP SUB-erg
Vperf
[
V
DO ]]]
[
V
DO ]]]
inherent ergative NOM
Figure 4.1 Case relations in Hindi perfective transitive clauses
selection … TFIN [AspP
Aspimperf
[vP
SUB
Vimperf
NOM
ACC
Figure 4.2 Case relations in Hindi imperfective transitive clauses
4.3.2 DOM in Hindi Hindi displays differential object marking (DOM) that interacts with the characterization of object case. Examples (5) and (6) provide examples of DOM objects in Hindi ergative constructions, while (9) and (10) illustrate that DOM objects can be found in non-ergative constructions as well. (5) Miiraa-ne laṛkii-ko kal dekh-aa thaa Meera-erg(fem) girl(fem)-DOM yesterday see-perf.masc be.pst.masc ‘Meera had seen the girl yesterday.’
11
In this chapter, I omit discussion of why the ergative subject does not block case assignment by T to the DO in Figure 4.1. Various ways of handling this ‘non-intervention’ effect have been proposed in the literature. For the purpose of this chapter, I will assume that ergative assignment makes the subject ‘inert’ and not visible to structural case assignment by T. See Mahajan (1990) and Bhatt (2005) for two of the possible analyses.
Accusative and ergative in Hindi 93 (6) Miiraa-ne kitaab-ko kal paṛh-aa thaa Meera-erg(fem) book (fem)-DOM yesterday read-perf.masc be.pst.masc ‘Meera read the book yesterday.’ (7) mε˜ (*-ne) laṛkii-ko har roz bulaa-taa hũ I(masc) girl-DOM every day call-imperf.masc.sg be.pres.IP.sg ‘I call the girl every day.’ (8) mε˜ (*-ne) kitaab-ko jəldii-se paṛhũ-gaa I(masc) book-DOM (fem) quickly read.IP-fut.masc.sg ‘I will read the book quickly.’ Broadly speaking, Hindi proper names and pronouns must always be followed by DOM and specific animate and inanimate objects are also followed by DOM. There is extensive literature on Hindi DOM that deals with various interpretational issues of DOM objects. (For formal proposals see, among others, Mahajan 1990, Butt 1993b, Mohanan 1994a, Bhatt and Anagnostopoulou 1996).12 The DOM marking in Hindi is the same as the case ending obligatorily carried by indirect objects. With a few exceptions, the Hindi DO -ko ending is identified as the accusative ending (as in Mohanan 1994a and Butt 1993b, 1995). Given this common and pervasive view, the -ko ending in (5) and (6) is taken to represent ACC. In view of the new data discussed in this chapter, the proposal is that the -ko ending in (5) and (6) is not morphological realization of ACC but simply DOM marking. This view was originally put forward for Hindi in Mahajan (1990). The licensing of the DOM morphology is not an issue discussed here, though see Bhatt and Anagnostopoulou (1996), and a more recent idea in Kalin (2014). My contention is that the case licensing of DOs in (5) and (6), and also (7) and (8) is obscured by the surface appearance of the DOM marker. In particular, I will argue, contrary to much work in Hindi linguistics, that the DO in (5) and (6) is case licensed by the T head while the DO in (7), and also in (8) (though I do not directly argue for that), is licensed by a little v. I leave open the issue of what governs the appearance of the DOM marking. The point I want to make is that the DOM marking is not a substitute for the structural case licensing requirement. This in turn raises yet another interesting issue of whether non-structural case marking can ever substitute for structural case licensing in terms of a condition of the sort envisaged by Vergnaud’s (2008) original proposal concerning case. My tentative answer to this would be that DOM does not substitute for structural case, or more specifically, DOM nominals must be structurally case licensed in the sense of the classical structural case requirement.13
12 Mohanan (1994a) disallows DOM endings on inanimate DOs. Other works such as Mahajan (1990) and Bhatt and Anagnostopoulou (1996) accept DOM inanimates. 13 This raises the issue of whether the inherent ergative case (for those who view it as inherent case) is sufficient to fulfill the case theory requirements. I will not discuss this here, though see Legate (2002) who
94 Anoop Mahajan
4.4 A New Empirical Domain in Resolving the Issue of Case in Hindi Perfective Environments Given that Hindi DOs can have null case ending or be DOM marked in various environments, it is hard to find crucial evidence that will help us decide the exact nature of case licensing in normal transitive clauses. My strategy here is to turn to a new empirical clausal domain where both perfective and imperfective verb forms that we see in Hindi ergative and non-ergative constructions are employed and isolate the DO licensing outcomes in those environments. The clausal domains are similar to ergative constructions with respect to the verbal forms (the lower vP clausal domain) but crucially different in that they filter out the possibility of case licensing by T since finite T is systematically unavailable in these environments. This in turn provides us with a useful control in looking at how the DOs may be licensed in structures that contain sub-parts of the ergative construction and provide a new insight into object case licensing. The empirical domain that we now turn to is that of non-finite prenominal relative clauses.
4.5 Perfective Participle Prenominal Relative Clauses 4.5.1 The Case of DOs In this subsection, I will present the core argument of this chapter suggesting that the perfective participles in Hindi are simply incapable of case licensing DOs in Hindi. Since the ergative construction by itself does not provide conclusive evidence, I turn to a somewhat similar construction that helps us resolve the relevant issue. Hindi prenominal relative clauses come in various varieties—the one we start with here is the form that is of most interest to us—the relative clauses that use a perfect participle form of the verb which is identical to the one found in Hindi ergative constructions. Consider the data below. The brackets indicate the limits of the prenominal relative clause; the head of the relative clause is the NP that appears after the right bracket.
presents some Warlpiri data with ergative subjects in infinitive constructions that may have consequences for this issue. Hindi ergative case is however only found in tensed environments. I discuss in note 18 one environment where the assignment of the ergative case could be attributed to a non-finite context, though the ergative nominal in that case can be argued to have raised to a finite clause.
Accusative and ergative in Hindi 95 (9) [Kabiir-kii likh-ii (huii)] kitaab Kabir-gen write-perf.fem be.perf.fem book(fem) ‘a/the book written by Kabir’ (10) [bazaar se aa-yii (huii)] taazii sabzii market-from come-perf.fem (be.perf.fem) fresh vegetable(fem) ‘fresh vegetables (which) arrived from the market’ (11)
[mar-ii (huii)] čhipkali die-perf.fem (be.perf.fem) lizard(fem) ‘a/the lizard that is dead’
(12)
[mez-par soy-ii (huii)] billii table-on sleep-perf.fem (be.perf.fem) cat(fem) ‘a/the cat sleeping on the table’
Examples (10)–(12) are cases of subject relativization.14 Example (9) is an instance of an object relativization. In (9), the subject of the relative clause is followed by the genitive postposition. The perfective participle verb and participial aux be (which is optional) agree with the relativized head noun (and the genitive postposition). The subject can also be marked as a by-phrase instead of with genitive (but can never be unmarked). (13)
[Kabiir-dwaara likh-ii (huii)] kitaab Kabir-by write-perf.fem (be.perf.fem) book(fem) ‘a/the book written by Kabir’
Example (9) is parallel to a normal perfective transitive ergative clause (14). Similarly, (10) is parallel to a normal intransitive perfective non-ergative (15).
14 There are interesting restrictions on intransitive subject relatives. Some unergative subjects that can optionally take ergative subjects cannot be relativized (Mahajan 1990), as in (i)–(iii) below. Though so-naa ‘to sleep’ can optionally take an ergative subject, it can appear in prenominal perfective relatives. This pattern is not fully understood and needs to be investigated.
(i)
*[tez dauṛ-aa (huaa)] fast run-perf.masc be.perf.masc ‘a/the boy who ran fast’
laṛkaa boy
(ii)
*[zor-se bhõk-aa (huaa)] Loudly bark-perf.masc be.perf.masc ‘a/the dog that barked loudly’
kutta dog
(iii) [zor-se soy-aa (huaa)] aadmii soundly sleep-perf.masc be.perf.masc man ‘a/the man who is sleeping soundly’
96 Anoop Mahajan (14) Kabiir-ne kitaab likh-ii (thii) Kabir-erg book(fem) write-perf.fem (be.pst.fem) ‘Kabir had written the book.’ (15)
bazaar se taazii sabzii aa-yii (thii) market-from fresh vegetables(fem) come-perf.fem (be.pst.fem) ‘The fresh vegetables had arrived from the market.’
The point that I wish to make is that the perfective prenominal relatives in (9)–(12) are structurally parallel to normal non-relative clauses such as (14) and (15) with a crucial difference: they use the same perfective participle verbal forms and have similar agreement patterns; but while (9)–(12) lack a finite T, (14) and (15) have a finite T (though it can be null). Prenominal relative clauses in Hindi cannot be finite. This is illustrated by the ungrammaticality of (16). (16) *[Kabiir-kii likh-ii thii / hɛ̃] kitaab Kabir-gen write-perf.fem be.pst.fem / be.pres book(fem) ‘a/the book that was/is written by Kabir’ Also, while a genitive subject is possible in (9), and a by-subject is possible in (13),15 an ergative subject is systematically impossible in prenominal relative clauses. (17) *[Kabiir-ne likh-ii (huii)] kitaab Kabir-erg write-perf.fem be.perf.fem book(fem) ‘a/the book written by Kabir’ (18) *[Miira-ne Kabiir-ko likh-ii (huii)] kitaab Meera-erg Kabir –dat write-perf.fem be.perf.fem book(fem) ‘a/the book written to Kabir by Meera’ The reason for the ungrammaticality of (17) and (18) is very likely due to the non- finiteness of the relative clauses, though the exact nature of the finiteness requirement (for ergative) in Hindi is not clearly understood.16 One possibility is that Hindi ergative
15
I leave aside the issue of whether the relative clause in (13) is a passive, since it has a subject form that is also found in passives. It should be noted though that (13) itself does not have the usual passive auxiliary which would be based on the verb form jaa-‘go.’ However, the be auxiliary of the examples like (9) and (13) can be replaced by the go auxiliary (though only in transitive prenominal perfective relatives) making them look more like passives even though (9) has a genitive subject, and regular clausal passives in Hindi do not mark the agent with a genitive. 16 Bhatt (2005: 767) includes finiteness as one of the three requirements for ergative licensing (the other two are perfectivity and transitivity).
Accusative and ergative in Hindi 97 subjects are only realized in finite environments and while it can be argued (as I do in Mahajan 2012) that Hindi ergative case is inherently assigned, there is a further connection between the ergative case and the finite T in this language.17, 18 It is also important to note that an unmarked subject is systematically excluded from the prenominal relative clause. (19)
*[Kabiir paṛh-ii (hui)] kitaab Kabir read-perf-.fem be.perf.fem book(fem) ‘a/the book read by Kabir’
(20) *[Miira Kabiir-ko likhii (hui)] kitaab Meera Kabir –dat write-perf.fem be.perf.fem book(fem) ‘a/the book written to Kabir by Meera’ The ungrammaticality of (19) and (20) is perhaps not surprising given the ungrammaticality of (16). Since prenominal relative clauses in Hindi are non-finite, and cannot have a finite auxiliary as shown in (16), the unmarked assumption would be that T case licensing is not available in such clauses, forcing the subjects to be either marked by a genitive postposition or an agentive postposition (or be null).19 17 The assignment and realization of the ergative nominal is a somewhat complex matter. So while Hindi infinitives (as well as prenominal relatives) do not allow ergative subjects, perfective participle if-conditionals do.
(i) agar tum-ne vah kitaab paṛh-ii, If you-erg that book read-perf.fem ‘if you read that book, then …’
to … then
If the generalization that finiteness is a requirement for ergative realization is true for Hindi, then one would have to argue that if-conditionals have a hidden tense (and they do optionally allow an overt finite tense auxiliary). 18 The example below (somewhat marginal for some Hindi speakers) supports the idea that ergative licensing is itself not dependent on finite tense (the right bracket is provided to indicate the clausal boundary between the raising verb and its complement; the left bracket is left out on purpose since its placement depends on discussion beyond the scope of this chapter). (i) Kabiir-ne bahut baṛii galtii kar d-ii (*thii)] Kabir(masc)-erg very big mistake(fem) do give-perf.fem be.pst.fem lag-tii thii seem-imperf.fem be.pst.fem ‘Kabir seemed to have made a big mistake.’ The ergative is surely assigned in the complement clause since that is a transitive perfective clause, and the matrix clause is not. The complement clause must however be non-finite, as shown by the inability to place a finite auxiliary inside it. This may provide evidence that ergative assignment itself is independent of finiteness. However, it is possible that the ergative phrase in (i) has undergone raising and is in the matrix finite clause thus meeting a possible requirement (in Hindi) that ergative subjects need to be in a finite clause. This type of raising is presumably blocked in (17) and (18) since that would involve raising out of a relative clause. 19 This does imply that the genitive and agentive subjects are distinct in their case requirements from the ergative subjects in the sense that they are more like real PPs.
98 Anoop Mahajan The important generalization seems to be that subjects and objects can be relativized in prenominal perfective relatives in Hindi. Furthermore, when an object is relativized, the subject cannot be unmarked—it must either have a genitive case postposition or an agentive postposition. The crucial observation is that the DO, when present because the verb is transitive, must be relativized. It cannot appear inside the perfective prenominal relative clause.20 Examples (21)–(24) involve attempted subject relativization with an overt DO inside the relative clause and they are ungrammatical. (21) *[kitaab paṛh-aa (huaa)] laṛkaa book(fem)write-perf.masc (be-part-masc) boy ‘a/the boy who has read the book’ (*even if agreement on V and aux is fem) (22) *[angrezii akhbaar khariid-aa (huaa)] aadmii English newspaper buy-perf.masc be.perf.masc man ‘a/the man who had bought the English newspaper’ (23) *[Kabiir-ko bahut pɛse diy-aa (huaa)] aadmii Kabir-dat lot money give-perf.masc be.perf.masc man ‘a/the man who gave a lot of money to Kabir’ (24) *[Miiraa-se us-kii kitaab maaŋg-aa (huaa)] aadmii Meera-from she-gen book ask-perf.masc be.perf.masc man ‘a/the man who asked Meera for her book’
20
An exception involving a DO inside the relative clause is given in (i). (i) [šaraab piyaa (huaa)] aadmii Liquor drink-perf.masc be.perf.masc man ‘a/the man who drank liquor’ (= a drunk man)
However, (ii) is ungrammatical. (ii) *[duudh/paanii/dawaaii pi-yaa (huaa)] aadmii milk/water/medicine drink-perf.masc be.perf.masc man ‘a/the man who drank milk/water/medicine’ My hunch is that (i) involves noun incorporation of the DO and (therefore) has an idiomatic meaning. Since one does not usually get drunk drinking milk/water/medicine, (ii) is ungrammatical. Furthermore, attempts to modify the object in (i) yield ungrammaticality. (iii) *[desii/videšii šaraab pi-yaa (huaa)] aadmii local/foreign liquor drink-perf.masc be.perf.masc man ‘a/the man who drank country/foreign liquor’
Accusative and ergative in Hindi 99 Examples (21)–(24) show that in Hindi an unmarked DO cannot be present inside a perfective prenominal relative clause while another nominal is relativized. We have already seen that unmarked subjects are not possible inside prenominal relative clauses in (19) and (20). I had suggested that (19) and (20) are ungrammatical because there is no source of NOM in the non-finite prenominal relative clause. The ungrammaticality of (21)–(24) can now similarly be attributed to the unavailability of any additional structural case inside the prenominal relative clause. In particular, the ungrammaticality of (21)–(24) tells us that perfective prenominal relative clauses are unable to license any non-PP arguments inside them. This can be taken to imply that a structural ACC is not available in these clauses.21 I don’t know of any other reason for the ungrammaticality of (21)–(24). The fact that a DO, if present within a perfective prenominal relative clause, must be externalized (i.e. must be the head that appears outside the relative clause) indicates that the heads of prenominal relative clauses are externally case marked. Thus, relativized DPs ((21)–(24)) must appear in a structural case position in a clause as in (25) (in a subject position) and in (26) (as an object of a preposition). The oblique ending of the plural head in (26) is due to the postposition -se, supporting the idea that these heads are externally case marked. (25) [Kabiir-kii paṛh-ii (huii)] kitaab kho ga-yii Kabir-gen read-perf.fem be.part.fem book(fem) lost go-perf.fem ‘A/the book written by Kabir was lost.’ (26) Mohan-ko [[Kabiir-kii likhii (huii)] Mohan-dat Kabir-gen write-perf.fem be.part.fem ḍar lagtaa hɛ fear feel-imperf.masc.sg be.pres ‘Mohan is afraid from a/the book written by Kabir.’
kitaabõ]-se books.obl(fem)-from
Further support for the proposal that the heads of the prenominal relative clauses are externally case marked comes from the relativization possibilities of phrases like locatives, manner phrases and instrumental phrases (and other oblique phrases) which must be case licensed by their own postpositions. The prediction is that such nominals cannot be relativized using the perfective prenominal relative clause construction. This prediction is borne out as illustrated in (27)–(29). (27) a. Mohan kũẽ-mẽ ḍuub ga-yaa Mohan well-in drown go-perf.masc ‘Mohan drowned in the well.’
21 As pointed out by an anonymous reviewer, English perfective participle relative clauses like in [[a book [written by John]] is on sale] display restrictions similar to Hindi. Part of the analysis being developed here for the Hindi facts may have potential consequences for the syntax of reduced participle relatives in English.
100 Anoop Mahajan b. *Mohan-kaa ḍuub-aa huaa kũãã/kũẽ (-mẽ) Mohan-Gen drown-perf.masc be.perf.masc well in ‘the well in which Mohan drowned’ (28) a. vo kaar-se ga-yaa he car-with go-perf.masc ‘He went with a car.’ (He used a car to go.) b. *us-ki ga-yii huii kaar he- gen go- perf.fem be.perf.fem car ‘the car in which he went’ (29) a. čor-ne laṛke-ko čaaku-se maar-aa thief-erg boy-DOM knife-with kill-perf.masc ‘The thief killed the boy with a knife.’ b. *čor-kaa laṛke-ko maar-aa huaa čaaku thief-gen boy-DOM kill-perf.masc be.perf.masc knife ‘the knife with which the thief killed the boy’ The proposal that the head in a Hindi prenominal relative clause must be externally case marked now helps us make sense of the restriction that in perfective prenominal relative clauses in Hindi, only subjects and DOs can be relativized, since they are the only type of arguments that can be structurally case licensed by clause internal (spinal) heads like T and v. The inability of a DO to survive inside a perfective prenominal relative clause must therefore follow from the lack of structural case inside the relative clause. Given that a finite T is clearly absent in Hindi prenominal relative clauses, the only other plausible source for the structural case would have been an accusative assigning little v. On the basis of the evidence that we have seen so far, I suggest that the little v present inside these clauses is unable to assign ACC, thus making these clauses very similar in ACC case assigning property as transitive perfective ergative clauses.22 To complete this line of argumentation, the big difference between a prenominal relative clause in (30) and a normal finite transitive clause like (31) is the absence/presence of finite T. The fact that the object is licensed in (31) and not in (30) must then be due to
22 The issue of why the subject of a transitive perfective relative clause cannot be ergative (and must be genitive, if present) remains unresolved here. If, as suggested by Bhatt (2005), the presence of a finite T is a further requirement for ergative case licensing (as mentioned in n. 17), then the lack of a finite T within prenominal relative clauses may be tied to the lack of ergative case. However, in view of the data in n. 17, it is not clear if this would be a desirable analytical direction. An alternative that I do not develop in this chapter could be that the little v that assigns case is distinct from the little v that introduces the external argument (as in Mahajan 2012). If one follows that analytical option, then one may be able to argue that the perfective prenominal relative clauses simply lack the case assigning little v head.
Accusative and ergative in Hindi 101 this finite T. Therefore, if finite T licenses NOM, then the object in (31) must have NOM, since we have already eliminated the possibility of the availability of ACC inside perfective clauses. (30) *[kitaab paṛh-aa (huaa)] laṛkaa book(fem) read-perf.masc (be.perf.masc) boy ‘a/the boy who has read the book’ (*even if agreement on V and aux is fm) (31)
laṛke-ne kitaab paṛh-ii boy-erg book(fem) write-perf-masc ‘The boy had read the book.’
thii / (*huii) be.pst.fem / (be.perf.fem)
A schematic sketch comparing the analyses of finite perfective transitive clauses and prenominal perfective relative clauses is presented in Figures 4.3 and 4.4 (Figure 4.3 repeated from Figure 4.1 in section 4.3.1): selection … TFIN
[AspP
Aspperf
[vP SUB-erg
Vperf
[
V
DO ]]]
inherent ergative NOM
Figure 4.3 Case relations in Hindi perfective transitive clauses
selection … TNONFIN [AspP
Aspperf
[vP
NOM NOT AVAILABLE
SUB-gen
Vperf
[
V
DO ]]]
therefore DO must be externalized
Figure 4.4 Case relations in Hindi prenominal perfective relative clauses
A by-product of this discussion is that it helps us make sense of why only certain kinds of grammatical function positions are accessible for relativization in the context of the typology of relativization as discussed in Keenan and Comrie (1977). Only subjects and (non- postpositional) DOs are accessible for relativization in Hindi perfective prenominal relative clauses because: (i) only those two need structural case licensing; (ii) structural case is not available in Hindi perfective prenominal relative clauses; and (iii) the relative clause head in Hindi perfective prenominal relatives is externally case marked. The fact that the indirect object and the obliques (PPs) in Hindi cannot be relativized in perfective prenominal relatives follows since they all receive a case from a postposition inside the relative clause.
102 Anoop Mahajan
4.5.2 PP DOs and DOM Objects Interestingly, at least some oblique DOs can appear inside the prenominal perfective relatives. (32) [dušman-se mil-ii (huii)] enemy-with join-perf.fem be.perf.fem ‘a/the girl who has joined (with) the enemy’
laṛkii girl
(33) [dušman-se ḍar-ii (huii)] senaa enemy-with fear-perf.fem be.perf.fem army ‘a/the army that is afraid of the enemy’ However, a differentially case marked object cannot appear within such relative clauses. This has obvious consequences for analyses of DOM. (34) *[laṛkii-ko dekh-aa (huaa)] girl-DOM see-perf.masc be.perf.masc ‘a/the man who saw the girl’
aadmii people
(35) *[dhyaan-se tasveer-ko dekh-aa (huaa)] aadmii care-with picture-DOM see-perf.masc be.perf.masc man ‘a/the man who carefully saw/examined the picture’ The contrast between (32)–(33) and (34)–(35) is interesting. Examples (32)–(33) tell us that the constraint on not having a DO inside a perfective prenominal relative clause is not about DOs per se but is about whether that DO is a PP or not. A PP DO does not need PP external case licensing, and we would expect it to survive inside a prenominal relative clause, which it does. The ungrammaticality of (34) and (35) appears to be telling us that the differential object marker -ko is treated differently from normal postpositions. In particular, it looks like DOM -ko objects require structural case licensing. Given that no structural case is available inside perfective prenominal relative clauses (as we have argued), the ungrammaticality of (34) and (35) can be attributed to the failure of case licensing of DOs in these examples. The consequence of this is that the DO -ko marking itself cannot be the morphological realization of structural ACC case (as is often assumed in Hindi linguistics).23 Furthermore, it also tells us that -ko marked objects in Hindi need to be structurally case licensed. Given that we have already argued that there are no structural case licensing heads inside perfective prenominal
23 See Bhatt and Anagnastapolou (1996) and Kalin (2014) for some relevant discussion, where it is argued that –ko may be assigned, or enters the derivation, higher than the base position of the DOs. Both of these proposals are compatible with the current proposal.
Accusative and ergative in Hindi 103 relative clauses in Hindi, it now becomes possible to entertain the possibility that the -ko objects in (36) and (37) (finite main clause counterparts of the prenominal relatives in (34) and (35)), are actually case licensed by the finite T, i.e. they have NOM structural case (Mahajan 1990). (36) lõgõ-ne laṛkii-ko dekh-aa people-erg girl-DOM see-perf.masc ‘The people had seen the girl.’ (37)
thaa be.pst.masc
aadmii-ne dhyaan-se tasveer-ko dekha man-erg care-with picture-DOM see-perf.masc ‘The man had seen the picture carefully.’
thaa be.perf.masc
I am obviously not suggesting that (all) -ko marked objects in Hindi always have NOM. In section 4.6, we will see that such objects may have ACC when ACC is available.
4.6 Imperfective Participle Prenominal Relative Clauses 4.6.1 The Case of DOs Hindi prenominal relatives can also be built on an imperfective participle and the properties of this type of relative clause provides more support for the proposal that I have developed so far. Recall that Hindi ergativity is crucially dependent upon perfectivity and that imperfective transitive constructions cannot have ergative subjects. This was shown in section 4.3.1 by the contrast between examples (2) and (3) (repeated below). (2) mε˜-ne vo akhbaarẽ jəldii-se beç-ĩĩ thĩĩ I-erg(masc) those newspapers(fem) quickly sell-perf.fem be.pres.fem ‘I sold those newspapers quickly.’ (3) mε˜ (*-ne) vo akhbaarẽ jəldii-se I(masc) those newspapers(fem) quickly ‘I (habitually) read those newspapers quickly.’
paṛh-taa hũ read-imperf.mas be.pres.IP
I argued in section 4.5, contra Bhatt (2005) and Legate (2008), that the DO in (2) does not have ACC (but has NOM) and that ACC is systematically unavailable in transitive perfective clauses. The issue that I now take up is the case of the identical-looking DO vo akhbaarẽ ‘those newspapers’ in the transitive imperfective clause in (3).
104 Anoop Mahajan For some relevant evidence, I once again turn to prenominal relative clauses. Examples (38)–(40) are prenominal relative clauses that contain imperfective participle main verbs.24 (38) [bhuukh-se mar-taa (huaa) / (*hε˜)] aadmii hunger-from die-imperf.masc be.perf.masc / be.pres people ‘a/the man (who is) dying of hunger’ (39) [saṛkõ-par so-te (hue) / (*hε˜)] log roads-on sleep-imperf.pl be.perf.pl/ be.pres.pl people ‘people (who are) sleeping on the roads’ (40) [tezii-se bhaag-tii (huii) / (*hɛ)] laṛkii quickness-with run-imperf.fem be.perf.fem be.pres.pl girl ‘girl (who is) running fast’ Examples (38)–(40) are all cases of subject relativization. Given: (i) that prenominal imperfective relative clauses must be non-finite, as shown by the ungrammaticality of attempts to insert a finite auxiliary inside them (compare with (3) where a finite auxiliary appears in a normal imperfective participle clause); and (ii) our discussion earlier showing that the relative clause head of prenominal clauses is case marked externally, we expect (38)–(40) to be grammatical, since the non-finite relative clause does not have the capability of licensing an argument with a case assigned by a finite T. When the subject is relativized in (38)–(40), under our assumptions, it can be externally case licensed, and therefore the grammaticality of (38)–(40) is in line with our expectations. The crucial difference between the imperfective prenominal relative clauses and the perfective prenominal relative is that the former allow a lexical DO inside them while the latter do not (as we observed in section 4.5). (41) [kitaab paṛh-tii (huii)] laṛkii book read-imperf.fem (be.perf.fem) girl ‘a/the girl (who is) reading the book’ (42) [sarkaar-se apne haq maaŋg-te (hue)] government-from self ’s rights ask-imperf.pl be.perf.pl ‘people demanding their rights from the government’
24
log people
The morphological shape of the optional auxiliary in (38)–(40) is the same as in prenominal perfective relatives. While I continue to gloss this auxiliary be.perf, the relative clauses in (38)–(40) are not semantically perfective. Interestingly, though, they have a progressive reading and they lack the habitual interpretation that is available with main clause imperfectives in Hindi. The precise nature, function, and representation of the optional auxiliary remains an open question.
Accusative and ergative in Hindi 105 (43)
[Mohan-ko lambe lambe patr likh-tii (hui)] paagal laṛkii Mohan-dat long long letters write-imperf.fem (be.perf.fem) crazy girl ‘the crazy girl writing long letters to Mohan’
Given that a finite T is absent in these clauses, the DO nominal can only be licensed by a different head, and I suggest that this head is the little v of the imperfective vP. This also implies that the case of the DO in main imperfective clauses like (3) is ACC. A sketch of the analyses of finite imperfective transitive clauses and imperfective prenominal relative clauses is provided in Figures 4.5 and 4.6 (Figure 4.5 repeated from Figure 4.2 in section 4.3.1).
selection … TFIN [AspP
Aspimperf
[vP
SUB
[
Vimperf
NOM
V
DO ]]]
ACC
Figure 4.5 Case relations in Hindi imperfective transitive clauses
selection … TNONFIN
[AspP
Aspimperf
[vP SUB
Vimperf
[
V
DO ]]]
ACC NOM NOT AVAILABLE
subject must be externalized
Figure 4.6 Case relations in Hindi prenominal imperfective relative clauses
If my suggestion that prenominal relative clauses are externally case marked is correct, then a clear prediction can now be made. The prediction is that unlike in perfective prenominal relative clauses, the imperfective prenominal relative clauses will disallow DO relativization, since that configuration will lead to case marking the head of the relative clause twice (inside the relative clause and also externally). This prediction is fulfilled. (44) *[laṛkii-kii paṛh-tii (huii)] kitaab girl-gen read-imperf.fem be.perf.fem book ‘the book that the girl is reading’ (45)
*[logõ-kaa sarkaar-se maaŋg-taa (huaa)] people-gen government-from ask-imperf.masc be.perf.masc ‘the right that the people are demanding from the government’
apnaa haq self ’s right
106 Anoop Mahajan (46) *[paagal laṛkii-ke Mohan-ko likh-te (hue)] lambe lambe patr crazy girl-gen Mohan-dat write-imperf.pl (be.perf.pl) long long letters ‘the long letters that the crazy girl is writing to Mohan’ In fact, the prediction is that only subjects can be relativized in imperfective relatives, given that the imperfectives have a little v that licenses DOs and oblique nominals like locatives and instrumentals are internally case marked PPs. These nominals are therefore internally case licensed and cannot move to a clause external case position. As shown below, PPs fail to be relativized using this strategy. (47) *[Mohan-kaa ḍuub-taa (huaa)] kũãã / kũẽ -mẽ Mohan-gen drown-imperf.masc (be.perf.masc) well well-in ‘the well in which Mohan is drowning’ (48) *[Raam-kaa čuhaa maar-taa huaa] čaaku /čaaku-se Ram-gen rat kill-imperf.masc.sg be.perf.masc.sg knife /knife-with ‘a/the knife with which Ram is killing the rat’ If the discussion in this section is on the right track, then we now have an answer to the question posed at the beginning of this subsection where we asked about the case of the DO nominal in (2) and (3) given that there is no morphological case distinction. The answer is that the DO in (3) has ACC, while the DO in (2) does not have ACC, but has NOM as suggested in section 4.5.
4.6.2 DOM Objects Following up on the parallel discussion on morphologically marked objects, we expect DOM objects to be fully well formed inside the imperfective prenominal clauses, since they will now be able to have ACC. This prediction is fulfilled. (49) [laṛke-ko ghuur-tii (huii)] laŗkii boy-DOM stare-imperf.fem be.perf.fem girl ‘a/the girl (who is) staring at the boy’ (50) [dhyaan-se tasveer-ko dekh-taa (huaa)] aadmii care-with picture-DOM see-imperf.masc be.perf.masc man ‘a/the man who was carefully looking at the picture’ If the DOM objects in (49) and (50) have ACC, a further prediction is that they cannot be relativized in these constructions. This prediction also holds.
Accusative and ergative in Hindi 107 (51)
*[laṛkii-ke ghuur-te (hue)] laṛke-ko girl-gen stare-imperf.obl be.perf.obl boy-DOM ‘the boy whom the girl is staring at’
(52)
*[aadmii-kii dhyaan-se dekh-tii (huii)] tasveer-ko man-gen care-with see-imperf.fem be.perf.fem picture-DOM ‘the picture that the man is carefully looking at’
To conclude this section, I hope to have shown that imperfective transitive constructions in Hindi always have an ACC available and that this case is assigned to both morphologically unmarked objects as well as to DOM objects. Thus, imperfective vPs in Hindi are crucially distinct from perfective vPs with respect to their case licensing capability.
4.7 General Discussion and Conclusions The basic question that this chapter addressed concerned the structural case assigned to the DO in Hindi ergative constructions such as (1) repeated below. (1) Kabir-ne vah laal gaaṛii jəldii-se beç-ii thii Kabir-erg(masc) that red car(fem) quick-with sell-perf.fem.sg be.pst.fem.sg ‘Kabir had sold that red car quickly.’ There is no morphological case on the DO in (1), a common situation in ergative languages where the absolutive case is often null. Legate (2008) suggests that it is possible to identify two types of languages with null absolutives. She distinguishes between ABS=DEF(ault) vs ABS=NOM type languages. She argues that Hindi is an ABS=DEF language. In particular, she proposes that the distinction between the two types of languages is located in the case assignment properties of little v. In ABS=DEF, little v assigns accusative case, while in ABS=NOM languages, little v does not assign accusative case (Legate 2008: 58). She further suggests that in ABS=DEF languages, nominative case is not assigned in transitive clauses and the subject receives an inherent ergative case while the DO receives ACC. Thus, with respect to (1), her specific proposal would be that: (i) the DO has ACC; and (ii) NOM is not assigned in (1). In this chapter, I have argued against both of these proposals by bringing in a new set of data that abstracts away from morphology and focuses on argument licensing in terms of abstract case licensing as originally envisaged within the GB framework (see Chomsky 1981 and Vergnaud 2008). I have argued that the DO in (1) does not have ACC, and that ACC is
108 Anoop Mahajan not licensed by perfective little v in Hindi. I have also argued that the case that the DO in (1) has is NOM.25 In trying to uncover the case licensing condition on DOs, we have observed that DOM objects behave like non-DOM objects in their case licensing in Hindi participle prenominal relative clauses. I have suggested that this implies that DOM objects must be structurally case licensed, that they can have a NOM (in perfectives) or ACC (in imperfectives), and that the presence of DOM -ko cannot be taken to reflect the same underlying structural case (usually suggested to be accusative). I have also argued that despite superficial appearances, the DO in Hindi imperfective constructions bears ACC. That is, the imperfective little v is an ACC licenser. Given the pattern of data that we have seen in this chapter, it appears that ERG and ACC in Hindi are in complementary distribution, though a satisfactory theory of this complementarity remains unclear.26
Acknowledgments Parts of this chapter were presented in the ‘Case by Case’ Workshop held at École Normale Supérieure, Paris, in October 2011, and at the Third Formal Approaches to South Asian Languages workshop at USC in March, 2013. My thanks to the participants of these workshops, including Adriana Belletti, Rajesh Bhatt, Jessica Coon, Amy Rose Deal, Hilda Koopman, Luigi Rizzi, Dominique Sportiche, and K. V. Subbārāo for their comments. I also thank an anonymous reviewer for helpful comments.
Abbreviations Abbreviations used in the glosses in this chapter: 1P, first person; ABS, absolutive; ACC, accusative; AGR, agreement node; AUX, auxiliary verb; DAT, dative; DO, direct object; DOM, differential object case marking; ERG, ergative; FEM, feminine gender; FUT, future tense; GEN, genitive; IMPERF, imperfective aspect; MASC, masculine gender; NOM, nominative; OBL, oblique; PERF, perfective aspect; PL, plural; PRES, present tense; PST, past tense; SG, singular; T, tense node.
25 In this chapter, I have not invoked the argument that the presence of object agreement on T is a reflex of NOM. For more discussion on that topic, see Mahajan (1990) and Bhatt (2005). 26 Marantz’s (1991) dependent case assignment account is a possible contender. Another possibility (a variation on Marantz’s proposal) is that a little v can license case once, either an inherent ERG or a structural ACC. It is not obvious how such accounts can be parametrized if ERG–ACC languages do exist.
The Nature of Ergative Case
Chapter 5
On inherent a nd de pendent th e ori e s of ergative c ase Mark C. Baker and Jonathan David Bobaljik
5.1 Introduction The phenomenon of ergative case, in which a specially marked case is found only on the subject of a transitive clause (see (1)) has long posed a special challenge for the ‘Case Theory’ of Government and Binding (GB) and its descendants. (1) a. Maria-nin-ra ochiti noko-ke. Maria-ERG-PRT dog find-PRF ‘Maria found the dog.’ b. Maria-ra Maria-PRT ‘Maria went.’
(Shipibo, Baker 2014a: 342)
ka-ke. go-PRF
This is partly because in this theoretical tradition the structural configurations posited as being relevant for the assignment of structural case are the same as the structural configurations that characterize grammatical functions. For example, nominative case is assigned to the subject of a finite clause, interpreted as either a particular position (Spec,IP) or a configurational relation with a functional head (governed by Tense). The problem, then, is that ergative and absolutive cases simply do not align with the grammatical functions of subject and object. Nevertheless, many theories of ergativity that are broadly consistent with a GB-style Case Theory have been offered. The most prominent such approach in recent years is one that treats ergative case as an inherent case. We call this the IC theory (ICT). This approach maintains a conservative version of Case Theory, but restricts its scope: the subject of a transitive clause in an ergative language
112 Mark C. Baker and Jonathan David Bobaljik receives case in a manner fundamentally different from either the subject or object in a nominative–accusative language. An alternative approach to ergative alignment, initiated by Marantz (1991) and developed extensively in Baker (2015), among others, is the dependent case theory (DCT). The syntactic configuration that matters for the DCT is not the absolute position of an NP or its relationship to a functional head, but rather its position relative to other NPs within a particular syntactic domain: if there is only a single NP in the domain (e.g., an intransitive clause like (1b)), then the NP bears unmarked case (NOM/ABS); if there are two NPs, then at least one of the NPs may bear a dependent case—so called because its appearance on one NP depends on the presence of another NP in the same domain. On this view, NOM–ACC and ERG–ABS/NOM systems are distinguished by a simple parameter: if the lower of two NPs in a domain is marked, then the dependent case is named accusative; if the higher is marked, as in (1a), it is named ergative. Languages may choose to mark one or the other (or both or neither; yielding further alignment types). For concreteness, we may express the DCT view of ergative as follows (see Baker 2015 for refinements; and the chapters by Nash, Coon and Preminger, and Baker, Chapters 8, 10, and 31 in this volume, for analyses making use of a DCT). (2) a. If NP1 c-commands NP2 and both are contained in the same domain (say, clause), then value the case feature of NP1 as ergative. b. Otherwise NP is nominative/absolutive. In this chapter, we put these two approaches to ergative case side by side, comparing some of their predictions and assessing them empirically.1 In doing this, we review some of the arguments that have been given for treating ergative as inherent case. We show that these arguments fail to generalize, and indeed provide evidence that ergative in some languages is not an inherent case; rather, it is best characterized under the dependent case viewpoint (at least for many canonical ergative languages). On the DCT side, (2) will already do as a starting point for comparison. There is, however, more to say about the roots of the ICT, so we begin by reviewing the defining characteristics of inherent case.
5.2 Inherent Case 5.2.1 The Roots of Inherent Case It is well known that in languages with even moderately rich systems of case morphology, there seems to be a rough division between structural cases, which are determined 1 There is, of course, a third major theoretical contender: the view that case is assigned to an NP by a functional head under a relationship of Agree. However, that sort of theory is known to be particularly hard to apply to ergative case (see, for example, Baker 2015: ch. 2), so we do not consider it here.
On inherent and dependent theories of ergative case 113 by the surface–syntactic context that an NP finds itself in, and inherent cases, which are more restricted, linked either to particular semantic functions (theta-roles) or assigned as a quirky lexical property of particular heads. Structural case is typically blind to thematic roles: a surface subject bears nominative, regardless of its exact theta-role, and an object bears accusative. The ECM/Raising-to-Object configuration (I believe him to be my friend) provides a canonical example of structural case: the subject of the non-finite complement of a verb like ‘believe’ may bear no thematic relation to the matrix predicate, only a structural one, yet because of that structural relation, the NP has accusative case. Conversely, passive and unaccusative clauses show that the thematic object (patient) of a verb can bear either nominative or accusative case, depending on the larger structure that contains it. Inherent case, by contrast, is tied to theta-roles and/or to specific predicates. Verbs meaning ‘help’ in many languages assign dative to their object, rather than accusative. As theta-roles are a property of an object’s base position, it is this base position, rather than any subsequent post-movement configuration, that matters. As a result, inherent case NPs typically do not undergo case-alternations. Thus, in a language like Icelandic, the internal argument of ‘help’ is dative not only in a simple active sentence, but also in the corresponding passive, in a passive under an ECM predicate like ‘believe,’ and in a passive under the passive of an ECM predicate (see Andrews 1990). Inherent case comes to this argument from the verb along with its thematic role.2 Since, in the normal instance, a given NP bears a unique theta-role, unchanged throughout the derivation, inherent case cues this ‘base’ position of the NP, and remains constant. Structural cases, by contrast, appear to alternate among closely related derivations, such as active and passive, where the theta role of an NP remains constant but its surface grammatical relation (highest A-position) varies.
5.2.2 Ergative as an Inherent Case Against this background, a prominent line of reasoning within the recent Chomskyan tradition holds that ergative is an instance of inherent case (see L. Nash 1996; Woolford 1997, 2006; Aldridge 2004, 2008a, 2012b; Anand & Nevins 2006; Laka 2006b; Legate 2006, 2008, 2012a; Massam 2006; Coon 2013a; Mahajan 2012; see also the chapters by Sheehan, Laka, Legate, and Woolford, Chapters 3, 6, 7, and 9 in this volume). These authors hold that ergative is assigned by v to the external argument in Spec,vP together with that argument’s theta role, much as dative case is assigned to the internal argument of ‘help’ along with the beneficiary theta-role in Icelandic. The difference between
2 This holds at least in the canonical A-movement environments. Inherent dative case is apparently not preserved in s-passives/unaccusatives, and ‘get’-passives (Zaenen & Maling 1984). It is debated whether these constructions involve simple NP-movement, or a more complex structure with a binding dependency between the nominative NP and the theta-position. See Shimamura (2014a) and Wood (2014) for contrasting views.
114 Mark C. Baker and Jonathan David Bobaljik a NOM–ACC alignment and an ERG–ABS alignment under this view can be schematized as in (3). (3)
ACC
NOM a.
[ [TP
T0
[ [vP
v0
EA
[VP
V
θ
IA
]]]
IA
]]]
θ
?NOM b.
[ [TP
T0
[ [vP
v0
EA θ,ERG
[VP
V θ
In the NOM–ACC alignment (3a), v assigns a theta-role to the external argument, but assigns structural accusative case to the internal argument. The external argument receives its case from (finite) T. By contrast, in the ERG–ABS alignment, the EA receives case along with its theta-role from v ((3b)). The IA is left to get case by other means: it may get nominative case from finite T, either at a distance via Agree or by moving to SpecTP; it may get (covert) accusative case from v (Legate 2008), or it could conceivably be left to get default case, as in the DCT. Proponents of such a view may point to ostensible differences between ergative subjects and nominative subjects. For example, they may claim (i) that ergative case is associated with a particular thematic role, such as agent, and (ii) that ergative does not alternate with other cases in ECM or raising-type constructions (e.g. Woolford 2006).3 Subsumed under point (i) are two ways in which ergative case in certain languages may depart from the canonical ergative pattern of being associated with all and only the subjects of transitive verbs. On the one hand, there are languages, such as Georgian, Basque, and Hindi, in which the subjects of (some) agentive intransitive verbs (unergatives) bear ergative case. Examples of this sort suggest that ergative case is tied to the external (agent, actor) theta- role, not to transitivity as such. On the other hand, in some ergative languages, there are subjects of two-argument verbs that fail to receive ergative case. This latter point can be formulated as the Ergative Case Generalization (ECG, Marantz 1991: 236): (4) Even when ergative case may go on the subject of an intransitive clause, ergative case will not appear on a derived subject. Legate (2012a: 183) emphasizes this generalization, noting that: “If the Ergative Case Generalization holds, it constitutes a powerful argument for the inherent analysis of ergative case—since this analysis predicts the generalization to hold. Ergative is assigned by v to thematic subjects, and thus should not appear on derived subjects.” If derived 3
Ergative and nominative subjects may also differ with respect to controlling agreement on the verb, or in their scope properties (Anand & Nevins 2006); these differences are not systematically correlated with ergative versus nominative alignment however. On the interaction of case and agreement, see Baker (2008, 2015) and Bobaljik (2008).
On inherent and dependent theories of ergative case 115 subjects move from a theta-position to a non-thematic position, then they will be ineligible to receive a theta-related case in their surface position. We can now put the ICT and the DCT side by side. According to the ICT, the main factor in ergative case assignment should be what head a given NP gets its thematic role from. It should not matter (much) how many other NPs are in the clause. As a result, one would expect ergative case assignment to be relatively stable, little affected by properties of the larger structure, just like dative case on the complement of ‘help’ in Icelandic. In contrast, according to the DCT, the main factor in ergative case assignment should be how many NPs are in the same local domain. It should not matter (much) what the thematic roles of those NPs are. As a corollary to this, we might expect ergative case assignment to be relatively fluid, with the same NP getting different cases in different clauses depending on its context. Indeed, there are some constructions in some languages that look promising for the ICT, and have helped get it fixed in the beliefs of many in the field, as mentioned above. However, we want to show that, looking more broadly at other languages, there is much evidence of the sort that supports the DCT and looks problematic for the ICT. Indeed, our assessment is that the bulk of the data for the ergative languages we know about is on the side of the DCT. We look first with some care at cases in which nonagents (non-external arguments) get ergative case despite having the wrong theta-role, as long as there is another NP around, contrary to the ECG. Then we look more briefly at situations in which agents (theta- marked by v) fail to get ergative case even though they have the right theta-role, because there is no other suitable NP in the vicinity. Finally, we take a broader typological view, pointing out that the ICT really predicts an active case pattern, rather than an ergative case pattern, but no paradigm instance of a dependent-marking language with an active alignment pattern is known to typology. This gap is however expected on the DCT.
5.3 Ergative Case on Internal Arguments Our most complete instance of ergative case on internal arguments comes from Shipibo, a uniformly ergative language of the Panoan family, spoken in Eastern Peru; see Valenzuela (2003) (PV) for a thorough description; our synopsis here follows Baker (2014a). In this language, it is particularly clear that derived, nonagentive subjects can receive ergative case because Shipibo happens to have productive, morphologically overt applicative constructions. Legate (2012a: 183) points out explicitly why such applicative constructions are significant for theories of ergative case. She writes that: An additional way around the confound would be a two-argument verb in which both arguments are internal, for example, the passive of a double object verb, or the applicative of an unaccusative verb. If the [ICT] holds, the subject of such verbs would not bear ergative case, despite the presence of two DP arguments.
116 Mark C. Baker and Jonathan David Bobaljik We accept this prediction, and the reasoning behind it, but observe that Shipibo shows it to be false, such that the ICT fails whereas the DCT succeeds for this language. In Shipibo, all dyadic verbs with agentive subjects get ergative case (-n, -nin, -kan) (see (1a)), but no intransitive subjects do in simple clauses. This is shown in (5a) for unergatives and in (5b) for unaccusatives.4 (5) a. Joni-bo-ra teet-ai; Rosa-ra bewa-ke. (see also PV: 336–337) person-PL-PRT work-IMPF Rosa-PRT sing-PRF ‘The people are working.’ ‘Rosa sang.’ b. Kokoti-ra joshin-ke; fruit-PRT ripen-PRF ‘The fruit ripened.’
Maria-ra mawa-ke. Maria-PRT die-PRF ‘Maria died.’
Shipibo also has three applicative affixes (-xon, -anan, -kin; see Valenzuela 2003: ch. 17); we focus on -xon, since it makes all the essential points. This affix can attach productively to transitive verbs and unergative verbs, adding an argument that is interpreted as affected by the event—either a benefactive or a malefactive. This argument is structurally lower than the agent, but higher than the theme. Notice that the unergative subject in (6b) with the applicative bears ergative case, in contrast to (5a). (6) a. Jose-kan-ra Rosa atapa rete-xon-ke. Jose-ERG-PRT Rosa hen kill-APPL-PRF ‘Jose killed a hen for Rosa.’ b. Papashoko-n-ra Rosa bewa-xon-ai. grandfather-ERG-PRT Rosa sing-APPL-IMPF ‘The grandfather is singing for Rosa.’
(also PV: 695–699)
(also PV: 689–690)
This is a high applicative in the sense of Pylkkänen (2008), and it fits well into the standard theory in which an applicative head theta-marks the applied object, takes a VP potentially including the theme as its complement, and itself serves as the complement of the agent- assigning head v. Now crucially, -xon can also attach to unaccusative verbs. Two examples are given in (7). Other attested examples have glosses like ‘spoil on,’ ‘get sick on,’ ‘grow up for,’ ‘turn sour (ferment) for,’ and ‘sink on’ (also PV: 691, 694).
4 For discussion of four possible exceptions to this (‘row,’ ‘pole,’ ‘breathe,’ and an alternative word for ‘die’ (rebes-), mentioned by Valenzeula (2003), see Baker (2014a: 350–352).
On inherent and dependent theories of ergative case 117 (7) a. Nokon shino-n-ra e-a mawa-xon-ke. (*shino-ra) my.GEN monkey-ERG-PRT me-ABS die-APPL-PRF monkey.ABS-PRT ‘My monkey died on me.’ b. Bimi-n-ra Rosa joshin-xon-ke. fruit-ERG-PRT Rosa ripen-APPL-PRF ‘The fruit ripened for Rosa.’
(*bimi-ra) (*fruit-PRT)
Note that the examples in (7) (and all similar examples) have theme arguments bearing ergative case; we do not, for example, get a double absolutive configuration in this construction. This goes against Legate’s (2012a) prediction, derived from the ICT. However, the DCT can account for this, since the examples in (7) have a second NP, not present in (5b). Hence it is not surprising that the theme argument is ergative in (7) but not (5b). More generally, comparing (5a) with (6b) and (5b) with (7) shows that what theta role an NP has (agent or theme), or what head it gets its theta role from (v or V), does not determine whether it is ergative or not, whereas how many NPs are in the clause (one or two) clearly does.5 Thus, we see that the ECG is spurious, although clear violations are seen only when various factors converge (see discussion of (13) below) Shipibo is special in that it has productive morphological applicative constructions, and we have independent evidence whether a verb is unaccusative or unergative (see Baker 2014a: 368–371). Other ergative languages may not present quite as clear a picture, either because we know less about them, or because they lack productive applicatives of unaccusatives. Nevertheless, we do not think that Shipibo is unique in any important sense; rather, there is evidence from other ergative languages that, although more fragmentary, seems to go in the same direction. For example, Kalaallisut (West Greenlandic) is another ergative language that has a morphological applicative (underlyingly -uti-, Fortescue 1984: 89–90). When this affix is added to an intransitive verb, the result is a transitive clause with ergative case on the argument that bears absolutive in the intransitive version. The affix is not fully regular and productive, but among the predicates that -uti-may attach to are non-agentive predicates of emotion: (8) a. kamap-p-uq angry-INDIC-3SG.ABS ‘He/she is angry.’ b. Arna-p angut kama-ap-p-aa woman-ERG man.ABS be.angry-APPL-INDIC-3SG>3SG ‘The woman is angry with the man.’ (Michael Fortescue, p.c.) 5 See Baker (2014a) for further details and refinements, including an explanation for why applicatives of unaccusatives are different from dyadic verbs with experiencer subjects, which do take two absolutive arguments and no ergative argument, in apparent support of the ECG (see also Baker, Chapter 31, in this volume, on analogous predicates in Burushaski).
118 Mark C. Baker and Jonathan David Bobaljik Note that in English, roots like ‘anger’ undergo a kind of transitivity alternation (Chris is angry vs. Pat angered Chris), suggesting that the experiencer can count as an internal argument. The corresponding experiencer is nevertheless marked for ergative case in the applicative version in (8b). Therefore, if predicates like ‘be angry’ are confirmed to be unaccusative in Kalaallisut (as claimed also by Spreng 2012), then (8b) replicates (7) from Shipibo—additional evidence for the DCT over the ICT. Chukchi is an ergative language that does not, apparently, have a productive morphological applicative that combines with unaccusatives, but it does have inchoative and locative alternations that seem relevant, according to Nedjalkov’s (1976) survey of Chukchi’s many diathesis alternations. First, Chukchi has a morphologically unmarked causative–inchoative alternation, which applies to verbs like ‘fill,’ among others: (9)
a. ətləg-e jərʔen-nin əʔtvʔet miml-e father-ERG fill-3SG>3.SG boat.ABS water-INSTR ‘Father filled the boat with water.’ b. əʔtvʔet jərʔet-gʔi miml-e boat.ABS fill-3SG water-INSTR ‘The boat filled with water.’
Comparison with (9a) strongly suggests that both ‘boat’ and ‘water’ are internal arguments of ‘fill’ in (9b), and thus that ‘fill’ with these two arguments counts as an unaccusative verb with a derived subject. Further evidence for this is the fact that either ‘water’ or ‘boat’ can incorporate into the verb ‘fill’ (Nedjalkov 1976: 189, 208; see also Baker 1988). Chukchi also happens to have conative and locative-type alternations, where a given argument can be projected as either a PP (realized as semantic case) or as a bare NP (Nedjalkov 1976: 193, 206, etc.). In particular, the locatum argument that is projected as an instrumental PP in (9a, b) can alternatively be projected as an NP. When this happens with the agentless version of ‘fill,’ it has two NPs as internal arguments—and one of them (the locatum argument) is crucially marked with ergative case: (10) əʔtvʔet jərʔen-nin boat.ABS fill-3SG>3.SG ‘Water filled the boat.’
miml-e water-ERG
(Nedjalkov 1976: 195, 206)
Note that instrumental case and ergative are syncretic on inanimate NPs in Chukchi, but they are distinguished by agreement: the verb agrees with an ergative NP, but not an instrumental one. In particular, ‘fill’ agrees with ‘water’ in (10) but not in (9b), confirming that ‘water’ is ergative in (10), despite it being a theme-type internal argument. (10) is thus another probable counterexample to the ICT (as noted in Bobaljik and Branigan 2006). In contrast, the DCT can explain why a theme subject gets ergative case if and only if there is another internal NP (not a PP) in the clause. ‘Cover’ is another verb that exhibits this diathesis pattern (Nedjalkov 1976: 195).
On inherent and dependent theories of ergative case 119 Yup’ik is yet another ergative language that seems relevant. It has a type of malefactive applicative in which the affix -i-is added to the verb along with an additional NP that expresses someone adversely affected by the event, as seen in (11b). Note that in this example the agent argument of ‘eat’ is ergative and the malefactee is absolutive, showing that the malefactee is a lower, internal argument. (11)
a. arnar neqa-mek woman.ABS fish-ABL ‘The woman ate a fish.’
ner’-uq eat-INDIC.3SG
(Mithun 2003: 565)
b. qimugte- m ner- i- a angun akuta- mek dog-ERG eat-APPL-3SG>3SG man.ABS mixture-ABL ‘The dog ate some “akutaq” on the man (ate the man’s akutaq).’ (Mithun 2000: 97) Now (12a) is a typical unaccusative clause, and (12b) is a related malefactive construction derived from the same verb root. Crucially the malefactive argument bears ergative case in (12b). (12)
a. Maklagaq kit’e-llru-uq bearded.seal.ABS sink-PAST-INDIC.3SG ‘The bearded seal sank.’ b. Ing-um maklagaq kic-i-lq-aa that.one-ERG bearded.seal.ABS sink-APPL-PAST-INDIC.3SG>3SG ‘The bearded seal sank on that guy.’ (Yup’ik; Woodbury 1981: 332–333)
We clearly cannot say that ‘that guy’ in (12b) is an external argument, or that it receives ergative case along with its malefactee theta-role by (a generalization of) the ICT, because ‘man’ gets the same theta-role from the same head (-i) in (11b) but does not have the same case. The DCT, however, can work, because there is another NP in (12b) (but not in (11b)) which ‘that guy’ c-commands, namely the theme ‘bearded seal.’ Hence the dependent-case rule in (2) can apply correctly. While (12) in Yup’ik is similar to (7) in Shipibo in that ergative case is assigned in the applicative of an unaccusative, the languages differ in which of the two internal arguments (theme or affectee) moves to Spec,TP and serves as the subject of the clause. Baker (2014a) argues that applied arguments in Shipibo are NPs embedded in a null-headed PP; this PP shell prevents the applied argument from moving to Spec,TP. Therefore, the theme argument must move, and it ends up c-commanding the applied argument and getting ergative case.6 Evidently, in the Yup’ik example in (12) (and similar examples in
6
Crucially the null P in this construction is not a phase head. That is why its complement is still visible to trigger ergative case on the theme after the theme moves to Spec,TP, whereas canonical PPs do not trigger ergative on the subject. Even overt Ps occasionally fail to be phase heads; see for example Baker & Vinokurova (2010: 623) on three such Ps in Sakha.
120 Mark C. Baker and Jonathan David Bobaljik distantly related Kalaallisut, Fortescue 1984: 269) the applied argument is truly an NP, so it can and does move to Spec,TP as the closer NP argument and gets ergative. Despite this difference, both languages provide instances of derived subjects receiving ergative case, contrary to the ECG. In this section, we have presented a series of examples that plausibly involve derived transitive subjects getting ergative case, which the ICT predicts to be impossible. Note that we are not claiming that applicatives of unaccusatives always yield an ERG–ABS array in ergative languages. The DCT leaves room for arrays other than ERG–ABS, thus permitting the analysis of a broader range of languages. Niuean provides a prominent example of an ergative language in which some applicatives yield an ABS–ABS array; (13) is an example (from Massam 2006: 33, cf. Legate 2012a).7 (13) Ne faka-kofu aki e PAST CAUS-cover with ABS ‘The canoe was covered with leaves.’
vaka canoe
e ABS
tau PL
lauakau. leaf
Note that this example is thematically similar to (10) from Chukchi, where ergative is assigned. Similar ABS–ABS patterns with nonagentive subjects are also found with reciprocals derived from ditransitive verbs and with a few psych verbs in Shipibo; these contrast with the applicatives of unaccusatives discussed above. Baker (2014a, 2015) develops one particular version of the DCT in which two arguments are ABS if three conditions hold: (i) the two arguments both start out inside the vP phase (hence are nonagentive), (ii) their initial c-command relationship is not reversed by NP-movement (as happens in Shipibo (7) but not Yup’ik (12b)), and (iii) ergative case is assigned at the spell out of TP but not VP. If these conditions do not hold, the higher argument will be ergative, even if it is nonagentive. Encouraging for a theory of this kind is the fact that the “location” argument appears to be the subject in Niuean (13) but the ‘locatum’ argument is the subject in (10) from Chukchi; this suggests that NP-movement does reverse the arguments in Chukchi but not in Niuean. See Baker (Chapter 31, this volume) for a review of this proposal and some additional discussion, making connections
7
This example is complicated by having a causative prefix in addition to the applicative aki, as Massam notes. This might suggest an alternative analysis in which it is a concealed ditransitive with a third, hidden (ergative) argument, hence an ERG ABS ABS array. Legate (2012a) also considers ERG– DAT and ABS–DAT arrays in Warlpiri, arguing that the distribution of verbs selecting these frames supports the ECG. She notes, however, that the difference in the arrays may be due to differing properties of the DAT argument in the two verb classes (NP versus PP), an analysis pursued in Baker (2015, Chapter 31, this volume). Niuean also has locative applicative-like transitive clauses formed from intransitive verbs that yield an ERG ABS array. Massam treats these as agentive intransitives, but some (‘sleep,’ ‘sit/stay’) may actually be unaccusatives, given that they show reduplication for plural number of the NP that is the subject of the intransitive, and the ergative of the derived transitives (Seiter 1980: 64). This process of participant number marking seems to be otherwise limited to internal arguments; no non-derived ergative triggers plural number marking on the verb.
On inherent and dependent theories of ergative case 121 to variation in ditransitive constructions. We can see, then, some room for variation in how derived subjects are case marked within a DCT, whereas in the ICT the ban on ergative case on derived subjects is expected to be quite rigid. Overall, then, the examples in this section fit with the DCT and go against a core prediction of the ICT. One could of course question whether these subjects are ‘derived’ in precisely the right sense picked out by the ECG. Since the advent of the VP-internal subject hypothesis, there is a sense in which all subjects are derived. What our examples show, we claim, is that there is no type of theta-role that is totally immune to ergative case: malefactives, locations, locatums, and even themes can all receive ergative case, if they end up as subjects with another NP lower in the same clause. Thus, what is relevant cross linguistically for ergative case assignment is transitivity within a local domain, not particular theta-roles. If these configurations are somewhat rare, it is because the lower an NP is on the theta-hierarchy, the more likely it is that, in a transitive context, the other NP will be the one that becomes the subject. If themes, for example, are quite low, then a configuration in which a theme becomes a transitive subject arises only when the higher argument fails to raise to subject position for some special reason, as in Shipibo.
5.4 Absolutive Case on External Arguments In section 5.3, we argued for the DCT over the ICT approach to ergative case by looking at themes and other internal arguments in noncanonical situations in which there is a second internal NP, showing that in many such situations they receive ergative case. In this section, we look more briefly at the converse situation: agents (external arguments) in ‘noncanonical’ situations in which an internal argument that would otherwise be there somehow becomes unavailable. According to the ICT, we would not expect this to matter much: the external argument presumably gets the same theta-role from the same head (v) as it does in canonical transitive constructions; therefore, it should get the same inherent ergative case from v, all things being equal. In contrast, the DCT takes the presence of a suitable second argument to be essential to the subject receiving ergative case, so if something happens to that second argument, the subject should receive default absolutive case, by (2). In section 5.4.1, we show that again it is the DCT that makes the right prediction. In section 5.4.2, we then critically discuss proposals by Woolford and others to patch the ICT by supplementing it with a transitivity condition.
5.4.1 Ergative Lost under Detransitivization One relevant thing that can happen to the internal argument of a transitive verb in some ergative languages is that it can incorporate into the verb. This happens productively in
122 Mark C. Baker and Jonathan David Bobaljik Chukchi, for example. (14) gives a prototypical noun incorporation (NI) pair (Polinskaja and Nedjalkov 1987: 240). (14) a. ətləg-e mətqəmət (kawkaw-ək) Father-ERG butter.ABS bread-LOC ‘The father spread the butter (on the bread).’
kili-nin spread.on-3SG>3SG
b. ətləg-ən (kawkaw-ək) mətqə-rkele-nen Father-ABS bread-LOC butter-spread.on-3SG>3SG ‘The father spread butter (on the bread).’ When the object is not incorporated, the subject is ergative, as expected ((14a)), but when the object is incorporated, the subject is not ergative but absolutive ((14b)). We know of no reason to say that the subjects in these two examples are theta-marked any differently, yet they differ in case. This is problematic for the ICT, since inherent case is supposed to remain constant when the surrounding syntactic context changes (as dative does in Icelandic). In contrast, this pattern follows from (2), as long as we assume that incorporated objects are inaccessible to the rule of dependent case assignment. This could be either because incorporation happens by compounding in the lexicon, so that there is no object present in the syntax at all, or (as we believe) because the trace of the syntactic movement that creates incorporation is invisible to rules of case and agreement.8 The same effect can be seen in Kalaallisut, with the slight complication that some verbal morphemes require NI in Kalaallisut and others forbid it, so one needs to compare different verbs with similar meanings in that language. A second thing that can happen to an internal argument in some ergative languages is that it can be projected not as an NP but as a PP (with P a phase head). We observed above that Chukchi is rich in these conative/locative-type alternations, and that is relevant here as well. In particular, the location argument which is expressed as a PP in (14b) can alternatively be expressed as a bare NP (compare English: I smeared butter on the bread vs. I smeared the bread with butter). When the locatum argument is incorporated but the location argument is projected as an NP, then the agent-subject is marked ergative again, as shown in (15). (15) ətləg-e kawkaw mətqə-rkele-nen Father-ERG bread butter-spread.on-3SG>3SG ‘The father spread the bread with butter.’ (Polinskaja and Nedjalkov 1987: 240) It is very possible that (15) means something slightly different from (14b), with ‘bread’ counting as a location in (14) but as a theme in (15). But it is very doubtful that the 8 For example, Baker et al. (2005) argue that the phi-features of the trace of head movement are deleted in Chukchi and some other languages, making it invisible to agreement. This deletion would bleed (2) if we understand ‘NP’ as ‘phrase bearing phi-features.’
On inherent and dependent theories of ergative case 123 subject gets different theta-roles in (14) and (15): it looks like a canonical agent in both. Comparing these three examples, then, it seems evident that the theta-role of the subject does not determine whether it is ergative or not, but whether there is another NP in the same domain as the subject does. In (15), ‘bread’ is another NP in the same domain as the subject, but in (14b) it is not, given that PPs are usually separate domains (phases), the internal constituents of which are invisible to the outside world for purposes of case and agreement (see Baker, Chapter 31, this volume, but see note 7 for some exceptions). A third thing that can happen to a theme object in some ergative languages is that it can be removed as an object by antipassive. Descriptively speaking, antipassive is a morphological process that removes the object of a transitive verb from the core syntax, leaving it either unspecified or expressed as an oblique/PP. Chukchi illustrates again: (16a) is a normal transitive; (16b) is the corresponding antipassive (Nedjalkov 1976: 201). (16) a. ətləg-e qərir-ə-rkən-en ekək. father-ERG seek-PRES-3SG>3SG son.ABS ‘The father is seeking the son.’ b. ətləg-ən ine-lqərir-ə-rkən father-ABS APASS-seek-PRES.3sS ‘The father is searching (for the son).’
(akka-gtə). son-DAT
Again, the subject is ergative in (16a) and absolutive in (16b), even though there is no detectable change in its theta-role—a problem for the ICT. However, whether a theme argument is syntactically present in the clause, and if so whether it is expressed as an NP or a PP, clearly does affect the case of the subject, exactly as expected under the DCT (although we do not commit to any particular view of the antipassive here). Similar facts can be given for Kalaallisut, and for various Australian languages.
5.4.2 Against Supplementing the ICT with a Transitivity Condition Unlike the facts surrounding the Ergative Case Generalization, the facts outlined in this section are well known and not in dispute. Prima facie, they seem (to us) to provide a strong argument against the ICT, although the issue is oddly under-discussed. Where the issue is addressed, proponents of the ICT contend that the thematic condition on ergative case assignment needs to be supplemented (in some languages) with a transitivity condition (Massam 2006: 32; Woolford 2006: 119–120; Legate 2012a: 182). However, the exact nature of this transitivity condition and its theoretical implications are usually left unexplored. (It is not obvious, for example, how v can ‘see’ whether there is an NP inside VP, and if so why that should affect its relationship with its specifier, as pointed
124 Mark C. Baker and Jonathan David Bobaljik out to us by Laura Kalin.)9 Canonical instances of inherent case assignment, such as dative in Icelandic, are patently not subject to a transitivity condition, since dative case in Icelandic is possible on the subjects of monadic predicates. Only Woolford (2006) makes a real effort to give independent motivation for a transitivity condition on inherent case, by arguing that a similar condition holds for dative case in Japanese and Basque. Much could be said about her particular examples. For example, in Japanese the putative transitivity condition is far from general, holding if at all in matrix clauses only (Shibatani 1977). Moreover, for languages in which it is true that dative arguments never occur as the sole NP in a clause, there are at least two ways in which this might be explained without recourse to a stipulated (and as yet unformulated) transitivity condition on inherent case assignment. First, it could well be that in some languages dative case is not an inherent case at all, but rather another instance of dependent case—specifically dependent case assigned to the higher of two NPs generated inside the same VP (see for analyses of this type Baker 2015; Baker & Vinokurova 2010; Bobaljik & Branigan 2006). Second, there could be languages in which dative case is inherent, but in which inherent case nominals may not function as true subjects. Icelandic and German famously contrast in whether datives may (Icelandic) or may not (German) serve as grammatical subjects (Zaenen et al. 1985). Rezac (2008) (see also Davison 2004a) argues that the Basque datives pattern with German rather than Icelandic with regard to subjecthood diagnostics. If there is a requirement of a syntactic (EPP) or morphological (possibly default agreement) subject in every clause, then clauses with dative arguments will always appear to have an additional, possibly null, argument. This effect arises without a stipulated transitivity condition on inherent case. Importantly, since ergative NPs cross-linguistically satisfy subjecthood tests, this reasoning about Basque datives cannot be extended to ergative. Thus, we conclude that there is no independent motivation for putting a transitivity restriction on the assignment of ergative case, as IC theorists need to do. Indeed, there are syntactic configurations in which the arguments that receive ergative in simple transitive clauses undergo case alternations even though there is no loss of transitivity. Rezac et al. (2014) present Basque perception verbs as part of an extended argument against the ICT. As in many languages, perception verbs may take a full CP complement ((17a)), or a reduced complement ((17b)): (17) a. Katu-ek sagu-ak harrapa-tu cat-PL.ERG mouse-PL.ABS catch ‘I saw that the cats caught the mice.’
dituzte-la ikusi dut. AUX-that see AUX.1SG
9 Omer Preminger suggests one possible way of working this out. He points out that a v that assigns accusative case to an NP in its domain under Agree could be considered a distinct lexical item from a v that does not. Given this, the agent-subject in a detransitivized structure (or unergative structure) is theta-marked by a different v from one in a transitive structure, even though the thematic role is the same. This could account for their different cases, if the first v is not an assigner of inherent ergative case whereas the second v is.
On inherent and dependent theories of ergative case 125 b. Katu-ak sagu-ak harrapa-tzen ikusi ditut. cat-PL.ABS mouse-PL.ABS catch-ING seen AUX.1SG>3PL ‘I saw the cats catch mice.’ (Rezac et al. 2014: 1280) The configuration (17b) is a species of ECM environment. Rezac et al. show carefully that the NP katu-ak ‘cats’ is thematically the external argument of the embedded clause, which is transitive, but it behaves syntactically as the object of the higher clause, and therefore bears absolutive case. On the DCT, (17b) is unremarkable; ‘cats’ does not get ergative case because it is in a case domain with the matrix subject (presumably because CP is missing, contrast (17a)) and it is not the highest NP in the domain.10 Yet on the ICT, the alternation is surprising: inherent case is supposed to be retained on arguments in ECM environments, as inherent dative case is in Icelandic, a model for the ICT. Stipulating that inherent ergative is only realized in a transitive clause—even in the manner of n. 10—does not avoid the problem in (17b), since both the matrix and embedded clauses are fully transitive. Causative constructions make a similar point. As in many languages, the causee (the embedded subject) in Basque is marked absolutive if the embedded predicate is intransitive but dative if the embedded predicate is transitive, as in (18). (18)
Jon-ek Xabier-ri Mikel hil-arazi dio Jon-ERG Xabier-DAT Mikel.ABS kill-CAUS AUX.3SG>3SG.3SG.DAT ‘John made Xavier kill Michael.’ (de Rijk 2007: 378; our gloss)
Here again, there is a case alternation (between ergative and dative) such that an agent argument gets different cases in different syntactic environments, contrary to the view that ergative case is an inherent case. Moreover, since ‘kill’ still has its theme argument in (18), no appeal to a transitivity restriction will explain why its subject ‘Xavier’ is not ergative in (18). Although ECM constructions like (17) may be rare across languages,
This approach in essence builds a version of the DCT (ergative is dependent on accusative in the same domain) into the lexical entries for v heads. It strikes us as providing no insight into why this condition might hold. For example, the same machinery would allow one to stipulate that only the v that does not assign accusative assigns inherent ergative to the NP it theta-marks. This would yield a pattern where only unergative subjects are ergative, while transitive (and unaccusative) subjects are nominative. This pattern cannot be readily described on the DCT approach, and as far as we know it never occurs. Moreover, even if workable, the theoretical possibility sketched here addresses only the narrow issue of how a transitivity condition on ergative might be formulated within an ICT; it does not address the other evidence we have collected in favor of treating ergative as a structural (dependent) case. 10 Omer Preminger very rightly asks why, even given that (17b) is a single case domain, the lower subject ‘cats’ does not get ergative simply by virtue of c-commanding the lower object ‘mice.’ For us, the ERG–ABS–ABS pattern in (17b) is exactly parallel to the ERG–ABS–ABS pattern found with simple ditransitive verbs in ergative languages ((almost) never ERG–ERG–ABS). However, there is indeed more to say about why this is so: Baker (2014a, 2015) attributes it to a cyclicity effect related to vP also being a phase; see Baker (Chapter 31, this volume) for a reprise.
126 Mark C. Baker and Jonathan David Bobaljik morphological causatives are fairly common; the same argument can be constructed in Inuit languages, for example.11 In sum, there is abundant evidence that the appearance of ergative case is tied to surface transitivity. On the ICT, this requires a special condition on the assignment of ergative case. We suggest that once the transitivity condition is brought fully into the light and elaborated properly, as it is in a DCT, there is no further need for a thematic condition that the ergative must be an agent—at least for the pure ergative languages we are focusing on here, like Shipibo, Chukchi, and Kalaallisut. Moreover, there is evidence from embedded environments that transitivity alone is insufficient as a condition on ergative case, even for canonical agents, whereas the DCT seems to draw the right distinctions.
5.5 Typological Considerations: on the Rarity/Absence of Active Case Systems 5.5.1 Preliminaries In fact, the challenges for the ICT mentioned in the previous section are arguably even more general. Following the ICT’s leading idea in its purest form naturally leads us to expect ergative case not only on the subjects of detransitivized transitive verbs, but also on the subjects of simple unergative verbs, given that they also receive the agent theta-role from v on standard generative accounts. In other words, the widespread view that v assigns inherent case to the agent NP that it theta-marks most naturally generates an active case pattern rather than a true ergative case pattern—a pattern in which one case appears on the subjects of transitive verbs and unergative verbs, and a different case appears on the objects of transitive verbs and the subjects of unaccusative verbs. This is not the situation in languages like Shipibo, Inuit, and Chukchi, for which the ICT must invoke something like Woolford’s (2006) transitivity condition. But if the ICT is the right leading idea, then we might expect to see it working in purer form in some other languages, unalloyed with a transitivity condition. If this is not so—if there are no genuinely active case marking languages—we may begin to doubt not only whether the ICT should be the primary theory of ergative case, but whether it is even allowed by universal grammar. With this in mind, we argue (contra,
11 Julie Legate (p.c.) reminds us that the causee of a causative construction might be projected as an internal argument of the causative verb in a control-like structure in some languages, rather than as the external argument of the lower verb in an ECM-like structure. When that happens, it is not expected to get ergative case even within an ICT. However, one would still expect the transitive causee to be invariantly ergative in some subset of ergative languages with a morphological causative (namely those with an ECM-style causative), whereas we know of no ergative language with this pattern.
On inherent and dependent theories of ergative case 127 e.g., Woolford 2015) that current typological knowledge does not offer any promising paradigm case of a dependent-marking language with a true active case system. Space limitations (and some knowledge limitations) prevent us from discussing any putatively active language in depth, but we outline what we see as the major issues, as a spur to further work. Typological sources say that the active–inactive/stative alignment pattern (also called a split-S or fluid-S pattern) is attested in languages of the world: see, for example, Merlan (1985); Mithun (1991b); Dixon (1994: 70–83); and Comrie (2005). But our certainty that this is so is marred by the fact that these discussions generally conflate data from morphological case marking on nouns with data from agreement patterns on verbs. Mithun (1991b) is a typical example: of the five active languages that she discusses at some length, four are head-marking languages (Lakhota, Guaraní, Caddo, Mohawk), and only one (Central Pomo) has overt case marking on NPs. Dixon’s discussion is similar, and he notes in passing (1994: 76) that “for most languages of this type morphological marking is achieved by cross-referencing on the verb”; see also Merlan (1985: 353). The only languages Dixon mentions as having active case marking on NPs other than (Eastern) Pomo are two Caucasian languages: Laz and Tsova-Tush. Data from the World Atlas of Language Structures confirms that there is a strong interaction between Nichols’s (1986) head-marking/dependent-marking distinction and the distinction between active languages and true ergative languages. Siewierska (2005) lists 26 out of 380 languages as having an active agreement pattern, a respectable 6.8 percent. Indeed, in agreement-oriented languages, an active system is slightly more common than a straight ergative system (19/380, 5 percent). In contrast, Comrie (2005) lists only 4 out of 190 dependent marking languages as having an active case pattern (2.1 percent), and in this language type a straight ergative pattern is much more common than an active one (32/190, 16.8 percent). For what it is worth, a rough chi-squared contingency table test confirms that marking type and alignment type are not independent in this data (χ2=25.998, p < 0.00001), suggesting that one should not combine head-marking and dependent-marking languages in discussions of this topic. Once we refrain from doing so, we face the fact that active dependent-marking languages are at best extremely rare. Comrie lists only Basque, Georgian, Imonda, and Drehu, to which we can add the Pomo languages discussed by Mithun and Dixon, Dixon’s Laz and Tsova-Tush, and Lhasa Tibetan discussed by DeLancey (1984a, 2011). Of these languages, the first two are well known to generativists, and have (together with Hindi) encouraged the idea that ergative case is inherent case; in contrast, the others are little known to generativists. We briefly survey why we do not think that any of these languages provides a good prototype for the ICT to build on.
5.5.2 Split Active Languages Our first observation is that there are many languages such as Shipibo, Inuktitut, Tsez, and Chukchi, whose morphological ergativity shows a clean pattern, in the
128 Mark C. Baker and Jonathan David Bobaljik sense that transitivity alone is the determining factor for ergative case. By contrast, in many potential active languages, ‘ergative’ case is conditioned not only on the subject of the clause having an agent thematic role (as the ICT would lead one to expect), but also on a variety of other factors. In Hindi, for example, cited in support of the ICT, only a small subclass of unergative verbs (‘cough,’ ‘bark’) permit ergative subjects, and even with those verbs ergative is optional. Moreover, Hindi is famously a split ergative language, in which ergative only occurs in the perfect aspect (see Butt and King 2003), never in the imperfective. Similarly, in Georgian, the subjects of transitive verbs and some intransitive verbs are famously ergative only in aorist and perfect clauses (see Harris 1981). Similar remarks hold for Lhasa Tibetan, in which subjects of agentive intransitive verbs are sometimes marked with the ergative particle, but only in perfective clauses, never in imperfectives, as illustrated in (19): (19) a. ŋa-s Seattle-la I-ERG Seattle-to ‘I went to Seattle.’
phyin-pa-yin. went-PRF-VOL
b. ŋa-(*s) Seattle-la I.NOM(*ERG) Seattle-to ‘I will go to Seattle.’
(DeLancey 1984a: 133)
’gro-gi-yin. go-FUT-VOL
Yet another example is Drehu, an Austronesian language spoken in New Caledonia (Tryon 1967; Moyse-Faurie 1983), which actually draws a three-way distinction among the subjects of present clauses (nominative), past clauses (marked nominative) and other tenses (canonical ergative, with rare exceptions). This interaction with tense–aspect is well known, but its theoretical implications are not always kept clearly in mind. It strongly suggests that Tense and Aspect heads are heavily involved in the assignment of case—or, for the DCT, in establishing the relevant case domains—in these languages, not the theta-role assigning head v. In none of these so-called active languages is the subject marked ‘ergative’ if and only if it receives an agent theta-role from v, as the ICT would expect.12 They are quite different in this respect from Icelandic, where verbs that take inherent dative case subjects do so in every tense–aspect, just as we would expect on theoretical grounds. Since ergative
12 We admit that we do not have a fully worked out DCT analysis for these “split active” languages. But see Baker (2015, Chapter 31, this volume) for a proposal in which some aspect heads are extra phase heads, affecting when two NPs count as being in the same domain for (2). See also Nash (Chapter 8, this volume) for a new DCT-style approach to the core facts of Georgian. Note also that in Drehu, noun incorporation of the object bleeds ergative case marking of the subject in relevant clause types (nonpast, nonpresent, Moyse-Faurie 1983: 159). We took parallel facts in Chukchi to be evidence in favor of the DCT in section 5.4.1.
On inherent and dependent theories of ergative case 129 case varies across clauses in this way, it seems very doubtful that it should be thought of as an inherent case on the Icelandic model.13 Tsova-Tush (aka Batsbi) also shows a split, but in terms of person, rather than tense– aspect. This language has been claimed to be a particularly good candidate for an active language, in that ergative case tracks the agentivity (volition and control) of the intransitive subject very closely (Holisky 1987; thanks to Omer Preminger for pointing this out). However, only first and second person pronouns may bear ergative in intransitives; third person pronouns and NPs are always nominative in this role, though they must be ergative as transitive subjects (Holisky 1987: 104–105, 119). In addition, nuances of volitionality and control, which Holisky characterizes as components of meaning distinct from thematic roles, come into play in case-selection only with intransitive subjects (1987: 122). Both the person split and the fluidity of interpretation are differences between transitive agents and intransitive agents that are entirely unexpected by the ICT, and they prevent Tsova-Tush from counting as its missing prototype.
5.5.3 Active Languages with Unmarked ‘Ergative’ The next group of putatively active dependent marking languages to consider includes the Pomo languages (Hokan family, spoken in California; see especially O’Connor 1987) and Imonda (Papuan; Seiler 1985). These languages are all quite similar in that the putative ergative case on agentive subjects is actually morphologically unmarked (Ø); what is overtly marked is a kind of nonagentive case found on (some) direct objects and (some) nonagentive subjects of intransitive verbs: –al on pronouns in Northern Pomo; –m in Imonda. This is already somewhat suspicious for the ICT, since it is extremely rare for ergative case to be morphologically unmarked in a canonical ergative language (Nias is perhaps the only attested example; see Baker 2015: ch. 3 for an analysis). Bittner and Hale (1996b) refer to these as “accusative active languages,” suggesting that it is an extension of the objective case that yields the pattern, rather than the existence of a thematically restricted but phonologically null ergative case. Indeed, O’Connor argues in some detail against the idea that Northern Pomo’s null-marked case forms are connected to a particular theta-role in the way that the ICT would hope, concluding that “the A case is semantically unmarked, it does not convey any information about volition, control, agentivity, etc.” (196). Most importantly, it is clear that the putative ergative marking found on intransitive subjects in these languages extends well beyond the canonical unergative class to include many typical unaccusative predicates: for example, ‘die,’ ‘sleep,’ and ‘misspeak’ in Northern Pomo, and ‘startle’ in Imonda. Indeed, Seiler (1985: 145–148) reports
13 We set aside here the possibility of a hybrid model, in which inherent case-marked NPs need abstract (structural) licensing in addition, as proposed for Icelandic by Cowper (1987); see Shimamura (2014b) for an extension of this to account for subject–object extraction asymmetries in syntactically ergative languages.
130 Mark C. Baker and Jonathan David Bobaljik that only eight known verbs take the nonagentive case marker on their subject in Imonda. These languages, then, are not good prototypes for the ICT either. Instead, our tentative analysis of these languages is that they are really neutral languages, in which neither dependent ergative case nor dependent accusative case is consistently assigned. Rather -al in Northern Pomo and -m in Imonda are fundamentally dative case markers. As such, their core use is on the goal arguments of ditransitive verbs (indeed Seiler’s gloss for -m is GL, short for ‘goal’). These case markers can also be used on the theme arguments of monotransitive verbs, but only as so-called differential object markers (Bossong 1985; Aissen 2003b). Thus, overt affixes like -al are used in Pomo only on pronouns, proper nouns, and (with a clitic or demonstrative) animate or human nouns used as direct objects; inanimate common nouns do not show overt case inflection. Similarly, in Imonda the case marker -m marks only high-animate objects (Seiler 1985: 163–165). This is very reminiscent of markers such as Hindi -ko, which serve both as differential markers for animate and/or specific direct objects, and in dative functions such as marking recipients (the most common DOM pattern, Bossong 1985). Once we think of the overt cases in these languages as datives rather than accusatives, an easy hypothesis becomes available for why they are found on the subjects of a proper subset of the unaccusative predicates: these are simply predicates that select for quirky dative case on their subjects, like those known from Icelandic (cf. Marantz’s (1984) generalization that only direct internal arguments—nonagents—of a verb can receive lexical case from that verb in Icelandic). This fits well with the fact that only eight known verbs have case-marked subjects in Imonda, not the whole class of unaccusatives. It also fits with O’Connor’s (1987) observation that having overt case-marking on the subject in Northern Pomo sometimes expresses ‘empathy’ with the subject rather than agentivity; we interpret such subjects as being experiencers rather than themes, with lexical dative case being assigned to experiencer arguments only. If this is right, then Imonda and the Pomo languages are not the result of v assigning null ergative to agents, but the result of dative case being extended to some objects via DOM and to some nonagentive subjects as an instance of dative subject constructions—two familiar developments that happen to come together in these languages. If this is on the right track, then these so-called active languages are not directly relevant to comparing the ICT and the DCT as theories of ergativity.
5.5.4 Languages with Concealed Transitives This leaves only Basque and Laz to consider. Basque also proves problematic for the ICT (though it is in some ways also challenging for the DCT). In Basque, ergative case is indeed found on the subjects of some intransitive verbs as well as on the subjects of transitive verbs; an example is (20). (20) Euskara-k noiz arte Basque-ERG when until ‘How long will Basque last?’
iraungo last
du? AUX.3SG>3SG (de Rijk 2007: 265; our gloss)
On inherent and dependent theories of ergative case 131 However, the generative literature on this language has shown clearly that the case of the subject does not match up perfectly with whether the verb is unaccusative or unergative, as the ICT would hope. The verb ‘last’ in (20), for example, is a likely unaccusative verb; ‘boil’ is another that takes an ergative subject. Basque even has raising predicates like behar ‘need’ that take ergative subjects, as argued at length by Rezac et al. (2014) (but see Laka, Chapter 7, this volume for an opposing view). Moreover, as Preminger (2012) notes, examples where nearly synonymous verbs—or even the same verb across different dialects— take subjects with different cases seem to speak against a view that ties case too closely to theta-roles. Instead, some kind of lexical idiosyncrasy seems called for, as in Pomo and Imonda. However, it does not seem right simply to say that verbs like ‘last’ assign quirky/lexical ergative case to their theme arguments. The reason is because ergative case on the subject of ‘last’ varies across syntactic structures, just as ‘regular’ ergative subjects do in Basque (see (17) and (18)). For example, (21) embeds ‘last’ in a causative construction: some varieties of Basque use dative case on the argument of ‘last,’ as in (21), others use absolutive case, but none preserve ergative case on this argument. (21)
Norbaite-k eta zerbaite-k iraun-arazi dio hizkuntz-ari. Someone-ERG and something-ERG last-CAUS AUX.3SG>3SG.3SD language-DAT ‘Someone and something has caused the language to last.’ (de Rijk 2007: 380; our gloss)
Similarly, intransitive verbs with ergative subjects can be embedded under a perception verb like ‘see’ to give sentences like ‘I saw the milk boiling.’ Then ‘milk’ has absolutive case (Karlos Arregi and Ikuska Ansola-Badiola, p.c.), the result of ECM, not ergative case. The case of the subject in examples like (20) thus behaves like structural case, not inherent case. Given this, our DCT view leads us to locate the lexical idiosyncrasy of verbs like ‘last’ elsewhere: we suggest that they are concealed transitives, taking a second argument that is approximately meaningless and phonologically null, but nevertheless counts for triggering ergative case on its coargument by the dependent case rule in (2). For the more unergative verbs in this class, especially semantically monadic predicates that are syntactically expressed as a light (or ‘compound’) verb construction (e.g., lo egin ‘sleep do’ [=‘sleep’]), this is a conventional analysis (see Bobaljik 1993a; Hale & Keyser 1993; Laka 1993b).14 This view predicts that verbs like ‘dance’ and ‘last’ should also behave like transitive verbs for other syntactic diagnostics (where available). In this connection, we take it to be significant that the causative form of ‘last’ in (21) has a causee in dative case,
14 Preminger (2012) argues that not all clauses with unexpected ergatives in Basque should be analyzed as having a null absolutive nominal. However, his most compelling point is that dummy absolutive agreement on the auxiliary is not a positive argument for this view. It is still possible that these clauses have a null argument that is active for dependent case but inert for agreement—indeed, so inert that it does not even count as a defective intervener, as dative NPs do in Basque. See Baker (2014a) for an analysis of this sort for the small class of verbs in Shipibo mentioned in n. 5.
132 Mark C. Baker and Jonathan David Bobaljik not absolutive case, in standard/conservative/Southern Basque dialects.15 This is crucially a property of transitive verbs, not intransitive ones (see (18)), supporting the claim that some kind of null object is indeed present with verbs of this type. If this hypothesis holds true in general, then a DCT approach to Basque may be tenable, whereas an ICT approach seems not to be. A concealed transitive approach may well be a promising analysis for the Kartvelian language Laz, as well. (22) shows an active-style contrast between unergative and unaccusative verbs in this language: (22) a. Himu-k i-bgar-s. s/he-ERG VAL-cry-PRES.3SG ‘S/he is crying.’
(Öztürk & Pöchtrager 2011: 26)
b. Him ulu-n s/he.NOM go-PRS.3SG ‘S/he is going.’ In related Georgian, ergative marking is limited to the aorist (perfect) tenses, but in Laz, it has spread to both major tense/aspect classes (Harris 1985). This lack of an aspect split leads Woolford (in Chapter 9, this volume) to cite Laz as the prime example of the pattern predicted by the ICT. Yet even without the split, there is an important point of comparison to Georgian: Nash (in Chapter 8, this volume) analyzes the prefix i-in Georgian unergatives parallel to (22a) as a reflexive marker, occupying the internal argument position and providing a case competitor for the subject NP. Note that the cognate prefix (glossed VAL for valency) is present in (22a) but not (22b), and Öztürk & Pöchtrager (2011: 68) suggests that this difference is (reasonably) systematic in Laz.16 If correct, this suggests that in Laz, as in Basque, it is formal–syntactic transitivity (possibly partially concealed), not theta-roles per se, that determines case marking, as expected under the DCT.
5.5.5 Theoretical Implications Whatever the ultimate account of some of the putatively active languages surveyed here turns out to be, we have found no clear case of a uniformly active dependent marking language in the literature. This leads us to conjecture about why active alignment 15
In innovative dialects where the causee is absolutive, we might tentatively say that ‘last-CAUS’ has evolved into a lexical item in its own right, not composed syntactically from ‘last’ and ‘CAUS.’ As such, its case pattern is simply that of an ordinary transitive verb. 16 In contrast, Harris (1985) argues that the correspondence between i-prefixation and ergative case- marking on unergatives is weaker in Laz than in Georgian. Harris provides evidence that even (some) non i-marked unergatives are concealed transitives, though she ultimately does not accept this as a general synchronic account.
On inherent and dependent theories of ergative case 133 patterns are attested in head-marking languages but not in dependent-marking languages. This is, of course, quite a mysterious distribution from the point of view of standard Chomskyan theory, which holds that case and agreement are two sides of the same coin; from that perspective, one expects the very same alignment patterns to show up in languages that realize case overtly and languages that realize agreement overtly. But dependent case theory is not committed to there being a deep parallelism between case and agreement. On the contrary, dependent case crucially indicates a relationship between two NPs, with possibly agreeing functional heads playing no direct role. As a result, the transitivity of the clause is crucial, but the absolute position of a single NP within the clause (the unergative–unaccusative distinction) will typically not be crucial. Therefore, we observe ergative case marking patterns, but few or no truly active case marking patterns. In contrast, agreement is crucially a relationship between an agreeing functional head and an NP. It is perfectly plausible, then, that the location of a single NP within a clause could make a difference for this: for example, the closest c- commanding head with agreeing features for a theme argument could well be different from the closest c-commanding head with agreeing features for an agent argument (it could be v as opposed to T, for example; see Baker, Chapter 31, this volume for examples from Burushaski). Therefore, the DCT can contribute to a plausible theoretical explanation of why active–inactive systems are not uncommon in languages with agreement systems, but are extremely rare or impossible in languages with overt case marking on nominals. In contrast, a straightforward ICT might expect languages with active case patterns to be at least as common as those with pure ergative patterns, contrary to fact.
5.6. Conclusion In this chapter, we have compared two contrasting theories of ergative case in some detail: a theory in which it is an inherent case assigned by v along with an external theta- role, on analogy with inherent dative case in Icelandic, and a dependent case theory in which it is assigned to the higher of two NPs in the same local domain. We have found many advantages for the second sort of theory: it can explain why nonagentive verbs with two NP arguments can have ergative case on one of those NPs when conditions are right, it can explain why agentive verbs that are detransitivized in one way or another typically lose ergative case on the subject, and it can explain why uniformly ergative languages are far more common than uniformly active case (as opposed to agreement) patterns throughout the world. Along the way, we have criticized the move of supplementing the ergative-as-inherent-case theory with a transitivity condition, claiming that this sneaks the dependent case idea in through the back door and renders the thematic condition superfluous. We therefore submit that the inherent-case view of ergative fails to find support over the dependent-case view, at least for many canonical ergative languages. Indeed, the current typological record makes us question whether the inherent-ergative case view is even allowed as an option by universal grammar.
134 Mark C. Baker and Jonathan David Bobaljik
Acknowledgments For discussion of the material presented here, we thank Jessica Coon, Laura Kalin, Julie Legate, Omer Preminger, Koji Shimamura, and Susi Wurmbrand, as well as audiences and seminar participants at CUNY, MIT, UConn, McGill, and Rutgers University.
Abbreviations We use the following abbreviations in glosses: ABL, ablative; ABS, absolutive; APASS, antipassive; APPL, applicative; AUX, auxiliary; CAUS, causative; DAT, dative; ERG, ergative; FUT, future; GEN, genitive; IMPF, imperfective; INDIC, indicative; ING, gerund(-like); INSTR, instrumental; LOC, locative; O+, objective/patientive case marking (Pomo); PL, plural; PRES, present; PRF, perfective; PRT (evidential), particle; SM, singular masculine; VAL, valence; VOL, volitionality; 1SG>PL indicates an unsegmented agreement morpheme for 1SG subject and 3PL object. Source abbreviation: PV = Valenzuela (2003).
Chapter 6
The l o cu s of ergative c ase Julie Anne Legate
6.1 Introduction In this chapter, I demonstrate that the factors governing the assignment of ergative case vary significantly from language to language, are multifaceted, and are low in the clause, centered around vP, in a wide range of languages. I consider two languages for which the governing factors prima facie seem high in the clause, and find that instead these factors are in fact low. It is left open whether all apparent high-ergative languages are similarly subject to reanalysis. The patterns discussed herein demonstrate that the assignment of ergative case cannot be reduced to a single factor and should not be oversimplified. In addition, the patterns are perhaps unexpected for two types of approaches to ergative case. One such approach treats ergative as a high case, based in the CP/TP domain (for example, Levin and Massam 1985; Bobaljik 1993a; Chomsky 1993; Bittner and Hale 1996a; Bobaljik and Branigan 2006). (Note that care must be taken with such a high-ergative analysis to ensure that we are not simply dealing with nomenclature—a proposal for a language whereby “ergative” is a structural case assigned by T and “absolutive” is a structural case assigned by v is a proposal that the language in question is in fact nominative- accusative.) The second approach for which the patterns discussed here are perhaps unexpected are dependent case approaches, whereby the assignment of ergative is dependent on the presence of another DP in the same domain not already marked with a (lexical) case (for example, Marantz 1991; Bittner and Hale 1996a, 1996b; Baker 2014a). Before beginning, I should note that ergative as a term is used descriptively by authors from many different traditions, sometimes due to the unique patterning of the transitive subject in the relevant language, sometimes due to the unified patterning of the intransitive subject and the transitive object, sometimes for other reasons. We cannot expect that everything labelled “ergative” will turn out to be instances of a single phenomenon.
136 Julie Anne Legate In this chapter, I leave aside the unified patterning of the intransitive subject and the transitive object; see Legate (2008) for my thoughts on this matter. I focus instead on the unique patterning of the transitive subject, and include the extension of this patterning to unergative intransitive subjects in some languages (these forming a subset of split-S languages). There are many ergative languages for which ergative case fails to be morphologically realized on a subset of nominals; I abstract away from this morphological realization here, but see Legate (2014a). However, I exclude languages that lack morphological realization of ergative case entirely—agreement/clitic patterns are not simply faithful representations of case patterns, and cannot be treated as such.
6.2 Low Ergative In this section, I review a number of unrelated languages in which ergative is assigned based on properties low in the clause, centered around vP. Indeed, this situation is well attested; I only provide a few representative examples here. The examples chosen also illustrate that the factors contributing to ergative case assignment differ across languages, and are often multifaceted within a language, and hence we cannot insist upon a simplistic, uniform analysis of ergative case assignment.
6.2.1 Tsova-Tush Tsova-Tush (aka Batsbi) (East Caucasian: Georgia) (Holisky 1984, 1987; Holisky and Gagua 1994)1 exhibits a case pattern whereby transitive subjects bear ergative case, while intransitive subjects bear either ergative or nominative.2 The case found on the intransitive subject is based on the θ-role borne by the DP, a property determined low, within the vP; to wit, subjects of unaccusatives bear nominative, whereas subjects of unergatives bear ergative. Holisky (1987) separates Tsova-Tush intransitive predicates3 into classes, based on the propensity to use nominative or ergative case; she notes, however, that the classes are fluid, depending on what situation the speaker has in mind. The first class, consisting of approximately 31 verbs, allows nominative only on S, and only an unaccusative interpretation is possible. Examples include a=reva(d)dalar ‘be confused,’ and h”abdalar
1
Thank you to Dee Ann Holisky for discussion of Tsova-Tush. I use nominative here as the traditional term for this morphological form of the verb; it appears on intransitive subjects and transitive objects. I leave aside whether this form corresponds to nominative from T for both the subject and object, or nominative from T for the subject and accusative from v for the object. For related discussion, see for example, Aldridge (2008a); Legate (2008a); Coon et al. (2014). 3 Including those that appear with an oblique, rather than nominative, object; see Holisky 1984: 192, n.10. 2
The locus of ergative case 137 ‘be mentioned, be remembered.’ Verbs in the second class, about 27 of them, preferentially use nominative, but allow ergative with a marked volitional interpretation (=unergative). For example of dah” dax:ar ‘drown, suffocate,’ a consultant suggested: “if a distraught rejected lover throws herself into a river and drowns, she could later, hypothetically, relate her death by using ergative marking” (Holisky 1987: 110). About 61 verbs appear either with nominative or ergative subjects; the corresponding meaning difference reflects an unaccusative versus unergative interpretation. For example, ʕopdalar with a nominative subject means ‘come to be hidden,’ in the context that something moves in front of you so that you end up hidden, whereas with an ergative subject means to hide oneself. Similarly, ‘fall’ can be understood as agentively with an ergative subject, or nonagentively with a nominative subject: (1)
a. (as) vuiž-n-as 1SG.ERG fell-AOR-1SG.ERG ‘I fell down, on purpose’ b. (so) vož-en-sO 1SG.NOM fell-AOR-1SG.NOM ‘I fell down, by accident.’
(Holisky 1987: 105)
For approximately 36 verbs, the subject is usually ergative, but may be nominative under a marked non-volitional interpretation. ga=rek’a(d)dalar ‘run very fast’ falls into this class; a consultant suggested for the nominative interpretation “a person doesn’t want or intend to run, but starting down a hill, finds himself running because it is very steep” (Holisky 1987: 112). Finally, approximately 78 verbs appear only with ergative subjects under an unergative interpretation, including lavar/levar ‘talk,’ and lap’c’ar ‘play.’ Furthermore, there is evidence of a structural difference between intransitives with nominative subjects and intransitives with ergative subjects. The intransitive marker - Dalar4 when added to a transitive eliminates the ergative subject, yielding an unaccusative. When added to an intransitive that normally takes an ergative subject, it yields an intransitive that takes a nominative subject, with an unaccusative “unintentional action” interpretation (Holisky and Gagua 1994). (2) a. ču Jiš-n-as PVB go.to.bed-AOR-1SG.ERG ‘I went to bed’ b. ču Jiš-Jal-in-sŏ PVB go.to.bed-INTR-AOR-1SG.NOM ‘I went to bed (unconsciously, without realizing it)’ (Holisky and Gagua 1994: [81b,c]) 4
See Holisky and Gagua for the (morpho)phonological rules that yield the surface forms of this and other verbal morphology.
138 Julie Anne Legate Thus, the pattern of ergative case assignment in Tsova-Tush shows significant sensitivity to the θ-position of the DP—external arguments receive ergative case, while subjects of unaccusatives receive nominative. This factor regulating ergative assignment is low in the clause, within the vP. However, the pattern is not uniquely determined by θ-position; transitivity of the predicate and person features of the DP also play a role. To wit, the above pattern is limited to first-/second-person DPs. Third-person DPs remain nominative with intransitive verbs,5 regardless of interpretation, whereas third-person DPs do bear ergative with a transitive verb.6 (3) a. bader dah” dapx-dalĩ child.NOM PVB undress-AOR ‘The child got undressed’
(Holisky 1987: 104)
b. as dah” japx- jail- n- as 1SG.ERG PVB undress-INTR-AOR-1SG.ERG ‘I got undressed’
(Holisky 1987: 105)
c. k’nat- ev bader dah” dapx-diẽ boy-ERG child.NOM PVB undress-TR.AOR.3 ‘The boy undressed the child’
(Holisky 1987: 104)
The transitivity of the verb is again a property determined low in the structure. The person features of the DP are first present low in the structure, although higher structural heads could also be sensitive to these features. In addition, lexical selection, a relationship established within the vP, also seems to play a role, in that there are predicates with a nominative-oblique case pattern, as well as predicates with an ergative-oblique case pattern. The ergative-oblique pattern is rare in the language, a “minor pattern” (Holisky and Gagua 1994, section 3.2.1.6), in contrast with the nominative-oblique pattern, which is found on a “large group” of verbs (Holisky and Gagua 1994, section 3.2.1.2). This contrast indicates that the presence of a nominative (rather than oblique) object is also a factor in ergative assignment. Compare the following.
5
Again, including verbs with an oblique object. Note that the contrast between (3b) versus (3c) is indeed due to the difference in person, not due to the difference in the status of the subject as a pronoun versus a full DP. Like full DPs, third-person pronouns (derived from demonstratives) do show ergative in transitives; hence in the following example the subject is the third singular ergative oqus rather than the third singular nominative o. 6
(i) oqus Jet: xen-ex Bexk’-in 3SG.ERG cow.NOM tree-CON tie-AOR.3SG ‘He tied the cow to the tree.’ (Holisky and Gagua 1994:[75a]) Thank you to an anonymous reviewer for raising this issue.
The locus of ergative case 139 (4) a. mastov mak qet-in enemy.NOM PVB attack-AOR ‘The enemy attacked us’
vai-n 1PL-DAT (Holisky and Gagua 1994:[76c])
b. tupliv h”eč’q’O son shoe.PL.ERG pinch 1SG.DAT ‘My shoes are pinching me’
(Holisky 1984: 188)
Lexical selection can also be found in intransitives like mušebadar “work” and gamarǰbadar “win, be victorious”, which obligatorily take ergative subjects even when the subject is third person. Holisky (1984: 189) attributes this exceptional pattern to them being borrowed from Georgian. Overall, we find that there is not a single deciding factor in the assignment of ergative case in Tsova-Tush, but rather a variety of factors play a role. These factors include the θ-position of the DP, the person features of the DP, the presence of an object, the case borne by the object, and lexical selection by the predicate; these factors are overwhelmingly clustered low in the clause, in the vP domain.
6.2.2 Nez Perce Nez Perce (Sahaptin: North Idaho) (Rude 1985; Woolford 1997; Deal 2010a, 2010b) illustrates a different cluster of low properties governing ergative case assignment. The basic case pattern is tripartite, with ergative -nim, accusative7 -ne, and zero nominative/elsewhere. These are illustrated in the following. (5) a. hi-pním-se 3SBJ-sleep-PFV ‘The cat is sleeping.’
pícpic. cat
b. ki-nm picpíc-nim pee-p-ú’ cu’yéem-ne this-ERG cat-ERG 3/3-eat-PROSP fish-ACC ‘This cat will eat the fish.’
(Deal 2010a: 77)
Ergative case fails to be assigned with pseudo noun incorporated objects (see Massam 2001 on pseudo noun incorporation, and Deal 2010b for discussion of the Nez Perce instance), as illustrated in (6).
7 It may be that this is better glossed as DAT. Note that in a double object construction, the goal must bear this case and the theme cannot. It is often referred to in the literature with the more neutral “objective.”
140 Julie Anne Legate (6) ’ipí hi-qn’íi-se 3SG 3SBJ-dig-PFV ‘He digs qeqíit roots.’
qeqíit. edible.root
(Crook 1999: 238)
The lack of ergative case in such constructions may have several potential sources. It may be that these are syntactically treated as intransitive; indeed Rude (1985) argues for such an analysis for Nez Perce. Also, the object is unmarked for case; since ergative subjects do not occur with unmarked objects, ergative could be tied to assignment of accusative case. The object also fails to trigger object agreement, thus, the presence of object agreement may be a crucial factor in ergative case assignment. Additional data demonstrate that more than simple intransitivity is at issue. Deal (2013), building on Rude (1985), discusses transitive clauses in which the object is a possessed DP. If the possessor is disjoint in reference from the subject, it bears accusative case and triggers object agreement; the subject bears ergative case. Deal (2013) analyses this as an obligatory possessor raising construction. If the possessor is bound by the subject, on the other hand, the possessor bears genitive case and fails to trigger object agreement. Like in the possessor raising construction, however, the possessed DP also does not bear accusative or trigger object agreement.8 In the absence of accusative case and object agreement, ergative case is not assigned. (7) a. Pit’iin’-imi paa-’yax̂-na’ny-∅-a ’ip-nek picpic girl-ERG 3SBJ-find-PR-PFV-REM.PST 3SG-ACC cat ‘The girli found his/herk/∗i cat.’ b. Pit’iin’i hi-’yaax̂-n-a ’ip-nimi picpic girl 3SBJ-find-PFV-REM.PST 3SG.GEN cat ‘The girli found heri/∗k cat.’ (Deal 2013: 413) Causatives provide potential evidence disambiguating whether lack of accusative or lack of object agreement is the crucial factor. In the causative of a transitive in Nez Perce, the causee does not bear ergative case. It is important to recognize that causees in ergative languages are only expected to bear ergative case if they are introduced into the structure in a vP identical to the vP that introduces agents; see for example Ippolito (2000) and Legate (2014b) for arguments that causees are rather introduced into the structure more like (high) applicative objects. Specifically to Nez Perce, provisionally assuming the causee to be introduced in the specifier of the right type of v, as sketched in (8), there is accusative case associated with this v, but no object agreement, and no ergative case.
8 This requires additional explanation; Rude (1986) and Deal (2010a) point out that a genitive possessor in the subject blocks ergative as well, yielding a clause with an unmarked subject and accusative object. Deal (2010a) notes that the possessor must be treated as closer to a higher probe than the containing DP in possessor raising constructions; she treats the lack of agreement for bound possessors (which do not raise) as an anaphor agreement effect (see Rizzi 1990a; Woolford 1999).
The locus of ergative case 141 (8)
vP
v'
Agent.ERG
v
vP
[uAGR] v'
Causee.ACC v
VP V
Theme.ACC
Thus, the structure exhibits one set of object agreement, associated with the v that introduces the agent and agrees with the causee, but two accusative objects: the causee and the theme. Consider the examples in (9), which exhibit both an accusative causee and an accusative theme; the ergative agents are pro-dropped. In (9a), the causee is plural and the theme is singular. The v that introduces the agent agrees with the causee, registering third-person plural agreement, and assigns it accusative case; the v that introduces the causee registers no agreement with the singular theme, but does assign it accusative case. In (9b), the causee is singular and the theme is plural. The v that introduces the agent agrees with the causee, registering third-person singular agreement, and assigns it accusative case; the v that introduces the causee registers no agreement with the plural theme, but does assign it accusative case. (9) a. ’e-nee-sepe-cukwe-n-e ha-’ayato-na Bessie-ne 3OBJ-PL.OBJ-CAUS-know-ASP-REM.PST PL-woman-ACC Bessie-ACC ‘I made the women know Bessie.’ ‘I introduced Bessie to the women.’ (Deal 2010a: 380) b. marsi- na ’e- sepee- twik- ce geyb Marcie-ACC 3OBJ-CAUS-accompany-IPFV.PRS Gabe kaa ceeki-ne and Jackie-ACC ‘I make Marcie accompany Gabe and Jackie.’ (Deal 2010a: 380) These constructions then provide potential evidence for object agreement rather than accusative case assignment as a determining factor in ergative case assignment in Nez Perce. The v that introduces the causee assigns accusative case but does not agree with the object, and hence the causee does not bear ergative case. The v that introduces the
142 Julie Anne Legate agent assigns accusative case and agrees with the object, hence the agent bears ergative case. On the assumption that we have been making (following Deal (2010a, 2010b) for Nez Perce, and tracing back to Chomsky (1995) more generally) that object agreement is associated with v, this property is again based low in the clause. The issue remains underdetermined, however, in that the position of causees in the structure must be clarified.9 Object agreement is not the only factor in determining ergative case assignment in Nez Perce, however; person is also relevant: ergative case is found only on third-person DPs. Deal (2016) uses the tests developed in Legate (2014a) to demonstrate that first- and second-person DPs are not assigned ergative case in Nez Perce. The Nez Perce data thus stands in contrast with the widespread pattern whereby ergative case is assigned to all DPs, but is only realized morphologically on a subset of DPs; see Legate (2014a) for details.10 The following examples illustrate the Nez Perce pattern. (10) a. ’Ip-ním pée-’pewi-se 3SG.-ERG 3/3-look.for-IPFV.PRS ‘She is looking for Mary.’
Méli-ne. Mary-ACC
b. ’Iin ’ipéwi- se Méli- ne. 1SG..NOM look.for-IPFV.PRS Mary-ACC ‘I am looking for Mary.’
(Deal 2016: [(2c)])
(Deal 2016: [(2a)])
In summary, the person features of the DP, and at least one of object agreement, accusative case, and the θ-position of the DP are primary determinants of whether ergative case is assigned in Nez Perce. Again, while the role of person is potentially ambiguous in height, the other factors are clearly low in the clause, associated with vP.
6.2.3 Warlpiri Warlpiri (Pama-Nyungan, South-West: Northern Territory, Australia) is often discussed in the literature, but the complex factors governing ergative case assignment are 9 Deal (2010a, 2010b) suggests an alternative explanation for the lack of ergative on causees, proposing that an ergative DP must enter an agreement relationship with T. In that this analysis is tantamount to a quirky case analysis of ergative, and given that ergative DPs do indeed undergo A-movement, raising to become the grammatical subject, as expected of quirky case-marked DPs, the analysis has plausibility. The necessity of agreement with T, however, cannot be established on the basis of causees alone, given that the lack of ergative on causees is subsumed under the requirement of object agreement for ergative assignment, and given that the θ-position of the causee may be distinct from that of more standard external arguments. 10 Tsova-Tush, discussed in the section 6.2.1, constitutes another language in which the person feature of the DP is a factor in syntactic case assignment, rather than in the morphological realization of case. Interestingly, in that instance it is the third-person DPs that fail to be assigned ergative rather than the first and second-person DPs.
The locus of ergative case 143 often glossed over. The presence of an object is indeed relevant. This is evidenced by the lack of a class of transitive verbs that have two absolutive arguments in their basic use (see for example, Swartz 1996), and by the fact that that many intransitive verbs take absolutive subjects.11 (11)
a. Mirni ka-lu yapa wangka-mi-lki thereabouts PRS.IPFV-3PL.SBJ person.ABS speak-NPST-now ‘There are people talking over there somewhere.’ b. Parnpa-ngka, malkarri-rla, ritual.type-LOC ceremonial.shield.design-LOC miirn-nyina-mi yapa panu work-sit-NPST person.ABS many.ABS ‘Many people work at the parnpa ceremony.’
ka-lu PRS.IPFV-3PL.SBJ
c. Ngawininyi ka wararrkurra- parnka. snake.species.ABS PRS.IPFV slither-run.NPST ‘The snake slithers away.’ However, an absolutive object is not required for ergative case assignment. For example, Warlpiri retains ergative with a dative unaffected object (see also for example, Djaru (Pama-Nyungan, South-West: Northern Territory, Australia) (Tsunoda 1981a), Gurindji (Pama-Nyungan, South-West: Northern Territory, Australia) (McConvell 1980)). (12)
a. Ngarrka-ngku ka marlu man-ERG PRS.IPFV kangaroo.ABS ‘The man is shooting the kangaroo.’
luwa-rni shoot-NPST
b. Ngarrka-ngku ka-rla-jinta marlu-ku luwa-rni man-ERG PRS.IPFV-3DAT.OBJ-3DAT.OBJ kangaroo-DAT shoot-NPST ‘The man is shooting at the kangaroo.’ (Hale et al. 1995: 1439)
Whether an object is required is less clear. It is difficult to unambiguously distinguish unergative verbs from transitive verbs in the language, given that the language has rampant pro-drop, given that third-person singular absolutive object agreement is null, and given that the language has productive applicative constructions (see for example, Legate 2003). It is worth noting that Swartz (1996) does not include for the language a class of verbs that are intransitive with an ergative subject. However, we do find interesting contrasts like the following, whereby yunparni ‘sing’ takes a subject in ergative case, even when apparently being used unergatively, whereas wirntimi ‘dance’12 takes an absolutive subject in the same context. 11 Warlpiri “absolutive” is nominative on S and accusative on O, see Legate (2008). Warlpiri data from Laughren et al. (2007) unless otherwise noted. 12 This is a particular style of dancing, typical of women.
144 Julie Anne Legate (13)
a. Yurapiti-rli ka yunpa-rni rabbit-ERG PRS.IPFV sing-NPST ‘The rabbit sings before sleep.’
jarda-kungarnti-rli. sleep-PREP.COMP-ERG
b. Wirnti-ja-lpa-lu karnta-patu-ju. dance-PST-PST.IPFV-3PL.SBJ women-PAUC.ABS-TOP ‘The women were dancing.’
Interestingly, Laughren et al. (2007) reports that there is dialectal variation on this point, whereby an ergative subject is used with wirntimi for some speakers from Lajamanu. It is clear that the presence of an object (absolutive or dative) is not the primary determinant of ergative case assignment, in that when an object is added to intransitives with an absolutive subject the subject remains absolutive. This is true whether the object is absolutive, (14a), or dative, as in (14b), which also illustrates that the dative passes objecthood tests in triggering object agreement and use of the object control complementizer -kurra (see Hale 1983; Simpson and Bresnan 1983). (14c) provides an additional illustration of an absolutive subject with an agreeing object, this time a high applicative dative object (see Simpson 1991 and Legate 2001 on dative high applicatives in Warlpiri). (14) a. Warlpiri ka- rna ngajulu wangka-mi. Warlpiri.ABS IPFV-1SG.SBJ 1SG.ABS speak-NPST ‘I am speaking Warlpiri.’ b. Karnta ka- rla wangka- mi ngarrka- ku woman.ABS PRS.IPFV-3DAT.OBJ speak-NPST man-DAT [jarnti-rninja-kurra](-ku). trim-INF-OBJ.COMP-(DAT) ‘The woman is speaking to the man trimming it.’ (Simpson and Bresnan, 1983: 54) c. Karnta ka-rla kurdu-ku parnka-mi. woman.ABS PRS.IPFV-3DAT.OBJ child-DAT run.NPST ‘The woman is running for the sake of the child.’ (Simpson, 1991: 381)
In Legate (2012a), I provide evidence that the θ-position of the DP also plays a role in whether ergative case is assigned. I examine Warlpiri verbs with two arguments, one of which receives dative case. For those verbs that are ergative-dative, the ergative is an external argument; examples are provided here for jinkami ‘support, help to walk’ and warrirni ‘seek.’ For those verbs that are absolutive-dative, in contrast, many of the absolutives are internal arguments, especially themes/ patients; examples are provided here for wiirr-parntarrimi ‘be a white film over’ and rdipimi ‘come upon.’
The locus of ergative case 145 (15)
a. Ngati- nyanu- rlu ka- rla kurdu- ku mother-ANAPH-ERG PRS.IPFV-3DAT.OBJ child-DAT nyanungu-nyangu-ku warru jinka-mi. 3-POSS-DAT around help.to.walk-NPST ‘The mother is propping up her child as he walks around.’ b. Wati- ngki ka- rla kurduku warri- rni. man-ERG PRS.IPFV-3DAT.OBJ child-DAT seek-NPST ‘The man is looking for the child.’
(16) a. Wiirr-parntarri-mi ka-rla kurdu-ku yurlkurrpa white-over.top-NPST PRES.IPFV-3DAT child-DAT soap.ABS palka-juku. body-still ‘The soap is still plastered over the child.’ b. Pardany- para- ja ngula- ji yangka chance.meeting-follow-PAST that.ABS-TOP like kuja- ka- rla yapa — wati marda, DECL.COMP- PRS.IPFV- 3DAT.OBJ person.ABS man.ABS maybe karnta marda, kurdu marda— rdipi- mi warna- ku woman.ABS maybe child.ABS maybe encounter- NPST snake- DAT marda, wardap i-ki marda, lungkarda- ku marda, maybe goanna- DAT maybe blue.tongued.lizard- DAT maybe ngurrpa marda, yangka marna- ngka- ku marda, ignorant.ABS maybe like grass- LOC- DAT maybe ngulya-ngka-ku marda. burrow-LOC-DAT maybe ‘Pardany-paraja is like when someone—a man or a woman or a child—comes across a snake or a goanna or a Blue Tongue lizard perhaps without knowing it was there—like in the grass or in a burrow.’ Experiencer subjects may have either ergative or absolutive case, depending on the predicate; examples are provided for pulka-pinyi ‘approve of, praise,’ which takes an ergative subject, and kapatimi ‘dislike,’ which takes an absolutive. (17)
a. Pulka- pi- nyi ngula- ji yangka approval-hit-NPST that.ABS-TOP like kuja- ka- rla- jinta ngati- nyanu- rlu DECL.COMP-PRS.IPFV-3DAT.OBJ-3DAT.OBJ mother-ANAPH-ERG manu kirda-nyanu-rlu kulu-parnta-ku, kurdu-nyanu-ku, or father-ANAPH-ERG anger-having-DAT child-ANAPH-DAT ngula yangka kuja- ka- jana yapa-kari pi-nyi that.ABS like DECL.COMP-PRS.IPFV-3PL.OBJ person-other hit-NPST
146 Julie Anne Legate kulu-parnta-rlu anger-having-ERG ‘Pulka-pinyi is when a mother or father gives approval to their child who fights, like when he fights and beats up other people’ b. Warrki-ki ka-rla kapati-mi yangka yapa. work-DAT PRS.IPFV-3DAT.OBJ dislike-NPST that.ABS person.ABS ‘That person doesn’t like work.’ The role of lexical selection is also apparent in that certain agents appear with absolutive case in the presence of a dative object, including the subjects of jaka-yirrarni ‘plan, plot’ and jurrurru-yarnkami ‘seize.’ (18)
a. Napanangka-rla jaka-yirra-rnu Napaljarri-ki. Kapu Napanangka.ABS-3DAT.OBJ plan-put-PST Napaljarri-DAT FUT.COMP nganta paka-rni kulu-ngku perhaps hit-NPST anger-ERG ‘Napanangka threatened Napaljarri that she would hit her in anger.’ b. Kulu ka- rla karnta fight.ABS PRS.IPFV-3DAT.OBJ woman.ABS jinta-kari-ki-rlangu-ku jurrurru-yarnka-mi watiya-ku one-other-DAT-for.example-DAT grabbing-grab-NPST wood-DAT karlangu-ku. digging.stick-DAT ‘In a fight a woman grabs hold of the other woman’s stick.’
In summary, we see that at least the presence of an object and the θ-position of the subject are relevant factors in the assignment of ergative case in Warlpiri. However, neither are determinative, and there is a significant role for lexical selection. These three factors are again low in the syntactic structure, within the vP, supporting an approach whereby ergative case assignment is determined within the vP.
6.2.4 Tshangla Tshangla (Tibeto-Burman: Bhutan) (Andvik 1999)13 contrasts the transitive subject marked with ergative -gi,14 with the intransitive subject and transitive object, which are morphologically unmarked (and unglossed) for case.15 13
In the Tshangla data, note that the morpheme glossed as a copula (COP) has several uses, including the imperfective; the morpheme glossed as a nominalizer (NMLZ) also has several uses, including marking past perfective in the finite affirmative; the stem extender (SE) is added to vowel-final verb roots. 14 The ergative also appears on instrumental and reason adjuncts; I leave this syncretism aside. 15 There is additional complexity involving information structure that I don’t consider here. Ergative is obligatory for focal transitive subjects, but optional for topical, and impossible for contrastively focused.
The locus of ergative case 147 (19)
a. Ji-gi shing 1SG-ERG tree ‘I cut the tree.’
chat-pa cut-NMLZ
b. Jang yi-pha 1SG sleep-NMLZ ‘I slept.’
(Andvik 1999: 200)
(Andvik 1999: 180)
Andvik (1999) examines the multiple factors involved in the appearance of ergative case, stating (1999: 193) “no single one of which is sufficient on its own to motivate agentive marking”. From the above examples we see that some notion of transitivity is relevant, but the details need to be determined. To begin, assignment of ergative is not dependent on an object that bears the unmarked case: verbs that select for locative/dative16 objects also take ergative subjects. For example, ‘to cheat,’ ‘to rebel against,’ ‘to bother,’ and ‘to scold’ fall into this class; examples follow. (20) a. Tsongpen-gi a-ha tem a-wa. merchant-ERG 1PL-LOC/DAT cheat do-NMLZ ‘The merchant cheated us.’
(Andvik 1999: 221)
b. Ro-ki ro-ka apa-ga ngolok a-na. 3-ERG 3-GEN father-LOC/DAT rebellion do-COP ‘He is rebelling against his father.’ (Andvik 1999: 222) c. Kuchi, ji- gi nan- ga trok- pa na. excuse 1SG-ERG 2SG-LOC/DAT bother-NMLZ PRT ‘Excuse me, I have bothered you.’ (Andvik 1999: 243) d. Ro-ki ja-ga brang-pa. 3-ERG 1SG-LOC/DAT scold-NMLZ ‘He scolded me.’
(Andvik 1999: 243)
Nor is the assignment of ergative dependent on a DP object. Verbs of cognition, including for example sele ‘to know,’ tsile ‘to reckon, consider,’ nale ‘to comply, agree,’ and verbs of utterance, including for example yekpe ‘to speak,’ and jime ‘to ask,’ take ergative whether used with a nominal complement, propositional complement, or a null complement. Two examples follow. (21)
a. Nyi nga-ba-ki “jang bu-i jang bu-i” yek-pa-la. PRT fish-PL-ERG 1SG take-HOR 1SG take-HOR speak-NMLZ-COP ‘And the fish said, “Take me! Take me!” ’ (Andvik 1999: 197)
Case morphology affected by information structure is found in non-ergative languages; see for example the well-discussed case of Japanese. 16 These are synthetic.
148 Julie Anne Legate b. Ji-gi ma-se-la. Jang shi-n nyok-nyi-sha ji-gi se-le. 1SG-ERG NEG-know-PRT 1SG die-SE receive-NF-PRT 1SG-ERG know-INF ‘I don’t know. Only if I had died would I know.’ (Andvik 1999: 196) The presence of a complement is relevant, though; predicates that normally appear as intransitive with an unmarked subject take ergative subjects when used with a complement (including null, as seen for the transitive in (21)). Andvik (1999: 215) characterizes this usage as “the action of the subject referent has consequences for another referent.” In the following examples, the same complex predicate (consisting of a light verb and noun), takes an unmarked subject with the unergative interpretation ‘pray,’ but an ergative subject with the transitive interpretation ‘entreat.’ (22) a. Nyi shepa phi-n chhum-deke, rokte sewu ta-phe PRT preach do- SE finish- NF 3PL prayer make- INF ren-pa-kap-nyi ... prepare-PTCP-with-NF ‘And when the preaching was finished, and the others were about to pray . . .’ (Andvik 1999: 221) b. Songo-ba-ki ro-ka sewu ta-pha-la. person-PL-ERG 3-LOC/DAT prayer make-N MLZ-C OP ‘The people entreated him.’ (Andvik 1999: 221) Another example involves the verb ‘walk,’ which takes an unmarked subject in its unergative use, but in the following takes an ergative subject on the interpretation ‘walk ahead of.’ (23) Apa-gi gum gum dang-nyi, jang tshin-ga lus chho-wa father-ERG ahead ahead walk-NF 1SG after-LOC leave stay-NMLZ ‘Father walking ahead, I was left behind.’ (Andvik 1999: 216) Similarly, an experiencer subject may appear with ergative case in the presence of a complement. Thus, in the following, a complex predicate (consisting of a light verb and a noun) appears with an unmarked subject with the meaning ‘be happy’ but an ergative subject with the meaning ‘be pleased with.’ (24) a. Ro kap-nyi chhas phi-nyi jang-ta shonang phi-wa. 3SG with-NF talk do-NF 1SG-PRT happy feel-NMLZ ‘Talking with him I feel happy.’ (Andvik 1999: 223) b. Nyi khaila onyen shi- deke, semchen thamche- ki ribong-ga namesame PRT tiger DEM die-NF animal all-ERG rabbit-LOC very shonang phe-nyi... happiness feel-NF ‘And after the tiger died, all of the animals were very pleased (with the rabbit).’ (Andvik 1999: 224)
The locus of ergative case 149 The presence of a complement is only one of the factors determining ergative case assignment. The θ-position of the DP is also relevant; thus, we find ergative only on thematic subjects, not two-argument unaccusatives. (25) a. Ama shi-n chhum-deke omchhang ata-ga unyu mother die-SE finish-NF another elder.brother-LOC/DAT DEM pruskin natsha-rang nyong-pa similar disease-EMPH receive-NMLZ ‘After mother died, the older brother also got the very same disease.’ (Andvik 1999: 212) b. Nyi unyu chhesung-gi not-dengai-la songo thamchen yi PRT DEM spirit-ERG harm-NF-PRT person all blood phros-nyi shile. vomit-NF die-INF ‘And if this spirit makes them sick, everyone will vomit blood and die.’ (Andvik 1999: 204) Furthermore, verbs that are normally unaccusative do appear with ergative subjects when used in a marked agentive context, again demonstrating the relevance of the θ-position of the DP. Compare the unaccusative use of ‘vomit’ in (25b), with the use with an ergative external argument in (26). (26) Shi-wa songo-gi bra songo-ga not-nyila, unyu die-NOM person-ERG other person-DAT/LOC make.sick-COND this shinang. ... Shinang tshebang- gi phros- pe. shinang shinang some-ERG vomit-NOM ‘If a dead person causes another person to become sick, this is “shinang”. ... Some shinangs vomit.’ (Andvik 1999: 204) Another example follows, in which the owner of a house is expecting a thief that night and thus is forcing himself to stay awake. (27) ro-ki onya binang-ga ma-yi-pha 3-ERG DEM night-LOC NEG-sleep-PTCP ‘on that night he will not sleep.’
(Andvik 1999: 205)
The following use of ergative on the subject of a normally unaccusative predicate Andvik (1999: 205) treats as a separate phenomenon, as “contrary to expectations of what is normal.” In that violating expectations may require a volitional act, and in that the point of this utterance seems to be that the DP has agentive control over the situation, it is likely that this type of example may be assimilated to the two previous. (28) Apa nan shuk chaka, nan-gi mar-be hang-rang mancha. father 2SG power COP 2SG-ERG be.sick.INF what-EMPH NEG.COP ‘Father, you are powerful; you are never going to get sick.’ (Andvik 1999: 205)
150 Julie Anne Legate Aspect is also relevant, and operates in the crosslinguistically expected direction: in the perfective, ergative on transitive subjects is obligatory, whereas in the imperfective, the ergative is optional. (29) a. Ji-gi/*Jang shing chat-pa. 1SG-ERG/*1SG tree cut-NMLZ ‘I cut the tree.’ b. Ji-gi/Jang shing chat-cha. 1SG-ERG/1SG tree cut-COP ‘I am cutting the tree.’
(Andvik 1999: 200)
(Andvik 1999: 200)
Here I note simply that aspect falls under our generalization that ergative assignment is affected by properties low in the clause, in that Aspect is in a selectional relationship with vP. However, see for example Laka (2006a), Mateo-Toledo (2008), Mateo Pedro (2009), Coon (2010a) for approaches whereby the imperfective functions as an intransitive matrix predicate; such approaches are compatible with the current discussion. In sum, ergative case assignment in Tshangla is determined by a cluster of factors that are low in the clausal structure, including at least presence of a complement, the θ- position of the DP, and aspectual marking.
6.2.5 Hindi In this section we consider another well- studied language, Hindi/ Urdu (Indo- Aryan: Pakistan, India) (Mahajan 1989; Mohanan 1994a; Butt and King 2004; Davison 2004a; among many others). The basic pattern in the perfective aspect is for the transitive subject to be marked ergative and the intransitive subject and transitive object to be unmarked.17 In other aspects, all of these core arguments are unmarked. (30) a. gaaḍii muḍii. vehicle turn.PFV ‘The vehicle turned.’
(Mohanan 1994a: 34) h
b. raam-ne darvaazaa k olaa Ram-ERG door open.PFV ‘Ram opened the door.’
(Mohanan 1994a: 8)
17 The unmarked case is unglossed; see Legate (2008) for arguments that it corresponds to nominative on the intransitive subject and accusative on the transitive object.
The locus of ergative case 151 c. raam darvaazaa kholegaa Ram door open.FUT ‘Ram will open the door.’
(Mohanan 1994a: 8)
The assignment of ergative is not dependent on the object bearing the unmarked case; dative objects are equally compatible with ergative subjects. (31)
raam-ne ravii-ko piiṭaa Ram-ERG Ravi-DAT beat.PFV ‘Ram beat Ravi.’
(Mohanan 1994a: 70)
Note that despite some inconsistency in the literature, ko on the object in Hindi/Urdu is indeed dative rather than accusative. Synchronically, ko is used in contexts that are unambiguously dative, including experiencer subjects and the indirect object in a double object construction. These uses are also historically prior, appearing in Old Urdu in 1200AD for the indirect object in a double object construction and the object of ‘seek’ (Butt and Ahmed 2011) (‘seek’ commonly appears with a dative object crosslinguistically, see for example the use in Warlpiri (section 6.2.3)), as well as for the goal of directed motion verbs (Butt, Ahmed, and Poudel 2008). Regarding the use of ko on objects in differential object marking (based on specificity/animacy), such marking is commonly dative crosslinguistically. See also Legate (2008) for arguments that Hindi/Urdu accusative is unmarked, morphologically syncretic with the nominative, and Bubenik (1998) for evidence that Old Indo-Aryan nominative and accusative cases became syncretic in Middle Indo-Aryan. While presence of a complement is relevant for ergative assignment, it is not determinative. A few intransitive verbs allow ergative subjects, with a corresponding interpretive difference, including ‘cough,’ ‘sneeze,’ ‘smile,’ ‘spit,’ ‘cry,’ ‘laugh,’ ‘sleep.’ (See for example, Tuite et al. 1985, Hook et al. 1987, Mohanan 1994a, Davison 1999 for discussion.) (32) a. (raam-ko acaanak šer dikhaa.) (Ram-DAT suddenly lion appear.PFV) (Ram suddenly saw a lion.) ‘He screamed.’ b. us-ne /*vah jaan buujkhkar cillaayaa. he-ERG he deliberately shout.PFV ‘He shouted deliberately.’
vah /*us-ne cillaayaa. he he-ERG scream.PFV
(Mohanan 1994a: 72)
In addition, there are lexical effects, whereby particular lexical verbs unexpectedly appear without ergative, most notably ‘bring’ (Mahajan 1989), but also ‘speak’ and optionally ‘understand,’ among others.
152 Julie Anne Legate (33) raam(*-ne) šiišaa laayaa Ram-ERG mirror bring.PFV ‘Ram brought the mirror.’
(Mohanan 1994a: 72)
Again, this cluster of factors is low in the clause, within vP. The relationship between ergative case assignment and the vP is strongly supported by Hindi light verb constructions (see for example Butt 1995; Mahajan 2012). In such constructions, the presence/absence of ergative case on the external argument is determined by the light verb. For example, when ‘bring’ as a lexical verb combines with the light verb ‘give,’ its subject does receive ergative case. (34) Kabir-ne vo kitaab laa dii Kabir-ERG that book bring give.PFV.F ‘Kabir brought that book.’
(Mahajan 2012: 208)
Conversely, when a lexical verb that normally takes an ergative subject combines with ‘bring’ as a light verb, its subject does not receive ergative case. The following illustrates with the lexical verb ‘cause to climb.’ (35) a. us-ne gaaRii pǝhaaRii-pǝr cǝRhaayii he-ERG car hill-on climb.CAUS.PFV.F ‘He took the car up the hill.’ (=He caused the car to climb the hill) b. * us-ne gaaRii pǝhaaRii-pǝr cǝRhaa laayii he-ERG car hill-on climb.CAUS bring.PFV.F ‘He took the car up the hill.’ (=He caused the car to climb the hill) c. vo gaaRii pǝhaaRii-pǝr cǝRhaa laayaa he car hill-on climb.CAUS bring.PFV.M ‘He took the car up the hill.’ (=He caused the car to climb the hill) (Mahajan 2012: 209) This is particularly clear evidence that the assignment of ergative case is determined low in the clause, within the vP. In summary, the assignment of ergative is dependent on perfective aspect, the presence of an object of the verb, the identity of the lexical verb (operating both ways, disallowing ergative with a transitive verb, and allowing ergative with an intransitive in a marked agentive interpretation), and the identity of the light verb. Again, this cluster of properties is low, centered around vP. To conclude this section, we have found substantial crosslinguistic variation in the distribution of ergative case, and multiple contributing factors within each language.
The locus of ergative case 153 However, in all examples the factors are low in the clause, centered around vP or the XPs in a selectional relationship with vP, including VP and AspectP.
6.3 High Ergative In this section, I discuss two languages for which the assignment of ergative has been described as dependent on properties high in the clausal structure. I demonstrate that for these two languages, at least, the descriptions should not lead us to posit a high source for ergative case. It may turn out that other more solid instances may be found of ergative dependent on a high source in the clause; if so, these would be a distinct phenomenon, and should be named differently in order to avoid confusion in the literature. “High ergative” may suffice.
6.3.1 Kurdish Past Tense Kurmanji Kurdish (Iranian: Turkey, Iran) (Bynon 1979; Payne 1980; Haig 1998; Thackston 2006) exhibits a pattern that may be initially described as ergative dependent on past tense. Note that the ergative is marked with a generalized oblique case while the nominative is unmarked.18 In the past, the intransitive subject is unmarked, the transitive subject is oblique, i.e. ergative, and the transitive object is unmarked. (36) a. ez ḑû-m 1SG.DIR go.PST-1SG ‘I went.’ b. min nan xwar 1SG.OBL bread eat.PST.3SG ‘I ate the bread.’
(Haig 1998: 157)
(Haig 1998: 160)
In the present, in contrast, the subject is unmarked, while the object is oblique, in essentially an accusative pattern. (37) ez nên di-xw-im 1SG.DIR bread.OBL DUR-eat.PRS-1SG ‘I am eating the bread.’
(Haig 1998: 160)
18 Case marking is neutralized when the noun is modified. “Nominative” here is used as a traditional term for the unmarked case. In pronouns, the distinction between the oblique and the unmarked is suppletive. Unmarked nouns are unglossed; unmarked pronouns are glossed as direct (DIR). See, for example, Dorleijin 1996 for discussion of dialect variation in case marking, including extension of the marked ACC of the present into the past, yielding a transitive OBL-OBL pattern.
154 Julie Anne Legate We should not, however, conclude that ergative is assigned by past tense T, and hence dependent on a projection high in the clausal structure. The notion of “past” that is relevant to ergative case assignment here is not clausal tense associated with TP. Instead, the “past” is an allomorph of the verb stem, which evolved from an Old Iranian perfect participle (see for example Payne 1980; Haig 2008). The past versus present allomorph of the stem appears inside negation, aspect and agreement morphology, confirming that it is indeed based low in the clause. (38) a. Goşt me ne-ti-xward. meat 1.PL.OBL NEG-DUR-eat.PST.3SG ‘Meat, we didn’t eat.’ b. pe-ǝt n-ā-le-ǝm to-2SG.DAT NEG-IPFV-tell.PRS-1SG.NOM a-ka-ā IPFV-do.PRS-3SG.NOM ‘I shall not tell you what Hiwa is doing.’
(Haig 1998: 160) Hiwa chi Hiwa what (Karimi 2010: 697)
Furthermore, use of the past stem does not necessarily correspond to a clausal past tense interpretation. The present perfective combines the past stem with the perfective aspect, thus showing a dissociation between the verbal allomorph and clausal tense. Importantly, it is the verbal allomorph that determines case: ergative is indeed assigned in the present perfective, despite the clausal present tense.19 (39) a. min nan xwar-iye 1SG.OBL bread eat.PST-PFV.3SG ‘I’ve eaten the bread.’
(Haig 1998: 159)
b. Min heta niha çar kitêb çapkir-ine. 1sg.OBL until now four books publish.PST-PFV.PL ‘Until now I have published four books.’ (Thackston 2006: 54) I conclude that the past allomorph of the verbal root is a factor in ergative assignment in Kurmanji Kurdish, not clausal past tense based in TP. The language thus confirms to the generalization that ergative is assigned based on properties low in the clausal structure.
6.3.2 Yukulta Irrealis Yukulta20 (Tangic: northwest Queensland, Australia) (McConvell 1981; Keen 1983) has been described as ergative dependent on the realis mood (see for example, Tsuonda 19
20
I have added glosses to the example from Thackston (2006). This language is also known as Gang(g)alidda.
The locus of ergative case 155 1981b). Lexical DPs in Yukulta have marked ergative and absolutive forms,21 pronouns show a single form for all of ergative, nominative, and accusative, while the clitic cluster has distinct agreement forms for each of ergative, nominative, and accusative. The following illustrate the basic pattern. (40) a. ṭir-iya -ka-nta snake-ERG -TR-PST.TR.R ‘The snake bit the boy.’
pa:tya maṇṭuwara bite.IND boy.ABS (Keen 1983: 205)
b. ṭir-a -ŋka paritya waḷmat̪-i kamar-i snake-ABS -PRS.INTR.3SG crawl.IND on.top-LOC stone-LOC ‘The snake is crawling over the stone.’ (Keen 1983: 206) A nominative-dative case frame is used for two-argument predicates that are not canonically transitive, including predicates with an experiencer subject and predicates with a goal object. (41) a. pulwitya -ka-ti ṭir-inytya feel.fright.IND -1SG.NOM-PRS.INTR.R snake-DAT ‘I’m frightened of snakes.’ (Keen 1983: 206) b. tyanitya -ka-ti look.IND -1SG.NOM-PRS.INTR.R ‘I’m looking for the boy.’
maṇṭuwara-n̪ t ̪a boy-DAT
(Keen 1983: 223)
The clitic cluster registers agreement, as well as information regarding transitivity, tense, and mood, under complex interactions. The past forms in the clitic cluster show a realis/ irrealis distinction, but importantly, this distinction does not affect the assignment of ergative case. In the following, the first example is past realis and the second past irrealis; both have an ergative first-person subject indicated in the clitic cluster and an absolutive object. (42) a. ṭat̪in-ta -ŋa-nta warunta kuritya there-ABS 1SG.ERG-PST.TR.R goanna.ABS see.IND ‘I saw a goanna over there.’ (Keen 1983: 202) b. walira-ŋa-nti NEG-1SG.ERG-PST.TR.IRR ‘I didn’t find your spear.’
kapa ŋumpanta miyaḷṭa find your.ABS spear.ABS (Keen 1983: 235)
This indicates that ergative assignment is not dependent on realis mood. 21 Ergative is syncretic with locative. Absolutive is used as a traditional term, referring to the morphological realization of nominative on the intransitive subject and accusative on the transitive object; see Legate (2008, 2014a) for discussion.
156 Julie Anne Legate Instead, the intransitive nominative-dative frame is used in two contexts signaling reduced transitivity, in the sense of an unaffected/goal object. The first context is negative non-past clauses. In the first example of the following pair, the affirmative present appears with an ergative-absolutive case frame; note that the object is a patient affected by the event. In the second example, the negative present appears with a nominative- dative case frame; note that the object cannot be affected by the non-occurring event. The clitic cluster registers the first as a transitive verb and the second as an intransitive verb, but shows no realis versus irrealis distinction. (43) a. ṭaŋka-ya -ka-ri ŋawu palat̪a man-ERG -TR-PRS.TR.R dog.ABS hit.IND ‘The man is hitting the dog.’ b. walira -ŋka ṭaŋka-ṛa NEG -PRS.INTR.3sg man-ABS ‘The man isn’t hitting the dog.’
(Keen 1983: 206)
ŋawu-n̪ t a̪ palat̪a dog-DAT hit.IND (Keen 1983: 206)
The second context of reduced transitivity marked by use of the nominative-dative case frame is desideratives. The use of nominative-dative in the desiderative is optional, and “can suggest that the expectation of an action being completed or experienced is reduced due to outside factors” (Keen 1983: 239). Note that the desiderative itself is low in the clausal structure, appearing as a form of the lexical verb, below the higher inflectional information registered on the auxiliary. The following examples illustrate the desiderative nominative-dative, in contrast with the ergative-absolutive. Again, the distinction between the two is registered in the auxiliary as a difference in transitivity, not mood; both appear in the realis. (44) a. kaṇata -ka-ti cook.DES 1SG.NOM-PRS.INTR.R ‘I’d like to cook some tucker.’ b. wuḷanta -ŋa-ri food.ABS 1SG.ERG-PRS.TR.R ‘I’m cooking tucker.’
wuḷan-inytya food-DAT
kaṇatya cook.IND
(Keen 1983: 221)
(Keen 1983: 221)
As an aside, note that this case frame is also employed when the object outranks the subject, according to a hierarchy whereby first-person nonsingular pronouns outrank first-person singular and second-person pronouns, which in turn outrank third-person pronouns and nominals. The following illustrates; note that the third-person subject ‘mosquito’ is in the nominative/absolutive, and the clitic cluster marks the clause as intransitive realis, and the first-person agreement clitic is the oblique form as triggered by a dative.
The locus of ergative case 157 (45) kuŋul-ta -tu̪ -yiŋka mosquito-ABS 1SG.OBL-PAST.INTR.R ‘A mosquito bit me.’
pa:tya bite.IND
(Keen 1983: 234)
To summarize, Yukulta ergative is not assigned by realis mood. Instead, ergative is assigned to the thematic subject of transitive verbs. In selected contexts of reduced transitivity, and in contexts in which the thematic object outranks the thematic subject, the intransitive nominative-dative case frame is used instead. This case frame is otherwise used for two-argument predicates that are not canonically transitive, including experiencer subject predicates, and predicates with a goal object. The factors of transitivity, the θ-role of the subject and object, and the case borne by the object, are all low in the clause. The language therefore is in fact consistent with the generalization that ergative assignment is dependent on factors low in the clause, within the vP.
6.4 Conclusion This chapter has had modest goals: to demonstrate that the assignment of ergative case is multifaceted, both within and across languages, and to demonstrate that ergative case in a range of unrelated languages is assigned based on properties low in the clause, centered around vP. We reexamined two languages, Kurmanji Kurdish and Yukulta, that have been described as exhibiting ergative dependent on factors high in the clausal structure, past tense, and realis mood respectively. We discovered that ergative case assignment in these languages in fact is not dependent on tense and mood, but rather on properties that are determined low in the vP: an allomorph of the lexical verb determined within vP, the θ-role borne by the subject and object, the case of the object, the desiderative form of the lexical verb, and the relative ranking of the person features of the subject and object. We leave open whether true “high ergative” languages may be found, that is languages in which assignment of ergative is dependent on factors high in the clausal structure, in the TP/CP domain. Note that only languages in which ergative can be clearly differentiated from nominative assigned by TP/CP would the label “high ergative” be appropriate for that case; otherwise, the case would simply be nominative. In addition to Kurmanji Kurdish and Yukulta, we examined five typologically disparate languages in which the ergative is assigned based on a variety of factors centered around vP: Tsova-Tush, Nez Perce, Warlpiri, Tshangla, and Hindi. Many of the factors identified may fall under the notion of transitivity broadly conceived, including the presence of a complement, the assignment of accusative case, the presence of object agreement, and the thematic interpretation of the subject and the object. Other
158 Julie Anne Legate low factors identified include the identity of the lexical predicate, the identity of the light verb, and the clausal aspect. These properties are clustered around vP, and are independent of higher projections in the TP/CP domain. I conclude that for a wide range of “low ergative” languages, the locus of ergative case is vP.
Acknowledgements Thank you to the editors, Jessica Coon, Diane Massam, and Lisa Travis, to an anonymous reviewer, to the audience at the 2013 LSA Annual Meeting, and to all those who have discussed ergativity with me over the years.
Abbreviations Abbreviations in glosses follow the Leipzig conventions, with the following additions: ANAPH = anaphoric, AOR = aorist, ASP = aspect, CON = contact, DES = desiderative, DIR = direct, EMPH = emphatic, HOR = hortative, NF = nonfinal verbal suffix, PAUC = paucal, PR = possessor raising, PREP = preparatory, PROSP = prospective, PRT = particle, PVB = preverb, R = realis, REM = remote, SE = stem extender.
Chapter 7
Ergative n e e d n ot spl it : An expl orat i on into the T otalE rg hy p oth e si s Itziar Laka
7.1 Introduction: The TotalErg Hypothesis Ergativity has received significant attention during the last decades in generative grammar; the number of languages and phenomena under scrutiny has increased, and as a consequence our knowledge about the properties and the range of variation in the ergative class has deepened considerably. As a result of this intensive focus, there has been an increasing trend toward convergence in the hypotheses as to what constitutes the core of ergativity. This convergence involves two different but related hypotheses, the sum of which I will name the TotalErg hypothesis: (1) The TotalErg hypothesis: (a) Ergative case is inherent. (b) Ergativity does not split. The inherent ergative hypothesis has been increasingly gaining empirical support, and has been defended for a large variety of ergative languages (Levin 1983; Mahajan 1989; Johns 1992; Oyharçabal 1992; Woolford 1997, 2001, 2006; Massam 1998, 2006; Holmer 1999; Legate 2002, 2008; Aldridge 2004, 2008a; Stepanov 2004; Anand and Nevins 2006; Laka 2006a, 2006b; Wiltschko 2008; among others, see Baker and Bobaljik (Chapter 5, this volume), Legate (Chapter 6, this volume), and Sheenan (Chapter 3,
160 Itziar Laka this volume) for an overview). Aldridge (2008a: 987), following Legate (2002, 2006), claims that ergative morphology is in fact this very property: “Morphological ergativity is defined by the uniform assignment of inherent case to the subject by transitive v, as proposed by Legate (2002, 2008).” One widespread implementation of this generalization that I will assume in this chapter is to say that ergative case is associated to the specifier of (a subset of) little vs (see Sheenan, Chapter 3, this volume); other mechanisms have also been proposed in the literature (see for instance the review by Polinsky and Preminger 2014). The inherent ergative hypothesis dissociates ergative case marking from structural Case licensing by Tense; ergative case is thus independent of Tense/ finiteness, unlike nominative case. The hypothesis that ergative case is inherent predicts that internal arguments cannot bear ergative case, and therefore that raising to ergative cannot exist, a claim originally made by Marantz (1984). To my knowledge, the earliest proposal that ergative case is inherent was put forth by Levin (1983) in order to provide an account of ergativity in Basque. Working within the Government and Biding Theory, Levin (1983) observed there was no evidence for a dissociation between theta roles and case morphology in Basque, the type of dissociation commonly found in nominative languages that justified the proposal of Case Theory as independent from Theta Theory (Chomsky 1981). The implementation proposed by Levin (1983) involved case assignment at D-structure, the level at which thematic relations were established in that model. Levin’s original proposal was not followed by subsequent generative studies on Basque in the 1980s and 1990s, probably because the assignment of case at D-structure represented too radical a departure from the standard GB claim that case was an S-structure licensing mechanism where Tense was crucially involved. Ortiz de Urbina (1989), for instance, argued that case in Basque was structural, and in particular that ergative was assigned to the specifier of Inflection by Tense. Many subsequent accounts have pursued the structural case hypothesis with variations (see, among many others, Laka 1993b; Fernández 1997; more recently see Rezac et al. 2014, and Berro and Etxepare, Chapter 32, this volume). The inherent ergative hypothesis for Basque has been defended by a minority of works in different forms, as an instance of lexical case by Oyharçabal (1992), who combines it with a structural licensing as well, or as inherent case related to AspP/vP (Holmer 1999; Laka 2006b). A second converging line of research on ergativity seeks to show that split ergativity is a misnomer (Laka 2006a; Coon 2010a, 2013a, 2013b; Coon and Preminger, Chapter 10, this volume). Under this hypothesis, linguistic phenomena labeled splits, suggesting a change from an ergative pattern to a nominative one, follow naturally given the grammatical properties of the languages under study, within the boundaries of an ergative system as understood here. Ultimately, this approach should provide a parsimonious understanding of ergativity where apparent changes in case morphology are not due to pockets of lexical or grammatical exceptions, but rather, they result from unexceptional processes acting across different syntactic structures. As Berro and Etxepare discuss in detail and with great nuance in this volume (Chapter 32), and originally pointed out by Levin (1983), ergative marking in Basque is sensitive to the unaccusative/unergative contrast. This type of ergative grammars have
Ergative need not split 161 been argued to instantiate a variety of split ergativity triggered by the lexical semantics of the predicates involved, and this is why they have been called Split-S languages by Dixon (1979, 1994), for instance. The account defended in this chapter claims that Split-S languages like Basque are in fact split-less: only external arguments carry ergative case, and that suffices to account for the contrast without any need to resort to claiming that the case pattern splits from one mode (ergative) to another (nominative). This, in turn, entails that case morphology in grammars like Basque is determined within the vP domain, because differences within the vP domain entail differences in case marking. As Sheenan (Chapter 3, this volume) and Berro and Etxepare (Chapter 32, this volume) discuss at length, languages and language varieties make different cutting points as to what subset of arguments are treated as external, that is, as to what subset of little vs assign ergative case; in particular, the set of arguments need not be restricted to those entailing causation and can include experiencers and holders as well. Another conclusion to be drawn at the end of the chapter is that agreement-based evidence is not necessarily valid to argue for case-related phenomena. I will claim that case and agreement are not always morphological reflexes of one syntactic operation; in particular, if ergative is inherent and vP related, as several authors argue in this volume, it is doubtful that agreement data can bear directly on discussions on case and ergativity. This is particularly relevant in a language like Basque, where agreement morphology is only licensed in finite configurations, but case morphology is insensitive to finiteness. Therefore, I will argue, evidence based on agreement facts does not necessarily hinge on case. This chapter presents an account of behar ‘need’ and associated syntactic configurations and predicates in Basque under the TotalErg hypothesis, which I argue provides a parsimonious account of the phenomena at stake. The interest of exploring this area of Basque syntax is due to the fact that it has become the focus of the debate between structural vs. inherent approaches to Basque ergativity, given recent claims by Rezac et al. (2014) that the behavior of this predicate provides crucial evidence in favor of a structural, T-dependent account of ergativity in this language. I will therefore consider the syntax of behar ‘need’ in Basque and argue that the changes in case assignment related to sentences headed by behar can be accounted for within the TotalErg hypothesis, without resort to a structurally assigned ergative case. In fact, I will argue that behar ‘need’ and similarly behaving predicates in Basque offer more empirical evidence supporting (a) and (b), thus converging with several other cross-linguistic accounts of ergativity. The chapter is structured as follows: in section 7.2, I briefly introduce the central phenomenon, seemingly involving a change in the case assigned to the subject of clauses when behar ‘need’ is introduced, and lay out the basics of the explanation of this contrast to be defended. In section 7.3, I discuss Etxepare and Uribe-Etxebarria (2012) and Harves and Kayne (2012), the two accounts upon which I base the derivation of the apparent split provoked by need. Although the accounts differ in certain details, the common ground they share suffices for the derivation I propose. In section 7.4, I discuss the categorical status of behar, and I also discuss the proposal in Rezac et al. (2014) that
162 Itziar Laka the presence of behar does indeed provoke a split and change the case assigning pattern of the clause. Finally, in section 7.5 I discuss the role that agreement-based evidence has on discussions of morphological case; I argue that agreement and case are dissociated though related phenomena, and hence agreement is not direct evidence for case. Finally, in section 7.6 I sum up the proposal and its consequences for the TotalErg hypothesis.
7.2 Ergative Need in a Nutshell The central grammatical phenomenon under discussion in this chapter is illustrated by the pair of sentences in (2):1 (2) a. zu Bilbo-ra joan zara you Bilbao-to gone 2sg.be ‘You have gone to Bilbao.’ b. zu-k Bilbo-ra joan behar duzu you-erg Bilbao-to gone need have.2sg ‘You need to go to Bilbao.’ The sentence in (2a) is headed by the unaccusative verb joan ‘go’ and the DP argument is case-marked absolutive (zero morphology). The introduction of behar ‘need’ in (2b) appears to change the case borne by the DP, which is now marked ergative (morpheme k). I will argue that the differences in case marking in (2) are the necessary consequence of the syntactic structure of each sentence, in a way very similar to what happens in the progressive, as discussed in Laka (2006a): while (2a) is a monoclausal structure, (2b) is biclausal, and the main clause predicate introduces its own external argument. To this end, I will follow the accounts in Etxepare and Uribe-Etxebarria (2012) and Harves and Kayne (2012), who agree in claiming that the noun behar ‘need’ in sentence (2b) introduces an external argument of its own. If this is so, and if ergative case is inherently assigned to external arguments in Basque (Levin 1983; Oyharçabal 1992; Holmer 1999; Laka 2006b), then the external argument introduced by behar necessarily carries ergative case, and the contrast between (2a) and (2b) follows.2
1 The contrast illustrated in (2) and discussed throughout the chapter does not hold for behar in western varieties, although it does for other predicates, like nahi ‘want.’ Thus, in western Basque, both (2a) and (2b) have absolutive subjects and intransitive auxiliaries. Western Basque appears to have grammaticalized behar as an element in the functional structure of the clause, in a pattern reminiscent of the grammaticalized ari progressive discussed in Laka (2006a, section 4.1) that lacked the case-changing phenomena generally induced by the progressive. 2 For the purposes of exposition, I focus on behar ‘need’ but it must be noted that nahi ‘want’ and a cohort of other predicates have the same nature and behavior. See Berro and Etxepare (this volume, section 32.5.1), for a fuller list and a description of the semantic classes they belong to.
Ergative need not split 163 If (2a) is monoclausal but (2b) is biclausal, then the absolutive DP in (2a) is the theme argument of joan ‘go,’ but the ergative DP in (2b) is the external argument introduced by behar ‘need.’ The key structural difference behind (2a, b) is schematized in (3a, b) respectively: (3) a. [theme go] b. [ext.arg.i need [themei go]] Given the structural differences in (2)–(3), it must be concluded that there is no split ergativity at play, no dissociation between semantic class and case morphology, and no change in the case assignment pattern from an ergative to a nominative one or vice- versa: the pronoun zu ‘you’ in (2b) carries ergative marking (-k) because it is an external argument, and it bears (zero) absolutive case in (2a) because it is an internal argument. The relation between the external and internal arguments in (2b) is one of control, as illustrated by the co-indexing in (3b), but it is possible to have referentially autonomous arguments in each clause (see later in the discussion, in e xample 11c). In sum, the general pattern behind the contrast between (2a) and (2b) is the same that has been attested cross-linguistically in so-called TAM Splits, which have been derived by showing that the seemingly split structures involve different complementation configurations which are at the base of the morphological changes observed (Laka 2006a; Coon 2010a, 2013a, 2013b; Coon and Preminger, Chapter 10, this volume).
7.3 Two Ways to Derive Transitive Need from a Noun I will now discuss in greater detail the account in Etxepare and Uribe-Etxebarria (2012) and compare it to that of Harves and Kayne (2012), showing that both accounts provide the necessary elements to derive the contrast in (2) within the tenets of TotalErg. Etxepare and Uribe-Etxebarria (2012) put forth an account of behar and associated syntactic configurations under the hypothesis that there is only one lexical entry for behar: it is always a noun. Since I will follow the basic structure provided by their account and focus on its consequences for the distribution and nature of morphological case in Basque and ergative morphology at large, I will lay it out and compare it to the cross-linguistic account of the syntax of transitive need defended by Harves and Kayne (2012),3 which also shares the initial hypothesis that transitive need originates as a noun. As we will see, the main difference between these two approaches involves the behavior of this noun during the derivation: whereas in Harves and Kayne (2012) the noun is an argument and incorporates into the verb have, in Etxepare and Uribe-Etxebarria (2012) it is a predicate and it does not incorporate. This difference, in turn, does not directly 3
Earlier versions of these proposals are found in Etxepare & Uribe-Etxebarria (2010) and in Harves (2008).
164 Itziar Laka bear on the discussion in this chapter, as we will see, because both accounts correctly predict that constructions with behar will necessarily yield ergative case-marked external arguments given the TotalErg hypothesis. Etxepare and Uribe-Etxebarria (2012) specifically discuss the syntactic structure associated with behar ‘need’ in sentences like (2a, b); they argue that behar is a noun acting as a predicate in a small clause [SC [DP/vP] behar] where it can combine either with a DP constituent (see example 11a for an illustration) or a vP constituent (see example 11b for an illustration). In this analysis, the ‘needer,’ which the authors conceive as an experiencer, and Berro and Etxepare (Chapter 32, this volume) categorize as a holder, is an external argument introduced in the specifier of an applicative head P that combines with the small clause, [PP experiencer P [SC]] and introduces the experiencer of the need, that is, the external argument, in its specifier position. (4) vP v
PP DPEXP
P' SC
DP/vP
be P
N
[HAVE be + appl]
appl
behar This PP, in turn, combines with the copula yielding [vP be [PP]]. The incorporation of the applicative P onto be yields the verb have (following Freeze 1992; Kayne 1993). Hence, (2a, b) are transitive due to the incorporation of P, the head that introduces the experiencer argument in its specifier (see also the structural descriptions in 5c, d). The account in Etxepare and Uribe-Etxebarria (2012) shares several basic features with the one in Harves and Kayne (2012), who put forth a cross-linguistic account of transitive need and relate it to auxiliary selection: languages with transitive need are necessarily languages with transitive have. According to Harves and Kayne (2012), transitive need emerges from the incorporation of the noun need onto the verb have: [VP have+needi [NP ei [DP]]. As the authors note, this derivation is compatible with the decompositional approach to have resulting from incorporation of P onto be (Freeze 1992 and Kayne 1993). Harves and Kayne (2012) explicitly include Basque within the group of languages where noun incorporation onto have yields transitive need, though they do not enter into the specifics of its syntax. Both accounts take as their starting assumption that behar ‘need’ is a noun; the difference between the two proposals lies in the syntactic configuration behar enters into, which in turn bears on whether this element incorporates into have or not. In Harves and Kayne (2012), transitive need results from noun incorporation (Baker 1988), and the constituent where need and its internal argument combine is a NP (see
Ergative need not split 165 the structural description in 5a). The incorporated noun does not require case, and the accusative case available for assignment from have can license the complement DP of the incorporated noun (5b). (5) a. [VP have [NP need [DP]] b. [VP have+needi [NP ei [DP]] c. be [PP experiencer P [SC[DP/vP] behar]] d. [[havebe+Pi] [PP experiencer ei [SC[DP/vP] behar]] In Etxepare and Uribe-Etxebarria (2012), behar is also a noun, but it acts as a predicate in a small clause. Therefore, it is not an argument and it does not require case (4), (5c), which in turn makes noun incorporation unnecessary (5d). The common ground shared by Etxepare and Uribe-Etxebarria (2012) and Harves and Kayne (2012) suffices to account for the transitive behavior of behar and its ergative marked external argument within the boundaries of TotalErg and without appeal to any exceptional or language-particular grammatical mechanism. Both accounts converge in arguing that the predicate structure of behar/need necessarily involves the introduction of an external argument independently of the structure of the complement (in the case of Harves and Kayne 2012) or sister constituent (in the case of Etxepare and Uribe-Etxebarria 2012) of the argument (in the case of Harves and Kayne 2012) or predicate (in the case of Etxepare and Uribe-Etxebarria 2012) of behar/ need. We compare the structural descriptions of a sentence like (2b) as argued for by each account: (6) a. Etxepare and Uribe-Etxebarria (2012)
b. Harves and Kayne (2012)
vP PP P'
DP zuk
SC
vP [PRO Bilbora joan]
vP v
DP
HAVE be + appl
zuk
P N
tappl
v' v
NP
vP [PRO Bilbora joan]
N
HAVE
behar
behar
As can be seen in the structural representations in (6), both accounts provide a specifier position within the vP domain where the external argument of behar/need is introduced. Given the inherent ergative hypothesis, that configuration requires the assignment of ergative case. Therefore, the syntax of behar as conceived by these accounts is compatible with the hypothesis that ergative case is inherently assigned to external arguments at the vP level, in accordance with the TotalErg hypothesis.
166 Itziar Laka
7.4 Is Basque Behar a Noun or a Raising Modal? I will now discuss the categorical status of behar, which has been subject to discussion in the recent literature. Two main hypotheses have been defended regarding the syntactic status of this element: (a) it is a lexical category, either a noun or a verb; (b) it is a functional category, a raising modal belonging in the inflectional structure of the sentence. Each hypothesis signals a different departure point and yields very different accounts; this is why it is important to consider what independent evidence can be found in support of one or the other. I will argue there is substantial independent evidence for a noun behar in Basque (as hypothesized by Etxepare and Uribe-Etxebarria 2012), and that the evidence in favor of the view that there is also a raising modal behar (as hypothesized by Rezac et al. 2014) is weaker. The more detailed traditional descriptions of Basque consider behar to be a lexical element, either a noun or a verb. De Rijk (2007: section 14.5) and Michelena (1990) extensively discuss the similarities and differences between behar the noun and behar the transitive verb, and the different types of constructions they can enter into. The generative literature on behar, for the most part, agrees with traditional descriptions of the language, and takes it to be a lexical category, a noun in some constructions and a verb in others, akin to English need. Proposals within generative grammar, like Goenaga (1985), and Ormazabal (1991) provide restructuring accounts of biclausal behar structures like (2b), and start from the assumption that in those structures behar is a verb. Goenaga (2006) discusses the categorical status of behar and concludes that it is underdetermined between a noun and an adjective. Ortiz de Urbina (2003: 300) refers to behar as a “semi auxiliary” verb, because “in some of their usages, seem to be ‘transparent’ with respect to their subordinate clausal complements, so that clause union effects may be perceived.” Haddican (2005) argues that behar is a ‘quasi-functional’ verb in the sense of Cardinaletti and Shlonsky (2004): it is a verb because it drives the selection of the auxiliary, but it is like a functional element because it is transparent to agreement with embedded arguments (clitic climbing). When considering the full array of constructions behar can enter into, some accounts have to assume that there is more than one behar in the mental lexicon of Basque speakers, each belonging to a different syntactic category. I will first consider the case of behar as a noun. There is agreement in the literature regarding the existence of a noun behar ‘need’ in the Basque lexicon (Michelena, 1990). As a common noun, behar can head argument DPs and behave as a nominal argument, as shown in (7):4
4
The change from a simple r in behar to a double rr in beharra is the orthographic reflex of the fact that in word final position the trill (grapheme rr) is neutralized (grapheme r).
Ergative need not split 167 (7) a. zu-k [DP[NP[PP liburu bat-en] beharr]-a] duzu you-erg book one-gen need-det have.2sg ‘You have the need of one/a book.’ b. zu-ki [DP[NP[PP PROi liburu bat irakur-tze-ko] beharr]-a] duzu you-ergi PROi book one reed-nom-gen need-det have.2sg ‘You have the need to read one/a book.’ The examples (7a) and (7b) only differ in the type of complement the noun behar ‘need’ takes. In (7a), the complement of behar is a genitive PP containing the DP liburu bat ‘one/a book’; in (7b) the complement is also a genitive PP containing a nominalized nonfinite clause liburu bat irakurtzeko ‘to read one/a book.’ As shown in (8), these DP arguments headed by the noun behar ‘need’ behave like any other DP argument; they carry a determiner (-a in the examples), and they bear ergative (8a, b), or dative case (8c, d) when required: (8) a. [DP[NP[PP liburu bat-en] beharr]-a]-k ez du gure ikerketa geldituko book one-gen need-det-erg not have.3SG our research.Det stop.asp ‘The need of one/a book will not stop our research.’ b. [DP[NP[PP PROarb liburua irakur-tze-ko] beharr]-a]-k ez du book.det read-nom-gen need-det-erg not have.3SG gure ikerketa geldituko our research.det stop.asp ‘The need to read one/a book will not stop our research.’ c. egoera hau [DP[NP[PP liburu bat-en] beharr]-a]-ri zor zaio situation this book one-gen need-det-dat owe be.3sg.3sg ‘This situation is due [Lit: owes] to the need of a book.’ d. egoera hau [DP[NP[PP PROarb liburu bat irakur-tze-ko] beharr]-a]-ri zor zaio situation this book one read-nom-gen need-det-dat owe be.3sg.3sg ‘This situation is due [Lit: owes] to the need to read a book.’
Michelena (1990), Goenaga (2006), De Rijk (2007), and Etxepare and Uribe-Etxebarria (2012) among others, offer a wide variety of examples and evidence supporting the existence of a noun behar, including instances like (9a), where behar combines with a locative postposition, (9b) where it is modified by an adjective, or (9c) where it bears the partitive determiner, as shown in these examples: (9) a. ni [PP zu-re laguntza-ren beharr-ean] nago I.abs you-gen help-gen need-in be/loc.1sg ‘I am in need of your help.’ b. ni-k [DP zu-re laguntza-ren behar gorria] dut I-erg you- gen help-gen need red have.1sg ‘Lit: I have a red need of your help.’ ‘I am in dire need of your help.’
168 Itziar Laka c. ni-k ez dut [DP zu-re laguntza-ren beharr-ik] I-erg not have.1sg you- gen help- gen need- part ‘I do not have any need of your help.’ I therefore take it as an uncontroversial fact that there is a noun behar ‘need’ in the Basque lexicon. This noun can take either DPs or non-finite clauses as complements, as shown in (7), (8), and (9). A full account of behar that postulates only one lexical entry is, all other things equal, preferable to an alternative account postulating various homophonous lexical entries. This is what Etxepare and Uribe-Etxebarria (2012) set out to do: they provide a full account of behar and associated syntactic structures by postulating a single lexical entry beharN. In the context of advocating for a structural assignment of ergative case in Basque, Rezac et al. (2014) discuss cases like (2b), which they refer to as the ‘INF+behar construction.’ Their hypothesis is that, in this particular construction, behar is a raising modal akin to English must. They follow the account in Wurmbrand (1999) who argued that modal verbs like English must are raising predicates involving syntactic movement, and not control structures. Hence, according to Rezac et al. (2014), the contrast in (2) involves the same thematic array of arguments: the unaccusative verb joan ‘go’ is the only thematic predicate in both (2a) and (2b), and the argument zu ‘you’ is a theme in both sentences, as shown in the derivations (10a) and (10b) respectively (adapted from Rezac et al. 2014: 1291, example 22): (10) a. [TP [DP zui] … [vP ti [PPBilbora] [vjoan] … zara] b. [TP [DP zuki] … [[vP ti [PPBilbora] [vjoan] …] behar] duzu] The presence of the modal element behar in (10b) triggers a change in the case assigned to the raised argument, which is now assigned ergative case by Tense. According to Rezac et al. (2014), this account of the contrast in (2) involving raising to ergative provides crucial evidence that ergative case is structural (Tense-dependent) and not inherent (vP-dependent) in Basque. I will argue that this raising account requires a more complex lexicon with different homophonous entries and the deployment of language-particular or exceptional mechanisms that burden our theory of grammar, and therefore, that the proposal in Etxepare and Uribe-Etxebarria (2012), combined with the TotalErg hypothesis, is both empirically and theoretically preferable. Rezac et al. (2014) do not discuss the lexical (nominal or verbal) behar; unlike Etxepare and Uribe-Etxebarria (2012), Rezac et al. (2014) consider only cases like (2b), restricting their account to that type of construction, which they distinctly label as INF+behar. However, they do not discuss other syntactic configurations behar enters into, like the ones shown and discussed in (7), (8), (9), or how their account of configurations like (2b) relate to them. Consider the following examples involving the predicate behar in (11):
Ergative need not split 169 (11) a. Zu-k liburu-a behar duzu you-erg book-det need have.2sg ‘You need the book.’ b. Zu-k liburu-a irakurri behar duzu You-erg book-det read need have.2sg ‘You need to read the book.’ c. Ni-k zu-k liburu-a irakur-tzea behar dut I-erg you-erg book-det read-inf need have.1sg ‘I need you to read the book.’ The proposal in Etxepare and Uribe-Etxebarria (2012) with behar as a noun, offers a unified explanation for the sentences in (11), which differ only in the type of constituent the predicate noun behar combines with in the small clause it heads: in (11a) it combines with a DP, in (11b) it combines with a vP, and in (11c) it combines with an AspP, as schematically shown in (12): (12) a. be [PP DP P [SC [DP] behar]] b. be [PP DP P [SC [vP] behar]] c. be [PP DP P [SC [AspP] behar]] The raising analysis of INF+behar (the construction it sets out to account for is illustrated in example 11b) begs the question of how it relates to cases where behar acts as a transitive predicate (11a, 12a), or cases where behar takes non-control infinitival complements whose external argument is not co-referential with the one in the finite main clause (11c, 12c). A raising account of behar for (11b) is also not straightforwardly compatible with the possibility of having non-finite clauses headed by behar, which is possible in Basque, as shown in (13): (13) normala da [ni-k liburu-a berriz irakurri behar iza-te]-a] normal is [I-erg book-det again read need be-nom]-det] ‘It is normal for me to need to/have to read the book again.’ (cf. English *it is normal for me to must read the book again) An account of (11b) in terms of a raising modal structure entails that (11a, c) and (13) involve at least one different lexical entry for behar which is homophonous with the raising modal deployed in INF+behar constructions as analyzed by Rezac et al. (2014). However, a nominal account of behar like the one put forth by Etxepare and Uribe-Etxebarria (2012) or Harves and Kayne (2012) does not need to postulate multiple homophonous lexical entries for behar; it can provide a full account of its syntax taking it to be a single lexical element of category N that enters into various syntactic combinations yielding the large array of constructions this nominal element can enter into, either as a predicate or as an argument.
170 Itziar Laka Rezac et al. (2014) argue that the behar+INF construction provides evidence that ergative is not inherently assigned in Basque: “Like English INF + must, Basque INF + behar proves to be a raising construction, but raising confers ergativity on the raisee even if it would otherwise be absolutive. Thus, ergativity is again dissociated from thematic relations” (Rezac et al. 2014: 1289). But if behar in (2b, 11b) is a raising modal with no external thematic role to assign, the assignment of ergative case to the subject is puzzling, because, as the authors acknowledge, other raising predicates like the copular verbs izan ‘be(individual level)’ egon ‘be(stage level)’ do not assign ergative to their subjects (as shown in 14b,d); they have absolutive subjects (as shown in 14a,c) (see Zabala 2003 for an extensive study of copular constructions in Basque). That is, if ergative case is structurally assigned in Basque by Tense, then all raising configurations should yield the same case for the raised subject. However, this is not so, as shown by the following examples of copular sentences: (14) a. Zu ikaslea zara you student be.2sg ‘You are a student.’ b. *zu-k ikaslea zara you-erg student be.2sg c. Gu Bilbo-n gaude we Bilbao-in are.1pl ‘We are in Bilbao.’ d. *Gu-k Bilbo-n gaude we-erg Bilbao-in are.1pl ‘We are in Bilbao.’ Rezac et al. (2014) do not discuss the reason why structures with behar assign ergative case to the raised argument. This issue is of significant relevance in a discussion on ergativity, because accepting that structures with behar + INF involve raising entails accepting that Basque stands out as a grammar involving ‘raising-to-ergative,’ a type of grammatical process that has been widely argued not to be possible in human language (Marantz 2000; Woolford 2006; Legate 2008, 2012a; Sheenan, Chapter 3, this volume, among others), and moreover, the account bears the burden of explaining why only some cases of raising in Basque yield this exceptional effect. Given the highly exceptional status of raising-to-ergative in grammatical theory, it is necessary to ascertain that other explanatory trails cannot be more successfully tread. The argument put forth in this chapter is that there is no need to postulate such an exceptional mechanism as raising to ergative to understand the syntax of behar and ergativity in Basque. By extension, a non-raising account of behar like that put forth by Etxepare and Uribe-Etxebarria (2012) is preferable both on empirical as well as theoretical grounds, because of the following reasons: (a) it does not need to postulate more than one lexical
Ergative need not split 171 entry for behar, revealed to be always a noun; (b) it does not need to postulate raising to ergative; (c) it does not need to postulate language particular exceptional processes to ensure that some raising structures will yield absolutive subjects while others yield ergative subjects; (d) it does not need to postulate invisible Tense projections in non-finite clauses like (11c, 13) where ergative case is perfectly grammatical; (e) it is compatible with the TotalErg hypothesis: inherent ergative and no splits. In contrast, the account in Rezac et al. (2014) must postulate at least two different lexical/categorical entries for behar: if there is a functional category behar akin to English ‘must,’ then there must also be at least a lexical noun behar in order to account for cases where behar is clearly akin to the English noun ‘need.’ It must conclude that raising to ergative is a possibility for Basque grammar, even though the reason why the raised argument is assigned ergative is not motivated, and by extension it yields a less constrained theory of grammar. The raising account relies on Tense as the sole assigner of ergative case, and in principle it begs the question of how to account for cases like (13), where ergative is possible in nonfinite and agreement-less clauses. Finally, the raising account of behar/need yields a highly exceptional and very language-specific picture of Basque that sets it apart from converging accounts of ergativity, and burdens our theory of grammar.
7.5 Agreement Issues: Little to Do with Case In this section, I briefly discuss the agreement morphology patterns generated by structures with behar. I argue that agreement morphology is dissociated from case morphology in Basque, a dissociation that is to be expected if ergative case is inherent and associated to vP configurations. In order to do that, I will first lay out some basic facts of agreement in Basque (see also Berro and Etxepare in this volume for a fuller view). As is well known, Basque has pluripersonal agreement (ergative, dative and absolutive arguments must obligatorily agree with verbal inflection), and agreement morphology, unlike case morphology, surfaces only in finite clauses, as shown in (15): (15) a. ni-k zu-ri liburua eman dizut I-erg you-dat book.abs given 3sg.have.2sg.1sg ‘I have given the book to you.’ b. [ni-k zu-ri liburua ema-te-a] ezinezkoa da I-erg you-dat book.abs give-nom-det impossible is ‘It is impossible for me to give you the book.’ (Lit: *it is impossible I you the book give)
172 Itziar Laka Both (15a) and (15b) contain case-marked DPs, but only the finite clause (15a) has agreement morphology referencing the case-marked DPs. In (15b), the clause containing the arguments is not finite and there is no agreement morphology in the nominalized verbal form ematea ‘to give’; the main clause contains an inflected auxiliary that does not agree with the DPs from the embedded clause. This contrast can straightforwardly be captured if case and agreement morphology are not manifestations of the same licensing condition, as claimed in Laka (2006b). Given a dissociation between case and agreement morphology, it is not surprising that children display different acquisition stages for case and agreement morphology (Ezeizabarrena 1996), that speakers suffering agrammatism produce many agreement errors but few case-morphology errors (Laka and Erriondo Korostola 2001), or that proficient non-native speakers of Basque who are native speakers of Spanish generate different electrophysiological components from native speakers when processing ergative case morphology, but not when processing ergative or absolutive agreement morphology (Zawiszewski et al. 2011; Zawiszewski, Chapter 28, this volume). Regarding the agreement patterns associated with behar, it has been widely noted in the literature that control configurations like (11b) above, repeated here as (16a) behave like clitic climbing structures in Romance, because the matrix auxiliary obligatorily agrees with the object inside the embedded clause (16b): (16) a. Zu-k liburu-a irakurri behar du-zu you-erg book-det read need have-2sg ‘You need to read the book.’ b. Zu-k liburu-ak irakurri behar d-it-u-zu you-erg book-detpl read need 3pl-have-2sg ‘You need to read the books.’ c. *Zu-k liburu-ak irakurri behar d-u-zu you-erg book-detpl read need have-2sg (You need to read the books) These agreement patterns have often been taken as direct evidence for case assignment, for instance by Etxepare and Uribe-Etxebarria (2012), who argue as follows: “Recall that at the beginning of this section we have shown that the matrix auxiliary has to display agreement with all the arguments of the embedded predicate. If this is correct, it suggests that, in contrast with other non-finite constructions, the embedded arguments cannot check their Case and agreement features within the non-finite clause and must look at the matrix domain to do it.” I depart from the assumption that case and agreement morphology necessarily reflect the externalization of a single grammatical process (Chomsky 2000), and follow instead the hypothesis that, at least in ergative languages, they are separate grammatical mechanisms. Note that this is a necessary consequence of the hypothesis that ergative is inherent, assigned within the vP domain.
Ergative need not split 173 Dissociating case from agreement provides a way of accounting for the numerous mismatches between case and agreement morphology attested cross-linguistically, many of which involve ergative languages, like Warlpiri or Chukchi, with both ergative– absolutive case morphology and nominative– accusative agreement (Dixon 1994), the Spurious Antipassive in Chukchi (Bobaljik and Branigan 2006), or Ergative Displacement in Basque (Laka 1988, 1993a), where agreement markers are dissociated from case morphology. This dissociation also provides a straightforward path to explain data from nonfinite environments where ergative case occurs freely in the absence of agreement and finiteness, like the examples provided in (11c), (13), (15b).
7.6 Discussion: No Need to Split The proposal in this chapter attempts to advance our understanding of ergativity with minimal appeal to language-particular exceptions, and assuming that ergativity is a uniform property manifested in some human grammars, a position that I have named the TotalErg hypothesis, bringing together two distinct but related hypotheses: (a) ergative case is inherent: its morphology signals external arguments associated to the specifier of the vP domain; (b) ergativity is a uniform grammatical phenomenon and it does not split; there are no nominative ruptures within ergative grammars. The TotalErg hypothesis refers to morphological case, and not to abstract Case, the overt NP licensing condition postulated in Government and Binding and the Minimalist Program (Chomsky 1981, 1995, 2000). As discussed in Bobaljik and Wurmbrand (2008) and Polinsky and Preminger (2014), morphological and abstract case were originally thought to be directly related, but this relation has become increasingly more tenuous in contemporary studies. I subscribe the suggestion in Bobaljik and Wurmbrand (2008) that once the dissociation of abstract Case and morphological case is in place, “this further level of abstraction leaves open the possibility that ergativity is best described as a morphological phenomenon, lying squarely outside the domain of Case Theory, and masking a (more) uniform syntax.” Bobaljik and Wurmbrand (2008) also note, quoting Chomsky (2000), that the relevance of abstract/structural case has been significantly reduced in Minimalism, given the feature-matching view of probe–goal relations. “The question arises still more sharply if matching is just identity, so that Case can never be attracted; operations are not induced by Case-checking requirements” (Chomsky 2000: 127). Within this view of what triggers grammatical operations, agreement (a φ-feature identity match) can be more adequately thought of as the externalization of finiteness-dependent operations than case. Morphological case, in turn, need not be a reflex of finiteness-dependent operations, but it can signal vP related configurations, as it does in the case of ergative languages, if the approach defended here is on the right track.
174 Itziar Laka
Acknowledgments I want to thank Richard Kayne, Martina Wiltschko, and Adam Zawiszewski for very helpful comments and suggestions, and the editors Jessica Coon, Diane Massam, and Lisa Travis for their patience and encouragement. The name TotalErg is inspired by an Italian company whose logo prompted me to think about the issues discussed in this chapter. Research funding from the European Community (EC FP7/SSH-2013-1 AThEME (613465)), the Basque government (IT665-13), and the Spanish government (FFI2012-31360) is gratefully acknowledged.
Abbreviations 1pl, first person plural; 1sg, first person singular; 2sg, second person singular; 3sg, third person singular; asp, aspect; dat, dative; Det, determiner; erg, ergative; gen, genitive; inf, infinitive; loc, locative; nom, nominalizer.
Chapter 8
T he structura l s ou rc e of split erg at i v i t y and ergati v e c ase in Georg ia n Léa Nash
8.1 Introduction The heterogeneous class of ergative languages shares one distinguishing property: the transitive subject bears a special marking, the ergative case, which does not appear on the intransitive subject. There is a large, albeit not unanimous, consensus in recent linguistic theorizing concerning the inherent nature of the ergative case, assigned by the thematic licenser, the little v (see Johns et al. 2006; Legate 2008; among others; and section 8.5.2). A question arises as to why the same category cannot assign the inherent ergative in all languages. Is this due to different parametric settings of case features of v, or is the ergative case epiphenomenal and conditioned by case-independent configurational factors at work in ergative systems? One of the reasons to opt for the second line of inquiry is that ergative languages are never fully ergative and display properties typical of nominative systems under certain structural circumstances. Languages vary in the degree of manifesting this mixed behaviour known as split-ergativity (Dixon 1994:2): Basque shifts to nominative only in the progressive aspect, and is ergative in all main tenses (Laka 2006a), while Hindi, Samoan, and Georgian shift in a wider range of constructions—they are nominative in imperfective aspects and ergative in perfective aspects. The existence of split ergativity leads to the reformulation of our initial questions: why can v assign an inherent case in some constructions but not in others in the same language? Why in a given configuration (e.g. perfective), is ergative available in certain languages but not in others? Why are some languages mixed ergative–nominative and others pure nominative?
176 Léa Nash In the present work, I attempt to single out the determining property that conditions the shift of an ergative language to nominative (or vice versa), on the basis of a close investigation of split ergativity in Georgian. This language, nominative in imperfective tenses and ergative in non-imperfective tenses, is particularly interesting because other than the change in case marking of main arguments, the shift does not affect its other structural properties: absence of auxiliary support, word order, number/person agreement, relativization strategies (Polinsky, in press), scope facts (Anand & Nevins 2006) are identical in ergative and nominative environments. My investigation of aspectual properties of imperfective clauses in Georgian that show nominative case alignment and their comparison to clauses underspecified for grammatical aspect that show ergative alignment leads me to the conclusion that ergative splits are conditioned by functional complexity in this language, and, by extension, universally. Concretely, I contend that the difference between a nominative and an ergative behaviour of the same language, and by extension between nominative and ergative languages, consists in the capacity of the transitive subject to be theta-licensed, and by consequence case-licensed, in a position outside vP only in the nominative type. An outcome of this difference is that the transitive subject in ergative languages is licensed in vP, which is also the minimal domain containing the direct object. As both arguments of the transitive verb stay in vP, they are case-licensed by the same c-commanding functional head, according to the mechanism of Dependent Case assignment as originally proposed by Marantz (1991) (see also Bobaljik 2008; Baker & Vinokurova 2010). (1)
If there are two distinct NPs in the same phase such that NP1 c-commands NP2 , then value the case feature of NP1 as ergative unless NP2 has already been marked for case.1 [from Baker & Vinokurova 2010: 595]
The reason why (1) holds in a language, which under the present analysis should be defined as a language with ergative (or at least with non-nominative) case alignment, is due to the functional impoverishment between T and vP. Specifically, a minimal functional spine of a nominative clause includes an aspectual category, which I will call Event borrowing the term from Ramchand (2013) (see also Ramchand & Svenonius 2014), it expresses the viewpoint aspect that ties eventualities in vP with the reference time in T and theta-licenses the subject of Event in its specifier, à la Kratzer’s (1994) Voice. The highest of vP arguments introduced in Spec,EventP, has its case valued by T, while the other argument, the object DP in vP checks case by Event. As this functional category is absent in ergative systems, T directly c-commands vP and case checking proceeds as in (1) (2) [TP T [EvP DPcase Ev [vP v […DPcase ]]]] (3) [TP T [vP DPerg v […DPcase ]]]
1
nominative system ergative system
It will be shown in section 8.5.2 that the notion of NP in the formulation should be understood in a larger sense, encompassing CPs and weak implicit arguments as in Landau (2010).
Split ergativity and ergative case in Georgian 177 While I contend that relations between arguments and the verbal predicate are identical in both systems, nominative systems provide a different thematic licensing position for the transitive subject, Spec,EventP. But how does the agent DP which is the argument of v, the highest of primitive verbal predicates in a decomposed transitive vP, get to Spec,EventP in nominative languages? Is this due to A-movement, and, if yes, what motivates it? Building on Kratzer’s (1994) original insight that the theta-licensing category, Voice, obtains information about the semantic role of the external argument via a semantic mechanism of Event Identification, I claim that the syntactic implementation of this idea is possible through the mechanism of semantic control, essential in restructuring contexts as proposed by Wurmbrand (2003) (see also Harley 2013). The Event category is then to be viewed as a functional head of the same type as an auxiliary verb: it constitutes a complex predicate with the lexical predicate and licenses the argument of the latter via restructuring. The unavailability of Event in ergative systems should have other repercussions than case-licensing, namely this absence should affect the temporal interpretation of an eventuality expressed by v, which I show to be the case in Georgian. In section 8.2, I show how nominative tenses in this language are unambiguously imperfective, while ergative tenses are aspectually deficient—neither perfective nor imperfective. In section 8.3, I analyse this aspectual underspecification in syntactic terms; I propose that a viewpoint aspectual category, which I label as Event, is absent in ergative tenses. Section 8.4 is devoted to justifying the presence of the category Event in nominative tenses and its effect on thematic licensing of the transitive subject via Event- v restructuring in the sense of Wurmbrand (2003). In section 8.5, I analyse the impact of Event’s absence on case licensing in ergative tenses. I first expose properties of ergative DPs in Georgian focusing on the fact that these arguments behave as structurally case- marked arguments, and then argue that in T-vP configurations, case assignment proceeds along the mechanism of dependent case as in Marantz (1991) whereby the highest of vP arguments is marked as ergative. In section 8.6, I compare the conclusions of this study to an alternative analysis of ergative split initiated by Laka (2006a) and show how both approaches hinge on one central idea pertaining to a richer structural articulation of temporal relations, by adding a clause or a functional category, in nominative systems but not in ergative systems.
8.2 Ergative Split in Georgian 8.2.1 Nominative and Ergative Tenses In the present, future, imperfective past tenses, referred henceforth as nominative tenses, Georgian shows nominative case alignment: the transitive subject is nominative, the direct object is accusative. The case affix on direct and indirect objects is identical in nominative tenses, but their licensing properties are not: unlike direct objects (theme
178 Léa Nash arguments), dative arguments (goals, benefactors, affected locations, experiencers) are licensed by applicatives and show constant case behaviour in nominative and ergative tenses. Therefore, I will continue to refer to the case on direct objects as accusative. In the aorist and subjunctive tenses, henceforth ergative tenses, Georgian behaves as a typical ergative language: the transitive subject is marked with the ergative case (also known in the Kartvelian grammatical tradition as narrative (see Tuite, Chapter 45, this volume) and the direct object is nominative. As the case borne by direct objects in ergative tenses is strictly conditioned by the verb’s finiteness (see section 8.5.1.2), I consider this morphologically absolutive case to correspond to the structural nominative checked by T rather than to the abstract “accusative” assigned by V (see Legate’s 2008 distinction of absolutive as abstract V or abstract T cases). nominative tenses: (4) a. vano-ø xaT-av-s / xaT-av-d-a mankana-s Vano-nom draw-ts-3s/ draw-ts-past-3s car-acc ‘Vano is drawing/was drawing a car’ b. vano-ø Ċam-ø-s / Ċam-ø-d-a kada-s Vano-nom eat-ts-3s / eat-ts-past-3s cake-acc ‘Vano is eating/was eating a cake’ c. vano-ø alag-eb-s / alag-eb-d-a otax-s Vano-nom tidy-ts-3s / tidy-ts-past-3s room-acc ‘Vano is/was tidying the room’ ergative tenses: (5) a. vano-m xaT-a / xaT-o-s mankana-ø Vano-erg draw-aor3s/ draw-sbjn-3s car-nom ‘Vano drew/draw a car’ b. vano-m Ċam-a / Ċam-o-s kada-ø Vano-erg eat-aor3s/ eat-sbjn-3s cake-nom ‘Vano ate/eat a cake c. vano-m alag-a / alag-o-s otax-i Vano-erg tidy-aor3s/ tidy-sbjn-3s room-nom ‘Vano tidied/tidy the room’
8.2.2 Meaning of Nominative and Ergative Tenses In Georgian, the present and the past imperfective tenses are used in progressive, generic, and habitual contexts. These tenses are imperfective as they successfully pass any test that shows that the event they convey contains the reference time of the utterance. In (6a), only
Split ergativity and ergative case in Georgian 179 simultaneity with the other temporal clause is implied, and in (6b), the sentence only has durative, but not inclusive reading. (6) a. roca Kar-i ga=v-aγ-e nino-ø Telepon-ze laParaK-ob-d-a when door prev=1-open-aor Nino-nom phone-on speak-ts-past-3s ‘When I opened the door, Nino was speaking on the phone’ (–> simultaneous talking and opening) b. nino-ø Telepon-ze ori-dan sam-amde laParaK-ob-d-a Nino-nom phone-on two-from three-until speak-ts-past-3s ‘Nino was speaking on the phone from two to three’ (–> continued after three) Although both the past imperfective and the aorist are used to describe situations that happened before the reference time, the past imperfective is employed to report simple general facts about the past, while the aorist is not felicitous in these contexts. So (7b) can only describe a specific event in the past, e.g. a TV talk show, where people spoke five languages. (7) a. adre xalx-i xut ena-ze laParaK-ob-da before people-nom five language-on speak-ts-past3s
past imperfective
b. #adre xalx-ma xut ena-ze i-laParaK-a aorist before, people-erg five language-on riam2-speak-aor3s ‘Before, people spoke five languages’
8.2.3 The Aorist Although it is commonly stated that Georgian switches to ergative case system in perfective tenses, a close look at the aorist reveals a more complex situation. The Georgian aorist does describe events terminated before the reference time. But although they are completed and do not continue unlike (6b) beyond the reference time, events in the aorist are interpreted as non-culminated in the sense that the natural endpoint inherent to the meaning of the predicate is not necessarily reached upon the termination of the event. In (8a), we tend to understand that the event has ended but the car is not drawn, the dress is not sewn in (8b), the table is not painted in (8c). (8) a. vano-m xaT-a mankana-ø vano-erg draw-aor3s car-nom ‘Vano drew a car’ 2
RIAM stands for the reflexive implicit argument marker and its properties will be discussed in section 8.4.3.1.
180 Léa Nash b. vano-m Ker-a es Kaba-ø vano-erg sew-aor3sg this dress-nom ‘Vano sewed a dress’ c. vano-m γeb-a magida-ø vano-erg paint-aor3s table-nom ‘Vano painted the table’ Non-culmination is not entailed but strongly implied, as both continuations of (9) are equally felicitous. (9)
vano-m Vano-erg
c’mind-a kotan-i, … clean-aor3s pot-nom a. … magram ver ga=c’mind-a but can’t prev=clean-aor3s b. … da Kargad ga=c’mind-a and well prev=clean-aor3s ‘Vano cleaned the pot … but (he) couldn’t clean it/and cleaned it well’
In (9) the pot can be understood as absolutely not cleaned as a result of the event, as partially cleaned, or completely cleaned. The only meaning entailed in (9) is that Vano started and ended some action with the pot that can be defined as cleaning but the outcome of his doings is not entailed. If the direct object is a plural DP as in (10), the sentence entails that Vano did something to each of the pots, without a further entailment that any of them got cleaned. The situation is very different in imperfective contexts with plural direct objects as in (11), where the sentence is true even if Vano was in the process of cleaning only one pot and in the end he cleans just three3 (see Singh 1998 for similar effects in Hindi neutral perfectives). (10) gušin, vano-m c’mind-a xut-i kotan-i yesterday, vano-erg clean-aor3s five-nom pot-nom ‘Yesterday Vano cleaned five pots’ (entails: 5 pots got affected by Vano’s activity) (11)
Vano-ø c’mind-av-d-a vano-nom clean-ts-past-3s ‘Vano was cleaning five pots’
xut-ø kotan-s five-acc pot-acc (does not entail: 5 pots were being cleaned)
3 The contrast between (10) and (11) undermines the idea that non-culminated accomplishments should be analysed in a similar fashion as imperfectives in terms of Dowty (1979)’s semantic notion of inertia worlds, as has been proposed by Bar-el et al. (2005) for St’át’imcets and Koenig and Muansuwan (2001) for Thai.
Split ergativity and ergative case in Georgian 181 As events without a culmination point are atelic, a speaker chooses to use the aorist form in order to convey the duration of a naturally telic accomplishment, as in (12). (12) a. vano-m Kargad da=alag-a otax-i, ori saati alag-a Vano-erg well prev=tidy-aor3s room-nom, two hours tidy-aor3s ‘Vano cleaned the room well, he cleaned it for two hours’ b. vano-m ver ga=Tex-a kotan-i, ori saati Ki Tex-a Vano-erg can’t break-aor3s vase-nom, two hours yet break-aor3s ‘Vano could not break the vase, yet he broke it for two hours’ From the left parts of assertions in (12) and in (13), we see that the culmination of terminated events is expressed by a perfectivizing morpheme (preverb) prefixed to the aorist forms. Georgian preverbs, which generally have a locative meaning and are in many aspects comparable to Slavic perfectivizing preverbs are not only dedicated to the aorist, they are added to imperfective verbs in the present and the past tense to shift their meaning to the future. They are also independent of finiteness and optionally surface in nominalizations, as in Slavic. (13) a. Vano-m da=xaT-a mankana- ø Vano-ERG prev=draw-aor3s car-nom ‘Vano drew (and finished) the car’ b. Vano-m še=Ker-a Kaba-ø Vano-ERG prev=sew-aor3s dress-nom ‘Vano sewed (and finished) a dress’ c. Vano-m ga=c’mind-a otax-i Vano-erg prev=clean-aor3s room-nom ‘Vano cleaned (and finished) the room’ On the basis of the exposed properties of the aorist, I conclude that this tense expresses events which took place and terminated before the reference time but it is not formally specified for the perfective aspect. The situation is different from the past and the present tenses where clauses are positively specified for the imperfective aspect.
8.3 Aspectual Deficiency as Structural Impoverishment 8.3.1 Neutral Aspect and Georgian Aorist The implication of non-culmination of terminated events, typical for the Georgian aorist, has been reported to exist in other languages (Singh 1998; Bar-el et al. 2005;
182 Léa Nash Tatevosov 2008) and naturally fits the description of neutral viewpoint aspect put forth in Smith (1991). According to the author, sentences in the neutral aspect, which is morphologically default, describe situations that include the initial point and at least one stage of the situation. This aspect shares some properties with the perfective, and others with the imperfective. As witnessed above and summarized in what follows, the Georgian aorist perfectly manifests all the properties ascribed to this type of viewpoint aspect (see also Iatridou et al. 2001) (a) Neutral aspect makes reference only to the beginning of the internal temporal structure of an eventuality. Similarly to the imperfective (14a), and in contrast to the perfective (14b), (14c) does not assert that a table gets painted. (14)
a. vano-ø γeb-av-d-a magida-s Vano-nom paint-ts-past-3s table-acc ‘Vano was painting the table’
imperfective
b. vano-m še=γeb-a magida-ø preverb+aorist vano-erg prev=paint-aor3s table-nom ‘Vano painted the table (and finished) c. vano-m γeb-a magida-ø aorist Vano-erg paint-aor3s table-nom ‘Vano painted the table’ (=was engaged in table painting) (b) Predicates in neutral aspect disallow completive adverbials (e.g. ‘in two hours’), (15)
a. *nino-m γeb-a magida-ø or saat-ši [ok with še=γeba (prev+aor)] Nino-erg paint-aor3s table- nom two hour-in ‘Nino painted the table in two hours’ b. *nino-ø γeb-av-d-a magida-s Nino-nom paint-ts-past-3s table-acc ‘*Nino was painting the table in two hours’
or two
saat-ši hour-in
(c) Neutral aspect allows both durative and inclusive interpretation of time intervals (e.g. between 3 and 4 o’clock), like the perfective, and unlike the imperfective. (16) a. nino-m γeb-a magida-ø sami-dan otx-amde Nino-erg painted-aor3s table-nom three-from four-to ‘Nino painted the table from 3 to 4’ (entails: Nino did not paint after 4)
Split ergativity and ergative case in Georgian 183 b. nino-ø γeb-av-d-a magida-s sami-dan otx-amde Nino-nom paint-ts-past-3s table-acc three-from four-to ‘Nino was painting the table from 3 to 4’ (does not entail: Nino did not paint after three) (d) the neutral patterns in temporal sequencing like perfective eventualities, and unlike imperfectives which only allow simultaneous readings with temporal clauses. (17) a. nino-m γeb-a magida-ø roca vano-ø c’a=vid-a Nino-erg paint-aor3s table-nom when Vano-nom prev=go-aor3s ‘Nino painted the table when Vano left’ (sequencing) b. nino-ø γeb-av-d-a magida-s roca vano-ø c’a=vid-a Nino-nom paint-ts-past-3s table-acc when Vano-nom prev=go-aor3s ‘Nino was painting the table when Vano left’ (simultaneity)
8.3.2 No Category of Aspect in the Neutral Aspect Smith (1991) questions whether neutral aspect is a third type of viewpoint aspect, along with the perfective and the imperfective. There is an ongoing debate concerning its ontological validity, and the effects ascribed to this aspect can be claimed to be deductible from other semantic properties such as telicity and cumulativity, which are independently required (Artshuler 2013). I contend that the deficient perfectivity of sentences in Georgian aorist is not due to the presence of a formal neutral viewpoint Aspect category, but is rather a result of the absence of any Aspect category in the clausal functional structure. As opposed, past and present imperfective tenses have their aspectual properties encoded by such a category. My view is grounded in the conception of grammatical viewpoint aspect as a distinct notion from the aspectual organization of vP (Aktionsart) and constituting a distinct syntactic category with a binary value (perfective–imperfective), situated in the clause between T and vP (see Smith 1991 for distinguishing the two notions of aspect; Depreatere 1995; see also Stowell 1996 for properties of ZeitP).
8.3.2.1 The Aorist and the Subjunctive Share Structural Deficiency: Imperative Clauses Additional support for my claim that sentences in the aorist do not contain a category expressing the viewpoint aspect comes from the properties that the aorist shares with the subjunctive in Georgian. In both “tenses”, the transitive subject is marked with ergative and the verb surfaces in its impoverished form to be discussed in section 8.4.1. Most significantly, aorist and subjunctive forms of the verb are used in imperative clauses as well.
184 Léa Nash As Georgian does not have infinitives subjunctive forms are used in modal contexts (after can/must/wish) and in counterfactuals, and are marked by a special irrealis mood affix -o-. As in the aorist, the perfectivizing preverb is added to the verb to convey the natural telicity of the event, otherwise the meaning of the verb is neutral. Example (18a) means that I wish Vano to engage in the event of car drawing, by drawing at least some lines that eventually, in normal circumstances, can yield a picture of a car. The sentence also serves to convey the duration of the event that I wish Vano to be engaged in. (18)
a. m-i-nda (rom) vano-m xaT-o-s mankana-ø (ori saati) 1o-appl-wish that Vano-erg draw-sbjn-3s car-nom (two hour) ‘I wish that Vano draw a car (for two hours)’ b. Kargi ikneba vano-m (rom) xaT-o-s mankana-ø good befut Vano-erg that draw-sbjn-3s car-nom ‘It would be great if Vano draw a car’
In imperative clauses, the subjunctive and the aorist forms share the labor: the subjunctive form is used in negative contexts (19a) and with 1st pl, 3rd person subjects (let us, him, them), (19b). The aorist form is used with 2nd person imperatives (19c): (19) a. ar ø-xaT-o mankana-ø neg 2-draw-sbjn car-nom ‘Don’t draw a car’ b. (man) xaT-o-s mankana-ø (he-erg) draw-sbjn-3s car-nom ‘Let him draw a car’ c. ø-xaT-e mankana-ø 2-draw-aor car-nom ‘Draw a car!’ Aorist and subjunctive forms are used with the imperative mood and in irrealis, non-referential temporal contexts with no temporal anchoring in the reference time. I conclude that these forms occur in temporally deficient contexts with impoverished functional structures (see Zanuttini 1997, among others, who analyses Romance imperative clauses as functionally deficient and lacking tense).
8.3.3 Interim Conclusion: Ergative–Nominative Asymmetry We have seen that in Georgian aspectually underspecified events are expressed by ergative tenses, and imperfective events by nominative tenses. This missing aspectual specification, typical of ergative clauses, generally serves to make a time flow transition between
Split ergativity and ergative case in Georgian 185 the speech time and the eventuality itself. Syntactically, this ‘bridging’ task is carried out by a special functional category Event (or Aspect, or Zeit), generated in the clausal functional spine between T (time) and vP (eventuality) (see Ramchand & Svenonius 2014). (20) nominative systems: [TPT..-..Ev-vP ] ergative systems: [TP T..-Ev-vP] The interpretative cost of structural deficiency in ergative tenses is that events are interpreted as distal, not anchored with respect to the reference time—the event is just named without referring to the specifics of its internal temporal organization as if it were “nominalized” in some sorts. The morphological consequence of (20) is a systematic absence of thematic suffixes, glossed as TS, in ergative tenses in Georgian. In section 8.4, I show that this suffix stands for imperfective aspect and as such formally fleshes out the category Event. We saw that its absence does not shift the verb into perfective in Georgian; what Georgian lacks, unlike English, is a zero morpheme positively marking the perfective aspect of the clause. This is the reason why an assertion in the perfective aspect such as John broke the window but could not break it sounds as a contradiction in English but not in Georgian.
8.4 The Category Event in Nominative Tenses 8.4.1 Morphological Evidence: TS as Markers of Imperfectivity In nominative tenses, transitive, unergative and unaccusative verbs show up with a suffix attached between the root and tense/agreement markers. Following Georgian grammarian tradition, I refer to them as thematic suffixes, ts, and show that below they spell out the imperfective aspect (see Tuite, Chapter 45, this volume). While unaccusative verbs all have the same ts -eb (see section 8.4.3.1), unergative and transitive verbs appear with a small set: -av,-eb, -i, -ob, -ø. Each ts can coarsely be tied to a semantic class: creation/destruction/reconfiguration transitive verbs—manner verbs in the sense of Rappaport Hovav & Levin (2010)—have -av, deadjectival causatives— Rappaport Hovav and Levin’s result verbs—come with -eb. As for unergatives, verbs of sound and noise emission take -eb, manner of motion -av, manner of behaviour -ob.4 Thematic suffixes also appear on corresponding nominalizations but must be followed by the nominalizer -a. 4
Stative verbs never take a ts, never show up in the aorist or in the future where only bounded, i.e. terminated, events appear.
186 Léa Nash (21) verbs (present) Root+ts+t/agr draw sow frighten whiten
xaT-av-s tes-av-s (a)šin-eb-s5 (a)tetr-eb-s
reign (act+king) act+nervous roll swim
mep-ob-s nerviul-ob-s gor-av-s cur-av-s
Nominalization transitive drawing sowing frightening whitening unergative reigning acting-nervous rolling swimming
Root+ts+n xat-av-a tes-av-a šin-eb-a tetr-eb-a mep-ob-a nerviul-ob-a gor-av-a cur-av-a
As thematic affixes are incompatible with non-imperfective ergative tenses, I take them to spell out the imperfective aspect of the eventuality in clauses.
8.4.2 Event as an Aspectual Head That Triggers Restructuring The imperfective aspect is formally represented as a syntactic category Event which has a double function cross-linguistically (see Ramchand 2013 and section 8.6.3): it selects vP and theta-licenses its highest argument via the mechanism of semantic control, as proposed by Wurmbrand (2003) for restructuring constructions. Semantic control results from Event’s binding an open variable in v, just as in restructuring contexts the higher of the two verbs binds a thematic variable of its V complement. (22) [EvP DPi Evi [vP vi [VP…]]]
complex predicate Ev-v
This head-to-head binding is at the heart of complex predicate composition and argument sharing. The capacity to theta-license an argument makes Event look as an argument taking predicate but it functions as an auxiliary because, unlike lexical verbs, it cannot have its own thematic properties but only inherit those of its complement. In Georgian, this privileged relation between Event and v affects the shape of ts, their form is sensitive to—in a sense “agrees” with—the Aktionsart properties of the selected verb. For example, it is noteworthy that all (anti)causative result verbs appear with -eb (see section 8.4.3.1). 5 The morpheme a-prefixed to the root in deadjectival causatives signals that the event is causative, it shows that in a bieventive verb, the upper causing event and the lower change event are not identical. It can be qualified, as a shorthand, as a causative marker, or as an applicative.
Split ergativity and ergative case in Georgian 187 While subjects of transitive verbs are theta-licensed in nominative systems by a functional category Event, as in (22), all thematic licensing proceeds at the level of vP in ergative tenses. Ergative case signals that agents are base-generated in vP, as specifiers of their lexical predicate. In nominative systems, the agent is external to its selecting predicate as its licensing happens after restructuring.6 A similar conception of agent licensing that implies that the highest lexical verbal predicate, labelled as v-caus, or as v, or as Cause, does not project its argument locally goes back to Hale and Keyser’s (1993, 2002) syntactic decomposition of verbs and has been repeatedly argued for by many (see Pylkkänen 1999; Schäfer 2012b; Harley 2013, among others). My analysis differs from these authors’ in not taking this property of v as a primitive but deriving it from structural requirements as in (30) which crucially hinge on the existence of the restructuring category Event.
8.4.3 Event Does Not Express Voice Distinctions Nash (1995) analyses Georgian thematic suffixes as realizations of Kratzer’s (1994) Voice, as their presence coincides with that of accusative objects, following Burzio’s generalization that the category that theta-licenses the external argument also checks accusative. As Voice is absent in ergative tenses, case marking of main arguments shifts and the transitive subject is marked with the inherent verb case. At that period of syntactic theorizing there was much confusion concerning the distinction between Kratzer’s Voice and Chomsky’s (1995) little v, the transitivizing head absent in unaccusative VPs. Nash (1995) does not provide a clear answer whether the absent head is a transitivizer in which case Georgian clauses in ergative tenses are predicted to have a radically different VP structure than in nominative systems—transitive VPs would be mono-level unaccusative structures with an adjoined agent marked with the inherent ergative. In nominative tenses, with Voice projected, Georgian transitive predicates are predicted to have a more articulated substructure and clearly show voice distinctions. Drawing on the conclusions of the present analysis about aspectual properties of the aorist in Georgian, I can maintain the same insight as in Nash (1995) concerning functional impoverishment of ergative tenses and show that the absent Event category, which only expresses temporal properties of the clause, does not affect the expression of voice distinctions which involve operations lower in the structure, at the level of vP. 6 Collins’s (2005) analysis of ergative–nominative case asymmetry in terms of movement from Spec,vP to Spec,VoiceP constitutes an alternative to the restructuring operation. I do not adopt it here as its motivation remains problematic—if DPs are lexically marked for case, movement for case in the process of which the lexical marker is absorbed by the selecting head is unwarranted. If the ergative case- affix is analysed as a preposition (see Polinsky in press), such movement operation reduces to preposition incorporation, comparable in spirit to Kayne’s (1993) treatment of auxiliary selection and Mahajan’s (1994) analysis of ergativity in Hindi. As shown, Georgian ergative nominals cannot be analysed as PPs and preposition incorporation in one tense but not in the other would require additional elaboration of the link between Voice and vP.
188 Léa Nash
8.4.3.1 Voice Distinctions in Georgian The principal reason why ergativity cannot be tied to voice distinctions in Georgian is that verbal predicates in non-active voice have the same shape across nominative and ergative tenses. There are three ways to express non-active voice: (i) non-active counterparts of agentive transitives are marked with the prefix i-, which also serves to mark reflexive verbs; by its polysemy, i- resembles the Romance SE/SI clitic (and is glossed as reflexive-implicit argument marker RIAM; see also section 8.5.1.3), (ii) anti-causative deadjectival verbs are formed by means of the suffix -d, the fientive primitive predicate become, (iii) some unaccusatives are unmarked, their semantic class is heterogenous containing both underived unaccusatives such as rč-eb-a (stay) and change of state verbs such as tb-eb-a (warm). (23) a. vano-ø xe-s γeb-av-s, a-tetr-eb-s, a-lP-ob-s Vano-nom tree-acc paint-ts-3s, appl-whiten-ts-3s, APPL-rot-ts-3s ‘Vano is painting/whitening/rotting the tree’ b. xe-ø i-γeb-eb-a, tetr-d-eb-a, lP-eb-a tree-nom riam-paint-ts-3s, white-become-ts-3s, rot-ts-3s ‘the tree is being painted, is whitening, is rotting c. xe-ø i-γeb-a, ga=tetr-d-a, da=lP-a tree-nom riam-paint-aor3s, prev=white-become-aor3s, prev=rot-aor3s7 ‘the tree was painted (but not finished)’, whitened, rotted’ (23b) shows that in nominative tenses non-active morphemes coexist with ts -eb, so they cannot spell out the same functional head. There is robust morphological evidence that non-active morphemes attach to the verb before ts. According to Baker’s (1985) Mirror Principle, the order of morphemes reflects the structural hierarchy of corresponding syntactic categories, therefore, Event scopes over the fientive predicate as the non-active -d attaches to the root before ts in (23b). The same cannot be as easily asserted for riam, a prefix. Yet, there is some indirect evidence that ts scopes over riam as well. As mentioned in section 8.4.1, all non-active verbs take ts -eb, while transitive verbs combine with different ts. Normally, transitive agentive manner verbs take -av but their corresponding non-active forms still show up with -eb, (24). If riam were affixed after the composition of V with ts, it is unclear how that would result in the change of -av to -eb. (24) a. vano-ø γeb-av-s / c’mind-av-s kotan-s Vano-nom paint-ts-3s / clean-ts-3s pot-acc ‘Vano is painting/cleaning the pot’ 7 Perfective preverbs are obligatory on unaccusatives but not on non-active forms which are derived from agentive transitives and carry the riam prefix. As unaccusatives are achievements, the expression of telicity becomes mandatory for this aspectual class.
Split ergativity and ergative case in Georgian 189 b. *kotan-i i-γeb-av-a / i-c’mind-av-a b’. kotan-i i-γeb-eb-a / i-c’mind-eb-a pot-nom riam-paint-ts-3s / riam-clean-ts-3s ‘The pot is being painted/cleaned’ The morphological evidence presented above suggests that voice related operations take place at the level of vP below Event, when this category is instantiated. The little v is active when it selects a referential DP argument, Agent or Causer, and it is non-active when its argument is not realized, or non-referential, in which case the predicate bears the riam i-(see Embick 1998; Alexiadou et al. 2006; Schäffer 2008). As for unaccusatives built from the fientive predicate -d become, I conjecture that they are monoeventive predicates just as their English translation suggests and are not to be viewed as truncated causatives. Below I summarize crucial morphological pieces which serve to build non-active and unaccusative predicates in Georgian. (25) a. i- —[reflexive implicit argument marking on] non-active v b. -d —fientive V become c. -eb —imperfective aspect realized in Event and absent in the aorist/subjunctive Active and non-active v are thematic predicates, while the category that guarantees the nominative behaviour of Georgian is functional, it provides aspectual information and theta-licenses the higher argument of the active vP. The transitive subject and the direct object in nominative tenses belong to two different domains, EventP and vP. The subject is case-licensed by the c-commanding T via standard structural case-Agreement mechanisms as in Chomsky (2000), while the direct object checks its case against the closest functional c-commanding category Event. (26) [TP T [EvP DPi case Evi [vP vi [VP DPcase V]]]] In the absence of Event in ergative tenses, T directly takes a vP as its complement. If the verbal domain contains one argument, its case is checked against T and is nominative. However, in transitive contexts where two arguments need to have their case checked, it is the lower of the two, the direct object, that case-Agrees with T. The higher agent argument is case-marked with the ergative case, and accounting for the source of this marking has been a notorious conundrum of linguistic theorizing. In a nutshell, the problem may be tackled from three angles: ergative can be analysed as (i) an inherent case (to be discussed in section 8.5.3), (ii) a structural case checked by a specific feature-content of some functional category (see section 8.5.4), or (iii) a dependent case, assigned when some specific structural configurational requirements are met. I will adopt the third path in section 8.5, first showing why it is unwarranted to view the ergative case in Georgian as inherent, contra Nash (1995).
190 Léa Nash
8.5 Case Assignment in Ergative Tenses 8.5.1 Properties of Ergative Arguments in Georgian In this section, I expose properties of arguments marked with the ergative case in Georgian which reveal that these arguments behave like their homologues in nominative tenses: both are sensitive to finiteness of the clause, both trigger number agreement with the verb and both case-concord with secondary predicates.
8.5.1.1 Ergative Nominals are DPs The ergative case morpheme, the suffix -ma/-m, has no other functions in Georgian, unlike other ergative languages where the affix or the adposition which marks the transitive subject is also used to mark locative arguments (e.g. Avar), instrumentals (e.g. Udi), genitives (e.g. Lak, Inuit) (see Tuite, Chapter 45, this volume). The ergative case marking in Georgian cannot qualify as a postposition because unlike other postpositions and like nominative and accusative/dative case affixes it is subject to a noun–adjective case concord within a DP (see Polinsky in press on analysis of ergative as a prepositional marker). (27) a. ert-ma lamaz-ma Kac-ma
oneerg handsomeerg manerg
b. ert-s lamaz-s Kac-s
oneacc handsomeacc manacc
c. ert-i lamaz-i Kac-i
onenom handsomenom mannom
d. ert(*-ze) lamaz(*-ze) Kac-ze one-on handsome-on man-on ‘on one handsome man’
8.5.1.2 Ergative DPs Occur in Finite Contexts Ergative, like nominative, dative and accusative, is unavailable in non-finite contexts, which in Georgian involve nominalizations. Nor does the ergative affix mark optional agents in passive constructions that are formed in Georgian as projections of a postposition mier specifically designated to mark passivized agents, (28a). In nominalizations, genitive is assigned to the sole argument of a nominalized predicate, and in case of nominalizations of transitive predicates genitive marks the theme and the agent optionally surfaces as a mier-PP of passive constructions (28b–c). (28) a. vano-ø iQo [nino-s mier] nakebi Vano-nom was Nino-gen by praise.past.p ‘Vano was praised by Nino’
Split ergativity and ergative case in Georgian 191 b. ga=v-igon-e [nino-s laParaK]-i / *[nino-m laParaK]-i prev=1-hear-aor Nino-gen talking-nom / Nino-erg talking-nom ‘I heard Nino(’s) talking’ c. ga=v-igon-e [[nino-s mier] vano-s keba]-ø prev=1-hear-aor Nino-gen by Vano-gen praising-nom ‘I heard Nino(’s) praising (of) Vano’
8.5.1.3 Ergative DPs Are Subjects of “Transitivized” Unergatives Subjects of unergative verbs are uniformly marked as ergative, which can be taken as an indication that this case is semantically tied to the expression of agentivity. Yet, unergative constructions are not syntactically monoargumental as clearly witnessed by the morphological shape of corresponding verbs which obligatorily contain the RIAM i-, which as we saw in section 8.4.3.1 is also employed in Georgian to signal the implicit argument in non-active contexts. The presence of this affix is a non-ambiguous indication that the internal structure of unergative predicates is complex in ergative tenses. Nash (2016) analyses them as transitive bieventive internally caused accomplishments where the causer is coreferential with the argument of the process whose structural presence is signalled by riam. This reanalysis of the aspectual class of activities into accomplishments is not directly tied to ergativity, as the same riam also appears on unergative verbs in the nominative future tense. It is rather forced by the mismatch between interpretative requirements to express completed events (the future is perfective in Georgian) and formal syntactic means to execute them: in the absence of a functional category that expresses perfectivity, this information must be read off the internal structure of vP: the structure of accomplishments with two subevents, activity and result, provides the required meaning of the completion of eventuality.8 (29) nino-m i-laParaK-a, i-varǰiš-a, Nino-erg riam-talk-aor3s riam-exercise-aor3s, ‘Nino talked, exercised, hissed’ literally, ‘Nino caused her talking, exercising, hissing’ ‘Nino had herself talk, exercise, hiss’
i-sisin-a riam-hiss-aor3s
8.5.1.4 Ergative DPs Trigger Number Agreement A good indication that ergative DPs are structurally case-marked is that they trigger number agreement with the verb in the same fashion as nominative subjects.9 8
Subjects of stative verbs in Georgian such as love, fear, hate are never marked with ergative; this class of verbs has a radically different behaviour than dynamic verbs and will not be treated in this work. It is sufficient to point out that the dynamic–stative split is an indication that ergative is tied to complex internal vP structure, namely to the availability of bieventive decomposition of the eventuality expressed by the verb. 9 Examples in (30–31) show that 3rd person nominative or accusative plural direct objects never agree with the verb. Number agreement in Georgian is subject-oriented, triggered by the highest argument (see also n. 12)
192 Léa Nash (30) a. Kac-i xaT-av-s kal-eb-s man-nom draw-ts-3s woman-pl-acc ‘a man draws women’ b. Kac-eb-i xaT-av-en man-pl-nom draw-ts-3pl ‘men draw a woman’ (31)
kal-s woman-acc
a. Kac-ma xaT-a kal-eb-i man-erg draw-aor3s woman-pl-nom ‘a man drew women’ b. Kac-eb-ma xat-es kal-i man-pl-erg draw-aor3pl woman-nom ‘men drew a woman’
8.5.1.5 Ergative and 1st/2nd Person Arguments Only 3rd person arguments can appear with the ergative case marking. First-and second-person pronouns do not bear any case endings in Georgian and are coreferenced on the verb by person prefixes which have been analysed as clitics by Nash- Haran 1992; Halle & Marantz 1993; Nash 1997. I conjecture that 1st/2nd arguments, due to their special status as discourse participants, are uniformly represented as referential pro in vP (or in EventP in nominative tenses) and are doubled by a corresponding clitic in T, similarly to the situation attested in many Romance languages (Manzini & Savoia 2005). The optional “caseless” full pronouns are henceforth best viewed as adjuncts to case-marked pro. This is particularly clear when we consider agreement patterns of depictive nominal and adjectival secondary predicates. The secondary predicate agrees with the silent pro marked as ergative in (32a, 33a) and as nominative in (32b, 33b), while the optional “caseless” full 1sg pronoun me does not change its form across tenses.
(32) a. bavš-ma (me) proerg da=v-Tov-e tbilis-i child-erg I prev=1-leave-aor Tbilisi-nom ‘I left Tbilisi (while) a child’ b. bavšv-i (me) pronom da=v-obl-di child-nom I prev=1-orphan-become-aor ‘I became an orphan, (while) a child’ (33) a. (me) proerg v-nax-e I 1-see-aor ‘Ii saw Vano drunki’
vano-ø mtvral-ma Vano-nom drunk-erg
Split ergativity and ergative case in Georgian 193 b. (me) pronom v-naxul-ob-di vano-s mtvral-i I 1-see-ts-past Vano-acc drunk-nom ‘Ii used to see Vano drunki’ Having illustrated that ergative arguments behave as structurally case-marked DPs with respect to subject–verb agreement and case concord, I turn in section 8.5.2 to laying out an account of ergative case assignment in functionally impoverished environments where one functional category, instead of two in nominative contexts, checks structural case of main arguments of transitive clauses.
8.5.2 Ergative as a Dependent Case According to the present analysis, Event is absent in the functional spine of clauses in ergative tenses, and Tense, the locus of reference time information, directly selects vP. If the verbal domain contains one argument, its case is naturally checked by the closest c-commanding functional head, T, and is nominative. But when the vP contains a transitive predicate, two arguments must have their cases checked against the same functional head. I propose that case checking in these circumstances proceeds along the algorithm of Dependent Case, as proposed in Marantz (1991) and further elaborated in Bobaljik (2008), Baker & Vinokurova (2010), Baker (2014a). The central idea of this principle is that a “dependent” structural case may be assigned to one argument if there is a second argument in the same domain that needs its case checked.10 In other words, two nominals in the same domain that have no other structural/lexical means to have their case licensed compete for the same case-checking source (see Bittner & Hale 1996a). The dependent case assigned to the higher of two is ergative. The notion of nominals is to be taken here in a larger sense and should apply to implicit and clausal arguments. The first reason is that in Georgian, the subject of the transitive clause is marked ergative even when the direct object is CP (34). Therefore, the dependent case mechanism should apply not only to DPs per se but to argument positions that nominal may occupy. (34) vano-m tkv-a [CProm bebia-ø ak vano-erg say-aor3s that grandma-nom here ‘Vano said that Grandmother is here’
ari-s] be-pres3sg
Second, some predicates take ergative and dative arguments, so the ergative subject does not “see” a case competitor, as dative is assigned by the applicative head. 10 Specifically, according to Marantz’s (1991) formulation, dependent case is assigned by V+I to a position governed by V+I when a distinct position governed by V+I is: a. not “marked” (does not have lexically governed case) b. distinct from the chain being assigned dependent case. Dependent case assigned up to subject ergative Dependent case assigned down to object accusative
194 Léa Nash (35)
a. vano-m u-Kbin-a nino-s Vano-erg appl-bite-aor3s Nino-dat ‘Vano bit (at) Nino’ b. vano-m u-daraj-a Vano-erg appl-guard-aor3s ‘Vano guarded the house’
nino-s Nino-dat
c. vano-m u-Qvir-a nino-s Vano-erg appl-shout-aor3s Nino-dat ‘Vano shouted at Nino’ These constructions seem problematic if we want to maintain the idea that ergative is dependent on the presence of another element as no other argument is required in (35), as their translations indicate. Yet, the presence of applicative u-implies that there is more than meets the eye in these configurations. u-generally adds possessive/benefactive arguments and can be regarded as a low applicative head that relates two individuals in the sense of Pylkkänen (2008). The sentences in (35) should then be interpreted not as their English translations suggest but rather as events of attribution (causing to have) of the result of the activity to another entity. So (35a) means Vano gave Nino a bite; (35b) Vano gave the house a guard(ing), (35c) Vano gave Nino a shout. I propose that in these constructions the dative possessor is related via the applicative u-to a weak implicit argument, with a cognate object meaning, as in Landau (2010). This implicit argument is syntactically represented as pro and counts as a case-competitor for the DP agent.11 (36) vano-m u-Kbin-a [Nino-s Vano-erg appl-bite-aor3s Nino-dat Lit.: ‘Vano caused Nino to have a bite
u- pro] appl
An independent proof that there is a hidden nominal in (37b–c) is that these predicates are compatible with the adjectival intensifier iseti ‘such … that’ that only modifies nominals, as in (37a). (37) a. nino-m [iset-i Kaba-ø] iQid-a [rom Qvela-ø ga=giž-d-a] N-erg such-nom dress-nom buy-aor3s that all- nom prev=crazybecome-aor3s ‘Nino bought such a dress that everyone went crazy’ b. suares-ma [iset-i pro] u-Kbin-a vano-s [rom Qvela-ø Suarez-erg such-nom appl-bite-aor3s vano-dat that all-nom ga=giž-d-a] prev=crazy-become-aor3s ‘Suarez bit Vano “such” that everyone went crazy’ 11
Referential pronominal argument drop is common in Georgian, in which case the missing argument is represented as pro. These silent pronouns count as visible case-competitors, obviously.
Split ergativity and ergative case in Georgian 195 c. nino-m [iset-i pro] u-Qvir-a vano-s [rom Qvela-ø N-erg such-nom appl-shout-aor3s Vano-dat that all-nom ga=giž-d-a] prev=crazy-become-aor3s ‘Nino shouted “such” at Vano that everyone went crazy’
I therefore conclude that in spite of some apparently problematic configurations where the direct object is not expressed as a “visible” nominal, ergative is always assigned in configurations where T sees two arguments with unvalued case features in the same vP domain and marks the higher of the two with the dependent case. I now turn to the discussion of why accounting for ergative as an inherent case, at least in Georgian and desirably more generally, is less warranted than the present analysis.
8.5.3 Ergative Is Not an Inherent Case The literature on ergativity abounds with analyses that treat ergative as an inherent case assigned within VP confines, by V or little v (Mahajan 1989; Jelinek 1993; Johns 1993; Nash 1995; Butt 1995; Massam 2002; Woolford 2006; Legate 2008). According to Woolford (2006: 12), structural cases are assigned via Agreement that a DP establishes with the closest c-commanding functional head (Chomsky 2000), while inherent cases are associated with certain θ-positions and are typically assigned by a lexical head to the DP base-generated in its specifier. There is good cross-linguistic evidence that ergative can qualify as an inherent case: (a) it is not assigned to derived subjects (Marantz 1991; Legate 2012a); (b) it is assigned to agents, causers, and instruments, i.e. to Proto-Agents in Dowty’s (1991) sense. The link between ergative and agentive participant is stronger than that between nominative and agent (Polinsky in press); (c) morphologically, ergative cross-reference markers are closer to the verb root than absolutive markers (Jelinek 1993 on Lummi). These properties of ergative can also be easily accounted for if it is analysed as a structural dependent case assigned to the highest of vP arguments, along Marantz’s algorithm. Properties (a) and (b) follow from the fact that dependent case is assigned within vP where each structural position is thematically motivated—there is no raising into theta-positions. Property (c) is a natural consequence of ergative as the first case to be assigned when two nominals compete for the case valuation from T. We can predict that in languages where ergative is a dependent case, the second DP also checks its case by T and is sensitive to its finiteness, as attested in Georgian (see section 8.5.1.2). On a more general level, the principal weakness of theories that treat ergative as inherent assigned by v pertains to the identification of the relevant property which enables the verb to assign ergative. If v universally assigns the agent theta role why is it a case assigner only in a subpart of languages, namely ergative languages? Is the reason to be sought in different thematic capacities or in different case-assigning capacities of
196 Léa Nash v in two case systems? Most accounts just state that v in ergative languages is endowed with the special [Case] feature. But if parametric variation hinges on this feature—two systems differ because there is a special case feature available in one system but not in the other system—I conclude that this amounts to a simple restatement of the parameter rather than to its explanation.12
8.5.4 Are Ergative and Nominative Checked by Different Functional Categories? Another line of thought that argues against the inherent nature of ergative and considers both nominative and ergative as structural cases checked by different functional categories has been developed in Murasugi (1992) and Bobaljik (1993b): languages obey Obligatory Case Principle which fixes which case must be obligatorily assigned in the structure—nominative systems set the subject case to be obligatory, while ergative systems identify the object case to be obligatory. This principle is subject to the same questioning; why do languages set the values as they do? If case is not a syntactic primitive but rather a derived notion which serves as a shorthand for a privileged structural relation between a (functional) category and a DP in the clause, OCP can be viewed as a restatement of the problem according to which different functional categories in each language type have certain case features without attributing this asymmetry to independent factors of their respective grammars. The present analysis attempts to pinpoint the triggering factor in one language, Georgian, which is responsible for the difference. This factor pertains to the structural means that a language disposes to link an eventuality to the reference time. Languages may dispose a Tense–Event–vP chain where a dedicated functional category, Event, expresses how the event described by the verb is articulated with respect to the reference time: whether it attains its natural end (perfective), or whether it still goes on (imperfective). In ergative systems, as witnessed by Georgian, this category may be absent and the eventuality is not explicitly located with respect to the reference time by functional means and is therefore understood as some dynamic process that happened (began and ended) without further entailments about attaining the natural telos or about its 12
A reviewer suggests that my account of ergativity in terms of functional deficiency can be made compatible with an analysis of the ergative case as an inherent case assigned by v. Under such an analysis, v universally assigns the inherent ergative case locally, to its specifier. But as in nominative systems Event- v reanalysis takes place, the agent is generated too far from v, in Spec,EventP, which blocks its marking as ergative. Although a theory along these lines is conceptually appealing (because no extra mechanism of dependent case is needed) it fails to explain, without additional assumptions, why the ergative is contingent on finiteness in Georgian. What would prevent an agent in nominalisations to surface with the inherent ergative instead of the genitive? More importantly, such an analysis fails to account for the number agreement triggered by ergative DPs. A way out cannot be sought in Multiple Agree mechanism (see Hiraiwa 2001) that allows not only a lower structurally marked DP to be visible for feature valuation by T, but an intervening inherently case-marked DP too. As shown in (31) and in n. 9, transitive ergative subjects trigger number agreement but nominative objects never do.
Split ergativity and ergative case in Georgian 197 continuation. A direct consequence of this structural deficiency leads to dependent case assignment in a language, as one functional category, T, has to check two unvalued cases of vP arguments. It ensues from this analysis that thematic properties of verbal predicates are identical in both case systems. Yet in nominative languages, thematic licensing of the agent is relegated to the specifier of the category Event via the mechanism of semantic control, as in (22).
8.6 An Alternative Account of Ergative Splits 8.6.1 Nominative Systems as Biclausal Structures On the basis of the analysis of progressive constructions in Basque, Laka (2006a) proposes that an ergative language can shift to a nominative system when the syntactic expression of the tense requires a biclausal configuration. In Basque, a transitive VP in the progressive tense surfaces as a nominalization embedded under a locative PP (38). The main verb ari (be engaged) is intransitive and its only argument is naturally marked with nominative. The nominative argument of ari syntactically controls PRO in the embedded nominalized VP. In a nutshell, Basque clauses in the progressive tense (39) are analysed as their English archaic homologues as in Johni is a(t) [PROi writing the letter], while other tenses, with the ergative transitive subject, are expressed by monoclausal structures. (38) [TPDPi T [VPV ariintrans [PP P [NP N [VPPROi DP Vtrans ] (39) emakume-ai [PROi ogi-a ja-te-n] ari da woman-det bread-det eat-nom-loc engage is ‘the woman is (engaged in) eating the bread.’ [Laka 2006a: 183, ex. 16] Building on Laka’s proposal, Coon (2010b), Coon and Preminger (2012) argue that the shift from ergative to nominative systems should not viewed as a shift of case-assigning features of functional heads but rather as a result of adding structure to clauses with ergative subjects: ergative clauses are monoclausal and nominative ones biclausal. The question may be raised as to whether a biclausal structure is always needed to tie a simple eventuality to the reference time. The authors answer positively to this question and propose that the progressive tense, which syntactically is a manifestation of imperfective aspect, may universally require the nominalization of the eventuality expressed by vP and its subsequent embedding under a locative P: X is in [NPV-ing]. As a consequence of this syntactic representation of imperfectivity, the external argument of the eventuality is realized in the nominalized VP as a PRO and the main verb is intransitive.
198 Léa Nash
8.6.2 Temporal Relations: Biclausal Configurations or Restructuring Importantly, the Laka–Coon line of thought and the present analysis hinge on one central idea: when a language shifts from ergative to nominative, the added structure— another clause or an extra functional category—serves to syntactically control as in (38) or semantically control, as in (22), the highest argument of the transitive predicate. There is no linguistic evidence to analyse Georgian imperfective tenses as biclausal: no auxiliaries, no nominalizations and no locative P are involved in building nominative clauses. The imperfective marker ts, which minimally distinguishes the verbal form in imperfective and non-imperfective aspects, cannot be further split into the nominalizing part and the re-verbalizing affix functioning as the main verb. And although ts appears in nominalizations as shown in (21), it is not a nominalizer by itself—an additional nominalizing affix is required to categorize the verb as a noun. Under a larger perspective, one may question whether an analysis of progressives as biclausal control structures, viable for Basque, can be easily extended to other languages. Does the periphrastic progressive tense in English also involve syntactic control or is the surface subject the argument of -ing verbs rather than the argument of be? Ramchand (2013) provides a convincing support for the second option and shows that -ing forms and main verbs surface in the same position.
8.6.3 There Is a Phase Head above vP in English: Ramchand (2013) Recent work by Ramchand (2013) and Ramchand and Svenonius (2014) on the semantics of functional inventory sheds new light on the embedding of functional categories in English. They argue that several syntactic tests, such as VP-fronting, pseudo-cleft constructions, British non-finite do substitution, show that English progressive -ing forms, -en passives, and main verbs behave differently than perfect participles or modal infinitives in English. This is an indication that these verbal forms originate low in the VP structure. I reproduce here just one test showing that do can substitute for an infinitive modal complement or a perfect participle (40a–b), but not for a progressive or passive participle. (40) a. b. c. d.
John might leave, and Mary might do also. John has left, and Mary has done also. John is leaving, and Mary is (*doing) also. John was arrested, and Mary was (*done) also
Split ergativity and ergative case in Georgian 199 Other tests cited by the authors leave little doubt that the syntactic size of the simple VP and of the progressive VP is identical, which goes against the idea, at least synchronically, that progressive clauses are biclausal configurations, unlike present perfect clauses in (40b). Ramchand (2013) introduces a frontier (signalled in (41) by #) in the functional spine of English clauses, where the perfect and T are separated from the progressive arguing that this cut corresponds to the phase boundary and tries to identify the category that demarcates the phase. (41) [TPT[have] ASPperf[ -en] #ASP[-ing] V[PASS-en] V…] Ramchand labels it as Event, the term that I have borrowed in the present work, and contends that this category “guarantees the sortal shift from event descriptions to temporally enriched situational descriptions and it also determines a choice of subject of predication for the whole complex event, the Topic of Event”. The author claims that the existence of Event does not impose the rejection of the little v, which must thematically license the agent (the initiator) and semantically delineate an eventuality. In this work, I attempt to show that the meaning Ramchand ascribes to this phasal head Event corresponds to the category spelled out by ts in Georgian. In nominative tenses, Georgian, just like English, has topics (syntactically, specifiers) of Events, while in ergative tenses, Georgian has topics of eventualities.
8.7 Conclusion On the basis of the analysis of ergative split in Georgian I have attempted to show that nominative and ergative tenses within the same language, and, by extension, in nominative and ergative languages in general, differ with respect to functional inventories that are available to express temporal relations syntactically in each system. Nominative systems involve a functional category Event, a phasal head above vP, which combines two functions. On the one hand, Event has the properties commonly attributed to Kratzer’s Voice as it thematically licenses the external argument of vP via semantic control with v. On the other hand, it also has the properties commonly ascribed to the Aspect category as it expresses the localization of the eventuality with respect to the reference time in T. The argument in Spec,EventP is marked as nominative by the c-commanding functional category T in nominative languages. Ergative systems dispose a smaller functional clausal spine as they lack the Event head. As a result, in these systems (i) the eventuality stays temporally opaque with respect to the time of utterance, and (ii) the transitive subject is generated in a different position and lower than in nominative tenses: it is thematically licensed by v in Spec,vP and is marked as ergative via the mechanism of dependent case assignment. To conclude, the ergative case on a DP is taken in the system I have elaborated in the present analysis to signal the absence of an intervening
200 Léa Nash functional category between T and v that serves as a thematic licenser of subjects in a nominative language.
Abbreviations 1st, first-person singular; 1o, first-person singular object; 2nd, second-person singular; 3s, third-person singular; ACC, accusative; AOR, aorist; APPL, applicative; ASP, aspect; DAT, dative; ERG, ergative; GEN, genitive; LOC, locative; NEG, negation; NOM, nominative; PAST.P, past participle; PAST, past tense; PL, plural; PRES, present tense; PREV, aspectual preverb; RIAM, reflexive implicit; TS, thematic suffix.
PA RT I I
C HA R AC T E R I ST IC S A N D E X T E N SION S
Characteristics
Chapter 9
Split ergat i v i t y i n sy ntax a nd at morphol o gica l spe l l ou t Ellen Woolford
9.1 Introduction A split ergative case pattern is one in which not all of the subjects that could be marked with ergative case actually are. Two well-known types are person splits where only third person subjects are marked with ergative case, and aspect splits where ergative case is only used in the perfective aspect.1 A language with a split ergative case pattern qualifies as a split ergative language under anyone’s definition, but there is a wide range of opinion among linguists as to what else would qualify a language to be called a split ergative language. Under the narrowest definition, which I favor, a split ergative language necessarily has a split ergative case pattern.2 In contrast, Aldridge (2008a) considers Warlpiri to be a split ergative language, even though its ergative case pattern is not split, nor is its agreement (which follows a purely nominative–accusative pattern). Warlpiri qualifies as a split ergative language under a broader definition wherein an ergative language is classified as split ergative unless both its case and agreement systems are fully ergative. Deal (2015) uses an even broader definition, under which even a language with a fully ergative case and agreement pattern would not qualify as fully ergative unless the language
1
There are many discussions/surveys of types of split ergativity in the literature. Some of the most well-known include Silverstein (1976); Comrie (1978); Dixon (1979, 1994); DeLancey (1981); and Song (2001). 2 This includes languages with a split ergative agreement pattern if the case system (overt or covert) is ergative.
206 Ellen Woolford also had syntactic ergativity.3 Dixon (1994: 161) does not lump morphological and syntactic ergativity together, but instead classifies the morphological and syntactic pattern of a language separately, describing Dyirbal as split with respect to case (because it has a person split), but fully ergative at the syntactic level. Dixon (1994) and McGregor (2009) classify languages as split ergative even if all main clauses are fully ergative, if embedded/dependent clauses are not. Another point of disagreement concerns whether active (Split S) languages count as split ergative.4 Dahlstrom (1983) maintains that active languages are not split ergative, but Dixon (1994) maintains that they are, although of a different type. Each view stems from a different definition of split ergativity. I will add my view here, that there are actually two different types of active languages, which I refer to as active and active ergative, but neither is split ergative. In fact, in my view, active ergative languages are the only ergative languages that are not split ergative. Perhaps the narrowest definition of split ergativity is that of Coon (2013a) who maintains that in a technical sense there is no split ergativity; instead what we describe as a split ergative pattern always involves two different syntactic constructions, only one of which allows/licenses ergative case. The question of how split ergativity is defined becomes important in evaluating the validity of typological claims such as the common statement in the literature that (almost) every ergative language is split ergative (e.g. Silverstein 1976; Moravcsik 1978b; Dixon 1979). It is also theoretically important in that there is a tendency to assume that full ergativity is the basic pattern for an ergative language whereas split ergativity needs to be explained. Sections 9.2 and 9.3 of this chapter focus on person and aspect splits and the literature and controversies concerning the proper theoretical analysis of these two well-known types of ergative splits. Two lesser known types of split ergativity are described in sections 9.4 and 9.5, one in Nepali based on Kratzer’s (1995) distinction between stage and individual level predicates and one in two unrelated languages, Folopa and Mongsen Ao, which is partially determined by social conventions. Section 9.6 attempts to clarify the controversy over whether active languages are split ergative, and the various definitions of split ergativity upon which this controversy is based. Section 9.7 turns a little-known type of ergative split mentioned in Silverstein (1976), but seldom included in subsequent surveys. This type of ergative case split is based on the features of the object. I agree with Silverstein that this is a type of split ergativity, even though the literature generally does not describe the relevant languages as split ergative. Languages with this type of ergative case split include Inuit (Bittner and Hale 1996b), Niuean (Massam 2000), and Nez Perce (Rude 1988; Deal 2015). For Inuit and Niuean, the object feature that matters is specificity, while in Nez Perce it is described as topicality (Rude 1988). Finally, section 9.8 3
A language is said to be syntactically ergative if syntactic rules (appear to) make reference to ergative case. 4 Active languages, also called Split S languages, have two classes of verbs. In one verb class, all subjects, transitive and intransitive are marked alike; in the other class, intransitive subjects are marked like objects.
Split in syntax and at morphological spellout 207 describes an ergative case split based on the proximate/obviate distinction in Sahaptin (Rigsby and Rude 1996).
9.2 Person/Animacy/NP Splits Person splits are often referred to as animacy splits or NP splits since they are not limited to person but can involve other features of NPs such as animacy. Person/animacy/ NP splits are “almost universal in Australian languages and well-attested elsewhere (in North America, and in some Siberian and Tibeto-Burman languages)” according to DeLancey (1981: 628). In the most common type of person split, ergative case marks only third person subjects, never first or second. We can illustrate this person split with examples from Marathi. There is a contrast between example (1) where the third person subject is marked with ergative case and example (2) where the first person subject is not marked with ergative case:5 (1) Ram-ne acɘvlɘ. Ram-erg washed.hands.neut.3sg ‘Ram washed his hands.’ (2) Mi acɘvlɘ. I washed.hands.neut.3sg ‘I washed my hands.’
[Marathi] (Comrie 1984: 862 (15)) [Marathi] (Comrie 1984: 862 (16))
9.2.1 The Silverstein Hierarchy and Markedness Discussions of person/animacy/NP splits always include Silverstein’s (1976) classic work on this topic, and his proposal that in such splits, ergative case marks subjects whose features are low the person/animacy hierarchy: (3) Person/Animacy Hierarchy (simplified) 1st > 2nd > animate > inanimate For example, one can characterize the Marathi person split shown in (1) and (2) by drawing a line between 2nd person and human, indicating that ergative case marks the 5 Note that the agreement in the Marathi example in (1) is the default 3rd person neuter which occurs when there is no nominative in the clause, and this default agreement also occurs in (2), indicating that the first person subject in (2) has ergative case in syntax, where agreement features are determined. This is an example of a purely morphological ergative split. (See the discussion in section 9.2.2.)
208 Ellen Woolford lower ranked third person subjects, but not the higher ranked first or second person subjects: (4) Marathi pattern 1st > 2nd │> animate > inanimate nominative ergative ------------→ However, not all person splits can be characterized by dividing this standard hierarchy into two neat parts. Silverstein discusses several more complex patterns. Dhirari marks all subjects ergative except 1st and 2nd person plural subjects. This is unexpected under the simple version of the hierarchy as outlined. Describing the Dhirari pattern requires a different hierarchy, with 1st and 2nd person plurals at the top/left. Using this (language specific) hierarchy, it is possible to characterize the pattern as divided into two neat parts, where ergative case marks only the lower elements: (5) Dhirari pattern 1st pl > 2nd pl │> 1st sg > 2nd sg > animate > inanimate nominative ergative ---------------------------------→ Silverstein describes a slight variant of the Dhirari pattern which occurs in Bandjalang, where all subjects except 1st plural are marked with ergative case. This pattern can be described using the same hierarchy as in Dhirari, but moving the line leftward: (6) Bandjalang pattern 1st pl │> 2nd pl > 1st sg > 2nd sg > animate > inanimate nom ergative ----------------------------------------------→ Silverstein (1976: 127) describes a more complex in Aranda, where only 1st singular pronouns and 3rd person inanimates are marked with ergative case (although Legate (2014a) points out based on the original data source that 3rd animates are also marked ergative in this language). Marking the lower inanimates with ergative case (or all third persons), but nothing higher would be an expected pattern under the animacy hierarchy approach; the problem is the fact that the very high 1st singular is also marked with ergative case: (7) Aranda pattern (with the correction noted in Legate (2014a)) 1st sg > │ 1st pl > 2nd sg > 2nd pl │> animate > inanimate ergative nominative ergative -------------→ Silverstein describes Aranda as having two separate split systems of ergative marking. That is, the pattern of this language can be described using two separate hierarchies, one separating local persons )1st and 2nd from 3rd persons, and one involving only local persons, which places 1st singular at the bottom:
Split in syntax and at morphological spellout 209 (8) Two hierarchy Approach to Aranda local │> 3rd nom ergative 2nd pl > 1st pl > 2nd sg │> 1st sg nominative ergative It is clear from Silverstein’s discussion that he does not claim to account for all person/ animacy/NP splits in all languages, but rather he proposes an approach to this kind of split as the basis for further work: “There are numerous ‘holes’ in the pattern and these mean that we have the opportunity for further constraint of the system” (1976: 125).6 Two important points that Silverstein makes are that ergative splits are not random and that the patterns of NP splits have to do with markedness.
9.2.2 Person/Animacy/NP Splits Occur at Morphological Spellout In the introduction to this section, I pointed out that the person split in Marathi is not present in syntax, based on the fact that it has no effect on agreement. Instead, this person split, and perhaps all person splits are only splits at the morphological level. Goddard (1982) argues that there is no split in ergative case in Australian languages, but only a split in ergative case marking, and he views these as instances of “case homonymy.” Baerman et al. (2005) argue more generally that all ergative splits of the person/ animacy/NP type are instances of case syncretism, which they define broadly to include any situation in which a case feature present in syntax is not morphologically realized/ distinguished. Legate (2014a: 183) presents detailed arguments for this position, that “split ergativity based on nominal type is a morphological phenomenon, not a syntactic one.” Although Legate notes that the formal analysis of this type of splits depends on one’s theory of morphology, she gives a formal account of this type of ergative split in the framework of Distributed Morphology. In that framework, there is a level between syntax and morpheme insertion where certain operations on features can take place. Legate proposes that at this level, the ergative case feature on the subject in syntax is deleted in certain contexts, such as in the presence of a local person feature. As a result, the ergative case feature is not morphologically spelled out/distinguished, in that context. Under this view, person/animacy/NP splits involving ergative case are much like what one might call a ‘gender split’ in English pronouns, where gender is morphologically marked in third person pronouns (he, she, it), but not in first or second person 6 Comrie (1981a) notes that although the Saibai dialect of Kalaw Lagaw Ya comes close to matching Silverstein’s hierarchy predictions, there are exceptions. One exception is that proper nouns behave as if they are higher ranked than pronouns. Additional exceptions to the strict predictions of the animacy hierarchy are pointed out in Filimonova (2005) and Legate (2014a).
210 Ellen Woolford pronouns.7 A more direct parallel to ergative person splits is the dative person split in Italian wherein pronominal clitics are only distinctly marked for dative case in the third person: (9)
Dative split in Italian pronominal clitics 1st mi accusative/dative 2nd ti accusative/dative 3rd lo accusative.masculine 3rd gli dative.masculine
9.3 Aspect/TAM Splits We can illustrate an aspect split with the Marathi examples in (10) and (11), where ergative case is used in the perfective aspect, but not in the imperfective. (10) Ravi-ni kavitaa vaac-l-i. Ravi-erg poem read-perf-3sg.fem ‘Ravi read the poem.’ (11)
Ravi kavitaa vaac-t-o. Ravi.nom poem read-impf-3sg.masc ‘Ravi reads a poem.’
[Marathi] (Gair and Wali 1988: 96 (18b)) [Marathi] (Gair and Wali 1988: 96 (18a))
Splits of this type are often called TAM splits because they are said to involve tense, aspect, or mood (e.g. Dixon 1994). However, some recent literature questions whether tense or mood splits actually exist. Coon (2013a: 189) cites literature suggesting that purported examples of tense splits are better analyzed as aspect splits. Salanova (2007: 47) in particular states this conclusion directly: “so-called tense-aspect-mood splits essentially boil down to aspectually conditioned splits.” Salanova also doubts that mood splits exist, as does Coon who notes that mood splits often involve imperatives and argues that all mood splits are “reducible to a clause type split” (2013a: 189). Of course, some would count clause type splits as ergative splits, but Coon’s position is that the syntactic structure of the two clauses in such splits differs such that one would not expect ergative case to be assigned in one of them. In Marathi and other languages which have agreement, it is clear that the aspect split is present in syntax, given that it affects the agreement. Note that the agreement in (10) is 3rd feminine, agreeing with the feminine nominative object, poem, while the agreement in (11) is 3rd masculine, agreeing with the nominative subject, Ravi. 7 A reviewer notes that in other languages with this gender split in pronouns, such as French and Russian, there is evidence in the form of predicate adjective agreement and participle agreement that gender features are present in syntax on first and second person pronouns.
Split in syntax and at morphological spellout 211 Coon (2013b) has an excellent recent survey of examples of ergative aspect splits, including some less often cited examples such as Samoan. Coon (2013b: 184) gives the following example from Milner (1973): (12) Na va’ ai-a e le tama le i’a. past look.at-perf erg the boy the fish ‘The boy spotted the fish.’ (13) Na va’ai le tama i le i’a. past look.at the boy at the fish ‘The boy looked at the fish.’
[Samoan] (perfective aspect)
(imperfective aspect)
While many examples of ergative aspect splits are like that of Hindi, shown in (10) and (11), which manifests a split in the case marking of transitive clauses depending on the aspect, some of the Mayan languages have an aspect split which is limited to intransitive clauses, as we see in the Chol examples from Coon (2013a) ((14)–(15)). The transitive pair in (14) and (15), which differ in aspect, both show the same agreement pattern: the set A prefix cross-references the subject and the set B suffix cross-references the object: (14) Tyi i-jats’-ä-yoñ. perf A3-hit-tr-B1 ‘She hit me.’
[Chol] (Coon 2013a: 11 (15a))
(15) Choñkol i-jats’-oñ. prog A3-hit-B1 ‘She’s hitting me.’
(Coon 2013a: 11 (16a))
In contrast, the intransitive pair in (16) and (17) are different in agreement marking. The subject in the perfective example in (16) is marked like the transitive object, while the subject in the imperfective example in (17) is marked like the transitive subject: (16) Tyi majl-i-yoñ. perf go-intr-B1 ‘I went.’
[Chol] (Coon 2013a: 11 (15b))
(17) Choñkol i-majl-el. prog A3-go-nml ‘She’s going.’
(Coon 2013a: 11 (16b))
There is currently no standard view as to how aspect splits should be analyzed. Existing approaches differ as to whether they situate some or all aspects in syntax or at morphological spellout. Davison (2004b) proposes a differential licensing approach for Hindi, postulating that the features/conditions required to license ergative case in syntax
212 Ellen Woolford are different in different aspects, and Ura (2006) pursues this type of approach as well. Woolford (2013) explores a differential faithfulness approach, drawing a parallel between aspect splits in syntax and positional/contextual faithfulness effects in phonology, e.g. Beckman (1998). Coon (2013a, 2013b) argues for a differential structures approach, under which ergative case cannot be licensed in imperfectives in some languages because the structure/verb is one that does not take an external argument (Coon 2013a, 2013b). Coon’s approach is partially motivated by Laka’s (2006) account of the lack of ergative case in the Basque progressive, as in (18), in contrast to the presence of ergative case on the subject with the same verb in other contexts such as the simple present in (19): (18) Emakume-a ogi-a jaten woman-det bread-det eating ‘The woman is eating (the) bread.’ (19) Emakume-a-k ogi-a woman-det-erg bread-det ‘The woman eats (the) bread.’
ari da. prog is jaten du. eating has
[Basque] (Laka 2006a: 173 (1b))
(Laka 2006a: 173 (1a))
Although this is not the familiar ergative aspect split between perfective and imperfective aspect, it does involve an aspect, the progressive. Laka argues that morpheme, ari, glossed as progressive aspect in the example in (18), is really a main verb, with a meaning something like ‘engaged in,’ and that this progressive example actually has a biclausal construction, as shown in (20): (20) Emakume-a ogi-a ja-te-n ari da. woman-det bread-det eat-nom-loc engaged is ‘The woman is (engaged in) eating the bread.’
[Basque] (Laka 2006a: 174 (2c))
Supporting evidence that verbs in the higher clause does not take an external argument, and cannot license ergative case, is the presence of the be auxiliary, which occurs in unaccusative constructions in Basque, as in (21), rather than the have auxiliary, which occurs with verbs that do take an external argument, as in the example with an ergative subject in (19). (21)
Emakume-a hurbil-tzen woman-det get/near-impf ‘The woman gets closer.’
da. is
[Basque] (Laka 2006a: 179 (7))
Coon (2013a) analyzes the aspect split in the agreement pattern of the Chol examples as involving very different syntactic structures in the two aspects: the perfective examples have an ordinary clausal structure, but the proposal is that the imperfective examples involve a possessive construction, so that a better translation of the progressive example in (17), repeated as (22) would be something like ‘Her going continues’:
Split in syntax and at morphological spellout 213 (22) Choñkol i-majl-el. prog A3-go-nml ‘She is going.’
[Chol] (Coon 2013a: 11 (16b))
In contrast, this aspect split, which occurs in some Mayan languages, but not others, is treated as a purely morphological phenomenon under the approach in Woolford (2013), involving a choice of which of the two cross-referencing series, A (true agreement) or B (a pronominal clitic), to use for an intransitive subject. An important generalization that the correct formal approach to aspect splits should capture is the correlation between ergative case and the perfective aspect in such splits: (23) “If a split is conditioned by tense or aspect, the ergative marking is always found either in past tense or in perfective aspect (Dixon 1994: 99).” Coon’s approach will capture this generalization if the hypothesis holds cross-linguistically that imperfectives in such splits always involve the verb ‘be,’ or some other verb or construction does not take an external argument, and can thus not license ergative case. As Coon notes, this approach is similar to that of Tsunoda (1981b) who tries to unite aspect splits with active–stative languages, arguing that both involve alternations between types of verbs which do and do not take agentive/external arguments.
9.4 Nepali: A Split Based on Stage versus Individual Level Predicates A new type of ergative split is described by Butt and Poudel (2007) in Nepali. The split involves the distinction between individual level and stage level predicates described in Kratzer (1995). The following minimal pair illustrates this ergative split:8 (24) Raam-le angreji jaan-da-cha. [individual level predicate, Nepali] Ram-erg English know-impf-nonpast.masc.3sg ‘Ram knows English.’ (Butt and Poudel 2007 (16a)) (25) Raam (aajaa) angriji bol-da-cha. [stage level predicate, Nepali] Ram.nom (today) English speak-impf-nonpast.masc.3sg ‘Ram will speak English (today).’ (Butt and Poudel 2007 (16b)) An individual level predicate “predicates a lasting/inherent property of a referent” (Butt and Poudel 2007: 4). In contrast, a stage level predicate “says something about a property 8
There is also a complex pattern of optionality of ergative marking in Nepali described in Li (2007).
214 Ellen Woolford of a referent that holds for a slice of that referent’s spatio-temporal existence” (Butt and Poudel 2007: 4). Additional evidence that Nepali makes a distinction between stage and individual level predicates comes from copular clauses, where are different copulas for the two types of predicates: (26) Saru Bhakta kabi hun. [individual level predicate, Nepali] Saru Bhakta poet be.nonpast.3masc.hon ‘Saru Bhakta is a poet.’ (Butt and Poudel 2007 (13)) (27) Saru Bjalta aaja khusi Saru Bhakta today happy Saru Bhakta is happy today.
chan. [stage level predicate, Nepali] be.nonpast.3masc.hon (Butt and Poudel 2007 (14))
In other types of clauses, lacking a copula to mark the type of predicate, ergative case functions to mark the predicate as individual level, as in (24), while the absence of ergative case indicates a stage level predicate, as in (25). However, this pattern is present only in the imperfective aspect/nonpast tense. Nepali is like Marathi in that in the perfective aspect/past tense, all the verbs that can take an ergative subject do so, regardless of whether the predicate is stage or individual level. The fact that Nepali has two cross- cutting ergative splits, one based on aspect and one based on predicate type, may present an additional challenge (or perhaps an additional clue) for those who attempt a formal account of this pattern.
9.5 Socially Controlled Ergative Case Splits Dixon (1994) discusses an ergative case split which occurs in two unrelated languages, Folopa (Anderson and Wade 1988) and Mongsen Ao (Coupe 2008). This type of ergative case split appears to be determined by social factors; using the ergative morpheme is interpreted in certain contexts as asserting or emphasizing the subject’s control/ownership of the object, which can be rude in some social situations, and indicates theft in others.
9.5.1 Folopa Anderson and Wade (1988) describe a language of Papua New Guinea, Folopa, which is a counterexample to the following claim in Dixon (1979): while languages can be Fluid-S with respect to intransitives (using either ergative or nominative case for subjects), no language is Fluid-S with respect to transitive clauses. Dixon (1994) modifies his typology and classifies as Folopa as Fluid-S in both intransitive and transitive clauses.
Split in syntax and at morphological spellout 215 Although Folopa does allow a choice of an ergative or a nominative subject for most transitive verbs, the primary usage pattern according to Anderson and Wade (1988) is to use ergative case. Using/spelling out the ergative case morpheme is the most frequent choice in texts and in judgments of sentences out of context. Anderson and Wade point out that ergative is the normal choice even if there is no possibility of ambiguity as to which argument is the subject, as in the following example: (28) U so-né ama wa teo doko-ta-pό. [Folopa] that woman-erg her string.bag unfinished weave-pres-ind ‘That woman is weaving her unfinished string bag.’ (Anderson and Wade 1988: 6 (16)) However, in situations involving food, using an ergative subject carries an implication of stinginess, because marking the subject with an ergative morpheme adds an interpretation an overt assertion of control or ownership. With respect to food, the norm is to share. “If a person uses the ergative and thus states his control, he will be considered stingy” (Anderson and Wade 1988: 11). Thus the socially correct version of the statement in (29) uses a nominative subject. (29) ę o faa-ta. I.nom sago break.open-pres ‘I am breaking this sago.’
[Folopa] (Anderson and Wade 1988: 11 (23))
In contrast, using I.erg yąlo instead of I.nom ę in the example (29) would be asserting control/ownership over the food and would be considered stingy. In situations involving things the speaker does not own, using ergative case on the subject also asserts control/ownership. In the version of the statement in (30), the speaker uses a nominative subject to indicate that he is not asserting control/ownership of his brother’s sago palm, i.e. he mistakenly cut it down. In the second version in (31), the speaker uses an ergative subject to indicate that he is asserting control/ownership, in that he intentionally cut down his brother’s sago palm. (30) No-ό kale naaǫ o make ę di-ale-pό. [Folopa] brother-voc the your sago young I.nom cut.down-past-ind ‘Brother, I (mistakenly) cut down your young sago tree.’ (Anderson and Wade 1988: 7) (31) No-ό naaǫ o make yąlo di-ale-pό. brother-voc your sago young I.erg cut.down-past-ind ‘Brother, I (intentionally) cut down your young sago tree.’ (Anderson and Wade 1988: 7) In section 9.5.2 we see this same split in an unrelated language, Mongsen Ao.
216 Ellen Woolford
9.5.2 Mongsen Ao The Mongsen dialect of Ao, a Tibeto-Burman language described in Coupe (2008), has an ergative split that looks very much like what we saw in Folopa. As in Folopa, using ergative case in certain contexts, involving food and other people’s belongings, is taken to indicate that the speaker claims or emphasizes ownership, which is rude in certain contexts, and indicates theft in other contexts. The following minimal pair is given by Coupe (2008) to illustrate a situation in which using/pronouncing ergative case on the subject indicates that the subject is willfully stealing. The first version of the sentence lacks an ergative case morpheme and lacks this interpretation; in contrast the second version has an overt ergative case morpheme and has the interpretation of willful theft: (32) A-hɘn a-tʃak tʃàɁ-ɘ̀ɹ-ùɁ. nrl-chicken nrl-paddy consume-pres-dec9 ‘The chickens are eating paddy.’
[Mongsen Ao] (Coupe 2008: 157 (5.22))
(33) A-hɘn nɘ a-tʃak tʃàɁ-ɘ̀ɹ-ùɁ. [Mongsen Ao] nrl-chicken erg nrl-paddy consume-pres-dec ‘The chickens are eating paddy.’ (implying that they are stealing it) (Coupe 2008: 157 (5.23)) Mongsen Ao is related to the Tibetan languages where ergative case marking is said to be optional and often used to disambiguate sentences. It seems likely that the morphological spellout of the ergative case feature in Folopa and Mongsen Ao is optional with respect to the grammar, and thus available for uses determined by the social context. Mongsen Ao appears to have another ergative split as well, one of the type described earlier for Nepali involving stage versus individual level predicates. We see this is the following minimal pair from Coupe (2008: 156). The first example, in (34), has a stage level predicate, one that refers to something that the subject is doing at the moment (chopping wood). The second example, in (35), has an individual level predicate that refers to a property of the subject, his occupation. Paralleling the pattern we saw in Nepali, only the example with an individual level predicate has ergative on the subject: (34) Nì a-sɘ́ŋ sɘ-ɘ̀ɹ. 1sg nrl-wood chop-pres ‘I’m chopping wood.’
9
[Mongsen Ao] (Coupe 2008: 156 (5.20))
The gloss nrl abbreviates ‘non-relational prefix,’ and the gloss dec abbreviates ‘declarative mood clitic.’
Split in syntax and at morphological spellout 217 (35) Nì nɘ a-sɘ́ŋ sɘ-ɘ̀ɹ. 1sg erg nrl-wood chop-pres ‘I chop wood.’ (i.e. habitually, as an occupation)
(Coupe 2008: 156 (5.21))
9.6 Are Active Languages Split Ergative? Active languages (often called active–stative languages) have two classes of intransitive verbs whose case or agreement marking patterns differ. The active pattern is usually described as follows: verbs in one class case mark or cross-reference intransitive subjects like transitive subjects, while verbs in the other class case mark or cross-reference intransitive subjects like transitive objects. An example of an active language is Choctaw. Intransitive verbs in the more active class (verb that take an external argument/initial 1/initial subject) cross-reference their subjects with what is labeled Series I in the Choctaw literature, as in (36). Intransitive verbs in the more stative class (which do not take an external argument/initial 1/initial subject) cross-reference their subjects with Series II, as in (37): (36) Hilha-li-tok. dance-1sg.seriesI-past ‘I danced.’ (37) Sa-niya-h. 1sg.seriesII-fat-tns ‘I am fat.’
[Choctaw] (Davies 1986: 14 (1a))
(Broadwell 2006: 33 (9))
In most transitive verbs, Series I cross-references subjects, while Series II cross- references objects: (38) Chi-pí-li-h. 2sg.seriesII-see-1sg.seriesI-tense ‘I see you.’
[Choctaw] (Broadwell 2006: 45 (71))
However, there are a few transitive verbs which do not take an external argument, and these verbs cross-reference both subject and object with Series II: (39) Chi-sa-yimmi-h. 2sg.seriesII-1sg.seriesII-believe-pred ‘I believe you.’
[Choctaw] (Davies 1986: 77 (30))
218 Ellen Woolford The generalization is that external arguments are cross-referenced with Series I and internal arguments are cross-referenced with Series II. The case marking pattern in Laz is also classified typologically as active (e.g. Dixon 1994; Song 2001), although here the pattern involves case rather than cross-referencing. In Laz, one class of verbs marks all subjects with ergative case, even intransitives as in (40), while in the other verb class, intransitive subjects get nominative case, as in (41): (40) Bere-k imgars. child-erg 3sg.cry ‘The child cries.’
[Laz] (Harris 1985: 52 (38))
(41) Bere oxori-s doskidu. child.nom house-dat 3sg.stay ‘The child stayed in the house.’
(Harris 1985: 53 (49))
There is disagreement in the literature as to whether active languages qualify as split ergative. As we will see in this section, this disagreement is due to a difference of opinion concerning the definition of split ergativity.
9.6.1 Dahlstrom 1983: Active Languages Are Not Split Ergative Dahlstrom (1983) argues that active languages are not split ergative. This conclusion is based on the view that a case or agreement pattern qualifies as split ergative only if the same verb manifests a nominative–accusative pattern in one context, but an ergative pattern in another context. This disqualifies active languages because each verb class always marks subjects in the same way, regardless of the context.10
9.6.2 Dixon 1994: Active (Split S) Languages Are Split Ergative In contrast, Dixon (1994) argues that active languages are split ergative. To understand why Dixon disagrees with Dahlstom, we need to understand the details of the pattern- based definition of split ergative that Dixon is using: if a language has both nominative– accusative case patterns and ergative case patterns, it qualifies as split ergative in Dixon’s view. It is important here to note that Dixon’s strictly pattern-based definition is 10 Some verbs appear to be members of both verb classes, but there is typically a difference in meaning involving control or intentionality. Languages with many such verbs are referred to as Fluid-S in the typological literature, e.g. Dixon (1994).
Split in syntax and at morphological spellout 219 independent of the actual identity of the cases involved. Under that definition, an active language such as Laz does qualify as split ergative because the verb class that uses ergative case for both intransitive and transitive subjects has a nominative–accusative case pattern in Dixon’s terms (because S and A are marked alike), while the verb class which use ergative case only for transitive subjects, has as an ergative case pattern (because S and O are marked alike):11 (42) Active ergative case patterns under Dixon’s definition of split ergative: Verb class 1: intransitive: transitive: Verb class 2: intransitive: transitive:
S-ergative A-ergative S-nom/abs A-ergative
[pattern: nominative–accusative] O-nom/abs [pattern: ergative–absolutive] O-nom/abs
However, Dixon still recognizes a difference between active languages and other split ergative languages: he considers active languages to be a distinct type of split ergative language which he calls Split S. Dixon’s motivation for using a purely pattern-based definition of ergativity and split ergative is sound from a typological perspective, especially given that at the time this approach was developed, in the 1970s, syntactic theory was in its infancy and provided little or no guidance in determining the identity of different cases, especially ergative. Moreover, there was also considerable disagreement in the descriptive literature as to how to label cases. In that situation, it was entirely reasonable to conclude that the only reliable way to type case systems would be in terms of their patterns. Moreover, that kind of typology is relevant to a question frequently explored in the typological literature: how do different languages mark/ differentiate subjects and objects? However, to answer the question of what syntactic and morphological conditions contribute to the distribution of ergative case, a question that I wish to explore, it is helpful if patterns described as ergative actually have ergative case. It is confusing if verbs or languages that mark all subjects with ergative case are coded as having a nominative–accusative case pattern. It would be better if we had separate terms for patterns versus cases. For example, since a nominative–accusative case pattern is simply one that marks all subjects alike, we could refer to this simply as a subject pattern or a subject– object pattern. Since ergative is a case, but absolutive is not (Legate 2008), we could refer to the ergative–absolutive pattern as simply an absolutive pattern.
9.6.3 Two Types of Active Languages: Neither Is Split Ergative I agree with Dahlstrom (1983) that active languages do not qualify as split ergative. However, I argue that Choctaw and Laz are really two different types of languages. I reserve the label 11 These two verb classes are sometimes unfortunately labeled unergative and ergative/unaccusative, which is very confusing since in active languages with ergative case, it is the unergative class that takes an ergative subject.
220 Ellen Woolford ‘active’ for languages such as Choctaw, and label languages such as Laz as active ergative (or simply ergative) for reasons which will become clear in the following.
9.6.3.1 Active Languages Under a strictly case-based definition, a language cannot be split ergative unless it is ergative, and it is not ergative unless it uses ergative case. I argue that active languages such as Choctaw are not split ergative because they are not ergative. They do not use ergative case. The case pattern of Choctaw is nominative–accusative (Broadwell 2006):12 (43) John-at tákkon(-a) chopa-h. John-nom peach(-acc) bought-tense ‘John bought a peach.’
[Choctaw] (Broadwell 2006: 39)
(44) Hattak-at alla-yã towa(-yã) ĩ-pila-tok. man-nom child-acc ball(-acc) appl-throw-past ‘The man threw the child the ball.’
(Davies 1986: 7)
I argue in Woolford (2010) that the difference between the Series I and Series II cross- referencing forms is that Series I is ordinary agreement, while Series II consists of nominative and accusative pronominal clitics, which are not morphologically distinguished by case. In Choctaw, ordinary agreement cross-references external arguments, while a nominative or accusative clitic is used to cross-reference internal arguments. Thus there is no disconnect between the case of arguments and the case of the cross-referencing elements in Choctaw.
9.6.3.2 Active–Ergative Languages: Fully Ergative In contrast to Choctaw, Laz uses ergative case and thus is an ergative language. Nevertheless, in my view, the Laz case pattern is not a split ergative pattern. Rather, Laz has the most fully ergative case pattern possible, where every verb that can license ergative case does so, regardless of the context, and regardless of whether or not there is an object in the clause. Another active ergative language is Warlpiri. Verbs in the verb class that can license ergative case always mark their subjects ergative, while verbs in the other verb class which cannot license ergative case never mark their subjects with ergative case. In (45) and (46), we see an example of each verb type, in roughly the same transitivity context, where an oblique object is present in the clause: (45) Ngajulu-rlu ka-ma-rla jurlarda-ku me-erg pres-1sg-3sg honey-dat ‘I am looking for honey.’ 12
warni-mi. [Warlpiri] look.for-nonpast (Bittner and Hale 1996b: 561 (49a))
The accusative morpheme need not be spelled out in Choctaw when the object is adjacent to the verb (Broadwell 2006).
Split in syntax and at morphological spellout 221 (46) Ngarrka ka-rla wangka-mi kurdu-ku. man.nom pres.3sg-3sg speak-nonpast child-dat ‘The man is speaking to the child.’ (Bittner and Hale 1996b: 561 (49b)) However, not even an oblique object is required in order for the subject to take ergative case in Warlpiri; verbs in the ergative case licensing verb class take an ergative subject even when no object is present: (47) Ngarrka-ngku ka yunpa-rni. man-erg pres sing-nonpast ‘The man is singing.’
[Warlpiri] (Hale 1982: 237 (30a))
Verbs that take an ergative subject have an external argument (called an initial 1 or an initial subject in other frameworks), and such verbs are classified as transitive under the Hale and Keyser (1993) definition, where all verbs that take an external argument are classified as transitive (whether or not an object is present in syntax, under their assumption that such verbs always have an object at the argument structure level). The connection between external arguments and ergative case is expected under the view that ergative is an inherent case like dative, licensed to external arguments in connection with theta marking (see Woolford 2006 and references cited therein). However, it is important to note that the class of verbs that takes an ergative subject/external argument varies from language to language, and is not entirely predictable from the semantics of the verb.
9.7 Object Shift Split Ergative In this section, we turn to a type of ergative split which is mentioned in Silverstein (1976: 125) but seldom found in other surveys of types of split ergativity. In this type, the case marking of the subject depends on the features of the object. Bittner and Hale (1996b) contrast Warlpiri, where all verbs with an external argument mark their subjects with ergative case, with Inuit, where the subject is ergative only when a [+specific] direct object is present in the clause. Based on evidence in Bittner (1994), they argue that specific objects move out of the VP in Inuit, but non-specific objects remain in situ inside the VP. Constructions with a specific object have an ergative subject, as in (48), while constructions with the same verb, but a non-specific object have a nominative subject, as in (49): (48) Juuna-p miiqqa-t paar(i-v)-a-i. [Inuit] Juuna-erg child-pl.nom look.after-ind-[+tr]-3sg.3pl ‘Juuna is looking after the children.’ (specific object)(Bittner and Hale 1996b: 544 (22b))
222 Ellen Woolford (49) Juuna atuakka-nik marlun-nik. pi-si-v-u-q Juuna.nom book-pl.ins two-pl.ins paid.for-get-ind-[-tr]-3sg ‘Juuna bought two books.’ (non-specific object) (Bittner 1994: 72 (45b)) This type of movement where the object moves out of the VP to a position below the subject is often called object shift; it is well known from work on Icelandic and other Germanic languages (e.g. Holmberg 1986; Vikner 1991; and Diesing 1996). Other languages where the subject gets ergative case only if object shift occurs include Nez Perce (Woolford 1997; Deal 2013) and Niuean (Massam 2000). These two languages differ from Inuit in that both the shifted and unshifted objects are clearly direct objects. We see this in the Nez Perce example: (50) Háama-nm pée-'wi-ye wewúkiye-ne. man-erg 3.3-shoot-asp elk-obj ‘The man shot an elk.’ (topical object) (51)
Háama hi-'wí-ye wewúkiye. man 3-shoot-asp elk ‘The man shot an elk.’ (non-topical object)
[Nez Perce] (Rude 1988: 552 (30))
(Rude 1988: 552 (31))
Niuean provides word order evidence that the object has moved out of the VP in ergative subject constructions. Both constructions have VP fronting, but in the VSO version in (52), the specific object moves before the VP fronts, while in the VOS version in (53), the non-specific object remains inside the VP and fronts with it (Massam 2000): (52) Ne inu e Sione e kofe. past drink erg Sione nom coffee ‘Sione drank the coffee.’ (specific object) (53)
Ne inu kofe a Sione. past drink coffee nom Sione ‘Sione drank coffee.’ (nonspecific object)
[VSO Niuean]13 (Massam 2000: 98 (2a)) [VOS [Niuean]] (Massam 2000: 98 (2b))
Although it is true that the presence of an ergative subject correlates with a feature of the object in this kind of ergative split, as Silverstein (1976) states, the relationship between the object features and the subject features is indirect: the features of the object determine whether or not it will undergo object shift, and in turn, moving the object to a position outside the VP results in the use of ergative case on the subject. There are currently 13
The ergative case morpheme for proper nouns in Niuean, e, happens to look just like the nominative/ absolutive case morpheme for common nouns, e.
Split in syntax and at morphological spellout 223 three hypotheses in the literature as to why ergative case is only used when the object moves out of the VP in this type of language: Bittner and Hale (1996b) propose that in this type of language, VP is a barrier, and that the object has to move across that barrier in order to be in the same domain as the subject, so as to serve as a case competitor for the subject, so as to allow/force ergative case to be licensed on the subject. Deal (2013) proposes that in Nez Perce, moving the object out of the VP causes the agreement to be portmanteau, which in turn allows/causes the subject to be ergative. Woolford (2015) proposes that moving the object above the base position of the subject/external argument interferes with the ability of Infl/T to probe past it to get to the base position of the external argument to assign it nominative case. Using ergative case on the subject instead, as a last resort, avoids this locality violation. Since ergative case is used only when object shift occurs in this type of ergative split, I refer to this as an object shift ergative split.
9.8 Proximate Obviate Ergative Case Split In this last section, we look at another type of ergative case split that is not often included in typological surveys of split ergativity. This type of split is based on the proximate/ obviate distinction.14 This proximate/obviate ergative split occurs in Sahaptin. Sahaptin clearly qualifies as a split ergative language in that subjects of some transitive clauses have nominative case, while subjects of other transitive clauses have ergative case. This ergative split is governed by whether the subject is proximate or obviate: proximate subjects are nominative as in (54), while obviate subjects are ergative as in (55): (54) Ɨwínš i-q̓ínun-a wapaanƚá-n. man.nom 3-see-past grizzly-obj ‘The man (proximate) saw a grizzly (obviate).’
[Sahaptin] (Rigsby and Rude 1996: 677 (21))
(55) Ɨwínš-in pá-tuχnana yáamaš-na. man-obv.erg 3.3-shot mule.deer-obj ‘The man (obviate) shot a mule deer (proximate).’ (Rigsby and Rude 1996: 676 (13)) Now, languages generally only mark the proximate/obviate distinction in clauses with two third person arguments. However, ergative case can mark the subject when the 14 The proximate argument is an established referent that is essentially the main character/focus of empathy in a narrative or span of discourse. An obviate argument is some other character.
224 Ellen Woolford object is first or second person. In this situation, Sahaptin uses different forms of the ergative case morpheme, one with and one without the feature [+obviate]: (56) Ɨwínš-nɨm=nam i-q̓ínu-ša. man-erg=2sgcl 3-see-imperf ‘The man sees you.’
[Sahaptin] (Rigsby and Rude 1996: 677 (22))
Thus one might say that Sahaptin qualifies as split ergative in a second, unusual sense as well: its ergative case morpheme is split into two.
9.9 Conclusion This chapter has described a range of types of ergative splits, including the well- known person and aspect splits, as well as several less well-known types. I have sided here with Dahlstrom’s (1993) view that active languages are not split ergative; however, I have divided active languages into two very different types: active languages, which are not ergative at all, and active ergative languages such as Laz and Warlpiri that are ergative. I have taken the unusual position of defining the ergative pattern in active ergative languages as the only fully ergative case pattern there is, based on the following definition of a fully ergative case pattern: the case pattern of a language is fully ergative if all verbs that can license ergative case do so in all contexts and regardless of transitivity. I support Silverstein’s (1976) view that there is a type of split ergative language wherein the ergativity of the subject depends on the features of the object, and illustrate this type with Inuit, Nez Perce, and Niuean. Following Massam (2000) and Deal (2013), I argue that this type of split is actually based on whether or not object shift occurs, which is in turn determined by the features of the object. This chapter has included discussion of several controversies as to whether particular types of patterns should count as instances of split ergativity, and has pointed out differences in the definition of split ergativity that are the source of these differences of opinion. The chapter has also included a discussion of various theoretical proposals in the literature to account for different types of split ergativity. A major theme of this chapter is determining whether the various types of ergative splits are present in syntax or are purely morphological. Most of the splits discussed here appear to be present in syntax, with the notable exception of person/animacy/NP splits, which are occur after syntax, at the stage of morphological spellout of the features from syntax.
Split in syntax and at morphological spellout 225
Abbreviations ASP, aspect; DAT, dative; DEC, declarative; DET, determiner; ERG, ergative; FEM, feminine; HON, honorific; IMPF, imperfective; IND, indicative; INTR, intransitive; LOC, locative; MASC, masculine; NEUT, neuter; NML, nominal; NOM, nominative; NRL, non-relational; OBJ, object; OBV, obviate; PAST, past tense; PERF, perfective; PL, plural; PRED, predicate; PROG, progressive; SG, singular; TR, transitive; VOC, vocative.
Chapter 10
Split ergati v i t y i s not a b ou t erg at i v i t y Jessica Coon and Omer Preminger
10.1 Introduction It has been frequently noted in the literature on ergativity that few—if any—ergative systems are purely ergative. Rather, many ergative languages exhibit a phenomenon known as “split ergativity” in which the ergative pattern is lost in certain parts of the grammar. No ergative language is fully consistent in carrying through the ergative principle throughout its entire morphology, syntax, and lexicon: all languages that exhibit ergative patterning in their commonest case-marking system also exhibit some accusative pattern somewhere in the rest of their grammar. [Moravcsik 1978b: 237]
The two most commonly described types of split ergativity are (i) aspectual splits, and (ii) person splits. (See also the contributions by Laka, Chapter 7, Nash, Chapter 8, Woolford, Chapter 9, this volume.) In the former, the ergative pattern is lost in some subset of non-perfective aspects (or possibly certain tenses; see Coon 2013b for discussion); and in the latter, the ergative pattern is lost with some particular combination of “highly-ranked” nominal arguments (we return to the details of such “rankings”). The central argument put forth in this chapter is that split ergativity is epiphenomenal, and that the factors which trigger the appearance of such splits are not limited to ergative systems in the first place. In both aspectual and person splits, we argue, the split is the result of a bifurcation of the clause into two distinct case/agreement domains; this bifurcation results in the subject being, in structural terms, an intransitive subject.1 Since intransitive subjects do not appear with ergative marking, this straightforwardly accounts 1 For other structural accounts of ergative splits, see, among others, Laka (2006a, Chapter 7, this volume) and Nash (Chapter 8, this volume).
Split ergativity is not about ergativity 227 for the absence of ergative morphology in those cases. But crucially, such bifurcation is not specific to ergative-patterning languages; rather, it is obfuscated in nominative- accusative environments because—by definition—transitive and intransitive subjects pattern alike in those environments, and the terminology in question (‘ergative’ vs. ‘non- ergative’) specifically tracks the behavior of subjects. Thus, Moravcsik’s generalization, as quoted, does not reflect any deep instability of ergative systems, nor a real asymmetry between ergativity and accusativity (contra Visser 2006, for example). In an ergative system that exhibits this type of split, ergative-absolutive alignment is always associated with a fixed set of substantive values (e.g. perfective for aspectual splits, 3rd person for person splits). The account we will present derives this universal directionality of splits by connecting the addition of extra structure to independently attested facts: the use of locative constructions in progressive and non-perfective aspects (Bybee, Perkins, & Pagliuca 1994; Laka 2006a; Coon 2013a), and the requirement that 1st and 2nd person arguments be structurally licensed (Béjar & Rezac 2003, Preminger 2014). The remainder of this chapter is organized as follows. Section 10.2 briefly reviews ergative and split ergative patterns. Section 10.3 focuses on aspectual splits, summarizing the different types of split patterns that emerge, and presenting a structural motivation for the loss of ergativity in certain aspects. In section 10.4 we turn to person splits, and present a structural account of this phenomenon, as well. Section 10.5 concludes.
10.2 Ergativity and Split Ergativity Typical ergative- absolutive and nominative- accusative alignment systems are represented in (1) and (2), where we follow Dixon (1979) in using the following labels: A = transitive subject; P = transitive object; and S = intransitive subject. (1) ERGATIVE– ABSOLUTIVE SYSTEM
transitive:
intransitive:
(2) NOMINATIVE– ACCUSATIVE SYSTEM
A
P
A
P
ERG
ABS
NOM
ACC
S
S
ABS
NOM
228 Jessica Coon and Omer Preminger As evidenced by the range of work in the present volume, accounts of ergative case and agreement patterns within the generative tradition have been numerous, and we do not aim to adjudicate among them here. In light of this wide array of approaches, we choose not to commit to any one particular analysis of ergativity itself, and instead aim to show how a structure-based account of split ergativity can account for non-ergative patterning in an otherwise ergative system, regardless of the specific theory of ergative case assignment adopted. Throughout this chapter, we will speak of ergative patterns, rather than ergative languages. As noted at the outset, a given language rarely shows a consistent ergative pattern of alignment throughout its entire grammar (see Silverstein 1976; Comrie 1978; Moravcsik 1978b; DeLancey 1981; Tsunoda 1981b for discussion; as well as numerous examples). As observed by Dixon (1994), the two main factors that one finds as triggers of split ergativity cross-linguistically are aspectual splits (discussed in section 10.3), and person splits (discussed in section 10.4). We take it to be a relevant and crucial fact that both in aspect-based and in person-based splits, one finds a universally fixed directionality (Dixon 1994, inter alia). What this means is that for each split-triggering substantive category, the ergative-absolutive alignment is always associated with a fixed value (or set of values) of that substantive category (e.g. perfective for aspectual splits, 3rd person for person splits). This fixed directionality is schematized for aspectual splits in (3) and for person splits in (4):2 (3) fixed directionality of aspectual splits ← ergative non-ergative → ≫
perfective
imperfective
(4) fixed directionality of person splits ← ergative common nous
≫
≫ progressive
non-ergative
proper ≫ demonstratives, nouns 3rd person pronouns
≫
→ 1st/2nd person pronouns
In this chapter, developing the proposal in Coon & Preminger 2012, we offer a unified account of aspect-based and person-based split ergativity, which captures the universal directionality across these two substantive categories. Crucially, the proposed account reduces both types of splits to structural factors that are not specific to ergative languages in particular—while explaining why, in a nominative-accusative language, these same factors would not result in what one would characterize as an “alignment split.” 2 While person splits are commonly described as involving this type of scale, we demonstrate in section 10.4 that the relevant cut may in fact always be between 1st/2nd person pronouns and all other nouns. We discuss factors that have obscured this distinction, particularly in the language Dyirbal.
Split ergativity is not about ergativity 229
10.3 Aspectual Splits This section provides a brief synopsis of aspectual splits, discussed in much greater detail in Coon 2013a,b.3 As (3) illustrates, different languages may make the split in different places along the aspectual scale, but ergativity crucially remains consistently anchored to the left side of this scale (i.e., to the perfective).4 Coon (2013b) argues that aspectual splits have two underlying causes: (i) the introduction of complex syntactic structure associated with non-perfective aspects (discussed in section 10.3.1); and (ii) reduced transitivity, manifested as demotion of the direct object, triggered in non- perfective aspects (discussed in section 10.3.2). In both scenarios, the transitive subject of a “split”-patterning non-perfective construction will not receive ergative marking because it is, structurally speaking, no longer a transitive subject (where being a transitive subject means being the higher of two non-oblique noun phrases in a single, non- bifurcated clause). Furthermore, as we will show, the factors responsible for these splits are not limited to ergative-patterning languages. We therefore need not think of split ergativity as a property that is particular to “ergative languages” to the exclusion of other languages. Rather, when the structural conditions for a split arise, the result is that transitive subjects pattern with intransitive subjects; this goes unnoticed in a non-ergative system because, in these systems, all subjects receive the same marking (i.e. nominative).
10.3.1 Added Structure In Basque, one finds an aspectual split that is demonstrably triggered by the addition of syntactic structure, in this case in the progressive aspect (Laka 2006a, Chapter 7, this volume). In the perfective and imperfective aspects, Basque shows an ergative- absolutive alignment in the forms of the suffixal article. Taking singular noun phrases as an illustration, A arguments take the article -ak (ergative sg.), while P and S arguments take the article -a (absolutive sg.): (5) Basque —perfective a. [A Ehiztari-ak ] [P otso-a ] harrapatu d-ϕ-u-ϕ. hunter-artsg.erg wolf-artsg(abs) caught 3.abs-sg.abs- ‘The hunter has caught the wolf.’
AUX -3sg.erg
3 Though these are sometimes called TAM (i.e. tense, aspect, mood) splits, Salanova (2007) and Coon (2013b) call into question the existence of tense-or mood-based splits that do not also involve aspect or clause-type. In what follows, we assume that aspect is the only true trigger of this type of split—though in practice, of course, perfective aspect frequently overlaps with past tense. 4 Note that since progressive aspect is a subtype of imperfective aspect, the hierarchy in (3) is predictable in another way: the split can either target just the progressive, or also the imperfective, which properly contains the progressive; see Coon 2013a for discussion.
230 Jessica Coon and Omer Preminger b. [S Otso-a ] etorrid-a. wolf- artsg(abs) arrived 3.abs‘The wolf has arrived.’
AUX
(3sg.abs)
[Laka 1996]
In the progressive, however, the ergative marking is lost. Here all singular arguments take the -a suffix, as shown in (6). Furthermore, while the auxiliary agrees with both the subject and the object in the ergative-patterning perfective (as in (5)), only the subject is agreed with in the progressive—even if the lexical verb is transitive, as in (6a):5 (6)
Basque —progressive a. [A emakume-a ] [P ogi-a ] ja-te-n ari d-a. woman-artsg(abs) bread-artsg(abs) eat- nml-loc prog 3.abs- AUX (3sg.abs) ‘The woman is eating the bread.’
b. [S emakume-a ] dantza-n ari d-a. woman-artsg(abs) dance- loc prog 3.abs‘The woman is dancing.’
AUX (3sg.abs)
[Laka 1996]
Note that the split in the Basque progressive crucially does not involve the language switching from an ergative-absolutive to a nominative-accusative pattern, though this is frequently how such splits are informally described. While it is the case that both subjects pattern alike in (6), the pattern seen here is more accurately described as “neutral”: all core arguments are in the unmarked absolutive form, including the transitive object (ogi-a “bread-artsg(abs)”). Similar patterns are found in Nakh-Daghestanian and Indo-Aryan languages (see Coon 2013b). Laka (2006a) argues that these “split” forms in the Basque progressive are in fact bi-clausal, involving a progressive auxiliary (ari; see (6a–b)) which embeds a locative- marked subordinate clause containing the lexical verb and its object. This proposal accounts elegantly for the properties shown above: the A argument is not marked ergative, because it is not a transitive subject; it is the subject of an auxiliary whose complement is something other than a non-oblique nominal—and therefore, this auxiliary is formally intransitive (in Basque, the complement of this subordinator is oblique, as reflected in the glosses of (6a–b)). The P argument does not trigger agreement because it is in a separate, lower clause. Similar proposals have been made for Nakh-Daghestanian languages; see Kazenin 1998, 2001c; Kazenin & Testelec 1999; Forker 2010; Gagliardi et al. 2014.6 This analysis is advanced further in Coon 2010b, 2013a, where it is argued 5 This is somewhat of a simplification. As shown by Arregi & Nevins (2008, 2012) and Preminger (2009), so-called “subject agreement” (or “ergative agreement”) and “indirect object agreement” (or “dative agreement”) in Basque are actually instances of obligatory clitic doubling. 6 The proposal that non-perfective aspects may involve added structure also provides an explanation for the behavior of aspectual splits in the Mayan family, for example in Yucatec (Bricker 1981), in Chol (Vázquez Álvarez 2002; Coon 2010b), and in Q’anjob’al (Mateo-Toledo 2003; Mateo Pedro 2009); see
Split ergativity is not about ergativity 231 that such clausal bifurcation is always found in non-perfective aspects rather than in the perfective, because it is precisely non-perfective aspects whose structure is constructed using spatial/locative building blocks (see also Bybee, Perkins & Pagliuca 1994; Demirdache & Uribe-Etxebarria 2000). There are two important consequences of this analysis. First, it provides a natural explanation for why no such “splits” have been observed in nominative-accusative languages. In many nominative-accusative languages, the progressive, and sometimes the imperfective, are built on locative constructions—just as they are in Basque. An example is given in (7). (7) Ik ben het huis aan het I am the house at the ‘I am building the house.’
bouwen. (Dutch) build [Demirdache & Uribe-Etxebarria 2000: 178]
The difference is that in a nominative-accusative system, this insulation of the subject for case purposes has no effect on the marking of subjects: A and S arguments in a nominative-accusative system receive the same marking (nominative) regardless of whether or not a direct object is syntactically accessible. Second, examples like (7), as well as their myriad cross-linguistic counterparts (Bybee, Perkins & Pagliuca 1994), illustrate that independently of split ergativity—or even ergativity more generally—it is typically the non-perfective aspects that are built using elements of locative morphosyntax not found in the perfective. In conjunction with the clausal bifurcation analysis presented here, this provides an explanation for the cross- linguistically fixed directionality of aspectual splits (recall (3)): given a language that is underlyingly ergative, the perfective—in which no structure is added to the basic clausal skeleton—will reveal the underlying ergative alignment. But the structure added in a non-perfective aspect could, if it bifurcates the clause into two separate case/agreement domains, result in what looks like a “shift” out of this underlyingly ergative pattern. Importantly, nothing said so far dictates that the structure added in a non-perfective aspect must bifurcate the clause in this manner. It is conceivable that the relevant locative elements used to construct non-perfective aspects would be syntactically opaque (e.g. phasal) in one language, but syntactically transparent in another. The parametric choice between these two options would yield, accordingly, the distinction between “split-ergative” and “consistently ergative”—now conceived of as an epiphenomenal, descriptive distinction.
10.3.2 Reduced Transitivity Like the languages surveyed in section 10.3.1, Samoan (Polynesian) has been described as a language with aspect-based split ergativity. The pattern in Samoan, however, looks Larsen & Norman 1979; and Dayley 1981 for overviews. Splits in these languages involve an “extended ergative” pattern, not discussed here for reasons of space, but see Coon 2013b for an overview.
232 Jessica Coon and Omer Preminger slightly different than what we have seen so far. The basic ergative pattern is shown in (8a–b), where the A argument in (8a) appears with the ergative case marker e; absolutive arguments, like the subject in (8b), are unmarked. (8) Samoan (Polynesian) a. Na fasi [A e le tama] [P Sina]. pst hit erg det boy Sina ‘The boy hit Sina.’ b. ’ olo ’ o moe [S le tama]. pres.prog sleep det boy ‘The boy is sleeping.’
[Ochs 1988: 89]
Samoan exhibits a split between the perfective and imperfective aspects, as illustrated by the pair of transitives in (9a–b). Just as we saw for Basque in section 10.3.1, the ergative marking on the A argument in (9b) is lost in the “split” pattern, and replaced with (unmarked) absolutive. However, while in the languages from section 10.3.1 we saw absolutive marking retained on the P argument (resulting in what we described as a ‘neutral’ pattern, with all core arguments marked absolutive), here the P argument takes an oblique suffix. We will refer to this as an ‘abs-obl’ pattern. (9) a. perfective na va ’ ai-a [A e le tama] [P le i’a] pst look.at-prfv erg the boy the Fish ‘The boy spotted the fish.’ b. imperfective na va’ai [A le tama] [P i le i’a] pst look.at the boy obl the fish ‘The boy looked at the fish.’
[Milner 1973]
In the linguistic literature on Polynesian, forms like those in (9a) are known as “ergative,” while those in (9b) are labeled “objective” (Milner 1973) or “middle” (Chung 1978).7 But despite the term “objective,” the object in the imperfective is marked with i, whose function throughout Polynesian is that of an oblique marker (Chung 1978: 26). The same essential pattern is found in Warrungu (Pama-Nyungan), and a related pattern is found in Adyghe (NW Caucasian), also discussed in Tsunoda 1981b
7
Though the distinction between forms like (9a) and (9b) has previously been treated as a voice contrast, Milner (1973) argues that the distinction is aspectual, and that the English translations of some pairs are often best captured by using distinct lexical items—e.g. spotted vs. looked at—one which emphasizes the “totality” of the action (i.e., perfective), and the other which focuses on “the action itself ” (i.e., imperfective); see Milner (1973: 631).
Split ergativity is not about ergativity 233 (citing Anderson 1976). Here, however, the variation (shown in (10a) and (10b)) is not per-aspect, but per-lexeme: -š’a- (“lead”) follows an ergative pattern, while -ža(“wait”) follows an abs-obl pattern (certain verbs in Adyghe can appear in both constructions). (10) Adyghe (NW Caucasian) a. [A a-š’ ] [P a-χe-r ] ϕ -ә-š’a-ɣe-χ he- erg he-pl-abs- o3sg-s3sg-lead-prfv-pl ‘He led them.’ b. [A a-χe-r ] [P a-š’ ] ϕ -je-ža-ɣe-χ he-pl-abs he-obl s3-io3sg-wait-prfv-pl ‘They waited for him.’ [Kumakhov & Vamling 2009: 94–96] In order to account for patterns like this, Tsunoda (1981b) proposes an Effectiveness Condition (‘EF-CON’)—which appeals to notions like “effectiveness,” “conclusiveness,” “definiteness,” “actualness,” and several others—meant to account both for aspectual splits like the one in (9), and lexeme-based splits as in the Adyghe data in (10a–b). (A similar proposal was developed independently by Hopper & Thompson 1980; see Malchukov 2005 for a recent survey.) The idea behind EF-CON is that failures to meet some portion of the factors associated with high transitivity have a common morpho-syntactic consequence—namely, the P argument surfacing as oblique—regardless of whether these factors are related to grammatical aspect, or alternatively, tied to the specific meaning of a given verb. Tsunoda concludes: “Verb-split and TAM-split are fundamentally no different from each other, their semantics and case-marking mechanisms involving common principles” (Tsunoda 1981b: 391). In non-perfective aspects, objects are generally more likely to be indefinite, non-referential, and less affected; in the perfective the focus is on the culmination of the event, and objects are more likely to be affected. We might also add to this group “antipassive” constructions, described in a number of ergative languages (including Iñupiaq, Chukchi, Salish, and Dyirbal; see for example Heath 1976 and Spreng 2010).8 Generally speaking, antipassive constructions have the effect of demoting the notional P argument, by relegating it to an oblique, by incorporating it into the verb, or by omitting it altogether. Crucially, this triggers a concomitant change in marking of the A argument from ergative to absolutive (see Polinsky, Chapter 13, this volume, for a review). In the domain of grammatical aspect, antipassives are typically found in imperfective or “unbounded” aspects; the internal argument of an antipassive is often described as non-specific and/or indefinite. Formally, then, these constructions appear strikingly similar to the aspectual and “verb-type” splits reviewed above. Indeed, Spreng (2010: 563) notes: “If we review the triggers for ergativity splits across languages, we find that the Antipassive occurs under some of the same conditions.” 8 While far more commonly described in languages that exhibit an ergative argument alignment, the antipassive construction is not exclusive to such languages (see, e.g., Dryer & Haspelmath 2013).
234 Jessica Coon and Omer Preminger While the abs-obl aspectual splits, “verb-type” splits, and antipassive constructions discussed in this section are most commonly noted in so-called “ergative languages,” Tsunoda (1981b) notes that the general phenomenon of object demotion in less affected environments is not limited to languages with ergative alignments. Consider the English conative alternation: in (11a), the bear is assigned a direct case (presumably, accusative/ objective), and is clearly an affected argument; in (11b), the Patient is expressed as a PP, and there is no requirement that the act of shooting be successful—that is, the bear may be totally unaffected by the event. In English, of course, subjects carry unmarked nominative regardless of whether they are transitive subjects or intransitive subjects; but if English were ergative, we would expect to find ergative on the (transitive) subject in (11a), but absolutive on the (intransitive) subject in (11b). See Kiparsky 1998a for a related discussion of partitive alternations that arise with Finnish objects. (11)
a. Sam shot [P the bear]. b. Sam shot [P at the bear].
A survey of the various proposals put forth to account for conative and related alternations is beyond the scope of this chapter (see, e.g., Levin 1993; Kiparsky 1998a; Borer 2005b). Let us assume that some account of these alternation is in place. That means we have a way of predicting that loss of affectedness (or some closely related notion) on the part of the P argument will correlate with oblique marking on that argument, as it does in (11a–b). Crucially, such oblique marking creates a very similar state of affairs, syntactically, to what we saw in section 10.3.1: in the Basque progressive, for example, the A argument was the subject of a syntactically intransitive aspectual auxiliary, because that auxiliary selected a PP as its complement. Here, in cases of reduced affectedness, it is the main lexical verb that selects a PP complement. The absence of ergative marking on the subject of these split constructions, then, simply reflects the fact that the A argument is no longer a transitive subject (in the sense that it no longer has a non-oblique clausemate object)—as was the case in the “split” patterns surveyed in section 10.3.1. In other words, the pattern shown here is triggered by the very same syntactic factor that triggered the aspectual splits: a PP layer separating the subject from the object.9
10.3.3 Summary In this section, we have examined several types of aspectual splits. Importantly, it is not clear that any of these patterns really instantiate a split between ergative-absolutive and 9
In recent work, Kalin & van Urk (2015) argue that a complex pattern of “agreement reversal” in certain Neo-Aramaic languages can also be captured through the addition of structure precisely in non-perfective aspects. Crucially, neither of the Neo-Aramaic patterns in question is characterized as “ergative”—lending further support to the proposal here that the structural factors which trigger the appearance of “split ergativity” are independent of ergativity or an ergative argument alignment.
Split ergativity is not about ergativity 235 nominative-accusative per se; that verdict might ultimately be a matter of terminology (e.g. whether one wants to call the marking on the P argument in non-ergative alignments ‘accusative’ even though it is identical to a marker that serves as oblique elsewhere in the language). The splits share the following two properties. First, the split is not the result of special rules of case assignment or agreement, active only in certain aspects and not others; case-assigning functional heads in different aspects need not bear different case/agreement features (cf. Anand & Nevins’ 2006 analysis of Hindi, as well Ura 2006). Second, the underlying mechanisms responsible for these splits are not a special property of ergative-patterning languages. The phenomena above are also found in predominantly nominative-accusative languages (for example, the progressive construction in Dutch and the English conative alternation). The difference between, e.g., Samoan on the one hand and, e.g., English on the other, is that—by definition—transitive and intransitive subjects are marked alike in a nominative-accusative system, making it impossible to see what would otherwise be a split in subject marking.
10.4 Person Splits In this section, we turn to person-or NP-based split ergativity. This refers to systems that exhibit a non-ergative alignment when certain types of arguments are involved (e.g. 1st/2nd person pronouns), but exhibit an ergative alignment otherwise. There are important parallels between person-based split ergativity and Differential Argument- Marking—which we use, for now, as a cover-term for both Differential Subject Marking (DSM) and Differential Object Marking (DOM); see Malchukov, Chapter 11, this volume. This refers to the relatively well-studied phenomenon whereby a certain class of arguments (e.g. definites, proper nouns), when they occur in a particular structural position (e.g. direct object), bear a case-marking that is atypical for arguments in that position (e.g. dative, instead of accusative). We believe that these parallels are significant and should be considered carefully. We therefore turn first to Differential Argument Marking, and outline a particular approach to this phenomenon and a proposal for how it arises. This proposal will then be related back to the structure-based account of split ergativity put forth in section 10.3, above.
10.4.1 Differential Argument Marking as a Configurational Phenomenon We start with a basic observation drawn from the work of Baker & Vinokurova (2010) on Sakha (Turkic): that DOM—at least in this language—is determined configurationally. In Sakha, the presence of overt accusative marking on the direct object covaries with specificity: specific objects bear overt accusative marking, non-specific ones
236 Jessica Coon and Omer Preminger do not.10 So far, this characterization would fit nicely within the foregoing description of ‘Differential Argument Marking’ (in particular, DOM). But Sakha also exhibits a correlation—familiar from other languages in the Turkic family and beyond—between specificity and the structural position of the object. In effect, what one finds, at least in simple cases, is a three-way correlation between position, specificity, and (overt) case- marking: the specific object in (12a) bears accusative case marking and appears outside of the VP (as demarcated by the adverb türgennik “quickly”), while the non-specific object in (12b) has no accusative marking and remains VP-internal. (12)
a. Masha salamaat1-*(y) [VP türgennik t1 sie-te ]. Masha porridge-*(acc) quickly eat-past.3sg.subj ‘Masha ate the porridge quickly.’
(Sakha)
b. Masha [VP türgennik salamaat-(#y) sie-te ]. Masha quickly porridge-(#acc) eat-past.3sg.subj ‘Masha ate porridge quickly.’ [B&V: 602; annotations added] This three- way correlation between specificity, structural position, and case- marking begs the question of cause and effect: is one of these three factors the underlying cause, from which the other two stem? We take short object movement of the kind shown in (12a) to be an operation that obligatorily applies to all specific noun phrases within the VP; and whose successful culmination depends on whether the noun phrase in question is in a position from which vacating the VP is possible (see Preminger 2011a, 2014 for discussion, building on Diesing 1992, 1997, Vikner 1997, and others). In this sense, the morpho-semantic feature of specificity is the underlying cause for object movement. Turning to the morphological side, the reason DOM manifests itself as case morphology is because case is assigned configurationally—and therefore, case depends on the positions of different noun phrases in the clause; see also Baker (Chapter 31, this volume) and Baker & Bobaljik (Chapter 5, this volume).11 When a (specific) object moves out of VP, it is in a local enough configuration with the subject to receive dependent case (viz. accusative), as shown in (13a). When the object remains inside VP, it receives a non-specific interpretation, and is not in a local enough configuration with the subject to receive dependent case, as shown in (13b).
10
An exception to this involves accusative-marked objects that are contrastively focused; we will not discuss this here (see Baker & Vinokurova 2010: 602). 11 Baker & Vinokurova’s (2010) actual claim is that the facts of case in Sakha require a ‘hybrid’ theory of case, consisting of a configurational component and a probe-goal component. However, Levin & Preminger (2015) have shown that this is incorrect, and that the facts regarding case in Sakha can be derived in a completely configurational manner, given certain independently motivated changes to Baker & Vinokurova’s theory of agreement.
Split ergativity is not about ergativity 237 (13) a. Masha
porridge1 - ACC
[VP . . . t1 . . . ]
dependent acc b. Masha . . .
[VP . . . porridge . . . ] dependent acc
The picture that emerges, concerning the causal relations between specificity, structural position, and case-marking, is therefore the following:12 (14) Causal Relations in Differential Argument Marking morpho-semantic property (e.g.non-specific/specific)
structural configuration (e.g. object inside/outside VP)
case-marking (e.g.bare/ACC)
Extending this approach to ergative systems, let us suppose that just like accusative on Sakha objects, ergative on subjects (at least in those languages that show differential ergative marking) depends on a second DP occupying a position in the same case- domain (following Marantz 1991; see also Baker 2015). If something were to disrupt this sufficiently local configuration of two DPs, it would bleed the assignment of ergative— just like the assignment of accusative is bled in the Sakha (12b/13b). In fact, it appears that precisely such a disruption is attested. As shown by Woolford (2008), Massam (2013), and Baker (2015), an object that remains inside VP can bleed the assignment of ergative to the subject of said VP. This is illustrated for Eastern Ostyak (Finno-Ugric, Siberia) in (15). In (15a), the specific object appears outside of the VP and the subject is marked with ergative case. In (15b), in contrast, the object remains VP-internal, and the subject appears in its nominative form. (15) a. mə-ŋən ləɣə1 [VP əllə juɣ kanŋa t1 aməɣaloɣ ]. (Eastern Ostyak) we-erg them large tree beside put.past.3pl.obj/1pl.subj ‘We put them (pots of berries) beside a big tree.’ b. mä [VP t’əkäjəɣlämnä ula mənɣäləm ]. we.dual(nom) younger.sister.com berry pick.past.1pl.subj ‘I went to pick berries with my younger sister.’ [Gulya 1966, via Baker 2015; annotations added]
12
See Merchant (2009) for a different (and predating) proposal—building on Aissen (1999a, 2003) and implemented in a ‘cartographic’ fashion—that nevertheless preserves the fundamental insight that position is the independent variable in Differential Argument Marking, whereas changes in interpretation and case-marking are derivative.
238 Jessica Coon and Omer Preminger Here, failure of the object to move into a sufficiently local configuration with the subject results in a failure to assign dependent case to the latter—since this is an ergative language. Crucially, though, this state of affairs is completely parallel to what we saw in Sakha, aside from the directionality of dependent case assignment; compare (13) with (16): (16) a. We-ERG them1 [VP . . . t1 . . . ] dependent erg b. We . . . [VP . . . berry . . . ]
dependent erg
In a similar vein, Massam (2013) shows that in Niuean, specificity/non-specificity of the object has the same effect on the case-marking of the subject (i.e. ergative vs. bare), even when the object in question is null—all but ruling out an account of this subject case alternation in terms of (pseudo-)noun-incorporation of the object. Overall, these patterns are a rather powerful demonstration that Differential Argument Marking is fundamentally a configurational phenomenon. Changes in the case-marking of a particular argument are not a response to changes in the semantic properties of that argument per se, except insofar as those semantic changes affect the argument’s structural position. Indeed, in the limiting case, changes in the semantic properties of a given argument can trigger changes in the case-marking on a different argument (if the case-marking on the latter depends on the structural relation between the two, as in (16)). A theory of Differential Argument Marking that attempts to derive the morphological marking of an argument directly from that argument’s own properties (e.g. Legate 2014a) thus fails to generalize to cases like (15a–b). But a configurational theory, where the independent variable is not semantic type (specific/non-specific) but relative position, does generalize to such cases. Before concluding this subsection, it might be instructive to consider a question of terminology: would the difference between the Sakha (12a) (with a specific, accusative- marked object) and (12b) (with a non-specific, unmarked object) have led anyone to characterize Sakha as a “split accusative” language? That is, to assert that in the presence of a non-specific object, the language “shifts” out of its normal accusative alignment, and into a neutral alignment (where all core arguments are unmarked)? Note that similar alternations in subject marking routinely result in a language being classified as “split ergative”; this terminological bias toward focusing on the marking of subjects, we contend, is behind the impression that ergative languages—but not nominative–accusative ones—are where “splits” tend to arise (see the discussion in section 10.1). We return to this terminological bias in section 10.4.7
Split ergativity is not about ergativity 239
10.4.2 “Binary Split” Systems and the Prospects of Unification with Differential Argument Marking In light of the results presented in section 10.4.1, it is obviously intriguing to explore the possibility of unifying person-based split ergativity with Differential Argument Marking, under a configurational approach to both. However, there is one type of person-based split ergativity, known as a “binary split” system, which seems to pose a serious challenge to such a unification. Dyirbal is an example of such a system. Focusing on clauses with lexical (i.e., non-pronominal) arguments, we find an ergative alignment, as shown in (17): the A argument is marked with the suffix -ŋgu, while P and S arguments are morphologically unmarked. (17) a. transitive: lexical subject, lexical object [P ŋuma] [A yabu-ŋgu ] bura-n father mother- erg see-nonfut ‘Mother saw father.’ b. intransitive: lexical subject [S ŋuma] miyanda-nyu father laugh-nonfut ‘Father laughed.’
(Dyirbal)
[Dixon 1994: 161]
Turning to clauses which involve 1st and 2nd person pronominals, as in (18), we find the opposite pattern. Here, only the P argument receives morphological case, resulting in a nominative-accusative alignment pattern. (18) a. transitive: 1st/2nd person subject, 1st/2nd person object [A ŋana] [P nyurra-na] bura-n we y’all- acc see-nonfut ‘We saw y’all.’ b. intransitive: 1st/2nd person subject [S ŋana] miyanda-nyu we laugh-nonfut ‘We laughed.’
[Dixon 1994: 161]
Importantly, and in contrast with the dependent case configurations examined in section 10.4.1, the choice of marking on a given nominal appears independent of properties of other nominals in the clause. As shown in (19), unmarked subjects may coincide with unmarked object as in (19a); and marked subjects may coincide with marked objects, as in (19b).
240 Jessica Coon and Omer Preminger (19) transitive: mixed clauses a. 1st/2nd person subject, lexical object [A ŋana] [P ŋuma] bura-n we father see-nonfut ‘We saw father.’ b. lexical subject, 1st/2nd person object [P ŋana-na] [A ŋuma-ŋgu] bura-n we- acc father-erg see-nonfut ‘Father saw us.’
[Dixon 1994: 130]
These facts are summarized in (20): (20) Dyirbal argument marking —a summary 1st/2nd person pronouns
other nominals
A
∅
-ŋgu (‘erg’)
S
∅
∅
P
-na (‘acc’)
∅
The crucial observations are as follows (Dixon 1994): ‘accusative’ and ‘ergative’ can co-occur, as in (19b); but neither depends on the presence of the other, as shown in (17a) and (18a). It is therefore tempting to view this as two separate systems—a subject system and an object system—both of which are sensitive to interpretation (1st/2nd person vs. 3rd person), but each of which operates independently of the other. This type of view is schematized in (21) (we note that there have been several attempts to derive (21), or something like it, from factors such as “prototypicality” and “iconicity”; see, e.g., Wierzbicka 1981; and see Silverstein 1981 for a critical discussion). (21)
a. subjects: • 1st/2nd person → unmarked • 3rd person → marked (‘erg’) b. objects: • 3rd person → unmarked • 1st/2nd person → marked (‘acc’)
Recall, however, that such a treatment would not generalize to the Eastern Ostyak and Niuean patterns discussed in section 10.4.1; ideally, we would want a configurational account of these Dyirbal facts, as well. In what follows, we propose such an account, based on the premise that Dyirbal has the kind of very short object movement proposed by Johnson (1991), inter alia:
Split ergativity is not about ergativity 241 (22)
··· µP
···
µ'
µ0
νP ν'
SUBJ ν0
VP V0
We follow Baker & Vinokurova (2010) in assuming that dependent case is assigned as soon as possible; and that it can be relativized to a particular cyclic domain (in Baker & Vinokurova’s analysis of Sakha, dative is assigned to the higher of two DPs within the VP domain, whereas accusative is assigned to the lower of two DPs within the CP domain). With this assumption in place, we can begin to derive the distribution of ‘accusative’ and ‘ergative’ in Dyirbal in purely configurational terms. Consider a first attempt, given in (23):13 (23) case in Dyirbal a. ‘accusative’ in Dyirbal: dependent case assigned to the higher of two DPs outside of VP. b. ‘ergative’ in Dyirbal: dependent case assigned to the higher of two DPs inside of vP. Given (23a), the object will only be ‘accusative’ if it undergoes the kind of short object movement that, in Sakha, was associated with specificity. Only then will there be two noun phrases outside of VP; and the higher of these two will be the object, at least prior to any subsequent movement of the subject (and recall that dependent case is assigned as soon as the conditions on its assignment are met). The only difference is that in Dyirbal, the relevant semantic property associated with short object movement is not specificity per se, but some other property that groups 1st and 2nd person pronouns together to the exclusion of all other nominals; we return to the precise nature of this property in section 10.4.5, below. 13 Varying the directionality of dependent case in (23a–b) creates, in both instances, non-differential assignment of case. Changing the direction of (23b) to lower results in invariable assignment of dependent case to the object (which, terminologically, would then be called ‘accusative’ rather than ‘ergative’). Changing the direction of (23a) to lower results in invariable assignment of dependent case to the transitive subject (‘ergative’, rather than ‘accusative’). Importantly, then, these parametric options do not seem to overgenerate.
242 Jessica Coon and Omer Preminger The problem with (22–23) as an account of case in Dyirbal pertains to (23b): as it stands, the subject will always be marked with ‘ergative’ case, since the subject and object satisfy (23b) in their base positions. The solution to this problem, we argue, lies in recognizing the unique agreement requirements that apply to 1st/2nd person arguments, which we turn to now.
10.4.3 The Formal Agreement Requirement on 1st/2nd Person Pronouns The Person Case Constraint (PCC), also known is the *Me-Lui Constraint, is a prohibition against certain combinations of arguments, usually affecting the two internal arguments of a ditransitive. There are at least four attested varieties of the PCC (see Nevins 2007 for a recent review); of interest here is the so-called “strong” PCC. The strong PCC prohibits 1st/2nd person direct objects from co-occurring with an indirect object. In other words, it amounts to the requirement that direct objects of ditransitives be 3rd person. (See Bonet 1991, Anagnostopoulou 2005, Nevins 2007, inter alia, for further discussion.) But when one looks at formal accounts of the (strong) PCC, what one finds is that they do not rule out 1st/2nd person direct objects per se—what they rule out is 1st/ 2nd person object agreement, or 1st/2nd person object clitics. This is for good reason: the PCC arises precisely where object-agreement/object-clitics are found (e.g. the Basque (24)), and its effects disappear when no such object marking is found (e.g. the embedded clause in (25), which is the infinitival counterpart of (24b)). (24) a. Zuk niri liburu-a saldu d-i-ϕ-da-zu. (Basque) you.erg me.dat book-artsg(abs) sold 3.abs- -sg.abs-1sg.dat-2sg.erg ‘You have the book to me.’ b. * Zuk harakin-ari ni saldu n-(a)i-ϕ-o-zu. you.erg butcher-artsg.dat me(abs) sold 1.abs- -sg.abs-3sg.dat-2sg.erg Intended: ‘You have sold me to the butcher.’ (25) Gaizki iruditzen ϕ-zai-ϕ-t [zuk wrong look-impf 3.abs- -sg.abs-1sg.dat you.erg harakin- ari saltzea]. butcher-artsg.dat sold-nmz-artsg(abs) ‘It seems wrong to me for you to sell me to the butcher.’
ni me(abs)
Thus, what the PCC rules out is not a given combination of (internal) arguments, but rather a given combination of object-agreement markers (or object-clitics). Without further provisions, however, the expectation arises that a 1st/2nd person strong pronoun in direct object position of a ditransitive would be just fine, just as long as the finite verbal element carried agreement morphology expressing 3rd person features (rather than the 1st/2nd person features of the actual object). But this expectation is of course false:
Split ergativity is not about ergativity 243 (26) * Zuk harakin-ari ni saldu d-i-ϕ-o-zu. you.erg butcher-artsg.dat me(abs) sold 3.abs- -sg.abs-3sg.dat-2sg.erg Intended: ‘You have sold me to the butcher.’ To handle this, accounts of the PCC are commonly supplemented with something along the lines of (27)—or, on Preminger’s (2014) formulation, (28).14 (27) Person Licensing Condition (PLC) Interpretable 1st/2nd person features must be licensed by entering into an Agree relation with an appropriate functional category. [Béjar & Rezac 2003] (28) Person Licensing Condition (PLC) —alternative formulation The feature [participant] on a pronoun must participate in a valuation relation. NB: ‘[participant]’ is the feature that distinguishes 1st/2nd person nominals from 3rd person ones; see Harley & Ritter (2002) and McGinnis (2005) for further discussion. As noted by Preminger (2011b), this requirement appears to be a sui generis requirement on marked person features, which does which does not extend to other φ-feature classes (number and gender; pace Baker 2008).
10.4.4 The Person Licensing Condition Meets Differential Subject Marking Recall the analysis of the Dyirbal “binary split” system put forth in section 10.4.2, repeated here: ···
(29)
[=(22)] µP
···
µ'
µ0
νP ν'
SUBJ ν0
VP V0
14 A principle like (27) or (28) will correctly rule out cases like (26), but it leaves open the question of how the local direct object pronoun in the infinitival clause in (25) satisfies its licensing requirements. See Preminger (2011b) for an explanation based on locality.
244 Jessica Coon and Omer Preminger (30) Case in Dyirbal a. ‘accusative’ in Dyirbal: [=(23a–b)] dependent case assigned to the higher of two DPs outside of VP. b. ‘ergative’ in Dyirbal: dependent case assigned to the higher of two DPs inside of vP. The problem with this version was that it predicted invariable ergative case on the subject, contrary to fact (see section 10.4.2 for details). This can be remedied, we propose, if—in light of the Person Licensing Condition—we add the following provision:15 (31)
a. If the External Argument bears [participant], agreement with v0 (upon first merge in [Spec,vP]) will satisfy its PLC requirement. b. [participant]-bearing v0 is phasal: yes/no. ← per-language parameterization
In a language where (31b) is set to ‘yes’, the assignment of ergative will be bled precisely when the subject bears [participant] features (i.e. when the subject is 1st/2nd person). That is because the phasehood of v0 will render the object, located inside the complement of v0, inaccessible for the purposes of (30b).16 And, crucially, (30a) will be unaffected by phasehood (or lack thereof) of vP, because the subject is already at the edge of vP. Thus, the assignment of ‘accusative’ will not be sensitive to the features of the subject, even if (31b) is set to ‘yes.’ This produces the attested behavior of the “binary split” system of Dyirbal. The object will be assigned ‘accusative’ only if it undergoes short object movement—which, by hypothesis, applies only to 1st/2nd person pronouns in Dyirbal. The subject will be assigned ‘ergative’ unless it is 1st/2nd person, in which case the phasehood of vP will prevent (30b) from applying successfully.
10.4.5 DSM vs. DOM There is a residual issue with the account of the “binary split” system that was put forth in section 10.4.4. If Differential Argument Marking is fundamentally a configurational phenomenon, as argued here, then it comes out as a coincidence that in Dyirbal, the same 15 One might wonder whether, and to what degree, the parameterization in (31b) conflicts with the idea that vP is a category whose phasal status alternates based on the transitivity of the verb (Chomsky 2001, et seq.). In the Baker & Vinokurova treatment of DOM—which we build upon—it is crucially VP (and not vP) that is the ‘phase’ for the purposes of case computation. We take this to mean that the “classic” vP phase is simply a poor fit for the kind of locality domain required in an adequate case-assignment algorithm. We therefore set aside the traditional, transitivity-based definition of vP phasehood for the present purposes. 16 Since case assignment rules apply as soon as possible (see section 10.4.1), it is conceivable that later movement of the subject out of vP will once again bring the subject and object into a single case domain, resulting in the assignment of ‘ergative’ to the subject (or even to the object, depending on their relative hierarchical configuration). To avoid such spurious case assignment configurations, we tentatively assume that the rule in (30a) does not apply outside of µP.
Split ergativity is not about ergativity 245 factors that regulate movement of the object out of VP also regulate the phasehood of vP— namely, the presence or absence of a [participant] feature. We contend, however, that this is actually a desirable state of affairs: we want this to be a Dyirbal-specific coincidence. To see why, we must first attend to the featural bases for DSM and for DOM, cross-linguistically. Both DSM and DOM have historically often been claimed to adhere to a scale like (32): (32) ← subj marked erg common proper ≫ ≫ nouns nouns
obj marked acc → demonstratives, 1st/2nd person ≫ 3rd person pronouns pronouns [Dixon 1994, Silverstein 1976, inter alia]
While this has proven to be an extremely useful (and influential) first approximation, it belies the finer typological differences one finds between DSM and DOM. To wit: Intuitively we could expect [the DSM] split to be found between humans and non-humans, or between animates and inanimates. Actually, no language places the split in such positions; most of them indeed single out 1/2 pronouns from the rest. [Cocchi 1999: 112] If [a unified approach to DOM and DSM is correct], we expect to find an equal diversity of types of subject and object splits in the world’s languages; however, that prediction is not borne out. Instead, there are very few kinds of subject splits, in contrast to an enormous diversity of object splits. For example, Comrie (1981b: 123) notes that while definiteness is frequently the basis of object splits, there is an embarrassing absence of clear attestations of the predicted marked indefinite subject. [Woolford 2001: 535]
And with respect to Australian languages (like Dyirbal) in particular: My key point is simply that [Hopper & Thompson (1980)] clearly establish that special accusative marking tends to occur with proper nouns, human and animate nouns, and definite, referential usages in many language families. There is no reason to link this phenomenon when it occurs in Australian languages with the incidental fact that Australian languages have ergative case. [Goddard 1982: 191]
Overall, while differential ergative marking on subjects (DSM) typically correlates with the 1st/2nd person vs. 3rd person distinction, differential case marking on objects (DOM) is typically sensitive to features like animacy, specificity, and definiteness.17
17
One potentially problematic case involves patterns of apparent DOM that are sensitive to person features per se, rather than the canonical DOM features discussed here. Roberta D’Alessandro (p.c.) points out that in the Italian dialect Ariellese, an a marker is added to 1st/2nd person object
246 Jessica Coon and Omer Preminger Consequently, a language like, e.g., Balochi (NW Iranian) might be a far better representative for how “binary split” systems usually work than Dyirbal is (we return to Dyirbal itself in section 10.4.6). In Balochi, DSM distinguishes 1st/2nd person from 3rd person, but DOM is sensitive to definiteness. This pattern is summarized in (33); see Farrell 1995: 224 for details.18 (33)
Balochi Argument Marking 1st/2nd person pronouns
3rd person pronouns, and other definites
indefinites
A
∅
‘erg’
‘erg’
S
∅
∅
∅
P
‘dat’
‘dat’
∅
To capture these crosslinguistic generalizations, which Balochi exemplifies, we propose the following generalization: (34) Differential Argument Marking Generalization a. DSM (i.e. of A) is based on the presence or absence of the feature [participant] b. DOM (i.e. of P) is governed by definiteness, specificity, and/or animacy Returning to our proposal regarding binary split systems, the basic structure of which is repeated again in (35), the generalizations in (34a–b) can be captured as follows. DOM (34b) is regulated by movement to [Spec,µP], and can therefore depend on one of a set of possible features—definiteness, specificity, and/or animacy— depending on what it is that µ0 probes for. DSM (34a), on the other hand, is regulated by the phasehood of vP, which on our proposal is determined by a single, fixed feature: [participant].
pronouns, but not to the ‘unmarked’ 3rd person pronoun esse (Ariellese also has a system of proximate/medial/distal 3rd person demonstratives, all of which apparently marginally tolerate the a marker). A similar though non-identical system, involving optionality of a with 3rd person pronouns (but obligatoriness of it with 1st/2nd person ones) is reported by Manzini & Savoia (2005) for the Canosa Sannita dialect. While nothing in our account specifically rules out person-sensitive DOM, the overwhelming tendency of DOM to be based on other features (in contrast to DSM, which is frequently—perhaps always—based on person distinctions) is still a very robust cross-linguistic generalization. We therefore tentatively set aside the pattern in Ariellese and Canosa Sannita for future work. 18 An essentially identical pattern—DSM based on 1st/2nd person vs. 3rd person, with DOM based on definiteness vs. indefiniteness—is found in Kham (Tibeto-Burman; see DeLancey 1981, Watters 2002, Merchant 2009).
Split ergativity is not about ergativity 247 ···
(35) ···
[≈(29)] µP µ'
µ0
νP ν'
SUBJ
VP
ν0 V0
10.4.6 A New (or Old) Take on Dyirbal Objects While the proposal in (29–30) successfully derives systems like the one exemplified by Balochi, DOM in Dyirbal now appears to be an exception to the rule—since it is described as tracking the 1st/2nd person vs. 3rd person distinction. In this subsection, we demonstrate that this may be based on a mischaracterization of what the relevant distinction really is. The crucial observation is that Dyirbal doesn’t actually have 3rd person pronouns. This means that we can recast (36) (repeated from earlier) as (37), without altering the empirical coverage whatsoever: (36) Dyirbal Argument Marking—Summary 1st/2nd person pronouns
[=(20)]
other nominals
A
∅
-ŋgu (‘erg’)
S
∅
∅
P
-na (‘acc’)
∅
(37) Dyirbal Argument Marking —Expanded version 1st/2nd person pronouns
3rd person pronouns
other nominals
A
∅
-ŋgu (‘erg’)
-ŋgu (‘erg’)
S
∅
∅
∅
P
-na (‘acc’)
-na (‘acc’)
∅
248 Jessica Coon and Omer Preminger To be clear, we are not saying that there is a Dyirbal-internal argument in favor of (37); only that given the lack of 3rd person pronouns in the language, there is no Dyirbal- internal argument against (37). It is worth noting, in this respect, that the kind of pattern we are asserting to exist in Dyirbal is attested—in a language with 3rd person pronominals, in Cashinawa (Panoan; Dixon 1994). As the table in (38) shows, the split in the marking of A arguments in Cashinawa divides 1st/2nd person from 3rd person, while for objects the distinction is between pronouns (regardless of person) and all other nominals. (38) Cashinawa (Dixon 1994: 86) 1st/2nd-person 3rd-person pronouns pronouns other nouns A
-∅
habũ
nasalization
S
-∅
habu
-∅
P
-a
haa
-∅
In fact, this view of Dyirbal itself is not without precedent, either (see also Legate 2008): Proper and some common nouns (usually just those referring to humans) can take the suffix -na, but only when they are in transitive object function. [Dixon 1972: 43]
We can therefore maintain the view that while DSM in Dyirbal is about [participant] vs. lack thereof, DOM in Dyirbal is about pronominality—or perhaps, given this quote from Dixon (1972), animacy. This brings Dyirbal into accordance with the Differential Argument Marking generalization given in (34) and, in turn, allows those mechanisms we have proposed to derive (34) to derive the behavior of Dyirbal, as well.
10.4.7 Summary We have argued that Differential Argument Marking, including DSM, is a configurational phenomenon—and crucially, that DSM encompasses what in an ergative system is characterized as “split ergativity”. First, morpho-semantic distinctions (such as specificity, definiteness, or animacy) may give rise to changes in the relative positions of core arguments, via the mechanism of Object Shift (or other very short object movement). This, in turn, may affect the way case is assigned in the clause—normally resulting in the morpho-semantic properties of an argument co-varying with its own case marking; but in certain cases, via the mechanism of dependent case, resulting in properties of one argument co-varying with the morphological marking of another (as in Eastern Ostyak and Niuean). Second, even subject marking alternations based on whether the subject is 1st/2nd person or 3rd person can be recast configurationally, given the independently motivated formal agreement requirement on 1st/2nd person
Split ergativity is not about ergativity 249 arguments (Béjar & Rezac 2003; Anagnostopoulou 2005). Finally, even a language like Dyirbal does not counterexemplify these generalizations, since the lack of 3rd person pronouns in the language does not actually allow one to distinguish between a pronominality/animacy distinction and a distinction based on 1st/2nd person vs. 3rd person. We now turn to a note on terminology. There is a common impression that DSM (also known as ‘split ergativity’) is more common in ergative languages than in non- ergative ones, an impression that might arise even from surveying the preceding subsections. Why would this be? Our answer—echoing the discussion of aspectual splits summarized in section 10.3.3—is that this is nothing more than a terminological bias towards properties of the subject (and not of the object) as the defining property of a morphological marking system. Recall that, on the current analysis, DSM is about the disruption (or non-disruption) of the dependent case relation that the subject participates in. In a language with an ergative-absolutive alignment, disrupting this relation alters the case-marking of the subject itself (namely, ‘ergative’). But in a nominative-accusative system, disrupting this relation does not affect the subject’s marking—instead, it affects the assignment of ‘accusative’ to the object. So, by fixating on the subject, we would not notice that anything “noteworthy” has changed (recall the discussion in section 10.4.1 of Sakha as a ‘split accusative’ language).19 If we look carefully, we should be able to find something that looks like the mirror image of the Eastern Ostyak and Niuean pattern, discussed in section 10.4.1: a scenario where the [participant] features on a 1st/2nd person subject—and the concomitant phasehood of v0—affect a morphosyntactic property other than the subject’s own morphological marking. As discussed in Coon & Preminger 2012, this might be exactly what is happening in the Abruzzese dialect of Italian: (39) a. Ji so’ magnate. I eaten.sg am(be) ‘I have eaten.’ b. Esse a magnate she has(have) eaten.sg ‘She has eaten.’
(Abruzzese)
[D’Alessandro & Roberts 2010: 54–55]
On the assumption that have is be + X0 (where X0 is some clausal particle of category D or P; see Freeze 1992, Kayne 1993), a phasal v0—caused by a [participant]-bearing
19
There is one noteworthy scenario where disrupting a dependent case relation in a nominative- accusative system would affect the case-marking on a subject—when the relation in question is between the subject and an even higher noun phrase. Accusative subjects in Turkic (see George & Kornfilt 1981; Kornfilt 1984, 2003, 2006 on Turkish; as well as Baker & Vinokurova 2010 on Sakha) may constitute precisely such a scenario.
250 Jessica Coon and Omer Preminger subject, in same manner proposed in section 10.4.4—would block the head-movement/ incorporation of X0 into be. This seems to be exactly the state of affairs in Abruzzese (see D’Alessandro & Roberts 2010).20
10.5 Conclusion We argued that split ergativity, of both the aspectual sort and the person-based sort, is epiphenomenal. The absence of ergative marking on (what appears to be) a transitive subject is the result of structural properties of the clauses in question, which affect the configurational basis upon which case and agreement depend. In the domain of aspectual splits (section 10.3), we showed that non-perfective aspects involve additional structure, either in the form of an aspectual auxiliary (e.g. Basque), or demotion of the P argument (e.g. Samoan). As for person splits (section 10.4), we showed that general properties of Differential Argument Marking are implicated in so-called person-based split ergativity. Specifically, while the marking of objects (DOM) is sensitive, in most if not all cases, to features such as animacy, definiteness, and specificity—the usual features responsible for Object Shift—the marking of subjects (DSM) is sensitive, in most if not all cases, to the distinction between 1st/2nd person and 3rd person. As we have shown, there is good reason to treat these patterns, too, as configurationally triggered. Object shift would bring the object into the same case domain as the subject, facilitating the assignment of dependent case. 1st/2nd person arguments are known to have unique licensing requirements, and if the projection satisfying these requirements on a 1st/2nd person subject is phasal, it will disrupt otherwise available dependent case relations. We have presented some specific case studies in how the addition of structure may arise; but note that these are not intended to be exhaustive. Our general proposal is not that every language with split ergativity is necessarily a variation on one of these specific cases. Rather, our proposal is that splits, in general, arise as the result of structural changes to the relevant clauses, and not as the result of special mechanisms or featural specifications not found in nominative-accusative languages. An illustrative example is provided by Nash (Chapter 8, this volume). She argues that while split ergativity in Georgian (Kartvelian) does not involve the addition of morphosyntactic structure of the type seen in section 10.3 (contra the suggestion in Coon 2013b), it does nonetheless fit the general proposal advanced here: ergative-patterning aorist tenses are in fact an example of neutral aspect, and lack a phasal EventP present in split-patterning non-aorist tenses. 20
What we have not yet been able to find is a language where [participant] features on the subject affect the case marking of the object per se—i.e. a language where the presence of a 1st/2nd person subject bleeds the assignment of accusative to the object. We are not sure, at this juncture, whether this is a true typological gap, given that its mirror image—properties of the object bleeding the assignment of ergative to the subject—was only found in two languages in our sample (Eastern Ostyak and Niuean).
Split ergativity is not about ergativity 251 We have also not examined in detail here proposals in which 1st and 2nd person arguments are base-generated in a higher clausal position (Nash, Chapter 8, this volume, on Georgian; Wiltschko 2006 on Halkomelem (Salish)), though this again fits the general pattern of splits being the result of different structures, and of special licensing requirements applying only to 1st/2nd person arguments. We have argued further that the structural differences that cause splits are not limited to ergative systems in the first place. Rather, the issue boils down to the defining characteristic used by linguists to distinguish ‘ergative systems’ and ‘accusative systems’ from one another. The simple fact that in a nominative-accusative system, both transitive and intransitive subjects receive the same marking (nominative), obscures the fact that some of these so-called transitive subjects may actually be intransitive subjects in the morphosyntactic sense (e.g. in the presence of a complex progressive aspectual construction, a demoted object, or phasal vP). In other words, the difference between a transitive and intransitive subject (as in the English John shot the bear vs. John shot at the bear, for example) does not catch our attention as readily in a nominative-accusative system as it does in an ergative one (see also Polinsky, Chapter 13, this volume). As noted above, this analysis gives us a handle on the seeming paucity of consistently ergative languages, compared to consistently accusative languages—an asymmetry which on the current view is rather superficial, and has more to do with our readiness to apply the terminology of ‘split’ than it has to do with any deep grammatical properties. Furthermore, the proposed analyses of both aspect-based and person-based split ergativity are able to account for the universal directionality of these splits: the fact that ergative-absolutive alignment and nominative-accusative alignment each remain anchored, cross-linguistically, to fixed ends of the relevant scale (be it an aspectual scale or an NP “prominence” scale). That is because in both domains (aspect and NP-type), there is independent evidence that the values on one end of the relevant ‘scale’—but not the other—are cross-linguistically associated with additional syntactic structure (in ergative languages and accusative ones alike). It is this additional structure, on the current account, that is responsible for the relevant bifurcation of the clause, in those languages where the added structure happens to be syntactically opaque; and it is this bifurcation that results, in a language that is ergative to begin with, in the appearance of a split, due to altering the configurational properties of the clause. Proposals which seek to account for split ergativity by making direct reference to the scales in (3) and (4) (e.g. DeLancey 1981) overgenerate: in the domain of person splits, for example, it was shown that most of the splits predicted by such a scale are in fact unattested. Overwhelmingly, the only category on this scale that is relevant to subjects is 1st/2nd person—a category which is independently known to require special licensing mechanisms (cf. PCC effects). While we make no new proposals about why some languages display an ergative- absolutive pattern and others display a nominative-accusative one, the resulting picture is one in which a given language consistently exhibits one or the other pattern—at least as far as the core principles of case assignment are concerned (see also Laka, Chapter 7, this volume). There is no ‘split’ in the set of case-assignment rules or principles operative
252 Jessica Coon and Omer Preminger in the language as a whole; different constructions may result in detransitivization, which as discussed, is not a phenomenon that is specific to ergative systems in the first place. This “demystification” of split ergativity contributes to a growing body of work that suggests there is no ergative macro-parameter, which would group together ergative and split-ergative languages under a single setting, in opposition to other, fully accusative languages. Furthermore, the same results cast doubt on the idea that there is anything especially marked about ergativity (cf. Visser 2006).
Acknowledgements The authors’ names appear in alphabetical order. Many thanks to Peter Arkadiev, Roberta D’Alessandro, Amy Rose Deal, Itziar Laka, Anoop Mahajan, Jason Merchant, David Pesetsky, Masha Polinsky, Milan Rezac, Norvin Richards, Ian Roberts, Andrés Pablo Salanova, and Thomas Weir, as well as to audiences at the University of Cambridge, Reed College, CLS 47, WCCFL 29, and SLE 2013, for feedback and comments at various stages of this project.
Abbreviations 1, 2, 3, 1st-, 2nd-, 3rd-person; ABS, absolutive; AOR, aorist; ASP, aspect; AUX, auxiliary; DET, determiner; ERG, ergative; FEM, feminine; FUT, future; GEN, genitive; HON, honorific; IMPF, imperfective; LOC, locative; MASC, masculine; NML, nominal; NONFUT, non-future; OBL, oblique; PL, plural; PRES, present; PRFV, perfective; PROG, progressive; PREP, preposition; PST, past; POSS, possessive; PTCP, participle; SG, singular. In some cases, glosses have been simplified or modified from those of the original authors for consistency. We follow the original authors’ choices regarding capitalization and punctuation. If no citation is provided, data is from the authors’ field notes.
Chapter 11
Ergativit y a nd differentia l c ase marki ng Andrej Malchukov
11.1 Introduction The issue of differential subject marking (DSM) in ergative languages has been primarily discussed by Silverstein (1976) and others in connection to patterns of split ergativity found in Australian languages. A typical Australian pattern of split ergativity makes a split between pronouns (sometimes, just 1st and 2nd person pronouns) vs. nouns: nouns show overwhelmingly an ergative pattern, while pronouns an accusative pattern. The study of differential object marking (DOM) has a somewhat different source. DOM has been introduced into general literature in the work by Bossong, Comrie, and Lazard among others and has been extensively documented by Bossong (1985) for Iranian languages. Later on, however, the two strands of research have started to converge, as the Australian pattern of split ergativity with pronouns showing the accusative pattern has been identified as one of classical cases of DOM. In accordance with the focus of the volume on ergativity, I will have more to say in this chapter about the DSM, which is more prominent in ergative languages as compared to DOM (see section 11.9 for discussion and explanation), yet for proper understanding of differential case marking a comparison of DSM and DOM phenomena is important. In what follows I discuss the phenomenon of differential case marking in ergative languages and propose an explanation for the attested patterns in terms of competing motivations as practiced in functional typology, which can be also formalized in Optimality Theory. Section 11.2 introduces DOM, and discusses a popular explanation of the attested patterns in terms of markedness. Section 11.3 extends the discussion of differential case marking to DSM in ergative languages, showing that markedness cannot provide a complete explanation for DSM patterns. In particular, it will be claimed
254 Andrej Malchukov that while in the domain of DOM we encounter just one pattern, complying with the predictions of the Silverstein hierarchy, in the domain of DSM we find two opposite patterns; a “Silverstein-pattern” and an “anti-Silverstein pattern.” Section 11.4 introduces a two-factor approach, in terms of indexing and distinguishability constraints, which is able to explain asymmetries between DOM and DSM. Section 11.5 extends the discussion to other features implicated in differential case marking involving definiteness and discourse factors. The following sections deal with varieties of differential case marking: they address symmetric and asymmetric case marking (section 11.6), split vs. fluid marking (section 11.7), and show how this taxonomy relates to the two case- marking strategies driven by indexing or distinguishability constraints (section 11.8). The next section (section 11.9) explores a relation between differential case marking and voice alternations such as passive and antipassive. Section 11.10 provides a discussion of diachronic scenarios behind differential case marking, and section 11.11 is a brief conclusion.
11.2 DOM: The Markedness Approach Among the phenomena known under the rubric of differential case marking, DOM is definitely better researched (Bossong 1985; Lazard 1998; Aissen 2003b; among others). These studies have shown that in many languages objects higher on the animacy/definiteness hierarchy tend to be (case) marked while those which are lower on the hierarchy need not be. It has also been observed that in some languages DOM is more sensitive to the definiteness dimension (e.g. in Persian, where the postposition -râ occurs mostly with definite NPs), while in some other languages DOM is more sensitive to animacy proper (e.g. in Guaraní, the postposition pe is normally found with animate NPs). In yet other languages DOM is sensitive to both features. One of the best studied cases is Hindi, where animate NPs should be marked by accusative/dative case, while inanimates are marked only if definite (Mohanan 1990): (1)
Hindi (Mohanan 1990: 104) Ilaa-ne bacce-ko / (*baccaa) uTaayaa. Ila-ERG child-ACC / (*child) lift.PERF ‘Ila lifted a/the child.’
(2) Ilaa-ne haar uTaayaa. Ila-ERG necklace lift.PERF ‘Ila lifted a/the necklace.’ (3) Ilaa-ne haar-ko uTaayaa. Ila-ERG necklace-ACC lift.PERF ‘Ila lifted the necklace.’
Ergativity and differential case marking 255 The usual explanation for DOM relies on the concept of markedness. This explanation, originally due to Silverstein (1976), is formulated by Comrie (1989: 128) as follows: “In a standard transitive scenario, A is animate and definite while P is inanimate /indefinite (or at least less animate and definite than A), so any deviation from this scenario should be marked.” This account correctly predicts that animate/definite objects, which manifest a marked (less natural) combination of role and semantic features, should be case- marked, while inanimate/indefinite objects, which manifest an unmarked combination, need not be marked. This markedness explanation of DOM patterns also lies at the heart of Aissen’s (2003) optimality–theoretic account of DOM. Aissen proposes to capture the markedness pattern through a harmonic alignment between a scale of grammatical functions (subject and object) and prominence scales incorporating the Animacy Hierarchy. On Aissen’s account the insight that animate arguments are less natural qua objects (than qua subjects), is captured by a constraint ranking, where a constraint against having unmarked animate objects is stronger than a constraint against having unmarked inanimate objects. Thus, the pattern in Hindi where only humans are obligatorily case-marked is captured in Aissen’s account in the following fashion (a simplified fragment of the constraint hierarchy is represented in (4)): (4) …*Oj/Hum & Øc >> *Case >>….>> *Oj/Inan & Øc… In this representation, the constraint against having zero marked human objects (*Oj/ Hum & Øc) dominates the economy constraint prohibiting any case marking (*Case), which in turn dominates a constraint against having zero marked inanimate objects (*Oj/Inan & Øc). As a result, human objects are obligatorily marked in Hindi, while inanimates may be left unmarked (see Aissen 2003b for a full account). Aissen’s approach has been highly influential and was followed up and refined in a number of respects in subsequent literature. de Swart (2007), de Hoop and Malchukov (2008), and Malchukov (2008a) proposed to subsume Aissen’s markedness constraints under Distinguishability (as instantiation of the discriminating function of cases). In this way they extended this analysis to cases where case assignment is ‘global’ (in terms of Silverstein 1976), that is, sensitive to properties of both arguments in a clause. Thus, de Swart (2007) cites an example form Malayalam, where object marking is usually lacking on inanimate arguments, but may appear on the P if the A is inanimate as well (in examples like ‘The waves destroyed the ship’). Malchukov (2008a) discusses a pattern of global DOM in the Papuan language Awtuw (Feldman 1986). In Awtuw, ACC is obligatorily used if a P equals or is higher than an A on the Animacy Hierarchy (as in (5)), and is left unmarked otherwise (as in (6)): Awtuw (Feldman 1986: 110) (5) Tey tale-re yaw d-æl-i. 3FS woman-ACC pig FA-bite-P ‘The pig bit the woman.’
256 Andrej Malchukov (6) Tey tale yaw d-æl-i. 3FS woman pig FA-bite-P ‘The woman bit the pig.’ This pattern is clearly due to the differentiating strategy; the ACC marks untypical Ps (animate, human), which may be confused with the subject otherwise. The distinction between local and global disambiguation is not clear-cut, though. Thus, while in Hindi markedness is local, in a genetically related Kashmiri it is (at least partially) global in that the P takes an object (ACC/DAT) case if the A is lower than the P on the Animacy/Person Hierarchy: Kashmiri (Wali and Koul 1997: 155) (7) Bı chus-ath tsı parına:va:an. 1SG.NOM am.2SG.ACC 2SG.NOM teaching ‘I am teaching you.’ (8) Su chu-y tse parına:va:an. 3SG.NOM is-2SG 2SG.ACC teaching ‘He is teaching you.’ In (7), the P is lower than the A on the Animacy/Person Hierarchy and remains in the NOM case; in (8), the P is higher than the A and therefore takes the object case. Given that the distinction between local and global distinguishability is not clear-cut, I consider both local and global distinguishability as manifestations of basically the same strategy. The relation between global and local distinguishability can also be conceived in diachronic terms: what starts as a pattern of global distinguishability, where use of a marker is optional (dependent on the context, which can always make its use dispensable), is eventually conventionalized as a pattern of local distinguishability (where, say, animate objects are always marked irrespective of the context). This is consistent with what we know about the extension of DOM in individual languages (see, e.g., Aissen 2003b for a diachronic discussion of DOM in Spanish). A related development occurred in Persian where an animacy based DOM (attested in early Judaeo-Persian texts) developed into definiteness based pattern (Stilo 2009). This conclusion is also compatible with analyses which suggest that ‘pragmatic’ DOM can eventually evolve into ‘semantic’ DOM (Zeevat and Jäger 2002). Since distinguishability relates to ambiguity avoidance, a natural move would conceive it as a comprehension constraint in OT semantics, rather than a production constraints in OT syntax. Indeed, Zeevat and Jäger (2002) proposed to reconstruct Aissen’s analysis in OT semantics (Hendriks and de Hoop 2001), and Malchukov and de Hoop (2008) use distinguishability as a comprehension constraint in a bidirectional OT model. Finally, from a psycholinguistic perspective, de Hoop and Lamers (2006)
Ergativity and differential case marking 257 explicitly argue for incremental optimization of case interpretation making use of distinguishability constraints. So, it appears that by now we have a complete story for differential case marking (DCM) based on the concept of markedness/distinguishability. Yet, as we will see in section 11.3, extending this analysis to DSM meets with mixed success.
11.3 DSM: A Problematic Pattern As aptly demonstrated in the literature (Moravcsik 1978a; Bossong 1985; Comrie 1989; Lazard 1998; Aissen 2003b; among others), DOM is both a pervasive and a cross- linguistically consistent phenomenon, even though languages differ in extensions of object marking along the animacy/definiteness hierarchy. Less studied is the phenomenon of DSM, which may be largely due to the fact that ergative languages, which provide stronger cases of DSM, became the subject of a systematic study only recently.1 Does DSM conform to the markedness pattern, as does DOM? The predictions here would be that inanimate and/or indefinite As which deviate from the prototype are preferably marked (Comrie 1989). Sometimes one indeed finds such a pattern. Consider the case of Qiang (Lapolla 2003: 125), where an A in a transitive causative clause does not take the “Agentive” case unless inanimate: (9)
Qiang (Lapolla 2003: 125) Moᴚu-wu qa da-tuə-ᶎ. wind-AGT 1SG DIR-fall.over-CAUS ‘The wind knocked me down.’
Yet, such a pattern, with only inanimate As case-marked, is rare. In most other split- ergative languages of Australia (as well as some Tibetan and Caucasian languages), markedness effects are manifested in a noun/pronoun split, where pronouns (or just the first and second person pronouns, as in Dyirbal; Dixon 1972), which are highest on the Animacy Hierarchy, lack the ergative case (Silverstein 1976; Comrie 1989). This pattern is consistent with Silverstein’s generalization, which predicts that accusative marking spreads from the nominals on the top of the Animacy Hierarchy to lower nominals, while ergative marking spreads in the opposite direction starting from the nominals
1
Note that in most accusative languages the syntactic status of “non-canonical subjects” is controversial (see the chapters in Bhaskararao and Subbārāo 2004 for a recent discussion). For example, Icelandic seems to be quite exceptional among Germanic languages in having oblique subjects, involved in a DSM pattern. The corresponding constructions in other Germanic languages (such as German Mir gefällt das Buch [me.DAT please.PRES.3SG the.N book.NOM.N] ‘I like the book’) are usually considered to involve object experiencers, rather than subject experiencers, since the experiencer fails the standard tests for subjecthood (see, e.g., Zaenen et al. 1985; Sigurðsson 2002 for discussion of Icelandic/German contrasts; see, however, Barðdal 2002 for a different view).
258 Andrej Malchukov on the bottom of the Animacy Hierarchy. This pattern can be straightforwardly captured through the Aissen-style constraints, as illustrated above. Consider the pattern in Dyirbal, where only 1st and 2nd person pronouns, highest on the Animacy/Person Hierarchy are left unmarked, while all other nominals feature ergative marking. This pattern can be captured by the constraint hierarchy in (10).2 (10) …*A/Inan & Øc >>*A/An & Øc >>…*A/3rd & Øc >> *Case >>….>> *A/1,2 & Øc… As in case of DOM, this discriminating account of case marking works well for cases when case-marking pattern is global, that is when availability of ERG case on the A depends on properties of the P argument. Consider a case of Fore (Scott 1978), another language with a global pattern of case marking, which is in a way a mirror image of the Awtuw pattern illustrated in (5)–(6) above. In Fore, the ergative marker (Scott’s “delineator”) is used only if the P is higher than the A on the Animacy Hierarchy (as in (12)), otherwise the A remains unmarked (as in (11)): Fore (Scott 1978: 116)
(11)
Yagaa wá aegúye. pig man 3SG.hit.3SG ‘The man hits (or kills) the pig.’
(12)
Yagaa-wama wá aegúye. pig-ERG man 3SG.hit.3SG ‘The pig hits the man.’
Also here the use of the case marker is motivated by the need to disambiguate: the ERG case appears on non-prominent (non-human) As which are more likely to be construed as objects than as subjects. Note also that ERG marking becomes dispensable if the arguments are disambiguated through the use of verbal agreement; on Foley’s account (1986: 173), this means that verbal (agreement) morphology takes precedence to the nominal (case) morphology as a disambiguation strategy. More discussion of the Fore pattern from the optimality–theoretic perspective can be found in Donohue (1999) and de Swart (2007). While the cases above provide evidence for the markedness/discriminating approach, there are also cases which directly contradict the predictions of the markedness approach. This is particularly clear when the DSM is attested for intransitive subjects, as also restrictedly attested in Hindi. As is well known, DSM in Hindi is primarily conditioned by aspect: ergative marking of transitive subjects is attested only in perfective
2 The representation in (10) does not appear in this form in Aissen (2003), since she relies on a general notion of subject, but appears in other studies that make a distinction between transitive (A) and intransitive (S) subjects (e.g. Stiebels 2002; Arkadiev 2008a).
Ergativity and differential case marking 259 tenses. There is also a group of intransitive verbs (Mohanan’s “class 2 verbs”) which allow a case alternation on subjects in the perfective, and where the choice between ERG and NOM depends on volitionality (see also Lee 2003 and de Hoop and Narasimhan 2005 for discussion): Hindi: Mohanan (1990: 94) (13) Vah cillaaya. he.NOM shout/scream.PERF ‘He screamed.’ (14) Us-ne cillaaya. He-ERG shout/scream.PERF ‘He shouted (deliberately).’ Note that ergative marking here is restricted to volitional and hence animate subjects, contrary to what is predicted under the markedness approach. This is not an isolated case. It has long been noted (see, e.g., DeLancey 1981), that DSM in many ergative languages (both consistently ergative languages and ergative languages with a tense/aspect based split) does not follow the markedness/economy predictions. For these languages, the presence of an ergative case on the subject is rather related to its agentivity (properties of volitionality/control). Similar counterexamples to the expected markedness pattern are attested elsewhere. Consider the following examples from Samoan, where animate As are consistently encoded by ergative case, while inanimate As (e.g. those referring to a natural force) allow for an alternation between ergative and oblique (locative) cases: Samoan (Mosel and Hovdhaugen 1992: 424–5) (15) Na tapuni e le matagi le faitoto’a. PAST close ERG ART wind ART door ‘The wind closed the door.’ (16) Na tapuni i le matagi le faitoto’a. PAST close LOC ART wind ART door ‘The wind closed the door.’ This case is also incompatible with the markedness account, since it is a more animate A that is obligatorily marked by the ergative. Thus, it seems that markedness can account for the DOM pattern, but fails to provide a general explanation for DSM (see also Woolford 2001). Indeed, in case of DOM, one finds just one consistent pattern, with more prominent Ps preferentially marked, while in the domain of DSM, we obtain two opposite patterns: a “Silverstein-pattern” with less prominent As preferentially marked (as in Dyirbal and Qiang), and an “anti-Silverstein pattern” with more prominent agentive As marked with an ergative case (as in Hindi and Samoan). In section 11.4
260 Andrej Malchukov I present an explanation for the puzzling asymmetry between DOM and DSM in that respect.
11.4 Explaining Asymmetries in Differential Case Marking: The Two Factor Approach In the functional–typological literature, two main functions of case marking have been identified: the indexing function, that is, cases are used to encode semantic roles, and the discriminating function, that is, the need to distinguish between the core arguments (subjects and objects) (Hopper and Thompson 1980; Mallinson and Blake 1981; Kibrik 1985; Comrie 1989). At first, recognition of these functions has given rise to two analyses of case marking that were viewed as alternative and even competing. Nowadays, however, it is generally acknowledged that both are indispensable to account for the cross-linguistic variation in case patterns (Mallinson and Blake 1981: 91–6; Song 2001: 156–67). For example, the indexing approach provides a better account of case marking of oblique arguments, and of the marking of core arguments in languages where case reflects semantic roles of arguments (‘role-dominated’ languages in terms of Van Valin and Lapolla 1997). On the other hand, the indexing approach on its own fails to account for a well-known tendency, attested both in accusative and ergative languages, to leave the single argument of an intransitive verb (S) as well as one of the core arguments of the transitive verb (A or P) unmarked. To account for this pattern one usually invokes other functional factors, such as the need to differentiate between the two arguments (that is, the discriminating function mentioned above), as well as economy that disfavors overt marking of (core) arguments (see Arkadiev 2008a for further discussion). Once these two motivations for case marking in general are recognized, we can consider what predictions follow for the domain of differential case marking. From a functional–typological perspective, indexing, and differentiating can be viewed as two (potentially) competing motivations (see Du Bois 1985; Croft 1990), or, in optimality– theoretic terms, two conflicting constraints (Prince and Smolensky 2004). I assume that ‘differentiating’ (Diff) and ‘indexing’ (Index) are two general constraints (or rather constraint families) in the domain of case marking (see Malchukov 2008a; de Hoop and Malchukov 2008). • Diff: The arguments (A and P) must be distinguishable. • Index: Encode semantic roles (A and P). Index may be conceived as a family of role-related faithfulness constraints, as commonly assumed in the literature (e.g. Woolford 2001; Lee 2003; Grimm 2011). Diff is similar to the markedness constraints introduced by Aissen (2003), yet its scope is broader as Diff
Ergativity and differential case marking 261 can be satisfied by other means than case such as agreement or word order (de Hoop and Lamers 2006; de Swart 2007). The question to be addressed is what consequences for the domain of case marking follow from recognition of these two general constraint families. Importantly, for the case of DOM, marking a prominent (animate) P is consistent with both constraints. Indeed, from a distinguishing perspective, preferential marking of animate Ps is understandable, as they are more likely to be confused with the subject. From an indexing perspective, marking of animate Ps is also favored, although for a different reason: Ps that are higher on the Animacy Hierarchy are more salient in discourse (are ‘prominent’ arguments in terms of Legendre et al. (1993), or ‘strong’ arguments, in terms of de Hoop 1996) and thus make a “better” patient (Hopper and Thompson 1980; Næss 2007). Table 11.1, as familiar from OT literature, shows cross-linguistic preferences in the marking of prominent (P) and less prominent (p) objects, given these two constraints. Table 11.1 Constraint interaction in DOM Index
Diff
P-marking
Diff
Index
*!
*
P-marking *!
p-marking
*
p-marking
As is clear from Table 11.1, both Index and Diff constraints penalize marking of less prominent objects in preference to prominent ones, so marking of prominent Ps is preferred irrespective of the ranking of the two constraints. Indeed, the absence of case marking of low-prominent (in particular, inanimate) patients can be attributed either to Index (under the constraint ranking Index-P >> Econ >> {Index-p, Diff-p, Diff-P}) or to Diff (under the constraint ranking Diff-P >> Econ >> {Diff-p, Index-P, Index-p}). As both constraints favour the same pattern, the cross-linguistic consistency of the DOM pattern is (correctly) predicted. With regard to DSM, the predictions are rather different. Diff will disfavor ergative marking of ‘strong’ (prominent, animate) As; since the latter are identified as agents more easily, ergative marking is dispensable. By contrast, Index penalizes ERG on ‘weak’ (inanimate/non-sentient) As, as they are less typical as agents. Thus, in case of DSM, these two constraints give conflicting predictions with respect to the marking of prominent/strong (A) or less prominent/weak (a) subjects, as shown in Table 11.2. Table 11.2 Constraint interaction in DSM Index
*
A-m arking a-marking
Diff
*!
Diff A-marking a-marking
Index
*! *
262 Andrej Malchukov Given that the two constraints are in conflict, and on a further assumption, that there is no universal fixed ranking between the two, this analysis correctly predicts a variation in DSM patterns across languages. Thus, it explains why DSM is less consistent cross- linguistically as compared to DOM. As predicted, two different DSM patterns are found. In some languages, more prominent nominals (e.g. pronouns, as in Dyirbal, or animates as in Qiang) will lack ERG case, in accordance with the differentiation/markedness approach. Other languages restrict the use of ERG case to more agentive (and hence animate) nominals (cf. the pattern in Hindi and Samoan), following the predictions of the indexing approach. Thus, the Dyirbal pattern results from a constraint hierarchy where Diff outranks Index (Diff-a >> Econ >> {Diff-A, Index-A, Index-a}), while the Samoan pattern results from the opposite ranking of the two constraints (Index-A >> Econ >> {Index-a, Diff-a, Diff-A}). (See also de Hoop and Malchukov 2008 for a bidirectional OT analysis of these patterns.) Thus, it seems futile to try to reduce both factors (constraints) to one: the fact remains that in the domain of DSM we find two opposite patterns, a “Silverstein- pattern” and an “anti-Silverstein” pattern, even if the latter has been largely disregarded in the literature on alignment splits. For the case of DOM, the two motivations converge, so some patterns (like in Hindi in (2)–(3)) are amenable to explanation either in terms of Distinguishability or Indexing. In other cases, the effects of the two constraints can be discerned. Thus, the pattern of the global case marking, as illustrated for Awtuw ((5)–(6)), is clearly due to Distinguishability. On the other hand, some other DOM patterns are arguably due to the indexing function. In particular, the role of Indexing becomes obvious when animacy effects in DOM are extended to DSM, as is the case in Central Pomo (Mithun 1991b). In Pomo the object (patient) case is only found on human patients, but also carries over to animate subjects of unaccusative intransitives (see Malchukov 2008a for discussion and illustration). This corroborates a connection between animacy, prominence, and affectedness, and implicates that some cases of DOM can be better explained under the indexing approach. Indeed, when the pattern of case marking in a transitive clause is extended to intransitive subject, the issues of distinguishability do not arise. Thus, animacy effects in differential case marking are epiphenomenal: in some cases, they are related to disambiguating function (and therefore can be compensated by agreement and/or word order), in other cases are implicated by some role-related features such as volitionality. Indeed, it is redundant to mark animacy per se.3
3 In a recent article dealing with DSM, Fauconnier and Verstraete (2014) argue that ‘motivity’ rather than animacy is a contributing factor to DSM (along with the features of Instigation and Directed Transmission). ‘Motivity’ is a broader concept than animacy as it relates to agents that possess an internal source of energy, including natural forces (such as fire, earthquakes, wind, or rain).
Ergativity and differential case marking 263
11.5 Extending the Two Factor Approach: Definiteness and Discourse Properties As noted above, Comrie (1981b) relied on markedness to explain why not only animate Ps, but also definite Ps tend to be marked in languages displaying a DOM pattern. We further observed that the role of animacy is different for DOM and DSM patterns, as in the latter case the indexing and differentiating strategies make different predictions. The question arises then whether we find a similar asymmetry between DOM and DSM with respect to the definiteness dimension of the nominal hierarchy, as well. In fact, Comrie (1981b: 129) notes an “embarrassing” absence of evidence for markedness effects in cases of differential encoding of definite and indefinite subjects. That is, one does not readily find cases where only indefinite As appear in the ergative case, as expected under the markedness approach. The same point has been repeated in subsequent literature, both functional–typological and optimality–theoretic (e.g., Woolford 2001; Næss 2004). This claim, however, is not entirely correct. Consider the case of Ika (Frank 1985), where we find exactly this pattern: new, indefinite As take the ergative case (see (17)), while given/definite As do not, unless they are emphatic/ contrastive (as in (18)): Ika (Frank 1985: 150) (17) Ikı gä-ža kua ikä-se’ gä-ža? man eat-MED or man-ERG eat-MED ‘They eat people or people eat them?’ (18) José-se’ eigui keiwı tšei-na. Jose-ERG also right.away shoot-DIST ‘Also José shot it.’ Frank (1985: 150) notes that A arguments that are newly introduced or contrastive are regularly marked by ERG in Ika, and explicitly relates this fact to the markedness pattern observed by Silverstein and Comrie. Ika is not exceptional in this respect. A similar pattern is found in many other languages displaying a phenomenon termed here ‘focal ergativity.’ Although the fact that in some languages ERG marking appears on emphatic, new, or contrastive As has not gone unnoticed in the literature (Plank 1979: 34; Van Valin 1992), its significance for the theory of case marking has not been sufficiently appreciated until recently. Yet, more recent work of McGregor and others (see, e.g., McGregor 2009; McGregor and Verstraete 2010) demonstrated a connection between optional ergativity and discourse structure beyond any doubt. For example, in Gooniyandi (McGregor 1992), given/predictable agents, when overt, need not be marked, but ERG
264 Andrej Malchukov regularly marks those transitive subjects that are unexpected as agents, as in the following example: (19) Gooniyandi (McGregor 1992: 289) Ned Colin-ngga ridim-ngangadda Ned Colin-ERG ride-3SG.A+3SG.P+1SG.D ‘Ned Colin rode my horse for me.’
yawarda. horse
As shown by McGregor (2009 and passim) patterns of focal ergativity are commonplace in Australian and Oceanic languages. Malchukov (2008a) discusses a number of examples from other linguistic areas; in particular, he cites Newari, where As take the ergative marker when rhematic (Givón 1984: 154). Thus, (20) would be appropriate as an answer to “Who is breaking the window?,” while (21) as an answer to “What is the man doing?” (Givón 1984: 154): Newari (Givón 1984: 154) (20) Wō ma̧nu-nã jhya tajya-na̧ co-na̧. the man-ERG window break-AUX be-AUX ‘The man is breaking the window.’ (21)
Wō ma̧nu jhya tajya-na̧ the man window break-AUX ‘The man is breaking the window.’
co-na̧. be-AUX
Importantly, both Malchukov (2008a) and McGregor (2009) explicitly relate this pattern to a violation of Du Bois’ “Given-A constraint.” According to Du Bois (1987b), A arguments are overwhelmingly given/definite, hence when they are new and/or indefinite they are expected to be eligible for special marking. Thus, markedness reversal effects in DSM are well attested with respect to definiteness, even though the data available indicate that the relevant distinction may be between topical/given and focal/new NPs, rather than between definite and indefinite ones. It should be noted though that the same conclusion emerges from recent studies of DOM: many DOM patterns which have been attributed to definiteness in earlier studies, have been reinterpreted in terms of information structure (topicality) in more recent work (Iemmolo 2010; Darlymple and Nikolaeva 2011).4 4
It is interesting to observe that this literature, while agreeing that more topical Ps are preferentially marked, offers very different explanations for the attested patterns. On the one hand, Darlymple and Nikolaeva (2011) attribute preferential marking of more topical objects to the fact that objects are ‘secondary topics’ (following Givón who associates subjects with primary topics and objects with secondary topics). This is more in line with the approach of Hopper and Thompson (1980) who attribute preferential marking of more prominent (in particular, animate) objects to their high(er) individuation. On the other hand, Iemmolo (2010) explains preferential marking of topical objects rather by the need to signal non- prototypicality of topical objects, which is obviously more in line with the markedness approach.
Ergativity and differential case marking 265 However, the opposite pattern where the ergative case is missing on low-prominent/ non-referential As is attested as well. Thus, Dixon (1994: 90) observes that some ergative languages restrict ergative marking to definite subjects. In Semelai (Kruspe 1999), specific As take the ergative case and are cross-referenced by the ergative proclitic, while generic As are neither marked nor cross-referenced: Semelai (Kruspe 1999: 253) (22) Cɔ jəl jkɔs. dog(s).DIR bark.at porcupine.DIR ‘Dogs bark at porcupines.’ (23) Jkɔs ki-jəl la-cɔ. porcupine.DIR 3SG-bark.at ERG-dog(s) ‘The dog(s) bark(s) at the porcupine.’ A similar pattern has been reported for Alsea (Mithun 1999), where ergative is restricted to referential As. While in the examples (22)–(23), availability of ergative marking correlates with semantic referentiality (specificity), in other languages it is rather related to discourse referentiality, that is, communicative importance or salience of an argument in discourse (see Givón 1984: 423ff. on a distinction between semantic and pragmatic referentiality). Thus, in Timbe (Givón 2001b: 215, citing Foster) discourse prominent As (persistent topics) are more likely to take the ergative case. And in Gooniyandi, one of the functions of the ergative case is “stressing referentiality of an actor” (McGregor 1992). Thus, in the domain of definiteness we find asymmetries between DOM and DSM patterns similar to those we observed in the domain of animacy. Again, the DOM pattern is cross-linguistically consistent: high-prominent patients, which are specific/referential and given/topical are preferentially marked. Preferential marking of specific patients can be naturally explained in terms of prominence, and therefore attributed to Index (given the constraint ranking: Index-P >> Econ >> {Index-p, Diff-p, Diff-P}). The role of prominence constraints is probably most obvious for incorporating languages where low-prominent (in particular, non-specific) objects tend to be incorporated (Mithun 1984). On the other hand, preferential marking of topical and/or given (hence, usually definite) objects can be attributed to Diff, as given/topical arguments are likely to be construed as otherwise (given the constraint ranking: Diff-P >> Econ >> {Diff-p, Index-P, Index-p}). Yet given a natural correlation between topicality/ givenness, definiteness, and specificity, the effects of the two strategies are often difficult to discern in the domain of DOM. In DSM we find less consistency with respect to definiteness effects. Thus, we find languages, like Semelai and Alsea, where ERG case is reserved only for referential subjects (strong As), which can be attributed to Index (Index-A >> Econ >> {Index-a, Diff-a, Diff-A}). In other languages (Ika, Newari), ERG marking is confined to non-topical, new, indefinite subjects. This pattern is readily explained by Diff (given the constraint ranking: Diff-a >> Econ >> {Diff-A, Index-A,
266 Andrej Malchukov Index-a}), since non-topical/new/indefinite arguments are more likely to be construed as objects than as (transitive) subjects, which are topics par excellence. Thus, asymmetries between DOM and DSM, in the domain of definiteness/topicality are parallel to those observed in the domain of animacy, and can be also accounted for in terms of the same two constraints.5
11.6 Case-M arking Strategies and Formal Classes of DCM We started with a discussion of different strategies of case marking—indexing and differentiating—and proceeded to a discussion of animacy effects peculiar to these strategies. An interesting question to address is whether the different strategies correspond to different patterns of differential case marking. Let me first introduce the fundamental distinction between two formal types of differential case marking, which are frequently confused in the literature, ‘asymmetric’ and ‘symmetric’ (de Hoop and Malchukov 2008; Malchukov 2008a). The ‘asymmetric’ type of differential marking involves an alternation of an overt case with a zero case (e.g. a nominative, in an accusative system, or absolutive, in an ergative system). The second type dubbed here ‘symmetric’ involves an alternation between two overt cases (such as an ergative/oblique alternation for DSM and an accusative/oblique alternation for DOM). Importantly, only the asymmetric type may be due to the differentiating strategy (and economy). It is hardly accidental that Aissen’s (2003) markedness approach was specifically designed to capture variation in asymmetric DOM patterns.6 On the other hand, considerations of disambiguation and economy hardly play a role in the symmetric type, as the opposition of the two markers does not manifest a clear markedness pattern. Indeed, both overt cases involved in a symmetric alternation satisfy the differentiating constraint, and neither of them, being overt, satisfies economy. The two formal types of differential case marking are independent of each other and may co-occur in a single language. By way of illustration, consider a three-way DOM in Finnish. As is well known, ACC in Finnish (formally identical to GEN on nouns) is involved in two different case-alternations (see, e.g., Sulkala and Karjalainen 1992: 212–19). On the one hand, the partitive case replaces the accusative in order to signal a less 5
In their cross-linguistic study of factors underlying DSM, Fauconnier and Verstraete (2014) confirm pervasiveness of patterns of focal ergativity, but claim that preferential marking of focal As can’t be viewed as a reversal of the DOM pattern with marking of topical Ps, because the focus cannot be regarded as the opposite of topicality. Yet what matters here is that A arguments are preferentially associated with (primary) topics/given information, while P arguments are rather associated with focus/ new information. Therefore, any deviation from this scenario (e.g. violation of the ‘Given A constraint’) may require extra marking on the markedness approach. 6 See, however, Keine and Müller (2015) for a proposal as to how this account (reconstructed in Distributed Morphology) can be extended to other cases, when the forms in variation reveal a formal asymmetry but both forms are overt.
Ergativity and differential case marking 267 affected/indefinite P or imperfective aspect (see, Kiparsky 1998a for discussion). On the other hand, ACC is replaced by the unmarked (nominative) form when an A argument is missing, as in impersonal and imperative constructions (Comrie 1975). As predicted, the symmetric alternation is due to the indexing strategy, while the asymmetric alternation is due to differentiating strategy. Note also that in the latter case, case-marking is “global” (e.g. case on P is dependent on the presence of the A argument), a frequent corollary to the differentiating strategy. Similar patterns observed above for two formal types of DOM, are also found in DSM. Here again it is important to distinguish between an asymmetric DSM type (an alternation between ergative and a zero case) and a symmetric alternation (between ergative and an oblique). Again, as in the previous case, only the former pattern can be readily explained in terms of markedness/differentiation. The asymmetric DSM pattern is found in many split-ergative languages of the Australian type, where, in accordance with Silverstein’s generalization, ERG case is not available for nominals high on the animacy hierarchy (Silverstein 1976; Blake 1977). The second—symmetric—DSM type can be exemplified by the “involuntary agent construction” in Lezgian (Haspelmath 1993). Note that ergative switches to an oblique (“adelative”) case to indicate lack of control volitionality on the part of the agent argument: Lezgian (Haspelmath 1993: 292) (24) Ajal-di get’e xa-na. child-ERG pot.ABS break-AOR ‘The child broke the pot.’ (25) Zamiira.di-waj get’e xa-na. Zamira-AdEl pot.ABS break-AOR ‘Zamira broke the pot (accidentally/involuntarily).’ Clearly this DSM pattern is related to the indexing function, as it pertains to encoding of volitionality/control. Similar patterns of symmetric DSM manifesting “involuntary agent construction” are attested elsewhere (see Kittilä 2002 for an overview). Thus, also in case of DSM we observe a correlation between a case-marking strategy and a formal pattern: an indexing strategy yields a symmetric DSM, while a differentiating strategy yields an asymmetric DSM.
11.7 Case-M arking Strategies and Distributional Types of DCM Above we have noted that differential case marking of the symmetric type shows correlation with the indexing strategy, while the asymmetric type with the differentiating strategy. Earlier a somewhat similar observation has been made by DeLancey (1981)
268 Andrej Malchukov who argued that one should distinguish between alternations of the ergative with an overt (oblique) case (as in many consistently ergative languages), and alternations of the ergative with zero (as in split ergative languages of the Australian type). According to DeLancey (1981), the former type is related to agentivity, while the latter type is not. Yet, as noted by Næss (2007), this is an oversimplification since alternations of the latter type (i.e. an asymmetric DSM) may be conditioned by agentivity as well. For example, in Tibetan (Denwood 1999) ERG marking is lost when A is non-agentive ( Kittilä 2002: 239; Næss 2007): Tibetan (Denwood 1999: 137) (26) Sang.nyin nga-s las.ka ‘di byed.kyi.yin. tomorrow I-ERG work this do.LINK.AUX.VOL ‘I shall do this work tomorrow (of my own free will).’ (27) Sang.nyin nga las.ka ‘di byed.kyi.yin. tomorrow I work this do.LINK.AUX.INVOL I shall do this work tomorrow (whether I like it or not).’ Note that even though the DSM in Tibetan is asymmetric, it is driven by agentivity distinctions, not unlike those which we observed for symmetric DSM in Lezgian. A similar point of course can be made for Hindi, where the choice between ergative and nominative case on intransitive subjects is sensitive to volitionality (as in (13)–(14)). Here, however, an important distinction between asymmetric alternations in split ergative Australian languages such as Dyirbal, on the one hand, and in Tibetan and Hindi, on the other hand, should be noted. The asymmetric marking in the Australian languages is of a “split” type, where different types of nominals select for different cases, while the alternation observed for other languages is of the “fluid” type, where the same nominal can alternatively take both case markers (the terms “split” vs. “fluid” have been adopted by de Hoop and Malchukov (2007) from Dixon’s (1994) classification of split- intransitive languages). Importantly, the “fluid” type of transitivity alternation correlates with the indexing strategy, even in situations where case alternation is asymmetric. The reason for this seems to be a general constraint against existence of two synonymous forms:7 if there are two alternative forms in variation there should also be some concomitant meaning differences associated with them. (We abstract away from cases where competition of two forms is indicative of language change.) In case two forms are in a complementary distribution they of course do not violate this constraint. This means, in effect, that whenever some two case forms are in alternation they will correlate with meaning differences, thus falling under the domain of the indexing strategy.8 7 This constraint is well known from the research in the field of lexical semantics/pragmatics, where it is held responsible for the phenomenon of (lexical) blocking (see Kiparsky’s 1982a constraint Avoid Synonymy). 8 See de Hoop and Malchukov (2007) for an argument that fluid alternations can be adequately described by bidirectional OT, which computes (super)optimal candidate meaning–form pairs. This is
Ergativity and differential case marking 269
11.8 Case-M arking Strategies and Typology of Differential Case Marking: Recapitulation Above we have observed certain correlations between the type of differential case- marking, on the one hand, and case-marking strategies, on the other. Thus, it was shown that only the asymmetric type can be due to the differentiating strategy. For symmetric DCM, issues of distinguishability and economy are irrelevant, as it does not manifest a markedness pattern. On the other hand, DCM of the fluid type shows a correlation with the indexing strategy. The latter correlation has been attributed to the fact that availability of paradigmatic opposition will always invite a semantic contrast, while there need not be such a contrast in DCM of the split type where two forms are in complementary distribution. Provided that the fluid DCM correlates with the indexing strategy, and the asymmetric DCM correlates with distinguishability, the connections between case-marking strategies and types of DCM—both formal and distributional—can be represented as in Table 11.3. Table 11.3 Differential case-marking types and case-marking strategies Symmetric
Asymmetric
Fluid
Indexing
Indexing/ Differentiating
Split
?
Differentiating
As is clear from Table 11.3, the two types of DCM which show diametrically opposite correlations in terms of strategies involved are the symmetric DCM of the fluid type, on the one hand, and the asymmetric split type, on the other hand. The former illustrated above by the subject alternation in Lezgian (as in (24)–(25)) is always due to the indexing strategy, while the latter as found in many split ergative languages of the Australian type is due to the differentiating strategy. The other two possible types (the asymmetric fluid type, and the symmetric split type) can’t be unambiguously associated with either
so because the bidirectional OT, as advocated by Blutner (2000) and others, is specifically designed to capture iconicity effects, when a marked (more complex) expression is associated with a more marked (nonstandard) meaning, and the unmarked expression is associated with a standard meaning, whenever the two expressions appear in direct competition.
270 Andrej Malchukov of the two strategies. As noted above, the asymmetric fluid DSM, can be either due to the indexing strategy (see DSM in Tibetan, as in (26)–(27)), or to the “global” differentiating strategy (see DSM in Fore, as in (11)–(12)). Furthermore, it can be arguably due to both the indexing and the differentiating strategy as observed in the canonical cases of DOM, explained by Silverstein and Aissen in terms of markedness (see, e.g., the DOM pattern in Hindi in (1)–(3), where ACC marking targets only animate and/or definite objects). An opposite combination of features, the symmetric split pattern appears to be outside the scope of either indexing or differentiating strategy. Since it is symmetric it cannot be due to the differentiating strategy. On the other hand, since the case forms do not alternate with each other, the choice between the two forms does not invite a semantic contrast, and is more likely to be a morphological matter. Thus, alternative forms for the object case with different classes of nominals, as familiar from Indo-European languages, are usually taken to represent different declensional types (Baerman 2009: 224), and the same seems to hold for the split realization of the A-case in ergative languages (e.g. according to Spencer 2006, the two different ergative markers used for animates and inanimates in Chukchi should be regarded as different morphological realizations of a single syntactic case).
11.9 Ergativity, DCM, and Diathetic Alternations It has been observed in the literature, that differential case marking is also dependent on language typology: in particular, DOM is more common in nominative–accusative languages, while DSM is more common in ergative languages (Bossong 1985; Drossard 1991: 452, 479). A correlation between accusative alignment and availability of DOM alternations, on the one hand, and ergative alignment and availability of DSM alternations, on the other hand, is also evident in the case of split ergative languages with a tense/aspect based split. Thus, in Indo-Aryan languages showing a tense-based split, DSM of the type illustrated for Hindi in (13)–(14) is only found in the ergative domain (perfective tenses), and is excluded from the accusative domain (imperfective tenses). Although DOM is found in many languages (including Hindi) in both domains, in some languages, like Kashmiri, it is restricted to the accusative domain (see Klaiman 1987: 77). To account for the skewed typology of DCM with respect to ergative and accusative languages, Malchukov (2006) introduces the economy constraints PAIP, which was originally an abbreviation of “Primary Actant Immunity Principle.” • PAIP: Avoid (case) marking of the primary argument. The primary argument (or primary term in Palmer’s 1994 terminology) refers to the argument of a transitive clause which is encoded like the intransitive subject (S). In
Ergativity and differential case marking 271 effect, PAIP penalizes case marking of an (otherwise) unmarked argument. Thus, in general, PAIP penalizes morphological case marking of the absolutive argument in ergative languages and of the nominative argument in nominative–accusative languages. Of course, this can be seen as a reformulation of a constraint that states that one argument should always bear the unmarked case, or to put it differently, that the unmarked case (either nominative or absolutive) is obligatorily present in every sentence of a language. This constraint thus resembles Tsunoda’s (1981b) “Unmarked Case Constraint” as well as Bobaljik’s (1993b) “Obligatory Case Parameter.” On Malchukov’s (2006) account the economy constraint PAIP competes with the iconically motivated Relevance Principle, which is formulated in reliance to Hopper & Thomson’s (1980) transitivity parameters: • Relevance Principle: mark a transitivity parameter on the ‘relevant’ constituent (i.e. on the constituent to which a parameter pertains) Relevance Principle may be seen as a manifestation of Indexing constraints (belonging to Identify family in de Hoop and Malchukov 2008), but in addition carries a locality requirement, which (trivially) assumes that an argument should be marked for its own theta-role. Relevance Principle would predict that an A-related parameter (such as subject volitionality, or sentience) will be preferentially encoded on the A argument, while a P-related parameter (such as P-affectedness, or P-individuation) would be preferentially marked on the P. This is, of course, the most usual case: manipulation in P-related parameters is most often manifested in DOM rather than DSM, while manipulation in A-related parameters is manifested in DSM rather than DOM (this is also clear from Kittilä’s cross-linguistic survey of transitivity alternations; Kittilä 2002). Yet, occasional counterexamples to this principle are also found (Næss 2004; Malchukov 2006). Consider the following examples from Shipibo-Conibo (Valenzuela 1997), also discussed by Kittilä (2002) and Næss (2004). In this Panoan language, an A must be marked by the ergative case only when a P is referential: Shipibo-Conibo (Valenzuela 1997; Kittilä 2002: 212) (28) e-n-ra yapa-ø pi-kas-ai. I-ERG-AS fish-ABS eat-DES-INCOMPL ‘I want to eat fish (referential only).’ (29) ea-ø-ra yapa-ø pi-kas-ai. I-ABS-AS fish-ABS eat-DES-INCOMPL ‘I want to eat fish (referential or non-referential).’ This case-marking pattern is thus ‘global’ in Shipibo-Conibo: as in the Fore examples (11)–(12) above, case marking of one argument depends on the feature of another clausal argument. Yet, differently from Fore, it is referentiality rather than animacy that is at issue here; therefore this pattern is less naturally explained by distinguishability. On the
272 Andrej Malchukov other hand, capturing such global patterns from an indexing perspective is not straightforward either.9 A more general solution would be to attribute this case alternation to economy. On this approach, a markedness/economy principle like PAIP (Malchukov 2006) would be held responsible for the fact that in an ergative language a P-feature could affect encoding of A rather than the primary absolutive argument. In the example above, Relevance is violated in order to satisfy PAIP, resulting in P- features being A-marked in an ergative language or A features being marked on P in an accusative language (Malchukov 2006). Similarly, PAIP can be violated in order to satisfy Relevance. An example of a violation of PAIP comes from Warlpiri (Hale 1973a): Warlpiri (Hale 1973a) (30) Njuntulu-lu npa-tju ŋatju. 2SG-ERG 2SG-1SG speared.1SG ‘You speared me.’ (31)
Njuntulu-lu npa-tju-la ŋatju-ku. 2SG-ERG 2SG-1SG-la speared.1SG-DAT ‘You speared at me.’/‘You tried to spear me.’
Note that violation of PAIP in the ergative-dative pattern in (31) is due to Relevance: the need to signal decreased affectedness of the P argument. This pattern is, however, rather exceptional for ergative languages which mostly adhere to PAIP (or more generally, to the Unmarked Case Constraint; Tsunoda 1981b). Indeed, in most ergative languages the ERG-OBL pattern is disallowed, and demotion of a P in a would-be DOM pattern triggers a shift to an antipassive construction. Warrungu (Tsunoda 1985) (32) Pama-ngku yuri nyaka-n. man-ERG kangaroo-ABS see-NF ‘A man saw (found) a kangaroo.’ (33) Pama yuri-wa nyaka-kali-n. man.ABS kangaroo-DAT see-APASS-NF ‘A man was (is) looking for a kangaroo.’
9 One approach to account for such cases has been developed by Næss (2004, 2007). Næss, who generally adopts an indexing approach, suggests that a semantically transitive clause should additionally satisfy the condition of Maximal Semantic Distinguishability of its arguments: if they are not maximally distinguished (e.g. the subject is non-volitional, or object is non-affected), the transitive pattern may shift to intransitive. On this approach, a correlation between ergativity and DSM, on the one hand, and accusativity and DOM, on the other hand, is not expected.
Ergativity and differential case marking 273 Thus an antipassive construction is regularly used in ergative languages to signal decreased affectedness or referentiality of an object (Cooreman 1994). By using the antipassive construction, the subject becomes the unmarked argument (in the absolutive case), which means that PAIP is fulfilled. Similarly, we see evidence for PAIP in the nominative–accusative languages as well. While in ergative languages a change in the strength/prominence of the subject can affect the form of the subject exclusively (resulting in DSM), in nominative–accusative languages, on the other hand, manipulating the strength/prominence of an agentive argument regularly leads to passivization. As noted by Shibatani (1985), a passive is primarily used for ‘agent-defocusing,’ when an agent is indefinite, non-specific or not important in the discourse. Similarly, in some languages passive forms are used to indicate non-volitionality of the subject (see Masica 1991 on Sinhala and Dhivehi). In ergative languages, on the other hand, lack of volitionality on the part of the A argument is more likely to be signaled through DSM (see (25) from Lezgian). Thus, features that trigger DSM in ergative languages may cause the use of a passive construction in nominative– accusative languages. Again, this can be straightforwardly explained by PAIP. To sum up, we have seen harmonic cases, where Relevance (Indexing) and PAIP reinforce each other. That is, both constraints can be satisfied in case of DOM in nominative– accusative languages and DSM in ergative languages. When the two constraints are in conflict, we commonly see that a voice alternation is a way to resolve the conflict. As predicted, passivization is applied when there is need to encode a subject alternation in Table 11.4 The relation between case and voice alternations Input
Output in a nominative–accusative language
Output in an ergative–absolutive language
A/a prominence
Active/passive voice alternation
DSM
P/p prominence
DOM
Active/antipassive voice alternation
a nominative–accusative language, while antipassivization applies when an object alternation must be encoded in an ergative language (Malchukov 2006). This is illustrated in Table 11.4 (adopted from de Hoop and Malchukov 2008).10
10 This is in line with Legendre et al. (1993) who also argue that passives apply when the input is aP (with a non-prominent subject), while antipassives apply when the input structure is Ap (with a non- prominent object). However, Legendre et al. do not account for the fact that passives are found more often in nominative–accusative languages, while antipassives are found more often in ergative languages.
274 Andrej Malchukov
11.10 Diachronic Issues In conclusion let us briefly address diachronic scenarios behind evolution of DSM in ergative languages. As is clear from the discussion above, there are different types of DSM available; in particular, the distinction should be drawn between alternations of the asymmetric and symmetric type, on the one hand, and alternations of the split and fluid type, on the other hand. We will not attempt a comprehensive coverage of all scenarios leading to these systems. This is because the explanation of the scenarios leading to an asymmetric type amounts to explanation of the developments of the ergative system, which is covered in the volume elsewhere (see McGregor, Chapter 19, on grammaticalization of the ergative case). Suffice it to note that one well attested path of development of an ergative pattern is through reanalysis of an intransitive (resultative or passive) pattern, with the ergative originating in an oblique. More to the point is the development of the symmetric system with an oblique alternating with the ergative case. Also in this case, it is clear that the pattern with an oblique case was originally intransitive. In fact, some of the cases described above in terms of DSM remain synchronically controversial. Thus, the Lezgian pattern with an adelative case in (25) has been described as basically intransitive (Haspelmath 1993), since it is not found with canonical transitives, but rather with verbs that are either intransitive or labile (like ‘break’ in (24)–(25)). While this analysis may indeed be appropriate for Lezgian, it should be noted that for some related languages, reanalysis of an intransitive to a transitive pattern proceeded further. Thus, in Agul (another Daghestanian language), the involuntary subject construction is also primarily found with intransitive and labile verbs, yet the adelative argument does not show any decrease of syntactic subject properties in comparison to the ergative subject (Ganenkov et al. 2008). At this stage, the non-volitional agent can be adequately analyzed as a non-canonical subject, that is, a subject showing exceptional marking, but displaying syntactic characteristics of a regular subject. Finally, at a still later stage, the variation between alternative case patterns may be ‘morphologisized’ and become associated with certain types of nominals, yielding an alternation of the split symmetric type. Thus, while the status of individual languages with respect to the cline may be controversial, the general stages of this development may be represented in the following way (see also Seržant 2013): • 1st stage: the initial construction is intransitive, with an oblique Cause/Agent • 2nd stage: the agent acquires syntactic subject properties (is reanalyzed as a non- canonical subject) • 3rd stage: the alternation is conventionalized as DSM (of the split symmetric type) • 4th (final) stage: the construction is realigned to standard coding This diachronic perspective can also throw light on a controversial question of the taxonomy of differential case-marking phenomena. It has been claimed by a number of authors, that differential case marking subsumes two different phenomena, some of them
Ergativity and differential case marking 275 being syntactic in nature, some other morphological (see, e.g., Woolford 2008; Legate 2014a; see also Spencer 2006 for a distinction between m-case and s-case). Other authors, however, treat differential case marking in a unified way, in syntax (e.g. Coon and Preminger, Chapter 10, this volume), or in morphology (e.g. Keine and Müller 2015). The same controversy is also found in the typological literature (see Goddard 1982), in particular, in the discussion of the relation between split ergativity and case syncretism (see also two contributions in the Oxford Handbook of Case by Iggesen (2009) and Baerman (2009) with rather different takes on this issue). From the discussion above, it should be clear that we view these distinctions as a part of a diachronic cline, which indisputably starts from different syntactic structures (possibly differing in transitivity value), which develops through a stage of noncanonical argument marking into a patterns of case syncretism. As also acknowledged by Baerman (2009: 229), who is generally a proponent of the syncretism approach, “the most widespread type of case syncretism, that of the core cases, may in many instances represent the outcome of desyntacticization, that is, the morphologized relic of what was once an active syntactic rule.” Another diachronic scenario leading to ergativity, or, rather, split intransitivity, involves transitive-to-intransitive reanalysis. This scenario involves reanalysis of transitive impersonal constructions (‘transimpersonals’; Malchukov 2008b) into a patient subject construction in a split intransitive language. (While split intransitivity might be viewed as a distinct phenomenon, it does yield a restricted ergative S=P alignment). As shown by Malchukov (2008b) this is a common grammaticalization path. This path starts from a transitive construction with a ‘weak’ (non-agentive, non-prominent, non- referential) A and involves reanalysis of a prominent P argument into a non-canonical subject (and later eventually into a canonical subject). Koasati (Kimball 1991), like many other Muskogean languages, may serve as an illustration. In Koasati, a number of intransitive (unaccusative) verbs pattern with transitives insofar as the experiencer argument is indexed through object agreement (rather than subject agreement used elsewhere). In fact, if not for the case marking (which operates on the nominative–accusative basis and independently of agreement), the patient subject pattern in (34) could be confused with a transitive pattern in (35): Koasati (Kimball 1991: 251) (34) (Anó-k) ca-libatli-t. (I-NOM) 1sg.obj-burn-past ‘I got burned.’ (35) Nihahci ikba-k ca-libatli-t. grease hot-NOM 1sg.obj-burn-past ‘The hot grease burned me.’ Given similarity of the two constructions, Kimball (1991) suggested that diachronically the patient subject pattern originated from reanalysis of transitive verbs with a third person (singular) agent. As independently observed by Mithun (2008) and Malchukov
276 Andrej Malchukov (2008b), the features which facilitated this reanalysis are: (a) the presence of object agreement on the verb; (b) the zero marking of a 3rd person subject. Now, as shown in both publications for Amerindian and other languages, these two features are typical of split intransitive languages, suggesting that transimpersonal constructions constitute an important historical source for split intransitivity (Malchukov 2008b; see Malchukov and Ogawa 2011 for further discussion). As noted by Malchukov (2008b), the functional motivation behind reanalysis of transimpersonal constructions into the patient subject construction is the functional pressure for realignment of the most prominent argument with the most prominent grammatical relation. The same explanation pertains to other cases of reanalysis, such as reanalysis of object experiencers into subject experiencers (Haspelmath 2001). Such cases of “please-to-like reanalysis” are well known from English (Allen 1995) and are also attested in other European languages (see Seržant 2013 for some discussion). Interestingly, the transimpersonal scenario above is arguably responsible for one of famous generalizations in generative literature, Burzio’s generalization. (36) Burzio’s generalization: All and only the verbs that can assign an external theta role to the subject can assign accusative Case to an object. (Burzio 1986: 178) The synchronic validity of this generalization may be contested; in fact, the transimpersonal constructions of the type illustrated above may be viewed as counterexamples, if taken at face value. Also, the theoretical status of this generalization remains controversial; in spite of its popularity there does not seem to be a consensus, how it can be derived from a more general principle (see Reuland 2000 for a variety of proposals). Now, I would like to suggest that this generalization makes a perfect sense diachronically, insofar as transimpersonal constructions lacking a referential A tend to be reanalyzed into intransitive structures. Thus the rationale behind this generalization should be stated in diachronic terms.11
11.11 Conclusions The present chapter argues that patterns of (differential) case marking can be explained in reliance to two general constraints related to (role)-indexing, on the one hand, and
11
In a recent article, Faarlund (2013) makes a somewhat similar suggestion relating reanalysis of object experiencers into subject experiencers to the loss of the pro-subject in mainland Scandinavian languages. However, Faarlund does not explicitly relate reanalysis of transimpersonal constructions in Scandinavian to Burzio’s generalization.
Ergativity and differential case marking 277 distinguishability (or markedness) on the other hand. This approach correctly predicts asymmetries between DOM and DSM with regard to animacy, definiteness, as well as discourse features. With regard to all of these features, DSM has been found less consistent than DOM, in particular, an ergative case may be associated with either less prominent arguments (the “Silverstein pattern”), or more prominent ones (the “anti- Silverstein pattern”).12 The same approach can be also used to explain functional import of different varieties of DCM, symmetric, and asymmetric, split and fluid. I also show how this approach can be extended to capture a relation between case and voice alternation, as well as briefly outline diachronic scenarios leading to different types of differential case marking. A number of questions have been left unaddressed in this chapter for space reasons. Thus, I have not discussed TAM-based split ergativity, which shows conceptual overlap with differential case marking. Yet, the approach outlined above can be naturally extended to cover such cases. Indeed, Malchukov (2015) and Malchukov and de Hoop (2011) present a comprehensive hierarchy of TAM-based ergativity splits, and show how these splits can be represented on the basis of the constraints assumed above (indexing, distinguishability, economy). In order to capture a distinction between languages with morphological and syntactic ergativity one needs to invoke additional factors, yet also this distinction can also be fruitfully explored in a typological model based on competing motivations (which can be recast as OT constraints). In particular, Malchukov (2014) explains how patterns of syntactic ergativity/accusativity can be captured through interaction of Bias constraints (embodying functional preferences) with Harmony constraints (embodying the analogical influence from coding to behavior). Other contributions to the same volume (MacWhinney, Malchukov, and Moravcsik 2014), provide further evidence for the power of the competing motivations approach.
Acknowledgments The research reported in this chapter has been carried out partially in collaboration with Helen de Hoop; I am grateful to Helen for useful feedback and discussions in the course of
12
There are still few large scale typological statistical studies testing these predictions. One of such studies addressing both DSM and DOM (Bickel et al. 2015) found that DSM is indeed less consistent than DOM, which is in line with the present approach. Yet, Bickel and coauthors further showed that even for DOM evidence for the hierarchy effects is weaker than expected and furthermore that DOM patterns are prone to areal diffusion. On the other hand, the study by Sinnemäki (2014), which is confined to DOM, confirmed that DOM is a typologically predominant pattern; in fact, it was found to be more common than a consistent object marking. However, also this study could not identify either animacy or definiteness as the main contributing factors behind DOM. This finding, if confirmed, is consistent with approaches viewing animacy effects as epiphenomenal (see Malchukov 2008a), as well as with approaches treating definiteness effects as derivative from discourse factors (see Iemmolo 2010; Darlymple and Nikolaeva 2011).
278 Andrej Malchukov our collaborative work. I am also grateful to the editors for editorial feedback as well as to the anonymous reviewer for the useful comments on an earlier version of this chapter. The usual disclaimers apply.
Abbreviations ABS, absolutive; ACC, accusative; AdEl, adelative; AGT, agentive; AOR, aorist; APASS, antipassive; ART, article; AS, ascertained (direct) evidential; AUX, auxiliary; CAUS, Causative; D, dative; DAT, dative; DES, desiderative; DIR, directional; DIST, distal deictic aspect; ERG, ergative; FA, factive aspect; FS, feminine singular; INVOL, involuntary; INCOMPL, incompletive; LINK, linker; LOC, locative; MED, medial deictic aspect; NF, non-future; NOM nominative; OBJ, object; OT Optimality Theory; P, patient; PAST, past; PERF, perfective; SG, singular; VOL, voluntary.
Chapter 12
Three-way syst e ms d o not ex i st Gereon Müller and Daniela Thomas
12.1 Background There are three main types of encoding systems (via case or agreement) for core arguments in the world’s languages, where core arguments are external (DPext) and internal arguments (DPint) of transitive (Vt) and intransitive verbs (Vi) that are not lexically case- marked. In ergative systems, DPext of Vt is marked in some special way, as ergative, that is different from the marking of DPint of Vt, DPext of Vi (unergative intransitive context), and DPint of Vi (unaccusative intransitive context), all of which bear absolutive; see (1-a). In contrast, in accusative encoding systems, the DPint of Vt is singled out (by bearing accusative), and the remaining core arguments are grouped together (nominative); see (1-b). Finally, in active systems, argument encoding is oblivious to transitivity: DPext is encoded in one way, and DPint is encoded in some other way, with both Vt and Vi; see (1-c). Such an active system of argument encoding may then in principle qualify either as an ergative system at its core (where the distribution of the ergative is extended to DPext of Vi; see Dixon (1994)), or as a modified accusative system (where the distribution of the accusative is extended to DPint of Vi; see Bittner & Hale (1996a,b)). (1) a. Ergative system
b. Accusative system
c. Active system
DPext-Vi DPint-Vi
DPext-Vi DPint-Vi
DPext-Vi
DPint-Vi
DPext-Vt DPint-Vt
DPext-Vt DPint-Vt
DPext-Vt
DPint-Vt
erg/nom
abs/acc
erg
abs
nom
acc
These basic systems of argument encoding have been given simple accounts in Principles-and-Parameters-type approaches. A common basic idea that has often been
280 Gereon Müller and Daniela Thomas pursued is that there are uniformly two functional heads (which for present purposes we may identify with T and v) that assign one structural case each in transitive contexts; and only one of the two case assigners remains active in intransitive contexts. On this view, of the four cases currently under consideration—ergative, absolutive, nominative, accusative—there are actually only two: one assigned by T, and one assigned by v. Throughout this article, we adopt a version of this approach according to which ergative is identified with accusative, and nominative with absolutive; see Murasugi (1992), Jelinek (1993), Ura (2000; 2006), Müller (2009), and Assmann et al. (2015).1 More specifically, given a TP-vP analysis of clause structure, ergative/accusative is a structural case assigned by v to either Specv or CompV in transitive contexts, and nominative/absolutive is uniformly the case assigned by T. As shown in (2), on this view ergative and accusative systems differ in transitive contexts. Here v assigns case to DPext in ergative systems, and to DPint in accusative systems. T assigns case to the remaining argument (i.e. DPint in ergative systems, and DPext in accusative systems).2 (2) Transitive context TP T'
T
nom abs
vP v'
DPext v acc
1
erg V
VP DPint
Alternatively, it has been suggested that ergative is to be identified with nominative, and accusative with absolutive; see Levin & Massam (1985), Bobaljik (1993a), Laka (1993b), Chomsky (1995, ch. 3), Rezac (2003), and Bobaljik & Branigan (2006), among others. Most of what follows could also easily be made compatible with this view. 2 Here and in what follows, accusative encoding is represented by dashed lines in syntactic trees, ergative encoding by full lines. Note also that in our reconstructions of existing approaches, we adopt an Agree-based approach (as in Chomsky (2001; 2008)) throughout, and disregard the possibility of case assignment being tied to movement of DPs.
Three-way systems do not exist 281 On this view, the parameter distinguishing ergative and accusative systems exclusively concerns v: Both upward and downward case assignment must be possible in principle, but there is a preference for upward case assignment in ergative systems, and a preference for downward case assignment in accusative systems.3 In contrast, ergative and accusative systems work in exactly the same way in intransitive contexts: Only T remains as a case-assigning head here, with both DPint (see (3-a)) and DPext (see (3-b)). This corresponds directly to tendencies of morphological marking: The case associated with T is typically morphologically less marked than verbal case assigned by v. (3) Intransitive unaccusative and unergative contexts a.
TP
b.
TP
T' vP
T nom
T'
v'
–
vP
T
v abs V
abs
VP DPint
v'
DPext
nom
v
VP V
–
This general approach can then be extended to active systems in various ways. For instance, one can postulate that whether T or v is the sole case assigner in intransitive contexts may be governed by different conditions, among which languages can choose. To sum up so far, this kind of analysis provides a simple account of the basic patterns of argument encoding in the world’s languages. Given the identification of two case- assigning heads with two structural cases, the question arises of how this approach fares with three-way (tripartite) systems of argument encoding, as they are schematically depicted in (4) (see Dixon (1994), among others). 3
Assmann et al. (2015) explicitly postulate a “Specifier-Head Bias” which, if accompanied by an opposite “Complement-Head Bias,” would capture the parametrization.
282 Gereon Müller and Daniela Thomas (4) Three-way system: DPext-Vi
DPint-Vi nom/abs
DPext-Vt
DPint-Vt
erg
acc
In a three-way system, there are three structural cases for core arguments, i.e. nominative/absolutive for the sole DP of Vi, ergative for DPext of Vt, and accusative for DPint of Vt. As described in Bittner & Hale (1996b), the Central Australian language Antekerrepenhe instantiates such a three-way system; see (5). (5) a. Arengke-le aye-nhe dog-erg me-acc ‘The dog bit me.’
ke-ke bite-pst
b. Apwerte-le athe arengke-nhe we-ke stones-ins I-erg dog-acc pelt-pst ‘I pelted the dog with stones.’ c. Arengke nterre-ke dog-nom run-pst ‘The dog ran.’ Three-way systems are cross-linguistically rare.4 However, they are potentially problematic for the type of analysis sketched above, where two case assigners (T, v) are responsible for two structural cases and each of {erg, acc} is identified with exactly one case of the other system.5 What can be done in view of this state of affairs? There are basically two kinds of solutions to the problem posed by three-way systems. One possible way out is to enrich the
4 They qualify as non-canonical from a typological perspective (see Corbett (2005); Corbett & Fedden (2014)). Note that this assessment presupposes that the voice systems of Austronesian languages do not instantiate three-way patterns (with an accusative in actor topic contexts and an ergative in theme topic contexts); see Aldridge (2004), Paul & Travis (2006) for discussion. 5 This conclusion does not necessarily hold for approaches where structural case assignment in transitive contexts is relational; see Marantz (1991), Bittner & Hale (1996b), Wunderlich (1997; 2006), Kiparsky (1999), Stiebels (2002), McFadden (2004), Schäfer (2012a), and Baker (2015). Here, there are two separate sources for the ergative and for the accusative; and nominative/absolutive is a default case for DPs in intransitive contexts. However, it is not clear whether these analyses have anything interesting to say about the cross-linguistic rarity of three-way systems. For this reason, we will disregard relational approaches in what follows.
Three-way systems do not exist 283 syntactic analysis with ad hoc assumptions about additional case features. Such a strategy will technically work (see Müller (2009)), but it does not strike us as particularly insightful. A second option is to leave the syntactic analysis as is, thus exclusively deriving two-way systems, and relocate the phenomenon to morphology. On this view, three- way case systems do not exist as a syntactic phenomenon. It is this latter view that we will pursue in what follows. The challenge for a morphological reanalysis of seemingly tripartite systems is then twofold. First, it must be shown that a simple morphological analysis can indeed be given, one that does not have to resort to unnatural classes (see Bierkandt (2006) for this objection, based on evidence from Diyari), and that is ideally independently corroborated. And, second, it must be shown that the hypothesis that syntactically there are only two cases where traditionally three have been postulated, can be maintained in view of the classes of DP arguments that syntactic operations access in the relevant languages. The main bulk of the present chapter addresses the first question in sections 12.2 and 12.3; we turn to the second question in section 12.4. A first indication that a morphological reanalysis of three-way systems might be worth pursuing is that independent evidence for distinguishing between case as a syntactic category and case as a morphological exponent has recently come to the fore. On the one hand, one and the same morphological case exponent may correspond to two different syntactic cases. Thus, as has been argued in Legate (2008), a zero exponent may either indicate a syntactic nominative, or it may act as the default realization of some other case (like accusative or ergative), depending on the language. On the other hand, one and the same syntactic case may correspond to two different morphological case exponents in a given language. In particular, it is argued in Keine & Müller (2011; 2015) that scale-based differential object marking should be realized as a morphological phenomenon (i.e. as scale-driven allomorphy associated with a single syntactic case). Crucially, three-way systems typically also involve scale effects, such that, e.g., only non-prototypical DPint arguments receive what at first sight looks like an accusative, or only non-prototypical DPext arguments bear what at first sight looks like an ergative. This situation obtains in Nez Perce (see Rude (1985), Woolford (1997)): There is (what has been called) an accusative exponent for DPint of Vt; there is (what has been called) an ergative exponent for DPext-3rd-person of Vt (but not for other, more typical DPext types with Vt, i.e. for 1st/2nd-person); and there is a nominative (zero) exponent for DPint, DPext of Vi, and for DPext-1st/2nd-person of Vt. All this is shown by the data in (6). (6) a. Kaa wéet’u’ núun- e ká’la hinéesqicxne and not 1pl-acc just 3nom.pldo.take.care.of.perf ‘And he just didn’t take care of us.’ (Rude (1985, 93)) b. ‘Iceyéeye- nm xáxaasna hináaswapci’yawna coyote-erg grizzly-acc 3nom.pldo.kill.perf ‘Coyote killed the grizzlies.’
(Rude (1985, 88))
284 Gereon Müller and Daniela Thomas c. (i) Núun ∅-papáayna we 1/ 2nom-pl.nom.arrive.perf ‘We arrived.’ (ii) Núun ’epe’wíye we 1/ 2trANS.shoot.perf ‘We shot him.’
(Rude (1985, 85))
Given this state of affairs, our main goal in what follows is to show that the morphological approach to differential object marking in terms of scale-driven impoverishment developed in Keine & Müller (2011; 2015) (on the basis of Aissen (1999b; 2003)) can be extended to three-way systems without major problems. It turns out that only one important new assumption is required: In addition to the standard prominence scales related to person, animacy, and definiteness (going back to Hale (1972) and Silverstein (1976)), we postulate that there is also a transitivity scale which participates in harmonic alignment processes that eventually bring about post-syntactic impoverishment. We will argue that, as a consequence of scale-driven impoverishment, case features are deleted in certain contexts (intransitive and “prototypical” transitive contexts in particular), and this leads to zero exponence with certain arguments bearing structural ergative or accusative case. Thus, we contend that what seem to be three-way systems on the surface are underlyingly common ergative or accusative systems with overt markers for each of the two cases that disappear in intransitive (and, typically, other) contexts.
12.2 Theoretical Assumptions The optimality-theoretic approach to prominence scale-based differential argument encoding developed in Aissen (1999b; 2003) in terms of harmonic alignment and local conjunction does not distinguish between case as a morphological category and case as a syntactic category; it predicts that variation in, say, differential object marking can only be between an overt case exponent (i.e. presence of case) and no exponent (i.e. absence of case). In Keine & Müller (2015), it is shown that variation in argument encoding that is governed by exactly the same prominence scales can also be between two different overt case exponents; so there are non-zero/non-zero alternations just as there are non-zero/zero alternations. To capture both kinds of effects, Aissen’s approach is reconstructed as an optimization procedure that applies at the interface between syntax and (post-syntactic) morphology, and that deletes certain (but not necessarily all) subfeatures of syntactic cases that must independently be assumed to capture instances of syncretism (see Jakobson (1936), Bierwisch (1967)). For instance, assuming that the accusative is composed of the abstract case features [+gov,–obl] in syntax, post-syntactic optimization may result in the deletion of [–obl]
Three-way systems do not exist 285 but not [+gov] in certain (prototypical) contexts, and subsequent vocabulary insertion (Halle & Marantz (1993)) may then choose a vocabulary item /α/bearing only the feature [+gov] rather than the otherwise expected, more specific vocabulary item /β/ characterized by the features [+gov,–obl] because the latter is not compatible with the insertion site anymore (it does not realize a subset of the features of the syntactic context). Thus, these optimization procedures can be viewed as principled versions of impoverishment rules as they have widely been adopted in Distributed Morphology (see Halle & Marantz (1993)). Of course, in many instances the less specific morphological exponent that must be chosen after feature deletion will be a zero exponent; but it does not have to be, and this provides an argument for a morphological (rather than syntactic) approach. All that said, whereas we will presuppose the morphological version of Aissen’s harmonic alignment approach to differential argument encoding developed in Keine & Müller (2011; 2015) in what follows, we will not actually consider data where there is a non-zero/non- zero alternation; rather, all alternations discussed below will be between non-zero and zero exponents. To begin with, suppose that the core structural cases are defined by the features in (7). (7)
Feature decomposition of cases a. ergative/accusative: [+gov–obl] b. absolutive/nominative: [–gov,–obl]
(assigned by v) (assigned by T)
Ergative and accusative are [+governed] cases assigned by v, and absolutive and nominative are [–governed] cases assigned by T. Both cases are structural (i.e. [–oblique]), which separates them from lexical and oblique cases; however, since the latter do not play a role for the data under consideration in this chapter, we will generally ignore the feature [±obl] in what follows. Next, consider the prominence scales in (8). The scales in (8-abc) go back to Hale (1972), Silverstein (1976), and Aissen (1999b; 2003). We take them to be ontological primitives in grammatical theory but will remain neutral as to their ultimate source (part of the language faculty or grounded in some extralinguistic domain). (8)
Scales: a. Person scale: Local Pers. (1,2) > 3. Pers. b. Animacy scale: Hum(an) > Anim(ate) > Inan(imate) c. Definiteness scale: Pro(noun) > Name (PN) > Def(inite) > Indefinite Specific (Spec) > NonSpecific (NSpec) d. Transitivity scale: vt(rans) > vi(ntrans)
286 Gereon Müller and Daniela Thomas The transitivity scale in (8-d) is a new scale that we postulate; it will play an important role in deriving three-way systems in a syntactic approach that only envisages two cases. The transitivity scale presupposes that transitive and intransitive v can be distinguished, in both ergative and accusative languages. This is unproblematic under the present syntactic analysis, where v is uniformly the inactive head in intransitive contexts. In addition to these scales (which can in principle have arbitrarily many members), a basic binary scale is needed in the theory of harmonic alignment. We assume that this is the DP case scale in (9), which simply states that DPs with the marked value for the feature [±gov] (which captures DPs bearing ergative/accusative) are more prominent than DPs with the unmarked value (which captures DPs with nominative/absolutive). This scale replaces the grammatical function scale in Aissen (1999b; 2003) and Keine & Müller (2011; 2015).6 (9)
DP case scale: DP[+gov] > DP[−gov]
Harmonic alignment (Prince & Smolensky (2004)) then takes a basic scale X > Y and another scale a > b > … > z, and produces a pair of complex markedness scales X/a ≻ X/b ≻ ... ≻ X/z and Y/z ≻ … ≻ Y/b ≻ Y/a; these scale are then transferred to fixed hierarchies of constraints with reversed order: *X/z ≫… ≫ *X/b ≫ *X/a; *Y/a ≫ *Y/b ≫ … ≫ *Y/z. Harmonic alignment of the binary DP case scale with the transitivity scale yields the two constraint hierarchies with invariant internal order in (10). (10) a. *DP[+gov]/vi ≫ *DP[+gov]/vt b. *DP[−gov]/vt ≫ *DP[−gov]/vi As it stands, constraints like *DP[−gov]/vt and *DP[−gov]/vi in (10-b) indiscriminately block configurations where a nominative/absolutive DP co-occurs with vt or vi in a clause. This is not yet adequate because it is not the DPs themselves that need to be deleted in certain (prototypical) contexts, but rather their case features. For this reason, following Aissen (1999b; 2003), a faithfulness constraint Max(case) demanding case feature preservation is locally conjoined with the members of the hierarchies derived by harmonic alignment, which is then counteracted by a general markedness constraint demanding case feature deletion.7 Local conjunction is a mechanism introduced by Smolensky (1995) (also see Legendre et al. (1998), Smolensky (2006)). 6
This then qualifies as a second, though minor, difference, in addition to the introduction of the transitivity scale. The underlying reason why the grammatical function scale (consisting of “subject,” i.e., Specv, and “object,” i.e., CompV) that Keine & Müller (2011; 2015) adopt from Aissen (1999b; 2003) needs to be replaced is that the notion of subject becomes unclear when ergative systems and intransitive contexts are taken into consideration (which the analysis in Keine & Müller (2011; 2015) is not concerned with). 7 Strictly speaking, the role of the faithfulness constraint Max(case) is played by the markedness constraint *Øcase in Aissen’s original approach, where case is not yet present in an input.
Three-way systems do not exist 287 Under local conjunction, two constraints A, B are combined to form a new constraint A&B which inherently outranks both A and B. A&B is violated if both conjoined constraints are violated (in a certain local domain). Importantly, local conjunction of a constraint A with members of a fixed constraint hierarchy B1 ≫ B2 ≫ … ≫ Bn derived by harmonic alignment preserves order. For the case feature [–gov], the two additional constraints that play a role are given in (11) (analogous constraints exist for [+gov]; see below). (11)
a. Max(case): Preserve case features. b. *[–gov]: Avoid the feature [–gov].
Max(case) is a faithfulness constraint that demands preservation of a case feature in a syntactic structure in the post-syntactic morphological component; this constraint can be conjoined with a constraint hierarchy derived from harmonic alignment. On the other hand, *[–gov] is a markedness constraint that forces deletion of [–gov] before vocabulary insertion—i.e. it brings about impoverishment; this constraint cannot be conjoined with a constraint hierarchy (see Aissen (1999b; 2003), Keine & Müller (2011; 2015)). The result of local conjunction of Max(case) with the fixed constraint hierarchy in (10-b) is given in (12). (12)
*DP[−gov]/vt & Max(case) ≫ *DP[−gov]/vi & Max(case)
The constraint *DP[−gov]/vt & Max(case) is violated by a post-syntactic (pre-vocabulary insertion) representation if there is a nominative/absolutive DP in a transitive clause that has its [–gov] feature deleted; similarly, *DP[−gov]/vi & Max(case) is violated if a nominative/absolutive DP in an intransitive clause gets its [–gov] feature deleted.8 If both these constraints outrank *[–gov], all syntactic [–gov] case features will be preserved 8
A technical question arises regarding the dual use of case features in a constraint like *DP[−gov]/ vt & Max(case): It must be ensured that a case feature like [–gov] that is deleted (thereby violating MAX(case)) can still be accessed so as to determine the violation (i.e., [–gov] is needed to characterize the class of DPs that are subject to the constraint). In principle, there would seem to be two possibilities. First, one could distinguish between deletion and erasure, as it is suggested in a structurally similar context in Chomsky (1995): On this view, deleted material would be inaccessible for morphological realization, but still accessible for constraint evaluation. Second, one can postulate that constraints like *DP[−gov]/vt & Max(case) are not only output-sensitive, but also input- sensitive (see Trommer (2006)). Thus, [–gov] in “*DP[−gov]/vt” refers to the input (i.e., the syntactic representation where feature deletion is not yet an issue), whereas [–gov] in “Max(case)” refers to the output (i.e., the post-syntactic representation in which feature deletion may or may not have applied). In view of the fact that the relevant constraints arise as a result of combining two separate constraints by local conjunction, and given that we are dealing with syntax/morphology interface optimization procedures, this second solution strikes us as vastly superior, and we will adopt it in the remainder of this chapter.
288 Gereon Müller and Daniela Thomas post-syntactically, and the regular exponents for nominative/absolutive will show up as a result of vocabulary insertion. If *[–gov] outranks both these constraints, all syntactic [–gov] case features will be deleted post-syntactically, and zero marking will often arise—or at least a “retreat to the general case” (see Halle & Marantz (1993)), depending on whether or not there are other, more underspecified non-zero exponents that can realize what is left (e.g. [–obl]). The most interesting situation is where *[–gov] is interspersed between *DP[−gov]/vt & Max(case) and *DP[−gov]/vi & Max(case) in a language, as in the ranking in (13). (Recall that the order of the two relativized faithfulness constraints themselves is fixed once and for all, as a consequence of harmonic alignment.) (13) *DP[−gov]/vt & Max(case) ≫ *[–gov] ≫ *DP[−gov]/vi & Max(case) Now [–gov] will be preserved post-syntactically in transitive contexts but deleted in intransitive contexts. Subsequent vocabulary insertion can then lead to a [–gov]-marked exponent as a case marker for DP in transitive contexts, but given that vocabulary insertion obeys the Subset Principle (Halle & Marantz (1993), Halle (1997)), it will have to resort to an underspecified (typically zero) exponent not bearing [–gov] in intransitive contexts. We would like to suggest that this represents one basic situation with apparent three-way systems:9 What at first sight looks like a separate occurrence of an accusative and an absolutive marker emerges as an allomorphic realization of an absolutive marker; there is no accusative present at any point in the derivation. At this point, one may think that a viable alternative to feature deletion via optimization based on harmonic alignment and local conjunction might be to postulate an appropriate impoverishment rule like the one in (14) (with the contextual information interpreted loosely, not necessarily requiring either adjacency or the linearization indicated here). (14) [–gov] → Ø / DP__[vi] However, whereas (14) simply stipulates the context in which deletion takes place, (13) derives this context. In addition, (13) (again in contrast to (14)) predicts that there can be no language where deletion of [–gov] takes place in transitive but not in intransitive contexts; and we take this to be a correct generalization. Another difference that strikes us as even more important will become clear below: As noted above, three-way systems typically also involve (other) scale effects; so it remains to be shown how harmonic alignment and local conjunction with the other scales can be brought into the picture. It will turn out that the optimization approach captures these multidimensional scale effects in a fairly straightforward way whereas a standard, rule-based impoverishment approach will face what look like insurmountable obstacles because the deletion contexts do not form natural classes.
9
The other possible situation is completely analogous, with [–gov] replaced by [+gov]; see below.
Three-way systems do not exist 289 Still, before addressing this issue by carrying out some case studies, a general question needs to be clarified that concerns locality. In order to evaluate a constraint like *DP[−gov]/vt & Max(case) or *DP[−gov]/vi & Max(case), both the properties of the DP (either DPext or DPint) and the properties of v must be taken into account. This suggests that the local domain for constraint evaluation at the interface might be the phase (see Chomsky (2001)), with feature deletion and vocabulary insertion applying cyclically.10 Against this background, let us now turn to some case studies. We will consider putative three-way systems in Kham, Djapu, Nez Perce, Upriver Halkomelem, and Dyirbal, and we will show how they can all be given simple accounts in terms of syntax/morphology interface optimizations via harmonic alignment and local conjunction.
12.3 Case Studies 12.3.1 Kham 12.3.1.1 Data The Tibeto-Burman language Kham has been argued to rely on a three-way system of argument encoding by case. The distribution of the case markers is shown in (15) (cf. Watters (2002, 66f.)). (15)
Distribution of case markers 1st
2nd 3rd, definite
3rd, indefinite
DPext-Vt
-∅
-∅
-e/-ye
-e/-ye
DPext/int-Vi
-∅
-∅
-∅
-∅
DPint-Vt
-lai -lai
-lai
-∅
This can be taken to mean that there is a nominative/absolutive case for sole DPext/int- Vi arguments, for prototypical (local person) DPext-Vt arguments, and for prototypical (3rd person indefinite) DPint-Vt arguments; an ergative case for marked 3rd person DPext-Vt arguments; and finally, an accusative case for marked (local person or 3rd person definite) DPint-Vt arguments. Clearly, a system of rules for syntactic case assignment designed to capture this distribution would necessarily be much more complex than is standardly assumed for structural case. In contrast, as indicated in (15), we argue that Kham basically exhibits a standard ergative system in the syntax, with -e/-ye as the 10 Postulating cyclic phase-based spell-out will also ensure that *DP [−gov]/vt & Max(case) cannot be
violated by deletion of some other case feature (e.g. of *DP[+gov]) in the presence of DP[−gov]/vt in the same phase: Spell-out of the complement XP of a phase head Y can only delete features in XP (but is sensitive to the properties of Y); and subsequent spell-out of the phase edge (as part of the next higher phase complement) cannot access previously spelled-out material again.
290 Gereon Müller and Daniela Thomas canonical ergative marker and -lai as the canonical absolutive marker. On this view, the simple person-based split in ergative contexts, and the more complex transitivity-/ definiteness-based split in absolutive contexts, are instances of allomorphic variation reducible to scale-driven optimization.11 Let us address the two cases in turn, beginning with the more complex situation with absolutive realization.
12.3.1.2 Absolutive Marking In addition to the binary DP case scale as the basic scale (see (9)), (15) illustrates that two scales are relevant, namely, the transitivity scale (see (8-d)) and the definiteness scale (see (8-c)). Harmonic alignment of the case scale with the transitivity scale and the definiteness scale yields the two constraint hierarchies for absolutive ([–gov]) DPs in (16). (16) a. *DP[−gov]/vt ≫ *DP[−gov]/vi b. *DP[−gov]/Pro ≫ *DP[−gov]/PN ≫ *DP[−gov]/Def ≫ *DP[−gov]/Spec ≫ *DP[−gov]/NSpec Next, the two hierarchies with fixed internal rankings thus derived are locally conjoined with one another, giving rise to two-dimensional local conjunction (see Aissen (2003)). Here, each constraint of one hierarchy is locally conjoined with each constraint of the other hierarchy, preserving original orders, as before. The new hierarchies that result are given in (17-ab).12 (17) a. *DP[−gov]/Pro/vt ≫ *DP[−gov]/PN/vt ≫ *DP[−gov]/Def/vt ≫ *DP[−gov]/Spec/vt ≫ *DP[−gov]/NSpec/vt b. *DP[−gov]/Pro/vi ≫ *DP[−gov]/PN/vi ≫ *DP[−gov]/Def/vi ≫ *DP[−gov]/Spec/vi ≫ *DP[−gov]/NSpec/vi c. *DP[−gov]/Pro/vt ≫ *DP[−gov]/Pro/vi d. *DP[−gov]/PN/vt ≫ *DP[−gov]/PN/vi e. *DP[−gov]/Def/vt ≫ *DP[−gov]/Def/vi f. *DP[−gov]/Spec/vt ≫ *DP[−gov]/Spec/vi g. *DP[−gov]/NSpec/vt ≫ *DP[−gov]/NSpec/vi Finally, the hierarchies in (17) are locally conjoined with Max(case), again preserving original orders. Thus, we end up with constraints like *DP[−gov]/Pro/vt & Max(case); this constraint is violated if there is an absolutive DP that is a pronoun, and the absolutive DP shows up with a transitive verb, and its [–gov] case feature is deleted—i.e. if the case feature of an absolutive pronoun is deleted in transitive contexts. 11
This implies that the absolutive can indeed be a non-zero case throughout in the languages that we consider here, notwithstanding its cross-linguistic tendency to be less marked segmentally. See Handschuh (2014). 12 Following Aissen (2003), we adopt a simplified notational variant where, e.g. “*DP [−gov]/Pro/vt” stands for “*DP[−gov]/Pro & DP[−gov]/vt” (where linear order of the two conjuncts is irrelevant).
Three-way systems do not exist 291 As a consequence, a two-dimensional system of argument encoding arises where some constraint pairs exhibit a fixed ranking, and others do not (such that languages simply can choose how they rank the constraints with respect to one another). Following Aissen’s (2003) conventions, fixed and variable rankings among the constraints generated by successive harmonic alignment and local conjunction with Max(case) can be represented as in (18). In this graph, constraints that stand in a domination relation invariantly have a fixed ranking (as a consequence of the mechanics of harmonic alignment and local conjunction), whereas constraints that do not stand in a domination relation are freely ordered with respect to each other. (18)
Absolutive allomorphy in Kham *DP[−gov]/Pro/vt I: /lai/
& Max(case)
*DP[−gov]/PN/vt
*DP[−gov]/Pro/vi
& Max(case)
& Max(case)
*DP[−gov]/Def/vt
*DP[−gov]/PN/vi
& Max(case)
& Max(case)
*DP[−gov]/Spec/vt
*DP[−gov]/Def/vi
& Max(case)
II: / /
& Max(case)
*DP[−gov]/NSpec/vt
*DP[−gov]/Spec/vi
& Max(case)
& Max(case)
*DP[−gov]/NSpec/vi & Max(case)
All the constraints in (18) demand case feature preservation. At this point, the ranking of the conflicting constraint demanding case feature deletion becomes relevant: *[–gov] leads to zero-marking for DPs with the feature combinations identified by the constraints that are ranked below it. In Kham, this constraint must be ranked above *DP[−gov]/Spec/vt & Max(case) and *DP[−gov]/Pro/vi & Max(case), and below *DP[−gov]/ Def/vt & Max(case) and *DP[−gov]/Pro/vt & Max(case), thereby separating the system in (18) into two discrete areas (given transitivity of ranking relations and the fixed rankings
292 Gereon Müller and Daniela Thomas established under harmonic alignment and local conjunction), which are here referred to as I and II. The absolutive case feature [–gov] is preserved in area I and removed in area II, which leads to the fully specified exponent /lai/in I configurations and to the elsewhere exponent /Ø/in II configurations. We can now ask whether this pattern could also be captured in a similar way by adopting a standard impoverishment rule as the source of case feature deletion. It turns out that this is not the case: One would have to postulate two separate impoverishment rules, as in (19), since the contexts in which [–gov] deletion takes place (intransitive clause and indefinite interpretation of DP) cannot be referred to as a natural class. Furthermore, (19) would give rise to redundancies with indefinite (specific or non-specific) DPs in intransitive contexts. (19) a. [–gov] → ∅ / DP__[vi] b. [–gov] → ∅ / DP[indef]__ Thus, in contrast to an approach in terms of genuine impoverishment rules, an optimality- theoretic approach employing scale-driven deletion makes it possible to refer to the diverse contexts where case feature deletion takes place as a natural class (defined by the relative ranking of the constraint demanding case feature deletion with respect to the ranking of the constraints demanding case-feature preservation in the various contexts).
12.3.1.3 Ergative Marking Turning next to allomorphy in ergative realization, the account is simple. It does not involve two-dimensional argument encoding because only the person scale is relevant for harmonic alignment with the binary basic case scale, of which the relevant member now is DP[+gov]. Harmonic alignment of the case scale and the person scale plus local conjunction with Max(case) yields the invariant hierarchy in (20). (20) *DP[+gov]/3 & Max(case) ≫ *DP[+gov]/loc & Max(case) Interleaving of *[+gov] between the two constraints of this hierarchy then produces zero exponence in cases where there is a prototypical external argument DP; see (21). (21) Ergative allomorphy in Kham *DP[+gov]/3 & Max(case)
I: /(y)e/
*[+gov] *DP[+gov]/loc & Max(case)
II: /Ø/
Three-way systems do not exist 293 Thus, overt ergative marking is blocked with transitive 1st or 2nd person subjects (because [+gov] is deleted here, and a retreat to the general elsewhere marker must take place), but available with 3rd person subjects (because [+gov] is preserved here, so that the vocabulary item /(y)e/that is characterized by the feature [+gov] can be inserted without violating the Subset Principle). To conclude, Kham has an ergative encoding system where both ergative and absolutive can be non-zero cases, and both ergative and absolutive can be zero-marked, as a consequence of case feature deletion in prototypical configurations.
12.3.2 Djapu 12.3.2.1 Data The Pama-Nyungan language Djapu has also been analyzed in terms of a three-way system comprising nominative/absolutive, ergative, and accusative. The distribution of case exponents is shown in (22) (see Morphy (1983, 34–35)). (22) Distribution of case markers Pron
+hu
–hu
DPext-Vt
-∅
-DHu
-DHu
DPext/int-Vi
-∅
-∅
-∅
DPint-Vt
-NHA
-NHA
-∅
Again, we assume that Djapu underlyingly exhibits an ergative system, with -DHu as the ergative marker and -NHA as the absolutive marker. Thus, overt absolutive marking is suspended in intransitive contexts and for non-human objects; overt ergative marking does not show up on pronominal transitive subjects.
12.3.2.2 Absolutive Marking The relevant scales determining the distribution of morphological case exponents are the transitivity scale and the animacy scale. Both are harmonically aligned with the basic case scale, yielding (23-a) and (23-b) for absolutive contexts. (23)
a. *DP[−gov]/vt ≫ *DP[−gov]/vi b. *DP[−gov]/Hum ≫ *DP[−gov]/Anim ≫ *DP[−gov]/Inan
Local conjunction among the members of these constraint hierarchies with fixed internal order produces the strict rankings in (24). (24) a. *DP[−gov]/Hum/vt ≫ *DP[−gov]/Anim/vt ≫ *DP[−gov]/Inan/vt b. *DP[−gov]/Hum/vi ≫ *DP[−gov]/Anim/vi ≫ *DP[−gov]/Inan/vi c. *DP[−gov]/Hum/vt ≫ *DP[−gov]/Hum/vi
294 Gereon Müller and Daniela Thomas d. *DP[−gov]/Anim/vt ≫ *DP[−gov]/Anim/vi e. *DP[−gov]/Inan/vt ≫ *DP[−gov]/Inan/vi Finally, order-preserving local conjunction with Max(case) gives rise to the two- dimensional system in (25). (25) Absolutive allomorphy in Djapu I: /NHA/
*DP[−gov]/Hum/vt & Max(case)
*DP[−gov]/Anim/vt
*DP[−gov]/Hum/vi
& Max(case) *DP[−gov]/Inan/vt
& Max(case) *DP[−gov]/Anim/vi
& Max(case)
& Max(case) *DP[−gov]/Inan/vi
II: /Ø/
& Max(case) Since all absolutive arguments except for human objects remain without an overt marker, the conflicting markedness constraint *[–gov] demanding case feature deletion must be located below *DP[−gov]/Hum/vt & Max(case), and directly above both *DP[−gov]/Anim/vt & Max(case) on the one hand (in transitive contexts), and *DP[−gov]/Hum/vi & Max(case) on the other (in intransitive contexts). Given that the absolutive marker /NHA/is specified for the feature [–gov], it can only show up in transitive contexts with a human referent interpretation of the internal argument DP. As before, an impoverishment account would be inferior because the contexts in which case feature deletion takes place do not form a natural class definable in terms of shared morpho-syntactic features.
12.3.2.3 Ergative Marking Again, allomorphic variation in the ergative system is somewhat simpler. The relevant scales are the case scale and the definiteness scale; see (26). (26) *DP[+gov]/Nspec ≫ *DP[+gov]/Spec ≫ *DP[+gov]/Def ≫ *DP[+gov]/PN ≫ *DP[+gov]/Pron Local conjunction with Max(case) and interleaving of *[+gov] between *DP[+gov]/PN & Max(case) and *DP[+gov]/Pron & Max(case) yields a distribution of the overt ergative exponent /DHu/that involves all DPext arguments of transitive contexts except for pronouns. This is shown in (27).
Three-way systems do not exist 295 (27)
Ergative allomorphy in Djapu: *DP[+gov]/Nspec & Max(case) *DP[+gov]/Spec & Max(case) *DP[+gov]/Def & Max(case)
I: /DHu/
*DP[+gov]/PN & Max(case) *[+gov] *DP[+gov]/Pron & Max(case)
II: /Ø/
12.3.3 Nez Perce 12.3.3.1 Data In Nez Perce (Penutian), sole arguments of intransitive verbs are unmarked, as are local person external arguments of transitive verbs; 3rd person external arguments of transitive verbs receive a special marker, and the same goes for all internal arguments of transitive verbs; see (28) (Rude (1985, 82f.)). As noted above, this is usually taken to imply a three-way system based on a general accusative, a general nominative/absolutive, and a scale-dependent ergative. (28) Distribution of case markers 1/2 pronouns
3 pronouns
proper names
common nouns
DPext-Vt
-∅
-(n(i))m
-(n(i))m
-(n(i))m
DPext/int-Vi
-∅
-∅
-∅
-∅
DPint-Vt
-ne
-ne
-ne
-ne
As before, we suggest that this pattern be reanalyzed as a canonical ergative system, with scale-driven allomorphy affecting both ergative and absolutive contexts.
12.3.3.2 Absolutive Marking This time, the absolutive alternation pattern is very simple: It results from a simple local conjunction of the transitivity and case scales, which yields the fixed order in (29).
296 Gereon Müller and Daniela Thomas (29) *DP[−gov]/vt & Max(case) ≫ *DP[−gov]/vi & Max(case) The conflicting markedness constraint demanding deletion of [–gov] is ranked between these two constraints in Nez Perce, which produces absence of [–gov] in the morphological component (hence, a zero exponent) in intransitive contexts and retention of [–gov] (hence, a non-zero exponent /ne/, which is specified for this feature) with absolutive DPs in transitive contexts. (30) Absolutive allomorphy in Nez Perce: *DP[−gov]/vt & Max(case)
I: /ne/
*[–gov]
*DP[−gov]/vi & Max(case)
II: /Ø/
12.3.3.3 Ergative Marking The scales that play a role in accounting for ergative allomorphy are the person and case scales. Local conjunction of the constraints derived by harmonic alignment with Max(case) gives rise to a strict hierarchy: (31) *DP[+gov]/3 & Max(case) ≫ *DP[+gov]/loc & Max(case) Next, interleaving of *[+gov] leads to zero exponence in local contexts (given that insertion of the ergative marker /nim/depends on the presence of [+gov]): (32) Ergative allomorphy in Nez Perce: *DP[+gov]/3 & Max(case)
I: /nim/
*[+gov]
*DP[+gov]/loc & Max(case)
II: /Ø/
Three-way systems do not exist 297
12.3.4 Upriver Halkomelem 12.3.4.1 Data The evidence from Upriver Halkomelem (Salish) differs from what we have seen so far in two respects. First, argument encoding proceeds by agreement rather than by case-marking; this is unproblematic against the background of the approach in section 12.1: If assignment of case by a functional head (v, T) to a DP is viewed as an instance of Agree (Chomsky (2001)), i.e. as a probe-goal relation, then the relevant case information will show up on both the head (v, T) and the DP, and can thus be morphologically realized either on the former or on the latter. Second, Upriver Halkomelem exhibits an accusative rather than an ergative basic pattern; in the approach adopted here, this implies that v structurally encodes DPint in transitive contexts, and T encodes all other core arguments in transitive and intransitive contexts. The agreement paradigm is given in (33) (see Galloway (1977, 141)). (33)
Distribution of cross-reference markers DPext-Vt
1SG
1PL
2SG
2PL
3SG/PL
-cəl
-cət
-cəxw
-cεp
-əs
-cεp
-∅
-álə
-∅/-ə xw
DPext/int-Vi
-cəl
-cət
-cəxw
DPint-Vt
-áyy
-álxw
-ámə
As illustrated in (33), the initial evidence for postulating a three-way system in Upriver Halkomelem is confined to zero exponence in 3rd-person intransitive configurations. Under present assumptions, this split is treated as an instance of non-zero/zero allomorphy with nominative exponents.13
12.3.4.2 Nominative Marking The distribution of nominative exponents (on T, as part of a complex verbal category) reveals harmonic alignment of both the transitivity scale and the person scale with the case scale, yielding the two-dimensional system in (34). (34) a. b. c. d.
*DP[−gov]/loc/vt ≫ *DP[−gov]/3/vt *DP[−gov]/loc/vi ≫ *DP[−gov]/3/vi *DP[−gov]/loc/vt ≫ *DP[−gov]/loc/vi *DP[−gov]/3/vt ≫ *DP[−gov]/3/vi
After order-preserving local conjunction with Max(case) (which applies to case features on both DP and T/v), a system of partially free, and partially fixed, rankings is 13 There is also an allomorphy with 3rd-person DP arguments in transitive environments. This is arguably int not of direct interest in the present context because it depends on the type of transitivity marker on the verb.
298 Gereon Müller and Daniela Thomas derived. As for the interleaving of *[–gov], it is clear that this constraint must outrank *DP[−gov]/3/vi & Max(case) since zero exponence occurs in this context, and must in turn be dominated by *DP[−gov]/loc/vi & Max(case) since the non-zero nominative exponent shows up here; in contrast, in transitive contexts all argument types are overtly encoded, so *[–gov] must be outranked by all pertinent faithfulness constraints. All of this is shown schematically in (35), with the areas marked I and II capturing the two domains.14 (35) Nominative allomorphy in Upriver Halkomelem *DP[–gov]/loc/vt & Max(case)
*DP[–gov]/3/vt
*DP[–gov]/loc/vi
& Max(case)
& Max(case)
*DP[–gov]/3/vi & Max(case)
II: /Ø/
12.3.4.3 Accusative Marking Accusative marking is consistent and not subject to scale effects; the constraint *[+gov] demanding case feature deletion on v is ranked below all faithfulness constraints resulting from harmonic alignment and local conjunction.
12.3.5 Dyirbal 12.3.5.1 Data Dyirbal (Pama-Nyungan) is also typically claimed to instantiate a three-way system, based on the distribution of argument-encoding exponents on DPs in (36) (see Dixon (1972; 1994)). We would like to suggest that Dyirbal case marking is best analyzed in terms of scale-dependent allomorphy on the basis of a simple ergative system.
14 Note incidentally that in this particular case, a standard impoverishment rule would also suffice since it is just one specific context in which case feature deletion needs to be brought about. Still, even here, the other two arguments against classical impoverishment given above remain valid.
Three-way systems do not exist 299 (36)
Distribution of Case markers 1st/2nd pronouns 3rd pronouns proper names common nouns DPext-Vt
-∅
-ŋgu
-ŋgu
ŋgu
DPext/int-Vi -∅
-∅
-∅
-∅
DPint-Vt
-∅
-∅
-∅
-na
12.3.5.2 Ergative Marking The relevant scales for ergative marking are the case scale and the person scale. Harmonic alignment plus local conjunction with Max(case) produces the constraints in (37). (37)
*DP[+gov]/3 & Max(case) ≫ *DP[+gov]/loc & Max(case)
By interleaving *[gov] between these two constraints and assuming /ŋgu/to be a vocabulary item specified as [+gov], allomorphic realization of the ergative is correctly predicted; see (38). (38)
Ergative allomorphy in Dyirbal: *DP[+gov]/3 & Max(case) *[+gov]
*DP[+gov]/loc & Max(case)
II: /Ø/
Note that this system is completely identical to the system of ergative allomorphy in Nez Perce.
12.3.5.3 Absolutive Marking Turning to variation in the realization of absolutive case next, the first thing to note is that the same scales are relevant as they are with nominative allomorphy in Upriver Halkomelem: The basic case scale is harmonically aligned with both the transitivity scale and the person scale, and subsequently, local conjunction with Max(case) applies to the two constraint hierarchies thus generated, yielding the multidimensional system in (39). Assuming that the exponent /na/is specified for [–gov], the markedness constraint *[–gov] must determine the optimal output (forming the input for morphological realization) in all intransitive contexts, and in 3rd person transitive contexts; i.e. *[+gov] is only dominated by *DP[−gov]/loc & Max(case). The resulting system is shown in (39). The area signaled by I shows non-zero absolutive exponence; the area marked by II has zero exponence.
300 Gereon Müller and Daniela Thomas (39) Absolutive allomorphy in Dyirbal *DP[−gov]/loc/vt I: /na/
& Max(case)
*DP[−gov]/3/vt
*DP[−gov]/loc/vi
& Max(case)
& Max(case)
II: /Ø/
*DP[−gov]/3/vi & Max(case) As in several of the cases addressed before, no single impoverishment rule could capture this distribution of zero and non-zero absolutive markers since the contexts in which deletion takes place do not form a natural class. Note also that even though Upriver Halkomelem and Dyirbal differ radically with respect to the basic argument encoding system employed in the language (accusative vs. ergative, agreement vs. case), deletion of the unmarked case (nominative/absolutive) involves (a) the same case feature ([–gov]), (b) identical derived faithfulness constraints with identical ranking restrictions ((39) and (35) are basically the same), and an identical markedness constraint demanding case feature deletion (*[–gov]). The only difference concerns the cut-off point between preservation and deletion; to highlight this difference, we have repeated the partition employed by Upriver Halkomelem in (39), in the form of a grey dotted line.
12.3.6 Interim Conclusion This concludes the first, major part of the analysis: We have shown that it is straightforwardly possible to reanalyze three-way case systems as standard two-way (ergative or accusative) case systems, with all the marker variation derived by scale-driven optimization operations at the syntax/morphology interface. Given that case marker allomorphy based on person, animacy, and definiteness is independently known to occur, and given that these effects also show up in all the languages that we have addressed here, the only additional assumption that is needed to capture all effects in a maximally simple way—i.e. the postulation of a transitivity scale on a par with other Hale/Silverstein scales—strikes us as well motivated.15 Furthermore, the fact that putative three-way systems are typically accompanied by Hale/Silverstein scale effects, and that the fact that 15
Also, the crosslinguistic rarity of three-way systems (see n. 4) can now be explained under the assumption that non-homogeneous post-syntactic case-feature deletion is inherently marked.
Three-way systems do not exist 301 these effects, though subject to implicational generalizations, are not uniform across languages, together pose what we take to be an enormous challenge for a syntactic approach recognizing three different cases; ambitious recent attempts notwithstanding (see in particular Deal (2014)) we would like to contend that it is hardly possible to come up with a comprehensive syntactic approach to the phenomenon that qualifies as both simple and elegant, and that covers both an individual language’s pattern in detail, and captures cross-linguistic variation as well. However, that said, the question of whether a morphological or a syntactic approach to apparent three-way systems is correct is also an empirical one; we address this issue in the final section of this chapter.
12.4 Syntactic Evidence The present morphology-based approach to three-way systems differs from syntactic approaches in that it reanalyzes what at first sight looks like an accusative DP (in Kham, Djapu, Nez Perce, and Dyirbal) as a non-zero-encoded absolutive DP, and what looks like an ergative DP (in Upriver Halkomelem) as a non-zero-encoded nominative DP. The prediction thus is that there might be independent evidence for the status of the pertinent DPs as absolutive/nominative (i.e. [–gov,–obl], assigned by T). More generally, we expect to find evidence for a morphological approach in terms of case allomorphy based on identical syntactic behavior of the non-zero-marked and zero-marked DPs; in the same way, different syntactic behavior might provide counter-evidence against the proposal. Let us address some relevant phenomena.
12.4.1 Topic Chaining in Dyirbal A first piece of evidence comes from the topic chaining construction in Dyirbal, an instance of what has sometimes been called syntactic ergativity (Dixon (1972; 1994)). As shown in (40), if the coordinative process of topic chaining combines a transitive clause (with DPext bearing ergative case and DPint bearing absolutive case) and an intransitive clause where the sole DPext/int argument is not overtly realized (here encoded by an empty pronoun pro, for expository purposes), then this latter DP, which must bear absolutive case, must be coreferent with the absolutive argument of the transitive clause, not with the ergative argument; see (40). (40) [CP1 ŋuma yabu-ŋgu bura-n] father- abs mother-erg see-nonfut ‘Mother saw father and he/*she returned.’
[CP2 pro pro-abs
banaga-nyu] return-nonfut
Thus, there is a case-matching requirement active in Dyirbal topic-chaining constructions, and we can test the predictions made by the two approaches. On the one hand, if
302 Gereon Müller and Daniela Thomas local person DPint arguments marked by /na/bear accusative case (as assumed under the standard three-way approach), they should not be able to corefer with a non-overt sole DP argument of an intransitive clause (which bears absolutive case) in a topic chaining construction; on the other hand, if local person DPint arguments marked by /na/bear absolutive case (as assumed under the present analysis), they should be able to corefer with a non-overt DP argument of an intransitive clause. The data in (41) (from Dixon (1972; 1994)) show that the latter prediction is in fact the correct one; as noted by Morgenroth & Salzmann (2013), this provides strong evidence against a separate accusative case, and hence against a three-way system, in Dyirbal. (41) a. [CP1 ŋana-Ø banaga-nyu] [CP2 nyurra-Ø pro we-abs return-nonfut you all-erg pro-abs ‘We returned and you all saw us.’
bura-n] see-nonfut
b. [CP1 nyurra-Ø ŋana-na bura-n] [CP2 pro banaga-nyu] you all-erg we-abs see-nonfut pro-abs return-nonfut ‘You all saw us and we returned.’ Note that (41-b) shows that the three-way approach not only fails to derive the possibility of coreference of the absolutive argument of the intransitive clause with the DPint argument of the transitive clause (because the latter shows non-zero exponence); assuming that non-zero exponence on DPext in the transitive clause indicates absolutive case (but see footnote 16 below), it also wrongly predicts coreference of the absolutive argument of the intransitive clause with the DPext argument of the transitive clause to be possible (since the latter exhibits zero-exponence).
12.4.2 Modifiers and Relative Clauses in Dyirbal This latter consequence with respect to zero-marked DPext arguments in transitive contexts also arises with modifiers in Dyirbal. The example in (42) (see Dixon (1972, 133), Mel’čuk (1979, 54)) illustrates that modifiers and relative clauses of a local person DPext argument of a transitive verb bear ergative case despite the lack of ergative marking of the pronoun itself. (42) ŋaɖa wuygi-ŋgu balan I.nom old-erg ncII.there.abs ‘I, old, hit the woman.’
ɖugumbil woman.abs
balga-n hit-nfut
This clearly shows that ergative is indeed assigned to DPext in (42) (even though it is not overtly realized by an ergative exponent on the head), and passed on to other DP-internal items via concord (with subsequent morphological non-zero realization on D). In the present analysis, this state of affairs can be addressed straightforwardly, by invoking standard assumptions about the order of operations involved here (see Müller (2009), Keine (2010),
Three-way systems do not exist 303 Georgi (2014)). First, there is assignment of ergative ([+gov,–obl]) case to DPext (hence, D) by v; we assume this case feature to be transferred automatically to a potential NP selected by D (this is irrelevant in the case of pronouns, as in (42), but it would be relevant with non-pronominal 3rd person DPs). Second, there is another operation (‘concord’) that copies the feature from the nominal spine (D, N) to AP and CP modifiers. And third, scale- driven optimization (leading to deletion of [+gov]) targets DP, deleting the feature on D but leaving it intact on AP and CP modifiers. This counter-bleeding effect (deletion would bleed concord but comes too late to do so) is shown in (43). (43)
a. Case assignment
b. Concord vP
vP
D:[+gov]
v'
DPext
v'
DPext AP v
VP
... DPint ...
D:[+gov]
AP:[+gov]
v
VP
... DPint ...
c. Scale-driven deletion vP
v'
DPext D: –
AP:[+gov] v
Ø
VP
... DPint ...
In contrast, if the case feature assigned to local person DPext of transitive verbs in Dyirbal is absolutive rather than ergative, the source of the ergative exponents on DP-internal items in (42) must remain a mystery.16 16 To be sure, adopting a three-way approach involving ergative, accusative and nominative/ absolutive for the distribution of markers in (36) does not per se imply that the source of zero exponence with local person DPext-Vt is the same as the source of zero exponence with DPext/DPint-Vi (or with 3rd person DPint-Vt, for that matter); in other words, one could in principle assume that Dyirbal has three structural cases, but some instances of zero marking involve a morphologically unrealized ergative rather than syntactically assigned nominative/absolutive. If so, an approach along the lines just sketched would also be available in a three-way approach, but it would miss an obvious generalization: The non- zero/zero alternation with DPext-Vt is clearly governed by the same factor that also restricts the non- zero/zero alternation with DPint-Vt (i.e. the person scale).
304 Gereon Müller and Daniela Thomas
12.4.3 Complex DPs in Djapu Further independent evidence for treating an apparent coexistence of absolutive and accusative as allomorphic realization of a single case (the absolutive) comes from Djapu. The following example from Morphy (1983, 136) illustrates that a non-zero exponent can show up on a DPint-Vt (glossed here as acc, following the literature) whereas a modifier phrase agreeing with this DP exhibits zero exponence (glossed here as abs). (44) dhuwal ŋarra yothu-n this.abs 1sg.nom child-acc ‘I can hear a child crying.’
yukurra lie.unm
ŋ-ä-ma hear-unm
[ŋäthi-nyar] [cry-nmlz.abs]
Similarly, Legate (2008, 77) observes that a demonstrative will show zero exponence where a non-zero exponent occurs with a (human) DPint-Vt. Following Legate, we take this to indicate that the same case feature is involved on the noun and the modifier or demonstrative, and that the variable exponence is a purely morphological phenomenon, reducible to the availability or lack of compatible vocabulary items. However, our analysis differs from Legate’s (2008) account in one crucial respect: Legate postulates a three-way system for Djapu and considers the source of zero exponence with DPext/int in intransitive contexts to be different from the source of zero exponence with DPint in transitive contexts (absolutive vs. unmarked accusative), which gives rise to the conceptual problem mentioned in footnote 16.
12.4.4 Relative Clauses in Kham Like Dyirbal, Kham has non-zero ergative exponence on relative clauses associated with local person DPext-Vt arguments that do not exhibit any morphological case marker themselves; cf. (45) (cf. Watters (2002, 201, 205)). (45) geː ma-ba-o-ra-e ge-ma-dəi-ye we neg-go-nml-pl-erg 1pl-neg-receive-impfv ‘Those of us who didn’t go didn’t get any.’ Again, this asymmetry follows directly under the present approach, along the lines of the derivation in (43); but it raises severe problems under a three-way approach.
12.4.5 Modifiers in Nez Perce Recall from section 12.4.2. that cases where zero and non-zero exponence of case cooccur in a single DP follow under the morphological approach but pose a challenge
Three-way systems do not exist 305 for a syntactic three-way analysis. Conversely, Deal (2014) presents an argument based on the reverse configuration in support of a syntactic rather than morphological analysis. In Nez Perce, modifiers can (optionally) take ergative marking. From a morphological perspective, one might therefore expect this marking to occur on the modifier if it is combined with a local-person transitive subject which does not bear ergative marking. As shown in (46), this is not the case (glossing follows Deal (2014, 17)). (46) a. Yú’s-nim ’iceyéeye-nm, wéet’u minma’í ’itúu-ne pée-p-se-∅ poor-erg coyote-erg neg prt what-acc 3/3-eat-imperf-pres ‘Poor coyote isn’t eating anything.’ b. Yu’c /*Yú’s-nim (pro), wéet’u q’o minma’í ’itúu-ne poor.nom /*poor-erg pro.1sg neg prt prt what-acc ’ee-pí-se-∅ 3obj-eat-imperf-pres ‘Poor me isn’t eating anything.’ However, closer inspection reveals that (46-b) is in fact unproblematic under the approach adopted here. Recall from (43) that we expect marker preservation on modifiers if the order of operations is (i) case assignment (of [+gov] to DPext by v), (ii) concord (of D and AP/CP), and (iii) scale-driven deletion, an instance of optimization. Suppose now that the last two operations can also apply in reverse order, such that feature deletion can precede concord; and that this is the case in Nez Perce (but not in, say, Dyirbal).17 If so, the absence of a case exponent on the modifier in (46-b) is derived: [+gov] is deleted on D before concord with AP is effected (an instance of bleeding).
12.4.6 Coordination in Nez Perce Deal (2014, 20f.) gives a second argument in support of a three-way analysis of Nez Perce. Coordination in Nez Perce is not subject to any restrictions when it occurs with DPext/int-Vi arguments or with DPint-Vt arguments. In these contexts, all kinds of person combinations are allowed. (Case marking may appear on both coordinates or just the final one.) However, there are restrictions in the case of coordination of DPext-Vt arguments. Coordinations of two local person pronouns and of two 3rd person pronouns/nouns are unproblematic (cf., e.g., (47-a)), whereas the combination of local person and 3rd person turns out to be ungrammatical (see (47-b)):
17
See Keine (2010), Doliana (2013) on early, inner-syntactic impoverishment-by-optimization.
306 Gereon Müller and Daniela Thomas (47) a. Kátie(-nim) kaa Hárold-nim pée-’pewi-six-∅ Múna-ne. Katie(-erg) and Harold-erg 3/3-look.for-imperf.pl-pres Muna.acc ‘Katie and Harold are looking for Muna.’ b. *Ángel- nim kaa ’iin ’e- nées- tecukwe- cix- ∅ (pro). Angel-erg and 1sg.nom 3obj-o.pl-teach-imperf.pl-pres (pro.3.pl) ‘I and Angel are teaching them.’ This can be derived if some DPext-Vt arguments bear absolutive case in an otherwise ergative system (based on conflicting demands of case-assignment); but closer scrutiny reveals that the present analysis (where all coordinated DPs in (47) bear ergative case) also does not face any particular difficulties. Assuming that & is the head of a coordinate structure &P, ergative case assignment by v will first instantiate [+gov] on &. As argued in section 13.4.5, scale-driven deletion applies next in Nez Perce, before &-internal concord (which transfers [+gov] to the two DPs in coordinate structures). However, at this point, there is incompatible contextual information: For the second DP in (47-b), optimization applying to & would rely on a local person feature; for the first, it would rely on a 3rd person feature; and the outcome of optimization would be different (deletion vs. preservation of [+gov]). It is conceivable that languages in principle may have the option to give preference to one of the two outcomes of optimization in this kind of situation (giving rise, e.g., to first-vs. last conjunct agreement); however, in Nez Perce, it leads to a breakdown of the derivation. In contrast, in other coordination environments (as in (47-a), or with DPext/ int-Vi, or with DPint-Vt), the two DP contexts for & uniformly demand deletion or preservation of the case feature; so no problem of conflicting instructions will arise. More generally, we may venture the hypothesis that the available syntactic evidence either directly supports the hypothesis that there are no three-way systems, or can at least be addressed on the basis of a syntactic approach relying on only two structural cases in an insightful way. From an even more general perspective, if what precedes is on the right track, we can conclude that by addressing instances of variation in case exponence in the morphological component of the grammar, the syntactic component can be kept simple and elegant, with three-way systems emerging as an artefact.
Acknowledgments For helpful comments, we would like to thank Sebastian Bank, Aaron Doliana, Doreen Georgi, Peter Staroverov, Michael Yoshitaka Erlewine, Lisa Travis, and one anonymous reviewer.
Three-way systems do not exist 307
Abbreviations abs, absolutive; acc, accusative; Anim, animate; Def, definite; DPext, external Determiner Phrase; DPint, internal Determiner Phrase; erg, ergative; gov, governed; HU, human; Hum, human; imperf, imperfective; impfv, imperfective; Inanim, inanimate; intrans, intransitive; loc, local person; Max, Maximality (constraint); neg, negation; nml, nominalizer; nmlz, nominalizer; nom, nominative; Nspec, nonspecific; nonfut, non-future; obl, oblique; perf, perfect; pl, plural; pldo, plural direct object; PN, proper name; pres, present; Pro, pronoun; prt, particle; pst, past; Spec, indefinite specific; trans, transitive; unm, unmarked; Vi, intransitive verb; Vt, transitive verb.
Chapter 13
Antipas si v e Maria Polinsky
13.1 Introduction Valency alternations follow from the mapping of logical arguments of a predicate into different clausal constituents. In morphologically accusative languages, active-voice clauses typically have a nominative subject and accusative object (1). Morphologically ergative languages, by contrast, have ergative subjects and absolutive objects; this alignment can be expressed by case-marking and agreement (2) or by agreement alone (3): (1) Did sxopyv grandfather.nom.masc grab.pst.masc ‘Grandfather grabbed the turnip.’
ripk-u. turnip-acc
(2) Anguti-up annak taku-janga. man-erg woman.abs see-3sg.subj:3sg.obj.prs ‘The man sees the woman.’ (3) Max-Ø s-tx’aj ix compl-abs3 erg3-wash clf ‘The woman washed the clothes.’
ix woman
an clf
pichilej. clothes
Ukrainian
Labrador Inuit (Smith 1982a: 164) Q’anjob’al
In passive-voice clauses in both accusative and ergative languages, the logical subject remains unexpressed or appears as a by-phrase; the subject position is filled by either the logical object ((4a), (5), and (6)) or by an expletive subject (4b): (4)
a. Repk- a bu- l- a sxople- n- a (did- om). Ukrainian turnip.nom.fem be-pst-fem grab-pst.ptcp-fem grandfather-ins ‘The turnip was grabbed (by the grandfather).’
Antipassive 309 b. expl repk-u bu-l-o sxople-n-o (did-om). Ukrainian turnip-acc be-pst-neuter grab-pst.ptcp-neuter grandfather-ins ‘The turnip was grabbed (by the grandfather).’ (5)
Labrador Inuit
Annak (anguti-mut) taku-jau-juk. woman.abs man-dat see-pass-3subj.prs ‘The woman is seen (by the man).’
(Smith 1982a: 165)
(6) Max-Ø tx’aj-lay an pichilej (y-uj ix ix). Q’anjob’al compl-abs3 wash-pass clf clothes gen3-rel.noun clf woman ‘The clothes were washed (by the woman).’ Broadly speaking, the defining property of the passive construction is the “suppression” or “demotion” of the highest-ranking argument to a by-phrase or a null existential argument without altering the truth conditions of the sentence (Goodall 1993, 1999; Bruening 2013). The logical object may appear as the syntactic subject of the resulting construction, but, as (4b) shows, it need not. In the foundational work on voice structures in Relational Grammar, “demotion” was a theoretical primitive (see Perlmutter 1980; Blake 1990b for overviews). In this chapter, I use this term pre-theoretically to indicate that an expression has been lowered vis-à-vis the hierarchy of grammatical roles: (7)
subject > object > non-core argument > non-argument
While subject demotion is the unifying factor among passive constructions cross- linguistically, motivation for this demotion varies across languages and structures; differences among passives are often attributed to distinct features of the passive morpheme or voice head. In the antipassive construction, the logical object of a transitive verb is demoted: either suppressed or represented by an expression lower on the grammatical hierarchy (7). In (8), the verb appears in the intransitive (reflexive) form, and the logical object receives accusative case from the preposition za. This prepositional object fails various direct- object diagnostics; for example, it cannot bind depictives or float quantifiers. (8)
Did sxopyv-sja grandfather.nom.masc grab.pst.masc-refl ‘Grandfather grabbed at the turnip.’
za at
ripk-u. turnip-acc
Ukrainian
In (9), the intransitive counterpart of (2), the logical object appears in an oblique form. The case of the subject changes from ergative to absolutive, and the verb shows only subject agreement. (9)
Angutik anna-mik taku-juk. man.abs woman-ins see-3sg.subj.prs ‘The man sees a woman.’
Labrador Inuit (Smith 1982a: 164)
310 Maria Polinsky Likewise, in (10), the counterpart of (3), the verb loses its object agreement.1 (10)
Max-Ø tx’aj-w-i (an) compl-abs3 wash-ap-itr clf ‘The woman washed (the) clothes.’
pichilej clothes
ix clf
ix. woman
Q’anjob’al
The term antipassive was coined by Michael Silverstein in his description of Chinook: I have termed this -ki-form the ANTIPASSIVE construction, playing upon its inverse equivalence to a passive of accusative languages, because the sense is clearly equivalent to a transitive, though the form is intransitive, with the grammatical function of the remaining NP reversed (ergator becomes nonergator). (Silverstein 1972: 395)
At the time of Silverstein’s writing, it was widely believed that the absolutive DP of a transitive clause was the grammatical subject (“nonergator” in the Silverstein quote) in ergative languages. This analysis has since been rejected (cf. Anderson 1976, and much subsequent work), but the term “antipassive” has remained. Many papers have lamented the term at length; in what follows, I set aside terminological debate and focus on analysis. Note that the term “antipassive” has two related uses in the linguistics literature; throughout this chapter, “antipassive” will be understood as in (11a). (11)
a. Antipassive: a clause with a transitive predicate whose logical object is demoted to a non-core argument or non-argument b. Antipassive: the form of a two-place predicate that appears in the antipassive clause
The antipassive predicate is semantically transitive, but does not project a direct object; hence, it is morphosyntactically intransitive.2 In what follows, I will be using the term “antipassive object” to signify the logical object of a clause. This logical object appears as a non-core argument of the antipassive construction (8), (9) or is suppressed (12b). (12) a. Aid opa-n matses that.one dog-erg people.abs ‘That dog bites people.’
pe-e-k. bite-npst-ind
b. Aid opa pe-an-e-k. that.one dog.abs bite-ap-npst-ind ‘That dog bites (me/us).’
1
Matses
(Fleck 2006: 559)
Also note that, unlike (3), the classifier on the object in (10) is optional, further signaling a change in its grammatical status. 2 The valency of an antipassive verb can be increased by adding an applicative head; this head then licenses an object, but crucially not the same direct object as the one selected by the non-applicative form (for applicative antipassives in Chukchi, see Nedjalkov 1980; Kozinsky et al. 1988).
Antipassive 311 Antipassives have been documented in a number of languages, but not all antipassives meet the same operational criteria. Two criteria are typically used to establish the existence of an antipassive: (i) oblique marking on the object and absolutive (not ergative) marking on the subject; (ii) structural/discourse “foregrounding” of the agent (see section 13.3; cf. Foley and Van Valin 1984 on foregrounding and backgrounding antipassives; Cooreman 1988 on discourse-based and structural antipassives). These criteria, based on morphological exponence and discourse effects, are not sufficient to fully characterize the antipassive construction. Concerning criterion (i), antipassive morphology need not be consistent within a given language: a particular morphological exponent may be used to mark antipassive in one use, but not in others; conversely, more than one type of morphological exponence may be associated with the antipassive. Concerning criterion (ii), it is problematic to define a construction on the basis of discourse effects, although such effects may arise as a consequence or entailment of a particular construction; furthermore, the same interpretative or discourse effect may be associated with more than one construction. Thus, neither morphological nor discourse effects seem independently capable of defining the antipassive construction. The demotion or removal of the object argument, however, is definitional of the antipassive—just as the demotion or removal of the external argument is definitional of the passive. While detailed descriptions of antipassives have been offered for individual languages (I reference some in the chapter; without such resources, this chapter could not have been written), the data are often insufficient to draw far-reaching conclusions. In such instances, I simply offer observations and make suggestions for further study. Consequentially, this chapter may be read as both an overview and promissory note; it is my hope that future analytical work on antipassives will follow. Section 13.2 describes typical morphological hallmarks of the antipassive; section 13.3 addresses its main interpretative characteristics. Section 13.4 summarizes the main existing analyses of the construction. Section 13.5 discusses syntactic side effects observed under antipassivization. Section 13.6 contrasts the true antipassive with certain “lookalike” constructions. Finally, section 13.7 addresses the question of whether the antipassive is unique to ergative languages.
13.2 Morphological Indicators of the Antipassive Morphological signs of antipassivization vary widely both within and across languages (see below on Diyari). No single morphological diagnostic exists, and morphology alone may not be sufficient to identify the antipassive. However, a number of noteworthy patterns exist. Let us begin with case-marking. An oblique case often signals the non-core status of the antipassive object, as in (8) and (9). In ergative languages, a change in object encoding (absolutive > oblique) typically corresponds to a change in subject encoding (ergative >
312 Maria Polinsky absolutive). However, at least Warlpiri (Hale 1973a: 366), Djaru (Tsunoda 1981a: 149), and Goonyandi (Tsunoda 1988: 627) preserve ergative marking in the antipassive: (13) a. Njuntulu-lu npa-tju 2sg-erg 2sg-1sg ‘You speared me.’
pantu-nu spear-pst
ŋatju. 1sg.abs
Warlpiri
b. Njuntulu-lu npa-tju-la pantu-nu ŋatju-ku. 2sg-erg 2sg-1sg-ap spear-pst 1sg-dat ‘You speared at me.’ Some researchers argue that the structure in (13b) is in fact an instance of D(ifferential) O(bject) M(arking), which does not entail detransitivization (Campana 1992; Malchukov 2006: 347; Malchukov, Chapter 11, this volume). I will return to DOM in section 13.6.1. Oblique object marking is neither necessary nor sufficient to identify an antipassive. The logical object may appear under pseudo noun incorporation (PNI), a syntactic process that generates a caseless object NP (not a DP) adjacent to the verb (Massam 2001). Unlike regular incorporation, PNI incorporates expressions larger than N°. PNI is more easily detectable in languages with overt absolutive marking. In the following examples, the PNI object includes an adnominal PP; in the Tongan example, a modifier (lelei) scopes over the entire NP. Overt case-marking of this expression is impossible.3 In both cases, the logical object (internal argument) is case-marked differently than in the corresponding transitive clause: (14) a. ‘Oku puke ‘e he pepe [DP‘a e me’a va’inga prs hold erg det baby abs det thing playing [PP mo e pulu] lelei]. com det ball good ‘The baby is holding a/the nice toy and ball.’ (transitive)
Tongan
b. ‘Oku puke (*‘a) [NP e me’a va’inga [PP mo e pulu] lelei] prs hold abs det thing playing com det ball good ‘a e pepe. abs det baby ‘The baby is holding a nice toy and ball.’ (PNI antipassive) (15)
3
a. Ka hoko te matua [DP ī ngā tīkiti aor buy det parent acc det.pl ticket [PP mo ngā tamariki]]. for det.pl child ‘The parent buys (the) tickets (intended) for the children.’
Māori
Additionally, the PNI expression cannot be separated from the verb. Due to space constraints, I do not show the ungrammatical data here.
Antipassive 313 b. Ka hoko [NP tīkiti [PP mo ngā tamariki]] te matua. aor buy ticket for det.pl child det parent ‘The parent buys tickets (intended) for the children.’ PNI is not limited to objects, and the analysis of PNI is a complex issue (Massam 2001; Ball 2008; Clemens 2014). For our purposes, the relevant point is that object PNI is a particular instance of antipassivization: the object does not receive structural case, and the resulting construction is intransitive. In languages without overt object marking (e.g. Diyari, an ergative language with unmarked absolutives), additional diagnostics are required to distinguish direct objects from NPs under PNI. Diyari has a general detransitivizing suffix -ṯadi that forms reflexives and antipassives (Austin 1981b: 71–72). In a subclass of verbs (Austin’s class 2C), both the logical subject and the logical object of the antipassive seem to be in the absolutive form, with SOV word order:4 (16)
a. ṉnulu kaṇa-li ṉiṉa ŋaṉṯi dem.erg person-erg dem.abs meat.abs ‘The man is eating this meat.’(transitive)
ṯayi-yi. eat-prs
Diyari
b. ṉawu kaṇa ṉiṉa ŋaṉṯi ṯayi-ṯadi-yi. dem.abs person.abs dem.abs meat.abs eat-ap-prs ‘The man is having a feed of this meat.’ (antipassive) (Austin 1981b: 154) However, significant differences exist between the (apparent) absolutive objects in (16a,b). While the object in (16a) can be separated from the verb and scrambled, the object position in (16b) is fixed. (16a) answers the question, “Who ate the meat?”, whereas (16b) answers, “What is the man doing?” (Austin 1981b: 154). The nominal demonstrative may be adjectival; the NP+verb unit seems syntactically inseparable, yet does not form a lexical item. This suggests that in (16b), the object ṉiṉa ŋaṉṯi is not absolutive, but a caseless PNI object—which explains its immobility.5 Finally, under regular noun incorporation (NI), the logical object is inaccessible to case-licensors because it has been absorbed into the predicate; several researchers have suggested that NI is a morphological realization of antipassivization (Baker 1988; Kozinsky et al. 1988). In morphologically ergative languages, NI is accompanied by a shift in subject marking from ergative to absolutive. Consider the following triplet from Chukchi: the 4
Another predicate type (Austin’s verb class 2B) undergoes alternation: ergative–absolutive marking with SOV word order in the transitive; absolutive–oblique marking with SVO order in the antipassive (Austin 1981b: 153–154). 5 Example (16b) may be a bi-absolutive (bi-nominative) construction. Forker (2010; 2012) analyzes such constructions across Nakh-Dagestanian languages uniformly as PNI, while Gagliardi et al. (2014) and Coon (2013a) show that the syntax of bi-absolutives varies cross-linguistically. Some bi-absolutives may be antipassives.
314 Maria Polinsky transitive construction in (17a) corresponds to a regular antipassive in (17b) and to an antipassive with NI in (17c). Assuming that the Diyari example in (16b) is a genuine case of PNI, we also note an important difference between PNI and NI: PNI (as in Diyari) is compatible with an antipassive suffix on the verb, but NI (as in Chukchi) is in complementary distribution with the antipassive prefix. (17) a. ʔətt-e melota-lɣən piri-nin. dog-erg hare-abs catch-aor.3sg:3sg ‘The dog caught a/the hare.’
Chukchi (Kurebito 2012: 184)
b. ʔətt-ən ine-piri-ɣʔi (melot- etə). dog-abs ap-catch-aor.3sg hare-dat ‘The dog caught (a/the hare/something).’ c. ʔətt-ən milute-piri-ɣʔi. dog-abs hare-catch-aor.3sg ‘The dog caught a/the hare.’ d. *ʔətt-ən ine-milute-piri-ɣʔi. dog-abs ap-hare-catch-aor.3sg Since the logical object of the antipassive is a non-core argument, it can always be omitted without significant change in meaning, as indicated in (17b). This optionality is yet another morphosyntactic sign of antipassivization. In terms of the grammatical role hierarchy, the antipassive object is always lower than the syntactic object in (7); object marking and (P)NI make this asymmetry visible. The antipassive can also be signaled by verbal affixation. In (17b), the Chukchi verb has the antipassive prefix ine-(note also the difference in word order between (17a) and (17b)). In the majority of languages that mark the antipassive verbally, the affix indexes other categories as well. Two typical patterns of syncretism are attested: (18) a. antipassive is syncretic with detransitivizing affixes such as anticausative, reflexive/reciprocal, middle, or passive markers, e.g., in Chukchi (Kozinsky et al. 1988), Diyari (Austin 1981b), Halkomelemem -Vm (‘middle’, per Gerdts and Hukari 2005, 2006), some Pama-Nyungan languages (Dixon 1972; 1977; Terrill 1997), Kiowa (Watkins 1984) b. antipassive is syncretic with aspectual markers, most commonly inchoative, inceptive, or iterative e.g. in Bezhta (Comrie et al. 2015) or Eskimo/Inuit6 (Spreng 2012; Basilico 2012) I have not observed languages with a non-syncretic antipassive marker. In principle, such a language could exist; however, the patterns in (18) are common enough to justify 6
I am using “Inuit” as a cover term for several languages/dialects in the North American Arctic and Labrador for which similar data have been reported by different researchers.
Antipassive 315 the prediction that a given antipassive marker will also serve as a general detransitivizing/aspectual affix. The syncretic patterns in (18) correlate with two main syntactic analyses of the antipassive, discussed in section 13.4.2. More than one antipassive verbal marker can exist in a single language. These markers can be stacked (doubled), although the effects of such doubling are poorly understood. In Chukchi, the productive antipassive prefix ine-and the semi-productive antipassive suffix -tko-can co-occur (Kozinsky et al. 1988: 661); likewise, in Halkomelem, the antipassive suffixes -m-(“middle,” per Gerdts and Hukari 2006) and -els-(“activity”) stack: (19)
Niʔ ḱwłeʔ-əm-els ʔə aux pour-ap-ap obl ‘John served some tea.’
ḱw det
ti tea
tθə John. Halkomelem det John (Gerdts and Hukari 2005, ex. (26))
The stacking of antipassive markers has no apparent syntactic consequences. By itself, the suffix -m serves to detransitivize the verb; the contrast in meaning between the stacked and single antipassives is unclear and warrants further investigation. Next, the antipassive is morphologically visible through agreement. Antipassives typically show intransitive agreement, since the object, as a syntactically non-core argument, cannot agree—compare (3) and (10) for Q’anjob’al; (17a) and (17b) for Chukchi. Changes in agreement are typically accompanied by changes in case-marking and/or the addition of a verbal marker; all three morphological cues appear in (17b). However, in some languages, the change in agreement is the only sign of antipassivization. Finally, antipassives sometimes differ from corresponding transitive constructions in word order (compare (17a,b)); however, this is a very weak diagnostic, especially because the oblique object in the antipassive is generally dispensable. Word order differences may be a side effect of the adjacency requirement imposed by PNI (see examples (14) and (14).
13.3 Interpretative Properties of Antipassives Several interpretative properties are commonly associated with the antipassive; none of these properties is essential, and none applies to all languages with purported antipassives. Thus, these properties are best viewed as concomitant, rather than as defining, characteristics of the antipassive, much as thematic discourse prominence is a concomitant but not defining property of the passive. To reiterate the discussion in section 13.1, antipassivization is a syntactic operation whose defining properties are independent of its interpretative characteristics and discourse uses. Antipassives often have special aspectual meaning: inchoative, inceptive, durative, progressive, imperfective, or even iterative (Tsunoda 1981b; Tchekoff 1987; Dowty 1991; Cooreman 1994; Dixon 1994; Spreng 2010). In some languages, the antipassive is
316 Maria Polinsky reanalyzed as durative and loses its syntactic detransitivizing properties (Comrie et al. 2015: 554). In each case, the antipassive may be associated with atelicity (the antipassive ~ imperfective correlation).7 This association, however, is not present in all languages; for example, it is not found in Chukchi, Chamorro, or Polynesian languages. The typicality of the antipassive ~ imperfective correlation predicts the following: (20) If an antipassive construction can have a perfective (telic) interpretation, it must also have an imperfective (non-telic) interpretation Another common interpretative effect of the antipassive is semantic or pragmatic SUBJECT prominence, a phenomenon referred to as “agent foregrounding,” “agent focus(ing),”8 or “agent maintenance.” Agent foregrounding likely follows from the activity interpretation of the antipassive and the absence/backgrounding of the logical object. If the object is omitted, then the subject (agent) naturally becomes the only salient participant. When the object is expressed, its oblique encoding has subtle interpretative consequences. In the antipassive’s transitive counterpart, new discourse referents are typically introduced as internal (accusative, absolutive) arguments, especially in presentational constructions (Du Bois 1987b; Prince 1992; Birner and Ward 1996, 1998), indicating that the new referent will likely be featured in the upcoming discourse (Givón 1983b). If a new referent appears as an oblique (or incorporated nominal), it signals that the referent will not be maintained in subsequent discourse (Polinskaja and Nedjalkov 1987). Thus, the antipassive construction facilitates a low-individuation interpretation (indefinite, non-specific) of the object participant (Foley and Van Valin 1984; Cooreman 1994): in other words, the object is presented as a prototypical ‘anti-topic’—see Kalmár (1979), Berge (2011), and Johns and Kučerova (Chapter 17, this volume) for Inuit. The foregrounding of the agent and concomitant backgrounding of the object is nevertheless far from a universal characteristic of the antipassive. While in some languages, the antipassive is impossible with highly individuated (first or second person) objects (see Gerdts 1988a: 157 for Halkomelem), some languages require the antipassive construction specifically with these roles (see section 13.5.3). Semantic and discourse functions of antipassives differ both within and across languages (Heath 1976; Comrie 1978; Cooreman 1988; 1994).
13.4 Analysis of the Antipassive Several analyses of the antipassive have been advanced within a number of theories. What follows is a brief survey; the reader is advised to consult the respective references for more 7 The correlation antipassive ~ imperfective on the one hand and passive ~ perfective on the other is yet another sign of the mirror-image correspondence between antipassive and passive. 8 Agent focusing should not be confused with the special agent focus construction, documented in Mayan languages (see section 13.6).
Antipassive 317 details. Although terminology and theoretical tools vary across approaches, analyses of the antipassive construction can be broadly divided into lexicalist and syntactic camps.
13.4.1 Lexical(ist) Approaches Even in languages with a robust antipassive, use of the construction is lexically specified. Typically, the antipassive occurs with “manner verbs”, i.e. verbs that denote actions performed in a particular manner with no entailed result-state (Rappaport Hovav and Levin 2010 and references therein). Objects of manner verbs can be readily “suppressed” or left implicit, producing an agent emphasis that parallels the focus effect reported for the antipassive. Arkadiev and Letuchiy (2008) systematically test and confirm the restriction of the antipassive to manner verbs in Adyghe, but supporting work in other languages remains to be done. In some languages, only a handful of verbs are incompatible with the antipassive construction. In Chamorro, Chung (1998: 39) reports three verbs that do not form antipassive predicates (without noting how many verbs there are in total). Such scenarios raise the question: what is the minimally sufficient number of exceptions necessary to motivate a lexical generalization?9 Given the availability of lexical rules, productivity of a construction does not necessitate a syntactic analysis; conversely, confinement of a construction to a particular semantic class does not rule out a syntactic analysis—such restrictions can be built into the features or meanings of the heads involved. Lexical(ist) approaches to the antipassive postulate a lexical rule that affects clausal argument structure, demoting the object/theme to an oblique nominal. The antipassive morpheme is then added to the verbal base as a concomitant to the lexical rule. See Grimshaw and Mester (1985) for the application of this approach to Inuit languages and Woodbury and Sadock (1986) for refinements and modifications. A clear discussion can be found in Farrell’s (1992) lexical account of the Halkomelem antipassive. Working in the framework of Relational Grammar, Farrell argues that antipassivization cannot be derived syntactically; significantly, he observes that non-eventive nominalizations can be formed from antipassive verbs but not from passive verbs:10 (21)
a. niʔ cən t’il-əm ʔə aux 1subj sing-ap obl ‘I sang the hymn.’ b. s-t’il-əm nmlz-sing-ap ‘song’
9
tə det
st’iʔwiʔəł. hymn
Halkomelem (Farrell 1992, ex. (16a))
(Farrell 1992, ex. (17a))
For some languages, the data vary dialectically. For instance, the variety of Chukchi described by Skorik (1948; 1961; 1977) and the one I have observed have a fully productive antipassive; meanwhile, according to Dunn (1999: 200–201, 70), the Telqep dialect has a low-productivity antipassive. 10 The same detransitivizing suffix is used to derive the passive and the antipassive (see (18a)), but in the passive, that suffix must follow the transitive suffix -Vt.
318 Maria Polinsky c. xws-t’il-əm agt.nmlz-sing-ap ‘singer’
(Farrell 1992, ex. (17b))
d. *s-t’il-ət-əm nmlz-sing-tr-pass If both the passive and antipassive constructions were formed syntactically, this contrast would be mysterious. If, however, antipassivization is a rule that modifies the argument structure of an input word, these facts follow straightforwardly. Gerdts and Hukari (2005; 2006) advance a lexicalist analysis that associates the antipassive with “agent maintenance” or prominence (see section 13.3); they propose a mapping rule that specifies antipassive verbs as two-place predicates selecting only an agent argument. (22) Do not link object to an argument and cancel the inflectional position for object if there is one Head-driven Phrase Structure Grammar (HPSG) characterizes the antipassive as a complex argument structure of the type shown in (23). The outer a(rgument)-subject, i.e. the leftmost argument on the (local) argument list, maps to the syntactic subject; optionally, a distinct inner argument maps to an oblique object: (23) Antipassive ARG-STR > These are just some available lexical analyses of the antipassive; other devices are also possible. The crucial consideration in each case concerns the language-internal evidence for treating the antipassive as lexically specified. In those languages where the construction is (at least partially) productive, syntactic alternatives warrant attention.
13.4.2 Syntactic Approaches to the Antipassive The absence of a single, unifying syntactic analysis of the antipassive should not be surprising, given the significant cross-linguistic variation we saw earlier. (Disagreements among researchers over the definition of “antipassive,” which I alluded to earlier, do not improve matters.) Analyses converge on the understanding that the difference between the transitive and the antipassive is in the status of the logical object. According to one family of analyses, accusative/absolutive case is absorbed in the vP; other analyses contend that the antipassive allows object licensing, but that the structural locus of that licensing differs from the transitive.
13.4.2.1 Case Absorption “Demotion” of the logical object can be achieved by base-generating a nominal element inside the VP. This abstract nominal constituent (AP) undergoes head movement and
Antipassive 319 absorbs the accusative/absolutive case. Since the VP-internal argument position is saturated, the verb cannot assign case to its object. Thus, if an antipassive object appears, it is analyzed as an adjunct. Schematically, the derivation proceeds as follows; the functional head v can assign only one case, and that case is assigned to the external argument. (24)
vP v’ v
VP ti
APi-V
Baker (1988) first observed the parallels between genuine NI and antipassivization, describing the antipassive morpheme as a type of incorporated object. Under Baker's analysis, the internal argument of the verb is filled by a non-maximal nominal expression (most likely N, although some researchers identify it as D; cf. Bittner and Hale 1996a, b). Several morphological realizations are possible: an abstract “antipassive” nominal (expressed through the antipassive morpheme), an incorporated internal argument, or an implicit argument. This nominal antipassive marker undergoes c-selection-triggered head movement into the lexical verb (some researchers suggest that phrasal movement may be implicated as well). The incorporated nominal expression does not receive case, and the V+AP complex is syntactically intransitive. External case assignment (by the inflectional head or the functional head v that dominates the V+AP complex) proceeds according to language-specific requirements. When the logical object is realized as an oblique expression, it is a genuine adjunct, never an argument; compare to the adjunct analysis of the by-phrase in passives (but see Goodall 1993, 1999 and Bruening 2013, for a different view of passives). The nominal analysis is particularly appealing for antipassives whose marker is syncretic with reflexive or middle morphology (see (18a)), as reflexive and middle morphemes saturate the internal argument position of a two-place predicate (Basilico 2004). Additionally, this approach predicts that NI and antipassivization are two sides of the same coin. Recall that the two operations are mutually exclusive in Chukchi: in fact, the Telqep dialect of Chukchi (Dunn 1999) uses NI in most instances where northern dialects employ a genuine antipassive (see n. 11). The nominal antipassive analysis is open to various interpretations with respect to the obligatory implicit argument—a covert element whose syntactic representation is far from clear. It is generally assumed, following Rizzi (1986), that there is a difference between syntactically projected objects (which can be realized as pro, for example, in Italian, or as bound variables, as posited for Mandarin or Portuguese), and another, less- well-defined category of lexically saturated implicit objects (IMPs), which need not be syntactically projected (as in English). Lexical saturation may proceed from different sources, including verbal semantics. The use of IMPs is often more lexically constrained than the use of syntactically represented objects.
320 Maria Polinsky The nominal antipassive analysis would be problematic if the silent object were expressed by a null pronominal; such a configuration would entail that a single theta- role be filled twice in the same predication. Since antipassives are often associated with lexical exceptions, this latter property suggests that objectless antipassives may in fact include IMPs. This possibility leads to a number of structural predictions extending from the differences between syntactically projected objects and IMPs. Specifically, only the former can participate in control, bind reflexives and be modified by adjunct small clauses. Although this aspect of antipassive constructions has not been systematically investigated, the available predictions could be put to the test in languages with implicit- object antipassives.
13.4.2.2 Different Object-Licensing Positions Although the case-absorption analysis has been popular, some researchers, in particular those working on Inuit, reject the connection between the antipassive and nominal head movement (Schmidt 2003; Spreng 2006; 2012). The main argument for this approach comes from the observation that in Inuit, noun incorporation and the antipassive marker are not mutually exclusive. Their co-occurrence is illustrated in the following example from West Greenlandic:11 (25) a. Meeqqat tujuulu-ler-pai. children.abs sweater-provide.with-ind.3sg.3pl ‘S/he dressed the children in sweaters.’ b. Meeqqat-nik tujuulu-li-i-voq. children-ins sweater-provide.with-ap-ind.3sg ‘S/he dressed the children in sweaters.’
West Greenlandic
(Schmidt 2003: 388)
In the absence of a nominal AP head, the transitive/antipassive distinction can be derived based on two related criteria: the way in which an object is licensed and the semantic contribution of the verbal root in terms of result or manner. Basilico’s (2012) analysis of Inuit ergative–antipassive alternation is a particularly clear instance of such an approach. In Basilico’s analysis, the absolutive object is introduced and licensed by the functional head v (the ergative is considered an inherent case on that analysis). The verb root in this structure provides the result component to the interpretation of a given event, and the absolutive object serves as the measuring-out argument (cf. (26a)). In the antipassive, the verb root provides the manner component, not the result component to the interpretation of a given event; this root combines with the light verb encoded by the antipassive morpheme, forming an intransitive. The resulting intransitive verb can optionally combine with a PP, whose interpretative 11 A number of researchers have suggested that what is referred to as noun incorporation in Inuit may involve different mechanisms than those found in familiar cases where the internal argument of a verb adjoins to the verbal head (Baker 1988; Johns 2009). If this proposal is on the right track, the argument against antipassivization as head movement loses its power.
Antipassive 321 contribution is in presenting the scale of a given event—consider the schematics in (26b): (26) a.
vP v’
DPobject
vP
v [θ: theme] [abs]
v
√ [RESULT]
vP
b. v
PP v antipassive affix
√ [MANNER]
Some other analyses postulate the presence of an extra functional projection in the antipassive verbal complex (Alexiadou 1999; Spreng 2012; Compton, Chapter 34, this volume). The extra projection licenses the internal argument and is aspectual in nature. Details of interpretation differ, but most scholars hold that this abstract antipassive functional head is aspectually specified as [-telic] or [+imperfective] (cf. (18b)). Its aspectual value is responsible for the persistent aspectual characteristics of the antipassive (section 13.3). This functional head v/Asp selects the verbal root and licenses either inherent case or accusative case on the internal object. The accusative analysis of the antipassive object has been particularly prominent in work on Inuit (Bok-Bennema 1991; Bittner 1994; Spreng 2012). Finally, the external argument of the antipassive receives case from the v/Asp head or from a higher head: vP
(27)
v’
DPexternal v
AspP
DPinternal
Asp’ Asp
√/V
While details of implementation may differ, this approach relies on the contrast between absolutive licensing (high in the structure) and accusative licensing (low). The antipassive object is always “deeper” in the structure than the absolutive object—but
322 Maria Polinsky if a case-licensing system allows for both accusative and absolutive licensing, both objects receive structural case (cf. Woolford 1997; Aldridge 2008a; 2011; Legate 2008). However, this analysis does not explain why antipassive objects often appear in an oblique case. It is also unclear what motivates object drop in the antipassive; even if the two objects are licensed in different places in the verb phrase, the regular omission of one object but not the other remains unmotivated. The object-omission problem does not arise on the case absorption analysis, which treats the antipassive object as an adjunct. The parallel between the object case assigned to the antipassive object and the more traditional accusative case is important in light of historical processes. It is often assumed that ergative languages can become nominative–accusative if the ergative construction is lost and the antipassive is reanalyzed as a transitive (Anderson 1977; 1988; Comrie 1978; Harris and Campbell 1995: ch. 9; Spreng 2010; Aldridge 2011). The structural similarity between the antipassive object case and the true accusative could facilitate such a reanalysis.
13.4.2.3 Subject Case and Antipassives In addition to the demotion of the logical object, in morphologically ergative languages, the case of the subject also changes in the antipassive construction, from ergative to absolutive. Most researchers agree that the subject of a transitive verb and its antipassive/intransitive counterpart are in the same position.12 Two main approaches have been explored. Some authors contend that the choice of subject case is a side effect of transitivity (Woolford 2006) and propose a special transitivity requirement which states that the ergative is incompatible with basic or derived intransitives. Bruening (2007) offers a detailed critique of this approach, protesting that “an additional transitivity requirement simply restates the definition of ergative case, making the explanation for its distribution include a statement of its distribution; it is inconsistent in its treatment of accusative case.” An alternative analysis (Marantz 1991; Bittner and Hale 1996b; Levin and Preminger 2015; Baker, Chapter 31, this volume; Baker and Bobaljik, Chapter 5, this volume) contends that case is determined by the presence of a case competitor. In this type of theory, nominative/absolutive case is unmarked, and ergative and accusative are only assigned in opposition to a nominative. If a predicate has only one argument, that argument will receive the unmarked nominative/absolutive case. If it has two arguments, one will receive the marked case—ergative or accusative—and the other will receive nominative/absolutive. Ergative is assigned to the higher of two co-arguments; accusative is assigned to the lower. An advantage of this approach is that it treats passive and antipassive consistently: in both structures, one argument is demoted (subject in the passive, object in the antipassive), leaving the other to receive default nominative/ absolutive. 12
See Bobaljik (1993a) for a different view, according to which transitive subjects move to a higher A-position than transitive or antipassive subjects.
Antipassive 323
13.5 Syntactic Side Effects of Antipassivization This section discusses syntactic phenomena associated with the use of the antipassive. Although many of these effects are still poorly understood, some may motivate antipassive use in particular languages.
13.5.1 Way-Station Effects In a number of languages, only (intransitive) subject or absolutive DPs can undergo A- bar movement, participate in control/ raising chains, and bind anaphors. Antipassivization can serve as a “way-station,” allowing the sole argument of a detransitivized verb to participate in such grammatical processes. In Chukchi, for example, relativization in participial clauses is possible only for absolutive arguments (Polinsky 1994): intransitive subjects and transitive direct objects can relativize directly with a gap, as shown in (28b) and (29b). For a transitive subject to relativize, it must first become an absolutive via antipassivization (30a); the subject of the antipassive is then relativized as in (30b).13 (28) a. ŋinqey pəkir-gʔi. boy.abs arrive-aor.3sg ‘The boy arrived.’
Chukchi
b. [ti pəkərə-lʔ-ən] ŋinqeyi arrive- ptcp-abs boy.abs ‘the boy that arrived’ (29) a. Tumg-e ŋinqey rəyegtetew-nin. friend-erg boy.abs save-aor.3sg.3sg ‘The friend saved the boy.’ (transitive) b. [tumg- e ti rəyagtala-lʔ-ən] ŋinqeyi friend-erg save-ptcp-abs boy.abs ‘the boy that the friend saved’ c. *[ ti ŋinqey rəyagtala-lʔ-ən] tumgətum boy.abs save-ptcp-abs friend.abs (‘the friend that saved the boy’) 13 Chukchi is often used in illustration, but it is by no means unique in its relativization behavior. For example, Lower Grand Valley Dani (Bromley 1981) and Inuit (Bittner 1994; Manning 1996) also require antipassivization as a way-station for subject extraction with a gap.
324 Maria Polinsky (30) a. Tumgətum ŋinqey-ək ine-nyegtele-gʔi. friend.abs boy-loc ap-save-aor.3sg ‘The friend saved the boy.’ (antipassive) b. [ti ŋinqey- ək ine-nyegtelewə-lʔ-ən] tumgətumi boy- loc ap-save-ptcp-abs friend.abs ‘the friend that saved the boy’ Relatedly, in a number of languages (not all of them ergative), non-finite control can only target the intransitive subject position. Again, the antipassive “way-station” can license control. In the following two examples from Jakaltec, control complement clauses must be intransitive (Craig 1977: 311–327). The embedded clause in (31a) is antipassive; transitive non-finite clauses with overt absolutives, as in (31b), are ungrammatical. (31) a. Ch-ach to [PRO col-wa-l asp-abs go help-ap-nom ‘You are going (there) to help him.’
y-iñ gen-to
Jakaltek
naj]. him
b. *Ch- ach to [PRO ha- col- al y- iñ naj]. asp-abs go erg-help-nom gen-to him
(Craig 1977: 318)
A similar restriction appears in control complements in Chukchi (Nedjalkov 1976; Skorik 1977) and in some languages of Australia (Dixon 1979; 1994). Aldridge (2008a; 2008c; 2011) extends this restriction to a number of Austronesian languages (e.g. Seediq, Tagalog) and suggests that it may be tied to ergativity; however, the ergative status of these languages is debatable.
13.5.2 Scope Effects Antipassives and transitives reportedly differ in scope readings in Inuit. Inuit antipassive objects are restricted to narrow scope, while transitive absolutive objects can take wide or narrow scope (Bittner 1987, 1994; Benua 1995). In (32a), the absolutive object takes wide scope, and in (32b), the antipassive object must take narrow scope: (32) a. Atuartu-t ila-at ikiur-tariaqar-p-a-ra. student-pl.erg part-3pl.sg help-must-ind-[+tr]-1sg.3sg ‘There is one of the students that I must help.’
West Greenlandic
b. Atuartu-t ila-an-nik ikiu-i-sariaqar-p-u-nga. student-pl.erg part-3pl.sg-ins help-ap-must-ind-[-tr]-1sg ‘I must help one of the students (any one will do).’ (Bittner 1994: 138)
Antipassive 325 Such scope differences, however, are not consistently associated with the antipassive cross-linguistically. For example, in Adyghe, both transitive and antipassive clauses with an oblique object are ambiguous between surface and inverse scope:14 (33)
a. pŝaŝe-m zeč’e-r-jә pjәsme-(xe-)r ∅-ә-txә-ʁ. Adyghe girl-erg all-abs-add letter-pl-abs 3abs-3sg.erg-write-pst ‘A/The girl wrote all the letters.’ (one > all; all > one) b. pŝaŝe-r zeč’e-m-jә pjәsme-(xe-)m ∅-txa-ʁe. girl-abs all-obl-add letter-pl-obl 3abs-write.ap-pst ‘A/The girl wrote all the letters.’ (one > all; all > one)
Likewise, in the English conative alternation (see section 13.7), both forms display scope ambiguities: (34) a. An athlete grabbed every curl-bar. (a > every, every > a) b. An athlete grabbed at every curl-bar. (a > every, every > a) Why is Inuit different from other languages? Let me offer some considerations. The crucial test case for inverse scope is found in doubly quantified sentences where ‘one/ a’ scopes over ‘every’ at the surface string. This configuration is present in (33) and (34) but not in (32); arriving at a robust comparison is difficult without additional West Greenlandic data. It is possible that fixed scope results from some interaction between quantification and modality (which is known to be quite complex; cf. Hacquard 2006: 118ff. and references therein), between quantification and negation, or between quantification and the defined aspectual properties noted for the Inuit antipassive (recall (27)). The bottom line is that differences in West Greenlandic scope readings may well be orthogonal to the transitive/antipassive contrast. Assuming a structure such as (27), there is no reason to expect antipassive scope relations to differ from transitive scope relations. Our empirical knowledge of these phenomena must be considerably enriched before analysis can move beyond speculation.
13.5.3 Antipassive and Agreement Antipassives may appear when the subject and object are in an “inverse” relationship— i.e. when the subject is lower than the object on one or more hierarchies: person (1>2>3); number (sg > pl); animacy (human > animate > inanimate). Languages vary in the combinations they consider “inverse,” and careful examination of a given agreement system is always needed; in general, however, detransitivizing morphology—passive 14 The ergative and the oblique in Adyghe have the suffix -(V)m. Arkadiev and Letuchiy (2008) treat all instances of -(V)m as oblique; here and in Caponigro and Polinsky (2011), I gloss -(V)m differently depending on its function.
326 Maria Polinsky or antipassive—can be used to mark inverse configurations. Inverse marking through antipassive is common in Australian languages. In Yukulta, a third person can act transitively upon another third person, but the antipassive is used when a third person acts upon a first person: (35) a. Kungul-i=ka-nt-a mosquito-erg=tr-nprs-realis ‘A mosquito bit him.’
paa-tya. bite-ind
Yukulta (Keen 1983, ex. (146))
b. Kungul- ta=thu=yingk- a paa- tya. mosquito-abs=1sg.dat-nprs.intr-realis bite-ind ‘A mosquito bit me.’ (Keen 1983, ex. (147)) Some instances of inverse ~ antipassive correlation may reveal a diachronic connection but be synchronically distinct. I will discuss one such case in section 13.6.2.
13.6 Related Constructions 13.6.1 Antipassive Lookalikes Above, I mentioned certain linguistic phenomena that border on or subsume antipassives: some instances of PNI, object NI constructions, and some bi-absolutive (bi- nominative) constructions. Here, I discuss two other constructions that resemble the antipassive but should be differentiated from it. The first is the agent focus (AF) construction, reported for a number of Mayan languages (see Stiebels 2006 for an overview). Although AF has some hallmarks of the antipassive, it is considered a transitive, not an intransitive, construction (Aissen 2011); AF is variously analyzed as a separate voice (Tonhauser 2003), an inverse form (Aissen 1999a), or a specialized agreement form for A-bar subject extraction (Stiebels 2006). Some Mayan languages, such as Q’anjob’al, have both genuine antipassives and AF—see (10) for the antipassive and Coon et al. (2014) for a discussion of AF. However, some early work, particularly Larsen and Norman (1979), identifies AF as an antipassive construction; it would be prudent to assess the AF ~ antipassive relationship in each language separately. Differential object marking (DOM) resembles the antipassive in that objects of transitive verbs receive different encoding depending on animacy, specificity, etc. However, in DOM, the logical object remains a syntactic direct object, regardless of overt marking (Aissen 2003b). Furthermore, semantic factors associated with DOM differ from those that condition antipassives. A similar superficial parallel arises between antipassives and differential subject marking (DSM) constructions; although the case of the subject shifts in both constructions, only in the antipassive does this change in subject case- marking crucially depend on the status of the object.
Antipassive 327
13.6.2 Spurious Antipassive As noted, antipassives may be used to indicate an atypical/inverse hierarchical relationship between agent and theme. Noting several unusual properties of this phenomenon, researchers have labeled the antipassive that serves this inverse function the “crazy antipassive,” “eccentric agreement” (Hale 2002), or “spurious antipassive” (Halle and Hale 1997). Spurious antipassives in Chukotko- Kamchatkan languages represent a well- documented case of such “inverse” agreement (Comrie 1979, 1980; Nedjalkov 1979; Bobaljik and Branigan 2006). Simplifying somewhat, these antipassives occur in contexts where a second-or third-person participant acts upon a first-person participant. Compare (36a) and (36b); see Dunn (1999: 181–184) and Bobaljik (2007) for complete paradigms. In (36b, c), the verb carries an antipassive prefix; agreement is intransitive, but pronominal case-marking remains transitive. Both antipassive affixes, ine-and -tku-, participate in inverse alignment: (36) a. ɣəm-nan ɣət 1sg-erg 2sg.abs ‘I saw you.’
tə-ɬʔu-ɣət. 1sg.subj-see-aor.2sg.obj
Chukchi
b. ə-nan ɣəm ine-ɬʔu-ɣʔi. 3sg-erg 1sg.abs ap-see-aor.3sg ‘S/he saw me.’ c. ɣət-nan muri ɬʔu-tku-∅. 2sg-erg 1pl.abs see-ap-aor.2sg ‘You (sg.) saw us.’ Bobaljik and Branigan (2006) and Bobaljik (2007) argue that (36b, c) are regular transitive clauses, with both subject and object licensed in a typical manner by the inflectional head T. The “spuriousness” pertains to the morphological interpretation of these clauses. Since these authors understand agreement to be post-syntactic, they locate antipassive insertion within the mapping of narrow syntax to the morphological component; the features of the lower argument (object) are deleted at T, rendering T apparently intransitive. This analysis eliminates “spuriousness” by maintaining the uniformity of the antipassive affix; however, it crucially depends on Bobaljik’s (2008) conception of agreement as a post-syntactic operation. If object features play no role in derivation, then it is surprising that spurious antipassives distinguish between singular and plural objects. ine- is used only for singular first-person objects, and -tku- for plural (cf. (36b, c); see Dunn 1999: 184). In addition, an inverse form marked with the prefix ne- is used to encode a plural third-person participant acting upon a (singular or plural) third-person participant. Most importantly, regular antipassive markers are preserved in action nominalizations, but the “spurious” ones are lost, just like all other agreement markers in
328 Maria Polinsky nominalizations.15 This suggests that spurious antipassives are categorically different from true antipassives. (37) a. Gəm ʔituʔit-ək 1sg.abs goose-loc ‘I saw a/the goose.’
ine-ɬʔu-ɣʔek. ap-see-aor.1sg
Chukchi
b. Gəm-in ʔituʔit-ək/*ʔituʔit-in ine-ɬʔu-wərg-ən 1sg-poss goose-loc/goose-poss ap-see-nmlz-abs ‘my seeing of a/the goose’ (38) a. ətɬʔəg-e ɣəm father-erg 1sg.abs ‘Father saw me.’
ine-ɬʔu-ɣʔi. ap-see-aor.3sg
b. ətɬʔən- in (ɣəm-in/*ɣəm) ɬʔu-wərg-ən/*ine-ɬʔu-wərg-ən father-poss 1sg-poss/1sg.abs see-nmlz-abs/ap-see-nmlz-abs ‘Father’s seeing me’ Bobaljik and Branigan’s generalization—that clauses such as (36b, c) are transitive— holds. However, ine- and -tku- here are not antipassive morphemes and have nothing to do with valency-changing morphology; instead, they are person/number agreement markers (see Abramovitz 2015 for a similar conclusion). The details of such agreement are yet to be worked out. It is possible that the two functions of ine-/–tku- are related historically; on the synchronic level, however, the Chukchi “spurious” antipassive is not antipassive at all.
13.7 Antipassive Is Not Unique to Ergative Languages Some researchers have suggested that antipassives are unique to ergative languages (Silverstein 1976; Dixon 1979; Spencer 1991: 24) or, even more narrowly, to syntactically ergative languages, that is, languages where the ergative cannot undergo A-bar movement (e.g. Otsuka 2000). However, there is nothing in the basic definition of the antipassive that predicts this restriction. Choctaw (Davies 1984), Chamorro (Cooreman 1988, 1994), Kiowa (Watkins 1984), Māori (Bauer 1983), German (Müller 2011), Romance (Postal 1977; Masullo 1992; Mejias-Bikandi 1999; Medová 2010), and Slavic (Say 2005;
15 Chukchi event (action) nominalizations typically allow just one possessive form, which can correspond to either the external or internal argument, making them ambiguous. The inclusion of both possessives within a nominalization is permitted but not preferable.
Antipassive 329 Medová 2010) are good examples of accusative languages with antipassives (for more examples from a survey of grammars, see Polinsky 2013). Within the framework of Relational Grammar, Postal (1977) specifically argues against the unique association of the antipassive with ergativity; see also Heath (1976), Givón (1984); Polinsky (2013). What is truly at issue here is visibility: in ergative languages, presence of the antipassive correlates with an obvious change of subject encoding from ergative to absolutive; thus, this construction is more noticeable in ergative languages than it is in accusative languages, but it is not limited to ergative languages. Conversely, passives are not impossible in ergative languages, contra Laka (1993b: 168), Dixon (1994: 152), van de Visser (2006). Quite a few languages exhibit both passive and antipassive constructions. Just within morphologically ergative languages, we can find both passive and antipassive in Halkomelem, Inuit, Georgian, several Mayan languages (Vapnarsky et al. 2012), and possibly Basque. It is true, however, that passives are generally less common in ergative languages (Kazenin 2001b: 926), leading Nichols (1992: 158) to propose the following universal: if a language has an antipassive but no passive, that language is ergative. In English, the closest parallels to the transitive ~ antipassive alternation are the conative alternation and the unexpressed object alternation, for which multiple lexical restrictions apply; Blight (2004) analyzes these alternations, as well as the preposition-drop alternation, as English antipassives. The conative alternation contrasts two instances of the same verb: one that takes a direct object, and another that takes a prepositional complement. Only the direct object is construed as affected by the verbal event (Levin 1993: 5–11, 41–42; Beavers 2011; Vincent 2013). For example, (39a) entails that the hunter hit the target, rendering the continuation infelicitous; in (39b), this entailment is absent: (39)
a. The hunter shot the bear #but he missed. b. The hunter shot at the bear but he missed.
Levin associates the conative alternation with verbs that describe change of state achieved through motion and contact (as opposed to verbs that denote pure change of state, such as break, verbs that denote contact only, and verbs that denote motion only). It may be possible to extend this generalization to antipassives with an explicit oblique complement, but more cross-linguistic work on change-of-state verbs and implicit- argument verbs is needed to determine the relevant lexical restrictions. It is true that, whereas use of a prototypical transitive verb entails a change of state in the object participant, the corresponding antipassive cancels such an entailment, and this cancellation correlates with the oblique marking of the object (Hopper and Thompson 1980; Tsunoda 1981b; Dowty 1991; Van Valin 1991; among others). Omission of the affected object participant can have two consequences. First, when a clause lacks overt mention of a participant affected by the event (incremental theme), the event is likely to be interpreted as incomplete (Dowty 1991; Spreng 2010; Basilico 2012). This consequence echoes the established correlation between the antipassive construction and the imperfective
330 Maria Polinsky interpretation. Second, in the absence of an affected patient, it becomes possible to interpret the agent as an affected participant of the event. The reflexive construction similarly permits a verbal event to cause change in the agent; as discussed earlier, antipassive and reflexive mechanisms do overlap in some languages. The unexpressed object alternation in English brings together the following types of sentences (Levin 1993: 33–38): (40) a. I have ironed the clothes. b. I have ironed. Levin (1993: 33) observes that this alternation occurs with a wide range of activity verbs where the missing object is understood as the “typical object” of a given verb; of course, the notion of “typical object” is vague, but it seems that a similar observation can be made for the unexpressed object of the antipassive construction. All things considered, it is probably better to account for the English conative alternation and unexpressed-object verbs via lexical rules (section 13.4.1); however, a syntactic approach like the one illustrated in section 13.4.2.1 is also possible.
13.8 Conclusions Antipassives have long been considered “exotic”—found in exotic languages and associated with exotic syntax. One of my goals in this chapter has been to illustrate that the antipassive is in fact well behaved, observable wherever the logical object of a transitive predicate appears as a non-core argument or an adjunct. Too often, we seek antipassives in places where they do not exist while ignoring obvious instances in our own “Standard Average European” backyard. Once we agree on a more manageable landscape for navigation, the task of identifying antipassives becomes much simpler. On the one hand, a number of constructions that resemble the antipassive are truly distinct; even if two constructions have similar discourse functions or similar morphological hallmarks, they need not be syntactically equivalent. On the other hand, certain neglected constructions, such as (P)NI objects, meet the structural definition of the antipassive and should be duly considered under that rubric. Based on the criteria discussed here, we must reject the notion that antipassives occur only in ergative languages. Likewise, it is unreasonable to maintain that antipassives and passives are mutually exclusive; nothing in the definition of the antipassive makes such a prediction, and a number of languages provide empirical evidence in support of anti/passive compatibility. The antipassive bears analysis either as a case frame of individual lexically specified verbs that alternate with regular transitives, or as the output of a syntactic operation. Only in the latter case can one assume, following Silverstein (1972), that “the sense [of the antipassive—MP] is clearly equivalent to a transitive.” Lexical analysis is plausible when only a subset of a language’s verbal lexicon participates
Antipassive 331 in the alternation and when independent evidence supports such an analysis, as in Halkomelem nominalizations. Within syntax, two main approaches are taken, which differ along two related dimensions: the character of the antipassive affix (nominal/verbal), and object licensing. Under the first approach, a nominal element saturates the internal argument of a two-place verb, and as a result, the verb cannot assign case to its object. Under the second approach, the difference between the transitive and the antipassive is reduced to a licensing distinction between absolutives (high objects) and accusatives (low objects). Often, such differential licensing is associated with an extra functional head with aspectual connotations. These two approaches attempt to account for the semantic and pragmatic effects of the antipassive without defining the construction by these properties.
Acknowledgments A portion of this work was supported by funding from NSF (BCS-114223, BCS-137274, BCS- 1414318), Harvard University, and the Max- Planck Institute for Evolutionary Anthropology. Thanks to Rafael Abramovitz, David Basilico, Grant Goodall, Ivona Kučerová, Beth Levin, Gesoel Mendes, Eric Potsdam, Jerry Sadock, Adam Singerman, and Bettina Spreng for helpful comments; I am solely responsible for any remaining errors. Unless stated otherwise, examples come from my own fieldnotes. I am grateful to Raxmet Yesheva (Adyghe), Peter Inenlikey and Vladimir Raxtilin (Chukchi), Winifred Bauer (Māori), Pedro Mateo Pedro (Q’anjob’al), Sisilia Lutui (Tongan), and Roksolana Mykhaylyk (Ukrainian) for linguistic examples.
Abbreviations abs, absolutive; acc, accusative; add, additive; agt, agent; ap, antipassive; asp, aspect; aux, auxiliary; clf, classifier; com, comitative; compl, completive; dat, dative; det, determiner; erg, ergative; fem, feminine; gen, genitive; ind, indicative; ins, instrumental; itr, intransitive; loc, locative; masc, masculine; n, non-; ni, noun incorporation; nmlz, nominalizer; nom, nominative; obj, object; obl, oblique; pass, passive; pl, plural; pni, pseudo noun incorporation; poss, possessive; prs, present; pst, past; ptcp, participle; refl, reflexive; rel.noun, relational noun; subj, subject; tr, transitive.
Chapter 14
Rem arks on th e re l at i on bet w een case -a l i g nme nt an d c onstitue nt orde r Tarald Taraldsen
14.1 Introduction This chapter focuses on a descriptive generalization which claims that no language both has basic SVO order and Erg(ative)/Abs(olutive) case-alignment (sometimes called “Mahajan’s Generalization”). I’ll call this the *SErgVOAbs generalization. I do not intend to offer a theoretical account of this descriptive generalization, which according to Richard Kayne (p.c.) goes back to Schwartz (1972).1 My aim is rather to isolate some empirical issues that bear on the proper formulation of the generalization, and to examine to what extent various theories of case-assignment provide a way of understanding it. Along the way, we will see how different theoretical interpretations enable us to isolate classes of languages that are expected not to fall under the generalization.
14.2 What is the *SErgVOAbs Generalization a Generalization about? There are different ways for an SVO order to arise, e.g. by V-movement into the CP- layer, by V-movement to T or some other functional head inside TP or by not raising 1
Schwartz (1972) does not indicate how he arrived at the generalization. Paumarí (see Aikhenvald 2009) and Zoque (see Faarlund 2012) seem to be the only known exceptions to it. There is a brief comment on how Paumarí and Zoque might be dealt with in section 14.4.5.
The relation between case-alignment and constituent order 333 complements of V in situ (assuming universal H-complement order as in Kayne (1994, 2010)). This is one factor in determining what the empirical coverage of the *SErgVOAbs generalization might be expected to be: Does it apply regardless how SVO order arises in a given language or sentence type, or is it restricted to languages/sentence types where SVO order arises in one specific manner? For each of these possibilities, some theoretical accounts of the generalization will be more plausible than others. Likewise, there may be different ways for Erg/Abs alignment to arise. For example, there may be an empirically motivated distinction between syntactic ergativity and morphological ergativity. If so, does the *SErgVOAbs generalization hold regardless how Erg/Abs alignment arises, or does it hold only when Erg/Abs alignment arises in one of the possible ways? Settling this issue will also impact on the choice of a theoretical account of the generalization. In this section, we look at the different derivational paths that may lead to SVO order as well as the ways Erg/Abs alignment might arise.
14.2.1 What Kind of SVO Might *SErgVOAbs Be a Generalization about? The first issue we will discuss, is what should be meant by SVO in relation to *SErgVOAbs. This linearization pattern may have different structural sources. Does *SErgVOAbs apply to all sentences with SVO order regardless of how the SVO order is created? If that were the case, Kashmiri is a counterexample, since Kashmiri can combine SVO order with Erg/Abs case-alignment in main clauses and a range of embedded finite clauses. The following examples adapted from Koul and Wali (2006) illustrate this, but they also illustrate that Kashmiri is a V2 language: (1)
Aslam-an dits kita:b Mohn-as Aslam-Erg gave.fsg book.fsg. Mohan-Dat ‘Aslam gave Mohan a book yesterday.’
(2)
Ra:th dits Aslam-an kita:b yesterday gave-fsg Aslam-Erg book.fsg ‘Yesterday, Aslam gave Mohan a book.’
ra:th yesterday Mohn-as Mohan-Dat
Hence, we discard the possibility that *SErgVOAbs is about just any SVO order. Since Kashmiri has SOV order non-finite clauses as well as well as in conditionals, alternative questions and relative clauses and, as we have seen, also has XVSO order in main clauses (and all embedded clauses that admit SVO order), it is natural to analyze it as a language with SOV order within TP and “verb second” V-movement, much like German or Dutch. Then, main clauses with SVO order have the V in C or higher, and we can eliminate the counterexample by restricting *SErgVOAbs to sentences with SVO order within TP (at some stage of the derivation).
334 Tarald Taraldsen However, SVO order within TP might arise in a number of different ways. A priori, both V and O might remain in VP both in SVO and SOV languages, in which case VO vs. OV would be determined by the setting of a directionality parameter: (3) a [VP V O ] b [VP O V ] Alternatively, the V remains in VP in both SVO and SOV languages, but the O is raised out of VP only in SOV languages, as in Kayne (1994): (4) a [VP V O ] b O… [VP V ] It is equally plausible, however, that there are SVO languages where an O raises out of the VP, but this is masked by V-movement across the raised O; cf. Johnson (1991), Taraldsen (2000): (5) O … [VP V ] → V … O … [VP ] It may be that the verb moves to T in languages where it ends up preceding sentence adverbs, as in finite clauses in French (Pollock 1989). But in SVO-languages like English and most Mainland Scandinavian varieties, the V must follow all adverbs inside TP, and even in those Mainland Scandinavian varieties that may have the V preceding adverbs inside TP, the V arguably doesn’t raise to T; cf. Bentzen (2006). In principle, there might be as many types of TP-internal V-movement as there are functional heads between C and VP. If all these possible derivations of TP-internal SVO order are actually realized in one language or another, it seems rather unlikely that the *SErgVOAbs generalization is valid for all SVO languages. Therefore, if the generalization does hold for all SVO languages, and we think this is not accidental, we will be led to restricting the set of possible derivations leading to TP-internal SVO order. Conversely, if the generalization turns out not to hold for all SVO languages, we will maintain the full range of possible ways of being SVO TP-internally and restrict the empirical scope of the *SErgVOAbs generalization so that it only should hold for SVO orders derived in certain specific ways.
14.2.2 The Empirical Scope of the Generalization and Theoretical Accounts Suppose the empirical scope of the *SErgVOAbs generalization includes SVO orders with V inside VP. If SOV languages also have both V and O inside VP and OV is determined merely by the setting of a directionality parameter, the account offered by Mahajan (1994) (see section 14.4.2) is the only existing proposal capable of differentiating between
The relation between case-alignment and constituent order 335 SVO and SOV sentences with respect to Erg/Abs alignment, since it is the only account in which only linear order plays a role. If SOV orders only result from moving O out of VP, and the derivation of SVO orders cannot involve moving the O out of VP, theoretical accounts may capitalize on a structural difference between SVO sentences and SOV sentences. There may be different ways that raising the object could interact with case-assignment. For example, it might enable T to assign nominative/absolutive case to the object instead of the subject; cf. the discussion in section 14.5.1. If the scope of the generalization includes TP-internal SVO orders derived by V- movement across O previously extracted from the VP, the raising of the object cannot be the sole decisive factor, but another analytical option becomes available. Raising V might make Erg/Abs alignment impossible. Since VSO-languages may have Erg/Abs alignment, one must then say that only V-movement placing the verb between the S and the O is incompatible with Erg/Abs alignment. This is the basic idea pursued in Taraldsen (2010), where it is developed in such a way that it is immaterial exactly which of the heads within TP the verb raises to, as long as it ends up in a position higher than a previously raised object, but lower than the subject position. In fact, Taraldsen’s account predicts that even an SVO sentence with both the V and O within VP disallows Erg/Abs alignment, since the subject is an intervener.
14.2.3 Ways of Being Ergative Just as there is more than one way to create an SVO order, there may well be more than one way to create an Erg/Abs case-alignment. This, in part, is due to the distinction between syntactic case and its morphological realization. A syntactic Erg/Nom/Acc pattern may surface as an Erg/Abs pattern as a result of Nom/Acc syncretism, a possibility investigated in Legate (2008, 2012a), for example. In addition to this, syntactically determined Erg/Abs alignments may fall into different categories. For example, ergative case may be assigned as a “semantic case” to any DP in Spec-vP in certain languages, while in other languages a DP in Spec-vP is assigned ergative case only if there is also an object. In the first type of ergative language (“active languages”), the subjects of certain intransitive verbs have ergative case, but in the second type of language, the subject of an intransitive verb never has ergative case. For languages of the second kind, both Bittner and Hale (1996a) and Baker (2015) appeal to the notion of “case competition” introduced by Marantz (1991); cf. section 14.4.4. But Bittner and Hale also assume that ergative case is assigned by I, and that an object can become a case-competitor visible to I in two different ways—either by raising the object to Spec-IP or by raising the V to I. In the former case, the language displays properties associated with syntactic ergativity, in Bittner and Hale’s sense, and in the latter case it does not.2 2
But Dyirbal, one of Bittner and Hale’s examples of a syntactically ergative language, is analyzed as syntactically tripartite Erg–Nom–Acc by Legate (2012a).
336 Tarald Taraldsen If surface Erg/Abs alignment can arise in all the different ways as mentioned, it becomes quite difficult to make theoretical sense of a descriptive *SErgVOAbs generalization understood to rule out all surface combinations of SVO order and Erg/Abs alignment. Thus, either there is essentially only one single way of being ergative or *SErgVOAbs should be expected to have counterexamples within some independently identifiable type of ergative languages.
14.2.4 Summary I have looked at different ways SVO orders may be created as well as different ways Erg/ Abs alignment might arise. The conclusion is that either there should be empirical evidence that the *SErgVOAbs generalization fails to hold in a range of well-defined cases, or else there should be only one way of obtaining Erg/Abs case-alignment and perhaps only one way being SVO TP-internally.
14.3 More on the Empirical Coverage of the Generalization The *SErgVOAbs generalization may be thought of as making a claim only about SVO sentences of a certain kind combining with an overtly marked Erg/Abs case-alignment. But its empirical scope might be broadened beyond this, and I will now consider the possibility that the descriptive generalization might extend to tripartite languages and to neutral languages with Erg/Abs agreement patterns.
14.3.1 Tripartite Languages One may wonder whether the *SErgVOAbs generalization holds (if it does) because ergative case cannot be assigned in (certain) SVO-structures, or because the object cannot receive absolutive case in such structures. Information about the so-called tripartite languages would bear on this. In tripartite languages, a transitive subject is assigned ergative case and an intransitive subject is assigned nominative/absolutive, as in languages with Erg/Abs alignment, but an object is assigned accusative. Hence, if *SErgVOAbs really is about the availability of absolutive case for the object, the generalization is consistent with the existence of tripartite SVO-languages, but if the generalization is about ergative case, there should be no such languages. In the WALS sample, there are no tripartite languages with dominant SVO order vs. two tripartite SOV languages. But there are also no tripartite VSO languages (vs. one
The relation between case-alignment and constituent order 337 ergative/absolutive VSO language). Two languages with no dominant order are classified as tripartite. One of these is Nez Perce, although the Nez Perce examples reproduced in Baker (2015) all seem to show SVO order. We also note that if there are tripartite SVO languages, it seems unlikely that *SErgVOAbs can be surface true, since there is no way of excluding a surface Erg/Abs alignment from arising from a Nom/Acc syncretism in an underlyingly tripartite language, and there is little reason to suppose that Nom/Acc syncretism would be sensitive to SVO vs. SOV and VSO. Assuming, for example, the case-theory of Caha (2009), where cases are represented as sets of features standing in a subset relation to each other, the syntax would produce (6) in a language L with underlying tripartite alignment:3 (6) a A-{ X, Y, Z } [VP V O-{ Y, Z } b S-{ Y } [VP V ] But if L has the exponents B ←→ { X, Y, Z } and C ←→ { Y, Z }, but no D ←→ { Z }, (6) will surface as (8), by the Superset Principle: (7)
The Superset Principle A morpheme A with the lexical entry A ←→ F, where F is a feature set, can spell out any feature set F’ which is a subset of F.
(8)
a A-B [VP V O-C ] b S-C [VP V ]
14.3.2 Ergative/Absolutive Agreement Patterns One may also wonder whether the *SErgVOAbs generalization should be expected to hold only for languages with overt case-marking. In particular, it seems conceivable that it might also hold for neutral languages with Erg/Abs agreement patterns. An agreement system is said to follow an Erg/Abs agreement pattern, if it treats S on a par with O to the exclusion of A. I will focus on the situation where some head only agrees with absolutive S and O. A Nom/Acc agreement pattern is one in which A and S are treated the same way to exclusion of O. For example, some head only agrees with A and S. Languages with Ergative/ Absolutive case- alignment can have either Ergative/ Absolutive agreement (three languages in WALS, e.g. Central Yup’ik) or Nominative/ Accusative agreement (12 languages in WALS, e.g. Berber). Importantly, neutral languages are also compatible with both agreement patterns (five languages in WALS
3 To avoid introducing orthogonal issues at this point, I don’t adopt here Caha’s claim that the sets of case features should be replaced with syntactic trees where each case-feature is a head. However, this ingredient of Caha’s proposal will assumed in section 14.4.3.
338 Tarald Taraldsen with Ergative/Absolutive agreement, e.g. Jacaltec (discussed in Woolford 1999), vs. 52 languages with Nominative/Accusative agreement, e.g. Nahuatl), but languages with Nominative/Accusative case-alignment cannot have Ergative/Absolutive agreement (no such languages in WALS). For our purposes, it is important to consider the possibility that agreement patterns are determined by prior case-assignment, i.e. agreement is case-sensitive in the sense of Bobaljik (2008) and Baker (2015). For example, a probe H may be able to agree only with DPs bearing absolutive case, leading to an Erg/Abs agreement pattern in any language with Erg/Abs case-alignment. To allow for the combination of Erg/Abs case-alignment with a Nom/Acc agreement pattern, one may follow Baker in adapting Bobaljik’s (2008) proposal: There is a case-hierarchy … Erg > … Acc > Nom/Abs and that if P can agree with a DP with case K in … Erg > … Acc > Nom, it can agree with a DP with any K’ below K. A language with Erg/Abs case-alignment and a Nom/Acc agreement pattern would then have H able to agree with a DP with ergative case, hence also with DPs with Absolutive case. That is, the existence of (9)a (where the subscript x marks agreement) in a language entails the existence of (9)b in the same language: (9) a Hx … b Hx …
[vP Ax-Erg [VP V O-Abs ]] [vP Sx-Erg [VP V ]]
By Relativized Minimality, H will then agree with both A and S, but never with O (unless O moves before H is merged).4 This type of account excludes the combination of Nom/Acc case-alignment with an Erg/Abs agreement pattern. For H to agree with O, H must be able to agree with accusative DPs, but then it can also agree with nominative DPs, and therefore Relativized Minimality guarantees that a Nom/Acc agreement pattern ensues (again provided the O doesn’t move before H is merged). Unlike accounts where agreement is subject to Relativized Minimality, but is not case- sensitive, this approach also allows one to account for failure of agreement with ergative S in languages where H agrees with O and absolutive S. It is sufficient to stipulate that H only agree with absolutive DPs. However, the descriptive generalization that no language has both Nom/Acc case- alignment and Erg/Abs agreement is based on surface case. Therefore, case-sensitive agreement will not capture the generalization if an underlying Erg/Abs case-alignment can surface as a Nom/Acc alignment. But there seems to be no way this could happen: The exponent associated with both ergative and absolutive case for subjects, will presumably also be associated with absolutive objects, so that the result is a neutral alignment. Things are more delicate if we start out from an underlying tripartite case- alignment. On the approach to case-sensitive agreement taken here, it should be 4
Notice that these assumptions make it impossible to derive a tripartite agreement pattern. I don’t know whether this is a good result.
The relation between case-alignment and constituent order 339 possible to have H agreeing with accusative O and nominative S, but not with ergative A. So, if the same exponent could be chosen both for ergative and nominative case, but not for accusative case, the result would be a surface Nom/Acc case-alignment combined with an Erg/Abs agreement pattern. But this possibility too is ruled out, if we adopt the case-hierarchy as suggested and Caha’s (2009) account of case-syncretism. On this account, each case K in the hierarchy corresponds to set of features properly including the set of features corresponding to any K’ below K in the hierarchy. For example, if the nominative corresponds to { Z }, the accusative corresponds to { Y, Z } and the ergative minimally corresponds to { Z,Y,Z }. A syncretism over the ergative and the nominative arises from the Superset Principle (repeated below) just in case there is an exponent A ←→ { Z, Y, Z }, but no B ←→ { Y, Z } and no C ←→ { Z }(because of the Elsewhere Principle in (10)): (7)
The Superset Principle A morpheme A with the lexical entry A ←→ F, where F is a feature set, can spell out any feature set F’ which is a subset of F.
(10)
The Elsewhere Principle If two morphemes A and B with entries A ←→ F and B ←→ G can both spell out a feature set H by the Superset Principle, but G is a proper subset of F, B must be chosen.
If so, an ergative/nominative syncretism must necessarily also include the accusative. Again, the result is a neutral language. If indeed an Erg/Abs agreement pattern can only arise in languages with an underlying Erg/Abs or Erg/Abs/Acc case alignment, but an underlying Erg/Abs(/Acc) case- alignment may surface as neutral alignment as a result of case-syncretism, the next question is whether the *SErgVOAbs generalization extends to neutral languages with Erg/Abs agreement patterns.5
14.3.3 Summary We have suggested that one should investigate whether some appropriately sharpened version of the *SErgVOAbs generalization which is valid for languages with overtly marked Erg/Abs case-alignment, also holds for tripartite languages and neutral 5
Past participle agreement in Standard Italian might be said to follow an Ergative/Abs (or Acc) agreement pattern, since the participle only agrees with O and unaccusative S. The order of the participle and the object is VO, but the object doesn’t actually trigger agreement unless it moves when the participle also has an external argument. That is, only object clitics and wh-moved objects may trigger agreement in transitive sentences; cf. Kayne (2000a). However, this might be because the ergative subject is a “defective intervener” for agreement, unless the object is raised to a position between the subject and the agreeing probe; cf. section 14.5.3.
340 Tarald Taraldsen languages with Erg/Abs agreement patterns. If the generalization does hold for tripartite languages, we conclude that the generalization really is to be understood as *SErgVO, but if it doesn’t, the generalization must be *SVOAbs. If the generalization is valid for neutral languages with Erg/Abs agreement patterns, we conclude that the generalization is really about syntactic case, as one would be inclined to suspect to begin with. However, although there are no neutral SVO languages with Ergative/ Absolutive agreement in the WALS sample, there are 3 neutral SVO languages with active agreement (Taba, Mountain Arapesh, Apurinâ).6 If the active agreement pattern reflects underlying Active/Inactive (or “Split-S”) case-alignment in these three languages, one may conjecture that *SErgVOAbs doesn’t hold for active languages, plausibly because of the way the ergative is assigned in such languages. But then, the fact that there are no SVO languages with overt Active/Inactive case-alignment in WALS, comes as a surprise.
14.4 *SErgVOAbs = *SErgVO? If a suitably precise version of the *SErgVOAbs generalization holds, one will ask why. We will assume that any interesting explanation will be based on assumptions regarding syntactic case rather than its morphological realization.7 If so, the generalization may hold either because the subject cannot have syntactic ergative case in certain structures linearizing as SVO (*SErgVO), or because the object cannot bear syntactic absolutive case in such structures (*SVOAbs) (or both). As mentioned earlier, information about tripartite languages should be relevant to the choice between these options. If SVO order is merely incompatible with assignment of syntactic absolutive case to the object, e.g. by T, there should be tripartite SVO languages (as well as SVO languages with surface Erg/Abs alignment arising from underlying Erg/ Nom/Acc by syncretism). But if SVO order is incompatible with the subject surfacing with ergative case, there should be no tripartite SVO languages (and no SVO languages with surface Erg/Abs alignment arising from underlying Erg/Nom/Acc by syncretism). At present, however, no conclusion can be drawn from this, since the data available to me seems insufficient. In this section, we will look at the major proposals regarding the assignment of syntactic ergative case and evaluate them with respect to the prospects of providing an account of *SErgVOAbs as *SErgVO.
6 Notice that participle agreement in Standard Italian also follows an Active/Inactive agreement pattern, since there is no agreement with the S unergative verbs. 7 Woolford’s (1997) rebuttal, endorsed by Legate (2012a), is based on the observation that datives too may exhibit a transitivity restriction, but her argument gratuitously presupposes that the dative is necessarily an inherent case rather than a dependent case requiring a case-competitor; cf. Baker (2015).
The relation between case-alignment and constituent order 341
14.4.1 Ergative Case Assigned by v Ergative case may plausibly be taken to be assigned by (certain subspecies of) “little v” to its Specifier; cf. Woolford (1997, 2006); Anand and Nevins (2006); Legate (2012a); among others. That is, the ergative is an “inherent case” in the sense that it is assigned together with a specific Θ-role. The main argument in favor of this view comes from the existence of languages in which ergative case is associated with every subject arguably born in Spec-vP, but never to derived subjects (in passives and with unaccusative verbs). On the other hand, this proposal is less well suited to account for languages in which only the subjects of transitive verbs can have ergative case.8 I also note that there may be counterexamples to the generalization that derived subjects cannot have ergative case, at least if Erg/Abs agreement requires a syntactic Erg/Abs case-alignment. Holmberg and Odden (2004) show that the Kurdish language Hawrami has Erg/Abs agreement in passivized double object constructions (but not in passivized monotransitive sentences), although the derived subject does not preserve dative marking. Likewise, Rezac et al. (2014) argue for raising-to-ergative in Basque and conclude that the ergative is not inherent in Basque. Conceivably, ergative case is assigned in different ways in different languages. Here, I will simply discuss how *SErgVO might be accounted for if ergative case is assigned by v. It seems highly unlikely that the properties of structures linearizing as SVO could interfere with assignment of ergative case by v to its Specifier. However, it is possible that such structures might force previously assigned ergative case to be canceled in ways I will now sketch.
14.4.2 Mahajan (1994) The basic idea developed in Mahajan (1994) is inspired by Kayne’s (1993) account of the have/be alternation with past participles in Romance. Kayne takes have to be the spell- out of BE with a preposition originating in C incorporated into it: (11)
[AuxP BE [CP P [IP DP [IP I … → [AuxP P+BE [CP P [IP DP [IP I …
Kayne’s system is set up so as to require a complementizer P only with transitive and unergative participles. Since this is reminiscent of a certain type of Erg/Abs alignment, Mahajan suggests extending Kayne’s analysis to Erg/Abs case-alignment in languages like Hindi by saying that P doesn’t incorporate into the auxiliary in such languages, and that an unincorporated P assigns ergative case to the subject. Adding that P universally incorporates into a left-adjacent auxiliary, he derives *SErgAuxVO. 8
Mahajan’s account is tailor-made for languages where Erg/Abs alignment only occurs with perfective verbs, and perfective verbs combine with auxiliaries, as in Hindi and Romance.
342 Tarald Taraldsen Since *SErgAuxVO does not entail *SErgVO, a further step must be taken if *SErgVO is indeed the right descriptive generalization.9 For example, one might say that the head X hosting the auxiliary in the compound tenses is also present in simple tenses, and that the verb adjoins to it, when there is no auxiliary, assuming that P+X is then spelled out by the verb. To capture the dependence of the P on the participle being transitive or unergative, one may replace (11) with (12) taking the P to be selected by v: (12) [AuxP BE [vP [PP P DP] [vP v … → [AuxP P+BE [vP [PP P DP] [vP v … But this seems equivalent to viewing P as a case-head K (= “ergative”) licensed by v. In this sense, Mahajan’s analysis provides a mechanism for depriving a subject of an ergative case-feature previously assigned to it. A potential weakness in Mahajan’s account is its reliance on linear adjacency, which is now generally supposed not to play a role in syntax. But even if linear adjacency does play a role in syntax (see Kayne 2010), there is a problem, if the merger of a head with a phrasal complement universally results in H-compl order (Kayne 1994, 2010). Then, an auxiliary will always be left-adjacent to the ergative P/K at the point where the auxiliary merges with a participial phrase. Since P-incorporation into a left-adjacent auxiliary must be obligatory in Mahajan’s system, there should therefore be no ergative languages, if P-incorporation applies at that point. Rather, P-incorporation must apply at a later stage of the derivation to give the participial phrase time to slip across the participle. But the timing is rather delicate. Even though the participle regularly precedes the auxiliary in German, the German have/be alternation seems rather similar to the one seen in Romance, which, from Mahajan’s perspective, must mean that P-incorporation must apply before the participial phrase is moved across the auxiliary in German. In other words, the movement of the participial complement of an auxiliary in German must be to a higher position than in ergative SOV languages like Hindi. The question is whether there is any independent evidence that this is in fact true. Another issue is why P-incorporation into a left-adjacent auxiliary should be universally obligatory.10 For that matter, one might also wonder whether the disappearance of ergative case can be blamed on P/K-incorporation into the auxiliary at all, since Basque seems to have a have/be alternation, but the subject carries an ergative case-affix even in the presence of have. 9 Notice that if interpreted this way, Mahajan’s account also provides a way of making an ergative surface as a nominative without appealing to syncretism. Hence, languages combining surface Nom/Acc alignment (rather than neutral) with an Erg/Abs agreement pattern depending on underlying Erg/Abs/ Acc case-alignment will not be ruled out by the assumptions about syncretism invoked in section 14.3.2. But past participle agreement in Romance may in fact be an instance of an Erg/Abs agreement pattern in languages with overt Nom/Acc case-alignment at least for pronouns. 10 Kayne (1993) suggests that P-incorporation must apply in languages which lack a spell-out for the oblique case that would otherwise be assigned by the P, i.e. the ergative on Mahajan’s interpretation. But if this were to be the only thing driving P-incorporation, no language with the morphological means to
The relation between case-alignment and constituent order 343
14.4.3 Taraldsen (2010) The approach taken by Taraldsen (2010) also assumes that a nominative DP may emerge in the syntax from a DP previously assigned ergative case in Spec-vP. But instead of becoming a nominative by losing its ergative K to incorporation, it strands its ergative case-layer by subextraction, a process dubbed “peeling” in Caha (2009): (13)
[NomP Nom … [vP [ErgP X [AccP Y [NomP Z DP ]]] [vP v … → [NomP [NomP Z DP] [NomP Nom … [vP [ErgP X [AccP Y [NomP Z DP ]]] [vP v …
This is based on an assumption already mentioned: The different cases are decomposed into sets of privative features linearly ordered by the subset relation. In addition, Taraldsen (2010) also follows Caha by assuming that each case-feature is a syntactic head so that sets of case-features are represented as trees. In (13), Nom is a head on the verbal spine attracting a NomP to its Specifier from the nearest DP. It is also assumed that an object always raises to an Acc-position between Nom and vP except in anti-passives, where the object surfaces inside VP with the oblique case initially assigned to it; see Medová (2009): (14)
[NomP Nom … [AccP Y [NomP Z DPO]] … [vP [ErgP X [AccP Y [NomP Z DPA ]]] [vP v …
If nothing further happens, the object will therefore be the DP closest to Nom, and an Erg/Abs alignment emerges with the object’s nominative layer in Spec-NomP: (15)
[NomP [NomP Z DPO ] [NomP Nom … [AccP Y [NomP Z DPO]] … [vP [ErgP X [AccP Y [NomP Z DPA ]]] [vP v …
But SVO within TP is taken to arise only from moving the remnant vP in between Nom and the previously raised object for reasons discussed in Taraldsen (2010): (16)
[NomP Nom …[vP [ErgP X [AccP Y [NomP Z DPA ]]] [v’ v … [AccP Y [NomP Z DPO]] … [vP [ErgP X [AccP Y [NomP Z DPA ]]] [vP v …
In (16), the nominative layer of the external argument is assumed to be closest to Nom, and by locality Nom/Acc alignment must emerge.11,12 Hence, no language with spell out ergative case would have P-incorporation and might therefore combine Erg/Abs alignment with SVO order. 11 It is not actually clear that NomP contained in the subject is closer to Nom than the object’s NomP, since the former is embedded under more case-heads than the latter. 12 Notice that this raises the question how Icelandic, an SVO-language, can have quirky subjects and nominative objects.
344 Tarald Taraldsen TP-internal SVO order can have Erg/Abs case-alignment. That is, *SVO+Erg follows even if v universally assigns ergative case to its Specifier.13 The most obvious problem for this account is perhaps that it predicts that the object should be higher than the external argument in all ergative languages, as in Bittner and Hale’s (1996a) account of Dyirbal and Inuit, but this does not seem correct. If the classical diagnostics for syntactic ergativity respond to structural properties of the sort assumed for ergative sentences by Taraldsen, not all ergative sentences can have these structural properties, since not all ergative languages display the hallmarks of syntactic ergativity.14 Notice also that while Taraldsen’s account makes the right prediction for case- alignment in those ergative languages in which only the subjects of transitive verbs have ergative case, it also incorrectly predicts that there should be no language in which both subjects of transitive verbs and subjects of unergative intransitive verbs have ergative case, unless unergative verbs come with a hidden object as assumed by Bittner and Hale (1996a) for Georgian. In this respect, it is similar to the case-competition accounts to which we now turn.15
14.4.4 Case-Competition The case-competition approach assumes that certain cases can only be assigned to a case-less DPi in a case-domain XP if XP also contains another case-less DPj, a “case- competitor.” A case of this sort is usually referred to as a “dependent case.” If DPi asymmetrically c-commands DPj, some languages will assign the dependent case to DPi, while others assign it to DPj.16 To illustrate, let us assume that vP is a case-domain. If a vP contains only a single case- less DP, as in (17), dependent case is not assigned and the DP must be case-licensed in one of the alternative ways sketched in section 14.5: (17) [vP DP [vP v [VP V ]]] But if vP contains a case-less object DP in addition to the external argument, as in (18), dependent case will be assigned either to the external argument or to the object depending on which way the language sets the relevant parameter: 13
Notice that on this account, case-sensitive agreement will be able to produce Erg/Abs agreement in a Nom/Acc language if the probe is sufficiently low, as it might be in Romance past participle agreement. 14 Aikhenvald’s (2009) observation that the “pivot” for coordination is always S or O, never A, in Paumarí regardless of whether the case-alignment is Erg/Abs or Nom/Acc, suggests that case-alignment is in fact quite independent of the properties underlying syntactic ergativity. 15 A related problem is that Taraldsen’s analysis cannot easily accommodate the tripartite languages. 16 Notice that this seems to go against the spirit of Kayne’s (1994) antisymmetry proposal, since saying that the dependent case go to either A or B when A asymmetrically c-commands B, subject to a language-specific choice, is similar to saying that A asymmetrically c-commands B can map onto A precedes B or A follows B, as dictated by a language-specific parameter setting.
The relation between case-alignment and constituent order 345 (18)
[vP DPi [vP v [VP DPj V]]]
Assume that the DP that remains case-less at the end of the vP-cycle, will eventually be assigned the same case as the case-less external argument in (17). Then, if dependent case is assigned to the external argument in (18), the language will have Erg/Abs case-alignment. If dependent case is assigned to the object DP, we will have Nom/Acc alignment.17 If the dependent case can be assigned both to the higher and the lower DP at the same time, as suggested by Baker (2015), a tripartite language emerges. This analysis comes with the transitivity condition built in. It predicts not only that an object will not have accusative case in the absence of an external argument, but also that an external argument has ergative case only in the presence of a case-less object.18 Therefore, this line of analysis has no problem with ergative languages where only the subject of a transitive verb can be ergative, but must posit covert objects for languages in which all external arguments bear ergative case. On the conceptual side, one may wonder where exactly the procedure for assignment of dependent case and the parameter associated with it resides in the grammar. The proposal in Taraldsen (2010) can be viewed as an attempt to answer these questions, i.e. as a specific implementation of the case-competition view. A different answer is provided by Bittner and Hale (1996a) who assume that ergative case is assigned by I while accusative is assigned by V. Given that a case-competitor must be case-less, it follows from this that ergative case will only be assigned in sentences where the verb fails to assign accusative case to the object. Hence, whether Erg/Abs or Nom/Acc alignment emerges, ultimately depends on a property of verbs.19 If the domain for assignment of dependent case is vP, the dependent case approach offers limited promise of understanding a *SVO+Erg generalization. Whether the dependent case is assigned to the higher or to lower one of two case-less DPs inside vP, should be totally independent of how the ingredients of the vP end up being linearized with respect to each other. Only analyses in which ergative case previously assigned can be lost, as illustrated, might do the trick. If the relevant case-domain is I/TP, however, additional options become available. For example, Bittner and Hale (1996a) assume that ergative case is assigned by I, and that a case-less object becomes a case-competitor enabling I to assign ergative case to the subject when either the object raises to Spec-IP or the verb raises to I. In the first case, the subject cannot also raise to Spec-IP so that the direct outcome within IP is an OSV
17
In Taraldsen’s (2010) analysis, the alignment type depends on which of the two DPs succeeds in reaching the Nom position. The effect of requiring a case-competitor comes from the fact that when there is a single DP, that DP will always reach the Nom position. 18 The promoted first object in a passivized double object construction is also expected to have ergative case, as in Hawrami, if both objects are case-less at the stage of the derivation where dependent case is assigned. 19 On this approach, too, tripartite languages call for additional assumptions, as in section 11 of Bittner and Hale (1996a).
346 Tarald Taraldsen order. In the second case, the subject doesn’t raise to Spec-IP, because it is case-licensed in situ and movement to Spec-IP applies only when needed for case-licensing. In this case, the outcome is a VSO order. Either way, underlying SVO order within VP will not be preserved at the IP-level. In a language with Nom/Acc alignment, on the other hand, the subject is case-less and may move to Spec-IP to be governed by C (depending on whether V has raised to I or not). Therefore, a sentence with Nom/Acc alignment may surface with SVO order, assuming the order VO within VP. Notice, however, that in order to accommodate ergative languages with rigid SOV, Bittner and Hale’s account must acknowledge the possibility that a subject may raise to a position above Spec-IP for reasons not related to case-licensing. This movement will derive SOV orders from OSV (within IP), but must still be unable to derive SVO orders from VSO (within IP). Therefore, even Bittner and Hale’s version of the case- competition approach needs extra assumptions to capture the *SErgVOAbs generalization, unless, of course, it turns out that SVO sentences in fact allow Erg/Abs alignment precisely when it can be shown that S is in the same position above Spec-IP where it must be placed in ergative SOV languages.
14.4.5 Kayne (2000b) Kayne (2000b) has a different way of connecting V-movement to ergative case-assignment as part of an attempt to understand *SErgVO in terms of general properties of adpositions. The basic axiom is that adpositions are functional heads on the verbal spine and therefore combine with what we take to be their complements via movement. The first step in a derivation putting together the sequence to John, for instance, would be as in (19): (19) [PP to [vP I [VP spoke John ]]] → [PP John [PP to [vP I [VP spoke John ]]]] Then, a head W is merged, and the P incorporates into W yielding: (20) [WP to+W [PP John [PP to [vP I [VP spoke John ]]]]] Finally, P+W attracts vP: (21) [WP [vP I [VP spoke John ]] [WP to+W [PP John [PP to [vP I [VP spoke John ]]]]] If W is not merged, the adposition surfaces as a postposition, and to capture the preference for postpositions over prepositions in verb-final languages, Kayne proposes that P+W universally attracts vP.20 20
An SXV language can then have prepositions just in case the V is extracted from vP, before vP is raised to Spec-WP.
The relation between case-alignment and constituent order 347 Suppose now that ergative case is assigned by a special P to a DP in its Spec. The derivation of a sentence with an ergative subject must then include a step similar to (19): (22) [PP P … [vP DPA [VP V DPO ]]] → [PP DPA [PP P … [vP DPA [VP V DPO ]]]] In a language where all Ps are prepositions, the derivation continues as in (20)–(21): (23)
→ [WP P+W [PP DPA [PP P … [vP DPA [VP V DPO ]]]] → [WP[vP DPA [VP V DPO ]] [WP P+W [PP DPA [PP P … [vP DPa [VP V DPO ]]
Since the last step produces a V-initial order (provided, of course, the subject does not subsequently move across the raised vP), *SVO+Erg follows for any language with prepositions rather than postpositions. In particular, it follows for all SVO languages, if adpositions are always prepositions in SVO languages.21 Notice, however, that nothing excludes the existence of V-initial sentences with ergative subjects in languages that otherwise have SVX order. If such mixed languages don’t exist, additional assumptions are called for. Notice also that there are a few SVO languages with postpositions. Hence, there might also be a few ergative SVO languages, i.e. *SErgVO should find exceptions among those SVO languages that have postpositions. In the exceptional languages, Kayne’s ergative P should be a postposition. This expectation is in fact met in the two exceptional languages I know of, viz. Paumarí and Zoque. Both languages have SVO-sentences with Erg/Abs alignment, but the ergative case-marker is enclitic to the noun. Judging from Faarlund (2012), all case-markers are in fact postpositional in Zoque. The following example from Paumarí comes from Aikhenvald (2009:113) (Tr = transitivity marker, Th = object agreement): (24) Kodi-jomahi-a bi-a-vi-kha-‘a-ha ada jao’oro 1sg.-dog(m)-Erg 3sgTr-get-Tr-get-Asp-Th.m Dem.m-agouti (m) ‘Our dog caught an agouti.’ Faarlund (2012:120) provides the following Zoque example: (25)
21
Te’ n-galyo= ‘is y-ni-tüp-‘üy-u te’ wedu DET M-cock= Erg 3A-front-jump-V-CP DET fox ‘The cock attacked the fox.’ (3A = 3rd person series A prefix, V = verbalizer, CP = completive aspect)
The proposal in Taraldsen (2010) might be recast in similar terms by saying that accusative case is associated with a preposition above vP in SVO-languages.
348 Tarald Taraldsen
14.4.6 Summary We have examined different proposals about the licensing of ergative case in order to determine whether they provide any hope of seeing the *SErgVOAbs generalization as reflecting the impossibility of assigning ergative case in structures that give rise to SVO order. We have seen that if ergative case is assigned vP-internally, either as a “semantic case” or as a dependent case, only analyses where a subject is deprived of previously assigned ergative case in the course of derivations leading to SVO order, have any reasonable chance of explaining the generalization as the effect of *SErgVO. But the two existing accounts that pursue this line of analysis have both been shown to be defective or incomplete. If ergative case is assigned by I, as in Bittner and Hale’s account, there is a better chance of reducing *SVO+Erg/Abs to *SVO+Erg, and in fact Bittner and Hale claim to have achieved exactly that, but we have also seen that their account too is incomplete in ways that threaten to undercut their conclusion. Finally, we have looked at Kayne’s (2000b) proposal, which makes a very specific prediction about the scope of *SErgVO—a prediction which in fact is consistent with the two known exceptions to the generalization.
14.5 *SErgVOAbs = *SVOAbs? We will now look at the major proposals for absolutive case with a view to determining whether they supply a plausible starting point for attributing *SErgVOAbs to failure of absolutive case-licensing for the object in structures that give rise to SVO order, i.e. for seeing *SErgVOAbs as *SVOAbs.
14.5.1 Abs Assigned by T In the analysis proposed by Marantz (1991), the absolutive is automatically available as a default for any DP to which inherent or dependent case has not been assigned. In this scenario, it is hard to see how the positioning of the verb might interfere with the licensing of absolutive case. If, on the other hand, absolutive case is assigned by T (like nominative case on standard accounts), locality may come into play. As mentioned in section 14.2.2, moving the object across the subject, might be a precondition for the object to be probed by T, and if the verb cannot be moved to a position between T and the raised object and the subject winds up in Spec-TP, *SVOAbs would follow. Even if the verb can move across a raised object, but only by remnant movement of vP carrying along the subject, as in Taraldsen (2010), *SVOAbs would follow provided this turns the subject into an intervener between T and the object.
The relation between case-alignment and constituent order 349 In addition to Relativized Minimality, phasehood might be relevant. This possibility is exploited by Lahne (2008). Taking the standard view that vP is a phase, she points out that if the PIC takes effect as soon as the vP-phase has been completed, T will not be able to value the object’s case-feature unless the object has moved to the vP-edge. If so, the object must always become an “inner Specifier” of v in ergative languages. This in turn means that there is no head position between the subject and the object which the verb could move to, provided the subject doesn’t move out of vP. Hence, *SVOAbs. But there is a problem with this. Icelandic is SVO, and therefore the S and the O cannot both be specifiers of vP at the end of the day. But in sentences with quirky subjects, an object has nominative case. If the nominative is valued by T, this should be impossible on Lahne’s assumptions. More generally, if the absolutive case and the nominative case are in fact the same case and this case is always licensed in the same fashion, Icelandic sentences with SVO order and nominative case on the O show that there is no hope of deriving *SErgVOAbs from *SVOAbs.
14.5.2 Nominative Is Independent of Agreement in Icelandic Actually, agreement facts suggest that the Icelandic nominative is best analyzed as a default case. In Icelandic sentences with quirky subjects, the verb may show overt number agreement with the nominative object, as in (26): (26) Mér líku∂-u me.Dat liked-3pl ‘I liked the horses.’
hestarnir horses-Nom-the-Nom.m.pl
But in some cases there is no agreement even though the object has nominative case.22 (27)
a Honum líka∂-i him.Dat liked-3sg ‘He liked us.’
vi∂ we.Nom
b Mér fannst hestarnir vera venir me.Dat seemed.3sg horses- the.Nom be beautiful- Nom.pl ‘I found the horses beautiful.’ And as shown in Holmberg and Hróarsdóttir (2000), a postverbal quirky subject blocks agreement with the nominative object: (28) †a∂ fannst/ *funnust einhverjum student tölvurnar ljótar It seemed.3sg/*3pl some-Dat student computers-the.Nom ugly- Nom.f.pl ‘Some student found the computers ugly.’ 22
Sentences like (27)a where the nominative object is not 3rd person are unacceptable to some speakers. But they become unacceptable to all speakers with plural agreement on the verb.
350 Tarald Taraldsen This strongly suggests that nominative case is not licensed under Agree with T in Icelandic, while it might perfectly well be analyzed as the default case. But if nominative case and absolutive case are the same thing, as is often assumed, then one would also expect that the absolutive is a default case in some language with Erg/Abs case- alignment. If we can identify such languages and find that the *SErgVOAbs generalization doesn’t hold for them, while it does hold for languages in which there is independent evidence that absolutive case is licensed by T, we should conclude that the generalization really is *SVOAbs after all. Conversely, if the *SErgVOAbs generalization is shown to hold also for languages with the absolutive as a default case, the conclusion must be that the correct generalization is *SErgVO.
14.5.3 Defective Intervention In (28), the dative DP acts as a “defective intervener” with respect to agreement with the nominative object. The probe is case-sensitive and cannot agree with datives, but nevertheless the intervening dative prevents the probe from reaching the nominative. The phenomenon of defective intervention is also relevant to the analysis of sentences with Erg/Abs alignment. To account for an Erg/Abs agreement pattern with agreement only with absolutive DPs, it is not sufficient to say that the probe is case-sensitive, once defective intervention is taken into account. Unless an absolutive object is raised to a position P above the subject before the probe X is merged, the subject will be a defective intervener, as in (28). The object may then also be assigned absolutive case by X or a head higher than X without the subject acting as an intervener. (If ergative case is assigned by a probe, this probe must be v or a head located between v and the raised object.) A Nom/Acc agreement pattern, however, will only arise if the object does not raise to P. But then absolutive case cannot be assigned to it by any head outside vP, since the ergative subject would act as an intervener. This suggests that absolutive case may be licensed by a vP-external head in ergative languages with agreement only with O and absolutive S, but is a default case in ergative languages with agreement only with A and S, just like the Icelandic nominative. Consequently, if *SErgVOAbs is to be understood as *SVOAbs, we may expect that the generalization doesn’t hold for ergative languages with Nom/Acc agreement patterns, since it seems impossible to make sense of *SVOAbs when the absolutive is a default case.
14.5.4 Summary The *SErgVOAbs generalization can be understood as *SVOAbs only if absolutive case must be licensed by a head outside vP, e.g. T. But even so, *SVOAbs seems implausible, if nominative case is also always licensed by T, since Icelandic has SVO sentences with nominative subjects. However, intervention facts suggest that the Icelandic nominative
The relation between case-alignment and constituent order 351 is really a default case, and therefore the possibility remains that *SVOAbs holds to the extent that the absolutive is licensed by T. But it seems likely that there are ergative languages in which the absolutive is a default case, just like the Icelandic nominative, and *SVOAbs is not expected to hold for such languages. If *SErgVOAbs nevertheless turns out to hold for all ergative languages, we conclude that *SErgVOAbs is best understood as *SErgVO, but if it holds only for those ergative languages where the absolutive is arguably not a default, *SErgVOAbs is really *SVOAbs. We have also suggested that the absolutive may be licensed by T in ergative languages with Erg/Abs agreement, but must be a default in ergative languages with Nom/Acc agreement.
14.6 Conclusion One conclusion is that if there are different ways for Erg/Abs alignment to arise as well as different ways for a sentence to come out with SVO order, one should expect the *SErgVOAbs generalization to have exceptions, hopefully of a well-defined kind. But if the generalization is valid for all sentences with Erg/Abs alignment, we are led to conclude that Erg/Abs alignment must arise in essentially the same way in all languages. We have also suggested that tripartite languages and neutral languages with Erg/Abs agreement patterns should be investigated with a view to determining whether such languages should fall under the scope of the *SErgVOAbs generalization. In particular, determining whether tripartite languages fall under the generalization, is crucial to deciding whether the generalization should be seen as *SErgVO or *SVOAbs. In the final sections of the chapter, I have tried to evaluate the major theoretical accounts of ergative and absolutive case with respect to how well they allow us to ground a suitably sharpened version of the *SErgVOAbs generalization in general principles of syntax.
Abbreviations 3A, third person series A prefix; Abs, absolutive; asp, aspect; CP, completive aspect; Dat, dative; Dem, demonstrative; DET, determiner; Erg, ergative; f, feminine; m, masculine; Nom, nominative; pl, plural; sg, singular; Th, object agreement; Tr, transitivity marker; V, Verbalizer.
Extensions
Chapter 15
Ergativi t y i n nom inaliz at i on Artemis Alexiadou
15.1 Introduction The correlation between ergativity and deverbal nominalization crosslinguistically has been discussed extensively in e.g. Alexiadou (2001), Salanova (2007), Imanishi (2014), among others; and see Williams (1987), Bok-Bennema (1991), and Johns (1992) for earlier similar ideas. According to these authors, nominalizations across languages show an ergative case-marking pattern, which is attributed to the presence of a defective v (or Voice) in their syntactic representation. In this chapter, after briefly summarizing the main arguments in favor of this view, I will address the following question: what is it about noun-ness that obligatorily triggers the presence of an ergative case pattern?
15.2 Deverbal Nominalizations Have Ergative Case Patterns Alexiadou (2001) explores the hypothesis that certain types of nominalizations exhibit ergative case patterns. This claim has been made on the basis of data from a variety of languages, which realize the theme argument of the nominalization in the genitive. Consider the Greek examples in (1). In (1a), the Agent is introduced by a prepositional phrase. This PP has the same form as the one we find with verbal passives (1b). (1) a. i katastrofi tis polis apo tus varvarus mesa se tris meres the destruction the city.GEN by the barbarians within three days ‘The destruction of the city by the barbarians within three days’
356 Artemis Alexiadou b. i poli katastrafike apo tus varvarus. the city.NOM destroyed.PASS by the barbarians ‘The city was destroyed by the barbarians.’ Irrespectively of the presence of the PP, the genitive argument bears the theme/patient theta-role. What is excluded is an agent bearing genitive case and co-occurring with a genitive theme, no matter in which order the two genitives appear (see Horrocks & Stavrou 1987): (2) a. *i silipsi the capture
tu Jani the John.GEN
tis astinomias the police.GEN
b. *tis astinomias i silipsi tu Jani In case of nominalizations derived from intransitive verbs, both unaccusative and unergative ones, their single argument surfaces with genitive:1 (3) a. i afiksi ton pedion the arrive the children.GEN b. to the
treksimo running
tu the
athliti athlete.GEN
It appears then that the patterns in (1) and (3) are reminiscent of an ergative case pattern, in which the single argument of an intransitive predicate surfaces with the same case as the internal argument of a transitive predicate, and the agent argument bears a different marking. Similar patterns are found in other languages as well, as can be seen by the following set of examples. For instance, in English internal arguments of nominalizations derived from transitive predicates as well as the single argument of intransitive nominalizations 1 Note that Grimshaw (1990) argued that nominalizations of type (3b) should not be considered as complex event nominals in her terms or argument supporting nominals in e.g. Alexiadou, Iordachioaia, & Soare’s (2010) or Borer’s (2013) terminology. In fact, in Alexiadou (2001), I provided ample evidence that unergative nominalizations should not be treated as argument supporting nominals, suggesting that the genitive case we see in these nominals is similar to that of the possessor in non-derived nominals (e.g. John’s book). If this is correct, nothing special needs to be said with respect to (3b) in the sense that the genitive is introduced in the nominal part of the structure, i.e. presumably by the n head that embeds a verbal substructure, and receives case in D, see also Imanishi (2014). The alternative would be to assume that data such as the ones in (3) suggest that nominalizations have case patterns similar to those ergative languages, such as Niuean, in which subjects of both unergative and unaccusative predicates are marked with absolutive. Later on in the chapter, I will discuss the requirement that nominalizations lack an external argument, see (39). In order for (3b) to conform with this requirement, one would need to assume that, in agreement with Massam (2009), subjects of unergatives are introduced within vP, at least in nominalizations. I will come back to this point in the last section of the chapter.
Ergativity in nominalization 357 are introduced by of and the agent in the case of the deverbal nominalization of a transitive predicate is introduced via a by-phrase: (4)
a. the destruction of the city by the barbarians b. the arrival of the policemen c. the jumping of the cow
What is not allowed is that both arguments are introduced via of:2 (5)
*the destruction of the city of the barbarians
Picallo (1991) reports that a similar pattern is found in nominalizations in Catalan, where it is not possible to realize both arguments of the nominalization in the genitive, as is illustrated by the unacceptability of (6). (6) *l’afusellament de l’escamot d’en Ferrer Guardia the execution of the squad of Ferrer Guardia ‘the squad’s execution of Ferrer Guardia’ Genitive arguments are always interpreted as themes in such cases. Thus, the following examples are grammatical, but only in the interpretation where the squad and the police are executed or captured respectively. The agent, when realized, must be realized as a PP (see 7c): (7)
a. l’afusellament de l’escamot the execution of the squad b. la captura de la policia the capture of the police c. l’afusellament d’en Ferrer Guardia per part de l’escamot the execution of Ferrer Guardia on part of the squad
Theme/*Agent Theme/*Agent
2 There is, however, an important difference between Greek and English nominalizations: the agent argument in English nominalizations derived from transitive verbs can also appear in the Saxon genitive, a pattern that is often referred to as the transitive variant of the nominalization (see Chomsky 1970, Kayne 1984, among others; cf. Koptjevskaja-Tamm (1993) who labels this pattern double possessive):
(i) a. John’s destruction of the city b. the judge’s reversal of the decision In this, Modern Greek differs from Classical Greek, which allowed double possessive nominalizations, see (ii), from Manolessou (2000): (ii) hê Phaiakôn proenoikêsis the.NOM Phaecians.GEN occupation.NOM the Phaeacians’ occupation of Corcyra T. 1.25.
tês Kerkuras the.GEN Corcyra.GEN
I will briefly discuss this pattern in section 15.5. Note that even in English there are certain restrictions on the formation of transitive nominalizations related to the type of external arguments these permit, see Alexiadou et al. (2013) for discussion and references.
358 Artemis Alexiadou In French, Spanish, and Italian (see Cinque 1980; Milner 1982; Zubizarreta 1987; Bottari 1992; among others) two ‘de-phrases’, cannot appear within nominalizations, and the agent argument is introduced by par, por and de parte de (8). (8) a. *la destruction de la ville des soldats the destruction of the city of the soldiers
French
b. la destruction de la ville par les soldats the destruction of the city by the solders c. *la captura de los fugitivos de los soldados the capture of the fugitives of the soldiers
Spanish
d. La captura de los fugitivos por los soldados the capture of the fugitives by the soldiers e. *La cattura del soldato del enemico the capture of the soldier of the enemy
Italian
f. La cattura del soldato da parte del enemico the capture of the soldier by the enemy As was the case in Greek and English, the agent is introduced by a PP which has a similar form to the PP found in passive sentences introducing the agent. In fact, the ergative pattern in nominalization is really pervasive. Salanova (2007) shows that even if a language permits ergative splits in the verbal domain, the nominal domain is obligatorily ergative. As shown below, Mẽbengokre event nominalizations, which are employed in a variety of constructions in the language, display an ergative– absolutive pattern, examples from Salanova (2007): (9) a. b^ kam i-mõr kuni forest in 1-go.PL.N all “all my goings into the woods” b. ijɛ ^ktirɛ krõr jã 1ERG hawk.people make.peace.N this “this (occasion in which) I was making peace with the Àktire” In the case of Mẽbengokre, contrary to what happens in English and in the other languages discussed here, there is morphological identity between the cases employed in nominalizations and those employed in ergative main clauses. Salanova stresses that ‘an important fact to note is that, contrary to other ergativity splits, which languages may or may not have, action nominalizations are normally ergative.’ The cross-linguistic observations with respect to ergativity in nominalization are summarized in Table 15.1, from Alexiadou (2001: 166).
Ergativity in nominalization 359 Table 15.1 Cross-linguistic observations and ergativity in nominalization N/A system
E/A system
Nominalization
A-argument
NOM
ERG/PP
S-argument
NOM ABS
GEN
P-argument
ACC ABS
GEN
This ties in nicely with observations made in other literature, where it is suggested that the ergative syntax observed in most e.g. Philippine and Formosan languages is the result of diachronic reanalysis of clausal nominalizations as finite verbal clauses, see e.g. Aldridge (2013a, Chapter 21, this volume), Kaufmann (Chapter 24, this volume) for recent discussion and references. Hence there must be something deeper explaining the connection between nominalization and ergativity. Before I raise the questions to be addressed in this chapter, three remarks are in order. First, not all types of nominalizations exhibit ergative case patterns, as we will see, and see Koptjevskaja-Tamm (1993) for a typological survey. Thus, ergativity characterizes only a subset of nominalizations, namely those that show a pattern of the type in Table 15.1. Second, it should be pointed out that in e.g. Mayan languages ergative is homophonous with genitive. In a subgroup of Mayan languages all subjects align with ergative/genitive. This is the case for example in Chol/Q’anjob’al, as discussed in Coon (2010b), but not in Kaqchikel, as discussed in Imanishi (2014), which has an alignment more similar to the one we find in Greek. The Chol pattern, by contrast, is closer to nominalized forms of the type encountered in English verbal gerunds, see section 15.4.3. for some discussion. Third, as we see in Table 15.1 the similarity between nominalization and ergativity emerges for those ergative languages that treat single arguments of all types of intransitive predicates alike. In other words, what is comparable are case patterns in Greek type nominalizations with ergative case patterns in languages such as Niuean, in which single arguments of both unergative and unaccusative predicates bear absolutive, and it is only the agent of a transitive predicate that surfaces with a distinct case marking, see n. 1 and the conclusions section. The question that arises then is: what is special about nominal forms that obligatorily gives rise to an ergative case pattern? In the next section, we will see that ergative patterns arise only in a subset of nominalizations in those nominative–accusative languages, which have more than one nominalization pattern. Crucially, such patterns obligatorily contain an n head. We will then turn to the question what is special about the morphosyntax of these nominalizations. The answer will be, as anticipated in Alexiadou (2001) and Salanova (2007), and see also Imanishi (2014), that it relates to the case assigning heads in nominal structure. Specifically, n-based nominalizations make room for one structural Case within the nP phase, and require a deficient VoiceP/vP complement, i.e. a Voice/vP projection that does not project an external/agent argument.
360 Artemis Alexiadou
15.3 Two Types of Deverbal Nominalizations 15.3.1 Mixed Extended Projections In the recent literature, several authors have discussed the criteria that can be used to determine that nominalizations come either with a verbal internal structure of the type in (12a) or with a mixed internal structure of the type in (12b); for instance, Alexiadou et al. (2011) have argued that given their eventive nature, all deverbal Argument Structure nominalizations (ASNs) involve vP, but differ with respect to the type of functional structure above the vP layer. Certain types of nominalizations contain only verbal functional projections and no nominal layers, while others involve nominal layers on top of optional extended verbal projections (see Borsley and Kornfilt 2000; Alexiadou 2001; Ehrich 2002): (10) a. [ DP [ extended Verbal FP [vP ... ]]] b. [ DP [ Nominal FP [ (extended Verbal FP) [vP ... ]]]]
verbal internal structure mixed internal structure
According to (10a–b), all ASNs have the syntactic distribution of a noun (i.e. which is provided by the DP layer), but they retain the syntactic and semantic characteristics of the input verb to a varying degree (vP and extended Verbal FPs), to which they may or may not add further Nominal FPs. The type of evidence that was used to diagnose the presence of these layers relates to the verbal and nominal properties of the different types of nominalizations. Under nominal properties, I understand the availability of adjectival modification, gender and plural marking, the presence of several types of determiners as well as the presence of genitive subjects. Under verbal properties, we can include the presence of arguments, the presence of auxiliaries, the licensing of adverbial modification, the availability of accusative Case on the internal argument, and nominative Case on the external argument. On this view, the distinction between Vs and Ns is not absolute, but gradual in nature: the V/N cut-off point of a nominalization can be located at any point in these scales. In the next section, I will apply these criteria to English gerunds.
15.3.2 English Gerunds As Greek only has one type of ASN, namely the type (10b), leaving nominalized clauses aside, and thus it is not so illuminating as to the division of labor between nominal and verbal properties within nominalizations, let us look at the English patterns in some detail. As is well known, English distinguishes between verbal and nominal gerunds. Nominal gerunds but not verbal gerunds accept all kinds of determiners.
Ergativity in nominalization 361 (11)
a. *That/*the/*a criticizing the book annoyed us. b. The/that/?a reading of the manuscript pleased us.
Moreover, the verbal gerund assigns accusative case (11a), while the nominal gerund, similar to the -ation nominal examples discussed in the previous section, realizes its theme as an of-PP (see Lees 1960; Chomsky 1970). As has been discussed by many authors, and is recently summarized in Alexiadou et al. (2011), further properties correlate with this distinction: the subject of the verbal gerund cannot be replaced by a determiner, but this is possible with the nominal gerund (12); the verbal gerund only allows adverbial modifiers, the nominal gerund only allows adjectives (13): (12)
a. *That/*The criticizing the book annoyed us. b. The jumping of the cows annoyed us.
(13)
a. Pat disapproved of my quietly/*quiet leaving the room. b. The careful/*carefully restoring of the painting took six months.
Under the standard assumption that adverbs modify verbal structures and adjectives nominal structures, this contrast suggests that the verbal gerund contains a verbal internal structure (Abney 1987; Borer 1993; Kratzer 1994; and others), while the nominal gerund has a nominal internal structure. This difference is maintained with respect to aspectual modifiers, as illustrated in (14). While the verbal gerund can license aspectual adverbs this is not the case for the nominal gerund: (14)
a. I am sick of her constantly/*constant criticizing me. b. He could not tolerate her constant/*constantly criticizing of him.
Following Alexiadou (1997) and Cinque (1999), this suggests that the verbal gerund hosts an AspP, which is absent in the nominal gerund. Thus, nominal gerunds cannot license this type of adverbials. For the purposes of this chapter, I assume, following Pustejovsky (1995), Alexiadou (2001) and Borer (2005a) that the verbal gerund -ing suffix carries imperfective aspect. To conclude, verbal gerunds lack nominal internal properties. As a consequence, they also lack plural morphology:3 (15)
*He could not stand her criticizings me.
3 For now, I will consider derived nominals to be identical to nominal gerunds, although I am aware of the fact that important differences exist between them. Since, however, they behave alike as far as the case patterns are concerned, I will assume that they are structurally rather similar. In section 15.4, however, I will offer some more details as to their internal structure.
362 Artemis Alexiadou In view of what was mentioned in section 15.3.1, we would expect the nominal gerund to allow plural morphology. However, as has been discussed in detail in Alexiadou et al. (2010), though plural morphology is indeed possible with some nominal gerunds, it is impossible with others. As shown in (16), some nominal gerunds allow plural, others disallow it (16b), and in some other cases, the nominal gerund is excluded in favor of some other nominalizer: (16) a. I heard of repeated killings of unarmed civilians. b. *The repeated fallings of the stock prices induced their further decline. c. The frequent late arrivals/*arrivings of the train made me take the bus. Alexiadou et al. argued that this is because the nominalizer -ing (different from the verbal gerund suffix -ing) is sensitive to the inner Aspect of the original verb and its competition with other deverbal nominalizers. This was noted in Borer (2005a, vol. 2: 239–245), who claimed that the nominalizer -ing is ungrammatical with telic events like the achievements in (17), but it is fine with atelic events (activities and semelfactives), as in (18): (17) a. *the arriving of the train b. *the erupting of Vesuvius c. *the exploding of the balloon (18) a. the sinking of the ships b. the falling of the stock prices c. the jumping of the cows Other affixes (19a), the verbal gerund (19b) and the progressive (19c) accept telic verbs: (19) a. the arrival of the train b. The train arriving at 5 pm is unlikely. c. The train is arriving. Alexiadou et al. (2010), following Jackendoff (1991), assume that only [+bounded] entities may pluralize. If the nominalizer -ing tends to select atelic events, which are [-bounded], we expect it to reject plural marking. This explains the contrast above. The ungrammaticality of (16c) is a consequence of the incompatibility of the nominal gerund with telic verbs, so the nominalizer -al is preferred (see also (19a)). Thus, crucially the inner aspect sensitivity of the nominalizer -ing also has to do with the availability of other nominal suffixes for the same structure.
15.3.3 Greek Nominalizations If we now compare the Greek nominalization data to the two types of English nominalizations, the following picture emerges: in Greek, all nominalizations show
Ergativity in nominalization 363 gender marking, and this seems to be sensitive to the type of nominalizer involved, - m-nouns are neuter, while -s-nouns are feminine. In agreement with e.g. Alexiadou (2004), Kramer (2014b), Alexiadou et al. (2015) and references therein, gender features are on n: (20) a. kathariz-o kathariz-m-a clean-verb cleaning b. anagnorizo recognize
anagnori-s-i recognition
The behavior of Greek -m-nouns is similar to English nominal gerunds, see Alexiadou (2011) for extensive discussion. As Kolliakou (2003) observed, prototypical state and accomplishment predicates do not give grammatical nominalizations when they combine with the affix -m-. (21)
a. * agapima (loving) *skepsimo (thinking) b. *dolofonima (assassinating) *katastrema (destroying)
Kolliakou (2003: 179)
She notes that a subset of -m-nouns denotes activities, as in (22a), while others denote concrete nouns, as in (22b): (22) a. Activities perpatao walk sprohno ena karotsi push a cart b. Concrete nouns paraskevazo produce
to perpati-m-a the walk to sprok-sim-o tu karotsiu the pushing of the cart paraskevas-m-a product/concoction
However, a closer look at the possible and impossible formations reveals finer details. Certain accomplishment predicates can build -m-nouns, as illustrated below. On the other hand, achievement predicates cannot build -m-nouns at all: (23)
a. Accomplishments htizo ena spiti build a house zografizo ena kiklo draw a circle
to htisimo enos spitiu the building of a house to zografisma enos kiklu the drawing of a circle
364 Artemis Alexiadou b. Achievements anagnorizo recognize ftano arrive ekrignio explode
i anagnorisi/*anagnorisma tu klefti the recognition of the thief i afiksi/*to afigma the arrival/the arriving i ekriksi/*to ekrigma the explosion
Achievement nominalizations with -m-do not receive a different interpretation. They are simply unacceptable. This suggests that -m-affixation does not introduce aspectual shift, it rather introduces non-culmination, extension of activity. Hence it gives grammatical results only with those accomplishments which can receive an extended interpretation. As Kolliakou notes, even if the source predicate denotes a bounded event, the -m- nominalization is interpreted as a process, i.e. as a non-bounded event. Note that the - s-affix seems to happily co-occur with achievements, being what one might call a telicity marker. We can conclude then that -m-nouns are similar to the English nominalizer -ing.
15.3.4 External Arguments in Nominalization Finally, let me turn to issues pertaining to the presence of external arguments in nominalizations. English nominal gerunds arguably contain Voice, which, however, does not project an overt external argument, i.e. it has passive like properties. As can be seen in (24), the ing-of gerund (24b) patterns with the verbal passive in (24a) in excluding a self-action interpretation, the standard diagnostic for verbal passives in Kratzer (1996, 2003). By contrast, derived nominals in (24c) allow a self-action interpretation indicating the lack of VoiceP, see Alexiadou et al. (2013) for discussion. (24) a. The children were being registered. i. *Th = Ag: The children registered themselves ii. Th ≠ Ag: The children were registered by someone b. The report mentioned the painfully slow registering of the children. Th ≠ Ag /*Th = Ag c. The report mentioned the painfully slow registration of the children. Th ≠ Ag /Th = Ag
Greek nominalizations behave somewhat different. In some cases, the genitive DP is interpreted only as an internal argument (Alexiadou 2001). Thus (25) is not compatible with reflexive interpretation: (25)
i the
anagelia announcing
ton the
kalesmenon guests.GEN
(agent≠theme)
Ergativity in nominalization 365 In some other cases, Greek nominals derived from alternating verbs are structurally ambiguous between the passive and the unaccusative form, which I take to signal the absence of Voice. Evidence for this view comes from the following domains, discussed in Alexiadou, Anagnostopoulou, & Schäfer (2009). First, nominals can but do not need to contain Voice. Nominalizations of internally caused verbs never license an agent PP/VoiceP: (26)
to sapisma the rotting
(27)
a. Ta fila sapisan apo tin igrasia / *apo ton kipuro the leaves rotted from the humidity / by the gardener b. *I the
ton the
filon apo tin leaves.GEN from the
igrasia / humidity /
*o the
kipuros gardener
igrasia / *apo humidity / by
sapise rotted
ta the
ton kipuro the gardener
fila leaves
Second, modification by adjectives/adverbs like spontaneous/suddenly gives us distinct interpretations: spontaneous gives a passive interpretation, while sudden is ambiguous between a passive and an unaccusative one. In the case of internally caused roots, however, it only has the unaccusative interpretation: (28)
(29)
(30)
a. to afthormito anigma tis portas the spontaneous opening the door.GEN
→
passive interpretation
b. to ksafniko anigma tis portas the sudden opening the door.GEN
→
ambiguous interpretation
a. to anigma tis portas afthormita → the opening the door.GEN spontaneously b. to anigma tis portas ksafnika → the opening the door.GEN suddenly
passive interpretation ambiguous interpretation
a. *to afthormito sapisma ton filon the spontaneous rotting the leaves.GEN
→
b. to ksafniko sapisma ton filon the sudden rotting the leaves.GEN
→ unaccusative interpretation
*passive interpretation
On the basis of the above diagnostic, we can conclude that while English nominal gerunds contain passive Voice, Greek nominals are ambiguous between an interpretation that contains passive Voice and one that lacks Voice. To conclude, both Greek nominalizations and English nominal gerunds (and certain English derived nominals, see section 15.4) can then be accounted by the structure in (30), and in addition (31) is available for certain Greek nominals and English derived nominals. Arguably, the structure (33) is involved in the formation of English verbal gerunds: (31)
[ DP [ (NumP) [ ClassP[±count] [nP [VoiceP [ vP …]]]]]]
366 Artemis Alexiadou (32)
[ DP [ (NumP) [ ClassP[±count] [nP [ vP …]]]]]
(33)
[ DP [ AspectP [ VoiceP [ vP [ Root ]]]]]
We can now establish the generalization that ergative case patterns in nominalization are related to the presence of n. As far as I can tell, there is no type of nominalization that contains a nominalizer and has a nominative–accusative Case pattern.4 The question is why should this be the case?
15.4 Explaining the Source of Ergativity in the Nominal Domain 15.4.1 General Remarks Coon & Salanova (2009) have argued that the source of ergativity is the separation of the predicate head from T°. Specifically, they propose that because of the presence of an intervening head between the predicate and TP, case assignment takes place within the nP domain, which results in the ergative case pattern. From their perspective, when v and T are in a local relationship, v activates T for obligatory nominative case assignment. When v and T are not in a local relationship, v itself triggers obligatory absolutive case. They argue that in the case of Mebengokre the disruption of the local relationship emerges through nominalization. I think that the basic intuition of their analysis is correct, i.e. the emergence of ergative case patterns has certainly something to do with the number of case licensing/checking heads within nP, and naturally the presence of nP itself; thus, their idea will be very helpful to understand why nominalizations which include n necessarily give rise to ergative case patterns. However, some refinements are necessary. Importantly, we cannot attribute the ergativity pattern to a disruption of a local relationship between vP and TP, as all types of nominalization discussed in section 15.3 lack T. For instance, the verbal gerund lacks T, but still it does not show an ergative case pattern, and disallows the internal argument to surface with genitive. Crucially, then, it is the presence of nP that is responsible for the ergative pattern. Before I turn to my assumptions concerning Case in general and in nominalization in particular, let me briefly remind the reader that Grimshaw (1990) offered a way to capture this. She argued that nominalization is a process akin to passivization. According to this 4
There is one exception to that known to me, namely the Romanian supine. This nominalization behaves similar to the English verbal gerund in all properties but one, namely the one related to its Case pattern, which is similar to that English derived nominals and nominal gerunds. Alexiadou, Iordachioaia, & Soare (2010) discuss this and relate it to the properties of the Romanian enclitic article, which creates a nominal defective environment, unlike English D. Note also that this type of nominalization is based on a verbal participle.
Ergativity in nominalization 367 view, nominalization, like passivization, includes a process of suppressing or demoting the external argument. Support for the proposal that these arguments are suppressed comes from the observation that both the possessive and the by-phrase are optional, see (34–35). (34) a. The enemy’s destruction of the city was unexpected. b. The destruction of the city was unexpected. (35)
a. The city was destroyed by the enemy b. The city was destroyed
If nominalization is like passivization, then we would expect an implicit argument to be present in nominals. Roeper (1987) argues that control properties indicate the presence of an agent in the subject position of nominals, see (36): (36) a. the use of drugs to go to sleep b. the PROi use of drugs [ to PROi go to sleep ] Borer (2013: 184) presents a series of arguments that nominals should be compared with verbal passives. For instance, as in clausal passives, “a generic interpretation for the implicit subject in ASNs is available in appropriate contexts, where it is, presumably, licensed by a generic operator”; this is shown in (37–38), Borer’s (10–11): (37)
a. In the middle ages, old people were particularly appreciated [by all]. b. In some parts of the world, girls are excluded from school [by all].
(38)
a. the appreciation of old people in the middle ages [by all] b. the exclusion of girls from school in some parts of the world [by all]
As we have seen in the previous two sections, nominalizations in other languages realize the external argument as a PP, similar to the one found in verbal passives, and allow for disjoint reference interpretations, exactly like verbal passives. We can thus conclude that indeed some nominalizations are passive-like. Others are derived from intransitive predicates and arguably lack an external argument introduced in VoiceP. From this perspective, then, nominalizations realize ergative structures basically because they are passive/unaccusative, as argued for in Alexiadou (2001), see (39a), from Alexiadou (2001: 111). A related formulation is given in (39b), from Imanishi (2014: 123), but see the remarks in n. 1: (39)
a. Nominalizations lack external arguments. b. The Unaccusative requirement on nominalization Nominalized verbs must lack an external argument.
Recently, Bruening (2013) proposed that NOM heads (or n in our terminology) in addition to changing category have properties very similar to those of passives. Thus,
368 Artemis Alexiadou in the presence of an n head, its task, when embedding a transitive verb, is to turn it into a passive. In particular, according to Bruening (2013: 31), the head Nom, which embeds a VoiceP, is like his Pass head “in requiring that all arguments be saturated. If there is an open argument, Nom, unlike Pass, will project it in its own specifier.” This would account for the double-possessive pattern of English nominalization, but not necessarily for the pattern containing a by-phrase. “If there is no open argument, Nom, like Pass, is semantically vacuous. The way the open argument of Voice can be saturated prior to merger of Nom is by adjoining a by-phrase”. In this chapter, I will assume a version of Bruening’s characterization. Specifically, I will assume that deverbal nominalizations are subject to principle (39), which could be implemented as having a particular kind of n in the structure. This type of selects for a structure that does not project an external argument in VoiceP.
15.4.2 Case In this section, I will briefly summarize my assumptions concerning Case, and in particular ergative Case. Building on Marantz (1991/2000) and Baker (2015), I assume that case is assigned on the basis of the principles of “dependent case” (cf. Yip, Maling, & Jackendoff 1987), see also Baker & Bobaljik (Chapter 5, this volume). Specifically, Marantz (1991/2000) argues that the distribution of morphological case is determined at PF, subject to the case realization hierarchy in (40): (40) case realization disjunctive hierarchy: i) lexically governed case, ii) “dependent” case (accusative and ergative), iii) unmarked case (environment-sensitive), iv) default case The more specific a case is it is assigned first taking precedence over the cases lower in the hierarchy. In this system, structural accusative Case is “dependent case” subject to the definition in (41):5 (41) Dependent case is assigned by V+I to a position governed by V+I when a distinct position governed by V+I is: a. not “marked” (not part of a chain governed by a lexical case determiner) b. distinct from the chain being assigned dependent case Baker (2015; 48–49), building on Baker and Vinokurova (2010), proposes a reformulation of this as in (42), cf. Alexiadou, Anagnostopoulou & Sevdali (2014: 24) for a reformulation of (41) in terms of Agree: 5 Originally dependent case assignment was regulated by the formation of a chain between V + I, see Marantz (1991/2000). More recent analyses embrace a phase-based model and attempt to restate the conditions on dependent case assignment, see e.g. (42) and Schäfer (2012a) among others and references therein.
Ergativity in nominalization 369 (42) a. If there are two distinct NPs in the same spell out domain such that NP1 c-commands NP2, then value the case feature of NP2 as accusative unless NP1 has already been marked for case. b. If there are two distinct NPs in the same spell out domain such that NP1c-commands NP2, then value the case feature of NP1 as ergative unless NP2 has already been marked for case. While Marantz (1991/2000) as well as Baker (2015) considered ergative case to be a dependent case assigned on the subject, see (40), other authors have argued that ergative case is not structural. For instance, Nash (1995) proposed that in accusative languages the subject is projected external to the VP as the specifier of a functional projection which selects VP, Voice in our structures. In ergative languages, however, the subject is projected VP-internally, as the highest adjunct of the lexical VP projection. Since no external argument is introduced in Spec,VoiceP, following Bok-Bennema (1991), ergativity then means that transitive verbs cannot assign dependent structural accusative case. The single argument receives unmarked absolutive Case, see also Alexiadou (2001), Woolford (2006), and more recently Legate (2012a). See also Legate (Chapter 6, this volume), Nash (Chapter 8, this volume), Sheehan (Chapter 3, this volume), and Baker & Bobaljik (Chapter 5, this volume) for further discussion on the nature of ergative case. What is important for us is that the morphological realization of case is sensitive to a particular local domain in which case is assigned. In the verbal domain, the relevant domain is Voice+v+I or rather the CP, see Baker (2015). In the nominal domain, in the absence of I, n and D are candidates for creating the relevant domains, i.e. the nP and DP layer.
15.4.3 Zooming in on n Let us then take a closer look at the properties of this n head. Following Embick (2010), I take n to be a phase head, in the sense that it fixes the interpretation of the structure as a nominal one. As soon as n is inserted, higher functional heads with verbal properties are not allowed, as n and the functional layers above n have to Agree in morphosyntactic features, cf. Imanishi (2014). n is a nominalizer, but also has properties similar to a passive head. Due to its category changing properties, n now creates a nominal Case domain. Since n either embeds a passive VoiceP or simply a vP, see (39), it now counts as the relevant head that determines case. According to (39), the unmarked case for arguments within nP is GEN; this is briefly mentioned in Marantz’s (1991/2000) article, see also Alexiadou (2001), and Baker (2015). The intuition behind the dependency view of ACC is based on the presence of another argument being assigned unmarked case. Let us assume that this intuition is correct, and moreover assume that dependent Case patterns arise when we have more than one head that can enter Agree with an argument, see Schäfer (2008), Bobaljik (2008). Crucially, what is excluded is a head entering multiple Agree with two arguments, see also Haegeman & Lohndal (2010) for discussion.
370 Artemis Alexiadou In the verbal domain, the internal argument of a predicate enters Agree with Voice, and surfaces with accusative, while the external argument enters Agree with T and surfaces with nominative (or alternatively receives unmarked nominative, see Baker 2015). A nominal reflex of this pattern is found in the case of the verbal gerund, the external argument enters Agree with D and surfaces in the Saxon genitive, while the internal argument enters Agree with Voice and surfaces with accusative. Consider now what happens in the case of nominal gerunds or derived nominals. In these cases, our domain of Agree for Case is now the nP. In all cases, we have examined, there is only one argument within the nP, which surfaces with genitive. (43)
DP nP n
vP/VoiceP-ext.arg./ DPgen
The only option allowed for the other argument, in the case of a nominalization derived from a transitive predicate, is either to surface as a by-phrase below nP, since strictly speaking the complement of nP has undergone passive formation, or to be merged higher than the nP domain, thus escaping nP. The latter is what we see in (44), where the external argument is in Spec,DP, and the single internal argument is in the genitive: (44) John’s destruction of the manuscript Data such as the ones in (44) crucially suggest that English D is different from that in other languages, e.g. the Romanian supine data briefly discussed in note 4. Unlike Romanian D, for instance, English D can enter Agree, and thus be responsible for licensing an extra Case within the nominal clause. In contrast, Romanian D is simply a nominalizer, and functions similar to n in (43). When the option in (44) cannot be chosen, due, for example, to the properties of the DP projection, as is the case in Greek, where Spec,DP is an A’-position, or because movement has not taken place, there is no other option available for the external argument but to resort to a PP realization. The nominalization structure in this case, unlike the verbal structure, simply does not make two heads within the nP available for the two arguments to Agree with and surface with distinct cases, thus the external argument will surface as a PP, i.e. take an ergative realization. This is precisely why we cannot have two genitives within the same nP domain, i.e. in one case domain. This seems to be a more general phenomenon that can be captured under the subject-in-situ generalization proposed in Alexiadou & Anagnostopoulou (2001), see (45), and also the Multiple Case Condition in (46), from Collins (2003: 16), cf. the notion of Distinctness in Richards (2010):
Ergativity in nominalization 371 (45)
The subject-in-situ generalization (SSG) By Spell-Out, vP can contain only one argument with a structural Case feature. Alexiadou and Anagnostopoulou (2001)
(46) Multiple Case Condition By Spell-Out VP can contain no more than one argument with a (valued) undeleted Case feature. In both (45), and (46), replacing vP and VP with nP will give us the required result. Since verbal gerunds lack n, they are not subject to (45) or (46), and the external argument obligatorily moves to Spec, DP in this case; dependent Case can be assigned as the external argument enters Agree with D, and no process of passivization has taken place: (47) John’s destroying the manuscript The important question that arises is why nominalizations that contain n or other elements that function as nominalizers cannot tolerate Tense. I believe the answer to this question is relatively simple and rather straightforward: derived nominals simply lack tense, as they cannot be located in time; only events can. Even for the languages that have been argued to contain nominal tense, it has been shown that this is either adverbial in nature (Wiltschko 2003), or that it takes over functions associated with determiners in languages that lack nominal tense (Lecarme 1996, and subsequent work; Alexiadou 2008). Summarizing the analysis: in nominalization, the presence of n locks the interpretation and the properties of the derived element as being nominal. Importantly, the presence of n has as a consequence that higher functional heads are nominal in nature, and not verbal. n creates a case domain requiring the DP argument of ASNs to surface with genitive. In the case of ASNs derived from transitive predicates, unless the external argument moves outside nP, it will be realized via a PP, and as a result an ergative pattern is enforced.
15.5 Conclusion In this chapter, I showed that the presence of a nominalizer head in a syntactic structure creates an environment in which only one argument can receive structural Case. As a result, an ergative case pattern emerges. Let me now briefly return to unergative nominalizations. As stated in footnote one, traditionally such nominalizations are not treated as argument supporting. Thus, the fact that their single argument bears genitive could be explained by assuming that this is introduced in the nominal part of the structure, i.e. it is not an argument of the nominalized verb. From this perspective then the fact that their single argument bears genitive is simply a result the default case pattern within nominals, and the nominalization itself is not really subject to the requirement in (39).
372 Artemis Alexiadou If the single argument of an unergative predicate is introduced in VoiceP, similar to the external argument of a transitive predicate, (39) would enforce a syntax similar to that of impersonal passives, which is clearly not what we have in the case of examples such as (3b). Thus we would be led to assume that the single argument of unergative nominals is introduced in nominal structure. If we, however, assume that ergativity always treats the subject of an intransitive and the object of a transitive predicate alike, while it treats the agent of a transitive predicate differently, we would be led to a different conclusion. The case patterns would be interpreted as suggesting that intransitive nominalizations do not distinguish between unergative and unaccusative predicates. This means that they show a case system of the type described for Niuean by e.g. Massam (2009) and reference therein. In this type of language, the single argument of both unergative and unaccusative predicates bears absolutive case, see Table 15.1. This is in contrast with what has been reported for e.g. certain dialects of Basque, where the subject of unergative predicates surfaces with ergative and the subject of unaccusatives bears absolutive case, see Tollan (2013) for a recent discussion and references. Massam (2009) explicitly argued that the Niuean case pattern can be derived by assuming that only transitive agents are introduced in Voice, while arguments of unergative verbs are introduced within the vP. If unergative nominalizations behave alike, then (39) basically enforces the blurring of the unergative vs. unaccusative distinction in the nominal domain. One could interpret this as suggesting that (39) should explicitly state that nominalizations must lack external arguments of the type introduced in VoiceP. The question that arises then is why not all ergative languages conform to the case patterns in Table 15.1 as far as their verbal domain is concerned (e.g. Niuean vs. dialects of Basque). While treating this issue is well beyond the scope of this chapter, two observations seem relevant here. First, as already discussed in Massam (2009), such patterns raise questions as to the treatment of (all) unergatives as concealed transitives across languages (Hale & Keyer 2002). Second, it points to a difference in the syntactic representation of the event/theta role of the subject of a transitive predicate as opposed to that of an unergative one, see Massam (2009) for some discussion.
Acknowledgments I am indebted to the editors of this volume, Terje Lohndal, and one anonymous reviewer for their comments. My research has been supported by a DFG grant for project B1 The Form and Interpretation of Derived Nominals of the collaborative research center 732 Incremental Specification in Context at the University of Stuttgart.
Abbreviations 1, first person; ABS, absolutive; ACC, accusative; ERG, ergative; GEN, genitive; NOM, nominative; PASS, passive; PL, plural; PP, prepositional phrase.
Chapter 16
Ergativit y a nd Au stronesia n-t ype voice syst e ms Michael Yoshitaka Erlewine, Theodore Levin, and Coppe van Urk
16.1 Introduction Many languages of the Austronesian family exhibit what has been called a “voice system”:1 a particular pattern of alternations in word order, case marking, and verbal morphology, which also interacts with Ā-extraction. The voice system has been a central concern in the study of Austronesian syntax. One influential proposal for such languages treats them as morphologically and syntactically ergative. The ergative hypothesis is attractive because it offers a way of mapping some of the morphosyntactic properties that look uniquely Austronesian, such as its voice morphology, to familiar features of non-Austronesian languages. In this chapter, we critique the ergative analysis of Austronesian-type voice system languages, using data from well- studied voice system languages, including Tagalog, Malagasy, and Atayalic languages, along with new data from Balinese and Dinka (Nilotic), a non-Austronesian language with all of the hallmark properties of an Austronesian voice system.2 On the basis of dissociations between case, voice, and
1
The “voice system” has been made famous by Philippine languages, although the basic description (section 16.1.1) also applies to a range of Formosan and Western Malayo-Polynesian languages. The voice system has been reconstructed for Proto-Austronesian (Wolff 1973; Starosta et al. 1982/2009). See also n. 4 on the term “voice system.” 2 We note that there are also Austronesian languages which are morphologically and syntactically ergative but do not exhibit the voice system described here. See for example Otsuka (Chapter 40, this volume) for discussion of Tongan.
374 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk extraction, we argue that there must be mechanisms other than ergativity that will yield the behavior associated with Austronesian voice. The chapter is organized as follows. We first introduce Austronesian voice systems and their treatment as morphologically and syntactically ergative. In section 16.2, we present new data from the Nilotic language Dinka, a non-Austronesian language with a voice system, in which dissociations between voice and case reveal a consistently nominative- accusative alignment. Section 16.3 documents evidence from multiple extraction in the Malay/Indonesian languages Balinese and Bahasa Indonesia that a voice system can exist in the absence of strict syntactic ergativity. In addition, we show that there are surprising restrictions on the licensing of in situ subjects in these languages, unexpected under an ergative analysis.
16.1.1 Properties of a Voice System Voice systems are characterized by the fact that a single argument of the clause—possibly a non-core argument, as we will see—is privileged in certain ways. This argument may be in a certain linear position or receive a particular morphological marking, and dedicated morphology on the verb indicates which argument of the verb was chosen for this special status. Furthermore, Ā-extraction is often limited to this privileged argument. By way of example, consider the sentences in (1). The sentences all describe Tali eating fish, but vary in word order, case marking, and verbal morphology. (1)
Voice alternation in Squliq Atayal (Liu 2004):3 a. M- aniq qulih qu Tali’. sv-eat fish qu Tali ‘Tali eats fish.’
Subject Voice (SV)
b. Niq-un na’ Tali’ qu qulih qasa. eat-ov gen Tali qu fish that ‘Tali ate the fish.’
Object Voice (OV)
c. Niq-an na’ Tali’ qulih qu ngasal qasa. eat-lv gen Tali fish qu house that ‘Tali eats fish in that house.’
Locative Voice (LV)
d. S- qaniq na’ Tali’ qulih qu qway. iv-eat gen Tali fish qu chopsticks ‘Tali eats fish with chopsticks.’ Benefactive/Instrumental Voice (B/IV)
In each example, one argument of the verb (in italics) is in sentence-final position preceded by the marker qu. Voice morphology on the verb (in bold) reflects this choice of 3
Glosses and translations are modified. It is most common in the Philippine and Formosan literature to refer to Subject Voice and Object Voice as “Actor Voice” and “Patient Voice,” respectively.
Ergativity and Austronesian-type voice systems 375 argument. It is common for Philippine and Formosan languages to have four or five distinct voices. Note that the subject in Non-Subject Voices is preceded by the genitive case maker na, which is also used for nominal possessors. This genitive marking of Non- Subject Voice subjects will become important later. We will refer to the special argument as the “pivot.” A terminological warning is immediately in order: we mean to use the terms “pivot” and “voice” as pre-theoretical labels for the privileged argument cross-referenced by verbal morphology in these languages and the morphology cross-referencing it. The use of “voice” should not be conflated with familiar active/passive alternations in non-Austronesian languages.4 An important property of voice systems is that Ā-extraction is typically restricted to the pivot argument, as illustrated by Squliq Atayal wh-questions (2–3).5 Subject Ā-extraction requires Subject Voice morphology (2a–b), while object Ā-extraction requires Object Voice morphology (3a–b). This pattern extends to other voices as well. (2)
Subject wh-question ⇒ Subject Voice: a. Ima (qu) wal m- aniq sehuy qasa? who qu past sv-eat taro that ‘Who ate that taro?’ b. *Ima who
(3)
(qu) qu
wal past
niq-un eat-ov
sehuy taro
qasa? that
Object wh-question ⇒ Object Voice: a. *Nanu (qu) wal m-aniq (qu) Yuraw? what qu past sv-eat qu Yuraw b. Nanu (qu) wal niq- un na Yuraw? what qu past eat-ov gen Yuraw ‘What did Yuraw eat?’
Atayal exhibits all of the hallmark properties of an Austronesian-type voice system. These properties are summarized in (4). It is worth noting, however, that not all Austronesian languages which could be or have been described as having a voice system clearly exhibit all four of these characteristics.
4 A range of different terms have been used in previous Austronesian literature for these same notions. For example, the terms “subject,” “focus,” “topic,” and “trigger” have all been used by some authors for what we call the “pivot” here. Similarly, the “voice system” is often called a “focus system,” among other terms. See Blust (2002); Ross and Teng (2005) for an overview of terminological use in the literature, also discussed in Blust (2013: sec. 7.1). 5 Wh-questions in Atayal and other Austronesian voice system languages have been variously analyzed as Ā-movement of the wh-word itself or a pseudocleft construction with the wh-word predicating a headless relative to its right; either way, we assume these examples involve Ā-extraction over the pivot argument.
376 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk (4) Characteristics of Austronesian-type voice systems: a. A privileged argument: One argument is designated the “pivot,” and is realized in a particular morphological form and/or structural position, regardless of its original grammatical function. b. Articulated voice morphology: Morphology on the verb varies with the choice of pivot, including options for taking certain oblique arguments as pivots. c. Extraction restriction: Ā-extraction (wh-movement, relativization, topicalization, etc.) is limited to the pivot argument. d. Marking of non-pivot subjects: Non-pivot subjects are morphologically marked, often coinciding with the form of possessors (i.e. genitive case). One of the main challenges of Austronesian syntax is to explain this unique constellation of properties. One prominent attempt to do so, which we will now review, is to analyze voice systems as morphologically and syntactically ergative. (See also Kaufman (Chapter 24, this volume) for further discussion of Austronesian voice systems and their analysis.)
16.1.2 The Ergative Hypothesis In the late 1970s and early 1980s, a line of work emerged suggesting that voice system languages should be analyzed as morphologically and syntactically ergative (De Guzman 1976, 1988; Payne 1982; Cooreman et al. 1984; Gerdts 1988b), and the hypothesis has been modernized and championed in the past decade by Aldridge (2004 and subsequent work). Payne (1982), for example, draws explicit parallels between the clause structure of Tagalog and that of the ergative language Yupik Eskimo. The central tenet of the ergative hypothesis is that the privileged argument of the clause (our “pivot”) carries absolutive case. Marked subjects in Non-Subject Voices are ergative arguments. Subject Voice clauses with transitive roots are analyzed as antipassive clauses, so that Voice morphology is, in the simplest cases, a marker of the verb’s syntactic transitivity. We illustrate the ergative hypothesis using the Squliq Atayal voice system described in section 16.1.1.6 Example (5a) repeats the Squliq Atayal examples (1b), reglossed according to an ergative analysis, together with an intransitive clause from Liu (2004) in (5b). Alterations are bolded. (5)
6
Squliq Atayal as an ergative language: a. Niq-un na’ Tali’ qu qulih qasa. eat-trans erg Tali abs fish that ‘Tali ate the fish.’
Object Voice (1b) = transitive
The presentation here follows ergative analyses of Atayal as in Huang (1994) and Starosta (1999); and the ergative analysis of the sister language Seediq (Atayalic) in Aldridge (2004).
Ergativity and Austronesian-type voice systems 377 b. Cyux m-’abi’ prog intrans-sleep ‘Tali is sleeping.’
qu abs
Tali’. Tali
Subject Voice = intransitive
In an ergative analysis, Object Voice clauses are analyzed as simple transitive clauses, in which the object is marked with absolutive case and the subject with ergative case (5a). Recall that in Atayal Object Voice—and more generally in Non-Subject Voices—the subject is morphologically marked in the same way that nominal possessors are, with the marker na. This is treated as a syncretism between ergative and genitive case, a common pattern cross-linguistically (Trask 1979). In this analysis, Subject Voice marks a syntactically intransitive clause, so that the prototypical case of SV is an example like (5b), in which the intransitive subject is morphologically marked in the same way as the transitive object in (5a): with the absolutive marker qu. The voice morphology glossed as OV and SV in (5) are then markers of the clause’s syntactic transitivity, transitive and intransitive, respectively. This analysis can be extended to Subject Voice clauses with transitive roots by treating them as antipassive constructions. (See Polinsky (Chapter 13, this volume) for an overview of the antipassive cross-linguistically.) The antipassive alternation takes the transitive verb in (5a) and demotes the object qulih ‘fish’ into an oblique, resulting in a syntactically intransitive verb with a single argument, Tali. The result is (6): the verb is now intransitive and therefore bears intransitive morphology (m-). Tali is now the subject of an intransitive verb and thus carries absolutive (qu). No morphology is associated with the antipassivization proper.7 The argument ‘fish’ which was demoted is, under this view, now an oblique. No oblique marking is observed in (6), but note that this argument would be preceded by a distinct marker in other Atayalic languages such as Mayrinax Atayal (Huang et al. 1998; Huang 2000). (6) Subject Voice with a transitive root is analyzed as an antipassive (ap): M-aniq qulih qu Tali’. intrans-eat(ap) fish(obl) abs Tali ‘Tali eats fish.’ Subject Voice (1a) = antipassive Additional voices beyond Subject and Object Voice can be analyzed as applicative constructions (Aldridge 2004), which introduce an adjunct or indirect object as the highest internal argument. It is this argument that is then picked out as the absolutive. Note that we will mainly concentrate on the contrast between Subject and Non-Subject Voices, taking Object Voice as our representative case. The final ingredient in an ergative analysis is syntactic ergativity. Recall that in Austronesian-type voice systems, only the pivot argument cross-referenced by voice morphology (in the descriptive terminology used in the previous section) can be 7 The lack of overt morphological evidence for the ergative hypothesis led some researchers to develop alternative, usage-based diagnostics for ergativity, for example based on the corpus frequency and acquisition of different voices. Such arguments will not be discussed here. See Cumming and Wouk (1987) for review and discussion.
378 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk Ā-extracted. Under the ergative hypothesis, this is described as a restriction that only absolutive arguments can be Ā-extracted. For example, a transitive object wh-question requires Object Voice, which is the regular transitive clause form, as the object is then the absolutive argument. A transitive subject wh-question requires Subject Voice, which uses antipassivization to turn the transitive subject into an absolutive argument.8 This type of extraction restriction is independently observed in many (though not all) morphologically ergative languages (e.g. Manning 1994). A classic example of this extraction restriction in an unambiguously morphologically ergative language comes from Dyirbal (Pama-Nyungan; Australia), in (7a–c). (7) Dyirbal relativization targets the absolutive (Dixon 1979: p. 128): a. ŋuma- ŋgu [duŋgara- ŋu]- ru yabu bura- n. father-erg cry-rel-erg mother(abs) see- past ‘Father, who was crying, saw mother.’ Intransitive subject relative b. ŋuma [yabu- ŋgu bura- ŋu] duŋgara- nyu. father(abs) mother- erg see-rel cry-past ‘Father, who mother saw, was crying.’ Transitive object relative c. ŋuma [bural- ŋa- ŋu yabu- gu] duygara- nyu. father(abs) see- ap-rel mother-dat cry-past ‘Father, who saw mother, was crying.’ Transitive subject relative ⇒ antipassive
Intransitive subjects and transitive objects (absolutive arguments) can be relativized without restrictions, but relativization of a transitive subject requires an antipassivization step first (7c), in order to make the subject an absolutive. This property that only absolutive arguments can be Ā-extracted is often referred to simply as “syntactic ergativity,” although the label originally referred to the presence of any syntactic process sensitive to the ergative/absolutive distinction. We will adopt this terminological choice here and refer to this syntactically ergative extraction asymmetry as “syntactic ergativity.” Consider how the ergative hypothesis captures each of the core properties of voice systems (4): (8) The ergative hypothesis for Austronesian-type voice systems, following (4): a. A privileged argument: Every clause has one absolutive argument. b. Articulated voice morphology: Morphology on the verb reflects the transitivity of the clause and any argument structure alternations, which correlate with the choice of absolutive argument. Applicatives introduce an additional argument as the highest internal argument, which will thus be absolutive.
8 See for example Aldridge (2004) for a detailed derivation of this extraction restriction. Note that, for Aldridge, the structural position of absolutive arguments makes it the unique target for movement; the extraction asymmetry is not a ban on movement of ergative or oblique arguments per se.
Ergativity and Austronesian-type voice systems 379 c. Extraction restriction: The language is syntactically ergative: Ā-extraction (wh-movement, relativization, topicalization, etc.) is limited to the absolutive argument. d. Marking of non-pivot subjects: Transitive subjects are ergative. Ergative may be syncretic with genitive. (In Subject Voice, antipassivization makes the subject absolutive.) The ergative hypothesis was illustrated here using the Atayalic language of Squliq Atayal, but it has also been considered in contemporary literature for other well-studied Austronesian voice system languages, including Tagalog (Aldridge 2004), Malagasy (Paul and Travis 2006), and Indonesian (Aldridge 2008b). The strength of the ergative hypothesis lies in the fact that it offers an explanation of voice system behavior that does not require postulating mechanisms that are unique to Austronesian. Under the assumption that Austronesian languages are syntactically ergative, this view allows us to recast voice systems as a particular grouping of argument structure alternations which are otherwise cross-linguistically well attested. In the next two sections, however, we show that there are voice systems which cannot plausibly be analyzed as ergative. In Dinka, a Nilotic language with all the properties of an Austronesian voice system, dissociations between voice and case uncover a consistently nominative-accusative alignment. In Balinese and Bahasa Indonesia, two Malay/Indonesian languages, multiple extraction in questions, topicalization, and relative clauses reveals the absence of syntactic ergativity in OV clauses. These facts suggest that neither morphological nor syntactic ergativity are necessary ingredients for an Austronesian-type voice system. We conclude then that, even if ergativity might underlie some Austronesian voice systems, there must be mechanisms other than ergativity that will bring about a voice system.
16.2 Dissociating Case and Voice in Dinka In this section, we introduce the voice system of the Nilotic language Dinka. As in Austronesian languages, voice morphology in Dinka correlates with restrictions on extraction and changes in case relations. Most importantly for our purposes, Dinka subjects display the same case pattern that has provided the impetus for the ergativity view of Austronesian voice, alternating between unmarked case in the Subject Voice and a marked case also used for possessors in all other voices, variously referred to as genitive, oblique, or marked nominative (Andersen 1991, 2002; König 2008b, Chapter 37, this volume). However, in Dinka, case marking on subjects is dissociable from voice morphology. We will show that there are several syntactic environments in which the voice system is suppressed, triggering the appearance of Subject Voice as a morphological default. In
380 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk these contexts, however, subjects still appear in the genitive case, the case that would be treated as ergative in an ergative analysis. In fact, these environments show a consistent nominative-accusative alignment. As a result, there must be a mechanism that yields these case marking patterns that is independent of voice morphology.9
16.2.1 The Dinka Voice System Dinka is a Nilotic language spoken in South Sudan. Data in this chapter is from Dinka Bor, the major dialect in the southeastern dialect group. Dinka is a V2 language, which, following van Urk (2015), we take to reflect a requirement of C that it must have a specifier, with concomitant movement of the highest verb/auxiliary up to C. Dinka, like Malagasy (e.g. Pearson 2001, 2005), has three voices, which reflect the grammatical function of the noun phrase in Spec-CP, or the pivot. Subject Voice is used when the subject is the clause-initial pivot (9a), Object Voice when it is the object (9b), and the Oblique Voice is employed for all other choices of pivot (9c). (9) Voice on second position verb: a. Bòl à- càm cuîin cé̤ pǎal. Bol 3s-eat.sv food p knife ‘Bol is eating the food with a knife.’
Subject Voice (SV)
b. Cuîin à-cέεm Bôl nè pǎal. food 3s-eat.ov Bol.gen p knife ‘The food, Bol is eating with a knife.’ c. Pǎal à-cέεmè Bôl cuîin knife 3s-eat.oblv Bol.gen food ‘With a knife, Bol is eating the food.’
Object Voice (OV)
Oblique Voice (OblV)
Voice morphology appears on the verb or auxiliary in C, which is the main verb in (9a–c). However, if an auxiliary is present, the highest auxiliary moves to second position, just as in Germanic V2 languages. In such constructions, voice distinctions are made on the auxiliary and not the verb. The examples in (10a–c) illustrate this for the perfect auxiliary cé.10 (10) Voice on second position auxiliary: a. BÒl à-cé̤ cuîin câam nè pǎal. Bol 3s-perf.sv food eat.nf p knife ‘Bol has eaten food with a knife.’
Subject Voice
9 See also König (2008b, Chapter 37, this volume) for descriptions of Nilotic languages with the same case pattern without voice morphology. 10 Dinka differs in this respect from many Austronesian languages, in which voice morphology appears on the verb. However, it is known that even related languages can vary in whether certain
Ergativity and Austronesian-type voice systems 381 b. Cuîin à - c í i Bôl câam nè pǎal. knife food 3s-perf.ov Bol.gen eat.nf p ‘Food, Bol has eaten with a knife.’ c. Pǎal à-cénè Bôl cuîin câam. knife 3s-perf.oblv Bol.gen food eat.nf ‘With a knife, Bol has eaten the food.’
Object Voice
Oblique Voice
As in Austronesian languages, voice has repercussions for case marking and extraction. The pivot always occurs in the unmarked case, regardless of its grammatical function, as evident in the examples in (9a–c) and (10a–c). In addition, voice restricts extraction, so that only the pivot can undergo wh-movement, for example (11a–c). (11)
Voice restricts extraction: a. Yeŋà càm/*cέεm/*cέεmè cuîin nè pǎal? knife who eat.sv/eat.ov/eat.oblv food p ‘Who is eating the food with a knife?’
Subject Voice
b. Yeŋó cέεm/*càm/*cέεmè Bôl nè pǎal? eat.ov/eat.sv/eat.oblv Bol.gen knife what p ‘What is Bol eating with a knife?’
Object Voice
c. Yeŋó cέεmè/*càm/*cέεm Bôl cuîin? eat.sv/eat.ov Bol.gen food eat.oblv/ what ‘What is Bol eating the food with?’
Oblique Voice
Just as in other voice systems, the case marking on subjects alternates with voice. In Subject Voice, the clause-initial subject occurs in the unmarked case (12a), but, in Object Voice or Oblique Voice, subjects appear in the genitive case (12b).11 (12)
Voice determines case marking on subjects: a. Àyén à-càm cuîin nè pǎal. Ayen 3s-eat.sv food p knife ‘Ayen is eating the food with a knife.’ b. Cuîin à-cέεm Áyèn nè pǎal. knife food 3s-eat.ov Ayen.gen p ‘The food, Ayen is eating with a knife.’
morphology is expressed on the verb or auxiliary. We therefore believe this difference does not preclude us from considering Dinka in the context of a broader discussion of Austronesian-type voice systems, as Dinka does exhibit the core properties summarized in (4). 11
Case marking in Dinka involves tonal alternations. See Andersen (2002) for an overview.
382 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk We refer to this case as genitive, because it also appears on possessors (see Andersen 2002 and König 2008b for discussion). In any case, the similarity with Austronesian voice systems is striking. It should be clear then that Dinka has all the properties that make an ergativity approach to voice appealing: a case alternation with subjects and restrictions on extraction that correlate with voice morphology.
16.2.2 Against an Ergative Analysis of Dinka We will now show that we can rule out an ergative analysis of Dinka. This leads us to the conclusion that, despite the advantages of the ergative analysis, there must be a different mechanism for arriving at an Austronesian-style voice system. The first problem with an ergative analysis of Dinka is that morphology encoding argument structure alternations has a different distribution than voice morphology. For example, Dinka has an antipassive construction, which is independent of the voice system described in section 16.2.1. As documented in detail by Andersen (1992), antipassive morphology appears on the verb and the object is demoted to an optional PP (13a–b).12 (13) Dinka has an independent antipassive: a. Àyén à-cé cuîin câam. Ayen 3s-perf.sv food eat.nf ‘Ayen has eaten the food.’ b. Àyén à-cé càm Ayen 3s-perf.sv eat.ap.nf ‘Ayen has eaten food.’
(è cuîin). p food
Antipassive morphology always appears on the lexical verb, even when an auxiliary is present, as (13b) illustrates. Voice morphology, in contrast, shifts to the highest auxiliary if one is present, as previously discussed. This difference is problematic for a view in which voice morphology is argument structure morphology, particularly if we treat Subject Voice as an antipassive. Another problem faced by an ergative view of Dinka is that the mechanisms behind voice morphology can be shown to be independent of the mechanisms behind genitive case marking. In particular, there are several syntactic environments in Dinka in which
12
There is also an independent applicative construction, which introduces benefactive arguments. Like the antipassive, this morphology is restricted to the verb. It is not unreasonable to think, however, that Dinka might have two distinct applicatives (along the lines of Pylkkänen 2002). Mark Baker (p.c.) asks whether we could think of Dinka as having two different antipassives, as has been claimed for some Mayan languages. However, it is not clear that all of the constructions that have been analyzed as antipassives in Mayan in fact are true antipassives—see e.g. Smith-Stark (1978), Grinevald Craig (1979), Aissen (1999a), Stiebels (2006) for arguments against viewing Agent Focus (formerly described as the “agentive/focus antipassive”) as an antipassive.
Ergativity and Austronesian-type voice systems 383 V1 order is possible, and where no phrase overtly moves to Spec-CP. In these contexts, subjects surface in the genitive case, but the clause is marked with Subject Voice. The V1 pattern is found obligatorily in yes-no questions and optionally in wh-in situ questions or after the finite complementizer è. Both environments involve full finite in allowing V1 order (14a–c). clauses, but differ from matrix declarative clauses (14)
V1 order in yes-no and in situ questions and embedded clauses: a. Cé Áyèn cuîin câam? perf.sv Ayen.gen food eat.nf ‘Has Ayen eaten the food?’ b. Cám Bôl ŋó? eat.sv Bol.gen what ‘What is Bol eating?’ cuîin câam]. c. A-yùukù luêeel [è cé Áyèn 3s-hab.1pl say.nf C perf.sv Ayen.gen food eat.nf ‘We say that Ayen has eaten the food.’
We propose that the V1 order arises because these constructions involve C heads that do not require V2.13 When these V1 orders are possible, every nominal in the clause is case-marked just as when it is not the pivot. Subjects are genitive, as in Object Voice and Oblique Voice clauses. Importantly, however, a V1 clause only allows Subject Voice morphology. This mismatch is surprising under an ergative analysis. If we take Non-Subject Voice morphology to reflect ergative alignment, genitive should not be able to surface in the absence of this morphology. It should not matter whether V2 is possible, since these clauses are big enough to host the requisite argument structure alternations. It is worth reflecting briefly on what kind of approach to voice morphology might fare better with regard to the facts in (14a–c). We think that, at least for Dinka, this pattern argues strongly for an analysis in which voice morphology is treated as extraction marking, as in wh-agreement or case agreement approaches (e.g. Chung 1994; Richards 2000; Pearson 2001, 2005; Rackowski 2002). If voice morphology is a by-product of extraction, then there should not be voice distinctions in clauses without extraction. We can then interpret the Subject Voice just as the default form in the voice morphology paradigm.14 The final problem for an ergative analysis we would like to discuss is that the genitive case on non-pivot subjects shows no sensitivity to properties of the verb, such as transitivity and 13
Another option is that some of the constructions involve silent operators that satisfy V2 but do not participate in the voice system, either because they are not nominal in nature or because they are base- generated in the left periphery and so have not undergone movement. 14 Certain nonfinite clauses in a range of Formosan languages can only occur in Subject Voice (Chang 2010), which suggests that Subject Voice is a morphological default in these languages as well. This view is strengthened by the analysis of such embeddings as restructuring, and the availability of so-called long passive constructions which show that the embedded Subject Voice morphology is not syntactically real. See Chen (2010, 2014) for such arguments from Mayrinax and Squliq Atayal and Wurmbrand (2015) for discussion.
384 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk unaccusativity.15 As long as the subject is not clause-initial, such as in a V1 clause, genitive case may surface in unaccusatives (15a), unergatives (15b), and antipassives (15c).16 (15)
Genitive case occurs with all intransitives: a. Bé lèc dhuôoŋ? fut.sv stick.gen break.inch.nf ‘Will the stick break?’ b. Thε´t Bôl? cook.sv Bol.gen ‘Is Bol cooking?’ c. Bé Bôl càm (è̤ cuî̤in)? fut.sv Bol.gen eat.ap.nf p food ‘Will Bol eat food?’
This pattern too is surprising under an ergative analysis, because it reveals a consistent case marking for subjects according to a nominative-accusative alignment. Taken together, these facts suggest that there are mechanisms other than ergativity that will yield an Austronesian-type voice system. Specifically, it seems clear that there are syntactic processes independent of ergativity that may lead to voice morphology as well as a case alternation involving genitive for subjects.
16.2.3 Genitive as a Repair In this section, we consider the question of what mechanism might lie behind the assignment of genitive case in Dinka. As discussed in the previous section, we assume that voice morphology in Dinka should be treated as a form of extraction marking, as in wh-agreement or case agreement proposals (Chung 1994; Richards 2000; Pearson 2001, 2005), given its independence from the processes behind case marking on subjects. The view of genitive case we want to pursue here is that it represents a strategy for licensing nominals not in a case position, and so functions as a type of repair (cf.
15
See Rackowski (2002) for similar argumentation in Tagalog. One of the ways in which we can tell that these are unaccusatives is that verbs like dhuòoŋ (‘break.inch’) participate in a inchoative/causative alternation (ia–b). 16
(i)
Inchoative/causative alternation: a. Léc à-bé dhuôoŋ. stick 3sfut.sv break.inch.nf ‘The stick will break.’ b. Bòl à-bé léc Bol 3s-fut.sv stick ‘Bol will break the stick.’
dhôoŋ. break.nf
Ergativity and Austronesian-type voice systems 385 Donohue and Donohue 2010; Imanishi 2014). To be precise, we follow Halpert (2012) in assuming that, in some languages, case morphology may be merged directly to a nominal to license it, if no other licensing strategy is available.17 See also Stowell (1981) on English of-insertion. We apply this to Dinka as follows. We propose, following van Urk (2015), that Dinka Spec-CP fulfills both of the functions traditionally associated with Spec-CP and Spec-TP, so that it is the landing site of Ā-movement, but also a case position. In Subject Voice, subjects receive case in Spec-CP and so appear in the unmarked case. In Non-Subject Voices, however, the subject needs to be licensed in a different way, because Spec-CP is occupied and T is not a case assigner. This is the role of genitive case morphology. In Dinka, this strategy is not necessary for other nominals. As van Urk and Richards (2015) show, there is a position for objects inside of the verb phrase where they may receive unmarked case. This analysis extends well to Austronesian languages. A number of Austronesian systems can be described in the same terms as Dinka. An example is the (Squliq) Atayal system described in section 16.1 (16a–c). (16) Voice in Squliq Atayal, repeated from (1): a. M-aniq qulih qu Tali’. sv-eat fish qu Tali ‘Tali eats fish.’
Subject Voice (SV)
b. Niq- un na’ Tali’ qu qulih qasa. eat-ov gen Tali qu fish that ‘Tali ate the fish.’
Object Voice (OV)
c. Niq- an na’ Tali’ qulih qu ngasal qasa. eat-lv gen Tali fish qu house that ‘Tali eats fish in that house.’
Locative Voice (LV)
In Atayal, voice morphology references the XP that moves to the position marked by qu.18 Aside from this, however, we see the same case alternations as in Dinka. Subjects are unmarked in Subject Voice and genitive otherwise, while objects are always unmarked. We might also expect to find voice languages in which the object may also be in need of such a licensing strategy when not in pivot position. This appears to be the case in Tagalog. In Tagalog, any subject or object not cross-referenced by voice morphology is marked genitive (17a–c).19 17
See Rezac (2011) for a technical implementation of the notion of repair, based on similar repairs in the context of violations of the Person-Case Constraint. 18 The details of the analysis of Atayal is further complicated by the fact that there are cases where qu marks an argument other than the pivot. Such cases constitute an argument against viewing qu as a case marker. See Erlewine (to appear) for details. 19 A notable exception is a process of differential object marking that targets proper names and pronouns in the context of subject extraction (e.g. McFarland 1978).
386 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk (17) Tagalog voice marking (adapted from Guilfoyle et al. 1992): a. Sino ang bili ng damit para sa bata? who ang sv.asp-buy gen dress for obl child ‘Who bought the dress for the child?’ Subject Voice (SV) b. Ano ang bili ng tao para sa bata? what ang ov.asp-buy gen man for obl child ‘What did the man buy for the child?’ Object Voice (OV) c. Sino ang i-bili ng tao ng damit? what ang bv.asp-buy gen man gen dress ‘Who was bought the dress (for) by the man?’ Benefactive Voice (BV) This fits well with the notion that genitive is available as an alternative case-licensing strategy. Under this view, the only difference between Tagalog and Dinka or Atayal is that Tagalog also lacks a licensor for objects in non-Object Voices, triggering genitive morphology there as well.20, 21 In fact, we can find evidence in Tagalog as well that the distribution of genitive is independent of voice morphology. As pointed out by McGinn (1988) and Schachter (1996), Tagalog, just like Dinka, has constructions without voice distinctions. In the recent perfective, no XP is marked with ang-morphology and no voice morphology shows up on the verb (18).22 (18) Kabi-bigay lang ng maestra ng libro sa bata. rec.perf-give just gen teacher gen book dat child ‘The teacher just gave a book to the child.’ (Schachter 1996: p. 7) Importantly, both the subject and the object still receive genitive marking. This construction is then essentially analogous to the V1 constructions of Dinka, and shows that the Tagalog genitive, regardless of whether it appears on the subject or object, is also not dependent on voice. This section has shown that the mechanisms behind voice morphology can be dissociated from those behind case marking on subjects. This is evidence that there must be routes independent of ergativity that lead to a voice system. We suggested that the marked case on Non-Subject Voice subjects reflects the presence of case morphology directly merged to the nominal, following Halpert (2012) (see also Imanishi 2014), in order to provide a way of case-licensing subjects outside of the voice system. 20
As Aldridge (2012b) points out, genitive marking on Tagalog objects has interpretive consequences. We adopt Aldridge’s proposal that this is result of the interaction between inherent licensing and the application of existential closure. See Aldridge (2012b) for details. 21 This kind of system could also be a source of Austronesian languages in which objects in non-Object Voices surface with accusative, if we allow for the “repair” case to have a different spell-out inside the VP. 22 As discussed in McGinn (1988) and Schachter (1996), any XP in a recent perfective clause may undergo extraction. This seems to fit well with the view, implicit in our discussion, that voice morphology is in essence cosmetic and does not impose extraction restrictions.
Ergativity and Austronesian-type voice systems 387
16.3 Dissociating Voice and Extraction in Malay/Indonesian As discussed in the previous section, one of the defining characteristics of Austronesian voice systems is that only the nominal cross-referenced by voice morphology, the pivot, is eligible for extraction. Within an ergative analysis of Austronesian voice, this correlation is attributed to syntactic ergativity. Extraction of the subject is only possible in Subject Voice where the subject receives absolutive case. In this section, we examine wh-movement in the Malay/Indonesian languages Balinese and Bahasa Indonesia. We show that contrary to the expectations of an ergative analysis, non-pivot subjects are not immobile. This indicates that syntactic ergativity is not a necessary condition of Austronesian-type voice systems. In addition, there are surprising conditions on the realization of non-pivot subjects in Balinese and Bahasa Indonesia, which suggests that they are subject to a strict head-head adjacency requirement with the verb. We argue that this reflects an alternative licensing strategy, much like genitive case in Atayal, Dinka, and Tagalog.
16.3.1 The Malay/Indonesian Voice System Balinese and Bahasa Indonesia, as well as other Malay/Indonesian languages like Javanese and Madurese, have three voices: Subject Voice, Object Voice, and an Indo-European- style passive voice. We will only be concerned with Subject and Object Voices here. In Balinese, Subject Voice is marked by a nasal prefix ng-, whose form is phonologically conditioned by the initial segment of the verbal stem (19a). Object Voice is marked by the absence of this prefix (19b). The preverbal position is the canonical pivot position, to the left of all auxiliaries. Non-pivot arguments are realized to the right of the verb. (19)
Balinese Subject Voice and Object Voice a. Tiang lakar numbas bawi-ne punika. I will sv.buy pig-def that ‘I will buy that pig.’
Subject Voice
b. Bawi- ne punika lakar tumbas tiang. pig-def that will ov.buy I ‘I will buy that pig.’
Object Voice
In Bahasa Indonesia, the prefix meng-, whose form is also phonologically conditioned, marks Subject Voice (20a). As in Balinese, Object Voice is marked by the absence of this morphology (20b). Unlike in Balinese, non-pivot subjects appear immediately to the left of the verb.
388 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk (20) Bahasa Indonesia Subject Voice and Object Voice (adapted from Cole et al. 2008): a. Tono tidak mem-beli buku di toko buku. Tono neg sv-buy book loc store book ‘Tono didn’t buy the book at the book store.’ Subject Voice b. Top ini sudah saya beli. hat this perf 1sg ov.buy ‘I bought this hat.’
Object Voice
Compared to the voice systems of Philippine and Formosan languages, the voice inventory of Malay/Indonesian languages is diminished. These languages do not employ separate voice morphology for cross-referencing benefactive, locative or recipient arguments. Rather, they utilize applicative marking that promotes such arguments to direct objects. As a result, any argument other than the external argument that serves as pivot is cross-referenced by Object Voice.
16.3.2 Wh-extraction in Indonesian Languages Balinese and Bahasa Indonesia, like other Austronesian-type languages discussed in the previous section, show wh-extraction asymmetries.23 When the verb bears Subject Voice, only the subject can be extracted. This restriction is easy to see in Bahasa Indonesia in which the marker yang co-occurs with wh-movement. Consider the dichotomy in (21a–b): (21) Subject Voice restricts extraction to subjects (Cole and Hermon 2005): a. Siapa yang ___ mem-beli buku di toko buku? who yang sv-buy book loc store book ‘Who bought a book at the book store?’ b. *Apa yang Tono mem-beli ___ di toko buku? what yang Tono sv- buy loc store book Intended: ‘What did Tono buy at the book store?’
23
In addition to wh-extraction, Malay/Indonesian languages permit wh-in situ constructions. In such constructions, either argument can be questioned regardless of voice, as illustrated for Balinese by the data in (ia–d). (i) In situ wh-questions show no voice asymmetries: a. Nyen meli montor anyar? who sv.buy car new b. Montor anyar beli car new ov.buy ‘Who bought a new car?’
nyen? who
Ergativity and Austronesian-type voice systems 389 Similar facts obtain in Balinese, although there is no analog of yang in the language. In Subject Voice, only the subject may be extracted (22a–b).24 (22) Subject Voice blocks object extraction: a. Nyen ngalih bawi-ne punika ditu who sv.seek pig-def that there ‘Who looked for that pig there yesterday?’
ibi? yesterday
b. *Apa ci ngalih ditu ibi? what you sv.seek there yesterday Intended: ‘What did you look for there yesterday?’ In Object Voice clauses, only the object can be wh-extracted. The dichotomy is again easily seen in Bahasa Indonesia (23a–b), and also holds in Balinese (23a–b).25 (23)
Object Voice restricts extraction to objects (Cole and Hermon 2005): a. Apa yang ___ akan kamu lihat? what yang will 2sg ov.see ‘What will you see?’ b. *Siapa yang buku ini akan ___ lihat? who yang book this will ov.see Intended: ‘Who will see this book?’
(24) Object Voice blocks subject extraction: a. Apa beli Nyoman? What ov.buy Nyoman ‘What did Nyoman buy?’
c. Cicing ngugut nyen? dog sv.bite who d. Nyen gugut cicing? who ov.bite dog ‘Who did the dog bite?’ Similar observations are made for Bahasa Indonesia by Cole et al. (2008). These facts indicate that voice only restricts extraction, and not whether the non-pivot argument can be questioned. 24 We could imagine analyzing (22a) as an instance of wh-in situ (as in n. 23). However, more complex examples involving object scrambling show that wh-subjects can undergo movement in the Subject Voice, as discussed by Arka (2004). 25 Here, we cannot actually be sure that (24a) involves wh-extraction rather than wh-in situ, since scrambling the subject before the verb in OV is independently ungrammatical (Artawa 1994; Clynes 1995; Wechsler and Arka 1998). Furthermore, given the relatively free word order of adverbial elements, we cannot be sure that the relative position of the wh-phrase with respect to an adverbial indicates overt movement. Nevertheless, given the availability of overt wh-movement in SV clauses, we take such movement to be possible here as well.
390 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk b. *Nyen montor anyar beli ___? who car new ov.buy Intended: ‘Who bought a new car?’ Based upon the extraction asymmetries illustrated in (23)-(24), it is often reported that wh- extraction only targets the nominal cross- referenced by the verb (e.g. Wechsler and Arka 1998, Arka 2004 for Balinese; Cartier 1979, Hopper 1983, Verhaar 1988, Arka and Manning 1998, for Bahasa Indonesia). Like the extraction asymmetries in Philippine and Formosan Austronesian languages which display more articulated voice systems, these facts are amenable to an ergative analysis of Austronesian voice, in which these extraction restrictions are attributed to syntactic ergativity although Balinese and Bahasa Indonesia both lack overt case morphology. However, the extraction restriction in Balinese and Bahasa Indonesia is not as rigid as it is in Atayal or Dinka. It is possible to extract both subject and object arguments, so long as the appropriate Voice morphology is realized.26 Unlike the ungrammatical (22b), the object and subject may simultaneously be extracted in Balinese when the verb bears OV morphology (25a–b). (25) OV permits non-subject extraction: a. Apa ci alih ___ ditu ibi? what you ov.seek there yesterday ‘What did you look for there yesterday?’ b. Buku nyen Nyoman lakar book which Nyoman will ‘Which book will Nyoman read?’
baca ov.read
___?
In (25), neither argument is in situ, because both are realized to the left of the verb.27 Such examples are problematic for an analysis of Austronesian voice which involves syntactic ergativity, because they show that non-pivot arguments may undergo movement. Extraction of an object over an already extracted subject is also attested in topicalization. In an SV clause, both the canonical SVO word order and the marked OSV word order are well formed (26) (Arka 2004).
26
See e.g. Cole et al. (2008) and Yanti (2010) for similar observations in related languages. Edith Aldridge (p.c.) asks whether the preverbal subjects in (25) could be clitics on the verb. They cannot be, for two reasons. First, Balinese does have a series of pronominal clitics but they follow the verb, as in (i) below from Wechsler and Arka (1998: p. 21). Second, these pronominal clitics are always hosted by the lexical verb, but the preverbal subjects can precede auxiliaries, as in (25b). 27
(i) Buku-ne jemak=a. book-def ov.take=3 ‘(S)he took the book.’
Ergativity and Austronesian-type voice systems 391 (26) Non-pivot object topicalization: a. Tiang nunas kopi-ne niki. 1 sv.take coffee- def this ‘I took this coffee.’ b. Kopi-ne niki tiang coffee-def this 1 ‘This coffee, I took it.’
nunas. sv.take
Again, in (26), neither argument is in situ, because both are realized to the left of the verb, illustrating that non-pivot arguments may undergo movement. A similar argument has been made from Bahasa Indonesia relative clauses (Chung 1976, 1978; Cole and Hermon 2005). As in matrix wh-questions, relative clauses display extraction asymmetries. An object cannot be relativized if the predicate of the embedded clause bears SV morphology and the subject is in pivot position (27). (27)
Subject Voice restricts extraction to subjects: *[Buku [yang Budi tidak akan mem-baca]] sangat menarik. [buku [yang Budi neg will sv-read ]] very interesting Intended: ‘The book that Budi will not read is very interesting.’
However, if the verb appears in OV, object relativization can accompany subject fronting (28). (28) Object Voice permits multiple extraction: [Buku [yang Budi tidak akan baca ]] [buku [yang Budi neg will ov.read]] ‘The book that Budi will not read is very interesting.’
sangat very
menarik. interesting
We can see that the subject has undergone movement, because it is realized to the left of auxiliaries and negation (cf. (20)a–b). Like Balinese matrix wh-questions, the behavior of Bahasa Indonesia relative clauses reveals that OV is dissociable from extraction. The observation that multiple arguments can be extracted in Malay/Indonesian languages indicates that syntactic ergativity is not a necessary condition on the formation of voice systems. Voice does not determine which arguments are available for extraction, as would be expected under a strict implementation of syntactic ergativity. Rather, Voice seems to indicate which arguments have been extracted to which positions.28 This characterization is, like the Dinka data in section 16.2, amenable to a view that Voice morphology is extraction marking, as in wh-agreement or case agreement approaches (e.g. 28 This second point is critical. Not all extraction is marked equally. Wh-extraction of the object over the subject requires OV morphology, as in (25) and (28). Topicalization requires SV morphology (26). We suggest that the positions targeted by these movements are distinct. Movement to the former results in a change of Voice; movement to the latter does not. Chamorro wh-agreement also displays a change in verbal morphology triggered by wh-movement (Chung 1994).
392 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk Chung 1994; Richards 2000; Pearson 2001, 2005). If voice morphology is a by-product of extraction, then multiple arguments should be able to extract, as is attested. Voice simply marks the results of the extraction process.29 What remains to be explained is why certain combinations of extracted arguments, like (24b), are unattested, which will be addressed in the following section.
16.3.3 The Behavior of Non-Pivot Subjects The ergativity hypothesis faces further complications when considering restrictions on non-pivot subjects. In this section, we show that there are constraints on what nominals are well formed as non-pivot subjects in Balinese.30 Specifically, such nominals must display head-head adjacency between the nominal head and verb (e.g. Baker 2014b; Levin 2015). We suggest that this represents an alternative method of subject licensing, in lieu of the genitive case in Atayal, Dinka, and Tagalog.31 In Balinese, in situ subjects do not appear in a dedicated case, as in Dinka or as in many other Austronesian languages. In fact, there is no overt case morphology in the language. Instead, in situ subjects are constrained in entirely different ways. These subjects can only be realized as pronouns (29a),32 proper names (29b), and indefinite NPs (29c). Definite descriptions are blocked from appearing in post-verbal position (29d) (Wechsler and Arka 1998). (29) The form of Balinese in situ subjects: a. Be-e daar ida. fish-def ov.eat 3sg ‘(S)he ate the fish.’ b. Be-e daar Nyoman. fish-def ov.eat Nyoman ‘Nyoman ate the fish.’ c. Be-e daar fish-def ov.eat ‘A dog ate the fish.’
cicing. dog
d. *Be-e daar cicing-e. fish-def ov.eat dog-def ‘The dog ate the fish.’ 29
This position is taken in Saddy (1991) and was later adopted by Cole and Hermon (1994, 1998) and Soh (1996). However see Aldridge (2008b) for an alternative proposal. 30 Similar facts hold of Bahasa Indonesia (Guilfoyle et al. 1992; Sneddon 1996). However, non-pivot subjects are limited to pronouns and proper names. 31 See Baker (1988) for a specific implementation of how adjacency of a nominal to a verb, or more accurately the adjunction process which yields adjacency, i.e. Head Movement, can license that nominal in the absence of case assignment. 32 The pronominal element is a clitic in low register and a free pronoun in high register speech.
Ergativity and Austronesian-type voice systems 393 This is not an instance of differential subject marking, because it groups together indefinite subjects, pronouns, and proper names, to the exclusion of definite subjects, unlike any process of differential argument marking (e.g. Aissen 2003b). Levin (2015) argues instead that what the acceptable subjects in (29a–d) have in common is that the highest nominal head (D0 in the case of (29a–b) and N0 in the case of (29c)) is surface adjacent to the verb. In contrast, definite subjects are headed by the suffix -e; the NP then intervenes between the verb and the highest nominal head. This reflects a more general pattern. Whenever linear adjacency of the verb and the highest nominal head is disrupted, ungrammaticality arises. Such intervention can be caused both by material outside of the nominal or by material inside of the nominal. Adverbs, which generally show freedom of placement in the clause (e.g. Wechsler and Arka 1998), cannot appear between the verb and OV subject (30). (30) NP-external intervention: *Be-e daar keras-keras ida/Nyoman/cicing. fish-def ov.eat quickly 3sg/Nyoman/dog ‘(S)he/Nyoman/A dog ate the fish quickly.’ Similarly, while modifiers are canonically realized to the right of the nominal they modify, some can be realized to the left (31a). Modifier-noun order is impossible with OV subjects, however, because the modifier intervenes between the verb and the subject (31b).33 (31) NP-internal intervention: a. (Liu) cicing (liu) ngugut Nyoman many dog many sv.bite Nyoman ‘Many dogs bit Nyoman. b. Nyoman gugut (*liu) cicing (liu) Nyoman ov.bite many dog many ‘Many dogs bit Nyoman. We propose that this reflects a strategy of subject licensing under adjacency, following Levin (2015). In particular, we suggest that, as in Dinka, there is no case position for non-pivot subjects, and so these subjects require an alternative method of licensing. This approach lets us capture the adjacency facts, but can also explain the limited set of nominals which can occur as non-pivot subjects. Only those nominals in which the nominal head is immediately adjacent to the verb will be well formed. This restricts definite DPs to pronouns and proper names, because any other DP will have (overt) 33 See Baker (2014b) and references cited therein for similar observations regarding pseudo-noun incorporation. In such constructions head-head adjacency is also required between the nominal head of a caseless NP and the verb. Intervention effects arise when NP-external and NP-internal material disrupts the required adjacency.
394 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk NP-material intervening between the D0 head, which appears to the right of the NP, and the verb. Furthermore, while NPs can be larger than N0, as in (31b), their head must be the leftmost element in the phrase. As in Dinka, there are then two possible means of licensing the subject in Balinese (and Bahasa Indonesia). Recall from our previous discussion of the genitive repair in Dinka that we take the pivot position to be the landing site of Ā-movement, but also a case position. In Subject Voice, subjects receive case in this position. In Non-Subject Voices, however, the subject needs to be licensed in a different way, because the pivot position is occupied. This is the role of licensing under adjacency. We believe that the general logic of Baker’s (1988, and subsequent work) account of licensing via Head Movement can be extended to these data. Specifically, we suggest, following Levin (2015), that adjunction of a nominal head to a verbal head renders it invisible to the Case Filter.34 Crucially, ungrammatical instances of multiple extraction can be captured under this view of licensing via adjunction. As noted, multiple extraction is possible when the object is extracted over an already extracted subject in the case of wh-movement (25) and topicalization (26). However, wh-movement (24b), repeated in (32a) or topicalization (32b) of a subject over an already extracted object is ungrammatical (32). (32) Subjects cannot extract across a fronted object: a. *Nyen montor anyar beli ___? who car new ov.buy Intended: ‘Who bought a new car?’ b. *Cicing ia uber ___. dog 3 ov.chase Intended: ‘A dog, it chased him/her.’ The ungrammaticality of these sentences can be captured as a failure to case-license the subject. We propose that only the pivot position is a case-position. All other positions in the left periphery are strict Ā-positions, unable to case-license arguments. If the subject is not extracted to pivot position, it must be licensed under adjunction in immediately postverbal position. Subsequent movement operations either cannot target the subject at all due to this requirement, or move the subject to a position in which licensing under adjunction is impossible, yielding ungrammaticality. Similar facts obtain in Austronesian languages with more voices, such as Malagasy. As discussed in Paul (1996) and Keenan (2000), non-pivot subjects undergo a form of compounding with the initial verb referred to as N-bonding,35 as the examples in (33a–c) demonstrate. 34
Levin (2015) assumes that adjunction can occur at various points in the derivation (e.g. Halle and Marantz 1993; Bobaljik 1995). See Embick and Noyer (2001) for an articulated account of the interaction of adjunction and derivational timing. In the case of Balinese, adjunction occurs very late in the derivation after linear order has been established, capturing the strict, linear head-head adjacency requirement. 35 N-bonding is also attested on possessors, again highlight the similarity of form shared by (non-pivot) subjects and possessors attested in many of the languages discussed in this chapter. This may suggest that possessors and non-pivot subjects in Malagasy both lack a licensor.
Ergativity and Austronesian-type voice systems 395 (33)
N-bonding in Malagasy: a. Hitan-ny lehilahy ny trano. ov.see-det man det house ‘The house was seen by the man.’ b. Hitan-ao ny trano. ov.see-2sg det house ‘The house was seen by you.’ c. Hitan-dRabe ny trano. ov.see-Rabe det house ‘The house was seen by Rabe.’
Just as in Balinese, these subjects cannot be separated from the verb (34a–b). (34) Malagasy non-pivot subjects must be right-adjacent to the verb (adapted from Pearson 2005): a. Nohanin-ny gidro haingana ny voankazo omaly. past.ov.eat-det lemur quickly det fruit yesterday ‘The lemur ate the fruit quickly yesterday.’ b. *Nohanin’ haingana ny gidro ny voankazo omaly. We propose that these subjects are licensed in the same way as Balinese non-pivot subjects, under adjunction, which yields head-head adjacency. Unlike Balinese, non-pivot subjects can be definite in Malagasy, as (33a) and (34a) show. Importantly, Malagasy differs from Balinese in that the D head is leftmost in the noun phrase and so is immediately adjacent to the verb.36
16.4 Concluding Remarks In this chapter, we have shown that there are languages with Austronesian-type voice systems that do not display any ergativity. We introduced novel data from the Nilotic language Dinka, a non-Austronesian language with a voice system, which has a consistent underlying nominative-accusative alignment. In addition, we documented a dissociation between voice and extraction in Malay/Indonesian languages, which argues against the idea that all voice systems display syntactic ergativity. On the basis of these facts, we conclude that ergativity cannot be the only route to a voice system. 36 Lisa Travis (p.c.) observes that in Malagasy Oblique Voice constructions, the non-pivot subject displays head-head adjacency with both unergative and unaccusative verbs. We assume that in both cases there is only one argument licensor. Burzio’s generalization holds. When a non-core argument is extracted to pivot, the subject, regardless of base position, must be licensed under adjunction with the verb.
396 Michael Yoshitaka Erlewine, Theodore Levin, Coppe van Urk At the same time, there is an important generalization in the behavior of these different voice system languages: non-pivot subjects are treated differently from other arguments. In Atayal, Dinka, and Tagalog, non-pivot subjects appear in genitive case. In Balinese and Malagasy, non-pivot subjects require adjacency with the verb. We can give a unified characterization to these two types of behaviors through a requirement that non-pivot subjects require a special form of licensing (Case). The two strategies observed are simply two different ways of licensing the non-pivot subject. This licensing requirement is shared between voice system languages which are more amenable to an analysis as morphologically ergative and those which are not. A remaining open question is why and how languages differ in the availability of these two repairs: a last-resort genitive case and licensing by adjacency. One final issue we would like to discuss relates to the analysis of voice morphology. The dissociations between voice and extraction we observed in Dinka and Malay/ Indonesian support a treatment of voice as extraction marking (e.g. Chung 1994; Richards 2000; Pearson 2001; Rackowski 2002). In ongoing work, we develop a theory for Austronesian-type voice systems as extraction marking, which also explains the need for exceptional licensing of non-pivot subjects.
Acknowledgments The work here is part of our ongoing work on the syntax of Austronesian-type voice systems within Austronesian and beyond; see also Erlewine, Levin, and van Urk (2015). We thank Julie Legate, David Pesetsky, Masha Polinsky, Norvin Richards, and audiences at the 21st meeting of the Austronesian Formal Linguistics Association, the 2015 LSA, and at McGill University for discussion and comments. We especially thank Edith Aldridge, Mark Baker, and Lisa Travis for extensive comments on this manuscript. Part of this work is supported by an NSF Dissertation Improvement Grant for the third author, BCS-1440427. Errors are ours.
Abbreviations abs, absolutive; ap, antipassive; asp, aspect; bv, Benefactive Voice; def, definite; erg, ergative; fut, future; gen, genitive; hab, habitual; inch, inchoative; intrans, intransitive; iv, Instrumental Voice; loc, locative; lv, Locative Voice; neg, negation; nf, nonfinite; obl, oblique; oblv, Oblique Voice; ov, Object Voice; p, preposition; past/pst, past; prf/pfct, perfect; rel, relative clause marker; sv, Subject Voice; trans, transitive.
Chapter 17
On the morpho syntac t i c re f l exes of in format i on structure i n t h e e rgative pat te rni ng of Inu it l ang uag e Alana Johns and Ivona Kučerová
17.1 Introduction The Inuit language1 is often characterized as an ergative language (Bok-Bennema 1991; Johns 1992, 2000, 2006; Manning 1996; Compton, Chapter 34, this volume; among others). Interestingly, the Inuit language exhibits a case assignment variability which, unlike traditional split ergativity, does not affect argument alignment, but instead concerns which—and how many—arguments trigger ϕ-feature agreement on finite verb. This chapter asks what is the nature of the relevant grammatical property, and how does it relate to the agreement properties that result. This case variability has not gone unnoticed in the current literature. Some proposals attribute the split to aspect, for instance, Bittner (1987), Spreng (2006, 2010, 2012), Clarke (2009), while others to information structure, or scope (Kalmár 1979; Bok-Bennema
1 The Inuit language is a branch of the Eskimo languages of the Eskimo-Aleut language family. There are four major dialect groupings: Inupiaq, Western, Eastern, and Greenlandic. The majority of linguistic work has been on the Kalallisut (West Greenlandic) dialect. Throughout this chapter we will refer to the Inuit language rather than use the term Inuit, as some do. Inuit is literally ‘people’ inu(k)-it person-plural. We follow the Nunavut government in our usage.
398 Alana Johns and Ivona Kučerová 1991; Bittner 1994; Manga 1996; Bittner and Hale 1996b; Hallman 2008; Berge 2011; among others). All these proposals point out some important property of the observed case variability. Yet, it is not always clear how the proposals derive the morphosyntactic properties of the Inuit language, neither do they provide insight into how these seemingly disparate properties relate to each other. We argue—closely following the proposal of Berge (2011)—that the ergative clause structure of the Inuit language is conditioned by information structure properties, more precisely by its topic-comment properties. We propose a formal model in which this morphosyntactic make-up directly follows from this information structure trigger. Furthermore, we will show that not only does the model account for the morphosyntactic properties of the Inuit language, namely, split case and agreement properties, but it also subsumes other relevant properties discussed in the literature, i.e. scope properties of objects and aspect. These will be seen to be byproducts of the information structure underlying the ergative split. Before presenting our argument, we will first provide evidence that the difference between so-called single (also called intransitive) and double (also called transitive) agreement is a difference of ϕ-agree vs cliticization (Johns to appear pace Compton 2014). We will link this conclusion to the second core observation (see Berge 2011), namely, that absolutive objects must be topics.2 More precisely, we will show that they must be (aboutness) topics in the sense of Reinhart (1981), which is a narrower notion of topics than that used in functionalist literature and does not necessarily entail discourse notions such as recency of mention or persistence in subsequent discourse.3 Crucially, this type of topic—which we will call sentential topic—is strictly realized at a sentential level and has a dedicated syntactic representation. This move will have several immediate consequences: Since topics have been associated with cliticization (Dočekal and Kallulli 2012), the analysis that double agreement is formed by cliticization, along with the independent analysis of the corresponding argument as sentential topic, provides an explanation for the clitic nature of double agreement (and for double agreement itself). Furthermore, since sentential topics must be at the edge of a phase,4 we will argue for a VP shell-like (applicative) structure of Inuit ergative clauses as a necessary precondition for objects being marked as topics (Basilico 2003). With this structural distinction in place, we will show that case marking straightforwardly follows from locality and the morphological case hierarchy (Marantz 1991), in
2 Berge argues that absolutive subjects can also be topics but we focus here on absolutive objects which are uniquely associated with double or transitive agreement. Related ideas, that some sort of discourse familiarity is relevant, are found in Kalmár (1979) and Manga (1996). 3 Berge (2011) refers to this type of topic as local topics and distinguishes them from global topics, which are persistent across a wider domain of discourse. 4 Our analysis is framed within the phase theory of Minimalist program (Chomsky 2001, 2008, 2013, among others). A phase in this theory refers to a syntactic derivational unit which is in a technical sense syntactically and semantically complete. The notion thus roughly corresponds to the notion of cycle in the earlier stages of generative grammar.
Morphosyntactic reflexes of information structure in Inuit 399 a way familiar from case alternations in English double-object constructions. Finally, since topics create the illusion of wide scope (Endriss 2009), and have ‘maximization’ properties, this allows us to explain the scope observations and the appearance of an aspect-based split (Borik 2002, Filip and Rothstein 2006, Filip 2008, Ramchand 2008). For reasons of space, portions of our analysis will only be briefly outlined. We hope, however, that the overall structure of the argument will emerge with sufficient clarity. Aside from this language-specific exploration, our proposal contributes to a more general debate on the nature of cross-linguistic differences in argument alignment. We argue that even though the underlying factors (e.g., agentivity, topic/comment, given/ new) may vary among languages, once we isolate the critical alignment factor, the rest of the system (locality, movement properties, case assignment properties) may be stated in universal terms (see also in this volume Baker and Bobajlik (Chapter 5), Du Bois (Chapter 2), Cheliah (Chapter 38), and Coon and Preminger (Chapter 10), for a similar position).
17.2 The Core Facts 17.2.1 Double Agreement as Marker of Ergativity In the Inuit language, a transitive verb may appear in two distinct patterns: the so-called ergative and the so-called antipassive.5 They differ not only in their case assignment but also in their agreement properties. As for the case properties, the subject of antipassive is morphologically marked as absolutive, while the object is marked with an oblique case (mik). In the ergative pattern, the subject is marked with the so called relative case6—a term that corresponds to ergative in the description of other ergative languages—and the object as absolutive. Note that while the relative case has an overt morphological realization, absolutive is zero marked. In this chapter we will use the term relative when referring to the case, and ergative when referring to the ergative clause pattern (including case and agreement of the transitive verb). As for finite agreement, the ergative pattern is associated with an agreement pattern which we will call here transitive double agreement. Unlike its intransitive counterpart, given in (1a), the transitive verb in (1b) inflects for person and number of two arguments, the subject and the object.
5
By antipassive we refer both to transitive verbs which require an antipassive morpheme to show intransitive agreement and to transitive verbs that do not require such a morpheme to show intransitive agreement. 6 Formally, the morphological form of relative case is identical to possessive case found in the nominal domain.
400 Alana Johns and Ivona Kučerová (1)
Baker Lake (Johns 1992, 58-59) a. angut ani- juq man-abs.s walk-INTR.PArt.3s ‘The man is walking.’ b. arna-up angut kuni-ga-a woman-erg.s man.abs.s kiss-tr.part-3s/3s ‘The woman kissed the man.’
(single)
(transitive double)
The antipassive pattern (Spreng 2012), sometimes also called semitransitive (Fortescue 1984), triggers an agreement pattern, which we could call transitive single agreement.7 Abstracting away from other properties of antipassives, the fact that interests us here is that the verb agrees only with the subject. This is seen in example (2) from Labrador Inuttitut. (2) Margarita Kuinatsa-i-juk Ritsati-mik Margarita.abs.s tickle-ap-intr.part.3s Richard-mod.s ‘Margarita is tickling Richard.’ (transitive singular) Labrador (Johns 2001b, 211) The description of the system is slightly complicated by the fact that the Inuit language is a pro-drop language. As a result, in many instances there is no overt DP, i.e. there is no overt case marking. In addition, the plural of the relative case is homophonous with the plural of the absolutive. Consequently case distinctions are less overt than agreement distinctions, which we contend are central to transitive constructions. Furthermore, there appears to be a dialect difference between Eastern and Western dialects of the Inuit language in Canada: while in Western dialects, objects may be overt DPs in either of the two transitive constructions (double or single agreement), in Eastern dialects, the object DP of a transitive double agreement construction is not usually overt.8 Finally, the ergative pattern is significantly less frequent in Eastern dialects than Western dialects. While we will have nothing to say about the frequency effect, a closer look at the morphosyntactic properties of the double pattern in section 17.2.2. will shed light on why absolutive object DPs might be dropped in one dialect group but not the other.
17.2.2 Double Agreement as Cliticization The intuition behind this and previous accounts is that while an object in the ergative pattern is a core argument and therefore able to relate to the verb directly, an object in 7 We will continue to use the term antipassive to refer to constructions with subject/object but only single agreement. 8 This distinction is not absolute and requires more research. See (Johns to appear) for more details.
Morphosyntactic reflexes of information structure in Inuit 401 the antipassive pattern is an oblique and must be outside of the domain of agreement. The question that interests us here is what structural representation underlies these morphosyntactic configurations. In order to answer this question we start by investigating the morphosyntactic properties of the agreement properties. Johns (to appear) suggests that objects in the ergative pattern must be salient from the previous discourse. The observation is based on data such as (3) from Labrador Inuttitut. Here we see that the first occurrence of Kajotta ‘cup’ as an object appears in the antipassive pattern, (3a) with single agreement. It is only in (3c) that the same object, this time not overtly realized, triggers double agreement on the verb. (3) The discourse requirement on double agreement (Labrador): a. John kata-i-juk Kajotta-mik John.abs.s drop-ap-intr.part.3s cup-mod.s b. amma-lu Kajottak siKumi-mmat, also-and cup.abs.s break-caus.3s c. âkKi-sima-janga nipi-ti-guti-mmut fix-perf.-tr.part. 3s/3s adhere-cause-instrument-instr.s ‘John dropped the cup and then when the cup broke, he fixed it with the glue.’ Johns (2013, to appear) uses this observation to argue that the transitive double agreement is not a result of ϕ-feature agree.9 Instead she proposes that object inflection is a clitic, following recent reanalyses of object agreement (Preminger 2009; Nevins 2011; Kramer 2014a). Since clitics require some form of saliency—her argument goes—the information structure restriction demonstrated in (3) immediately follows. Furthermore, an analysis of Inuit object inflection as clitic immediately affords a perspective on the dialect difference introduced in 17.2.1 as a familiar difference in clitic doubling (Anagnostopoulou 2006). As Anagnostopoulou shows, languages differ in whether or not the clitic’s full DP associate is overt. In other words, while all clitic languages morphologically realize the clitic, only some realize simultaneously both the clitic and the doubled DP. Thus if we analyze the object agreement as a clitic, we can explain the restriction on overtness of object DPs as a dialect variation in clitic doubling: while Western dialects allow clitic doubling and hence double agreement with overt object DPs, clitic doubling in Eastern dialects is severely limited, so we expect to see double agreement only when there is no overt DP.10 We follow Johns’s proposal and provide additional evidence that agreement with the object of the ergative construction contains a clitic (contra Compton 2014).11 Our 9 A combination of pro and ϕ-agree analysis was proposed by Merchant 2011 for Aleut, a distant relative of the Inuit language. 10 Interestingly, this entails that Labrador Inuttitut is no longer ergative by definition, since transitive clauses and ergative patterning are not equivalent. 11 Though in section 17.2.3 we will disagree with Johns (to appear) on her information-structure characterization of objects in the ergative pattern.
402 Alana Johns and Ivona Kučerová argument is based on the discussion of ϕ-agree versus cliticization in Nevins (2011). Nevins argues that clitics, unlike ϕ-agree, are tense invariant, display gaps in morphological paradigms, are reminiscent of PCC effects, and have omnivorous number. As we will see, even though there are non-trivial issues in determining whether double agreement is tense-invariant, the double agreement pattern in the Inuit language displays the other two properties characteristic of cliticization. An additional argument will come from the domain of semantic interpretation (Dočekal and Kallulli 2012). Let us start with the question of tense invariance. As Compton (2014) points out, even though there is no tense-sensitivity in the Inuit language, there is a morphophonological sensitivity to mood. For example, transitive participial (declarative) mood inflection for 2s/1s is jar-ma, while transitive interrogative equivalent is -vi-nga. Both ma/-nga indicate first person singular object. While for Compton this sensitivity indicates ϕ-agree, we find problems with this argument. First, tense invariance is not a direct by-product of category (pronoun vs agreement). Nevins (2011) indeed argues that clitichood needs to be based on syntactic, not on purely morphophonological evidence. For the conclusion about tense invariance to hold, it must be a consequence of which type of element is in closer proximity to T. In languages discussed by Nevins, object clitics which undergo object shift tuck in under the subject in spec,vP. They are higher, but still not as close to T as subjects are. As Compton (2014) points out, however, Mood (not T) is the major clausal category in the Inuit language. Furthermore, as Compton and Pittman (2010) argue, a word in the Inuit language is a phase, bounded by mood at the phase edge. Arguments evacuate before the final phase is completed. Inflectional arguments will therefore adjoin to mood. Assuming Distributed Morphology (Halle and Marantz 1993), at the point at which the form of the object pronoun undergoes morphological insertion, it is adjacent to the mood morpheme. It is therefore not surprising if there is some morphophonological sensitivity based on variance between mood morphemes. Because of these complexities, tense variance cannot be used as a determining property for the status of double agreement.12,13
12
There are additional issues; for example while double agreement of independent moods resembles possessive agreement found on possessums, double agreement of dependent moods resembles possessive agreement found when the possessor is itself possessed, as in (26) from the Labrador dialect. (26) anêna-ma mother-poss.1s/3s ‘my mother’s hats’
nasa-ngit hat-3s/p
Finally subject agreement varies between single and double agreement, so it is not inconceivable that this could affect object form. 13 The reader may have noticed that double agreement is not always clearly segmentable into subject and object at the morphophonological level, unlike some other languages where object agreement has claimed to be clitic (Kramer 2014a). We do not view this as an issue, given that we rely on syntactic evidence, following Nevins (2011). Merchant (2011), in his analysis of Aleut, describes the portmanteau effects of clitics, citing similar effects in Basque (Arregi and Nevins 2008).
Morphosyntactic reflexes of information structure in Inuit 403 Let us now turn to the question of Person Case Constraint (PCC) effects, i.e. a restriction on 1/2 person objects in certain environments (Béjar and Rezac 2003; Rezac 2008; among others). The Inuit language displays systematic person constraints within object marking. This is entirely unexpected if the object marker were a genuine instantiation of ϕ-agree. As observed in Johns (1996), the Labrador dialect indeed displays a restriction on the person marking of the object in the ergative pattern. While 1>3 may be found in both participial and indicative moods, (4), 3>1 is possible only in the indicative mood, as in (5).14 (4) a. nigi-jaga eat-TR.PArt.1s/3s b. nigi-vaga eat-TR.ind.1s/3s ‘I ate it.’ (5) a. *taku-jânga see-tr.part.3s/1s b. taku-vânga see-tr.ind.3s/1s ‘He saw me.’ The example in (6), from the South Baffin dialect shows a similar restriction but with a different result. The first example shows a 1>3 transitive participial mood verb form with double agreement. In the second example, however, the *1>3 prohibition forces a single agreement verb form with the object as an independent pronoun (in fact an antipassive construction). (6) a. mali-langa-si-jara follow-going.to-incept.-tr.part.1s/3s ‘I am going to follow him.’ b. Jaani uvannit ikaju-ruma-nngit-tuq John.abs 1s.pro-mod help-want-neg.-intr.part.3s ‘John does not want to help me.’ Compton (2014) argues against a clitic analysis for verbs based on his claim that possessive agreement, which is closely related to transitive double agreement, shows default agreement in the South Baffin dialect in oblique case environments. Default agreement
14 The indicative mood and the participial mood are very close semantically, with the indicative mood usually adding a sense of vividness. Where the participial mood is grammatically prohibited, the indicative mood is used in its place, but without the vividness.
404 Alana Johns and Ivona Kučerová is a hallmark of ϕ-agree (Preminger 2009), so if we accept that possessive agreement and transitive verb agreement are closely related, this would be an argument against transitive verb agreement having a clitic. The crucial data involving possessives is in (7). (7) a. iksivauta-ra chair-1s.poss/s ‘my chair’ b. uvanga iksivauta-nga-nit 1s.pro chair-3poss-ablative ‘from my chair’ In (7a) we see possessive agreement on the head nominal that indicates both features of the possessor and number of the nominal. In (7b) where the nominal is in oblique case, we see a periphrastic construction, but also a third person possessor marking on the head nominal. Compton claims that this is a default possessor marking, leading to his conclusion that possessor/transitive double agreement cannot involve clitics. While the data are interesting, we do not believe it refutes an analysis of transitive verb agreement as involving clitics. Note that the morphology shown in (7b) does not appear on verbs. Furthermore, Yuan (2014) provides critical data that shows that (7b) is an instantiation of a PCC repair, a property consistent with clitics, not agreement (Rezac 2008). The default agreement analysis makes a clear prediction. If the third person possessive marking in (7b) indeed is default agreement, the number marking on the nominal head should always be third person singular. Yuan (2014) shows that the third person marking on the head nominal can be plural as well. The examples in (8), from Yuan (2014), show an absolutive possessed noun on the left, with an oblique case marked possessive on the right. Only the latter has the periphrastic form with possessor inflection ngit- on the possessum, indicating not only that it is a possessed form, but that it is a plural possessed form. (8) qimmi-kka → uvanga qimmi-ngin-nut dog-1s/3p 1s.pro dog-3p-allative ‘my dogs’ ‘to my dogs’ Consequently, Yuan argues that this is not default agreement but simply the reflex of the possessum features. As a result, Yuan characterizes this construction as a PCC effect where the oblique case blocks co-occurrence with first/second person objects. From this perspective the presence of a first person independent pronoun on the right in (8) can be seen as a repair, a property consistent with clitics, not agreement (Rezac 2008). If possessive forms are relevant evidence to clitics in transitive double agreement, then this evidence supports a clitic analysis of transitive agreement. In summary, there is ample evidence of PCC effects involving object marking (and also possessor marking), leading to the conclusion they are clitics.
Morphosyntactic reflexes of information structure in Inuit 405 Let us turn to the third syntactic property associated with object clitic marking in Nevins (2011), namely omnivorous number, where the same plural marking may denote either the subject or the object as plural. While omnivorous number is clear in Aleut (Merchant 2011), it is not as clear in the Inuit language. However consider the partial transitive indicative paradigm from Harper (1974) for North Baffin dialects in (9). (9)
Subject Object 3s 3d 3p
3s
3d
3p
vaa or vanga vaangik vait or vangik vangak
vaangik
vangik
vaat or vangat vaangik vait or vangit
In (9) we see that the regular third plural marker for nominals -(i)t marks not only the 3plural of the object in 3/3p -vait (singular subject) but also the plural of the 3plural agent in 3p/3s -vaat (singular object). Thus the Inuit language shows some evidence of omnivorous number, supporting the clitic analysis. Our final argument supporting the clitic nature of object marking in the Inuit language is that its presence is associated with a special meaning. This property has been acknowledged as a diagnostic of clitichood by Dočekal and Kallulli (2012), Anagnostopoulou (2006), Kramer (2014a), among others. The presence of ϕ-agree never relates to special meaning. For example in Albanian, as Dočekal and Kallulli (2012, p. 117) show, the object of a verb requires clitic doubling in contexts where a topical interpretation exists, as in (10)–(11). At the same time, clitic doubling is prohibited in contexts where the object of the verb is not topic, as in (12). Note that this example involves existential ‘have.’ (10) A: Who read the book? B:
Ana *(e) lexoi librin. Anna cl.acc.3s read book.the
(11) A: What did Ana do with/to the book? B: Ana *(e) lexoi librin. Anna cl.acc.3s read book.the (12) (*I) kishte minj n. gjith. apartamentin. cl.acc.3p had mice in all apartment.the ‘There were mice all over the apartment.’ The data from the Labrador dialect that parallel Dočekal and Kallulli’s examples (10)–(12) demonstrate the topic status of the absolutive objects. As we can see in (13), if the object is a topic, the agreement pattern is ergative (double agreement) and there is no overt DP.15 15
Note that names often do not reflect relative case, but this is not important, as our claim focusses on the single/double agreement distinction.
406 Alana Johns and Ivona Kučerová (13)
A: kina iKalu-mmik iga-sima-jong? who.abs fish-mod.s cook-perf-intr.part.interr.3s ‘Who cooked the fish?’ B: Peta iga-sima-janga Peter.abs cook-perf-tr.part.interr.3s/s ‘Peter cooked it.’
At the same time we see that definite DPs which are not topics are found in the antipassive (single agreement), as in (14). (14) A: Su-sima-jong what.do-perf-intr.part.interr.3s ‘What did Sally do?’
Sally? Sally
B: Sally atua-tsi-sima-juk alla-mik. Sally.abs read-ap-perf-intr.part.3s book-mod.s ‘Sally read the book.’ Since the Labrador dialect does not favor clitic doubling, a topic with an overt DP will also be in the antipassive (single agreement), as can be seen in (15). (15)
A: kina Kuki-nni-jong nanu-mming? who.abs shoot-ap-intr.part.interr.3s polar.bear-mod.s.interr ‘Who shot the polar bear?’ B: Davide Kuki-nni-juk nanu-mmik David.abs shoot-ap-intr.part.3s polar.bear-mod.s ‘David shot the polar bear.’
Kramer (2014a) shows that clitic doubling in Amharic is related to special meaning as well. It can be found with wh-words but only if d-linked, and is also found as a sort of emphasis. Kramer (2014a, p. 624) suggests that topichood may be an underlying factor, but acknowledges that more fieldwork is needed. The Inuit ergative construction is also known to have a special meaning. The exact nature of the meaning difference and the structure underlying varies. It has been attributed to specific (Manga 1996), wide scope (Bittner 1987, 1994), given (Kalmár 1979, Johns to appear), and topic (Berge 2011). The exact semantic nature of the double agreement pattern will be discussed in section 17.2.3 but for now it suffices to say that, that unlike objects of transitive verbs which do not bear special nominal interpretation as a result of being transitive objects, the nominal in absolutive case in the ergative pattern in the Inuit language has a distinct interpretation from that it would get otherwise. We conclude that the semantic properties of the double object agreement yet again point in the direction of a clitic, not ϕ-agree.
Morphosyntactic reflexes of information structure in Inuit 407 To summarize, we have shown that the object markers in the Inuit language clearly display all the characteristics of object clitics. This conclusion supports Nevins’ (2011, p. 967) tentative contention, based on a suggestion by Woolford, that in all languages displaying agreement with both subject and object, object agreement should be reanalyzed as a pronominal clitic.
17.2.3 The Case for Topics The fact that objects in the ergative and antipassive pattern are not semantically equal has not gone unnoticed in the literature on the Inuit language and other ergative languages. In this section we follow Berge (2011) and her analysis of Western Greenlandic and argue that objects in the ergative pattern are best characterized as topics. More precisely, we will argue that absolutive objects which Berge refers to as locals topics correspond to (aboutness) topics in the sense of Reinhart (1981). We will call these topics sentential topics in order to indicate that unlike their broader counterpart, they have a designated clause-bound syntactic representation. Thus our notion of sentential topic will be strictly used as a label for the part of a structure the sentence is about, instead of the broader notion of topic often used in the functionalist literature (global topics or non-topics in Berge’s terminology).16 In contrast, objects of antipassive are best characterized as anti-topics (non-topics in Dočekal and Kallulli 2012), i.e. backgrounded elements.17 In section 17.3 we will show that not only is this characterization empirically more accurate than other previously suggested characterizations but it also straightforwardly derives the core morphosyntactic properties of the Inuit case and agreement system we investigate here. Bittner (1987) and following work observed that absolutive objects in the Inuit language, unlike their oblique counterparts, take wide scope. The example in (16), originally from Bittner (1987, (40)–(41)), demonstrates this through the relation between an indefinite noun phrase ‘kayak’ and sentential negation. In the ergative pattern, as in (16a), the absolutive object scopes above negation, which results in a specific reading. In contrast, whether definite or indefinite, its oblique counterpart in (16b) must be interpreted within the scope of negation (‘He/she doesn’t use a kayak anymore’). While this fact is undisputed, it is not clear what underlying grammatical property is responsible for the scopal interaction. Note the contrast could be a result of a genuine wide scope, but equally it might have arisen from another semantic factor, 16
Berge’s notion of local topic is related to the notion of theme in the Prague school terminology. The notion of global topic, which Berge uses to characterize objects in the ergative pattern, is closer to Schwarzschild (1999)’s notion of givenness. As for aboutness/sentential topics, we use Reinhart (1981)’s formalization—more precisely we will adopt the formal implementation of Endriss (2009)—because it is easier to implement within the generative framework we assume here. We refer the interested reader to Hajičová et al. (1998) for an attempt to reconcile these two distinct generative traditions in semantic terms. As for the information status of subjects, we refer the reader to Berge (2011)’s description. 17 Similar claims have been made by other authors for other ergative languages, mostly in the functionalist tradition (Dixon 1972; Blake 1976; Mallinson and Blake 1981; Du Bois 1987b; Authier and Haude 2012; among others).
408 Alana Johns and Ivona Kučerová e.g. specificity, definiteness, or perhaps givenness modeled as anaphoricity (Kalmár 1979; Manga 1996; Johns to appear).18 (16) a. qajaq atur-unnaar-paa kayak.abs use-no_longer-ind.3s/3s ∃ [x is a kayak & ¬(he uses x)] b. qaanna-mik atur-unnaar-puq kayak.mod.s use-no_longer-ind.3s ¬∃x[x is a kayak & he uses x]
abs: ∃ > ¬
ins: ¬ > ∃ (Hallman 2008, 10)
Let us have a closer look at the hypothesis that the difference between absolutive and oblique objects is indeed based on a genuine scope interaction. If the semantic contribution of the oblique case is narrow scope, then we expect that non-scoping elements such as proper names, personal pronouns, or rigid designators, such as ‘my father,’ should always appear only in absolutive case as they obligatorily scope over the type of scope operators Bittner bases her argument on.19 However, as these Canadian Inuit examples in (17) from Hallman (2008) demonstrate, this prediction is not borne out. Thus the fact that non-scoping items may appear with both case markings is entirely unexpected under a genuine scope hypothesis. The explanation of the appearance of wide scope must lie elsewhere. (17)
a. qimmiq taiviti-mit kii-si-qqau-juq dog.abs David.mod.s bite-ap-past-INTR.part.3s ‘A dog bit David.’ b. qimmiq uvannit kii-si-qqau-juq dog.abs me.mod.s bite.-ap-past-INTR.part.3s ‘A dog bit me.’ c. qimmiq ataata-nit kii-si-qqau-juq dog.abs father-my.mod.s bite-ap-INTR.part-ind.3s ‘A dog bit my father.’
(Hallman 2008, 51)
18 Hallman (2008) proposes that the transitive object is either anaphoric or introduces new arguments because an NP in absolutive case (a) has existential assertion, rather than supposition, and (b) a uniqueness presupposition. 19 The logic of our argument crucially relies on Fox (2000)’s notion of scope economy. (See also Fox 1995; Reinhart 1995, 2006.) According to Fox, nominal expressions can undergo a scope-taking operation, only if the scope-taking structural change (be it quantifier raising or something else) yields an interpretation that would not be available otherwise. Since here, the interpretation associated with the wide scope is already available in situ, there is no reason for a structural change to take place. One could argue that there is another feature present in the structure that is in and of itself independent of wide scope; however, we prefer the simpler analysis presented in the main text.
Morphosyntactic reflexes of information structure in Inuit 409 Could definiteness or givenness, modeled as discourse anaphoricity (Schwarzschild 1999), be a possible explanation, as suggested in Kalmár (1979), Manga (1996), Johns (to appear)? Interestingly, as Hallman (2008) points out, although absolutive case often reflects definite or anaphoric meaning, absolutive object DPs can also appear in out of the blue contexts, as in (18), an example from Kalaallisut (Bittner 1987, (12)), taken from the Gospel according to Matthew in the New Testament. (18) figiqussuar-lu aqqusirnup sania-niit-tuq taku-gamiuk fig.tree.abs-and road.erg its.side-loc-part.3s see-when.3r/3s ‘and as he saw a fig tree standing at the side of the road’ The ‘figtree’ in absolutive case is a new entity in the discourse, assuming the English translation is a guide. This is unexpected if absolutive objects have to be discourse salient. In other words, notions such as given/new, discourse salient do not seem to provide an accurate description of the facts. Neither does the distinction between definiteness and indefiniteness, at least not in the sense assumed for English (see the discussion in Hallman 2008). Furthermore, as Compton (Chapter 34, this volume) argues, oblique case (MOD) cannot be equated with indefinite properties since it is found on names of objects in some dialects and it is found on names under some circumstance in all dialects. The examples given in (17) make the same point. We conclude that neither a genuine scope analysis, nor ones based on definiteness or discourse saliency are empirically accurate. Instead we propose that absolutive objects are sentential topics, following the core proposal of Berge (2011). However, before we demonstrate the validity of this hypothesis, we find it useful to first outline what we mean by sentential topics, as there has been a significant amount of confusion about the semantic contribution of topics and how they relate to notions such as givenness or discourse saliency. As pointed out by Molnár (1993), there are in principle three distinct levels of information packaging (the following formulation is based on Endriss 2009). (i) a level that distinguishes what is said (the comment) from what this comment is about (the topic of the utterance); (ii) a level that differentiates between things that are new to the hearer (the rheme) and things that are already known (the theme); and (iii) a level where the utterance is divided into what is important or relevant from the speaker’s viewpoint, i.e. the focus, and what is not as important and thus constitutes the background. We concur with Reinhart (1981), Molnár (1993), and Endriss (2009), among others, that a sentential topic is what a sentence is about.20 Thus this is a notion that strictly operates at the sentential level and may be directly encoded in the morphosyntactic representation. From the semantic point of view, this means that for something to be a sentential topic, it needs to be associated with a referential address which in and of itself is
20
The distinction between topic and comment thus syntactically corresponds to the distinction between thetic and categorical statements (Kuroda 1972; Ladusaw 2000; Basilico 2003; among others). We will return to this distinction in section 17.3, where we discuss the syntax of topics.
410 Alana Johns and Ivona Kučerová associated with the common ground. The common ground association may come about in two distinct ways: either the item is already in the common ground, or its existence needs first to be asserted, and then added to the common ground. The final interpretation results once predication takes place over the anchored item.21 A side point, which is going to be important in section 17.3 is that when the address is being established, a ‘maximized’ interpretation is necessary (Endriss 2009). The crucial point for us is that topics must be associated with a referential address, i.e. they are modeled as an address for the context update. The consequence of this is that topical material cannot be interpreted in the predicative part of the sentence. As Endriss (2009) carefully explores, this results in an appearance of wide scope and/or specificity, often associated with topics. Note that the notion of referential address often coincides with familiarity/discourse saliency but it is not identical. Thus this narrow notion of topicality is compatible with indefinites and other non-salient elements such as quantifiers. In contrast, the broader notion of topics, often used in the functionalist literature, requires topics to be recently mentioned, persistent in subsequent discourse, etc., and as a result is not a good fit for the data we investigate here. We are now in a position to come back to our Inuit language data. Recall that we have seen that absolutive objects take wide scope (although oblique objects may take wide scope as well), and we have also seen that although absolutive objects tend to be definite and discourse salient, they can be novel as well, as in the example in (18). While this example was problematic for the hypothesis that absolutive objects are discourse salient or given, the example is consistent with the object being a topic. Under such analysis, ‘figtree’ may be understood as ‘cataphoric,’ i.e. being a sentential topic, with the rest of the utterance being its comment (see also Berge 2011, who describes an example of double agreement as cataphoric). A topic analysis will allow absolutive objects to be new, but only if the rest of the utterance predicates over them (e.g. predicates of appearance on the stage or noteworthy in the sense of Ionin 2006). The English example in (19) demonstrates this type of interaction: strictly speaking, the definite DP this guy is discourse new, however, the demonstrative form indicates that the definite DP is going to be commented upon (see Ionin 2006 for a more detailed discussion). (19) I entered the subway and there was this guy. a. #I was lucky to find a seat and I read a book until we reached my station. b. He immediately started talking to me. In the same way, the fact that the ‘figtree’ will and must play a role in the narrative is indicated through the presence of absolutive case. To summarize, we argue that all the facts we have seen so far are compatible with absolutive objects being sentential topics. What about oblique objects? According to Berge (2011) they are anti-topics. What that means is that they are part of the comment structure of an utterance. If this is correct, 21
For a formal implementation, see Endriss (2009, 245).
Morphosyntactic reflexes of information structure in Inuit 411 we expect that they will take narrow scope, unless their lexical content will make them scope outside of the predicative structure. This is precisely what we have seen in (16) and (17), respectively. Furthermore, even though comment is often new information, it is compatible with a given/discourse salient element as well, as long as the given element is not the sentential topic. This prediction is borne out as well, as seen in (20). Here, ‘Peter’ is the subject of the first sentence, i.e. a global topic if we were to use the terminology of Berge (2011). However, in the following utterance, ‘Peter’ occupies a non-topic position and even though it is clearly discourse-salient/given, it appears as an oblique adjunct. (20) a. pita-up qukiq-qau-ngit-tanga peter.erg shoot-past-neg-TR.part.3s/3s ‘Peter didn’t shoot a seal.’
natsiq seal.abs
b. qimat-si-qqau-juq pita-mit qukiuti-nga-nit flee-ap-past-INTR.part.3s peter.abl gun-his.mod.s saku-li-gasua-liq-tillugu cartridge-make-try-prog-conj.4/3s ‘It fled from Peter while he was trying to put a cartridge in his gun.’ (Hallman 2008, 46) To summarize, all these properties are compatible with the topic/comment distinction. Further supporting evidence for this conclusion comes from Murasugi (2014). Murasugi conducted a behavioral study showing that 1/2 person objects are more likely to be in absolutive case than in oblique case.22 This finding is compatible with the idea that oblique objects are comment, while absolutive objects are topics, as the speaker/ hearer is more often the topic of a sentence than the comment.
17.3 Putting the Pieces Together Let us summarize what we have learned so far. First, the Inuit language exhibits an information-structure driven case and agreement split. More precisely, the object of a transitive structure may be in absolutive or oblique case. If it is in absolutive case, then the object must be a sentential topic. If it is in oblique case, then the object must be in the comment part of the structure. Other semantic effects associated with absolutive objects, such as wide scope or the tendency to be definite or discourse salient, are a direct consequence of sentential topics being associated with a referential address. Furthermore, we have shown that if the object is sentential topic, then it triggers double agreement on the verb. Crucially, the object marker is not a 22
This finding needs to be compared with seemingly contradictory restrictions, such as those in (4)–(6).
412 Alana Johns and Ivona Kučerová morphological exponent of ϕ-agree (pace Compton 2014). Instead, it is a nominal clitic, adjoined to the verbal complex. In this section, we will use the sentential-topic analysis of absolutive objects to explain the inflectional properties and case properties of the Inuit split. As for double agreement, our findings confirm other proposals that argue that clitic doubling is always conditioned by information structure, including object markers on verb (Anagnostopoulou 2006; Kramer 2014a; among others). Specifically, we follow the proposal made in Dočekal and Kallulli (2012) who argue that only sentential topics trigger clitic doubling.23 Thus under the clitic analysis of the verbal object marker, the fact that only absolutive objects trigger double agreement is unsurprising, since all absolutive objects are sentential topics. This conclusion also straightforwardly derives another fact, namely, the dialectal difference between Eastern and Western dialects. As we discussed in section 17.2.1, while in Western dialects objects may be overt in either transitive pattern (ergative or antipassive), in Eastern dialects, the object DP of a transitive double agreement construction is not usually overt. As Anagnostopoulou (2006) and work cited there shows, there is indeed a large body of cross-linguistic and dialectal variation in the domain of clitic doubling which affects whether or not the full DP is overt. We argue that the dialectal variation attested in the Inuit language dialects can be subsumed under this common variation in the domain of clitic doubling. We thus conclude that the topic analysis of absolutive objects captures not only their semantic properties, but it also provides an insight into the agreement split and the nature of dialectal variation associated with it. We will now turn to the more fundamental question underlying the current discussion which asks what it is about absolutive objects that requires them to be interpreted as sentential topics. We argue that this property is another direct consequence of their semantic import, that is of their requirement to be associated with a referential address. Specifically, we argue that the referential requirement forces sentential topics to be at the edge of a phase since otherwise they would not be accessible to the interpretive module (Chomsky 2000, 2001, 2013; among others). The associated XP movement changes locality properties for the purposes of case assignment, and in turn yields a change in the morphological realization of direct objects. Rizzi (1997, 2004), Grohmann (2003), among others, established that topic-like interpretations, including contrastive topics,24 move to CP, i.e. the edge of the CP phase. Interestingly, there is evidence that even within vP, topics must move to the phase edge 23 Dočekal and Kallulli (2012) closely follow the formalization proposed in Endriss (2009). The core of their argument comes from the observation that only a very specific type of quantifier (for the semantically savvy reader, only those that can be mapped on a minimal witness set) can function as sentential topics—and consequently can be clitic doubled. Crucially, no other information-structure dimension, e.g., familiarity, correctly identifies the right group of quantifiers. 24 Note that even though contrastive topics by name resemble sentential topics, semantically they are quite different, as contrastive topics bring about focus interpretation. See, for instance, Kučerová and Neeleman (2012) for syntactic consequences of the additional semantic import.
Morphosyntactic reflexes of information structure in Inuit 413 as well. This has been most convincingly shown for German by Frey (2000) and related work. Furthermore, we know that even if topics stay lower in the structure, they must move to the edge of the phase at LF (see, for instance, Polinsky and Potsdam (2001) for intriguing evidence from long-distance agreement). The question is why this should be so. We argue that the obligatory movement to the edge of the phase is a direct consequence of topics requiring to be associated with a referential address. We argue that since the referential address is created through anchoring to the common ground in the interpretive component (CI), topic movement makes items accessible to minimal search by the CI component at the point of Transfer (Chomsky 2008, 2013, Narita 2011; cf. von Fintel 2004).25 While in languages like English, if there is more than one argument, sentential topics coincide with grammatical subjects, and consequently, the relevant locality domain is CP, in the Inuit language sentential topics coincide with objects, and consequently the relevant locality domain for establishing referential anchoring is vP.26 This conclusion raises a non-trivial issue. If we take seriously the semantic analysis of sentential topics such as that of Endriss (2009), then sentential topics are first anchored, and then they are predicated upon by the rest of the structure. If sentential topics were anchored at CI before CP is built (which follows from vP being sent to Transfer), and only then are they predicated upon, the resulting derivation would be counter-cyclic. We argue that the solution lies in an intuition that underlies much of previous and existing research on ergative languages, namely, the idea that ergative systems are at some level of abstraction passive or unaccusative structures (Fillmore 1968; Hale 1970; Marantz 1984; Bok-Bennema 1991; Johns 1992; among others). Technically, what this amounts to is that an ergative subject is not merged in the same position as that of the external argument in nominative/accusative systems. The reason is that either υ is entirely missing (L. Nash 1995, 1996), or it is defective (Bok-Bennema 1991; Johns 1992; Alexiadou 2001). We can rephrase this conclusion in terms of phases and their Transfer (Chomsky 2001, 2008).27
25 Minimal search is the current Minimal Program formalization of the idea that syntactic objects are accessible to further operations only if they appear at the edge of their local domain, cf. the Phase Impenetrability Condition of Chomsky (2000) or the concept of specified subject in the GB framework. The main difference is that the minimal search allows for feature interactions across modules, i.e. not only between syntactic domains. See Narita (2011) for an extensive exploration of consequences of this very property. 26 Note that if an internal argument is to be interpreted as a sentential topic in nominative/accusative languages, the structure often undergoes a significant change to yield the necessary alignment. For instance, in English, topical internal arguments are typically realized as subjects of passives, while in languages like German or Czech they are A-scrambled to the edge of their local domain. See, for example, Kučerová (2007) and references cited therein. 27 Note that a similar generalization underlies Compton and Pittman (2010)’s proposal that for the purposes of the morphophonological mapping, CP is the smallest accessible domain in the Inuit language.
414 Alana Johns and Ivona Kučerová (21) Absolutive topic generalization: For an internal argument to be minimally searchable and anchored to the common ground, it must be sent to Transfer at the propositional level (CP), i.e. vP phase is not sent to Transfer prior to the completion of CP.28 Note that this conclusion is complicated by the fact that the lower structure of the transitive clause in the Inuit language is not a familiar vP of nominative/accusative languages, but either an applicative29 or a nominalized structure (Johns 1992 and much following work; cf. also Alexiadou 2001 and Chapter 15, this volume and Haig, Chapter 20, this volume). Recall that to achieve the topic/comment sentential partition, the topic must be asserted and the comment subsequently must be predicated over it. According to some authors, it is this very partition that underlies the distinction between so-called categorical and thetic statement distinction (Kuroda 1972; Ladusaw 2000; Basilico 2003; among others). Here we adopt the view advocated in Basilico (2003), namely, that the topic/comment structure requires a syntactic partition (see Diesing 1992 and following work for the idea that semantic partitions map directly on syntactic structure). As Basilico points out, while the topic interpretation of a subject can be achieved by raising the subject from vP to TP, it is not immediately clear how to create the same syntax-semantics partitioning effect with an object. We argue that for an internal argument to be interpreted as a sentential topic, it must raise from its base-generated position to a higher functional projection within the same phase—analogous to the raising of subjects. We further argue that for such a movement to be possible, the internal argument cannot be merged within a simple VP projection. Instead, the VP part of the structure must be more complex in order to facilitate the required raising. We follow a suggestion made in Basilico (2012) and argue that the internal argument is merged in an applicative-like structure. More precisely, the internal argument is merged as a sister of a low applicative head (Pylkkänen 2002). If the internal argument is not a sentential topic, then it remains within the applicative projection. If, however, it is to obtain a sentential topic interpretation, it must raise to the specifier of VP—analogous to the raising of topical subjects from the specifier of vP to the specifier of TP. The trees in (22) schematize the basic structural distinction between the ergative and the antipassive pattern. Note that the truncated structure does not provide a direct representation of the nominalized character of the vP/VP part, neither does it contain higher functional projections, such as MoodP and TP.30
28
Meaning there is no v at all, or it is defective. For a recent discussion of applicatives, see Carrier (2014). 30 Notice that our applicative structure introduces an internal argument of the direct object sort, unlike high applicatives proposed for indirect objects in the Inuit language (Carrier 2014). Though we do not fully understand the relation between nominalizations and applicative structures, it may be that the reason that the Inuit language has only high applicative structure for indirect objects is because there is a competing low applicative structure which introduces direct object. 29
Morphosyntactic reflexes of information structure in Inuit 415 (22) a.
Ergative pattern: vP v
VP INTERNAL ARG
V’ V
ApplP Appl
t
b. Antipassive pattern: vP v
VP ApplP
V
Appl
INTERNAL ARG
What does this structural difference entail for case assignment? We argue that the structural difference is similar to that of Dative shift in English. If the object remains within the applicative projection, its case is determined by the applicative head. The result is an oblique case (MOD). In contrast, if the internal argument raises to the spec,VP, it can be assigned case by whatever the appropriate structural case assigner is.31 As a result, the internal argument surfaces as absolutive. (23) The Dative shift analogy a. Give money to him. b. Give him money.
∼ antipassive pattern ∼ ergative pattern
Note that we assume a Distributed-morphology style of case assignment (Marantz 1991); that is, the morphological realization of case reflects the morphological-case hierarchy, and hence only indirectly the underlying syntactic structure. Furthermore, after 31
It is possible that the corresponding case assigner is v. However, since v is defective, it depends on the implementation of defectiveness whether or not it might act as a case assigner.
416 Alana Johns and Ivona Kučerová the internal argument raises out of the applicative projection, it must trigger cliticization within the higher functional complex. The proposed analysis thus straightforwardly ties the information-structure properties of the Inuit structures and their case realization. Before we conclude, we will shortly return to a couple of outstanding properties, namely, aspectual properties of the ergative and antipassive structures, and dialectal differences in terms of what type of DPs may appear in the antipassive structure. It has been suggested that the ergative vs antipassive split is aspect-based (Spreng 2006, 2010, 2012; Clarke 2009). Even though a careful investigation of this suggestion extends far beyond this chapter, we would like to suggest that the role of aspect is secondary, and in fact it relates to the topic/comment distinction we argue for here. See also Berge (2011). There is a large body of syntax-semantics literature that makes a connection between aspect, more precisely telicity,32 and some form of definiteness (Krifka 1998, Borik 2002; Filip and Rothstein 2006; Basilico 2008; Filip 2008; Ramchand 2008). The basic intuition can be approximated by a comparison of English and Czech strictly incremental verbs (Filip 2008), as can be seen in examples (24) and (25). The readings in (24a) and (25a) are atelic, they are neutral as to whether or not the event in question was completed, e.g. we don’t know if all the apples were eaten. On the other hand, the readings in (24b) and (25b) are associated with a telic interpretation, i.e. all the apples were eaten. (24) a. Peter ate apples b. Peter ate the apples (25) a. Petr jedl jablka Petr ate.imperf apples ‘Peter ate apples.’ b. Petr snědl jablka Petr ate.perf apples ‘Peter at the apples.’ When we closely look at these examples, we see that in Czech the grammatical source of the telic interpretation is the perfective marking on the verbal morphology. Even though the English translation indicates that the object is to be interpreted as definite, the noun phrase itself does not have any definiteness marking. In other words, the aspectual marking in Czech and the definiteness marking must have the same—or a very similar—semantic denominator. We follow Filip and Rothstein (2006) and Filip (2008) and argue that the common denominator is best modeled in terms of the maximization 32 Though the terms perfectivity and telicity may coincide, they are distinct concepts (Giorgi and Pianesi 2001). While perfectivity denotes the inclusion of the event time in the topic time—using the terminology of Klein (1994) and the denotation of Paslawska and von Stechow (2003) and others— telicity determines whether the actual event has been completed.
Morphosyntactic reflexes of information structure in Inuit 417 of the event/direct object. Note, furthermore, that for an event to be maximized, it must be first mapped to a scale. In fact, Basilico (2012) proposes that it is the scalar properties of events which underlie the formation of antipassives in the Inuit language. How does this relate to topics? According to Endriss (2009), a sentential topic requires a maximized interpretation, irrespective of whether it involves a definite, indefinite, or quantifiers.33 We suggest that it is this very property that drives the telic interpretation of the ergative pattern and yields an imperfective-like interpretation of its antipassive counterpart. This is similar to the English telic effect involving the maximization interpretation of the definite argument which we observed in (24b). In other words, there is a direct connection between topical interpretations and aspect, a relation which likely underlies the aspectual properties attested in the Inuit patterns. We thus conclude that the topic analysis subsumes the aspect analysis, and consequently, a direct reference to aspect is unnecessary. The last remaining issue has to do with dialectal variation in the domain of oblique DPs that may appear in the antipassive construction. While in the Western Inuit dialects and Western Greenlandic, the antipassive construction requires the DP in oblique case (MOD) to be indefinite—or more precisely it excludes referential DPs, there appears to be no such restriction in the Eastern Canadian dialects. We find this reminiscent of restrictions on Scandinavian Object Shift (Thráinsson 2001 and literature cited there). Object Shift is A-movement of certain object DPs to the specifier of vP. Even though Object Shift somewhat resembles A-scrambling in Germanic and Slavic languages, it seems to be structurally more restricted (Holmberg 1986, 1999). Furthermore, while some Scandinavian languages allow Object Shift only of pronouns (for example, Danish); others (e.g. Icelandic) allow optional movement of full definite DPs as well. Finally yet other varieties (for instance, some Norwegian dialects, Nilsen 1997) additionally allow Object Shift of indefinites. Furthermore, languages differ as to whether they allow just Object Shift or whether they extend it to allow semantically motivated movement (A-scrambling), as in Icelandic. In contrast, there are languages, such as English that allow topic movement to the left periphery, but their semantic movement within vP is restricted only to certain ditransitive verbs. The full exploration of this possible connection however awaits future research. To summarize, we have argued that the most adequate characterization of absolutive objects in the ergative pattern in the Inuit language is in terms of sentential topics. Even though our analysis has not fully explored all consequences of this hypothesis, we have shown how it derives the major morphosyntactic properties of the ergative and the antipassive pattern, and we have sketched how the topic analysis ties together various seemingly independent observations about the nature of the split, i.e. its information structure properties, scope properties, and aspectual properties. Even though more work needs to be done, especially in the domain of dialectal variation, the proposed
33
The main contribution of Endriss (2009) is that she models topicalized quantifiers to a minimal witness set representation, i.e. the maximal set of which a certain property must hold.
418 Alana Johns and Ivona Kučerová analysis raises various issues about the nature of ergative languages in general. First of all, we suggested that there is a connection between the passive-like properties of the Inuit language and the necessity for the vP phase not to be sent to Transfer prematurely, if the internal argument is to be interpreted as a sentential topic. One question that immediately arises is what is the trigger and what is the consequence. The other question is whether a similar connection might hold in other ergative languages. Another conclusion with possible consequences for other ergative languages is that the topic interpretation might be the source of certain aspectual interpretations associated with the split. Aspect is a common property associated with ergative splits in general, and it is possible that the proper characterization of these splits indeed lies elsewhere. Before one jumps too quickly to this conclusion, though, it is important to keep in mind that aspectual splits tend to involve re-alignment of arguments, which is not what we see in the Inuit language. In other words, aspectual distinctions in this language do not condition whether we get ergative or antipassive patterning. If the connection to other types of ergative splits is real, it is not trivial. Yet in our mind it is worth exploring in future research, as is the relation to nominalized structures, only touched upon here.
Acknowledgments We want to express our appreciation to Elizabeth Cowper and especially to Judith Aissen for comments and suggestions on this chapter. We also wish to thank Saila Michael, as well as Katie E. Winters and other Nunatsiavut Inuit for data. Thanks also to SSHRC for funding this research project in a grant (435-2015-0979) to Johns, Kučerová and Lampe and a grant (435- 2012-1567) to Kučerová.
Abbreviations 1, 1st person; 3, 3rd person; ABS, absolutive case; ACC, accusative case; AP, antipassive; CAUS, causative; CL, clitic; CONJ, conjunctive mood; D, dual; ERG, ergative case; IMPERF, imperfective; INCEPT, inceptive; IND, indicative mood; INSTR, instrumental case; INSTRUMENT, instrument affix; INTERR, interrogative intonation; INTR, intransitive; LOC, locative; MOD, modalis case; NEG, negation; P, plural; PART, participial mood; PERF, perfective; POSS, possessive; PRO, pronoun; PROG, progressive; R, reflexive; S, singular; TR, transitive.
Chapter 18
Ergative c onst e l l at i ons in the stru c t u re of speech ac ts Martina Wiltschko
18.1 Introduction Ergativity is a heterogeneous phenomenon (Bittner & Hale 1994). The surface constellations associated with it indicate that we need to distinguish two structural layers: a layer where arguments are introduced (henceforth argument–structure) and a layer where grammatical relations (case) are introduced (henceforth case–structure) (cf. Williams 2003). The goal of this chapter is to demonstrate that we find similar structural constellations in the layer of structure where speech act relations are introduced (Speas & Tenny 2003; Haegeman 2013). In particular, I argue that speech act structure (henceforth SA- structure) consists of a grounding layer, where the speaker’s or the addressee’s commitment toward the proposition are encoded. The second layer of SA-structure is dedicated to the response system of language: e.g. what the speaker (henceforth S) wants the addressee (henceforth A) to do with the utterance. Each of these layers can come in different guises, in much the same way as argument–structure can be transitive, ergative, or unergative. This chapter is organized as follows. In section 18.2, I introduce ergative constellations in the domain of argument–structure. In section 18.3, I introduce background assumptions regarding the syntax of speech acts. And in sections 18.4 and 18.5 I show that the logic of ergativity can be equally applied to SA-structure. In section 18.6, I conclude.
420 Martina Wiltschko
18.2 The Heterogeneity of Ergativity and the Logic behind It An ergative/absolutive system is typically described as follows: intransitive subjects are marked like transitive objects but differently from transitive subjects. In particular, transitive subjects are marked as ergative while intransitive subjects and objects are marked as absolutive. This contrasts with nominative/accusative systems in which transitive and intransitive subjects are identically marked as nominative while objects are marked differently, namely as accusative (Dixon 1979). While all ergative systems have in common that transitive subjects differ from intransitive subjects in some ways, there are also many ways in which such systems differ from each other. Hence, it has become clear that ergativity is a surface phenomenon that can come about in different ways and it does not come as a surprise that many different analyses have been proposed (see Coon & Adar 2013; Deal 2015, for an overview). From a generative perspective, it is not surprising that surface ergativity comes about in different ways: the key notions that identify ergative constellations are not primitives. That is, transitive subjects, transitive objects, and intransitive subjects are all derived concepts. To understand ergativity from a generative point of view, we need to understand these notions. Within the generative tradition, subjects are not a unified concept. Rather, they are defined across various levels of structure. That is, one of the key insights within the Principles and Parameters framework (e.g. Chomsky 1981) is the distinction between thematic roles and grammatical (case) roles. Thematic roles are assigned to arguments by the verbs that introduce them, while grammatical roles are assigned by functional categories (such as INFL). This assumption accounts for the fact that arguments that are realized as grammatical subjects can bear different thematic relations to their verbs. For example, in the active voice, the agent is realized as the grammatical subject (1)a, while in the passive voice it is the theme or patient which is realized as the grammatical subject (1)b. (1)
a. The dog was catching the ball. b. The ball was caught by the dog.
This type of mismatch between thematic and grammatical roles is precisely the motivator for separating the thematic domain from the domain of grammatical relations and structural case-assignment. The separation between thematic and grammatical roles was not always complete. Rather, initially, it was assumed that in passive sentences the thematic object role and the grammatical object role were both assigned in the complement of V, while the
Ergative constellations in the structure of speech acts 421 thematic subject role and the grammatical subject role were both assigned in SpecIP. The mismatch between these roles observed in passives was analyzed as involving the absorption of accusative case via passive morphology (Baker et al. 1989), which in turn results in the unavailability of the agent role (Burzio 1986). Given the assumption that all sentences must have subjects and that all arguments must receive case, the object argument was assumed to move to SpecIP where it receives nominative case. Hence, this is a situation where the thematic role is assigned in a different position than the grammatical role. (2)
The GB-style analysis of passive IP VP
Subj AG
V' V
TH
Since this first deconstruction of thematic and grammatical roles, however, there have been two seminal assumptions, which resulted in a complete structural distinction between thematic and grammatical roles. These two assumptions are the VP-internal subject hypothesis (Koopman & Sportiche 1991) and the assumption that accusative case, like nominative case, is assigned by a functional category above the VP (Borer 1994, 2005b; Megerdoomian 2000).1 As for the VP-internal subject hypothesis, in its current incarnation it is usually assumed that agents are introduced by a semi-functional head v (Chomsky 1992; Kratzer 1996). Thus, there is a one-to-one correspondence between heads and their arguments. This yields a complete separation of argument–structure from case–structure as illustrated in (3) (Williams 2003; Wiltschko 2014).
1
Not everyone subscribes to this conceptualization of the case domain. In particular, it is often assumed that accusative is assigned by v—the same (semi-) functional head which introduces the external argument (e.g. Chomsky 1995). However, evidence from systems where accusative case- assignment is sensitive to aspectual properties suggests otherwise (Kiparsky 1998a). Even if accusative case is assigned by v, the points about ergativity to follow still hold. What is important is that thematic roles are structurally separate from grammatical roles.
422 Martina Wiltschko (3)
A two-layered system IP
Subj
Case-structure AspP vP
Object
argument-structure
AG
VP TH
Thus, we can identify two notions of subjects: thematic subjects realized in SpecvP and grammatical subjects realized in SpecIP. And similarly, we can identify two notions of objects: thematic objects, realized within VP and grammatical objects realized in SpecAsp. The assumption that both subjects and objects come in (at least) two different types (thematic and grammatical) has important implications for our understanding of ergativity. Note that the classic description of ergativity makes no distinction between thematic and grammatical roles. Once this division is in place, however, we can distinguish between two notions of ergativity. On the one hand, we can define ergativity in terms of thematic relations only. And, on the other hand, we can define ergativity in terms of grammatical relations. In his seminal paper, Perlmutter 1978 identifies two types of intransitive verbs. Those whose sole argument starts out like the thematic object of a transitive verb. These are known as unaccusative verbs. In a nominative/accusative system, such arguments still behave like transitive subjects, at least on the surface. This is because verbs that lack an external argument fail to assign accusative case (hence the label unaccusative). Consequently, the underlying object moves to the position of grammatical subjects, just like the object of a passive verb as in (4)b. In contrast, there is also a class of verbs whose sole argument starts out like the subject of a transitive verb and hence this argument shares some properties with transitive subjects (4)a. These are known as unergative verbs, a label which reflects the fact that the behavior of the contrasting class of verbs (unaccusative verbs) is akin to ergativity. And indeed some scholars refer to unaccusative verbs as ergative verbs (e.g. Burzio 1981; den Besten 1981), a convention I will follow. (4) a. unergative verbs
b. unaccusative (=ergative) verbs
IP Subj
IP VP
AG
Subj
VP
V' V
V' V
TH
Ergative constellations in the structure of speech acts 423 Within the minimalist program, theory-internal considerations lead Chomsky 1992 to adopt an alternative conceptualization of unergatives, namely that of Hale & Keyser 1993 (henceforth H&K). According to H&K, unergative verbs are best analyzed as concealed transitives: they come with an internal argument, which incorporates into the verb as in (5). (5)
Unergatives as concealed transitives vP v'
AG v
VP V
TH
Thus, even languages that are otherwise nominative/accusative show effects of ergativity: in terms of argument–structure properties, intransitive subjects may behave either like transitive subjects or like transitive objects. Since this type of ergativity is exclusively a matter of argument structure, I refer to it as argument–structure ergativity. It contrasts with case–structure ergativity which arises as a matter of grammatical marking. That is, case–structure ergativity is not a matter of how many arguments are introduced at the argument–structure layer, but instead it concerns the marking of those arguments in terms of their grammatical relations (often in the form of case or agreement). In a nominative/accusative system the subject of an intransitive is marked in the same way as the subject of a transitive (i.e. nominative) and the transitive object receives special case (i.e. accusative). In contrast, in an ergative/absolutive system the subject of an intransitive is marked in the same way as the object of a transitive (i.e. ergative) and the transitive subject receives special case (i.e. ergative). Thus, case–structure ergativity has in common with argument–structure ergativity the fact that transitive subjects are special. As discussed at length in this volume, there are three core analyses to account for the difference between ergative and accusative case systems. According to early approaches (Campana 1992; Murasugi 1992; Bobaljik 1993a) case-assignment is correlated with agreement triggered by functional heads. In a nominative system, the higher functional head is always active whereas the lower functional head is only active if there is an argument associated with the higher head. Hence transitive objects receive a different case (ACC) than transitive and intransitive subjects, as in (6). In an ergative system, the lower functional head is always active whereas the higher head is only active if there is an argument associated with the lower head. Hence transitive subjects will receive a different case than intransitive subjects (which remain associated with the lower position), as in (7).
424 Martina Wiltschko (6) nominative system a. [F2 argnom [F1 [arg v [VP V ]]]] b. [F2 argnom [F1 argacc [arg v [VP V arg]]]]
→ intransitive → transitive
(7) ergative system a. [F2 [F1 argabs [arg v [VP V ]]]] b. [F2 argerg [F1 argabs [arg v [VP V arg]]]]
→ intransitive → transitive
Several problems have been identified with this line of approach, including the fact that case doesn’t always correlate with agreement (see for example Baker 2014a) and some of the movements necessary for the analysis are not motivated under minimalist assumptions. The two prevalent approaches in current minimalist theorizing are the dependent case approach and the inherent case approach. According to the former (which has its roots in Marantz 1991), the ergative/accusative contrast follows from a parameter akin to (8). There is a structural case which can only be assigned if another argument appears in the same domain (e.g. see Bobaljik 2008; Baker & Vinokurova 2010, Baker 2014a; Coon & Preminger, Chapter 10, this volume) (8)
The dependent case analysis If there are two distinct NPs in the same clause (“governed by V+I”) then: (i) Mark the lower one with dependent case (accusative) and/or (ii) Mark the higher one with dependent case (ergative). (iii) Otherwise, mark NPs with unmarked/default case (called nominative or absolutive). Adapted from Baker 2014a (3)
According to the inherent case approach, ergative case is treated as an inherent case rather than a structural case and thus correlates with thematic role assignment (see L. Nash 1996; Woolford 1997, 2006; Aldridge 2004, 2008b, 2012b; Anand and Nevins 2006; Legate 2006, 2008, 2012; Laka 2006a; Massam 2006; Mahajan 2012). Hence on this view, ergative case is assumed to be assigned in the domain of argument–structure, much lower than any of the structural cases. In the remainder of this chapter I explore the question as to whether ergative properties are found in SA-structure as well.
18.3 Introducing Speech Act Structure Speech acts are often considered a purely pragmatic phenomenon. That is, traditionally syntax is taken to be the module that regulates the composition of meaningful units of
Ergative constellations in the structure of speech acts 425 language into larger constituents and phrases. The unit of investigation for many syntacticians is typically a sentence expressing a proposition, which is in turn the maximal unit of analysis of many semanticists. There is however a body of research that seeks to incorporate elements of speech acts into the domain of syntax. The purpose of this section is to review this literature. It will serve as the backdrop against which we explore ergative constellations in SA-structure.
18.3.1 Ross’s (1970) Performative Hypothesis Within the generative literature, the primacy of the sentence for syntactic investigations has first been called into question by Ross (1970). He proposes a layer of structure that dominates the root clause and encodes the illocutionary force of a given utterance. Ross takes this layer to be another type of clause, consisting of a predicate (e.g. a verb of saying) a subject (the speaker) and an object (the addressee). This analysis is known as the performative hypothesis because its core insight is that even declarative clauses can be viewed as being performative, i.e. by uttering a declarative S does something, namely performing an act of speaking (see also Sadock 1969, 1974). Ross’s analysis is schematized in Figure 18.1. Deep Structure (interpretation)
Surface Structure (form)
S SAstructure I tell you that
S
Performative Deletion
pstructure I have a dog
S pstructure I have a dog
Figure 18.1 Ross’s (1970) performative hypothesis
The surface declarative clause is analyzed as being embedded in a superordinate structure, which can roughly be paraphrased as I tell you that. Since none of the postulated components of the superordinate structure are overtly marked, Ross (1970) assumes a rule of performative deletion according to which the superordinate structure is deleted deriving the surface form I have a dog.2 If there is no overt marking of the superordinate structure, how do we know that it is still part of syntax, rather than simply being a matter of pragmatic inferencing? Ross (1970) discusses a series of arguments having to do with (i) the presence of a higher 2
This analysis is couched within a framework that took Deep structure (DS) to be the input for interpretation.
426 Martina Wiltschko first person subject (the speaker); (ii) the presence of a verb of saying above the matrix clause; (iii) the presence of a higher 2nd person indirect object (the addressee); and (iv) the possibility to modify the performative clause. If there is indeed another layer above the root of the clause, and if this layer can be characterized in terms of another predicate–argument–structure, then we expect to find the hallmark of ergative constellations. However, before we can explore whether this is indeed the case, we need to be sure that we have an adequate framework within which to pursue this question. That is, it is well known that Ross’s (1970) analysis faces serious problems, which were sufficient for the field to reject the performative hypothesis (Anderson 1971; Grewendorf 1972; Fraser 1974; Gazdar 1979; Mittwoch 1977; Newmeyer 1986). However, the arguments Ross presented, as well as the general insight that speech acts ought to be syntactically represented didn’t die out completely. Its revival is in part made possible by the discovery of functional categories. That is, much like the insights of generative semantics into the decomposition of events has been reanalyzed in terms of postulating a series of (semi-)functional categories within the VP-domain (e.g. see Ramchand 2011) the decomposition of speech acts has been reanalyzed in terms of postulating a series of functional categories above the CP domain. Among the categories postulated by different scholars we find PragP (Hill 2006), SpeechActP (SAP; Hill 2007a, 2007b; Krifka 2013), AttitudeP (Paul 2014), and PartP (Haegeman 2014; Zu 2015). I turn to the most prominent incarnation of this type of analyses in the next subsection.
18.3.2 Functional Categories in the Speech Act Domain The current revival of the performative hypothesis—let us call it the neo-performative hypothesis—presents some overt evidence for the existence of SA-structure. In particular, recent syntactic analyses focus on units of language that directly encode SA- structure. These include evidential markers (Speas & Tenny 2003), sentence-peripheral particles (Haegeman 2013), Vocatives (Hill 2013), and response particles (Krifka 2013), among others. These units of language are incorporated into syntactic structures in ways that indicate the workings of the syntactic component: they display linear ordering restrictions, pronominalization patterns, scope effects, and agreement patterns. For reasons of space, I limit the discussion to agreement effects. As mentioned in Ross (1970), some languages display S-agreement in gender. Thai is such a language as shown in (9). The sentence peripheral marker khráp is used with male speakers while kâ is used for female speakers. (9) a. Khaw maa khráp. he come spkr=male ‘He is coming.’
Ergative constellations in the structure of speech acts 427 b. Khaw maa kâ. he come spkr=female ‘He is coming.’ (from Oyharçabal (1993) presented in Miyagawa (2012): ex 5) On the assumption that agreement is syntactic (Chomsky 1981, 1995, 2001) S must be represented in the clausal architecture (see Giorgi 2009 for explicit arguments that S coordinates are represented in the CP-layer). And similarly, there are languages that display A-agreement (i.e. allocutive agreement; Miyagawa 2014). For example, in the Basque examples in (10), the sentence-final auxiliary differs in form depending on the gender and number of A as well as the nature of the relation between S and A such that formal status is explicitly encoded (10). (10) Basque A-agreement a. Pettek lan egin dik Peter.erg work.abs do.prf aux.2masc ‘Peter worked.’ b. Pettek lan egin Peter.erg work.abs do.prf ‘Peter worked.’
din aux.2fem
c. Pettek lan egin Peter.erg work.abs do.prf ‘Peter worked.’
dizü aux.2formal
d. Pettek lan egin Peter.erg work.abs do.prf ‘Peter worked.’
du aux.2pl (Miyagawa 2012: (8), cited in Oyharçabal 1993)
Again, if agreement is syntactically conditioned, it must be the case that A is represented syntactically (see Zanuttini 2008). In sum, evidence from agreement patterns lends support to the syntactic encoding of speech act participants. Thus, despite the problems that the performative hypothesis faces, there are still empirical generalizations that need to be accounted for. This is precisely what neo-performative hypotheses seek to achieve: they capture (among other things) the evident presence of speech act participants in the syntactic structure but they are not vulnerable to the same criticism as Ross’s original proposal (see Speas & Tenny 2003: 338 for explicit comparison). While Ross (1970) took the superordinate structure to be a run-of-the-mill matrix clause of the type S[NP VP], neo-performative analyses take the superordinate structure to be an extension of the functional projection of the clause. For example, Speas & Tenny 2003 propose a complex speech act phrase (saP) consisting of two layers, as illustrated in (11). The higher head introduces S in its specifier
428 Martina Wiltschko position and takes a lower saP as its complement, which in turn hosts the utterance content in its specifier and A (labeled Hearer in Speas and Tenny 2003) in its complement. (11)
Speech act structure in the extended clausal projection saP sa
(Speaker) sa
(Utterance content)
sa* sa* sa*
(Hearer)
This model preserves the main insight of the performative hypothesis, in that it postulates a superordinate SA-structure. While this structure is part of the functional architecture, Speas & Tenny 2003 suggest that it still follows the same logic as argument–structure, adopting the framework of Hale and Keyser 1993. According to this proposal, SA-structure is likened to the double object construction: S serves as the agent of the speech act; the utterance content serves as the theme and A as its goal. Thus, they maintain the assumption that—in the context of an assertion—SA-structure encodes something like ‘I give the utterance to you.’ Speas & Tenny 2003 further argue that the restriction on the number of speech acts we observe across languages follows from the available structural configurations. In particular, interrogatives are derived by a passive-like movement such that the hearer moves above the speaker, and imperatives and subjunctives are defined by their non-finite utterance content. Note that subjunctive is not typically included in the list of speech acts. It appears that Speas & Tenny 2003 take clause-type or mood to be the defining property of speech acts. However, things are more complicated in that clause-type alone is not a reliable predictor of illocutionary force. For example, rising declaratives, extensively discussed in Gunlogson 2003, 2008, are formally declaratives but are associated with rising intonation which triggers a questioning interpretation. Thus, while Speas & Tenny’s analysis of SA-structure doesn’t face the problems of Ross’s (1970), it comes with its own problems (Gärtner & Steinbach 2006). Here I address one potential problem that arises in the context of the present hypothesis. Specifically, given the argument–structure approach toward SA-structure we would expect to find argument–structure ergativity in the domain of SA-structure (see Haegeman 2013). Speas and Tenny 2003 however do not discuss this possibility. And given their particular implementation of SA-structure, this is not surprising. What we would expect based on their analysis is to find two types of intransitive speech acts: those that have an external (S) argument only would parallel unergative predicates while those that have an internal (A) argument only, would parallel ergative (aka unaccusative) predicates. Given that every utterance requires the presence of S, the latter
Ergative constellations in the structure of speech acts 429 option seems non-sensical. And this may be the reason Speas & Tenny 2003 didn’t pursue this possibility. Thus, either ergativity plays no role in SA-structure or else we have to rethink the composition of SA-structure. In what follows, I pursue the latter option showing that there are independent reasons to postulate a different model of SA-structure.
18.3.3 The Complexity of Speech Acts There are two problems with the neo-performative hypothesis as conceived of in Speas & Tenny 2003 and subsequent work. First, there is empirical evidence that A is structurally higher than S (Lam 2014, Heim et al. 2014). For example, Cantonese, has a series of sentence-peripheral discourse particles. Relevant for the present purpose are two such particles. Me13 combines with declaratives and derives a request for confirmation with a negative bias on behalf of S, as in (12)b. Hence, me1 is classified as an S-oriented particle. The second particle, ho2 also derives a request for confirmation, but it introduces an additional meaning component, namely that S assumes that A believes p, (12)c. Hence ho2 is classified as an A-oriented particle (see Lam 2014 for detailed discussion). (12) a. zi3ming4 jau5 fu6ceot1 gwo3 si4gaan3 Jimmy have devote asp time ‘Jimmy has spent time (on the project).’ b. zi3ming4 jau5 fu6ceot1 gwo3 si4gaan3 me1? Jimmy have devote Asp time prt ‘Jimmy has spent time (on the project), has he?’ c. zi3ming4 jau5 fu6ceot1 gwo3 si4gaan3 gaa3 ho2? Jimmy have devote Asp time prt prt ‘Jimmy has spent time (on the project), right?’ Crucially, the two particles can co-occur but their ordering is restricted such that A- oriented ho2 has to follow S-oriented me1, as in (13). (13) Jimmy is the first of a long taxi queue. A taxi is coming, but someone not from the queue opens the door of the taxi saying loudly that he is in a hurry. Everyone in the queue is angry. Jimmy says this to the second person in the queue: a. daai6 seng1 zau6 dak1 gaa3 laa3 me1 ho2 loud voice then okay prt prt prt prt ‘What can one get by just by being loud? I assume you’d agree it’s a valid question, right?’ 3
Numbers indicate tone.
430 Martina Wiltschko
b. *daai6 loud
seng1 zau6 dak1 gaa3 laa3 ho2 voice then okay prt prt prt
me1 prt
Lam 2014: 64 (6)
Assuming that the further to the right a particle appears, the higher it must be in the hierarchical structure (Lam 2014), we have to conclude that the A-argument is structurally higher than the S-argument.4 This suggests that SA-structure is organized differently than assumed under current neo-performative analyses.5 This further implies that the common conceptualization of declarative clauses in terms of I give p to you cannot be on the right track. Interestingly, pragmatic analyses of speech acts have changed since Ross (1970). His view of declaratives reflects the assumption that declarative assertions have the following conditions of use: (14) Conditions of use for declarative assertions (i) S believes the proposition (p) conveyed by her utterance. (ii) S wants A to adopt p into her set of beliefs Bach and Harnish 1979 This view doesn’t capture all declaratives, however. In particular, there are (at least) two ways in which a declarative can be modified. First, S may modify the commitment to p thereby changing what is being said. Much of Speas’ (2003) work explores this type of speech act modification. For example, S can indicate, by means of an evidential marker, that they don’t have direct evidence for the truth of p. Second, S can also modify what s/he wants A to do. According to (14), by uttering a typical declarative, S expects A to believe p and thus asks A to adopt p into their set of beliefs. However, this is not the only thing S can do with declarative clauses. As we have seen above, the addition of a sentence-final particle in Cantonese renders a declarative into a request for confirmation. The particle modifies what S expects A to do with the utterance. Beyssade & Marandin (2006) refer to this aspect of the speech act as the Call on Addressee (henceforth CoA). We observe a similar pattern of speech act modification in English. Consider the difference between a regular declarative (15)a and one that is modified by the sentence-final particle eh (15)b. Given its function, Heim et al. (2014) refer to this type of particle as conformational, and I follow this convention. (15) a. You are leaving. b. You are leaving, eh? 4 An anonymous reviewer points out that the linear order would also be compatible with an analysis whereby particles are head-initial and the utterance content moves above the particles. In this case, S would have to be associated with the higher position, as in Speas & Tenny 2003. However, the interpretation of the particles suggests that the A-oriented one scopes above the S-oriented one: A is asked to respond to S’s belief. 5 See Thoma (2016) for additional arguments based on discourse particles in Bavarian German.
Ergative constellations in the structure of speech acts 431 According to the conditions of use for a declarative assertion (15)a would be used if the conditions in (16) hold. (16) Conditions of use for declaratives (i) Bel(S,p) (ii) CoA: Bel(A,p) Given that (15)a is a statement about A, its discourse conditions are slightly marked. Some special circumstances must hold for S to felicitously tell A something about themselves. For example, S may use (15)a as an indirect command. The conditions of use for the eh-modified declarative in (15)b differ. By uttering (15) b, S does not inform A that p and hence it cannot be used as an indirect command. Rather (15)b is used to confirm that S’s belief is accurate. More precisely, S is not fully committed to the truth of p—though S has a bias toward believing p. I indicate this with a disjunctive commitment to p with opposite polarity values. Boldface on the positive belief is used to indicate the bias. In addition to modifying the commitment toward p, eh also modifies CoA. In particular, with the use of eh, S requests A to confirm her biased belief (17(ii)). (17) Conditions of use for eh-modified declaratives (i) Bel (S,p) ∨ Bel (S,¬p) (ii) CoA: Confirm (Bel (A,p)) In sum, eh appears to modify both S’s commitment as well as CoA. Based on the properties of speech act modifiers such as eh, I propose that there are two distinct layers that comprise SA-structure (see Heim et al. 2014 for more detailed discussion): a layer which is responsible for encoding the commitment of the speech act participants to p and a second layer where CoA is encoded. As indicated in (18), I refer to the lower layer as GroundP and the higher layer as ResponseP. The label GroundP is meant to evoke Clark & Brennan’s 1991 notion of grounding as well as the notion of the common ground (cf. Heim et al. 2014, Thoma 2016, Wiltschko & Heim 2016). (18)
A complex speech act structure RespP CoA Responding
GroundP propostional attitude
S "sentence"
432 Martina Wiltschko Moreover, speech act modifiers may also modify declarative clauses such that they express (S’s assessment of) A’s commitment to p. Consequently, Lam (2014) argues that GroundP comes in two guises: one is relativized to S’s set of beliefs (Ground-S) while the other is relativized to A’s set of beliefs (Ground-A). Accordingly, the two particles introduced in (12) and (13) occupy different syntactic positions: Ground-S (me1) and Ground- A (ho2), respectively. (19)
An articulated grounding layer RespP CoA
groundP
Ground-A
GroundP Ground-S
S p-structure
With this model of SA-structure in place, we can now explore whether the hallmarks of ergativity can be detected in this domain. We shall see that argument–structure ergativity plays a role in both the grounding layer as well as in the response layer. As we shall see, interpretable roles are assigned in each of these layers (grounding and responding roles); hence both structures are more akin to argument–structure than to case– structure and hence we expect to find argument–structure ergativity in both layers. However, before I show that this prediction is indeed borne out, I briefly discuss the relation between p-structure, grounding structure, and response structure.
18.3.4 The Syntax of Assertions Recall that according to Ross (1970), assertions are performative speech acts which encode something akin to “I give p to you.” According to the present analysis, which is based on contemporary understanding of speech acts, assertions are complex moves. That is, in a typical assertion, S conveys their commitment to the truth of p. At the same time, they request from A to adopt the same belief. I assume, following Truckenbrodt 2003, that the latter is encoded by means of a falling intonation contour. Thus, by means of a declarative, S asserts that the proposition is in their belief set (Ground-S) and the assertive falling intonation asks A to do the same thing. This contrasts with rising intonation, which asks A to respond to the utterance. The syntax of typical assertions is schematized in (20).
Ergative constellations in the structure of speech acts 433 (20)
Syntax of assertions RespP CoA GroundP Ground-S
S p-structure
Note that not every utterance needs to include a CoA or an assessment about A’s belief set. That is, S may simply utter a declarative clause to indicate their commitment toward the truth of p, without requesting for A to share this commitment. Consider the example in (21). (21) John runs into Mary and notices that she is walking a young puppy. He knows that she had wanted to get a new dog for a while. He exclaims: You have a new dog! Here John does not wish to inform Mary that she has a new dog. Instead, with the use of this declarative, John lets Mary know that he now knows that she has a new dog. In this way, the declarative clause-type serves as an exclamation. Note crucially that the intonational contour of a non-informative assertion differs from that associated with informative assertions. This indicates that the response structure need not be projected. We can thus explore ergative constellations associated with the grounding structure independent of those associated with the response structure. In what follows I show that we do indeed find ergative constellations in both these structures. I start with a discussion of ergative speech acts based on the grounding structure (section 18.4) and then I explore ergative speech acts based on the responding structure (section 18.5).
18.4 Ergativity and the Grounding Structure Let us consider a basic declarative with Ground-S only. These constructions will serve as the basis relative to which we explore ergativity because they present us with the grounding structure version of transitive predicates. The internal argument corresponds to p-structure whereas the external argument corresponds to Ground-S, as in (22).
434 Martina Wiltschko Transitive grounding structure
(22)
GroundP Ground-S external argument
Ground Ground
S p-structure internal argument
Ground-S thus parallels the initiator of the grounding event, just like in Speas & Tenny’s 2003 conceptualization of declaratives. However, the present proposal differs in the way the role of A is conceptualized. Within the grounding structure A is not viewed as the goal of the speech act but instead it may be understood as a role akin to a causative argument. It won’t play a role in the syntax of ergative speech acts to which we now turn. That is, we expect there to be two types of intransitive grounding structures: unergative and ergative (aka unaccusative) structures.
18.4.1 Unergative Intransitive Structure We expect unergative speech acts to consist of an external argument only, with the utterance being incorporated. I propose that this is the case in imperatives. In particular, I follow Portner (2004) in assuming that an imperative clause-type denotes a property. Since properties cannot function as arguments, the complement is incorporated, as in (23). (23)
Imperatives as unergatives. GroundP Ground-S
Ground Ground
S p-structure
Just as incorporated nouns denote properties rather than arguments, so does the propositional content of imperatives. In addition, I assume that Ground-S associated with
Ergative constellations in the structure of speech acts 435 imperatives contains a set of intentions, rather than a set of beliefs.6 Thus, the clause-type of the complement has an effect on the interpretation of Ground-S. For completeness note that a full-fledged SA-structure for imperatives would also contain the response structure, with a CoA which marks S’s request that A put the same p-content into their set of intentions.
18.4.2 Ergative Intransitive Speech Acts Given the logic of our analysis, we expect that ergative speech acts lack the external argument of the grounding structure (Ground-S) but instead consist of the internal argument only. I propose that this is the case in a speech act known as presentation. This is a type of speech act not often discussed. The term is due to Faller (2002: 16), who argues that sentences with evidential markers have presentative rather than assertive force (see also von Fintel 2003; Portner 2006; Déchaine 2007). According to Déchaine et al. (2014), this speech act has fewer commitments than assertions. In particular, its force is that of presenting a proposition for consideration without making a truth-claim about it. In other words, S puts forth p without committing to the truth of p (Déchaine et al. 2014: 6). According to Faller 2002 this type of speech act is triggered by evidential marking. The example in (24) is one of Faller’s examples from Cuzco Quechua and Faller’s description is as follows: the speaker brings the embedded proposition into the conversation for consideration. That is the current speaker’s speech act is one of presentation of another speaker’s assertion. […]There is no condition that the speaker believes p, and the illocutionary act is that of present (2002: 198f.). (24) Para-sha-n-si rain-prog-3-si p= ‘It is raining” ill = present (p)
Faller 2002: 199 (165)
Within our analysis, this amounts to saying that S does not express a propositional attitude toward the proposition—hence it doesn’t enter into Ground-S. Thus, I hypothesize that a presentative is a clause-type which lacks the external argument of the grounding layer (Ground-S), as in (25). (25)
Presentatives as ergatives GroundP Ground Ground
S p-structure
6
This corresponds to Portner’s (2004) To-Do-list and Han’s (2001) Plan-set.
436 Martina Wiltschko In this way, presentatives are the most basic speech acts: they contain the least amount of structure compared to all other speech act types. This echoes Portner’s 2004 semantic analysis of this speech act, which takes it to be the most basic kind of update.
18.4.3 Summary We have now seen that some of the commonly discussed speech acts (assertions and requests) as well as the less commonly discussed presentation can be analyzed as instantiating the three types of speech acts expected on the assumption that speech act structure parallels argument structure. Without taking into account the second layer above GroundP for now, we can summarize the parallelism as in Table 18.1. Table 18.1 The parallelism between argument–structure and speech act structure Argument–structure
Speech act structure
Transitive predicates
Declarative assertion
Unergatives predicates (qua concealed transitive)
Imperative
Unaccusative (ergative) predicates
Presentation
In what follows I show that similar considerations apply to the response structure.
18.5 Ergativity and the Response Structure In section 18.3.3, I have introduced the idea that the grounding structure is dominated by a dedicated layer that encodes CoA. CoA refers to the aspect of a speech act which can be viewed as requesting a response. Hence I refer to this structure as the response–structure. I propose that the response–structure itself is complex: it consists of a projection that hosts the subject of the response. In line with the convention established for argument–structure, where the head which introduces the external argument is labeled (little) v, I use the label resp for this layer. The internal argument of the response layer hosts the object of the response. The full response structure is illustrated in (26).
Ergative constellations in the structure of speech acts 437 (26)
A complex response-structure respP RespSUBJ
resp resp
RespP Resp
RespOBJ Resp
Comm p
If there is indeed another layer above the grounding structure, we expect to find two ergative constellations in the second layer as well. That is, just as there are two types of intransitive argument–structures and grounding structures, we expect to find two types of intransitive response–structures. In this section I show that this is indeed the case. I argue that transitive response structures are realized as requests for confirmation (5.1). Furthermore, I argue that polar questions are best analyzed as unergative structures (5.2). As for ergative response–structures, I argue that this is instantiated by utterance- initial ja in Austrian German, which marks the utterance as a response (5.3).
18.5.1 Requests for Confirmation as Transitive Response Structures In a transitive construal, we expect that the utterance is marked for who is the responder as well as what it is that the responder is supposed to respond to. I argue that this is precisely the configuration we find in confirmation requests marked by sentence-final confirmationals, such as in (27). (27) You are leaving now, huh? I assume that the particle huh marks what S wants A to respond to (namely her believe that p). Thus, huh marks the utterance as the object of the requested response. In this way, Resp has a similar function as v in that it marks the sister of its complement as an object (in this case the object of the response). I further assume that the rising intonation realized on the confirmational encodes CoA (see Heim et al. 2014 for more detailed
438 Martina Wiltschko discussion). I further assume that A is associated with the specifier of respP by external merge.7 This is illustrated in (28), where rising intonation is represented as /. Transitive response structure
(28)
respP resp
Adr resp-S /
Resp Resp-O huh
GroundP
Evidence that the sentence-final particle huh marks the object of the response stems from the fact that there are different types of confirmationals. While all of them are used to request confirmation, they differ according to what S wants confirmation for (i.e. what A is asked to respond to). Consider the examples in (29). (29) Mary has been planning to go on a trip for a while but she had never set a date. One day, she decides to pack and hop on the next train. Her roommate John witnesses her packing. a. John: You are leaving now, {eh/right?} b. Mary: I’m leaving now, {eh /#right?} Consider first (29)a. In this context, John is pretty certain that Mary is leaving, though there is a chance that she is just packing her clothes to bring them to the dry-cleaner. John can use (29)a to request confirmation of his belief that Mary is leaving. This differs from (29)b. Mary knows very well that she is leaving, and she is pretty certain that John also knows. However, there is a chance that John may think she is just preparing her clothes for the dry cleaner. Here Mary can use (29)b to confirm that John knows that she is leaving. In other words, S is requesting confirmation that her belief about A’s belief is correct. In both utterances, eh is used to request confirmation. The request itself is encoded by the rising intonation on the particle. However, the utterances differ in what they request to be confirmed. Eh has two possible uses. It can be used to request
7
It is for this reason that response–structure cannot be equated with case–structure. In the domain of case–structure, argument positions are filled by internal merge rather than external merge (with the exception of expletives).
Ergative constellations in the structure of speech acts 439 confirmation of S’s belief regarding p but it can also be used to request confirmation of S’s assumption about A’s belief. Thus, eh marks the propositional attitude of S or A as the object of the response. In contrast, right doesn’t mark the propositional attitude toward the proposition as the object of the response but instead the proposition itself. In other words, with the use of right, S requests confirmation that a given proposition is true. Therefore, the use of right cannot be relativized to what A believes. In sum, while requests for confirmation share several aspects of meaning with regular questions (i.e. both require a response and thus involve a CoA) they also differ from regular questions, as I discuss in the next subsection.
18.5.2 Polar Questions as Unergative Response–Structures Given the complex SA-structure in (26), we expect that unergative response–structures mark the subject of response only. I argue that this is the case in standard questions marked with rising intonation. As in requests for confirmation, rising intonation marks the subject of the response (i.e. CoA), but the object of response is not explicitly marked. Hence, questions are compatible with contexts where S is asking for information because s/he doesn’t know the answer as in (30)a. But questions are also compatible with contexts in which S knows the answer. This may be in contexts where S wants to test A and hence inquires whether A knows the answer, as in (31). Similarly, in (32) S knows perfectly well that A does not know the answer. This question is only asked to solicit a response from A. (30) Mary has been planning to go on a trip for a while but she had never set a date. Her roommate John wants to know whether she is still planning to go. He asks: Are you still going on your trip? (31)
History teacher to student in grade 5: Did Columbus discover America?
(32) “You know who lives there? Huh? No you wouldn't know who lives there I’m just saying, and you know who lives there?” (scene from Martin Scorsese’s Taxi Driver) In sum, I argue that yes/no questions with rising intonation are used to mark the utterance as one that requires a response. That is, Ground-S indicates that the p-structure of the utterance is in the set of questions (recall from section 18.4.1 that the clause-type determines the relevant discourse component in GroundP). Furthermore, I assume that rising intonation associates with respP to mark that A should respond (Heim 2015). Since the object of the response is not explicitly marked (unlike in requests for confirmation marked by tags) this corresponds to an unergative intransitive constellation, as shown in (33).
440 Martina Wiltschko (33)
Unergative intransitive response structure
respP resp
Adr resp-S /
GroundP Ground-S
S p-structure
18.5.3 Ergative Response–Structure: Marking the Utterance as a Response Finally, we also expect to find utterances in which only the object of response is marked but not the subject. I argue that this is the case in utterances which are themselves marked as a response without requesting a response. Note that exclamatives are S-oriented speech acts: no response is requested. Nevertheless, exclamatives can still be embedded under the response structure. In particular, in spoken German, any utterance (including exclamatives) can be introduced by a sentence-initial response particle (jo). It marks the utterance it introduces as a response. To see this, consider the example in (34) and its representation in (35). In (34) it is the absence of a predicted event that triggers A’s exclamation. While the presence of jo is not obligatory, its use is strongly preferred. I suggest that this is because jo explicitly marks the utterance as a response. (34) Context. A and B are co-workers. Their working hours are fixed and they always go home at 4.30. Typically, they get ready to leave at 4.25 so they can be out the door by 4.30. Today B is not showing any signs of getting ready even at 4.25. A comments: Jo du oaweit-st heit long!? jo you work- 2sg today long ‘You’re working long today.’
Ergative constellations in the structure of speech acts 441 (35) Ergative response-structure RespP Resp-O jo
GroundP Ground-S
S p-structure
Note that this use of the response particle ja differs from its use as an affirmative response marker in form and function. The sentence-initial response marker discussed here is prosodically integrated with the utterance it introduces and it cannot stand alone. In its use as an affirmative response marker jo is associated with its own prosodic contour and can thus stand alone. Furthermore, as an affirmative response marker it can only respond to yes/no questions whereas as a marker of response jo can respond to any speech act as well as non-linguistic events as in (34). For further discussion of the grammar of this type of response particle see Wiltschko (in press).
18.6 Conclusion In this chapter I have explored the idea that ergative constellations are detectable in the syntax of speech acts. This idea is a logical consequence of Ross’s (1970) performative hypothesis and its more recent incarnations (Speas & Tenny 2003; Haegeman 2013). This is because the logic of ergativity operates on the composition of predicate–argument structures and speech act structure can be conceived of as a special kind of such structure. In particular, the ergative constellations we observe mirror those found in the domain of argument–structure. I have argued that SA-structure is best analyzed as consisting of two distinct layers. The lower layer (GroundP) consists of the utterance and a level where the speech act participant’s commitment to that utterance is expressed, i.e. their propositional attitudes. We have also seen evidence for another layer (RespP) which is dedicated to encoding the Call on Addressee in the sense of Beyssade & Marandin (2006). In particular, I have suggested that this can be understood as a layer responsible for the responsive aspect of language. With this framework in mind, we were able to explore ergative constellations In the domain of argument–structure ergativity it is widely acknowledged that we need to distinguish not only between transitive and intransitive predicates but also
442 Martina Wiltschko between two types of intransitive predicates: unergatives and ergatives (aka unaccusatives). I have argued that we find the same division in the domain of speech act structure. In particular, in the grounding domain the internal argument is the proposition whereas the external argument corresponds to the participant whose commitment toward p is expressed. On this view, declaratives are analyzed as transitive clause-types, imperatives are analyzed as unergatives, and presentatives are analyzed as ergative speech acts. In addition, we have seen evidence for an articulated response structure, where both subject and object of response can be marked. Both are marked in requests for confirmation (by means of confirmationals) but only one the subject of the response is marked in regular questions. We have further seen that there is a dedicated marker in German which serves to mark the utterance as a response. In the absence of a CoA (as is the case with exclamatives) this results in an ergative constellation in that only the object of the response is marked. For obvious reasons, this chapter has to remain programmatic in nature. The main point to take away from it is that, everything else being equal, we expect ergative constellations not to be restricted to the domain of argument–and case–structure. If we are indeed dealing with a matter of structural configurations, as assumed by most work in the generative tradition, then these structural configurations should not be restricted to one particular domain. We have further seen that this avenue of research is fruitful in that it brings to light different types of speech acts including some that are not standardly discussed. However, due to the exploratory nature of this enterprise we only discussed a limited number of such speech ac types. However, if the proposed grammar of speech acts is indeed on the right track, we expect a more complex and fine-grained typology of speech acts as is typically assumed. It is my hope that this chapter will inspire others to explore the structure of speech acts from this angle.
Abbreviations 2, second person; abs, absolutive; aux, auxiliary; erg, ergative; fem, feminine; Jo, response marker “jo”; masc, masculine; prf, perfective; prog, progressive; prt, particle; spkr, speaker.
pa rt i i i
A P P ROAC H E S TO E RG AT I V I T Y
Diachronic
Chapter 19
Gr amm atical i z at i on of ergative case ma rk i ng William B. McGregor
19.1 Introduction The grammaticalization of case markers has been discussed in a number of chapters in previous volumes in the Oxford Handbooks in Linguistics series (Heine 2003, 2009; Kulikov 2009; Johanson 2009; König 2011). The present chapter focuses on ergative case markers, which is also touched on in most of the previously mentioned chapters; here we provide a more elaborate and nuanced account. It is beyond the scope of this chapter to examine the grammaticalization of other aspects of ergative patterning, including other types of morphological ergativity (e.g. ergative agreement and cross-referencing systems, as found in e.g. Sacapultec—Du Bois 1987b) and syntactic ergativity (e.g. Dixon 1994: 143–181). Grammaticalization has been conceived of in a wide variety of different ways by scholars (Narrog and Heine 2011: 2–3). We are concerned here with the development of grammatical items. The standard textbook conceptualization is thus appropriate: grammaticalization is a diachronic processes whereby a lexical item becomes grammatical or a grammatical item becomes more grammatical (e.g. Kuryłowicz 1975: 52; Hopper and Traugott 1993: xv; Lehmann 2002: vii; Matthews 2007: 164). Two qualifications are in order. First, unlike some writers (e.g. Heine 2009), I will not assume that grammaticalization necessarily begins with lexical items: the available historical evidence simply does not support such a strong claim. Second, in discussions of the grammaticalization of case-marking morphemes metaphor is frequently invoked as an explanatory mechanism. Conceptual motivation is often proposed whereby an abstract domain is construed in terms of a concrete domain (Heine et al. 1991). Not all instances of the grammaticalization of case markers are so motivated, or indeed cognitively motivated in any significant sense (see also Gildea 2004). As will be seen, the grammaticalization of ergative case markers reveals a far richer set of pathways than expected from the literature (e.g. Trask 1979; Estival and Myhill 1988; Dixon 1994; Palancar 2002). This diversity is consistent with the enormous variety within the phenomenon of ergativity itself (DeLancey 2006; Gildea 2004).
448 William B. McGregor Ergative case markers can also be the sources of other grammatical markers, as discussed in 19.3. Section 19.4 overviews what can happen to ergative case marking in language contact situations, including language shift and endangerment.
19.2 The Emergence of Ergative Case Markers Four sources of ergative case markers are attested in varying degrees of frequency and with varying degrees of plausibility: Lexical items Markers of other cases Indexical elements Directional elements We examine these sources in order in the following subsections. They are not the only sources that have been identified. One frequently mentioned source is in discourse tendencies: ergative case-marking systems arise due to tendencies in discourse relating to the flow of information (Du Bois 1987b, 2003b; Hopper and Traugott 1993: 162). The problem is that case markers require some material source (see also Queixalós and Gildea 2010: 13, fn. 20). Nevertheless, 19.2.3 will reveal that a discourse tendency is sometimes relevant. Another possible source is in noun class markers. Sands (1996: 48– 49) proposes such an origin for ergative case markers in some non-Pama–Nyungan languages of Australia. However, there are more plausible sources in indexical items that simultaneously mark nominal class (Greenberg 1978/1990; McGregor 2008: 307–308; Kulikov 2009: 448; and section 19.2.3, this chapter).
19.2.1 Lexical Sources of Ergative Case Markers This is a poorly attested source of ergative case markers, and convincing examples are thin on the ground. Heine and Kuteva (2002: 165–166) suggest that the lexeme ‘hand’ may develop into an agent marker in passives, though no subsequent development into an ergative marker is attested. Hopper and Traugott (1993: 163) suggest that the preposition oleh is an ergative marker in Malay, and has origins in a verb meaning ‘get, obtain, do, manage, return’ which originally occurred in a separate clause. However, it seems that this preposition is in fact a marker of the agent of a passive construction in Modern Malay. Two ergative markers in Nungali (Mirndi, northern Australia) may perhaps have a lexical source in the Wanderwort mayi ‘vegetable food’. One is the allomorph of the case and class marking prefix mi-which occurs on adjectives modifying nominals of the vegetable class (Hoddinott and Kofod 1976b: 398; Green and Nordlinger 2004: 306). The other is the suffix -mayi used on vegetable class nouns (Hoddinott and Kofod 1976b: 398–399). Sands
Grammaticalization of ergative case marking 449 (1996: 48–49) proposes a development for the latter allomorph from a class marker to an ergative marker,1 though as already remarked, a more plausible pathway of development is from a determiner or pronominal marking the vegetable class. In this scenario, the original lexical source is irrelevant to the development of the ergative marker. The crucial development lay in a single process whereby a paradigm of class-marking indexical elements were reanalysed as case markers; the ultimate source of each had nothing to do with the process. Coupe (2011a: 32–34) suggests a lexical source for the ergative marker in the Ao group in the Tibeto-Burman family. He suggests that the most likely diachronic source of Mongsen Ao ergative -nə and its cognates within the Ao group, is a relational lexeme ‘side’ in (pre-)proto-Ao. This generalized marker of location, Coupe suggests, subsequently developed into a general oblique marker *na in proto-Ao, and ultimately into an ergative marker (via a process such as discussed under 19.2.2). Coupe (2011a: 33) cites Peterson (2002) on a similar lexical origin for the ergative morpheme in Chin languages.
19.2.2 Markers of Other Cases The most commonly proposed source for ergative case markers is other case markers, especially case markers with more local or concrete meanings. Four are reasonably well attested: instrumental, genitive, oblique, and ablative (e.g. Lehmann 2002: 98–99; Heine and Kuteva 2002: 180; Palancar 2002; Heine 2009: 467; Malchukov and Narrog 2009: 525). Here we overview a sample; there is no attempt to be comprehensive. Other possible sources include locative and comitative case markers. The former is not infrequently mentioned, though convincing examples are hard to find. In the instances that I have seen, this historical development is presumed merely on the basis of a shared formal marker, and the assumption that local case markers will extend to core cases rather than the reverse. Moreover, the same form usually also marks instruments, so the change may well be via an intermediate instrumental stage. This also seems to be the case for all instances of comitative marker to ergative (Lehmann 2002: 99). Grammaticalization studies have, until recently, tended to focus on the development of individual morphemes in isolation, paying little heed of the constructions in which they occur. The grammaticalization of ergative markers represents an exception. In the early days of awareness of this case-marking phenomenon linguists were inclined to see ergative constructions as passives (e.g. Pott 1873; Müller 1887; Schuchardt 1895; Uhlenbeck 1916; see further Seely 1977: 197–199). Although such views are no longer widely held by linguists, even in recent times one finds some who maintain passive origins for ergative constructions. For instance, Estival and Myhill (1988: 445) propose ‘the hypothesis that in fact all ergative constructions have developed from passives’. Slightly less restricted proposals have been made by many others. Trask (1979) proposes two source constructions, passives (for what he calls Type A ergative systems) and perfectives (for Type B ergative 1
Sands’ scenario is one of simple replacement. She assumes that all Australian languages are related, and that proto-Australian was an ergative language. For some unknown reason the inherited ergative marker was lost in the development of Nungali, and the class marker was pressed into use as its replacement.
450 William B. McGregor systems). Palancar (2002: 221) suggests that ergative constructions developed from transitive interpretations of formerly intransitive (or reduced transitivity) constructions of various types, not necessarily passives or perfectives. Queixalós and Gildea (2010: 13–17) propose that ergative patterns arise in main clauses via reanalysis of biclausal constructions such as nominalizations, and marked voice constructions such as passives. In these scenarios, ergative markers arise via reanalysis of markers of the (possibly optional) agent phrases in the marked constructions, which became increasingly common until they ultimately replaced the ordinary transitive construction. However, there are many languages for which there is no evidence that ergative markers arose within marked constructions such as passives or nominalizations. This is the case for ergative languages of Australia, where in no known language did an ergative case marker demonstrably arise in a passive or nominalization (e.g. Heath 1980: 907; McGregor 2009: 498). This also seems to be the case for ergative languages of the Eastern Sudanic groups (e.g. Nilotic, Surmic, and the Jebel language Gaahmg) and Tima (Niger- Congo) (Gerrit Dimmendaal, p.c.). On the other hand, languages of the Indic branch of Indo-European are generally considered to provide a ‘textbook’ case for the development of ergative case marking through reinterpretation of a passive construction (Anderson 1977; Comrie 1978; Payne 1980; Dixon 1994: 190; see Stroński 2009: 78–89 for a detailed overview). The received scenario has it that the mother language Sanskrit was an accusative language, and that the agent NP in an early passive construction was in instrumental case. The passive was reinterpreted as a perfect, and correspondingly the instrumental case of the agent was reinterpreted as ergative case of the Agent in the new transitive construction.2 This change did not occur in the imperfect, which retained nominative–accusative case marking. Despite its plausibility there are serious difficulties with this proposal (Klaiman 1987; Bynon 2005; Butt 2006a, 2008; Stroński 2009). First, there are problems with the notion that the source construction was a passive. Second, historical–comparative evidence reveals that the ergative of most modern Indic languages cannot be a reflex of the old instrumental marker (Butt 2006a: 77, 2008). Nevertheless, instrumental case markers are among the most commonly suggested sources of ergative markers, and represent the most plausible sources according to Heine and Kuteva (2002: 180). Garrett (1990) proposes that the Anatolian ergative marker *-anti developed from an earlier instrument/ablative marker. He proposes that this development occurred in a transitive construction with an instrumental NP but no subject NP, like ‘(s/he) strikes it with a dagger’, which he suggests must have existed in Anatolian. This was interpreted, he suggests, as ‘a dagger strikes it’, with the instrument reanalysed as subject. As Garrett (1990: 285) observes, this development does not require that the instrument/ablative marker was ever a causal marker (as is sometimes presumed—e.g. Palancar 2002); nor does it invoke metaphoric transfer. 2
In this chapter I follow the terminology I employed in McGregor (2009: 480) whereby Agent indicates the subject of a transitive clause, Actor the subject of an intransitive clause, and Undergoer the object of a transitive clause. This is a simplification of my proposals (McGregor 1997, 1998, 2002a, 2014), and these roles should not be equated with the familiar but problematic labels S, A, and O of linguistic typology—see also Mithun and Chafe (1999). Lower-case terms indicate etic relations.
Grammaticalization of ergative case marking 451 Foley (1986: 107) observes that in a number of Papuan languages (including Enga, Kewa, Kâte, Dani, and Selepet) the same case marker is used on Agent and instrument NPs, and suggests that the ergative usage represents an extension from the instrumental. The problem is that there is no supporting historical–comparative evidence. Garrett (1990: 280–285) presents some such evidence for the ergative marker in the Gorokan languages of the Eastern Highlands of Papua New Guinea. Internal evidence suggests that a single marker in the proto-language had both ergative and instrumental uses. Garrett (1990: 285) suggests further that the Gorokan morpheme is the reflex of a Common Eastern Highlands morpheme that had only instrumental uses. Genitive markers are also common sources for ergative case markers. In the Iranian branch of Indo-Iranian a possessive perfect construction was reanalysed as an active main clause; the genitive-marked possessor in the original nominalized construction was correspondingly reanalysed as an ergative marker (e.g. Benveniste 1952/1971; Allen 1964; Anderson 1977; Trask 1979; Dixon 1994: 191; Stroński 2009: 79–80). Bynon (2005) suggests that the original ergative marker in Indic languages was also a genitive, and that it was subsequently replaced by an instrumental marker (see also Stroński 2009: 95–96).3 Dimmendaal (2014: 12) suggests that ergative markers in the few morphologically ergative African languages derive from genitives (Nilotic languages such as Päri—cf. however 19.2.2) or instrumentals (Surmic languages, e.g. Majang). Furthermore, the Shilluk (Nilotic) ergative preposition yɪ is identical in form with the widespread oblique preposition of Western Nilotic (Dimmendaal 2014: 13). Ablative sources of ergative markers have been proposed for some languages, often via a causal (sometimes genitive) intermediary. The metaphorical basis is apparent: agents are sources and causers of actions and events; the target domain of agentivitity is plausibly construed in terms of a spatial source domain. Palancar (2002: 253–255) observes that the Basque ergative marker -k is formally similar to the ablative -tik, and suggests that the ergative derives from an earlier ablative, while the modern ablative involves additional phonological material. In support of this, he observes that remnantal ablative uses of -k survive in the Guipuzkoan dialect and that a causal sense survives in some idiomatic expressions. In some languages, ablative markers assign focus to Agent NPs. Thus in the Mirndi languages Jaminjung and Ngaliwurru (Schultze-Berndt 2000: 168), the ergative marker can be replaced by the ablative to assign contrastive focus to an Agent NP. In nearby Kija (Jarrakan), which is not morphologically ergative, the ablative case marker can be employed on an Agent NP to the same effect. These examples indicate that there is no need for a causative intermediate stage. The above instances concern developments from markers of non-core cases to markers of core cases. In some ergative languages of Africa ergative markers may have sources in markers of core cases, in particular in markers of nominative case. In East Africa, a number of languages are found in which the nominative is the marked case (König 2006, 2008a, 2008b, 2009). The relatively few ergative languages of the African continent occur in the 3
In view of the historical–comparative facts mentioned this is improbable. However, it could well be that some other marker later replaced the genitive. This could be a dative marker, as suggested by Butt (2006a, 2008)—though Butt cautions that the origin of the form is uncertain.
452 William B. McGregor same region. In some of them an origin of the ergative case marker in a nominative marker is plausible; the reverse direction may also have occurred in some languages—see 19.3. Päri shows a number of ergative characteristics including ergative case marking of Agent NPs in transitive indicative clauses, as shown in (1) (Andersen 1988). The same marker, -ɪ ̀ ~ -ɩ ̀ ~ -è ~ -ɛ ̀ (here I gloss over details of allomorphy), is a nominative case marker in other clause types (including most subordinate clause types, imperative clauses, and indicative clauses with marked focus). (1)
dháagɔ̀ á-yàaɲ ùbúrr-ɩ ̀ woman C-insult Ubur-ERG ‘Ubur insulted the woman.’
Päri (Andersen 1988: 292)
Andersen (1988) argues that the marked nominative pattern was historically prior, previously found in all clause types, and that the ergative marker arose through reanalysis of the nominative marker. He suggests that the earlier system showed VSO and VS word orders, like some modern languages. A topicalization pattern emerged in which postverbal NPs were fronted, and lost their case marking. This topicalization pattern became normal for Actors and Undergoers, giving rise to the modern unmarked word orders SV (intransitive clauses) and OVS (transitive clauses). The result of these changes would have been an ergative system as found in modern Päri, with Agent marked by the old nominative case marker, and Actor and Undergoer not case marked. Consistent with this, the Päri ergative marker may be cognate with the nominative singular marker -i ~ -e of distantly related Murle. König (2006: 706–708) agrees that the ergative marker of Päri derived from a nominative marker. She suggests a slightly different diachronic scenario: from definite marker to nominative marker, and thence to ergative marker. She observes that the Päri ergative marker is cognate with a definiteness marker in Anywa (which is in a dialectal relation to Päri), that is optionally used on post-verbal Actor NPs in both transitive and intransitive clauses.4 Significantly, Andersen (1988: 318) gives an example of an intransitive Actor NP in Päri marked by the ergative in what he suggests is a type of cleft construction. One wonders whether it might not be possible to account for the ergative and the marked nominative markers in Päri as independent developments from an earlier definiteness marker as per the scenario outlined in the next section.
19.2.3 Indexical Elements Oblique markers in passives, possessive perfects, and nominalizations are the most over-rated sources of ergative case markers. Indexical items such as pronominal 4
Dimmendaal (2014: 8–9) questions the analysis of this marker as definiteness marker, and proposes that it is an ergative marker. The problem is that both Agent and Actor NPs can take the marker when in post- verbal position (Reh 1996: 137). This is not to suggest that I find Reh’s analysis entirely convincing. Her claim that the morpheme in question is not a case marker seems dubious: possibly Anywa has an optional marked nominative, or the marked nominative may be a portmanteau morpheme that also indicates definiteness.
Grammaticalization of ergative case marking 453 elements and determiners are by contrast the most underrated sources. Indeed, in many general survey treatments they either fail to be mentioned at all or are mentioned only in passing, and fail to be incorporated into the primary story (e.g. Dixon 1994: 182–206; Heine and Kuteva 2002; Palancar 2002; Gildea 2004; Heine 2009: 467; König 2011). This is despite the considerable body of evidence that has accumulated over the past century or so, including in sources used in some of these works. Furthermore, it is remarkable that the very person who appears to have been the first to use the term ergative, albeit for a local case, the missionary George Taplin (1872: 85– 86; see Lindner 2014), remarks that in the Ngarrindjeri language of the Adelaide region, South Australia: The causative [i.e. ergative in modern terminology] is formed in the singular by the affix il—evidently an abbreviation of the pronoun kili (by him); as, Kornil mempir napangk, or inangk nap—The man struck his wife. (Taplin 1880: 8)
Support for Taplin’s proposal comes from the formal identity of the ergative marker and -il, the third person ergative bound pronoun (Cerin 1994), cited in Simpson et al. (2008: 104). As this example indicates, the indexical item in question often marks something else as well as indexicality, often case, sometimes noun class. However, as suggested below, it is the indexical component that is principally relevant to this grammaticalization process. As mentioned in 19.2.1, in some members of the Mirndi family nominals take portmanteau affixes indicating both ergative case and noun class. For instance, Jingulu has ergative allomorphs -(r)ni on masculine nominals, and -nga on feminine nominals (Chadwick 1976: 394–395; Pensalfini 1999: 227; Green and Nordlinger 2004: 306). These are plausible reflexes of earlier gender markers *(r)ni masculine and *ngayi feminine, that were simultaneously third person forms, possibly oblique case forms of demonstratives or pronouns (Chadwick 1976: 394; see also Blake 1988: 12–13, 33). Similarly, Nungali has corresponding masculine and feminine ergative prefixes nyi-and nganyi-~ nyanyi- (Hoddinott and Kofod 1976b: 397–398), which may also derive from earlier demonstratives or pronominals that distinguished masculine and feminine genders. This renders more plausible the suggestion already made that the ergative formatives occurring on vegetable class nouns and modifying adjectives might have originated in indexicals. For the remaining class in Nungali, things are not quite so clear, though there is a recurrent (w)u that might be a reflex of an indexical item marking this class (such a class is found in some other languages of northern Australia). In most other Mirndi languages ergative markers do not show different forms according to nominal class. Nonetheless, the elsewhere ergative allomorph is -ni in Jaminjung and Ngaliwurru (Schultze-Berndt 2000: 54–56) and Wambaya (Nordlinger 1998a: 83). This is presumably cognate with the masculine form in Jingulu and Nungali. Nyulnyulan languages have a bound ergative postposition showing a variety of allomorphs, including -ni ~ -rni ~ -in ~ -nim ~ -nimaa ~ -nma ~ -na. These forms could be reflexes of the *(r)ni masculine indexical referred to in the previous paragraph. Significantly,
454 William B. McGregor there is evidence from the modern languages suggestive of a third person singular pronominal form *(yi)ni in proto-Nyulnyulan (Stokes and McGregor 2003; McGregor 2008: 309). Some Daly River languages also show an ergative form that might have derived from an earlier pronominal element ni, e.g. Ngangikurrunggur -ningke and Ngan’gityemerri –ninggi (cf. Blake 1988: 45, who interprets the initial ni element as a class marker). An interrogative indexical may be the source of the ergative marker in Garrwa, a non-Pama–Nyungan language of the Gulf of Carpentaria region (see further below). The elsewhere ergative allomorph has the form –wanyi, which is identical with the free nominative interrogative wanyi ‘what, who’. The other allomorph, -nyi, could be a truncation. Proto-Pama–Nyungan was almost certainly an ergative language, and most languages of this family have ergative markers inherited from the proto-language (Blake 1987a; Sands 1996: 1–39). In a few languages, however, the ergative markers appear to be recent innovations, deriving from indexicals. Wangkumara distinguishes two noun classes: masculine singular and non- masculine- singular, encompassing everything else (Breen 1976). Case- marking of arguments is by bound pronouns attached to the relevant nominal. Three case forms are distinguished for all nominals and free pronouns: nominative (for Actors), ergative (for Agents), and accusative (for Undergoers). The masculine singular ergative marker is -(u)lu (a reduced form of the free masculine singular ergative pronoun nhulu)5 while the non-masculine-singular ergative is -(a)ndru (a reduced form of the free form nhandru), as shown in (2) and (3). (2)
(3)
karna-ulu kalka-nga thithi-nhanha person-ERG hit-PST dog-ACC ‘The man hit the bitch.’ makurr-andru nganha kabiliba-nga tree-ERG 1SG.ACC block-PST ‘The trees blocked me.’
Wangkumara (Breen 1976: 337) Wangkumara (Breen 1976: 337)
In the Pantyikali and Southern Paakantyi dialects of Paakantyi bound forms of two ergative demonstratives, ‘this’ and ‘this here’, can be attached to nouns as ergative markers (Hercus 1982: 63), as in (4). The free form can be used instead, as in (5), which apparently express an emphatic meaning. (4) yartu-thurru kaanti-t-urru-ana wind-DEM:ERG carry-FUT-3SG.NOM-3SG.ACC ‘The wind will carry it along.’
Southern Paakantyi (Hercus 1982: 63)
5 This form of the ergative shows the need for caution in identification of forms on the basis of allomorphs in isolation. Seen by itself one might wrongly conclude that the -(u)lu allomorph is a direct inheritance from the proto-Pama-Nyungan ergative marker *-lu (e.g. Blake 1988: 27).
Grammaticalization of ergative case marking 455 (5) kajiluku marlibarlu irnuruu-dinga wadu-ji-na Southern Paakantyi small boy DEM:ERG-EMP take-PST-3SG.ACC ‘It was the small boy who took it.’ (Hercus 1982: 63) Similarly in Paarruntyi in addition to the inherited system another mode of expression is available for singular Agents: a free resumptive ergative pronoun alongside of (though not necessarily adjacent to) the unmarked Agent NP, as in (6). (6) nganha karli-wa thatya wuthu-rru nguma my dog-EMP bite PST:3SG-ERG 2:SG:ACC ‘My dog bit you.’
Paarruntyi (Hercus 1982: 62)
There are a number of less certain cases of this developmental pathway. The Umpithamu ergative marker -mbal may be a reflex of a widespread demonstrative form pala (Jean- Christophe Verstraete, pers. comm.). The ergative markers in Wardaman are -yi ~ -ji ~ -nyi; similar forms occur in Gunwinjuwan varieties: -yih (Dalabon, Rembarrnga, Ngalakgan), -yi ~ -yih (Kune, Manyallaluk Mayali), and -yi (Warray). The first allomorph in Wardaman is identical in form with one of the three nominal class-marking prefixes, yi-for animates and humans. Moreover, yi occurs as a component of third person pronominal forms in many languages from the Kimberley to Arnhem Land, and as an agreement class-marker in various languages, including the Gunwinjuwan language Nunggubuyu (yii-). It seems more probable that these ergative forms developed from a class-marked indexical element than that they are reflexes of a proto-Australian ergative allomorph *-DHu (McGregor 2008: 314; cf. Sands 1996: 66). McGregor (2008: 310–313) suggests an explanation for this grammaticalization scenario. Things might have begun with a source construction comprising an NP in apposition with a determiner (definite or indefinite/interrogative) or pronoun (see also Green and Nordlinger 2004: 304). Such a construction might have been used to single out unexpected Agents in discourse, while expected Agents could have either been ellipsed entirely, or have been represented by unmarked NPs. That such a construction might be dedicated to singling out Agent NPs to the exclusion of NPs in other grammatical roles is not unreasonable if we assume the preferred argument structure (PAS) constraint was in operation (Du Bois 1987b, 2003b). According to the PAS, the Agent role is strongly associated with given information. If an Agent NP conveys new information this marked pragmatic circumstance might well be formally marked. On the other hand, Actor and Undergoer are cross-linguistically preferred sites for introduction of new items into discourse, and thus of unexpected NPs. There should thus be no strong need to signal that NPs in these roles are unexpected. Thus juxtaposition of a resumptive pronominal to NPs in these roles would be unnecessary. The PAS constraint concerns information status of the Agent. If the indexical item (and cleft construction) began to be used with new Agent NPs, it is not implausible that it might generalize to marking Agent NPs that are unexpected for other reasons, such as low animacy or relative powerlessness. We cannot exclude the possibility that the
456 William B. McGregor initial motivation for use of the cleft construction was not something else, for instance the animacy hierarchy (Silverstein 1976). It is possible that some traces of different initial points might remain. Thus, in the Cape York Pama–Nyungan language Wik-Mungkan the ergative marker for ordinary NPs is the clitic -ang, which is formally identical to the accusative marker for pronouns (Kilham 1977: 56); similarly, in Kala Lagaw Ya -n is an ergative marker on common nouns and an accusative on singular pronouns (Kennedy 1984: 156). Assuming that these are innovations that can be traced back to cleft constructions, an explanation is possible for these unusual systems. The cleft may have been used with phrases that were unexpected in Agent and Undergoer roles according to animacy—inanimate NPs in the Agent role and pronominals in the Undergoer role. Right from the very beginning the indexical item in the cleft construction would have been closely associated with the grammatical role of Agent. Over time, this construction might have increased in frequency, and begun to lose its association with unexpectedness. The presence of the resumptive item might become the norm in Agent NPs. The resumptive item might not occur on Agents that are expected, but be used otherwise, as seems to be the case in Warrwa and other Nyulnyulan languages. Alternatively—or subsequently—it might become obligatory. The emergence of indexical-based ergative markers in Paakantyi varieties appears to be recent. The inherited ergative marker -rru is rare in most dialects (except Paarruntyi), possibly a consequence of other changes in the grammar such as rigidification of word order. The inherited marker may thus be in the last stages before complete disappearance. In this context, the PAS constraint might motivate the appearance of the new constructions involving resumptive items. Although many of the source forms proposed above involve markers of case and/or class as well as indexicality, it is the latter component that evidently drives the grammaticalizations. The beginning is in a type of cleft construction that assigns focus to unexpected Agents. The other grammatical categories marked by the form are relevant to the development insofar as they are selected by the grammatical context. This scenario also provides a plausible account for the development of the ergative marker from an interrogative determiner in Garrwa. It is not suggested that this explanation accounts for all instances of the rise of ergative markers from indexicals. It does not explain the situation in Wangkumara, where nominative and accusative case-markers have also grammaticalized from free pronouns. PAS does not predict marked treatment of unexpected (or expected) Actors or Undergoers. Perhaps in this language the inherited ergative marker wore down phonologically, and (in the absence of rigidification of word order), the pronominal items in the appropriate case forms came to be used to mark (indeed, over-mark) grammatical relations. Alternatively, it may have been that unexpected or unpredictable NPs in any role came to be highlighted—after all, according to the PAS Actor and Undergoer roles are associated with both given and new information, and there could well have been a need to mark the difference. Over time and with increased usage these markers could have lost their focal values and became the markers of the three core grammatical relations. Formal similarity of ergative case markers and indexical items is not restricted to the Australian continent.
Grammaticalization of ergative case marking 457 In Trans-New Guinea Tauya the ergative suffix -ni is formally identical with the third person singular pronoun ni (MacDonald 1990: 322). MacDonald (1990: 322–323) suggests that transitive clauses showed SOV word order, and that those with third person singular objects in ni became increasingly frequent. This led to the reanalysis of ni as a suffix to the subject of transitive clauses with third person singular objects; this suffix was subsequently generalized to occur in all transitive clauses. The proto-Polynesian ergative preposition *e is, according to Kikusawa (2002: 155), most likely the personal noun marker *i or *ii. The ergative preposition e in the Oceanic language Roviana may derive from the proto-Western Oceanic personal article *e (Corston 1996: 61). Ergative case markers -mə and -n in the North-West Caucasian languages Kabardian and Ubykh respectively may trace back to mə and jəna ‘this’ (Kumaxov 1971: 43, 158, 1989:31f, cited in Kulikov 2006, 2009). The Georgian ergative suffix -ma ~ -m evidently derives from Old Georgian -man, which is identical in form with a third person ergative pronominal (Boeder 1979: 457–458).
19.2.4 Directional Elements An unusual source of ergative marker is a directional element indicating ‘hither’. The Sahaptian language Nez Perce of north-western USA has the ergative marker -nim ~ -nm—as shown in (7)—which has such a source in a ‘hither’ marker according to Rude (1991, 1997).6 Correspondingly, the accusative marker -ne may have grammaticalized from a ‘thither’ marker. (7) wewúkiye-ne pée-’wi-ye háama-nm elk-ACC 3ERG:3ACC-shoot-PST man-ERG ‘The man shot an elk.’
Nez Perce (Rude 1991: 25)
Given the metaphoric source of ergatives in ablatives frequently alluded to (see 19.2.2) this is an unexpected pathway of grammaticalization. Rude (1991), however, provides a plausible explanation. He suggests (Rude 1991: 45) that the Sahaptian ergative marker can be traced back to a locative *m in pre-Sahaptian-Klamath. This developed in two directions: to a genitive case marker, and to the directional marker *ɨm ‘hither’ in proto- Sahaptian. The latter was subsequently reanalysed in two ways: as a verbal suffix ‘hither’, and separately as -(n)im ergative when the object was a speech act participant, i.e. first or second person. In a transitive clause describing action directed to a speech act participant it is possible that the ‘hither’ marker was employed on both the verb and the Agent NP. In the former case, this would indicate direction towards the speech act participants. In the latter case, being restricted to transitive clauses, the marker could well be reinterpreted as 6
Nez Perce does not show prototypical ergative–absolutive patterning in its case-marking system. First and second person NPs follow a nominative–accusative case-marking system. Third person NPs have a three-way case-marking system distinguishing ergative (marking Agents), nominative (marking Actors), and accusative (marking Undergoer) (Rude 1991: 25). This does not render the label ergative inappropriate for the case marker of the Agent NP: it is an ergative marker in a tripartite system.
458 William B. McGregor a marker of Agent when the event is directed to a speech act participant. Significantly, in modern Sahaptin the ergative marker -(n)ɨm—a reflex of -(n)im—is restricted to transitive clauses in which a third person Agent is acting on a first or second person Undergoer (Rude 1991: 27). In Nez Perce the nominal marker may have undergone further reanalysis and generalized to transitive clauses regardless of the person of the Undergoer. Consistent with this grammaticalization scenario first and second person pronouns do not take ergative marking. With a speech act participant as Agent the ‘hither’ directional element should not be used, the action being directed away from the speech act participant. This would remain an inexplicable restriction if the ergative had arisen via grammaticalization of the genitive.
19.3 Developments from Ergative Case Markers Ergative case markers can disappear virtually without trace, as seems to have happened to the inherited marker in some Pama–Nyungan languages (as discussed in 19.2.3). They may also be reanalysed as markers of other cases and other grammatical categories. Below I discuss a few examples; considerations of space preclude comprehensive treatment. A few Pama–Nyungan languages have lost their ergative morphology, and have become nominative–accusative. This happened in many Ngayarda languages of the Pilbara region of Western Australia, including Panyjima, Ngarluma, Yinjiparnti, and Martuthunira; others, e.g. Ngarla, Nhuwala, Nyamal, have remained ergative. A plausible grammaticalization pathway is adumbrated in Dench (1982). He proposes that alongside the ordinary transitive construction there was a marked intransitive alternant, in which the erstwhile Agent was unmarked, while the Undergoer was in the dative case. Such a construction is attested in various languages of the region, and is often associated with lack of affectedness of the Undergoer. This may have become an increasingly popular means of expression and ultimately have been reanalysed as the transitive construction.7 The result would be a language with an accusative case-marking system, where the accusative case was identical with the former dative case. This is shown in (8), where -ku is also a marker of datives and benefactives, and a reflex of the proto-Pama–Nyungan purposive marker. (8) ngunha marlpa yukurru-ku wiya-rna that man dog-ACC see-PST ‘That man saw the dog.’
Panyjima (Dench 1982: 47)
Reflexes of the proto-Pama–Nyungan ergative marker remain in the accusative Ngayarda languages. One is the instrumental marker, which has a rather restricted usage—e.g. in 7 Dench (1982: 58) suggests that the increasing popularity of this mode of expression may have been motivated by a pressure for the morphological case-marking system to align with the syntactically accusative system.
Grammaticalization of ergative case marking 459 Panyjima it is used only on inalienably possessed instruments (Dench 1982: 47). Another is in the marker of the agent of the various types of passive constructions, which include two inflectional passives (as in (9) and (10)) and a derivational passive (as in (11)). As Dench (1982: 55–56) observes, the inflectional passives highlight the affectedness of the semantic undergoer; they would have been incompatible with the semantics of the construction in which the undergoer was marked by the dative. Thus they retained the ergative case-marking system of the proto-language. After the dative construction had been reanalysed as the unmarked transitive construction, they were reanalysed as passives. The case marking of the derivational passive may have been modelled on the inflectional passives. If Dench is right, the marking of agent of the passive does not derive from the instrumental use of the original ergative marker, but from its Agent-marking use.8 This would thus be a genuine instance of development from an ergative case marker. (9)
panha yalha ngayi-rnaanu wirrpi-ngku that shed throw-PFV.PASS wind-INS ‘That bough-shed has been wrecked by the wind.’
Panyjima (Dench 1982: 49)
(10) ngatha pilanyja-ku katama-lpuru ngunha-jirri-lu 1SG.NOM frightened-PRS hit-POT.PASS that-PL-INS jilyantharri-lu children-INS ‘I’m frightened I might get hit by those children.’ (11)
yukurru ngunha-ngku marlpa-ngku wiya-nnguli-nha dog that-INS man-INS see-PASS-PST ‘The dog was seen by that man.’
Panyjima
Panyjima (Dench 1982: 47)
Dimmendaal (2010a: 34, 2014) suggest—contrary to Andersen (1988) and König (2006, 2008b, 2009: 546–547)—that nominative markers in all Nilotic and Surmic languages have origins in ergative markers that extended in usage to cover Actors. He proposes that ergative marking of Agent NPs in post-verbal position extended to marking of post-verbal Actor NPs, thus giving rise to marked nominative systems. The latter order appears to be an innovation. Tennet subordinate clauses show pre-verbal Actor NPs, and 8
Most Nyulnyulan languages have a medio-active construction which shows identical case marking as ordinary transitive clauses but the verb cross-references the unmarked NP rather than the ergatively marked NP as in ordinary transitive clauses (McGregor 1999b). This construction is semantically almost identical with the perfective passive of Ngayarda languages. One wonders whether the perfective passive—and perhaps also the potential passive—might not be traceable back to earlier marked constructions that were not ordinary transitives. This does not affect the observation that it was the ergative use of the case marker that was reanalysed as marker of the agent in passives, not the instrumental. However, Dench’s scenario implies that an ergative marker of the Agent role has become a marker of the agent of a passive, which change must be a degrammaticalization, given the usual assumption that the change from marker of agent of a passive to an ergative marker is an instance of grammaticalization. My alternative does not invoke such a change, since the Agent of the medio-active is not a core participant in the clause (see McGregor 2014 for further discussion).
460 William B. McGregor post-verbal Agent NPs (Dimmendaal 2014: 9). Given that subordinate clauses are typically more conservative than main clauses, they may well reflect the typical word order of the proto-language. It would show an ergative case-marking system: the post-verbal Agent NP would be marked, while the pre-verbal Actor in intransitive clauses would not be marked morphologically. If main intransitive clauses innovated a word order in which Actors could occur post-verbally (possibly motivated by information distribution), the ergative marker might occur on these NPs, resulting in a marked nominative system. While this seems plausible, there remain doubts about its universality in Nilotic and Surmic languages. Thus, as we have seen, in Päri evidence from some subordinate clause types—exactly the same type of evidence employed by Dimmendaal (2014)—suggests that the marked nominative system was earlier, and developed into an ergative system. This grammaticalization pathway is not restricted to the African continent. In the Kartvelian language Mingrelian the erstwhile ergative marker, restricted to the aorist, extended to cover intransitive Actor NPs, thus resulting in a marked nominative system (Dixon 1994: 202). Similarly, in Pitta-Pitta the ergative marker is -lu, a reflex of the proto- form; this is used only in non-future contexts. In the future, -ngu, evidently a reflex of the ergative allomorph *-ngku of proto-Pama–Nyungan, is a nominative marker occurring on both Agent and Actor roles (Blake 1979c: 193). As remarked in 19.2.2 the usual interpretation of the situation in which Agent of a transitive clause is marked in the same way as an instrument is that the instrumental use is historically primary, and that this developed into an ergative marker. In some languages in which the ergative derives from an indexical item we also find a single form with both ergative and instrumental uses, e.g. Jaminjung (Schultze-Berndt 2000: 54). Given the grammaticalization scenario of 19.2.3 it seems improbable that use as an instrumental marker preceded use as an ergative marker: it is unlikely that a cleft-like construction would begin with instrumental NPs, and expand to Agent NPs. Moreover, PAS accounts for a beginning in Agent marking. In such circumstances the ergative marker appears to have extended to cover instrumental NPs. Wangkumara lends support to this scenario. Instrumental NPs in this language are marked by the non-masculine-singular ergative allomorph -(a)ndru; the masculine-singular is not attested in this function, apparently regardless of the gender of the instrument (Breen 1976: 338). It is improbable that the instrumental use, in which the bound pronominal is attached to an inanimate noun, would predate use on an animate noun. Furthermore, Breen (1976: 338) provides an example of this marker extending into the accompaniment domain usually marked by the comitative -bartu: (12) karnia yanthangaga pakarranyi-andru, marli-andru kandigandi-andru man go/PST/? boomerang-ERG spear-ERG axe-ERG ‘The man came here with his boomerang, his spears, and his axe.’ Wangkumara Markers of ergative case sometimes give rise to markers of other grammatical categories, including subordinate clause markers and tense/aspect markers; see 19.4 for possible development into a focus marker. In Newari (Tibeto-Burman) the ergative/instrumental case marker -na is used as a temporal clause subordinator, as in (13). Genetti (1991) argues
Grammaticalization of ergative case marking 461 that the subordinate clause usage developed from the case-marking use, though it is not clear whether or not this was via the ergative or instrumental use or something else such as comitative—cf. ‘with his coming to the house …’. Heine (2009: 468) suggests that the case marker first extended from nouns to nominalized verbs, and then to subordinate clauses. (13) chẽ-ku yer-na wā ām-e naku moŋ-an Newari house-LOC come-SUB EMP 3SG.M-GEN cheek swell-PRT coŋ-gu stay-3SG.PST.HAB ‘When he came to the house his cheek was swollen.’ (Genetti 1991: 227) In a number of Australian Aboriginal languages, the ergative marker is also found on secondary predicates, typically on Agent NPs (McGregor 2005), external inalienable possessions (McGregor 1999a), and sometimes on manner adverbials (not necessarily in transitive clauses). This is the case in the Bunuban language Gooniyandi, where the same marker, -ngga ~ -ga, is also found on non-finite clauses indicating attributes of Actor NPs (McGregor 1990: 178, 396–397). A plausible line of development is that the ergative marker first came to be used on secondary predicates on Agents, perhaps to make it clear which role the secondary predicate applied to. From there an extension to marking body parts and the like in inalienably possessed constructions (a type of secondary predicate construction in Gooniyandi) and nominals indicating a quality of the Agent relevant to the event (e.g. speed, size, condition, mental state) is not unreasonable. The marker could then have been extended further to non-finite verbs that indicated a concomitant state of the Agent, or a concomitant event the Agent was engaged in. Why this extended to non-finite verbs indicating concomitant states of Actors in intransitive clauses remains a puzzle. Admittedly this scenario is entirely hypothetical; nevertheless, it is based on the type of changes that are not infrequently found in grammaticalization. In Kala Lagau Ya virtually all case markers, including the ergative, are used as markers of tense and aspect on verbs (Kennedy 1984). The ergative -n marks completive aspect. However, the ergative also marks accusative on singular pronouns, and it is just as plausible that the extension into the verbal domain began with the accusative as the ergative (Blake 2001: 180).
19.4 Ergative Case Marking in Language Contact and Contact Languages There can be little doubt that language contact can be a factor in the emergence and development of ergative case markers and systems. The areal clusterings of languages with ergative case marking attests to this. It has been suggested that ergativity in
462 William B. McGregor Indo-Aryan originated in the contact situation within the South-Asia linguistic area (Stroński 2009: 103). However, there are difficulties with this proposal, and convincing morpho-syntactic convergences with known possible substrate languages are lacking (Stroński 2009: 104). Contact with Dravidian languages may have contributed to the gradual loss of ergative patterning in Indic languages. Ergative case marking in non- Pama–Nyungan languages of northern Australia may result from contact with Pama– Nyungan languages. The fact that ergative and marked nominative languages in Africa are areally restricted suggests that these phenomena may be contact-based. Ergative case markers are perhaps sometimes borrowed. Anderson (1977) suggests that the Sinha ergative was borrowed from Tibetan. The Hindi/Urdu ergative postposition ne has been attributed to borrowing from Tibeto-Burman languages, a number of which show an ergative form like na (Zakharyin 1979). Coupe (2011a: 31) proposes that the ergative marker of Chungli (of the Angami-Ao group in Tibeto-Burman) was borrowed from Chang (Bodo-Konyak-Jinghpaw). In extreme circumstances of language contact morbidity of one or more of the languages of a speech community can result. In many Australian Aboriginal communities in post-contact times there has been marked shift away from the indigenous languages towards English lexicalized varieties such as Kriol and Aboriginal English. The traditional languages sometimes undergo grammatical changes as speakers use them in increasingly reduced circumstances. Case markers, including ergative case markers, often show reduced allomorphy. Schmidt (1985: 383–386) documents the reduction in allomorphy of the ergative case marker in the language of young Dyirbal speakers in the early 1980s, initially to a single allomorph, later to its complete disappearance in the youngest generation of speakers. Other Australian languages show similar losses of the ergative, with or without allomorphic reduction. In the case of Nyulnyul, now without mother-tongue speakers, the last two fluent speakers appear to have used the traditional optional ergative system (McGregor 2002b). However, two semi-speakers I interviewed in the 1980s showed new systems. One had apparently lost the ergative marker entirely, using English-based word order to distinguish arguments. The other generalized ergative marking to all Agent NPs obligatorily. A different set of changes occurred in Jingulu. Traditional Jingulu was a morphologically ergative language in which the ergative marker was used obligatorily on all Agent NPs: nouns showed an ergative–absolutive system of marking, while pronouns were marked according to a tripartite system (Chadwick 1976; Pensalfini 1999: 226). In the late 1960s when Neil Chadwick began his work on Jingulu, emphatic or focal markers appeared that had not been present when Ken Hale did fieldwork on the language a decade earlier. These markers are formally identical with ergative and dative allomorphs. Pensalfini (1999: 238) suggests that they indeed derive from these case markers by reanalysis that was initially induced by contact with Kriol. In the Jingulu of the 1990s the erstwhile ergative marker occurs not just on Agent NPs, but also on Actors and Undergoers. However, when it occurs on an Agent NP, the focal marker follows an instance of the same marker. Pensalfini (1999: 238) suggests that when the present-day speakers were children learning the language in an environment in which Kriol was becoming increasingly prominent they first analysed Jingulu as a nominative–accusative language, and reanalysed the ergative and dative markers as markers of discourse prominence. Later, with more exposure to the traditional
Grammaticalization of ergative case marking 463 language they learnt that the ergative and dative markers were indeed case markers, but the discourse usages had already been established, and were not unlearnt. Thus ergative and dative markers coexisted alongside of the prominence markers that they gave rise to. The Amazonian language Tacana, unlike the other members of the Tacanan family, shows a system of optional ergative marking. Guillaume (2014) suggests that this is most likely an innovation of Tacana, possibly resulting from language obsolescence and/or influence from Spanish, and that the proto-language had a system of obligatory ergative marking. New languages sometimes arise in contact situations (Bakker and Matras 2013). There appear to be no instances of pidgins or creoles with ergative case markers. However, a few mixed languages with ergative case marking are attested. The Australian mixed languages Gurindji Kriol and Light Warlpiri are examples (McConvell and Meakins 2005; Meakins and O’Shannessy 2010; Meakins 2011; 2014a). These languages arose in contact between the traditional Aboriginal languages Gurindji and Warlpiri, both Pama–Nyungan, and Kriol, an English-lexicalized contact variety that arose in Northern Australia sometime in the twentieth century. Meakins and O’Shannessy (2010) argue that the competition between the two systems of argument marking, the ergative case-marking systems of Gurindji and Warlpiri and word order in Kriol, gave rise to a new system in the mixed languages. This is an optional ergative system, in which use of the ergative marker accords prominence to the Agent NP. There are other differences between the ergative case-marking system of the mixed and traditional languages. In Gurindji Kriol ergative marking has generalized to all Agent NPs—any Agent NP may be ergatively marked regardless of its animacy—whereas traditional Gurindji showed split ergativity, whereby free pronouns did not take ergative marking. In traditional Warlpiri, ergative marking is obligatory on all nominal types except first and second person singular pronouns, on which it is optional. Both Gurindji Kriol and Light Warlpiri show reduced allomorphy in the ergative case marker compared with the traditional languages (Meakins 2011).
19.5 Conclusions The chapter shows that there are many ways in which a language can become ergative morphologically, consistent with the synchronic diversity in ergative case marking (Gildea 2004). Many of these processes of grammaticalization are not cognitively motivated in terms of transference from one conceptual domain to another, as is often presumed. Contrary to Gildea (2004), however, it does not seem that the evidence supports the conclusion that grammaticalization of ergative case marking is a mechanical process involving constructions that just happen to have oblique case marked Agent NPs. Oblique sources account for a proper subset of ergative case markers. The discussion of a number of instances of grammaticalization of and from ergative markers in this chapter reveals the crucial role of constructions (see e.g. Hopper and Traugott 2003: 1–2; Himmelmann 2004: 31). We have seen, however, that caution is required: evidence is needed for constructions just as much as for markers. It is not
464 William B. McGregor sufficient to hypothesize grammaticalization within constructions that are presumed on the basis of well-known languages. The range of constructions in which ergative case marking may have emerged is as wide as the range of sources of the case markers themselves. Contra Queixalós and Gildea (2010: 13) biclausal constructions (such as nominalizations) and marked voice constructions (like passives) are not the only constructions in which ergative case marking can emerge. A number of grammaticalization scenarios show connections with focus: this is fundamental to the development of some ergative case markers from indexical items (via clefts); it appears to also be a possible consideration in the development of some ablative case markers to ergative markers, and in the case of Jingulu it seems that ergative markers can become markers of focus. Synchronically a strong connection between ergative case marking and focus is sometimes apparent—e.g. McGregor and Verstraete (2010); Rumsey et al. (2013). I have not found convincing cases in which a focus marker serves as the historical source for an ergative case marker.9 This is, however, a possible origin for the ergative markers of Bunuban languages (McGregor 2008: 313) and Kuikuro (Southern Carib, Brazil) (McGregor 2012: 239; cf. Franchetto 2010). The discussion of sections 19.2.3 and 19.2.5 reveals that one needs to be cautious in drawing conclusions from ‘polysemies’ of case markers. While abstract and more grammatical uses of case markers sometimes develop from more concrete uses, we cannot presume this in every instance. It is also possible that such a situation emerged historically through independent development of another marker. And in some languages (e.g. Wangkumara) concrete senses appear to be secondary in diachrony.
Acknowledgments I am grateful to Barry Blake and Gerrit Dimmendaal for useful comments on an earlier draft of this chapter, and to Alan Rumsey and Lila San Roque for providing me with unpublished material.
Abbreviations 1, first person; 2, second person; and 3, third person; ACC, accusative; AG, agent; C, completive; DEF, definite marker; DEM, demonstrative; EMP, emphatic; ERG, ergative; FUT, future; GEN, genitive; HAB, habitual; INS, instrumental; LOC, locative; M, masculine; NP, noun phrase; PAS, preferred argument structure; PASS, passive; PFV, perfective; PL, plural; POT, potential mood; PRS, present; PRT, partitive; PST, past tense; SG, singular; SUB, subordinate.
9 Kilian-Hatz (2008: 55, 2013: 376–377) and König (2008b: 276) argue that the postposition à in Khwe (Khoe-Kwadi, Botswana) developed from a copula into a focus marker, and thence to an optional marker of accusative case. (See however McGregor forthcoming for an alternative scenario in which the accusative marker developed from a previous attention-directing indexical item.)
Chapter 20
Dec onstru c t i ng Iranian erg at i v i t y Geoffrey Haig
20.1 Introduction The Iranian languages constitute one of the major branches of Indo-European. They are most closely related to the Indo-Aryan branch, from which they separated some 4,000 years ago. The oldest reliably datable attestations of Iranian, the Old Persian inscriptions, are approximately 2,500 years old, but the earliest Avestan texts represent an Iranian language probably spoken several centuries earlier, though the dating remains contentious because they were committed to writing at a later date (Skjærvø 2009: 43). Today, Iranian languages are spoken across a vast swathe of Asia, from the Pamir mountains in the border region of Tajikistan and China, to central Anatolia in today’s Turkey. Traditionally they are genetically classified into two main branches, East Iranian and West Iranian, and the latter is further divided into north-west and south-west Iranian. However, many details of the subgroupings remain controversial; see Windfuhr (2009) for recent discussion. The following typological features characterize most modern Iranian languages:
• OV word order; • differential object marking, though absent in Kurdish and Zazaki; • a very high frequency of complex predicates, based on a small set of light verbs; • a tense-based alignment split, affecting transitive verbs,1 and involving some kind of ergativity in those clauses based on verb stems etymologically derived from a participle.
1 Generally, tense-based alignment splits in Iranian are only relevant for transitive verbs, but there are two counter-examples: Wakhi (East Iranian, Tajikistan, and Afghanistan), where in the Hunza dialect intransitive subjects may occur in variant case forms (apparently restricted to first/second person
466 Geoffrey Haig This chapter focuses on the last feature. Though absent in some modern Iranian languages, most notably Persian, most Iranian languages have, at some point in their history, passed through a stage with an alignment split, involving ergative structures in the past tense. In the case of Persian, the relevant structures are well attested for the Middle Persian period (Jügel 2015), but Persian has since ironed out this wrinkle and returned to a unified accusative alignment in all tenses. Elsewhere in Iranian, however, the characteristic split is still evident in various aspects of the morphosyntax. Given the extent of the Iranian languages, this chapter can only consider a selection of languages and themes related to ergativity. A common thread running through the presentation is the massive variation in what is loosely termed ergativity in Iranian (see section 20.2 on terminology). In the literature, a number of additional holistic labels for alignment types have been coined with the aim of capturing the variation (double oblique, tripartite, neutral, etc.). I will suggest that such labels, while useful for taxonomic purposes, actually obscure more important generalizations and commonalities cross-cutting supposedly distinct alignment types. A more insightful account can be framed in terms of distinct sub-aspects of alignment, hence the ‘deconstruction’ in the chapter’s title. From its inception, the literature on ergativity in Iranian has focused on the two interrelated issues of micro-variation, and origins, the former pioneered in Payne’s (1980) study of alignment variation in Pamir languages, and the latter in Bynon’s (1980) paper on the diachronics of ergativity (see section 20.3). Synchronic analyses from the perspective of formal syntax models are less prominent, possibly due to the ubiquitous issue of variation, but see Dorleijn (1996) for an attempt to confront formal approaches to ergativity with dialectal variation. The current chapter continues the tradition of relating micro-variation to a diachronic scenario. The chapter is built around three case studies of dialect clusters, all from the West Iranian branch of the family: Kurdish, Balochi, and Taleshi. For East Iranian languages, the respective chapters of Windfuhr (ed.) may be consulted. Kurdish, Balochi and Taleshi have been chosen because each exhibit a rich range of micro-variation with regard to ergativity, they are geographically divergent, and the sources are reasonably accessible. Figure 20.1 shows the approximate locations of the three dialect clusters (note that Balochi is also spoken by around 500,000 speakers on the other side of the Persian Gulf in Oman). In section 20.2, ergativity in the Iranian context is introduced and terminological conventions laid down, while section 20.3 looks at the origins of ergativity. Sections 20.4, 20.5, and 20.6 contain surveys of Kurdish, Balochi and Taleshi respectively, while
pronouns, cf. Bashir (2009: 842–843). The second exception appears to be the Mutki dialect of Zazaki (West Iranian, East Anatolia), though the analysis is still preliminary (Öpengin & Anuk 2015). Outside of these marginal cases, the restriction to transitive verbs is stable, and constitutes a major difference between the ergative structures of Iranian, and those of the neighboring East Aramaic languages, where (in some dialects at least) the special marking of the A is shared by the S arguments of certain unergative intransitive verbs (see Doron & Khan 2012; Khan, Chapter 36, this volume). The difference is noteworthy given that the East Aramaic brand of ergativity is generally considered to have developed under Iranian influence.
Key: NK=Northern Kurdish; CK=Central Kurdish; SK=Southern Kurdish; SB=Southern Balochi; WB=Western Balochi; EB=Eastern Balochi. Sources: Kurdish: Haig & Öpengin (2014); Taleshi: Paul (2011); Balochi: Jahani & Korn (2009)
Deconstructing Iranian ergativity 467
Figure 20.1 The approximate locations of the Kurdish, Balochi, and Taleshi dialect clusters
468 Geoffrey Haig section 20.7 summarizes the main points and proposes some more general principles behind the observations of the preceding sections.
20.2 Conceptual and Terminological Issues Throughout this chapter I will adopt the following abbreviations: S = core argument of an intransitive verb, bearing the least-marked morphological case form A = core argument of a transitive verb, semantically exhibiting the most (potential) control P = core argument of a transitive verb, semantically exhibiting the least control
The classic definition of ergativity is a grammatical system in which ‘S is marked in the same way as O[=P] and differently from A’ (Dixon 1994: 16), abbreviated here as A≠S, S=P. Overwhelmingly, A≠S involves additional marking of the A in comparison to the S, rather than the other way round; I am unaware of any language in which S is consistently more morphologically marked than A. We can restrict the notion of ‘marked in the same way’ to the following two domains of morphology (i) case marking, e.g. inflectional affixes or adpositions that indicate predicate/argument relations, and are located on the relevant NP; (ii) agreement (Dixon’s cross-referencing), which involves the replication of features such as gender or person of a particular argument on another constituent. In addition to the above morphological criteria, we can gauge the similarities between S, A, and P in terms of their implication in syntactic rules, such as control of reflexives, or coreferential deletion across coordinated clauses. The latter kinds of similarities, generally termed ‘syntactic ergativity’, turn out to be largely irrelevant in the Iranian context. There is no reliably attested example of an Iranian language which consistently groups S and P in terms of syntactic rules, while there are many well-documented cases of the grouping of S and A. Thus, most of Iranian syntax appears to work on a S/A-pivot, in Dixon’s (1994) terms, regardless of the alignment of its morphology (see e.g. Haig 1998 on Northern Kurdish).
20.2.1 Deconstructing Ergativity It is well known that ergativity generally characterizes only subdomains of a grammar, where it coexists with non-ergative alignments of other parts (see the extensive literature on various kinds of ‘split’). As a consequence, the holistic view of ergativity as a deep and fundamental trait characterizing an entire grammar has largely been abandoned in favour of more contingent approaches, which focus on individual constructions and their interplay (Haig 2010). As Bickel (2010: 442) puts it, ‘once popular expressions like ‘ergative language’ are simply senseless’. This, then, is one way in which the concept of
Deconstructing Iranian ergativity 469 ergativity can be deconstructed. But there is a second, and more fundamental, fault line running through the concept of ergativity. Recall that the classic definition of ergativity involves two independent statements: A≠S, and S=P. As I hope to show in this chapter, there are sound empirical reasons for treating the two sub-aspects distinctly. Turning first to A≠S, one way of conceiving it is in terms of a special marking, or singling out, of transitive subjects as a morphologically marked category among subjects generally. This most typically occurs in the case system, but there are often differences in the agreement system between S and A as well. Diachronically, marked A, or A≠S, appears to be the most persistent aspect of ergativity in Iranian, and is relatively stable across the micro-variation surveyed here. The second component of ergativity, S=P, is, in its pure form (i.e. full identity in case and agreement) exceedingly rare in Iranian. While in case marking, many languages do unite S and P, agreement systems almost always make a distinction. In a sense, then, agreement is the weak link in the ergativity chain, a fact that echoes the typological generalization that agreement systems favour accusative alignment, while case systems are more prone to ergativity (cf. Haig (2008: 301–304) for discussion of Iranian, and Bickel (2010: 442) for a typological perspective). But the failure of S and P to unite is not restricted to agreement. In many varieties of both Balochi and Taleshi, the case marking of P also differs from that of S. But crucially, these differences in case marking between S and P are generally independent of the presence or absence of A≠S. It can thus be demonstrated that the two aspects of ergativity, A≠S and S=P, are in principle independent of one another. For the purposes of this chapter, I will refer to any grammatical subsystem with A≠S, i.e. where transitive subjects are marked relative to intransitive subjects, as ‘ergative’, but the reader should be aware that this is merely a terminological convention, and in most cases construction-specific details diverge from the standard definition of ergativity cited at the outset of this section.
20.3 The Origins of Iranian Ergativity, and Its Principal Manifestations In Old Iranian, transitive clauses with all finite verb forms had a unified accusative alignment, regardless of the tense of the verb: the A was in the nominative case, the verb agreed with A, and the P was marked by the accusative case. This is demonstrated with a transitive clause from Old Avestan (Haig 2008: 25, transcription and glosses adapted): (1)
aṯ zī θwā fšuyantaē=cā and indeed 2sg.acc cattle.breeder.dat=and θwōrəšta tatašā fashioner.nom has.created.3sg
vāstrāi=cā herdsman.dat=and
‘And indeed the Fashioner(A) has created you(P) for the benefit of the cattle-breeder and the herdsman’ (Old Avestan, Yasna 29,6).
470 Geoffrey Haig Alongside the system of finite verbs, Old Iranian also had a set of participles, generally involving a final -ta. These participles were originally ‘verbal adjectives, with a resultative sense’ (Haig 2008: 41), in their semantics comparable to English participles such as broken, or fallen. They were already widely attested in Old Avestan; examples include vista- ‘known’, mərəta- ‘died, dead’, or bərəta-‘carried, brought’ (Jügel 2015: 101–104). These participles occurred both in attributive, and predicative function, and agreed in gender, case, and number with the noun they were associated with. Participles could be derived from both transitive and intransitive verbs, but in Old Iranian, the attested examples of participles with intransitive verbs appear to be restricted to change-of-state predicates (Jügel 2015: 101–104). The following Young Avestan example illustrates a participle used predicatively (and here without the copula), and negated via a verbal negation particle: (2) āaṯ yaṯ =hē zasta nōiṯ frasnāta then when=clpro.3sg.gen hands not washed.ptcpl ‘For if his hands are not washed […]’2 (Young Avestan, Jügel 2015: 66, glosses added, transcription simplified). In Old Persian, much of the Indo-European system of aspectual distinctions for verbs had disappeared; for past tense reference in narratives, the default verb form was the imperfect, while the Old Iranian Aorist is only marginally attested, and presumably already largely defunct. Clauses with a ta-participle as predicate became increasingly integrated into the system of aspectual distinctions, and according to Jügel (2015) they contributed a resultative, perfective aspectual nuance. The following example contains two clauses headed by the ta-participle, and a third with imperfective. Jügel (2015: 70) suggests that the first two clauses express the speaker’s deeds from the perspective of completed events (‘deeds’), while the third clause, containing the imperfect form akunavam, focuses on the process itself, which involved accompaniment by the will of Ahuramazda. (3) a. taya that.which.n.sg
manā 1sg.gen/dat
b. taya=mai that.which=clpro.1sg.gen/dat c. awa wisam wašna that all will.instr
kərtam do.ptcpl.n.sg
apataram kərtam, afar do.ptcpl.n.sg
auramazdāhā akunav-am Ahuramazd.gen do.ipfv-1sg
a. ‘That which I did here, b. and that which I did afar, c. all that I did through the will of Ahuramazda’ (Old Persian, Jügel 2015: 70, glosses added) 2
idā utā here and
[participle, non-finite] [participle, non-finite] [imperfective, finite]
The translation here follows Jügel’s German translation ‘Denn wenn seine Hände nicht gewaschen (sind) […]’. An alternative interpretation is taken up below.
Deconstructing Iranian ergativity 471 Aside from the nuances of aspectual interpretation, (3) also serves to introduce the key syntactic issues surrounding ergativity in Iranian: first of all, the predicate in (3a) and (3b) is the participle kərtam, which agrees in gender and case with its underlying object (‘that which’). In the first clause, the doer is expressed through the independent first person singular pronoun manā. It is in a case that is historically derived from the old Iranian Genitive, but which had absorbed all the old functions of the Dative (the Dative no longer existed as a distinct formative in Old Persian). In the second clause, the doer is expressed by the clitic version of this pronoun, =mai, which is hosted by the first element of its clause in compliance with well-known principles of second-position clitic placement (Wackernagel’s law). By the Middle Iranian period, loosely covering the first millennium ad, constructions such as (3a–b) had become the sole means of expressing past time reference. The verb system had broken down to a basic binary opposition between two stems: what I will term a ‘present stem’, and the reflex of the resultative participle in -ta, termed in this study the ‘past stem’. The syntax of transitive clauses based on the past stem in Middle Iranian basically continued what we have illustrated for Old Persian, but with a crucial difference. Case morphology had largely been lost, so S, A, and P appeared in the same unmarked case (with the exception of the first person singular and certain kinship terms in Parthian and early Inscriptional Middle Persian texts). The A, therefore, appeared in the common, unmarked case form. But more frequently, the A was expressed through a clitic pronoun in a Wackernagel position, the etymological continuation of pronouns such as =mai in (3b). The Middle Iranian (Middle Persian and Parthian) forms for the clitics are shown in Table 20.1; they are still recognizable in the modern languages discussed below (though the third person singular has a variant form in some languages): Table 20.1 Pronominal clitics in Middle Iranian Singular
Plural
First person
=m
=mān
Second person
=t/d
=tān/=dān
=š
=šān
Third person
Source: Jügel (2015: 222), simplified
The verb itself (often an auxiliary) generally agreed with the P, quite in keeping with its participial origins. The following Middle Persian examples illustrate the clitic pronoun expressing a past A, as well as the agreement of the verb (here an auxiliary) with the P: (4) u=š druyist [fn. omitted, GH] kird hēd and=clpro.3sg complete do.pst(3sg) aux.2pl ‘and he created you(pl)’ (Middle Persian, Jügel 2015: 373, glossed added)
472 Geoffrey Haig The use of the pronominal clitic to express the A generally only occurred when the A was not overtly present in the clause. Thus in the first part of (5) no pronominal clitic occurs, but it does in the second: (5) ud kū šāhān and when king.pl
šāh ēn paykar dīd king dem statue see.pst(3sg)
u=š dād … and=clpro.3sg give.pst(3sg) ‘And when the King of Kings saw this statue, and gave […]’ (Middle Persian, Jügel 2015: 534, glosses added) In an intransitive past clause, on the other hand, the S is never expressed through a pronominal clitic,3 and the S controls agreement on the verb. The same applies to the A of a present transitive clause (not shown here): (6) āmad hēnd gāwiraz-ān come.pst aux.3pl throne.preparer-pl ‘Those who prepare the throne have come’ (Middle Persian, Parthian, Jügel 2015: 260, glosses added) The Middle Iranian construction has generally been analysed as ‘ergative’, due to the agreement of the verb with the P, and the fact that A has distinct pronominalization patterns (and in the earliest texts, a distinct case) when compared to S. Thus the uniform accusative alignment that characterized finite verbs in Old Iranian (cf. (1)) had evolved into a system with split alignment in Middle Iranian: accusative alignment in the present, and ergative alignment in the past. The question that has been debated for decades concerns the nature of the mechanisms behind the transition from the Old Iranian accusative alignment to the Middle Iranian ergative alignment. Two main lines of explanation have been proposed. The first assumes that the Old Persian construction with a Genitive/Dative A was basically an agented passive, i.e. passive verb form+by-phrase, yielding for (3a) ‘that which was done by me’. Over time, this was reanalysed into an active construction, and the syntactically non-core argument (the by-phrase) develops into a fully fledged subject. Following Bynon (1980), this approach gained currency in diachronic syntax generally, and the Iranian case was added to the list of languages apparently supporting the ‘well-attested
3 The corpus of Middle Iranian texts investigated by Jügel actually contains isolated examples of pronominal clitics with S (13 cases among 3,522 intransitive subjects), but Jügel (2015: 328) considers them most likely to be scribal errors. In the later languages, pronominal clitics are occasionally attested with intransitives (e.g. Balochi of Turkmenistan, Axenov 2006: 108–109), but this appears to be a marginal extension, with no obvious systematic regularities. As a general rule holding across Iranian, wherever indexing of subjects with pronominal clitics is attested, it is restricted to past tense A’s.
Deconstructing Iranian ergativity 473 […] reanalysis of earlier passive constructions’ as a source of ergativity (Lightfoot 1999: 140). Haig (2008: Ch. 2) notes that the account in terms of origins from an agented passive construction suffers from an empirical flaw, namely the striking lack of evidence for any of the assumed intermediate stages. He develops an alternative account, according to which the Old Persian construction was the extension of already existing Non- Canonical Subject constructions, involving a Genitive/Dative marked Benefactive, Experiencer, or Possessor. These already exhibited semantic and pragmatic features typical of subjecthood (topicality, high animacy), thus the necessity for assuming wholesale restructuring of the construction is reduced, because the Benefactives, etc. already shared crucial properties with transitive subjects. The link between possession/benefaction and agency has been pointed out many times; the Young Avestan example (2), repeated here as (7), provides a neat illustration of the constructional polysemy that was already available in Old Iranian, and which can be considered pivotal in the extension of this kind of construction to clauses involving agency: (7)
āaṯ yaṯ =hē zasta nōiṯ frasnāta then when=clpro.3sg.gen hands not washed.ptcpl ‘For if his hands are not washed […]’
The translation given here follows Jügel (2015). But it is not hard to see how one could read: ‘For if he has not washed his hands’, with the clitic pronoun expressing not only the possessor of ‘hands’, but the A of ‘washed’. The following example from Old Persian likewise illustrates the overlap of possession and agency (Haig 2008: 95): (8)
avaθā=šām thus=clpro.3pl.gen
hamaranam kartam battle do.ptcpl
While Kent (1953) still translates with a passive plus by-phrase ‘thus by them battle was done’, one could equally translate ‘their battle was fought’, or more naturally in the actual context of this example, ‘they engaged in battle’. On this account, the main change is not one of reanalysis or internal restructuring of the construction, but of frequency of occurrence, and markedness within the entirety of the tense system. Thus while in Old Persian, clauses headed by the ta-participle occupied a relatively minor aspectual niche (see (3)), structurally they were already largely identical with their Middle Iranian ‘ergative’ descendants. The difference is simply that by Middle Iranian, the ta-participle verb forms were the only available alternative for clauses with past time reference, hence the unmarked choice for past transitive clauses. The ‘change’ thus involved the extension of an existing construction into new contexts, brought about by the disappearance of the competitors through loss of finite verb forms such as the Aorist. A similar conclusion is recently reached by Jügel (2015), who goes as far as to classify the Old Persian m.k. construction with an overt agent as ‘ergative’, and likewise refers to the extension of a construction with a Benefactive to cover constructions with Agents
474 Geoffrey Haig (2012a: 464).4 For more detailed coverage, see Haig (2008: ch. 2) and two recent in-depth studies on Middle and Old Iranian, Jügel (2015) and Bavant (2014).5 The reduction of the verb system to two stems, one of which was in origin a resultative participle, had far-reaching consequences for the syntax of the daughter languages. These participles, like their counterparts in English, such as broken, fallen, etc., were unable to assign accusative case, and this defect was inherited by all the daughter languages. For participles from transitive verbs, the default assignment of nominative case was to the underlying object, with which the participle would agree, though agreement with the P is not consistent, even in Middle Iranian. With the nominative already assigned, and an accusative blocked, the expression of the A had to follow a different pattern. As we have seen, the case used for this purpose in Old Iranian was the Genitive/ Dative case, the case otherwise used for Benefactives, Possessors and Recipients.6 In the daughter languages, the reflex of this case continues to mark the A of the ergative. Furthermore, many of the later languages maintained the system of indexing an A via a pronominal clitic, already illustrated in (4) and (5). Thus the most salient aspects of ergativity in Iranian were already clearly discernible in the syntax of Old Iranian, and most later developments can be plausibly reconstructed back to the earlier constructions. In sum, three structural components that characterized ergativity throughout Iranian languages can clearly be traced back to developments that had their seeds in Old Iranian: (i) the binary opposition in the verb stems, with the past member based on the old participle; (ii) an inherited case system, generally reduced to a two-way opposition between an unmarked ‘Direct’ case, and a marked ‘Oblique’ case, the latter the continuation of the old Genitive/Dative. In some languages, the latter has disappeared entirely (Persian, Central Kurdish), and in some, additional cases based on grammaticalized adpositions have developed. (iii) The remarkably persistent use of clitic pronouns to index the A of a past tense transitive clause, already evident in Old Persian (3b), and still found in many modern West Iranian languages, where they exhibit forms cognate with the Middle Iranian paradigm shown in Table 20.1. 4 A reviewer points out that while examples such as (7) and (8) illustrate the overlap of Benefactive/ Possessor readings with Agency, it is not readily apparent how the much-discussed Old Persian examples such as (3), with the phrase ‘that which I did’, can be reconciled with Benefactive/Possessor readings. There are two responses to this. The first would be that the extension to general agency had already occurred in the Old Persian texts, hence the participial construction was already an available option with any transitive verb. The second is to note that even these examples can be viewed from a possessive/ resultative perspective, to be read as ‘these are my deeds’. Ultimately, the restricted size of the Old Iranian corpus means that reliable reconstruction of all the assumed intermediate stages cannot be achieved. 5 Bavant (2014) and Jügel (2015) provide detailed philological analysis and develop more articulated accounts of the Old Iranian verb system, incorporating additional data from Avestan. Notably, neither scholar advocates a return to the agented-passive interpretation. Bavant concludes that the question of the origins of Iranian ergativity is ultimately unanswerable, given the restricted nature of the Old Iranian corpus, while Jügel’s analysis largely confirms Haig (2008), though differing in details and terminology. 6 The two principal ideas behind this analysis, namely that ergativity in Iranian (i) primarily results from the intrinsic inability of the participles to assign accusative case; and (ii) exhibits both structural and diachronic parallels to constructions with ‘Dative’, or ‘Non-Canonical Subjects’, were around for decades in Indo-European linguistics (Benveniste 1952/1966), albeit in rather different terminologies, before being explicitly developed in Haig (2008). It is interesting to note that more recently, and quite independently, the same ideas are espoused in Minimalist approaches to ergativity; see Mendívil-Giro (2012).
Deconstructing Iranian ergativity 475
20.4 Kurdish Kurdish is a cover term for a group of closely related north-west Iranian languages (or dialects) spoken across a large region of Eastern Turkey, North Iraq, Syria, Armenia, and west Iran. The main dialects are Northern Kurdish, also termed Kurmanji, and Central Kurdish (also called Sorani). These two dialects differ with regard to ergativity very strikingly. Most of this section is devoted to Northern Kurdish, which exhibits a rather clear case of ergativity, while in 20.4.2 I will briefly illustrate the relevant constructions in Central Kurdish.
20.4.1 Northern Kurdish: Canonical Ergativity Northern Kurdish is not the official language of any nation state, but in three parts of the Northern Kurdish speech zone, different versions of a more or less standard written language have developed: in the former Soviet sphere, mainly in Armenia, Kurmanji was written in the Cyrillic script; the Kurds of Turkey and Syria adopted a Roman-based script for their version of the standard language, while in North Iraq, a modified version of the Arabic script is used for the Kurmanji dialects of that region. The Roman- based alphabet has become the most widely used standard for print and internet media, and examples here are based on that standard. Estimates of the number of Kurmanji speakers are unreliable, and vary between 12–20 million, most of whom live in (or have recently migrated from) today’s Turkey. Dialectal variation across Kurmanji is discussed in Öpengin & Haig (2014). In Kurmanji generally, pronominal clitics are entirely absent, though transitional dialects between Northern and Central Kurdish exhibit them in some contexts. Ergativity, then, where it is found, is solely manifested in case morphology, and in verbal agreement morphology. Nouns and pronouns occur in two forms, the unmarked Direct, and the marked (or suppletive) Oblique. The paradigm for case and number on nouns is given in Table 20.2, and for pronouns in Table 20.3, which give the forms of the Roman-alphabet based standard, with some dialectal variants in brackets. Table 20.2 Case and number on nouns in Northern Kurdish Singular, masculine and feminine Masculine singular ‘village’ Definite
Feminine singular ‘woman’
Indefinite
Dir.
Obl.
Dir.
Obl.
gund
gund-î
gund-ek
gund-ek-î
Definite
Indefinite
Dir.
Obl.
Dir.
Obl.
jin
jin-ê
jin-ek
jin-ek-ê
476 Geoffrey Haig
Plural ‘women’ Definite Dir jin
Indefinite
Obl.
Dir.
jin-an
jin-in
Obl. jin-(in-)an
Table 20.3 Case and number on pronouns in Northern Kurdish Singular
Plural
Direct
Oblique
Direct
Oblique
1
ez
min
em
me
2
tu
te
hûn
we
3
ew
(e)wî (m.)/(e)wê (f.)
ew
wan
With regard to the structure of the NP, determiners and quantifiers precede the head, while lexical modifiers (adjectives and possessors) follow the head and are linked to it by means of a vocalic particle traditionally termed the Ezafe (ez). Ezafe constructions are well known from Persian, but unlike Persian, the Ezafe particle in Kurmanji is sensitive to the gender and number of the head noun. Possessors in Ezafe constructions take the Oblique case, while adjectives do not inflect for case: (9) bajar-ek-î mezin ‘a big town’ town-indf-ez.m big (10) mal-a me house-ez.f 1pl.obl
‘our house’
The southern and south-eastern dialects of Kurmanji exhibit a fairly consistent brand of ergativity, which I will refer to as canonical ergativity. In canonical ergativity, S and P display identical features with respect to both case marking and agreement, while A is distinct in both respects. Canonical ergativity is restricted to past tense transitives. In intransitive clauses, the S is in the direct case, and the verb agrees with it; this alignment is impervious to a change in tense, as shown in (11): (11)
her sê t-ê-n-e / hat-in-e mal-ê each three indic-come.prs-pl-drct come.pst-pl-drct house-obl.f ‘All three are coming/came home’ (Southern Kurmanji, dialect of Midyat)
Deconstructing Iranian ergativity 477 In transitive clauses, however, alignment is sensitive to tense. In the present tense (12a), the A is in the Direct and agrees with the verb, while the P is in the Oblique. In (12b), the A is in the Oblique, while the P is in the direct, and agrees with the verb: (12)
a. ez 1sg.dir
wî 3sg.obl.m
na-bîn-im neg-see.prs-1sg
b. min ew ne-dît 1sg.obl 3sg.dir neg-see.pst(3sg) ‘I (a) don’t/(b) didn’t see him’ Example (13) has a plural A and a singular P. Number indexing on the verb targets the A in the present tense (13a), and the P in the past tense (13b): (13) a. gundî tişt-ek-î na-bêj-in villager(dir.pl) thing-indf-obl neg-say.prs-pl b. gundî-yan tişt-ek ne-got villager-obl.pl thing-indf(dir) neg-say.pst(3sg) ‘The villagers (a) aren’t saying/(b) didn’t say anything.’ The pragmatically neutral word order, however, remains APV, regardless of the morphology. In terms of case marking and agreement, then, the canonical ergative construction of Northern Kurdish in the past tenses involves S=P, S≠A. The system is summarized in Table 20.4. Table 20.4 Canonical ergativity, Northern Kurdish Type of verb
Case of core arguments
Present transitive
A-Direct
P-Oblique
With A
Past transitive
A-Oblique
P-Direct
With P
Intransitive, all tenses
S-Direct
Verbal agreement
With S
Analytical passives can be formed from transitive verbs, using the verb hatin ‘come’ as auxiliary coupled with the infinitive, which is based on the past stem, for example hat kuştin(-ê) ‘lit. (he/she) came (to) killing (-obl)’ = was killed’. Passives can be formed from transitives in both the past and the present tenses, with no apparent restrictions applying to past transitives.
20.4.1.1 Deviations from Canonical Ergativity The dialects of Northern Kurdish diverge from the scheme shown in Table 20.4 in various ways. The most widespread is the behaviour of verbal agreement with respect to
478 Geoffrey Haig plurality. There is a strong tendency for a plural A, particular if not overtly present in the clause, to trigger plural agreement on the verb—even when the P (which would normally control agreement) is singular. Plural agreement with a past A is actually found in all varieties of Kurmanji, spoken or written, when the A is not overtly present, and the preceding clause is an intransitive clause with the same subject. An example from the written language (the poem Ji Biçukan re, by Cegerxwîn), is typical: (14) herdu çû-n-e cem rovî both go.pst-3pl-drct to fox ___ doz-a xwe jê ra got-in ø case-ez.f refl adp.3sg.obl adp say.pst-3pl ‘The two of them went to the fox (and) put their case before him’ (lit. told him self ’s case) The past transitive verb form gotin has a plural agreement marker, although its direct object is singular (doz ‘case’, identified as singular by the singular form of the feminine ezafe). The plural agreement here thus reflects the plural number of the A herdu, which is not overtly expressed in the second clause. In fact, in this example (and similar ones), it would be impossible for the verb to agree with its singular object doza xwe ‘their case’. Thus under the conditions of co-referential deletion of a plural A, verbal agreement generally switches its alignment to the A, rather than the P. The same pattern may extend to clauses with an overt, Oblique marked A. The following has an overt A, and an explicitly singular P, yet the verb agrees in number with the A:7 (15)
pîrek-a(n) jî hirç-ik dî-n woman-pl.obl add bear-indf see.pst-pl ‘The women too saw a bear’ (Haig 2008: 234; note the non-standard form of the indefinite suffix)
This pattern is particularly common with those transitive verbs that generally lack a referential, nominal object, such as gotin ‘say’ (the ‘object’ of this verb is usually a clausal complement, usually direct speech), see Haig (2008: 231–242) for discussion of other relevant factors. The second well-documented deviation from canonical ergativity is the so-called double-oblique construction, familiar from Payne’s (1980) account of East Iranian Pamir languages. In this construction, both and A and P are in the Oblique case, and the verb agrees with neither, generally taking the default third person singular form. Under 7
The plural agreement with the A in this example would not be acceptable in all dialects of Northern Kurdish, for example Shemzinan (Ergin Öpengin, p.c.).
Deconstructing Iranian ergativity 479 the loose definition of ergativity given at the outset of this chapter, Double Obliques still qualify as a kind of ergativity because A≠S (A is Oblique). But it is not canonical ergativity, because S and P have different case and agreement properties. Double Oblique constructions are typical of north-eastern dialects of the Northern Kurmanji cluster, in particular the dialect of Muş, and is also found in the writings of authors from this region (examples from Haig 2008: 226, glosses slightly modified): (16) gundî-yan wan bizor ji hev kir villager-pl.obl 3pl.obl by.force from recipr do.pst(3sg) ‘The villagers separated them by force’ (lit. did them from each other) (17) îşev min keç-ek-ê last.night 1sg.obl girl-indf-obl.f di xewn-a xwe da di-dît adp sleep-ez.f refl adp ipfv-see.pst(3sg) ‘Last night I was seeing a girl in my dream’ A final deviation from canonical ergativity occurs in tight sequences of same-subject clauses, where the first verb is intransitive and the second transitive. The first verb is typically a motion verb çûn ‘go’, or hatin ‘come’. In such clause sequences, the subject of the intransitive verb may appear in the Oblique case, thus anticipating the case assignment of the second verb. Agreement on the first verb is usually default 3sg, though there is some variation here. This pattern only seems to be possible when the two clauses concerned are not linked by any overt coordinator (e.g. û ‘and’). An example is the following: (18) min çû cot-ek sol li bazar-ê kirî 1sg.obl go.pst(3sg) pair-indf shoe at market-obl buy.pst(3sg) ‘I went and bought a pair of shoes at the market’ (Haig 2008: 250) A very similar pattern also occurs in Balochi (Korn 2009: 68). One way of interpreting these sequences is as a kind of verb serialization, with the second verb determining the case assignment of the entire sequence, but there has been little research on these constructions (see Haig (2008: 248–252) for discussion).
20.4.1.2 Non-Canonical Subjects In the south- eastern dialects, predicates expressing physical and mental sensations, desire, and possession, may have their experiencer/possessor argument in the Oblique case. The use of Oblique experiencers is semantically not fully predictable, but is connected to certain predicates rather idiosyncratically, cf. (19) with an Oblique experiencer, and (20) with the experiencer in the Direct case and agreement with the predicate:
480 Geoffrey Haig (19) min sar e 1sg.obl cold cop.3sg
‘I am cold.’
(20) ez birsî me 1sg.dir hungry cop.1sg ‘I am hungry.’ In the south-eastern dialects, expressions of desire use the verb viyan ‘be necessary, desirable’. An overtly expressed ‘wanter’ is clause initial, and in the Oblique case. If there is an overt nominal ‘wanted’ in the clause, it is in the Direct case, and agrees with the verb (21); otherwise the verb takes the default third person singular ending, shown in (22): (21)
te ez na-vê-m 2sg.obl 1sg.dir neg-be.necessary.prs-1sg ‘You do not want me’ (lit. to.you I not.desirable.am’) (MacKenzie 1961a: 192, glosses supplied, transcription adapted)
(22) min di-vê-t [bi-ç-im] 1sg.obl indic-be.necessary.prs-3sg [irr-go.prs-1sg] ‘I want [to go]’ (lit. ‘to.me is.desirable I go’) Predicative expressions of possession show the same pattern: (23) min du bira he-ne 1sg.obl two brother existant-cop.3pl ‘I have two brothers.’ The fronted Obliques in (21)–(23) exhibit syntactic pivot properties, and can thus reasonably be considered to be subjects. In this respect, these constructions resemble the ergative constructions discussed in the preceding section. However, unlike the ergative construction, they are not restricted to past tenses, i.e. there is no tense-sensitivity. As a general tendency, the frequency of such non-canonical subjects decreases as one heads northwards and westwards from the south-eastern periphery of the Kurmanji speech zone. Thus in Northern Kurmanji, expressions of desire use the transitive verb xwastin ‘want, request’, and the wanter appears in the direct case in the present tenses (see Haig (2006) on the areal distribution of non-canonical subjects in Northern Kurdish).
20.4.1.3 Summary of Northern Kurdish (Kurmanji) Clauses based on past tense verbs in Northern Kurdish shows a fairly clear case of canonical ergativity, with S=P in case marking and agreement, and distinct from A. However, dialectally a number of deviations from this scheme can be discerned: plural agreement triggered by a plural A, and Double Oblique constructions, lacking agreement with A or P. Both of these deviations affect agreement on the verb, and the latter also affects the
Deconstructing Iranian ergativity 481 case marking of P. The Oblique case of the A, however, is remarkably consistent across all dialects. Among early first-language learners (up to 6 or 7 years old), constructions with a Direct-marked A can be observed, and in the dialect of Gilan, a dialect of north- west Iran isolated from the rest of Northern Kurdish, this appears to be the community norm (Shojai 2005). Whether these phenomena are portentous of a general drift back towards accusativity is a moot point. Among those mature speakers who acquired Northern Kurdish in the social context of an intact speech community, particularly in the southeast part of the Northern Kurdish speech zone, the ergative construction is alive and well. However, the political unrest of the last decades, accompanied by mass migration from the region, means the number of persons currently acquiring Northern Kurdish in an intact speech community is declining rapidly (Çağlayan 2014), and many young speakers now show uncertainties and variation in the relevant constructions.
20.4.2 Central Kurdish: the Role of Pronominal Clitics Central Kurdish, spoken to the south of Northern Kurdish in North Iraq and West Iran, differs in its system of alignment quite strikingly from Northern Kurdish. In the Suleimaniye dialect, the best-described variety (MacKenzie 1961, 1962), all nouns and pronouns lack overt case marking, thus S, A, and P are not distinguished at all by case marking. The morphosyntax of past transitive clauses nevertheless still differs from that of present transitive clauses. The principal difference is that in past transitive clauses, the A is obligatorily indexed via a mobile pronominal clitic, while in present transitive clauses, A is obligatorily indexed via a verbal suffix, in just the same manner as an S. Intransitive verbs, all tenses: (24) min a-rrō-m / rrōišt-im 1sg indic-go.prs-1sg / go.pst-1sg ‘I go, am going/I went’ (MacKenzie 1961a: 106, glosses and transcription modified) Transitive verbs, present tense: (25) min sag-aka na-kuž-im 1sg dog-def neg-kill.prs-1sg ‘I am not killing the dog’ In past transitives, on the other hand, the A is obligatorily indexed via a mobile pronominal clitic, etymologically the descendants of the Middle Iranian clitics from Table 20.1. Unlike the Middle Iranian clitics, though, these do not gravitate to the Wackernagel position, but are hosted by the first available constituent of the VP (see Öpengin (2013) for a more detailed account). Different possibilities are illustrated in (26)–(28). In (26), the clitic
482 Geoffrey Haig is hosted by the P, undoubtedly the preferred option, but in the absence of the P, the clitic drifts rightward to the first stressed element of the predicate, in (27) the negation prefix. In the absence of any other hosts, the clitic will land on the verb stem itself, as in (28). (26) min sag-aka=m 1sg dog-def=clpro.1sg ‘I didn’t kill the dog’
na-kušt neg-kill.pst(3sg)
(27) min na=m=kušt 1sg neg=clpro.1sg=kill.pst(3sg) ‘I didn’t kill (it)’ (28) min kušt=im 1sg kill.pst=clpro.1sg ‘I killed (it)’ Note that ‘every single past transitive construction requires an A-past clitic’, regardless of the presence or absence of an overt A constituent in the same clause (Haig 2008: 288). In other words, despite the evidently clitic nature of the marker itself, functionally, it is an agreement marker (see Samvelian 2007 for this position). The agreement marker for S, as in (24), or for a present A as in (25), on the other hand, is a verbal suffix, and cannot be hosted by any other constituent. Thus the present tense version of (25), using a mobile clitic to index the A, is hopelessly ungrammatical for the meaning provided: (29) *min sag-aka=m na-kuž-ē 1sg dog-def=clpro.1sg neg-kill.prs-3sg Intended: ‘I am not killing the dog’ The question of whether the past transitive constructions of Central Kurdish qualify as ‘ergative’ cannot readily be resolved. As mentioned, case marking is irrelevant, as there is no overt case morphology. With respect to agreement, one could argue for A=S (i.e. accusative), because both involve obligatory agreement. On the other hand, it could be argued that A≠S, because the nature of the markers used for agreement with A and S are different (mobile pronominal clitics versus verbal suffixes). As there is no principled way to resolve the matter, the choice of taxonomic label is of no theoretical import (Haig 2008: 305). Turning now to agreement, the basic system is that a verbal suffix, of the same paradigm as those used to index an S, indexes a P, but only under the condition that an overt P is not present in the clause. Thus P-indexing is not obligatory agreement, in the sense that A-indexing is, but is alternating, or conditioned indexing. The difference is shown by the following, taken from the Mukri dialect of Central Kurdish (Öpengin 2013, glosses and transcription slightly modified): (30) a. dena de=y-kušt-in otherwise ipfv=clpro.3sg-kill.pst-3pl
Deconstructing Iranian ergativity 483 b. dena ewāni=ī otherwise them=clpro.3sg ‘otherwise he would kill them’
de-kušt *-ini ipfv-kill.pst *-3pl
However, a plural agreement suffix may appear on the verb if the P is nominal, rather than pronominal, though the details remain somewhat obscure (Mukri Kurdish, Öpengin 2013; note the vestiges of the Oblique case in this dialect in (32), kuř-eke-ī): (31) estēre-k-āni=mān de-bižārt-ini star-def-pl=clpro.1pl ipfv-count.pst-3pl ‘(we would sit at night and) count the stars.’ (32) nāme-k-āni=ī dā-ni be kuř-eke-ī letter-def-pl=clpro.3sg give.pst-3pl to boy-def-obl ‘he gave the letters to the boy.’ A further complication with agreement in past transitive clauses is that the set of verbal suffixes that would normally index a P may be co-opted to index a non-core argument. The non-core argument concerned is most typically a recipient, benefactor, or external possessor (33), but is often an adpositional complement, in which case the adposition itself remains ‘stranded’, while its complement is expressed through the verbal suffix, cf. (34) and (35): (33) bač-ka-kān=ī a-xward-im child-def-pl=clpro.3sg ipfv-eat.pst-1sg ‘It used to eat my children’ (MacKenzie 1961: 115, glosses and transcription modified) (34) ew beserhāt-e=ī bo gērā-m-ewe dem adventure-def-clpro.3sg to narrate.pst-1sg-asp ‘(he) narrated this adventure to me …’ (Mukri dialect, Öpengin 2013) (35) feqet qise=m lē de-pirsī-y only issue=clpro.1sg from ipfv-ask.pst-2sg ‘I would only ask council of you.’ (Mukri dialect, Öpengin 2013) This construction, solidly attested in Middle Iranian (Jügel 2015: 378), is one of the most intensely discussed features of Central Kurdish syntax; see MacKenzie (1961), Samvelian (2007), Haig (2008), and Öpengin (2012, 2013). Co-opting an agreement suffix for non- core arguments is not possible in present tense transitive clauses, where verbal suffixes are restricted to agreement with an A. I would interpret this as additional evidence for the fundamental instability of P-agreement in past transitive clauses: it is either missing entirely, or is facultative, or it is co-opted for other constituents.
484 Geoffrey Haig
20.5 Balochi: the Interplay of Inherited and Innovated Case Balochi is a cover term for a bundle of related north-west Iranian languages spoken in southeast Iran (Sistan and Balochistan province), and in the neighbouring regions of Afghanistan and Pakistan, including a sizeable speech community in Karachi (see Figure 20.1). Baloch have also migrated further afield, with speech communities in Turkmenistan, the Gulf states, East Africa, and more recently to North America, Europe, and Australia. Balochi is divided roughly into three dialect groups: Western Balochi, Southern Balochi, and Eastern Balochi. Like Kurdish, Balochi is not the official language of any particular state, and competing versions of more or less standardized written forms coexist (see Jahani & Korn (2009) for a more detailed overview). Balochi exhibits a clear case of tense-based alignment split, with the typical concentration of variation in the realm of past transitive clauses, while the morphosyntax of present-stem clauses exhibits the unbroken monotony of accusative alignment. This section largely draws on the surveys by Korn (2008, 2009) and Jahani (2015), though the terminology and mode of presentation differs in some respects from the sources, and in the interests of brevity a number of complexities have been omitted; readers should consult the originals in case of doubt.
20.5.1 Case Marking in Balochi With regard to the case system, there are three crucial differences between Balochi and Kurdish. First, all dialects of Balochi have developed additional innovated structural cases; second, the case marking of first and second person pronouns often differs from third person nominals; third, the marking of P is often mediated by DOM. Taken together, these factors yield systems of some complexity, and among scholars of Balochi, there is no agreement regarding ‘the number of cases and what they should be called’ (Jahani & Korn 2009: 651). Table 20.5 shows the system which Korn (2009: 46) assumes to be the common underlying case system for nouns (I have omitted the vocative case, and the footnotes; there is no grammatical gender in Balochi). Table 20.5 Case and number marking on nouns in Balochi Direct Singular Plural
-ø
Oblique
Objective
Genitive
-ā
-ārā
-ay
-ān
-ā(n)rā
-ānī
Deconstructing Iranian ergativity 485 Pronouns of the first and second person show considerable cross-dialect variation, and according to (Korn 2009: 47), a common system cannot be reconstructed for them. It is not clear whether the distinction between Direct and Oblique, which is assumed in Table 20.5 for nouns, should be also be maintained for first and second person pronouns.8 For illustrative purposes, Table 20.6 provides the Southern Balochi forms only, and does not assume a Direct/Oblique distinction (based on Jahani & Korn (2009: 653–654), but ignoring the locative). Table 20.6 Case and number on first and second person pronouns, Southern Balochi Direct/Oblique
Objective
Genitive
1p sg
man
m(a)nā, manārā
m(a)nī
1p pl
mā
mār(ā)
maē, mē
2p sg.
taw, tō
t(a)rā, tarārā
taī
2p pl
šumā
šumārā
šumē
The Objective case is a clear example of an innovated object marker (Haig 2008: Ch. 3), as opposed to the inherited Oblique case illustrated in 20.4.1 for Northern Kurdish. The Objective displays an obvious phonological resemblance to the Persian object clitic =rā, which is known to have grammaticalized from a postposition with basically benefactive senses. Whether the Balochi Objective case is the result of an independent, but parallel grammaticalization, or was borrowed into Balochi from Persian, or both processes worked in parallel, is not clear. The Balochi case system is rendered more complex by the parallel existence of a dedicated ‘Genitive’ case, used for prenominal possessors, though never for S, A, or P. With the exception of some dialects of Iran and Turkmenistan (see below), most dialects of Balochi show some version of A≠S in their case marking alignment in past tense clauses. However, no dialect of Balochi has canonical ergativity in the sense outlined in 20.4.1 for Northern Kurdish. In Balochi, S and P are distinct in their agreement systems (see below) and frequently also differ with regard to case marking. The A in the past tense is generally Oblique/Objective marked across Balochi, with the exception of some dialects of Iran and Turkmenistan (see below). The P, if third person, goes into the Direct case. Together this yields ‘model ergative’ case marking (Korn 2008: 252). With first and second person pronouns, the case system works differently
8 Table 4 of Korn (2008) and table 0.3 of Korn (2009) imply that a three-way distinction Direct vs. Oblique vs. Objective is relevant in the paradigm, at least for some dialects, while Jahani & Korn (2009) collapse Direct and Oblique to a single column, the convention followed here.
486 Geoffrey Haig (see below). (36) and (37) illustrate third person arguments, while (38) illustrates the use of a clitic pronoun with the A: (36) sābir-ā ē hawāl uškit Sabir-obl dem news(dir) hear.pst(3sg) ‘Sabir heard this news’
(Korn 2008: 252, W. Balochi)
(37) āy-ā gōk kušt dem-obl cow(dir) kill.pst(3sg) ‘He/she killed the cow
(Korn 2008: 252, S. Balochi)
(38) pīālā=ȭ bowl(dir)=clpro.1sg ‘I have taken the bowl’
(Korn 2008: 254, S. Balochi)
zurt-a seize-perf(3sg)
20.5.2 Variation in the Marking of P: the Impact of DOM All dialects of Balochi exhibit DOM in their present tenses, with discourse identifiability (definiteness) as the main triggering factor. Examples (39) and (40) contrast an overtly marked (Objective) P with an unmarked P (Bohnacker & Mohammadi 2012: 67–68, transcription follows the source, glosses modified): (39) man wtī dōst-ā har rōč a gend-on 1sg.dir refl.gen friend-obl every day ipfv see.pres-1sg ‘I see my friend every day’ (40) man ya davār-ē gend-on 1sg.dir one house-indf see.pres-1sg ‘I (can) see a house’ Bohnacker & Mohammadi (2012: 69) note that DOM in Balochi ‘has not been studied systematically’, but suggests that the nature of the determining factors are very similar to those of Persian, which undoubtedly involve some notion of discourse recoverability and identifiability. Obviously, a first or second person P will always be marked, as it is inherently definite and identifiable. The system can be summed up in the following hierarchy, a variant of the Silverstein, or Animacy, Hierarchy discussed in numerous publications: (41) Hierarchy for DOM in Balochi: decreasing probability of overt marking of P 1st/2nd person pronouns >third person definite >third person indefinite Here ‘third person’ refers to all nouns, including pronouns, demonstratives, etc. The question that we need to address concerns the impact of DOM on the P in an ergative construction, which as noted earlier, is generally in the Direct case. In fact, DOM does
Deconstructing Iranian ergativity 487 impact on the marking of P in both the ergative and the accusative constructions, but its effects occur at different points on the hierarchy given in (41). The system is summed up in Table 20.7, showing case marking of direct objects in present and past tenses, with the shaded regions contrasting the effects of DOM in the tenses. Table 20.7 The respective range of DOM in past and present transitive constructions Position on hierarchy present tense past tense
1st/2nd person
3rd person definite
P is Objective P is Objective
3rd person indefinite P is Direct
P is Direct/Obj /Obl
For the present tense, we observe what was discussed earlier, namely that a definite P, regardless of person, will be overtly marked with the Objective case. In past transitives, we see that while first and second person P are always marked with the Objective, and indefinite third person are never marked with the Objective, for definite third person P, there is variation, indicated by pale shading. Essentially what is happening is that DOM operates in both tenses, but with different cut-off points. All first and second person P’s are overtly marked, but in the past tenses, overt P-marking has not diffused down the hierarchy to consistently affect definite third person P. What Table 20.7 does is to identify the ‘variation hot-spot’ in the case marking of P, namely third person definites, in past tenses. These points are illustrated in the following. First, a first or second person pronominal P in the past tense is always in the Objective case, across all dialects: (42) bādšāh-ā man-ã: khušth-a king-obl 1sg-obj kill-perf(3sg) ‘The king has killed me’ (Korn 2008: 263, Eastern Balochi; the object is glossed ‘obl’ in the original but I have adapted the gloss in line with Table 20.6) (43) rāh-ā mn-ā tunn-ā jat-a way-obl 1sg-obj thirst-obl strike-perf(3sg) ‘On the way, thirst has struck me’ (Korn 2008: 263, Western Balochi) For a third person P in the past tense, on the other hand, there is considerable variation. First, it may occur in the Direct case, and this appears to be the normal option in Southern Balochi outside the Karachi dialect (Carina Jahani, p.c.): (44) mard-ā bačakk-Ø dīst-ant man-obl child(pl) see.pst-pl ‘The man saw the children’
(Jahani 2015)
It may also occur in the Oblique, though this is apparently ‘not very common’ (Korn 2008: 261), and only possible when the P is definite. Finally, the P may be in the Objective case: Korn (2009: 263) suggests that ‘in Balochi of all major groups’, a P that is
488 Geoffrey Haig human may be in the Objective case if definite, though this has since been questioned for Southern Balochi (Carina Jahani, p.c.). The factors determining the variation in the case marking of a definite third person P are not fully understood. The important point for this overview presentation is that we identify the underlying commonalities across all dialects, namely that DOM is operative in all tenses, but it ‘lags behind’ in the past tenses, where it only consistently affects first and second person objects. Third person definites are the realm of variation.
20.5.3 Verbal Agreement As mentioned, verbal agreement in past transitives works differently from intransitives: only plural number appears to be regularly reflected in verbal agreement, while the category of person is not involved (Jahani and Korn 2009: 663). Jahani (2015), qualifies this statement, pointing to Southern Balochi examples where person marking indexing a P is possible. However, this only occurs in the absence of an overt P in the clause, suggesting that this kind of P-indexing is pronominal rather than agreement. These examples do not change the fundamental fact that in the past tenses, verbal agreement with P works differently to verbal agreement with S, the latter being always obligatory, across all dialects. If a P is plural, then plural number may be indexed on the verb, though this seems to be most frequent under the condition that overt plural marking (via a plural suffix) of the P is absent. Lack of overt coding of plurality can arise under two conditions: first, the object itself is not expressed at all (because its reference is contextually recoverable), as in (45). Second, the object is overtly expressed, but is in the direct case and hence carries no overt morphological signal of plurality (cf. Table 20.5), as in (46):9 (45) nũ: gwāt čanḍ-ēnt-ē now wind swing-caus.pst-3pl ‘Now the wind swung them (=the clothes)’
(Korn 2008: 256, Iranian Balochi)
(46) ã: hī-ā kull-ẽ: bandī yala kuθ-aɣ-ant dem-obl all-adj prisoner(dir.pl) free do-perf-3pl ‘He has freed all the prisoners’ (Korn 2008: 253, E. Balochi) If the plurality of the P is overtly expressed, plural agreement on the verb is not necessary: (47) mā mard-ãnrā jaθ-a 1sg.obl man-obj.pl strike-perf(3sg) ‘I struck the men’
(Korn 2008: 261, East Balochi)
9 Carina Jahani (p.c.) points out that plurality of the P can be indicated in other ways, for example through quantifiers, and the verb may still index plural number of the P. The relevant condition thus seems to be lack of the plural suffix on the P itself, not general lack of contextual indication of plurality.
Deconstructing Iranian ergativity 489 The above-mentioned condition for number agreement with the P is a necessary, but not a sufficient one. Even when plurality is morphologically expressed on the P, plural agreement may still appear on the verb, though this appears to be unusual, and subject to dialectal variation. In sum, verbal agreement with a P is (i) restricted to the feature of number (excluding the pronominal indexing of Southern Balochi (Jahani 2015); and (ii) is in part dependent on the presence or absence of overt plural marking on the P itself. It is thus not the kind of obligatory agreement in person and number that characterizes verbal agreement with S in all dialects of Balochi. This again confirms the findings from Kurdish, that agreement with the P is the least stable and consistent aspect of ergativity.
20.5.4 Iranian Balochi: Loss of Ergativity The Western Balochi dialects spoken in Iran have alignment systems that differ from the ergative systems just outlined. Two types are illustrated here: the Sarawani dialect of southeast Iran, described in Baranzehi (2003), and the Sistan dialect from the southeast periphery of Iran (Delforooz 2010). Turning first to the case system for nouns in Sarawani, we find a two-way distinction between Nominative, and Objective, shown in Table 20.8 (Korn 2008: 255). Table 20.8 Case system for Sarawani dialect of Iranian Balochi Nominative
Objective
Singular
-Ø
-ā(rā)
Plural
-ān
-ānā
In these varieties, the past A is indexed via a pronominal clitic, though the clitic does not co-occur with a co-referential free pronoun in the same clause (48). It is typically hosted by the first lexical category of the VP, as in (49) and (50). (48) mō dāt-ē ramazān-ā […] 1sg(nom) give.pst-3pl Ramazan-obj ‘I gave (them) to Ramadan’ (Sarawani dialect, Baranzehi 2003: 83) (49) zekk-ī tālān kort … goat.skin=clpro.3sg pouring do.pst(3sg) ‘She poured (oil from) the goatskin […]’ (Sarawani dialect, Baranzehi 2003: 83) Example (48) also illustrates the ‘anaphoric’ usage of plural agreement with the P, discussed earlier, which also characterizes this dialect. In the past tenses, a third person P is apparently consistently unmarked even when definite (50), while first and second
490 Geoffrey Haig person P, in line with the generalization summarized in Table 20.7, are Objective marked (not illustrated here). (50) Alī Hasan=ī zat Ali(nom) Hasan=clpro.3sg strike.pst(3sg) ‘Ali struck Hasan’ (Korn 2008: 258, Iranian Balochi, Lashari dialect) In the present tenses, however, a definite third person P is marked with the Objective case: (51)
[…] dān-ān-t-a be-bār … grain-pl-poss2sg-obj imper-take(2sg) ‘(come and) take your grain(s)!’
(Baranzehi 2003: 82)
Thus with regard to case marking of P, the Sarawani dialect complies with the system outlined in Table 20.7: we find DOM in both tenses, but the cut-off point in the past is higher, with only first and second person objects overtly marked. The most remarkable aspect of the Sarawani dialect is the lack of overt marking of A, already illustrated. Here then, we have S=A in case marking, though not in agreement. To what extent this is ergative is again a matter for (futile) debate. The system shows considerable parallels to that of Central Kurdish, except that a case distinction is still available, and used to mark P in the present tense. In the past tense, however, we have the same lack of case marking, and frequent use of pronominal clitics to index the A. The final pattern to be considered is the complete absence of ergativity, illustrated with the Sistan dialect of Balochi (Delforooz 2010). Delforooz provides no sketch grammar or case paradigms, but the basics of the system are readily apparent from the abundant text material. A very similar system, at least with regard to alignment, is also documented for Balochi of Turkmenistan (Axenov 2006). In these varieties, we find A=S in all tenses: both are nominative, and both control verbal agreement. The marking of P is consistent across all tenses, and mediated by DOM: somewhat simplified, the rule is that if definite, P is marked with the Objective case, whereas an indefinite P is unmarked for case. Examples for the direct-marked A, and an indefinite, unmarked P is (52); an example for an Objective-marked, definite P is (53): (52) tajjār p=ēšān gwarag=ē kušt merchant for=dem.pl.obl lamb=indf kill.pst(3sg) ‘The merchant slaughtered a lamb for them’ (Delforooz 2010: 344) (53) gwālag-ā purr=ē kurt=u sack-obj full=clpro.3sg do.pst(3sg)=and ‘(he) filled the sack and …’
(Delforooz 2010: 165)
The occasional use of a third person pronominal clitic to express a past A is illustrated by purr=ē ‘full=clpro.3sg’ in (53). With the exception of the latter phenomenon, alignment is basically identical to that of Persian, with unified accusative case marking in both tenses, unified S/A agreement, and DOM mediated by definiteness.
Deconstructing Iranian ergativity 491
20.5.5 Summary of Balochi The Balochi system illustrates the interplay of the inherited case system and an innovated object marker. Common to all of Balochi is the possibility of innovated object markers to mark a definite P, regardless of tense, though what counts as ‘definite’ varies from dialect to dialect, and according to tense. All varieties case-mark a first and second person P in all tenses; most mark a definite third person P in the present tense, but there is variation across the dialects with regard to case marking of third person, definite P in past tenses. In no dialect is the cut-off point on the Animacy Hierarchy for marking the P higher in the present tenses than in the past tenses. The overt marking of first and second person P can be related to the origins of the Objective case, typically a grammaticalized postposition that marked recipients. Recipients are very frequently first and second persons, thus the link to first/second person direct objects (textually actually rare) is quite obvious, and reflected in similar systems, e.g. DOM in Spanish. From the first/second person, Objective marking encroaches downwards to third person definite objects, at dialectally differing paces, and differing according to the tense of the verb. Interestingly, the addition of an innovated object marker to a basically ergative case-marking system does not automatically lead to breakdown and replacement of the earlier ergative system. Instead, we find systems combining overt marking of A (inherited Oblique case) with Objective marking of the P (with an innovated object marker). However, the Sarawani dialect shows a development otherwise unattested: the complete abandonment of the overt marking of A, yet the maintenance of the zero-marking of third person P.10 Notably, this dialect makes widespread use of pronominal clitics to index the A, exhibiting a striking similarity to Central Kurdish (20.4.2). But the nature of the case system in Sarawani is undoubtedly unusual in the Iranian context.
20.6 Ergativity in Iranian Taleshi Taleshi refers to the north-west Iranian language spoken by a group of speech communities along an approx. 100 km-long strip of the south-west Caspian coast, from just north of Rasht in Iran and extending into Azerbaijan (see Figure 20.1). The following data stem from Paul (2011), who focuses on the three main dialect groups of Taleshi in Iran: the northern region (Anbarâne Ardabil, here abbreviated Anb.), central (Asâlam, abbreviated Asâ.), and southern (Mâsâl Šânderman, abbrev. Mâs.). Of these, the dialect of Mâsâl is considered the ‘purest’ representative of the group by native speakers (Paul 2011: 18). Taleshi exhibits the two-stem verb system, familiar from the preceding discussion, and in the interests of comparability, I will continue to employ the labels ‘present’ and ‘past’. However, these labels are less appropriate for Taleshi, for the following reasons. First, in the northernmost dialect, Anbarani, the stem distinction has been lost on all but seven 10 Geoffrey Khan (p.c.) points out that some North-Eastern Neo-Aramaic dialects show a parallel development in their perfect verb forms.
492 Geoffrey Haig high-frequency verbs, yielding a single stem for all tense and aspect forms (in most cases a reflex of the old past stem, cf. Paul 2011: p. 104, table 23). In cases of uncertainty, the labels past/present are omitted from the glosses. Second, all dialects have verb forms expressing past imperfective aspect, but based on the present stem of the verb. In Anb. a preverbal augment precedes the stem (54), while in Mâs. no such augment occurs (55): (54) gândəm devan a-k-im wheat scythe aug-do.pres-1sg ‘I was scything the wheat.’ (55)
(Anb. dialect, Paul 2011: 129)
zua-te-i=na vâ-in pisakula boy-dim-indf=loc say.pres-3pl baldy ‘(She had a little boy.) They used to call the little boy baldy.’ (Mâs. dialect, Paul 2011: 131)
In accordance with the general rule for Iranian (Haig 2008: 10), alignment with such verb forms is dictated by the verb stem (the present stem), rather than the actual time reference of the entire predicate (past). Constructions such as (54) have thus predictably accusative alignment. Taleshi has also developed periphrastic progressive forms, used in both past and present contexts. Either a clitic form of ‘be’ is used, or a frozen form of kârd ‘do’. Alignment in such clauses is also accusative, regardless of actual time reference (Paul 2011: 103, table 22).
20.6.1 Case Marking in Taleshi The case marking of nouns is provided in Table 20.9; it shows obvious parallels to that of Northern Kurdish and some dialects of Balochi. The marking of plural differs, however: unlike Northern Kurdish and Balochi, we find distinct forms for singular Direct (zero), and plural direct (-e). The exception is Anb. dialect, which has gone the same way as Central Kurdish, generalizing the Oblique plural form to the Direct case, hence neutralizing the case distinction in plural number. Table 20.9 Case and number on nouns in Taleshi
Singular
Plural
Direct
Oblique
Anb.
-ø
-ə
Asâ.
-ø
-i
Mâs.
-ø
-i
Anb.
-un/-ün/-ân
Asâ.
-e
-un/-mun
Mâs.
-e
-ân
Source: Adapted from Paul 2011: table 15
Deconstructing Iranian ergativity 493 Table 20.10 (adapted from Paul 2011: 81) provides the paradigms for first and second person pronouns. It will be noted that the Oblique/Direct distinction is only overtly realized in the first person singular. A second important difference to the case system with the third person (see Table 20.9) concerns the functional range of the Oblique case. With nouns, the Oblique is also the case for prenominal possessors (i.e. a ‘Genitive’). With first/second pronouns, however, an additional form is used for the Genitive, which is generally assumed to derive from the fusion of the Oblique form of the pronoun with a preposition related to Old Iranian *haca ‘from’, for example Anb. cəmân ‘1sg.gen’. The same kind of formation is also found with demonstrative pronouns. In the interests of brevity, the special possessives are not treated further here; see Paul (2011: 84) for discussion. Finally, an additional ‘Accusative’ form is entered in the table for the Oblique of Anb. This point is taken up below. Table 20.10 Case and number on first and second person pronouns in Taleshi Direct 1S
Oblique
Anb.
Asâ.
Mâs.
Anb.
Asâ.
Mâs.
âz
az
az
mâ(n)/mânə (Acc.)
mən
mə(n)
2S
tə
tə
tə
tə
tə
tə
1P
ama
ama
ama
ama
ama
ama
2P
šəma
šəma
šəma
šəma
šəma
šəma
In Taleshi, ergative case marking is restricted to transitive clauses in the simple past (based on the past stem), and the perfect tenses (based on a participle, secondarily derived from the past stem). In all other environments, accusative alignment is found. Accusative alignment shows no variation; it is ‘the same in all three dialects’ (Paul 2011: 92). The A is unmarked (Direct), P is marked (Oblique). The marking of P is mediated by DOM: only definite direct objects (in the sense of ‘identifiable’, as defined in Lambrecht 1994: 77) are overtly case marked. Paul (2011: 69) refers to ‘specificity’ as a triggering factor, but his examples contain specific, indefinite objects which are not case-marked, as in: (56) əm əšta=râ ka=i sâz-ə dem self=for house=indf build-3sg ‘This one builds a house for herself.’ (Paul 2011: 71, Mâs. dialect, glosses adapted) In view of these and other examples, I will continue to refer to ‘definiteness’, rather than specificity as the relevant factor behind DOM in Taleshi. Obviously, pronouns of the first and second person are definite, and are overtly case marked when in object function. Ergative patterns, found with simple past, and perfect tenses, show the by now familiar variability. The main dialect division lies between Anb. and Asâ. dialects on the one hand, and Mâs. on the other, and I treat these two groupings separately before drawing more general conclusions.
494 Geoffrey Haig
20.6.2 Anbarane and Asâlem Dialects: Case Marking in Past Tense Transitives The A is consistently in the Oblique case, while P is consistently in the Direct. The effects of DOM are restricted, and only affect a P that is first or second person. The case marking of a third person P, on the other hand, is invariably Direct, regardless of definiteness. Consider the following: (57) əm pis-i əštan kis-e iâ nâ=n dem baldy-obl self ’s bag-pl.dir here put=tr.pl camun kis-e=əš ž=in əštan asb-i poss.3pl bag-pl.dir=clpro.3sg load=tr.pl self ’s horse-obl ‘This baldy put his own bags here and loaded their bags onto his horse.’ (Paul 2011: 93, Asâ. dialect, glosses adapted) Both the P’s in this example are definite, but are in the Direct case (plural -e). Note further that they trigger number agreement with the verb. This example also illustrates the use of a clitic pronoun to express the A in the second clause (see 20.6.3). Even a pronominal P (third person) remains in the Direct case, cf. av ‘him’ in (58):11 (58) səpa užnan av gat=e dog(obl) again 3sg bit=tr ‘The dog bit him again.’
(Paul 2011: 93, Anb. dialect, glosses adapted)
Anb. and Asâ. dialects thus show fairly consistent Oblique marking of the A, in all persons, Direct marking of the P (if third person), and number agreement with P. As in Balochi, a P that is first or second person, always goes into the Oblique case (only actually visible for the first person singular, cf. Table 20.10): (59) užna=š=an again=clpro.3sg=also ‘He struck me again.’
mân 1sg.obl
ža hit.tr (Paul 2011: 98, Anb. dialect, glosses modified).
(60) av-ə mân nəfin kârd=e […] 3g-obl 1sg.obl curse did=tr ‘He cursed me so [...].’ (Paul 2011: 99, Anb. dialect, glosses modified). Past transitive verbs do not agree with a first/second person (singular) P. It would be interesting to see whether such verbs agree with a first/second person plural P, at least in 11
Overt expression of the Oblique singular appears to be systematically suppressed on nouns ending in -a, hence the absence of the expected Oblique case on səpa ‘dog’ in (58).
Deconstructing Iranian ergativity 495 number. Unfortunately, no clear examples of such clauses were found, so this question remains unanswered. In sum, the case marking of A is consistently Oblique,12 so we have consistent A≠S. The case marking of P in the past tenses exhibits the same tendency identified for Balochi in Table 20.7. The difference is that in Taleshi, overt marking of a P in the past tenses is completely restricted to first and second persons. The system is illustrated in Table 20.11. Table 20.11 Case marking of P and the interaction of DOM and tense in Taleshi, Anb., and Asâ. dialects Position on hierarchy present tense past tense
1st/2nd person
3rd person definite
3rd person indefinite
P is Oblique P is Oblique
P is Direct P is Direct
20.6.3 Ergativity in the Mâsal Dialect In the Mâs. dialect, there are two quite distinct constructions available for past transitive clauses. The first is identical to that illustrated above for Anb. and Asâ. dialects, and need not be discussed further (see Paul 2011: 95). The second construction is very different and involves, quite remarkably, fully accusative case marking. The A takes the Direct case, the P (if definite) the Oblique. This construction is apparently contingent on the presence of a pronominal clitic, indexing the A, and hosted by the verb (note that the pronominal clitic occurs inside the so-called transitivity marker (tr) on the verb). Thus we have: A: Direct P: Oblique, if definite Pronominal clitic indexing the A, and hosted by the verb (61) a əštan tüng-i avi â-kard=əš=a 3sg: dir self ’s jug-obl loss pvb-caused=clpro.3sg=tr ‘He lost his jug.’ (Paul 2011: 95, Mâs. dialect, glosses modified) (62) az kuf-i tâ kard=əm=a 1sg: dir felt-obl fold did=clpro.1sg=tr ‘I folded the felt.’ (Paul 2011: 101, Mâs. dialect, glosses modified)
12
The ‘Accusative’ form of the Anb. first person singular is also used to mark the P in present-stem clauses (Paul 2011: 98). I have no explanation for the distribution of this form.
496 Geoffrey Haig These constructions are almost identical to those of the Sarawani dialect of Balochi, as discussed. But here, it appears that the Direct case of the A is contingent on the pronominal clitic being hosted by the verb. Why the presence of the pronominal clitic on the verb should have such a drastic impact on the case marking is not immediately obvious, though a correlation between pronominal clitics and case marking is discernible across Iranian, and is taken up in the summary (20.6.4). One can of course note that when the pronominal clitic is hosted by the verb, as in (62), the clause has an undeniable surface similarity to a transitive clause in the present tense, where the A is also indexed on the verb, via agreement suffixes.13 Indeed, in many parts of the paradigm, the pronominal clitics and the corresponding agreement suffixes are near-identical in form, thus heightening the parallels. One might conjecture, then, that surface similarity in the agreement system left the present and past constructions so close that only a minor change (a change in case marking of A, and of definite P) was required to bring the past construction in line with that of the present. The change can be seen as resulting from the pressure towards cross-system harmony (Haig 2008: 192–198).
20.6.4 Summary of Taleshi Like Balochi, the Taleshi dialects surveyed here show a range of case-marking strategies in their past transitives, from clearly ergative to fully accusative. And in both groups, DOM mediates the marking of P. Likewise, both groups have preserved, to varying degrees, the system of indexing an A through a pronominal clitic. The differences stem from the fact that there is no innovated object marker in Taleshi, while in Balochi, marking of objects is largely (perhaps exclusively) effected via an innovated object marker. Although the overall range of variation in the Taleshi dialects surveyed here appears less than that of Balochi, that is probably an artefact of the smaller number of dialects sampled and the more restricted scope of the material available.
20.7 Conclusions and Outlook Ergativity emerged during the transition from Old to Middle Iranian. I have suggested that the main mechanism involved the extension of constructions already existing in Old Iranian, to become the sole means for expressing past-time reference in the daughter languages. The main mechanism is thus extension, rather than restructuring or reanalysis. 13 The available material in Paul (2011) does not permit a satisfactory analysis of this construction. Furthermore, example (243) on p. 101 has the clitic on the verb, but the A in the Oblique case. It seems possible that some of the A’s could be seen as examples of left-dislocated topics, which would go into the Direct case, and are then resumed by the pronominal clitic on the verb.
Deconstructing Iranian ergativity 497 Although ergativity in the sense defined here has been a remarkably persistent trait of much of Iranian, there is nevertheless an undeniable sense of instability when one surveys the family. On the one hand we have straightforward accusative alignments, with little variation in case and agreement, throughout the present tenses. This relative stability contrasts sharply with the situation obtaining in past transitive constructions, where we are confronted with a range of distinct variants, even for relatively small dialect clusters like Taleshi. Why should the Iranian ergative constructions so consistently exhibit this degree of inter-and intra-language variability, and tendency towards diachronic change? One explanation is in terms of pressure towards cross-system harmony (Haig 2008). On this view, the instability of the ergative construction arises not from anything inherently instable in ergativity itself, but from the fact that speakers must essentially master two grammars for case and agreement alignment for transitive clauses, one applying to the present tenses and the other to the past. Now there is no obvious semantic or functional connection between tense and case marking: Case marking is concerned with predicate/ argument relations, while tense is concerned with temporal reference. In other words, the presence of a tense-based alignment split introduces unmotivated and opaque structural complexity into the grammar. All other things being equal, we might imagine that over time, minimal changes accrue that successively iron out the differences, ultimately returning to a unified alignment across all tenses. This is in fact what has generally happened: all the attested changes in the ergative construction of the past transitives can be viewed as bringing them closer to the corresponding present tense constructions. A shift in the other direction would of course have been a logical possibility, that is, we might have expected present tense constructions to adapt towards the ergative structures of the past. To the best of my knowledge, this has not happened anywhere in Iranian. The variation we have witnessed in the preceding sections nevertheless allows for certain generalizations regarding the possible pathways of change. First, we can fairly reliably reconstruct a common constructional schema behind all the variants of ergativity discussed here, as in Table 20.12 (see Haig (2004, 2008: 100) for justification).
Table 20.12 The proto-ergative construction of early Middle Iranian A
P
Verb
Oblique case, expressed through:
Direct:
agrees with P in
*N-closed, unrounded vowel (sg)
*N-ø (sg)
person and number
*N-ān (pl)
*N-ø (pl)
alternatively, the A is expressed via a pronominal clitic, cf. Table 20.1
(N=nominal host of case marking)
498 Geoffrey Haig The A was overtly case-marked, or was expressed through a pronominal clitic, P was unmarked, and the verb agreed with P. If we turn to the various systems we have considered, most can be relatively simply derived from the proto-ergative construction in Table 20.12 via fairly straightforward processes of morphosyntactic change, which have ready parallels outside of ergativity: (i) Loss of the pronominal clitics The ‘canonical ergative’ construction of Northern Kurdish has precisely the case- marking pattern shown in Table 20.12, but it has completely abandoned the pronominal clitics. Complete loss of clitics is also found in Zazaki (not discussed here). The Balochi of Karachi also seems to have largely lost the clitics, at least none are visible in Farrell’s (2003) material. Persian, on the other hand, has retained pronominal clitics, but not in the function of indexing an A past. Taleshi has retained them, but not in the function of adnominal possessors. Thus changes in the clitic system, including complete loss, suggest that they are a relatively autonomous subsystem, subject to various changes which may produce epiphenomenally variants of the ergative construction. (ii) Weakening of agreement with P Northern Kurdish dialects show several deviations from canonical ergativity, most notably affecting agreement with the P, which is either lost, or is controlled by the A. In Balochi, agreement with the P is largely reduced to the feature of number, and is not obligatory. Similarly, agreement with a P only seems to reflect number in Taleshi (though the material is insufficient to pass final judgement on this); agreement with a first/second person P is certainly not possible. In Central Kurdish too, we find weakening of verbal agreement with P: with very few exceptions (e.g. (31)) it is anaphoric, rather than agreement, and the relevant suffixes may be co-opted for indexing other arguments. (iii) Case marking of the P Consistent direct marking of the P is still found in Northern Kurdish, but in all other varieties surveyed, there are exceptions. The commonest occur in connection with DOM: the highest positions on the Animacy Hierarchy (41) are generally overtly marked, rather than Direct. The common pattern for Taleshi and Balochi, and quite possibly for other languages with DOM and ergativity in the past tenses, is shown in Table 20.13. Table 20.13 Common scheme for case marking of P in Balochi and Taleshi Position on hierarchy present tense past tense
1st/2nd person
3rd person definite
P is overtly marked P is overtly marked
Variation
3rd person indefinite P is zero-marked P is zero-marked
Deconstructing Iranian ergativity 499 (iv) Case marking of the A The Oblique marking of the A is surprisingly stable, and is found in all varieties surveyed, with the exception of Central Kurdish, which has lost all case marking. From these facts, some more general principles can be formulated. First of all, it is agreement with the P that is the least stable component, synchronically and diachronically, in Iranian ergativity. Relatively stable P-agreement occurs in East Iranian Pashto, and in some varieties of West Iranian Zazaki, but in both languages it also involves gender agreement, and there is good evidence that gender and person agreement have distinct characteristics. Consistent P-agreement in the feature of person is unknown to me in any coherent dialect group within Iranian. Consistent person agreement with S, or present tense A, on the other hand, is the norm. The second least stable aspect of ergativity in Iranian is the case marking of the P. In many cases, variation can be attributed to the impact of DOM (cf. Table 20.13), but it is not only the effects of DOM. In the double-oblique construction of Northern Kurdish, a P is Oblique marked regardless of definiteness, etc. Synchronically and diachronically more stable than the Direct marking of the P is the Oblique marking of the A.14 Throughout this chapter, I have attempted to analyse the ergative construction in Iranian not in terms of changes from one alignment type to another (e.g. ergative to accusative). Rather, I see the changes in terms of the partially independent workings of distinct subsystems. If we look at the data in this manner, the variation finds a natural explanation: variant constructions are epiphenomena, labels attached to the many contingent combinations that arise through the changes in the respective subsystems. This is not to claim that the individual subsystems are fully independent. On the contrary, there is a loose interdependency that makes itself felt in certain correlations. For example, it is probably no accident that the only language to have completely grammaticalized the clitic pronouns into agreement markers for the A, Central Kurdish, has also completely lost case marking. But outside of this extreme case, several different combinations of clitic pronouns and case marking are found (e.g. in Balochi), so the correlation is weak at best. Whether ergativity itself can be considered in any sense a privileged constellation within Iranian appears doubtful.
Acknowledgments I am indebted to Carina Jahani, Ergin Öpengin, and an anonymous reviewer of this volume for numerous insightful comments and corrections on earlier versions of this chapter. None of these people bear any responsibility for the remaining errors.
14 Speculatively this fact might be related to the high frequency of zero anaphora realizations of A in actual discourse (Haig & Schnell, 2016). Thus, in actual usage, case marking of the A is seldom overtly expressed, hence contributes comparatively little to the differences between the grammars of present and past tense transitive clauses.
500 Geoffrey Haig
Abbreviations ACC, Accusative case; ADD, Additive particle; ADJ, Adjective; AUG, Augment (verbal); CLPRO, Clitic pronoun; COP, Copula; DEF, Definite; DEM, Demonstrative; DIM, Diminutive; DIR, Direct case (traditional term for unmarked case in two-term Iranian case system); DOM, Differential Object Marking; DRCT, Directional particle; EB, Eastern Balochi; EZ, Ezafe particle (adnominal linking element); F, Feminine; GEN, Genitive; IMPER, Imperative; INDF, Indefinite; INDIC, Indicative; IPFV, Imperfective; M, Masculine; N, Neuter; NK, Northern Kurdish; NOM, Nominative; OBJ, Objective case; OBL, Oblique (inherited marked case form in two-term Iranian case systems); POSS, Possessive; PTCPL, Participle; SB, Southern Balochi; SK, Southern Kurdish; TR, Transitive; WB, Western Balochi.
Chapter 21
In transitivit y a nd t h e deve l opment of e rg at i v e alignme nt Edith Aldridge
21.1 Introduction Ergatively aligned languages pose a challenge to the generative approach to argument licensing and case marking. According to Chomsky (2001 and subsequent works), the functional heads finite T and transitive v are each merged with an unvalued ɸ-feature. D(P)s enter the derivation with valued ɸ-features and an unvalued case feature. (1)
Accusative language vTr: [uɸ] vIntr: No [uɸ] TFin: [uɸ]
The unvalued ɸ-feature on T or v acts as a probe and seeks a matching counterpart in its c-command domain. As soon as it finds an appropriate goal, i.e. a valued ɸ-feature set on a DP, the ɸ-feature on T is valued, and the DP supplying the valued ɸ-features is valued for case. Consequently, transitive v values accusative case on the structurally most prominent VP-internal DP (i.e. the object), while T values nominative case on the highest DP in the clause, e.g. the subject. (2)
a. She[NOM] walks. b. She[NOM] sees him[ACC].
502 Edith Aldridge c.
TP
T [u
vP
]
DP[
v’
]
v[u
VP
]
V
DP[
]
The challenge presented by ergative languages is the fact that nominative case does not always appear on the DP structurally closest to T. Specifically, the object has nominative case in a transitive clause rather than the subject. Seediq (Aldridge 2004: 78) (3) a. Wada kudurjak ka qedin=na. pst flee nom wife=3sg.gen ‘His wife ran away.’ b. Wada bube-un na Pihu ka dangi=na. pst hit-tr gen Pihu nom friend=3sg.gen ‘Pihu hit his friend.’ A common approach to this locality problem is to propose that subjects in transitive clauses are assigned some type of inherent case (genitive in Seediq) and consequently do not require nominative case from T. This allows T to ignore the subject and value nominative case on the direct object (Bok-Bennema 1991; Bittner and Hale 1996a; Woolford 1997; Ura 2000; Alexiadou 2001, Chapter 15, this volume; and others). (4)
TP
T [u ]
vP v’
DP [ , INH] v
VP V
DP [ , NOM]
A further consequence is that v cannot have its own set of unvalued ɸ-features, since there is no VP-internal DP to value them. This in turn predicts that ergative alignment arises diachronically in an accusative language as the result of two parameter settings: (1) v in a semantically transitive clause lacks the ability to license structural case;
Intransitivity and the development of ergative alignment 503 and (2) this v is able to assign inherent case to the external argument in its specifier. It is important to point out that each of these properties is independent of the other. In differential object marking languages like Finnish, objects receive structural accusative case in bounded events but not in unbounded events. Note, however, that the subject continues to surface with structural nominative case. Finnish (Kiparsky 1998a: 3) (5) etsi-n karhu-a/#karhu-n seek-1.sg bear-part/bear-acc ‘I’m looking for the (a) bear.’ A shift from accusative to ergative alignment will only be observed when the lack of accusative case for an internal argument is accompanied by the availability of inherent case from v for the external argument. The necessity of the convergence of these two changes also suggests a reason for the relative rarity of ergatively aligned languages, as opposed to the relative commonality of differential object marking in general. In this overview, I summarize how an analysis of this sort has been or can be made to account for accusative-to-ergative change in a variety of languages or language families.
21.2 Passive-to-E rgative Hypothesis A classic approach to the origin of ergative alignment is positing a passive construction as the source. In a passive clause, the external argument—if it surfaces at all—is marked by an adposition rather than structural case, while an internal argument—typically the theme or patient—has nominative case. This bears superficial resemblance to a transitive clause in an ergative language, having an obliquely marked external argument and nominative internal argument. EA
IA
(6) a. DPOBL DPNOM V
(passive clause)
b. DPINH DPNOM V
(ergative clause)
Anderson (1977) has proposed just such an origin for ergative clauses in the perfective aspect in Indo-Aryan languages. Many modern Indo-Aryan languages exhibit a split-ergative alignment, whereby imperfective aspect follows an accusative pattern, and perfective aspect is ergatively aligned. Note the ergative suffix on the subject in the Hindi example in (7b). Other DPs appear without a case marker. Note further that the verb shows agreement with the nominative subject in (7a) and the nominative object in (7b).
504 Edith Aldridge Hindi (Mahajan 1990: 72–73) (7) a. raam roTii khaataa thaa. Ram(m).nom bread(f) eat.ipfv.m was.m ‘Ram (habitually) ate bread.’ b. raam-ne roTii khaayii thii. Ram(m).erg bread(f).nom eat.pfv.f was.f ‘Ram ate bread.’ The ergative pattern in the perfective is generally traced to a construction in Sanskrit built on the participle -ta (Proto-Indo-European *-to), exemplified in (8b). Note the case on the external argument, glossed as “instrumental.” Classical Sanskrit (Klaiman 1978: 205) (8) a. naro vedān pat̥hati man.nom.sg Veda.acc.pl.m recites.3sg ‘The man recites the vedas.’ b. narena vedāh̥ pat̥hi-tāh man.ins.sg Veda.nom.pl.m recite-TA.pl.m ‘The man recited the vedas.’ Anderson suggests that the reanalysis of passive to ergative was motivated by the loss of the inflected perfect and its replacement by the participle, on the basis of the semantic similarity between perfect and passive in that both present a state resulting from a completed action (Anderson 1977: 336). The passive-to-ergative analysis has been adopted by Pray (1976), Bubenik (1989), Hook (1991), and others for Indo-Aryan. Cardona (1970), Payne (1980), Bubenik (1989), and others have made similar claims for the related Iranian languages, which manifest ergative alignment in the past tense. However, questions have also been raised regarding the empirical basis for positing a passive source. Specifically, the source constructions for both the Indo-Aryan and Iranian ergative clause types do not appear to have the characteristics of a canonical passive. Pray (1976) points out that the nominative object in Sanskrit remains in its base position between the agent and verb rather than moving to clause-initial subject position, as can be seen in (8b). Furthermore, the agent functions as a subject in being able to control into a gerundive embedded clause, as in (9a), and serving as the addressee of certain types of imperatives, as in (9b, c). Additionally, an intransitive verb can be inflected with participle -ta, as in (9c). Sanskrit (9) a. tapah̩ kr̩tvā mayā devo ārādhitah̩ austerity having.done by.me god.nom was.propitiated ‘Having performed austerities, I propitiated the god.’ (Pray 1976: 202)
Intransitivity and the development of ergative alignment 505 b. mañjūs̩ām ānāyya pr̩chyantām devatās tvayā chest.acc having.caused.to.bring let.them.be god.nom.pl you.ins ‘Having sent for your chest, inquire of the gods.’ (Pray 1976: 203) c. tvayā gamyatām you.ins let.it.be.gone ‘Please go.’
(Pray 1976: 203)
Haig (2008, 2010, Chapter 20, this volume) shows that the corresponding construction in Old Persian likewise did not have the properties of a passive. Here, too, the genitive agent behaves syntactically as a subject and not as an adjunct by-phrase. Not only is the agent frequently highly topical and animate, this argument can also be the pivot for clausal coordination. Old Persian (Kent 1953; DB III, 47–49; Haig 2008: 52) (10) avaθā=šām hamaranam kar-tam utā thus=3pl.gen battle do-ptcp and avam Vahyazdātam agarbāya utā that Vahyazdata take.prisoner.pst.3pl and ‘They fought battle and (they) took that Vahyazdata prisoner and ….’ In short, there is a lack of evidence that the external argument was ever a demoted adjunct. Facts of this nature prompt Klaiman (1978, 1987) to propose that clauses built on -ta participles were already ergative in Sanskrit times. However, this view raises the question of where the ergative construction came from, a topic which I turn to in the next section. Before concluding this section, I briefly summarize one more well-known passive- to-ergative proposal. Hale (1968), Hohepa (1969), and Chung (1978) have proposed that ergative alignment in Polynesian languages like Tongan and Samoan also arose through the reanalysis of a passive construction. Maori as an accusative Polynesian language with a passive formed by adding a suffix “-Cia,” where “C” is a lexically determined consonant.1 Maori (Chung 1978: 170) (11) a. Ka inu te tangata i te wai. uns drink the man acc the water ‘The man drinks the water.’ 1 Historically, this consonant was part of the base to which the suffix -ia attached. Final consonants were lost in Proto-Central Pacific, a subgroup of Oceanic containing Polynesian, Fijian, and Rotuman, so this consonant surfaced only when followed by a suffix. Eventually, the consonant was reanalyzed as being part of the suffix (Pawley 2001: 196).
506 Edith Aldridge b. Ka inu-mia te wai e te tangata. uns drink the water obl the man ‘The water is drunk by the man.’ In contrast, Tongan and Samoan have transitive ergative constructions which very closely resemble the Maori passive. Note the “e” marker on the external argument. (12)
a. Na’e taa’i ‘e Mele ‘a Sione. pst hit erg Mary abs John ‘Mary hit John.’
(Tongan; Chung 1978: 53)
b. Sā ‘āmata-ina e lātou le pese. pst begin-tr erg they the song ‘They began the song.’
(Samoan; Chung 1978: 55)
Pawley (2001) provides support for the passive analysis of -Cia by showing that a wide range of Oceanic languages both within and outside of the Polynesian subgroup have a suffix -a, which attaches to a transitive verb to derive an intransitive, stative verb. The following examples from the Southeast Solomonic language Arosi show a verb transitivized with the suffix -Ci and then made stative by the further addition of -a. Arosi (Pawley 2001: 200–201; from Fox 1970) (13)
a. age age-ri age-ri-a ‘thatch’ (V) ‘thatch something’ ‘thatched’ b. hunu hunu-‘i hunu-’i-a ‘kill’ ‘kill something’ ‘dead’
However, it is less clear that a diachronic connection can be made between passive -Cia in languages like Maori and transitive morphology in Tongan and Samoan. The transitivizing suffix employed in Tongan is not -Cia but -‘i. Tongan does have -Cia, but this suffix creates intransitives which are often stative, consistent with Pawley’s (2001) findings sketched in (13). Tongan (Pawley 2001: 204) (14) a. Na’a ku tanu-‘i (‘a e kappa). pst 1sg bury-tr abs the can ‘I buried it (the can).’ b. Na’a ku tanu-mia. pst 1sg bury-cia ‘I was buried.’
Intransitivity and the development of ergative alignment 507 Likewise, in Samoan -Cia is not generally used as a transitivizing suffix. Transitive verbs are either bare or take the suffix -ina, as shown in (12b). As in Tongan, when -Cia appears, it creates stative verbs. Samoan (Pawley 2001: 206) (15) a. alu alu-mia ‘go, go out’ ‘be in great demand, sell quickly’ b. fuli fuli-sia ‘turn over, roll over’ ‘turned over’ In short, the intransitive, non-agentive *-Cia seems to be retained broadly in this function in modern Eastern Oceanic languages, including the Polynesian languages, regardless of whether they have ergative or accusative alignment, and it is difficult to see a direct connection between this suffix and transitivizing morphology in the ergative Polynesian languages. It may be more reasonable to posit ergative alignment as conservative with Clark (1973, 1976), Kikusawa (2002, Chapter 23, this volume), and Otsuka (2011a), which may help to reconcile Proto-Polynesian alignment with the non- accusative type of alignment found in higher-order subgroups of the Austronesian family of the Philippines, Taiwan, and elsewhere. To summarize this section, passive is one intransitive source which has been claimed to give rise to ergative alignment. However, empirical evidence does not favor this claim for Indo-Iranian or for Polynesian languages. In section 21.3, I discuss an alternative analysis for Indo-Aryan and Iranian ergative clauses as arising from an active intransitive clause type, specifically a possessive construction.
21.3 Possessive Origin Like their Indo-Aryan relatives, Iranian languages also have split-ergative alignment in which ergative clauses are employed in the past tense, as in Modern Kurdish. Transitive subjects take oblique marking, while intransitive subjects and direct objects are nominative. Modern Northern Kurdish (Haig 2010: 258) (16) a. min tu dît-î 1sg.obl 2sg see.pst.2sg ‘I saw you.’ b. te ez dît-îm 2sg.obl 1sg see.pst.1sg ‘You saw me.’
508 Edith Aldridge c. ez zarok bû-m 1sg child cop.pst.1sg ‘I was a child.’ Old Persian, on the other hand, was a language with uniform accusative alignment. Subjects in both transitive and intransitive clauses are nominative. Old Persian (Haig 2008: 25) (17)
a. pasāva adam kāram frāišayam Bābirum thereupon 1sg.nom army.acc send.pst.1sg to.Babylon ‘Thereupon, I sent an army to Babylon.’ (Kent 1953; DB III, 84) b. yaθā Dārayavahauš xšāyaθiyā abava when Darius.nom king become.pst.3sg ‘when Darius became king’ (Kent 1953; XPf, 36–37)
As in the case of Indo-Aryan, the source for the ergative clause type in the Middle Iranian past tense was the participle ending -ta, from PIE *-to. In this construction, the object had nominative case, and the subject was marked with genitive case. Old Persian (Kent 1953; DB I, 28–29; Haig 2008: 26) (18)
ima tya manā kar-tam that which.nom 1sg.gen do-ptcp pasāva yaθā xšāyaθiyā abavam after when king become.pst.1sg ‘This (is) that (which) was done by me after I became king.’
Benveniste (1952) proposes that this participle construction was originally a possessive construction expressing the perfect, similar to the English “I have eaten.” Like Benveniste, Haig (2008, Chapter 20, this volume) likewise analyzes this construction as expressing (external) possession. This construction was reanalyzed as finite and transitive as a result of the loss of past tense verbal inflection, the participle being co-opted in order to express the past. Old Persian (Kent 1953; DB IV, 75; Haig 2008: 62) (19) utā=taiy tauhmā vasity biyā and.also=2sg.gen seed much may.be ‘and may you have much seed (offspring)’ (lit: ‘and may to you/for you much seed be’) Bynon (2005) has proposed a similar analysis for Indo-Aryan. She claims that the agent was a raised possessor in an anticausative construction that served as an evidential in Vedic. Bynon argues that the subject in the older construction had genitive
Intransitivity and the development of ergative alignment 509 case and that the later Sanskrit construction employing instrumental agents represents an innovation. Vedic (Bynon 2005: 56; MS 1.4.13:62.10; Kulikov 2001: 310) (20) átha yásya kapā́lam bhidyéta and who.gen pot.nom break.3sg.prs.opt tát sám dadhyāt it.nom pv put.3sg.prs.opt ‘And if someone’s dish would break, he should mend it.’ See Butt and Deo (Chapter 22, this volume) for a proposal that the Indo-Aryan ergative construction developed from a result-stative construction. Regarding the structure of the possessive construction, it was clearly intransitive in certain respects. The Proto-Indo-European resultative participle *-to had the distributional properties of an adjective (Haig 2008: 41–42), from which it can be inferred in both Indo-Iranian and Indo-Aryan, only one structural case (nominative) would have been available, this going to the internal argument in semantically transitive constructions, while the external argument was expressed as a possessor with genitive case. One possible formal analysis of this construction is that proposed by Mahajan (1997), as it relates ergativity in languages like Hindi to the HAVE–BE alternation in possessive constructions. In transitive perfect (perfective) constructions in Hindi, the subject is marked with an oblique case, labeled ergative, as already discussed. In contrast, perfect constructions in French (and English) have nominative subjects and accusative objects. Note, too, the difference in the auxiliary: BE in Hindi and HAVE in French. (21) a. Rɑɑm-ne bhinɖiiyɑ̃ɑ̃ pɑkɑɑyii hɛ̃ Ram-erg.masc okra.fem.pl cook.prv.fem.pl be.pres.fem.pl ‘Ram has cooked okra.’ (Hindi; Mahajan 1997: 40) b. Jean a cuit les tomates. ‘Jean has cooked the tomatoes.’
(French; Mahajan 1997: 39)
Following Freeze (1992) and Kayne (1993), Mahajan proposes that possessor subjects in all languages are underlyingly PPs, and the auxiliary is universally BE. HAVE is the result of incorporation of the preposition introducing the subject into BE. But oblique subjects will surface if incorporation fails to take place, and the auxiliary will remain BE. (22)
IP VP PP
BE VP
V
DP
510 Edith Aldridge Mahajan suggests that incorporation is blocked in languages with verb-peripheral word order like SOV, assuming that both government and adjacency are necessary for incorporation. In SVO languages, in contrast, the VP-internal subject occupies the specifier of the VP selected by the auxiliary in Infl. Given the head-initial word order, the auxiliary both governs and precedes the subject, so incorporation can take place. An obvious shortcoming of this approach is that it relies on directionality rather than structural relations alone. The unaccusative structure in (22) also raises some questions in light of the subject-like behavior of the external argument observed in the past two sections. As an alternative, Whitman and Yanagida (2012) opt instead for a semantically transitive structure along the lines of that proposed for modern Hindi by Anand and Nevins (2006). Following Woolford (1997) and Ura (2000), they analyze the external argument as a DP assigned inherent case in the specifier of v. This DP then moves to [Spec, TP] to satisfy the EPP property of T. But nominative case is valued on the object. This structure accounts easily for the subject properties of ergative arguments. Except for the movement of the ergative subject, this proposal is essentially identical to the analysis of ergative alignment in (4). (23)
TP DP[ , ERG]
T’ vP
T [u ]
tERG
v’
v
VP
V
DP [ , NOM]
21.4 Instrumental to NP Split-E rgativity Garrett (1990) proposes an analysis of the origin of NP split-ergativity in Anatolian. The Australian language Dyirbal is an example of a language with NP split-ergativity. Third- person nominals are marked according to an ergative–absolutive alignment, as shown in (24a, b). Intransitive subjects and transitive objects have no overt case-marking, while transitive subjects are marked with a suffix. In contrast to this, first-and second- person pronouns in Dyirbal are marked according to a nominative-accusative pattern. The transitive subject in (24c) is bare, while the object takes a suffix. See also Coon and
Intransitivity and the development of ergative alignment 511 Preminger, Woolford, and Peterson (Chapters 10, 9, and 41, respectively, this volume) for more illustration of person splits in ergative alignment. Dyirbal (Dixon 1994: 161) (24) a. yabu banaga-nyu mother.nom return-nfut ‘Mother returned.’ b. nguma yabu-nggu bura-n father.nom mother-erg see-nfut ‘Mother saw father.’ c. nyurra ngana-na bura-n you.pl.nom we-acc see-nfut ‘You all saw us.’ Silverstein (1976) observes that the Dyirbal facts are part of a larger cross-linguistic pattern. First and 2nd person pronouns and 3rd person animate nominals are more likely to be case-marked nominative-accusative, while those marked ergative–absolutive are more often found at the other end of this animacy hierarchy. (25) NP Hierarchy (Silverstein 1976; revised by Dixon 1994: 85) 1st/2nd Person Pronoun Dem/3rd Person Pronoun Proper N Common N Nom/Acc marking ←============================→ Erg/Abs marking (26) is a Hittite transitive clause with an ergative subject (marked with the ablative marker -anza) and a nominative object. According to Garrett (1990), NP split-ergativity manifests itself in Anatolian in the sense that the ergative suffix attaches only to neuter singular nouns. Hittite (KUB 14.4+rev 23’ [Hymnes 196]; Garrett 1990: 266) (26) nu KUR URU Ḫatti=ya apāš išḫan-anza arḫa namma ptcl land Ḫatti=and that.nom.sg blood-erg.sg pv moreover ‘and that murder moreover ended the land of Ḫatti’ For the source construction, Garrett (1990: 277) posits a transitive clause with an instrumental adjunct and no overt subject. Hittite (Garrett 1990: 277) (27) n=at witenanza parkunuzi ptcl=3sg.acc water.abl.sg pure.caus.prs.3sg ‘S/he purifies it with water.’
512 Edith Aldridge Garrett proposes that an instrumental adjunct was reinterpreted as an agent when the subject did not overtly appear. Another key ingredient of the reanalysis is that it had to have taken place in transitive clauses. This is because of the functional overlap between instruments and agents. Garrett (1990: 265) notes that it is rare—if even possible—for a theme to be packaged as an instrument. In English, for example, instruments can function as subjects only in transitive clauses but not in intransitive clauses. (28) a. b. c. d.
John opened the door with the key. The key opened the door. John walks with a cane. *A cane walks.
Consequently, if an instrument is reinterpreted as a subject, it will always be a transitive subject. In short, this analysis conforms to the type of change sketched in section 21.1. It involves a reduction in the number of structural cases available, since a nominative subject is replaced by a PP external argument. And projecting the external argument as a PP would have allowed T to value nominative case on the object, resulting in ergative alignment.
21.5 Nominalization Source Synchronic or diachronic connections between ergative clauses and nominalizations have been proposed for a wide variety of languages. This is unsurprising, given that nominalizations clearly have the formal properties illustrated in (4), most notably the assignment of inherent—specifically, genitive—case to the external argument. For additional examples, see Alexiadou, Chapter 15, and Salanova, Chapter 43, in this volume)
21.5.1 Inuit Johns (1992) proposes that transitive clauses in the Inuit language Inuktitut are derived synchronically from nominalizations. First note that possessors are marked with the same case as transitive subjects, glossed as “relative.” Inuktitut (29) a. anguti-up nanuq kapi-ja-a-0 man-rel bear.abs stab-pass-ptcp-3sg/3sg ‘The man stabbed the bear.’
(Johns 1992: 61)
Intransitivity and the development of ergative alignment 513 b. anguti-up qimmi-a man-rel dog-3sg ‘the man’s dog’
(Johns 1992: 69)
Johns proposes that transitive verbs combine with a passive participle, -ja in (29a), which nominalizes the verb root. The external argument is merged within the nominal projection and assigned genitive case, while the internal argument is base generated outside the nominalization in subject position. The absolutive argument is assigned both case and Ɵ-role as a result of verb movement to AgrV. (30)
AgrPV (=IP) NP nanuq ‘bear’
Agrv’ AgrPN
Agrv -0
AgrN’ NP anguti-up AgrN ‘man-Rel’ N kapi-ja -a ‘stabbed one’
Since ergative NPs function as the subject of the clause and generally precede absolutives in linear order, Johns proposes that this argument moves to a position above the absolutive NP. The motivation for this movement is case, since its case-assigner AgrN has moved to AgrV, and case is assigned in a spec–head configuration. (31)
AgrPv (=IP)
NP2 anguti-up
NP1 nanuq
AgrPV Agrv’ AgrPN AgrN’
N
tkapi-ja
Agrv kapi-ja-a-0 AgrN
tkapi-ja-a
Aside from the outdated theoretical assumption that case must be assigned in a spec–head configuration, another disadvantage of this proposal is its requirement that the ergative NP move past the absolutive NP, which violates Relativized Minimality (Rizzi 1990b), since both of these NPs are located in case-licensing A-positions. Another question is whether a synchronic derivation involving nominalization is truly
514 Edith Aldridge warranted, an issue which I take up again in relation to Tagalog ergative alignment in section 21.5.3.
21.5.2 Old Japanese Active Alignment Modern Standard Japanese is a canonical accusative language: nominative case appears on subjects in both transitive and intransitive clauses, while objects in transitive clauses have accusative case. Modern Japanese (32) a. Taroo-ga hasit-ta. Taro-nom run-pst ‘Taro ran.’ b. Taroo-ga ringo-o tabe-ta. Taro-nom apple-acc eat-pst ‘Taro ate an/the apple.’ According to Yanagida (2012) and Whitman and Yanagida (2012), Old Japanese of the eighth century likewise showed accusative alignment in finite root clauses. At this time in the history of the language, nominative and accusative cases were not morphologically marked. Old Japanese (Yanagida 2012) (33) a. 我期大王國所知良之 [Wa-ga opo-kimi] kuni siras-u rasi. I-gen great-lord country rule-conc seem ‘My great lord rules seems to rule the country.’ b. 烏梅能波奈伊麻佐加利奈利 [Ume-no pana] ima sakari nar-i. plum-gen blossom now at.peak be-conc ‘The plum blossoms are now at their peak.’
(Manyoshu 933)
(Manyoshu 933)
In contrast to this, case-marking of subjects in nominalized clauses exhibited an active alignment (see also Khan, Malchukov, Woolford, Chapters 36, 11, and 9, respectively, this volume, for other examples of active Split-S alignment). All external arguments in nominalized clauses appear with genitive case, while internal arguments—including subjects of unaccusative predicates—are bare. The genitive case marker in (34a) is ga, which in modern Japanese has been reanalyzed as nominative. However, its genitive use in this period is clearly in evidence in (33a) and (34a) marking the possessor wa-ga ‘my.’
Intransitivity and the development of ergative alignment 515 Old Japanese (34) a. 我背子之求流乳母尒 [wa-ga seko-ga motomu]-ru omo-ni I-gen lord-gen ask-adn nurse-dat ‘as the wet nurse that my lord asks for’ b. 久木生留清河原尒 [pisaki opu]-ru kiyo-ki kapara-ni catalpa grow-adn clear-adn river.bank-on ‘on the banks of the clear river where catalpas grow’
(Manyoshu 2926)
(Manyoshu 925)
In addition to the assignment of inherent genitive case to external argument subjects, Whitman and Yanagida further propose that the nominalized v lacks the ability to structurally license the object. Yanagida (2006), Yanagida and Whitman (2009), Yanagida (2012), and Whitman and Yanagida (2012) have shown that there was an asymmetry between null case-marked objects in nominalized clauses and those taking the particle wo, which is the historical precursor of the modern accusative particle o: wo-marked (accusative) objects are interpreted as specific and are required to precede a genitive subject, while bare objects remain in their base positions immediately preceding the verb. Old Japanese (35) a. 佐欲比賣能故何比列布利斯夜麻 (Manyoshu 868) [vP Sayopimye=no kwo=ga [VP pire puri]]-si yama Sayohime=gen child=gen scarf wave-pst.adn hill ‘the hill where the girl Sayohime waved her scarf ’ b. 蜻野叫人之懸者 (Manyoshu 1405) [AspP Akidu nwo=wo [vP pito=no [VP tObj kakure-ba]]] Akizu field=acc man=gen speak.of-when ‘When a man speaks of the moorland of Akizu …’ Yanagida and Whitman (2009) and Whitman and Yanagida (2012) propose that the v in the nominalized clause assigns inherent genitive case to its specifier but is unable to value structural case on an internal argument. A nonspecific object remains in its base position immediately preceding the verb. They point out that these objects are generally N0-level categories, and they analyze them are incorporated to the verb. But when the object is specific, it raises from its base position to the specifier of an aspect projection dominating vP, where it values structural accusative case. As for bare internal argument subjects in unaccusatives, which are potentially phrasal, Whitman and Yanagida (2009) propose that T values nominative case on these DPs, given that vP is defective in unaccusatives. In this way, Whitman and Yanagida (2012) implement the proposal that non- accusative alignment emerges in an otherwise accusative language when the external
516 Edith Aldridge argument in a transitive clause is assigned inherent case, while the object is denied structural licensing by this same v head.
21.5.3 Austronesian Tagalog has been claimed by Payne (1982), De Guzman (1988), Aldridge (2004, 2012b), and others to be a language with ergative alignment. This can be seen in the contrast between (36a) and (36b), in which the object in a transitive clause takes the same ang case-marker as the subject in an intransitive clause. Example (36c) is analyzed as an antipassive, a semantically transitive but syntactically intransitive clause type (see also Polinsky, Chapter 13, this volume, on the characteristics of antipassives). Like the simple intransitive in (36b), the subject (external argument) receives ang marking in the antipassive, while the object appears with inherent genitive case. Tagalog (36) a. Bili ng babae ang isda. buy gen woman nom fish ‘The woman bought the fish.’ b. Dating ang babae. arrive nom woman ‘The woman arrived.’ c. Bili ang babae ng isda. buy nom woman gen fish ‘The woman bought a fish.’
21.5.3.1 Synchronic Approach Kaufman (2009a, Chapter 24, this volume) offers an explanation for the use of genitive case to mark DP arguments that are not absolutives by analyzing Tagalog clauses as nominalizations embedded in a copula construction. Kaufman proposes that Tagalog lacks a v functional category. Consequently, lexical roots merge with n and project a nominal predicate. The external argument of a transitive clause is treated as a possessor merged in [Spec, n] and assigned genitive case. The nominalized predicate is a relative clause with a null operator in the specifier of PredP. T is treated as a null copula, which mediates a predication relation between PredP in its specifier and its DP complement. Tagalog (37) a. Bili ng babae ang isda. buy gen woman nom fish ‘The woman bought the fish.’ (lit. ‘The fish is what the woman bought.’)
Intransitivity and the development of ergative alignment 517 b.
TP PredP
OPi
T’ Pred’
Pred
T DP
binili ng babae bought gen woman
DPi ang isda nom fish
This analysis offers an account for the constraint in Tagalog and many other Austronesian languages (as well as syntactically ergative languages generally) that all DPs other than the absolutive are ineligible to undergo A’-movement, since genitive DPs are all contained within the relative clause DP, an island to extraction. Example (38a) shows that a relative clause can be formed on the absolutive object in a transitive clause. However, the ergative subject cannot be extracted in this way, as shown in (38b). Tagalog (38) a.
isda-ng2 bili ng babae fish-lk buy gen woman ‘fish that the woman bought’
b. *babae-ng bili ang isda woman-lk buy nom fish ‘woman who bought the fish’ However, this biclausal analysis suffers from a number of shortcomings. Aside from the stipulation that all finite clauses are copula constructions, the structure in (37b) makes incorrect predictions about word order in the language, as pointed out by Aldridge (2009). For example, this structure does not allow the absolutive DP to intervene between predicate-internal constituents. Example (39), however, shows an absolutive appearing between the verb and a genitive object (39a) and a goal PP (39b). Since these constituents are arguments of the verb, I assume they would be base generated in the PredP on Kaufman’s analysis. Consequently, their dislocation to clause-final position should invoke an island violation along the lines of (38b), contrary to fact.
2 The linker appearing between the head NP and the modifying clause is spelled the same way as the genitive case marker. But the two differ in pronunciation, the linker pronounced as the velar nasal, and are not the same morpheme.
518 Edith Aldridge Tagalog (Aldridge 2009: 53–54) (39) a.
[PredP Bili tDP ] ang babae ng bahay. buy nom woman gen woman ‘The woman bought a house.’
b. [PredP I-bi-bigay=ko tPP ] ang bulaklak kay Maria. appl-red-give=1sg. gen nom flower to Maria ‘I will give the flowers to Maria.’ In the next subsection, I propose that PP extraction from a predicate nominal is actually possible in certain Austronesian languages. However, there is a crucial structural difference between my approach and Kaufman (2009a). In Kaufman’s approach, extraction takes place from a relative clause in a specifier position, which should result in a violation of Huang’s (1982) Condition on Extraction Domain. In the structure I propose, the predicate nominal is the complement of a Pred head. I draw a parallel between this type of extraction and the lack of opacity effects in subextraction from indefinite objects in English.
21.5.3.2 Diachronic Approach Starosta et al. (1982/2009) take a diachronic approach to a possible connection between nominalizations and the type of ergative morphosyntax observed in most Philippine and Formosan3 languages. Specifically, they propose that many affixes appearing on transitive verbs in the modern languages were nominalizers in Proto-Austronesian (PAn). The nominalizer exemplified in (40) is *-an,4 which attaches to a verb and projects a relative clause predicated of the matrix subject in a copula construction. (40)
S NP N ‘climb’-an
NP ‘John’ (GEN)
NP ‘mountain’ (NOM)
‘The place where John climbed is the mountain.’ (Starosta et al. 1982/2009: 313) In time, biclausal copula constructions like (40) were reanalyzed as transitive verbal clauses like (41). Nominalizers like *-an were consequently became verbal affixes. A reflex of *-an continues to function solely as a nominalizer in Puyuma and Rukai but 3 The term “Formosan languages” refers collectively to the Austronesian languages spoken in Taiwan, but does not indicate a subgroup. 4 Starosta et al. reconstruct the morpheme as *-ana, but *-an is the commonly accepted reconstruction in more recent work by Austronesian historical linguists.
Intransitivity and the development of ergative alignment 519 is employed as a locative applicative attaching to verbs in many other Formosan and Philippine languages.5 (41)
S V ‘climb’-an
NP ‘John’ (GEN)
NP ‘mountain’ (NOM)
‘John climbed the mountain.’
(Starosta et al. 1982/2009: 314)
An advantage over Kaufman (2009a) is that the resulting structure is monoclausal and does not require that all root clauses be analyzed as copula constructions. However, the detailed steps in the reanalysis are not spelled out. Nor is a clear motivation for the change identified. In what follows, I develop the insight of Starosta et al. (1982/2009), Kaufman (2009a), and others that there is a connection between nominalization and ergative clauses in Philippine and most Formosan languages and propose an explicit account of how a nominalized relative clause in a copula construction could have been reanalyzed as a finite ergative clause. I also adopt Ross’ (2009) proposal that the reanalysis of nominalizations to root clauses took place in what Ross terms the Nuclear Austronesian subgroup and should not be attributed to Proto- Austronesian or Pre-Austronesian. This subgroup encompasses all Austronesian languages except for Rukai, Puyuma, and Tsou, the last three still spoken in Taiwan. (42)
(based on Ross 2009)
Austronesian Rukai
Tsou
Puyuma
Nuclear An (NAn)
Much of Ross’ evidence for this grouping comes from the fact that affixes forming finite ergative verbs in NAn languages bear resemblance to morphemes only employed in nominalizations in extra-NAn languages. Since the reanalysis of nominalizations as finite verbs has not taken place in the extra-NAn languages, this change can be regarded as an innovation defining the NAn subgroup. As an example, the Puyuma finite transitive verb in (43a) takes the suffix -aw, an affix which never appears on a verb in a relative clause. Verbs in relative clauses must be nominalized, as in (43b), where the verb takes perfective aspect marker and the nominalizer -an. Puyuma (43) a. tu=trakaw-aw na 3.gen=steal-tr def.nom ‘Isaw stole the money.’ 5
paisu kan isaw money sg.obl Isaw
(Teng 2008: 147)
See Gildea (1998) for a similar account of the reanalysis of nominalizations as ergative clauses in Cariban languages.
520 Edith Aldridge b. ala amuna sadru [[tu=trekelr-an] maybe because many 3=drink-nmlz ‘Maybe because the milk he drank is a lot.’
na asi] def.nom milk (Teng 2008: 105)
These morphemes surface only on nominalized verbs in Puyuma but appear on finite verbs in NAn languages like Tagalog. Note that -an has been reanalyzed as a locative applicative in Tagalog. Tagalog (44) Bilh-an ng babae ng buy-appl gen woman gen ‘The woman bought a/the fish at my store.’
isda fish
ang nom
tindahan=ko. store=1.sg.gen
For the reanalysis of nominalizations as verbal clauses, I assume with Starosta et al. (1982/2009) that the input structure was a nominal predication. In Aldridge (2004, 2016), I presented evidence for the following structural analysis of nominal predication in Tagalog. The predicate-initial word order is derived by moving the predicate NP to a position above the subject. The trace position of the NP is indicated with angled brackets.6 Tagalog (45) a. Importante-ng miyembro ng Sizzlers si Gilbert. • important-lk member gen Sizzlers nom Gilbert ‘Gilbert is an important member of the Sizzlers.’ b.
CP
C
TP T
PredP NP
Pred’ Gilbert
Pred’ Pred
importante miyembro 6
N’ Sizzlers
See Massam (2000, 2001, 2003), Rackowski and Travis (2000), Pearson (2001), Aldridge (2004), Cole and Hermon (2008), and others for derivations of Austronesian predicate-initial word order by means of phrasal fronting of all or part of the predicate.
Intransitivity and the development of ergative alignment 521 This analysis can also accommodate cases in which part of the nominal predicate has been stranded before fronting. In the following example, the complement PP vacates the predicate to be fronted and is stranded inside the PredP. Cole and Hermon (2008) propose a very similar analysis of stranding accompanying predicate-fronting in Toba Batak. Below, I suggest an important role for stranding in the reanalysis of nominal predications as verbal clauses. Tagalog (46) a. Importante-ng miyembro si Gilbert ng Sizzlers. Important-lk member nom Gilbert gen Sizzlers ‘Gilbert is an important member of the Sizzlers.’ b.
CP
C
TP T
PredP NP
Pred’ Gilbert
Pred’ Pred
XP sizzlers
importante miyembro
N’ tsizzlers
I propose that the input to the reanalysis in question was a nominal predication in which the predicate was a nominalized relative clause, specifically the type Krause (2001) terms “reduced relatives with genitive subjects.” The structure contains a position for aspect but not tense. As Chen (2008: 96) shows that the nominalizer -ane (cognate with the nominalizer -an in Puyuma) attaches closer to the root than the perfective marker -nga, I place AspP outside nP. The external argument and nominalizing affix are merged in the nP layer. Internal arguments are merged together with the root.
522 Edith Aldridge Budai Rukai (Chen 2008: 84)7 (47) a. Ta-badh-ane ki tina-ini ki lalake-ini nonfut-give-nmlz gen mother-3sg.gen obl child-3sg.gen ka laimai. nom clothes ‘The clothes are what the mother gave her child.’ b.
CP
C
TP T
PredP DP
Pred’ laimai
Pred’ Pred
D
AspP
ta-badh-ane
nP
tina-ini
n’ √P
Interestingly, stranding is also possible from within this type of predicate nominal. In (48), the dative argument has been moved out of the relative clause and appears in a position following the matrix subject. I assume that the PP moves from complement position in the root phrase and exits the DP via the edge this phase (and possibly also the nP) in order to avoid violating the Phase Impenetrability Condition of Chomsky (2001 and subsequent works). Budai Rukai (Chen 2008:82) (48) a. Ta-badh-ane ki tina-ini ka nonfut-give-nmlz gen mother-3sg.gen nom
laimai clothes
ki lalake-ini. obl child-3sg.gen ‘The clothes are what the mother gave her child.’ 7 Chen (2008) does not analyze this construction as a nominalization but rather as “object voice.” However, he admits that affixes like -ane are clearly nominalizers in the language. He also attributes stative semantics to the construction. So it is difficult for me to understand the rationale for not analyzing this construction as a nominalization.
Intransitivity and the development of ergative alignment 523 CP
b. C
TP T
PredP DP
Pred’ laimai
Pred’ Pred
XP PP
D' D
AspP
ta-badh-ane
nP n'
tina-ini n
It may seem surprising here that the predicate nominal is not an island to extraction. However, two points are worthy of note here. First, extraction takes place from complement position. This is a crucial distinction between the current analysis and Kaufman’s (2009a) proposal in (37b), in which the predicate is in a specifier position. Second, extraction from DP is in fact possible if certain conditions are met. For example, indefinite DPs in object position in English allow subextraction, but definite DPs do not. Regarding the copula constructions in Rukai, since the DP in question is predicational and not referential, it cannot be definite.8 (49) a. Who did you meet [some friends of ]? b. *Who did you meet [those friends of ]? Returning to the diachronic reanalysis of copula constructions as verbal clauses, I suggest here that the possibility of remnant predicate fronting may have played a key role. Specifically, when material from the relative clause is stranded to the right of the matrix subject, the utterance comes to strongly resemble a monoclausal verbal construction in its word order, as shown in (50b). Example (50a) repeats the stranding example in (48a). 8 Note further that reduced relative clauses in Tagalog existential constructions allow subextraction. See Aldridge (2012a) for discussion.
√P
524 Edith Aldridge Budai Rukai (50) a. Ta-badh-ane ki tina-ini ka laimai nonfut-give-nmlz gen mother-3sg.gen nom clothes ki lalake-ini. obl child-3sg.gen ‘The clothes are what the mother gave her child.’
(Chen 2008: 82)
b. Wa-bai ku laimai ka kineple ki cegau. nfut-give acc clothes nom Kineple obl Cegau ‘Kineple gave clothes to Cegau.’
(Chen 2008: 40)
The reanalysis is straightforward. The mechanisms involved are relabeling and pruning, as proposed by Whitman (2000). Because of the resemblance to monoclausal constructions, the nominalized verb is interpreted as the main verb. As a result, the nP in the relative clause is relabeled as vP. (51)
CP C
TP T
PredP DP
Pred’ laimai
Pred’ Pred
D
AspP
ta-badh-ane
nP > vP
tina-ini
n’ > v’ √P
Since the construction as a whole has been parsed as verbal and monoclausal, there is no longer any evidence for the DP and PredP layers, so these are pruned away, with the result that the AspP dominating vP will be directly selected by T. This clause will exhibit an ergative case-marking pattern, because the genitive-assigning functional head in the erstwhile relative clause is now the matrix v. As a former nominal category, this v also lacks the ability to license structural accusative case on the object. Consequently, the object will enter into an Agree relation with T and value nominative case in order to be
Intransitivity and the development of ergative alignment 525 licensed. In this way, the ergative v emerged in Proto-Nuclear Austronesian with precisely the properties expected in the analysis in (4). (52)
CP C
TP T
AspP Asp
vP v’
DP[GEN] v
√P √
DP[NOM]
Identifying a nominalized relative clause structure as the historical antecedent of the ergative clause type has the added advantage of providing an explanation for the extraction restriction observed above. Recall that the internal argument in a transitive clause can be the head of a relative clause, as in (53a), but not the external argument, as shown in (53b). A subject can only be extracted from an intransitive clause. Example (53c) is an antipassive, in which intransitive morphology appears on the verb and the object receives inherent genitive case rather than structural nominative. On the other hand, extraction of non-DPs is not subject to this constraint, as shown in (53d). Tagalog (53)
a.
isda-ng bili ng fish-lk buy gen ‘fish that the woman bought’
babae woman
b. *babae-ng bili ang isda woman-lk buy nom fish ‘woman who bought the fish’ c. babae-ng bili ng isda woman-lk buy gen fish ‘woman who bought a/the fish’ d. Saan bili ng babae ang isda. where buy gen woman nom fish ‘Where did the woman buy the fish?’
526 Edith Aldridge I propose that this locality restriction is a direct consequence of the properties of relativization in Austronesian languages. First consider Rukai, which is an accusative language9 and has not undergone the reanalysis of nominalizations as verbal clauses. In Rukai, relative clauses formed on subject position are finite CPs. This is evidenced by the fact that they contain the same tense markers as finite clauses. Tanan Rukai10 (54) a. luða ay-kɨla ku tina=li tomorrow fut-come nom mother=1.sg.gen ‘My mom will come tomorrow.’ b. [kuaDa ay-suwasuwaw] ka muka-baru-barua dem fut-clean top girl ‘The one who will clean is the girl.’ On the other hand, object relatives are nominalized with the suffix -anɨ, transcribed as -ane for the Budai dialect in the preceding discussion. But they do not carry tense marking. The prefix a-expresses imperfective aspect. Tanan Rukai (55) w-aga=su sa aga sa [a-kanɨ-anɨ=ta ki maum] past-cook=2.sg indef food indef imprv-eat-nmlz=1.pl.inc p night ‘Did you cook dinner (the food that we will eat tonight)?’ The analysis I propose here accounts for the restrictions in both Tagalog and Rukai. The strict locality between DPs observed in the Tagalog examples in (53) suggests that the feature driving movement in the relative clause is a feature which specifically attracts DPs. In Aldridge (2004), I proposed that Austronesian languages do not have A-movement to [Spec, TP]. Rather, the EPP feature that appears on T in other languages is carried instead by C in Austronesian languages. This accounts for the extraction restriction, since movement to [Spec, CP] in Austronesian languages is subject to the same category sensitive locality observed in subject movement to [Spec, TP] in other languages.
9 Interestingly, Puyuma and Tsou are ergative languages. Aldridge (2015, 2016) proposes a revision to Ross’ (2009) subgrouping hypothesis by positing an Ergative Austronesian subgroup as sister to Rukai and parent to Puyuma, Tsou, and Nuclear Austronesian. Rukai retains the accusative alignment of Proto-Austronesian. Ergativity was first innovated in irrealis clauses in Proto-Ergative Austronesian. This alignment is retained in irrealis clauses in the Nuclear-Austronesian languages. These languages have further reanalyzed nominalizations as ergative clauses in the realis paradigm. 10 Data for the Tanan dialect of Rukai were collected during fieldwork in Taiwan in 2013 and 2014 with support from the Chiang Ching-Kuo Foundation.
Intransitivity and the development of ergative alignment 527 In this chapter, I restate this analysis within Chomsky’s (2005, 2008) proposal of C-T inheritance. Specifically, the lack of A-movement in languages like Tagalog and Rukai is accounted for by proposing that C-T inheritance does not take place, so all movement targets [Spec, CP]. Another consequence of the lack of C-T inheritance is that the unvalued ɸ-feature which licenses the subject remains on C. Since this feature will probe for a set of valued ɸ-features, which are only visible on DPs, category sensitive locality between DPs will be observed in relative clause formation.11 In a finite CP, the subject enters into an Agree relation with the ɸ-probe on the C/T complex and value its case feature. Since the languages under discussion are verb-initial, I assume that the subject does not generally raise to [Spec, CP]. However, in cases of A’-movement, the subject does move. I assume that this movement takes place directly from the subject’s base position in vP, as per Ouali’s (2006) adaptation of C-T inheritance. Tense features are also retained on the C head. This accounts for the fact that subject relatives in Rukai can be formed on finite clauses and need not be nominalized. Extraction of the external argument in antipassives in Tagalog takes place in exactly the same way.
(56)
DP D
CP
DP[CASE:NOM] C’ vP
C
v’
v
VP
Relativization is not possible on object position in a full CP. If the object were to move to the edge of vP and consequently become visible to the ɸ-probe on the C/T complex, the derivation would not result in an object relative clause. There are two possible reasons for this. If the object’s ɸ-features entered into an Agree relation with C/T, the object would move to [Spec, CP]. However, this would leave the subject’s case feature unvalued, and the derivation would crash. Alternatively, if we assume the Activity Condition of Chomsky (2001), then the object would not be visible to the probe on C/T, since its case feature has already been valued. The derivation would not crash, but the subject would still be the DP to move to [Spec, CP].
See Carstens (2005) and Henderson (2006, 2011) for other proposals that ɸ-features can drive relative clause formation. 11
528 Edith Aldridge (57)
CP C
TP vP
T
DP[CASE:ACC]
v’ v’
DP[CASE: ] v
√P √
In order to extract an object over the subject, the competition for ɸ-feature valuing must be eliminated. This is accomplished in a nominalization, since the subject is given inherent genitive case, and there is also no C/T layer. Rather, it is the nominalizer itself which creates the gap to form the relative clause. This probe undergoes Agree with the object and raises it to the edge of nP.12 This accounts for the fact that object extraction in Rukai requires a nominalization. Nuclear Austronesian languages like Tagalog have inherited this relativizing n as the transitive (ergative) v, accounting for why internal arguments are extracted in transitive clauses.13 DP
(58)
D
AspP Asp
nP DPOBJ
n’ n’
DP[GEN] n[
√P
:]
√
12 I assume that the moving constituent in a relative clause is a null operator, which will not be spelled out with phonetic content, and that movement to the highest strong phase edge suffices to create a structure which can be interpreted as a lambda abstraction at the interface. But minimal revision could also accommodate a head raising approach by further moving the DP to a peripheral position. 13 This proposal entails that transitive v always raises an object to its outer specifier, which is a welcome consequence of this analysis. It is generally assumed that absolutive objects in Tagalog raise to the vP phase edge when the verb bears transitive morphology (Rackowski 2002; Aldridge 2004; Rackowski and Richards 2005) but is spelled out in its base position if it does not undergo further movement to [Spec, CP]. Raising of the object accounts not only for its ability to undergo A’-movement but also for its presuppositional, wide scope interpretation.
Intransitivity and the development of ergative alignment 529 Since the probe driving the movement can only agree with a DP, non-DPs can dislocate freely, presumably motivated by focus. It should now be clear how nominalization enables object relativization by eliminating the competition to value ɸ-features on the relativizing functional head. Interestingly, this DP locality restriction on A’-movement is widely observed in syntactically ergative languages, particularly those in which morphological ergativity arguably resulted from the reanalysis of a clausal nominalization. Recall from section 21.5.1 that Inuit languages exhibit a syncretism between ergative and genitive case, and ergative clauses have been argued to be derivationally related to nominalizations. Mayan is another language family which has both the ergative/genitive case syncretism and the DP extraction restriction. The preceding proposal speaks to this correlation by tracing the source of ergativity to a nominalized relative clause in which the inherent case-marking on the transitive subject is a strategy for allowing the object to enter into an Agree relation with a higher functional head without intervention by the subject. See Aissen, Otsuka, and Erlewine et al. (Chapters 30, 40, and 16, respectively, this volume) for other approaches to the extraction restriction in ergative languages.
21.6 Conclusion Ergative alignment can be characterized as involving one structural case, which appears on intransitive subjects and transitive objects, and one inherent case, which is assigned to transitive subjects. In this chapter, I have summarized various accounts of the emergence of ergative alignment and shown that they are compatible with an analysis which traces the source of ergativity to an intransitive v which assigns inherent case to its specifier and does not structurally license an internal argument. I have further shown that the strict locality exhibited between DPs in syntactically ergative languages results when the diachronic source of ergativity is a nominalizing v which facilitates movement of an internal argument over an external argument in relative clause formation.
Abbreviations 1, first person; 2, second person; 3, third person; abl, ablative; abs, absolutive; acc, accusative; adn, adnominal; appl, applicative; caus, causative; conc, conclusive; cop, copula; dat, dative; def, definite; dem, demonstrative; erg, ergative; f, feminine; fut, future; gen, genitive; indef, indefinite; inc, inclusive; inh, inherent; ins, instrumental; intr, intransitive; ipfv, imperfective; lk, linker; m, masculine; N, noun; nmlz, nominalizer; nom, nominative; nfut, nonfuture; obl, oblique; opt, optative; P, preposition; part, partitive; pass, passive; pfv, perfective; pl, plural; prs, present; pst, past; ptcl, particle; ptcp, participle; pv, preverb; red, reduplication; rel, relative case; sg, singular; top, topic; tr, transitive; uns, unspecified tense/ aspect/mood; V, verb.
Chapter 22
Devel opments i nto an d ou t of e rg at i v i t y: Ind o-A ryan diac h rony Miriam Butt and Ashwini Deo
22.1 Introduction This contribution provides an overview of what is known about the diachrony of ergativity in Indo-Aryan. The Indo-Aryan language family provides an ideal situation for a study of case and ergativity. Old Indo-Aryan (Vedic, Sanskrit) did not have an ergative case, but many of the New Indo-Aryan languages do. Some of the modern languages further show evidence of having lost ergative alignment. The written record for Indo-Aryan goes back to about 1900 bc and is the longest documented continuous historical record available to us in historical linguistics. Table 22.1 provides a time line as relevant for the discussions in this chapter.1 The standard modern languages tend to be spoken by millions of people and have a rich literary record. They are thus in principle easily accessible for linguistic study and show an interesting variety of Table 22.1 Indo-Aryan chronology
1
timeline
stage
language
1900 BC–1100 BC
I
Early Old Indo-Aryan
1000 BC –200 BC
I
Later Old Indo-Aryan
300 BC –700 AD
II
Middle Indo-Aryan
1100 AD–present
III
New Indo-Aryan
Approximate dates are based on Alsdorf (1936); Witzel (1999); Jamison and Witzel (2002).
Indo-Aryan diachrony 531 patterns in the domain of ergativity. Although much has been written about the patterns found in Indo-Aryan, a complete understanding of the diachrony of ergativity in Indo-Aryan remains to be attained. The issues surrounding ergativity are complex. In addition to the mere absence or presence of an ergative case, ergativity has been discussed in terms of alignment patterns (cf. Aldridge (Chapter 21, this volume)), which tend to prominently involve not only case marking but also agreement and other linguistic phenomena that are sensitive to alignment issues. It is still a matter of current debate whether Old Indo-Aryan (OIA) was already ergatively aligned, indeed there are proponents of an ergative analysis of Proto Indo-European (cf. Bauer 2000). In addition to the inherent complexity attached with an understanding of ergativity, our understanding of the diachronic phenomena has also partly been hampered by too ready an acceptance of an overly simplistic scenario in some of the mainstream literature (cf. section 22.2.1). Another complicating factor is the comparative lack of historical evidence for the crucial time in which, following the erosion of the older case system, new case markers began to be innovated, namely the early New Indo-Aryan (NIA) period from about 1100 ad–1300 ad. A final obstacle to a definitive understanding is the comparative dearth of modern work on NIA languages. In this chapter, we first lay out some general background on the diachrony of ergativity in Indo-Aryan (section 22.2). This discussion functions as a backdrop to the more detailed presentation in the following sections on the diachronic trajectory as far as it is known (section 22.3). We include a section thematizing further perspectives on ergativity, including proposals regarding markedness, the role of clausal and lexical semantics, language contact, and competition among case markers (section 22.4.4). The chapter ends with a discussion and an outlook onto where we believe future efforts at understanding ergativity in the history of Indo-Aryan should be concentrated (section 22.5).
22.2 General Background Two central linguistic sources have been implicated cross-linguistically in the rise of ergativity: passive and possessive constructions (Benveniste 1952; Anderson 1977; Plank 1979; Garrett 1990; Dixon 1994; Harris and Campbell 1995; Bynon 2005; Haig 2008). We discuss both here because both have been proposed for an understanding of the Indo- Aryan situation.
22.2.1 Passive-to-Ergative In the passive-to-ergative reanalysis scenario, ergative alignment is taken to arise when a passive participial construction that expresses the resultative or perfect aspect is reanalyzed as an active, ergative clause with perfective aspectual reference (e.g. Dixon 1994;
532 Miriam Butt and Ashwini Deo Harris and Campbell 1995). As part of this reanalysis, the former oblique agent adjunct of the participle (generally an instrumental) is reanalyzed as an ergative subject. The basic idea with respect to a reanalysis of former passives is illustrated in (1) with the constructed example from Sanskrit (based on Garrett 1990, 263). (1) a. ahi-r indr-eṇa ha-ta-ḥ serpent-nom.m.sg Indra- inst.sg kill-ptcpl-nom.m.sg ‘The serpent has been killed by Indra.’ Actually: The serpent is one killed by Indra.’
Sanskrit
b. Reanalyzed as: serpent-nom.m.sg Indra-erg.sg kill-perf-m.sg ‘Indra has killed the serpent.’ As illustrated in (1), the participial implicated in the development of an ergative is the ta marked adjectival passive participial form. It assigns nominative case to the patient argument of the verb while the agent appears with instrumental case marking. The participle, which has resultative semantics, allows reference to states that hold of objects as a result of the occurrence of some event. With transitive unaccusative verbs as in (2), such a participle predicates a result state of the sole argument while with transitive change-of-state verbs, it predicates a result state of the patient argument. It is this resultative semantics that generates the ergative-like configuration in which objects of transitive verbs and subjects of intransitive verbs come to be marked identically. Over time, this passive-like stative construction comes to be reanalyzed as active and the former instrumental is reinterpreted as an ergative. (2) ahaṃ sarayūtīra-m ā-ga-ta-ḥ I.nom Sarayu.bank-acc pv-go-ptcpl-nom.m.sg ‘I came to the bank of the Sarayu.’
(Rāmāyaṇa 2.58.12) (from Bynon 2005, 11)
There are many indications that the ta participle already had an active interpretation in Epic Sanskrit. In (3), we provide an example taken from Bynon (2005, 13), who shows that an active analysis not only does justice to the narrative structure of the text, but also makes syntactic sense given that the instrumental agent is controlling the subject of the gerund clause. The example is taken from the Rāmāyana, a text dated to ∼200 BC, and written in what is known as Epic Sanskrit. For further examples, see Bynon (2005). (3) śabda-m ālakṣ-ya mayā … visṛṣ-ṭo nārāca-ḥ sound-m.acc aim-ger I.inst … release-ptcpl.nom.m.sg arrow-nom.m.sg ‘Aiming at the sound I released an arrow.’ (Rāmāyaṇa 2.58.15) Along with evidence for an active interpretation of the participle, there is also good reason to believe that the ta participle was integrated into the finite verbal system of the
Indo-Aryan diachrony 533 languages—being used robustly to describe past, culminated events. An example from the Mahābhārata, another Epic Sanskrit text, shows that the participial form may be used as the main verbal predicate in the clause with past-referring frame adverbials (Deo 2012, 10). (4)
purā devayug-e ca eva dṛṣ-ṭaṃ sarvaṃ formerly god.age-loc.sg and ptcl see-perf.n.sg everything mayā vibho I-inst.sg lord-voc.sg ‘Lord, formerly, in the age of the Deva (Gods), I saw everything.’ (Mahābhārata 3.92.6a)
Bynon (2005, 11) provides further evidence from the Rāmāyana in support of this use of the participle. Here, a story is being told and the same action is alternatively described using the aorist or the ta participle. The aorist was used to express resultative perfect and recent past meanings (Kiparsky 1998b; Condoravdi and Deo 2014). A reasonable conclusion to draw from the existence of alternations such as (5) is that the ta participle could already be used to describe events in the recent past in Sanskrit. For more examples, again see Bynon (2005). (5)
a. aśrauṣam … ghoṣam hear.1.sg.aor … noise.acc ‘I heard a noise.’ b. ṣru-to hear-ptcpl.nom.m ‘I heard a sound.’
mayā I.inst.sg
(Rāmāyana 2.57.16) śabdo sound.nom.m.sg
(Rāmāyana 2.58.13)
Bynon considers register as a possible conditioning factor in the use of the ta participle vs. the aorist form. Be that as it may, the diachronic record clearly shows that the ta participle replaced the older inflecting tense forms over time. Bynon cites Bloch (1906, 48, 58) as counting about 150 main clause past participle predicates vs. 1,033 finite verbs at the time of the Mahābhārata (Epic Sanskrit). In a text from a later period this proportion is reversed: 790 finite verbs vs. 1,750 morphologically non-finite expressions. The reanalysis of the ta participial marking into active inflectional past/perfective morphology is undisputed. The modern NIA languages tend to show split-ergativity in that the ergative case predominantly appears in conjunction with this past/perfective morphology (see discussion in section 22.3.3). This type of split-ergativity is part of a wider cross-linguistic pattern (Trask 1979), though one that is still not wholly understood. From the perspective of the passive-to-ergative hypothesis, the historical explanation is straightforward: the ergative is naturally only associated with the past/perfective morphology because it is a direct continuation of the old agentive instrumental adjunct. The adjunct gets reanalyzed as the subject of the clause with the original instrumental case being reanalyzed as ergative. This hypothesis is supported by the fact that cross-linguistically ergatives are often form-identical with instrumentals. However, the
534 Miriam Butt and Ashwini Deo hypothesis is problematic within the Indo-Aryan context because it also tends to assume an instrumental-to-ergative change. The instrumental-to-ergative hypothesis was first proposed by Trumpp (1872, 113) for Indo-Aryan. Despite being debunked immediately and roundly by Beames (1872–79), it has continued to persist, being presented as textbook knowledge in works such as Dixon (1994) and Harris and Campbell (1995). What Beames (1872–79) objects to is Trumpp’s claim that the OIA instrumental case marking was the direct ancestor of the modern ergative case forms. This scenario is improbable because of the erosion of the OIA case system as attested in Indo-Aryan diachrony. In fact, at least some of the Indo-Aryan languages went through a period in which there was no overt marking of ergative case but an ergative alignment of clauses (see section 22.3.2). The instrumental-to-ergative hypothesis historically has been closely associated with the passive-to-ergative hypothesis. The former involves the case marker per se, the latter notions of how clauses are structurally aligned (see Butt 2006b for background and discussion). It is therefore worth mentioning that while the ta participle is often referred to as “passive” in the literature, it must be distinguished from the actual passive found in Sanskrit. As illustrated in (6), this passive was formed with the morpheme -ya. It was an integral part of the language, but crucially appears to have absolutely nothing to do with the realization of either an ergative case or an ergative structural alignment (Klaiman 1978). (6) devadattena kaṭaḥ kri-ya-te/kri-ya-nte Devadatta.inst.sg mat.nom do-pass-3.sg/do-pass-3.pl ‘by Devadatta a mat/mats is/are made’ Sanskrit (adapted from Hock 1986, 16) Given the idea that the rise of ergativity is tied to a reanalysis of a passive in passive-to- ergative hypothesis, why did the Sanskrit -ya- passive not give rise to the ergative pattern? One answer to this question may be the frequency of the appearance of the instrumental agent. Like the modern NIA languages, Sanskrit was able to drop arguments freely. A comparison of the instrumental agents occurring with the deverbal adjectival -ta participle and the passive -ya- shows that for both Sanskrit (Gonda 1951, 22) and the later Pāli (Peterson 1998), the instrumental agents in passives were rarely expressed. With the deverbal participle -ta construction, on the other hand, the instrumental agent was almost always expressed overtly. The -ta participle was also part of the overall tense/aspect paradigm in Sanskrit (see earlier discussion in this section) and is the ancestor of inflectional tense/aspect morphology in NIA. The passive -ya- did not enter this system and has generally been lost in NIA. Implicating passive, that is the voice system, as a conditioning factor for ergativity is thus misleading, at least in the Indo-Aryan context. Rather, stative expressions that are being drawn into the tense/aspect system of a language appear to furnish the starting point of an ergative system. These stative expressions begin by being nominal/adjectival and therefore have a “demoted” agent. As they are reanalyzed as verbal, the status of the agent argument is concomitantly reanalyzed and it is syntactically realized as the subject of a verbal predication. Research with a strong South Asian focus or background has tended to acknowledge this complex diachronic scenario and has also sought to understand the origin of the
Indo-Aryan diachrony 535 modern ergative markers in nuanced detail. The ergative is homophonous with the instrumental in some NIA languages—the semantic connection here is obvious since agents and instrumentals share many semantic characteristics. However, we see that the ergative case clitic in languages like Urdu/Hindi and Punjabi is also cognate to the dative case clitic in other languages like Gujarati and Rajasthani (Tessitori 1913, 1914; Montaut 2003, 2006; Butt 2006a; Butt and Ahmed 2011), a development that does not have an obvious explanation. See section 22.4.4 for further discussion.
22.2.2 Possessives Bynon (2005) also sees difficulties for the passive-to-ergative hypothesis in the explanation of how the instrumental/agent came to occupy a clause initial position in the default word order (vs. the clause internal position illustrated in (1b)) and in the appearance of the agent/instrumental on intransitives. She proposes an alternative hypothesis which looks to developments in Indo-Iranian and which invokes a possessive construction as the ancestor of ergativity in Indo-Aryan. The diachronic scenario generally assumed for Iranian is that there was an accusative aligned ancestral language that shifted to ergative alignment. This then shifted back to an accusative language, namely modern Persian, which is morphologically and syntactically accusative (but see Haig (2008) for a more differentiated discussion and Haig (Chapter 20, this volume)). The shift in alignment in Iranian is associated with a possessive construction exemplified by a sentence involving manā kartam ‘done of (by) me’ (paraphrasable as ‘my done thing’). The participle involved is the same -ta participle as the one implicated in the OIA system, the common Proto Indo-European ancestor has been reconstructed as *to/no. The posited reanalysis is illustrated in (7) for Old Persian (based on Kent (1953)). (7)
a. ima tya manā kartam pasāva yaθā xšāyaθiya that which 1.s.g.gen do.ptcpl after when king abavam become.past.1.sg Old Persian ‘This (is) that (which) was done by me after I became king.’ Reanalyzed as: ‘I did that after I became king.’ b. avaθ=šām hamaranam kartam thus=3.pl.gen battle do. ptcpl ‘Thus by them battle was done.’ Reanalyzed as: ‘Thus they did battle.’
Old Persian
In the manā kartam construction, the agent is realized as a genitive possessor. As with the passive-to-ergative hypothesis, the overall idea is that the original genitive-marked possessor argument was reanalyzed as an ergative-marked agent, and that the stative/
536 Miriam Butt and Ashwini Deo passive participle was reanalyzed as an active verb that was part of the finite verbal paradigm of the language. Bynon (2005) argues that the ergative pattern in Indo-Aryan similarly arose out of a possessive construction via a mechanism of possessor raising. In particular, she argues that the source of the ergative construction in Indic and Iranian was anticausative, but not passive. Understanding the construction in terms of anticausativity has the advantage that the appearance of the ergative with intransitives can be explained readily. Furthermore, she argues that the construction functioned as an evidential, signaling that the event in question was inferred or reported rather than witnessed directly. This idea has the advantage of bringing in a semantic component to the understanding of the ergative case, potentially being able to explain modal and other effects associated with the ergative in languages such as Urdu/Hindi and Nepali (see section 22.4.4). Bynon also makes the point, already argued for by many others, that the OIA -ta participle was not passive, but already reflected an ergative alignment in Sanskrit, albeit in the absence of an ergative case. This argument has been made by Hock (1986) for Sanskrit, who talks about the -ta participle as a “P(atient)-oriented” construction. Similarly, Peterson (1998) argues that MIA showed ergative alignment, but in the absence of an ergative case (cf. Klaiman 1978; Bubenik 1996). The real innovation in NIA then, is the development of an overt ergative marker. Why and how this should have happened is not quite understood. Indeed, while Bynon’s argumentation is very persuasive and suggestive, she also does not address the questions of how or why the NIA languages innovated an ergative marker. She does suggest that the instrumental marking associated with the -ta participle in Sanskrit is an innovation that replaced an older Persian-type manā kartam pattern. The advantage of this account is that an immediate parallelism is drawn to the closely related Iranian branch. However, as far as we are aware, there is no evidence for such a change in the historical record.
22.3 The Diachronic Trajectory The previous section served to set the scene and discussed the two major linguistic sources that have been adduced as relevant to the development of ergativity in Indo- Aryan: passives and possessives. This section provides more information on the diachronic trajectory. After having thus delved into considerable detail, we return to a more high level discussion of other factors that have been implicated in understanding Indo- Aryan ergativity in the following sections.
22.3.1 Old Indo-Aryan The affix -ta (allomorph -na) described in the previous section, inherited from Indo- European, is attested at all stages of Old and Middle Indo-Aryan. It attaches directly
Indo-Aryan diachrony 537 to the root, creating an adjectival stem, and inflects for number and gender like any other adjectival form. As already mentioned, this affix has been reconstructed for Indo- European as *to/-no. In the oldest Vedic texts, the -ta based form of the verb serves to describe a result-state brought about by a preceding event when it is used predicatively in an adjectival passive construction. The -ta forms (bold-faced) in (8a) agree with the nominative patient while the agent remains unexpressed. In (8b), the agents and instruments are overtly expressed in the instrumental case. (8)
a. stīr-ṇáṃ te barhíḥ strew-perf.n.sg you.dat.sg Barhis.nom.n.sg su-tá indra sóma- ḥ kṛ-tā́ press-perf.m.sg Indra.voc.sg Soma-nom.m.sg do-perf.m.pl dhānā́ át- tave te hā́ri-bhyāṁ barley.nom.m.pl eat-inf you.gen.sg horse-dat.sg ‘The Barhis has been strewn for thee, O Indra; the Soma has been pressed (into an extract). The barley grains have been prepared for thy two bay-horses to eat.’ (Ṛgveda 3.35.7) b. nṛ-bhir dhū-táḥ su-tó áśna-iḥ man-inst.pl wash-perf.m.sg press-perf.m.sg stone-inst.pl áv- yo vā́ra-iḥ páripū-taḥ wool-gen.sg filter-inst.pl strain-perf.m.sg ‘It (the Soma) has been washed by men, pressed with the help of stones, strained with wool-filters.’ (Ṛgveda 8.2.2)
As shown in (9), the -ta form agrees with the sole (nominative) argument of intransitive verbs. This results in a difference in the marking of the subject arguments of transitive and intransitive verbs. In (8) the verb does not agree with the instrumental agentive arguments. In (9), in contrast, the verb śri-taḥ has a nominative subject soma and agrees with it in number and gender. (9)
div-i somo adhi heaven.loc.sg soma.nom.m.sg on Soma rests (is supported) in the heaven.
śri-taḥ rest-perf.m.sg
(Ṛgveda 10.85.1)
This periphrastic resultative -ta construction is the source of the ergative pattern observed in the perfective aspect in the later languages. In later stages of OIA, the construction was extended to marking the perfect aspect and it exhibited existential as well as universal perfect readings (Condoravdi and Deo 2014). By the time of Epic Sanskrit (late stage of OIA), the -ta construction became a frequently used device for marking past perfective reference. The agent argument in these cases is most frequently overt and marked with instrumental case. The examples in (10) are from (Deo 2012, 10). Past eventive reference is indicated by the presence of past referring frame adverbials like purā ‘formerly’ and tadā ‘then’.
538 Miriam Butt and Ashwini Deo (10) a. purā devayug-e ca eve dṛṣ-ṭaṃ sarvaṃ formerly god.age-loc.sg and ptcl see-perf.n.sg. everything mayā vibho I-inst.sg lord-voc.sg ‘Lord, formerly, in the age of the Deva (Gods), I saw everything.’ (Mbh. 3.92.6a) b. hṛ-tā gau-ḥ sā tadā steal-perf.f.sg cow-nom.f.sg that-nom.f.sg then t- ena prapāta- s tu na tark-itaḥ he-inst.3.sg fall-nom.m.sg ptcl neg consider-perf.m.sg ‘Then he stole that cow, but did not consider the fall (consequences).’ (Mbh. 1.93.27e)
22.3.2 Middle Indo-Aryan The main change between Epic Sanskrit and the later MIA stage of the language concerns the erosion and simplification of the rich tense-aspect system (Pischel 1900). Inflectional past referring forms such as the aorist, the inflectional perfect, and the imperfect disappeared from the language, leaving the -ta construction as the only past referring device.2 This loss of the inflectional system has often been cited as a reason for the increase in the frequency and scope of the participial construction, which in turn led to the unmarking of the stative nature of the construction, and resulted in an active, ergative clause in late MIA (Hock 1986; Bubenik 1998). The examples that follow from an archaic MIA Mahāraṣṭrī text Vasudevahiṃḍī (ca. 500 ad) shows this ergative alignment. The verb agrees with the nominative subject in (11a). In (11b) the verb agrees with the nominative marked object while the agentive argument (that running one) appears in the instrumental. (11)
a. pat-to ya seṇiyo reach-perf.m.sg and S.nom.m.sg ta-m paesa-m that-acc.sg place-acc.sg ‘And King Seniya reached that place.’
rāyā king.nom.m.sg
b. t-eṇa palāyamāṇ-eṇa purāṇakuv-o that-inst.sg running-inst.sg old.well-nom.m.sg taṇadabbhaparichinn-o diṭ-ṭho grass.covered-nom.m.sg notice-perf.m.sg ‘That running one noticed an old well covered with grass.’
(VH.KH. 17.1)
(VH.KH. 8.6)
2 Traditional grammarians do provide instances of the inflectional perfect and the aorist during this period, but they only occur as isolated, unanalyzed forms for a few verbs like āha-‘say-aor’ and akāshi -‘do-aor’.
Indo-Aryan diachrony 539 Evidence that the agentive argument is indeed the subject of the clause comes from control in gerundial clauses. In (12), the controller of the gerundial clause having bought the buffalo is the argument that appears in the instrumental/ergative case ṇ-eṇa ‘by him’ in the main clause. The morphology on the main verb ‘kill’ is a cognate of the ta participial. (12)
tamm-i ya sama-e... so mahiso that-loc.sg and time-loc.sg that buffalo.nom.sg kiṇe-uṇa mār-io buy-ger kill-perf.m.sg ‘And, at that time, having bought that buffalo, he killed it.’
ṇ-eṇa he-inst.sg (VH:KH 14:21)
Another critical change between OIA and MIA is the restructuring of the case system— particularly the loss of morphological contrast between nominative and accusative as well as between the genitive and the dative cases. The syncretized paradigm is given in Table 22.2. Table 22.2 Syncretized case paradigm in MIA (Masica 1991, 231) Singular
Plural
-u, a, aṁ -eṁ, iṁ, he, hi
-a, aĩ -e(h)ĩ, ehi, ahĩ
-hu, ahu, aho
-hũ, ahũ
Genitive/Dative
-ho, aho, ha, su, ssu
-na, hã
Locative
-i, hi, hiṁ
-hĩ
Nominative/Accusative Instrumental Ablative
With respect to ergative alignment, syncretism is also observed between the nominative and instrumental forms of the first and second person plural pronouns. This is shown in Table 22.3. Table 22.3 Syncretisms in the MIA Pronominal System aspect
person
Non-perf Perf
number singular
plural
1 1
haũ maĩ
amhaĩ/amhẽ amhaĩ/amhẽ
Non-perf
2
tuhũ
tumhaĩ
Perf
2
taĩ
tumhaĩ
Non-perf
3
so
te
Perf
3
tẽ, teṇẽ
tehĩ
540 Miriam Butt and Ashwini Deo Despite this syncretism, agreement is uniformly with the nominative argument—with the nominative object in constructions based on the -ta form and with the nominative subject elsewhere. The examples in (13) illustrate this pattern with the first person plural pronoun amhẽ. These are taken from the Paumacariu, a Jaina rendition of the epic Rāmāyana, ca. eighth century ad). (13a) contains the syncretized pronoun amhẽ which triggers agreement in the imperfective aspect while the same form fails to trigger agreement in (13b). In (13c) the second person plural syncretic form similarly fails to trigger verb agreement in the perfective. (13) a. amhẽ jāe-va vaṇavāsa-ho we.syncr go-impf.1.pl forest.dwelling-dat.sg We are going to our forest-exile b. ki-u amhẽ do do-perf.m.sg we.syncr what What crime have we done?
avarāh-o crime-nom.m.sg
c. tumhẽ jaṃ cint-iu you.syncr what.rel.m.sg think-perf.m.sg taṃ hū-a that.correl.m.sg happen-perf.m.sg That, which you thought (would happen), happened.
(PC 2.23.14.3)
(PC 1.2.13.9)
(PC 3.47.9.6)
The proto-ergative system that crystallizes in Late MIA thus has the following properties: (14) a. It exhibits an aspect-based split in ergative alignment. Changes in the semantics of the –ta form (which starts out as a marker of result-stative aspect) render it a marker of perfective aspect (Condoravdi and Deo 2014), which contrasts with imperfective forms in the language. b. It exhibits partial syncretism within the instrumental marking paradigm and syncretism in nominative/accusative case, making agreement the overt indicator of ergative alignment in many clauses.
22.3.3 New Indo-Aryan The NIA languages display a variety of patterns and a comprehensive discussion of these can be found in Deo and Sharma (2006) and Verbeke (2013a). We identify three major patterns: in one set of languages, the ergative is one of the several new case markers innovated (e.g. in Hindi and Nepali); in another set of languages the loss of ergative case leads to the loss of ergative agreement patterns (e.g. in Bengali and Oriya), while for a third set of languages, the original pattern persists despite changes in morphosyntactic marking (e.g. Marathi, Gujarati). We review each of these patterns in the following sections.
Indo-Aryan diachrony 541
22.3.3.1 Renewal of Overt Ergative Marking Data from Early Hindi suggests that the original instrumental marking observed on transitive subjects for the MIA ergative system is lost in NIA. Examples here are from the opus of Kabir, a Bhakti poet from the fifteenth century ad. Note that the transitive subject arguments in the perfective clauses in (15a) and (15c) carry no overt marking but agreement is uniformly with the feminine object argument (explicit or unpronounced) chādar ‘sheet’. (15)
a. jo chādar sura-nara-muni which sheet.nom.f.sg gods-men-sages.∅erg Gods, men, and sage, all wore this sheet
oḍh-i wrap-perf.f.sg
b. oḍhi-ke mail-i kin-i chadariyā wrap-ger dirty-f.sg do-perf.f.sg sheet.nom.f.sg Having worn it, they (invariably) made it dirty (defiled it). c. dās kabir jatan=se oḍh-i servant Kabir.∅erg care=with wrap-perf.f.sg (Your) servant Kabir wore it with great care d. jyon ki tyon dhar deen-i just as it was hold-ger give-perf.f.sg He has given it back (to you) just as it was.
chadariyā sheet.nom.f.sg
In the later language, the subject of a transitive perfective clause uniformly receives overt ergative marking across all persons and numbers via an invariant case-clitic ne. This is an innovation that is observed in some languages of the Western and Central subgroups of Indo-Aryan, including Modern Standard Hindi. The modern Hindi paradigm and agreement pattern (uniformly with nominative object) is shown in Table 22.4. An example is provided in (16).
Table 22.4 Modern Hindi pronominal paradigm and agreement pattern aspect
person
Non-perf Perf
number singular
plural
1 1
maĩ mai-ne
ham ham-ne
Non-perf
2
tum
āp
Perf
2
tum-ne
āp-ne
Non-perf
3
vah, yah
ve, ye
Perf
3
is-ne, us-ne
inho-ne, unho-ne
542 Miriam Butt and Ashwini Deo (16) a. sītā rām-ko sītā.f.nom Rām.m=acc ‘Sita hits Ram.’ b. mai=ne Pron.1.sg=erg ‘I saw a sparrow.’
pīṭ-tī hit-impf.f.sg
chiḍiyā bird.f.nom
c. ram=ne chiḍiyā Ram.m=erg bird.f.nom ‘Ram saw a sparrow.’
hai be.pres.3.sg
dekh-ī see-perf.f.sg dekh-ī see-perf.f.sg
The path of change in this case is thus one of strengthening of the morphosyntactically weakened ergative alignment pattern of Older Hindi via the innovation of explicit marking. Nepali, a language that shares this part of the change with Hindi, innovates further a pattern of agreement in which the verb uniformly agrees with the subject—regardless of whether it is marked ergative or nominative (Deo and Sharma 2006).
22.3.3.2 Loss of Ergative Marking and Agreement Pattern Bengali and Oriya, which are languages of the Eastern subgroup of Indo-Aryan, show a contrast to the strengthening of ergative marking observed in Hindi and Nepali. Bengali and Oriya exhibit a trajectory that has led to a complete loss of ergativity. As Chatterji (1926, 742–743) notes, Old Bengali texts show a minimal distinction between nominative and ergative (instrumental) case-marked arguments, but the agreement pattern is clearly ergative. The only difference between the nominative and ergative case marking is the presence of nasalization on the ergative case marker.3 The verb agrees with the nominative subject in non-perfective clauses and with the nominative object in perfective clauses. (17) a. kānhē pothī kānha.m.sg.nom book.f.sg.nom Kānha reads the book. b. kānhẽ pothī kānha.m.sg.erg book.f.sg.nom Kānha read the book.
paḍh-ai read-pres.m.sg paḍh-ilī read-perf.f.sg
Old Bengali
Old Bengali
The complete syncretism of nominative and ergative case marking over time went hand in hand with the loss of the ergative agreement pattern. Modern Bengali (and also Oriya) have lost all traces of ergative marking on the subject and the verb uniformly agrees with the subject in perfective (18a-b) and non-perfective (18c) clauses. The agreement marking has been innovated through auxiliary incorporation. 3
Chatterji constructs the minimal pair in (17) to illustrate the Old Bengali pattern but also gives several examples that are directly taken from the textual corpus.
Indo-Aryan diachrony 543 (18)
a. āmī boi I.sg.nom book.f.sg.nom ‘I saw a book.’
dekh-lām see-perf.1.sg
b. rām boi rām.sg.nom book.f.sg.sg ‘Rām saw a book.’
dekh-lo see-perf.3.sg
c. āmī boi I.sg.nom book.f.sg.nom ‘I am looking at a book.’
dekh-chi see-prog.1.sg
Bengali
Bengali
Bengali
The loss of ergative marking and the corresponding agreement pattern is also attested in Indo-Aryan texts from a different period and provenance. Jamison (2000) describes the intriguing pattern of ergative, perfective clauses in the Niya documents, a collection of texts found in the early twentieth century around Niya on the Southern Silk Route, datable to the third century ad. The perfective paradigm based on -ta in this linguistic system innovates a set of new endings (through incorporated auxiliaries) that obligatorily agree in person and number with the clausal subject. In many cases, the subject argument of a transitive perfective clause is nominative (Jamison calls this the absolutive) as in (19a). However, in some cases, as in (19b), the subject argument exhibits the older instrumental/ergative case. (19)
a. ahu sumiṃna I.nom.sg dream.nom.sg ‘I saw a dream.’
triṭh-emi see-perf.1.sg
(Niya 157, Jamison 2000, 71)
b. tade supi-yehi aǵasit-aṃti there.abl Supi-inst.pl carry-perf.3.pl ‘The Supis carried (them = mares) off from there.’ (Niya 212, Jamison 2000, 76). The data from the Niya documents that Jamison presents shows that it is a complex system with variation between nominative and ergative marking on the subject being sensitive to the animacy of both arguments (agentive and patientive) of the clause. Overt instrumental/ergative marking is “essentially obligatory” when both the agent and the patient/theme are human (Jamison 2000, 73) but may be optional elsewhere. In fact, there are attested examples where subjects in non-perfective clauses receive overt marking. (20) y-ena vṛchā who.inst.sg tree.nom.pl Whoever cuts down trees
chiṃn-ati cut-pres.3.sg
(Niya 482, Jamison 2000, 76)
544 Miriam Butt and Ashwini Deo This picture of the MIA ergative system points to an agentivity, animacy driven use of ergative case, i.e. a move toward semantic rather than purely structural determinants of ergative alignment. This aspect of the use of the ergative is elaborated upon in section 22.4.
22.3.3.3 Persistence of ergative agreement In Marathi, a language of the Southern subgroup of Indo-Aryan, nominative and ergative case-marked pronouns in the first and second person are entirely syncretized (an analogical extension of the syncretism in the first and second plural observed in MIA— see Table 22.5). Table 22.5 Marathi pronominal paradigm aspect
person
Non-perf Perf
number singular
plural
1
mī
āmhī
1
mī
āmhī
Non-perf
2
tū
tumhī
Perf
2
tū
tumhī
Non-perf
3
to/tī/te
te
Perf
3
tyā-ne, ti-ne
tyā-nī
The agreement pattern, however, persists—the verb fails to agree with the subject of a transitive perfective clause despite lack of overt morphosyntactic evidence for an ergative pattern. In (21a), a non-perfective clause, the verb agrees in number and gender with the subject (assuming the speaker in the utterance context is male) while in (21b), a perfective clause, the verb agrees with the object. (21) a. mī ek chimṇī I.sg.nom one sparrow.f.sg.nom ‘I am watching a sparrow.’ b. mī ek I.sg.erg one ‘I saw a sparrow.’
chimṇī sparrow.f.sg.nom
bagha-toy see-prog.pres.m.sg baghit-lī see-perf.f.sg
Gujarati (Central Indo-Aryan) exhibits yet another pattern of agreement in ergative clauses. In understanding this pattern, it is necessary to take into consideration a diachronic change in NIA that has only recently received attention in discussions of ergativity from a diachronic perspective. The NIA languages innovated differential object
Indo-Aryan diachrony 545 marking (DOM) —a pattern in which objects receive overt marking in a subset of cases, typically to mark animacy, definiteness, and/or specificity properties of the object denotation (Bossong 1985). DOM is observed in both perfective and imperfective transitive clauses. The examples in (22) are from Urdu/Hindi, which employs the marker ko to mark specificity on objects as seen in (22b) (Butt 1993b). The ko marker is also obligatory with animate human denoting objects (22c). (22) a. ram gari Ram.m.sg.nom car.f.sg.nom ‘Ram will buy a/the car.’
xarid-e-ga buy-3.sg-fut-m.sg
b. ram gari=ko xarid-e-g-a Ram.m.sg.nom car.f.sg=acc buy-3.sg-fut-m.sg ‘Ram will buy the car (a specific car).’ c. ram sita=ko Ram.m.sg.nom sita.f.sg=acc ‘Ram will see Sita.’
dekh-e-g-a see-3.sg-fut-m.sg
Urdu/Hindi
Urdu/Hindi
Urdu/Hindi
In perfective transitive clauses, the presence of DOM together with ergative case marking has an effect on the ergative agreement pattern. In most NIA languages, in clauses with ergative subjects and accusative marked objects, the verb agrees with neither argument exhibiting default (masculine or neuter singular) agreement instead. In the perfective sentence in (23), we see that the verb carries default masculine agreement and thus fails to agree with the overtly marked accusative object. (23)
gita=ne gita.f.sg=erg ‘Gita saw Sita.’
sita=ko sita.f.sg=acc
dekh-a see-perf.m.sg
Urdu/Hindi
Gujarati provides an example of an interesting variation of this pattern. Gujarati also has animacy/specificity determined DOM marking. However, unlike Urdu/Hindi, it exhibits a persistence of ergative agreement even in clauses with overtly marked accusative objects. The examples in (24) show that the verb agrees with the feminine, accusative marked object argument Sita. (24) a. rām-e rām.m.sg-erg ‘Ram saw Sita.
sitā=ne sita.f.sg=acc
b. sītā-e rām=ne sita.f.sg-erg ram.m.sg=acc ‘Sita saw Ram.’
jo-i see-perf.f.sg jo-yo see-perf.m.sg
Gujarati
Gujarati
546 Miriam Butt and Ashwini Deo
22.3.4 Summary The ergative pattern in Indo-Aryan languages thus emerges as a consequence of a result- stative construction becoming the default construction for describing past completed eventualities. Over time, there is reduction in overt marking of ergative subjects, resulting in either renewal of ergative clitics (as in Hindi) or the loss of ergativity in its entirety (as in Bengali). The Niya documents, representing a distinct late MIA system, exhibit yet another system in which subject agreement is innovated in the perfective paradigm with semantically conditioned variability in the ergative marking on transitive subjects. In other NIA languages, ergative agreement is retained despite syncretism between the nominative and ergative cases in parts of the paradigm (Marathi) and the innovation of differential object marking (Gujarati). The focus in this section has been on the structural aspects of the development of ergative patterns in Indo-Aryan. In the next section, we discuss further factors that have been adduced with respect to ergative case marking in the Indo-Aryan context, namely markedness and clausal and lexical semantic factors.
22.4 Further Factors The picture that emerges from the overall literature on the history of the ergative in Indo-Aryan is a complex one. The previous sections have focused mainly on structural factors such as changes of alignment and agreement. However, even while attempting to adhere to a strictly structural perspective, discussions with respect to Differential Case Marking (DCM) have crept in. The DCM involved semantic factors: animacy in the case of the Differential Subject Marking (DSM) found with the Niya documents and the typical DOM involving specificity that was needed to understand the NIA Gujarati patterns. In this section, we elaborate on such further perspectives and suggest that they hold a key part of the answer to understanding the diachrony of ergativity in Indo-Aryan.
22.4.1 Markedness Malchukov and de Swart (2009) and de Hoop (2009) provide comprehensive surveys of the state of the art with respect to DCM of arguments and notions of markedness. One of the main issues this work addresses is the distribution of case marking across core arguments of verbs. The underlying question is: why do languages exhibit nominative- accusative systems vs. ergative systems vs. mixed systems? The literature on DCM and markedness suggests that the emergence of the various observed patterns is rooted in the maximization of distinctions between the core arguments in a clause.
Indo-Aryan diachrony 547 DSM in particular is taken to imply the presence of two different considerations, which may sometimes conflict with one another (see Malchukov (Chapter 11, this volume) for a related discussion). 1. Distinguishing strategy: in order to distinguish subjects from objects, mark non- prototypical subjects (i.e., subjects which could be mistaken for objects). 2. Indexing strategy: identify proto-typical subjects (agents) and mark this particular semantic role. Ergative languages are taken to be one typical result of these two basic strategies. The split-ergative pattern in Indo-Aryan can be seen as a result of the indexing strategy, whereby the proto-typical subject is marked, though only in the presence of perfective morphology. As far as we are aware, the question of why split-ergativity tends to go hand in hand with perfect/perfective or past morphology has not yet been fully resolved. We suspect, along with Trask (1979), that the solution to the puzzle lies in the fact that modern perfect/perfective/past forms in ergative clauses tend to come from old participles and that these participles expressed stative situations. The reinterpretation of the stative participles as eventive verbal predications thus may also necessitate a special marking of the “unexpected” or “marked” agents (since stative situations are not expected to have agents). As such, the employment of ergative markers in these situations may also alternatively be due to the distinguishing strategy whereby non-prototypical subjects, i.e., agentive subjects of originally stative situations, are marked.
22.4.2 Complex Case Marking Patterns Invoking markedness as a conditioning factor for ergativity has the potential of contributing significantly to the overall understanding of ergative marking and ergative alignment. However, Indo-Aryan exhibits patterns that are unexpected from the markedness perspective. Consider the examples in (23) and (24) presented in the previous section. An ergative marks the agent/subject, but additionally the object also carries an overt case marker. For purposes of indexing or distinction of arguments, having two overt case markers is a situation of overkill. Furthermore, one finds examples in NIA in which the same case form marks both subject and object. From the perspective of indexing and distinction of arguments, this is again unexpected as disambiguation of arguments is not being achieved when both arguments carry the same case forms. Consider the examples from Haryani and Kherwada Wagdi in (25) and (26), which involve polysemy between ergative and dative/ accusative.4
4
In Kherwarda Wagdi, ne and ṇe are allophonic variants of one another (Phillips 2013).
548 Miriam Butt and Ashwini Deo (25) man=ne sahab=ne Pron.1.sg=acc/dat Sahib.m.sg=erg ‘The Sahib hit me.’ (Shirani 1987)
mar-a hit-perf.m.sg
(26) va-ṇe ve-ne dekh-yu Pron.3.sg=erg Pron.3.sg=acc see-perf-n.sg ‘He/she saw him/her.’ (Phillips 2013)
Haryani
Kherwada Wagdi
Both examples (25) and (26) can only be disambiguated by context. The question thus arises —why mark both of the core participants and yet not distinguish? Examples like this are unexpected not only from the markedness perspective, but also from the perspective of the passive-to-ergative and possessive-to-ergative alignment theories. In particular, while ergative/instrumental or ergative/genitive polysemy is expected, polysemy between ergative and dative/accusative is not. The next two sections offer further perspectives that allow for an understanding of examples of this type.
22.4.3 Differential Case Marking One crucial piece of the overall picture involves semantically motivated case alternations. Section 22.3 included examples of semantically motivated DCM, namely animacy of both agents and patients conditioning the appearance of the ergative marker in the MIA Niya documents and specificity conditioning DOM in the NIA languages Hindi/Urdu and Gujarati. Indeed, examples of DCM on subjects, objects, and obliques abound in NIA. Butt and Ahmed (2011) observe that DOM is an old part of the language and argue that the strategy of using case marking to express semantic/pragmatic information cannot be ignored in developing an understanding of how new case markers, including the ergative, developed in the NIA languages. A typical example of DOM involving partitivity in OIA is provided in (27). Other examples include accusative vs. instrumental marking to signal degrees of affectedness on causees or dative vs. accusative to signal abstract vs. concrete movement (Butt and Ahmed 2011). (27) a. pibā soma-m drink.imp soma-acc ‘Drink soma. (all of the quantity)’ b. pibā soma-sya drink.imp soma-gen ‘Drink (of) soma.’
Vedic (Ṛgveda VIII.36.1, from Jamison 1976)
Vedic (Ṛgveda VIII.37.1, from Jamison 1976)
Indo-Aryan diachrony 549 As far as we know, OIA did not contain DSM and the nominative was the only case used to mark subjects. This situation changed in MIA, where DSM is attested. One example is constituted by the dialect recorded in the Niya documents. Another example involves a genitive/instrumental alternation on agents documented by Andersen (1986) for Aśokan inscriptions. Andersen finds that the genitive is rarer and can only apply when the agent is animate. No such restriction applies to the instrumental. The instrumental thus appears to function as the default case, while the genitive is used to mark animacy of the agent. Bynon (2005, 17) points to work by Jamison (1979a,b) to show that Vedic, the oldest attested state of Indo-Aryan, already allowed for variable agent marking: (1) compounded in together with the verb; (2) instrumental; (3) genitive. The instrumental realization is by far the most frequent, but as part of her overall possessive-to-ergative view, Bynon argues that the genitive is originally the default marker (contra the received wisdom by Cardona (1970) and Jamison (1979a,b)) and draws a connection to evidentiality. The Sanskrit grammarian Pāṇini considered this to be part of the verbal system, with the aorist and imperfect encoding the reporting of events the speaker can attest to, while the perfect encoded events not witnessed by the speaker. Bynon (2005) thus sees the perfect as a modally marked form and the ergative ultimately as encoding a modally marked evidential. This understanding of the ergative is interesting in the light of modern patterns by which DSM is used to signal differences in modality. (28) provides an example from Bengali, (29) from Urdu/Hindi (see Butt and King (2004) for detailed discussion). (28) a. ami toma=ke cai I.nom you=acc wants ‘I want you.’ (Klaiman 1980, 279)
Bengali
b. amar toma=ke cai I.gen you=acc wants ‘I need you.’ (Klaiman 1980, 279)
Bengali
(29) a. nadya=ne dakhane Nadya.f.sg = erg post office.m.sg.obl ‘Nadya wants to go to the zoo.’
ja-na hε go-inf.m.sg be.pres.3.sg Urdu
b. nadya=ko dakhane Nadya.f.sg=dat post office.m.sg.obl ‘Nadya has/wants to go to the zoo.’
ja-na go-inf.m.sg
hε be.pres.3.sg Urdu
In Bengali there is no ergative. The DSM is between an unmarked form (glossed as nominative) and the genitive. In Urdu the contrast is between ergative and dative. Also note that the ergative here appears without being licensed by perfective morphology. This use of the ergative appears to be an innovation (Bashir 1999; Montaut 2003, 2006, 2009), but the use of case morphology in conjunction with non-finites to express
550 Miriam Butt and Ashwini Deo modality is an old part of Indo-Aryan (Bynon 2005) (and Indo-European as a whole), see (30). (30) samprati gan-tavyā puri now go-ger city.nom.f.sg ‘now I want to go to the city of Benares’
vārāṇasī Benares.nom.f.sg
mayā I.inst
Sanskrit (from Speijer 1886: section 41)
The diachronic and synchronic evidence in Indo-Aryan thus points to a long tradition of DCM. Every instance of DCM that has been documented expresses a semantic distinction. The semantic parameters range from fairly well known ones like partitivity, specificity, and animacy to the expression of modality. Since the ergative case is often implicated in DCM, this observation about Indo-Aryan needs to be taken into account for an understanding of ergativity.
22.4.4 Lexical Semantics, Borrowing, and Competition Butt (2001, 2006a) and Butt and Ahmed (2011) set out to understand the development of the NIA case system primarily from a lexical semantic perspective of case. The new NIA case markers have primarily been drawn from a set of originally spatial terms, e.g., ‘with, at, from, in, on’ (Beames 1872–79; Kellogg 1893; Chatterji 1926; Tessitori 1913, 1914; Montaut 2003, 2006; Hewson and Bubenik 2006; 2009, Reinöhl 2015). In the hundreds of NIA languages and dialects, the same handful of case markers crops up again and again. Particularly common for the marking of core arguments are k-forms (e.g. ko, ke, khe, ku), n-forms (e.g. ne, nai, ni, nũ), and l-forms (e.g. lai, le, la). The forms distribute across the case paradigms and can take on different roles in different languages. In some languages, case polysemy can be observed. In Marathi the ne form furnishes both the ergative and the instrumental. In Haryani (cf. (25)), Kherwada Wagdi (cf. (26)) and Rajasthani (Allen 1960) ne expresses both the ergative and the dative/accusative. The overlap in forms is one of polysemy and not syncretism since distinctions have not been lost over time. Rather, the originally spatial term changes and expands its lexical semantic space over time. Understanding exactly how this happens remains a general problem to be solved within historical linguistics.5 Butt and Ahmed take a Lexical-Functional Grammar (LFG) perspective under which they treat the case clitics as instantiating individual lexical items (see also the discussion in Butt and King (2003, 2004) and the computational treatment of Urdu, Butt and King (2007)). These lexical entries are associated with semantic information and, as with any lexical item, both
5 Cf. McGregor (Chapter 19, this volume) who suggests that identifying direct lexical sources for ergative markers is difficult cross-linguistically. If correct, this has interesting theoretical implications as it suggests that ergatives are derived from other case markers
Indo-Aryan diachrony 551 semantic and grammatical information may be polysemous, ambiguous, or underspecified and is subject to change over time. Butt and Ahmed see semantically based DCM as a major motivating factor in drawing new case markers into a grammatical system. Hewson and Bubenik (2006) observe that languages which develop an article system tend not to redevelop case markers (e.g. English, Romance). Case marking clearly fulfills a function beyond indexing and marking and where this function is not redistributed onto other grammatical means, new case markers are drawn into the system. In Indo-Aryan this involves semantically based DCM. Butt and Ahmed concentrate on Urdu/Hindi and on the following case markers: dative/accusative ko, ergative ne and instrumental se. ko appeared in the language some two hundred years before ne did. Early on, ko was involved in DCM involving the (non)-attainment of goals. ne appears to have entered the language through borrowing via language contact (Beames 1872–79). The agent was generally not marked explicitly in Old Urdu/Hindi, but it was marked indirectly via agreement patterns (see discussion in section 22.3). In addition, the language engaged in systematic alternations where agency/control contrasted with involuntariness (cf. Montaut 2003, 2006, 2009). When the language came into contact with another language which explicitly marked agents, it was natural to borrow that marker, thus bringing ne into the language. However, only the agency reading of ne was borrowed since the other semantic domains were already taken up by existing case markers. A possible ablative meaning of ne, found in sister languages, was expressed by se and the dative/accusative meaning was already expressed by ko. Butt and Ahmed suggest that ne is found with a more restricted usage in Urdu/ Hindi than in languages like Rajasthani, Haryani, or Kherwada Wagdi because the other potentially available meanings were blocked by existing established case markers. Thus, the range of semantic functions a given case marker is taken to be determined jointly by the meaning of the original spatial term, the constraints imposed by the existing case system and the existing case alternations. If case markers already exist in the language that block part of the available meaning space, then the full range of possible semantic functions of new, incoming case form are not expressed. Butt and Ahmed paint a complex picture of the development of case marking systems. The picture involves factoring in lexical and clausal semantic information expressed by case marking, in particular, by DCM. It also involves an understanding of blocking and competition among case markers in a language. While Butt and Ahmed place their argumentative emphasis on semantic factors, the structural agreement and alignment factors discussed in sections 22.2 and 22.3 also play a central role, as does the fact that ergatives tend to be associated with the reanalysis of stative participial predications.
22.5 Conclusion and Discussion In conclusion, much more work on the effects of historical change on case systems needs to be done. The Indo-Aryan situation is a particularly promising one for achieving a
552 Miriam Butt and Ashwini Deo deeper understanding of the diachrony of ergativity. It is characterized by the longest historical record available for any one language family. Synchronically, the large modern languages tend to be spoken by millions, the smaller ones by thousands rather than by hundreds. The modern languages and dialects show interesting variation that can be exploited for comparative analyses. Original fieldwork by Deo on the central Indian language Bhili and the dialectal variation found in Pawri, Dehawali, Ahirani, and Kokana has shown that these retain MIA patterns with respect to tense/aspect, there is thus also the possibility of synchronically investigating patterns attested in the historical record. Beyond doing more original historical and field research, we see the most promising future research on the topic as being of the type that acknowledges the complex interplay between structural and semantic concerns and seeks to investigate this further.
Acknowledgements Ashwini Deo gratefully acknowledges the support of the National Science Foundation (NSF BCS-1255547/BCS-1660959). Miriam Butt gratefully acknowledges Opus Magnum funding from the Volkswagenstiftung.
Abbreviations ‘-’, stands for a morpheme boundary; ‘=’, for a clitic boundary; abl, ablative; acc, accusative; aor, aorist; dat, dative; erg, ergative; f, feminine; fut, future; gen, genitive; ger, gerund; Imp, imperative; IMPF, imperfective; inf, infinitive; inst, instrumental; loc, locative; m, masculine; n, neutrum; neg, negation; nom, nominative; pass, passive; perf, perfective; plural, plural; ptcl, discourse particle; ptcpl, participle; prog, progressive; pv, verb particle; sg, singular; syncr, syncretic; voc, vocative.
Chapter 23
E rgativit y an d l a ng uag e change in Au st rone sia n l anguag e s Ritsuko Kikusawa
23.1 Introduction The focus of this chapter is change that takes place in the case-alignment patterns found in pronominal systems in Austronesian languages. I will discuss how ergativity related systems are described and relevant phenomena can be analyzed in the context of syntactic reconstruction and language change. Although the notion “ergativity” has been applied to capture patterns observed in various linguistic aspects (Dixon 1994), the focus of this chapter will be on changes in case-alignment patterns of nominals, or the coding of case, that is relevant to ergativity in Austronesian languages. In particular, how pronominal arguments are aligned to express S, A, and P of sentences will be examined.1 In Figure 23.1, proto-typical ergative and accusative systems are presented. An ergative system is one where S and P are formally marked alike. The case form by which they are marked is commonly labeled as absolutive or nominative. In this system, A is marked differently and is commonly labeled as ergative, but also as instrument, genitive, etc., depending on other functions that the form carries in the language. On the other hand, the accusative system is where S and A are formally marked alike and the case is labeled as nominative. It is P that is marked differently in this system and the case form is labeled as accusative. 1
The following abbreviations are used to indicate syntactic roles of the arguments of transitive and intransitive sentences, following Comrie (1978) and Dixon (2000): A=Agent of transitive verbs P= Patient of transitive verbs S=Subject (actor/undergoer) of intransitive verbs E=Extended argument of intransitive verbs
554 Ritsuko Kikusawa
Intransitive
Transitive
S
S
A
P
ERG
NOM
Ergative
A NOM
P ACC
Accusative
Figure 23.1 Ergative and accusative alignments
Case-marking strategies in Austronesian languages are typically found in paradigmatic pronominal sets, and/or sets of typically short morphemes sometimes referred to as case-marking determiners, specifiers, or adpositions that introduce lexical NPs. Another strategy is simply word order. The languages show a variety of alignment patterns, including ergative, accusative, and various other types that could be referred to as some kind of “split.”2 There are several geographical areas where ergative languages are commonly found in the Austronesian family; namely Taiwan, island Southeast Asia, including the Philippines and northern Sulawesi, Madagascar, and Polynesia. In the geographical areas between, a wide range of variation in alignment patterns is found, some of which will be discussed in this chapter. The geographical positions of the languages referred to in this section are shown in Figure 23.2, and their genetic relationships are presented in Figure 23.3. Two different case alignment patterns, namely ergative and accusative, involving pronominal systems are presented as examples in what follows. The first example provides sentences from Betsimisaraka Malagasy spoken in Madagascar (a western Malayo- Polynesian language); (1a) is an intransitive sentence, and (1b) is a transitive sentence. Here, S and P are expressed by a nominative pronoun, while A is expressed by a genitive pronoun, thus showing an ergative pattern. (1) Betsimisaraka Malagasy sentence examples (1) a. Intransitive Mandry [izy]S. be.asleep 3sg.nom ‘S/he is asleep.’
2
(Kikusawa 2008: 49)
No language is ever fully ergative or accusative, for various kinds of lexical and/or grammatical splits also exist within cases that are characterized as one type or another, and also alignment systems are not confined to the relations between what are referred to as S, A, and P. In this sense, it is not accurate to include “split system,” as though the term carries equal status to ergative and accusative (Bickel and Nichols 2009).
Figure 23.2 Geographical positions of the languages referred to in this chapter
Ergativity and language change in Austronesian languages 555
556 Ritsuko Kikusawa
Austronesian Eastern Formosan Malayo-Polynesian (Extra-Formosan) (Western Malayo-Polynesian)* Nuclear Malayo-Polynesian Ibaloy Tagalog Sunda-Sulawesi (Central-Eastern) Mamuju Pendau Indonesian Balinese Central Eastern Malayo-Polynesian Malayo-Polynesian Tetun South-Halmahera/ Oceanic West New Guinea Taba Admiralty Western Central-Eastern Oceanic Oceanic Southeast Solomonic
Eastern (Remote) Oceanic Central Pacific Polynesian
Fijian Rotuman Tongan West Futuna-Aniwan
East Futunan, East Uvean
Samoan Eastern Polynesian
Māori
Figure 23.3 Languages referred to in this chapter (in italic and boxed) and their genetic relationships
Ergativity and language change in Austronesian languages 557 b. Transitive Tia[=ko]A [izy]P like=1sg.gen 3sg.nom ‘I like him/her/it.’
(Kikusawa 2008: 49)
In (1a), it should be noted that the sole argument is labeled as nominative, rather than absolutive, since the term “nominative” is defined here as the case form that marks the sole argument of a basic intransitive sentence, regardless of whether the language shows an ergative or an accusative pattern. This definition is based solely upon distributional features within a language. This is unlike the traditional labels for case forms in European languages in that it does not imply any historical or etymological origin. In (1b), the form expressing A is labeled as genitive rather than ergative, since it is common in western Austronesian languages that the forms marking the agent of transitive sentences are identical to those marking the possessor of a noun, that are typically referred to as “genitive.” Austronesian languages that show an ergative pattern often have another dyadic sentence structure in addition to the transitive one, such as the sentence shown in (2). The general meaning of the sentence (‘I am fond of him/her/it’) is the same as that of (1b) (‘I like him/her/it’), however, the case alignment of the sentence is different. The agent ‘I’ is expressed by nominative instead of ergative, and the patient ‘him/her/it’ is expressed by oblique instead of nominative. (2)
Betsimisaraka Malagasy sentence Extended intransitive Tia [ananjy]E [zaho] like 3sg.obl 1sg.nom ‘I am fond of him/her/it.’
(Kikusawa 2008: 49)
A non-transitive dyadic sentence such as (2) is labeled in this chapter as “extended intransitive” and the non-nominative argument in this structure is labeled as “E,” following Dixon 1998. This structure has been analyzed as forming part of a voice alternation system and has been referred to as “anti-passive” in some literature (Payne 1982; de Guzman 1988; Gerdts 1988b; Manning 1996; Aldridge 2012b; and others). However, to avoid dealing with the fundamental question of whether the structure should be analyzed as such, and to keep structural comparisons free from the different implications that are associated with this term, it will not be used in this chapter, unless it is part of the terminology used in the original source. The typical ergative case alignment pattern of pronominal systems is summarized in Table 23.1.
558 Ritsuko Kikusawa Table 23.1 Typical case alignment patterns in Austronesian ergative pronominal systems actor Intransitive
undergoer S (nom)
Extended intransitive
S (nom)
E (obl/loc)
Transitive
A (gen)
P (nom)
Accusative systems are commonly found in Oceanic languages. Example (3) shows one such system from Rotuman (spoken on Rotuman Island in the north of the Republic of Fiji). Here, unlike the Malagasy example, there is a single set of pronouns and the case is marked by the relative position to the verb. The form for 3pl is iris, which expresses S or A when preceding the verb (3a–c), and P when it follows the verb and is not preceded by an adposition (3c). A common view of case-marking is to include marking by word order, and following this view, Rotuman can be said to show an accusative pattern. (3) Rotuman sentence examples a. Intransitive [Iris]S ‘ā 3pl eat ‘They ate. /They were eaten.’ b. Intransitive [ŋou]S pæ 1sg sit ‘I sat down.’
sio down
c. Transitive [ŋou]A fesiʔen 1sg hate ‘I hate them.’
[iris]P 3pl
(Churchward 1940: 123)
(Vamarasi 2002: 38)
(Churchward 1940: 22)
Just as in languages showing an ergative pattern, Rotuman also has a dyadic intransitive sentence structure, an example of which is presented in (4). Here, the form for 3pl is irisa, which is preceded by a locative preposition (se) and expresses E. There are two sets of pronouns in Rotuman, namely, short forms (or “incomplete phase”) and long forms (or “complete phase”). Short-form pronouns occur in nominative and accusative case positions, while long-form pronouns occur when preceded by a preposition (Churchward 1940). (4) Rotuman intransitive dyadic sentence example c. Intransitive + PP [ŋou]S fesiaʔ [se irisa]E 1sg hate loc 3pl ‘I hate (toward) them.’
(Churchward 1940: 22)
Ergativity and language change in Austronesian languages 559 Argument structure of Rotuman sentence structures can be summarized as in Table 23.2. Figure 23.4 schematically shows ergative and accusative alignments including extended intransitive sentence structures. Table 23.2 Typical case alignment patterns in Austronesian accusative pronominal systems actor Intransitive
undergoer S (nom)
Extended intransitive
S (nom)
E (loc)
Transitive
A (nom)
P (acc)
It should be noted that these are just two of a wide variety of ergative and accusative patterns found in Austronesian pronominal systems.3 Clarifying the ways in which so many typologically different systems emerged from a single proto-system makes a fascinating scientific endeavor. The rest of this chapter is organized as follows. In section 23.2, methodology, terminology, and related issues are discussed. In section 23.3, three different sets of changes that resulted in a change in the case-alignment of Austronesian languages will be presented. For each set of changes, those that affected the interpretation of the case system are discussed, and how the results may have influenced other grammatical features are considered. Section 23.4 presents concluding remarks.
Monadic
Intransitive
S
Extended Intransitive
S
S E
E
S
Diadic Transitive A
P
ERG
NOM
Ergative
LOC/ OBL
A
P
NOM
ACC
LOC
Accusative
Figure 23.4 Ergative and accusative alignments including the extended argument 3
For specific examples of various alignment systems and some discussion about their differences from a historical perspective, see Kikusawa (2015b).
560 Ritsuko Kikusawa
23.2 Methodology Syntactic comparison and reconstruction was long considered to be a hazardous if not impossible endeavor. However, increasing interest is seen today in finding ways to apply the principles of the Comparative Method, and promising results are being published (Harris 2008, Gildea 2000, and others; see Barðdal 2015 for an overview). In such attempts, the following appears to be the consensus among researchers. First, the comparanda between languages need to be based on surface structures, where changes take place and are directly observed. The comparanda in this chapter are abstracted sentence patterns describing the argument structures of selected languages. Second, syntactic changes take place as a result of a cascade of changes (Roberts 2007; de Smet 2015). As Fried (2008: 47) states, “the gradualness of change [of grammatical phenomena] consists in discrete partial changes that involve specific features or aspects of a larger pattern before they affect the full pattern completely.” Thus, there is no holistic change such as a language switching from accusative to ergative, or ergative to accusative overnight. Approaches in this chapter follow this view. Since changes that take place in each form and category are discrete from other changes, in conducting historical examination, linguistic features forming part of the examined system have to be decomposed and analyzed separately. In dealing with case-alignment systems, changes in personal pronouns are examined separately from changes in the case-marking forms on lexical NPs. Moreover, word order change is perceived as a separate change, with the recognition that its results may interact with those of case-marking systems. Verb morphology is by definition not part of a case-alignment system, and comments on verb morphology will be restricted to those cases where an identified change in the coding of case affects or is affected by the verb morphology. On the other hand, person and number agreement markers, which in many languages compose part of the verbal morphology, often develop from pronominal forms that marked case and need to be examined in this context. Such an approach may appear to obscure the typological characteristics of each language and their ancestral languages, however, any system a language exhibits today is the end result of a collection of changes that took place in various grammatical features over time, changes which exist in layers. Having the above in mind and understanding that syntactic reconstruction is the reconstruction of patterns and not the reconstruction of actual sentences used in a proto-language (cf. Harris 2008), the following two aspects are considered to be particularly important in this work: (i) identifying cognacy and sentence pattern correspondences; (ii) identifying the direction of change (Barðdal 2015; Luján, Barðdal et al. to appear). In this chapter, cognacy among sentence structures is identified based on morpho-functional correspondence; more specifically, the occurrence of the ergative/ genitive pronouns associated with the A or actor of syntactically transitive sentences. It was mentioned earlier that in Austronesian ergative languages, it is often found that A is expressed by a genitive form (cf. (1)). Even when not, it is often found that a trace of what was formerly a genitive pronominal set is identifiable, as will be seen in section 23.3,
Ergativity and language change in Austronesian languages 561 and these are invaluable in identifying cognate structures. This is a method inductively developed, rather than theory oriented, based on the fact that such pronouns are found to be more stable than other relevant components. It is not clear what the reason is for this stability; it may be because of the salient nature of the referents of the phrase marked as ergative, or simply because this was the only set of pronouns that carried clitic status in their shared proto-language. Whatever the reason, tracing the occurrence of these pronominal forms in various sentence structures enables us to trace sentence structure changes and identify the directionality of change. Analyses and glossing follow the source description of each language, unless otherwise specified. It should be noted that case is usually determined according to typological criteria. The form or marking on S is by definition nominative. If A receives a different marking from that on S, it will be referred to as ergative, while if it is P that receives a different marking, it will be referred to as accusative. As has been mentioned earlier, the marking on A and the form of the associated pronouns is often shared in Austronesian languages with that of the possessor of a noun in noun phrases, and is consequently labeled as genitive (rather than ergative) in such cases. Typologically defined terms do not necessarily reflect etymological relationships, and the functional change of each case needs to be traced, based on formal correspondences. Finally, a few words are made here regarding macro-and micro-comparison. Languages compared in this chapter are all daughter languages of Proto-Malayo-Polynesian, the parent of the subgroup of Austronesian languages comprising all those that are geographically located outside of mainland Taiwan. Those that are directly compared with each other are often distantly related. The advantages of conducting such a macro-comparison is that it makes it easier to identify direct inheritance. Historical examination of closely related languages (micro-comparison) is often complicated by having to deal with a mixture of direct and indirect inheritance (that is borrowing from closely related languages), as well as sporadic local innovations, where earlier features are obscured by layers of change that took place subsequent to the split of the languages. It should be remembered that the comparison and reconstruction of lexical items and sound systems, which is today conducted applying “bottom-up” methodology, was initially done by macro-comparison, which set the basis for detailed bottom-up micro-comparison (cf. Blust 1990: 137-138). Needless to say, follow-up modification of any proposed hypothesis is always necessary, based on new data and the results of micro-comparison. I consider that macro-comparison is a reasonable way to start the investigation.
23.3 Ergativity Related Change in Austronesian Languages In this section, three sets of changes that resulted in a shift from an ergative to a different alignment system are described. In 23.3.1, a case of change from an ergative to a
562 Ritsuko Kikusawa bi-transitive system via an inverse one will be presented. This was probably triggered by a word order change, which eventually led to a system where word order alone marks grammatical relations. Among the sequence of relevant changes, it will be argued that it was the loss of the contrast between two distinct dyadic sentence structures that resulted in a change in the case-alignment of the language. In 23.3.2, a case of change from an ergative to an accusative system as a result of a merger of two pronominal sets will be provided. It is argued that this change resulted from the generalization of earlier genitive pronouns marking A to mark S, and the genitive pronominal forms subsequently began to be replaced with forms from other pronominal sets. In 23.3.3, an ergative to accusative change as a result of change in the distribution of morphological forms will be summarized. Languages in which the same set of case markers on lexical NPs can be interpreted as either ergative or accusative will be examined. For each set of changes discussed, the mechanisms by which the changes took place and their preconditions are described. Since the methodology for morphosyntactic comparison and reconstruction is not yet well established, I will explain how the changes described relate to the general principles of comparative (historical) linguistics. The combination of the results of the discussion in 23.3.1 through 23.3.3 gives a general overview of some of the case-alignment changes in the Austronesian language family.
23.3.1 Change from a Morphologically Marked Ergative System to a Word Order Marked Bi-transitive System Basic argument structures compared here are those in Ibaloy spoken in Luzon, the Philippines, based on Ruffolo (2004), and Pendau spoken in Sulawesi, based on Quick (2007). This is a comparison of an ergative pattern pronominal system and what has been described as an inverse system respectively. It will be shown that a historical comparison and reconstruction reveals that the shared proto-system was an ergative pattern and that the inverse system developed as a result of word order change. It will be also shown that the Balinese system shares different stages of the same line of development. Ibaloy example sentences are presented in (5). Sentence (5a) is an intransitive sentence with the S is expressed with a nominative clitic pronoun. Sentence (5b) is an extended intransitive sentence where the S is expressed by the absence of marking implying a third person singular nominative (as is the case in many Philippine languages) and the E is expressed with an oblique PP. Sentence (5c) is a transitive sentence, where the A is expressed by a genitive pronoun and the patient by a nominative pronoun.4 The system as a whole thus shows an ergative pattern contrasting S and P vs. A, following the pattern described in Table 23.1. 4
The sentences are cited from Ruffolo (2004), where all example sentences are extracts from texts and therefore the sentences do not show matching pronouns to show the case marking contrast.
Ergativity and language change in Austronesian languages 563 (5)
Ibaloy sentence examples with pronominal arguments a. imakadkami ali ʔakad[=kami]S ʔali go=1+/nom toward ‘We went back home.’ b. on’aseba ʔon-ʔasǝwa[Ø]S ActV/ipf-marry=3sg/nom ‘She will get married to him.’
son [so=n obl=gen/pers
(Ruffolo 2004: 175) si’kato siʔgato]E 3/ind (Ruffolo 2004: 150)
c. tinodonganchaka toloŋ-an[=da]A[=ka]P help-lv=3pl/gen=2sg/nom ‘They helped you.’
(Ruffolo 2004: 418)
Pendau sentence examples are presented in (6). Here, there are two sets of pronouns, which are labeled as “absolute” and “genitive.” Absolute pronouns occur marking S in intransitive sentences as in (6a), the actor (preceding the verb) and the undergoer (following the verb) as in (6b), and the undergoer (preceding the verb) as in (6c, d). The genitive pronoun expressing the actor occurs following the verb in (6c, d). As can be seen in the examples, there are two dyadic constructions in Pendau, one in which the actor precedes the verb and the undergoer follows the verb (6b), and the other in which the order is reversed and the undergoer precedes the verb and the actor follows the verb (6c, d). (6) Pendau sentence examples with pronominal arguments5 a. Sampanyo [jimo]S neosa sampanyo jimo N-pe-osa after.that 3pl.abs r-dy-rest ‘After that they rested.’ b. [Io]Aav neng-ebiling 3sg.abs av/r-leave ‘He left me.’
5
[’a’u]Pav 1sg.abs
c. [’A’u]Piv ni-ebiling 1sg.abs iv/r-leave ‘He left me.’
[-onyo]Aiv -3sg.gen
d. [Io]Piv ni-ebiling 3sg.abs iv/r-leave ‘I left him.’
[-o’u]Aiv -1sg.gen
(Quick 2007: 235)
(Quick 1997: 467)
(Quick 1997: 467)
(Quick 1997: 467)
In Quick (1997), the absolute and genitive cases are labeled as proximate and obviate respectively. The glossing here follows that in Quick (2007).
564 Ritsuko Kikusawa Quick (2007: in particular 360–387) analyzes the language as showing an inverse system based on an “emic analysis.” Both of the dyadic structures in this language are analyzed as syntactically transitive, and the structure where the actor precedes the verb and the undergoer follows the verb such as (6b) is referred to as “active voice,” while the one where the undergoer precedes the verb and the actor follows the verb such as (6c, d) is referred to as “inverse voice.” The arguments of each structure are indicated as Aav ‘A of active voice’, Pav ‘P of active voice’, Aiv ‘A of inverse voice’ and Piv ‘P of inverse voice’. The case marking system of Pendau is summarized in Table 23.3. Table 23.3 Case alignment pattern in the Pendau inverse system actor Intransitive
undergoer S (abs)
transitive (actor voice)
A (abs)
P (abs)
Transitive (inverse voice)
P (abs)
A (gen)
In what follows, to order to compare and reconstruct the sentence structures of the two languages described above, abstracted sentence structures are first presented, then cognate structures are identified and the changes that took place in their pronominal systems will be discussed. The changes presented here are considered to be generally shared between pronominal and other NP marking systems, but in cases where they are not, the differences will be indicated. Abstracted argument structures of Ibaloy are presented in (7). Personal pronouns are represented by the case names in each position. Genitive pronouns mark the A of a transitive construction, and nominative pronouns mark S in an intransitive construction and P in a transitive one. The NP expressing E in structure (7b) is marked by an oblique case-marking form so. The pronominal P may be expressed by a clitic pronoun (7c) or an independent pronoun (7d). (7) Abstracted argument structures in Ibaloy (1) Personal pronouns (without Aux) a. Vi[=nom]S b. Vi[=nom]S [so-n ind]E c. Vt[=gen]A[=nom]P d. Vt[=gen]A [nom/ind]P Abstracted argument structures of Pendau are presented in (8), showing that in this language there is one intransitive structure (8a), and two transitive structures (8b, c). Sentence structure (8b) is referred to as “actor voice (av),” and the initial argument is A, and the argument following the verb is P. Sentence structure (8c) is referred as “inverse voice,” where the initial argument is a P, and the argument following the verb is A. In this
Ergativity and language change in Austronesian languages 565 language, the A of the inverse voice (Aiv) is expressed by genitive forms, while others (S, Aav, Pav, Piv) are expressed by “absolute” (equivalent of nominative) forms. (8)
Abstracted argument structures in Pendau (1) Personal pronouns a. Vintr [abs]S b. [abs]Aav Vav [abs]Pav c. [abs]Piv Viv[-gen]Aiv
The etymological correspondences among the sentence structures of Ibaloy and Pendau are shown in Figure 23.5. In the figure, cognate structures are boxed together with dotted lines. Structures (7a) and (8a), structures (7b) and (8b), and structures (7c, d) and (8c) are identified to have developed from the same proto-structure and thus are cognate structures. These cognate structures are first identified here by the occurrence or non- occurrence of genitive pronominal forms associated with the marking of A and are then confirmed by the distribution of other grammatical forms (Kikusawa 2003, Kikusawa to appear b). In Ibaloy sentence (7c) and its alternate (7d), and Pendau sentence (8c), a genitive pronoun expresses the A and these are thus identified as cognate structures. The monadic sentence structures (7a) and (8a) are assumed to have developed from the same intransitive structure, and sentence structures (7b) and (8b) are also assumed to be cognate structures based on the fact that they are dyadic with no genitive pronoun marking A occurring in the structure. These sentence correspondences are supported by various types of evidence, in particular by the distribution of various verbal affixes. For example, the distribution of the verb prefix ʔon-‘imperfective, actor verb’ (5b) in Ibaloy, and neng-‘realis, actor voice marker’ (6b) in Pendau are restricted to the intransitive or actor voice structures (7a, b) and (8b). This can be associated with the syntactic distribution of the commonly found meN-/n eN-verb alternation in western Austronesian languages (e.g., məN-/nəN-verbs in Ibaloy and mong-/nong- verbs in Pendau),6 although space does not allow me to go into details here. Likewise, the Ibaloy
Figure 23.5 Cognacy among sentence structures in Ibaloy and Pendau 6
Vowels in the prefixes listed here for the two languages alternates depending on the phonological environment.
566 Ritsuko Kikusawa infix ‘perfective, locational voice’ (5c), and the Pendau prefix ni-‘realis, inverse voice marker’ (6c, d), which are probably related, occur only in transitive structures such as (7c, d) and (8c). Based on the identified cognate relationships, it becomes possible to reconstruct how the different systems developed. For the sake of convenience, I will refer to the two systems as systems (7) and (8). Here I claim that the general direction of the change is as from (7) to (8). This is based on the fact that this direction of change is readily explained while the reverse is not, as summarized in what follows.7 The existence of the two systems is explained as the result of word order change and subsequent changes where morphological contrast was lost. The sentence structures shown in (9) represent the assumed proto-system, which resembles the Ibaloy system (7). From this system, (8) is considered to have developed with the nominative NP acquiring pre-main-verb position. The new word order is a result of fronting, which was a topicalized position in the proto-system, becoming an unmarked position for nominative NPs.8 The process of the change from the proto-system (9) to the Pendau system (8) is schematically shown in Figure 23.6. (9) Reconstructed shared proto-system of Ibaloy and Pendau a. Vi[=nom]S b. Vi[=nom]S [ni ind]E c. Vt[=gen]A [nom]P In the proto-system, there are two dyadic constructions but with the nominative argument having a different macro-role in each. The nominative NP in (9b) is actor while in (9c) it is undergoer. Thus, the fronting of nominative NPs yielded two sentences where the positions of the actor and the undergoer are reversed, such as the one described in (6b, c–d) and repeated here as (10a, b). (10) Two dyadic sentence examples in Pendau (repeated from (6)) a. [Io]Aav neng-ebiling [’a’u]Pav (=(6b)) 3sg.abs av/r-leave 1sg.abs actor undergoer ‘He left me.’ b. [’A’u]Piv ni-ebiling 1sg.abs iv/r-leave undergoer ‘He left me.’
7
[-onyo]Aiv (=(6c)) -3sg.gen actor
(Quick 1997: 467)
(Quick 1997: 467)
See Kikusawa (2003a, to appear b) for details of the relevant discussion. The same process is proposed by Aldridge (2004, 2008b, 2010) but from a different theoretical perspective. 8
Ergativity and language change in Austronesian languages 567
Figure 23.6 Change from a shared proto-system to the pre-Pendau system
To illustrate the claimed change, Ibaloy extended intransitive sentences are presented in (11) and transitive sentences in (12), where nominative NPs are fronted to be topicalized. When the nominative NPs are fronted, the semantic actor (S) precedes the verb in extended intransitive sentences (11), and the semantic undergoer (P) precedes the verb in transitive sentences (12), showing the two word orders as in (10). According to Ruffolo (2004: 470), “When the topicalized constituent corresponds to the Nominative complement of a verb, the clause usually contains a resumptive Nominative pronoun which is co-referential with the topic. The pronoun is a bound form. When the topic refers to a third person singular entity no resumptive pronoun surfaces in the clause.” A resumptive pronoun occurs in examples (11a) and (12a) where the topicalized constituent is 3pl and a resumptive clitic pronoun is ʔida. In examples (11b) and (12b), the topicalized constituent is 3sg and there is no pronominal form occurring in the main clause. (11)
Ibaloy intransitive sentences with a topicalized nominative NP a. sota enganak tan sota nankaama, [sota ʔəN-ʔanak tan sota nanka-ʔama]TS nom/rec ActV/pft-sponsor and nom/rec StaPatV/pl-old.man menginom ira ni tapey. məN-ʔinom [ʔida]S [ni2 tapəy]E ActV/ipf-drink 3+/nom gen rice.wine ‘as for the sponsors and the old men, they drink rice wine’ (Ruffolo 2004: 471) b. nem sota embalangan aso, [nəm sota ʔən-balaŋa=n ʔaso]TS but nom/rec StaV/en-red=lk dog timiyed chi toktok ni chontog tijəd[=Ø]S [di toktok ni dontog]E climb loc head summit gen mountain ‘but as for the red dog, it climbed to the top of the mountain.’ (Ruffolo 2004: 471)
568 Ritsuko Kikusawa (12)
Ibaloy transitive sentences with a topicalized nominative NP a. say bebaknang ket in’obdaanto ira [saj cv-baknaŋ]TP kət ʔin-ʔobla-an[=to]A [ʔida]P top pl-rich TpLk BnfV/ipf-wok-BnfV=3/gen 3+/nom ‘as for the rich people, he works for them’ (Ruffolo 2004: 473) b. emin ya kanenmi [ʔəmin ja kan-ən=mi]TP all lk eat-PatV/ipf=1+/gen ‘all our food, he eats’
ket kət TpLk
kanento kan-ən[=to]A eat-PatV/ipf=3/gen (Ruffolo 2004: 473)
When the topicalized constituent is pronominal, it is the independent pronoun which occurs in the fronted position, with a corresponding resumptive pronouns. In (13), for example, the nominative pronoun si-Kam occurs in the fronted position with the nominative clitic pronoun (=2sg) occurring as a resumption pronoun. The sentence structures with topicalized nominative NPs are presented in (14). It can be hypothesized that from this system, the Pendau system must have developed via the one where the resumptive pronouns are optional as in (15) and (16). (13) Ibaloy sentence with a topicalized pronominal NP [Si-Kam,]TS daw[=ka]S [shi Bagiw]E 2sg.ind go=2sg.nom loc Baguio ‘You, you go to Baguio.’
(Ballard 2011: 796)
(14) Sentence structures with topicalized S and P in Ibaloy a. [ind]S Vi[=nom]S b. [ind]S Vi[=nom]S [ni ind]E c. [ind]P Vt[=gen]A (15)
Pre-Pendau system (1) a. [ind]S Vi(=nom) b. [ind]S Vi(=nom) c. [ind]P Vt[=gen]A
(16) Pre-Pendau system (2) a. [ind]S Vi b. [ind]S Vi c. [ind]P Vt[=gen]A
[ni ind]E
[ni ind]E
Note that this change is naturally motivated if we assume that the proto-system was not only morphologically ergative but also syntactically ergative. In the Philippine-type languages today, including Ibaloy and Tagalog, nominative NPs are the only core NPs that can be fronted for topicalizing (or, “extracted,” cf. Kaufman, Chapter 24, this volume). The fronted position is marked for these languages with either an intonational break, or a
Ergativity and language change in Austronesian languages 569 formal signal following the topicalized phrase labeled as a “topic linker,” while in Pendau and other languages in Indonesia, pre-verbal position is default for these NPs. Thus, the word order change between the two systems is readily explained by the originally allowed fronting of nominative arguments becoming a fixed position. If we assume the direction of the change was the reverse, it would be difficult to explain both motivations and the actual procedure. We would have to assume that, first, the sentence initial NPs, the macro-role of which (actor or undergoer) had been marked by word order and somehow started to follow the verb at a certain stage. It would have to be assumed that in this process, the nominative NP expressing the actor in the actor voice sentence had acquired post-verbal position, the position that marked the other NP as expressing the undergoer. The pre-verbal position would have had to be retained to become the marked position for topicalization. All these processes are much more difficult to explain and support than the proposed hypothesis. The proposed change appears to apply to lexical NPs as well as pronominal NPs as described in what follows. In Ibaloy, the case marking of lexical NPs shows the same case alignment as the pronominal system. Abstracted sentence structures are presented in (17). The Ibaloy forms si and ʔi (=y) expressing S and P, mark the difference between personal and common nouns respectively, and so are nen and ni expressing A. Two dyadic sentence examples are given in (18), where nominative NPs S and P are underlined. Thus, in Ibaloy, both personal pronouns and NP markers show an ergative pattern. As Kaufman (Chapter 24, this volume) states, “according to the criteria of case and transitivity, Tagalog, and the vast majority of Philippine languages, are clearly ergative,” and Ibaloy is not an exception.9 The relative position of the two lexical NP arguments is free, except that the genitive NP expressing A normally precedes the nominative complement (Ruffolo 2004: 417). (17)
9
Abstracted argument structures in Ibaloy (2) Lexical NPs (personal/non-personal)10 a. Vi [si/ʔi NP]S b. Vi [si/ʔi NP]S [so=ni2, di NP]E c. Vt [nǝn/ni1 NP]A [si/ʔi NP]P
In the description of Philippine languages, attempts have been made to simultaneously capture the system combining two different grammatical features, namely, case-marking patterns and the alternation of verb forms. Results of such attempts include symmetrical voice analyses, focus analyses, among others. Although an ergative analysis and a voice analysis are not mutually exclusive, those who take an either-or position, rejecting the former, often do so questioning the typological acceptability of the status of what are analyzed in this chapter as “extended intransitive,” as syntactically intransitive. See Kaufman (Chapter 24, this volume) for further discussion. 10 Ruffolo (2004: 471) glosses both ni and ni as genitive. However, in this chapter, because of the 1 2 functional difference and also semantic difference (an NP marked by ni1 is always definite while one marked by ni2 is always indefinite), the two forms are differentiated and the former is analyzed as genitive, while the latter as oblique. Incidentally, the NP marked with ni2 (oblique) does not alternate with a genitive pronoun, as ni1 (genitive) does.
570 Ritsuko Kikusawa (18)
Ibaloy sentence examples with lexical NPs a. engoney i aki ni ʔǝN-ʔonǝj [ʔi ʔaki]S [ni2 ActV/pft-see nom monkey obl ‘the monkey saw a mouse’
otot ʔotot]E mouse
(Ruffolo 2004: 238)
b. naon’an ni dedaki sota bibiid Batan na-ʔonǝj-an [ni1 cv-laki]A [sota cv-biʔi=d batan]P Pot.LocV/pft-see-LocV gen pl-man nom/rec pl-woman=loc Batan ‘the men happen[ed] to see the women of Batan’ (Ruffolo 2004: 306) The case-alignment pattern is shared by pronouns and lexical NPs in Pendau as well. Schematic Pendau sentence structures with lexical NP arguments are presented in (19), followed by two sentence examples illustrating (19b, c) in (20). The forms marking lexical NPs show the same pattern as pronouns, and carry the same case, with si and Ø marking respectively personal and common nouns when absolutive/nominative. The forms ni and nu marking respectively personal and common nouns when genitive (19). The post verbal positions of (19b, c) are fixed. The positioning of other NPs (including the sole NP in (19a)) is flexible and may precede the V or occur in sentence final position following the other NP (Quick 2007: 366–369). (19) Abstracted argument structures in Pendau (2) Lexical NPs (proper/common) a. Vintr [si/Ø NP]S b. [si/Ø NP]Aav Vav [si/Ø NP]Pav c. [si/Ø NP]Piv Viv [ni/nu NP]Aiv (20) Pendau actor and inverse voice sentence examples with lexical NP arguments a. [Si kai]A neng-ita-i [si be’e]P abs/pnm grandfather av/r-see-loc abs/pnm grandmother ‘The grandfather saw the grandmother.’ (Quick 1997: 466) b. [Si be’e]P ni-ita-i abs/pnm grandmother iv/r-see-loc ‘The grandfather saw the grandmother.’
[ni gen/pnm
kai]A grandfather (Quick 1997: 466)
By comparing the pre-Pendau system in (21), which is extracted from Figure 23.6, and the current Pendau system repeated here in (22), the following changes are inferred: (i) nominative NP started to occur in the sentence-initial position relatively freely; (ii) the position of the non-fronted NP in the extended intransitive structure (E in (21b) was fixed to the post-verb position; (iii) the sentence-initial position became the (semi-)default position for nominative NPs, and (iv) the syntactic distinction between extended intransitive and transitive ((21b) and (21c)) was lost. Of these, the last point is discussed in some detail in what follows.
Ergativity and language change in Austronesian languages 571 (21)
Pre-Pendau system a. [nom]S Vi b. [nom]S Vi [ni ind]E c. [nom]P Vt[=gen]A
(22) Argument structures in Pendau with pronouns (=(8)) a. Vintr [abs]S b. [abs]Aav Vav [abs]Pav c. [abs]Piv Viv[-gen]Aiv (23) Argument structures in Pendau with lexical NPs (=(19)) a. Vintr [si/Ø NP]S b. [si/Ø NP]Aav Vav [si/Ø NP]Pav c. [si/Ø NP]Piv Viv [ni/nu NP]Aiv As has been shown, both Ibaloy and Pendau have two kinds of dyadic sentence structure. The difference in the case alignment analysis of the two languages resides in the fact that, in Ibaloy, only one of the two dyadic structures is considered to be a canonical transitive, while in Pendau, both of them are analyzed as transitive. The Ibaloy analysis depends primarily on the morphology of the verb. Non-canonical transitive, verbs typically match the morphological structure of monadic intransitive verbs, hence their analysis as extended intransitive constructions. For example, verb affixes on- and məN-and their alternating forms occur in clear intransitive and extended intransitive sentences as in (24) and (25). It can be seen that the prefix ʔon-both in a monadic (24a) and dyadic (24b) sentences, while the prefix maN-occurs in a monadic sentence (25a) and its alternate naN-in a dyadic sentence (25b).11 (24) Examples of Ibaloy prefix on-in monadic and dyadic intransitive sentences a. onchakchak [i chanom]S ʔon-dakdak ʔi danom ActV/ipf-boil nom water ‘the water will boil’ b. onbono [sota too]S [ni ʔon-bono sota toʔo ni ActV/ipf-kill nom/rec person gen ‘the person will kill a snake (or snakes)’
11
oleg]E ʔoləg snake
(Ruffolo 2004: 216)
(Ruffolo 2004: 218)
The vowels in Ibaloy maN-/m əN- and naN-/n əN-alternate depending on word stress.
572 Ritsuko Kikusawa (25) Examples of Ibaloy prefix məN-in monadic and dyadic intransitive sentences a. mantejaw [ira]S ni pigen adew maN-tajaw ʔida ni piga=n ʔakəw ActV/ipf-traditional.dance 3+/nom gen several=lk day ‘they will dance for several days’ (Ruffolo 2004: 243) b. nangda naN-ʔala[=Ø]S ActV/pft-get=3/nom ‘he got some sugar cane’
ni ni gen
onas ʔonas sugarcane
(Ruffolo 2004: 242)
At the same time, the referent of the oblique NP of such constructions has to be indefinite or partitive, while the second NP of canonical transitive constructions are always interpreted as definite. In Pendau, however, Pav, the argument that historically corresponds to the oblique in Ibaloy, is “referentially definite or indefinite, but when it takes a demonstrative…it must be definite” (Quick 2007: 366). Quick (1997: 462) concludes that the two dyadic sentence structures such as (20) should both be analyzed as two primary transitive clauses in this language. The selection of which depends on contextual and structural constraints, which can be explained using functional/pragmatic parameters. Unlike Ibaloy where there is a semantic difference between two corresponding dyadic structures, the difference between the two structures has apparently been lost in Pendau. The applicative verb morphology occurring on verbs in both active and inverse voice structures is probably the result of an innovation subsequent to the loss of this distinction (see Kikusawa to appear a for details). The change is schematically shown in (26), which applies to both pronominal and lexical NPs.12 From the earlier intransitive verb (Vi), the Pendau actor voice verb develops (Vav) and from the earlier transitive verb (Vt), the Pendau inverse voice verb (Viv) develops. (26) Illustration of word order change from system (9) to system (8) proto-system nominative fronted inverse system Vi S E → S Vi E → A Vav P Vt A P → P Vt A → P Viv A The generalization of the change that has been discussed above is from a tripartite case contrast system (Ibaloy and other Philippine languages, with nominative, genitive and oblique cases marking core arguments) to a bipartite one (Pendau, with genitive and 12
The reconstruction of case-marking forms is commonly more difficult than general lexical reconstruction for various reasons, and although relevant to the current discussion, I take it as out of scope of this chapter. Case-marking forms are typically short (commonly monosyllabic) and tend to undergo sporadic sound changes. Such paradigmatic forms are susceptible to replacement, and their functional interpretation changes when the system changes, obscuring cognate relationships (cf. Reid 1978).
Ergativity and language change in Austronesian languages 573 nominative cases marking core arguments), with neutralization of the contrast between earlier nominative case and oblique case forms and the functional difference between them being marked by word order. The bi-partite case-marking system is found in many languages in Indonesia today, in which only residues of the earlier genitive pronoun are found within a single pronominal system. In such a system, word order is the major means of case marking. Example sentences from Balinese are presented in (27). It can be seen that the same pronoun occurs expressing the A of AVP pattern dyadic sentence (27b) and the A of PVA pattern dyadic sentence (27c), the latter of which would be expressed by a genitive pronoun in languages like Pendau. (27)
Sentence example from Balinese (Wechsler and Arka 1998: 388) a. [Ia]S pules (s)he sleep (S)he is sleeping. b. [Tiang]A numbas I av.buy ‘I bought the pig.’ c. [Bawi-ne punika]P pig-def that ‘I bought the pig.’
[bawi-ne pig-def
punika]P (high register) that
tumbas ov.buy
[tiang]A (high register) I
The sequence of changes is in line with the knowledge that in verb-medial languages, “dispreference for [morphological] case marking” is commonly observed and explanations for this tendency have been offered (Siewierska and Bakker 2009; cf. Hawkins 2004 for possible explanations). If in fact word order change took place and languages became verb-medial, the languages can then be said to have followed this path toward less case marking. Whatever the motivations are, with no semantic difference between two dyadic sentences, and the system being typically described as showing a “voice system” with actor/agent voice and patient voice sentence structures, the system can no longer be analyzed as showing ergative. The change presented in this section is from a morphologically marked ergative system to a word order marked bi-transitive system. The discussion is summarized in Figure 23.7.
23.3.2 A Development of an Accusative Pattern Clitic Pronoun System from an Earlier Ergative System In this section, data will be presented from Tongan and Samoan to demonstrate the development of an accusative clitic pronoun system from an earlier ergative system. In section 23.3.1, a shared proto-system of Ibaloy and Pendau was reconstructed as having developed from an ergative system as in (28).
574 Ritsuko Kikusawa
Figure 23.7 Changes from an ergative system that resulted in a bi-transitive system
(28) Shared proto-system of Ibaloy and Pendau (=(9)) a. Vi [nom]S b. Vi [nom]S [ni ind]E c. Vt[=gen]A [nom]P From this system, it was shown how an inverse system with two pronominal contrasts in Pendau and a bi-transitive system with a single-pronominal system in Balinese developed. In this section, it will be shown that the development of accusative pattern pronominal systems are also accounted for by assuming a similar proto-system, but with the following preconditions: (i) nominative pronouns expressing S were clitic pronouns; (ii) second-position, or Wackernagel clitics had acquired the pre-main verb position as their default position. This is illustrated by Ibaloy examples in (29). In (29a), the clitic pronoun =mo ‘2sg.gen’ expressing the A follows the main verb kalat-ən ‘to bite s.o.’ In (29b), where an auxiliary verb ʔəg occurs, the genitive pronoun expressing the A (the form =da in (29b)) “climbs up” to the second position, to follow the auxiliary. Example (29c) shows a case where both the genitive clitic expressing A (=to) and the nominative clitic expressing P (=ka) occurring in the second position. (29) Ibaloy sentence examples with auxiliary verbs a. nem kedatenmo ira nəm kalat-ən[=mo]A [ʔida]P but bite-PatV/ipf=2/gen 3+/nom ‘but you will bite them.’
(Ruffolo 2004: 179)
Ergativity and language change in Austronesian languages 575 b. egcha kedaten ʔəg[=da]A kalat-ən neg=3+/gen bite-PatV/ipf ‘they did not bite them.’
ira [ʔida]P 3+/nom
c. egtoka kegtinan ʔəg[=to]A[=ka]P gətin-an neg=3/gen=2/nom ipf-step-LocV/ipf ‘he will not step on you.’
(Ruffolo 2004: 178)
(Ruffolo 2004: 179)
It should also be noted that a nominative pronoun in Ibaloy can be replaced by an independent pronoun (as ʔida in (29a, b). In such a case, an independent pronoun “carries a more emphatic or contrastive meaning” (Ruffolo 2004: 174). An abstracted Ibaloy pronominal system illustrating constructions with an auxiliary verb is presented in (30). It should be noted that the pronominal forms that occur as clitic pronouns are limited to nominative and genitive. (30) Abstracted argument structures in Ibaloy with clitic/independent pronoun alternation Clitic pronouns Independent pronouns a. Vaux[=nom]S Vi e. Vaux Vi [ind]S b. Vaux[=nom]S Vi [son ind]E f. Vaux Vi [ind]S [son ind]E c. Vaux[=gen]A[=nom]P Vt d. Vaux[=gen]A Vt [nom]P g. Vaux[=gen]A Vt [ind]P Languages compared with this system in this subsection are Tongan and Samoan, which retain the earlier sentence structures with auxiliary verbs, the ones that are shown in (30). Tongan example sentences are presented in (31) and (32). Sentences in (31) show the clitic pronoun system. Here, S and A are expressed by the form ku, which precedes the main verb, while P is expressed by the form au following the main verb. Thus the clitic pronoun system shows an accusative pattern. (31)
Tongan sentence examples with clitic pronouns (Otsuka, Chapter 40, this volume) a. Na‘a [ku]S kata PST 1SG laugh ‘I laughed.’ b. Na‘a [ku]A ma‘u PST 1SG get ‘I caught a fish.’
[‘a ABS
e SPEC
ika]P fish
576 Ritsuko Kikusawa c. Na‘e taa‘i [au]P PST hit 1SG ‘John hit me.’
[‘e ERG
Sione]A John
Sentences in (32) present example sentences with independent pronouns, where pronominal arguments are underlined. When an argument is expressed with an independent pronoun, the pronoun is preceded by a case marking form, kiate ‘dative’ (32a), ‘e ‘ergative’ (32b, c), or ‘a ‘absolutive’ (32b). (32) Tongan sentence examples with independent pronoun arguments a. ‘E tokoni [‘a Sione]S [kiate koe]E fut help abs Sione to-person 2sg ‘Sione will help you.’ (Otsuka 2000: 258) b. Na‘e taa‘i [‘e pst hit erg ‘Sione hit you.’
Sione]A John
[‘a abs
c. Na‘e tala mai [‘e ia]A pst tell dir.1 erg 3.sg ‘He told me (that) it was correct.’
koe]P 2sg (Otsuka, Chapter 40, this volume) ‘oku tonu prs correct (Otsuka, Chapter 40, this volume)
Sentence structures with Tongan personal pronouns are presented in (33). Tongan personal pronouns occur in two different patterns: (i) a set of clitic pronouns marking both S and A, with an independent pronoun marking P of a transitive clause, and thus occurring in an accusative case-alignment pattern as in (31); (ii) independent pronouns receiving the same case marking as personal lexical NPs, showing an ergative pattern as in (32). According to Otsuka (Chapter 40, this volume), the use of independent pronouns in lexical NP slots is, like in Ibaloy, “marked and has an effect of emphasis.” (33) Tongan pronominal system Clitic pronouns a. Vaux [cltc]S Vi b. Vaux [cltc]S Vi … c. Vaux [cltc]A Vt [cltc]P
Independent pronouns e. Vaux Vi [ʔa ind]S f. Vaux Vi [ʔa ind]S [kiate ind]E g. Vaux Vt [ʔe ind]A [ʔa ind]P
The Tongan system as described is here compared with the Ibaloy sentence structures with clitic and independent pronouns, which are repeated in (34). (34) Abstracted argument structures in Ibaloy with clitic/independent pronoun alternation (=(30)) Clitic pronouns Independent pronouns a. Vaux[=nom]S Vi e. Vaux Vi [ind]S b. Vaux[=nom]S Vi [son ind]E f. Vaux Vi [ind]S [son ind]E
Ergativity and language change in Austronesian languages 577 c. Vaux[=gen]A[=nom]P Vt d. Vaux[=gen]A Vt [nom]P
g. Vaux[=gen]A
Vt [ind]P
Comparing the structures in (30) and (33), the following differences can be pointed out: (i) clitic pronouns expressing S and A form a single set in Tongan, while in Ibaloy, S is expressed with nominative and A with genitive; (ii) A can be expressed by an independent pronoun marked by ʔe in Tongan, while in Ibaloy, A is always expressed by a clitic pronoun and never by an independent pronoun, which alternates with a lexical NP; (iii) in Tongan, independent pronouns have to be morphologically case marked when occurring in an argument position. Accounting for the development of accusative pattern clitic pronoun sets (expressing S and A) in Oceanic languages, such as the one in Tongan, it has been hypothesized that the clitic pronouns showing an accusative pattern system are the result of a merger of the earlier genitive set (marking A) and nominative set (marking S) (Kikusawa 2002). The outline of this hypothesis is summarized in (35). (35) Development of a split pronominal system from an Ibaloy-type ergative pronominal system Clitic pronouns
Independent pronouns
a. Vaux[=nom]S
Vi
a. Vaux
Vi
[ind]S
b. Vaux[=nom]S
Vi [son ind]E
b. Vaux
Vi
[ind]S
c. Vaux[=gen]A
Vt [nom]P
c. Vaux [=gen]A
a. Vaux[=cltc]S
Vi
a. Vaux
Vi
[ind]S
b. Vaux[=cltc]S
Vi [son ind]E
b. Vaux
Vi
[ind]S
[son ind]E
c. Vaux[=cltc]S
Vt [nom]P
c. Vaux
Vt
[ind]A
[ind]P
[son ind]E Vt [ind]P
The original pronominal system is assumed to have been one similar to that in Ibaloy (and other Philippine-type and Formosan languages) with sentence structures containing auxiliary verbs. At this stage, clitic pronouns and independent pronouns must have alternated depending on various syntactic and pragmatic factors. When A was expressed by a pronoun, it had to be realized by a genitive clitic pronoun and no independent pronoun could be used. Starting from this system, a clitic pronoun set with a new function (covering both S and A) developed by merging the earlier nominative and genitive sets. The precondition of this change was probably the position of the second order clitic pronouns that became fixed in the pre-main-verb position. Then, S and A contrasted E and P by their relative position to the main verb, and because the S and A share the semantic role actor, while E and P share the semantic role undergoer, morphological marking was no longer necessary (Kikusawa 2003b, 2015).
578 Ritsuko Kikusawa It should be noted that merger is unidirectional and thus implies strong directionality. Observation of the forms of clitic pronouns supports this hypothesis. For example, Lynch, Ross and Crowley (2002: 68) reconstructed three sets of “subject proclitics” for Proto Oceanic, which are labeled as Set I, II, and III. Regarding the existence of so many sets for a single function, they state “[a]lthough subject proclitics (or prefixes) occur in many well distributed Oceanic languages and we can infer their presence in P[roto-] Oc[eanic], their forms vary considerably and a number of competing reconstructions can be made. We organize these into three sets … we see that Sets I and II respectively reflect the P[roto-]M[alayo-]P[olynesian] nominative and genitive clitics …” Meanwhile, genitive clitic pronouns expressing A started to alternate with the corresponding independent pronouns marked as lexical NPs in situ. The general change is from an ergative system to a system with accusative pattern clitic pronouns and independent pronouns marked in the same way as lexical NPs. The pronominal sets changed from a system with genitive clitics (expressing A), nominative clitics (expressing S and P), and independent pronouns (expressing S and P) to a system with a clitic pronoun set (expressing the actor), an independent set (expressing S, A, and P), and a genitive pronoun set, which no longer occur on verbs but only on nouns expressing possessors. In such a system, actual pronominal forms are often shared between different sets. The results in Samoan have been analyzed as an ergative language with an accusative pronominal pattern (Mosel and Hovdhaugen 1992), or an ergative language with an accusative “raising” pattern (Chung 1978), for example. A possible scenario with specific developmental stages to account for these changes is the following. First, some of the genitive pronouns that originally only expressed A in transitive constructions, were generalized also to express S in intransitive constructions. Genitive pronouns were then gradually replaced by corresponding forms from (an)other pronominal set(s). The end result of this sequence of changes varies depending on the language. In some languages, most if not all the clitic pronoun forms are the retention of forms in the original genitive set, while in some languages, clitic pronoun forms show a mixed etymology, and yet in other languages, all the forms have been replaced with independent pronominal forms as will be seen with Tongan and Samoan data. Table 23.4 presents the forms of Tongan pronouns. There is often more than a single form for each function. When we closely examine them, it can be seen that clitic and genitive pronouns share the same in common. This reflects the split of the earlier genitive set splitting to a clitic set occurring on the verb and a genitive set occurring on the noun. It can be also seen, however, that in non-singular pronouns, the longer genitive form and part of the corresponding independent form are identical. This appears to indicate the spread of the function of independent forms. The independent form spreading to cover the function of the other pronoun sets is more obvious in Samoan (Table 23.5). Like Tongan, Samoan personal pronouns are commonly described as consisting of three sets, namely, pre-main-verb pronouns
Ergativity and language change in Austronesian languages 579 Table 23.4 Tongan personal pronounsa Clitic 1sg
Genitive
Independent
ou, u ku
ku au
2sg
ke o, u koe
3sg
ne
ne no, na ia
1dl.incl
ta
ta taua
1dl.excl
ma
ma maua
2dl
mo na tau mau mou nau
(ki)mautolu
mou moutolu
3pl
(ki)tautolu
mau mautolu
2pl
(ki)naua
tau tautolu
1pl.excl
(ki)moua
na naua
1pl.incl
(ki)maua
mo moua
3dl
(ki)taua
(ki)moutolu
nau nautolu
(ki)nautolu
a. The longer possessive forms are used for emphasis.
(corresponding to “clitic pronouns”), independent pronouns, and possessive pronouns. Here, it can be seen that in some plural forms, a long form, or the earlier independent form, occurs to cover all the three functions. Like Tongan, Samoan shows a split system as shown in (36) followed by sentence examples in (37). Sentence (37a) exemplifies the intransitive structure (36b), while sentences (37b–d) exemplify the transitive structure (36)c.
580 Ritsuko Kikusawa Table 23.5 Samoan personal pronouns (based on Mosel and Hovdhaugen 1992: 121–124) 1sg.gnrl
preverbal
possessive
=ʔu
=ʔu
independent
ʔou, oʔu aʔu 1sg.em
ta
=ta, =tā taʔita
2sg
ʔe, e, ʔē, ē =u ʔoe
3sg
na
=na
ia 1dl.excl
tā
ia tā tāʔua
1dl.incl
mā
mā māʔua
2dl
lua
(ʔi)tāʔua
(ʔi)māʔua
lua =ulua
3dl
ʔoulua
ʔoulua
lā
lā
ʔoulua
lāʔua
(ʔi)lāʔua
1pl.excl
tātou
tātou
(ʔi)tātou
1pl.incl
mātou
mātou
(ʔi)mātou
2pl
tou
tou =utou
3pl
ʔoutou
ʔoutou
ʔoutou
lātou
lātou
(ʔi)lātou
(36) Samoan pronominal system Pre-main-verb pronouns a. Vaux [cltc]S Vi b. Vaux [cltc]S Vi … c. Vaux [cltc]A Vt …
Independent pronouns a. Vaux Vi b. Vaux Vi c. Vaux Vt
[ind]S [ind]S [ind]P
[(ʔ)i(ā) ind]E [e ind]A
Ergativity and language change in Austronesian languages 581 (37)
Samoan sentence examples with pronominal arguments a. olo [oʔu]S galue t 1sg work ‘I am working.’ b. ʔua [na]A fasi-a perf 3sg hit-tr ‘He hit the girl.’
[le the
teine]P girl
c. Na [ia]A opo-ina past 3sg hug-tr ‘He hugged the girl.’
[le the
teine]P girl
d. Na opo [e ia]A past hug erg 3sg ‘He hugged the girl.’
[le the
teine]P girl
(Clark 1976: 91)
(Mosel 1987: 461, my analysis)
(Cook 1991: 82, my analysis)
(Cook 1991: 82, my analysis)
The 1sg and 2sg forms in Samoan show similar patterns to those that are found in Tongan, with forms (ʔou, oʔu and ʔe, e) exclusively occurring in the clitic pronoun set, and others shared with the genitive set, or with both the genitive and independent sets. By comparing with Tongan, it can be seen that independent pronouns have extended their distribution to clitic pronoun position. What is particularly interesting in the given context is that the use of ia in a pre-main-verb position is apparently a recent innovation. The 3sg pre-main-verb pronouns =na and ia go back etymologically to earlier genitive forms (PAn *=ni-á, POc *=ña (ACD)) and nominative forms (PAn *=ia (ACD), POc *=ia (Ross 1988)), respectively. This is supported by the fact that in Tongan (a closely related language), the form ia is restricted to the independent set.13 According to Mosel (1987: 461), in Samoan, the 3sg clitic pronoun =na occurs only to express A, and reflects the original distribution of POc *=ña. In addition, she claims that the occurrence of the 3dl and 3pl forms is restricted to A of a transitive construction and only to the S of an intransitive construction that is an actor, but not to the S that is an undergoer.14 This appears to be a transitional stage prior to the point where the same set of clitic pronouns is used for A and all of the exponents of S. If the earlier genitive pronouns were in fact generalized from marking A to also mark S, as I claim, we would expect to see this reflected in languages where a cross-referencing system developed, and this is in fact what we find in Tetun (East Timor) and Taba (a language of southern Halmahera in North Maluku province of Indonesia). Taba, for example, has actor cross-referencing proclitics (showing an accusative pattern), the forms of four of 13
Note that in Samoan this 3rd person pronoun is the only one that is shared exclusively between the pre-main-verb and independent sets, and not with the genitive set. 14 Mosel and Hovdhaugen (1992: 697–698) comment, however, that they found “counter-examples for all proposed constraints” regarding the occurrence of third person pre-main-verb pronouns expressing S.
582 Ritsuko Kikusawa which (the three singular forms and 1pl.incl) are shared with the endings of the corresponding possessive ligatures (Table 23.6). Example sentences are presented in (38). Table 23.6 Taba agreement forms and possessive pronominal endings (Bowden 2001: 188–189)a Actor cross- referencing proclitics
possessive ligatures
independent pronouns
1sg
k=
nik
yak
2sg
m=
nim
au
3sg
n=
ni
i
1pl.incl
t=
nit
tit
1pl.excl
a=
amam
am
2pl
h=
memeu/mmeu
meu
3pl
l=
nidi
si
a. Whether the cross-referencing clitic forms for 1pl.excl and 2pl etymologically
relate to the corresponding possessive ligature and independent pronoun requires investigation. The form for 3pl appears to have developed by the same process as the other forms did, namely, retaining the part following ni of the possessive ligature.
(38) Taba sentence examples i. Mapot i be.heavy 3sg ‘It’s heavy.’
(Bowden 2001: 102)
ii. Yak kwom yak k=wom 1sg 1sg.cr=come ‘I’ve come.’
(Bowden 2001: 187)
iii. Yanti ncung um yanti n=sung um Yanti 3sg.cr=enter house ‘Yanti entered the house.’
(li)15 (li) (loc)
(Bowden 2001: 102)
15 I could not find a sentence example with all the arguments expressed with a pronoun for this construction. An example of a locative complement phrase expressed with a pronoun (yak li) is as follows:
Malusa nim wlo maduga m=ha-lusa nim wlo m=ha-duga 2sg=caus-say 2sg.poss liver 2sg=caus-only ‘You said your heart was only for me.’
yak yak 1sg
li. li loc (Bowden 2001: 323)
Ergativity and language change in Austronesian languages 583 iv. Am aamsi am a=am=si 1ex.pl 1ex.pl=see=3pl ‘We already saw them.’
do do real
(Bowden 2001: 35)
Cross-referencing marking is obligatory for the A and the S of “actor intransitive verbs,” while optional for the S of “undergoer intransitive verbs” (Bowden 2001: 118– 119, 188), thus showing a transitional stage where the verb agreement marking originally marked A is extending its distribution to cover part of the intransitive structure. Bowden states that although there are times when either actor intransitives or undergoer intransitives can be used to refer to the same events, “actor intransitives are more commonly associated with animate Actors and Undergoer intransitive more commonly associated with inanimates” (Bowden 2001: 199–200). Note that among Samoan clitic pronouns, reflexes of 1st and 2nd person pronouns occur expressing both S and A, while the occurrence of 3dl and 3pl is restricted to S that expresses an undergoer and A, and 3sg only to S. Both the Samoan and Taba data could be related to the semantic saliency hierarchy.
23.3.3 Ergative and Accusative Interpretation of Polynesian Sentence Structures Changes discussed in the previous two subsections both involved structural changes that resulted in a new interpretation of the case-alignment system. The change that is going to be briefly discussed in this subsection, is one that took place in the marking of lexical NPs in Polynesian languages, where it is believed that the earlier ergative system changed to an accusative system. However, this change does not involve changes in surface-structures. Instead, according to Otsuka (2011a), it is the combination of the distribution of certain verb forms and sentence types that resulted in the change in the interpretation of their sentence structures. It has been commonly known that languages that are analyzed as ergative (such as Tongan) and those as accusative (such as Māori) share the same set of sentence structures shown in (39).16 The judgment as to the pattern of the case-alignment system of these languages has depended on which of the dyadic sentence structures serves as the canonical transitive. More specifically, when the structures (39b) and (39c) are analyzed as the canonical transitive structure, the language is analyzed as accusative. On the other hand, when the structure (39d) is analyzed as the canonical transitive 16
Whether the Proto-Polynesian system was ergative and changed to accusative in some languages, or whether it was accusative and changed to ergative in some languages has produced extensive debate in Austronesian linguistics (see Ball 2007: 130–132 and Pucilowski 2006 for a summary).
584 Ritsuko Kikusawa structure, the system is analyzed as ergative. These are illustrated with specific examples in what follows. (39) Abstracted argument structures of lexical NPs in Polynesian languages a. Vaux Vi [NP]S(actor/undergoer) b. Vaux V [NP]actor [i NP]undergoer c. Vaux V [NP]actor [ki NP]undergoer d. Vaux V [e NP]actor [NP]undergoer Tongan and Samoan systems are analyzed as ergative, as shown in (40). In these languages, structure (39b) carries a partitive or “less affected” (Otsuka 2011a) reading, showing semantically lower transitivity. Thus, structure (40c) is analyzed as the canonical transitive and structure (40b) as an extended intransitive. This analysis is parallel to that of Ibaloy and other Philippine languages where the extended intransitive construction also carries partitive and less affected senses. It should be noted that the analysis is also parallel to that of Rotuman, an accusative language ((3) and (4)). (40) Abstracted argument structures of lexical NPs in Samoan a. Vaux Vi [NP]S b. Vaux Vie [NP]S [i/ʔi NP]E c. Vaux Vt [e NP]A [NP]P (41) Samoan sentence examples with lexical NPs a. ‘ua ‘ata [le tama]S perf laugh the boy ‘The boy laughed.’ b. Sa fa’afetai [le teine]S [i pst thank the girl obl ‘The girl thanked her mother’
lona her
c. ‘olo’o fafao [e le tama]A prog pack erg the boy ‘The boy is packing the banana-case.’
[le the
(Milner 1976: 26) tina]E mother pusafa’i]P banana-case
(Chung 1978: 217)
(Milner 1976: 59)
On the other hand, Māori and other Eastern Polynesian languages are commonly analyzed as accusative, and in such analyses, it is structure (39b). that is analyzed as the canonical transitive as shown in (42). According to Otsuka (2011a), even in the languages which have been analyzed as accusative, the i marked NP in a structure such as (42b) is less affected, just as in Tongan and Samoan, one of the facts that Gibson and Starosta used to claim that Māori is an ergative language (Gibson and Starosta 1990).
Ergativity and language change in Austronesian languages 585 They claim that if consistent criteria had been applied exclusively to transitivity and the case-marking of lexical NPs, it is likely that all Polynesian languages would be analyzed the same—showing an ergative system. (42) Abstracted argument structures of lexical NPs in Māori a. Vaux Vi [NP]S b. Vaux Vt [NP]A [i NP]P c. Vaux Vpass [NP]S [e NP]agent (43) Māori sentence examples with lexical NPs a. kua haere [a Hone]S tam go pers Hone ‘Hone has gone.’
(Pucilowski 2006: 13)
b. e kai ana [ngā tamariki]A [i ngā āporo]P tam eat tam the.pl child acc the.pl apple ‘the children are eating the apples’ (Bauer 1997: 40) c. i patua [te kurī]S [e tam hit the dog by ‘the dog was hit by the child’
te the
tamaiti]agent child
(Bauer 1997: 41)
The two systems presented in (40) and (42) are compared in Figure 23.8. What is it then that make Tongan and Samoan more readily analyzed as ergative and Māori and other Eastern Polynesian languages as accusative? Otsuka (2011a) argues that in Tongan and Samoan, verbs with highly transitive meanings do not occur (or are very limited) in sentences with the pattern shown in(39b) and (39c), but in (39d), while in Māori and other Eastern Polynesian languages, any verb could occur in structures (39b) and (39c). Thus, with proto-typical transitive verbs, some analysts find the (39d), the sole structure in which such verbs occur, as the canonical transitive, while (39b) and (39c) as derived from (39d). As a result, Tongan and Samoan are more likely to be analyzed as ergative. Otsuka demonstrates the details of the developmental paths of Māori and other Eastern Polynesian systems from the clearly ergative system found in Tongan and Samoan. This includes the generalization (extension) of the distribution of the *-Cia suffix (which I claim originated from the earlier canonical transitive structure but appears to occur in both dyadic sentences), and the affectedness alternation associated with the existence or non- existence of the suffix. Although I have a reservation in accepting her statement that “any dyadic verbs [sic] may occur in at least two types of constructions freely” in Philippine-type languages, the developmental paths that she proposes is in line with what is expected based on the cognacy of sentence structures.
586 Ritsuko Kikusawa
Figure 23.8 A comparison of ergative and accusative analyses
Ergativity and language change in Austronesian languages 587
23.4 Concluding Remarks In this chapter, I have shown three sets of changes each of which resulted in a change in the case-alignment system. The first case was a change in the alignment system which is claimed to have started with word order change. The resulting structures allowed the reading of both of the two dyadic sentence structures as transitive, and the system is no longer analyzed as ergative, but a “voice system,” an Austronesian linguistic term for languages with two transitive structures where the positions of the two core arguments are reversed. In the second case, a merger of earlier nominative pronouns marking S and genitive pronouns marking A resulted in a change from an ergative-to an accusative-pattern pronominal system. In the third case, proto-typical transitive verb forms extended their distribution from what was earlier a canonical transitive structure to dyadic intransitive structures. This resulted in the change in the interpretation of the case-alignment system. The comparison and reconstruction was mainly based on a macro-comparison between a Philippine-type ergative language and geographically distant Malayo-Polynesian languages. The generally assumed working hypothesis, that the Philippine-type language, Ibaloy, is structurally more conservative and that other languages belonging to lower branches in the family tree show innovations, is supported by the results obtained in this study. The changes referred to in this chapter all involve what appear to be a change in case- alignment from an ergative system to an accusative one. Whether languages in the higher-order subgroups of Austronesian exhibit an ergative system or a voice system is not an issue here, for whichever of the possible analyses one follows, a sequence of the changes by which other languages developed needs to be clarified and explained, and if the proposed hypothesis is correct, it should be translatable to the framework which yields the other analysis. The approach followed in this chapter is to examine case-marking systems and the verb systems separately, considering the independent and gradual nature of syntactic change. However, by combining the results of the comparison and reconstruction of the case system with that of the verb system, we should be able to capture how some 1,200 languages that are typologically so diverse developed in the Austronesian language family. Ball (2007: 140–142) lists a few Oceanic languages where a change from accusative to ergative appears to have taken place. However, each of these is obviously a relatively recent independent innovation as Ball claims, and they must have taken place subsequent to the general flow of ergative to accusative change that took place in Austronesian languages, after the Oceanic languages had developed new accusative systems. Such changes all involve developments in the marking of lexical NPs, and what is seen is the susceptibility of case marking on lexical NPs, sporadic or borrowed, to change or reanalysis contrasting with the relative stability of pronominal sets that were originally distinguished by case, in particular the genitive set. An anonymous reviewer questions the validity of determining structural cognacy by the occurrence or partial occurrence of the genitive pronouns associated with the actor, since the realis construction in Proto-Austronesian was historically a nominalization, while the irrealis paradigm is claimed to have a different source (Aldridge, Chapter 21,
588 Ritsuko Kikusawa this volume), so that genitive pronouns could be sourced to different structures. While the details of the reconstruction of Proto-Austronesian morphosyntax is still highly controversial and the subject of much ongoing research, this chapter has dealt mainly with case-alignment systems, and the syntactic reconstruction of patterns within Malayo-Polynesian. The typological characteristics that each language exhibits is a sum of the results of a large number of accumulated changes. By clarifying developmental paths of components consisting of languages, we come to learn explanations as to how existing systems, typologically common or unusual, developed. Thus, in the context of diachronic examination, the focus is not on whether a system changed from ergative to accusative, or vice versa, but how each of these changes contributed to the analyses that linguists propose.
Abbreviations -, affix boundary; /, may alternate with; [ ]A, expressing A; [ ]E, expressing E; [ ]P, expressing P; [ ]S, expressing S; =, (clitic boundary); A, agent of transitive verb; abs, absolute, absolutive; ActV, actor verb, Aux, auxiliary verb; av, av, actor voice; BnfV, beneficiary-oriented verb; cltc, clitic pronoun; dy, dynamic verb; E, extended argument of intransitive verbs; erg, ergative; gen, genitive; excl, exclusive; genr, general; incl, inclusive; ind, independent pronoun; ipf, imperfective aspect; iv, inverse voice; lk, linker; loc, locative; lv, locational voice; nom, nominative; NP, noun phrase; obl, oblique; ov, object voice; P, patient of transitive verbs; pass, passive; PatV, patient-oriented verb; pft, perfective; pfv, perfective; pl, plural; pnm, personal; pst, past tense; pot, potential; r, realis; rec, recognitional demonstrative; reciprocal marker; S, subject (actor/ undergoer) of intransitive verbs; sg, singular; spec, specific; ta, tense & aspect; tp, topicalized P; TpLk, topic linker; ts, topicalized S; Vaux, auxiliary verb; Vav, agent voice or actor verb; Vi, Vintr, intransitive verb; Viv, inverse voice verb; Vpv, patient verb; Vt, transitive verb.
Chapter 24
Lexical cat e g ory and alignm e nt i n Austron e sia n Daniel Kaufman
24.1 The Alignment of Philippine-Type Languages Philippine-type languages are often cited as exemplifying a cross-linguistically unique voice system, in which verb morphology can select not only an agent or patient, but also locative, instrumental, and other adjunct type relations as subject.1 Current syntactic treatments characterize this phenomenon alternatively as a rich voice system, a rich applicative system, “case agreement” or thematic nominalization. Relatedly, there is disagreement as to whether Philippine languages are best analyzed as ergative, accusative, active, or symmetrical. The uncomfortable position of Philippine languages in regard to more common alignment systems has been a long-standing topic in the typological literature (some earlier treatments include DeWolf 1988; Shibatani 1988; Himmelmann 1991).2 Here, I present a critical review of three current syntactic analyses of Tagalog: the ergative analysis (Gerdts 1988b; De Guzman 1988; Liao 2004; Aldridge 2004), the case-agreement approach (Richards 2000; Rackowski 2002; Rackowski 1
I use the term “Philippine-type language” here to refer to a subset of Austronesian languages spoken in the Philippines as well as northern Sulawesi and Borneo. Proto-Malayo-Polynesian, the ancestor of all Austronesian languages spoken outside of Taiwan is reconstructed as a Philippine-type language although some Austronesian languages of Taiwan may also be considered Philippine type (e.g. Amis, Paiwan, Seediq). See Himmelmann (2005) for discussion. 2 Richards (2013:fn.2), for instance, states, “I suspect that the debate about whether Tagalog is ergative will prove to be a terminological one; Tagalog resembles ergative languages in some respects, and differs from them in others, and the only question is how vague we want the technical term ‘ergative’ to be.”
590 Daniel Kaufman and Richards 2005) and a nominalism analysis (Starosta et al. 1982/2009; Kaufman 2009a,b), while advocating for the latter. I also present new data from Mamuju (South Sulawesi subgroup, Indonesia) to demonstrate that canonical ergative languages also exist within the Western Malayo-Polynesian subgroup of Austronesian. I argue that this retreat to the canonical ergative type is a result of the historical re-emergence of verbal predication. The typical Austronesian “voice” paradigm is exemplified in (1) and (2) with Tagalog.3 Note that the use of each morpheme correlates with the selection of a different argument or adjunct as the “ang phrase.” (1) a. Bili ng bulaklak ang bátà para um.beg-buy ng flower ang child for ‘The child bought a flower for the prisoner.’
sa sa
b. Bili-∅ ng bátà ang bulaklak para buy-in ng Juan ang flower for ‘A/the child bought the flower for the prisoner.’ c. I-bili ng bátà ng bulaklak ang i-give ng child ng flower ang ‘A/the child bought the prisoner a flower.’
b. I-bigay ng bátà sa bilanggô ang i-give ng child sa prisoner ang ‘A/the child gave the flower to the prisoner.’ 3
sa bilanggô sa prisoner bilanggô prisoner
d. Bil-han ng bátà ng bulaklak ang give-an ng child ng flower ang ‘A/the child bought a flower from the prisoner.’ (2) a. Nag-bigay ng bulaklak sa bilanggô mag.beg-give ng flower sa prisoner ‘The child gave a flower to the prisoner.’
bilanggô prisoner
bilanggô prisoner
ang ang
bátà child
bulaklak flower
The different verbal morphemes glossed act for actor voice/agreement, pat for patient voice/agreement, conv for conveyance voice/agreement and loc for locative voice/ agreement. The glosses should be self-explanatory except for the conveyance voice. The types of arguments and adjuncts selected by the conveyance voice do not seem to form a natural semantic class. They include instruments, benefactives, and themes. I adopt the term conveyance voice from Wolff (1973) based on its selection of objects that are conveyed away from the agent. It is also referred to as instrumental, benefactive and circumstantial voice in the literature. Note that the actor voice is indicated by in (i) but by nag-in (ii). The latter form is best analyzed as a combination of and pag-, a transitivity related prefix, but this will not concern us here. Regarding spelling conventions, the Tagalog genitive case marker ng is an abbreviation for /naŋ/. I follow the accentual conventions of (Wolff et al. 1991) where final stress is treated as the unmarked default and penultimate stress/length is marked with an acute accent. The grave accent is used to indicate word-final glottal stop. Thus, bátà indicates /ˈbaːtaʔ/.
Lexical category and alignment in Austronesian 591 c. Bigy-an ng bátà ng bulaklak ang give-an ng child ng flower ang ‘A/the child gave the prisoner a flower.’
bilanggô prisoner
The proper analysis of the alternating morphology on the predicates (1) and (2) (glossed neutrally as , nag-, i-, -an, etc.) as well as the markers introducing the arguments (ang, ng, sa) are at the crux of the debate on Philippine-type languages. As emphasized by Foley (2008), what makes the system atypical is its morphologically symmetrical nature; note that in each of the examples (1) and (2), one and only one voice marker appears.4 Following the typologically oriented work of Himmelmann (1991) and Foley (1998), I trace this symmetry to the root level, specifically, the lack of a clear contrast between entity- denoting and event-denoting roots. Crucially, when this contrast develops, as in many languages of Indonesia, more canonical alignment patterns emerge as well (Kaufman 2009b).5
24.1.1 The Ergative Analysis of Philippine Languages Under Aldridge’s (2004) ergative analysis of Tagalog, our example sentences in (1) would be analyzed as in (3), following the interpretation of the morphology presented in Table 24.1. (3)
a. Bili ng bulaklak ang bátà antipass-buy obl flower abs child ‘The child bought a flower the prisoner.’
para for
b. Bili-∅ ng bátà ang bulaklak buy-tr erg child abs flower ‘A/the child bought the flower for the prisoner.’
4
sa prep
para for
sa prep
bilanggô prisoner bilanggô prisoner
c. I-bili ng bátà ng bulaklak appl-give erg child obl flower ‘A/the child bought the prisoner a flower.’
ang abs
bilanggô prisoner
d. Bil-han ng bátà ng bulaklak give-appl erg child obl flower ‘A/the child bought a flower from the prisoner.’
ang abs
bilanggô prisoner
Note that the symmetry referred to here is primarily morphological. The different “voices” are clearly not interchangeable in the majority of cases as they correspond to different possibilities for the definiteness of arguments. The term “symmetrical” has also been used in the sense of (Bresnan and Moshi 1990) to refer to the treatment of ditransitive objects. This is also unrelated to the morphological symmetry discussed here. 5 In later historical developments, this ergativity has also given way to a more accusative syntax, as it does in Indonesian and Malagasy. See Aldridge (2008a) for a formal treatment of accusative features in Austronesian.
592 Daniel Kaufman The sentence in (3b) contains a canonical transitive clause with the zero allomorph of the transitive suffix -in.6 In this clause, the agent is marked with ergative case and the patient is marked with absolutive case. The sentence in (3a) must then be analyzed as an antipassive construction in which the agent is marked with absolutive and the patient is marked with oblique. In Tagalog, there is no formal difference between oblique and ergative case marking. Other Philippine languages, such as Ivatan (Reid 1966), distinguish these functions with separate case markers. Table 24.1 Ergative analysis of Tagalog morphology Argument marking
Predicate marking
ang ng1
absolutive ergative
ng2
oblique
sa
preposition
antipassive1
mag-
antipassive2
-in
transitive
-an
applicative1 (directional)
i-
applicative2 (instrumental, benefactive, etc.)
In support of the antipassive analysis of (3a), we can note the indefinite/non-specific interpretation of the object, a fact which has been commented upon in all descriptions and analyses of Tagalog (Bloomfield 1917; Schachter and Otanes 1982: 76; Wolff et al. 1991; Kroeger 1993; Maclachlan and Nakamura 1997: 310; Richards 2000; Rackowski 2002; Kaufman 2005; as well as older descriptive works). As pointed out by proponents of the ergative analysis, this is a typical property of antipassive patients and is seen clearly in numerous Inuit and Mayan languages. In Tagalog, the objects of putative antipassive verbs are robustly indefinite (except when the agent is extracted, see Adams and Manaster- Ramer 1988). For instance, in (4a) we see that a definite demonstrative is not felicitous (without a partitive reading) in the object position of an antipassive verb. There is no such constraint on the clause in (4b), with what is a canonical transitive verb on the ergative analysis. (4) a. *?Káin nito ang bátà eat obl.this abs child (For, ‘The child ate this.’ OK for ‘The child ate from this.’) 6 The zero allomorph of -in is used in conjunction with the aspectual infix . The infix also has a zero allomorph which is used in the prospective aspect. Because both of these voice markers have zero allomorphs in their aspectual paradigm, neither can be said to be less marked than the other.
Lexical category and alignment in Austronesian 593 b. Káin-∅ ng eat-tr erg ‘A/the child ate this.’
bátà child
ito abs.this
According to the criteria of case and transitivity, Tagalog, and the vast majority of Philippine languages, are clearly ergative. The problem, as noted by Foley (1998, 2008) and Himmelmann (1991, 2005) among others, is that there are few if any ergative languages outside the Austronesian family that use the same morphology for putative antipassives as well as monadic (underlyingly intransitive) predicates. In Philippine languages, not only is used in antipassive contexts as in (5a), it is used in underlyingly intransitive predicates as in (5b). The same problem is noted by Paul and Travis (2006: 321) for Malagasy man-and holds throughout Philippine-type languages. (5)
a. Patay kill ‘Galvan killed a goat.’ b. Lákad walk ‘Galvan walked.’
ng obl si abs
kambing goat
si abs
Galvan Galvan
Galvan Galvan
Furthermore, is used on meteorological verbs, which differ from canonical intransitives in disallowing an overt absolutive argument, as shown in (6). In this case then it seems that simply functions to create an event-denoting predicate rather than reducing valency or relating a particular thematic role to the ang phrase. (6) a. áraw sun ‘It was sunny.’
(*ang araw) abs sun
b. ulan rain ‘It rained.’
(*ang ulan) abs rain
c. Lindol earthquake ‘There was an earthquake.’
(*ang lindol) abs earthquake
The question of the Tagalog antipassive is discussed at length by Aldridge (2012b), who argues that antipassives are not derived but rather combine with intransitive verbal morphology. Aldridge (2012b: 198–200) claims that a demotion analysis of antipassives requires downward movement of the demoted object, a possibility that is excluded by Chomsky (1995, 2005). While this is easy to argue from a Philippine perspective, it
594 Daniel Kaufman remains to be explained why antipassive morphology only attaches to transitive verbs in other well-known ergative languages.7 The two other verbal alternations involving i-and -an are treated as applicatives on the ergative approach (Aldridge 2004). The first promotes benefactives, instrumentals and conveyance objects to absolutive while the latter promotes directional and locative arguments. There are two difficulties with treating these morphemes as applicatives. The first is that, unlike traditional applicatives, they are always required to introduce arguments that correspond to their functions, whether selected by the predicate or not. For example, a predicate like bigay ‘give’ requires applicatives to promote both the theme and the recipient to absolutive position, as shown in (7). (7) a. Bigy-an mo ng give-appl 2s.erg obl ‘Give the child a mango.’
mangga mango
b. I-bigay mo ang mango appl-give 2s.erg abs mangga ‘Give the mango to the child.’
ang abs
bátà child
sa prep
bátà child
This is typologically unusual, as one of the two alternations is generally treated as basic, but in Tagalog and other Philippine languages all such arguments are derived as there is no unmarked verb. The second difficulty is that we expect applicative morphology to co- occur with -in, which under the ergative analysis must be treated as a simple transitive marker. Yet no Philippine language employs combinations such as i-STEM-in or STEM- in-an for these putative applicative constructions. On the other hand, a strong piece of supporting evidence for the ergative view is the well- known restriction on extraction in Philippine and other Austronesian languages (as first discussed in a generative context by Keenan 1972). Ng-marked arguments in Tagalog cannot be topicalized, relativized, or clefted. This is demonstrated via ungrammatical topicalization of an ergative argument in (8) and an antipassive patient in (9). In both cases, topicalization of the corresponding absolutive argument is fully acceptable, as shown in the (a) sentences. (8) a. Ang bulaklak ay bili-∅ abs flower top buy-tr ‘The flower, the child bought.’ b. *Ng bátà ay bili-∅ ang erg child top buy-tr abs (For, ‘The child, bought the flower.’) 7
ng erg
bátà child
bulaklak flower
The Chol intransitive marker -i and transitive marker -V may, however, be relevant here. The -i suffix appears with underlying monadic predicates as well as derived intransitives, such as passives. Antipassives are formed with a transitive light verb and thus do not display -i. See Coon (2012) for discussion.
Lexical category and alignment in Austronesian 595 (9)
a. Ang bátà ay bili abs child top buy ‘The child, bought a flower.’ b. *Ng bulaklak ay bili obl flower top buy (For, ‘A flower, the child bought.’)
ng obl ang abs
bulaklak flower bátà child
Similar restrictions are attested in Mayan and Inuit languages among others. It is these facts, together with the reduced transitivity of “actor voice” verbs marked with and mag-, that form the strongest arguments for the ergativity analysis of Philippine- type languages. To review, the most basic arguments for the ergative view are shown in (10) and those against it, in (11). (10)
a. indefiniteness (and low scope) of “actor voice” objects b. restrictions on the extraction of “actor voice” objects and transitive agents c. case marking pattern (e.g. gen/erg syncretism) common to other ergative languages
(11)
a. absence of unmarked transitive and intransitive predicates b. complementary distribution of “transitive” -in, and “applicatives” i-conv and -an loc. c. absence of a true de-transitivizing antipassive
24.1.2 The Case Agreement Analysis of Philippine Languages Chung (1998), Richards (2000), Rackowski (2002), and Pearson (2001) present variations of an agreement approach to the Austronesian verbal alternations seen above in which the morphology instantiates agreement with an argument in an A-bar position (see also Erlewine, Levin, and van Urk, Chapter 16, this volume, for a similar approach employing an A-position). I restrict my focus here to the analyses of Richards (2000), Rackowski (2002) and Rackowski and Richards (2005) as the languages investigated by Chung (1998) and Pearson (2001) (Chamorro and Malagasy, respectively) differ from Tagalog in some important respects. In case agreement approaches, all arguments have their case checked in their respective A-positions after which object shift may occur. Object shift moves a DP to an outer specifier of vP above the merged position of the external argument. The verb then agrees with the highest argument, not for canonical agreement features, but rather for its case features. Rackowski (2002) proposes that complements of verbs receive accusative case and verbs register accusative agreement with -in (or its zero allomorph). External arguments receive nominative case from T with the corresponding verbal agreement being
596 Daniel Kaufman or mag-. Nominative agreement with the external argument only takes place when no other argument has moved to a higher position. This is the case when the direct object is indefinite and there are no applicatives present. On this analysis, Tagalog also has null high and low applicatives (cf. Pylkkänen 2002) which promote benefactives, instrumentals, locatives, and other types of arguments and adjuncts. The high applicative assigns oblique case and the verb displays oblique agreement with the prefix i-. The low applicative assigns dative case and the verb displays dative agreement with the suffix -an. The merged structure of a benefactive (high) applicative construction is shown in (12) and the result of subsequent movement in (13). The applicative argument ends up on the edge of vP and enters into an agreement relationship with the verb which is spelled out as i-. (12)
vP v’
DPEA v
ApplP DPBen
Appl’ Appl
VP V
(13)
DPDO
vP vP
DPBen i
v’
DPEA v
ApplP ti
Appl’ Appl
VP V
DPDO
This analysis is significantly more abstract than that of the ergative analysis, in which voice morphology is interpreted directly as applicatives and transitivity markers. Here, the actual applicatives are only detectable indirectly through distinct agreement on the predicate. Note also that the cases with which the predicate agrees are abstract. As seen earlier, there are only three phonologically distinct argument markers in Tagalog but for the case agreement analysis to hold, there must be distinct oblique and dative cases underlyingly that the verb agrees with. The interpretation of Tagalog’s core functional morphology under this analysis is shown in Table 24.2.
Lexical category and alignment in Austronesian 597 Table 24.2 Case agreement analysis of Tagalog morphology Argument marking
Predicate marking
ang ng
agreement trigger default case
sa
preposition
nominative agreement1
mag-
nominative agreement2
-in
accusative agreement
-an
dative agreement (directional)
i-
oblique agreement (instrumental, benefactive, etc.)
The ang marker, interpreted as absolutive case on the ergative analysis, is treated as a topic marker of sorts under case agreement analyses.8 The ng-case, which instantiates both ergative case and oblique case on the ergative analysis is simply glossed as case in the case-agreement analysis. It is not clear what kind of case it is and the details of its assignment remain to be explained. As with the ergative analysis, there are also difficult typological and theoretical questions that arise with ostensive applicatives in the case agreement approach. First of all, applicative objects must somehow retain their underlying (oblique and dative) case for case agreement to take place but in familiar nominative-accusative languages applicatives promote arguments to a position where they receive accusative case and behave as canonical direct objects.9 Second, it remains unclear why applicative objects must shift to the edge of vP and block agreement with the external argument. Without such movement, we expect to find structures like that in (14b), where the ang phrase
8
However, the use of “topic” cannot be taken too literally, as the ang phrase has been shown conclusively to not be an actual discourse topic (Adams and Manaster-Ramer 1988; Kroeger 1993; Kaufman 2005). For instance, in the following mini-dialogue, an ang phrase appears within the focus of a completely felicitous response. But see Richards (2000) for arguments that the ang phrase is similar to the topic position in Icelandic and other Germanic languages. (i) A: Ano ang gá∼gawà what ang imprf∼do ‘What is Jojo doing?’
ni case
B: Hú∼hugás-an niya imprf∼wash case.3s ‘He’s washing the dishes.’
ang ang
Jojo Jojo? mga pl
pinggan. dish
9 The literature on Bantu applicatives bears this out as well as studies of non-accusative languages such as those of the Salish family. Kiyosawa and Gerdts (2010: 41), for instance, shows that applicative objects across Salish languages are extracted in exactly the same manner as direct objects, not as obliques, which require nominalization of the clause.
598 Daniel Kaufman benefactive fails to trigger case agreement on the verb. As pointed out by Rackowski and Richards (2005), such promotion without further movement is ungrammatical. They offer potential analogues from other syntactic domains but little if any cross-linguistic support exists for applicative objects blocking agreement with an external argument in nominative-accusative languages.10 (14) a. Ni-lútò-Ø ni Romeo ang adóbo para sa babáe beg-cook-Acc case Romeo ang adobo for prep woman ‘Romeo cooked the adobo for the woman.’ (Rackowski and Richards 2005: ex.13a) b. *Ni-lútò-Ø ni Romeo ng babáe ang adóbo beg-cook-Acc case Romeo case woman ang adobo ‘Romeo cooked the adobo for a woman.’ (Rackowski and Richards 2005: ex.11b) The theory makes crucial use of Pylkkanen’s (2002) distinction between low and high applicatives but the Tagalog facts do not accord well with the theory’s prediction. Low applicatives relate a promoted argument to a direct object and are crucially dependent on the presence of a direct object. Rackowski (2002) argues that this is the case based on the data in (15). (15)
a. Bigy-an ko ang give-dat 1s ang ‘I gave the father his child.’ b. *Bigy-an ko ang give-dat 1s ang ‘I gave (to) the father.’
ama father
ng case
anak. child
ama. father
However, the judgment reported for in (15b) seems better attributed to pragmatic rather than syntactic factors. There can be no act of giving without a theme and there is thus an expectation that the theme will be included in an out-of-the-blue context. Similar examples could be produced without a theme but more conclusive is the data shown in (16), which demonstrates that monadic predicates of all types can take -an, counter to the expectations of a low applicative in Pylkkanen’s framework.
10
An analogy is made to the wager-class of verbs in English, in which an argument must move from its underlying position with no obvious motivation. Collins (2002) and Epstein and Seely (2006:83–5) question the judgments surrounding this phenomenon in English and it remains to be seen if the proposed link can be further substantiated. (i) a. *John wagered Mary to have won the race. b. Maryi was wagered ti to have won the race.
Lexical category and alignment in Austronesian 599 (16)
a. Ni-lakár-an ng bátà beg-walk-dat case child ‘The child walked the road.’
ang ang
daan road
b. Sigaw-an ni shout-dat case ‘Romeo shouted at Jojo.’
Romeo Romeo
si ang
Jojo Jojo
c. ubu-han ni cough-dat case ‘Romeo coughed at Jojo.’
Romeo Romeo
si ang
Jojo Jojo
d. Na-ulan-an si sta.beg-rain-dat ang ‘Jojo was rained on.’
Jojo Jojo
The role of specificity shift is crucial in both the ergative and case agreement approaches to Tagalog. Both predict, in their own ways, that a definite/specific object should not be possible for a matrix clause verb marked with or mag-. On the ergative analysis, these verbs are antipassives whose objects are assigned inherent oblique case from v and remain in VP. In VP, they are subject to existential closure (Diesing 1992) and receive an existential indefinite interpretation. In the case agreement analysis, specificity triggers object shift, which puts a shifted object above the external argument and thereby triggers accusative agreement rather than nominative agreement with or mag-. Neither approach, however, seems to predict all the specificity facts of applicative constructions correctly. Rackowski (2002) and Rackowski and Richards (2005) explicitly claim that the underlying direct object in applicative constructions can be specific, offering the example in (17) as evidence. (17)
I-pinaglútò ni Romeo ng adóbo ang babáe obl-cook case Romeo case adobo ang woman ‘Romeo cooked (the) adobo for the woman.’ (Rackowski and Richards 2005: 570)
However, there is no independent evidence for the definite interpretation of the object in structures like (17). Nothing in (17) forces a specific interpretation and, in fact, evidence points to definite direct objects being just as marked in structures such as (17) as they are as object of verbs marked with or mag-. This becomes clear by using pronominal arguments as diagnostics. As Rackowski and Richards (2005) show with (18), a pronominal object of an marked verb is ungrammatical. Unexpectedly, the same ungrammaticality appears with direct objects of -an and i-marked verbs, as in (19). (18)
*Sampal ko ang mandurukot. slap 1s.case ang pickpocket (For, ‘The pickpocket slapped me.’) (Rackowski and Richards 2005: 568, ex.4b)
600 Daniel Kaufman (19) Bigy-an ni Maria (ng pera cook-dat case Maria case money Juliette Juliette ‘Maria gave money/*Romeo to Juliette.’
/*ni Romeo) si / case Romeo ang
In a regular matrix clause with a definite object and an oblique, the object must surface with ang case and the oblique must surface as an oblique/prepositional object, as shown in (20). (20) I-bigay ni Maria obl-cook case Maria ‘Maria gave Romeo to Juliette.’
si ang
Romeo Romeo
kay obl
Juliette Juliette
Rackowski and Richards (2005: 568) follow Chomsky (2001) in positing that: object shift occurs as the result of an EPP-feature on v that is present only when it has an effect on semantic outcome. There is an effect on semantic outcome because the position at the edge of the vP is assigned a specific interpretation, while everything internal to vP is assigned a nonspecific interpretation.
But there is an issue of derivational look-ahead in this account as well as an empirical problem. As ng-marked objects cannot be specific/definite in a regular matrix clauses, we must assume that they are barred from leaving vP. This means that the presence of an applicative object must arbitrarily preclude an EPP-feature on v on this account. Barring any larger generalization which can be extracted from this coincidence, it merely describes the facts. Regarding ng-marked agents, Rackowski & Richards recognize the need to account for both an existential indefinite as well as a specific/definite reading for them. As seen in (21), a ng-marked agent can felicitously be a proper name or an indefinite pronoun. (21) Hindi bigy-an (ni neg give-dat case pera money ‘Jose/nobody gave Maria money.’
Jose/ninuman) si Jose/case.anyone ang
Maria Maria
ng case
The reasoning for this ambiguity is that only internal arguments have the opportunity to undergo object shift. For arguments that do not have such an opportunity, specificity is unpredictable. But as we have seen in (19), this empirical claim cannot be upheld. A separate treatment is required for ng-marked objects, which cannot be definite/specific, and ng-marked agents, which can. This partially supports the ergative analysis in that ergative arguments are not known to show the kind of definiteness effects attested for antipassive patients. However, the restricted interpretation of direct objects of -an and
Lexical category and alignment in Austronesian 601 i-verbs is equally problematic. On the ergative analysis, non-specific readings are only forced for antipassive patients but something extra must be said for -an and i- verbs, which can only be analyzed as transitive. We find that the case-agreement analysis captures several important truths about Philippine voice systems, enumerated in (22). On the other hand, there remain theoretical issues with the analysis, some of which are summarized in (23). (22) a. no unmarked verbs, (virtually) all inflected verbs show “voice morphology” b. “voice morphology” is only marked once per verb c. range of meanings associated with “voice morphology” resembles that of case marking (23) a. incorrect predictions regarding definiteness/specificity b. unusual properties of (null) applicatives c. unclear what morphological case is and how it is assigned
24.1.3 The Nominalization Analysis of Philippine Languages Yet another interpretation of Philippine-type verbal alternations views them as participant nominalizations. This was first proposed by Starosta et al. (1982/2009) on a historical basis and further developed by Ross (2002, 2009) and, in synchronic terms, by Kaufman (2009a).11 On this approach, each “voice” indicates a different participant nominalization, as shown in Table 24.3.12 The three argument markers are interpreted as case, as in traditional descriptions. The ang marker indicates subject case, which can be termed either nominative or absolutive. The ng-case is first and foremost genitive case, that is, the case of possessors, but is recruited for other purposes due to the regular use of nominalized predicates. Finally, sa is an oblique case marker (and not a preposition as in the two approaches reviewed above).13 11
This also builds on an important body of work beginning with Bloomfield’s (1917) Tagalog grammar and furthered by Capell (1964), Naylor (1980), Himmelmann (1987, 1991, 2008), DeWolf (1988), and Gil (1993, 1995, 2000) who propose precategorial analyses for Tagalog, or, in the case of Himmelmann (2008), a distinction that cross-cuts traditional categories. The technical side of the analysis has a close analogue in Johns’ (1992) nominalization-based approach to Inuktitut. See also Coon (2014) for a similarly v-less analysis of partially overlapping facts in the morphosyntax of Chol, a Mayan language. 12 For reasons of space, I do not discuss the difference between and mag-although I treat mag-as containing an inner causative as suggested by Travis (2000). This is supported in Kaufman 2009c where I show comparative evidence for mag-being composed historically, and potentially synchronically, of the morphemes pa-caus-. 13 For all locality based approaches to the extraction restriction, including case agreement and the ergative analysis, there must in fact only be two cases. The oblique case, marked by sa, is treated as a preposition by these approaches because sa phrases can typically be fronted and topicalized regardless of voice. In other words, oblique phrases are effectively invisible for calculations of locality under previous approaches. But there are many reasons to believe that sa is a case marker rather than a preposition. It is in complementary distribution with the other case markers and is selected by bona fide prepositions, e.g.
602 Daniel Kaufman Table 24.3 Nominalist analysis of Tagalog morphology Argument marking
Predicate marking
ang
nominative/absolutive
ng
genitive
sa
oblique
actor nominalization
mag-
inner causative + actor nominalization
-in
patient nominalization
-an
locative nominalization
i-
circumstantial nominalization
This view differs from the previous two approaches in treating all predication in Philippine languages as inherently copular. The interpretation of the ang phrase is not derived through the use of applicatives but rather through copular identification with the predicate itself. To compare with Johns’ (1992) analysis of Inuktitut, she proposes that predications are formed compositionally in the manner shown in (24). (24) a. kapi-jaq stab-pass.part ‘the stabbed one’ b. anguti-up kapi-ja-a man-erg stab-pass.part-3s/3s ‘the man’s stabbed one’ OR ‘the one that the man stabbed’ c. anguti-up nanuq kapi-ja-a man-erg polar bear.abs stab-pass.part-3s/3s ‘The polar bear is the man’s stabbed one.’ OR ‘The man stabbed the polar bear.’ (Johns 1992) Applying this approach to Tagalog, we arrive at the following literal translations for the basic “voice” alternations: (25) a. Káin ng dagà eat gen rat ‘The cat was the eater of a rat.’
ang nom
púsà cat
gáling ‘from’, patúngo ‘towards.’ The oblique also displays a human/non-human distinction (sa for non- humans vs. kay for humans) which is a unique feature of the two other case markers and not found on any of the bona fide prepositions. These facts alone should dispel any notions that the oblique could be a preposition but see Kaufman (2009a: 40) and Gerassimova and Sells (2008: 196–197) for further arguments.
Lexical category and alignment in Austronesian 603 b. Káin-∅ ng púsà ang eat-pat gen cat nom ‘The rat was the eaten one of the cat.’
dagà rat
c. Kaín-an ng púsà ng dagà ang pinggan eat-loc gen cat gen rat nom plate ‘The plate was the cat’s eating place of the rat.’ d. I-káin ng púsà ng dagà ang conv-eat gen cat gen rat nom ‘The dog was the cat’s “eating benefactor” of the rat.’
áso dog (Kaufman 2009a: 6)
This avoids the difficulties of theta-linking approaches discussed by Rackowski (2002) since nominalization, like case, is well known to reflect thematic roles imprecisely. More importantly, nominalization also makes sense of the fact that Philippine predicate morphology does not distinguish between argument and adjunct. Barker (1998: 714) discusses at length the way in which English -ee nominalizations select participants that are not part of the argument structure of the corresponding verb.14 This offers an excellent analogue to one of the more typologically difficult aspects of Philippine-type voice. As argued in Kaufman (2009a), the nominal nature of event-denoting predicates in Tagalog can also go a long way in explaining other curiosities of Philippine syntax, including: (26) a. Near identical syntactic distribution of canonical event-denoting predicates (i.e. “verbs”) and entity-denoting predicates (i.e. “nouns”). b. Identity between the case of ergative agents and possessors across Philippine languages. c. Ungrammaticality of extracting genitive marked arguments corresponds with the difficulties of extracting possessors and other nominal dependents cross-linguistically. d. Symmetric nature of Philippine “voice” –only one instance per predicate, as would be expected of nominalization morphology. 14 Barker utilizes the notion of “episodic linking” rather than argument structure to account for how participant nominalizations in English identify their referent.
… the meaning of the verb amputate guarantees the existence of a person undergoing amputation, even though there is no syntactic argument that corresponds to this participant… the fact that the person undergoing amputation is a participant of every amputation event is sufficient to enable a set of amputation events to characterize the -ee noun amputee: for each amputation event e, there exists an individual x which is a participant in e such that x is (becomes) an amputee. Thus amputee is episodically linked to the meaning of amputate despite the fact that there is no corresponding syntactic argument position. Similarly, Aronoff (1980) discusses the role of Gricean principles in certain morphosyntactic alternations.
604 Daniel Kaufman Event-denoting predicates in Philippine-type languages can always serve as the direct complement to case markers.15 There is no strong evidence from Philippine languages for null complementizers or headless relatives in (27b) and (c).16 (27) a. Ang áso nom dog ‘the dog’ b. Ang lú∼lutú-in nom imprf∼cook-pat ‘the thing to be cooked’ c. Ang mag lú∼lutò nom act-imprf∼cook ‘the one who will cook’ The syntax of ng-marked possessors and ng-marked agents appears identical. That is, there is no reason to believe that the addition of the morphology in (28b) leads to a substantial difference in syntactic structure between (28a) and (b). (28) a. Súlat ni Juan write gen Juan ‘That is Juan’s letter.’ b. Súlat-∅ write-pat ‘Juan wrote that.’
ni gen
iyan that:nom Juan Juan
iyan that:nom
Johns (1992) makes the same claim for Inuktitut pairs such as those in (29). Inuktitut ja in (29b) functions similarly to the Tagalog patient nominalizer -in/∅. In both cases, a NP or DP predicate combines with a structural subject through a copular structure to yield a predication such as (28b).17 (29) a. anguti-up qimmi-a man-gen dog-3s/3s ‘the man’s dog’
15
Kaufman (2009a: 25) points out an important exception to this pattern with the descendants of the original (non-nominalized) verbs in certain dialects of Tagalog. 16 But see Richards (2009) for some arguments to this effect and Kaufman (forthcoming) for a defense of the symmetry. 17 This is similar to Pearson’s 2005 proposal of base generating the equivalent of the ang-phrase in Malagasy in an A-bar position located in a high rightwards specifier and co-indexed with a lower null operator in an A-position.
Lexical category and alignment in Austronesian 605 b. anguti-up kapi-ja-a man-gen stab-pass.part-3s/3s ‘the man’s stabbed one’ OR ‘the one that the man stabbed’ There are two potentially independent conditions conspiring against fronting genitive arguments. On one hand, extracting ergatives should bear similarity to sub-extraction from NP in other languages, a highly constrained operation. Indeed, in Tagalog itself extraction of possessors is subject to precisely the same constraints as ergative extraction as shown by the illicit topicalizations in (30).18 (30) a. Bili-∅ ni Maria buy-pat gen Maria ‘Maria bought Juan’s car.’
ang nom
kótse car
b. *Ni Juan ay bili-∅ gen Juan top buy-pat (For, ‘Juan, Maria bought (his) car.’)
ni gen
ni Maria GEN Maria
Juan. Juan ang kótse ____. nom car
c. *Ni Maria ay bili-∅ ____ ang kótse gen Maria top buy-pat nom car (For, ‘Maria, (she) bought Juan’s car.’)
ni Juan. gen Juan
18
Richards cites Cena’s (1979) examples in (i) of apparent possessor extraction as an argument against constraints on linking ergative extraction to possessor extraction. (i) a. Kasama ng doktor companion gen doctor ‘The child is with the doctor’
ang nom
anak child
b. ang doktor [na kasama nom doctor lnk companion ‘the doctor that the child is with’
ang nom
anak] child (Cena 1979; Richards 2009)
As shown in (30), possessor extraction of the normal type is subject to exactly the same constraints as ergative extraction. A minimal pair for Cena’s example with an ostensibly verbal predicate, shown in (ii), demonstrates clearly that relativization of the ergative argument is no more marked than relativization of the possessor. I leave the analysis of this construction to further work noting only that Cena’s examples do not entail different treatments for possessors and ergative arguments. (ii) a. sa∼sama-han ng imprf∼accompany-loc gen ‘the doctor accompanies his/her child’
doktor doctor
ang nom
b. ang doktor [na sa∼sama-han nom doctor lnk imprf∼accompany-loc ‘the doctor that accompanies his/her child’
anak child ang nom
anak] child
606 Daniel Kaufman Potentially separate from the issue of extraction from NP is the general ban on genitive predicates. This is independently necessary in Tagalog as shown in (31) and has clear cognates cross-linguistically. Genitive phrases can be modifiers, as in (31a) but not predicates, as shown in (31b). As in Hungarian and many other languages (Szabolcsi 1983), predicate and extracted possessors must be expressed as obliques or datives, as in (32). (31) a. Ang súlat ni nom write gen ‘The letter of Juan’s’
Juan Juan
b. *Ni Juan ang súlat gen Juan nom write (For, ‘The letter is Juan’s.’) (32) Kay Juan ang obl Juan nom ‘The letter is Juan’s.’
súlat write
Argument questions in Philippine-type languages have been widely analyzed with the interrogative phrase as the predicate of the clause (Paul 2001; Aldridge 2002; Oda 2002; Massam 2003; Potsdam 2006, 2009; Gerassimova and Sells 2008). As we have already seen that genitive predicates are banned, we also correctly predict that genitive case interrogatives should be ruled out, as seen in (33). Note that the genitive interrogative is allowable in an in-situ post-nominal position, as shown in (34) (see also Erlewine, Levin, and van Urk, this volume, Chapter 16, n.21). (33) a. Kaníno ang súlat? obl.who nom write ‘Whose is the letter?’ b. *Nino ang súlat? gen.who nom write (For, ‘Whose is the letter?’) (34) Ang súlat nino?!? nom write gen.who ‘The letter of who?!?’ Similarly, if event-denoting predicates in Philippine-type languages are nominalized, we can rule out interrogatives like those in (35b) and (c), which attempt to make a predicate out of a genitive agent. The ungrammaticality in (35b) ensues from the constraint against genitive predicates and (35c) is ruled out on the simple basis of case preservation. What should be a genitive marked agent, as in (35a), cannot be expressed as an interrogative in the nominative (ang) case. A copular predication has at most two nominative marked
Lexical category and alignment in Austronesian 607 arguments, one in the predicate position and one in the subject position. The nominal approach to Philippine-type languages is thus uniquely able to unify the ungrammaticality behind (34) and (35), which is common to the vast majority of Philippine-type languages. (35)
a. Sú∼sulát-in ni imprf∼write-pat gen ‘Jojo will write the book.’
Jojo Jojo
ang nom
b. *Nino ang sú∼sulát-in gen.who nom imprf∼write-pat (For, ‘Who will write the book?’) c. *Sino ang sú∼sulát-in nom.who nom imprf∼write-pat (For, ‘Who will write the book?’)
libro book ang nom
ang nom
libro? book libro? book
On this approach, genitive marked arguments in Philippine-type languages bear a certain resemblance to English of-phrases. Just like ng-phrases, of -phrases are highly restricted in predicate position (36), and also display restrictions on their extraction from NP, as shown in (37).19 (36) The team is (*of) Juan’s. (37)
a. Juan was an employee of Rizal. b. *?Of whom was Juan an employee?
In (38), we see that these constraints do not hold for instrumental agents of passive verbs. (38)
a. Juan was employed by Rizal. b. By whom was Juan employed?
As pointed out in Kaufman 2009a, this could hold the key to why some ergative languages show a restriction on extracting the ergative argument while others do not (see Dixon 1994 and Manning 1996 for discussion). If the ergative argument is treated as a nominal dependent, we expect it to share a morphological case with possessors and resist extraction. If the ergative argument is a verbal dependent, we
19 See Davies and Dubinsky (2003) for a review of approaches to extraction from NP. Davies and Dubinsky rely on participant structure to derive the complex pattern of extractability from English NPs but they do not consider extraction from participant nominalizations. Other approaches have implicated the left branch condition (Ross 1967), the ECP (Chomsky 1981), the case filter (Huang 1982), movement of non-constituents (Bošković 2005), and most recently, the treatment of nP or DP as a strong phase (Chomsky 2001; Svenonius 2004).
608 Daniel Kaufman expect it to share a morphological case with instrumentals or obliques and to allow extraction more freely. In particular, it seems that a genitive-ergative syncretism in combination with a pseudo-cleft strategy for interrogatives is what conspires to constrain extraction in ergative languages. In languages with the same syncretism but in-situ interrogatives, as Inuktitut, constraints against ergative interrogatives are not attested, as evidenced by (39). Indeed, even in the rare contexts that Tagalog allows wh-in-situ, as in (40), genitive interrogatives become acceptable (compare above). (39) Kia Alaana who.gen Alana.nom ‘Who bit Alana?’
kii-ja-nga? bite-pass.part-3s/3s
(40) Gawà-∅ nino make-pat gen.who ‘Who made those shoes?’
ang nom
sapatos shoe
(Yuan 2013) na iyon? lnk that (Schachter and Otanes 1982: 512)
The typological correlation is worth investigating on a larger scale but here we are most concerned with what happens when a bona fide v category develops from n and its consequences for ergativity. The crucial properties of the v versus n heads which determine lexical category are the following. (41) n properties: a. possessor is projected [Spec, n] b. genitive case to nP-internal phrases c. strong island properties (42) v properties: a. agent is projected in [Spec, v] b. accusative case to object c. islandhood dependent on v features The v category projects an agent and more generally, v is associated argument structure that is arguably not present in low nominalizations. While apparent arguments can nonetheless be expressed in low nominalizations, they are not distinguished by a distinct object case nor are they obligatory. The general assignment of genitive case within nP can be seen in (43), where the agent, instrument and a manner adverbial are all marked with ng-case. (43)
nP [Buks-an
ko ng súsi ng maingay] open-loc 1s.gen gen key gen loud ‘I opened the door loudly with the key.’
ang nom
pintuan door
Lexical category and alignment in Austronesian 609 Note also that the ergative argument is not obligatory nor must it be interpreted as pro when omitted in line with nominalizations cross-linguistically (Abney 1987; Himmelmann 1991: 22; Alexiadou 2001). The sentence in (44), for instance, can be uttered in an out-of- the-blue context without any implication regarding an agent. Optionality is expected for a possessor but not for the external argument of a transitive verbal projection.20 (44) Mukhang kain-∅ ang lahat seem:lnk eat-pat nom all ‘It seems that everything has been eaten.’ Relatedly, there are also patient voice predicates that function as adversatives and cannot take agents of any sort. Two such examples are shown in (45) and (46).21 (45)
Ni-langgam-∅ ang asúkal beg-ant-pat nom sugar ‘The sugar was infested with ants.’
(46) In-ú∼ubo-∅ ang bátà beg-cough-pat nom child ‘The child is (effected by) coughing.’ As discussed earlier, the full range of definiteness/specificity effects have presented difficulties to previous treatments of Tagalog. In the simplest case, which all theories have an account for, actor voice patients tend to be non-specific and strictly disallow pronominals, as seen in (47). More difficult to account for is the parallel constraint on themes of locative and circumstantial voice predicates, as seen earlier in (19). (47) a. Nag-pi∼pinta act.beg-imprf∼paint ‘John paints faces.’ b. *Nag-pi∼pinta act.beg-imprf∼paint (For, ‘John paints me.’)
ng gen ko 1s.gen
mukha face
si nom
si nom
Juan Juan
Juan Juan
20 As Lawrence Reid points out (p.c.), this is not the normal state of affairs in the Cordilleran languages, spoken in North Luzon, Philippines. In many of these languages it seems that an ergative agent is obligatory with transitive predicates. I take this property to be a later historical development based on its narrow distribution but nonetheless very important with regard to the viability of a nominal analysis for the Cordilleran subgroup. 21 We can again draw a parallel to English patient nominalizations with -ee in that they appear to operate outside of argument structure (Barker 1998), attaching to monadic bases (standee, escapee) as well as nominal bases (hoaxee).
610 Daniel Kaufman The nominalization approach suggests a parallel between these facts and similar constraints on definite of-phrase objects of nominals in English, as shown in (48).22 (48) a. John is a painter of portraits b. *John is a painter of me In English, this effect extends to themes of recipient nominalizations, as shown in (49). While (49a) shows that a definite theme is possible, a pronominal theme is completely unacceptable. Definite genitive case themes are thus possible in both English and Tagalog (under the right circumstances), but both languages strictly disallow pronominals in this position. A formalization of these facts will not be offered here but these parallels are clearly promising. (49) a. John was the awardee of the nobel prize. b. *John was the awardee of it. In regard to the semantics of the nominative/absolutive phrase, it seems far more felicitous to attribute its referential properties to the case marker itself as an operator rather than strictly to object shift (as originally advocated by Himmelmann 1991). Out-of-the- blue exclamations such as those in (50) support such an approach. It would make little sense to derive the definiteness of (50b) via object shift or agreement with T. On the analysis suggested here, ang contains both case and definiteness features.23 (50) a. Dagà! rat ‘A rat!’ b. Ang pangúlo! ang president ‘The president!’ In the next subsection, we look at root level phenomena to demonstrate that much of the syntax attributed to the voice/agreement system is already present at the first phase of word building, further suggesting the absence of v in the functional inventory. 22
Note that John is a painter of mine is acceptable on the possessive reading where I have a painter. According to several speakers I have consulted, this reading is also acceptable for the Tagalog constructions as in (47b) although I have not found naturally occurring examples of an aspect-inflected actor voice predicate with a possessor. 23 This approach is further supported by other Central Philippine languages which distinguish [±specific] variants for each case marker (see McFarland 1974 and Zorc 1977 for examples). Of course, an analysis in which ang functions as a specific/definite determiner is not necessarily incompatible with object shift. As Edith Aldridge points out (p.c.), Austronesian languages which still show the same specificity/definiteness pattern without overt case markers suggest that a structural approach might still be necessary.
Lexical category and alignment in Austronesian 611
24.1.4 Ergativity on the root level Foley (1998) argues that Tagalog roots are fundamentally different from those of English: The lexeme give in English is a verb, with a corresponding argument structure . The Tagalog root bigay ‘give’ however is precategorial; it lacks a true argument structure, but does have a precategorial semantic structure like ‘the giving of something to someone by someone.’
On this view, it is the voice morphology which imbues words with argument structure rather than anything in the underlying semantic representation. This explains why notional valency has little consequence for the voice marking potential of roots in Philippine-type languages. There are, however, problems with the claim that roots lack category and argument structure altogether. In contrast to early claims made about Tagalog roots, they appear regularly in their bare form as both predicates and arguments. Unlike the predictions of the precategorial analysis, bare roots are restricted to particular readings and do not display flexibility in their distribution of thematic roles to genitive and nominative marked arguments. As discussed by Himmelmann (1991: 40) and Kaufman (2009a), the reading of a bare root is consistently that of a patient-oriented (or “proto-object”) entity. In Table 24.4, we find a list of roots and their glosses in the first two columns. In the following columns we see the event denoting predicates formed from these roots through the use of voice morphology. Table 24.4 Tagalog root meanings Root
Gloss
Actor voice form
Gloss
WALK EAT
lákad káin
‘a walk, an errand’ ‘eating, a meal’
lákad káin
‘to walk’ ‘to eat’
THINK
ísip
‘thought, thinking’
mag-isip
‘to think’
KILL
patay
‘corpse’
patay
‘to kill’
BREAK
básag
‘a break’
básag
‘to break’
TEACH
túrò
‘lesson, teaching’
mag-túrò
‘to teach’
SAY
sábi
‘what is said’
mag-sábi
‘to say’
BUY
bili
‘price bought for’
bili
‘to buy’
TAKE
kúha
‘taken object’
kúha
‘to take’
Most surprising is the interpretation of stems with the causative prefix, as exemplified in Table 24.5 (adapted from Schachter 1976: 105). Here, we find that even when embedded under a causative head, these roots maintain their patient- oriented interpretation.
612 Daniel Kaufman Table 24.5 Tagalog causative meanings Causative stem
Gloss
CAUS - HAND.OVER CAUS - BRING
pa-ábot pa-dala
‘something caused to be handed over’ ‘something caused to be brought’
CAUS - MAKE
pa-gawà
‘something caused to be made’
CAUS - COOK
pa-lútò
‘something caused to be cooked’
CAUS - KEEP
pa-tágò
‘something caused to be kept’
The patient-oriented nature of roots is projected to the clause level in predications, which is analyzed as essentially copular. Thus, even with a bare root predicate as in (51), the agentive argument surfaces with genitive case and the patient/theme argument surfaces with nominative/absolutive case. (51) Háwak ni Jojo ang hold gen Jojo nom ‘Jojo holds the money.’
pera money
Just as with fully inflected event-denoting predicates, extraction obeys the predicted pattern. The genitive argument is trapped within the domain of the nominal predicate and cannot be topicalized or otherwise extracted, as shown in (52). (52) a. Ang pera ay háwak nom money top hold ‘The money, Jojo holds.’ b. *Ni Jojo ay háwak gen Jojo top hold (For, ‘Jojo, holds the money.’)
ni gen ang nom
Jojo Jojo pera money
So what does nominalization add to an already nominal root? Primarily, it changes its reference to that of a potential participant. Again, we can compare English -ee in a word like hoaxee where it similarly attaches to a nominal stem to select a notional participant. Second, it also allows the root to combine with aspect inflection, which bestows an agentive reading to the genitive/ergative argument.24 This is highly reminiscent of
24 Aspect inflection is generally impossible without voice with one exception. Predicates inflected for the recent perfective show no voice marking and assign genitive case to all arguments. This is probably related to the exclamative function of the recent perfective in Tagalog, as discussed in Kaufman 2011.
Lexical category and alignment in Austronesian 613 what has been described by Davis and Demirdache (2000: 100) for St’át’imcets, a Salish language. In St’át’imcets, notionally bivalent roots surface as unaccusative predicates when used in their bare form. This is seen in (53), where the bare predicate qam’t ‘hit’ assigns the patient role to the subject ti sqáycw-a ‘the man.’ Similar to Tagalog, an agent must be licensed by a voice-related morpheme, as seen in the contrast between (54a) and (b). In (54b), the suffix -en introduces an agent and creates a transitive predicate. For Davis and Demirdache (2000), roots like qam’t and máys as well as “classic” unaccusative roots contain an underlying causer (following Chierchia 2004) but no agent. This accords relatively well with Tagalog bare root predicates in that they license causers but not true agents.25 (53)
qam’t ti sqáycw-a hit det man-det ‘The man was hit (with something thrown).’
(54)
a. Mays ti tsítcw-a built det house-det ‘The house got built.’ b. Máys-en ti tsítcw-a built-dir det house-det ‘(They) built the house.’
Let us briefly consider one theory of how Tagalog predicates are built from the root up. The category determining head n introduces a referential index (in the sense of Baker 2003). When the inserted root is entity-denoting, as in (55), this is entirely straightforward. In (55), the merged n+√ stem simply takes on a referential index associated with an instance of ⟦dog⟧. But determining how the reference is set with an event-denoting root, as in (56), is more complex, as it is often a particular participant rather than the event which the bare stem denotes on the surface. (55)
nP Possessor n0
n’ √PEntity √dog
25 The lack of an agentive reading with bare root predicates can be gleaned from (i), where the adverb nang sindayà ‘intentionally’ is infelicitous.
(i) Bigay ko ang bulaklak (?*nang give 1s.gen nom flower gen ‘The flower is my gift (?*intentionally).’
sinadyà) intentionally
614 Daniel Kaufman (56)
nP Possessor
n’
n0
√PEvent Theme
√’ √give
Recipient
Despite lacking an argument structure, an event-denoting root projects associated thematic roles as Davidsonian event arguments (or “participants,” in the terminology of Grimshaw 1990). In the case of GIVE this will include at least a theme and a recipient, as shown. Let us then posit that when √P combines with n, the root raises to n and n probes downwards to select the nearest thematic role, which determines the reference of the stem. This will naturally exclude an agentive interpretations of roots, as Agents are introduced by the higher functional projection, VoiceP (Kratzer 1996). It will also exclude possessor interpretations as the possessor role is projected to the specifier of nP and is thus outside the c-command domain of n. We thus have a mechanism for deriving a phrase such as that in (57), where the root bigay GIVE is identified with the theme of the event, i.e. a gift. The possessor and the recipient are expressed as genitive and oblique phrases, respectively.26 (57) Bigay ni Nonoy kay Neneng give gen Nonoy obl Neneng ‘The flower is a gift of Nonoy to Neneng’
ang nom
bulaklak flower
The higher functional category VoiceP licenses an agent and hosts the voice marking in Voice0, which allows the predicate to be identified with one of several thematic roles via episodic linking (Barker 1998) as in participant nominalization. The AspP projection houses aspect morphology and certain aspectual adverbs. These two functional categories undergo morphological merger, as evidenced by several points of syncretism in the Voice/Aspect paradigm and by the fact that voice marking is a prerequisite for aspect marking. We see a fully inflected analogue of (57) in (58).
26
Note that this predicts patient-orientation for all low nominalizations. This has been argued for independently by Salanova (2007) and is supported by the broad comparative study of Koptjevskaja- Tamm (1993). It may even be present in English bare nominalizations of bivalent roots (e.g. take as in ‘The take was $500’), which seem to only refer to the patient and never the agent.
Lexical category and alignment in Austronesian 615 (58)
I-bigay ni Nonoy kay Neneng conv-give gen Nonoy obl Neneng ‘The flower was given by Nonoy to Neneng’
ang nom
bulaklak flower
The peculiar meaning of causative stems seen earlier in Table 24.5 is also predicted here by the fact that the causative head lies beneath the voice head. This is corroborated by the order of prefixes: the conveyance voice prefix and actor voice prefix are always external to the causative marker: i-pa- conv-caus-, mag-pa- act-caus-. The referential index is set by n and is not affected by further movement to the causative head, pa-. So while the causative projection introduces a causer, the causer is introduced too late to be identified with the meaning of the bare stem. CausP
(59) Causer
Caus’ pa-
nP Possessor
n’
n0
√PEvent Theme
√’ √give
Recipient
The structure in (59) underlies the predicate phrase in examples like (60). (60) Pa-bigay ni Nonoy ang caus-give gen Nonoy nom ‘The flowers are Nonoy’s caused gift’
bulaklak flower
This section has shown that ergativity in Tagalog is present on the root level even before the attachment of voice morphology. This poses an additional challenge to both the case agreement and canonical ergativity approach reviewed above. Under those approaches it is left to explain why object shift should occur in the absence of a VP and, if it occurs, why it is not registered as agreement or transitivity marking on the predicate. With this brief introduction to Tagalog word structure, we are now ready to examine Mamuju, an Indonesian language which has maintained an ergative pattern while developing a strong noun-verb distinction.
616 Daniel Kaufman
24.2 Mamuju and the re-emergence of v In the following, we observe the basic morphosyntactic alternations of Mamuju, an Austronesian language of the South Sulawesi subgroup, to show the consequences of v on the syntax.27 What differentiates Mamuju from Tagalog is the loss of the nominal properties of event denoting predicates and the development of a strong Noun/Verb contrast (Himmelmann 2005: 128–131). Two important consequences of this are that agreement and valency changing operations make reference to underlying argument structure and that constraints on ergative extraction are loosened. Mamuju displays clear differences from Tagalog on the level of root and word. One simple distinction between nouns and verbs in Mamuju is that only verbs can be the complement of the tense marker na future. As seen in (61b), nominal predicates cannot follow na. In Tagalog, there are no tense/aspect markers whose distribution distinguishes root classes. (61) a. na menjari=aʔ guru fut become=1sg.nom teacher ‘I will be a teacher there at Udayana.’ b. (*na) guru=aʔ jao fut teacher=1sg.nom there ‘I am a teacher there at Udayana.’
jao there di prep
di prep
Udayana Udayana
Udayana Udayana
Conversely, verbal roots cannot combine directly with a possessor, unlike nouns, as shown in (62). An overt nominalizing head must combine with a verbal stem before prior to modification by a possessor. As seen earlier, this is not the case in Philippine- type languages. (62) a. *langi-na swim-3.gen b. bau-na fish-3.gen ‘his/her fish’ If the symmetric nature of Philippine-type alignment systems are due to the nominal nature of the roots, we expect that the emergence of v will allow for bona fide 27
Mamuju has so far only been described in a single article by Stromme (1994), who also published interlinearized text collections (Stromme 1991). That work has been supplemented by elicitation with native speaker Husni Husain over the course of a semester at the CUNY Graduate Center in 2009 as well as two weeks of field work in the Mamuju area itself. I thank Husni Husain for generously sharing his knowledge with us as well as my graduate students Eva Szymanski, Josh Gray, Ji Young Shim for their valuable input.
Lexical category and alignment in Austronesian 617 argument structure to be projected from the root and an absence of symmetry in the derivational morphology. Recall that in Tagalog, the notion of valency plays almost no role in determining a root’s morphological potential. This is seen clearly in Table 24.6, where a notionally monovalent root, langoy ‘swim’ is compared with a bivalent root patay ‘kill.’ Both roots can take the full range of voice morphology suggesting that they are essentially of the same type at the point where they merge with Voice. Table 24.6 Tagalog word classes √LANGOY ‘swim’
√PATAY ‘kill’
actor voice -in patient voice
langoy languy-in
‘nom to swim’ patay ‘to swim to nom’ patay-in
‘nom to kill’ ‘to kill nom’
-an locative voice
languy-an
‘to swim in nom’ patay-an
‘to kill from nom’
This contrasts starkly with Mamuju and other South Sulawesi languages. In Table 24.7, we see that the cognate Mamuju roots are in complementary distribution with regard to their morphosyntactic potential. An unergative verb like langi ‘swim,’ requires prefixation with mo to form an intransitive predicate. In contrast, a bivalent root like patei ‘kill,’ cannot take the intransitive prefix but rather must be prefixed with antipassive mang- to enter into a intransitive predication. Bivalent stems must be prefixed with ergative agreement when forming transitive verbs, as seen with ku-patei ‘I kill (X).’ This agreement marking is impossible with monovalent roots like langi.
Table 24.7 Mamuju word classes √LANGI ‘swim’
√PATEI ‘kill’
mo-active mang-antipassive
mo-langi *man-langi
‘abs to swim’ —
ku-1sg.erg
*ku-langi
—
*mo-patei mam-patei ku-patei
— ‘abs to kill’ ‘erg to kill abs’
This development can be understood as part of the emergence of a robust v category which converts the thematic structure of an event-denoting root into actual argument structure. Note also that, unlike in Philippine-type languages, we now find an unmarked class of intransitive and transitive verbs. Unaccusatives, like tama ‘enter’ in (63), do not require derivational morphology when surfacing as intransitive verbs. Similarly, bivalent roots are only prefixed with ergative agreement when functioning as transitive
618 Daniel Kaufman verbs, as seen in (64). All predicates can host second-position pronominal clitics reflecting the absolutive argument. (63) tama=do=ʔ di enter=already=1.abs prep ‘I already entered the room’
songi room
(64) na-kita=ko 3.erg-see=2.abs ‘S/he sees you.’ Note also the divergence in possessor and ergative agreement. The former is expressed via a set of suffixes and the latter through a set of verbal prefixes. In all Philippine-type languages, these two functions are expressed with the same set of pronouns, the typically second-position genitive clitics. Importantly, Mamuju has a robust antipassive, which only combines with transitive predicates and is used to introduce indefinite objects, as can be seen in the comparisons in (65) and (66).28 In the transitive clause, a missing patient argument would be interpreted as a null pronoun retrievable from discourse. The ergative argument typically follows the verb directly although scrambling may also occur to yield an ABS ERG order. (65) a. na-kande i Husni 3.erg-eat art Husni ‘Husni is eating the fish.’ b. mang-kande (bau) i antipass-eat fish art ‘Husni is eating (fish).’
bau fish Husni Husni
(66) a. mu-kita=a’ 2.erg-see=1.abs ‘You see me.’ b. mang-kita=a’ antipass-see=1.abs ‘I see a fish.’
bau fish
In addition to the anti-passive, a limited number of verbs appear to take a “super- antipassive,” which has also been described for the neighboring Seko Padang language by 28 The antipassive mang-prefix derives historically from a distributive/pluractional infix * (Kaufman 2009c). The path from Proto-Malayo-Polynesian distributive marker to an antipassive marker in the South Sulawesi languages is straightforward. Just like antipassives, distributive verbs are bivalent but cannot take a specific or definite object. The Makassarese cognate aN(N)-is discussed by Jukes (2013).
Lexical category and alignment in Austronesian 619 Payne and Laskowske (1997). The super-antipassive also expresses the subject as an absolutive but does not allow for any objects, whether definite or indefinite, as shown in (67).29 (67) kande=ko eat=2.abs ‘You eat.’
(*bau) fish
Mamuju also has a passive, marked on the verb with the prefix ni-, shown in (68b). Ergative agreement is obligatory on transitive verbs and thus the passive allows for backgrounded or impersonal agents.30 (68) a. Apa na-kande todapa? what 3.erg-eat people ‘What do the people eat?’ b. Apa ni-kande (*todapa)? what pass-eat people ‘What is being eaten?’ Mamuju has bona fide applicatives that promote benefactives and oblique arguments to absolutive. Furthermore, the division between low and high applicatives does not run into the difficulties noted earlier for Tagalog. In (69a), the applicative -ang attaches low and transitivizes an intransitive verb by introducing a theme, thereby feeding ergative agreement. When -ang merges as a high applicative, as in (69b), it attaches to a transitive stem (pa-lamme) and introduces a benefactive. (69) a. ku-lamme-ang buku 1s.erg-fall-appl book ‘I dropped a book.’
29
The super-antipassive is more useful than it appears at first sight. With the loss of case marking on DPs, the difference between antipassive and super-antipassive is all that distinguishes a post-verbal object from a post-verbal subject in sentences like (i). (i) a. mangapa bongi itte ampe’ why/when night dem conj ‘What time did the fish eat last night?’
kande eat
b. mangapa bongi itte ampe’ why/when night dem conj ‘What time did he eat fish last night?’
mangande antipass:eat
bau? fish
bau? fish
30 The existence of a passive in Philippine-type languages is a somewhat difficult question. Reid and Liao (2004) and Tanangkingsing and Huang (2007) claim that ma-verbs are essentially passives in Bontok and Cebuano, respectively, because of their inability to license agents. On the other hand, the ma-prefix indicates non-volitionality and is obligatory on certain unaccusative predicates like ‘to fall’,
620 Daniel Kaufman b. ku-pa-lamme-ang=ko 1s.erg-caus-fall-appl=2s.abs ‘I dropped a book for you.’
buku book
Although the data on double applicatives is still unclear for Mamuju, other languages of Sulawesi allow for promotion of multiple adjuncts to arguments through applicative stacking. Tukang Besi, for instance, allows the comitative and benefactive applicative combination shown in (70). Among the South Sulawesi languages, Selayarese allows double applicatives when both the theme and recipient of a ditransitive verb are definite, as shown in (71) and Sirk (1996: 82) discusses an identical construction for Bugis. Recall that no Philippine-type language allows double applicatives of this type (on the analysis of the voice/nominalization morphology as applicative marking).31 (70) No-wila-ngkene-ako te ina-no 3.rl-go-com-appl core mother-3poss ‘She went with Wa Ki’i for her mother.’ (71) Ku-kiring-i-ang=ko 1s.erg-send-appl-appl=2.abs ‘I sent you the money.’
te core
doe’-injo money-def
Wa Ki’i. Wa Ki’i (Donohue 1999: 248)
(Basri 1999: 313)
As shown in (72b), applicatives are incompatible with the antipassive. Aldridge (2012b) predicts this behavior via the requirement that applied objects require structural case, which is systematically lacking in antipassives.32 which makes it look quite different from a traditional passive. (See Kaufman 2012 for a discussion of the historical development of ma-verbs). I thank Laurie Reid for bringing this point to my attention. 31 An apparent counterexample is found in the Cordilleran languages that employ the circumfix i- -an for the benefactive. While these were historically two morphemes (*i-conveyance and *-an locative), they have been reanalyzed as one. This is clear from the syntax of benefactives in the relevant languages, in which only one argument, the benefactive, is “promoted” to the nominative/absolutive. 32 Note that in Bajau, and potentially other Austronesian languages, antipassives can both host definite objects, as seen in (i), and freely allow combinations of actor voice and applicatives, as shown in (ii). The prefix must still be considered antipassive as it only attaches to transitive stems to make an absolutive argument from the agent.
(i)
Nga-daka’ manu’ iru antipass-catch chicken that ‘I caught that chicken for my father.’
aku 1s
(ii) Nga-daka-an uwa’-ku manu’ antipass-catch father-1s.gen chicken ‘I caught that chicken for my father.’
pugay do.for iru that
uwa’-ku. father-1s.gen aku. 1s (Donohue 1996: 789)
Bajau thus represents an intermediate stage between Mamuju and modern Indonesian, where the cognate meng-prefix has extended its domain to include many monadic verbs as well (e.g. Indonesian menangis /meN-tangis/ ‘cry’, menyala /meN-nyala/ ‘to be on/alight’).
Lexical category and alignment in Austronesian 621 (72)
a. Mam-baca=aʔ buku antipass-read=1.abs book ‘I read a book for Husni.’ b. *Mam-baca-ang=aʔ antipass-read-appl=1.abs (For, ‘I read Husni a book.’)
bua for
Husni Husni
buku book
Husni Husni
Antipassives are, however, allowed to combine freely with applicatives when the agent is extracted, as in (73). This is the same condition under which definite patients are possible for actor voice predicates in Philippine-type languages. (73)
Sema mam-baca-ang who antipass-read-appL ‘Who read Husni a book?’
Husni Husni
buku? book
The Mamuju data observed thus far suggests that a verb’s underlying argument structure largely determines its morphosyntactic potential. The v category combines with event- denoting roots and projects a basic argument structure from the thematic structure. This structure can then be adjusted by transitivizing and detransitivizing morphology. In Philippine-type languages, to the extent that we can speak of a fixed argument structure, it is created by the nominalizing/voice morphology rather than inherited from the root. I argued earlier that the simplest account of extraction constraints in Philippine-type languages unifies the patterns found across event-denoting and entity-denoting predicates. In Mamuju, where event-denoting predicates no longer have any nominal properties at all, we would expect a loosening of these extraction constraints. This expectation is borne out for topicalization, as shown in (74). Here, the ergative argument is extracted to a pre-verbal topic position, which was shown earlier in (8) to be generally ungrammatical in Tagalog.33 (74)
Baco na-patei Ali Baco 3.erg-kill Ali ‘Ali killed Baco.’ OR ‘Baco killed Ali.’
However, relativization and question formation are still restricted in the usual way, as seen in (75), which only has a single interpretation. 33 Edith Aldridge (p.c.) points out that the extraction might be expected if na-functions as a resumptive pronominal clitic. From a morphological perspective, ergative person marking in Mamuju appears to be standard agreement in that it is obligatory, attaches consistently to the left edge of the verbal stem and can co-occur with a co-referential pronominal argument. Nonetheless, it is possible that agreement plays a role in licensing topicalization, especially given that antipassive objects, which do not trigger agreement, are more highly constrained. See Kaufman (2008) for details.
622 Daniel Kaufman (75) Sema na-kita Ali? who 3.erg-see Ali ‘Who did Ali see?’ NOT ‘Who saw Ali?’ It would seem then that extraction constraints can, historically speaking, outlast the nominal features of the predicate. Mamuju would seem to be an ideal candidate for an ergative analysis along the lines of Aldridge (2004) and would not pose any of the empirical hurdles found with Philippine-type languages enumerated in (11). The mang-prefix would be a true antipassive while the formatives me-, mo- and mu-could be treated as instantiations of intransitive v. Unlike as in Philippine languages, polyvalent roots would give rise to unmarked transitive verbs and the suffixes -i and -ang behave like true applicatives attaching to an unmarked transitive verb and, with agent extraction, to an antipassive. Nonetheless, I would argue that the restrictions on extraction in Mamuju obtain the best explanation as historical residue from Philippine-type morphosyntax. Recall that two potentially independent factors account for extraction restrictions in Philippine-type languages: the constraints on extraction from NP and the ban on genitive predicates. As shown earlier in detail, Mamuju has developed a strong N/V distinction and thus constraints on extraction from NP have become irrelevant for ergative extraction. This explains why ergatives can be freely topicalized to the preverbal position as in (74). The second factor, however, remains firmly in place as seen in (76), which shows the only way a would-be genitive argument can be extracted. The “dummy” predicate ampunna ‘owner’ allows a possessor to be expressed as an absolutive argument. (76) Sema ampunna ku’bur itte di bao di who owner grave that prep down prep Timbu me- loda batu? Timbu av.have-roof stone ‘Whose is that grave down in Timbu with the stone roof?’ (Stromme 1991:Maradika Lasalaga) Although there is no overt morphological case marking of DPs in any of the languages of the South Sulawesi subgroup, we find that the interrogative and relative marking elements themselves bear unambiguous traces of absolutive case. The interrogative pronoun sema ‘who,’ which is plausibly derived from Proto-Austronesian *si-ima (Blust et al. 2010), contains a reflex of the personal nominative/absolutive case marker *si-.34 34
The presence of Proto-Malayo-Polynesian nominative/absolutive *si can also be found in Indonesian/Malay siapa ‘who,’ among many other Austronesian languages. See Ross (2006) for a detailed reconstruction of the relevant Austronesian case markers. An alternative source of Mamuju sema is a hypothetical sai-ma ‘who-relt.’ Blust et al. (2010) reconstructs *sai ‘who’ for Proto-Malayo-Polynesian and ma is found in Bajau and Kambera as a relativizer (although not attested for Mamuju). The nominative case element on sai is not as easy to isolate but given the reconstructions of the nominative singular proper name marker *si and its plural counterpart *sa, it is not far-fetched to also relate the s(a) of sai to the nominative case function.
Lexical category and alignment in Austronesian 623 Crucially, no genitive/ergative variant (which we would expect to come out as nema) exists in Mamuju (nor in any of the South Sulawesi languages). Recall that genitive marked interrogative pronouns do exist in Philippine-type languages but can only be used with in-situ interrogative phrases, as seen earlier in (40), repeated here in (77). (77) Gawà-∅ nino make-pat gen.who ‘Who made those shoes?’
ang nom
sapatos shoe
na iyon? lnk that (Schachter and Otanes 1982: 512)
Similarly, the Mamuju relative marker anu is clearly cognate with the Tagalog nominative interrogative ano (underlyingly also /anu/). As the interrogatives themselves show morphological traces of absolutive case, it follows that only absolutive arguments can be clefted given case preservation.35 The proper morphological analysis then of interrogative sentences such as (75) is shown in the glossing of (78), where the case marked interrogative constrains the types of arguments that can be extracted. Similarly, the grammatical voice of verbs in relative clauses is constrained by the case of the relativizer, as exemplified in (79). (78) Sema na-kita Ali? abs.who 3.erg-see Ali ‘Who did Ali see?’ NOT ‘Who saw Ali?’ (79) Ku-pe-ingarangng-i=mo apa 1.erg-tr-remember-app=emph abs.what sambongi. last.night ‘I remember all that I dreamt last night.’
anu abs.relt
ku-so’na 1.erg-dream
(Stromme 1991:Alibe Niso’na)
This analysis, although syntactically trivial, offers a morphologically well-founded basis for the difference between relativizations and question formation on one hand, which both require case marked operators, and topicalization, which does not. As we have seen, the former operations are just as constrained in Mamuju as they are in Philippine- type languages while the latter operation is freer, as constraints against extraction from NP are no longer relevant for event-denoting predicates.36 35 The morphological argument is less clear with ‘what’ than ‘who’ as nominative ‘what’ can be treated as unmarked. Compare Tagalog ano ‘what,’ ng ano ‘gen what,’ saan ‘obl.what/where.’ However, the a-initial in the Mamuju relative marker anu and interrogative apa ‘what’ is found in a wide range of nominative/ absolutive interrogatives throughout Philippine-type and Formosan languages. In Tagalog, we can compare the argument interrogatives a-no, a-lin ‘which’ and si-no ‘nom.who’ which all begin with the putative nominative markers a-or si-(for proper names) to the adjunct interrogatives ilan ‘how many,’ kailan ‘when,’ bakit ‘why,’ magkano ‘how much’ none of which begin with a-. Ross (2006) presents the comparative data but tentatively reconstructs the *a-formative with demonstrative features rather than case features. 36 Note however that this approach requires a view of the grammar that attributes more power to the lexicon than is currently popular in generative theorizing. In particular, a lexical gap in the case paradigm for certain operators must be able to block syntactic movements. This would not be
624 Daniel Kaufman Note that this analysis extends beyond Mamuju and can account for a similar pattern in Indonesian, a language which is generally considered to have adopted a nominative- accusative alignment pattern (Chung 1976, 2008; Cole et al. 2008; Aldridge 2008a). As can be seen in (80), actor voice objects can be topicalized but not relativized in Indonesian. (80) a. Orang itu, saya meng-ajak ke sini. person that 1sg act-invite to here ‘As for that person, I invited (him/her) here.’ b. *Orang itu yang saya meng-ajak person that relt 1sg act-invite (For, ‘That person whom I invited here.’)
ke sini. to here (Arka and Manning 2008: 53)
Again, we can plausibly attribute (80b) to the relativizer, which Adelaar (1992) reconstructs as ia-ng 3s.nom-lnk. Indeed, ia functions as a strictly nominative case pronoun in modern Indonesian as well. Compare in (81) the distribution of nominative ia with dia, a case neutral pronoun which can function as either subject or object of an actor voice clause (Musgrave 2001). (81) Dia/ia 3s.neut/3s.nom ‘He sees him.’
me-lihat act-see
dia/*ia 3s.neut/3s.nom
Were Indonesian to have in its functional inventory a case neutral relativizer (hypothetically diang), we might expect that extraction could take place from a wider array of syntactic positions. The prediction here is that Austronesian’s well-known “subjects only” condition on extraction can easily dissipate once the nominal features of event-denoting predication are lost, as it is then “relic” properties of functional items that keep the restriction in place rather than the strong island characteristics of Philippine-type predicates. While constraints on ergative extraction are relatively tenacious among Indonesian languages, we do find several ergative languages that allow ergative extraction in a manner unknown among Philippine-type languages. This is exemplified here by Sumbawa Besar (as described by Shiohara 2013) and Selayarese (as described by Basri and Finer 1987; Basri 1999).37 The transitive verbal clause in Sumbawa employs an unmarked verb countenanced in a framework such as Distributed Morphology where the lexicon is reduced to almost nothing and the putative lexical gap relied upon above would itself have to be derived synchronically in the syntax. On the other hand, it is a perfectly natural analysis in Combinatory Category Grammar (Steedman 2001) where Mamuju anu would be of the category: (N\N)/(S/NPABS), that is, a predicate which seeks a clause with a nominative argument gap to its right and a noun to its left to yield another noun via predicate modification. 37
See also Erlewine, Levin, and van Urk, Chapter 16, this volume, who discuss Balinese pre-verbal subjects in object voice.
Lexical category and alignment in Austronesian 625 stem with prefixal ergative agreement and introduces the ergative argument with the case marker léng, as seen in (82).38 (82)
ka=ku=inóm kawa past=1sg=drink coffee ‘I drank the coffee.’
nan that
léng by
aku 1sg
(Shiohara 2013: 132)
Antipassives are formed with the nasal prefix N-and subject agreement (which is similar, but not identical to ergative agreement). As shown in (84), antipassives do not allow for objects. (83)
a. ka=ku=ng-inóm past=1sg=antipass-drink ‘I drink (something).’ b. *ku=ng-inóm kawa 1sg=antipass-drink coffee (For, ‘I will drink (the) coffee.’)
aku. 1sg (nan) (that)
(Shiohara 2013: 132)
Crucially, the ban on antipassive objects (much like Mamuju’s super-antipassive in (67)) appears to be indefeasible and thus trumps the restriction against extraction of the ergative argument. The result is that agent extraction proceeds from a transitive clause as shown in (84a) in the presence of an object. The antipassive is used in subject questions only when no object is present, as in (84b). (84) a. sai adè ka=tumpan’ jangan=ta who nom past=get fish=this ‘Who caught the fish?’ b. sai adè ka=n-umpan’? who nom past=antipass-get ‘Who already had a catch (in fishing)?’
(Shiohara 2013: 135)
Topicalization of the ergative argument to a preverbal position is also possible, as expected, although interestingly, Shiohara shows that the ergative marker must be omitted in this construction, as shown in (85) (see n. 38).
38
The pattern of a leng-phrase agent triggering prefixal agreement appears very similar on the surface to that found in Acehnese (Durie 1985; Legate 2012b, 2014b). However, while Sumbawa clearly follows an ergative pattern, Acehnese is typically described as having an active-stative alignment type which Legate (2012b, 2014b) derives from properties of vP. Shiohara (2013) in fact glosses leng as ‘by’ but because leng seems obligatory on external arguments of transitive verbs, I re-gloss leng as erg. Shibatani (2008) argues that there is a significant difference between the verb in agent extraction, as in (84a), and a canonical declarative like (82). Space does not allow a more in depth discussion of the complex Sumbawa facts.
626 Daniel Kaufman (85) (*leng) aku (ku=)inóm erg 1sg 1sg=drink ‘I drink the coffee.’
kawa coffee
nan. that
(Shiohara 2013: 137)
Selayarese presents a similar state of affairs. The alternation in (86) shows that person marking and alignment are identical to Mamuju in the simple case. A definite object requires a transitive clause with ergative agreement and a second-position absolutive clitic. An indefinite object must be introduced with an antipassive/intransitive verb. (86) a. Ku-halli’=i sapo=ɲjo 1.erg-buy=3.abs house=def ‘I bought the house.’ b. M-mali=a sapo intr-buy=1.abs house ‘I bought a house.’
(Mithun 1991a: 175)
As with Sumbawa and Mamuju, the external argument of both a transitive and intransitive predicate can be topicalized, as shown in (87). (87) a. I Baso’ la-alle=i pm Baso 3.erg-take=3.abs ‘Baso took the money.’ b. I Baso’ (a)ng-alle(=i) pm Baso intr-take=3.abs ‘Baso took some money.’
doe’=ɲjo money=def doe’ money
(Finer 1994: 159)
Like Sumbawa, but unlike Mamuju, this extends to cases of agent extraction in transitive clauses, as seen in (88), where the verb still takes ergative agreement rather than antipassive/intransitive morphology. Just like in the dialect of Sumbawa described by Shiohara (2013), this is the only option when the object is definite.39 (88) Inai la-sumbele=i who 3.erg-slaughter=3.abs ‘Who slaughtered the buffalo?’
tedong=injo? buffalo=def
(Hasan Basri 2006 p.c.)
39
Note that the extraction of antipassive objects in Selayarese is still restricted just as in more morphosyntactically conservative languages. Recall that in Indonesian, where the cognate prefix meN- marks transitive active voice verbs, topicalization (but not relativization) of the object is permissible, as shown earlier in (80). (i) *Doe (a)ng-alle=i money intr-take=3.abs (For, ‘Money, Baso took.’)
i pm
Baso’ Baso (Finer 1994)
Lexical category and alignment in Austronesian 627 This also extends to relative clauses, as seen in (89), which are optionally introduced by the relative marker nu. (89) a. asu (nu) n-datalaʔ dog relt intr-chase ‘the dog that chased a pig.’
bahi pig
b. asu (nu) la-datala-ɲjo=i dog relt 3.erg-chase-def=3.abs ‘the dog that chased the pig.’
bahi=ɲjo pig=def (adapted from Basri 1999: 292)
While this diachronic analysis of case marked relativizers and interrogative operators can extend to many Austronesian languages transparently there are inevitably exceptions. Bajau as described by Donohue (1996), for instance, shows the classic extraction restrictions but has a relativizer ma that alternates with ∅. Bajau ma is homophonous with the oblique case marker/preposition rather than an absolutive/nominative one and yet it is only the absolutive argument which can be relativized. The converse problem is found in Malagasy, where a focus marker which almost certainly also derives from PMP *anu allows for clefting of prepositional phrases. (Note also that the Selayarese relativizer in (89) derives from anu yet allows relativization of the ergative argument.) (90) T-amin-ny antsy no nanapaka ity pst-with-det knife foc pst.act.cut this ‘It was with the knife that Sahondra cut this tree.’
hazo tree
ity this
Sahondra Sahondra (Paul 2001)
Clearly, the etymology of relativizers and interrogatives is only suggestive of the solution proposed here. We cannot expect that the features and structures involved in questions and relative clauses will not diverge from their etymological roots. Nor should we expect that the features of a relative marker or interrogative may not be inherited by a lexical item that comes to replace it.
24.3 Conclusion I have argued here that lexical categories play a critical role in accounting for the difference between Philippine-type language and related ergative languages of Indonesia. In particular, symmetrical voice systems in which every event-denoting predicate is marked with voice morphology emerge from participant nominalizations. True argument structure with unmarked intransitive and transitive event-denoting predicates are rooted in verbal categories. Mamuju was presented as a canonical ergative language with a highly developed N/V contrast. As a result, it displays canonical antipassives and
628 Daniel Kaufman applicatives of a kind not found in Philippine-type languages. When event-denoting predicates lose their nominal properties and become verbalized, certain syntactic properties of ergative alignment (e.g. extraction constraints) are more likely to be lost as predicted by Manning’s (1996: 21) hypothesis that “syntactic ergativity” results from nominalization, in contrast to more surface-oriented “morphological ergativity.” Further research in the comparative syntax of Indonesian languages is necessary to better understand the correlates between alignment-type, lexical category and extractability. The bewildering assortment of agreement and argument marking patterns (see, for instance, Kikusawa (Chapter 23, this volume) as well as the papers in Adelaar and Himmelmann (2005), Wouk and Ross (2002), Arka and Ross (2005) and Adelaar (2013) for a sample) will likely require decades to fully sort out. One of the purposes of this chapter has been to suggest new potential correlations for further investigation as we expand our empirical scope to the many under-described Austronesian languages of Indonesia and beyond.
Acknowledgements I thank Lisa Travis, Edith Aldridge, Laurie Reid and an anonymous reviewer for extensive comments on this chapter which led to considerable improvements. None of them should be held responsible for the views expressed herein, for which I am alone to blame.
Abbreviations ABS, absolutive; AV, actor voice; ACC, accusative; ACT, actor nominalization, active voice; ANTIPASS, antipassive; APPL, applicative; ART, article; BEG, begun aspect (a component of both the perfective and progressive); CAUS, causative; COM, comitative; CONJ, conjunction; CONV, conveyance nominalization; CORE, core argument; DAT, dative; DEF, definite; DEM, demonstrative; DET, determiner; DIR, directive transitivizer; EMPH, emphatic; GEN, genitive; ERG, ergative; FOC, focus; FUT, future; IMPRF, imperfective INTR, intransitive; LOC, locative nominalization; LNK, linker; NEG, negation; NEUT, neutral case; NOM, nominative; OBL, oblique; PASS, passive; PASS.PART, passive participle; PAT, patient nominalization; PL, plural; PM, personal marker; POSS, possessive; PREP, preposition; RELT, relativizer; RL, realis; STA, stative; SUPERANTIPASS, super anti-passive; TOP, topic; TR, transitive.
Aquisition
Chapter 25
The ac quisi t i on of e rgativit y: An ov e rv i ew Edith L. Bavin
25.1 Introduction An important focus for researchers of child language has been identifying properties that seem to facilitate the acquisition of structures relative to a specific language, structures which indicate events and who is doing what to whom. In order to acquire a language, and so become productive in using it, children extract patterns from the input and generalize to new instances. Crosslinguistic research is important in generalizing the factors that influence how children achieve this. Some research focuses on developmental paths, for example, when children start producing grammatical morphemes, when they produce them in appropriate contexts, and which ones appear early and which late. Other research focuses on cues in the input, for example, morphology and word order that might help in identifying and acquiring grammatical morphology. In this chapter I discuss some of the variation found in ergative alignment and properties that might impact on how readily a child acquires the ergative system and usage patterns of the ambient language. In addition, I summarize data reported for three ergative languages to illustrate language specific features and their influence on the acquisition of the ergative features of the language. Overall, overgeneralizing or undergeneralizing in the use of ergative morphology is not frequent in relation to the acquisition of other morphology (as pointed out by Pye 1990). Based on research from a number of ergative languages representing different language families (see Bavin and Stoll 2013), it is clear that children acquiring ergative languages acquire the system rapidly and by age 2;6 or 3 years have knowledge of the system.1
1
Note: more information about the acquisition of ergative languages is presented by Austin, Chapter 26, this volume, and Pye and Pfeiler, Chapter 27, this volume.
632 Edith l. Bavin
25.2 Alignment and acquisition 25.2.1 Alignment A major distinction across languages is how events are encoded: whether the language uses nominative-accusative alignment or ergative-absolutive alignment. Ergativity is typically associated with transitivity. In ergative-absolutive languages, transitive subjects (A) and intransitive subjects (S) are distinguished; S and objects (O) are treated similarly. In nominative-accusative languages A and S are treated similarly. In the prototypical transitive clause, an agent fills the subject slot but other semantic roles may do so; in ergative languages the A is identified as ergative. Ergativity is generally shown morphologically rather than syntactically. The distinction between A and S is evident in case marking (as in Hindi; Narasimhan, 2005, 2013), verb agreement (as in Mayan languages; Brown et al. 2013; Pye, Pfeiler, and Mateo Pedro 2013) or both case and verb agreement (as in Basque; Austin 2013). Case marking is more common than verb morphology in marking ergativity. Ergative languages rarely show ergative alignment in their syntactic operations (for examples from Dyirbal, a syntactically ergative language, see Dixon 1979), so that the child learner needs to distinguish two systems: morphological and syntactic. Warlpiri, a Pama Nyungan language spoken in central Australia, is an ergative language in terms of its case marking (see also Laughren, Chapter 39, this volume). Ergative alignment is illustrated in (1) by the case marking on the A arguments), while the O and S arguments are unmarked (absolutive case). Note that the arguments can be interpreted as definite or indefinite and word order is variable. Case marking is used on free- standing pronouns, which register person and number, but it is not obligatory on first and second person singular pronouns. Clitics, which appear in second position in the clause, register the person and number of core arguments. However, unlike the case marking system, clitics representing the S and A arguments are treated alike and are distinguished from those representing O arguments. Except for dative third person, the third person singular clitics are null in form (see Hale 1982). Examples (2a and 2b) illustrate pala, the third person dual clitic for A and S. In (2a) the object clitic for the first singular object, ju, is distinguished from rna, the first person singular subject (S and A) form, which is used in (2c) and (2d). (1)
a. Luwa-rnu wati-ngki marlu shoot-PST man-ERG kangaroo ‘The/a man shot the/a kangaroo.’ b. Nyina-mi ka wati sit-N.PST IPFV man ‘The/a man is sitting.’
(2) a. Nya-nyi ka-pala-ju karnta-jarra-ngku See-N.PST IPFV-3DU.SUBJ-1SG.OBJ woman-DU-ERG ‘Two women are looking at me.’
An overview of the acquisition of ergativity 633 b. Nyina-mi ka-pala sit-NPST IPFV-3DU.SUBJ ‘They (two) are sitting.’ c. Nya-ngu-rna-ngku see-PST-1SG.SUBJ-2SG.SUBJ ‘I saw you.’ d. Nyina-mi ka-rna sit-NPST IPFV-1SG.SUBJ ‘I am sitting.’ Syntactic ergativity is shown if syntactic operations, such as clause conjoining, follow ergative-absolutive alignment (see Dixon 1979, 1994). However, in Warlpiri S and A arguments are treated similarly in that the same complementizer, karra, is used to indicate that the main clause S (shown in 3a) or A (shown in 3b) controls the subordinate subject (examples from Hale 1983), whereas for a main clause object as controller of the subordinate subject, the complementizer is kurra. (3) a. Karnta ka-ju wanka-mi yarla karla-nja-karra Woman IPFV-1SG.DAT speak-N.PST yam dig-INF-COMP ‘The woman is speaking to me while digging yams.’ b. Ngarrka-ngku ka purlapa yunpa-rni karli jarnti-rninja-karra-rlu man-ERG IPFV corroboree sing-N.PST boomerang trim-INF-COMP-ERG ‘The man is singing a corroboree while trimming the boomerang.’
25.2.2 Semantic Bootstrapping Pinker’s (1984, 1989) bootstrapping theory presents a possible explanation for how syntactic roles are acquired. Children are assumed to draw on basic (innate) knowledge of syntactic categories, semantic roles and linking rules (the expected links between semantic and syntactic functions). Initially, children would link an agent thematic role to the subject of a verb. Once that link is established they would extend subjecthood to include other thematic roles. That is, semantics is assumed to provide an entry into syntax. If the role of agent helps bootstrap children into acquiring syntactic structures, a prediction would be that children acquiring a language with ergative case marking would identify prototypical agents first and then extend the ergative marking to other semantic roles, irrespective of the verb’s transitivity, and so extend the case marking to S arguments (see also Pye 1990; Van Valin 1992; Siegel 2000; Van Valan 1992). Pinker did not perceive ergative languages as posing a problem; his stated view (1984: 372) was that children would either notice similarities in the encoding of the agent (A) in a transitive clause and actor (S) in an intransitive clause (nominative-accusative languages), or similarities for encoding the intransitive actor (S) and transitive patient
634 Edith l. Bavin (O) (ergative–absolutive languages), suggesting that children would need to distinguish verb type and associated morphological marking for the arguments for each, that is identify distributional patterns in the input using the available language specific cues. This, however, simplifies the problem since the use of ergative alignment varies across and within languages. Ergative languages typically have split systems, with ergative- absolutive alignment used in some contexts but not all, and the use of ergative morphology may be variable in languages, (as discussed in section 25.3). If the input to young children includes a large proportion of highly transitive clauses, it might be expected that children would use a high proportion of transitive verbs in their early utterances. So rather than assuming that if children first use ergative marking with highly transitive verbs they are drawing on a link between agent and ergative marker before extending the marker to other A arguments (as predicted from the semantic bootstrapping approach), an alternative explanation is that the input favours such verbs and children are acquiring the patterns they hear in the input. Children need to identify the specific linguistic contexts, and social contexts, for using case morphology or verb agreement associated with ergative-absolutive alignment on the basis of patterns in the input. Rumsey, San Roque and Schieffelin (2013) reported on the acquisition of ergative marking in three Trans New Guinea languages. In conversations and monologues from Duna, one of the languages studied, ergative marking was used on only 54% of the A nominals; in Ku Waru, a related language, approximately 62% of A arguments in a sample of adult speech were marked with ergative marking. Clearly, when there is variability in whether an A argument is marked with ergative morphology or not, the opportunity is reduced for children to determine the contexts where ergative marking is applicable. In another study, on the acquisition of Samoan, Ochs (1982) found that the young children studied used ergative marking in less than 5% of obligatory contexts. The use of ergative case marking in Samoan is constrained by social factors, the social distance between speaker and listeners, and so is rarely used when communicating with family members. Thus the young children would have restricted opportunities to hear ergative marking and identify the conditions for its use.
25.3 Variability, Pragmatic Functions and Multifunctionality 25.3.1 Split Systems and Noncanonical Use of Ergative Marking A general problem in determining where ergative morphology applies is that ergative alignment does not generally apply overall in a language. The prototypical pattern for ergative alignment is for a single argument of a monovalent verb to be identified as absolutive and the primary argument of a bivalent verb to be identified as ergative. However, ergative marking varies within languages and split systems are common. Comrie (2013c) discusses some of the variability in the encoding of ergativity, questioning whether it represents syntactic or semantic alignment. Ergative morphology may also be found for the single argument of a monovalent verb in some languages.
An overview of the acquisition of ergativity 635 Split alignment systems are conditioned by various factors. For example, nouns and demonstratives may show ergative-absolutive alignment while pronouns do not. Splits may also be restricted to certain pronouns based on person or number. In one such language, Arctic Quebec Inuktitut (Allen 2013), the split system is based on number; singular ergative and absolutive case forms on nouns and demonstratives are distinguished, but not the plural forms. Verb aspect is the basis for the split system in Hindi; ergative alignment is applied in clauses with verbs in perfective aspect. In Trans New Guinea languages (Papua New Guinea), ergative case marking depends on word order; it is used in OAV word order, but in AOV the A is marked ergative only if both the A and O arguments are proper names or kin terms. The specific contexts for the application of ergative marking need to be identified based on language specific distribution patterns. Split systems would be problematic if a child assumes ergative marking is associated with any A argument of a bivalent verb. Another potential problem for acquisition is that for some languages ergative marking is extended in specific contexts to S arguments, as in the Mayan languages discussed by Pye, Pfeifer and Mateo Pedro (2013). In Nam, the infrequent extension of ergativity to S arguments is restricted to complex clauses. In contrast, the extension of ergativity to S arguments in Yukatec is frequent, occurring after a progressive verb. In some contexts, ergative morphology can also crossreference the O argument of a transitive verb (see Pye et al. 2013 for details). There is no strong evidence that split ergative alignment systems within a language lead to delayed acquisition. For example, Narasimhan (2013) reports that children do not extend the ergative case to agents of transitive clauses with non-perfective aspect marking. Nor are children found to extend ergative marking generally to S arguments if this is not a feature of the input language. The acquisition data support the view that children rely on distributional evidence in the input to identify where ergative marking applies, and its acquisition is relatively error free.
25.3.2 Pragmatic Functions Although the term ‘ergative language’ is typically used in reference to the grammatical or semantic conditioning of alignment patterns, ergative marking often serves pragmatic functions, indicating, for example, focus, contrast, or individuation. McGregor (2010) illustrates that there are two relevant features of the S (agentive and referential) in using ergative marking in intransitive clauses in Warrwa, a Nyulnyulan language of North West Australia. The referential use indicates that an ergative marked argument is unexpected in the particular context. In Hindi, the S argument of a small number of intransitive verbs, including verbs of ‘bodily emission’ and others (e.g. chĩĩk ‘sneeze’, ro ‘cry’, cillaa ‘shout’, nahaa ‘bathe’) can be marked as ergative in perfective contexts and when the S is viewed as being in control, that is, in volitional contexts (Narasimhan 2013). DeLancey (2011) discusses noncanonical ergativity in Tibeto-Burman languages; the case marker is optional on A arguments and in specific contexts is used on some S arguments (also see LaPolla 1995a). He argues that while a language may show a split ergative pattern in elicited forms, in natural discourse, ergative marking is found only in some clauses, usually with pragmatic functions -indicating emphasis or contrast.
636 Edith l. Bavin
25.3.3 Multifunctionality In addition to canonical and pragmatic functions, ergative case morphology may have other functions. It is frequently used for instrumental case, as in Darma (Sino Tibetan) and Warlpiri, as illustrated in (4). In Chintang (Sino-Tibetan; Stoll and Bickel 2013) it marks instrumental case and also cause and source. In Kaluli, the ergative case marking is also used for instrumental case as well as genitive case, and in Mayan languages ergative morphology is also used on nominal possessors (Brown, Pfeiler, de León and Pye 2013). Marking nominals associated with the A argument in Warlpiri is another function for Warlpiri ergative case forms, as shown in (5b) in which wita ‘small’ is associated with wirriya ‘boy’. Even if the A argument is not overt in the clause, as in (5c), associated words will be marked as ergative. In this example, the ergative marker on the locative marked pirli ‘rock/hill’ indicates that the A (ellipsed in this clause) was at that location. (4) Paka-rnu kurlarda-rlu hit-PST spear-INS ‘(Someone) hit (something) with a spear.’ (5) a. Wirriya nya-ngu wita-ngku boy see-PST small-ERG ‘The small one saw the boy.’ b. Wita-ngku nya-ngu wirriya-rlu small-ERG see-PST boy-ERG ‘The small boy saw something/someone’ c. Marlu pantu-rnu pirli-ngka-rlu kangaroo spear-PST rock/hill-LOC-ERG ‘(Someone) on a rock/hill speared the kangaroo.’ In the data from ergative languages reported to date, instrument marking tends to be less frequent in utterances to very young children than ergative case, which means exposure to the instrumental function of ergative forms is reduced, and it is typically not reported to be observed in children’s speech until after it has been used on an A argument.
25.4 Determining when Ergative Marking is Acquired 25.4.1 What Counts as Acquisition? That children have acquired ergative morphology is evident if they map thematic roles onto syntactic roles using the language specific morphology or word order and in the
An overview of the acquisition of ergativity 637 expected contexts. Following Roger Brown (1973), a measure that is frequently used to determine acquisition is the percentage of use in obligatory contexts. Brown chose 90% as indicating acquisition in research on the acquisition of grammatical morphemes in English. When a single ergative form appears, or just a few forms appear, it is not strong evidence that the child has identified the function of ergative marking since the word + case marking could be extracted from the input as a whole unit. However, if the child also used the nominal without the case marker in an S or O role, or if the child produced an appropriate ergative morpheme with a number of different lexical items it does indicate some knowledge of its function, although not necessarily all. When a child marks ergative case but does not use the appropriate allomorph, it is indicative that the child has identified the function, but the distribution of the particular forms is not yet mastered. Although children may use ergative morphology in similar proportions to the speakers who provide input, suggesting mastery of ergative alignment, they may in their early productions use only a limited set of verbs, as reported by Stoll and Bickel (2013) in their study of the acquisition of ergative marking in Chintang. A distribution of ergative markers with a larger set of verbs would provide stronger evidence of generalization, that is, acquisition. In a language that allows ellipsis of core arguments and a single unmarked nominal is included in a child’s utterance, there may be some uncertainty as to whether a child is failing to use an ergative case form on an A argument or appropriately representing an O argument since absolutive case typically has null marking. The potential ambiguity might be resolved from the context, but not necessarily. However, accounts of the acquisition of both absolutive and ergative morphology for languages that use overt absolutive forms in cross-referencing arguments on verbs are available for Mayan languages (Brown et al. 2013) and Basque (Austin 2013).
25.4.2 Sampling Child Language Much of the research on the acquisition of ergative language is based on naturalistic data –children are recorded in daily activities interacting with family members and others. One advantage of this type of data collection is that language input to the children is also available. This provides information about how others use language in the presence of children and which structures the children hear, rather than relying on a description of the mature system. Data from the same children collected over a period of time allows researchers to identify individual developmental patterns. A major problem in collection longitudinal data, however, is that the resources required are not readily available and when such research is conducted, data samples tend to be small, collected from a few children over a limited period. Then there may be few examples of ergative marking in the children’s productions, particularly if the language has frequent ellipsis of core arguments and the proportion of ergative marked nominal in the input is low. With just a few examples the researcher is limited in being able to draw conclusions about acquisition.
638 Edith l. Bavin Stoll and Bickel (2013) were able to collect a large data base in their research on Chintang, so enabling a meaningful comparison of the ratio of ergative use for adults and children. The ratios were found to be similar, as was the distribution in terms of agent versus other functions. However, some differences were evident in usage; for children ergative use was more item specific -with fewer lexical items. Comparisons such as made by Stoll and Bickel are not reliable with only a few examples and since there is variability in language development between individual children, small data sets limit the generalizations that can be made. Pooling comparable data can help overcome the small sample issue. Alternatively, cross sectional data from children of different ages can provide an overview of developmental stages over a longer time period. Individuals do not produce all they know, which justifies the use of elicited data, including narratives, in acquisition research. As well as narratives from adults, they have been collected from children beyond the early stages of acquiring a language—children who are still mastering linguistic properties of the language. Narratives are useful for documenting how reference is maintained in extended discourse, and identifying contexts where argument ellipsis is less likely or more likely to occur (see, for example, Allen 2000; Bavin 2000) and in identifying pragmatic contexts from ergative marking. Comparing acquisition patterns across closely related languages with structural similarities and differences and similar cultural norms allows the researcher more opportunity to focus on whether differences between the languages account for different development patterns. For example, Brown et al. (2013) report on the emergence of ergative marking for children acquiring four Mayan languages. The languages have many similarities -for example they use verbal inflection to cross-reference the ergative argument. In all languages the verb morphology is complex. While there are similarities there are also differences, including the form of the pre-consonant ergative marker and its position in relation to other verb morphology. For two of the languages, the non-syllabic ergative marking used with consonant initial roots was found to be produced later than the ergative marker in the other two languages. As discussed by the authors, the non-syllabic form makes it less salient but, when they are acquired, aspect inflections form a syllable with the ergative marker so making it easier to perceive. Acquiring ergative morphology is not done in isolation; the data show that the acquisition of ergative forms may be combined with or influenced by other morphology in the language. Brown et al. found other language differences in the acquisition of case morphology and suggest these might result from difficulties in processing adjacent affixes, but experimental data rather than naturalistic data would be required to confirm this.
25.4.3 The Social Context While comparing research findings across typologically different languages is important for understanding the factors that influence language acquisition, it can be difficult
An overview of the acquisition of ergativity 639 because of different socialization practices and cultural expectations of the child. It is important to document the culturally specific input and interaction routines between caregivers and children. For any language, social context will influence how much opportunity there is for children to hear, process, and use specific language features. Different types of verbs are used with different activities, for example (see Hoff and Naigles 2002) and this will impact on which case markings are used. Similarly, the use of pronouns in conversations is influenced by the number of participants involved. The linguistic forms used by others in the presence of the child influence the patterns detected and acquired. Communication between child and caregiver(s) is often predominantly biased to first and second person agents. In a language system in which ergative case marking is not marked on these pronouns, there will be reduced opportunity for children to hear ergative case morphology. As an example, in Arctic Quebec Inuktitut, first and second person pronouns are only indicated by verb inflection, not free standing pronouns (Allen 2013), and so ergative case marking will not be available for the child to hear when these pronouns fill the argument slots. Schieffelin (1985) argued that the acquisition of ergative marking on focused pronouns in Kaluli is facilitated by pragmatic scaffolding. Through interactions which socialize children to appropriate behavior exchanges, such as giving and sharing, adults model appropriate communication which clearly distinguish the speaker and listener. In these modelling routines children are exposed to the obligatory contexts for ergative marking with focused pronouns. In Ku Waru, through modeling of appropriate language by adults in multi-turn exchanges with young children and correction when the children omit required ergative markers (Rumsey, San Roque, and Schieffelin), children are exposed to appropriate ergative marking
25.5 Developmental Patterns for Three Ergative Languages: Kaluli, Arctic Quebec Inuktitut, and Warlpiri 25.5.1 Kaluli Kaluli has a split ergative system, dependent on word order and pragmatic factors (Schieffelin 1981, 1985; Rumsey, San Roque and Schieffelin 2013). Neutral (absolutive) case is used with AOV word order unless both the A and O are proper names or kin terms; then the A argument is marked with ergative case. Ergative marking is obligatory on nouns, deictics, and demonstrative A arguments in (O)AV word order. Ellipsis of a noun argument from three constituent clauses is common. A focused set of pronouns is used for ergative arguments of transitive verbs but in AV clauses ergative marking is not required; it is used where new information is provided by the A, as in responses to ‘who’ questions.
640 Edith l. Bavin Ergative marking used with genitive function appeared first in the naturalistic data from the three children studied, aged approximately two years at the start of data collection. Inconsistent marking for A arguments in AV clauses followed; later ergative marking was used on OAV clauses. Some errors were noted from one child. There were some overgeneralizations of A marking to AOV word order, but by age 2;8 the children were sensitive to the word order constraints of ergative marking (Scheiffelin 1985) although for AV clauses the children continued to omit the case marking in contexts where it was appropriate to use it. A trend for using ergative marking in AV order in clauses with past tense, highly transitive, and positive rather than negated verbs was noted. For focused pronouns, no errors were found for OAV clauses and no overgeneralizations were made by using focused pronouns in AOV order. Possible explanations provided by Schieffelin for the apparent ease of acquiring appropriate ergative marking with focused pronoun arguments compared to non-pronominal arguments were: they appear consistently in preverbal position; they are independent, unlike case markers; and they have one function. In addition, first and second person arguments are frequent in interactions with young children, in which conversations are about the here and now and focus on the speaker and listener.
25.5.2 Arctic Quebec Inuktitut Ergative is associated with transitive constructions but ‘detransitivization’ processes can change the role of the arguments for clauses containing a bivalent verb, as in Arctic Quebec Inuktitut. Allen (2013) discusses two main influencing factors influencing the use of ergative marking by children acquiring the language: ellipsis of core arguments and detransitivizing processes. Another possible influencing factor is that ergative case marking is distinguished from absolutive case only in singular forms for nouns and demonstratives, not dual or plural, so the split system could potentially lead to overgeneralisation in the use of ergative case markers. Detransitivizing reduces the opportunities to use ergative case marking. Allen reports naturalistic data from four children aged between 2;0 and 2;10 at the start of the study, which continued over a period of nine months. Only 16 examples of transitive constructions were identified; of these, only seven had overt third person subjects. First and second person A arguments (which do not require ergative case marking) were more common than third person; third person A arguments were typically ellipsed. However, ergative case marking was present when required in the child productions. In contrast, intransitive clauses numbered 197, reflecting the high frequency of detransitivized constructions for bivalent verbs: passive, noun incorporation and antipassive constructions. In antipassives and noun incorporation constructions the agent takes on the S role; in passive constructions the O takes on the S role and the agent is treated as an adjunct. As with the child data, the input showed frequent use of detransitivized constructions with noun and demonstrative agents.
An overview of the acquisition of ergativity 641 In narratives elicited from older children (8–9 and 15–16 years) and adults, ergative case was present when required, but as with the naturalistic data, obligatory contexts were infrequent. Third person A arguments were more frequent in intransitive (detransitivized) constructions than transitive. The data from Inuktitut indicates that frequency of ergative case markers is not necessarily a requirement for acquisition. Frequency has not been found to be predictive of ergative use in other languages also. For example, Pye et al. (2013) found that frequency did not predict acquisition of extended ergativity in 2–3-year- old children. If ergative marking is used infrequently in the input, it may focus children’s attention to the contexts when it is used; this point was made by Stoll and Bickel (2013) in relation to their findings on the acquisition of ergative marking in Chintang. They argue that since argument ellipsis is frequent in that language, overt noun phrases will be salient.
25.5.3 Warlpiri 25.5.3.1 Some Features of the Language Warlpiri has variable word order and there is extensive ellipsis of core arguments, and thus case morphology is not always available to aid the child in identifying the alignment patterns of the language. However, as indicated previously, this may make ergative marking more salient when ergative marked nouns do appear. There are a number of factors that could potentially influence how readily children are able to acquire the distribution patterns for ergative marking. Since the cross-referencing of core arguments with clitics follows a nominative-accusative pattern, as shown in the examples in (2), they will not be of value in determining the case frame of the verb. In addition, there are allomorphs for ergative case and children need to identify that these forms mark the same function but in different environments. The allomorphs children heard in the input at the time of data collection were ngku, nkgi, rlu, and rli, with vowel harmony conditioning the use of rli or rlu and ngku or ngki. Words longer than two syllables had rlu/rli and nkgu or ngki was used on two syllable words. Examples are wati-ngki ‘man- ERG’, karnta-ngku ‘woman-ERG’, maliki-rli ‘dog-ERG’, and wirriya-rlu ‘boy-ERG’. Another potential influencing factor is multifunctionality. The forms of the ergative case marking are also used for instrumental case, as well as agreement between an A argument and nominal associated with it, as illustrated in example (4). There are also homophones for ngku and rli. The ergative form ngku is homophonous with the second person singular cross-referencing clitic, which is illustrated in example (2c) and rli is the first person inclusive dual subject clitic. In conversations and stories there is lot of repetition, using different word orders. In this style of communication some of the information is repeated and more information is added. In the repetitions core arguments will sometimes be overt and sometimes not, which may help focus children’s attention on how the arguments, when they are overt, are marked when included with particular verbs. It also exposes them to the variability of including or not including core arguments.
642 Edith l. Bavin A feature of the language that might assist children in learning which verbs require an A argument, is the small number of verb roots. These are divided into five verb classes based on verb inflections for tense, mood, and finiteness. There is high correlation between valency and verb class; almost all verbs in class one require a single S argument (e.g. nyinami ‘sit’), and most in class 2 have an ergative-absolutive case frame (e.g. luwarni ‘shoot’). The small number of verbs included in classes 3, 4, and 5 include bivalent and univalent verbs. A verb has a specified case frame, with the ergative-absolutive case frame used for prototypical transitive verbs such as pakarni ‘hit’ but others also (e.g. nyanyi ‘see’. However, a verb’s case frame can be modified, resulting in a change of meaning. If the absolutive object of a bivalent verb is replaced by a dative object, the verb is interpreted as attempted action. In addition, a number of preverbs are used in the language and these modify the verb’s meaning and can also modify the case frame (see Nash 1982).
25.5.3.2 The Data Naturalistic cross sectional data Bavin collected naturalistic data from young
Warlpiri children interacting with at least one other child and adults close by, and narrative data from children and adults (previously discussed in Bavin 1992, 2000, 2004, 2013). The naturalistic data showed no use of verbs before the age of 2;0. At age 2;0 locatives, deictics, nouns and some pronouns were frequent. From age 2 to 3 years, there was great variability in how many verbs were produced but the majority were imperatives. More verbs were used overall from age three, and these included a proportion of verbs with ergative-absolutive case frames. A arguments of ergative verbs were not overt until around age three. From the data from the 3-year-olds (N=9) the contexts for use of ergative marking ranged from 0 to 9. With the exception of one young 3-year-old who used no verbs, the number of verb tokens ranged from 10 to 59; for ergative-absolutive case frame verbs the range was a 6–30. That is, the majority of A arguments were not overt, either because the verb form was imperative or the argument was ellipsed. In all obligatory contexts, an ergative case marker was included. For the 4-year-olds (N=6) there were 18–56 verb tokens (15–33 ergative-absolutive verbs) and the contexts for ergative marking on A arguments ranged from zero (for three of the children) to four, and there were also two instrumental contexts and one context for ergative agreement–marking on a nominal associated with a non overt A (watiya-ngku ‘with a stick’). All had ergative marking. The verbs in these clauses varied. They included pakarni ‘hit’, nyanyi ‘see’, ngarni ‘eat’ warrirni ‘look for’, kanyi ‘carry’, and others. Ergative case marking was not restricted to past tense. The child who used ergative forms for instrumental and agreement functions at age 4 had used no ergative forms in an earlier session at age 2;1. However, age is not a reliable way of predicting how much a child will use ergative case forms. One child produced nine ergative forms at age 3;6 but none six months later. Even though 20 ergative-absolutive verbs were included in her utterances, there were no contexts for ergative case to be used. Although children used ergative case when required from about age 3, the allomorph used was not always appropriate. Vowel harmony caused no confusion but length of word as a conditioning factor for which allomorph to use did, for example, ngku was
An overview of the acquisition of ergativity 643 used on words longer than two syllables, even though previously a word had been used with the appropriate rlu case form. It could be that ngku/ngki are heard more in the early input (e.g. on wati ‘man’ ngati ‘mother’ and kurdu ‘child’) and used as the default. In elicited narratives, discussed in the next section, case forms were found to be confused sometimes also by older children.
Elicited narratives Narratives were collected using picture books. The first of two
books comprised eight pictures drawn to show people going hunting, shooting a kangaroo, coming home with the animal, and cooking it. In her narrative, one child (aged 5;2) used six ergative-absolutive verbs; of these five had overt A arguments and all six had an overt object. However, a second child, aged 5;0, used eight verbs but no overt core arguments. Another child aged 6;4 used 12 verbs, including six ergative-absolutive verbs, but only one had an overt A argument, with the eleventh verb she used. It was at a stage in the story when the hunters had returned home with a kangaroo and a woman was cooking the animal on the fire. So there was a new subject (agent), which may have motivated the inclusion of the argument. Other children, 6-and 7-year-olds, also tended to use an overt A with ergative marking for this particular scene although they had not included an A argument elsewhere. Possibly because it was a highlight of the story, the overt A was used as a way of focusing on the event Book 2, with 12 coloured pictures, was part of a published book. Narratives were collected from 28 children aged 4; 8–124 and six adults, parents of some of the children. The six adults used ergative forms (numbering between 4 and 12) with some of these used for instrument and agreement functions. There was a great deal of variation across speakers as to whether and where A arguments were included, from zero instances of seven possible contexts from a 5-year-old to 11 in 12 possible contexts for a 23-year- old. All of the adults included some ergative arguments with appropriate case marking, and the older children used more than the younger children. From 9 years up all but one child used some ergative case markers; below that age, eight of 14 children used at least one. The number of verb tokens included in the narratives ranged from 5 to 41, so some stories were lengthy, even from the younger children. When ergative arguments were included, ergative case marking was used. A arguments were used sometimes when new characters appeared, and sometimes, it seems, to highlight part of the story; and possibly for emphasis or contrast. The data indicate that pragmatic factors need to be taken into account in documenting when children do include ergative arguments when argument ellipsis is common in the language.
25.6 Conclusion Available crosslinguistic research findings show that children with typical development become attuned to features of the input language from an early age. This is evident from
644 Edith l. Bavin data in a number of collected volumes (e.g. Slobin 1985, 1992, 1997; Berman and Slobin 1994; Bowerman and Brown 2008), and journal articles. Ergativity is not represented in the same way across all ergative languages and the acquisition of morphology associated with it, as with the acquisition of other case morphology and verb inflections, is influenced by a number of factors. However, overall, reports on children’s acquisition of ergative languages show clear evidence that by age 3;0 years, often earlier, children are attuned to patterns of ergative marking in the input language. In summary, overgeneralizations are not frequent, although if there are different allomorphs to indicate ergativity some overgeneralizing of one form may occur. Ergative marking may have other functions, but this does not appear to significantly delay acquisition. Ergative markings may also be homophonous with other morphemes, but children seem to keep them separate based on distribution. Frequency of specific constructions in the input can facilitate the detection and acquisition of usage patterns but low frequency of ergative marking in the input, the result of ellipsis or the use of detranisitivized constructions, does not necessarily delay the acquisition of ergative features of a language. Children do not seem to be significantly hindered by having to acquire two systems—one morphological and one syntactic—nor do split morphological systems and variable use of ergative marking seem to be problematic overall, although they may impact on developmental patterns, as shown in Kaluli where ergative marking depends on word order and is not always obligatory. More child language research on ergative languages will provide opportunities for greater understanding about the impact of the type of ergative marking, its position, the conditions for use, variability in use and input styles. One aspect that needs to be considered is that many ergative languages are undergoing rapid change. For example, in Kurmanji Kurdish (Mahalingappa 2013) a shift in the case marking system has resulted in inconsistent input to children. However, by age 2;6 children are reported to use the system in a similar way to adults. Language change occurs in contact situations; an interesting example of this is discussed by Meakins (2015). Gurindji Kriol, an optional ergative language, has emerged from contact between a Pama Nyungan ergative language, Gurindji, and Kriol, a nominative-accusative language. The Gurindji ergative marker merged with the verb system of Kriol, along with other case markers, and its function underwent a series of changes from ergative marker to optional nominative marker. Children are a medium for change and acquisition research of ergative languages in contact and bilingual situations can add to our understanding of factors that influence changes in alignment and changes in the functions of ergative case marking.
An overview of the acquisition of ergativity 645
Abbreviations 1, first person; 2, second person; A, ergative- like subject of a transitive verb; COMP, Complementizer; DAT, dative; DU, Dual; ERG, ergative; INF, Infinitive; INS, instrumental; NPST, nonpast; OBJ, object; PST, past; SUBJ, Subject; S, Intransitive subject; SG, Singular; V, verb. In the chapter O is used for object (when referring to transitive versus intransitive sentences and Subject Verb Object word order).
Chapter 26
The role of de fau lts in the ac qui si t i on of Basque erg at i v e a nd dative morph ol o g y Jennifer Austin
The use of default agreement plays a key role in morphological theories from diverse perspectives, as well as in many analyses of child language acquisition. In this chapter, the development of ergative and dative agreement and case in 20 bilingual and 11 monolingual Basque-speaking children between 2;00–3;06 years old is examined. I propose that the most commonly produced errors in child Basque involve the substitution of unmarked absolutive forms for ergative and dative case and dative verbal morphemes; for independent reasons, the absolutive is considered to be unmarked inflection in adult Basque (Arregi and Nevins 2012). These errors suggest that in early stages of morphological acquisition, children learning Basque use default forms which encode a subset of the morphemes as a “best match” to support their developing language when they are unable to produce or retrieve target forms.
26.1 Introduction The notion of a default, or a linguistic form which surfaces as the result of a general rule in cases where a more specific rule cannot apply, has a long history in linguistic analysis. In Kiparsky’s (1973) analysis of disjunctive ordering in phonological rules, he termed this phenomenon the Elsewhere Condition. In many theoretical approaches to morphology, default inflection also plays a critical role. For example, in the Distributed Morphology (DM) framework, vocabulary items are underspecified and compete with
Acquisition of Basque ergative and dative morphology 647 each other for insertion. Default (elsewhere) morphological forms are selected when there is no exact match between morphosyntactic features and phonological forms (Harley & Noyer 1999). Defaults also figure prominently in the Network Morphology Framework (Brown & Hippisley 2012), which distinguishes between two types of morphological defaults: normal case defaults, which generally apply in the absence of a more specific rule, and exceptional case defaults, which are lexically specified and used to revert back to the default form from a specific rule which otherwise applies (Aronoff 2013). In several theoretical models of child morphological development, default inflection is used as a mechanism for building of morphological paradigms in language development rather than serving as elsewhere forms. According to Bybee (1985, 2007) and Albright (2002), inflectional paradigms usually have a default or base form that children use for creating other parts of the paradigm. While Bybee proposed that this basic or default form is the least inflected member of the paradigm, Albright suggested that the morphological form with the greatest information is selected to construct the rest of the paradigm. In a similar vein, researchers using a constructivist approach view children’s paradigm building as a means to organize and advance the learning of morphology in language acquisition (Dressler & Karpf 1995; Aguirre 2003, 2006; Bittner, Dressler, & Kilani-Schoch 2003; Dressler 2005). In this chapter I will argue that the use of default forms can serve another purpose in child grammars, namely as a kind of repair strategy while children are still in the process of acquiring inflectional morphology. I develop the analysis introduced in Austin (2009, 2010) and propose that children’s inflectional errors in Basque are default forms selected by the Subset Principle (Halle, 1997) as the closest match when a target form cannot be retrieved or produced. My claim is that this approach provides an account of patterns seen in the development of both nominal and verbal inflection. I will begin by presenting the morphological characteristics of Basque that are germane to this chapter as well as evidence for the use of default morphology in adult Basque speakers (Arregi & Nevis, 2012). This overview will be followed by a discussion of data from previous studies of child Basque and the details of my proposal regarding children’s use of default inflection in acquiring Basque.
26.2 Basque Verbal Morphology and Case Basque is a language with an ergative case and agreement system. In Basque, there are distinct agreement morphemes for the subjects of transitive verbs (ergative agreement and case), and for the subjects of intransitive verbs and direct objects (absolutive agreement and case). This pattern is shown in sentences in (1a–b). While the subject of a
648 Jennifer Austin transitive verb such as entzun ‘to listen’ in (1a) is marked with the ergative morpheme –k, the direct object receives absolutive case, which is zero-marked (Ø). In contrast, the subject of an unaccusative verb such as ibili ‘to walk’ in example (1b) has the same absolutive case marker as the direct object of a transitive verb (Ø): (1)
a. Transitive verb: Ni-k berri-a-Ø entzun du-t I-ERG news-DET-ABS hear AUX.ABS.3SG-ERG.1SG ‘I have heard the news’ b. Unaccusative verb: Ni-Ø ibili naiz I-ABS walk AUX.ABS.1SG ‘I have walked’
However, in Basque the subjects of unergative verbs receive ergative and not absolutive case, as seen in example (2). (2) Unergative verb:
Ni-k korritu du-t I-ERG run AUX.ABS.3SG-ERG1SG ‘I have run’
Verbs in Basque are obligatorily inflected for person, number, gender, tense, aspect, and mood (gender is only marked for second person informal forms with the ergative and dative morphemes). In a periphrastic verb, the participle is inflected for aspect, while the auxiliary is marked with the morphemes for person, number, and tense: (3) liburu asko ema-ten d-izki-o-t book many give-IMP AUX-ABS3PL-DAT3SG-ERG1SG ‘I (often) give him/her many books’ The order of the morphemes on a finite Basque verbs (present tense) can be seen in (4): (4)
Absolutive-Root-Dative-Modal-Ergative-Tense
(Laka 1988)
Basque verbal agreement distinguishes between three persons, as well as singular and plural number, as illustrated in Table 26.1 (adapted from Ezeizabarrena 1996: 45).
Acquisition of Basque ergative and dative morphology 649 Table 26.1 Verbal person markers in Basque (present tense)
Prefixes for absolutive agreement (unaccusative argument structure)
Suffixes for dative agreement (dative experiencer argument structure) and ergative agreement (monotransitive argument structure)
1st singular
n- Example: naiz ‘ I (ABS) am’
-t Examples: zait ‘To me (DAT)’ dut ‘I (ERG) have’
2nd singular (familiar)
h- Example: haiz ‘you are’
-k(masculine) -n (feminine) Examples: zaik, zain ‘to you (DAT)’ duk, dun ‘you (ERG) have’
3rd singulara
d- Example: da ‘he/she/it is’
-o Example: zaio ‘to him/her (DAT)’ -Æ Example: du ‘he/she/it (ERG) has’
1st plural
g- Example: gara ‘we are’
-gu Examples: zaigu ‘to us (DAT)” dugu ‘we (ERG) have”
2nd plural
z- + -te (plural marker) Example:
-zue Examples: zaizue ‘to you pl. (DAT)’ duzue ‘you pl. have (ERG)’
zarete ‘you(pl) are’ 3rd plural
d- Example: dira ‘they are’
-e Example: zaie ‘to them (DAT)’ Example: -te dute ‘they have (ERG)’
a Some authors consider -d the realization of tense or mood, rather than a third person agreement
morpheme. See Laka (1988) and discussion in Ezeizabarrena (1996).
Number agreement for plural direct objects takes the form of an infix that is inserted into the auxiliary root, as seen in (5), where the morpheme -it is inserted between the
650 Jennifer Austin auxiliary root prefix d-and before the root vowel u and the final first person ergative marker -t. (5) singular: plural
du-t AUX.Erg1SG ‘I have it’ d-it-u-t AUX.ABS3pl.AUX.ERG1SG ‘I have them’
The two Basque auxiliaries: izan ‘to be’ and edun ‘to have’, are selected according to the argument structure of the main verb. If the verb takes an ergative argument, edun is projected. If it does not, izan is selected. The number of arguments that can be inflected is up to three (for edun). The possible combinations of arguments that can be inflected are shown in (6): (6) Auxiliary 1 argument 2 arguments 2 arguments 3 arguments
izan ‘to be’ ABS ABS/DAT ------ ------
edun ‘to have’ ERG ERG/ABS ERG/DAT ERG/DAT/ABS
26.2.1 Basque Case Marking In Basque three nominal cases can agree with the verb: the ergative, dative, and absolutive cases. An example of a ditransitive verb inflected with all three of these agreement markers is provided in (7): (7) Irakaslea-k ikasle-ei lan asko The teacher-ERG the students-DAT work a lot eman d-i-e. give AUX-ABS3SG-DAT.3PL-ERG.3SG ‘The teacher has given the students a lot of work.’ Examples with the NP neska (‘the girl’) inflected with the ergative, dative, and absolutive case are shown in Table 26.2: Three of these morphemes are homophonous, ending in - k: absolutive plural, ergative singular, and ergative plural. In Basque, absolutive singular case is zero-marked, as is generally true in ergative languages (Dixon 1994; Bittner & Hale 1996a).
Acquisition of Basque ergative and dative morphology 651 Table 26.2 Basque case suffixes Case
Morpheme
Singular (a)
Plural (k)
Absolutive
-Ø
DP: nesk-a ‘the girl’
DP: nesk-ak ‘the girls’
Ergative
-k
DP: nesk-a-k ‘the girl’
DP: nesk-ek ‘the girls’
Dative
-ri, -ei
DP: nesk-ari ‘to the girl’
DP: nesk-ei ‘to the girls’
26.2.2 Default Case and Agreement in Adult Basque Several recent analyses of Basque morphosyntax assume that absolutive, which is zero- marked, is the morphologically default case in Basque (Preminger 2011b; Arregi & Nevis 2012; Rezac et al. 2014). Although Basque absolutive morphology appears in some of Schütze (2001)’s default case contexts (Rezac et al. 2014), others of Schütze’s tests such as dislocation, ellipsis, and gapping cannot be applied Basque, since case matching permits both ergative and absolutive morphology to be in used in non-finite contexts, such as (8) and (9). (8) Left dislocation: Nik, (nik) tipo hau ezagutzen dut. I-ERG I-ERG guy this-ABS know AUX- 1sgERG3sgABS “Me, I know this guy” (9) Ellipsis:
Nork nahi du izozki bat? Who-ERG want AUX-3sgERG3sgABS ice cream one-ABS “Who wants an ice cream?” Nik/ *ni I-ERG/I-ABS “Me.” Nik ez/*ni ez. I-ERG NEG/I-ABS NEG “Not me.”
Both absolutive and ergative case are even possible in contexts where there is no finite verb in the discourse context, such as in the song title “Guk euskaraz, zuk zergatik ez?” (by Urko and Gabriel Aresti): (10) Gu-k euskara-z, zu-k zergatik We-ERG Basque-INS you-ERG why “We (speak) in Basque, why not you?”
ez? NEG
652 Jennifer Austin According to Preminger (2011b), the availability of absolutive case in both finite and non-finite contexts in Basque demonstrates that “absolutive case in freely available in Basque, and is not dependent on agreement with-or even the presence of-any particular functional node” (Preminger 2011b: 13). Arregi and Nevins’ (2012) analysis of the morphological structure of Basque auxiliaries is couched in the Distributed Morphology framework, and outlined in Figure 26.1. SYNTAX Merge & Move Agree-Link Cliticization Absolutive Promotion
POSTSYNTAX Exponence Conversion Agree-Copy Fission ... Feature Markedness Participant Dissimilation Plural Clinic Impoverishment ... Morphological Concord Have-Insertion Complementizer Agreement ... LINEARIZATION Linear Operations Clitic Metathesis and Doubling ... VOCABULARY INSERTION ...
Figure 26.1 Morphological structure of Basque auxiliaries Source: Reproduced from Arregi and Nevins (2012) with permission of Springer.
Rather than relying on the lexicon to supply the syntax with an assembled word, in Distributed Morphology, there are post-syntactic vocabulary items which get selected; each vocabulary item has its own schema. This schema pairs a morphosyntactic feature bundle with the phonological string that represents the morpheme as in (11):
Acquisition of Basque ergative and dative morphology 653 (11) Vocabulary Item Schema signal context of insertion Example: /-k/ [____, +ergative]/DP Under this model, morphemes are joined to form words at any syntactic level through head movement, adjunction, or head merger, rather than via separate word-assembling mechanisms in an independent lexicon. In their analysis, Arregi and Nevins (2012) propose that absolutive arguments are caseless in the syntax and are assigned case features at the Exponence Conversion Stage (post-syntactically). Thus, absolutive case is not related to a specific functional head, unlike ergative and dative case, which are assigned structurally in a particular position in Basque. As seen in (Figure 26.2), Arregi and Nevins claim that dative case is assigned in spec ApplP and ergative case is assigned in spec of vP rather than TP, a hypothesis which is compatible with the existence of both finite and non-finite clauses containing ergative arguments in Basque. Ditransitive sentences
vP v’
AErg VP App1P IODat
v V
App1’
DOAbs
App1
Figure 26.2 Dative case assignment in Basque ditransitives Source: Reproduced from Arregi and Nevins (2012) with permission of Springer.
In addition, Arregi and Nevins argue that Basque auxiliaries are formed of clitics rather than agreement markers which are hosted by C and T, as shown in the auxiliary structure in (Figure 26.3). C T (C1Abs/Dat)
C TAgr
(C1Erg)
C Agr
C
Figure 26.3 The structure of Basque auxiliaries Source: Reproduced from Arregi and Nevins (2012) with permission of Springer.
654 Jennifer Austin The clitics are assembled as bundles of abstract features in the syntax, and then sent to post-syntactic modules where morphological rules arrange them in the correct order, and provide them with a phonological exponent at the Vocabulary Insertion stage, where the exponent with the greatest number of matching features is selected. Based on evidence from Person-Case Constraint effects in Basque, Arregi and Nevins claim the dative and absolutive clitics are in competition for cliticizing to T, and that absolutive agreement (which is default) surfaces when there is no competitor. In section 26.4 I will propose that this theoretical framework which can account for several patterns of inflectional errors seen in child Basque.
26.3 The Order of Acquisition of Case and Verbal Morphemes in Child Basque Research into the acquisition of Basque began with the pioneering work of Larrañaga (1994), Barreña (1995), and Ezeizabarena (1996). As part of a collaborative project between the University of the Basque Country and the University of Hamburg, these researchers collected longitudinal natural speech data from two bilingual children, Mikel and Jurgi. The data were collected every month in Basque and in Spanish between the ages of 1;06 to 4;00 years (Mikel) and 1:06 to 4;01 years (Jurgi). Larrañaga (1994) studied the development of Spanish and Basque nominal case in Mikel from age 1:06 up to the age of 2;06, and found that the use of absolutive case preceded the use of ergative case in child Basque. Ezeizabarrena and Larrañaga (1996) compared Mikel’s production of case to his production of verbal agreement in Basque. They found that Mikel began target-like production of ergative agreement at 2;00, but did not begin producing ergative case regularly (in 87% of obligatory contexts) until 2;04. There was no corresponding discrepancy for absolutive singular case, but because this case is zero-marked in the singular, it is difficult to know for certain when it is acquired. Absolutive singular case and verbal agreement emerged at the same time, in contrast to ergative case and agreement, although the absolutive plural was slow to be acquired both in verbal inflection and nominal marking. Ezeizabarrena and Larrañaga found that pre- consonantal contexts favored the children’s omission of both the absolutive plural determiner and the ergative case marker. In addition, these authors found the ergative case morpheme was omitted more often than the absolutive plural determiner before vowels. They concluded that the protracted development of ergative case in Basque was not due to an inherent difficulty of ergativity per se, since Mikel produced ergative verbal agreement from an early age. Rather, they suggested that there is a delay in discovering the relation between verbal agreement and case marking, leading the child to express the argument structure of the verb through verbal inflection only at first. In research based on these same data, Ezeizabarrena (1996) studied the development of verbal agreement morphology in Mikel and Jurgi. Barreña (1995) studied the development of Basque verbal agreement and case in three children: Mikel, Peru (another
Acquisition of Basque ergative and dative morphology 655 Spanish/Basque bilingual child) and Oitz, a monolingual child learning Basque. The pattern that emerged from these studies was that absolutive verbal agreement was produced first, followed by ergative agreement, and dative agreement last of all (there was a discrepancy between the age at which the two authors considered Mikel to have first used absolutive agreement), as shown in Table 26.3. Table 26.3 The order of production of Basque verbal agreement morphemes in child speech Child
Age of first use of absolutive agreement
Age of first use of ergative agreement
Age of first use of dative agreement
Oitz
2;00
2;02
2;04
Mikel
1:10 (Barreña) 1;07 (Ezeizabarrena)
1:10
2;04
Peru
1;11
2;04
3;00
Jurgi
2;04
2;08
3;03
Sources: Barreña (1995); Ezeizabarrena (1996)
In previous research, I examined the production of finite verbs (Austin 2009, 2010, 2012, 2013) and case marking (Austin, 2007, 2013) in natural speech samples from 20 Basque/Spanish bilingual children and 11 monolingual children learning Basque. These cross-sectional data were collected in the Spanish Basque Country, and information about the children’s MLUs in each language, number of utterances, and length of sessions are provided in Tables 26.A1 and 26.A2 in the Appendix. Austin (2013) found that children acquiring Basque do not produce dative case and agreement in tandem, in addition to the discrepancy between the production of ergative case and agreement reported by Ezeizabarrena and Larrañaga (1996). In particular, many children omitted dative indirect object agreement while producing dative case. Interestingly, some children were able to produce dative experiencer agreement but not dative indirect object agreement. This can be seen in examples from RB, who produces an utterance with missing indirect object inflection in (12a), yet can produce target-like dative experiencer agreement, as in (13). The auxiliary in (12a) is also missing the final morpheme (t) which indicates first person singular agreement. (12) a. *Bai esan-go d-u (t) aita-ri eta amatxu-ri yes tell-FUT AUX.ABS.3SG-ERG.1SG dad-DAT and mommy-DAT ‘(I) will tell Dad and Mommy’ (RB 2;08) b. Target: Bai esan-go di-e-t aita-ri eta amatxu-ri yes tell-FUT AUX.ABS.3SG-DAT3PL-ERG.1SG dad-DAT and mommy-DAT ‘(I) will tell Dad and Mommy’
656 Jennifer Austin (13)
(Ez) z-ai-o (NEG) AUX.ABS.3SG-DAT.3SG ‘(Doesn’t) like’
gusta-tzen. like-IMPERF
(RB 2;08)
26.3.1 Inflectional Errors in Child Basque Although children acquiring Basque begin to use finite verbs before two years of age, their ability to produce target-like morphological forms continues to develop over the course of several additional years, during which time they continue to make inflectional errors. Austin (2009) found that children produced four types of errors with verbal agreement most often: (1) the omission of an entire auxiliary, which occurred in 58/1548 or 4% of their utterances with verbs; (2) the omission of an ergative agreement person marker, which occurred in 44/656 possible contexts, or 7% of the time; 3) the omission of dative agreement indirect object agreement, in 12/23 or 52% of possible contexts; and 4), and the substitution of singular for plural agreement, which occurred in 8/74 utterances (11% of the time). Children also frequently produced case errors in Basque. Bilingual and monolingual children omitted ergative case in 84/188 obligatory contexts (45% of the time). In addition, bilingual children omitted dative case in 4/23 or 17% of obligatory contexts. The most common case error of omission was the failure to produce ergative case with transitive verbs, as seen in example (14): (14) zein(ek) harrapatu which-Ø catch ‘Which has caught it?’
du? AUX.ERG3SG.ABS3SG
[IA 3;02]
There were only two examples in which a child substituted ergative for absolutive case, as shown in (15): (15)
*ni-k, nere ohea-n I-ERG my bed-in ‘I gets in my bed’
sartu da get be.ABS3SG
[TC 2;05]
At younger ages, children substituted infinitives for finite verbs and 3rd person singular forms for other person and number combinations. These results are similar to the patterns of inflectional errors from child Spanish reported in Radford and Ploennig-Pacheco (1995); Clahsen, Aveledo and Roca (2002); Davidiak and Grinstead (2004) in which children substitute non-finite verbs or 3rd person singular verbs for other parts of the verbal paradigm, but not vice-versa.
Acquisition of Basque ergative and dative morphology 657
26.3.2 Factors Affecting the Production of Inflection in Child Basque Phonological difficulty and morphological complexity have been reported to influence the acquisition of case and agreement morphemes in children learning Basque. For phonological reasons, ergative case in Basque may be particularly difficult for children to produce. In Basque, ergative case is marked with the suffix -k, and in adult Basque speech, consonants tend to undergo deletion when they are in pre-consonantal contexts, a process known as Stop Deletion (Hualde 1991; Hualde & Ortiz de Urbina 2003). In early stages of acquiring Basque, monolingual and bilingual children omit the ergative case marker in all phonological contexts (Barreña 1995; Ezeizabarrena and Larrañaga 1996). Children continue to omit this morpheme well after they are producing ergative verbal agreement. In Austin (2007) I compared ergative case-marking in child and adult Basque speech, and found that whereas adults only dropped the ergative case marker in pre-consonantal position, children omitted the ergative case marker in pre-vocalic and utterance-final positions too. That study, which compared natural speech production in Basque in eight monolingual and 20 bilingual children age 2;00– 3:06, also found that bilingual children omitted ergative case significantly more than monolingual ones; bilingual children omitted ergative case in 52% of possible contexts, compared to 31% for monolingual children and 4% of possible contexts in the case of adults. By comparison, dative case, which is marked by a suffix that forms an open syllable (-ri) was omitted to a far lesser extent than ergative case; bilingual children omitted dative case 17% of the time, and dative case was not omitted at all by monolingual children nor by adult speakers (Austin 2013).1 When children omitted ergative or dative, they produced a zero-marked DP, which is homophonous with the absolutive singular form, as seen in (16): (16) inor(k) ez du ikus-ten no one-Ø NEG AUX.ERG3sg.ABS3sg see.IMP ‘No one can see’
(IA 3;02)
Phonological factors may also contribute to the late production of dative indirect object agreement in child Basque relative to other agreement morphemes. Dative inflection is a non-final suffix, and crosslinguistic research indicates that middle affixes are
1
The data examined for this study were collected from the same 20 bilingual and 8 monolingual children examined in Austin (2007) with the addition of three more monolinguals (for a total of 11 monolingual children).
658 Jennifer Austin slower to be acquired than prefixes or suffixes, perhaps in part because they are less phonologically salient (Slobin 1973; Pye 1983; Pye et al. 2007). The third type of agreement error is shown in example (17), the dative agreement morpheme is omitted and a transitive auxiliary is being used instead of a ditransitive one, even though there is a dative marked argument present: (17)
Dative agreement missing: *ba ematen d-e-t txorixu-∅ well give-IMPERF AUX-ERG1SG.ABS.3SG chorizo-ABS ‘Well, I give chorizo (sausage) to the cat’
katu-ari cat-DAT (NC 2;04)
In Austin (2010, 2013) I found that morphological complexity predicted the order in which auxiliaries were acquired in Basque (morphological complexity was defined as the number of morphemes encoded by the auxiliary). In child Basque, verbs that are more complex morphologically (such as ditransitive auxiliaries) were produced later in development than ones that encoded fewer arguments, such as unaccusative verbs. Children produced significantly more root infinitives with verbs that required ergative/absolutive agreement than with verbs that required only absolutive agreement. When children repeated adult utterances (these instances were excluded from consideration in the analyses reported in this chapter), children made the same types of inflectional errors as in their natural speech, such as the omission of dative verbal agreement seen in example (18): (18)
Adult:
Aber zer let’s see what konta-tzen d-io-gu-n! tell-IMPERF AUX-ERG.1PL-DAT.1SG-ABS.3SG ‘Let’s see what we are telling him!’
Child:
Aber xxx kontatu-ko d-egu! let’s see tell-FUT AUX-ERG1PL.ABS.3SG ‘Let’s see what we will tell (him)’
[TC 2;05]
Austin (2013) examined frequency in the adult input as a potential additional factor which could influence children’s production of morphology. As seen in Table 26.4, I found that the order of acquisition of agreement and case morphemes in child Basque was not predicted by their frequency in adult speech:
Acquisition of Basque ergative and dative morphology 659 Table 26.4 Adults’ production of case and agreement in Basque Type of verbal agreement
Ergative
Absolutive subject agreement
Percentage produced by adults
684/1488 45.9%
Order of production by children
Dative experiencer
Ditransitive agreement
573/1488 38.5%
42/1488
83/1488
Second
First
Third
Fourth
Type of case
Ergative case
Absolutive case
Dative case
Percentage produced by adults
100/1488 6.7%
596/1488 40%
17/1488 1%
Order of production by children
Second
First
Third
3%
6%
These results suggest that the order of acquisition of these morphemes in not driven primarily by frequency in the adult input, but by morphological complexity.
26.4 The Use of Default Morphology as a Repair Strategy in Child Basque The two most common inflectional errors in child Basque were the omission of ergative case, seen in (19a), and the omission of dative ditransitive agreement, seen in (19b). (19) a. Omission of ergative case: *Ni(k) baduakat hau etxi-an I-ÿ have-ERG1sgABS3s that-ABS house-in ‘I have that at home’
(LA 3;0)
b. Omission of dative ditransitive agreement: *Bai esan-go d-u (t) aita-ri eta amatxu-ri yes tell-FUT AUX.ABS.3SG-ERG.1SG dad-DAT and mommy-DAT ‘(I) will tell Dad and Mommy’ (RB 2;08)
660 Jennifer Austin In errors such as these, children omit either verbal inflection or nominal case but not both. This suggests that the errors are not due to a syntactic deficit, in which case the child’s utterances would lack finiteness or certain functional categories altogether; instead, these errors seem to be attributable to a problem in producing target morphological forms. In each case, an unmarked absolutive form is substituted for a marked ergative or dative form. I propose that these inflectional errors in Basque find a natural explanation in the Distributed Morphology framework as elsewhere forms that surface when a child has difficulty selecting the appropriate form for whatever reason (phonological difficulty or complexity, for instance). Under such circumstances, the Subset Principle (Halle 1997: 428) allows the child to insert the next best fit, a vocabulary item which contains only a subset of the morpheme’s features: The phonological exponent of a vocabulary item is inserted into a morpheme … if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the vocabulary item contains features not present in the morpheme. Where several vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen.
When no closer match can be found, a default vocabulary item is inserted via the Elsewhere Condition. In example (19a), the child has produced absolutive rather than ergative case. Under the analysis proposed here, the child has not selected the correct phonological exponent seen in (20a), and the absolutive elsewhere form is selected (20b):2 (20) a. ergative: b. absolutive:
/k/ [+motion, -peripheral] /Ø/ [-motion, -peripheral]
The availability of the absolutive elsewhere form when a child runs into morphological trouble would explain why children’s case errors involve substituting absolutive for ergative or for dative, but not the opposite. This outcome (moving from marked to unmarked feature values) is similar to the result of ergative impoverishment rules in adult Basque discussed in Arregi and Nevins (2012). Another possible case error, substituting ergative for dative case (or vice-versa), was never produced by any children. In examples such as (19b) where the child’s utterance is missing dative agreement, the child has selected a morphological form with a subset of the features of the ditransitive target auxiliary. The presence of dative case-marked arguments (aita-ri ‘Dad’and amatxu-ri ‘Mommy’) in this utterance suggests that dative agreement has been generated in the syntax, allowing for dative case to be assigned, but morphologically something has gone astray in producing the target form. In Arregi and Nevins’ theory, this 2
Case features proposed by Calabrese (2008).
Acquisition of Basque ergative and dative morphology 661 case-assigning configuration (KP) would have the structure in (Figure 26.4) for ergative and dative case (nonparticipant arguments are third person), with a clitic generated in the spec of KP. Nonparticipant arguments with dative or ergative case KP Dcl
K’
DPArg –participant
K ±peripheral
–author
+motion
±singular
Figure 26.4 Case assignment in Basque Source: Reproduced from Arregi and Nevins (2012) with permission of Springer.
While there may several reasons why child has substituted default absolutive inflection for the target dative agreement, the presence of dative case suggests that the problem is one of morphological rather than syntactic competence. This analysis proposed here assumes that missing inflection in child language is the result of a morphological system which is still developing rather a syntactic system which has missing or underspecified functional projections, as was proposed by the earliest generativist accounts of inflectional development. These approaches interpreted missing verbal inflection to reflect a lack of syntactic knowledge of inflection on the part of the child, or a truncated clause in early stages of development (until age 2:06 or so) consisting of a VP with the upper functional projections missing (Rizzi 1994). While these earlier theories were based on data from English and other languages with little morphology, they made incorrect predictions for Basque and other highly inflected languages. The Truncation Hypothesis would predict the emergence of inflectional morphemes in children’s language to correspond to the hierarchy of functional morphemes in the Basque inflectional phrase (absolutive-> dative-> ergative). However, the pattern that we find in child Basque is different (absolutive–ergative–dative), and does not correspond to the order which we would expect if clauses were being built in increments from the bottom up. The Optional Infinitive Hypothesis (Wexler 1994, 1995) predicts that there should be no optional infinitives in pro-drop languages, but children acquiring Basque produce them. Another advantage of the analysis proposed here is that it can account for a wide array of children’s inflectional errors, rather than just their production of nonfinite verbs in finite contexts. This account correctly predicts that children’s inflectional errors will be characterized by the substitution of less specified forms for more specific ones. It also predicts that the production of nonfinite forms will gradually decline in child grammars, rather than ending abruptly (Austin 2010).
662 Jennifer Austin
26.5 Conclusions The acquisition of verbal morphology and case in L1 Basque is a lengthy process that proceeds incrementally. Children’s inflectional errors in Basque consist of producing default inflection or morphological forms that contain a subset of the abstract features of the target forms. These errors suggest that in early stages of morphological acquisition children use default forms which encode a subset of the morphemes as a “best match” of the target forms to support their developing language when they are unable to produce or retrieve target forms. Children’s systematic substitution of default morphological forms for more complex ones suggests that they know enough about the properties of the abstract features encoded by the morphemes they are learning to avoid using them in the wrong context. This adherence to the Subset Principle also suggests that they are guided early on by knowledge of the argument structure of the verbs that they are acquiring; children never substituted ergative inflection for dative inflection, for example, despite the fact that ergative agreement is produced more frequently by adults in the input (see Table 26.4). The ability of children to avoid non-subset errors supports the hypothesis that in L1 acquisition “syntax comes first, morphology later” Blom and Wijnen (2006). The inflectional errors that children make in acquiring Basque indicate that early in development, they are generating combinations of verbs and auxiliaries not found in the adult input, rather than using complex morphological forms in lexicalized chunks. Children’s production of these inflectional errors presents a challenge to usage-based models of grammatical development (e.g. Tomasello 2003), since children are producing innovate combinations rather than replicating patterns found in the adult input. The complex possibilities afforded by Basque morphology make visible patterns in morphological development which cannot be observed in the acquisition of more commonly studied languages which have less rich verbal inflection. In the case of an highly inflected language like Basque, for instance, children produce auxiliaries with a simpler argument structure and fewer morphemes earlier than more morphologically complex ones; the acquisition of inflection is not necessarily an “all or nothing” process, as was implicitly assumed in earlier analyses of the acquisition of finiteness in English such as the Optional Infinitive Hypothesis (Harris & Wexler 1996). The ergative nature of Basque morphology also permits us to compare the development of subject and object inflection in child acquisition as well as the effect of verbal argument structure on the acquisition of different types of subjects. The finding that the acquisition of case marking and verbal agreement does not occur in parallel in Basque would not be possible to observe in a language with simpler morphology; for example, the production of ergative agreement precedes the production of ergative case, whereas the opposite pattern obtains for dative case and agreement. Insights such as these highlight the fact that while
Acquisition of Basque ergative and dative morphology 663 research on the development of morphology in English has yielded many important insights that are universally applicable, nevertheless it is important for comprehensive theories of L1 development that we also consider data from less commonly studied languages such as Basque.
Appendix Table 26.A1 Number of utterances and mean length of utterance in each language for bilingual children Child’s initials
Sex
GG
Recording length (in minutes)
Age
Basque utterances
Basque MLU
M
2;01
283
1.85
60
NI
F
2;01
78
1.18
60
NC
F
2;04
85
2.44
60
LH
M
2;05
71
1.95
60
TC
F
2;05
413
1.53
90
AI
F
2;06
205
2.14
60
ME
F
2;06
200
2.65
60
RM
M
2;07
64
2.20
90
IC
M
2;07
20
2.22
60
RB
F
2;08
364
4.55
100
AR
M
2;08
67
1.85
60
OH
M
2;08
362
3.16
90
IU
M
2;10
126
3.93
60
XO
M
3;00
315
3.16
90
LA
F
3;00
151
4.26
60
DG
M
3;01
67
3.15
60
MA
M
3;01
279
2.48
60
IA
M
3;02
237
2.95
100
AM
F
3;02
115
3.33
60
AB
F
3;04
86
3.24
30
664 Jennifer Austin Table 26.A2 Number of utterances and mean length of utterance for monolingual children in Basque
MLU
Length of recording (in minutes)
Child’s initials
Sex
Age
Total utterances in sample
AC
F
2;01
221
1.95
60
JH
M
2;01
126
1.88
60
MA
M
2;03
183
1.73
60
EC
F
2;05
319
1.65
90
AH
F
2;05
162
2.46
60
MC
M
2;08
278
1.80
60
AG
M
3;00
155
3.24
60
EG
M
3;00
117
2.42
45
ME
M
3;01
177
3.67
60
NS
F
3;02
282
5.07
60
AB
M
3;03
257
2.83
60
Acknowledgments Many thanks to the children and parents in the Basque Country who participated in this study, as well as the research assistants who worked with me for their invaluable help with data collection and transcription. I am also grateful to Edith L. Bavin, Susana Bejar, and Lisa Travis for their very helpful comments on this chapter. All remaining errors are my own.
Abbreviations 1, first person; 3, third person; ABS, absolutive; AUX, auxiliary; DAT, dative; DET, determiner; ERG, ergative; FUT, future; IMP, imperative; IMPERF, imperfective; INS, instrumental; NEG, negative; PL, plural; SG, singular.
Chapter 27
A c omparati v e st u dy of the ac qu i si t i on of nom inative and e rg at i v e al ignm ent in E u rope a n and M ayan l a ng uag e s Clifton Pye and Barbara Pfeiler
27.1 Introduction This chapter uses language data on the acquisition of nominative person markers in two European languages as an external test of the nature of person marking in four Mayan languages. Specifically, we compare the acquisition of nominative person markers in French and Spanish with the acquisition of person marking in the Mayan languages Wastek, Yukatek, Ch’ol, and K’iche’. This comparison clarifies the clitic status of the Mayan person markers and provides a better understanding of the acquisition of person marking in all six languages. Spanish uses verb suffixes to mark person, number, and tense (1). (1) Spanish person marking a. Bail-o. dance-1.sg.present ‘I dance.’ b. Bail-as. dance-2.sg.present ‘You dance.’ c. Bail-a. dance-3.sg.present ‘She/he/it dances.’
666 Clifton Pye and Barbara Pfeiler The person marking suffixes in Spanish are phonetically distinct and make the overt use of pronouns unnecessary in most contexts (cf. Grinstead 2004). In comparison with Spanish, the suffixes on French verbs are phonetically indistinct. French makes use of obligatory pronominal clitics to mark person (2). (2) French person marking a. Je danse. ‘I dance.’ b. Tu danses. ‘You dance.’ c. Il/Elle danse. ‘He/She dances.’ The Spanish person marking suffixes have some features in common with the French pronominal clitics. The person suffixes and the pronominal clitics both realize nominative person marking. Neither the suffixes nor pronominal clitics can be used in isolation (3). (3) Question responses in Spanish and French a. Spanish ¿Quién vino? *-a. Who came? He/She. b. French Qui est venu? Who has come?
*Il/Elle. He/She.
Despite these similarities, there are two crucial differences between the Spanish person marking suffixes and the French subject clitics. First, the French subject clitics occur in a preverbal position, whereas the Spanish person suffixes occur in a postverbal position. Second, the Spanish person suffixes may not be separated from the verb stem by a negation marker, whereas French places the negation marker ne between the subject clitic and the verb (4). Kayne (1975) discusses other properties of the French pronominal clitics. (4) Je ne danse I neg dance ‘I do not dance.’
pas neg
In this chapter, we use acquisition data on person marking in French and Spanish to make an operational distinction between agreement and clitic doubling. We compare
The acquisition of nominative and ergative alignment 667 the acquisition data in French and Spanish with acquisition data on person marking in four Mayan languages in order to investigate whether the acquisition of person marking on Mayan verbs resembles the acquisition of nominative clitics in French or nominative agreement markers in Spanish. Our study is partly motivated by generative accounts of language acquisition that attribute different developmental patterns to the affix/clitic distinction in the adult languages (cf. Hamann et al. 1998: 329; Grinstead 2004). Our study is also motivated by proposals that identify ergative or absolutive person markers in Mayan and other languages with nominative person marking (cf. Johns 2006). Proposals that identify ergative marking as nominative include the Absolutive- S-as-Object approach of Trager (1946) and Bobaljik (1993b). Proposals that identify absolutive marking as nominative include the Passive approach of Schuchardt (1895) and Hale (1970) as well as the Inverse approach of Dixon (1972), Trechsel (1982), and Marantz (1984). More complex proposals include the Absolutive-as-Clitic approach of Woolford (2000), the Absolutive as Nominative and Accusative approach of Legate (2006) and the High and Low Absolutive approach of Coon et al. (2014). The precise nature of the distinction between agreement markers and pronominal clitics presents both descriptive and theoretical challenges (Spencer and Luís 2012). Clitics are notoriously difficult to describe precisely because they have properties of both bound and unbound morphemes. Descriptions of person markers use the term clitic or affix on the basis of one or two linguistic features instead of considering the full constellation of phonetic, morphological and syntactic diagnostics. The ambiguity between clitic doubling and agreement presents a challenge for linguistic description as well as for linguistic theory (cf. Kramer 2014a). This challenge is increased by grammaticalization processes that over time convert pronominal clitics to agreement markers. The literature on these topics is large and continues to expand. Generative theories analyze agreement as a realization of phi features on a phrasal head, whereas clitic doubling has been analyzed as the movement of a D head. Kramer (2014a: 593) notes that ‘In principle, these two phenomena are distinct, but in practice they can be difficult to distinguish.’ Spencer and Luís (2012) provide an overview of research on clitics, Corbett (2006) provides an overview of research on agreement and Siewierska (2004, 2011) provides overviews of research on person marking. Koopman and Sportiche (1991) provide a classic discussion of the relation between Case assignment and agreement. Our concern in this chapter is with the acquisition of person marking rather than these theoretical accounts. For this reason, we use the acquisition of nominative person marking in French and Spanish as evidence for the clitic status of the person markers in Mayan languages. The following section of the chapter presents the acquisition data for French and Spanish, and discusses its analysis in acquisition theories. The third section of the chapter presents acquisition data for the ergative person markers in the four Mayan languages. The fourth section of the chapter presents acquisition data for the absolutive person markers in the four Mayan languages. In the conclusion to the chapter
668 Clifton Pye and Barbara Pfeiler we discuss the implications of our findings for research on the acquisition of person marking.
27.2 The Acquisition of Person Marking in European Languages Research on the acquisition of grammatical inflections relies upon a measure of morpheme production relative to the morpheme’s obligatory contexts (Brown 1973). Following children’s percentage use of inflectional morphemes in their obligatory contexts allows researchers to compare morphological development between 2-year-old children, who do not produce many verbal utterances, and 3-year-old children, who produce many utterances during the course of a day. Children acquiring Italian, Spanish, and Catalan produce nominative person marking suffixes on verbs in over 90% of the obligatory contexts by the age of 2;0 (Hoekstra and Hyams 1998; Grinstead 2000). This claim was challenged by Gathercole et al. (1999), who observed that while children acquiring Spanish produce many verbs with subject markers, the vast majority of their verb forms are the third person present tense form (cf. Aguirre 2003). Acquisition researchers have appealed to the contrasting forms of the Spanish person marking suffixes to explain why children produce Spanish verb suffixes so early (Hyams 1986, 1989; Hamann 2002). Hoekstra and Hyams (1998: 87) ascribe Spanish children’s success to the contrasting person markers on verbs, which make visible a tense chain linking a tense operator in Comp to the tense inflection. The contrasting person markers on Spanish verbs allow children acquiring Spanish to rapidly converge on the adult grammar. Grinstead (2000: 132) simply states that ‘The Person Phrase … does appear to be active from the very beginning.’ Wexler (1998) proposed the Unique Checking Constraint (UCC) to account for the Spanish children’s relative success in producing inflected verbs. The UCC references a D-feature on agreement and tense, and proposes that 2-year-olds can only check the D-feature of a determiner phrase against a single functional category, i.e. tense or agreement, but not both. Wexler proposes that the D-feature of nominative agreement in Spanish is [+interpretable] and therefore is not checked. The grammar of children acquiring Spanish only has to check the D-feature of tense and therefore 2-year-old children can produce the agreement markers on verbs without violating the UCC. Children acquiring French exhibit a significant delay in the production of the pronominal subject clitics relative to the Spanish children’s production of person marking on verbs. Hamann et al. (1998: 329) provide information on the forms of the French child Augustin’s production of subject clitics. Table 27.1 shows the frequency of Augustin’s subject clitic production in their obligatory contexts.
The acquisition of nominative and ergative alignment 669 Table 27.1 Augustin’s use of French subject clitics in obligatory contextsa Age
Verbal Utterances
Subject clitic
Null subject
N
%
N
%
Percent use in obligatory context
2;0.2
57
17
29.8
16
38.1
51.5
2;0.23
30
4
13.3
6
42.9
40.0
2;1.15
22
4
18.2
2
25
66.7
2;2.13
55
16
29.1
11
28.9
59.3
2;3.10
45
12
26.6
9
31
50
2;4.1
62
10
16.1
25
52.1
26.5
2;4.22
54
11
20.4
14
36.8
44
2;6.16
116
25
21.6
30
32.3
45.5
2;9.2
175
80
45.7
28
21.1
74.1
2;9.30
115
99
63.4
12
9.6
89.1
a Based on Hamann et al. (1998)
The acquisition theories of Hoekstra and Hyams (1998) and Wexler (1998) also explain why children’s production of French pronominal clitics is delayed relative to children’s production of Spanish person suffixes. Hoekstra and Hyams would claim that the suffixes on French verbs do not make the tense chain visible to children. Wexler would claim that the D-feature of nominative agreement in French is not [+interpretable] and therefore competes with the D-feature of tense which in turn leads to UCC violations. We will use the results of the research on the acquisition of person marking in French and Spanish to analyze the acquisition of person markers in Mayan languages. We assume that if Mayan children acquire nominative person markers as agreement affixes they should display developmental changes that resemble the acquisition of the nominative agreement suffixes on Spanish verbs. In other words, the children should produce nominative agreement affixes in 90% of obligatory contexts by the age of 2;0. If Mayan children acquire nominative person markers as pronominal clitics they should display developmental changes that resemble the acquisition of nominative pronominal clitics in French. In this case, we expect Mayan children to produce nominative pronominal clitics with approximately 50% of verbs up to age 2;6 and in 90% of verbs at 3;0. We state this hypothesis in (5). (5) Child production of nominative person marking 1. Children produce nominative agreement on 90% or more of verbs at 2;0. 2. Children produce nominative pronominal clitics on approximately 50% of verbs up to 2;6 and in 80% of verbs at 2;9.
670 Clifton Pye and Barbara Pfeiler While there are many other factors that affect children’s inflectional development (cf. Pye 1983), the acquisition theories of Wexler and Hoekstra and Hyams tie children’s inflectional development directly to the structural realization of person marking. The hypothesis in (5) should hold to the extent that these acquisition theories are correct. In testing whether these theories account for the acquisition of person marking in Mayan languages we simultaneously test whether the theories provide a complete account of inflectional development. The acquisition data also provide evidence for identifying the Mayan person markers with the nominative person markers of French and Spanish. In the next section, we provide general information about the Mayan children and their language samples.
27.3 The Acquisition of Person Marking in Four Mayan Languages The Mayan language family has 28 languages that are still spoken in Mexico, Guatemala, Belize and Honduras. The family is divided into six major subgroups: Wastekan, Yukatekan, Greater Mamean, Greater K’iche’an, Greater Tzeltalan, and Q’anjob’alan (Kaufman 1990a; Robertson 1992). We will compare children’s acquisition of ergative and absolutive subject markers in languages from four of these major subgroups: Wastek (Wastekan), Yukatek (Yukatekan), Ch’ol (Greater Tzeltalan), and K’iche’ (Greater K’iche’an). In this section, we provide general information about the children and their language samples.
27.3.1 The Language Samples The language samples for the four languages were recorded in and around the children’s homes and each recording session lasted approximately one hour. The participants included the children, various members of their families, the investigators, and visitors. The mothers and siblings were generally present during the recordings, but the fathers only participated occasionally. The families live in rural villages, and the children spend most of their day within the family compounds. The investigators were native speakers of the languages who interacted with the children to different degrees. The recording sessions included play with toys, natural objects, and picture books.
The acquisition of nominative and ergative alignment 671
27.3.2 Wastek Pfeiler and Pye analyzed 13 hours of recordings from a longitudinal database of three children acquiring Wastek living in the vicinity of Aquismón and Tancanhuits, in the state of San Luis Potosí, Mexico. The children were being raised in monolingual Wastek households although Spanish is widespread outside of the home. Table 27.2 provides basic statistics for these recordings. Table 27.2 Basic statistics for Wastek Children ELV
VLA
CAR
No. of utterances
No. of verbal utterances
Proportion
Duration (hrs)
2;3.11
490
199
.41
1
2;5.27
664
401
.60
1
2;11.26
952
742
.78
1
3;0.17
579
353
.61
1
2;2.26
285
220
.77
1
2;6.2
285
143
.50
1
2;6.14
235
139
.59
1
2;11.19
248
131
.53
1
3;0.19
426
211
.50
1
2;4.24
382
100
.26
1
2;4.27
143
55
.38
1
2;7.7
592
289
.49
1
3;0.8
213
137
.46
1
Age
27.3.3 Yukatek The Yukatek recordings were made in Yalcobá in the eastern part of the state of Yucatán, Mexico in the 1990s (Pfeiler and Martín Briceño 1997; Pfeiler 2003). For this study Pfeiler selected 17 hours of data from her longitudinal database of three children at two, two and a half and three years of age (Table 27.3).
672 Clifton Pye and Barbara Pfeiler Table 27.3 Basic statistics for Yukatek Children
Age
No. of utterances
No. of verbal utterances
Proportion
Duration (hrs)
SAN
2;0
380
109
.29
2
2;6
339
98
.29
1.5
3;0
741
125
.17
1.5
2;0
327
100
.31
2.5
2;6
325
64
.20
2.5
3;0
443
84
.19
3
2;0
429
99
.23
1.5
2;6
293
217
.74
1.5
3;0
249
158
.63
1
ARM
DAV
27.3.4 Ch’ol The Ch’ol analyses are based on a set of longitudinal recordings that were made in Tila in Chiapas, Mexico between 2005 and 2010 by native Ch’ol speakers. The investigators recorded three children living in a small neighborhood who were acquiring Ch’ol as their first language. Table 27.4 provides basic statistics for the Ch’ol language samples. Table 27.4 Basic statistics for Ch’ol language samples Children EMA
MAR
Age
No. of utterances
2;1.30
583
186
.32
1
2;6.16
519
147
.28
1
2;9.30
547
211
.39
1
3;0.17
584
228
.39
1
2;0.21
263
32
.12
1
65
17
.26
1
2;5.27
200
21
.10
1
2;11.23
507
187
.37
1
3;0.6
506
168
.33
1
1;11
230
63
.27
1
1;11.18
200
65
.33
1
2;7.3
83
7
.08
1
3;0.3
119
31
.26
1
2;2
MA
No. of verbal utterances
Proportion
Duration (hrs)
The acquisition of nominative and ergative alignment 673
27.3.5 K’iche’ The K’iche’ analyses are based on a set of longitudinal recordings that Pye recorded in the town of Zunil, Guatemala (Pye 1980; 1992). Table 27.5 provides basic statistics for the K’iche’ language samples. Table 27.5 Basic statistics for K’iche’ language samples Children
Age
No. of utterances
No. of verbal utterances
Proportion
Duration (hrs)
TIY
2;1
574
131
.23
2
2;7
594
214
.36
2
2;10
605
255
.42
2
LIN
2;0
501
159
.32
2
CHA
2;9
713
221
.31
2
3;0
825
432
.52
2
27.4 The Acquisition of Ergative Person Marking in the Mayan Languages In this section, we first describe the ergative person markers in Wastek, Yukatek, Ch’ol, and K’iche’ before we present an analysis of the children’s acquisition of ergative person marking. We follow this description with an analysis of how children acquire the ergative person markers in the four languages. We compare the developmental profiles for ergative person markers in these four Mayan languages in order to establish whether the children exhibit a common pattern of ergative person marking. We then compare the children’s ergative person marking to children’s nominative person marking in French and Spanish.
27.4.1 Ergative Person Marking in the Mayan Languages The ergative person markers in Mayan languages have properties of both affixes and clitics. The Mayan ergative markers are traditionally described as prefixes, e.g. Kaufman (1990a: 71); Lehmann (1993) analyzes the ergative markers of Yukatek as enclitics. Mayan languages use the ergative markers to cross-reference the subjects of transitive verbs as well as the possessors of nouns. Woolford (2000: 172) identifies the ergative markers in the Mayan language Popti’ as nominative agreement prefixes. Coon et al. (2014) represent the ergative markers in Mayan languages as agreement prefixes that are marked for inherent case by the transitive v0.
674 Clifton Pye and Barbara Pfeiler The ergative markers in Yukatek, Ch’ol, and K’iche’, but not in Wastek, have allomorphs like affixes. Due to space limitations, we restrict our analysis in this chapter to the acquisition of the preconsonantal ergative allomorphs in Yukatek, Ch’ol, and K’iche’. Brown et al. (2013) present data on the acquisition of the prevocalic ergative allomorphs. The first person ergative marker in Ch’ol has the preconsonantal allomorphs k~j.1 Wastek uses the same set of ergative markers on verbs that begin with consonants and vowels, but requires a separate set of portmanteau markers for transitive verbs with first and second person objects. We excluded the Wastek portmanteau morphemes from our analysis. Table 27.6 lists the preconsonantal allomorphs of the ergative markers in the four languages. Table 27.6 Preconsonantal ergative person forms in Wastek, Yukatek, Ch’ol, and K’iche’ Number
Person
Wastek
Yukatek
Ch’ol
K’iche’
Singular
1
u
in
k~j
in
2
a
a
a
a
3
in
u
u
u
1 inclusive
i
in … o’on-e’ex
la=k~j
qa
1 exclusive
i
in … o’on
k~j … -lojoñ
qa
2
a … (chik)
a … e’ex
la=a
i
3
in … (chik)
u … o’ob
u … -ob
ki
Plural
The ergative markers in Wastek, Yukatek and Ch’ol also have several attributes of clitics. Wastek, Yukatek and Ch’ol, but not K’iche’, allow a limited set of adverbial modifiers to occur between the ergative marker and the verb stem (6). (6) Preverb stem adverbs in Wastek, Yukatek and Ch’ol a. Wastek taam ti-ø k’al-e ø-in ejeet kinin-iy-al (Zavala 1994: 47) then ti-3sg.abs go-it 3sg.abs-3sg.erg slowly pull-tt-inc ‘Then he went pulling it slowly.’ b. Yukatek Ts’o’ok in chéen hats’-ik-ech (based on Verhoeven 2007: 102) term 1sg.erg just beat-inc-2sg.abs ‘I have just beaten you’ 1 All Mayan words are shown in the practical orthography developed by the Proyecto Lingüístico Francisco Marroquín (Kaufman 1976a) with a single exception: we use rather than for the glottal stop. The other orthographic symbols have their standard IPA values except: