The Cambridge Handbook of Role and Reference Grammar 9781107130456, 9781316418086, 2022056326, 2022056327, 9781107571440

Citation preview

The Cambridge Handbook of Role and Reference Grammar Role and Reference Grammar (RRG) is a theory of language in which linguistic structures are accounted for in terms of the interplay of discourse, semantics and syntax. With contributions from a team of leading scholars, this Handbook provides a field-defining overview of RRG. Assuming no prior knowledge, it introduces the framework step-by-step, and includes a pedagogical guide for instructors. It features in-depth discussions of syntax, morphology, and lexical semantics, including treatments of lexical and grammatical categories, the syntax of simple clauses and complex sentences, and how the linking of syntax with semantics and discourse works in each of these domains. It illustrates RRG’s contribution to the study of language acquisition, language change and processing, computational linguistics, and neurolinguistics, and also contains five grammatical sketches which show how RRG analyses work in practice. Comprehensive yet accessible, it is essential reading for anyone who is interested in how grammar interfaces with meaning. delia bentley is Professor of Romance Linguistics at the University of Manchester. She is the author of Split Intransitivity in Italian (2006) and a coauthor of Existentials and Locatives in Romance Dialects of Italy (2015). ricardo mairal usón is Full Professor of Linguistics at Universidad Nacional de Educación a Distancia. He has co-authored or co-edited a number of books including Nuevas perspectivas en Gramática Funcional (1999), which was awarded the Nation Research Price AEDEAN 1999, and Linguistic Universals (with Juana Gil, Cambridge, 2006). wataru nakamura is Associate Professor of Applied Linguistics at Tohoku University. He is the editor of New Perspectives in Role and Reference Grammar (2011) and the co-editor of Objectivization and Subjectivization: Typology of Voice Systems (2012). robert d. van valin, jr. is Professor Emeritus in the Departments of Linguistics at the University at Buffalo (SUNY) and the Heinrich Heine University Düsseldorf. He is the co-author of Functional Syntax and Universal Grammar (with William A. Foley, Cambridge, 1984), the co-author of Syntax: Structure, Meaning and Function (with Randy J. LaPolla, Cambridge, 1997), and the author of Exploring the Syntax– Semantics Interface (Cambridge, 2005).

Published online by Cambridge University Press

cambridge handbooks in language and linguistics

Genuinely broad in scope, each handbook in this series provides a complete state-of-the-field overview of a major sub-discipline within language study and research. Grouped into broad thematic areas, the chapters in each volume encompass the most important issues and topics within each subject, offering a coherent picture of the latest theories and findings. Together, the volumes will build into an integrated overview of the discipline in its entirety.

Published titles The Cambridge Handbook of Phonology, edited by Paul de Lacy The Cambridge Handbook of Linguistic Code-switching, edited by Barbara E. Bullock and Almeida Jacqueline Toribio The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin and Letitia Naigles The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and Julia Sallabank The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt The Cambridge Handbook of Language Policy, edited by Bernard Spolsky The Cambridge Handbook of Second Language Acquisition, edited by Julia Herschensohn and Martha Young-Scholten The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K. Grohmann The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken The Cambridge Handbook of Communication Disorders, edited by Louise Cummings The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul Kockelman and Jack Sidnell The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and Randi Reppen The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger, Gaëtanelle Gilquin and Fanny Meunier The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian Cook The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and Päivi Pahta The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul Dekker The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg Stump The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian Roberts The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald and R. M. W. Dixon The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes Ortega

Published online by Cambridge University Press

The Cambridge Handbook of Systemic Functional Linguistics, edited by Geoff Thompson, Wendy L. Bowcher, Lise Fontaine and David Schönthal The Cambridge Handbook of African Linguistics, edited by H. Ekkehard Wolff The Cambridge Handbook of Language Learning, edited by John W. Schwieter and Alessandro Benati The Cambridge Handbook of World Englishes, edited by Daniel Schreier, Marianne Hundt and Edgar W. Schneider The Cambridge Handbook of Intercultural Communication, edited by Guido Rings and Sebastian Rasinger The Cambridge Handbook of Germanic Linguistics, edited by Michael T. Putnam and B. Richard Page The Cambridge Handbook of Discourse Studies, edited by Anna De Fina and Alexandra Georgakopoulou The Cambridge Handbook of Language Standardization, edited by Wendy AyresBennett and John Bellamy The Cambridge Handbook of Korean Linguistics, edited by Sungdai Cho and John Whitman The Cambridge Handbook of Phonetics, edited by Rachael-Anne Knight and Jane Setter The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching, edited by Hossein Nassaji and Eva Kartchava The Cambridge Handbook of Experimental Syntax, edited by Grant Goodall The Cambridge Handbook of Heritage Languages and Linguistics, edited by Silvina Montrul and Maria Polinsky The Cambridge Handbook of Arabic Linguistics, edited by Karin Ryding and David Wilmsen The Cambridge Handbook of the Philosophy of Language, edited by Piotr Stalmaszczyk The Cambridge Handbook of Sociopragmatics, edited by Michael Haugh, Dániel Z. Kádár and Marina Terkourafi The Cambridge Handbook of Task-Based Language Teaching, edited by Mohammed Ahmadian and Michael Long The Cambridge Handbook of Language Contact: Population Movement and Language Change, Volume 1, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Language Contact: Multilingualism in Population Structure, Volume 2, edited by Salikoko Mufwene and Anna Maria Escobar The Cambridge Handbook of Romance Linguistics, edited by Adam Ledgeway and Martin Maiden The Cambridge Handbook of Translation, edited by Kirsten Malmkjær The Cambridge Handbook of Chinese Linguistics, edited by Chu-Ren Huang, Yen-Hwei Lin, I-Hsuan Chen and Yu-Yin Hsu The Cambridge Handbook of Intercultural Pragmatics, edited by Istvan Kecskes The Cambridge Handbook of Role and Reference Grammar, edited by Delia Bentley, Ricardo Mairal Usón, Wataru Nakamura and Robert D. Van Valin, Jr.

Published online by Cambridge University Press

Published online by Cambridge University Press

The Cambridge Handbook of Role and Reference Grammar Edited by Delia Bentley University of Manchester

Ricardo Mairal Usón Universidad Nacional de Educación a Distancia

Wataru Nakamura Tohoku University

Robert D. Van Valin, Jr. University at Buffalo, The State University of New York and Heinrich-Heine-Universität Düsseldorf

Published online by Cambridge University Press

Shaftesbury Road, Cambridge CB2 8EA, United Kingdom One Liberty Plaza, 20th Floor, New York, NY 10006, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia 314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India 103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467 Cambridge University Press is part of Cambridge University Press & Assessment, a department of the University of Cambridge. We share the University’s mission to contribute to society through the pursuit of education, learning and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781107130456 DOI: 10.1017/9781316418086 © Cambridge University Press & Assessment 2023 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press & Assessment. First published 2023 A catalogue record for this publication is available from the British Library. Library of Congress Cataloging-in-Publication Data Names: Bentley, Delia, editor. | Mairal Usón, Ricardo, 1965– editor. | Nakamura, Wataru, editor. | Van Valin, Robert D., Jr., editor. Title: The Cambridge handbook of role and reference grammar / edited by Delia Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. Van Valin, Jr. Description: Cambridge, United Kingdom ; New York, NY : Cambridge University Press, 2023. | Series: Cambridge handbooks in language and linguistics | Includes bibliographical references. Identifiers: LCCN 2022056326 (print) | LCCN 2022056327 (ebook) | ISBN 9781107130456 (hardback) | ISBN 9781107571440 (paperback) | ISBN 9781316418086 (epub) Subjects: LCSH: Role and reference grammar. | LCGFT: Essays. Classification: LCC P166 .C36 2023 (print) | LCC P166 (ebook) | DDC 415.01/834–dc23/eng/20230313 LC record available at https://lccn.loc.gov/2022056326 LC ebook record available at https://lccn.loc.gov/2022056327 ISBN 978-1-107-13045-6 Hardback Cambridge University Press & Assessment has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Published online by Cambridge University Press

This volume is dedicated to Charles Fillmore and Michael Silverstein.

Published online by Cambridge University Press

Published online by Cambridge University Press

Contents

List of Figures List of Tables List of Contributors Pedagogical Guide Robert D. Van Valin, Jr.

page xi xix xxii xxiii

Introduction Delia Bentley Part One Overview 1 Principles of Role and Reference Grammar Robert D. Van Valin, Jr. Part Two Topics in RRG: Simple Sentences 2 Lexical and Grammatical Categories in RRG John M. Peterson 3 A Conceptually Oriented Approach to Semantic Composition in RRG Ricardo Mairal Usón and Pamela Faber 4 Semantic Macroroles Rolf Kailuweit 5 Grammatical Relations Randy J. LaPolla 6 Argument Structure Alternations James K. Watters 7 Case Assignment Wataru Nakamura 8 Morphology in RRG: The Layered Structure of the Word, Inflection and Derivation Francisco J. Cortés-Rodríguez 9 Adverbs, Mimetics and Ideophones Kiyoko Toratani 10 Adposition Assignment and Adpositional Phrase Types in RRG Sergio Ibáñez Cerda 11 The RRG Approach to Information Structure Delia Bentley 12 Information Structure and Argument Linking Anja Latrouite

and Robert D. Van Valin, Jr.

Published online by Cambridge University Press

1

17

181 218 242 269 292 318 368 403 428 456 488

x

CONTENTS

Part Three Topics in RRG: Complex Sentences 13 The Structure and Semantics of Complex Sentences Toshio Ohori 14 Linking Syntax and Semantics in Adverbial (Adjoined) Clauses Lilián Guerrero 15 Cleft Sentences and Relative Clauses Luis París 16 Extraction Restrictions in Complex Sentences Mitsuaki Shimojo

525 557 591 616

Part Four Applications of RRG 17 Role and Reference Grammar and Diachronic Syntax 18 19 20 21

Ranko Matasovi´c Functional Acquisition Processes in Child Language: Links to Role and Reference Grammar Richard M. Weist Grammatical Aspects of Language Processing in the Brain: A Role and Reference Grammar Perspective Robert D. Van Valin, Jr. Formalization of RRG Syntax Laura Kallmeyer and Rainer Osswald Computational Implementation and Applications of Role and Reference Grammar Brian Nolan

Part Five Grammatical Sketches 22 A Grammatical Sketch of Cheyenne (Plains Algonquian, USA) Avelino Corral Esteban 23 A Grammatical Sketch of Yimas (Lower Sepik, Papua New Guinea) William A. Foley 24 A Grammatical Sketch of Avatime (Kwa, NigerCongo, Ghana) Saskia van Putten and Rebecca Defina 25 A Grammatical Sketch of Amele (Papuan, Papua New Guinea) John R. Roberts 26 Case and Voice in Amis (Austronesian, Taiwan) Joy J. Wu Index

Published online by Cambridge University Press

647 666 693 737 785

823 861 889 914 954 983

Figures

1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10 1.11 1.12 1.13 1.14 1.15 1.16 1.17 1.18 1.19 1.20 1.21 1.22 1.23 1.24 1.25 1.26 1.27

The monostratal nature of RRG page 19 The layered structure of the clause (preliminary) 23 The structure of (3) 23 PrCS and PrDP in English 24 Constituents of the LSC 27 The constituent and operator projections 34 IF indicators in (16) 35 Examples of the ordering of aspect and tense markers 36 in different languages Passive be in (25a) and progressive be and passive be in (25b) 40 Default placement of phrasal adjuncts in Japanese 43 Periphery for individual adverb in (37a) 45 German predicative PPs 49 Non-predicative PPs in English and Bulgarian 50 The layered structure of the MP 52 The structure of the RP in (12a) 53 Structure of RP with multiple modifiers 54 General structure of RPs 58 English RPs 59 The structure of (10b) from Tagalog 59 Some syntactic templates for English 60 Combining syntactic templates 62 Layered structure of the word following Everett (2002) and English refusals 63 The structure of wiˇcháwakat’iŋkte ‘I will beat them to 65 death’ in (55a, a0 ) The structure of (55a) 66 The constituent projection for the multi-juncture 70 utterance in (63) The structure of (64e): English clausal coordination 76 The structure of (65b): English clausal cosubordination with two shared operators 77

Published online by Cambridge University Press

xii

LIST OF FIGURES

1.28 The structure of (65c): English clausal cosubordination with one shared operator 1.29 The structure of (60b): Mandarin nuclear cosubordination 1.30 Proposed structure for (80b) in Wari’ (Everett 2008) 1.31 Interclausal Syntactic Relations Hierarchy 1.32 The Interclausal Relations Hierarchy 1.33 The constituent, operator and information structure projections of the clause 1.34 Syntactic templates required by the output of Step 3 1.35 The output of the semantics-to-syntax linking for (109) 1.36 Summary of the linking from semantics to syntax for (109) 1.37 Constructional schema for German ‘plain’ passive 1.38 Constructional schema for Sama antipassive 1.39 Constructional schema for Japanese plain passive 1.40 Constructional schema for Japanese adversative passive 1.41 Constructional schema for basic clause structure in English 1.42 Constructional schema for English displaced wh-questions 1.43 Constructional schema for English peripheral adjuncts 1.44 Parser output in syntax-to-semantics linking 1.45 Output of step 1 in the syntax-to-semantics linking 1.46 Output of step 2 in the syntax-to-semantics linking 1.47 Output of steps 3 and 4 in the syntax-to-semantics linking 1.48 The linking from syntax to semantics in terms of the algorithm in (128) 1.49 The structure of (148b) 1.50 Core cosubordination in (142a) 1.51 Core coordination in (142b) 1.52 Core cosubordination in (143a) 1.53 Core coordination in (143b) 1.54 Core coordination in (142d) 2.1 Constituent projections for (2a), (2b) and (2d) (adapted from Van Valin 2008: 166) 2.2 Structure of RP with MP containing sentential nucleus (adapted from Van Valin 2008: 173) 2.3 Non-predicative (a) and predicative (b) prepositional phrases in English (Van Valin 2008: 171) 2.4 The Kharia lexicon (Peterson 2011a: 78) 2.5 TAM/Person- and Case-syntagmas with the same semantic base 2.6 The constituent projection of (47) 2.7 Constituent projection of (52) (adapted from Everett 2008: 398) 2.8 The sentence base and sentence modality 2.9 Constituent and operator projections for two English sentences (adapted from Van Valin 2005: 14)

Published online by Cambridge University Press

78 78 81 84 88 115 121 123 124 126 128 131 131 134 135 136 137 137 138 139 140 154 155 156 157 157 159 185 188 189 191 195 197 200 201 203

List of Figures

2.10 Constituent and operator projections in a Japanese sentence (adapted from Van Valin 2005: 14) 2.11 Constituent and operator projections in Kharia 2.12 Layered structure of the RP in English (adapted from Van Valin 2005: 25) 2.13 The structure of the Dyirbal sentence in (65b) (adapted from Van Valin 2005: 29) 3.1 Tripartite concept lattice (Pustejovsky 2001) 3.2 Qualia structure of violin (Pustejovsky and Batiukova 2019: 162) 3.3 FunGramKB Ontological Hierarchy for $STERILIZE_00 3.4 FunGramKB conceptual entry for PASTEURIZE 3.5 FunGramKB Ontological Hierarchy for concepts linked to $OBSOLETE_00 3.6 FunGramKB conceptual entry for $OBSOLETE_00 3.7 FunGramKB Ontological Hierarchy for +INVOICE_00 3.8 FunGramKB conceptual entry for +INVOICE_00 4.1 Actor-Undergoer Hierarchy (modified) 4.2 Four types of languages, accusative and ergative alignment (Haspelmath 2008) 7.1 The constituent structure of the simple clause in English 7.2 Actor-Undergoer Hierarchy (adapted from Van Valin and LaPolla 1997: 146) 8.1 The layered structure of the word 8.2 The LSW of friends’ 8.3 The LSW of the OE complex word bellringestre ‘bell ringer’ 8.4 The LSW of a recursive complex word 8.5 Feature percolation in RPs 8.6 Feature percolation in the LSC 8.7 The LSW of Lakhota verb wiˇchá-wa-k’u 8.8 The LSC of Lakhota sentence wiˇchá-wa-k’u 8.9 The LSC of Spanish sentence cantaban villancicos 8.10 Exocentric analysis of compound bookseller 8.11 Endocentric analysis of compound bookseller 8.12 ‘Acentric’ Spanish compound lavaplatos 9.1 Example of LSC (modified from Figure 1.13, Van Valin 2005: 22) 9.2 Operator projection (adapted from Van Valin 2005: 12) 9.3 Structure of (7a) (modified from Figure 1.14, Van Valin 2005: 22) 9.4 The mimetic as a nuclear adverb (adapted from Toratani 2007: 333) 9.5 The mimetic as a core adverb (modified from Toratani 2007: 334) 9.6 Three possible representations for (27c) 10.1 Universal oppositions underlying clause structure (Van Valin 2005: 4)

Published online by Cambridge University Press

203 207 209 210 223 224 230 232 233 233 235 235 251 258 320 322 372 372 373 374 378 379 380 380 381 392 393 394 405 406 409 420 421 422 429

xiii

xiv

LIST OF FIGURES

10.2 Components of the layered structure of the clause (Van Valin 2005: 4) 10.3 Layered structure of John gave a book to Mary in the library 10.4 Actor-Undergoer Hierarchy 10.5 English non-predicative PP 10.6 English predicative PP 11.1 The cognitive states of referents in discourse (Van Valin and LaPolla 1997: 201) 11.2 The Topic Acceptability Scale (Van Valin and LaPolla 1997: 204) 11.3 Coding of referents as topic and focus (Van Valin and LaPolla 1997: 205) 11.4 Pragmatically motivated positions in the layered structure of the clause 12.1 Components of the focus structure projection (Van Valin 2005: 77) 12.2 (Unmarked) Narrow focus in English 12.3 Revised IS-projection 12.4 The constituent, operator and information structure projections of a sentence 12.5 A simple example of DRSs 12.6 Derivation of predicate focus 12.7 Derivation of (unmarked) narrow focus 12.8a Representation of pragmatic presupposition for (3b) 12.8b Steps 1 and 2 of the semantics-to-syntax linking in (3b) 12.8c Step 3 of the semantics-to-syntax linking in (3b) 12.8d Step 4 of the semantics-to-syntax linking in (3b) 12.8e Summary of linking from semantics to syntax in Japanese verb-less numeral quantifier construction in (3b) 12.9 Summary of linking from syntax and discourse to semantics in Japanese verb-less numeral quantifier construction 12.10 Linking from DRS in cross-speaker ‘VP’-ellipsis in English 12.11 Constructional schema for English conjunction reduction 12.12 Analysis of conjunction reduction in English 12.13 Ungrammatical conjunction reduction in English 12.14 Restricted PFD in some SVO languages 12.15 Syntactic templates for English wh-question and locative inversion constructions 12.16 Split PFD in Italian 12.17 The pervasive role of discourse-pragmatics in grammar 13.1 The layered structure of the clause 13.2 Clausal coordination in Maori 13.3 Core coordination in French 13.4 Nuclear coordination in Tukang Besi 13.5 Core coordination in Turkish with operator projection

Published online by Cambridge University Press

430 430 432 435 435 460 464 464 465 489 490 491 491 492 493 493 496 496 496 496 497

497 499 501 501 502 512 512 513 517 526 528 529 530 532

List of Figures

13.6 Clausal subordination (clause-to-core/complement clause) in Huallaga Quechua 13.7 Clausal subordination (clause-to-periphery/adverbial clause) in Huallaga Quechua 13.8 Clausal subordination (clause-to-noun/relative clause) in Huallaga Quechua 13.9 Core subordination in English 13.10 Nuclear subordination in Khwe (Pavey 2010: 235) 13.11 Clausal cosubordination in Udihe with operator projection 13.12 Core cosubordination in English 13.13 Nuclear cosubordination in French 13.14 Sentential coordination in English 13.15 Sentential subordination (parenthetical) in English 13.16 Sentential subordination (PrDP) in Japanese 13.17 Clausal subordination (PrCS) in Japanese 13.18 Clausal subordination (clause-to-core periphery/ad-core) in English 13.19 Form–meaning correlation in complex structures 14.1 Interclausal Relations Hierarchy (Van Valin 2005: 209) 14.2 Conditional ad-clausal subordination in English 14.3 Simplified structure of reason relations as sentential coordination in Yaqui 14.4 Sequential temporal relations as ad-core subordination in Spanish 14.5 Sequential temporal relations as core coordination in Spanish 14.6 Simplified structure of sentential subordination in Spanish 14.7 Simplified structure of purposive core cosubordination in Yaqui 14.8 Simplified structure of purposive core coordination in Yaqui 14.9 Nuclear cosubordination in Yaqui 14.10 Core cosubordination in Yaqui 15.1 Externally headed relative clauses 15.2 Internally headed relative clauses (from Van Valin 2012) 15.3 Non-restrictive relative clauses 15.4 Pronounless relative clauses 15.5 Cleft sentence 15.6 The contrast between cleft sentences and relative clauses 16.1 Potential focus domain in clausal (daughter) subordination (Van Valin 2005: 214) 16.2 Potential focus domain in ad-core subordination (Van Valin 2005: 216) 16.3 Layered structure of the clause and potential focus domain 19.1 Language-related areas of the left hemisphere

Published online by Cambridge University Press

533 534 534 535 535 537 538 538 540 540 542 542 543 546 564 571 574 576 577 578 583 584 586 586 596 598 600 602 605 609 619 620 623 696

xv

xvi

LIST OF FIGURES

19.2 19.3 19.4 19.5 19.6 19.7 19.8 19.9 19.10 19.11 19.12 19.13 19.14 19.15 19.16 19.17 19.18 20.1 20.2 20.3 20.4 20.5 20.6 20.7 20.8 20.9 20.10 20.11 20.12 20.13 20.14 20.15 20.16 20.17 20.18 20.19 20.20 20.21 20.22 20.23

Simple English transitive linking template Intransitive template (verbal) Intransitive template (non-verbal) Template for three-place predicates Template for ‘double-object’ construction Passive template for transitive verbs Passive template for three-place predicates English subject wh-Q template English non-subject wh-Q template Long-distance wh-question linking template Infinitival complement template Croatian linking template for (9) Croatian linking template for (10) Bock and Levelt’s model of grammatical encoding and RRG Condensed semantics-to-syntax linking RRG as an abstract model of grammar RRG as a language processing system Examples of syntactic representation and syntactic templates in RRG Operator marking by features Periphery marking by features RRG templates Elementary RRG trees and their composition Lexical anchoring of elementary tree templates Schematic sketch of simple substitution (a) and sister adjunction (b) Sister adjunction of periphery elements at different layers Two possible syntactic representations of wh-extraction from complements Wh-extraction via simple substitution (a) and wrapping substitution (b) Wrapping substitution D-edge in final derived tree for Figure 20.10b Discontinuous complex predicates Basic transitive predication template for English with variants Example specifications of syntactic fragments Constituent structure and operator projection for (7) Constituent structure and operator projection for (9) Constituent structure and operator projection for (10) Encoding the operator structure in node features for (9) and (10) Obligatory adjunction of a tense operator Final feature unifications Obligatory adjunction of a tense operator: derived tree Keeping track of the operator projection in edge features

Published online by Cambridge University Press

701 702 702 704 704 705 705 708 709 709 710 713 714 724 725 728 728 738 741 742 744 745 745 746 747 749 749 750 752 754 755 756 759 759 760 761 762 763 764 765

List of Figures

20.24 Derived tree for (10) before (top) and after (bottom) final edge feature unification 20.25 RRG template schemas for non-subordinate nexus types 20.26 Two ways of compositionally deriving cosubordination constructions 20.27 Core cosubordination via sister adjunction: analysis of (13) 20.28 Core cosubordination analysis of (14) derived by wrapping substitution 20.29 Syntactic representations of the examples in (15) 20.30 Composition of templates by wrapping substitution for (15a) and (15b) 20.31 Composition of the syntactic representation of (16a) 20.32 Subordination via simple and wrapping substitution for (17a) and (17b), respectively 20.33 Derived tree for (18) 20.34 Derivation for (18) 20.35 Derivation for (19a) 20.36 Derivation for (21) 20.37 Syntactic tree for (24) 20.38 Derivation for (25a) 20.39 Linking syntax and semantics: derived tree 20.40 Anchoring the default transitive template with ‘smashed’ 20.41 Syntactic composition for (26) after argument linking 21.1 The Vauquois triangle (based on Dorr et al. 2006: 384) 21.2 MT–RRG-based interlingua bridge approach (from Nolan and Salem 2011: 315) 21.3 The RRG interlingua for UniArab (from Nolan and Salem 2011: 315) 21.4 The conceptual architecture of the UniArab system (from Nolan and Salem 2011: 321) 21.5 Dialogue management framework for a language understanding system supporting conversational agents and a dynamic common ground 21.6 Conceptual framework of the conversational software agent (from Panesar 2017: 190) 21.7 The avatar model (from Murtagh 2019b: 95) 21.8 The hand articulator in 3D space with nodes on fingers and thumb (from Murtagh 2019b: 98) 21.9 The architecture of FunGramKB (source: www.fungramkb .com) 22.1 Verb structure in Cheyenne (modified from Corral Esteban 2017: 310) 22.2 Layered structure of the clause with constituent and operator projections 22.3 Representation of direct core arguments (cf. 28) 22.4 Representation of an applicative construction with a derived two-place verb

Published online by Cambridge University Press

766 767 767 768 769 769 770 770 771 772 773 774 775 776 777 778 779 779 794 795 795 796

800 801 802 803 806 825 826 836 837

xvii

xviii

LIST OF FIGURES

22.5 Representation of a clause including an argumentadjunct and an adjunct 22.6 Linking from semantics to syntax in Cheyenne 22.7 Linking from syntax to semantics in Cheyenne 22.8 Information structure-based sentence template 22.9 Unmarked predicate-focus structure in Cheyenne 22.10 Unmarked sentence-focus structure in Cheyenne 22.11 Unmarked narrow-focus structure in Cheyenne 22.12 Interaction between the Person and Semantic Function hierarchies 23.1 Clause structure with constituent, operator and focus projections 23.2 Phrase structure with constituent and operator projections 23.3 Linking from semantics to syntax in (11) of Yimas 23.4 Constituent structure of a non-finite nominalization in Yimas 23.5 Constituent structure of a simple finite nominalization in Yimas 23.6 Constituent structure of a relative clause (28b) in Yimas 24.1 The layered structure of the sentence in example (7) 24.2 A sentence with pre-detached position and pre-core slot (example 11) 24.3 Sentence from example (20) with constituent and operator projections 24.4 Complex RP with constituent and operator projections (example 24) 24.5 Constituent projection of a PP (example 25) 24.6 The Interclausal Relations Hierarchy (Van Valin 2005: 209) 25.1 Basic syntax of the clause in Amele 25.2 The layered structure of the clause 25.3 Optional syntactic structures 25.4 The basic structure of the RP in Amele 25.5 The structure of the possessive RP in Amele 25.6 The PrDP and the potential focus domain 25.7 The PoDP and the potential focus domain 25.8 Semantics-to-syntax linking in a transitive sentence with acc marking 25.9 Semantics-to-syntax linking in an impersonal verb construction 25.10 Semantics-to-syntax linking in a ditransitive predicate 25.11 SVC with cosubordinate core juncture 25.12 SVC with nuclear subordination 25.13 Extended coordinate SVC 25.14 Cosubordinate clause chain structure, example (50) 25.15 Core subordinate DUn clause, example (56a) 25.16 Layered cosubordination in (57)

Published online by Cambridge University Press

837 840 840 843 845 845 846 852 866 867 875 878 883 886 893 894 898 900 900 907 916 923 923 930 931 933 934 936 936 937 939 940 940 943 945 947

Tables

1.1 1.2 1.3 1.4 1.5 1.6 2.1 2.2 2.3 2.4 3.1 4.1 4.2 4.3 4.4 7.1 7.2a 7.2b 7.3 7.4 7.5 7.6 8.1

Operators page 33 Operators in the RP 57 Nexus types 76 Juncture and nexus in complex RPs 92 Lexical representations for Aktionsart classes 100 Valence, macrorole number and M-transitivity 113 Possible alignments between lexical and syntactic 184 categories (Himmelmann 2008: 264) Operators in the layered structure of the clause, from Van 202 Valin (2005: 9) Predicate types with respect to ‘finiteness’ marking (Maas 205 2004: 379, modified) Operators in the layered structure of the RP (adapted 208 from Van Valin 2005: 24) Inventory of conceptual logical structures 229 Logical structures of states and activities and thematic 245 relations (see Van Valin and LaPolla 1997: 115) Actor-Undergoer Hierarchy (see Van Valin and LaPolla 1997: 127, 146; Van Valin 2005: 61) 246 Features and values for activity clusters (from 259 Kailuweit 2013) Activity Hierarchy of arguments of selected predicates 261 Jakobson’s (1936/1984) featural definitions of nominative, 335 accusative, dative and instrumental cases Case syncretism in Kabardian 342 Case syncretism in Palauan 342 Case syncretism in Romanian 343 Case marking of O arguments in Estonian 344 Dative domain in Northern Sámi, Estonian, Finnish and 351 Tolmaˇ ci Karelian Syntax–morphology mapping in Tagalog 354 Operators in the LSW (a partial classification) 375

Published online by Cambridge University Press

xx

LIST OF TABLES

10.1 Semantic units underlying the syntactic units of the layered structure of the clause (Van Valin 2005: 5) 10.2 Lexical representations for Aktionsart classes 12.1 Interaction of focus structure markedness with linking markedness 14.1 Constructional schema for Yaqui ‘topic chains’ 14.2 Constructional schema for ‘I bought frybread in order for me to eat’ in (27a) 18.1 Layered structure of the clause (LSC) and operator scope in RRG 18.2 Lexical representations for four basic Aktionsart classes 18.3 The percentage of verb tokens produced with -ta and with -te i- by Sumihare for two predicate categories and three age periods 18.4a The average age of the emergence of tense-aspect forms for (atelic/telic) predicates (future imperfective) 18.4b The average age of the emergence of tense-aspect forms for (atelic/telic) predicates (‘present’ progressive) 18.5 The average age of the emergence of tense-aspect, tense, and aspect contrasts summing over values of lexical aspect 18.6 The percentage of agreement and tense contrasts for the three predicate types in Polish and English 18.7 Melissa Bowerman’s observations of her daughter Christy’s temporal utterances 18.8 The initial occurrence of the temporal adverbs in three sets of adverbs 18.9 Sentence examples in the acquisition of existential meaning in Korean 18.10 Typology of restricted neutralization of semantic roles (Van Valin and LaPolla 1997: 269) 18.11 Average age and range of initial person agreement contrasts in English and Polish 18.12 The acquisition sequence for canonical and noncanonical constructions 20.1 Operators in the layered structure of the clause (cf. Van Valin 2005: 9) 21.1 The German bracket structure construction (based on Diedrichsen 2014: 109) 23.1 Yimas agreement affixes for local persons 23.2 Yimas agreement affixes for non-local persons 24.1 Paradigms of PSA prefixes (most noun classes omitted) 24.2 Characteristic properties of Avatime SVC subtypes 24.3 Semantic functions of SVCs in each subtype 25.1 M-intransitive posture verbs 26.1 Voice markers and the applicative markers in Amis (adapted from Wu 2006: 289)

Published online by Cambridge University Press

431 431 515 570 584 669 670

672 674 674

675 675 678 679 680 681 682 684 757 790 871 871 895 906 911 917 969

List of Tables

26.2 The relationship between the juxtaposed verb in the BLAQ evaluative construction, SVCs and commentative CCCs in Squlip Atayal: a comparison (from Yeh 2015: 138)

976

TABLEAUS 7.1 7.2 7.3 7.4 7.5

Transitive constructions in Icelandic ‘Dative-subject’ constructions in Icelandic Transitive constructions in Warlpiri Case syncretism in Kabardian Case syncretism in Old Persian

Published online by Cambridge University Press

page 333 333 334 353 353

xxi

Contributors

Delia Bentley – University of Manchester Avelino Corral Esteban – Universidad Autónoma de Madrid Francisco J. Cortés-Rodríguez – Universidad de la Laguna Rebecca Defina – University of Melbourne Pamela Faber – Universidad de Granada William A. Foley – University of Sydney and Columbia University Lilián Guerrero – Universidad Nacional Autónoma de México Sergio Ibáñez Cerda – Universidad Nacional Autónoma de México Rolf Kailuweit – Heinrich-Heine-Universität Düsseldorf Laura Kallmeyer – Heinrich-Heine-Universität Düsseldorf Randy J. LaPolla – Beijing Normal University at Zhuhai, China, and Nanyang Technological University, Singapore Anja Latrouite – Heinrich-Heine-Universität Düsseldorf Ricardo Mairal Usón – Universidad Nacional de Educación a Distancia Ranko Matasovi´ c – University of Zagreb Wataru Nakamura – Tohoku University Brian Nolan – Technological University Dublin Toshio Ohori – Keio University Rainer Osswald – Heinrich-Heine-Universität Düsseldorf Luis París – Consejo Nacional de Investigaciones Científicas y Técnicas John M. Peterson – Christian-Albrechts-Universität zu Kiel Saskia van Putten – Radboud University, Nijmegen John R. Roberts – SIL International Mitsuaki Shimojo – University at Buffalo, The State University of New York Kiyoko Toratani – York University Robert D. Van Valin, Jr. – University at Buffalo, The State University of New York, Heinrich-Heine-Universität Düsseldorf James K. Watters – SIL International Richard M. Weist – SUNY Fredonia Joy J. Wu – National Taiwan Normal University

Published online by Cambridge University Press

Pedagogical Guide to The Cambridge Handbook of Role and Reference Grammar These suggestions are for using the Handbook as a text for beginning and advanced courses in RRG. Exercises for a beginning course can be found in Emma Pavey’s The Structure of Language: An Introduction to Grammatical Analysis (Cambridge, 2010). Exercises for both beginning and advanced courses can be found in Van Valin and LaPolla’s Syntax: Structure, Meaning and Function (Cambridge, 1997).

Introduction: §1.1 The structure of simple sentences: Basic clause structure: §1.2–1.2.1 Lexical and syntactic categories: §1.2.2 Operators: §1.2.3–1.2.3.2, 1.2.3.4 Status of copular be: §1.2.3.3 Peripheries: §1.2.4 Phrasal adjuncts: §1.2.4.1 Non-phrasal adjuncts: §1.2.4.2 Structure of phrases: §1.2.5 Syntactic templates: §1.2.6 Formalization Structure of words: §1.2.7.0 Lexical integrity hypothesis: §1.2.7.1 Head-marking: §1.2.7.2

Beginning

Advanced

x

x

x x (Ch. 2) x (Ch. 2)

x

x x (Ch. 10) x (Ch. 9) x (Ch. 10) x x

x (Ch. 2) x x x (Ch. 10) x (Ch. 9) x (Ch. 10) x x (Ch. 20) x (Ch. 8) x x

Semantic representation of simple sentences Lexical representation: §1.4.2.1 Semantic roles: §1.4.3

x x

x (Ch. 3) x (Ch. 4)

Information structure Basic concepts: §1.5

x (Ch. 11)

x (Ch. 11)

Linking syntax and semantics in simple sentences: Grammatical relations Case assignment Linking from semantics to syntax: §1.6.1

x (Ch. 5) x (Ch. 7) x

x (Ch. 5) x (Ch. 7) x

Published online by Cambridge University Press

xxiv

PEDAGOGICAL GUIDE

(cont.)

Linking from syntax to semantics: §1.6.2 Constructional schemas: §1.6.3 Constructional schemas and linking: §1.6.4 Information structure and linking Patterns of argument realization Semantic motivation and Cross-linguistic variation: §1.6.5 Structure of complex sentences: §1.3 Juncture: §1.3.1 Nexus: §1.3.2 Issues concerning juncture and nexus: §1.3.3 Syntax–semantics interface in complex sentences: §1.3.4 Interclausal syntactic relations hierarchy: §1.3.4.1 Interclausal semantic relations hierarchy: §1.3.4.2 Representing the semantics of clause linkage: §1.4.2.2 Representing constructional meaning: §1.4.2.3 Interclausal Relations Hierarchy: §1.3.4.3 English causative verbs: §1.3.4.4 Complex RPs: §1.3.5 Linking syntax and semantics in complex sentences: §1.7 Modifications of the linking algorithm: §1.71 Argument identity in complex sentences: §1.7.2 Constraints on extraction Applying RRG Historical linguistics Language acquisition Neurolinguistics Computational linguistics Grammatical sketches

Published online by Cambridge University Press

Beginning

Advanced

x x

x (Ch. 6)

x x x x (Ch. 12) x (Ch. 6)

x

x

x (Ch. 13) x (Ch. 13) x (Ch. 13) x

x (Ch. 13) x (Ch. 13) x (Ch. 13)

x (Ch. 13)

x (Ch. 13)

x (Ch. 13)

x (Ch. 13, 14)

x (Ch. 13)

x (Ch. 13)

x (Ch. 15)

x x (Ch. 13, 14) x x (Ch. 15)

x (Ch. 14, 15)

x (Ch. 14, 15)

x

x x (Ch. 16)

x (Ch. 13)

x (Ch. 17) x (Ch. 18) x (Ch. 19) x (Ch. 21) x (Ch. 22–26)

Introduction Delia Bentley

Aims and Scope of the Cambridge Handbook of Role and Reference Grammar For Role and Reference Grammar (henceforth RRG), capturing the extent of variation in the grammars of the world’s languages is as important a goal of linguistic theory as identifying and explaining the properties that all grammars share. The first work which aimed to provide a rigorous, comprehensive and coherent analysis of the syntax of a number of typologically dissimilar languages, introducing many of the constructs and principles which would then be developed into tenets of RRG, was Foley and Van Valin’s (1984) Functional Syntax and Universal Grammar. Almost four decades after the publication of that volume, the Cambridge Handbook of Role and Reference Grammar sets forth to fulfil a more ambitious set of objectives. The volume is an up-to-date presentation of the framework, assuming very little familiarity, if any, on the part of the reader, while also introducing the many developments which RRG has undergone since the publication of the following manuals and collections: Van Valin and LaPolla (1997), Van Valin (2005), Pavey (2010) and Mairal Usón et al. (2012) (for further relevant work see the bibliography available here: https://rrg.caset.buffalo.edu/). In pursuing this objective, the volume is meant to be, quite literally, a handbook, that is, the most complete current treatment of RRG and the first point of reference for any researchers and teachers interested in this framework. Van Valin’s Chapter 1 provides a broad overview of the principles and workings of the framework, while the chapters in Parts II and III adduce more detailed discussions of simple and complex sentences, respectively. Part II deals with the lexicon, lexically motivated alternations, the structure of the word, and semantic decomposition (Cortés-Rodríguez; Mairal Usón and Faber; Peterson; Watters); semantic macroroles and grammatical relations (Kailuweit; LaPolla); important facets of the linking (Latrouite and Van Valin; Nakamura); adpositional, adverbial and mimetic constructions

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

2

DELIA BENTLEY

(Ibáñez Cerda; Toratani), and the encoding of information structure (Bentley; Latrouite and Van Valin). In Part III a discussion of the RRG theory of predicate and clause linkage (Ohori) is followed by chapters on adverbial sentences (Guerrero), relative clauses and clefts (París) and extraction restrictions in complex sentences (Shimojo). Many of these chapters offer specialist contributions on important issues in syntax, morphology, lexical semantics, discourse, and the interfaces between these levels of analysis. Therefore, they will be of interest to researchers who want to compare different perspectives on specific topics in the study of language. Although it remains a principal objective of RRG to provide an adequate set of tools for the description of the syntax of the world’s languages, RRG now also aims to offer an explanatory framework for the study of language acquisition, language change and processing and computational linguistics, and it has made contributions in the domain of neurolinguistics. The extensions of RRG to these fields are presented in Part IV: Matasovi´ c argues that the constructs of RRG, which were primarily developed to conduct synchronic work, also lend themselves to capture several aspects of language change. Weist explores the acquisition of key components of the RRG architecture of grammar within a cognition and communication approach to language acquisition. Van Valin uses the tools of RRG to explain the ability of split-brain patients to provide grammaticality judgements with their isolated right hemisphere, developing a proposal which could potentially also capture the decoupling of grammaticality judgements and interpretation in agrammatic aphasics. Finally, Kallmeyer and Osswald develop an RRG system for the formalization of syntactic and semantic composition operations, with potential advantages for the computational implementation of the framework, while Nolan discusses the merits of RRG in natural language processing. This part of the handbook will give the reader an opportunity to evaluate the potential of RRG in applied domains, and to gauge the psychological plausibility of the framework, as compared with approaches which make very different analytical assumptions. Over the years, the framework has been adopted by a large number of fieldworkers and scholars committed to the documentation and investigation of lesser-known and endangered languages. The concluding chapters, in Part V, are prime examples of this endeavour, featuring grammatical sketches of languages spoken in Papua New Guinea (Yimas, Lower Sepik, described by Foley; Amele, Papuan, treated by Roberts), Africa (Avatime, Kwa, Niger-Congo, discussed by van Putten and Defina) and North America (Cheyenne, Algonquian, dealt with by Corral Esteban), as well as a treatment of voice and case in Amis, Austronesian, by Wu. The handbook is aimed at a very large readership. Along with those who wish to adopt the RRG framework in their own research, this includes all advanced researchers in syntax, morphology and the discourse–semantics– syntax interface, teachers of syntax at higher education institutions, computational linguists, cognitive neuroscientists, and linguistic fieldworkers.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

It is hoped that these different audiences will find the handbook instructive and stimulating, and that the descriptions and analyses presented here will engender constructive scientific debate both within the RRG community and further afield. The volume is the result of long-term collaborative efforts of many scholars worldwide. The editors would like to thank the reviewers of the handbook proposal, and of the individual chapters, the many chapter authors, and last but by no means least, the whole community of RRG linguists, who enthusiastically and unfailingly supported the project through its many stages and vicissitudes, helping the editors to see it to its successful completion. At Cambridge University Press, the editors thank Helen Barton, Isabel Collins and Stephanie Taylor for their prompt, effective and friendly support.

RRG in Modern Linguistic Theory An important concern of RRG is to engage in cross-theoretical debate. To give but few examples, many chapters in Van Valin and LaPolla (1997) include discussions and mentions of relevant work in different theoretical approaches, while Van Valin (2001) systematically and explicitly compares the perspectives of Relational Grammar, Principle and Parameters theories and Lexical Functional Grammar with that of Role and Reference Grammar (see also Butler 2005a, b). As noted in Van Valin (2009), this type of evaluative and contrastive debate, which had previously been central in linguistics, has become less intense since the late 1970s, although we note that Farrell (2005) and Dalrymple (forthcoming) (including Bentley and Vincent forthcoming) offer prime examples of such comparisons. In this section, we therefore reflect on the place of RRG in modern linguistic theory without aiming at exhaustivity, but rather selecting some of the themes which are at the very core of the RRG conception of the architecture of grammar and assessing the perspective of RRG vis-à-vis that of other approaches. The principal characteristic of RRG which we hope to highlight in the discussion is that it seeks to explain the similarities and differences in the syntax of the world’s languages in terms of the interplay of syntax with lexical-semantic and informationstructural representation, and it is with reference to the interfaces that it makes its predictions. Therefore, RRG is a prime example of those that Jackendoff (2002) called parallel architecture theories, that is, linguistic theories which keep syntax separate from the other levels of analysis while placing the interplay of these levels at the forefront of linguistic investigation. Alongside Lexical Functional Grammar (Bresnan 1982, 2001; Börjars et al. 2019) and Head-Driven Phrase Structure Grammar (Pollard and Sag 1994; Sag and Wasow 1999), RRG is also one of the monostratal syntactic theories, which means that it does not rely on movement, or promotions and demotions, and, instead, it derives any cross-constructional comparisons and generalizations from facets of the linking of syntax with semantics and discourse.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

3

4

DELIA BENTLEY

Whilst placing due emphasis on relational constructs, RRG has distanced itself since its very inception from the framework which claimed grammatical relations to be primitives of syntactic theory and universals of human language, namely Relational Grammar (for Relational Grammar see Perlmutter 1983; Perlmutter and Rosen 1984; Perlmutter and Joseph 1990; for relevant discussion, see Van Valin 1977, 1981; Foley and Van Valin 1984: 388–389; Van Valin and LaPolla 1997: 242–285). RRG recognizes that individual languages may privilege the argument which bears unmarked case or is placed in a particular position in the clause, treating it as the source of agreement relations, the antecedent in anaphoric relations, and the controller – or the controllee – in interclausal cross-reference relations. However, no principle of RRG syntax requires that such coding and behavioural properties, or any language-specific subsets thereof, should cluster together, thus defining the subject, the object, etc. of a given language (see Chapters 1 and 5). Grammatical relations are thus not universal in RRG (see Chapter 5 and, for a discussion of comparable views, Farrell 2005: 14–38), but rather have to be defined with reference to specific constructions, where a restricted number of semantic relations or pragmatic functions is neutralized for syntactic purposes. It is worth pointing out here that although Lexical Functional Grammar has a comparable notion of semantically unrestricted grammatical functions – subject and objects (Börjars et al. 2019: 332), this framework differentiates between restricted and unrestricted functions, whereas no grammatical relation is postulated in RRG, if there is no restriction. The restricted neutralizations of semantic relations which are relevant to the definition of grammatical relations in RRG are captured with reference to the generalized semantic relations actor and undergoer, which are universal. These macroroles are defined on the basis of a hierarchy of five grammatically salient positions in the semantic representation of predicates (Van Valin and LaPolla 1997: 146). Actor generalizes across positions at the high end of the hierarchy, while undergoer generalizes across positions at the low end (see below and Chapters 3 and 4). The alignment of the privileged syntactic argument of a construction with the actor, or with the high end of the said hierarchy, characterizes accusative alignment, whereas ergative alignment aligns the privileged syntactic argument of a construction with the undergoer, or the low end of the hierarchy. Passive and, respectively, antipassive voice flag the marked choice in each alignment type (see Chapter 26). While the construction-specific account of grammatical relations readily captures split alignment, some languages, which are well represented in the Indo-European family, tend to select the actor as the privileged syntactic argument consistently across constructions. The notion of subject ensues precisely from this consistency, and it is in terms of this consistency that it can be understood and defined in RRG (Van Valin and LaPolla 1997: 175), although it has no cross-linguistic validity. The subject in English and other languages is defined by the restricted neutralization (A(ctor of transitive),

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

S (actor or undergoer of intransitive), derived-S (of passive)), which, crucially, leaves out U(ndergoer of transitive). The RRG account of grammatical relations as the synchronic neutralization, and the diachronic grammaticalization, of semantic relations turns out to be cogent in the analysis of all-focus constructions, where the controller of V-S agreement cannot align with the topic and is instead purely selected on semantic grounds (Bentley 2018; Bentley and Cennamo 2022). Clause structure has distinctive properties which set RRG apart from most syntactocentric and parallel architecture theories alike. The framework aims to do justice to the aspects of phrase structure that are comparable across languages, while also representing the broad range of variation observed in the phrase structure of the world’s languages. Since all languages distinguish structurally between predicating and non-predicating elements, the clause is composed of the following units or layers: (i) the Nucleus, which hosts the predicate, (ii) the Core, which includes the Nucleus and the referential phrases required by the predicate (the core arguments), and (iii) a Periphery for each layer, containing adjunct modifiers of that layer (see Chapters 1, 9 and 10). These three semantically defined layers are the only universal components of the clause. The building blocks of the clause, and hence the nodes of the syntactic projection called constituent projection, are not named after the major phrasal categories NP, AP, VP. This is in part the consequence of the absence of rewrite rules like S ! NP VP, although the main rationale of this virtually unique feature of RRG syntax is the empirical observation that noun phrases can predicate, although they are normally referential, and in fact in some languages they predicate without the support of a verb, while, in turn, verbs can be referential expressions (Van Valin 2008: 163–164 and Chapter 2 of this volume). Grammatical relations have no configurational definition in RRG (compare the notions such as Specifier of IP in Chomskyan generative theories) and the VP is ruled out as a universal feature of clause structure, as is also the case with Lexical Functional Grammar (Börjars et al. 2019: 5–6). To capture discontinuous constituency and the languages with completely unconstrained word order, there is no no-crossing condition in the constituent projection. In addition, the bound pronouns of head-marking languages, as well as the person and number morphology borne by the verb in null-subject languages, are linked to the Core node in the constituent projection, thus reflecting their referentiality and, ultimately, their status as core arguments. As was briefly mentioned, RRG syntax cannot rely on movement. Rather, each language has an inventory of syntactic templates, which are drawn upon in parsing to differentiate passive structures from their active counterparts, wh-questions from their declarative counterparts, etc., while the correspondences between these pairs of structures are captured at the interfaces of syntax with semantic representation and information structure. The syntactic templates in the syntactic inventory of each language

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

5

6

DELIA BENTLEY

also reflect the linear ordering requirements of the language and the whole range of clausal layers that it is endowed with (for the non-universal ones, which symmetrically occur to the left and the right of the Core, see Chapter 1). These templates are one of the constructional features of the framework, and we shall return to other such features below. The scope of aspect, negation, tense, deontic and epistemic modality, etc. over different layers of the clause is a language universal for RRG (see Chapter 9 for similar considerations regarding the scope of adverbs and ideophones). Indeed, Foley and Van Valin (1984: 208–224) (see also Van Valin and LaPolla 1997: 46–51; Van Valin 2005: 12) make the strong empirical claim that the linear order of operators in syntax follows from their scope. This is the Natural Serialization Principle, which was supported by the results of a typological survey reported in Bybee (1985). Thus, operators with narrower scope (e.g. aspect) are predicted always to be closer to the nucleus than operators with wider scope (e.g. deontic modality). Although similar claims are made in Functional Syntax (Dik 1978, 1980, 1989), as well as in Cartographic proposals (Cinque 1999), it is essential to note that the constituent projection can only host predicative and referential units in RRG, and, thus, it cannot accommodate any functional projections, comparable to IP, or indeed Mood(. . .)P, Tense(. . .)P, Aspect(. . .)P, etc. Instead, the universal, and strictly hierarchical, array of operators is represented in the operator projection (see Chapter 1). The fact that operators are represented separately from the clausal layers upon which they have scope has brought to light an interesting, and hitherto unexplained, property of language change (Matasovi´ c 2008 and Chapter 17). In accordance with the principle of unidirectionality, which has been explored from many theoretical perspectives, operators of narrower scope tend to grammaticalize as operators of broader scope (aspect > tense; deontic modality > epistemic modality). In RRG terms, this means that nuclear operators can only develop from nuclear operators, but can themselves develop into nuclear, core or clause operators. In turn, core operators can only develop from nuclear or core operators, but can themselves develop into core or clause operators. This is what Matasovi´ c (2008) calls the centrifugal direction of grammaticalization. However, the grammaticalization of verbal forms proceeds from lexical units to clitics and affixes, thus moving in a direction which is, in effect, centripetal, and opposite to that of the change in semantic scope. Thus, by disentangling operators from the syntactic units upon which they have scope, work in RRG has uncovered the opposite directions of the two unidirectional processes which characterize grammaticalization. This is an important contribution of the framework to the study of language change. The layers of clause structure, and the respective operator scope, play a key role in the RRG theory of predicate and clause linkage (for which see Chapters 13, 14, 15, 16). This theory relies on a distinction which, to our knowledge, has no direct equivalent in other frameworks, namely that

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

between nexus and juncture. Nexus is the relationship established between two clausal layers: RRG makes a trifold distinction between coordination, co-subordination, and subordination, the last of these being further subdivided into complement and adverbial subordination, in accordance with other scholarship. Juncture amounts to the issue of which layers are joined together by one of the nexus types. All in all, there are nine nexus–juncture combinations, and operator scope is one of the principal diagnostics that are available to identify them. Not only has RRG produced important insights on predicate and clause linkage cross-linguistically, for example by highlighting differences in the number of nexus–juncture combinations which are available across languages, but it has also advanced knowledge on the interplay between the semantic relation between the units in a construction (causative, aspectual, psych-action, purposive, etc.) and the degree of syntactic cohesion that is established between them. Building upon Silverstein (1976) and Givón (1980), RRG has in fact developed an Interclausal Relations Hierarchy (Van Valin and LaPolla 1997: 481–483; Van Valin 2005: 209, see also Casti 2012), which juxtaposes an array of semantic relations with a range of nexus–juncture types, both being listed in decreasing order of cohesion. The mapping between the two sides of the Interclausal Relations Hierarchy is many to one, and, in fact, the hierarchy brings to light the possibility and tendency for the more cohesive semantic linkage types (e.g. causation) to be expressed not only by tight morphosyntactic linkages, but also by looser ones. Importantly, RRG makes the strong falsifiable prediction that the tightest syntactic linkage realizing a particular semantic relation in a given language should be higher than or as high as the tightest syntactic linkage realizing lower semantic relations on the hierarchy in the same language. In light of the pivotal role played by predicate and clause linkage in the syntax of natural languages, it is to be hoped that the work conducted in RRG will inspire further research on this topic in the future. Although we cannot go into any of the relevant details (for which we refer the reader to Chapter 8), we wish to mention here that RRG has a theory of word structure, which parallels its theory of clausal structure. The three layers Nucleus, Core and Word are projected onto the constituent projection of the word, while the operators which have scope over each of the word layers figure in the operator projection of the word. The RRG approach to inflectional morphology has been characterized as inferential-realizational, in the sense of Stump (2001), and an interesting debate which has taken place in recent years concerns this type of morphology. Whereas some place inflection in the constituent projection of the word as daughters of the Core layer (see Everett 2002; Van Valin 2013), Martín Arista (2009: 90) proposes instead to treat inflection as part of the operator projection. We dedicate the concluding part of this introduction to the interplay of syntax with lexical semantics and discourse. RRG has a bidirectional semantics–syntax and syntax–semantics linking, which attempts to

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

7

8

DELIA BENTLEY

reproduce the process of language production and language comprehension, respectively. The steps of the linking that are concerned with semantics, called ‘the lexical phase’, are argued to be more regular and comparable cross-linguistically than the other steps of the linking, called ‘the morphosyntactic phase’. The claim is therefore that the more semantically motivated a linguistic phenomenon is, the less cross-linguistic variation there is, and vice versa (see Section 1.6.5). Discourse constitutes an independent component of grammar, whose role is pervasive in the linking. Indeed, the interplay of discourse with syntax and semantics is claimed to be the site of a great deal of cross-linguistic variation (Van Valin 2014): different languages may encode the same discourse roles in different ways: prosodically, syntactically, morphologically and even by particular lexical choices. There is, therefore, no universal association of syntactic positions or projections with specific discourse functions (see notions such as Top(ic)P, Foc(us)P, etc.). Such associations are of course admitted, but only on a language-specific basis. In fact, the only analytical assumption that the theory makes about the discourse–syntax interface is that the outermost positions in the layered structure of the clause, which are not universal, may host topics and afterthoughts, while foci figure closer to the core than topics do. From this point of view, the RRG treatment of information structure differs substantially from that of other syntactic theories, notably Cartography (Rizzi 1997 and subsequent literature), while other parallel architecture frameworks have not, in the past, devoted as much attention to the interaction of syntax with discourse as RRG has (see Zaenen forthcoming for Lexical Functional Grammar). As will be explained in Chapter 11, the RRG understanding of information structure draws heavily upon Lambrecht’s (1994) distinction between, on the one hand, the role played by each information unit in pragmatic presupposition and assertion, and, on the other, the status of the discourse referent of each information unit in the minds of the discourse participants, including how this status changes in discourse or text. Discourse in RRG is also crucially involved in the retrieval of arguments and predicates that are not overtly expressed in syntax, such as the silent predicates of Japanese and the null arguments of pro-drop languages which lack rich morphology, for example Chinese. While ruling out phonologically null elements in its constituent projection, RRG has a Completeness Constraint, which requires that all the arguments that are specified in semantic representation must be represented in syntax, and vice versa. RRG thus resolves the challenging case of what in other frameworks are analysed as phonologically silent elements by allowing arguments and predicates to participate in direct semantics–discourse and discourse–semantics linking. To this effect, it adopts the formalism of Discourse Representation Theory (Kamp and Reyle 1993). We shall not continue this discussion here, but rather we refer to Chapter 12 and the work of Shimojo (2004, 2008, 2016), among others, for further discussion.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

By way of conclusion, we consider the contribution of RRG to the discussion which has come to be known as the projectionist–(neo-)constructionist debate (see Levin and Rappaport Hovav 2005 for an overview). RRG clearly belongs to the projectionist theories of grammar, although constructions also play a role in the linking, as will be pointed out below. While fully embracing the programme launched by Gruber (1965) and Fillmore (1968), which aimed at deriving the syntactic realization of the arguments from semantic properties which they have qua arguments of a particular verb, RRG also began to address the problems faced by the theories of thematic roles earlier than such problems came to the fore in the scientific debate. We refer in particular to the difficulty of diagnosing thematic roles, and the vexed issue of the granularity and the ordering of such roles in thematic role hierarchies. Thematic roles were never defined intuitively in RRG. Rather, drawing upon Jackendoff (1976) and Dowty (1979), Foley and Van Valin (1984: 47–63) considered the semantic relations of arguments to be derivatives of decomposed predicate structures. The positions of arguments on the decomposed structure of their predicate defined a cline of accessibility to each macrorole status, actor or undergoer, and the syntactic realization of the arguments was captured in terms of these generalized semantic relations. These ideas were later developed into Van Valin and LaPolla’s (1997: 90–158) fully-fledged theory of lexical-semantic representation and macrorole assignment, where labels such as agent and patient are nothing but mnemonics (ibid., p. 116). In this theory, there are only five universal, hierarchically arranged, thematic positions that are relevant to macrorole assignment and, therefore, to the syntactic treatment of the arguments. It is, therefore, our contention that RRG takes care of the vexed questions mentioned above, viz. the diagnostic problem and the granularity and ordering issues. Of course, it does so by making precise analytical choices, and, in particular, by relying on a system of lexical-semantic decomposition which takes the Vendlerian Aktionsart classes as foundational (Vendler 1967[1957]; see Van Valin 2005: 42 for the addition of semelfactives; Smith 1997). While we do not think that the criticisms which have been advanced against theories of thematic roles and thematic hierarchies could justifiably be raised against RRG, it is undeniable that the linking proposed by this approach is less economical than the single projection proposed in purely syntactic theories of event structure (see, by way of example, Borer 2005a–c; Ramchand 2008). While valuing Occam’s razor in scientific investigation, RRG seeks to address the question of how the interaction of syntax with semantics and pragmatics in different grammatical systems can best be captured. It is the very pursuit of this goal that leads RRG to adopt the less economical solution. Suffice it to mention the contrast between different types of alignment: this speaks in favour of the disentanglement of the event structure hierarchy from its syntactic realization across languages. Whilst

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

9

10

DELIA BENTLEY

accusative alignment privileges the high end of the hierarchy in syntax (see above), ergative alignment privileges the low one. In split alignment, the high end provides the privileged syntactic argument of some constructions, though not others. The event structure hierarchy is one and the same in all grammars, and, indeed, there is reason to consider it to be the linguistic correlate of a cognitive universal. However, the study of argument realization across languages indicates that both ends can be syntactically unmarked in different languages or constructions, and this, in our view, supports the parallel architecture approach. Interestingly, evidence from neuroscience also corroborates the idea that the lexicon is an independent module in linguistic competence (see Chapter 19). Therefore, the linking will remain at the very centre of the RRG conception of grammar, and indeed work in RRG has over the years sought to refine the system of lexical decomposition which the event-structure hierarchy is a derivative of (Mairal Usón and Faber 2002; González-Orta 2002; Van Valin and Mairal Usón 2014; Bentley 2019, among others). As for the role of constructions, these are templates which define the unique syntactic, morphological, semantic and pragmatic features of individual constructions in individual languages. These templates – called ‘Constructional Schemas’ – need not include any of the general principles which are valid in the grammar of a given language, or across languages, but rather are sets of specific instructions which, combined with the general linking principles, constitute the grammar of a particular language. Constructions thus play a different role in RRG than in Construction Grammar (Goldberg 1995, 2006), and, crucially, they do not in any way reduce the role of the linking or of its components, including the lexicon. In this section we have introduced key aspects of the formalism developed by RRG to explore how different languages express linguistically salient meaning and communicate it in context. We have shed light on similarities and differences with other frameworks, and we have reflected on the role of RRG in current debates in linguistic theory. We hope that these observations will encourage researchers of various theoretical persuasions to read further.

References Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence from Italo-Romance. Linguistics 56(6): 1245–1301. Bentley, Delia. 2019. The logical structure of verbs of quantized and nonquantized change. Paper delivered at the International Conference on Role and Reference Grammar, University at Buffalo (SUNY), 19–21 August 2019. Bentley, Delia and Michela Cennamo. 2022. Thematic and lexico-aspectual constraints on V-S agreement: Evidence from Northern Italo-Romance.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

In A. Ledgeway, N. Vincent and J. C. Smith (eds.), Periphrasis and Inflexion in Diachrony. A View from Romance, 335–361. Oxford: Oxford University Press. Bentley, Delia and Nigel Vincent (forthcoming). LFG and Role and Reference Grammar. In Mary Dalrymple (ed.), The Handbook of Lexical Functional Grammar (Empirically Oriented Theoretical Morphology and Syntax). Berlin: Language Science Press. https://langsci-press.org/catalog/book/312. Borer, Hagit. 2005a. Structuring Sense I: In Name Only: Oxford: Oxford University Press. Borer, Hagit. 2005b. Structuring Sense II: The Normal Course of Events. Oxford: Oxford University Press. Borer, Hagit. 2005c. Structuring Sense III: Taking Form. Oxford: Oxford University Press. Börjars, Kersti, Rachel Nordlinger and Louisa Sadler. 2019. Lexical-Functional Grammar: An Introduction. Cambridge: Cambridge University Press. Bresnan, Joan. 1982. Control and complementation. In Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, 282–390. Cambridge, MA: MIT Press. Bresnan, Joan. 2001. Lexical-Functional Syntax. Oxford: Blackwell. Butler, Christopher S. 2005a. Functional approaches to language. In Christopher S. Butler, María de los Ángeles Gómez-González and Susana Doval Suárez (eds.), The Dynamics of Language Use: Functional and Contrastive Perspectives, 3–17. Amsterdam: John Benjamins. Butler, Christopher S. 2005b. Functionalist theories of language. In K. Brown, (ed.), Encyclopaedia of Language and Linguistics (2nd ed.), Vol. 1–14, 696–704. Oxford: Elsevier. Bybee, Joan. 1985. Morphology. Amsterdam: John Benjamins. Casti, Francesco. 2012. Testing the Interclausal Relations Hierarchy: Aspectual and Modal Periphrases in Modern Sardinian. PhD dissertation, University of Manchester. Cinque, Guglielmo. 1999. Adverbs and Functional Syntax: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Dalrymple, Mary. Forthcoming. The Handbook of Lexical Functional Grammar. (Empirically Oriented Theoretical Morphology and Syntax). Berlin: Language Science Press. https://langsci-press.org/catalog/book/312. Dik, Simon. 1978. Functional Grammar. Amsterdam: North-Holland. Dik, Simon. 1980. Studies in Functional Grammar. London: Academic Press. Dik, Simon. 1989. The Theory of Functional Grammar, Part 1. Dordrecht: Foris. Dowty, David R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Everett, Daniel. 2002. Towards an RRG theory of morphology. Lecture delivered at the International Conference on Role and Reference Grammar, University of La Rioja. Farrell, Patrick. 2005. Grammatical Relations. Oxford: Oxford University Press. Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert T. Harms (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart & Winston.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

11

12

DELIA BENTLEY

Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Givón, Talmy. 1980. The binding hierarchy and the typology of complements. Studies in Language 4: 333–377. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Goldberg, Adele E. 2006. Constructions at Work. Oxford: Oxford University Press. González-Orta, Marta Maria. 2002. Lexical templates and syntactic variation: The syntax-semantics interface of the Old English speech verb secgan. In Mairal Usón and Pérez Quintero (eds.), 281–302. Gruber, Jeffrey S. (1965) Studies in Lexical Relations. PhD dissertation, MIT, Cambridge, MA. Jackendoff, Ray. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7(1): 89–150. Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA: Kluwer. Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representation of Discourse Referents. Cambridge: Cambridge University Press. Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization. Cambridge: Cambridge University Press. Mairal Usón, Ricardo and Pamela Faber. 2002. Functional Grammar and lexical templates. In Mairal Usón and Pérez Quintero (eds.), 39–94. Mairal Usón, Ricardo and María Jesus Pérez Quintero (eds.). 2002. New Perspectives on Argument Structure in Functional Grammar. Berlin: Mouton de Gruyter. Mairal Usón, Ricardo, Lilián Guerrero and Carlos González (eds.). 2012. El funcionalismo en la teoría lingüística. La Gramática del Papel y la Referencia. Introducción, avances y aplicaciones. Madrid: Akal. Martín Arista, Javier. 2009. A typology of morphological constructions. In Christopher Butler and Javier Martín Arista (eds.), Deconstructing Constructions, 85–116. Amsterdam: John Benjamins. Matasovi´ c, Ranko. 2008. Patterns of grammaticalization and the layered structure of the clause. In Rolf Kailuweit et al. (eds.), New Applications of Role and Reference Grammar, 45–57. Newcastle upon Tyne: Cambridge Scholars Publishing. Pavey, Emma L. 2010. The Structure of Language. An Introduction to Grammatical Analysis. Cambridge: Cambridge University Press. Perlmutter, David M. (ed.). 1983. Studies in Relational Grammar 1. Chicago: University of Chicago Press. Perlmutter, David M. and Brian D. Joseph (eds.). 1990. Studies in Relational Grammar 3. Chicago: University of Chicago Press.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Introduction

Perlmutter, David M. and Carol G. Rosen (eds.). 1984. Studies in Relational Grammar 2. Chicago: University of Chicago Press. Pollard, Carl and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago, IL: University of Chicago Press. Ramchand, Gillian. 2008. Verb Meaning and the Lexicon. A First-Phase Syntax. Cambridge: Cambridge University Press. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman, (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer. Sag, Ivan and Thomas Wasow. 1999. Syntactic Theory: A Formal Introduction. Stanford: CSLI Publications. Shimojo, Mitsuaki. 2004. Quantifier float and information processing: A case study from Japanese. Journal of Pragmatics 36: 375–406. Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less numeral quantifier construction in Japanese. In Robert Van Valin (ed.), Investigations of the Syntax–Semantics–Pragmatics Interface, 285–304. Amsterdam: John Benjamins. Shimojo, Mitsuaki. 2016. Saliency in discourse and sentence form: Zero anaphora and topicalization in Japanese. In M. M. Jocelyne FernandezVest and Robert Van Valin Jr. (eds.), Information Structure and Spoken Language in a Cross-Linguistic Perspective, 55–75. Berlin: Mouton de Gruyter. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon (ed.), Grammatical Categories in Australian Languages, 112–171. Canberra: Australian Institute of Aboriginal Studies. Smith, Carlotta. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Reidel. Stump, G. 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 1977. Ergativity and the universality of subjects. Communication and Linguistics Studies 13: 689–706. Van Valin, Robert D., Jr. 1981. Grammatical relations in ergative languages. Studies in Language 5: 361–394. Van Valin, Robert D., Jr. 2001. An Introduction to Syntax. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface: An Introduction to Role and Reference Grammar. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic categories in Role and Reference Grammar. In Robert D. Van Valin, Jr. (ed.). Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2009. Linguistics past and present: A view from the Rhine. In Zarina Estrada Fernández, Albert Álvarez González and María Belén Carpio (eds.), Ser lingüista: un oficio diverso y polifacético. Diez años de una Maestría en Lingüística, 155–164. Hermosillo, MX: Editorial Unison. Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory. In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

13

14

DELIA BENTLEY

Timberlake (eds.), Language Typology and Historical Contingency, 91–124. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2014. On the place of information structure in a grammar. In Ana Díaz Galán et al., Comunicación, Cognición, Cibernétic@. Actas del XXXI Congreso de AESLA, 88–108. www.acsu.buffalo.edu/~rrgpage/ rrg.html. Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Ricardo Mairal Usón. 2014. Interfacing the lexicon and an ontology in a linking algorithm. In María Ángeles Gómez, Francisco Ruiz de Mendoza and Francisco Gonzálvez-García (eds.), Theory and Practice in Functional-Cognitive Space, 205–228. Amsterdam: John Benjamins. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press, 97–121. (Previously published in Vendler, Zeno. 1957. Verbs and times. Philosophical Review 56: 143–160.) Zaenen, Annie. Forthcoming. Information structure. In Mary Dalrymple (ed.), The Handbook of Lexical Functional Grammar (Empirically Oriented Theoretical Morphology and Syntax). Berlin: Language Science Press. https://langsci-press.org/catalog/book/312.

https://doi.org/10.1017/9781316418086.001 Published online by Cambridge University Press

Part One

https://doi.org/10.1017/9781316418086.002 Published online by Cambridge University Press

Overview

https://doi.org/10.1017/9781316418086.002 Published online by Cambridge University Press

1 Principles of Role and Reference Grammar Robert D. Van Valin, Jr.

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: A AOR ASP ATV AUH AV CL CS ECS EVID FIN IF GER INGR INT IRH LS(s) LSC LSMP LSPP

actor aorist aspect active voice Actor-Undergoer Hierarchy actor voice clause, clitic, classifier constructional schema extra-core slot evidential final stage of a process or action illocutionary force gerund

MP NEUT NMR NUC PN PoCS PoDP POT PrCS PrDP PRED

modifier phrase neuter non-macrorole nucleus proper noun post-core slot post-detached position potential pre-core slot pre-detached position predicator

PROC PSA

ingressive interrogative Interclausal Relations Hierarchy logical structure(s) layered structure of the clause layered structure of the modifier phrase layered structure of the adpositional phrase

PSBL RE RLS RP S SEML

process privileged syntactic argument possible referring expression realis reference phrase sentence, subject semelfactive

U

undergoer

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

18

ROBERT D. VAN VALIN, JR.

LSRP LSW

layered structure of the reference phrase layered structure of the word

MEP

macro-event property

1.1

UV

undergoer voice

XP

categorially unspecified phrase

Introduction

A linguistic theory is animated by specific issues, the consequences of which distinguish it from other theories.* The general issues, such as ‘What is a possible human language?’, ‘How is language acquired?’, are common to most approaches. The specific questions which stimulated the development of Role and Reference Grammar (RRG) were, ‘What would a linguistic theory look like if it were based on the analysis of languages with diverse structures, such as Lakhota, Tagalog, Dyirbal and Barai (Papua New Guinea), rather than on the analysis of English and similar languages?’ and ‘How can the interaction of syntax, semantics and pragmatics in different grammatical systems best be captured and explained?’ The two questions highlight the profound implications of the analysis of typologically diverse languages for the formulation of a linguistic theory, and they indicate that the resulting theory will be one in which semantics and pragmatics play significant roles. In other words, RRG is a theory of the syntax–semantics– pragmatics interface. Many of the constructs postulated in RRG are rather different from those in other theories, and this is precisely because of this starting point. For example, theories starting from English and other familiar Indo-European languages assume that endocentric phrase structure of the kind found in X-bar syntax is an appropriate model for syntactic structure universally, whereas the facts of some Philippine, Amazonian and Native American languages call this seriously into question and require a different conceptualization of phrase structure, as will be shown below. In terms of the classification of linguistic theories proposed in Jackendoff (2002), RRG is a parallel architecture theory in which syntax, semantics and discourse-pragmatics are represented independently, and in which each may interact directly with another. Numerous examples of these interactions can be found throughout this volume. Three additional theoretical assumptions deserve mention and will be elaborated on later in this chapter. The first is that RRG is a monostratal theory, in that it posits a single

* This research was supported in part by a Fellowship from the Max Planck Society and from the German Research Foundation via CRC 991 ‘The structure of representations in language, cognition and science.’ I would like to thank Delia Bentley, Jürgen Bohnemeyer, Fabricio Gerardi, Ranko Matasovic,´ Wataru Nakamura, Rainer Osswald, Mitsuaki Shimojo, and Jan Ullrich for comments on an earlier draft.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.1 The monostratal nature of RRG

syntactic representation for a sentence; there are no syntactic derivations of any kind. Rather, there is a direct mapping between the syntactic representation and the semantic representation by the RRG linking algorithm. This is represented graphically in Figure 1.1. ‘Discourse-pragmatics’ parallels the linking algorithm, and this signifies that it plays a role in the linking between syntax and semantics, an interaction which varies across languages with significant typological consequences. The second theoretical assumption concerns the nature of this syntactic representation: it is concrete, not abstract, reflecting the actual order of lexical items and grammatical morphemes. A crucial consequence of this concreteness is a ban on phonologically null elements in the syntax; RRG does not permit empty elements in the syntax (i.e. no traces, unpronounced copies of NPs, null pronouns, null light verbs, etc.) These two assumptions strongly constrain the theory. The third assumption concerns the nature of the grammar as a whole. Since the early 1970s there has been an opposition between constructional approaches (e.g. the Standard Theory (Chomsky 1965), Relational Grammar) and principles-and-parameters approaches (government and binding (GB), the Minimalist Program) which deny the theoretical validity of the notion of grammatical construction. Starting in the late 1980s, a variety of neo-constructional approaches appeared under the heading of ‘construction grammar’, all of which reject the principle-and-parameters approach and treat grammatical constructions as central to grammatical description and linguistic theory. RRG has been constructional from the outset and therefore is a kind of construction grammar, albeit one which differs in many ways from ‘name brand’ Construction Grammar. These assumptions will be discussed further and their consequences illustrated in this chapter. Linguistics is an integral part of cognitive science and, accordingly, theories have implications for the understanding of language acquisition and language processing. RRG poses the questions ‘Can language acquisition be accounted for without recourse to an autonomous Language Acquisition Device?’ and ‘Can a model of grammar that answers the typological and theoretical questions posed above provide any insights into the neurocognitive processing of language?’ The last chapter of Van Valin and LaPolla (1997)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

19

20

ROBERT D. VAN VALIN, JR.

summarizes RRG work on first language acquisition up to that point, and the tentative conclusion is that the acquisition of a variety of core grammatical phenomena can be explained in RRG terms, including some for which it has been argued that there is no evidence available to the child regarding them (e.g. subjacency; see Van Valin 1998, 2001, 2002; Weist, this volume, Chapter 18). In the last few years there has been research on applying RRG to language processing, both computational and neurolinguistic. Computational implementation of RRG is a relatively new undertaking, and the results are intriguing (see Nolan, this volume, Chapter 21, Kallmeyer and Osswald, this volume, Chapter 20). With regard to sentence processing, one of the distinctive attributes of RRG is the bidirectionality of the linking between syntax and semantics, which is represented by the double-headed arrow in Figure 1.1. The RRG linking algorithm maps from semantics to syntax and from syntax to semantics. This is an idealization of what a speaker does (semantics to syntax) and what a hearer does (syntax to semantics). Hence the design of the theory makes it readily amenable to psychoand neurolinguistic sentence processing models, as argued in Van Valin (2006, this volume, Chapter 19). The presentation will proceed as follows. Section 1.2 will be concerned with the syntactic representation of simple sentences, followed by the discussion of the syntactic representation of complex sentences in Section 1.3. Section 1.4 is devoted to the semantic representation of simple and complex sentences. The RRG approach to information structure is briefly introduced in Section 1.5. The next two sections present the linking algorithm for simple sentences (Section 1.6) and aspects of linking complex sentences (Section 1.7).

1.2

The Structure of Simple Sentences

1.2.1 Constituent Structure There are a number of options available to linguistic theories regarding the representation of syntactic structure: phrase-structure trees (Chomsky 1957, 1965), relational networks (Relational Grammar), X-bar trees (Chomsky 1970), functional structures (f-structures, Lexical Functional Grammar), binary-branching X-bar trees (GB, Minimalism), etc. All of the phrase-structure approaches, regardless of whether they assume X-bar or binary branching, maintain that crossing branches in a tree is not allowed. None of these options are compatible with the assumptions laid out in the previous section. To begin with, relational networks and f-structures, both of which are based on grammatical relations, are abstract and do not meet the concreteness requirement. More significant is the incompatibility of the various phrase-structure approaches with the constraints that RRG imposes on syntactic representations. RRG posits a single, relatively

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

concrete syntactic representation for a sentence; no derivations involving multiple representations or phonologically null elements are permitted. For approaches in the Chomskyan tradition, many syntactic phenomena require multiple, derivationally related representations, for example whdisplacement, free phrase and word order phenomena, and three-place predicates (binary-branching models only). Moreover, these phenomena are normally analysed as crucially involving phonologically null elements, namely, traditional traces (GB) or unpronounced copies of XPs (Minimalism). Hence none of the approaches mentioned so far could be used in RRG, because they are incompatible with the fundamental assumptions the theory makes. In Foley and Van Valin (1984) the first ideas for an RRG-compatible theory of clause structure, termed the layered structure of the clause (LSC), were articulated, and subsequent work (Johnson 1987; Watters 1993; Van Valin 1993) led to the development of more sophisticated versions of the LSC. The LSC is semantically based, in that the central and universal constructs are motivated by semantic distinctions. The theory posits that clause structure is based on two fundamental oppositions, which are universal and grounded in the fact that language is a system for communication between human beings: the opposition between predicating and non-predicating elements, on the one hand, and that between nonpredicating elements that are licensed by the predicating element and those that are not. In other words, all languages distinguish predicates, arguments, and adjuncts. This follows from the nature of language as a system of human communication: one of the communicative functions of language is to convey information, and this information is composed of propositions, which involve reference and predication. Languages therefore have elements that are specialized for referring (arguments) and predicating (predicates), and these are further distinguished from the optional modifiers (adjuncts) that enrich the propositions being communicated. This can be represented as in (1). (1)

[PROPOSITION [PREDICATE X ][ARGUMENT (Y1), (Y2), . . .][ADJUNCT (Z1), (Z2), . . .]]

These semantic units underlie the central syntactic units of the layered structure: the notion of predicate motivates the nucleus, which houses the predicating element (the predicator), the core, which consists of the nucleus plus the arguments of the predicating element, and the peripheries, which house the adjuncts. This can be represented as in (2) and applied to an English example in (3). (2)

[CLAUSE [CORE (Y1) [NUCLEUS X ] (Y2), . . .][PERIPHERYcore (Z1)] CORE][PERIPHERYclause (Z2)] CLAUSE]

(3)

[CLAUSE [CORE Mary [NUCLEUS ate ] a cookie ][PERIPHERYcore after lunch] [PERIPHERYclause despite her diet.] CLAUSE]

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

CORE]

21

22

ROBERT D. VAN VALIN, JR.

Nucleus, core and periphery are syntactic units which are relatively motivated by the semantic constructs in (1) but are not purely semantic. While in the prototypical case, the nucleus and the predicating element (predicator) are co-extensive, it is possible to have non-predicators in a nucleus. A classic example of this is noun incorporation, in which one of the arguments of the predicator occurs in the nucleus along with the predicator itself, for example Lakhota [ˇcháŋ kiŋ [NUC kaksá]] (wood the chop) ‘chop the wood’ vs. [[NUC ˇchaŋkáksa]] ‘chop wood’, ‘do wood-chopping’. In the non-incorporated form the object argument ˇcháŋ kiŋ ‘the wood’ occurs as an independent argument, whereas in the incorporated form the noun ˇcháŋ ‘wood’ occurs without an article and is compounded with the predicator kaksá ‘chop’, yielding ˇchaŋkáksa. Multisyllabic words in Lakhota tend strongly to have stress on the second syllable, as with kaksá ‘chop’, and in the incorporated form the stress has shifted to the second syllable, which is to be expected due to the incorporation of ˇcháŋ ‘wood’ (which loses its own accent), which creates a trisyllabic word. With respect to the core, in the prototypical case, a semantic argument of the predicator in the nucleus occurs in it, and an XP in the core is a semantic argument of the predicator. However, there are cases in which a semantic argument does not occur in the core, for example the agent in a passive construction appears in the core-level periphery, or a wh-expression may occur in a position outside of the core in some languages (e.g. Figure 1.4). Conversely, there are XPs occurring in the core which are not semantic arguments of the predicator in the nucleus. Examples include dummy arguments, like those that occur with meteorological verbs in some languages (e.g. It is raining, Il pleut (French), Es regnet (German)), and ‘raised’ arguments in ‘raising constructions’ (e.g. Mary seems to like okra, in which Mary occurs as a syntactic argument of the core headed by seems but is a semantic argument of like). Peripheries canonically contain adjuncts which modify a specific layer of the clause. In (3), for example, the temporal adjunct after lunch modifies the core Mary ate a cookie and occurs in the core-level periphery, and the concessive adjunct despite her diet modifies the clause Mary ate a cookie after lunch and occurs in the clause-level periphery. There is at least one case, however, where the XP in a periphery is not an adjunct modifier but rather a semantic argument of the predicator in the nucleus, namely, the agent in a passive construction. Moreover, there are constructions in which the adjunct represents a participant in the state of affairs which is not a semantic argument of the nucleus, for example the deputative beneficiary marked by for in Bill went to the store for Mary. The distinctions in (2) are better represented in a tree diagram, as in Figure 1.2. The nucleus contains the predicator and may be modified by an adverb in the nuclear-level periphery.1 The core minimally contains the nucleus (and

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.2 The layered structure of the clause (preliminary)

Figure 1.3 The structure of (3)

its periphery, if present), which would be the case for predicators which have no arguments (e.g. Lakhota maˇ gážu ‘it is raining’), but usually there are one or more argument expressions. It may optionally be modified by adverbs or adjunct adpositional phrases in the core-level periphery. The clause contains the core (plus its periphery, if present), and, like the nucleus and core, can be modified optionally by peripheral adverbs or adpositional phrases. The sentence node reflects the traditional idea that a sentence may consist of multiple clauses; its function will be clarified below. Linear order is not relevant to the definition of the units of the LSC; the order of elements in Figure 1.2 was chosen for clarity and is in no way privileged. This figure presents the universal features of the LSC. The structure of (3) is given in Figure 1.3. Eat is the predicator in the nucleus, Mary and a cookie are the core arguments, and the two PPs, after lunch and despite her diet, occur in peripheries. There is no VP in the structure, despite the well-known evidence that some constructions in English clearly involve a ‘VP’-like grouping, for example I expected to find someone washing the dishes, and [‘VP’ washing the dishes] was Bill. This is because ‘VPs’ are not universal; some languages have constructions with them and others do

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

23

24

ROBERT D. VAN VALIN, JR.

Figure 1.4 PrCS and PrDP in English

not, and so from an RRG perspective, ‘VP’ is a derivative category which occurs in particular constructions and whose existence in a language is the result of the interaction of the LSC with other components in the grammar. Specific examples will be discussed later in this chapter, as well as in Latrouite and Van Valin (this volume, chapter 12), where an analysis of ‘VP’-ellipsis in English is presented. The two PPs modify different layers of the clause: the temporal PP after lunch tells when the event in the core occurred,2 while the concessive PP despite her diet modifies the clause containing the core and the core-level periphery (see Section 1.2.4). The order of the PPs is not free, as the ordering reflects the layering: the higher-level modifier follows the lower-level modifier. The constituents in Figure 1.2 are semantically motivated and universal. There are other constituents which are not semantically motivated and accordingly are not universal. The English example in Figure 1.4 illustrates two of them. It is important to notice that there is no empty element in the core representing the displaced wh-expression what; there is no trace or unpronounced copy of it. The linking algorithm will link what directly to the correct place in the semantic representation for its interpretation; see Section 1.6. The position that displaced wh-expressions occur in when they appear at the beginning of a clause, as in English, German, Icelandic and many other languages, is known as the Pre-Core Slot (PrCS). It is not restricted to whexpressions; non-wh XPs functioning as contrastive topic, as in (4a), or contrastive focus,3 as in (4b), can appear in the PrCS. (4)

a. A: What do you think of the new novels by Suzanne Collins and John Grisham? B: Collins’ book I really enjoyed, but Grisham’s I found boring. b. A: Have you read anything good lately? B: Suzanne Collins’ new book I really liked, but John Grisham’s latest offering I didn’t care for.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

In English the subject is the first (non-oblique) argument in the core, and so a non-wh XP appearing immediately before it is outside of the core and is therefore in the PrCS. XPs in the PrCS, regardless of whether they are whexpressions or not, are not set off by an intonation break, and they never cooccur with a resumptive pronoun in languages like English.4 The existence of the PrCS in languages with nucleus-initial default word order (e.g. Tagalog (Austronesian, Schachter and Otanes 1972); Tzotzil (Mayan, Aissen 1987); and Wari’ (Chapakuran, Brazil, D. Everett 2008)) is straightforward to diagnose: an XP before the nucleus without an intonation break or an independent resumptive pronoun is in the PrCS. This is illustrated by the Tagalog examples in (5) from Schachter and Otanes (1972: 496–498). (5)

a. Tawa siya sa kaniya kahapon. laugh 3sg.nom dat 3sg.obl yesterday ‘She laughed at him yesterday.’ b. Kahapon siya tawa sa kaniya. yesterday 3sg.nom laugh dat 3sg.obl ‘It was yesterday she laughed at him.’ b′. Sa kaniya siya tawa kahapon. dat 3sg.obl 3sg.nom laugh yesterday ‘It was at him she laughed yesterday.’

The default word order in Tagalog is given in (5a), in which the nucleus containing the predicator tumawa ‘laughed’ occurs initially in the clause. It is followed by siya ‘3sg.nom’, which is a clitic pronoun that must occur in second position in the clause; in (5a) second position is immediately following the nucleus. In (5b) the temporal adverb kahapon ‘yesterday’ occurs before the nucleus, and it is followed by siya, the second position clitic. Based on clitic placement and a number of other criteria, Latrouite and Van Valin (2021) show that the adverb is in the PrCS in (5b). The same considerations lead to the conclusion that sa kaniya ‘at him’ in (5b′) is likewise in the PrCS. It should be noted that the PrCS element has a contrastive narrow focus reading like the English example in (4b). Languages with nucleus-final or extremely flexible word order present something of a challenge with respect to establishing the existence of a PrCS, because there is no fixed reference point for determining the onset of the core, unlike in Tagalog or English. The best evidence is the obligatory occurrence of wh-expressions at the beginning of the clause in wh-questions. It is well known, however, that nucleus-final languages rarely have obligatory displacement of wh-expressions to the beginning of the clause; they either have only wh-in situ or optional displacement. The latter situation, namely optional displacement to the beginning of the clause or in situ, is in principle distinguishable from ‘free’ word or phrase order, since in the latter situation a wh-expression could occur in any position in the clause; in

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

25

26

ROBERT D. VAN VALIN, JR.

contrast, with optional displacement, it can occur in only two positions, in situ or clause-initial. This is evidence for a PrCS, albeit weaker than the evidence in languages with obligatory displacement. Many nucleus-final languages, and some with other word orders as well, have a Post-Core Slot (PoCS), which, like the PrCS, is outside the core but inside the clause. It is multifunctional from a cross-linguistic perspective. In Japanese, according to Shimojo (1995), focal XPs, like those in the PrCS in English, cannot occur in the PoCS, but in Dhivehi (Indo-Aryan; Cain and Gair 2000), wh-expressions can appear in situ (6a) or in the PoCS (6a′), and contrastive focal XPs occur in it as well (6b, b′).5 (6)

a. Al¯ı k¯ıke bun¯ı ta? Ali what say.pst.foc q ‘What did Ali say?’ a′. Al¯ı bun¯ı kīke ta? Ali say.pst.foc what q ‘What did Ali say?’ b. M¯ ale ulun¯ıma aharen bon¯ı AIS KURĪMU. ˙ Male be.pst.prog.when 1sg drink.pres.foc ice cream ‘When in Male, it is ice cream that I eat.’ b′. M¯ ale ulun¯ıma ais kur¯ımu bon¯ı AHAREN. ˙ Male be.pst.prog.when ice cream drink.pres.foc 1st ‘When in Male, it is I who eat ice cream.’

In Figure 1.4, there is an XP (as for Mary) before the wh-expression, and it has quite different properties from XPs in the PrCS. It is normally set off by an intonation break, represented orthographically by a comma, and if the XP is an argument of the predicator in the nucleus of the clause, as in this sentence, there is obligatorily a resumptive pronoun in the core of the clause, in this case she. This position is termed the Pre-detached position (PrDP). It is usually termed ‘the left-detached position’ in the linguistic literature, including previous work in RRG. However, there is no left or right in spoken language, only before and after. Only in written language can one speak of left and right, and the label ‘left-detached position’ reflects the fact that European languages are written from left to right. For scholars writing in languages like Arabic and Farsi, on the other hand, it is the ‘rightdetached position’. There is no such problem with PrCS and PoCS, since ‘pre-’ and ‘post-’ refer to order in speech, not order on the printed page. Accordingly, the problematic terms ‘left-detached position’ and ‘right-detached position’ are replaced by PrDP and Post-detached position (PoDP), in order to avoid confusion.6 The two detached positions are outside of the clause and are direct daughters of the sentence node. Both are normally set off by an intonation break, and both require a resumptive pronoun if the XP in the detached position is an argument of the predicator in the nucleus of the clause. Tagalog presents clear evidence that the PrDP is outside of the clause, as exemplified in (7) from Schachter and Otanes (1972: 489).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (7)

Si May (ay), kailan ba (siya) ba~balik dito? nom.pn (top) when q (3sg.nom) ipfv.return here ‘As for May, when will (she) come here?’

The topic expression, si May, is set off by an intonation break and optionally marked by the topic marker ay (Nuhn 2021; Latrouite and Van Valin 2021). It must be in the PrDP, because it is followed by the adjunct whexpression kailan ‘when’, which is in turn followed by two second-position clitics, the interrogative marker ba and the optional resumptive pronoun siya ‘3sg.nom’. The placement of the clitics shows that kailan ‘when’ must be clause-initial, namely in the PrCS, as in (5b,b′), and this means that si May is outside of the clause and therefore in the PrDP. The structure of (7) is given in (8). (8)

[S [PrDP Si May (ay)], [CL [PrCS kailan ]¼ba¼(siya) [CORE [NUC babalik] dito?] CL] S]

The constituents of the LSC, both obligatory and optional, minus the peripheries, are given in Figure 1.5. These constituents can be viewed in terms of whether they are universal or non-universal. It has been argued that the nucleus, core and clause ‘spine’, plus peripheries, is semantically motivated and is based ultimately on the necessary properties of language as a system of human communication. All languages make use of the nucleus, core and clause spine. On the other hand, not all languages have detached positions or special positions outside of the core and inside the clause. They are pragmatically motivated rather than semantically motivated. While it would be impossible to form propositions without predicators and arguments, it is possible to communicate effectively without these special positions. Lakhota, for example, has a PrDP but no PrCS or PoCS. Dyirbal, with its extremely (grammatically) free word order, has no fixed positions with reference to which the PrDP or PrCS could be defined; moreover, it lacks markers indicating topics and foci akin to Japanese wa and ga, for example, which could be associated with these distinctions (see Shimojo 2011). The packaging of information in terms of special fixed positions for topics and foci is a useful but not an indispensable feature of human language.

Figure 1.5 Constituents of the LSC

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

27

28

ROBERT D. VAN VALIN, JR.

1.2.2 Lexical and Syntactic Categories In Figures 1.2 and 1.5 the abbreviation ‘XP’ is used in the possible core argument positions, because the syntactic category of the phrase is not fixed; it could be a noun or determiner phrase, an adpositional phrase, a core or a clause. In this sense it is a variable over syntactic categories, like in the term ‘X-bar syntax’, and it will continue to be used in this way. In Figures 1.3 and 1.4, on the other hand, the XPs are Mary, a cookie, what and she, and on standard assumptions these would be analysed as NPs or determiner phrases (DetPs).7 What this assumes here is the endocentric nature of syntactic categories: a noun projects an NP, a verb projects a VP, a preposition projects a PP, etc. Most syntactic theories assume that most, if not all, syntactic categories are endocentric; for theories which assume Merge as the central operation in syntax all syntactic categories are necessarily endocentric. RRG rejects endocentrism as a fundamental feature of constituent structure. Rather, the two key notions are the nucleus, which, viewed from a cross-linguistic perspective, is not restricted to any particular lexical category or even to being a head (it can be phrasal), and reference phrase (RP), which is a potentially referring expression, which may be headed in principle by a range of lexical categories (Van Valin 2008). In the examples examined thus far, the predicator in the nucleus has been a verb, but this is not the only possibility, as the examples in (9) show. (9)

a. Max is a very good lawyer. b. Max is (extremely) tall. c. Max is in the house

Predicator ¼ a very good lawyer [RP] Predicator ¼ (extremely) tall [ADJP] Predicator ¼ in the house [PP]

Be is not the predicator in these sentences; it is an auxiliary required for nucleus formation when the predicator is non-verbal or passive. If it were claimed that be is the predicator, then it would be necessary to claim that in languages which lack such a verb there is a phonologically null verb be which serves as the predicator. This, however, is a violation of the concreteness restriction in RRG. Consequently, the predicator in (9a) is the RP a very good lawyer, in (9b) the adjective phrase (extremely) tall, and in (9c) the PP in the house. These examples illustrate the fact that the nucleus is not a projection of verb, or any other category, but is category-neutral, and moreover they also show that the nucleus need not be a head but can be a phrase. A more extreme example of a phrasal nucleus will be given in §1.3.3.2. The usual label for argument expressions is NP, or in some approaches, DetP. The problem with labelling referring expressions as NPs can be seen in (10) from Tagalog and (11) from Lakhota. (10)

a. [CORE [NUC Tawa ] [RP ang bata] CORE] laugh nom child ‘The child laughed.’ b. [CORE [NUC Bata ] [RP ang tumawa] CORE] child nom laughed ‘The one who laughed is a child.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Tagalog

Principles of Role and Reference Grammar

(11)

a. [CLAUSE [RP Wíŋyaŋ kiŋ hé] [NUC waˇ chí-kte-šni]CLAUSE]8 woman the that dance-pot-neg ‘That woman will not dance.’ chí kiŋ hé] [NUC wíŋyaŋ-kte-šni ]CLAUSE] b. [CLAUSE [RP Waˇ dance the that woman-pot-neg ‘That dancer will not be a woman.’

Lakhota

In (10a) the predicator in the nucleus is a verb, and the single core argument is a noun, suggesting that it is correctly analysed as an NP. In (10b), on the other hand, the predicator is a noun and the head of the argument expression is the verb tumawa ‘laughed’. Importantly, there is no change in the form of the words in the two sentences: bata ‘child’ is an argument in (10a) and the predicator in the nucleus in (10b), and tumawa ‘laughed’ is the predicator in the nucleus in (10a) and the head of an argument expression in (10b). This contrasts with the English translation of (10b), in which the verb-based argument expression is expressed as a relative clause and the nucleus requires the auxiliary be due to the nonverbal predicator. Thus, unlike in English, a verb can head a referring expression without modification, and a noun can function as the predicator without modification. The same pattern holds in the Lakhota sentences in (11); both predicators take the same inflectional morphology, regardless of the lexical category of the head, and the head of the ‘NP’ waˇchí kiŋ hé ‘that dancer’ is the verb waˇchí ‘dance’. These examples are no problem for the category-neutral notion of nucleus, and they show that a category-neutral syntactic category for referring expressions is needed as well. Van Valin (2005: 28) made an initial suggestion which was further developed into the notion of RP in Van Valin (2008).9 It might be objected that there is no need for this notion, because these examples can be handled in terms of the already existing notion of DetP. The argument expressions in (10) and (11) would be DetPs, not NPs, and the complement to the determiner head of the phrase could be of different categories. However, DetP is a problematic construct, given RRG’s assumptions about syntactic structure. In order to have a DetP there must be a determiner functioning as the head of the DetP, and in many languages there are minimal determiner systems; Russian, Mandarin and Japanese, for example, have demonstratives but no articles. In such languages many sentences have arguments which lack determiners, and therefore to justify calling them DetPs it would be necessary to posit phonologically null determiners as heads of the DetPs. This is incompatible with the RRG prohibition against phonologically null elements in the syntax. RP, on the other hand, requires no such postulation of null determiners. Thus in (10) and (11) the arguments are RPs, regardless of whether the head is a noun or a verb. Modifiers, in particular those modifying within RPs, do not necessarily follow an endocentric pattern. The following examples are from D. Everett (2008).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

29

30

ROBERT D. VAN VALIN, JR.

(12)

a. The God is dead philosophers are mostly dead. b. My grandson likes to give me the who’s the boss now, silly old grandpa wink frequently.

In (12a) the modifier in the RP is a clause, while in (12b) it is a sentence. Neither can be analysed as a kind of adjective phrase headed by a lexical adjective, unlike the very tall man, and accordingly, there is a need for a category-neutral construct for modifiers analogous to the nucleus for predicators. Van Valin (2008) proposed the notion of modifier phrase (MP) to accommodate attributive modifiers of all types, as well as adverbial modifiers and adpositional phrase modifiers. Thus, the primary syntactic categories in RRG are RP, nucleus, PP and MP.10 Phenomena like those in (10) and (11) have led some analysts to suggest that these languages do not differentiate nouns from verbs, and therefore roots are precategorial and are assigned a category on the basis of how they are used in a clause. This is problematic, because it confuses category and function, as well as lexical and syntactic categories. Assuming endocentrism and associated phrase structure, bata ‘child’ in (10b) heads a VP, a syntactic category, and because syntactic categories are projections of lexical categories, bata ‘child’ must be a verb in this sentence. In (10a), on the other hand, the same reasoning leads to the conclusion that bata is a noun, since it is heading an NP functioning as an argument. In order to avoid claiming that there are two lexical items, one a noun and the other a verb, one solution is simply to avoid assigning lexical categories to words.11 This is very problematic, because there is evidence that both Tagalog and Lakhota differentiate nouns from verbs, albeit weakly. Himmelmann (2008) distinguishes nouns from verbs in Tagalog based on a morphological criterion: verbs can take voice morphology, while nouns cannot. Boas and Deloria (1941: 23) note that there are morphophonological differences between nouns and verbs in Lakhota, for example vowel contraction (nouns tend to contract adjacent vowels while verbs do not), reduplication (common with verbs, very rare with nouns). There are morphosyntactic differences as well. The examples in (13) are the same as those in (11) except for the plurality of the core argument. (13)

a. [CLAUSE [RP Wíŋyaŋ kiŋ hená] [NUC waˇ chí-pi-kte-šni] CLAUSE] woman the those dance-pl-pot-neg ‘Those women will not dance.’ chí-pi-kte-šni] CLAUSE] a′. [CLAUSE [RP *Wíŋyaŋ-pi kiŋ hená] [NUC waˇ woman-pl the those dance-pl-pot-neg ‘Those women will not dance.’ chí-pi kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE] b. [CLAUSE [RP Waˇ dance-pl the those woman-pl-pot-neg ‘Those dancers will not be women.’ chí kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE] b′. [CLAUSE [RP *Waˇ dance the that woman-pl-pot-neg ‘Those dancers will not be women.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Lakhota

Principles of Role and Reference Grammar

An interesting feature of Lakhota is the lack of marking number on a noun when it is an argument; plurality can be signalled by means of a plural demonstrative in an RP or a quantifier, but it cannot be marked directly on the noun itself. This is illustrated in (13a, a′): that wíŋyaŋ ‘woman’ is plural is indicated primarily by the plural demonstrative hená ‘those’ and also by the plural agreement on the predicator in the nucleus, but marking plural directly on wíŋyaŋ renders the sentence ungrammatical, as in (13a′). However, when the RP argument is headed by a verb, as in (13b, b′), plural marking on it is obligatory, as in (13b). The sentence is ungrammatical without it, as in (13b′). It is not the case that nouns like wíŋyaŋ ‘woman’ can never take plural marking; when they function as the predicator in the nucleus of the clause, as in (13b), plural marking is obligatory, just as it is for verbs with the same function, as in (13a). Thus nouns and verbs are treated differently with respect to number marking when functioning as arguments (referring expressions) but not when they serve as predicators. Hence, languages like Tagalog and Lakhota do distinguish nouns from verbs, albeit weakly. The issue, rather, is the existence of restrictions on what function a lexical item can have. How can RRG approach the issue of lexical categories, given the range of cross-linguistic variation, from languages like Tagalog and Lakhota, which seem to make only weak distinctions between nouns and verbs, to languages like Latin, Russian and Dyirbal, which make strict divisions between the two main categories, with English somewhere in the middle? One possibility is to claim that at the most basic level, lexical items fall into one of two classes: they are either referring expressions (REs) or predicates. This is related to the fundamental opposition motivating the layered structure of the clause, namely, the opposition between predicating and non-predicating elements. This distinction derives from the nature of language as a system of communication: communication involves conveying information, which involves propositions, which involve reference and predication, hence REs and predicates. REs and predicates can have one of three grammatical functions: argument, predicator or modifier. These functions are related to the three syntactic categories in RRG, namely, RP, nucleus and MP. In (10a) and (11a) a predicate is functioning as the predicator in the nucleus and an RE as the core argument, whereas in (10b) and (11b) an RE is the predicator in the nucleus and a predicate serves as the core argument. The traditional categories of verb, adjective, adverb and adposition are semantically predicates and grammatically predicators or modifiers, as defaults. The traditional category of noun encompasses REs serving as arguments as their default function.12 Defining lexical categories in terms of their morphophonological and morphosyntactic properties leads to the observation that Tagalog, Lakhota and other such languages show that languages need not make ‘deep’ lexical category distinctions, and this seems to follow from the ability of lexical items to function as an argument, predicator or modifier rather freely. This

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

31

32

ROBERT D. VAN VALIN, JR.

ability is an important morphosyntactic property which many or all lexical items share, and it is precisely the morphosyntactic properties that different words do not share that is the basis for assigning them to different lexical categories. In languages with well-defined lexical categories there are strict constraints on the grammatical function that a given lexical category may have: nouns (REs) can be arguments, verbs (predicates of certain semantic types) can be predicators, and adjectives and adverbs (predicates of certain semantic types) can be modifiers. These restrictions are among the morphosyntactic properties which define the category. If a category is to have a different grammatical function, for instance noun or adjective as predicator or verb as argument, special morphosyntactic treatment is necessary, such as the use of an auxiliary verb or verbalizing derivational morphology for nonverbal predicators, or nominalizing derivational morphology for nonnominal arguments. These special treatments are also part of the properties defining the different categories. This pattern is found in many European languages, and it has been taken as the norm for human language. This view has led some linguists, when confronted with Tagalog-type languages, to posit derivational morphology for verbalization and nominalization, all of which is marked by zero morphemes. Thus, on this type of analysis, in (10b) bata ‘child’ has undergone zero-marked verbalization, yielding ‘to be a child’, and tumawa ‘laughed’ has undergone zero-marked nominalization, yielding ‘the one who laughed’. The RRG analysis of (10b) given above does not invoke any zero-marked derivational morphology. Languages also differ as to the semantic domains encompassed by the nouns and verbs in them. While words referring to objects, entities, people, animals, etc., will typically be realized as nouns, and those referring to actions, events, states of affairs, etc., will typically be instantiated as verbs, others may be realized in different ways in different languages. Consider property expressions that indicate characteristics of entities and objects (e.g. tall, little, red, black, rich, happy, etc.). In some languages (e.g. English) they constitute a word class all their own, namely, adjectives. In other languages, however, they do not constitute an independent word class but rather are treated as a subtype of noun or verb. In Dyirbal (Dixon 1972) property words are assimilated to the category of noun and are treated morphosyntactically like nouns. In Lakhota, on the other hand, they are construed as a subclass of verbs and take the same inflectional morphology as verbs, as illustrated in (14). (14)

Wíŋyaŋ kiŋ hená háŋska-pi-kte-šni. woman the those tall-pl-pot-neg ‘Those women will not be tall.’

Lakhota13

The status of lexical categories in RRG is grounded in the universal semantic distinction between REs and predicates, which underlies the noun–verb dichotomy. The lexical categories beyond noun and verb in a

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

language are differentiations of the functions of predicates as modifiers (i.e. as adjectives and adverbs), and all of the distinctions must be justified morphophonologically and/or morphosyntactically. Adpositional predicates can be arguments, as with verbs like put, or adjunct modifiers. Thus, from an RRG perspective, lexical categories are language-specific but with a universal semantic foundation. Moreover, they play a rather different role in the nonendocentric syntax of RRG than they play in the endocentric syntax of other theories.

1.2.3 Operators In Figure 1.4, which represents the structure of As for Mary, what did she eat after lunch?, the auxiliary verb did is not attached to the LSC tree in any way. The reason for this is that the LSC trees introduced so far contain only lexical words and morphemes. Closed-class grammatical morphemes indicating functional categories like aspect, tense, modality and mood (illocutionary force) are not represented. These notions are termed operators in RRG, and they modify specific layers of the LSC. The operators discussed in RRG are given in Table 1.1. A detailed discussion of them can be found in Peterson (Chapter 2 of this volume), and the emphasis in this section will be on selected cases of the interaction of operators and the LSC. The most commonly encountered operators cross-linguistically are aspect, deontic modality, negation, tense, and illocutionary force (IF).14 Following a suggestion in Johnson (1987), they are represented in a separate projection of the clause (the operator projection) which is more or less a mirror image of the lexical tree, which is normally referred to as the constituent projection. A revised version of the tree in Figure 1.4 with the operator projection added is given in Figure 1.6. Table 1.1 Operators Nuclear operators Aspect Negation Directionals (only those modifying the orientation of an action or event without reference to participants) Core operators Directionals (only those expressing the orientation or motion of one participant with reference to another participant or to the speaker) Event quantification Modality (root modals, e.g. ability, permission, obligation) Internal (narrow scope) negation Clausal operators Status (epistemic modals, external negation) Tense Evidentials Illocutionary force (IF)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

33

34

ROBERT D. VAN VALIN, JR.

Figure 1.6 The constituent and operator projections15

The operator projection is not an exact replica of the constituent projection in two ways. First, it contains only the nucleus-to-sentence spine and does not include the other positions and the peripheries. Second, if there is more than one operator at a given layer, that node is duplicated so that each operator modifies a distinct node in the operator projection. In this example, there are two clause-level operators, tense and IF, each modifies a separate clause node, and the structure makes clear that IF has scope over tense. The evidence for this will become clear in the discussion of complex sentences in Section 1.3. Both operators are linked to the auxiliary did, because it expresses both of them, tense via its form (past tense of do) and IF by virtue of its position in the clause. English signals IF by means of the linear position of the tense morpheme in the main clause: core-medial tense indicates declarative IF, as in Figure 1.3; tense before the first constituent of the core signals interrogative IF (e.g. Did Mary eat a cookie after lunch?); and the lack of tense in the main clause indicates imperative IF. This way of signalling IF is relatively uncommon cross-linguistically. More common and straightforward is the Lakhota approach, which involves particles to mark IF. If one wants to change (11a) into a question, it is only necessary to add he, the question particle, to the end of the sentence, as in (15a). If one wants to tell someone ‘Don’t dance!’, then the imperative final particle would be added, as in (15b). (15)

a. Wíŋyaŋ kiŋ hé waˇ chí-kte-šni he? woman the that dance-pot-neg int ‘Will that woman not dance?’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Lakhota

Principles of Role and Reference Grammar b. Waˇ chí-šni yo/ye! dance-neg imp.Male/imp.Female ‘Don’t dance!’

Lakhota IF particles index the sex of the speaker (Ullrich 2011: 820–822): those ending in -o are for male speakers, while those ending in -e are for female speakers. He in (15a) was originally a female form but has come to be used by both sexes; the male form is huwó. The base form for the assertion marker for men is yeló, while women use ye; kštó is used by both sexes but primarily by women (Ullrich 2011). These morphemes raise an important issue with respect to the concreteness restriction on syntactic representations in RRG. The interrogative and imperative particles are obligatory in questions and commands, because without them the utterance would not be interpreted as having those IF values. The assertion particles are different; if they are not used, as in (11a), the utterance is still interpreted as an assertion due to the lack of the interrogative and imperative markers. Hence there are two ways to indicate an assertion in Lakhota: use of an overt assertion marker, or use of no IF particle of any kind. The first is unproblematic: there is an overt form, kštó for example, which can be linked to the operator projection as the IF operator with scope over the clause layer. The problem arises in the second case in which there is no IF marking of any kind, yet the sentence has a definite IF value. It is necessary to represent this, especially in the analysis of complex sentences; this will be discussed in Section 1.3. This is illustrated by (16), the structure of which is given in Figure 1.7; only the operator projections are represented. (16)

Wíŋyaŋ kiŋ hé waˇ chí-šni kštó/Ø woman the that dance-neg decl.Female/decl ‘That woman does/did not dance.’16

Lakhota

The issue with the second representation is whether it violates the prohibition against phonologically null elements in the syntax. There are good grounds for concluding it does not. First, the proscribed null elements are lexical categories or projections thereof; for example, null pronominals or verbs are considered to be members of the lexical categories of pronominal and verb. The zero morpheme in (16) is a closed-class grammatical (function) morpheme and is not a free form which can be manipulated in the syntax,

Figure 1.7 IF indicators in (16)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

35

36

ROBERT D. VAN VALIN, JR.

unlike null pronominals or verbs. It can be viewed as part of a paradigm of IF morphemes.17 (17)

Lakhota IF morphemes: Male speaker: Female speaker: Neutral:

declarative yeló ye ~ kštó Ø

interrogative huwó he he

imperative yo ye —

Zero morphemes which are part of a paradigm of closed-class grammatical categories do not violate the prohibition on phonologically null elements in the syntax, which applies only to phonologically null lexical items.

1.2.3.1 The Order of Morphemes Expressing Operators and the LSC One of the implications of treating operators as modifying different layers of the LSC is that the linear order of the morphemes expressing the operators reflects their scope in the LSC. When the relevant morphemes are on the same side of the predicator in the nucleus, those expressing nuclear operators will be closer to the nucleus than those expressing core operators, and they in turn will be closer to the nucleus than those expressing clausal operators. This hypothesis was first put forward in Foley and Van Valin (1984). In right-branching languages, these morphemes will typically be realized by prefixes, whereas in left-branching languages they are usually realized by suffixes. This is illustrated in Figure 1.8. Because aspect is a nuclear operator and tense a clausal operator, when they both occur on the same side of the nucleus, the morpheme expressing the aspect operator occurs closer to the nucleus than the morpheme expressing the tense operator, and this is what is found in the four languages in Figure 1.8.18 This generalization is very robust, and accordingly there are very few exceptions to it. If in a language tense is a prefix and aspect a suffix (or vice versa), this is not an exception or counterexample to this claim,

Figure 1.8 Examples of the ordering of aspect and tense markers in different languages

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

because no ordering relation between the two can be established; hence the ordering constraint does not apply in this case. This constraint applies to the ordering of operators modifying different layers of the LSC, but it does not restrict the order of multiple operator morphemes in a given layer. At the nuclear and core layers the possibilities are limited, namely, aspect and negation at the nuclear layer, and negation and deontic modality at the core level. With respect to aspect and negation, it would seem that the only logical possibility is negation > aspect; in a system with perfective and imperfective aspect, NEG [V-(im)perfective] is meaningful (i.e. ‘not [(in)complete state of affairs]’, e.g. ‘not [finished raining]’), but the other possibility is not (i.e. ‘(in)complete [NEG state of affairs]’, e.g. ‘finished [not raining]’). Van Valin and LaPolla (1997: 46) give an example of the interaction of aspect and negation from Qiang, a TibetoBurman language, and as predicted, negation has scope over aspect, and the order is negation-aspect-V. ɦɑ-m -tɕi-qɑ down-neg-asp-go.1sg ‘I haven’t gone yet.’ e

(18)

Qiang

At the core level, on the other hand, both scopes are possible. In a sentence like John must not leave the door open, the deontic modal can have scope over negation (i.e. ‘John is obliged not to leave the door open’), or the deontic modal can be in the scope of negation (i.e. ‘John is not obliged to leave the door open’). Interestingly, the second reading is stronger if the negation is contracted, as in John mustn’t leave the door open. This variation in scope is not reflected in the order of operator morphemes in a simple sentence, unlike in the paraphrase which contains two cores (see Section 1.3). It is, rather, signalled prosodically. The two readings can be distinguished clearly by different prosodic chunking of the sentence: {John must}{not leave the door open} vs. {John must not}{leave the door open}. The negative morpheme in English makes up for its fixed position by being able to scope in either direction. Multiple operators regularly co-occur at the clause level, as in Figure 1.6 in which tense and IF co-occur. There are four clausal operators (IF, evidentials, status (clausal negation, epistemic modality) and tense), and they fall into two groups: tense and status, on the one hand, and evidentials and IF, on the other. Tense and status are propositional modifiers and occur in both main and subordinate clauses. Evidentials and IF are utterance modifiers and are restricted to main clauses. In agglutinative languages in which operator morphemes occur in sequence, as in Lakhota, Kewa, Turkish, Tiwi and Qiang, the IF morpheme occurs farthest from the nucleus, if it has a fixed position; this is illustrated in the Lakhota examples in (15) and (16). In terms of the LSC operator projection, the IF and evidential operators modify the clause node which is immediately dominated by the sentence node; there is no such restriction on tense or status, hence they can occur freely in subordinate clauses. There is no

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

37

38

ROBERT D. VAN VALIN, JR.

scope asymmetry between tense and status; they modify at the same level, but because the tense and status morphemes must be linearized, languages show variation in the order of these morphemes.

1.2.3.2 Clausal Operators and the PrDP and PrCS The important distinction between the PrDP and the PrCS was characterized in terms of two primary differences: the PrDP is set off by an intonation break, and if the XP in the PrDP is a semantic argument of the predicator in the nucleus, then there can, and in some languages, must be a resumptive pronoun in the core, whereas the PrCS has neither of these features. The explanation for this opposition can be found in Figure 1.6 in the interaction between the constituent and operator projections. These differences follow from the fact that the PrCS is inside the clause and therefore in the scope of the IF operator, whereas the PrDP is outside of the clause and consequently is not in the scope of the IF operator. Being within the scope of the IF operator means that the XP in that position can be asserted, questioned or negated, possibilities denied to XPs outside the scope of the IF operator. With respect to wh-question formation, the following contrast is quite clear. (19)

a. [S [CL [PrCS What ] {did} [CORE Sally [NUC borrow ] from Bill? CORE] CL] S] b. [S [PrDP *As for whati ][CL{did} [CORE Sally [NUC borrow ] iti from Bill? CORE]

CL] S]

In (19a) what is in the PrCS, which is clause-internal and therefore in the scope of the IF operator. There is a single intonation contour over the clause, and there is no break after what. There is no need for a resumptive pronoun, because all of the arguments of the predicator are in the clause. The situation is very different in (19b), which is ungrammatical. Here the wh-expression occurs outside of the clause and therefore outside of the domain of the IF operator. It is not part of the speech act associated with this sentence. It cannot be interpreted as the focus of the utterance, unlike what in the PrCS in (19a). The intonation break following the PrDP is related to the IF operator. The intonation contour of the primary speech act begins at the first element in the clause19 and ends at the end of the clause. The PrDP is not part of the clause and accordingly is not part of the intonation contour. It is a prosodic unit of its own, and the intonation break is the transition from the prosodic contour of the PrDP and that of the clause. The usual test for whether an XP is asserted or not is the denial test with only the new information given, as in (20). (20)

Speaker 1: Mary bought a new Tesla. Speaker 2: No, a Mercedes.

This felicitous pair of utterances shows that the RP a new Tesla can be the focus of the assertion and therefore denied. This is not possible with an RP in the PrDP, as in (21a), but is fine with an XP in its default position, as in (21b), or in the PrCS, as in (21c).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (21)

a. Speaker 1: On Thursday, Mary will meet the journalist at the library. Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday b. Speaker 1: Mary will meet the journalist at the library on Thursday. Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday c. Speaker 1: Thursday Mary will meet the journalist at the library. Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday

The same restrictions hold with respect to forming questions. (22)

a. Speaker 1: On Thursday, will Mary meet the journalist at the library? Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday b. Speaker 1: Will Mary meet the journalist at the library on Thursday. Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday c. Speaker 1: Thursday will Mary meet the journalist at the library. Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday

Thus, the features distinguishing the PrDP and the PrCS are directly related to the fact that the scope of the IF operator is the clause. The same holds for the PoDP and PoCS; as the Dhivehi examples in (6) show, the PoCS can host whexpressions analogous to (19a) and contrastive focus phrases analogous to (4b). The PoDP, like the PrDP, cannot be asserted or questioned, as illustrated in (23a, b), nor can it host wh-expressions, as shown in (24). (23)

a. Speaker Speaker b. Speaker Speaker

(24)

a. a′. b. b′.

1: Mary bought them ice cream cones, the kids. 2: No, some popcorn/No, Bill did/??No, the adults. 1: Did Mary buy them ice cream cones, the kids? 2: No, some popcorn/No, Bill did/??No, the adults.

*Mary bought them ice cream cones, who? *Did Mary buy them ice cream cones, who? *Mary bought it for the kids, what? *Did Mary buy it for the kids, what?

Thus, recognizing the IF operator as having clausal scope leads to a deeper understanding of the differences between the PrDP and PoDP, on the one hand, and the PrCS and PoCS on the other.

1.2.3.3 The Status of Copular Verbs Like English be Copular verbs like English be present interesting issues regarding their representation in the LSC. In the Lakhota examples (11b), (13b) and (14) there is no copula of the kind required in their English translation. This suggests that in such constructions be is not the predicator in the nucleus in English but rather is just an auxiliary accompanying the actual predicator, namely, woman or tall. The same holds for the Tagalog example in (10b): bata ‘child’ is the predicator in (10b) and its English translation. Thus, this use of the English copula is not found in many other languages, and it would be odd to say that in English be is the predicator in these examples, but in Tagalog and Lakhota the predicator is the nominal or stative verb. Rather, the predicator in the constructions in all three languages is ‘woman’, ‘child’ or ‘tall’, and English requires the auxiliary be when the predicator is non-verbal.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

39

40

ROBERT D. VAN VALIN, JR.

Figure 1.9 Passive be in (25a) and progressive be and passive be in (25b)

There is another construction in English in which auxiliary be is obligatory, namely, the passive construction. It is possible to have two be auxiliaries in the same sentence, as in (25b), but the representation of the two is quite different, as can be seen in Figure 1.9. (25)

a. Max was interviewed by CNN during the insurrection. b. Max was being interviewed by CNN during the insurrection.

In (25a), the tense operator is realized on the passive auxiliary be, and its core-internal linear position marks declarative IF. In this construction was is attached to both projections: the tense and IF operators connect it to the operator projection, while the be to which these operators are attached is structurally required for passive nucleus formation.20 In (25b), the auxiliary was is attached to the operator projection three times: (1) it is part of the two-part progressive aspect morpheme, beþ-ing; (2) it signals past tense; and (3) its core-internal position in the linear string indicates declarative IF. Passive be is not connected to the operator projection; it hosts -ing, which is part of the expression of progressive aspect and is attached to the operator projection. Thus, passive be is always part of the constituent projection and may also be connected to the operator projection, whereas progressive be can only be connected to the operator projection. This follows from the fact that progressive be occurs only when the progressive aspect is expressed. Passive be, like copular be, is structurally required for proper nucleus formation and occurs even when there are no operators, as in the infinitives in (26). (26)

a. Max wanted/tried to be interviewed by CNN during the insurrection. b. Max wants to be stronger/a lawyer.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

It should be noted that there are other auxiliary verbs which can fulfil be’s structural role while contributing semantically to the predication. For clauses with adjectival predicates, seem and appear can replace be as in Bill is/seems/appears happy/stronger/healthy. Become and get can also serve this function: Mary became despondent, Sally got stronger. Get can also replace be in a passive construction, as in Max got interviewed by CNN during the insurrection.

1.2.3.4 Universal Operators Of all the operators discussed in RRG, only two are universal, found in every language: negation and IF. In every human language it is possible to make assertions, ask questions and give commands. That is fundamental to language as a system of communication. Hence IF is an operator in every language. This is not to say that the inventory of speech acts is exactly the same in every language; rather, it is a claim that in every language it is possible to make the three essential speech acts: asserting, questioning, commanding. Negation is essential for both communication and reasoning. With respect to the former, denials (e.g. ‘X is not the case’, ‘X did not do Y’) and negative commands (e.g. ‘Don’t do Z’) are essential and very important types of utterances. With respect to the latter, practical reasoning involves negation (e.g., ‘if I do X, I will not get Z, but if I do Y, I will get Z’). The exact detail of how negation works is not necessarily the same in every language, for instance a language may or may not have negative polarity items, but negation is part of the system in every language. A distinctive property of negation is that it is the sole operator that can modify all of the layers of the LSC: nuclear negation has just the nucleus in its scope, core negation has the core or some subpart, typically a core argument, in its scope, and clausal negation has the whole clause in its scope. One might expect that languages would typically have three negative morphemes, one for nuclear negation, one for core negation, and one for clausal negation, but this is in fact quite rare. Some languages have two negative morphemes with different scopes; for example, Barai, a Papuan language, has a negative morpheme ba which is the only element that can occur after the verb in Barai and has clausal scope (external negation), and another one, naebe, which occurs before the verb and can have nuclear or core scope (internal negation) (Foley and Van Valin 1984: 192–193). Korean also has two negative constructions, one which has nuclear scope, as shown in (27a), and one with core or clausal scope, as in (27b), from Yang (1994: 127). (27)

a. Chelswu-ka pap-lul manhi an - mek-ess-ta. -nom dinner-acc much neg-eat-pst-decl ‘Chulsoo did not eat much dinner (But he ate a little).’ b. Chelswu-ka pap-lul manhi mek-ci.ahn-ess-ta -nom dinner-acc much eat-neg-pst-decl ‘It is not the case that Chulsoo ate much dinner,’ ‘Chulsoo did not eat much dinner .’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

41

42

ROBERT D. VAN VALIN, JR.

The negative morpheme is an-/ahn-, and when it is prefixed to the verb, as in (27a), it has nuclear scope, but when it is suffixed to the verb, as in (27b), it may be interpreted as having core or clausal scope. Both Barai and Korean have two negative constructions, but the opposition in Barai is clausal vs. non-clausal, while in Korean it is nuclear vs. non-nuclear. The usual case is like that in English and other languages with a single negative in a fixed position with an ambiguous interpretation. (28)

Max didn’t buy a new car. a. He leased one. b. He just went window shopping c. He bought a new motorcycle. d. Bill did. e. It’s not the case that Max bought anything.

Nuclear scope Core scope Core (argument) scope Core (argument) scope Clausal scope

The scope of negation is tied to the focus structure of the clause, and consequently the interpretation depends on the context in which the sentence occurs.

1.2.4 Peripheries In the earlier versions of the LSC there was only a single periphery, which was attached at the core level. This proved to be wholly inadequate, and in Van Valin (2005) it was proposed that each of the layers of the spine of the LSC was modified by a periphery containing modifiers appropriate to the layer. Moreover, a distinction was made between phrasal and non-phrasal adjuncts, phrasal adjuncts being PPs and non-phrasal adjuncts being adverbs, in simple sentences.21 In complex sentences, phrasal adjuncts also include what are traditionally called adverbial clauses (to be discussed in Section 1.3). 1.2.4.1 Phrasal Adjuncts The two types of adjuncts are subject to different ordering constraints. Languages tend to have restrictions regarding the placement of phrasal adjuncts. In English, phrasal adjuncts occur after the core, as in (3), whereas in languages with primarily nucleus-final word order, they appear after the subject and before the remaining core arguments, for example Japanese. (29)

Hanako ga tosyokan de hon o yon-da. nom library in book acc read-pst ‘Hanako read a book in the library.’

The adjunct PP tosyokan de ‘in the library’ is linearly within the core, and the LSC diagram necessarily involves crossing tree branches, as in Figure 1.10. The PP tosyokan de ‘in the library’ is in the core-level periphery, because it expresses the location of the event described by the core, that is, Hanako’s reading a book, and consequently the representation of the hierarchical structure and the linear sequence of words leads to the crossing branches.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

43

Figure 1.10 Default placement of phrasal adjuncts in Japanese

Such crossing branches are allowed in RRG, because the tree represents semantic dependencies primarily rather than purely syntactic constituent structure. German presents an interesting case, as there has long been a controversy as to whether it is fundamentally verb-medial or verb-final. With respect to the default placement of adjunct PPs, it patterns with Japanese and not with English. (30)

a. Max hat nach der Auseinandersetzung mit Maria eine ganze Flasche Wein getrunken. has after the argument with a entire bottle wine drunk ‘Max drank an entire bottle of wine after the argument with Mary.’ b. Max trank nach der Auseinandersetzung mit Maria eine ganze Flasche Wein. drank after the argument with a entire bottle wine ‘Max drank an entire bottle of wine after the argument with Mary.’

In (30a) the nucleus is core- (and clause-) final, as in Japanese, and the order of arguments and adjuncts is the same, too: subject, core-level adjunct PP, direct object. In (30b) the nucleus is in second position in the core, as in English, but the default order of arguments and adjuncts is the same as in (30a) and Japanese. On the other hand, the word order in (30b) is impossible in English; nothing is allowed to occur between the nucleus and the following direct core argument (‘direct object’).22 Phrasal adjuncts may also appear in positions other than their default position. In English and many other languages, they may appear in the PrDP and function as a frame-setting topic. (31)

a. After the party, Sam went home and watched TV for a while. a′. After the party, where did Sam go? b. Because of the cold weather, vaccine distribution has been slowed down significantly. b′. Because of the cold weather, has vaccine distribution been adversely affected?

The adjunct PP in the PrDP does not express the topic of the sentence, unlike in Figure 1.4 or (7), because (31a) is not about the party but rather about Sam and (31b) is not about the cold weather. Rather, after the party and because of the cold weather set the frame of reference in terms of which the following

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

44

ROBERT D. VAN VALIN, JR.

clause(s) should be interpreted. In the questions the initial PP is not in the scope of the IF operator, hence is outside the clause in the PrDP. It is also possible for adjunct PPs to occur in the PrCS as contrastive framesetting topics, as in (32). (32)

a. At the mall Sally couldn’t find what she wanted, but in a small boutique she found it. b. Before lunch Bill drank only coffee, but after lunch he ate a lot of snacks.

These initial PPs are not set off prosodically like the ones in (31) and are part of the assertion, that is, they are within the scope of the IF operator, and are therefore in the PrCS. This entails that they do not occur in question, because the PP is in the actual focus domain (AFD; see Chapter 11). (33)

a. ??Before lunch did Bill drink only coffee, but after lunch did he eat a lot of snacks? b. *Before lunch what did Bill drink, but after lunch what did he eat?

Both examples are fine if there is a pause after the initial PPs, which indicates that the structure is the same as in (31), not (32). It was mentioned with respect to (3) in Figure 1.3 that the order of adjunct PPs is constrained by the layer they modify, with core-level modifiers preceding clause-level modifiers in a right-branching language like English. When there are multiple adjunct PPs which modify at the same level, however, the order is ‘free’ and influenced by information structure, as in (35). (34)

a. a′. b. b′.

Mary ate a cookie after lunch despite her diet. ??Mary ate a cookie despite her diet after lunch. Sally met Max in the library because of the cold weather. ??Sally met Max because of the cold weather in the library.

[¼(3)]

(35)

a. a′. b. b′.

Mary ate a cookie after lunch in the garden. Mary ate a cookie in the garden after lunch. Sally met Max in the library after the lecture. Sally met Max after the lecture in the library.

(36)

a. Vaccine distribution has been adversely affected despite the government’s best efforts because of the cold weather. b. Vaccine distribution has been adversely affected because of the cold weather despite the government’s best efforts.

PPs headed by after and in are core-level modifiers, while those headed by despite and because of are clause-level modifiers, and consequently the ordering restriction in (34) and the lack thereof in (35) and (36) follows from the analysis of phrasal adjuncts in terms of the LSC.

1.2.4.2 Non-Phrasal Adjuncts The class of non-phrasal adjuncts consists primarily of adverbs, and they have the same possibilities of occurrence as phrasal adjuncts and then some, as in (37).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (37)

a. b. c. d. e. f.

Mary quietly ate a cookie after lunch in the garden despite her diet. Mary ate a cookie quietly after lunch in the garden despite her diet. Mary ate a cookie after lunch quietly in the garden despite her diet. Mary ate a cookie after lunch in the garden quietly despite her diet. ??Mary ate a cookie after lunch in the garden despite her diet quietly. Quietly(,) Mary ate a cookie after lunch in the garden despite her diet.

Quietly is a type of manner adverb and is a core-level modifier, which accounts for the oddity of it following the clause-level adjunct PP in (37e), just as in (34a′, b′). If it is replaced by a clause-level adverb (e.g. evidently), the situation is somewhat different. (38)

a. b. c. d. e. f.

Mary evidently ate a cookie after lunch in the garden despite her diet. ?Mary ate a cookie evidently after lunch in the garden despite her diet. ?Mary ate a cookie after lunch evidently in the garden despite her diet. Mary ate a cookie after lunch in the garden evidently despite her diet. Mary ate a cookie after lunch in the garden despite her diet evidently. Evidently, Mary ate a cookie after lunch in the garden despite her diet.

The oddity of (38b) and (38c) results from having a clause-level modifier occurring in the midst of core-level modifiers. There is a striking contrast in acceptability between (37e) and (38e), which is a function of the difference in the level of modification (core vs. clausal) of the adverbs. The two ordering possibilities in (38d, e) show that it makes no difference with respect to the sequence whether the adjuncts are phrasal or not, as long as they are modifiers of the same layer.23 As (37) and (38) make clear, adverbs can occur in positions which are not part of the default periphery for phrasal adjuncts, and it was proposed in Van Valin (2005) that all adjuncts not in the default periphery, the Pr/PoDP or the Pr/PoCS are in a periphery specific to them. This is exemplified in Figure 1.11. The LSC is crucial for the analysis of the scope of modification of both phrasal and non-phrasal adjuncts, but it has significant limitations, namely,

Figure 1.11 Periphery for individual adverb in (37a)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

45

46

ROBERT D. VAN VALIN, JR.

there are only three layers but there are many more types of adverbs. This is illustrated in (39); the list of types of adverbs is not exhaustive. (39)

Adverbs and the LSC a. Clausal modifiers: 1. Speech act modifiers: candidly, honestly, . . . 2. Evidential: evidently, allegedly, . . . 3. Speaker attitude/judgement: appallingly, unfortunately, . . . b. Core modifiers: 1. Temporal: yesterday, tomorrow, . . . 2. Manner a. Pace: quickly, slowly, . . . b. Intent: deliberately, intentionally, . . . c. Performance: carefully, sloppily, . . . d. Action: violently, weakly, . . . c. Nuclear modifiers: 1. Aspectual: completely, continuously, . . .

Two points deserve mention. First, a given adverb can be used in more than one category, as in the well-known example of cleverly as in Mary hid the jewellery cleverly (manner-performance) vs. Cleverly, Mary hid the jewellery (speaker attitude/judgement) vs. Mary cleverly hid the jewellery (ambiguous). Second, even though the semantic domains of operators and adverbs are similar, adverbs are not operators; adverbs are open-class lexical items, whereas operators are closed-class grammatical elements. This is particularly important when it comes to distinguishing the operator tense and temporal adverbs; the former is a clausal operator, whereas the latter is a core-level modifier. This can be seen clearly in sentences like (40). (40)

[CLAUSE [CORE1 Fred told Sally][ yesterday] [CORE2 to meet him][ in Paris Friday]].

This sentence consists of a single clause with a single tense operator which is made up of two cores, each of which is modified by a different temporal adverb. Each temporal adverb modifies the core it occurs with, while the tense operator has scope over the entire clause. It is reasonable to speculate that the largest class of adverbs is manner adverbs, which has a number of subtypes, all of which are core-level modifiers. Despite them all being core modifiers, the ordering among them does not appear to be unconstrained. (41)

a. b. c. d. e. f. g.

Mary will intentionally speak more clearly tomorrow. ?Mary will tomorrow speak more clearly intentionally. Mary will speak more clearly intentionally tomorrow. Mary will speak intentionally more clearly tomorrow. *Mary will more clearly speak intentionally tomorrow. ?Mary will intentionally speak tomorrow more clearly. ?Mary will tomorrow speak intentionally more clearly.

There are two manner adverbs, intentionally (intent) and clearly (performance), and one temporal adverb, tomorrow, in these sentences. Clearly seems

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

to be the most constrained, because it is strongly preferred after the verb, whereas intentionally can appear either before or after the verb. Tomorrow is best in final position, and it is the only one of the three that can appear in initial position without changing its meaning; intentionally can’t appear at all initially, and clearly is interpreted as an expression of speaker attitude/ judgement rather than manner. (42)

a. Tomorrow(,) Mary will intentionally speak more clearly. b. Clearly, Mary will intentionally speak more intelligibly tomorrow. c. *Intentionally, Mary will speak more clearly tomorrow.

Thus, core-modifying adverbs differ in terms of their scope of modification within the core. Temporal adverbials appear to have the widest scope, and the relative scope relations among the manner adverbs remain to be worked out. This situation is reminiscent of the constraints on the ordering of adjectives, all of which modify the same level of the RP (see Section 1.2.5). (43)

a. b. c. d.

the expensive big red hand-made scarf *the big hand-made red expensive scarf *the expensive hand-made red big scarf *the red expensive big hand-made scarf

1.2.4.3 Extended Peripheries It was pointed out in Section 1.2.4.1 that right-branching languages like English tend strongly to have the default periphery for phrasal adjuncts, PPs and clauses, after the core, while left-branching languages like Japanese and German have it after the subject and before the direct object and the nucleus.24 It has been discovered that there are instances in which the periphery for phrasal adjuncts gets extended to before the nucleus in right-branching languages. Examples can be found in Italian (Bentley 2008: 278) and in Tagalog (Latrouite and Van Valin 2021). The example from Italian is given in (44). (44)

a. Il libro, per sbaglio, Luca ha strappato, non il quaderno. the book by mistake has torn not the workbook ‘It is the book that Luca tore by mistake, not the workbook.’ b. [[[PrCS Il libro], [PERIPH per sbaglio], [CORE Luca ha strappato]], [PoDP non il quaderno]]

In (44) there is an adjunct PP, per sbaglio ‘by mistake’, which appears between the PrCS RP il libro ‘the book’ and the core-initial subject Luca. Normally, phrasal adjuncts occur after the core, but in this case, the adjunct PP is in its own periphery, analogous to the non-phrasal adjunct in Figure 1.11. The Tagalog example is from Schachter and Otanes (1972: 498–499); it involves a construction they label ‘non-emphatic inversion’, in which an adverbial is displaced from its normal post-core position to the beginning of the clause.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

47

48

ROBERT D. VAN VALIN, JR.

(45)

a. Pasok nila ang silid nang bigla. enter 3pl.gen nom room lnk suddenly ‘They suddenly entered the room’, ‘They entered the room suddenly.’ b. Bigla nila-ng pasok ang silid. suddenly 3pl.gen-lnk enter nom room ‘Suddenly, they entered the room.’

In the discussion of (5) from Tagalog, it was pointed out that the personal pronouns are second-position clitics, and in (45b) nila ‘3plgen’ follows bigla ‘suddenly’, which shows it is in clause-initial position. It might be concluded that it is in the PrCS, but it lacks the contrastiveness of PrCS XPs in Tagalog. In fact, (46) shows that it is in the same place as per sbaglio in (44): between the PrCS and the core. (46)

a. Nag-trabaho ang lahat nang mabilis kanina-ng umaga. av.rls-work nom all lnk quickly this-lnk morning ‘Everyone worked quickly this morning.’ b. Kanina-ng umaga mabilis na nag-trabaho ang lahat. this-lnk morning quickly lnk av.rls-work nom all ‘It was this morning that everyone worked quickly.’ c. [CLAUSE [PrCS Kanina-ga umaga][PERIPHERY mabilis-na][CORE nagtrabaho ang lahat]]

The default position for the two adjuncts is given in (46a); they both follow the single core argument. In (46b), on the other hand, they appear before the nucleus, with the temporal adjunct kanina-ng umaga ‘this morning’ in clause-initial position in the PrCS, as indicated by the contrastive interpretation (cf. (5b)), and the adverbial mabilis ‘quickly’ between it and the coreinitial nucleus nagtrabaho ‘worked’. A comparison with (44b) reveals that the structures are completely parallel. It was mentioned earlier that in German the default location for adjuncts, both phrasal and non-phrasal, is after the subject and finite verb or auxiliary and before the remaining elements in the core. One of the relatively unique features of German main clause structure is what is traditionally characterized as the ‘prefield’, the clause-initial position before the finite verb or auxiliary which must be obligatorily filled. Wh-expressions normally occur in it in questions, suggesting it is the PrCS, but in many sentences, the XP in it is not contrastive or information-structurally special in any way. In (30), for example, Max is a normal, topical subject in the prefield. In (47) it is occupied by an adjunct which need not have any special informationstructural status. (47)

a. Nach der Vorlesung hat Dieter in der Bibliothek Almuth getroffen. after the lecture has in the library met ‘After the lecture Dieter met Almuth in the library.’ b. Gestern hat Dieter in der Bibliothek Almuth getroffen. yesterday has in the library met ‘Yesterday Dieter met Almuth in the library.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Ruhnau (2011) gives an RRG analysis of the prefield and argues that in (47a, b) the clause-initial adjunct is a peripheral constituent, not in the PrCS. The prefield is thus rather heterogeneous structurally; it can be a core-argument RP, as in (30), or PP, it can be the PrCS in wh-questions and contrastive topic or focus sentences, or it can be an adjunct in a periphery.

1.2.5 The Structure of Phrases The LSC is the model for the analysis of structure in RRG. Accordingly, phrases smaller than a clause are also analysed as having a layered structure, in particular PPs, MPs and RPs. 1.2.5.1 Adpositional Phrases The structure and function of prepositional and postposition phrases is discussed in detail in Ibáñez Cerda (Chapter 10), and consequently the discussion in this chapter will be brief. The most important distinction made regarding the structure of PPs is the contrast between predicative and non-predicative PPs, adopted from Bresnan (1982). Predicative PPs, as the name implies, have an adpositional predicate in the nucleus of the PP; this will be discussed in the section devoted to the semantic representation of sentences in RRG. The object of the adposition is a core argument in the layered structure of the adpositional phrase (LSPP), as illustrated in Figure 1.12 with the German PPs aus dem Haus ‘out of the house’ and dem Haus gegenüber ‘across from the house’. German has both prepositions (e.g. aus ‘from, out of’) and postpositions (e.g. gegenüber ‘across from’); in both structures there is an adpositional predicate in the nucleusP, and the object is in the coreP. This is the only endocentric structure posited in RRG. There are a few significant differences between the LSC and the LSPP. First, there is only one layer above the coreP, the PP layer, unlike in the LSC with clause and sentence above the core. Second, there are no extra positions like the Pr/PoDP or the Pr/PoCS. Third, there is no operator projection, because there are no operators modifying PPs. Fourth, there

Figure 1.12 German predicative PPs

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

49

50

ROBERT D. VAN VALIN, JR.

aren’t peripheries at every level, as in the LSC, as there are very few lexical modifiers for PPs. The best examples are adverbs like right or immediately, as in right behind the chair or immediately before the accident. They would appear to be coreP-level modifiers. Predicative PPs share the property of recursion with clauses in a limited way, as in The mouse ran out from under the bed. (48)

Structure of the recursive PP out from under the bed [PP[DIR out[COREp[NUCp from][COREp[NUCp under] the bed COREp]COREp]

DIR] PP]

In locative inversion constructions, the whole complex PP must be inverted, that is, Out from under the bed ran a mouse, not, for example, *Under the bed ran a mouse out from/Under the bed ran out from a mouse, but From under the bed a mouse ran out is fine, showing that out is a directional modifier rather than a preposition. Non-predicative PPs, on the other hand, do not have a full layered structure as in Figure 1.12. The adposition in the PP is not a semantic predicate, and therefore the object is not a semantic argument of the adposition. Rather, the adposition is functionally like a case marker, and the resulting structure is not endocentric. In RRG non-predicative adpositions are either assigned by rule or listed in the lexical entry of the verb. The classic example of a non-predicative adposition is the to marking the recipient argument of a transfer verb. This is illustrated in Figure 1.13 from English and Bulgarian; in both Old English and Old Church Slavonic the third argument of a transfer verb carried the dative case, and after English and Bulgarian lost their case systems, the dative in this context was replaced by a preposition, to in English and na in Bulgarian. The primary content in these phrases is the RPs, and therefore these are exocentric phrases. The adpositions are basically case markers,25 and in English and many other languages, their lack of semantic significance is shown by the fact that they are not obligatory, as in (49). (49)

a. a′. b. b′.

Nancy gave the flowers to the woman. Nancy gave the woman the flowers. Sam presented the trophy to the winner. Sam presented the winner with the trophy.

Figure 1.13 Non-predicative PPs in English and Bulgarian

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Other examples of alternating non-predicative prepositions can be found in (50). (50)

a. a′. b. b′. c. c′.

Bill loaded the vehicle with grapes. Bill loaded the grapes into the vehicle. Max drained the water from the swimming pool. Max drained the swimming pool of its water. Mary sprayed paint on the wall. Mary sprayed the wall with paint.

In some cases each member of the alternation occurs with a different verb, as in The gang stole $5000 from the bank vs. The gang robbed the bank of $5000. Non-predicative adpositions cannot be stacked the way predicative adpositions can be, as in the earlier example out from under the bed. The distinction between predicative and non-predicative adpositions is very important for the analysis of many grammatical phenomena crosslinguistically.

1.2.5.2 Modifier Phrases Canonically, nouns are modified by adjectives and nouns, while verbs and adjectives are modified by adverbs, but this is not always the case. The examples in (12) are repeated here, as they do not fit the traditional scheme. (12)

a. The God is dead philosophers are mostly dead. b. My grandson likes to give me the who’s the boss now, silly old grandpa wink frequently.

In (12a) the modifier in the ‘adjective slot’ is a clause, and in (12b) it is a sentence complete with PrCS and PoDP. That they are in an ‘adjective slot’ and are not some kind of special relative clause or the like is shown by the fact that they may co-occur with regular adjectives in a sequence of modifiers, as in (12′). (12′)

a. The notorious [God is dead] continental philosophers are mostly dead. b. My grandson likes to give me the triumphant [who’s the boss now, silly old grandpa] sly wink frequently.

There are other examples of phrasal modifiers in a position which is reserved for adjectival heads: English *the very proud of her daughter woman is impossible, but it is possible in German, die auf ihre Tochter sehr stolze Frau [the on her daughter very proud woman] ‘the woman very proud of her daughter’. MPs account for these modifying expressions, since they are nonendocentric like RPs and nuclei. MPs occur in the peripheries of the element modified; they are not involved in predicative uses of nouns, adjectives and adverbs. The German example is important, because it shows that MPs have a layered structure like other phrases. The nucleus of the MP is the adjective stolz ‘proud’, which takes a single core argument, the PP auf ihre Tochter ‘of [on] her daughter’. The structure of the whole RP die auf ihre Tochter sehr stolze

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

51

52

ROBERT D. VAN VALIN, JR.

Figure 1.14 The layered structure of the MP

Frau ‘the woman very proud of her daughter’ is presented in Figure 1.14. Two points need to be mentioned. First, degree modifiers like sehr ‘very’ do not head phrases and therefore are not in an MP. Second, the structure of RPs, including the status of definite articles like die ‘the’ will be discussed in the next section. Like the predicative PP, the MP has one less layer than the LSC, and there are no operators modifying it. There are peripheries at the nuclear level for degree modifiers (illustrated) (see also Fleischhauer 2016) and at the core level for manner adverbial modifiers, for instance the rapidly contained wildfire, the beautifully embroidered dress, the carefully constructed maze. In the English translation of this RP, the MP, which is a reduced relative clause, follows the head noun; when the MP consists of a participle, some, but not all, may occur either pre-head or post-head, for example the murdered man vs. the man murdered, the running man vs. the man running, the embroidered sweater vs. *the sweater embroidered (but the sweater embroidered by Mary). It was pointed out that the unusual phrasal modifiers in (12) cannot be considered some kind of exotic relative clause or the like, as the examples in (12′) show. What is so unusual about them is that they exhibit asymmetrical embedding, that is, a larger unit (clause, sentence) is embedded in a smaller unit (nucleus); this will be discussed in Section 1.3. The structure of the first RP in (12a) is given in Figure 1.15; the structure of the RP with the sentence as modifier in (12b) can be found in Van Valin (2008: 173). The fact that the embedded clause is under the nucleus rather than being under the MP node follows from the examples in (12′), in which the phrasal modifier occurs in the middle of a series of modifiers. The structure of RPs with multiple modifiers, for example the very expensive beautifully embroidered

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.15 The structure of the RP in (12a)

red sweater, involves multiple sequentially embedded coreM nodes, as illustrated in Figure 1.16. The recurring embedded units are coreM nodes for the reason given earlier: the modifying expression can be modified by manner adverb, which is a core-level modifier.

1.2.5.3 Reference Phrases RPs have a more complex layered structure than PPs and MPs. They have operators, one of which, definiteness, is represented in Figures 1.14–1.16. They have peripheries at three levels, nucR, coreR, and RP, and they have optional positions analogous to the Pr/PoCS and the Pr/PoDP. However, like PPs and MPs, the layered structure of RPs has one layer less than the clause, as is clear in the figures mentioned above. As with the operators modifying the layers of the LSC, the RP operators are closed-class, grammatical elements. The operators for each layer are associated with a specific semantic domain: the nuclearR operators express qualitative features of the referent, the coreR operators express quantitative characteristics of the referent, and the RP-level operators locate the referent within the immediate common ground, which includes the discourse

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

53

54

ROBERT D. VAN VALIN, JR.

Figure 1.16 Structure of RP with multiple modifiers

context and the physical environment. These distinctions derive from the proposals in Rijkhoff (1990, 2002). The primary nuclear operator in the clause is aspect, which is realized prototypically by the imperfective– perfective opposition, and a number of linguists, for example Jackendoff (1990) and Talmy (2000), have argued that the imperfective–perfective contrast for verbs has, as its analogue among nouns, the count–mass distinction, with mass and imperfective, on the one hand, and count and perfective, on the other, going together. If one needs to render a mass noun (e.g. water) countable, languages make use of measure phrases or numeral classifiers, for example two glasses/bottles/drops of water. A related use of nominal classifiers is to specify the relevant semantic properties of a referent. Becker (1975), for example, gives the following Burmese examples of classifiers contributing to the interpretation of an RP. e

‘river one place’ (e.g. destination for a picnic)

e e e e e e

(51)

‘river one line’ (e.g. on a map) ‘river one section’ (e.g. a fishing area) ‘river one distant arc’ (e.g. a path to the sea) ‘river one connection’ (e.g. tying two villages) ‘river one sacred object’ (e.g. in mythology) ‘river one conceptual unit’ (e.g. in a discussion of rivers in general) ‘river one river’ (the unmarked case)

a. myiɂ t yaɂ river one CL b. myiɂ t tan c. myiɂ t hmwa d. myiɂ t ′sin e. myiɂ t θwɛ f. myiɂ t ′pa g, myiɂ t khu′ h. myiɂ t myiɂ e

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

This nuclearR operator will be termed nominal aspect, and it covers the count–mass distinction and the use of classifiers, numeral or nominal. The core-level operators in the RP are concerned with quantity and quantification. The primary operator notion at this level is number. It is unquestionably closed class and grammatical. Less clear-cut is whether numerals and quantifiers are operators belonging to this domain. In previous work in RRG, numerals were assumed to be coreR operators, but there are significant problems with that analysis. First, numerals are not a closed class; in fact, they are an infinite open class. Second, grammatically they behave like lexical items, not grammatical morphemes. For example, they can function as arguments in elliptical contexts and sometimes as part of the predicator in the nucleus; number cannot. (52)

a. I’ll take two. b. Q: How many people will participate? A: Seven should./The expected number is seven.

Quantifiers present a range of behaviour, and this makes it impossible to give the whole class a uniform, consistent analysis. There are those that act like lexical items, as well as modifiers (e.g. Our problems are many, our options are few vs. We have many problems and few options). There are pairs of quantifiers which are very close in terms of meaning but which have quite different grammatical behaviour. For example, each and every are such a pair; each dog barked and every dog barked are both fine, but only each can stand alone in an elliptical context, for instance What did the dogs do? Each barked/*Every barked vs. Each/every one barked. They are a closed class but share important properties with lexical items. Accordingly, they should be analysed as lexical modifiers at the coreR-level, together with numerals. Hence grammatical number is a coreR operator and is linked to the operator projection of the RP, while the other two categories are lexical modifiers which occur in the coreR-level periphery. The other possibility for analysis as a coreR-level operator is negation, as in no book or its German counterpart kein Buch. Given that negation is uncontroversially an operator in the clause, it should be analysed as one in the RP. As in the clause, negation can modify at more than one level in the RP. NuclearR negation is expressed by no, e.g. no (brown) dogs were seen today, as it takes scope only over bare nouns or adjective þ noun. It cannot have scope over quantifiers or RP-level modifiers, e.g. demonstratives; not must be used (e.g. *no every student vs. not every student, *no that book vs. not that book). Interestingly, no can take numerals in its scope, as in no two snowflakes are alike, which suggests that numerals are more akin to adjectives (nuclearR-level modifiers) than to quantifiers (coreR-level modifiers). The third layer for operators is the RP level, which locates the referent of the RP in the immediate common ground (Berio et al. 2017). The immediate common ground consists of the physical and social environment in which the interaction takes place and the discourse context. The two notions that

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

55

56

ROBERT D. VAN VALIN, JR.

are relevant here are definiteness and deixis. Definiteness covers several related oppositions: referential vs. non-referential, specific vs. non-specific, and definite vs. indefinite. These distinctions are canonically expressed by articles, but it is well known that many languages lack articles. Nevertheless, these notions are relevant in article-less languages and are expressed by word order, demonstratives, and/or the numeral meaning ‘one’ (Balogh et al. 2020), for example. Lakhota has a particularly rich system of articles, with two definite articles, and an astonishing number of indefinite articles (see Ullrich 2011: 810–815). The central distinctions are illustrated in the following dialogue (Ullrich and Black Bear 2018: 108). (53)

a. Okíˇ chize olówaŋ waŋží yéksuya he? War song art you.remember q ‘Do you remember a war song?’ b. Haŋ, okíˇ chize olówaŋ waŋ wéksuye. yes war song art I.remember ‘Yes, I remember a (certain) war song.’ ˇ c. Olówaŋ kiŋ líla onáh’uŋ wašté. song art very good.to.listen.to ‘The [war] song is really beautiful’ ˇ d. Yéksuya he? John Vietnam etáŋhaŋ glí yuŋkháŋ you.remember q from return.home and.then olówaŋ k’uŋ hé kᡠge. song art that he.made.it ‘Do you remember? John came back from Vietnam, then he made the aforementioned [war] song.’

Waŋží in (53a) is the singular indefinite non-specific article (plural etáŋ); an alternative translation would be ‘Do you remember any war songs?’, which is clearly non-referential and avoids the ambiguity of English a(n). In (53b) waŋ is the singular specific indefinite article (plural eyá), hence the translation ‘a certain’. In the next sentence, (53c), the song is referred to again and is marked by kiŋ ‘the’, while in (53d) the other definite article, k’uŋ ‘the (aforementioned)’ is used. In terms of the immediate common ground, kiŋ can be used to refer to something in the physical environment which has not been explicitly mentioned before, as in ‘Please shut the door’, or, as in (53c) it can refer to a discourse antecedent. It is the default definite article. The other definite article, k’uŋ ‘the (aforementioned)’, can only be used to refer to something that has been explicitly mentioned in the discourse; it could not be used in the ‘Please shut the door’ situation described above, nor can it be used in associative anaphora, as in ‘I saw Mary’s new house. The kitchen was huge’. These articles would be linked to the operator projection of the RP at the RP-level, as shown in Figures 1.14–1.16. In languages without articles and which use word order to signal the discourse status of an RP, there is no definiteness operator to be linked to the operator projection. Rather, information-structural principles will supply the interpretation. This is analogous to languages without tense which rely on temporal

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Table 1.2 Operators in the RP NucleusR CoreR RP

Nominal aspect Number, negation Definiteness, deixis

adjuncts to express tense-related notions, but there is no tense operator in the operator projection of the LSC in them. The other RP-level operator is deixis, which is primarily instantiated by demonstratives, and whose canonical function is to locate referents spatially. This prototypical function relates to the physical environment part of the immediate common ground, but demonstratives can also refer to elements in discourse. Languages differ as to whether they allow articles and demonstratives to co-occur in the same RP. English, for example, does not allow it (e.g. *that the dog/*the that dog), but Lakhota clearly does, as shown in (11), (13), (14), (15a), (16) and (53d). Pure definite articles carry no deictic value, but demonstratives indicate that an RP is minimally referential and specific and would normally be construed as definite; hence they carry both operator values when they occur without an accompanying article. The operators in the layered structure of the reference phrase (LSRP) may be summarized as in Table 1.2. It was mentioned earlier that RPs, unlike PPs and MPs, have a position somewhat analogous to the positions outside of the core in the LSC (for English, the PrCS and the PrDP). Its canonical use is to host the possessor RP in a possessive construction (e.g. John’s book), but it is not limited to possessive constructions. In constructions involving complex nominals, the possessor position may host the equivalent of the core-internal subject of a proposition, as in Mary’s rejection of the marriage proposal, the rioters’ assault on the Capitol, a wh-element in the PrCS, as in Whose acceptance of the offer surprised the faculty the most?, Which assault on the Capitol are you talking about?, or a frame-setting topic in the PrDP, for example Tomorrow’s march on Washington will be a very tense affair, Yesterday’s destruction of the town by the earthquake shocked the nation. This position is called simply the RP-initial position (RPIP), and it collapses the core-initial subject position, the PrCS and the PrDP into a single position. That there is only a single position in an RP in contrast to a sentence is shown by the contrast between Yesterday, who won the match? vs. *Yesterday’s whose winning of the match. Some left-branching languages have a corresponding RP-final position (RPFP). Languages like English do not allow articles or demonstratives to co-occur with possessors in the RPIP, as shown by *the John’s book/*John’s the book; the possessor RP signals that the RP is definite, and therefore the genitive RP or pronoun in the RPIP is attached to both constituent and operator projections. In order to separate definiteness from possession, it is necessary for the possessor to occur after the head noun (e.g. the book of John’s vs. a book of

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

57

58

ROBERT D. VAN VALIN, JR.

Figure 1.17 General structure of RPs

John’s). In languages like Portuguese, in which articles and possessors cooccur, the possessor RP would be attached to just the constituent projection. The derived nominal destruction inherits the argument structure of the verb destroy from which it is derived (Nunes 1993). If both actor and undergoer arguments are realized, there are three possible patterns: the destruction of the village by the earthquake, the earthquake’s destruction of the village, and the village’s destruction by the earthquake. This illustrates how a nominal can have core arguments, and it can have locative and temporal adjuncts, for example the destruction of the village by the earthquake yesterday, the construction of the bridge in New York City. These adjuncts are in the coreR-level periphery in the RP, just as in the LSC. There are also RP-level and nuclearR-level peripheries, and they are associated with restrictive modification (nuclear) and non-restrictive modification (RP). The general structure of the LSRP in a right-branching language is given in Figure 1.17; peripheries are omitted. Two English RPs are given in Figure 1.18; the first one is a derived nominal with two semantic arguments and a locative adjunct, while the other illustrates an RP with modifiers. (‘NML’ stands for ‘numeral’; it is not in an MP because it does not potentially head a phrase.) The examples presented thus far are from familiar languages in which RPs are for the most part headed by nouns or derived nominals. The Tagalog sentence in (10b) presents a rather different situation; in this case, the predicator in the nucleus is a noun, while the head of the RP argument expression is a verb, and there is no nominalizing or verbalizing derivational morphology. Rather, the occurrence of the verb tumawa ‘laughed’ with the nominative case particle (CSP26) ang induces a type-shift, yielding the referring expression ‘the one who laughed’. The structure of (10b) is presented in Figure 1.19. The declarative IF operator is placed in the immediate post-nuclear position, because if it were a yes/no question, the question clitic ba would occur

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.18 English RPs

Figure 1.19 The structure of (10b) from Tagalog

in second position, that is, Bata ba ang tumawa?’Is the one who laughed a child?’ (cf. (7)). In Figures 1.14–1.16 and 1.18 the modifiers expressing properties or characteristics of the head of the RP occur in the nuclear periphery, which is the locus of restrictive modification in the RP. Non-restrictive modification, on the other hand, is associated with the RP-level periphery. In simple sentences this is primarily realized through appositive modifiers, for example Fred Jones, a lawyer, was arrested by the FBI on terrorism charges. In complex sentences involving relative clauses, restrictive relative clauses occur in the nuclear periphery and non-restrictive ones in the RP-level periphery. This

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

59

60

ROBERT D. VAN VALIN, JR.

correctly predicts that when a complex RP contains a restrictive and a nonrestrictive relative clause, the restrictive clause must be closer to the head of the RP than the non-restrictive one. This is related to the ordering principle involving nuclear and clausal operators discussed in Section 1.2.3.1. (54)

a.

I called the woman that I met on the train yesterday, who turned out to be a potential investor, to set up a meeting with our company. b. *I called the woman, who turned out to be a potential investor, that I met on the train yesterday to set up a meeting with our company.

1.2.6 Syntactic Templates and the Syntactic Inventory Syntactic structures must be represented and stored in a grammar. One way to do this is via phrase structure rules, as in classical transformational grammar. RRG takes a different approach, representing syntactic structure as templates and storing them in what it calls the syntactic inventory (SI). Examples of syntactic templates are given in Figure 1.20. Syntactic templates are pieces of the LSC which combine to form the structure of sentences; see Figure 1.21. The templates are language-specific, but they are composed of the universal and non-universal pieces of the LSC in Figure 1.20. There are five core templates, three for verbal predicators and two for non-verbal predicators. The range of core templates in the syntactic inventory of a language reflects the range of valence patterns of verbs and other predicators in that language. The most basic ones are given here: a one-argument template, which would canonically be used with oneplace intransitive verbs (but see below); a two-argument template, which could be used with a simple transitive verb (two RPs) or with an intransitive

Figure 1.20 Some syntactic templates for English

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

verb with a second oblique argument such as sit on, arrive at, pray for; a three-argument template, which can accommodate three RPs (Mary gave Chris a present), two RPs and a PP (Mary gave a present to Chris) or one RP and two PPs (Sam ran from the school to the fire station). These possibilities highlight the advantage of having underspecified templates; if the RPs or PPs were specified in the core templates, then many more would be needed. Because English word order is so rigid, especially with respect to the position of the nucleus, it is possible to include the nucleus in the core template. That would not be possible in German or Croatian, for example, because the nucleus can occur in a variety of positions within the clause. The other two core templates are for non-verbal predicators, which require an auxiliary, normally be, as discussed in Section 1.2.3.3. The one-place template is for sentences like The beer is cold or The weather is bad, and the two-place template is for sentences such as Mary is happy about the election results or Sam is upset with Sally. Below them are two operator projection templates, the first with the obligatory operators in English, tense and IF, and the second one adding an aspectual operator, which could be either progressive or perfect. In the upper right are the templates for PPs, both predicative and non-predicative, and there are three RP templates: the first two are for pronouns and proper nouns, which do not take operators or modifiers, and accordingly lack a layered structure. The third template is for common count nouns; the obligatory operators, definiteness and number, are included. Finally, below them are the most frequent ‘extra-spinal’ positions, the PrDP and the PrCS, as well as the adjunct and modifier hosting peripheries and the MP. If the sentence does not have any extra positions (PrDP, PrCS), then the ‘cap’ at the bottom is employed. The templates are stored in the SI, and when a semantic representation is to be mapped to a possible sentence structure, the syntactic representation is created by assembling the pieces from the SI according to the syntactic template selection principles in the linking algorithm (see Section 1.6.1). The construction of the syntactic structure for the sentence in Figure 1.6 is presented in Figure 1.21. The arrows indicate how the templates fit together; they do not constitute a ‘derivation’ of the sentence. There is a PrDP, and it contains a PP with a proper noun. The CLAUSE node in the PrDP template joins with the CLAUSE node in the PrCS template, and the PrCS hosts a wh-word, what, which is a type of pronoun. The core node in the PrCS template merges with the core node of one of the core templates. The semantic representation of the predicator is the basis for selecting the appropriate core template, as will be seen clearly when the linking algorithm is introduced later in the chapter, and so one would expect a core with two core argument slots, since eat is a transitive verb. However, one of the arguments of eat is a wh-expression, and the default situation is for it to appear in the PrCS; consequently a one-place core is selected. The appropriate operator

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

61

62

ROBERT D. VAN VALIN, JR.

Figure 1.21 Combining syntactic templates

projection is selected, and the adjunct temporal PP is assembled and added to the structure. After insertion of the lexical items, the result is the tree structure in Figure 1.6. More will be said about template selection in the description of the semantics-to-syntax linking in simple and complex sentences in later sections of this chapter, as well as in Chapter 19.

1.2.7 The Layered Structure of the Word It has been shown that clauses and phrases have layered structures, and the same is true for the internal structure of words. Cortés-Rodríguez (Chapter 8) presents a detailed discussion of RRG approaches to morphology, and consequently only two issues will be addressed here, namely, the lexical integrity hypothesis and the phenomenon of head-marking in the morphosyntax. One of the first proposals for an RRG approach to morphology was made in Everett (2002), and an example of the layered structure of the word (LSW) following his proposal is given in Figure 1.22. Like phrases (RP, PP, MP) and unlike clauses, the LSW spine has only three layers: nucleusw , corew, and word. The meaning-bearing root or stem is in the nucleusw (there may be multiple roots) along with any derivational morphemes. Inflectional morphology is at the corew-level, and clitics are attached at the word-level analogous to the Pr/PoDP. In refusals in Figure 1.22, the verb root refuse and the derivational suffix -al constitute the nucleusw,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.22 Layered structure of the word following Everett (2002) and English refusals

and the plural marker is an element in the corew. The difference between the plural -s in English cats and the possessive clitic -’s in cat’s is reflected in the LSW: cats would be [WORD [COREw [NUCw /kæt/] {PL}]], while cat’s would be [WORD [COREw [NUCw /kæt/]][{POSS}]]. As with the LSPP and LSMP, there are no operators modifying the layers of the word, hence there is no operator projection in Everett’s model of the LSW. (See Cortés-Rodríguez (Chapter 8) for a different view.)

1.2.7.1 The Lexical Integrity Hypothesis The lexical integrity hypothesis (LIH) concerns the ability of the syntax to have access to the internal structure of words. It is often phrased in terms of ‘can the syntax see into the internal structure of words?’ There is a range of views on this issue, from the morphology is completely opaque to the syntax to various degrees of transparency. The completely opaque view seems to be untenable, because, for example, verb/auxiliary agreement processes require that the syntax be able to identify the person and number of the subject, which are coded morphologically. The LSW provides a simple answer to this question: syntax can see into the corew of the word, but not into the nucleusw of the word. So from the point of view of the syntax, there is no difference between refusals and cats, since both are plural count common nouns. The information that refusals has an internally complex nucleusw and cats does not is unavailable and plays no role in any syntactic processes. But the corew-level inflections are accessible. Clitics should also be accessible, since they are outside of the nucleusw. 1.2.7.2 Head-Marking The visibility of inflectional morphology to the syntax is perhaps most clearly observable in head-marking languages. RRG has been concerned with head-marking grammar since its inception, since Lakhota, one of the ‘founding languages’ of the theory (Section 1.1), is consistently headmarking. Van Valin (1977) presented an initial attempt at an analysis of head-marked clause structure, followed by a more detailed account in Van Valin (1985).27 The basic phenomena are illustrated in (55).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

63

64

ROBERT D. VAN VALIN, JR.

(55)

a. Zuzéˇ ca kiŋ wiˇ chá-wa-ka-t’iŋ-kte. snake the 3pl.anim.u -1.a-by.striking-die-pot ‘I will beat the snakes to death.’ a′. Wiˇ chá-wa-ka-t’iŋ-kte. ‘I will beat them to death.’ ˇ b. Zuzéˇ ca kiŋ Ø-Ø-yahtáka-pi. snake the 3sg.u-3a-bite-pl ‘The snakes bit him/her.’ ˇ c. Wiˇ chá-ka-t’a iúthiŋ-kte. 3pl.anim.u-by.striking-die try-pot ‘I will try to beat them to death.’ ˇ c′. *Wiˇ chá-wa-ka-t’a iúthiŋ-kte. 3pl.anim.u-1SG .A -by.striking-die try-pot d. (Zuzéˇ ca kiŋ) wiˇ chá-wa-kte-kte. Snake the 3pl.anim.u-1sg.a-kill-pot ‘I will kill the snakes/them.’ ˇ d′. Wiˇ chá-kte iúthiŋ-kte. 3pl.anim.u-kill try-pot ‘I will try to kill them.’

The verb kat’Á ‘beat to death’ (typically with an axe, club, baseball bat, etc.) is made up of the stative verb t’Á ‘die, be dead’ and the instrumental prefix ka- ‘by striking’; a more literal translation would be ‘cause to die by striking’.28 In (55a) there is an object (undergoer) RP zuzéˇca kiŋ ‘the snake (s)’, which is cross-referenced on the verb by the third-person plural animate undergoer prefix wiˇchá-. The subject (actor) is signalled by the first-person singular actor prefix wa-; no independent pronoun is required. There is no case marking on the RP; it has the same form in (55a) as undergoer as it has in (55b) as actor. The difference in interpretation derives from the different ways it is cross-referenced on the predicator. In (55a′) the verb constitutes the entire sentence, and it is with respect to the analysis of examples like this that theories differ in important ways. Many posit null pronouns occupying the same positions in the syntax as full RPs which trigger the agreement/cross-reference on the verb. They are in effect treating these languages as being dependent-marking with a lot of agreement morphology (see Van Valin (2013) for detailed discussion). Such an analysis is impossible in RRG because of the concreteness condition on syntactic representations introduced in Section 1.1. Rather, as it has since Van Valin (1977), RRG recognizes the bound affixes on the predicator as being the core arguments that the syntax is sensitive to. An example of the syntax ‘seeing into the inflectional morphology of a word’ is the obligatory control construction in (55c). The first-person singular actor prefix on the linked verb, in this case wa-, must be omitted, and the actor of iyútˇ hA ‘try’ must be interpreted as the actor of the linked verb as well. If both verbs have actor marking, the result, as in (55c′), is ungrammatical. No RPs or independent pronouns are required. Only the bound argument markers on the predicator matter for this construction. Moreover, the construction is oblivious to whether the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.23 The structure of wiˇcháwakat’iŋkte ‘I will beat them to death’ in (55a, a′)

predicator is internally complex or not; the mono-morphemic verb kté ‘kill’ is treated exactly the same morphosyntactically as the bi-morphemic derived verb kat’Á ‘beat to death’, as illustrated in (55d, d′). The structure of the verb in (55a, a′) is given in Figure 1.23. If it is the bound argument markers on the predicator which count as the core arguments for the syntax, a question immediately arises as to the status of the RPs like zuzéˇca kiŋ ‘the snake(s)’ in (55). If the valence of the predicator is saturated by the bound affixes, then there are no valence ‘slots’ available in the core for full NPs or independent pronouns. If the RP were to be analysed as being in the core of the clause, then the undergoer argument in (55a) would be doubly instantiated (RP zuzéˇca kiŋ ‘the snake(s)’ and wiˇchá‘3planimU’), something that is normally not allowed. A number of solutions have been offered (see Van Valin (2013) for a critical overview), and RRG has consistently maintained that the RPs are within the clause but outside of the core; they are not in clause-external detached positions. Straightforward evidence for this can be found in wh-questions like (56). (56)

ˇ a. Zuzéˇ ca kiŋ tuwá Ø-Ø-yahtáka-pi he? snake the who 3sg.u-3a-bite-pl q ‘Who did the snakes bite?’ b. Tuwá zuzéˇ ca kiŋ wiˇ chá-Ø-ka-t’a he? who snake the 3planimu-3sgA-by.striking-die q ‘Who beat the snakes to death?’ c. Hokšíla kiŋ zuzéˇ ca kiŋ wiˇ chá-Ø-ka-t’a he? boy the snake the 3pl.anim.u-3sg.a-by.striking-die q ‘Did the boy beat the snakes to death?’

Lakhota has wh-in situ; there is no displacement of wh-expressions to the PrCS. This is significant, for two reasons. First, tuwá ‘who’ must be outside of the core but inside the clause, because it is in the scope of the IF operator he in either initial or pre-core position. In the polar question in (56c) the focus of the question can be on the RP hokšíla kiŋ ‘the boy’, which means that it is clause-internal as well; see the discussion in Section 1.2.3.2 regarding location in a sentence and IF scope. Second, this shows that Lakhota lacks a PrCS. The positions outside of the core but inside the clause are termed extra core slots (ECSs) in Van Valin (2013), where they are

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

65

66

ROBERT D. VAN VALIN, JR.

Figure 1.24 The structure of (55a)

differentiated from PrCSs in terms of a number of criterial features. The structure of (55a) is given in Figure 1.24. Having ECSs does not preclude the possibility of having a PrCS or PoCS as well; for example, Kihara (2017) shows that Kikuyu, a thoroughly head-marking Bantu language, has both ECSs and a PrCS. This analysis raises an interesting question for binding theory, namely, to what category do the bound argument markers belong? It is assumed in some generative approaches (e.g. Jelinek 1984; Pensalfini 2004) that they are pronouns, but this is untenable, because pronouns in an argument position cannot be bound clause-internally, meaning that RPs and independent pronouns would have to be in detached positions, which is incompatible with the facts of polar and wh-questions. It is not necessarily the case that the bound argument markers are all of the same type. Bresnan and Mchombo (1987) propose that in the Bantu language Chichewa the object markers are pronouns, because they cannot be bound clause-internally, but the subject markers are ambiguous between being agreement when there is a subject RP present and being a pronoun when there is no subject RP. The same pattern holds in Kikuyu (Kihara 2017). A similar situation obtains in socalled ‘pro-drop’ languages like Spanish (Belloro 2007; Kailuweit 2008): the subject marker counts as agreement when there is a subject RP present, otherwise as a pronoun. The accusative clitic, however, is pronominal in standard varieties of Spanish.29 ˇ

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (57)

a. (Juan) compr-ó un regalo para María. buy-3sg.pst a present for ‘Juan/He bought a present for María.’ b. (Juan) lo compr-ó para María, el regalo. 3sg.acc for the present ‘Juan/He bought it for María, the present.’ c. *(Juan) lo compr-ó el regalo para María. *‘Juan/He bought it the present for María.’

The accusative clitic lo (singular) ~ los (plural) cannot co-occur in the same clause with an overt object RP; they are in complementary distribution, as in Chichewa and Kikuyu. If both are to appear in a single utterance, the RP must be in a detached position, either before or after the clause. The subject marker on the finite verb or auxiliary, on the other hand, can co-occur with an RP in the core or refer like a pronoun if there is no subject RP. Rather than analysing these morphemes as sometime being agreement and sometime being pronominal, Van Valin (2013) argued that they should be considered to be pronominal anaphors: they can be bound locally by an argument, like an anaphor, or they can refer independently, like a pronoun, when not bound locally by an argument. In Chichewa, Kikuyu and Spanish, the subject affixes are pronominal anaphors, while the object markers are pronouns. An interesting twist is provided by the dative clitics in Spanish: they are pronominal anaphors, like the subject markers, and unlike the accusative clitics, as (58) shows. ˇ

ˇ

(58)

a. La música les gusta a los chicos. the music 3pl.dat like.3sg.pres dat 3pl.m boys ‘The boys like the music.’ b. La música les gusta. the music 3pl.dat like.3sg.pres ‘They like the music.’ c. *La música gusta a los chicos. the music like.3sg.pres dat 3pl.m boys ‘The boys like the music.’ gusta la música? ]] d. [CL [PrCS A quién][CORE le dat who 3sg.dat like the music ‘Who likes the music?’

In (58a) the dative clitic co-occurs in the core with a dative RP, indicating that it can be bound locally. It can also refer independently, as in (58b), and (58c) shows that the clitic is obligatory with the verb gustar ‘like’. The whquestion in (58d) shows that it can also be bound within a simple clause, which is predicted by the pronominal anaphor analysis. Lakhota is different from Chichewa, Kikuyu and Spanish in that all of the argument markers, not just the subject markers, are pronominal anaphors. They are bound within the clause when there are RPs, and when there are none, they refer like pronouns. The contrast between Chichewa ˇ

ˇ

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

67

68

ROBERT D. VAN VALIN, JR.

and Kikuyu, on the one hand, and Lakhota, on the other, is significant, because all three are consistently and thoroughly head-marking, yet they differ in terms of the binding properties of their argument markers. This is clearly a parameter of cross-linguistic variation in this type of language. Another parameter of variation has been proposed in Bohnemeyer et al. (2016) based on data from Yucatec Maya, another thoroughly head-marking language. They argue that in Yucatec, RPs are in the core, not in ECSs as in Lakhota. While they present their conclusions as tentative rather than definitive, they are nevertheless very intriguing. There are at least two ways to think about this. On the one hand, it could be taken as evidence that the bound argument markers do not necessarily saturate the valence slots in the core completely, as in Figure 1.24, but rather the RP and the affix together satisfy them. Alternatively, this could be viewed as being analogous to the variation in the distance between the antecedent (binder) of a reflexive anaphor and the anaphor itself. English and German permit only a short distance (at most a core) between the two, for example Nancy bought a new dress for herself but not *Nancy told Bill to buy a new dress for herself, whereas this sentence would be perfectly grammatical in Icelandic. With regard to the status of RPs in the clause, the pronominal anaphors in Spanish, Chichewa, Kikuyu and Yucatec permit the antecedent to be within the core, whereas Lakhota does not. This constraint on the binding of pronominal anaphors in Lakhota does not affect reflexive constructions, because they do not involve any binding; they are purely lexical (Van Valin and LaPolla 1997, §7.5.1). Thus, two parameters of typological variation among consistently headmarking languages have been identified, and this highlights the need for serious investigation of the typology of head-marking languages, as well as double-marking languages, which have not been discussed here. (See Van Valin (2005: 18–19) for some preliminary remarks.) As Nichols commented in her seminal 1986 paper (116): ˇ

I have argued that the theoretical apparatus of classical, traditional, structural and formal grammar is heavily based on dependent-marked syntax. If the hypothesis of the universally preferred nature of headmarked patterns holds true, then we will have to recognize that describing the world’s languages in standard theoretical terms is not merely Eurocentric distortion, but in fact forces the unmarked grammatical structure into a framework devised for the marked type.

1.3

The Structure of Complex Sentences

The term ‘complex sentences’ covers two distinct albeit related phenomena, namely, clause linkage, which involves the combining of predicator-based units, on the one hand, and complex RPs, which

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

prototypically includes modification by a relative clause of either the head of the RP (restrictive) or the RP as a whole (non-restrictive) or complementation, as in the rumour that Mary stole Bill’s wallet, on the other. In the analysis of clause linkage, there are three main issues which a theory must address. First, what are the units that are linked? Second, what are the possible structural relations between the units in the linkage? And third, what are the semantic relations that obtain between the units in the linkage? RRG gives unique answers to all three questions, and they are presented in Ohori (Chapter 13) and Guerrero (Chapter 14). In Sections 1.3.1, 13.2 and 1.3.3 just the basics will be presented and several outstanding issues in the theory of clause linkage will be discussed. The basic analysis of complex RPs will be the focus of Section 1.3.4, with the detailed presentation in París (Chapter 15).

1.3.1 Juncture The answer to the question ‘what are the units of clause linkage’ is simple and straightforward in RRG: the units of the LSC. This is summarized in (59). (59)

a. b. c. d.

[CORE . . .[NUC . . . ]. . . þ . . .[NUC . . . ] . . .] [CLAUSE . . .[CORE . . .]. . . þ . . .[CORE . . .] . . .] [SENTENCE . . .[CLAUSE . . .]. . . þ . . .[CLAUSE . . .] . . .] [TEXT. . .[ SENTENCE. . .]. . . þ . . .[ SENTENCE. . .] . . .]

Nuclear juncture Core juncture Clausal juncture Sentential juncture

Nuclear junctures are made up of multiple nuclei within a single core; they are complex predicators created syntactically instead of through derivational morphology. This opposition can be seen clearly in the contrast between the Lakhota kat’Á ‘beat to death’(i.e. ‘cause to die by striking’) in (55), repeated below, and the Mandarin nuclear juncture dˇ a sˇı ‘beat to death’. (60)

a. Zuzéˇ ca kiŋ wiˇ chá-wa-ka-t’iŋ-kte. snake the 3pl.anim.u-1sg.a-by.striking-die-pot ‘I will beat the snakes to death.’ b. Wˇ o dˇ a sˇı le hˇendu¯ o de shé. 1sg beat die pfv many prt snake ‘I beat many snakes to death.’

(¼(55a))

Mandarin

Lakhota kat’Á ‘beat to death’ is the result of a productive derivational pattern in the morphology of the language and is not a nuclear juncture, whereas the Mandarin complex predicate is not formed through derivational morphology but rather consists of two verbs, dˇ a ‘beat’ and sˇı ‘die’ which combine to create ‘cause to die by beating’, the same meaning as Lakhota kat’Á. Core junctures have multiple cores in a single clause. (61)

a. Hans hat versucht, die Tür aufzumachen. has tried the door to.open ‘Hans tried to open the door’.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

German

69

70

ROBERT D. VAN VALIN, JR.

b. Maria hat Hans überredet, das Auto zu waschen. has convinced the car to wash ‘Maria convinced Hans to wash the car.’

In (61a) the clause contains two cores, Hans hat versucht and die Tür aufzumachen, while in (b) the two cores are Maria hat Hans überredet and das Auto zu waschen. Clausal junctures have multiple clauses in a single sentence, as in (62a), and sentential junctures have multiple sentences in a single discourse unit, which is labelled here as ‘Text’, as in (62b). (62)

a. I’m aware Sam needs money to buy his wife a new car, but why did he blow all that money betting on horses? b. As for Germany, the response to the Covid-19 pandemic has been surprisingly problematic, and as for the EU as a whole, the situation is grim.

In (62a) there are two clauses in the sentence, each with its own IF operator; the first clause is an assertion and the second a question. In (b), each clause has its own frame-setting topic in the PrDP, and that means each of the conjuncts is a sentence, not just a clause. It is possible to have all four juncture types in a single utterance, as in (63). (63)

Poor Sally, she tried to fix the broken picture frame and have it done when Fred got home, but the glue she tried, it didn’t hold the pieces together properly.

The first sentence in the sentential juncture contains a clausal juncture, and the first clause in it contains a core juncture (tried to fix), while the second clause houses a nuclear juncture (have it done), and a temporal adverbial clause (when Fred got home) modifying the core containing the nuclear juncture. The second sentence contains a nuclear juncture (hold the pieces together), which, like the one in the first sentence, is an example of an English resultative construction, like wipe the table clean, paint the fence white. A rough approximation of the constituent projection of (63) is given in Figure 1.25. Details of phrases are not spelled out, and the conjunctions are not represented.

Figure 1.25 The constituent projection for the multi-juncture utterance in (63)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

1.3.2 Nexus The structural relations between units in clause linkage are termed nexus relations in RRG. Traditionally, two nexus relations are recognized, namely, coordination and subordination. With respect to clauses, coordination means that the two clauses are independent of each other and each can stand on its own as an independent utterance. It is a relation between two wholes. Subordination, on the other hand, is a part–whole relationship, in which the subordinate clause functions either as an argument of the matrix clause or a modifier of it in some way; subordinate clauses cannot stand on their own as independent utterances. RRG has introduced two innovations with respect to nexus relations. First, it applies them to the four layers of the clause: nucleus, core, clause and sentence. Hence there is a contrast between the two traditional nexus types at each of the four levels. Second, it posits a third nexus type, cosubordination, which is a kind of dependent coordination.30 The dependence is operator sharing: the linked unit must share an operator at the level of juncture with the licensing unit. For English clauses, the relevant operators are tense and IF. It is crucial, first and foremost, to distinguish operator sharing from operator identity. This is illustrated in (64) and (65). (64)

a. b. c. d. e.

Bill bought a new car yesterday, and he sold his old one, too. Bill bought a new car yesterday, and did he sell his old one, too? Bill bought a new car yesterday, and he will sell his old one tomorrow. Bill bought a new car yesterday, and will he sell his old one tomorrow? Did Bill buy a new car yesterday and will he sell his old one tomorrow?

(65)

a. b. b′. c. d.

Bill bought a new car yesterday and sold his old one, too. Did Bill buy a new car yesterday and sell his old one, too? *Did Bill buy a new car yesterday and sold his old one, too? Bill bought a new car yesterday and will sell his old one tomorrow./*? *Did Bill buy a new car yesterday and will sell his old one tomorrow?

In (64a) both clauses are past tense, and both have declarative IF; this is a case of operator identity, not operator sharing. The two clausal operators happen to be the same, but they are not obligatorily the same. This is shown in (64b), where the tense is the same in both but IF differs, in (64c) where tense differs but IF is the same, and in (64d) where both are different in each clause. Both clauses can be questions, as in (64e), but the interrogative IF must be indicated separately in each clause. There are no dependencies, operator or otherwise, between the two clauses. Coreference between Bill and he is not obligatory, despite being strongly preferred; in a context in which Sam is the main discourse topic and the issue of car selling and buying, he could be interpreted as referring to Sam rather than Bill. This contrasts sharply with the examples in (65), which differ from those in (64) in that the ‘subject’ of the second clause has been omitted under identity with the ‘subject’ of the first clause.31 The sentences in (64a) and (65a) look superficially similar, since they both have the same tense and IF in

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

71

72

ROBERT D. VAN VALIN, JR.

both clauses, but they are not the same. In (65) the IF must be the same in both clauses: in (65b) the tense operator is at the beginning of the clause signalling interrogative IF, and the verbs in both clauses are bare infinitives, which means they obligatorily share tense and IF in this interrogative utterance. Sharing tense is not obligatory with declarative IF, as (65c) shows, but the second clause cannot be interpreted as a question (hence ‘*?’), unlike (64d). The examples in (65b′), (c) and (d) are all ungrammatical, because the two clauses differ in IF, and this contrasts with (64b) and (d). Thus, the sentences in (64a) and (65a) are both clausal junctures, but they differ in nexus type: (64a) is coordination, while (65a) is cosubordination, because the linked clause must have the same IF as the licensing clause and is therefore dependent on it. There is another important contrast illustrated here. Both (64) and (65) are instances of conjunction, which is a kind of grammatical construction. Coordination and cosubordination, as well as subordination, are abstract linkage relations which are instantiated by a range of formal construction types. The Mandarin construction in (60b) is an example of cosubordination at the nuclear level; the two verbs dˇ a ‘beat’ and sˇı ‘die’ are followed by the perfective aspect marker le, which obligatorily has scope over both of them. This is an instance of an obligatorily shared operator at the level of juncture (perfective aspect is a nuclear operator), and consequently the nexus type is cosubordination. The formal construction type is serial verbs, not conjunction. There is no one-to-one correspondence between juncture and nexus combinations and formal construction types. A useful analogy is to traditional grammatical relations: they are abstract relations which are instantiated formally in a variety of ways, namely, word order, case marking, cross-reference (as in head-marking languages), agreement (as in ‘pro-drop’ languages) and even tone (e.g. Maasai), but ‘subject’ is not necessarily identified or correlated with a particular case or agreement or a position in the clause. The same is true of clause-linkage relations. Coordination and cosubordination are not always realized via conjunction, just like subordination is not always associated with embedding, as will be shown later in this discussion. This brings up a fundamental difference between the RRG theory of complex sentences and other approaches. The following equivalence is widely assumed: dependent ¼ embedded ¼ subordinate. Terms like ‘dependent clause’, ‘embedded clause’ and ‘subordinate clause’ are used as synonyms. In the RRG system, on the other hand, ‘dependent’, ‘embedded’ and ‘subordinate’ are not synonyms. Neither are ‘conjunction’ and ‘coordination’, despite common usage of them as synonyms. Coordination may be realized in constructions which do not involve conjunction, as will be seen in the discussion of core-level negation and at the end of this section. Crucially, a dependent unit is not necessarily embedded or subordinate. This can be seen clearly in (65); the second clause is dependent upon the first clause for its IF value; if the first clause is an assertion, as in (65a, c), then the second clause must be interpreted as an assertion as well, and if the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

first clause is a question, as in (65b), then the second must be a question, too. This is a particularly telling example, because the verb in the second clause is a bare infinitive and accordingly the second clause is dependent on the first clause for its IF value but also for its tense. If the tense in the first clause changes, as in Will Bill buy a new car tomorrow and sell his old one, too?, the tense interpretation of the second clause necessarily changes, too. Thus, the second clause in (65) is dependent on the first clause, but it is neither embedded nor subordinate. The RRG treatment of sub-clausal junctures, specifically core junctures, is also very different from that of other approaches. These include what are known as ‘obligatory control’ and ‘raising’ or ‘exceptional case-marking’ constructions. These core junctures have an important property: they require a shared core argument. This is illustrated in (66). (66)

a. Sam tried to fix the car.

Shared argument: Sam ¼ actor of try, fix

a′. *Sam tried (for) Bill to fix the car. b. Sam persuaded Bill to fix the car.

Shared argument: Bill ¼ undergoer of persuade, actor of fix b′. *Sam persuaded Bill (for) Tom to fix the car. (67)

a. Sam seems to have fixed the car.

Shared argument: Sam ¼ core argument of seem, actor of fix

a′. *It seems (for) Sam to have fixed the car. b. Bill believes Tom to have fixed the car.

Shared argument: Tom ¼ core argument of believe, actor of fix b′. * Bill believes Tom (for) Sam to have fixed the car.

There are two types of argument sharing in non-subordinate core junctures: (66) illustrates a controller–pivot relationship, whereby the controller in the licensing core, Sam in (66a) and Bill in (66b), supplies the interpretation for the missing argument in the linked core. The ungrammatical examples lack a shared argument in the linked core. In (67) the nature of the sharing is different, in that a semantic argument of the predicator in the linked core is realized as a syntactic core argument in the licensing core. The ungrammatical examples lack a shared argument. Subordinate core junctures may but need not have a shared argument, whereas peripheral ad-core subordinate cores require one. (68)

a. Susan regretted insulting Donald. b. Susan regretted the teacher’s insulting Donald.

(69)

a. Chris was stopped after running a red light.

Core subordination

Peripheral ad-core subordination a′. *Chris was stopped after his/Mary’s running a red light. b. Sally brushed her teeth before meeting Max for dinner. b′. *Sally brushed her teeth before her/Sam’s meeting Max for dinner.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

73

74

ROBERT D. VAN VALIN, JR.

Argument sharing will be discussed in more detail in section 1.7.2, on linking syntax and semantics in complex sentences, as well as in Guerrero (Chapter 14). There are two more ways in which the RRG analysis of (66) and (67) departs from the mainstream. First, the structures in (68) and (69) are examples of core subordination, and the standard analysis of (66) and (67) is that they involve subordination as well. The subordinate gerund in (68) can be passivized and it-clefted.32 (70)

a. a′. b. b′.

Insulting Donald was regretted by Susan. The teacher’s insulting Donald was regretted by Susan. It was insulting Donald that Susan regretted. It was the teacher’s insulting Donald that Susan regretted.

The infinitive in (66a) cannot. (71)

a. *To fix the car was tried by Sam. b. *It was to fix the car that Sam tried.

The passive test is not applicable to the structures in (66b) and (67), but the itcleft test is applicable; they all fail. (72)

a. *It was to fix the car that Sam persuaded Bill. b. *It is to have fixed the car that Sam seems. c. *It is to have fixed the car that Bill believes Tom.

This appears to be clear evidence that the infinitives in these constructions are not subordinate but rather they are non-subordinate core junctures, contra the mainstream analysis. Interesting evidence in support of this conclusion comes from try, which can license more than one type of linked core. It can also take a gerund as a linked core, and like the gerund with regret, it passes both tests. (73)

a. Sam tried fixing the car. b. Fixing the car was tried by Sam. c. It was fixing the car that Sam tried.

There is thus a clear contrast between try þ infinitive vs. try þ gerund with respect to nexus. This illustrates an important point: a given predicator can license more than one juncture–nexus combination (see Ohori, this volume, Chapter 13, Van Valin 2005: 210). The second difference also concerns the nexus type of these constructions. The difference between the two non-subordinate nexus types is: if there is a core operator (e.g. deontic modality) in the clause, it must have scope over all of the cores in cosubordination but not in coordination. This only applies when the primary participant (‘subject’) is the shared argument between the cores, as in (74a–c); it does not apply to constructions where the primary argument is not the shared argument, as in (74d–e).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (74)

a. Sam must [is obliged to] try to fix the car. b. Sam can [is able to] try to fix the car. c. Sam must [is obliged to] want to be able to walk again, before he will be able to do so. d. Sam must [is obliged to] persuade Bill to fix the car. e. Sam can [is able to] persuade Bill to fix the car.

In (74a, b) Sam is the shared argument, and Sam’s obligation or ability is to try to fix the car, not just try something unspecified, and accordingly the deontic modal has scope over both cores. In (74c) Sam is again the shared argument, but in this sentence the scope of the deontic modal operator in the first core does not extend to the second core, which is shown by the fact that it has a different deontic operator. In (74d, e) his obligation or ability is to persuade Bill to fix the car, but the modals do not affect the relationship between Bill and fixing the car; in other words, (74d) doesn’t mean ‘Sam is obliged to persuade Bill to be obliged to fix the car’, nor does (74e) mean ‘Sam is able to persuade Bill to be able to fix the car’. The modals in the first core do not have scope over the second core, and in fact it is possible to have distinct deontic operators in each core (i.e. Sam can [is able to] persuade Bill to be obliged to fix the car). Thus, there is obligatory operator dependence at the level of juncture in (74a, b) but not in (74c–e), and therefore the former are instances of cosubordination and the latter coordination. This conclusion is supported by the scope of core-level negation, the default interpretation of negation. The issue is whether negation in the first core can license a negative polarity item (NPI) in the second core; NPIs include any (Bill didn’t buy any flowers vs. *Bill bought any flowers) and a red cent (Bill didn’t spend a red cent vs. *Bill spent a red cent). (75)

a. b. c. d.

Bill didn’t try to buy any/some flowers. Bill didn’t try to spend a red cent. Bill didn’t persuade Sam to buy any/some flowers. ?Bill didn’t persuade Sam to spend a red cent.

(any > some) (some > any)

(‘x > y’ means ‘x is preferred over y’.) In the cosubordinate examples, (75a, b), negation in the first core can clearly license an NPI in the second. On the other hand, in (75c, d) the licensing of the NPI in the second core is less acceptable than in the first two examples. Core-level negation is more easily shared across the cores in the cosubordinate rather than the coordinate core juncture. This is a case of coordination not being realized by conjunction.

1.3.3 Some Issues Concerning Juncture and Nexus Coordination and subordination are found at all four levels of juncture, as noted at the beginning of the previous section. Cosubordination is not possible at the sentence level, because there are no sentence-level operators, and accordingly obligatory operator sharing is ruled out. Hence there are

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

75

76

ROBERT D. VAN VALIN, JR.

Table 1.3 Nexus types þembedded (in a core, clause or periphery) ‒embedded, þdependent (obligatory operator sharing at the level of juncture) ‒embedded, ‒dependent (no operator sharing at the level of juncture)

Subordination Cosubordination Coordination

Figure 1.26 The structure of (64e): English clausal coordination

only eleven juncture–nexus combinations instead of twelve. These will be discussed more in Section 1.3.4. The three nexus types may be characterized as shown in Table 1.3.

1.3.3.1 Representing Juncture–Nexus Types Since cosubordination is distinguished from coordination by obligatory operator sharing at the level of juncture, the operator projection is very important for the representation of complex sentences. The constituent and operator projections are more or less mirror images of each other. The only difference is that when there are multiple operators at a given level, each of those operators is represented by a distinct node. In the examples in Figures 1.26 and 1.27, there are two clausal operators, tense and IF, and so there are two clause nodes in the operator projections of these sentences. Coordination and cosubordination differ in terms of operator sharing, and this is represented by having a superordinate node at the level of linkage to which the operator applies and thereby has scope over the units. This can be

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.27 The structure of (65b): English clausal cosubordination with two shared operators

seen clearly in Figure 1.27, in which the superordinate clause nodes indicate that tense and IF have scope over both clauses. There is no such superordinate node in Figure 1.26, because each clause has its own tense and IF operators. The structure of (65c) is interesting in that each clause has its own tense but shared IF indicated by the position of the tense operator in the first clause; it is given in Figure 1.28. At sub-clausal levels the same patterns hold. The structure of (60b), a case of nuclear cosubordination in Mandarin, is given in Figure 1.29. In sub-clausal junctures, all higher layer operators are necessarily shared; in a nuclear juncture like this one, all core and clausal operators have scope over the nuclei, and similarly, in a core juncture all clausal operators have scope over the cores.

1.3.3.2 Symmetry in Clause Linkage The picture of clause linkage given in (59) is not the whole story. The combinations of nucleus þ nucleus, core þ core, etc., reflect the flat structures of coordination and cosubordination. These two nexus types are inherently symmetrical: only units of the same type can be linked in coordinate and cosubordinate relations. The same is not true of subordination, however. Complementation is perhaps the prototypical case of subordination,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

77

78

ROBERT D. VAN VALIN, JR.

Figure 1.28 The structure of (65c): English clausal cosubordination with one shared operator

Figure 1.29 The structure of (60b): Mandarin nuclear cosubordination

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

that is, the use of a clause as a core argument, and it is asymmetrical: a larger unit is embedded in a smaller unit. (76)

a. [CORE Larry believes [CLAUSE that Sam doesn’t like ice cream]] b. [CORE [CLAUSE That Sam doesn’t like ice cream] surprised his friends]

In a tree diagram the embedded clause would be a daughter of the core node, and so this type of subordination will be referred to as daughter subordination. These examples also illustrate a point around which there has been some confusion. The issue is, when there is asymmetric linkage, is the linkage type defined in terms of the size of the linked unit or the size of the licensing unit? In (76), the linked unit is a clause, and the licensing unit is a core. The clauses function as core arguments of the predicator in the nucleus, and therefore these are core junctures; in asymmetrical linkages the juncture type is determined by the licensing unit. Otherwise the analysis is very inconsistent. Symmetrical core subordination is illustrated in (77). (77)

a. [CORE Larry regrets [CORE offending Max]] b. [CORE [CORE Offending Max] is the least of Larry’s worries.]

The linked unit in these examples is a gerund, which is a nominalized core, and accordingly since a core is embedded in a core and functions as a core argument, this is clearly a core juncture. If one analyses (76) as a core juncture, then (76) and (77) are parallel constructions: both have a phrasal core argument larger than a simple RP, and despite the difference in the size of the linked unit, they are both core junctures. If, on the other hand, (76) were to be analysed as a clausal juncture, then (76) and (77) would not be parallel constructions, which is unsatisfactory. Further support for the analysis of (76) and (77) as core junctures comes from the variants in (78), which are symmetrical like (77). (78)

a. [CLAUSE [CORE Larry believed] [PERIPHERY yesterday] [CLAUSE that Sam doesn’t like ice cream CORE] CLAUSE] b. [CLAUSE [CORE It surprised his friends] [CLAUSE that Sam doesn’t like ice cream] CLAUSE]

These are extraposition constructions, and one way to look at them is as a way to resolve asymmetrical linkage as symmetrical. In (76a) the complement clause appears to be within the core, but the structure is actually ambiguous, because the embedded clause can either be inside the core, as represented in (76a) or it could be outside of the core and a direct daughter of the higher clause node. There is evidence supporting the position of the embedded clause being outside of the core. It is given in (78a). English normally restricts adjuncts in the core-level periphery to appear after all of the core elements, and accordingly the fact that the temporal adverb yesterday is strongly preferred before the embedded clause (cf. ??Larry believed that Sam doesn’t like ice cream yesterday) indicates that it is outside of the core. Because RRG is a monostratal theory with no derivations or movement rules,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

79

80

ROBERT D. VAN VALIN, JR.

the option of generating the clause core-internally and then moving it to an extraposed position after the adjuncts in the core-level periphery is not available. The appropriate option in RRG is to link the embedded complement clause to the position outside of the core, which then accommodates the placements of the adjuncts. This also results in a symmetrical linkage. There is no ambiguity with respect to the core-internal position of the embedded clause in (76b), which, while fully grammatical, is unlikely to occur in conversation. Rather the preferred form is (78b), in which the embedded clause is located outside of the core, yielding a symmetrical linkage. The other type of subordination is adverbial subordination (see Guerrero, Chapter 14, for detailed discussion). In RRG the general term ‘adverbial’ can be replaced by ad-nuclear, ad-core or ad-clausal subordination, and the cover term for these is peripheral (ad-)subordination, since the clauses occur in the periphery associated with the unit being modified. The question of symmetry of linkage is not as clear-cut with peripheral subordination; the core- and clause-level peripheries can house adverbials as in (78a), PPs as in (34)–(36), and ad-subordinate clauses, as in (63). The adsubordinate clauses are outside of the core or clause, depending on the level they modify, and there do not appear to be any issues in relation to symmetry. As noted in Section 1.2.4.1, there are ordering constraints on them, such that core-level adjuncts occur closer to the core than clause-level adjuncts (see (34)–(36), Figure 1.11), and the same holds true for ad-core and ad-clausal subordinate clauses. (79)

a. Susan was upset with Paul for not washing the dishes after he got home from the gym even though he had promised to do so. b. Susan was upset with Paul for not washing the dishes even though he had promised to do so after he got home from the gym.

There are two ad-subordinate clauses in these sentences: after he got home from the gym, a temporal ad-core subordinate clause, and even though he had promised to do so, a concessive ad-clausal subordinate clause. The two sentences don’t have the same meaning: in (79a) the ad-core clause modifies the main clause, so that the dishwashing was supposed to happen after the gym due to a prior promise, whereas in (79b) the promise was made after the gym and the time of the dishwashing is unspecified. The clause-level concessive adjunct blocks the core-level temporal adjunct from modifying the main clause, and it must be interpreted as modifying the core of the ad-clausal clause, hence the difference in meaning. It appears there are strong restrictions on what can modify the nucleus; no nuclear-level PPs or ad-subordinate clauses modifying the nucleus have thus far been encountered. Adjuncts in the nuclear-level periphery appear to be restricted to adverbials. There are some quite extreme instances of asymmetrical embedding; two examples were given in (12), (12′) and Figure 1.15, in which whole clauses

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

and sentences are used as attributive modifiers in the MP, filling a slot that would normally be filled by an adjectival modifier, possibly with a degree modifier. A radically asymmetrical linkage can be seen in what Everett and Kern (1997) and D. Everett (2008) term ‘intentional state constructions’ in Wari’, the last viable Chapakuran language spoken in the Amazon in Brazil and Bolivia. A simple Wari’ sentence is given in (80a) and an example of an intentional state construction is given in (80b), taken from D. Everett (2008). Basic word order is V-CL OS, and stress is on the last syllable of each word (indicated by italics). (80)

a. Mi′ noni -onj con hwam hwijima′i mon tarama’i. give 3pl.s.rp/p-3pl.o.m prep.3sg.m fish children coll man ‘The men gave the children fish.’ co mao nai -inj Guajaráj] b. [Ma′i that.prox.hearer m/f.rp/p go(sg) 3sg.rp/p-3n Guajará (Brazilian city) ′oro narima′l taramaxiconk. nak -naml 3sg.rp/p-3pl.f coll woman chief ‘“Who went to Guajará?” (said) the chief to the women.’

In (80a) the sentence begins with a verb followed by a clitic cluster crossreferencing the direct core arguments, which is followed by an oblique core argument followed by the two direct core arguments. In the intentional state construction in (80b) the sentence begins with the wh-question ‘who went to Guajará?’ followed by the clitic cluster and the two direct core arguments. What is striking about (80b) is that the main clause has no verb or predicate in it, and the slot for the nucleus is filled by the clause representing the content of the speech. Everett presents a number of arguments in favour of this analysis and shows that alternative analyses all fail to capture the Wari’ facts. A simplified representation of (80b) is given in Figure 1.30. Here a clause is embedded in a nucleus, an extreme violation of the tendency toward symmetrical embedding, but it is allowed by the theory

Figure 1.30 Proposed structure for (80b) in Wari’ (Everett 2008)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

81

82

ROBERT D. VAN VALIN, JR.

as a highly marked construction. The RRG analysis that Everett gives accounts in a principled way for all of the highly unusual properties of this construction.

1.3.4 The Syntax–Semantic Interface in Clause Linkage It was mentioned at the beginning of the previous section that there are eleven juncture–nexus combinations. They may be organized into a hierarchy in terms of the tightness of the relation between the units; this is known as the Interclausal Syntactic Relations Hierarchy, and it will be discussed in Section 1.3.4.1. The semantic relations holding between the units in the linkage can also be organized into a hierarchy, yielding the Interclausal Semantic Relations Hierarchy, which is the topic of Section 1.3.4.2. The two will be combined to create the Interclausal Relations Hierarchy in Section 1.3.4.3. (See also Ohori, Chapter 13.) 1.3.4.1 The Interclausal Syntactic Relations Hierarchy Ranking the juncture–nexus combinations in terms of how tightly bound the units are to each other in the linkage yields the Interclausal Syntactic Relations Hierarchy (ISynRH). One aspect of it is straightforward: in terms of the levels of juncture, nuclear junctures are tighter than core junctures, core junctures are tighter than clausal junctures, and clausal junctures are tighter than sentential junctures. With respect to nexus types, coordination is the ‘loosest’ because there can be no dependencies at the level of juncture between or among the units linked. That leaves the two types of subordination, daughter and peripheral, and cosubordination. The clearest comparison is at the clause level, because clauses have more freedom of placement than cores and nuclei, and since there is no cosubordination at the sentence level, it is irrelevant to this comparison. Accordingly, at the sentence level, coordination is looser than subordination. Of the three nexus types peripheral ad-subordinate clauses have the greatest freedom in terms of where they can occur. It has been suggested that a universal property of these clauses is that they can occur either before or after the licensing (‘main’) clause; this is not the case with daughter subordinate clauses, which in some languages have a fixed position at one edge of the clause or the other, depending on the direction of branching, and in others can appear in extraposed portions, as exemplified in (76) and (78). The core- and clause-level peripheries are less rigidly organized than the core or clause, and their constituents are more loosely related to each other and to the core or clause they modify. Peripheral ad-subordinate clauses need not share any operators with the modified clause; as far as IF is concerned, they are normally presupposed therefore are not the focus of assertions or questions (see Bentley (Chapter 11)). They are optional modifiers rather than semantic arguments of the predicator in the nucleus like

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

daughter subordinate clauses. Hence it is reasonable to conclude that peripheral ad-subordinate clauses are less tightly linked to the other unit in the juncture than daughter subordinate clauses. Peripheral ad-subordinate clauses are clearly more loosely linked than cosubordinate clauses, which have little or no possibility of variable locations in the sentence. They are flat structures which must be dependent on the licensing unit for an operator at the level of juncture. Neither peripheral nor daughter subordinate clauses are required to share any operators with the licensing clause. Thus it appears that the ranking of tightness of the nexus types is coordination (least) > peripheral ad-subordination > daughter subordination > cosubordination (most). This ranking, together with the one for juncture levels, generates an interesting prediction, first made in Foley and Van Valin (1984), namely, there is a default or unmarked nexus type for each level of juncture, such that if a language has only one nexus type at a given level of juncture, then it will be the unmarked one. So if a language has only one type of nexus at the clause or sentence level, it will be coordination. If a language has only one nexus type at the core level, it will be subordination, and the question then arises, which kind? The prototypical subordinate clause is a complement clause which is a semantic argument of the predicator in the nucleus. However, it has been claimed that there are languages that lack this kind of subordination (e.g. Hale ~ but do have ‘adjoined’ struc(1976) for Warlpiri, D. Everett (2005) for Piraha) tures, which would be analogous to peripheral ad-subordination. Finally, at the nuclear level, the unmarked nexus type would be cosubordination, and it seems very often to be the case that a language has only one type of nuclear juncture, which reveals itself to be cosubordination. The ISynRH is given in Figure 1.31. It is important to emphasize that languages vary in terms of which juncture–nexus types they exhibit; most languages do not have all of them, and which ones a given language has is an empirical question. Above nuclear cosubordination in terms of tightness is derivational morphology, which is outside of the domain of complex sentences. Thus, the Lakhota example in (55) and the Mandarin example in (60b), both of which mean ‘beat to death’, contrast structurally in that the Lakhota form kat’Á is a morphological construction whereas Mandarin dˇ a sˇı is a syntactic construction, an instance of nuclear cosubordination (compare Figures 1.24 and 1.29).

1.3.4.2 The Interclausal Semantic Relations Hierarchy In Section 1.3.2 an analogy was made between juncture–nexus types and traditional grammatical relations based on the shared property that they are abstract relations that can be realized via a variety of formal means. Another way they are similar is that they are both syntactic in nature but serve to convey semantic information: grammatical relations code semantic roles like agent and patient, and similarly juncture–nexus types express semantic

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

83

84

ROBERT D. VAN VALIN, JR.

Figure 1.31 Interclausal Syntactic Relations Hierarchy

relations like causality, purposive, jussive, etc. These relations can be thought of as the ‘thematic relations’ of clause linkage. The following is a revision of the interclausal semantic relations in Van Valin (2005, §6.6), and they are organized into clusters based on shared semantic properties. (81)

Interclausal semantic relations a. Single actions 1. Causative [1]: one state of affairs makes another state of affairs happen, and the two are subparts of a single state of affairs, e.g. (60b), Mary wiped the table clean. 2. Modifying sub-actions i. Method: the way an action is carried out, e.g. Mary wiped the table clean, Tom sliced the box open. ii. Manner: the manner in which a motion event is carried out, e.g. Bill entered the house limping. 3. Phase: a separate verb describes a facet of the temporal envelope of a state of affairs, specifically its onset, its termination, or its continuation, e.g. Sam started singing, Nancy kept crying, Oliver finished eating dinner. b. Multiple actions 4. Simultaneous i. Non-motion, e.g. I can eat dinner watching TV, Marie falls asleep listening to music.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

c.

d.

e.

f.

ii. Motion: motion accompanying another action, e.g. Lakhota yuhá ‘have’ þ hí ‘arrive coming’ ¼ ‘come while holding/possessing something’ iii. Position: stance while doing an action, e.g. Sally stood singing, Bill sat reading the newspaper. 5. Sequential, e.g. Close and lock the door, Sally will prepare and cook the fish. 6. Causative [2]: one state of affairs causes another state of affairs to happen, and the causing state of affairs is distinct but unspecified from the caused state of affairs. i. Unmediated: Mary made Sam wash the dishes. ii. Voluntary: Mary had Sam wash the dishes. iii.Facilitating: Mary let Sam wash the dishes. iv. Assisting: Mary helped Sam wash the dishes. Endeavour 7. Attempt, e.g. Fred tried/attempted/endeavoured to make the cake without a mix. 8. Success, e.g. Fred managed/happened to make the cake without a mix, Fred succeeded/lucked out in making the cake without a mix. 9. Failure, e.g. Fred failed to make the cake without a mix. Intentions 10. Refusal, e.g. Fred refused/declined to make the cake without a mix. 11. Psych-action: a mental disposition regarding a possible action on the part of a participant in a state of affairs, e.g. Mary decided to take the job offer, Sally forgot to close the window, Andy wants to be examined by a specialist. 12. Purposive: one action is one with the intention of realizing another state of affairs, e.g. Harry went to the store to buy some beer, Susan brought her own wine to drink. Bringing about 13. Causative [3]: the bringing about of a state of affairs by a distinct state of affairs, e.g. Maria taught Fred to speak Brazilian Portuguese, Marilyn’s arrogant attitude caused Sam to leave the reception early. 14. Jussive: a verbal causative, i.e. the expression of a command, request or demand, e.g. The teacher asked the student to stop talking, The general ordered the troop to attack the enemy. 15. Permissive: making a state of affairs possible by removing a barrier to it or directly licensing it, e.g. The rancher opened to gate to allow the cattle to move to a new feedlot, The city council permitted the demonstrators to gather in the park for their demonstration. 16. Injunctive: making a state of affairs impossible by putting up a barrier to it or directly enjoining it, e.g. The security fence prevented the rioters from reaching the Capitol, The fireman stopped Sam from going back into the burning house. Perception 17. Direct: the unmediated apprehension of some action, event or situation through the senses, e.g. The security guard saw Tom leave work early. 18. Indirect: the deduction of some action, event or situation from evidence of it, e.g. (looking at an empty executive parking place) I see that Tom left work early today.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

85

86

ROBERT D. VAN VALIN, JR.

g. Intentionality 19. Propositional attitude: the expression of a participant’s attitude, judgement or opinion regarding a state of affairs, e.g. Many people believe that politicians are dishonest, Max considers Don to be a loser, Sports fans want very much for their team to win. 20. Cognition: an expression of knowledge or mental activity, e.g. Bill knows that he is in the wrong this time, Sally is thinking about how to fix the problem with the microwave. 21. Emotion: an expression of the content of an emotional state, e.g. Sam is happy/loves it that the pandemic is finally over. Larry is afraid/fears that the corona virus will return. h. Speech 22. Indirect discourse: an expression of reported speech, e.g. Frank said that his friends are corrupt. 23. Direct discourse: The direct quotation of a speech event, e.g. Frank said, ‘My friends are corrupt.’ i. Locational 24. Space: the spatial location of a state of affairs, e.g. Matilda lived her whole life near where she was born. 25. Time: the temporal parameters of a state of affairs, e.g. Sally met Bob for a drink after she left the office. j. Circumstances 26. Reason: the motivation or cause for a state of affairs, e.g. The baby cried, because it was hungry. 27. Conditional: an expression of what consequence would hold, given a particular state of affairs, e.g. If it rains, we’ll have to stay home. 28. Concessive: the state of affairs specified in the main clause holds unexpectedly, given the state of affairs specified in the subordinate clause, e.g. Tim made it to work, even though it was snowing heavily. k. Temporality 29. Simultaneous: one state of affairs is temporally coterminous with another, e.g. Lisa yawned loudly while Dan was talking. 30. Sequential: one state of affairs follows another temporally, with or without any temporal overlap, e.g. Marge finished breakfast, and then the repairman was at the door. 31. Unordered: the temporal relation between states of affairs is unspecified, e.g. Sam watched a football match, and Sally chatted with her mother.

There are a number of shared semantic properties reflected in this list. First, for the most part, groups (a) through (h) are either verb-governed (e.g. (c), (e), (f ), (g) and (h)) or there are restrictions on one or both of the predicators involved (e.g. (a) and (d)). This contrasts sharply with (i) through (k), which do not depend upon the content of the clauses being linked. Second, the notions expressed in (i) and (j) are the same as ones expressed by the core-level and clause-level peripheral adjuncts discussed in Section 1.2.4.1. Third, the relations in (j) and (k) are essentially discourse relations of the kind posited in Rhetorical Structure Theory (Mann and Thompson 1988),

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

and thus the hierarchy constitutes a link between clause-level syntax– semantics and discourse relations (see also Winther-Nielsen 1995, 2021). Fourth, groups (a)–(c) have the Macro-Event Property (MEP) (Bohnemeyer and Van Valin 2017), while (d) through (k) lack it. The order of the relations in (81) is not random but rather reflects how close or loose the semantic relation between the unit is. In group (a) the units in the linkage are taken to be subparts of a single action, whereas all other groups involve multiple actions. One piece of evidence in favour of ranking the relations in group (a) at the top of the list is the fact that these relations are lexicalized in many cases, or expressed by derivational morphology, for instance Lakhota instrumental prefixes express a combination of causative [1] and method. These interclausal semantic relations can be represented in a hierarchy with the relations expressing the closest semantic ties at the top and those indicating the loosest relationship at the bottom. This is a complex hierarchy, and it has been argued that it is motivated by a number of more basic hierarchies, which are discussed by Guerrero in Chapter 14.

1.3.4.3 The Interclausal Relations Hierarchy It has been observed, since at least Silverstein (1976), that there is a fundamentally iconic relation governing the interaction of syntax and semantics in clause linkage. Specifically, the closer the semantic relationship is between the units in the linkage, the tighter the morphosyntactic integration of the units will be. Thus, closeness of meaning correlates with the extent of the morphosyntactic fusion of the units. This is captured in the Interclausal Relations Hierarchy (IRH) in Figure 1.32, which is composed of the syntactic and semantic hierarchies in Figures 1.31 and (81). One fact about clause linkage that jumps out from them is that there are many more semantic relations (31) than juncture–nexus types (14), and because of this disparity, the semantic side of the IRH in Figure 1.32 contains the cluster labels in (81), rather than the individual interclausal relations; where there are ordering constraints within a cluster, they are specified. Moreover, few if any languages have all of the syntactic relations, but presumably all languages can express the semantic distinctions. From this it follows that there can be no one-to-one matching between the hierarchies; rather, there is a many-to-one relationship between semantics and syntax. This is hardly surprising; the analogy to grammatical relations is relevant here, as there are many more thematic relations than grammatical relations, and so the mapping from thematic to grammatical relations is likewise many-to-one. The complexities of the many-to-one mapping between the syntactic and semantics sides of the IRH are discussed in Ohori (Chapter 13) and Guerrero (Chapter 14), as well as in Van Valin (2005, §6.6).33 The basic iconic principle entails that the tightest syntactic linkage type in a language will instantiate the closest semantic relationship. Accordingly, if

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

87

88

ROBERT D. VAN VALIN, JR.

Figure 1.32 The Interclausal Relations Hierarchy

causative [1] is realized syntactically, and if the language has constructions instantiating nuclear cosubordination, then causative [1] will be expressed via nuclear cosubordination, as in (60b) from Mandarin. However, it may be the case that causative [1] is expressed exclusively by means of derivational morphology. This is not a counterexample to the IRH, because, as mentioned earlier, morphological constructions are even more tightly fused than nuclear junctures, and this is in line with the principle of iconicity that motivates the IRH. Another complexity alluded to earlier is the fact that many verbs can license more than one juncture–nexus type, with each coding a different interclausal semantic relation. Want, for example, can be a psych-action predicate realized through core coordination, as in Bill wants to visit Nancy tomorrow, or a propositional attitude predicate expressed via core subordination, as in Bill wants very much for Sam to visit Nancy tomorrow. (See Chapters 13, 14, Van Valin and Wilkins 1993, Van Valin 2005: 210–211.) The IRH will be revisited in subsequent sections on semantic representation and on linking syntax and semantics in complex sentences.

1.3.4.4 English Causative Verbs The cluster in (81b) contains the English verbs make, have, let and help, which as a group have some intriguing properties and which pose certain challenges for the IRH. These properties are as follows. First, in the active voice the infinitive following them cannot (make, have, let) or may not (help) be

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

marked by to. Second, make and help require to in the passive voice, while have and let do not occur in the passive voice with or without it.34 This is illustrated in (82). (82)

a. a′. b. b′.

Sally made/had/let Tom (*to) clean the kitchen. Sally helped Tom (to) clean the kitchen. *Tom was made/had/let/helped clean the kitchen by Sally. Tom was made/*had/*let/?helped to clean the kitchen by Sally.

It has long been observed that for many English speakers the two versions in (82a′) can have different interpretations (e.g. Quirk et al. 1972; Bolinger 1975; Dixon 1984, 1991): in the to-less version Sally participated in the cleaning with Tom, whereas in the version with to she need not have done any of the cleaning but rather could have just given Tom some cleaning supplies. In other words, with the bare infinitive Sally’s helping and Tom’s cleaning temporally overlap, implying that she did some of the cleaning, while in the to-infinitive version they need not overlap. This illustrates a claim made in Van Valin and LaPolla (1997, §8.4.2) that units expressing temporally overlapping states of affairs are often not joined by any kind of linkage marker. In (82a) the unspecified causing/letting action is simultaneous with the caused/permitted action, that is, they overlap temporally, and this is consistent with the bare infinitive complement predicate. This is the marked case; the default is for infinitives in English to be marked by to. The juncture–nexus type of these examples is not readily apparent. Two factors suggest that this construction might be a nuclear juncture. First, it implicates unmediated causation with make, and in many languages constructions with an explicit causative verb are nuclear junctures, for example French faire ‘make, do’ þ infinitive, Italian fare ‘make, do’ þ infinitive, Jakaltek a’a’ ‘give’ þ infinitive. Second, nuclear junctures almost never have any kind of linkage marker. In English, for example, there is a clear difference between Nancy seems happy and Nancy seems to be happy, such that the former is a nuclear juncture, a complex predicate formed in the syntax, and the latter a core juncture, which requires to (and be) and is not a complex predicate. Evidence against (82a) being a nuclear juncture comes from reflexivization.35 English reflexivization has often been referred to as ‘clause bound’, that is, the controller and the reflexive have to be in the same clause, but that is not correct in terms of the LSC. In a simple sentence with a single core, this is an accurate description. However, if the clause contains two cores, the controller must be a semantic co-argument of the reflexive, as illustrated in (83). (83)

a. b. c. d.

Sami seems to have injured himselfi/him*i/j. Sami managed not to injure himselfi/him*i/j. Sami told Donnaj not to injure herselfj/her*j/k. Sami told Donnaj to help himself*i/himi/k.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Semantic co-arguments Semantic co-arguments Semantic co-arguments Not semantic co-arguments

89

90

ROBERT D. VAN VALIN, JR.

In (83a) Sam and himself are both semantic arguments of the verb injure, and despite the occurrence of Sam in a different core, the reflexive is obligatory. The same is true in (83b): Sam is the shared argument between manage and injure, and therefore Sam is a co-semantic argument of injure with the reflexive anaphor. In (83c) Donna is in the core headed by tell but is the shared argument with injure, and therefore Donna and herself are both semantic arguments of injure. In the final example, however, Sam and himself are not semantic arguments of the same verb and are in different cores, and consequently Sam cannot be interpreted as the binder of himself and the plain pronoun him must be used to refer to Sam. The crucial sentence is given in (84a), with possible structures as both nuclear (84b) and core (84c) junctures.36 (84)

a. Bill made/had/let/helped Mary measure him(self ). Reflexive possible? b. [CLAUSE [CORE Bill [NUC made ] Mary [NUC measure ] him(self ) CORE] CLAUSE] c. [CLAUSE [CORE Bill [NUC made ] Mary CORE] [CORE [NUC measure ] him(self ) CORE] CLAUSE]

If the constructions in (82) are nuclear junctures, then the structure would be as in (84b), and because Bill and himself are within the same core, reflexive binding between Bill and himself should be possible. On the other hand, if they are core junctures, then the structure would be as in (84c), and because Bill and himself are in different cores and are the semantic arguments of different verbs, reflexive binding should not be possible. The sentence in (84) is quite ungrammatical with the reflexive but compatible with a plain pronoun, which shows that it is a core juncture, not a nuclear juncture. The next question concerns the nexus type. Daughter subordination can be ruled out for the reasons given in Section 1.3.2, for example *It was clean the kitchen that Sally made/had/let/helped Tom, and it is clearly not an adjunct in the ad-core periphery. It must, therefore be coordination or cosubordination. For the reasons discussed with reference to (74) in Section 1.3.2, the deontic modal scope test does not apply to this construction, but the scope of negation test in (75) is applicable. It is illustrated in (85). (85)

a. a′. b. b′. c. c′.

The teacher didn’t make the students buy anything/??something. The teacher made the students buy *anything/something. No teacher had/let the students buy anything/??something. The teacher had/let the students buy anything/something. The teacher didn’t help the students (to) buy anything/??something. The teacher helped the students (to) buy *anything/something.

The results support a cosubordinate analysis, because in (85a–c) the corelevel negation in the first core licenses the NPI anything in the second core, which is strongly preferred over the non-NPI something. In the non-negative contexts in the primed examples, the NPI is predicted to be impossible, which is the case in (a′) and (c′) but curiously not in (b′); this does not affect the conclusion that a core-level operator is shared across both cores.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Additional support comes from the MEP and the LSC (Bohnemeyer and Van Valin 2017). Bohnemeyer and Van Valin argue that the RRG notion of the (verbal) core is the natural unit for expressing macro-events. Crucial for this discussion is the fact that macro-events can have at most one temporal positional modifier (TPM) (excluding relative clauses, which are RP-internal and may be a separate macro-event), as illustrated in (86). (86)

a. a′. b. c.

Mary ate pancakes for breakfast [TPM today]. [TPM *Yesterday], Mary ate pancakes for breakfast [TPM today]. [TPM This morning], Mary regretted kissing Bill at the party [TPM last night]. Mary asked her mother [TPM yesterday] to make pancakes for breakfast [TPM today].

The simple sentence in (86a) has the MEP, because it allows only one TPM, not two as in (a′). In the other two examples, a case of core-level daughter subordination in (b) and core coordination in (c), neither construction has the MEP, because each of them has two TMPs. The MEP, they argue, is preserved in complex sentences only in core cosubordination. In (74) and (75) try þ infinitive constructions were shown to be cases of core cosubordination, and in (87), it can be seen that they have the MEP. (87)

*Sam tried [TPM in the morning] to fix the bicycle [TPM in the afternoon].

The following examples show that the causative [2] constructions have the MEP. (88)

a. b. c. d. d′.

*Sam made Bill [TPM in the morning] fix the bicycle [TPM in the afternoon]. *Sam had Bill [TPM in the morning] fix the bicycle [TPM in the afternoon]. *Sam let Bill [TPM in the morning] fix the bicycle [TPM in the afternoon]. *Sam helped Bill [TPM in the morning] fix the bicycle [TPM in the afternoon]. ?Sam helped Bill [TPM in the morning] to fix the bicycle [TPM in the afternoon].

The first four examples do not allow more than one TPM, supporting the conclusion that the nexus type is cosubordination. The last example, while odd, is not as bad as the first four, and this is consonant with the claim that the help þ bare infinite form involves temporally overlapping actions while the help þ to-infinitive does not necessarily involve temporally overlapping actions, which renders the two TPMs less unacceptable. There is a clear contrast in acceptability between the examples in (86b, c) and those in (88), and taken together with the contrast between (75) and (85), they strongly support the analysis of these constructions as being instances of core cosubordination. It should be noted that they are very unusual examples of core cosubordination, for two reasons. First, they are the only cases of core cosubordination in which the ‘subject’ of the licensing verb is not the shared argument with the linked verb, and second, they are the only ones which take bare infinitives rather than to-infinitives in the linked core, which is a feature of a tighter morphosyntactic bond between the cores. This suggests that this is

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

91

92

ROBERT D. VAN VALIN, JR.

the tightest version of core cosubordination, one step up from nuclear junctures, as befits their causative semantics.

1.3.5 Complex RPs It was stated at the beginning of Section 1.3 that the notion of ‘complex sentences’ includes two components, clause linkage and complex RPs, and the analysis of complex RPs borrows substantially from the theory of clause linkage, which has received considerably more attention in RRG than the theory of complex RPs. In particular, some of the juncture and nexus concepts can be applied to RPs. It was shown in Section 1.2.5.3 that RPs have a layered structure with three layers, each with a periphery: the nucleusR, the locus of restrictive modification, the coreR, the domain of quantification, and the RP-level, the locus of non-restrictive modification. The most common complex RPs contain relative clauses, restrictive ones being housed in the nucleusR periphery and non-restrictive ones in the RP-level periphery. The RRG analysis of relative clauses is presented in París (Chapter 15), and the discussion here will focus on the application of juncture and nexus to complex RPs. It should be mentioned that considerably less work has been done on this issue than on clause linkage and relativization. Table 1.4 summarizes juncture and nexus in complex RPs. All three nexus types are found at each level of juncture. Coordination at the RP-level involves linked RPs with no operator dependence, which in this case would be definiteness and deixis, and in this case RP-coordination is realized by a conjunction construction, as are some of the other juncture–nexus combinations. In the big dog and a little cat, each of the linked units is a complete RP with its own RP-level operator. In the cosubordinate linkage, however, there is a single definiteness operator which has scope over both of the linked units. The second unit, little cat,

Table 1.4 Juncture and nexus in complex RPs Level of juncture

Nexus type

Example

RP

Coordination Cosubordination Subordination (peripheral) Coordination Cosubordination

the big dog and a little cat the big dog and little cat the Pope, who is from Argentina, . . . the big dogs and little cat ang mga aso at pusa Tagalog NOM PL dog and cat ‘the dogs and cats’ the claim that the election was fraudulent the many big dogs and little cats the two mugs of beer and glasses of wine the two barrels of beer and ale bottle opener, fire engine, duck hunter the girl who sang, the big dogs and cats

CoreR

NuclearR

Subordination (daughter) Subordination (peripheral) Coordination Cosubordination-1 Cosubordination-2 Subordination (peripheral)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

cannot easily stand on its own, for example ??I brushed little cat, ??Little cat meowed, unless it is interpreted as a name. Subordination at the RP-level involves non-restrictive modification, and non-restrictive relative clauses are housed in the RP-level periphery. At the coreR-level the relevant operators are negation and number. In coordination, each unit in the construction has independent operators at the level of juncture; higher level operators are by definition shared. In the big dogs and little cat, the first unit is plural and the second singular, showing operator independence at the level of juncture. This also works for negation, as in No men and three women survived the crash. This is trickier in cosubordination, because the number morpheme is a bound affix and is required. So in English it is not possible to mark number on one unit and have it apply to another unit, that is, *the dogs and cat-__, where cat is interpreted as plural. This is, however, possible in Tagalog (Kolmer 1998), because the plural marker is not an affix but a free, albeit clitic-like, element, as in the equivalent to the example just given, ang mga aso at pusa, in which it has scope over both of the linked units; the RP cannot be interpreted to mean ‘dogs and cat’, only ‘dogs and cats’. There is operator dependence across the linked units, hence it is cosubordination. With respect to negation, there appear to be forms in which negation is obligatorily shared across the units (e.g. neither dogs nor cats, *either dogs nor cats, ??neither dogs or cats). Both types of subordination are found at this level. Nominals derived primarily from verbs of intentionality and speech in (81) can take propositional complements, e.g. the belief/claim/rumour/assertion/fear that . . ., which would be coreR daughter subordination. Quantificational lexical modifiers, namely, quantifiers and numbers, occur in the coreR-level periphery, as discussed in Section 1.2.5.3. The nuclearR-level operator is nominal aspect, which includes nominal and numerical classifiers and the mass/count distinction. In the table it is represented by English measure words, which are a kind of numerical classifier. NuclearR coordination is straightforward, and nuclearR cosubordination-1 is analogous to it in that it involves sharing nominal aspect (numerical classifiers) across the two nuclei instead of each nucleus having its own classifier. NuclearR cosubordination-2, on the hand, is analogous to the nuclear juncture in (60b) from Mandarin, in which the two verbal nuclei form a complex predicate which takes a single aspect operator. It consists of compounding two nominal nuclei to form compounds, for example wood chopper, air conditioner, fire station, knife sharpener, computer repairman, cat litter, dog poop, which have a single value in terms of the mass/count distinction, which is a nuclearR-level operator, despite not having an overt morphological manifestation. It is, rather, what Whorf called a ‘covert category’, one that is signalled by its morphosyntactic ‘reactance’. In these compounds, the value is determined by the final word in the group, which means there is operator dependence across the two nuclei: wood is a mass noun, chopper is a count noun, and wood chopper is a

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

93

94

ROBERT D. VAN VALIN, JR.

count noun; knife is a count noun, sharpener is a count noun, and knife sharpener is a count noun; cat is a count noun, litter is a mass noun, and cat litter is a mass noun. The other nuclearR-level juncture–nexus type is peripheral ad-subordination, which, as discussed earlier, is the locus of restrictive modification. There is a further class of complex RPs which clearly involve juncture– nexus distinctions but have proven difficult to characterize: they are nominalization of jussive and other verbs which take an infinitive complement, for example the (commander’s) order to the troops to attack the fortress, the promise to the workers from the boss to increase their wages/the boss’s promise to the workers to increase their wages, the request from the teacher to the students to work quietly/the teacher’s request to the students to work quietly. There is a shared argument in these constructions, just as in the corresponding verbal predications discussed in Section 1.3.2, and accordingly these would be non-subordinate core junctures. By virtue of the deontic modal scope test, it was determined that the verbal predictions are core coordination, but there are no comparable tests applicable to these complex RPs to distinguish the nexus types. It seems reasonable, then, as a preliminary analysis, to categorize them as coreR coordination, based on the strong parallels with the corresponding verbal predications. More research is needed on these constructions.

1.4

Semantic Representation

1.4.1 Introduction In Figure 1.1 there are two representations, one syntactic and the other semantic, and the nature of the syntactic representation was explicated in Sections 1.2 and 1.3. The spotlight in this section is the semantic representation, which contains two major parts: the system of lexical representation, on the one hand, and the theory of semantic roles, on the other. The basics of each will be presented, as more extensive discussion can be found in Mairal Usón and Faber (Chapter 3) and Kailuweit (Chapter 4). Then some extensions of these ideas will be presented.

1.4.2 Lexical Representation RRG has long had a very rich lexical representation, based on lexical decomposition, in contrast to other approaches which represent the semantics of verbs by a list of thematic relations. 1.4.2.1 Lexical Decomposition Foley and Van Valin (1984) adopted the system of lexical decomposition put forward in Dowty (1979), using Dowty’s formalization of the Aktionsart categories originally proposed in Vendler (1967). There were four categories,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

namely states, activities, achievements and accomplishments, and they were formalized as in (89), and the representations are known as logical structures (LSs). (89)

a. State: predicate′ (x) or (x, y), e.g. dead′ (rat) ‘The rat is dead’, know′ (Bill, French) ‘Bill knows French’ b. Achievement: BECOME predicate′ (x) or (x, y), e.g. BECOME dead′ (rat) ‘The rat died’, BECOME know′ (Bill, French) ‘Bill learned French’ c. Activity: DO predicate′ (x) or (x, y), e.g. DO dance′ (Sally), ‘Sally danced’, DO eat′ (Bill, pizza) ‘Bill ate pizza’. d. Accomplishment: ϕ CAUSE ψ (where ϕ is normally an activity predicate and ψ an achievement predicate, e.g. [DO predicate′ (x)] CAUSE [BECOME predicate′ (y) or (y, z)], e.g. [DO do′ (cat)] CAUSE [BECOME dead′ (rat)] ‘The cat killed the rat’, [DO do′ (Sally)] CAUSE [BECOME know′ (Bill, French)] ‘Sally taught French to Bill’37

There are several appealing features of this approach. First, the main lexical semantic content of verbs is represented by state and activity predicates only, with change-of-state (achievement) and caused-change-of-state (accomplishment) verbs being derived via the addition of elements like BECOME and CAUSE. Second, the state $ achievement $ accomplishment pattern is directly reflected in the derivational morphology of verbs in many languages. Third, there is a set of syntactic and semantic tests for identifying the class to which a particular use of a verb can be assigned (see Van Valin 2005, §2.1.1). One of the serious complications in the analysis of verb semantics is that a given predicate, for example cool, can be used in more than one way, as in The soup is cool (state), The soup cooled (achievement) and The ice cooled the soup (accomplishment), and consequently what one is identifying in most cases is the interpretation of a predicate in a particular sentence. The question of how to analyse such a predicate is left open: one could claim that there are three entries in the lexicon for the verb form cool (cool1 ¼ state, cool2 ¼ achievement, cool3 ¼ accomplishment), or one could claim that there is only one entry, cool ¼ cool′ (x), with the other two forms derived by means of lexical derivational rules. RRG has always favoured the latter approach (e.g. Van Valin 2012a). The decomposition system was basically the same from 1984 through 1997, when Van Valin and LaPolla introduced a major revision of it. The primary features of the revision involved returning to the original distinctions as proposed in Vendler (1967)38 and factoring out causation as an independent parameter: all of the Aktionsart types have causative and noncausative versions. With respect to state predicates, more distinctions among subtypes were recognized, following Schwartz (1993). Activity predicates were marked with do′, in order to distinguish them from states (e.g. see′ (Mary, child) ‘Mary saw the child’ [state] vs. do′ (Mary, [see′ (Mary, child)]) ‘Mary watched the child’ [activity]). Change-of-state predicates went from one category, achievements, to three: achievements (punctual

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

95

96

ROBERT D. VAN VALIN, JR.

changes of state, marked by INGRessive (e.g. INGR popped′ (balloon) ‘The balloon popped’)), accomplishments39 (non-punctual changes of state with an inherent endpoint, marked by BECOME (e.g. BECOME melted′ (ice) ‘The ice melted’)), and processes (non-punctual changes of state or location without an inherent endpoint, marked by PROC (e.g. PROC melt′ (ice) ‘The ice was melting’)).40 In fact, BECOME can be decomposed into PROC & INGR, ‘&’ meaning ‘and sequential’, so The ice melted would be PROC melt′ (ice) & INGR melted′ (ice), which can be abbreviated as ‘BECOME melted′ (ice)’ (Van Valin 2005: 44). Both activities and processes have temporal duration, and both lack inherent endpoints (they are atelic); they differ in that activities are dynamic and processes are not, for example The house shook violently/intensely during the earthquake vs. *The ice melted violently/ intensely during the storm. Processes with endpoints are (process) accomplishments; what are activities with endpoints? Such predications exist: Bill jogged in the park (no endpoint) vs. Bill jogged to the park (endpoint), Max ate pizza (no endpoint) vs. Max ate (up) the pizza (endpoint). Activities with endpoints are termed active accomplishments in RRG, and unlike in other approaches, they are treated as a distinct category, as argued for in detail in Van Valin (2018a). It has been objected that they do not constitute a distinct category, because there are supposedly no lexical active accomplishments, but this is incorrect: devour is a lexical active accomplishment, as is enter (Bill jogged into the room ¼ Bill entered the room (jogging)). Active accomplishments always involve an incremental theme or incremental path. Incremental themes are associated with consumption and creation verbs, while incremental paths are a feature of motion verbs. The three types of verbs have slightly different LSs (from Van Valin 2018a). (90)

a. Consumption predicates, e.g. devour [do′ (x, [eat′ (x, y)]) ^ PROC consume′ (y)] & INGR consumed′ (y) b. Creation predicates, e.g. write (a poem) [do′ (x, [write′ (x, y)]) ^ PROC create′ (y)] & INGR exist′ (y) c. Motion predicates, e.g. enter [do′ (x, [move′ (x)]) ^ PROC cover.path.distance′ (x, (y))] & INGR be-LOC ′ (z, x)

The LS in (90a) has the interpretation ‘x eats y, which simultaneously undergoes a process of consumption, both of which terminate and lead to the result that y is consumed.’ The one in (90b) should be read as ‘x writes y, which simultaneously undergoes a process of creation, which terminates and leads to the result that y exists’. The final one signifies ‘x moves and simultaneously effects a process of covering distance, which terminates, which leads to the result that x comes to be located at/in/on z’. Osswald (2021) presents a critical review of the decomposition system in RRG, and two of his points are particularly relevant to the LSs proposed for both process and active accomplishments. The first concerns the nature of the predicate in the process LS, PROC predicate′. The decomposition of the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

process accomplishment melt given above is PROC melt′ & INGR melted′, and the question arises, if melted′ is a state, what is melt′? The system has only two types of basic predicates, states and activities, and melt′ fits into neither category. The same issue appears in the active accomplishment LSs in (90a, b) with PROC consume′/create′ (y). Osswald notes that the intended meaning in (90) is being.consumed′ and being.created′. Processes are non-punctual changes of state or location, and therefore the argument-bearing lexical predicate at the heart of the representation must be either one of state or condition, or one of location. Many processes do not have an incremental theme argument which is being consumed or created or an incremental path; rather, they involve the position of the referent on a scale. This is particularly true of what Dowty (1979) termed ‘degree achievements’ such as cool, warm, widen, narrow, grow, melt, freeze, redden (all [M]-intransitive), all of which assume a scale of some kind. For example, there is a temperature scale with ‘hot’ at the top and ‘cold’ at the bottom, as can be seen in the contrast between cool down/*up vs. warm up/*down. Osswald notes that as a process the soup cooled means ‘the soup became cooler’, not ‘the soup became cool’. Hence the incremental theme of these verbs is associated with movement on a scale rather than consumption or creation. This can be represented as PROC becoming.higher/lower.on.[α]scale′ (x). These representations carry over to their [M]-transitive causative counterparts. Nothing has been said about motion active accomplishments, which involve an incremental path rather than an incremental theme. Crucially, an incremental theme undergoes a change of state, but an incremental path does not; it measures out the distance the x-argument covers, which means the proposed predicate, covering.path.distance′, cannot be a predicate of state or condition. It is, rather, a change-of-location predicate, which can co-occur with motion activity predicates, as well as pure process predicates like fall, as in He fell fifty metres into the water. In motion active accomplishments there is an activity simultaneous with a process, the former characterizing the motion itself and the latter expressing the distance along the incremental path, whereas in the example with fall there is an uncontrolled change of location in a specific direction covering a specified distance before reaching a final location. Hence the LS for an active accomplishment like stumble into the room or enter the room stumbling should not be (90c) but rather [do′ (x, [move′ (x)]) ^ PROC covering.path. distance′ (x, (y))] & INGR be-LOC ′ (z, x).41 Consequently, the LS for motion active accomplishments parallels the LSs for creation and consumption predications.42 On the other hand, the LS for the example with fall would be [PROC moving.downward′ (x) ^ PROC covering.path.distance′ (x, (y))] & INGR be-LOC ′ (z, x), where ‘moving′’ means ‘changing location’. Moving′ differs from do′ (x, [move′ (x)]), in that it necessarily does not involve any effort on the part of the referent of its argument, is primarily associated with direction rather than manner and is not subject to the agentivity

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

97

98

ROBERT D. VAN VALIN, JR.

implicature, even with a human referent (see §1.4.3.1), whereas do′ (x, [move ′ (x)]) differs on all three of these characteristics. The second issue raised by Osswald is what he calls the ‘and-then anomaly’. He points out that representations like ‘PROC melt′ (x) & INGR melted′ (x)’ don’t accurately represent the situation. This one means ‘there was a process of melting, and then after the process ended there was a transition to the state of being melted’, but the state of being melted does not come about after the process ended; rather, it is the final stage of the process itself. Similarly, in active accomplishments like Sam devoured the pizza, it is not the case that the activity of eating and the process of being consumed terminated and then the pizza was consumed; the pizza was consumed at the termination of the eating, not after it. A possible solution to this anomaly is to introduce a function FIN ‘final stage of a process or action’ and change the connector from & ‘and then’ to ^ ‘and simultaneously’, which is adapted from Osswald’s proposal. Thus The ice cream melted would be PROC becoming. lower.on.[solid$liquid]scale′ (ice cream) ^ FIN melted′ (ice cream), meaning ‘the ice cream underwent a process of becoming lower on the “solid-toliquid” scale and the final stage is that the ice cream is melted’. For Sam devoured the pizza, the LS would be do′ (Sam, [eat′ (Sam, pizza)]) ^ PROC being. consumed′ (pizza)] ^ FIN consumed′ (pizza), meaning ‘Sam eats pizza and at the same time the pizza undergoes a process of consumption whose final stage is the pizza is consumed’. The introduction of the FIN function offers a way to capture the contrast between quantized and non-quantized changes of state. Following Beavers (2013), Bentley (2019) argues that the previous decomposition system does not express the difference between quantized change-of-state predicates like die, which lexicalize a specific result state, and non-quantized changeof-state predicates, which are of two types: those like melt and fill, which entail only that a specific result state (e.g. being liquid, being full) might but need not necessarily be reached, and those like widen or shorten which have no specific final state. The following represents the result state of the two subclasses of non-quantized change (D. Bentley, personal communication): (i) melt: PROC becoming.lower.on.[solid $ liquid]scale′ (ice cream) ^ FIN melted.by.some.amount′ (ice cream) (the ice cream was not melted to begin with and now is melted by some amount, which may be at the end of the scale, i.e. the ice cream was completely melted) (ii) widen: PROC becoming.higher.on.[wide $ narrow]scale′ (crack) ^ FIN wide.by.larger.amount′ (crack) (the crack was wide by some amount and now it is wide by a larger amount) This notation yields the necessary contrast with verbs of quantized change, which entail the reaching of a specific final state (e.g. FIN dead′ (x)). The final addition to the array of Aktionsart types is semelfactives (Smith 1997), which are punctual events with no change of state or result

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

99

state; they contrast primarily with achievements, which are punctual changes of state with a result state. For example, it is possible to say Sam flashed the light, and then he re-flashed it again but not *Sam popped the firecracker, and then he re-popped it again. Flashing a light does not cause it to undergo a change of state, and consequently it can be flashed again. In contrast, popping a firecracker causes it to undergo a change of state, and accordingly it cannot be popped again. Hence flash in this example is a (causative) semelfactive, while pop is a (causative) achievement. The LS for semelfactives is SEML do′ (x, [pred′ (x, (y))]). Because they are events, they are related to activities rather than states. The lexical representations for the Aktionsart categories are given in Table 1.5.43 Examples of each type are given in (91). (91)

a. STATES The window is shattered. John saw the picture. b. ACTIVITIES The children cried. Carl ate pizza. c. ACHIEVEMENTS The window shattered. The climber reached the summit. d. SEMELFACTIVES Mary coughed. Dana glimpsed the picture. e. ACCOMPLISHMENTS The ship sank. Mary learned French. f. ACTIVE ACCOMPLISHMENTS Carl ate the pizza.

shattered′ (window) see′ (John, picture) do′ (children, [cry′ (children)]) do′ (Carl, [eat′ (Carl, pizza)]) INGR shattered′ (window) INGR be-at′ (summit, climber) SEML do′ (Mary, [cough′ (Mary)]) SEML do′ (Dana, [see′ (Dana, picture)]) BECOME sunken′ (ship) BECOME know′ (Mary, French)

do′ (Carl, [eat′ (Carl, pizza)]) ^ PROC being.consumed′ (pizza) ^ FIN consumed′ (pizza) Chris ran two miles to the park. do′ (Chris, [run′ (Chris)]) ^ PROC covering.path.distance′ (Chris, two miles) ^ FIN be-at′ (park, Chris) g. CAUSATIVES The dog scares the boy. [do′ (dog, Ø)] CAUSE [feel′ (boy, [afraid′])] The submarine sank the ship. [do′ (submarine, Ø)] CAUSE [BECOME sunken′ (ship)] The cat popped the balloon [do′ (cat, Ø)] CAUSE [INGR popped′ (balloon)] The conductor flashed the light. [do′ (conductor, Ø)] CAUSE [SEML do′ (light, ([flash′ (light)])] Felix rolled the ball. [do′ (Felix, Ø)] CAUSE [do′ (ball, [roll′ (ball)])] Mary fed the pizza to the child. [do′ (Mary, Ø)] CAUSE [do′ (child, [eat′ (child, pizza)]) ^ PROC being.consumed′ (pizza) ^ FIN consumed′ (pizza)]

The causative LSs typically have an activity predicate as the first argument of CAUSE,44 and it is quite common cross-linguistically for that activity to be unspecified, as in all of the examples in (91). This unspecified activity has heretofore been represented as ‘[do′ (x, Ø)]’. An alternative way of representing this activity but with more semantic content is the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

100

ROBERT D. VAN VALIN, JR.

Table 1.5 Lexical representations for Aktionsart classes Verb class

Logical structure

STATE ACTIVITY ACHIEVEMENT SEMELFACTIVE PROCESS

predicate′ (x) or (x, y) do′ (x, [predicate′ (x) or (x, y)]) INGR predicate′ (x) or (x, y), or SEML do′ (x, [predicate′ (x) or (x, y)]) PROC being.consumed′/created′ (y) or (x, y) PROC becoming.higher/lower.on.[α]scale′ (x) PROC moving(.direction)′ (x) PROC covering.path.distance′ (x, (y)) BECOME predicate′ (x) or (x, y), where BECOME ¼ PROC ^ FIN

ACCOMPLISHMENT

ACTIVE ACCOMPLISHMENT do′ (x, [pred′ (x, y)]) ^ PROC being.consumed′ (y)] ^ FIN consumed′ (y) do′ (x, [pred′ (x, y)]) ^ PROC being.created′ (y)] ^ FIN exist′ (y) do′ (x, [pred′ (x)]) ^ PROC covering.path.distance′ (x, (y)) ^ FIN be-loc′ (z, x) CAUSATIVE α CAUSE β, where α, β are LSs of any type

following: [do′ (x, [affect′ (x, y))] CAUSE [BECOME/INGR pred′ (y)]], which signifies ‘x does something unspecified which affects y causing y to undergo a change of state’. This has the advantage that it ties the unspecified causing activity to the affected participant in the second LS in the causative LS. For example, the LS for The burglar smashed the window would be ‘[do′ (burglar, [affect′ (burglar, window))] CAUSE [INGR smashed′ (window)]’. Another alternative along the same lines is ‘[do′ (burglar, [act.on′ (burglar, window))] CAUSE [INGR smashed′ (window)]]’. For the examples in (91g) any of the three would be appropriate, but there are some interesting cases where either of the latter two would be better. They involve predicative prepositional phrases. In Section 1.2.5.1 the distinction between predicative and non-predicative PPs was introduced (see also Ibáñez Cerda, Chapter 10). Non-predicative adpositions are oblique core arguments of the predicator in the nucleus of the core, and they are assigned by rule or constructionally and are not directly represented in the LS of the predicator. Predicative adpositions, on the other hand, are predicates and are represented as such in the semantic representation of the sentence. In a sentence like Pam saw Felix in the library after the lecture, the LS would be be-after′ (lecture, [be-in′ (library, [see′ (Pam, Felix)])]), in which after the lecture (be-after′ (x, y)) and in the library (be-in′ (x, y)) are realized as adjunct PPs in the core-level periphery. The first argument is the spatial or temporal location of the event represented by the second argument. The default situation is that non-predicative PPs occur in the core as oblique core arguments and predicative PPs occur in a periphery as an adjunct. There are, however, two situations in which this default distribution of PPs does not hold. The first is the by-PP in a passive construction; it is a nonpredicative PP because it is constructionally motivated and is not represented in the LS, but it occurs in the core-level periphery. It is a case of a

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

non-predicative PP in a periphery rather than the core. The second situation involves verbs like put, which have a requirement for a locative element but do not specify what it should be. In the case of English put, there is a range of possibilities, all contributing to the meaning of the LS: a locative demonstrative (put it here/there), an intransitive preposition (put it down/away), or a full predicative PP (put it behind the chair/on the table/under the bed, etc.). These locative expressions contribute to the meaning of the clause and are represented as predicates in the LS. Thus, they satisfy a valence requirement of a predicate and occur in the core, like an argument, but they are also independent semantic units which are not determined by the predicate and contribute semantically, like an adjunct. Hence they are termed argument-adjuncts. In a sentence like Max put the book in the box, the RP the book is an argument of both put and the PP in the box. The LS for put would be [do′ (x, Ø)] CAUSE [INGR be-LOC ′ (. . ., z)], where ‘be-LOC ′’ can be realized by any of the three possibilities given above, and the usual LS for this sentence would be [do′ (Max, Ø)] CAUSE [INGR be-in′ (box, book)]. This represents the book primarily as an argument of be-in′, the predicative preposition, and the fact that Max does something to the book to cause it to move into the box is not overtly represented. Here is where the alternative formulations of the unspecified causing activity have an advantage: they both explicitly represent the argument sharing between the verb and the preposition, that is, [do′ (Max, [act. on′ (Max, book)])] CAUSE [INGR be-in′ (box, book)]. This argument sharing between the two predicates distinguishes pure adjunct PPs, which take the LS of the core as a whole as an argument, from argument-adjuncts, which take one of the arguments in the LS of the core as an argument. Adjectives and adverbs are also represented as predicates in the semantic representation. Following Schwartz (1993), the LS for attributive predications is ‘be′ (x, [pred′])’ (e.g. be′ (grass, [green′]) for The grass is green). The semantic representation for attributive modification is similar, for example Bill saw the green grass would be see′ (Bill, [be′ (grass, [green′])]); the underlining of grass signals that as head of the argument RP it functions simultaneously as an argument of the embedded attributive predication and as an argument of the main predicate see. This same double function is found in possessive constructions, for example Bill saw Mary’s car would have the LS see′ (Bill, [have′ (Mary, car)]), and Bill saw Mary’s new car would have two embedded propositions in the second argument position, that is, see′ (Bill, [have′ (Mary, [be′ (car, [new′])])]). In this LS car functions as an argument in three predications.45 Three points need to be mentioned. First, the fact that the attributive LS contains ‘be′’ in no way implies that a language necessarily has a copular verb analogous to English be, nor is it a claim that there is some kind of underlying copular verb. It is a part of the attributive LS, nothing more. It is analogous to CAUSE: it represents causation in a causative LS, but it does not imply that every language has a causative verb like French faire or a

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

101

102

ROBERT D. VAN VALIN, JR.

causative morpheme like Lakhota ka- ‘cause by striking’. Second, the predicate in an attributive LS can also occur in another type of LS. The adjective black can occur as an attributive predicate, as in be′ (door, [black′]) The door is black, an attributive modifier, as in like′ (Mary, [be′ (door, [black′])]) Mary liked the black door, and also in a causative change of state LS, as in [do′ (fire, [act. on′ (fire, door)])] CAUSE [BECOME black′ (door)] The fire blackened the door. In this sentence black′ (door) is the result state of a causative accomplishment, not an attribute, and accordingly the representation is different. Third, attributive predicates can have second arguments, for instance Sam is proud of his son, Nancy is angry at her husband, and the former would be represented as be′ (Sam, [proud.of′ ([have.as.kin′ (3sgM, son)])]). Adverbs have a simple LS, as they are one-place predicates, for example Bill carefully closed the door slowly yesterday would be yesterday′ ([do′ (Bill, [careful′ (act.on′ (Bill, door))])] CAUSE [slow′ (BECOME closed′ (door))]). As this example illustrates, adverbs may have the whole LS as their argument or a subpart, depending on the level and scope of modification, as discussed in Section 1.2.4.2. The semantic representation of nominals does not treat them as predicates, and their semantic representation is based on the qualia analysis proposed in Generative Lexicon Theory (Pustejovsky 1995). The qualia themselves are represented in terms of the decomposition system used in RRG, which differs from that used by Pustejovsky. See Van Valin (2005, §2.3) for further discussion. Operators have a place in the semantic representation of both RPs and clauses. The RP operators and the semantic representation for Sally’s two red parrots in Figure 1.18 are in (92). (92)

a. ⟨DEF  ⟨DEIX PROX/DISTAL ⟨NEG  ⟨NUM SG/DL/PL ⟨NASP COUNT/MASS ⟨[N]⟩⟩⟩⟩⟩⟩ b. ⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot, [red′])]) [two′])])]⟩⟩⟩⟩

The operators over the clause are summarized in (93a) and the semantic representation for The dogs may be barking at Sally’s two red parrots is given in (93b). These semantic representations must ultimately be interpreted within a formal semantic framework. (93)

a. ⟨IF DECL ⟨EVID Ø ⟨TNS PAST ⟨STA IRR ⟨NEG  ⟨MOD ABLE ⟨DIR Ø ⟨ASP IMPF ⟨[LS]⟩⟩⟩⟩⟩⟩⟩⟩⟩ ⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[dog (x)]⟩⟩⟩⟩ \ b. ⟨IF DECL ⟨TNS PRES ⟨STA PSBL ⟨NEG – ⟨ASP PROG ⟨[do′ (x, [bark.at′ (x, y)])]⟩⟩⟩⟩⟩⟩ / ⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot (y), [red′])]) [two′])])]⟩⟩⟩⟩

1.4.2.2 Representing the Semantics of Clause Linkage The idea of using the system of lexical decomposition introduced in the previous section to formally characterize the interclausal semantic

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

relations of the IRH in (81) was originally suggested in Ohori (2001), and Van Valin (2005: 207–208) used it to formalize the version of the IRH assumed then. Before the revised system is presented, a few notes of clarification are in order. First, ‘VERB ′ ’ stands for the class of predicates of this type; thus, know′ is the verb know but KNOW ′ symbolizes the class of cognition verbs. Causative [1] has the same LS as the lexical causative verbs in (91g) and need not involve a causative verb like English make, German lassen or French laisser, whereas causative [2] typically does feature an explicit causative verb; causative [3] always involves two verbs, one of which may be explicitly causative, such as English cause, force, or not, for example English teach in teach Sam to fix bicycles. ‘pred′ (x, [LS . . . x . . .])’ means that the participant denoted by x is involved in both the matrix and embedded LSs, as in (66). ‘pred′ (x, [LS . . . y . . .])’ also signals that the participant denoted by y is involved in both the matrix and embedded LSs but in a different way, as in (67). (94)

a. Single actions 1. Causative [1] [subevent1] CAUSE [subevent2] 2. Modifying sub-actions i. Method do′ (x, [pred1′ (x, y)] CAUSE [INGR/BECOME state.pred2′ (y)]) ii. Manner do′ (x, [MOTION ′ (x)] . . . ^ [MANNER . OF . MOTION ′ (x)]) 3. Phase do′ (x, [ONSET ′ (x, [LS . . . x . . .])], do′ (x, [CONTINUE ′ (x, [LS . . . x . . .] )], do′ (x, [TERMINATE ′ (x, [LS . . . x . . .])] b. Multiple actions 4. Simultaneous i. Non-motion do′ (x, [pred1′ (x, y)]) ^ do′ (x, [pred2′ (x, y)]) ii. Motion do′ (x, [MOTION ′ (x)] . . . ^ [pred2′ (x, (y))]) iii. Position do′ (x, [STANCE ′ (x)] ^ [pred2′ (x, (y))])46 5. Sequential do′ (x, [pred1′ (x, y)]) & do′ (x, [pred2′ (x, y)]) 6. Causative [2] i. Unmediated [do′ (x, [MAKE ′ (x, y)])] CAUSE [. . . pred2′ (y, (z))]) ii. Voluntary [do′ (x, [HAVE ′ (x, y)])] CAUSE [do′ (y, [pred2′ (y, (z))])]) iii. Facilitating [do′ (x, [LET ′ (x, y)])] CAUSE [ … pred2′ (y, (z))]) iv. Assisting [do′ (x, [HELP ′ (x, y)])] CAUSE [do′ (y, [pred2′ (y, (z))])]) c. Endeavour 7. Attempt do′ (x, [ATTEMPT ′ (x, [LS . . . x . . .])]) 8. Succeed do′ (x, [SUCCEED ′ (x, [LS . . . x . . .])]) 9. Failure do′ (x, [FAIL ′ (x, [LS . . . x . . .])]) d. Intentions 10. Refusal do′ (x, [REFUSE ′ (x, [LS . . . x . . .])]) MENTAL . DISPOSITION ′ (x, [LS . . . x . . .]) 11. Psych-action 12. Purposive want′ (x, LS2) ^ DO (x, [ [LS1] ◇CAUSE [LS2]]) e. Bringing about 13. Causative [3] [do′ (x, [VERB 1 ′ (x, y)])] CAUSE [. . . pred2′ (y, (z))]) 14. Jussive [do′ (x, [SAY ′ (x, y)])] CAUSE [MENTAL.DISP′ (y, [LS . . . y . . .])] 15. Permissive [do′ (x, [PERMIT ′ (x, y)] CAUSE [ … pred2′ (y, (z))]) 16. Injunctive [do′ (x, [STOP ′ (x, y)])] CAUSE [NOT . . . pred2′ (y, (z))])

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

103

104

ROBERT D. VAN VALIN, JR.

f.

Perception 17. Direct 18. Indirect g. Intentionality 19. Prop. attitude 20. Cognition 21. Emotion h. Speech 22. Indirect discourse 23. Direct discourse i. Locational 24. Space 25. Time j. Circumstances 26. Reason 27. Conditional 28. Concessive k. Temporality 29. Simultaneous 30. Sequential 31. Unordered

PERCEIVE′ (x, [LS . . . y . . .]) PERCEIVE′ (x, [LS]) BELIEVE′ ((x,) [LS]) KNOW′ (x, [LS]) FEEL′ (x, [LS]) do′ (x, [say′ (x, [LS ⟨TNS . . . ⟩])]) do′ (x, [say′ (x, [LS ⟨IF . . . ⟩])]) be-LOC′ ([LS1], [LS2]) be-TEMP′ ([LS1], [LS2]) [LS1] BECAUSE′ [LS2] [LS1]  [LS2] [LS1] IN.SPITE.OF′ [LS2] [LS1] ^ [LS2] [LS1] & [LS2] [LS1] þ [LS2]

These representations make it possible to give semantic representations for complex sentences involving clause linkage.

1.4.2.3 Representing Constructional Meaning The system of lexical decomposition is designed to represent the meaning of lexical items and their combinations in sentences. There are instances in which the meaning is not a function of the lexical items but is constructional in nature. The English resultative construction, for instance Mary painted the door black is an example of this; there is no lexical item coding causation, just Mary painted the door and the secondary predicate black, and the combination leads to the causative interpretation. The same is true in the Mandarin construction in (60b); the lexical verbs are dˇ a ‘beat’ and sˇı ‘die’, and the combination has the meaning ‘beat to death’. This raises the question, how is constructional meaning to be represented? Should it be represented the same as lexical meaning? In (94) causation is signalled the same way using ‘CAUSE’, regardless of whether it is lexical or constructional. Other cases of constructional meaning involve more than adding a causative meaning to two co-occurring predicators; a prime example is possessive predications in languages which lack a verb like have. Examples are given in (95)–(97). (95)

a. U menja est’ knig-a. at 1sg.gen exist book-nom ‘I have a book.’ [Literally: ‘at me exist a book’] b. U menja net knig-i. at 1sg.gen not.exist book-gen ‘I don’t have a book.’ [Literally: ‘at me not.exist a book’]

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Russian

Principles of Role and Reference Grammar

(96)

a. Bayan nhangu wunaa. Guugu Yimidhirr house 3sg.dat exist ‘He/she has a house’ [Literally: ‘his/her house exists’, ‘a house exists to him/her’] b. Mayi gabiir-bi guya. food girl-dat not.exist ‘The girl has no food.’ [Literally: ‘the girl’s food does not exist’, ‘food doesn’t exist to the girl’]

(97)

a. May libro ang babae. Tagalog exist book nom woman ‘The woman has a book.’ [Literally: ‘a book exists the woman’] b. Wala-ng libro ang babae. not.exist-lnk book nom woman ‘The woman doesn’t have a book.’ [Literally: ‘a book doesn’t exist the woman’]

None of these languages has a lexical verb expressing possession akin to English have, French avoir, Lakhota yuhá or Mandarin yˇou.47 In all of them an existential verb is crucial to the expression of possession and a negative existential verb for lack of possession. This is perhaps not surprising, given the close connection between existential expressions and those of having; in French, for example, the verbal part of the existential il y a is a ‘has’, and in Mandarin yˇou (same character) is also used in existential constructions, for example Zhu¯o sh` ang yˇou y¯ı bˇen sh¯ u [table on exist one CL book] ‘There is a book on the table’ vs. Wˇo yˇou s¯ an bˇen sh¯ u [1sg have three CL book] ‘I have three books’.48 In (95) from Russian, the possessor is the genitive object of the locative preposition u ‘at’, and the possessed argument is formally the subject in the nominative case, although the predicate est’ ‘exist’ is impersonal and does not agree with the nominative RP. When the verb is negated, as in net (< ne ‘not’ þ est’ ‘exist’), the single argument appears in the genitive case. In Guugu Yimidhirr, an Australian language spoken in northern Queensland (Haviland 1979), an alienable possessor in an RP is in the dative case, as in bayan nhangu [house 3sgdat] ‘his/her house’, mayi gabiir-bi [food girl-dat] ‘the girl’s food’. In a clause, the possessor can be analysed as RP-internal, yielding the first readings given above, or as a core argument, yielding the second readings. The latter parallels the Russian pattern, namely ‘a book exists at me’ vs. ‘a house exists to him/her’. The Tagalog pattern, while similar, differs in two striking respects: the possessor is neither dative nor oblique but nominative, and the possessed is obligatorily not case-marked at all. The issue concerns the LS for these predications. One possibility is to say that since all of them are used to express the equivalent of constructions with have, they should all get the LS have′ (x, y), just like in languages with a possessive verb like have. While this captures the semantic commonality among the forms, it will engender a lot of arbitrary linkings between syntax and semantics in the languages, since other two-place state predicates follow the standard linking pattern(s) for such predicates in the language.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

105

106

ROBERT D. VAN VALIN, JR.

Because of this problem, it might be suggested that the LSs should reflect the properties of the specific constructions. On this view, the LS for (95a) would be something like be-at′ (1sg, [exist′ (knig-)]), which can be readily accommodated in terms of the regular Russian linking and case assignment rules; (96a) would be exist′ ([have′ (3sg, bayan)]), where the possessor can be linked as an RP-internal possessor or linked via ‘possessor raising’ as an argument in the core of the clause; and (97a) would include exist′ (libro), but it’s not clear how the possessor is related to it, as it seems to be simply juxtaposed to it (i.e. ([exist′ (libro)] þ babae)), which is unsatisfactory. Also unsatisfactory is the semantic interpretation of these LSs, since they should mean the same as their counterparts in the languages mentioned above. In Latrouite and Van Valin (2014) a solution to this problem is proposed: constructional meaning is represented by the same basic decompositional system as lexical meaning, but the notation for each is distinctive. Lexical meaning continues to be represented as before, with elements in the semantic metalanguage in boldfaceþprime (e.g. have′). Constructional meaning, on the other hand, will be represented by small capsþprime (e.g. have′). In building representations of constructional meaning, the components INGR, PROC, BECOME and SEML have the same function as in the representation of lexical meaning, but CAUSE must now be treated differently in the two representational notations, because it can be lexically realized, as in the examples in (91g), still symbolized by CAUSE, or it can be constructionally induced, as in the examples discussed earlier, symbolized by cause in small caps, for example [do′ (1sg, [beat′ (1sg, shé)])] cause [INGR dead′ (shé)] for (60b) from Mandarin, [do′ (Mary, [paint′ (Mary, door)])] cause [INGR black′ (door)] for Mary painted the door black. Accordingly, the LS in (94a1) should have CAUSE*, which indicates both are possible, and (94a2i) should have cause rather than CAUSE, since it typically does not contain a lexical coding of causation, unlike the causative [2] relations in (94b6). With respect to the possessive predications in (95)–(97), the LSs would be be-at′ (1sg, [exist′ (knig-)]) ^ have′ (1sg, knig-) for Russian in (95a), exist′ ([have′ (3sg, bayan)]) ^ have′ (3sg, bayan) for Guugu Yimidhirr in (96a), and exist′ (libro) ^ have′ (babae, libro) for Tagalog in (97a). The Tagalog case is the most interesting one, because, as noted earlier, there is no obvious way to relate the possessor to the existential predication, and the constructional LS supplies the connection. Latrouite and Van Valin (2014) give an analysis of what are called ‘event existential constructions’ in Tagalog, which, they argue, are a recent innovation and are related to possessive constructions, roughly, ‘the woman has a book which was bought’ ) ‘there was an event of book buying by the woman’. Their RRG account of the transition from possessive to event existential construction crucially involves both the existential and the possession components in the LS. Thus, these representations make explicit what is similar and what is different with respect to the meaning of comparable constructions across languages.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

1.4.3 Semantic Roles RRG makes use of two types of semantic roles: traditional thematic relations, such as agent, patient, instrument, recipient, goal, theme, etc., and semantic macroroles, of which there are only two, actor and undergoer.49 See Kailuweit (Chapter 4) for a detailed discussion of semantic roles. 1.4.3.1 Thematic Relations RRG follows the approach pioneered by Gruber and Jackendoff of deriving thematic relations from decomposed lexical representation for verbs. It does not, however, use the same decomposition system as Gruber and Jackendoff; rather it uses the Aktionsart-based system presented in Section 1.4.2.1. In the RRG system only state and activity predicates have arguments, and therefore different semantic classes of verbs reflect the subclasses of state and activity predicates. Thematic relations are defined in terms of these classes. The full list of classes is given in Kailuweit (Chapter 4), so only a few examples will be given here (see also Van Valin 2005: 55). (98)

a. State predicates 1. State or condition 2. Pure location 3. Perception 4. Cognition 5. Internal sensation 6. Possession b. Activity predicates 1. Unspecified activity 2. Motion 3. Consumption 4. Creation 5. Directed perception 6. Use

dead′ (x) be-LOC ′ (x, y) see′ (x, y) know′ (x, y) feel′ (x, y) have′ (x, y)

x ¼ Patient x ¼ Location, y ¼ Theme x ¼ Perceiver, y ¼ Stimulus x ¼ Cognizer, y ¼ Content x ¼ Experiencer, y ¼ Sensation x ¼ Possessor, y ¼ Possessed

do′ (x, Ø) do′ (x, [walk′ (x)]) do′ (x, [eat′ (x, (y))]) do′ (x, [build′ (x, (y))]) do′ (x, [see′ (x, (y))]) do′ (x, [use′ (x, y)])

x ¼ Effector x ¼ Mover50 x ¼ Consumer, y ¼ Consumed x ¼ Creator, y ¼ Creation x ¼ Observer, y ¼ Stimulus x ¼ User, y ¼ Implement

An important feature of the RRG system is that the basic state and activity predicates take no more than two arguments, and this entails that threeplace predicates must be composed of more than one basic predicate. The primary way this is done is through forming causative LSs (see Table 1.4), and the vast majority of three-place predicates are in fact causative. The argument structure of a verb is the sum of the arguments of its constituent state and activity predicates. For example, Bill knows algebra would be know′ (Bill, algebra). Bill learned algebra would be BECOME know′ (Bill, algebra), with no change in the argument structure. Teach is a threeplace predicate, as in Mr Smith taught Bill algebra, and as predicted, it has a causative LS, namely, [do′ (Mr Smith, Ø)] CAUSE [BECOME know′ (Bill, algebra)]. The argument structure of teach is the combined argument structures of the constituent state and activity predicates, namely effector, cognizer and content. This has important theoretical consequences, as discussed in Kailuweit (Chapter 4) and Van Valin (2005, §2.4.1).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

107

108

ROBERT D. VAN VALIN, JR.

1.4.3.2 Semantic Macroroles The two semantic macroroles, actor and undergoer, are key elements in the linking algorithm. They correspond to the pre-theoretical notions of ‘logical subject’ and ‘logical object’, or alternatively, to the general notions of ‘agent’ and ‘patient’. The latter are often used in two different senses. ‘Agent’ can be narrowly construed to refer to the intending, wilful, controlling and initiating participant, and ‘patient’ can be narrowly interpreted as the participant undergoing a change of state or condition, in which case a verb like kill would take an agent and a patient but see or know would not. On the other hand, there is the general sense of ‘agent’ and ‘patient’, more or less equivalent to ‘logical subject’ and ‘logical object’, such that kill, see and know all take agent and patient arguments, because they are transitive verbs. The narrow sense corresponds to them as thematic relations, and the general sense corresponds to them as macroroles. With respect to morphosyntactic coding, the actor and undergoer arguments are (almost always) direct arguments. In a language like English which relies on word order to signal who is doing what to whom, it means they are not marked by adpositions. In languages with case systems like German and Dyirbal, the macroroles are nominative or accusative (German, Dyirbal first- and second-person pronouns), or they are ergative or absolutive (Dyirbal third-person arguments). Macroroles do not appear in the dative, instrumental or other cases as a default.51 In particular, so-called ‘dative subjects’ in languages like German and Icelandic are not actors; see Van Valin (1991, 2018b) for arguments to this effect. The same holds for the ‘dative objects’ of verbs like ‘help’ and ‘thank’ in numerous languages; they are not undergoers (see Van Valin 2018b: 123, fn. 7). Rather, the dative arguments are non-macrorole direct core arguments. The major exception to the claim that macroroles are always direct arguments is the constructionally specified status of the ‘demoted’ actor in a passive construction; it is always an oblique of some kind: for example English by PP, German vonþdat PP, Japanese ni(yotte) PP, Russian instrumental case RP, Jakaltek -u PP, Bahasa Indonesia untuk PP. Macroroles are related to thematic relations by means of the ActorUndergoer Hierarchy (AUH) in (99). (99) Actor-Undergoer Hierarchy

The actor is the most agent-like argument in the LS, namely the leftmost, while the undergoer is the most patient-like argument in the LS, namely the rightmost. There is an interesting asymmetry between actor and undergoer

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

here: the actor is always the leftmost argument in the LS, whereas the undergoer as the rightmost argument is, for some verbs in some languages, only the default situation. This follows from an asymmetry in lexicalization. The meanings of the macroroles are very general: the actor is the participant who is responsible for the state of affairs, in the sense that there can’t be an event of running without a runner, there can’t be a cognitive event without a cognizer, a perceptual event without a perceiver, etc., whereas the undergoer is the most affected participant in the state of affairs. Languages lexicalize responsibility but not necessarily affectedness. As examples of variable affectedness, consider the pairs of examples in (100). (100)

a. Marie taught French to the students (but they didn’t learn a word of it). a′. Marie taught the students French (?but they didn’t learn a word of it). b. The workmen loaded the hay on the truck. (all of the hay, the truck may or may not be full) b′. The workmen loaded the truck with the hay. (the truck is full (preferred), all of the hay may or may not have been loaded) c. The company shipped the package to Henry (but he never received it). c′. The company shipped Henry the package (?but he never received it).

In all of these examples, the undergoer is the direct RP immediately following the verb. The default selection for undergoer is given in the first member of each pair: French, the hay and the package would all be the rightmost argument in the LS. It is possible with most three-place verbs in English to select the second to rightmost argument as undergoer, as in the second member of each pair, and there is a semantic contrast: the students were affected by the teaching in (a′), that is, they learned French; the hay is the primarily affected participant in (b) but the truck is in (b′); and (c) doesn’t imply that Henry received the package while (c′) does. How a participant is affected differs from verb to verb, but undergoer choice can have semantic consequences. This leads to a more general point: macrorole status has semantic consequences, and this applies to actors as well as undergoers. The thematic relation ‘agent’ was not discussed in the previous section, and that is because a full account of ‘agent’ in RRG terms requires the notion of actor. In (89) and fn. 37, it was mentioned that in Dowty’s system agent arguments were indicated by DO. This makes having an agent argument a lexical property of the verb, but this is problematic in some cases. The verb kill would seem to be a good candidate for a verb that has an agent argument, but that is questionable, as the examples in (101) show. (101)

a. b. c. d.

The hunter killed the deer. The hunter intentionally killed the deer. The hunter inadvertently killed a large dog. Chronic wasting disease has killed many deer.

The default interpretation of (101a) is that shown in (b), not the one in (c), but both are possible. If the actor of kill is an agent, understood as the wilful,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

109

110

ROBERT D. VAN VALIN, JR.

intending and controlling instigator of an action, then (101c) is unexpected. Even more unexpected is (d), which has an inanimate subject. Contrast these examples with those involving the verb murder. (102)

a. b. c. d.

The hunter murdered the game warden. ?The hunter intentionally murdered the game warden. *The hunter inadvertently murdered the game warden. *Falling tree branches have murdered many game wardens.

The picture is very different here. The actor of murder must be an agent as described above, and consequently (b) is redundant, (c) is a contradiction, and (d) is nonsensical. This is different from the actor of kill, which may, but need not be, an agent. Rather, following Holisky (1987) and Van Valin and Wilkins (1996), the actor of kill is an effector, which can be interpreted as an agent if it is human and there is no evidence to the contrary. Accordingly, the default interpretation of (101a) is (101b), as noted above, in which the adverb intentionally specifies that the actor is agentive. In (c) the adverb inadvertently constitutes evidence to the contrary and blocks the agent implicature, which does not apply in (d) due to the nonhuman effector. Three points deserve mention. First, the agent implicature applies only to effector arguments. The human actor of receive or see is not construed as an agent. Receive has a non-volitional recipient-actor and contrasts with accept, which has a volitional recipient-actor. See has a non-volitional perceiveractor, but watch and look at, which are activity verbs of directed perception (see fn. 50), are subject to the agent implicature. Second, agent is fundamentally different from all other thematic relations in that it is always an overlay on top of another thematic relation, most commonly effector. Third, languages differ in the extent to which they lexicalize agentivity. English seems to rely on the agent implicature, as does Tsova Tush (Caucasian; Holisky 1987), whereas in Japanese the human actor of a transitive verb must be interpreted as an agent (Hasegawa 1996), and the agent implicature plays no role. There is also an implicature with undergoer. The first thing to point out is that in English the undergoer with an active voice transitive verb is the direct RP immediately following the nucleus, and this means that Sam is the undergoer in (103a) and the book is the undergoer in (a′). This does not correspond to the traditional analysis that claims that the book is the direct object in both sentences and Sam is likewise the indirect object in both. Evidence comes from passivization, which has the undergoer as the subject in English. There are two candidates for the passive form of (a), as in (103b, b′); there is only one possible passive of (a′), The book was given to Sam by Mary. (103)

a. a′. b. b′.

Mary [A] gave Sam [U] the book.52 Mary [A] gave the book [U] to Sam. Sam [U] was given the book by Mary [A]. ??The book was given Sam [U] by Mary [A].

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Hudson (1992) conducted a survey of varieties of English and found that (103b) is acceptable in all varieties but (b′) is not. Many speakers either interpret it to mean ‘the book received Sam’, which is nonsensical, or found it to be ill-formed. These facts support the claim that the first RP is the undergoer in the so-called ‘double-object construction’. The third arguments in these examples are non-macrorole (NMR) core arguments. The implicature with undergoer concerns the interpretation of change of possession vs. change of location, and is exemplified in (104). (104)

a. a′. b. b′. c. c′. d. d′.

Mary gave Sam the book (*but he never received it). Mary gave the book to Sam (*but he never received it). #Mary gave Boston the book. #Mary gave the book to Boston. Mary sent Sam the book (?but he never received it). Mary sent the book to Sam (but he never received it). ?Mary sent Boston the book. Mary sent the book to Boston.

Give is a change-of-possession verb, while send is a change-of-location verb, and the issue is the interpretation of a human argument as a possessor with change-of-location verbs. Just as some verbs lexicalize their highestranking argument as an agent, there are verbs which unequivocally lexicalize their second highest ranking argument as a possessor. The LS for give would be [do′ (x, Ø)] CAUSE [INGR have′ (y, z)] (alternatively, [do′ (x, [act.on′ (x, z)])] CAUSE [INGR have′ (y, z)]). The success of the transfer of the book to Sam cannot be denied, as the impossibility of adding but he never received it shows. When the y-argument is a location, as in (b, b′), the result is a semantic anomaly, since locations are not normally construed as possessors (but see Section 1.4.2.3). The LS for send, by contrast, is [do′ (x, Ø)] CAUSE [INGR be-LOC ′ (. . ., z)] (alternatively, [do′ (x, [act.on′ (x, z)])] CAUSE [INGR be-LOC ′ (. . ., z)]), with the result state predicate being locational rather than possessional. This means that Sam in (104c, c′) is a location rather than a possessor, strictly speaking, but if the change-oflocation is successfully completed, the primary interpretation is that Sam possesses the book. Undergoer selection is crucial for the interpretation of these examples. When the RP Sam is coded as a non-macrorole oblique argument, as in (c′), the change-of-location may or may not have been successfully concluded. The default interpretation is that it was successful, but adding but he never received it does not generate a contradiction. If successful, Sam can be seen as the possessor of the book. If, on the other hand, the RP Sam is expressed as the undergoer of send, then the strong implication is that the change-of-location was successful, hence the oddity of adding but he never received it, and Sam is construed as the possessor of the book. With a location as the y-argument, as in (d′), and with it coded as a non-macrorole oblique core argument, it is interpreted as the endpoint of a change of location only. Interestingly, when the y-argument functions

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

111

112

ROBERT D. VAN VALIN, JR.

as the undergoer, as in (d), the possession implicature is so strong that it can coerce a reading whereby the RP Boston is construed as some kind of quasi-animate entity, for instance the Boston office of a company, which can possess things. These phenomena have been discussed before in the literature (e.g. Pinker (1989), Rappaport Hovav and Levin (2008), Beavers (2010, 2011), among others), and the usual solution is to claim that, in RRG terms, the LS of the verb changes, that is, ‘. . .CAUSE [INGR be-LOC ′ (y, z)] ) . . .CAUSE [INGR have′ (y, z)]’. This doesn’t capture the flexibility of the interpretation, which is more appropriately viewed as an implicature which can be cancelled or weakened. The ‘human location as possessor’ implicature is strengthened when the human location is the undergoer, as in (104c). The general meaning of undergoer, as mentioned earlier, is the most affected argument, and in sentences like (104c) this leads to the interpretations of the change-oflocation as successful and the new human location as a possessor of the moved entity. It is worth noting that the constructions in (95)–(97) in Section 1.4.2.3, which involve the constructional construal of locative-type arguments as possessors, only get that interpretation when the locative-type argument refers to a human.

1.4.4 Transitivity and the Lexicon The notions of valence and transitivity are distinct, albeit related. The former concerns the number of arguments a verb or other predicate takes, as represented in its LS in RRG, while the latter is typically understood as the number of direct core arguments associated with the predicator in the nucleus of a core: one RP for intransitive predicates, two for transitive and three for ditransitive. RRG takes a somewhat different approach to transitivity from other theories. It distinguishes two types of transitivity: syntactic transitivity [S-transitivity] and macrorole transitivity [M-transitivity]. S-transitivity is basically the traditional notion characterized above, namely one direct core argument ¼ S-intransitive, etc. Far more important for the RRG linking system is M-transitivity, a concept originally proposed in Narasimhan (1998). The relationship between valence and M-transitivity is given in Table 1.6. The meteorological verb snow has no semantic arguments and consequently no macrorole arguments; hence it is M-atransitive. The change-ofstate verb faint and the activity verb limp have one semantic argument and one macrorole argument; hence they are M-intransitive. Consumption and creation activity verbs may occur with or without a second argument (e.g. drink (coffee), write (poetry)), but either way they take only one macrorole and are M-intransitive; see Van Valin and LaPolla (1997, §§3.2.3.3, 4.2) and Van Valin (1990a, 2004) for detailed discussion. It is the active accomplishment versions of these verbs (e.g. drink a cup of coffee, write a poem) which have two macrorole arguments and are M-transitive.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Table 1.6 Valence, macrorole number and M-transitivity

snow faint limp drink [ACT] drink [ACTACC] smash show give

Valence

Macroroles

0 1 1 1 or 2 2 2 3 3

0 1 1 1 2 2 2 2

M-atransitive M-intransitive M-intransitive M-intransitive M-transitive M-transitive M-transitive M-transitive

In general, two-argument activity predicates without a specific incremental theme argument are M-intransitive, while their active accomplishment counterparts are M-transitive, and one piece of evidence for this is the oddity of Pizza was eaten by Sam for ten minutes (activity) as compared with The pizza was eaten by Sam in ten minutes (active accomplishment). There are two major exceptions to this generalization, namely, activity verbs of directed perception, such as watch, and of use, such as use; they are M-transitive as activities. Significantly, they do not have an incremental theme argument and therefore do not participate in the activity– active accomplishment alternation. The remaining verbs (smash, show, give) have either two or three semantic arguments, but all are M-transitive, because there is no such thing as ‘Mditransitive’, since there are only two macroroles; see Van Valin (2004) for extensive argumentation against the existence of a third macrorole. The third semantic argument of a three-argument verb is a non-macrorole core argument, as mentioned earlier, which may be direct (e.g. Max [A] gave/ showed Anna [U] a photo of Sam [NMR]) or oblique (e.g. Max [A] gave/showed a photo of Sam [U] to Anna [NMR]). M-transitivity plays the same role in RRG that syntactic subcategorization information plays in theories that employ it; they both define the relevant notions of transitivity in the different approaches. Syntactic subcategorization properties have long been thought to be highly idiosyncratic, although Chomsky (1986) famously claimed that c[ategory]-selection should follow from s[emantic]-selection. M-transitivity is not random, as macrorole assignment for the majority of verbs follows the Default Macrorole Assignment Principles in (105). (105)

Default Macrorole Assignment Principles a. Number: the number of macroroles a verb takes is less than or equal to the number of arguments in its logical structure 1. If a verb has two or more arguments in its LS, it will take two macroroles. 2. If a verb has one argument in its LS, it will take one macrorole. b. Nature: for verbs which take one macrorole, 1. If the verb has an activity predicate in its LS, the macrorole is actor. 2. If the verb has no activity predicate in its LS, the macrorole is undergoer.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

113

114

ROBERT D. VAN VALIN, JR.

There are two principles here, the first one governing the number of macroroles a verb takes based on the number of arguments specified in its LS, while the second governs which macrorole an M-intransitive predicate takes. The crucial idea in (105b) is that actor macroroles are tied to the presence of an activity predicate in the LS of M-intransitive verbs. There is an important asymmetry between these two principles: there are very few exceptions to (105b) concerning the identity of the single macrorole, but there are numerous exceptions to (105a), which reflects the fact that the same basic semantic content can be expressed more than one way. An example from the domain of possession in English and German is own and besitzen (M-transitive) vs. belong (to) and gehören (M-intransitive). All of these verbs have basically the same LS, have′ (x, y), and own and besitzen follow the default principle in (105a). On the other hand, belong (to) and gehören do not, and this can be indicated by specifying the macrorole number in the LS, namely, have′ (x, y) [MR 1]. Nothing else needs to be specified; all of the morphosyntactic properties of the four verbs follow from the linking algorithm, given their LS and macrorole number, as will be shown in Section 1.6. (106)

a. b. c. d. e. f.

kill receive own belong (to) arrive go

g. h. i. j. k. l. m. n.

seem see watch show run drink melt afraid

[do′ (x, [act.on′ (x, y)])] CAUSE [INGR dead′ (y)] INGR have′ (x, y) have′ (x, y) have′ (x, y) [MR 1] INGR be-at′ (x, y) do′ (x, [move.away.from.ref.point′ (x)]) ^ PROC covering.path. distance′ (x, (y)) ^ FIN be-LOC ′ (z, x) seem′ (x, y) [MR 0]53 see′ (x, y) do′ (x, [see′ (x, y)]) [do′ (w, Ø)] CAUSE [INGR see′ (x, y)] do′ (x, [run′ (x)]) do′ (x, [drink′ (x, y)]) BECOME melt′ (x) feel′ (x, [afraid′ (y)])

The lexicon is a vital component in the RRG system, and it is more than just a storehouse for words and morphemes. It can be divided into two components, which may be metaphorically labelled ‘the warehouse’ and ‘the workshop’. The workshop is the location where lexical and derivational rules apply, and therefore it is where the LS for the semantic representation is assembled, among other operations. Crucially, the lexical items to fill the argument positions in the semantic representation are selected at this point. More will be said about this in Section 1.6.

1.5

Information Structure

The third major representation RRG assigns to a sentence is its information structure; see Bentley (Chapter 11) for an introduction to the RRG approach

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.33 The constituent, operator and information structure projections of the clause

to information structure. In Figure 1.1 it is instantiated by the arrow paralleling the linking algorithm, which symbolizes the fact that information structure interacts with the linking algorithm in important ways, which is discussed by Latrouite and Van Valin (Chapter 12). RRG syntactic representations have two projections, the constituent projection and the operator projection, and one of the advantages of a multi-projection representation is that it is always possible to add more projections.54 Figure 1.33 is Figure 12.4 from Chapter 12, which adds the information structure projection; it is explicated in detail in that chapter. The components of the information structure projection are, first of all, the information units (IU), which correspond to the amount of information contained in a simple wh-expression, that is, what, who, when, where (Lambrecht 1994). The bracketing around the IU nodes represents the topic– comment division of the utterance; in this example, he, the topic expression, refers to the topic referent, and bought a new car constitutes the comment about the topic referent. Below that is the partition of the utterance into focus and background. The heavy broken black line indicates the potential focus domain, the syntactic domain in which the focus can potentially fall, and the triangle represents the AFD, the part of the utterance that is the focus of the assertion in the context in which it occurs. The part of the utterance that is outside of the AFD is the background. The default relationship between the two bifurcations of the sentence is that the topic expression occurs within the background, and the AFD occurs within the comment. These notions will be explained in detail in Chapter 11, and in Chapter 12 their role in the linking between semantics and syntax will be illustrated. It is an important aspect of RRG that the analysis of grammatical phenomena can refer to one or more of these three projections as well as the LS and semantic roles. This will be amply illustrated in the chapters of this book.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

115

116

ROBERT D. VAN VALIN, JR.

1.6

Linking between Syntax and Semantics in Simple Sentences

The heart of the RRG system is the linking algorithm which maps between the semantic and syntactic representations. It is bidirectional: it maps the semantic representation into the syntactic representation, and it maps from syntax to semantics as well. This reflects the fact that language users are bidirectional. A speaker goes from a message to be conveyed to the formal packaging of it which is to be uttered (semantics to syntax), while a hearer analyses the formally packaged message and gives it an interpretation (syntax to semantics). As discussed in Chapter 19, RRG is not a neurocognitive processing model but has the potential to form the basis for one. Specific aspects of the linking algorithm are discussed in depth in different chapters (e.g. grammatical relations in Chapter 5, case assignment in Chapter 7, and argument structure alternations in Chapter 6). The purpose of this section is to give an overview of the system as a whole. Each direction of mapping will be presented separately, starting with semantics-to-syntax linking. The linking algorithm is governed by a very important principle, the Completeness Constraint. (107)

Completeness Constraint: All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence.

The application of this constraint will be exemplified in the linkings discussed below.

1.6.1 Linking from Semantics to Syntax The steps of the semantics-to-syntax linking algorithm are given in (108). (108)

The linking algorithm from semantics to syntax a. Step 1: Construct the semantic representation of the sentence based on the LS of the main predicator. b. Step 2: Assign actor and undergoer, following the AUH. c. Step 3: Determine the morphosyntactic coding of the arguments of the main predicator. i. Select the privileged syntactic argument (PSA), following the PSA Selection Hierarchy (see (111)). ii. Assign the arguments the appropriate case markers and/or adpositions. iii. Assign agreement marking to the main verb or auxiliary, as appropriate. d. Step 4: Select the syntactic templates for the sentence, following the syntactic template selection principles. e. Step 5: Assign the nucleus, the arguments and the adjuncts to positions in the syntactic representation of the sentence.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Steps 1 and 2 take place in the lexicon, specifically in the ‘workshop’. The central move in step 1 is the selection of the predicator in the nucleus and its LS. The argument positions in it must be filled by referring expressions, and the choice is determined by the activation status of the referents of the referring expressions; see Chapter 11. In addition, any adjunct LSs must also be incorporated into the representation, along with the relevant operators. As an example, the situation to be described is one in which a woman draws a man’s attention to something in a work situation subsequent to an online meeting. The verbal predicator selected is show, and the referring expressions to fill the argument positions are Linda, the salesman, and what. In addition, there are two predicative prepositions, after and in, which take one referring expression each, namely the meeting and the office, respectively; the other argument position is filled by the LS of show. Finally, the operator values are determined, specifically past tense and interrogative IF. The result is the representation in (109). (109)

⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (Linda, Ø)] CAUSE [INGR see′ (salesman, what)]])])]⟩⟩

Step 2 is to assign macroroles, and while Linda must be the actor, the speaker has a choice with respect to the assignment of undergoer. The default is that what would be the undergoer, and that choice will be reflected in the following discussion. The result of Step 2 is (110). (110)

⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (A: Linda, Ø)] CAUSE [INGR see′ (NMR: salesman, U: what)]])])]⟩⟩

It should be noted that this is not a distinct level of representation; the semantic representation from (109) is merely informationally enriched with macrorole information. The remaining steps take place outside of the lexicon. Step 3 concerns the morphosyntactic coding of the arguments in the LS, and the most important part of it is PSA selection. (See LaPolla, Chapter 5, for explication of the RRG theory of grammatical relations.) The PSA, as is clear from the name, is the most important core argument, and in many cases it corresponds to the traditional ‘subject’. The most significant difference between them is that the notion of PSA is construction-specific, that is, it is defined as the privileged argument in a particular construction, whereas the notion of ‘subject’ applies to the language as a whole. Thus, one can talk about the ‘PSA of a construction’ but not ‘the subject of a construction’. The traditional notion of ‘subject’ is a generalization across the PSAs of particular constructions in languages in which the PSAs of the majority of constructions are the same. (See Watters, Chapter 6, for an analysis of the most important constructions involving argument alternations.) There are numerous examples of languages in which different constructions have different PSAs (e.g. Tagalog, Schachter 1976; Jakaltek, Van Valin 1981), leading to the conclusion that

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

117

118

ROBERT D. VAN VALIN, JR.

there is no predominant PSA in the system and therefore no true ‘subject’ in the traditional sense. The PSA Selection Hierarchy and principles are stated in (111). (111)

a. Privileged Syntactic Argument Selection Hierarchy: arg of DO > 1st arg of do′ > 1st arg of pred′ (x, y) > 2nd arg of pred′ (x, y) > arg of pred′(x) b. Privileged syntactic argument selection principles: 1. Accusative constructions: Highest-ranking argument in (111a) (default) 2. Ergative constructions: Lowest-ranking argument in (111a) – default c. Restrictions on PSA in terms of macrorole status: 1. Languages in which only macrorole arguments can be PSA: Croatian, . . . 2. Languages in which non-macrorole direct core arguments can be PSA: Icelandic, . . .

The hierarchy in (111a) resembles the AUH, except that it is unidirectional: ‘arg of DO’ is the highest-ranking argument, ‘arg of pred′ (x)’ the lowest. In accusative constructions, the highest-ranking argument is the (default) choice for PSA. The qualification ‘default’ is in parenthesis, because in some languages, those that lack voice systems, the highestranking argument is the only possible choice for PSA. In languages with a voice system, the highest-ranking argument is merely the default choice of accusative constructions. In ergative constructions, on the other hand, the lowest-ranking argument is always only the default choice, and there exist grammatical means to allow the highest-ranking argument to be the PSA, normally an antipassive voice. (See Chapters 5 and 6 for discussion of voice in RRG.) A very important consideration in PSA selection is given in (111c), namely, whether only macrorole arguments can be selected as PSA, or whether nonmacrorole arguments can function as the PSA in a syntactic construction. This issue is standardly characterized as ‘are so-called dative subjects real subjects?’, because dative is the default case for non-macrorole direct core arguments (see Nakamura 2021, this volume, Chapter 7; Van Valin 1991, 2018b). Both German and Icelandic have verbs which appear to have a dative ‘subject’, as shown in (112). (112)

a. Mir schmeck-t das Brot. 1sg.dat taste-3sg.pres the.nom.neut.sg bread ‘I like the bread,’ lit. ‘The bread tastes [good] to me.’ lík-ar matur-in-n b. Þeimi 3pl.dat like-3sg.pres food-def-m.sg.nom ‘They like the food,’ lit. ‘The food likes to them’.

German

Icelandic

In both languages, the default word order is for the dative RP to be first in the core, and the finite verb agrees with the nominative RP. If we put these clauses into a conjunction reduction construction, as in (113), the different status of the dative RPs is clear (Rögnvaldsson 1982; Zaenen et al. 1985).55

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar (113)

a. Þeimi lík-ar matur-in-n og proi borð-a mikið. 3pl.dat like-3sg.pres food-def-m.sg.nom and eat-3pl.pres much ‘They like the food and eat much.’ das Brot und proi essen viel. b. *Ihneni schmeck-t 3pl.dat taste-3sg.pres the.nom bread and eat much ‘*They like the bread and eat a lot.’ das Brot und siei essen viel. b′. Ihneni schmeck-t 3pl.dat taste-3sg.pres the.nom bread and they.nom eat much ‘They like the bread and they eat a lot.’

In a conjunction reduction construction in a right-branching language, there are two PSAs: the missing argument in the second clause (the pivot) and the argument in the first clause which supplies the interpretation of the missing argument (the controller). In Icelandic, the dative argument in the first clause can serve as the controller of the nominative pivot in the second clause, as the (113a) example shows, but this is not the case in German, which does not permit dative controllers in this construction; consequently, there must be a nominative pronoun in the second clause, as in (113b′). In this construction, Icelandic permits non-macrorole direct core arguments in the dative case to function as a PSA, whereas German does not; it restricts PSA-hood to macroroles only. (It is also possible in this construction for the controller to be nominative and the pivot to be the dative argument of the verb in Icelandic but not in German.) In both languages, however, person agreement on a finite verb or auxiliary is restricted to macrorole arguments, namely the nominative RP. This highlights the construction-specific nature of grammatical relations. In German, the controller of person agreement on finite verbs and auxiliaries is the same as the controller in the conjunction reduction construction, namely the highest-ranking macrorole core argument. In Icelandic, on the other hand, the controller in the conjunction reduction construction is the highest-ranking direct core argument, regardless of whether it is a macrorole or not, which is different from the controller of person agreement on the finite verb or auxiliary, which is the highest-ranking macrorole core argument.56 Accordingly, German is consistent in having the same PSA for more than one construction, while Icelandic is not consistent, with different constructions having different PSAs. The other two parts of step 3 in (108c) concern case and preposition assignment, on the one hand, and agreement, on the other, both of which were touched on in the above discussion. The RRG approach to case assignment is discussed in Nakamura (Chapter 7), and preposition assignment is presented in Ibáñez Cerda (Chapter 10). The basic case assignment rules for direct core arguments are given in (114); they refer to case assigned by verbs, not case assigned by adpositions. (114)

a. Accusative constructions: 1. Assign nominative case to the highest-ranking macrorole. 2. Assign accusative case to the other macrorole. 3. Assign dative case to non-macrorole arguments (default).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

119

120

ROBERT D. VAN VALIN, JR.

b. Ergative constructions 1. Assign absolutive case to the lowest-ranking macrorole. 2. Assign ergative case to the other macrorole. 3. Assign dative case to non-macrorole arguments (default).

When there is only one macrorole in a core, it counts as the highest ranking for (114a) or the lowest ranking for (114b). The English preposition assignment rule that is relevant to the example in (109) is given in (115). (115) To-assignment rule:

Assign to to the non-macrorole a argument in the LS configuration BECOME/INGR/FIN predicate′ (a, b)

This says that when a two-place state predicate is in a change-of-state LS and the first argument is not a macrorole, it is marked by to. Finally, the rule for triggering person agreement on a finite verb or auxiliary is the same in English as the rule in German and Icelandic discussed above. Accordingly, the output of step 3 of the linking algorithm is given in (116). (116)

⟨IF INT ⟨TNS PAST [be-after′ (meeting [ACC], [be-in′ (office [ACC], [[do′ (A: PSA [NOM] Linda, Ø)] CAUSE [INGR see′ (NMR: to [ACC] salesman, U: [ACC] what)]])])]⟩⟩ {ACT: 3sg}

A number of points need to be clarified. First, nominative case and accusative case in English are vacuous on full RPs and realized only on pronouns. Second, English prepositions assign accusative case to their objects, vacuously in this example. Third, ‘ACT: 3sg’ under CAUSE indicates ‘active voice’ and the person and number of the PSA. Fourth, the to-rule in (115) comes into play, because the salesman is the non-macrorole first argument of . . . INGR see′ (y, z) . . . Fifth, as with the output of step 2, this is not a new level of representation but an informationally enriched version of the original representation in (109), now containing both semantic and morphosyntactic information. The next step is the selection of the syntactic templates which constitute the syntactic representation of the sentence. Syntactic templates were introduced in Section 1.2.6, and some of the templates used in English are given in Figure 1.20. The selection of the syntactic templates is based on the output from step 3. There are five RPs: meeting, office, salesman (common sortal nouns), Linda (proper name), and what (interrogative pronoun). These require three common noun templates, one proper noun template, and one pronoun template. There are three PPs, two predicative and one nonpredicative, and the appropriate templates are given in Figure 1.20. The core template is the most important template, and the selection principles for it are given in (117). (117)

a. Syntactic template selection principle: The number of syntactic slots for arguments within the core is equal to the number of distinct specified argument positions in the semantic representation of the core.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar b. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre- or post-core slot reduces the number of core slots by 1 [may override (1) above].

The principle in (117a) is common sense: the number of specified arguments in the LS of the predicator corresponds to the number of slots for arguments required in the core template. There is also a slot required for the nucleus. This principle reflects the default situation, and there are languagespecific qualifications in (117b). The first is exemplified by languages which require dummy ‘subjects’ for argument-less meteorological verbs (e.g. English it is raining, French il pleut, German es regnet). The second concerns voice constructions in which the default choice for PSA is either omitted (unspecified) or appears as a peripheral adjunct. The third applies to languages which can displace XPs to the PrCS or PoCS. The second and third qualifications can combine to override the first qualification, for example in Who was arrested by the police?, who is in the PrCS and by the police is in the core-level periphery, leaving just the nucleus within the core. In the example under discussion, the occurrence of an interrogative pronoun (what) as one of the arguments of show invokes (117b3) and also requires a PrCS template. One of the options for linking a temporal adjunct is to link it to the PrDP where it functions as a frame-setting topic. In the example under discussion, this option will be exercised for the temporal adjunct after the meeting. Consequently, it will be necessary to have a PrDP template. Finally, the operators in (109) are interrogative IF and past tense, and this means that there needs to be an auxiliary do in order to have tense occur core-initially as a signal of interrogative IF. Thus, the templates needed to give the output of Step 4 are given in Figure 1.34.

Figure 1.34 Syntactic templates required by the output of Step 3

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

121

122

ROBERT D. VAN VALIN, JR.

The assembly of the syntactic representation follows the example in Figure 1.21 and sets the stage for the final step. Step 5 in (108e) says ‘Assign the nucleus, arguments and adjuncts to positions in the syntactic representation of the sentence’, but this a major simplification; the full step is given in (118). (118)

Step 5. Assign the nucleus, arguments and adjuncts to positions in the syntactic representation of the sentence. a. Assign the nucleus to the appropriate position in the core. b. Assign the [WH] arguments to the appropriate positions in the clause. c. If there is a [þWH] XP, 1. assign it to the normal position of a non-wh-XP with the same function, or 2. assign it to the pre-core or post-core slot, or 3. assign it to a position within the potential focus domain of the clause (default ¼ the unmarked focus position). d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to information structure restrictions (optional). e. Assign the argument(s) of LS(s) other than that of the predicator in the nucleus to 1. the core- or clause level periphery (default), or 2. the pre-core or post-core slot, or 3. the pre-detached or post-detached position.

The formulations in (118a, b) are intentionally vague and for good reason: every language does it differently. While there are well-known cross-linguistic tendencies with respect to word (or phrase) order, languages differ in the very fine-grained constraints on word order, even if they follow these tendencies with respect to larger-grained phenomena. As mentioned earlier, it is possible to include the nucleus in the English core templates because its position is virtually always the same: after the PSA and before the other core arguments and phrasal adjuncts. This is not possible in languages like German, Croatian and Dyirbal, in which the nucleus has no fixed position, and in these three languages the linearization constraints on its position in the core are different. With respect to the linearization of non-wh argument expressions, the constraints on their order vary dramatically across languages, as a brief comparison of, for example, English, Croatian, Dyirbal and German would show. The situation regarding wh-expressions is considerably simpler: there are basically three possibilities. First, they could occur in situ, that is, the normal position of a non-wh-expression with the same function. Second, they could appear in the PrCS or PoCS, as in English and Dhivehi (see (6)), and third, they could appear in the unmarked focus position in the clause (see Chapter 11), as in Turkish and Basque. It should be noted that occurrence of a semantic argument in the PrCS or PoCS does not involve movement, as the wh-expression is linked directly from the semantic representation to the PrCS or PoCS; furthermore, there is no co-indexed empty RP or

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.35 The output of the semantics-to-syntax linking for (109)

unpronounced copy of it in the core, as the tree structure in Figure 1.6 clearly shows, following (117b3). As mentioned before, appearance in the PrCS or PoCS is not restricted to wh-expressions, as the statements in (118d) and (118e2) make clear. Direct linking to the PrDP or PoDP is possible only for adjuncts, because non-wh syntactic arguments in a detached position normally require a resumptive pronoun in the clause. The output of step 5 is the finished sentence, which is given in Figure 1.35. A condensed representation of the linking is given in Figure 1.36. The operator projection is omitted, and the added information, namely macrorole status, PSA-hood, case, etc., is represented above the argument to which these properties are added for clarity of exposition. The numbers refer to the steps in the linking algorithm.

1.6.2 Linking from Syntax to Semantics The essence of the syntax-to-semantics linking algorithm can be summed up in the following steps. (119)

Linking from syntax to semantics (summary) a. The parser outputs a labelled tree structure. b. The first step is to derive as much information from the overt morphosyntactic features of the clause: case marking/word order, the voice of the verb, adpositions. c. The second step is to retrieve the LS of the verb from the lexicon and assign macroroles where possible.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

123

124

ROBERT D. VAN VALIN, JR.

Figure 1.36 Summary of the linking from semantics to syntax for (109)

d. The information from these steps should link everything in the core to the argument positions in the LS; if there is an element in the PrCS, it will be linked last, to the remaining unlinked argument position in the LS. If there are peripheral adjuncts, retrieve the LS of the predicative adposition and add this LS to the semantic representation.

The linking presupposes that the parser has output a labelled syntactic tree. This is an idealization that is appropriate for an abstract grammar, but it does not reflect the nature of real-time sentence processing, in which interpretation is incremental and begins with the recognition of the first word of the sentence and does not wait for the entire sentence to be uttered before starting the interpretive process. (See Chapter 19 for a proposal which attempts to account for incremental interpretation in an RRG-based sentence processing system.) The first step is to glean all information from the morphosyntax of the sentence. In a language with a voice opposition, the voice of the predicator in the nucleus is crucial, since it signals the function of the PSA, in English, the first RP in the core, and in German, the nominative RP. With intransitive predicators, on the other hand, voice and the PSA Selection Hierarchy are irrelevant, because there is only one direct core argument, which can be either actor or undergoer, depending on the semantics of the predicator. In addition, there is a wh-RP in the PrCS; it carries no clues as to its function in the sentence in English, as in principle it could be the PSA (e.g. What upset Mary?), a non-PSA undergoer (e.g. What did Bill buy?), a nonmacrorole direct core argument (e.g. What did Sally show Linda?), or an oblique core argument (e.g. What did they load the truck with?). In addition, there are three PPs, one in the core, one in the core-level periphery, and one in the PrDP.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

The second step is to call up the LS of the predicator in the nucleus, in this case, show, and assign macroroles. The x-argument is assigned actor, but it is not possible to assign undergoer, due to the fact that either y or z could be the undergoer with this verb. The final step is to reconcile the results of the first two steps in order to satisfy the Completeness Constraint in (107). There is one direct connection: the RP Linda is the actor, the actor is the x-argument in the LS of show, and therefore Linda is the x-argument. The other RP in the core, the salesman, is the object of the non-predicative preposition to and must be linked to either the y- or the z-argument in the LS. The key here is the preposition assignment rule in (115), which states ‘Assign to to the non-macrorole a argument in the LS configuration BECOME/INGR/FIN predicate′ (a, b)’. In terms of the LS for show, the y-argument is the one corresponding to the ‘non-macrorole a argument in the LS configuration INGR predicate′ (a, b)’ , and accordingly the RP the salesman must be linked to the y-argument, as shown in Figure 1.47. That leaves one unlinked RP argument in the PrCS, what, and one unlinked argument position in the LS, namely the z-argument. In order to satisfy the Completeness Constraint, the wh-RP must be linked to the zargument, and this yields the correct interpretation of it. The Completeness Constraint is not yet fully satisfied, however, as there are two unlinked PPs in the sentence. These are adjunct predicative PPs which license their arguments and must be incorporated into the semantic representation of the sentence. Since the prepositions are predicative, they have lexical entries in the lexicon, and accordingly it is necessary to go back to the lexicon and retrieve their LSs. The question arises as to the order of composition of the two prepositional predicates. The locative PP is inside the clause in the core-level periphery, while the temporal PP is a frame-setting topic outside of the clause in the PrDP. Consequently, the locative LS is added, taking the LS of show as one of its arguments, and then the temporal LS is added, taking the locative LS as one of its arguments. When the operators are added (not shown), the result is the semantic representation in (116).

1.6.3

Representing General Principles vs. Language-Specific Grammatical Information The linking algorithms as presented in (108) and (119) contain both crosslinguistically valid general principles as well as language-specific grammatical rules, constraints and principles. For example, the syntactic template selection principles in (117) have one very general principle in (117a) and three language-specific qualifications in (117b). It is reasonable to include these exceptions, since they are quite common, but on what basis can one motivate their applicability to a given language? This can best be achieved by having language-specific representations of morphosyntactic information, and in RRG these are known as constructional schemas

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

125

126

ROBERT D. VAN VALIN, JR.

(CSs). As an example, consider the German ‘plain’ passive construction, illustrated in (120). (120)

a. Hans ha-t den Rechner kaputtgemacht. have-3sg.pres the.acc.sg computer broken.made ‘Hans broke/ruined/destroyed the computer.’ b. Der Rechner ist von Hans kaputtgemacht worden. the.nom.sg computer be.3sg.pres from broken.made pass.aux ‘The computer was broken/ruined/destroyed by Hans.’

In order to formulate a CS for the construction in (120b), it is necessary to introduce briefly the RRG theory of voice (see also Watters, Chapter 6; Wu, Chapter 26). RRG posits two abstract voice constructions: PSA-modulation voice, which links an argument other than the default choice to be the PSA, and argument-modulation voice, which gives non-canonical treatment to the default choice for PSA. These characterizations are independent of the passive vs. antipassive opposition, since canonical passive and antipassive constructions involve both. Crucially, however, they need not co-occur; it is not uncommon for a passive construction to involve only argument modulation, for example Ute, Sama, impersonal passives of intransitive verbs in German and Dutch, or for an antipassive construction to have only PSAmodulation, for example Jakaltek; see Van Valin (1981, 2005: 115–120, esp. fn. 16). The CS for the ‘plain’ German passive with M-transitive verbs is given in Figure 1.37; it shares a number of features with the ‘impersonal’ passive with M-intransitive verbs, and there is also the ‘bekommen-passive’, which has

CONSTRUCTION: German ‘plain’ passive with M-transitive verbs

SYNTAX: PSA: (111b1, c1); Variable [±pragmatic influence] Linking: PSA-modulation, Argument-modulation (core-level peripheral PP, Ø) Template(s): (117b2) MORPHOLOGY: Verb: Past participle AUX: werden (PERF: worden) P: von SEMANTICS: PSA is not instigator of state of affairs but is most affected participant (default) PRAGMATICS: Illocutionary force: Not imperative Focus structure: Predicate focus (default) Figure 1.37 Constructional schema for German ‘plain’ passive57

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

quite different properties from the ‘plain’ werden-passive (Van Valin 2003; Diedrichsen 2004). The schema specifies the defining syntactic, morphological, semantic and pragmatic properties of the construction.58 The first line under ‘syntax’, ‘PSA’, states that it is an accusative construction by referring to the PSA Selection Hierarchy in (111b1) and that the PSA in this construction must be a macrorole argument, (111c1). ‘Variable [pragmatic influence]’ indicates that the PSA is not fixed and that the choice of which macrorole functions as PSA may be influenced by information-structure factors (see LaPolla (Chapter 5), Bentley (Chapter 11), Latrouite and Van Valin (Chapter 12)). The next line identifies the construction as a voice construction involving both PSAand argument-modulation and specifies that argument modulation is instantiated either by the default choice for PSA occurring in a PP in the core-level periphery or by it being omitted. This is indicated by inserting a ‘Ø’ in the leftmost argument position of the LS. The third line states that the exception in (117b2) for argument-modulation voice construction comes into play in the selection for the syntactic template. Thus, the entries under ‘syntax’ in the CS give the language-specific values of the cross-linguistically valid principles in the theory. The next entries under the heading ‘morphology’ give the relevant morphological properties of the construction. The first entry gives the required form of the verb, namely the past participle, and the second identifies the auxiliary used in the construction, namely werden ‘become’, which has an irregular past-participle form when it is used as a passive auxiliary, worden, which lacks the normal ge- prefix: Es ist wärmer geworden ‘It has got warmer’ (*Es ist wärmer worden) vs. *Der Rechner ist von Hans kaputtgemacht geworden (cf. (120b)). The final part specifies the P in the adjunct PP which contains the actor, the default choice for PSA. It is von ‘from, of’, and the fact that it takes a dative object need not be specified here, as that is a general property of von. The characterization of the meaning of the construction under the ‘semantics’ heading is quite general but appropriate for the construction. There are two entries under ‘pragmatics’. The first states that this construction can be used in assertions and questions but not in imperatives, and the second reflects the fact that the default focus structure is predicate focus, with a topical PSA and a focal predicate phrase (see Chapter 11), and accordingly in a passive construction, just as in an active voice form, the nominative PSA is interpreted as the default topic, although it must be emphasized that this is only the default interpretation. CSs thus represent language-specific information, formulated in terms of the appropriate general, cross-linguistically valid principles. The schemas play an important role in the linking between form and meaning. In Van Valin (2005) there is a detailed discussion of the integration of the linking algorithm with CSs in Sama wh-questions (pp. 132–135, 157). In order to illustrate the interaction of CSs with the linking algorithm, the schema for the Sama antipassive is taken from Van Valin (2005) and presented here in

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

127

128

ROBERT D. VAN VALIN, JR.

CONSTRUCTION: Sama antipassive

SYNTAX: PSA: (111b2, c1); Variable [±pragmatic influence] Linking: PSA-modulation, argument-modulation: optional if 3p (Ø), obligatory if ~3p] (Ø, core PP) Template(s): (117b) [only with Ø argument-modulation]

MORPHOLOGY: Verb: N- + stem P: ma

SEMANTICS: PSA is instigator of state of affairs If predicate is stative, then it shifts to activity in actor voice If lowest ranking argument is not marked by determiner, then preferentially indefinite or non-referential PRAGMATICS: Illocutionary force: Unconstrained Focus structure: Predicate focus (default) Figure 1.38 Constructional schema for Sama antipassive

Figure 1.38 is an updated form, along with the relevant data in (121) (Walton 1986, personal communication). The homorganic nasal prefix Nwhich marks actor voice is realized as Nga- when prefixed to a verb beginning with /n/. (121)

a. Ø-Nda’ ku/ d’nda onde’ hi’. Undergoer voice (active) atv-see 1sg.erg/woman child det ‘I/the woman sees the child.’ a′. Nga-nda’ aku/ d’nda onde’ hi’. Actor voice (antipassive) antip-see 1sg.nom/woman child det ‘I/the woman is looking at/for the child.’ b. Ø-Nda’ onde’ hi’ aku. atv-see child det 1sg.nom ‘The child sees me.’ b′. Nga-Nda’ onde’ ma aku. antip-see child obl1sg.nom ‘The child is looking at/for me.’ c. Say Ø-nda’ d’nda hi’? who atv-see woman det ‘Who does the woman see?’ (‘*Who is looking for/at the woman?’) c′. Say Nga-nda’ d’nda hi’? who antip-see woman det ‘Who is looking at/for the woman?’ (‘*Who does the woman see?’)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

The relevant facts about actor and undergoer voices are summarized in the pairs of examples in (121a, b); the first example is undergoer voice (active) and the second is actor voice (antipassive). In the (a, a′) examples the actor is either first person (aku/ku) or third person (d’nda ‘woman’), and the undergoer is third person (onde’ ‘child’). Sama has case marking only on pronouns. In the undergoer (active) voice the full RPs are distinguished by word order, while the pronominal actor occurs in the ergative form. In the actor (antipassive) voice, the pronominal actor occurs in the nominative form, signalling that it is the PSA. A striking feature of this construction is that when the undergoer is third person, it may occur as a direct core argument.59 If, however, the actor is third person and the undergoer is first or second person, as in the (b, b′) examples, the situation is very different. In the undergoer voice, the first-person undergoer occurs in the nominative form aku, as in (121b), but in the actor voice the first-person undergoer is marked by the oblique preposition ma plus the nominative pronoun, as in (121b′). The CS for this voice construction is given in Figure 1.38. The PSA entry specifies that this is an ergative PSA restricted to macroroles, while the linking entry states that the primary function of the construction is PSA-modulation, with argument-modulation optional for thirdperson arguments (omission), but obligatory for non-third-person arguments, the options being omission or occurring as an oblique core argument in a PP as in (121b′). The final entry states that the default syntactic template selection principle in (117a) applies unless there is optional argumentmodulation, and then only when the choice is omission, since the PP for non-third persons is in the core, not a periphery. The relevance of this CS to wh-questions in Sama is clearly illustrated in (121c, c′). Wh-expressions like say ‘who’ or ay ‘what’ occur in the PrCS and must be the PSA of the clause. Accordingly, if the verb is in undergoer voice, as in (121c), the wh-expression must be interpreted as the undergoer, and d’nda ‘woman’ must be interpreted as the actor. On the other hand, if the verb is in actor (antipassive) voice, then the wh-expression is the actor and the postverbal RP d’nda’ must be interpreted as the undergoer. The CS for wh-questions invokes the antipassive CS when the wh-expression is an actor. Thus, the schema for the Sama antipassive, like the one for the German passive, combines cross-linguistically motivated theoretical principles with language-specific information. The interaction of CSs with the linking algorithm will be discussed in more detail in the next section. The issue to be investigated in the remainder of this section is the range of ways that language specific-grammatical information can be represented. CSs are one of the primary devices for representing language-specific grammatical information. Some cases of word-order variation lend themselves readily to a constructional analysis, for instance the occurrence of a phrase in an extra-core position (PrDP, PrCS, PoCS, PoDP) subject to informationstructural constraints. Some basic word-order patterns also seem amenable to a constructional analysis, even though a given word order need not be

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

129

130

ROBERT D. VAN VALIN, JR.

associated with a specific semantic content, for example in English [CORE RP [NUC V ] PP ] can instantiate ‘Actor – Activity Verb – Oblique’ (e.g. Sam sang to Sally. Tom talked to Max. Bill ran to the beach. Fred looked at the parade.) as well as ‘Undergoer – Process Verb – Oblique’ (e.g. The gravy spilled on the table. The napkin fell on the floor. The ice cream melted onto the counter.). In a language like English or French, which has relatively rigid word order, the syntactic templates capture many of the basic word-order patterns. More interesting is the situation in languages like German, Croatian and Warlpiri, which have very free phrase order together with one fixed point of reference, namely second position in the clause. The factors which influence the ordering of the constituents other than the one associated with second position are complex and involve information structure, phonological weight, inherent lexical content (‘animacy’), among other things.60

1.6.4

Interaction of Constructional Schemas and the Linking Algorithm In the next two sections the interaction of CSs with the semantics-to-syntax and syntax-to-semantics linking algorithms will be illustrated. 1.6.4.1 Linking from Semantics to Syntax Revisited The linking algorithm was given in (108) and (118), and (118c) and (e) contain options which different languages may utilize. These can be replaced by language-specific CSs, the range of options being a meta-generalization across the CSs. In principle, CSs can contribute language-particular information at each one of the steps in (108). To illustrate this, the linking from semantics to syntax in two Japanese passive constructions will be illustrated, namely, the plain passive in (122) and the adversative passive in (123)–(125) (Kuno 1973; Imai 1998; Toratani 2002), each with its own CS in Figures 1.39 and 1.40. (122)

a. Hanako ga yakuza o korosi-ta. nom gangster acc kill-pst ‘Hanako killed the gangster.’ b. Yakuza wa Hanako ni(yotte) koros-(r)are-ta. Gangster top ‘by’ kill-pass-pst ‘The gangster was killed by Hanako.’ c. [do′ (Hanako, [act.on′ (Hanako, yakuza)])] CAUSE [INGR dead′ (yakuza)]

(123)

a. Hanako ga nekom-da. nom become.bedridden-pst ‘Hanako became bedridden.’ a′. INGR bedridden′ (Hanako) b. Taroo wa Hanako ni nekom-(r)are-ta. top dat become.bedridden-pass-pst ‘Taro was affected by Hanako’s becoming bedridden.’ b′. feel.affected′ (Taroo, [INGR bedridden′ (Hanako)])

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

CONSTRUCTION: Japanese ‘plain’ passive

SYNTAX: PSA: (111b1, c2); Variable [±pragmatic influence] Linking: PSA-modulation, argument-modulation: (Ø, core-peripheral PP) Template(s): (117b) MORPHOLOGY: Verb: Stem + (r)are P: ni(yotte) SEMANTICS: PSA is not the instigator of state of affairs but undergoes the action of the verb. PRAGMATICS: Illocutionary force: Not imperative Focus structure: Predicate focus (default) Figure 1.39 Constructional schema for Japanese plain passive

CONSTRUCTION: Japanese adversative passive

SYNTAX: PSA: (111b1, c2) Linking: Highest ranking argument of stem verb —> NMR Template(s): (117a) MORPHOLOGY: Verb: Stem [renyookee ‘infinitive’ form] + (r)are Case: Assignment follows Imai (1998) SEMANTICS: -(r)are adds feel.affected′ (x, [y]) [MR Ø] to LS of stem verb y PSA feels negatively affected by state of affairs PRAGMATICS: Illocutionary force: Not imperative Focus structure: Predicate focus (default) Figure 1.40 Constructional schema for Japanese adversative passive

(124)

a. Hanako ga odot-ta. nom danced-pst ‘Hanako danced.’ a′. do′ (Hanako, [dance′ (Hanako)]) b. Taroo wa Hanako ni odor-are-ta. top dat dance-pass-pst ‘Taro was affected by Hanako’s dancing.’ b′. feel.affected′ (Taroo, [do′ (Hanako, [dance′ (Hanako)])])

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

131

132

ROBERT D. VAN VALIN, JR.

(125)

a. Hanako ga piza o tabe-ta. nom pizza acc eat-pst ‘Hanako ate the pizza.’ a′. do′ (Hanako, [eat′ (Hanako, piza)]) b. Taroo ga Hanako ni piza o tabe-rare-ta. nom dat pizza acc eat-pass-pst ‘Taroo was affected by Hanako’s having eaten the pizza.’ b′. feel.affected′ (Taroo, [do′ (Hanako, [eat′ (Hanako, piza)])])61

Despite the superficial similarities of the two constructions, their CSs show clearly how different they are. Both constructions appear to be marked by the same morpheme, -(r)are, but in fact -(r)are has different functions in the two constructions. The plain passive involves a relinking of the actor and undergoer in (122a), with the undergoer appearing as the PSA and the actor as an optional adjunct in the core-level periphery for phrasal adjuncts as in (122b); the LS of the verb is not affected, as both (122a) and (b) have the same LS, given in (c). In the adversative passive, on the other hand, the addition of -(r)are changes the semantic representation of the core by adding a predicate to the LS of the verb (feel.affected′ (x, [. . .]) [MR Ø]) and licensing an additional argument, and therefore it is a valenceincreasing construction. This is highly unusual for a voice construction and is more akin to what is found in an applicative construction. There are no ‘active’ forms with the same arguments as the adversative passive, and this raises the question of whether this construction should even be labelled a ‘passive’ construction. Another apparent similarity between the two constructions is that the plain voice PSA appears to be marked by the same morpheme, ni, but in fact in (122b) ni is a postposition and can have the alternative form niyotte, whereas in (123)–(125) the ni is the dative case marker and niyotte is not possible in this construction. Adversative -(r)are can occur with both transitive and intransitive verbs, and with intransitive verbs which have an undergoer argument, as in (123), as well as to those which take an actor, as in (124). The plain passive, on the other hand, is restricted to transitive verbs. Macrorole assignment in the adversative forms is different from that in (122). The highest-ranking argument of the base verb is not assigned a macrorole but rather is a non-macrorole argument; this predicts that it should receive dative case, which is correct. The macrorole status of the argument added by -(r)are is an interesting issue. To begin with, Japanese allows non-macrorole arguments to be the PSA of some constructions, for instance as the antecedent (controller) of a reflexive (Kuno 1973), and there are cases in which a non-macrorole can appear as the nominative PSA in a passive (Imai 1998); moreover, since it is the leftmost argument in the LS, regardless of its macrorole status, it will be the PSA according to the PSA selection principles in (111b1, c2), which is specified in the CS. Thus there is no reason to assign it a macrorole to account for its selection as the PSA, hence the ‘[MR Ø]’ notation in Figure 1.40 (Toratani 2002); this applies only

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

to the argument licensed by -(r)are, since, with a transitive verb like taberu ‘eat’ in (125), the accusative RP is an undergoer.62 Because the non-macrorole argument introduced by -(r)are is the PSA, it does not take dative case, following the more complex set of case assignment rules proposed for Japanese in Imai (1998) instead of the ones given in (114a). The steps in which the CS in Figure 1.39 plays a role in the linking from semantics to syntax for (122b) are summarized in (126); the steps were given in (108). (126)

Linking in (122b) a. Step 1: LS formation. Not involved. b. Step 2: Macrorole assignment. Not involved. c. Step 3: Morphosyntactic coding. Involved in PSA modulation, argumentmodulation (if it involves omission of actor, then the actor argument in the LS will be blocked from occurring overtly, or if the realization of the actor is as a core-level adjunct, it is marked by P ni(yotte)), adds suffix -(r)are to verb stem. d. Step 4: Syntactic template selection. Choices affected by argumentmodulation in step 3. e. Step 5: Assignment of elements to syntactic trees. Actor PP assigned to core-level periphery for phrasal adjuncts, undergoer RP normally in initial position.

The situation is very different in (123)–(125). (127)

Linking in (123)–(125) a. Step 1: LS formation. Embed the LS for the verb into the second argument position in the LS for -(r)are. b. Step 2: Macrorole assignment. The argument of -(r)are and the highestranking argument of the embedded LS are not assigned a macrorole. c. Step 3: Morphosyntactic coding. Involved in case assignment. d. Step 4: Syntactic template selection. (117a) is based on derived LS. e. Step 5: Assignment of elements to syntactic trees. Dative NMR is assigned to the core, the argument of -(r)are is in initial position.

As noted in Section 1.6.1, steps 1 and 2 take place in the lexicon, while steps 3–5 take place in the syntax, outside of the lexicon. In the plain passive in (122), both active and passive versions of the sentence have the same LS, given in (122c), and accordingly it does not involve steps 1 and 2. By contrast, steps 1 and 2 are crucial for the adversative construction, as the combining of the LSs takes place at step 1 and macrorole assignment (or the lack thereof ) at step 2. The remaining three steps in (127) reflect the consequences of steps 1 and 2. In (126), on the other hand, all of the relevant steps are outside of the lexicon, namely, steps 3 through 5. The adversative construction is, therefore, considered to be a lexical construction, while the plain passive is a syntactic construction. This distinction has important consequences for certain phenomena, for example reflexive binding (see Van Valin 2005, §5.2).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

133

134

ROBERT D. VAN VALIN, JR.

1.6.4.2 Linking from Syntax to Semantics Revisited The algorithm for linking from syntax to semantics in simple sentences given in Van Valin (2005: 149–150) includes both cross-linguistically valid principles and language-specific features, and it has been argued that language-specific information should be represented in CSs rather than in the linking algorithm directly. Accordingly, there must be schemas representing the basic clause structure of a simple sentence for a language, as well as for the voice oppositions, if any, that a language has. Preliminary schemas for basic clause structure, displaced wh-questions, and peripheral adjuncts in English are given in Figures 1.41 to 1.43. The CS in Figure 1.41 summarizes the basic facts about a simple sentence with a single core containing a single nucleus. It does not include the PrDP or PrCS, since they are not part of a basic simple clause. The CS in Figure 1.42 highlights the striking contrast between the simplicity of linking from semantics to syntax in displaced wh-questions in English and the complexity of linking from syntax to semantics. Regardless of whether the wh-expression is a core argument or a peripheral adjunct, the linking from semantics to syntax is straightforward: the wh-expression occurs in the PrCS. The CS in Figure 1.43 is concerned with peripheral adjuncts, and there are two types. The first includes one-place adverbial modifiers, such as tomorrow in tomorrow′ ([do′ (Sam, [visit′ (Sam, Meredith)])]), Sam visits Meredith tomorrow, where the LS of the core is the single argument of the adverbial predicate. The second type is predicative PPs, in which the object of the predicative P is the first argument and the LS of the core is the second, as in Sam visited

CONSTRUCTION: English basic clause structure

SYNTAX: PSA: (111b1, c2) Linking: Passive construction (optional), variable undergoer selection (default) Template(s): (117b1,2); RP, PP templates, see CSs for RP, PP Order: [S [CL [C RPPSA {AUX …}[N PRED ] (RPU) (RP/PP)] Adjunct(s)]] MORPHOLOGY: Auxiliary: [±past] PSA agreement; [-tense] bare infinitive, participle[±past] Verb: [±past] PSA agreement; [-tense] bare infinitive, participle[±past] RP Case: [+PRO] NOM, ACC, GEN; [-PRO] GEN SEMANTICS: LS of PRED in nucleus plus arguments and adjuncts PRAGMATICS: Illocutionary force: Unspecified Focus structure: Predicate focus (default) Figure 1.41 Constructional schema for basic clause structure in English

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

CONSTRUCTION: English displaced wh-questions SYNTAX: PSA: none Linking [semantics to syntax]: WH-XP in PrCS [syntax to semantics]: Arguments: After all of the XPs in the core have been linked, link the WH-XP to the remaining unlinked argument position in the semantic representation of the core. Adjuncts: If there are no unlinked argument positions in the semantic representation of the core, treat the WH-word like a predicative preposition and follow the CS for linking adjuncts Template(s): (117b3), PrCS MORPHOLOGY: X[+tense] must be the first element in the core linearly. SEMANTICS: Contains an open proposition with variable α , WH-XP = α. PRAGMATICS: Illocutionary force: Interrogative Focus structure: Narrow focus on PrCS Figure 1.42 Constructional schema for English displaced wh-questions

Meredith in the hospital, be-in′ (hospital, [do′ (Sam, [visit′ (Sam, Meredith)])]), or Sam visited Meredith after the party, be-after′ (party, [do′ (Sam, [visit′ (Sam, Meredith)])]). In both examples the first argument is the spatial or temporal reference point and the second argument is the event or state of affairs being located in space or time. While they all have a default periphery in which they occur, they are also subject to displacement and occurrence in the PrCS and PrDP. In addition, there can be multiple adjuncts, and their ‘scope’-like relations are constrained by both semantic and pragmatic factors. Many of the aspects of the CS in Figure 1.43 would be applicable to other languages; some would require only a superficial reformulation in terms of word order, for example a language which has postpositions rather than prepositions, but some would be quite substantial, for example languages which lack a PrCS and/or a PrDP, and, more significantly, languages such as Dyirbal that lack adpositions altogether. In languages with rich case systems, very often it is cases which mark adjuncts rather than adpositions (e.g. Hungarian), and so it is necessary to recognize that there can be predicative cases as well as predicative adpositions (see Nakamura 2021 for a discussion of predicative cases in Russian). It is now possible to formulate the syntax-to-semantics linking algorithm in such a way that it contains only cross-linguistically valid steps, the language-specific information being contained in a language’s CSs. The goal here is to redo the linking in Section 1.6.2 based on information supplied by the CSs in Figures 1.41–1.43.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

135

136

ROBERT D. VAN VALIN, JR.

CONSTRUCTION: English peripheral adjuncts SYNTAX: PSA: none Linking [semantics to syntax]: Assign the argument of an LS other than that of the predicator in the nucleus to a predicative PP headed by the predicative P, or identify a one-place adverbial predicate, and instantiate them in 1. a periphery [(81i) = core, (81j) = clause] (default), or 2. the PrCS or the PrDP, subject to information-structural conditions (see PRAGMATICS). [syntax to semantics]: If there are no unlinked argument positions in the semantic representation of the clause, retrieve the LS for the predicative preposition or he adverbial from the lexicon; if the predicative element is a one-place adverb, insert the LS of the core into its argument position, and if it is a two-place preposition, insert the LS of the core into the second argument position and insert the object of the P as the first argument. Template(s): periphery, predicative PP, adverb, PrDP, PrCS MORPHOLOGY: none SEMANTICS: Semantic relations: (81i, j)/(94i, j) When there are multiple adjuncts, the order of linking reflects (1) clausal periphery > core periphery, (2) PrDP > PrCS > default periphery, or (3) [PER AJTTOP > AJTFOC ], where the lower ranking adjunct is linked first. PRAGMATICS: Illocutionary force: Unspecified Focus structure: Frame-setting topic in PrDP Contrastive focus/topic in PrCS Neutral comment, narrow focus in periphery Figure 1.43 Constructional schema for English peripheral adjuncts

(128)

Linking algorithm: syntax to semantics Step 1: Determine the syntactic and semantic functions of the core arguments, based on the morphosyntactic properties of the clause. See the CS for the language’s basic clause structure and the CSs invoked by it. Step 2: Retrieve from the lexicon the LS of the predicator in the nucleus and execute the step in (108b), subject to the following proviso: If a language has verbs which allow variable undergoer assignment, and the core contains more than one choice for undergoer, do not assign undergoer to an argument in the LS. Step 3: Link the arguments identified in step 1 with those identified in step 2. Step 4: For variable undergoer-selection verbs, determine the linking of the non-macrorole core argument: a. If there is a two-place state predicate in the LS and if the non-macrorole core argument is marked by a locative adposition or dative or a locativetype case, then link it with the first argument position in the state predicate in the LS and link the other non-actor core argument (if there is one) to the second argument position in the state predicate, or

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar b. If there is a two-place state predicate in the LS and if the non-macrorole core argument is not marked by a locative adposition or dative or a locative-type case, then link it with the second argument position in the state predicate and link the other non-actor core argument (if there is one) to the first argument position in the state predicate. c. If neither (a) nor (b) applies, see the relevant CS for the language. Step 5: If there is a wh-XP in the Pr/PoCS, see the relevant CS for the language. Step 6: If there is a peripheral adjunct, see the relevant CS for the language.

It might be objected that step 4 contains considerable language-specific information, but in fact it is surprisingly valid across languages, having been applied to Croatian, Dyirbal, English, German, Russian, and many others. The detailed linking from syntax to semantics for After the meeting, what did Linda show to the salesman in the office? begins with the parser output, namely, in Figure 1.44. The output of step 1 is given in Figure 1.45. It makes crucial reference to the CS ‘Basic clause structure’ in Figure 1.41, which states that the PSA is accusative and not restricted to macroroles and that it is the core-initial RP, that the default position of the undergoer in an M-transitive core is immediately after the nucleus, and that there is the possibility of a passive

Figure 1.44 Parser output in syntax-to-semantics linking

Figure 1.45 Output of step 1 in the syntax-to-semantics linking

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

137

138

ROBERT D. VAN VALIN, JR.

Figure 1.46 Output of step 2 in the syntax-to-semantics linking

construction, thereby potentially invoking the passive CS.63 However, because the voice of the verb is active, the PSA is an actor, and according to the CS in Figure 1.41, the RP Linda as the first RP in the core, must be the actor. There is no direct argument immediately following the nucleus, hence no undergoer in the core, but there is a PP headed by to with the object the salesman and two adjunct PPs. Step 2 involves calling up the LS of the predicator in the nucleus, in this case [do′ (x, Ø)] CAUSE [INGR see′ (y, z)] for show and assign macroroles as in the comparable step in (108b) in the semantics-to-syntax linking (Figure 1.46). Accordingly, the x-argument is the actor, but because show is a variable undergoer verb, it is not possible to assign undergoer. Step 3 is rather trivial in this case, despite the complexity of the example: in step 1 Linda is identified as the actor, and in step 2 the x-argument in the LS is assigned the actor macrorole; ergo, the RP Linda in the syntactic representation links to the x-argument in the LS of show. Step 4 deals with the problem of determining the linking of the non-actor arguments when there is more than one candidate for undergoer. Some languages mark the choice of undergoer overtly, for instance German schenken ‘to give as a gift’, has the theme (z) as the undergoer, as in (129a), whereas beschenken ‘to gift someone’, has the recipient (y) as the undergoer, as in (129b). (129)

a. Der Mann hatder Frau Blumen geschenkt. the.nom.sg man has the.dat.sg woman flower.acc.pl given.as.a.gift.’ ‘The man gave flowers to the woman as a gift.’ b. Der Mann hat die Frau mit Blumen beschenkt. the.nom.sg man has the.acc.sg woman with flowers.dat.pl gifted. ‘The man gifted the woman with flowers.’

Among its many functions the verb prefix be- serves to signal a non-default choice for the argument to function as undergoer. Most languages which have variable undergoer selection are not so straightforward. Rather, the identity of the argument functioning as undergoer must be deduced from the way the non-macrorole argument is coded. The following English examples illustrate this. (130)

a. a′. b. b′. c.

Mary gave the award to Bill. Mary gave Bill the award. Mary presented the award to Bill. Mary presented Bill with the award. [do′ (x, Ø)] CAUSE [INGR have′ (y, z)]

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

In (130a) the non-macrorole argument is Bill, an oblique core argument, and therefore step 4 applies to the examples in (130). In (130a) the third argument is marked by to, which is a locative preposition, and according to Step 4a this means that the object of to should be linked to the first argument position (y) in the two-place state predicate in the LS in (130c), and consequently the other argument, the award, must be linked to the zargument in the LS. In (130a′), on the other hand, the third argument, the award, is not marked by a locative adposition or dative-type case, and therefore its third argument must be linked to the z-argument, leaving the recipient to be linked to the y-argument. The same is true in (130b, b′), which differ from (a, a′) only in the third argument being an oblique core argument. In (130b′) the third argument is marked by with, which is not a locative adposition, and therefore the object of with is linked to the zargument in the LS in (130c). The outputs of Steps 3 and 4 are given in Figure 1.47. There remain three unlinked elements in the syntactic representation, two predicative PP adjuncts and a wh-expression. They are handled by Steps 5 (wh-questions) and 6 (adjuncts), each of which refers to a languagespecific CS. Step 5 must be executed in order to link the wh-RP in the PrCS to the semantic representation. There is only one unlinked argument position in the LS of the core, the z-argument, and accordingly the RP what is linked to the z-argument in the LS of show, and this is the correct linking, because what refers to the content that is shown. Because there are two adjuncts, they must be composed in a specific order reflecting their relative ‘scope’. After the meeting occurs in the PrDP and functions as a frame-setting topic, while in the office is in its default position in the core-level periphery with no special information-structural function.

Figure 1.47 Output of steps 3 and 4 in the syntax-to-semantics linking

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

139

140

ROBERT D. VAN VALIN, JR.

Figure 1.48 The linking from syntax to semantics in terms of the algorithm in (128)

Hence following the CS in Figure 1.43, in the office will be added to the semantic representation first, followed by after the meeting. Thus the entire linking is summarized in Figure 1.48, which labels each step of the detailed linking algorithm in (128) and the relevant CSs invoked in each. Step 4c in (128) refers to instances to which Steps 4a, b do not apply. Two examples of this from Icelandic (Van Valin 1991) and Yaqui (Uto-Aztecan; Guerrero and Van Valin 2004) are given in (131) and (132). (131)

a. Ég skila-ð-i henni pening-un-um. 1sg.nom return-pst-1sg 3f.sg.dat money-def-dat ‘I returned her the money.’ b. Henni va-r skil-að pening-un-um. 3f.sg.dat be.pst-imper return-ptcp money-def-dat ‘She was returned the money.’ b′. *Pening-un-um va-r skil-að henni. money-def-dat be.pst-imper return-ptcp 3f.sg.dat ‘The money was returned to her.’ [intended meaning] c. Ég skila-ð-i pening-un-um til hennar. 1sg.nom return-pst-1sg money-def-dat to 3f.sg.gen ‘I returned the money to her.’ c′. Pening-un-um va-r skil-að til hennar. money-def-dat be.pst-imper return-ptcp to 3f.sg.gen ‘The money was returned to her.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Icelandic

Principles of Role and Reference Grammar

(132)

a. Joan-Ø Peo-ta Ɂuka vaci-ta miika-k. Juan-nom Pedro-acc det.acc corn-acc give-prf ‘Juan gave Pedro the corn.’ b. Peo-Ø Ɂuka vaci-ta miika-wa-k. Pedro-nom det.acc corn-acc give-pass-prf ‘Pedro was given the corn.’ c. *UɁu vaci-Ø Peo-ta miika-wa-k. det.nom corn-nom Pedro-acc give-pass-prf ‘The corn was given [to] Pedro.’ [intended meaning]

Yaqui

The issue here is clear: both non-actor arguments are in the same case in the active voice, and therefore the principles in Steps 4a, b fail to distinguish between the two arguments in the linking. The Icelandic verb skila ‘give back, return’ is a three-place predicate whose non-actor arguments exhibit two different patterns: two dative RPs (recipient, theme), as in (131a), and a dative RP (theme) and a PP (recipient), as in (131c). The latter follows the rule in Step 4a, since linking the object of the preposition til ‘to’ to the y-argument in the LS yields the correct result. This requires that the preposition til ‘to’ outweigh the dative case on the theme argument in Step 4a, and consequently this language-specific feature must be expressed in a CS concerning linking with three-place predicates. The passive in (131c′) illustrates the fact that the PSA in Icelandic does not have to be a macrorole; peningunum ‘the money’ has the syntactic privileges of a nominative PSA, despite not having the coding privileges of one (i.e. nominative case, ability to be an agreement trigger) (see Van Valin 1991). The pattern with two dative RPs in (131a) renders Steps 4a, b inapplicable, since the two RPs bear the same case. There is, nevertheless, an important difference between the two direct core arguments: one is human and the other is inanimate, and this difference correlates with the way the RPs link to the LS. The human RP is the recipient (y), while the inanimate RP is the theme (z). This can be captured in the following step in the Icelandic CS for three-place predicates. (133) Animacy Principle: When the two non-actor RPs in the core are both in the same case, link the human (animate) RP with the first argument position in the two-place state predicate in the LS.

This linking rule comes into play in Icelandic only when there are two dative RP objects in the core; otherwise steps 4a, b in (128) are operative. A comparison of the Yaqui examples in (132) with the Icelandic examples in (131a–c) suggests that a principle like (133) is part of the Yaqui linking algorithm as well, with double-accusative RP objects instead of two dative RPs. The Yaqui situation is more complex, however, due to sentences involving valence-increasing constructions, such as derived causative verbs, as illustrated in (134).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

141

142

ROBERT D. VAN VALIN, JR.

(134)

a. UɁu maejto usi-ta mansana-ta yoem-ta miik-tua-k. det.nom teacher child-acc apple-acc man-acc give-caus-prf ‘The teacher made the child give the man the apple.’ a′. UɁu usi-Ø mansana-ta yoem-ta miik-tua-wa-k. det.nom child-nom apple-acc man-acc give-caus-pass-prf ‘The child was made to give the man the apple.’ a′′. [do′ (maejto, Ø)] CAUSE [[do′ (usi, Ø)] CAUSE [INGR have′ (yoem, mansana)]] b. UɁu maejto yoem-ta mansana-ta usi-ta miik-tua-k. det.nom teacher man-acc apple-acc child-acc give-caus-prf ‘The teacher made the man give the child the apple.’ b′. UɁu yoem-Ø mansana-ta usi-ta miik-tua-wa-k. det.nom man-nom apple-acc child-acc give-caus-pass-prf ‘The man was made to give the child the apple.’ b′′. [do′ (maejto, Ø)] CAUSE [[do′ (yoem, Ø)] CAUSE [INGR have′ (usi, mansana)]]

These examples have three core arguments in the accusative case, two of which are human, and only one can be interpreted as the causee and only one as the recipient. The sentences in (134a, b) are unambiguous: in (a) usi‘the child’ can only be interpreted as the causee and yoem- ‘the man’ as the recipient, and vice versa in (b). Accordingly, the correct linking in this construction cannot refer simply to case or to animacy. It was noted in fn. 52 that Yaqui verbs show three different undergoer linking patterns, termed patterns A, B and C, following Guerrero and Van Valin (2004). Pattern A is the usual direct–indirect object pattern, exemplified by nenka ‘sell’ in (135a), in which the lowest-ranking argument in the AUH is the undergoer, is in the accusative case, and can be the PSA in a passive construction, as in (135a′). Pattern B is the primary object pattern (Dryer 1986), illustrated by miika ‘give’ in (132), in which the undergoer is the second highest-ranking argument in the AUH, all nonactor core arguments are in the accusative case, the undergoer normally immediately follows the nominative actor, and it alone can be the PSA in a passive. Pattern B not only applies to miika ‘give’ and the other verbs in this class, but also to derived forms involving valence-increasing constructions, namely causative, applicative, desiderative, reported speech and propositional attitude. (See Guerrero and Van Valin 2004 for exemplification.) Pattern C is defined by variable undergoer selection, alternating between Patterns A and B, as shown in (135b, c) with chijakta ‘sprinkle’. (135)

a. Joan-Ø Peo-ta-u ju-ka kaba’i-ta nenka-k. Juan-nom Pedro-acc-dir det-acc horse-acc sell-prf ‘Juan sold the horse to Pedro.’ a′. Ju-Ø kaba’i-Ø Peo-ta-u nanka-wa-k. det-nom horse-nom Pedro-acc-dir sell-pass-prf ‘The horse was sold to Pedro.’ b. Empo kafe-ta mesa-po chijakta-k 2sg.nom coffee-acc table-loc sprinkle-prf ‘You sprinkled coffee on the table.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar b′. Kafe-Ø mesa-po chijakta-wa-k coffee-nom table-loc sprinkle-pass-prf ‘Coffee was sprinkled on the table.’ c. Empo kafe-ae mesa-ta chijakta-k 2sg.nom coffee-ins table-acc sprinkle-prf ‘You sprinkled the table with coffee.’ c′. Mesa-Ø kafe-ae chijakta-wa-k table-nom coffee-ins sprinkle-pass-prf ‘The table was sprinkled with coffee.’

Actor and undergoer selection are best handled in terms of a CS, since undergoer selection has several language-specific facets. In terms of semantics-to-syntax linking, step 2 in (108b) is the relevant step, and pattern A follows the default pattern of the AUH. Pattern B, on the other hand, does not and specifies that the undergoer is the second-highest-ranking argument in the LS. In a two-argument LS (e.g. jamta ‘break’ [do′ (x, [act.on′ (x, y)])] CAUSE [INGR broken′ (y)]), ‘lowest ranking’ and ‘second highest raking’ pick out the same argument, namely y, but in a three-argument LS like miika ‘give’ ([do′ (x, Ø)] CAUSE [INGR have′ (y, z)]) they select different arguments as undergoer, z in pattern A and y in pattern B. The verbs which exhibit just pattern A or B do not alternate with respect to undergoer selection, and they each occur with different case marking. Pattern A verbs like nenka ‘sell’ in (135a) have an accusative undergoer/theme, (z), while the recipient (y) carries the postposition -u, making it an oblique core argument. Pattern B verbs like miika ‘give’ take all of their non-actor direct core arguments in the accusative case. The undergoer is distinguished by word order: it normally immediately follows the nominative actor, and if it is not immediately after the NOM RP, it is the first one after it satisfying the selectional restrictions of the verb. Pattern C verbs like chijakta ‘sprinkle’ show the same case alternation found in many other languages, namely ACC (U) – LOC (locative oblique core argument) vs. ACC (U) – INS (instrumental oblique core argument), as in (135b, c). In all three patterns the undergoer appears in the accusative case; it is the case marking of the other core arguments that varies across the three classes of verbs. The Animacy Principle was introduced in (133) to account for the Icelandic examples in (131), and based on the superficial similarity between the sentences in (132) and (131) it was suggested that it might be operative in Yaqui, too. It could be argued that it applies to (132a), for example, but this is also readily accounted for by Step 4 in (128). More problematic for the Animacy Principle are the sentences in (136), from Guerrero and Van Valin (2004). (136)

a. Aurelia-Ø Karmen-ta u-ka ili usi-ta bit-tua-k. Aurelia-nom Carmen-acc det-acc little child-acc see-caus-prf ‘Aurelia showed Carmen the child’, or ‘Aurelia showed the child Carmen.’ b. Tibu-Ø u-ka wikoi-ta juiwa-m64 u’ura-k. Tibu-nom det-aka rifle-acc bullets-pl take.away-prf ‘Tibu emptied the rifle of bullets.’

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

143

144

ROBERT D. VAN VALIN, JR.

In (136a) there are three human RPs, and the sentence is actually ambiguous. It could be argued that the Animacy Principle correctly predicts the ambiguity of (136a), since it makes it possible for either the RP Karmen or the RP usi ‘child’ to be interpreted as the first argument in the two-place state predicate in the LS. With respect to four-place derived predicators, as in (134), they are not ambiguous, and this can be accounted for by having the Animacy Principle apply after the actor and undergoer have been identified. Accordingly, in (134a), after maejto ‘teacher’ is determined to be the actor and usi- ‘child’ the undergoer, the Animacy Principle can apply to the remaining two core arguments, and the result is that yoem- ‘person’ is correctly analysed as the recipient/experiencer, and mansana ‘apple’ is correctly analysed as the thing given (theme) in the central LS of the events described in (134). The sentence in (136b), however, cannot be accounted for by (133), because all of the non-NOM RPs are inanimate. Hence (133) does not apply, and this is also the case when the sentence is causativized, as in (137). (137)

Goyo-Ø Tibu-ta u-ka wikoi-ta juiwa-m u’ura-tua-k. Goyo-nom Tibu-acc det-aka rifle-acc bullets-pl take.away-caus-prf ‘Goyo made Tibu empty the rifle of bullets.’

After determining that Goyo is the causer, actor and PSA and Tibu is the causee and undergoer, there are still two inanimate arguments left to be accounted for. All verbs have selectional restrictions constraining the type of RPs that can function as their arguments, and in RRG they are expressed as qualia properties (Pustejovsky 1995), both of the verb and of the RPs, such that the qualia properties of the RPs must be compatible with the selectional restrictions of the verb’s arguments. This can be formulated as the ‘Qualia Principle’ as in (138). (138)

Qualia Principle: In order for an RP to function as an argument of a predicator, its qualia properties must be compatible with the selection restrictions of that argument the predicator.

In (136b) and (137), the crucial part of the LS is ‘CAUSE [INGR NOT be-LOC ′ (y, z)]’, where y is the ‘container’ and z are the things in it. With respect to rifles and bullets, it is clear that rifles can have bullets in them, not the other way around, and therefore wikoi- ‘rifle’ is the y-argument and juiwa- ‘bullet’ is the z-argument. Thus the interpretation of (136b) and (137) derives from the very specific qualia properties of these RPs. By contrast, invoking the Animacy Principle in (133) to account for (131a) involves only very general qualia properties, namely animacy. Thus, the Animacy Principle can be subsumed under the Qualia Principle. The difference in complexity between what is required for the semanticsto-syntax linking and the syntax-to-semantics linking is striking. The

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

former is summarized in (139) and the latter in (140). The steps referred to in (139) are those in (108), whereas the steps in (140) are those found in (128). With respect to (139), it is important to keep in mind that patterns A and B do not involve variable undergoer selection; that is solely a property of pattern C. (139)

Semantics-to-syntax linking for the three verb patterns A. Pattern A 1. Step 2: AUH 2. Step 3: Default case assignment rules (DAT is realized by a postposition.) B. Pattern B 1. Step 2: U is second-highest-ranking argument in LS 2. Step 3: U has ACC case, all non-actor direct core arguments carry ACC case. C. Pattern C: 1. Step 2: AUH default, U can be second-highest-ranking argument in LS. 2. Step 3: U is ACC, third core argument is oblique. D. Derived pattern B verbs (e.g. causatives) 1. Step 2: U is second-highest-ranking argument in LS, regardless of base pattern 2. Step 3: U bears ACC, all non-actor direct core arguments carry ACC case, regardless of base pattern.

(140)

Syntax-to-semantics linking for the three verb patterns A. Pattern A 1. Step 1: If active voice, NOM PSA is actor, ACC RP is undergoer, oblique core argument marked by P. If passive voice, NOM PSA is undergoer, actor is omitted. 2. Step 2: In LS assign actor and undergoer (lowest ranking argument). If step 1 finds passive voice, replace highest ranking argument in LS with Ø. 3. Step 3: Link arguments identified in steps 1 and 2. B. Pattern B 1. Step 1: If active voice, NOM PSA is actor, first ACC RP following NOM RP which satisfies the selectional restrictions of the predicator is U. If passive voice, NOM PSA is undergoer, actor is omitted. 2. Step 2: In LS assign actor and undergoer (second-highest ranking argument). If step 1 finds passive voice, replace highest ranking argument in LS with Ø. 3. Step 3: Link arguments identified in steps 1 and 2. C. Pattern C 1. Step 1: If active voice, NOM PSA is actor, ACC RP is undergoer, oblique core argument marked by P. If passive voice, NOM PSA is undergoer, actor is omitted. 2. Step 2: In LS assign actor only. If step 1 finds passive voice, replace highest ranking argument in LS with Ø. 3. Step 3: Link arguments identified in steps 1 and 2. 4. Step 4: Determine the linking of the oblique (non-macrorole) core argument following Step 4 in (128) .

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

145

146

ROBERT D. VAN VALIN, JR.

D. Derived pattern B verbs (e.g. causatives) 1. Step 1: If active voice, NOM PSA is actor, first ACC RP following NOM RP which satisfies the selectional restrictions of the predicator is U. If passive voice, NOM PSA is undergoer, actor is omitted. 2. Step 2: In LS assign actor and undergoer (second highest ranking argument). If step 1 finds passive voice, replace highest ranking argument in LS with Ø. 3. Step 3: Link arguments identified in steps 1 and 2. 4. Step 4: Determine the linking of the remaining arguments, if there are any. a. Base verb is pattern A: follow Step 4 in (128) b. Base verb is pattern B: follow the Qualia Principle in (138) . c. Base verb is pattern C: follow Step 4 in (128)

The point of this discussion of Icelandic and Yaqui is to show what kinds of principles, both language-specific and cross-linguistically valid, are needed. The Qualia Principle is a very general principle governing the relationship between the core arguments and the predicator in verbal cores, whereas the specification that all non-PSA direct core arguments with pattern B verbs must have accusative case is clearly idiosyncratic to Yaqui. The information in (138) and (139) would be stored in CSs, under the heading of ‘Linking’ in the syntax section of the CSs.

1.6.4.3 Linking from Semantics to Syntax Revisited Again Step 5 in the semantics-to-syntax linking algorithm ranged from extremely vague (‘assign [. . .] to the appropriate positions in the . . .’) to very specific (e.g. ‘three options for [þWH] CPs’). It is repeated below. (118)

Step 5. Assign the nucleus, arguments and adjuncts to positions in the syntactic representation of the sentence. a. Assign the nucleus to the appropriate position in the core. b. Assign the [WH] arguments to the appropriate positions in the clause. c. If there is a [þWH] XP, 1. assign it to the normal position of a non-wh-XP with the same function, or 2. assign it to the pre-core or post-core slot, or 3. assign it to a position within the potential focus domain of the clause (default ¼ the unmarked focus position). d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to information structure restrictions (optional). e. Assign the argument(s) of LS(s) other than that of the predicator in the nucleus to 1. the core- or clause level periphery (default), or 2. the pre-core or post-core slot, or 3. the pre-detached or post-detached position.

Rather than specifying the range of cross-linguistic possibilities, it is preferable to customize the linking algorithm for the language under investigation, and this can be done by replacing the disjunctive offerings

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

of (118) with language-specific information from CSs. It may be reformulated as follows. (141)

Step 5: Assign the nucleus, arguments and adjuncts to positions in the syntactic representation of the sentence. a. See CS for English basic clause structure [Figure 1.41] (¼a, b) b. See CS for English displaced wh-questions [Figure 1.42] (¼c) c. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to information structure restrictions (optional). [a yet to be formulated CS] d. See CS for English peripheral adjuncts [Figure 1.43] (¼e)

This better represents what English speakers know about their language and yet still maintains a balance between cross-linguistically valid linguistic concepts and principles and the language-specific data that speakers rely on when they actually use language.

1.6.5 Semantic Motivation and Cross-Linguistic Variation The semantics-to-syntax linking algorithm in (108) has two phases: steps 1 and 2 take place in the lexicon, while steps 3 through 5 take place outside of the lexicon in the syntax. There is a striking asymmetry between steps 1 and 2, which will be called ‘the lexical phase’ of the linking, and steps 3 through 5, which will be called the ‘morphosyntactic phase’ with respect to the amount of cross-linguistic variation each exhibits. In a nutshell, there is much greater cross-linguistic variation in the morphosyntactic phase of the linking than in the lexical phase, and from this fact comes the claim that the more semantically motivated a linguistic phenomenon is, the less cross-linguistic variation there is, and vice versa. This can be illustrated for each step. Step 1 concerns the decomposition of predicating elements, and all languages make the same fundamental distinctions. Where there is variation, it concerns the analysis of specific verbs, not variation in terms of the Aktionsart categories that mould the system. Languages may vary, for example, as to whether causative change-of-state predicates are basic or derived from change-of-state predicates via causativization (Nichols et al. 2004), but regardless of the direction of derivation, the categories involved are the same. Step 2 concerns actor and undergoer assignment. There is more variation than in step 1, and the main variation is as follows. With respect to actor selection, there are languages in which the actors of transitive verbs must be animate (e.g. Lakhota, Japanese), in which the animate actors of transitive verbs must be interpreted as agentive (e.g. Japanese), and in which the interpretation of whether an actor is agentive is an implicature for most verbs, not a lexical property (e.g. Tsova-Tush, English, German). There is more variation with respect to undergoer selection, and this follows from the different semantics of actor and undergoer: as noted in Section 1.4.3.2,

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

147

148

ROBERT D. VAN VALIN, JR.

actors are responsible for the state of affairs, in the sense that in order to have a perceptual event there must be a perceiver, for a cognition state of affairs a cognizer, for an eating event an eater, etc., while undergoer is associated with affectedness. It appears that languages tend strongly to lexicalize responsibility but not affectedness, leaving the choice of which participant is most affected to the speaker in some languages. Languages vary with respect to whether they allow variable undergoer selection or not. If they do, there is variation in the number of verbs allowing the variable selection, and there are differences in terms of the definition of undergoer in terms of the AUH. As shown earlier, Yaqui has a class of verbs for which the undergoer is the lowest-ranking argument in the LS of verbs and another class in which it is the second-highest-ranking argument. Step 3 involves PSA selection and case marking, and here the crosslinguistic variation dwarfs that found in steps 1 and 2. The variation in PSA selection includes syntactically accusative languages without a voice system, syntactically accusative languages with a productive voice system, syntactically ergative languages with a productive voice system, mixed systems with first and second person being syntactically accusative and third person being syntactically ergative, inverse systems, split-intransitive systems, Philippine-type languages with multiple voice systems, which can be syntactically accusative, ergative or symmetrical, and systems which defy categorization in terms of traditional categories (e.g. Barai, PapuaNew Guinea; Van Valin 2009). Case-marking systems are likewise highly varied, and they don’t always match the syntactic pattern, for instance it has long been known that most morphologically ergative languages lack ergative syntax. Step 4 is concerned with the selection of the syntactic templates that constitute the syntactic representation of the sentence, and here too there is great diversity, given that the templates must embody the syntactic structures of the full range of human languages. The components of the LSC fall into two basic groups: those that are semantically motivated (see Section 1.2.2) and are universal (nucleus, core, clause, peripheries) and those that are not semantically motivated but rather are pragmatically motivated and are not universal (PrDP, PrCS, PoCS, PoDP). Step 5, as shown earlier, incorporates language-specific CSs capturing the full sweep of linearization options, displacement options, placement of adjuncts, etc. Thus, the post-lexical steps in the linking algorithm exhibit much greater cross-linguistic variation than the lexical steps, and this is indicative of the inverse relationship between semantic motivation and cross-linguistic variation. Two further examples will be presented to illustrate this point. The first concerns the LSC. There are obligatory and optional constituents in it. The obligatory ones are the nucleus, the core, the clause and the sentence, and within the core are the arguments of the predicator in the nucleus. As argued in Section 1.2.2, the fundamental distinctions in the LSC follow from

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

the nature of human language as a system of communication. Thus the obligatory aspects of the LSC are semantically motivated. The optional aspects, the Pr/PoDP and the Pr/PoCS, which are not universal, are not semantically motivated; rather, they are motivated by discoursepragmatic considerations. The second example is reflexive binding in languages with coreference reflexives, such as English, German, Icelandic, Japanese, Mandarin. There are two fundamental features of these constructions that must be accounted for: (1) the hierarchical relationship between the antecedent and the reflexive, and (2) the distance between them. The second issue is uncontroversially syntactic, whereas both syntactic and semantic accounts of the hierarchical relationship have been proposed. The RRG approach to coreference reflexives is semantic (see Van Valin and LaPolla 1997, §7.5.2, Van Valin 2005, §5.3), and the fundamental claim is that the reflexive anaphor must not be higher on the PSA Selection Hierarchy than the antecedent. In other words, oversimplifying somewhat, actors bind undergoers, not the other way around. In most cases involving coreference reflexives the syntactic c-command approaches and the semantic approach make the same prediction, but there are a few cases where they make different predictions, and it turns out that the semantic approach always makes the correct prediction (see Van Valin 2005, §5.3). There is some variation in coreference reflexives, but the basic principle, namely that a higher-ranking argument binds a lower-ranking argument on a semantic hierarchy, holds up. With respect to the issue of distance between the antecedent and the reflexive, there is substantial cross-linguistic variation, ranging from English, which does not allow binding across a core boundary (e.g. Mary asked Tom to help her/*herself), while Icelandic happily permits this, to Japanese and Mandarin in which the reflexive anaphor can be bound across clause boundaries, where the equivalent of Mary said that Tom had kissed herself would be possible in them. Here again there is substantial cross-linguistic variation with respect to the aspect of reflexivization which is not semantically motivated and very little with respect to the aspect which is. Thus, the features of human language which are most likely to be universal, or nearly so, are the ones with the strongest semantic motivation.

1.7

Linking between Syntax and Semantics in Complex Sentences

The linking algorithms for semantics-to-syntax and syntax-to-semantics linking in simple sentences were presented in Section 1.6, and the adaptation of them for complex sentences is discussed in detail in Guerrero (Chapter 14). (See also Van Valin 2005, §7.) Linking in complex RPs, primarily restrictive relative clauses, is presented in París (Chapter 15) (see also Van

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

149

150

ROBERT D. VAN VALIN, JR.

Valin 2012b). Constraints on the formation of wh-questions in complex sentences are the topic of Shimojo (Chapter 16). The main aspects of the algorithm that need modification for complex sentences will be discussed in Section 1.7.1. The final section will examine two critical areas of the complex sentence linking algorithm that have no analogue in simple sentences.

1.7.1 Modifications of the Linking Algorithm for Complex Sentences There are four levels of juncture (nuclear, core, clausal, sentential), and yet three of the four do not require substantial modifications of the linking algorithm. Nuclear junctures link like simple clauses with a multiargument predicate; there is a single core node and a single clause node, as in, for example, the Mandarin nuclear juncture in Figure 1.29. Clausal and sentential junctures, on the other hand, contain constituent clauses or sentences which link in most instances just like independent clauses or sentences. An exception regarding clausal junctures will be discussed in the following section. The major complexity in linking syntax and semantics is found in non-subordinate core junctures. Core subordination involves either a clause or a core used as a core argument (e.g. That Bill’s cat ran away shocked him vs. Bill’s cat’s running away shocked him) or a peripheral ad-subordinate clause (e.g. Bill’s cat ran away after she got bored with him). The units involved in these sentences follow the linking algorithm for simple clauses, and they would link like simple RPs into core positions or like simple adjunct PPs into the core or clause-level periphery. Non-subordinate core junctures require that the two cores have an argument in common, as illustrated in (142) and (143). (142)

(143)

a. Max tried to tune the piano. b. Sally persuaded Max to tune the piano. c. Sally promised Max to tune the piano. d. Sam brought wine for everyone to drink.

Max is the actor of try and the actor of tune. Max is the undergoer of persuade and the actor of tune. Sally is the actor of promise and of tune.

a. Sam seems to really like wine.

Sam is a syntactic core argument (and PSA) of the core headed by seem but is a semantic argument of like. Sam is a syntactic core argument of the core headed by believe but is a semantic argument of like.

b. Sally believes Sam to really like wine.

Wine is the undergoer of bring and drink.

In (142) each clause contains two cores, and there is a single RP which is a semantic argument of both predicators in the nucleus of each core. In (143), on the other hand, there is no single RP which is a semantic argument of each predicate; rather, there is an RP which is a syntactic core argument, but not a semantic argument, of one core, and a semantic

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

argument of the predicator in the linked core. Thus, there are two distinct ways for two cores to have an argument in common: the constructions in (142) involve what is traditionally known as (obligatory) control, while those in (143) are traditionally known as raising.65 The argument in common is technically known as the shared argument in the construction. In an obligatory control construction, a participant functions as a semantic argument in each core but is instantiated in only one of the cores. In sentences like (142a) there is only one core argument in the licensing core, and it must be the shared argument. In (142b, c) there are two core arguments in the licensing core, and there has been much discussion over the past almost sixty years regarding the explanation for the difference between verbs like persuade (so-called ‘object control’) and promise (so-called ‘subject control’). The solution was given in Foley and Van Valin (1984: 306–311), and it was shown in Van Valin and LaPolla (1997, §9.1.3.1) and Van Valin (2005, §7.3.1) to be valid for a typologically wide range of languages. It is semantic: if the licensing verb has causative or jussive semantics, the undergoer is always the controller, whereas if it does not have causative or jussive semantics, the actor is the controller. Because this principle is semantic, it would be expected in light of the discussion in Section 1.6.5 that it would be universally valid, and as far as is known, this is the case. This semantic account makes a prediction that syntactic accounts do not: if a verb can have either jussive or non-jussive semantics, the choice of controller should change with the semantic change in the verb. This is illustrated in (144). (144)

a. The teacher asked the rowdy student to leave the classroom. b. The sick student asked the teacher (to be permitted) to go home.

In (144a) the verb ask has jussive semantics, and the undergoer is the controller, as predicted. In (144b), on the other hand, it does not have jussive semantics, as it is a request, and the actor is the controller, as predicted. Because there is a shared argument in these constructions, the linked core has one core argument less than if it were a core in a simple sentence, and therefore the syntactic template selection principles in (117) must be modified. Unlike the language-specific principles in (117b2, 3), this one is universal. There is one additional principle affecting non-subordinate core junctures, which is illustrated here. (145)

a. a′. b. b′.

Bill taught Mary calculus. Bill taught Mary to swim. Sam forgot his keys. Sam forgot to lock the door.

[3 [2 [2 [1

core arguments] core arguments þ infinitive] core arguments] core argument þ infinitive]

Teach, for example, is a three-place predicate, and one of the arguments can be an RP, as in (145a), or a proposition, do′ (z, [swim′ (z)]), as in (145a′). In

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

151

152

ROBERT D. VAN VALIN, JR.

Section 1.3.2 it was shown that the infinitival cores in these constructions are not embedded, and therefore they are not core arguments in the licensing core. Accordingly, the core containing the licensing predicate in a nonsubordinate core juncture has one less core argument than when it is a simple core, as the contrast between (145a) and (a′) shows. The same holds for two-place predicates like forget in (145b, b′): when it heads a simple core, as in (145b), it has two core arguments, but when it is the head of the licensing core in a non-subordinate core juncture, as in (145b′), it has only one. Thus, the verb in the licensing core in a non-subordinate core juncture has one less core argument than when it is the head of a simple core with RP or PP arguments only. The LSs for (144a′, b′) are given in (146). (146)

a. [do′ (Bill, Ø)] CAUSE [BECOME know′ (Maryi, [do′ (zi, [swim′ (z)])] b. forget′ (Sami, [[do′ (yi, Ø)] CAUSE [INGR locked′ (door)])])

Both LSs contain an embedded proposition with a lexically unfilled argument, and it will be manifested as the infinitival core in the core juncture.66 This is a syntax–semantics mismatch: the LS of the linked core is embedded as an argument of the LS of the licensing core, but in the syntax the infinitive is not embedded as a core argument of the licensing core. The fact that the LS of the infinitival core is an argument of the licensing verb is significant, because when the infinitival core is not an argument of the licensing verb, then the reduction in core arguments does not occur. This is illustrated by (142d), Sam brought wine for everyone to drink. The infinitival core is missing an argument, but the LS of drink is not an argument of bring. Accordingly, there is no reduction of the number of core arguments in the core headed by bring. The revised syntactic template selection principles are given in (147); the principles unique to complex sentences are in (147b). (147)

Syntactic template selection principles (revised formulation): a. Basic principle: The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core. b. Qualifications of the principle in (a): The number of slots in a core is reduced by 1 if: 1. It is the licensing core in a non-subordinate core juncture in which the linked core is a semantic argument of the licensing verb. 2. It is the linked core in a non-subordinate core juncture. c. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre-/post-core slot reduces the number of core slots by 1 (may override c1).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

There seems to be an interesting asymmetry between (147b1) and (b2), namely, there seem to be systematic exceptions to the first principle and few to the second. It will be argued in the following section that the construction in (143b) is the result of a violation of (147b1), hence its relative rarity in the world’s languages. Even rarer is a violation of principle (b2), and it is found primarily in ‘backward control’ phenomena in some Caucasian languages (e.g. Tsez, Polinsky and Potsdam 2002; Kabardian; Matasovi´ c 2008, 2009, 2010). Examples of ‘forward’ (a) and ‘backward’ (b) control in Tsez are given in (148). (148)

a. Uži-r [t’ek-Ø magazin-y¯ ay yis-a] šuλ’i-s. boy-dat book-abs store-abl take-inf forget-pst.evid ‘The boy forgot to buy a book from the store.’ [forget′ (užii, [ . . . do′ (yi, [buy′ (yi, . . . ])]) b. Kid-b¯a ziya-Ø b-išr-a y-oq-si. girl.II-erg cow.III-abs III-feed-inf II-begin-pst.evid ‘The girl began to feed the cow.’ do′ (xi, [begin′ (xi, [[do′ (kidi, Ø)] CAUSE [do′ (ziya, [eat′ (ziya, Ø]])])

In (148a) the controller and shared argument uži- ‘boy’ is in the licensing core, which is shown by the fact that it carries dative case assigned by the predicator šuλ’i- ‘forget’, and it is co-indexed with the lexically unfilled variable in the embedded LS corresponding to the linked core. This example adheres to (147b1, 2). In (148b), on the other hand, the situation is very different, as the shared argument and controller is in the linked core, not the licensing core, as shown by the fact that it carries ergative case, being the actor of the transitive verb ‘feed’, rather than absolutive case as the single core argument of the intransitive verb -oq‘begin’. As the LS shows, the lexically unfilled variable is in the LS of the licensing core, and the lexically filled argument is in the LS of the linked core. Both constructions satisfy the Completeness Constraint, and there is nothing in the theory which precludes the two possibilities in (148). The question of why the one option is so much more frequent than the other remains. An interesting feature of the (b) example is that the controller in the linked core triggers noun class agreement on the verb in the licensing core. This is readily accounted for: ‘begin’ shows noun class agreement with the xargument, and the x-argument is co-indexed with kid ‘girl’ (class II) in the embedded LS, and accordingly it will show class II agreement. The structure of (148b) is given in Figure 1.49. This structure violates (147b2), because the linked core has its full complement of RP arguments, unlike its English translation. It could also be construed as violating (147b1), because it lacks the core argument in the licensing core that the principle predicts it should have: -oq- ‘begin’ has two arguments in its LS, and when it occurs in the licensing core of the juncture, it should have one core argument.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

153

154

ROBERT D. VAN VALIN, JR.

Figure 1.49 The structure of (148b)

Matasovi´ c (2008, 2009, 2010) adduces similar phenomena in Kabardian and proposes an RRG account based on a distinction between ‘head-first’ vs. ‘dependent-first’ application of grammatical principles. In (148), forward control in (a) is the result of head-first filling of argument positions with lexical items and applying case and agreement principles. Backward control in (b), on the other hand, results from a dependent-first application of these same operations. Matasovi´ c extends this account to the issue of the domain of case assignment in core junctures. In Van Valin and LaPolla (1997, §9.2.1) it was argued that some languages have the clause as the domain of case marking (e.g. English), while others have the core as the domain (e.g. Icelandic). He shows that the difference between English and Icelandic in this regard can be explained in terms of the head-first vs. dependent-first contrast, thereby eliminating the need for two case assignment domains and extending the explanatory value of this opposition. Framing the issue in this way opens the door to a possible answer to the question of why one option is so much more frequent than the other; it has long been recognized that heads, be they phrasal or sentential, play morphosyntactically significant roles in grammar, and it should therefore not be surprising that headfirst rather than dependent-first is the more prevalent orientation for the application of grammatical rules and principles.

1.7.2 Argument Identity in Complex Sentences An issue that affects only complex sentence linking is the way argument identity across different units of the construction is handled.67 There are a couple of interacting factors. The first is the level of juncture. In a nuclear juncture involving coordination or cosubordination, the arguments of the component nuclei are ‘pooled’ in a single derived argument structure, which is often the same as lexical verbs with a similar meaning. For example, the Mandarin nuclear juncture dˇ a sˇı ‘beat to death’ in (60b) is composed of dˇ a ‘beat’ and sˇı ‘die’; the LS for (60b), as proposed in Section 1.4.2.3, is [do′ (1sg, [beat′ (1sg, shé)])] cause [INGR dead′ (shé)],

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

which has three components: dˇ a ¼ [do′ (1sg, [beat′ (1sg, shé)])], sˇı ¼ [INGR dead′ (shé)], and cause, which represents the constructionally invoked meaning of causation. The derived LS is very similar to that of a lexical causative verb, for example smash as in ‘The burglar smashed the window’ ¼ [do′ (burglar, [act.on′ (burglar, window)])] CAUSE [INGR smashed′ (window)]. Nuclear junctures with the second predicate being transitive link like lexical ditransitive verbs, as in the French example in (149a) (Hyman and Zimmer 1976). (149)

` a. Je fer-ai mang-er les gâteux a Jean. 1sg.nom make-1sg.fut eat-inf the.m.pl cakes dat John. ‘I will make John eat the cakes.’ (Literally ‘I will make eat the cakes to John’) b. Je donner-ai les gâteux ` a Jean. 1sg.nom give-1sg.fut the.m.pl cakes dat John ‘I will give the cakes to John’. c. [do′ (1sg, Ø)] CAUSE [do′ (Jean′, [eat′ (Jean, cakes)])] ¼ (a) d. [do′ (1sg, Ø)] CAUSE [INGR have′ (Jean, cakes)])] ¼ (b) Actor NMR Undergoer

Macrorole assignment works the same with reference to both LSs, yielding the prototypical ditransitive linking pattern for both. The cores in non-subordinate core junctures necessarily ‘share’ a core argument, as shown in the examples in (142) and (143). The shared argument was termed the ‘argument in common’ earlier. The constituent projections of (142a, b) are given in Figures 1.50 and 1.51. (See Section 1.3.2.) If persuade in Figure 1.51 were replaced by promise, as in (142c), the structure would not change, because the difference between actor control and undergoer control is captured by the theory of obligatory control as part of the linking algorithm. The important point for the discussion of argument sharing is that there is no empty argument position or zero element in the

Figure 1.50 Core cosubordination in (142a)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

155

156

ROBERT D. VAN VALIN, JR.

Figure 1.51 Core coordination in (142b)

linked core. The essential features of the linking from semantics to syntax are given in (150). (150)

do′ (Maxi, [try′ (Max, [[do′ (xi, Ø)] CAUSE [INGR tuned′ (piano)]])]) Actori Undergoer Actori PSA DCA

The x-argument of the embedded LS is lexically unfilled and co-indexed with the controller in the matrix LS. The co-indexing satisfies the Completeness Constraint, because the x-argument is realized indirectly via its controller. See Van Valin (2005, §7.3.1) for detailed discussion. The ‘raising’ constructions in (143) parallel the obligatory control constructions in (142a–c): the intransitive ‘raising to subject’ version with seem is core cosubordination, whereas the transitive ‘raising to object’ version is core coordination. This is a change from the analysis in Van Valin (2005, §7.3.2), where it was argued that because deontic modals are unacceptable in seem constructions (e.g. *Sam must/ought to/ can seem to like wine), the nexus is coordination. But other considerations support a cosubordination analysis. First, in constructions like (143a) negation in the licensing core can readily trigger an NPI in the linked core, whereas with the transitive forms the NPI seems awkward. (151)

a. Sam doesn’t/didn’t seem (to Sally) to like (any of ) the wines. b. Sam doesn’t/didn’t seem (to Sally) to be eating any cookies. c. Sam doesn’t/didn’t seem (to Sally) to have saved a red cent.

(152)

a. Sally doesn’t/didn’t believe Sam to like (?any of ) the wines. b. Sally doesn’t/didn’t believe Sam to have eaten ?any/the cookies. c. ?Sally doesn’t/didn’t believe Sam to have saved a red cent.

Second, Bohnemeyer and Van Valin (2017) argue that core cosubordination constructions have the macro-event property and therefore they cannot take

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

multiple temporal positional modifiers, while core coordination constructions allow them and therefore lack the MEP. (153)

a. *This morning Sam didn’t seem to be eating pizza in the afternoon. b. This morning Sally believed Sam to be eating pizza for lunch at noon.

Taken together, the facts in (151)–(153) suggest that constructions like (143a) are core cosubordination, while those in (143b) are core coordination. The constituent projections of these two examples are given in Figures 1.52 and 1.53. Here again there are no empty argument positions in the linked core. ‘Raising to subject’ constructions are made possible by M-atransitive verbs like seem and appear. The LS for (143a) would be seem′ (Sally, [like′ (Sam, wine)]) [MR Ø], in which neither of the semantic arguments of seem can be

Figure 1.52 Core cosubordination in (143a)

Figure 1.53 Core coordination in (143b)

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

157

158

ROBERT D. VAN VALIN, JR.

realized as a direct core argument. From this LS two different patterns can come, depending on whether there is a tense operator in the embedded LS. If there is one, the result is It seems to Sally that Sam likes wine. If there is only one tense operator over the entire LS, the result is (143a), in which the PSA of the linked core appears as the PSA of the licensing core. This is a function of the principles in (147b2) and (147c1) together with the M-atransitive nature of seem: (147b2) requires that the linked core is missing a core argument position, and (147c1) requires that the licensing core have at lease one core argument position, which cannot be filled by one of the arguments of seem. In order to avoid a Completeness Constraint violation, the ‘orphaned’ argument of like is linked to the open core argument slot in the licensing core, yielding (143a). This is a different kind of argument sharing from that in control constructions. It is a very common construction in the languages of the world. The ‘raising to object’ construction, on the other hand, is not. The LS for (143b) would be believe′ (Sally, [like′ (Sam, wine)]), and as before, there are two possible realizations, depending on whether there are two tense operators or just one over the entire LS. The first possibility yields Sally believes that Sam likes wine, and the second yields (143b), which violates the principle in (147b1). Believe and other ‘raising to object’ verbs are exceptional in not reducing their S-transitivity as required by (147b1), thereby leaving an open core argument slot which can be filled by the ‘orphaned’ argument of the linked core verb, in this case like, thereby avoiding a Completeness Constraint violation. The nature of argument sharing in this construction is the same as that in the seem construction. (See also Van Valin 2005, §7.3.2.) The purpose construction in (142d), Sam brought wine for everyone to drink, has an obligatory shared argument, unlike the rationale construction in (154), which is a clausal juncture. (154)

a. Sam brought wine in order for Sally to drink it. b. *Sam brought wine in order for Sally to drink. c. Sam brought wine in order to drink it.

This construction does not allow a shared ‘object’ argument in the linked clause, but does permit an optional omitted PSA coreferential with the PSA in the first clause. The structure of (142d) is given in Figure 1.54; it is core coordination, as shown by the fact that each core can take an independent temporal positional modifier, for example Sam brought wine in the afternoon for everyone to drink at the party during the evening. The LS for this sentence is [do′ (Sam, [bring′ (Sam, winei)])] PURP [do′ (everyone, [drink′ (everyone, xi)])].68 This construction is significant, because it involves a controller–pivot relationship in which neither the controller nor the pivot is a traditional subject (see §1.3.2, Van Valin 2009). Because the dependent LS is not an argument of the matrix LS, the reduction in S-transitivity in the licensing core mandated by (147b1) does not apply.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Figure 1.54 Core coordination in (142d)

Despite the differences from the constructions in (142a–c), this construction has an obligatory shared argument of the same type. Thus, in nuclear junctures the arguments of the constituent nuclei are pooled, just as in an LS of a multi-argument lexical verb. In non-subordinate core junctures, there are two types of argument sharing. In control constructions, one argument is shared between (or among) the cores; it occurs overtly on one core and is co-indexed with its corresponding argument position in the LSs of other cores. In ‘raising’ constructions, on the other hand, a semantic argument from the linked core appears as a syntactic core argument in the core headed by the ‘raising’ verb. In neither case is there an empty syntactic slot in the linked core for the shared argument, nor is there a null pronominal (e.g. PRO) representing the shared argument. Argument sharing is only possible across core boundaries; across clause boundaries argument identity involves coreference. Controller–pivot relations may involve argument sharing, as in obligatory control constructions, or coreference, as in conjunction reduction constructions (e.g. (64), (65), (113)). The following examples illustrate an important contrast. (155)

a. Sallyi was talking to Samj, and shei/hej waved to Margaret as she walked by. b. Sallyi was talking to Samj, and proi/*j waved to Margaret as she walked by. c. Sallyi was talking to Sam, and waved to Margaret as she walked by.

The first example involves coreference between overt pronouns, and one of the major constraints in English is the requirement that third-person pronouns must agree in gender in order to be interpreted as coreferential. The second example lacks an overt pronoun in the PSA position in the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

159

160

ROBERT D. VAN VALIN, JR.

second clause, and most theories would posit null pronominal, pro, as the PSA of the second clause. In (155b) only Sally can be interpreted as the one who waved to Margaret; Sam is ruled out by a syntactic constraint on coreference to the effect that in conjunction reduction constructions, the PSA of the first clause must be the controller of the pivot (missing PSA) in the second clause. It makes sense to talk of coreference between the controller, RP Sally, and the phonologically null pro as pivot. But RRG does not permit phonologically null lexical items in its syntactic representations, and accordingly (155b) is not a possible RRG analysis.69 Rather, the structure RRG posits is (155c), with no null pronominal. This immediately raises the question, what is Sally coreferential with? If there’s nothing in the syntax instantiating the PSA of the second clause, how can this be a case of coreference? Nevertheless, the PSA of the second clause is interpreted to be the same as the PSA of the first clause. The solution cannot be formulated over the constituent projection of the LSC alone but must include discourse. In Chapters 11 (Bentley) and 12 (Latrouite and Van Valin) the RRG approach to the representation of discourse will be presented, and it is essential to the analysis of conjunction reduction in particular and for controller–pivot relations in clausal junctures in general. See Section 12.4 in Chapter 12 for detailed presentation of the RRG analysis of conjunction reduction. Controller–pivot relations in clausal junctures need not always be grammatically constrained, as they are in English, Icelandic, German, and Croatian, for example. Archi, a Caucasian language (Kibrik 1979), lacks syntactic or semantic constraints on controller–pivot coreference in clausal junctures. The following examples from Kibrik (1979) illustrate this. In the structural schemas, the italicized expressions with strikethrough do not occur overtly in the sentence. (156)

a. Adamli [k’ar¯ aši xit’-boli] cˇ’ele goıroı-abti. man.erg down push-ger stone.abs roll-aor. ‘The man rolled the stone, having pushed it down.’ a′. man.erg [man.erg stone.abs push down] stone.abs roll b. Os Helekulin lo [jab liqi’ili one hen.gen child[IVsg].abs this[IIIsg].erg eagle[IIIsg].erg o¨ xali] oqıali]. IVsg.carry_away.ger [IVsg].disappear ‘One of the chicks disappeared, carried away by an eagle.’ b′ chick.abs [eagle.erg chick.abs carry away] disappear ˇsu. c. Bošor [k’oaHan soli] weır˙ man[Isg].abs [stick[IVsg].abs hold.IVsg.ger] run.Isg.aor ‘The man ran, holding the stick.’ c′. man.abs [man.erg stick.abs hold] run

There are two controller–pivot relations in (156a): an ergative transitive actor controlling an ergative transitive actor, and an absolutive transitive undergoer controlling an absolutive transitive undergoer. In (156b) an

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

absolutive intransitive undergoer controls an absolutive transitive undergoer, and in (156c) an absolutive intransitive actor controls an ergative transitive undergoer. Thus intransitive and transitive actors can function as controllers (a, c) and as pivots (a, c), and similarly intransitive and transitive undergoers can function as controllers (a, b) and as pivots (a, b). Thus there are no constraints on coreference other than that the controller– pivot relations should make sense in the context (A. E. Kibrik, personal communication). In core and clausal junctures some of the constructions express argument identity by only overtly instantiating the first of the possible occurrences of the arguments. This is traditionally described in terms of ‘zero anaphora’. Moreover, there may be structural constraints on the interpretation of argument identity. It should be clear from this discussion that the nature of ‘zero anaphora’ in core junctures is very different from that of ‘zero anaphora’ in clausal junctures. This is summarized in (157). (157)

a. Argument-sharing [linking ‘zeros’] vs. coreference [discourse ‘zeros’] b. Coreference: structurally constrained (155c) vs. structurally unconstrained (156)

As shown in Figures 1.50 to 1.54, there is no gap or empty argument position in the non-subordinate core junctures, and therefore describing them in terms of ‘zero anaphora’ is quite incorrect from an RRG perspective. There are no zeros in these constructions. In the clausal junctures, on the other hand, there are lacunae in the linked clauses; in (155c) there is no PSA in the second clause, and in the Archi examples in (156) one or more core arguments is missing from the linked clause. These lacunae are analysed as zero elements in some approaches but not in RRG. What is significant is the contrast between the structurally constrained coreference in English and many other languages and the structurally unconstrained coreference in Archi. With respect to argument sharing in core junctures, the interpretation in (142a, b, c) is a function of the theory of control, and in (142d) it follows from the constructional schema. The argument sharing is constrained primarily lexically rather than by morphosyntactic structure.

References Aissen, Judith. 1987. Tzotzil Clause Structure. Dordrecht: Reidel. Balogh, Kata, Anja Latrouite and Robert D. Van Valin, Jr. 2020. Nominal anchoring: Introduction. In Kata Balogh et al. (eds.), Nominal Anchoring: Specificity, Definiteness, and Article Systems Across Languages, 1–18. Berlin: Language Sciences Press. Beavers, John. 2010. The structure of lexical meaning: Why semantics really matters. Language 86: 821–864.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

161

162

ROBERT D. VAN VALIN, JR.

Beavers, John. 2011. An aspectual analysis of ditransitive verbs of caused possession in English. Journal of Semantics 28: 1–54. Beavers, John. 2013. Aspectual classes and scales of change. Linguistics 51: 681–706. Becker, Alton. 1975. A linguistic image of nature: The Burmese numerative classifier system. Linguistics 13: 109–122. Belloro, Valeria. 2007. Spanish Clitic Doubling: A Study of the Syntax–Pragmatics Interface. PhD dissertation, University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/]. Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: Mouton. Bentley, Delia. 2008. The interplay of focus structure and syntax: Evidence from two sister languages. In Van Valin (ed.), 263–284. Bentley, Delia. 2019. The logical structure of verbs of quantized and nonquantized change. Paper presented at 2019 International Role and Reference Grammar Conference, University at Buffalo (SUNY). Bentley, Delia. 2022. Internally caused change as change by inner predisposition: Comparative evidence from Romance. Ms., University of Manchester. Berio, Leda, Anja Latrouite, Robert D. Van Valin, Jr. and Gottfried Vosgerau. 2017. Immediate and general common ground. In Patrick Brézillon, Roy Turner and Carlo Penco (eds.), Modeling and Using Context, 1–14. Heidelberg: Springer International Publishing. Bickel, Balthasar. 2010. Capturing particulars and universals in clause linkage: A multivariate analysis. In Bril (ed.), 51–101. Boas, Franz and Ella Deloria. 1941. Dakota Grammar. Memoirs of the National Academy of Sciences, Vol. 23. Washington, DC: US Govt. Printing Office. Bohnemeyer, Jürgen. 2019. Operator, information: Revisiting the operator projection in RRG, with special emphasis on tense, aspect, and finiteness. Paper presented at the 2019 RRG Conference, University at Buffalo (SUNY). Bohnemeyer, Jürgen and Robert D. Van Valin. Jr. 2017. The Macro-event Property and the layered structure of the clause. Studies in Language 41: 142–197. Bohnemeyer, Jürgen, Lindsay K. Butler and T. Florian Jaeger. 2016. Headmarking and agreement: Evidence from Yucatec Maya. In Jens Fleischhauer, Anja Latrouite and Rainer Osswald (eds.), Explorations of the Syntax– Semantics Interface, 169–207. Düsseldorf: Düsseldorf University Press. Bolinger, Dwight. 1975. Concept and percept: Two infinitive constructions and their vicissitudes. In World Papers in Phonetics: Festschrift for Dr. Onishi Kiju, 65–91. Tokyo: Phonetic Society of Japan. Bresnan, Joan. 1982. Control and complementation. In Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, 282–390. Cambridge, MA: MIT Press. Bresnan, Joan and Sam A. Mchombo. 1987. Topic, pronoun and agreement in Chichewa. Language 63: 741–782. Bril, Isabelle (ed.). 2010. Clause Linking and Clause Hierarchy. Amsterdam: John Benjamins. ˇ

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Cain, Bruce and James Gair. 2000. Dhivehi (Maldivian). Languages of the World/Materials 63. Munich: Lincom Europa. Chief, Liangchen. 2007. Scalarity and Incomplete Event Descriptions in Mandarin Chinese. PhD dissertation, University at Buffalo (SUNY). Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam. 1970. Remarks on nominalization. In Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar, 184–221. Waltham, MA: Ginn & Co. Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Diedrichsen, Elke. 2004. The German ‘bekommen-passive’ and RRG. In Brian Nolan (ed.), Linguistic Theory and Practice: Description, Implementation and Processing. Proceedings of the 2004 RRG Dublin Conference, 49–72. Diedrichsen, Elke. 2011. The theoretical importance of constructional schemas in RRG. In Nakamura (ed.), 169–197. Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. Dixon, R. M. W. (ed.). 1976. Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies. Dixon, R. M. W. 1984. The semantic basis of syntactic properties. BLS 10: 583–595. Dixon, R. M. W. 1991. A New Approach to English Grammar, on Semantic Principles. Oxford: Oxford University Press. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Dryer, Matthew. 1986. Primary objects, secondary objects, and antidative. Language 62: 808–845. Everett, Caleb. 2006. Patterns in Karitiâna: Articulation, Perception, and Grammar. PhD dissertation, Rice University, TX [available on Information Structure in Amazonian languages website]. Everett, Caleb. 2008. Constituent Focus in Karitiâna. Unpublished ms. [available on Information Structure in Amazonian languages website]. Everett, Daniel. 2002. Notes on an RRG Theory of Morphology. Unpublished ms. Everett, Daniel. 2005. Cultural constraints on grammar and cognition in ~ Another look at the design features of human language. Current Piraha: Anthropology 76: 621–646. Everett, Daniel. 2008. Wari’ intensional state constructions. In Van Valin (ed.), 381–412. Everett, Daniel and Doris Kern. 1997. Wari’. London: Routledge. Fleischhauer, Jens. 2016. Degree Gradation of Verbs. Doctoral dissertation, Heinrich Heine University Düsseldorf. Foley, William A. 2010. Clause linkage and nexus in Papuan languages. In Bril (ed.), 27–50.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

163

164

ROBERT D. VAN VALIN, JR.

Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Franklin, Karl. 1971. A Grammar of Kewa, New Guinea. (Pacific Linguistics C-16). Canberra: Australian National University. Goldberg, Adele. 2009. Constructions at work. Cognitive Linguistics 20: 201–224. González Vergara, Carlos. 2006. Las construcciones no reflexivas con ‘se’: Una propuesta desde la Gramática del Papel y la Referencia. PhD dissertation, Universidad Complutense de Madrid [available on the RRG website, https:// rrg.caset.buffalo.edu/]. González Vergara, Carlos. 2009. One rule to rule them all: Logical structures for Spanish non-reflexive se sentences. In Guerrero Valenzuela, Ibáñez and Belloro (eds.), 361–379. González Vergara, Carlos. 2011. Se-incompatible predicates in Spanish: A RRG explanation. In Nakamura (ed.), 134–142. Guerrero Valenzuela, Lilián, Sergio Ibáñez and Valeria Belloro (eds.). 2009. Studies in Role and Reference Grammar. México: UNAM. Guerrero Valenzuela, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of primary object languages. International Journal of American Linguistics 70: 290–319. Hale, Kenneth. 1976. The adjoined relative clause in Australia. In Dixon (ed.), 78–105. Hasegawa, Yoko. 1996. A Study of Japanese Clause Linkage: The Connective TE in Japanese (Studies in Japanese Linguistics 5). Stanford, CA: CSLI Publications. Haviland, John. 1979. How to talk to your brother-in-law in Guugu Yimidhirr. In T. Shopen (ed.), Languages and Their Speakers, 160–239. Cambridge, MA: Winthrop. Himmelmann, Nikolaus. 2008. Lexical categories and voice in Tagalog. In Paul Kroeger (ed.), Voice and Grammatical Relations in Austronesian Languages, 247–293. Stanford, CA: CSLI Publications. Holisky, Dee Ann. 1987. The case of the intransitive subject in Tsova-Tush (Batsbi). Lingua 71: 103–132. Hudson, Richard. 1992. So-called ‘double objects’ and grammatical relations. Language 68: 251–276. Humboldt, Wilhelm von. 1836. Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss aft die geistige Entwicklung des Menschengeschlects. Berlin: Dümmler. Hyman, Larry and Karl Zimmer. 1976. Embedded topic in French. In Charles Li (ed.), 189–211. Ikegami, Yoshihiko. 1985. ‘Activity’ – ‘accomplishment’ – ‘achievement’ – A language that can’t say ‘I burned it, but it didn’t burn’ and one that can. In A. Makkai and A. Melby (eds.), Linguistics and Philosophy: Essays in honor of Rulon S. Wells, 265–304. Amsterdam: John Benjamins. Imai, Shingo. 1998. Logical Structures and Case Marking in Japanese. MA thesis, University at Buffalo (SUNY) [available on RRG website, https://rrg.caset. buffalo.edu/].

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press. Jackendoff, Ray. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Jelinek, Eloise. 1984. Empty categories and non-configurational languages. Natural Language and Linguistic Theory 2: 39–76. Johnson, Mark. 1987. A New Approach to Clause Structure in Role and Reference Grammar. Davis Working Papers in Linguistics 1, 55–59. Davis: University of California. Kailuweit, Rolf. 2008. ‘Floating plurals’, pro drop and agreement: An optimality-based RRG approach. In Van Valin (ed.), 179–202. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kibrik, Alexander E. 1979. Canonical ergativity and Daghestan languages. In Frans Planck (ed.), Ergativity, 61–78. London: Academic Press. ~ u ~ (Kikuyu) Complex Sentences: A Role and Kihara, C. Patrick. 2017. Aspects of G~ıkuy Reference Grammar Analysis. PhD dissertation, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/]. Kolmer, Agnes. 1998. Pluralität im Tagalog (Arbeitspapiere des Instituts für Sprachwissenschaft der Universität zu Köln, 31.) Cologne: Institut für Sprachwissenschaft, Universität zu Köln. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA: MIT Press. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: Cambridge University Press. Langacker, Ronald. 2009. Cognitive (Construction) Grammar. Cognitive Linguistics 20: 167–176. Latrouite, Anja and Robert D. Van Valin, Jr. 2014. Event existentials in Tagalog: A Role and Reference Grammar account. In I. Wayan Arka and N. L. K Mas Indrawati (eds.), Argument Realizations and Related Constructions in Austronesian Languages, Papers from 12-ICAL, Vol. 2, 161–174. Canberra: AsiaPacific Linguistics [available on RRG website, https://rrg.caset.buffalo.edu/]. Latrouite, Anja and Robert D. Van Valin, Jr. 2021. An RRG account of aspects of the information structure-syntax interface in Tagalog. In Van Valin (ed.), 257–283. Li, Charles (ed.). 1976. Subject and Topic. New York: Academic Press. Mann, William C. and Sandra A. Thompson. 1988. Discourse Description. Amsterdam: John Benjamins. Matasovi´ c, Ranko. 2008. Transitivity in Kabardian. In Van Valin (ed.), 59–73. Matasovi´ c, Ranko. 2009. A new typology of control constructions within the framework of Role and Reference Grammar. In Guerrero Valenzuela, Ibáñez and Belloro (eds.), 305–318. Matasovi´ c, Ranko. 2010. Kabardian causatives, reflexives and case marking domains. Suvremena Lingvistika 69: 45–63. Matasovi´ c, Ranko. 2017. Agreement in Role and Reference Grammar: A typology of possible targets. Suvremena Lingvistika 84: 157–171. Nakamura, Wataru (ed.). 2011. New Perspectives in Role and Reference Grammar. Newcastle upon Tyne: Cambridge Scholars Publishing.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

165

166

ROBERT D. VAN VALIN, JR.

Nakamura, Wataru. 2021. A neo-Jakobsonian account of default cases: Instrumental vs. dative. In Van Valin (ed.), 135–168. Narasimhan, Bhuvana. 1998. A lexical semantic explanation for ‘quirky’ case marking in Hindi. Studia Linguistica 52: 48–76. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62: 56–119. Nichols, Johanna, David A. Peterson and Jonathan Barnes. 2004. Transitivizing and detransitivizing languages. Linguistic Typology 8: 149–211. Nolan, Brian and Elke Diedrichsen (eds.). 2013. Linking Constructions into Functional Linguistics: The Role of Constructions in Grammar. Amsterdam: John Benjamins. Nuhn, Patrick. 2021. Ay-Inversion in Tagalog: Information Structure and Morphosyntax of an Austronesian Language. Düsseldorf: Düsseldorf University Press [available on RRG website, https://rrg.caset.buffalo.edu/]. Nunes, Mary L. 1993. Argument linking in English derived nominals. In Van Valin (ed.), 375–432. O’Connor Rob. 2008. A prosodic projection for Role and Reference Grammar. In Van Valin (ed.), 227–244. Ohori. Toshio. 2001. Some Thoughts on a New Systematization of Interclausal Semantic Relations. Unpublished ms. [available on RRG website, https://rrg. caset.buffalo.edu/]. Osborne, C. R. 1974. The Tiwi Language. Canberra: Australian Institute of Aboriginal Studies. Osswald, Rainer. 2021. Activities, accomplishments and causation. In Van Valin (ed.), 3–30. Park, Ki-seong. 1993. Korean Causatives in Role and Reference Grammar. MA thesis, University at Buffalo (SUNY) [available on RRG website, https://rrg. caset.buffalo.edu/]. Pensalfini, Rob. 2004. Towards a typology of non-configurationality. Natural Language and Linguistic Theory 22: 359–408. Pinker, Steven. 1989. Learnability and Cognition. Cambridge, MA: MIT Press. Polinsky, Maria and Eric Potsdam. 2002. Backward control. Linguistic Inquiry 33: 245–282. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik. 1972. A Grammar of Contemporary English. New York: Seminar Press. Rappaport Hovav, Malka and Beth Levin. 2008. The English dative alternation: The case for verb sensitivity. Journal of Linguistics 44: 129–167. Rijkhoff, Jan. 1990. Explaining word order in the noun phrase. Linguistics 28: 5–42. Rijkhoff, Jan. 2002. The Noun Phrase: A Typological Study of Its Form and Structure. Oxford: Oxford University Press. Rögnvaldsson, Eiríkur. 1982. We need (some kind of a) rule of conjunction reduction. Linguistic Inquiry 13: 557–561. Ross, John Robert. 1967. Constraints on Variables in Syntax. PhD dissertation, MIT.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Ruhnau, Arne. 2011. Interpretation of the Topological Field Model of the German Clause in Terms of Role and Reference Grammar. MA thesis, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/]. Sag, Ivan A. 2012. Sign-Based Construction Grammar: An informal synopsis. In Hans C. Boas and Ivan A. Sag (eds.), Sign-Based Construction Grammar, 69–202. Stanford, CA: CSLI Publications. Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic, actor-topic, or none of the above. In Li (ed.), 491–518. Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley: University of California Press. Schwartz, Linda. 1993. On the syntactic and semantic alignment of attributive and identificational constructions. In Van Valin (ed.), 433–463. Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: wa and ga, and Word Order Flexibility. PhD dissertation, University at Buffalo (SUNY) [available on the RRG website, https://rrg.caset.buffalo.edu/]. Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japanese. In Wataru Nakamura (ed.), New Perspectives in Role and Reference Grammar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Dixon (ed.), 112–171. Smith, Carlotta. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Reidel. Talmy, Leonard. 2000. Toward a Cognitive Semantics. Cambridge, MA: MIT Press. Toratani, Kiyoko. 2002. The Morphosyntax and Logical Structures of Compound Verbs in Japanese. PhD dissertation, University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/]. Ullrich, Jan. 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN: Lakota Language Consortium. Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Constructions in Lakota. PhD dissertation, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/]. Ullrich, Jan and Ben Black Bear, Jr. 2018. Lakota Grammar Handbook (2nd ed.). Bloomington, IN: Lakota Language Consortium. Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics Interface. PhD dissertation, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/]. Van Valin, Robert D., Jr. 1977. Aspects of Lakhota Syntax. PhD dissertation, University of California, Berkeley. Van Valin, Robert D., Jr. 1981. Grammatical relations in ergative languages. Studies in Language 5: 361–394. Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota clause. In Johanna Nichols and Anthony Woodbury (eds.), Grammar Inside and Outside the Clause, 363–413. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 1990a. Semantic parameters of split intransitivity. Language 66: 221–260. Van Valin, Robert D., Jr. 1990b. Functionalism, anaphora and syntax. (Review article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

167

168

ROBERT D. VAN VALIN, JR.

Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and grammatical relations. Natural Language and Linguistic Theory 9: 145–194. Van Valin, Robert D., Jr. (ed.). 1993. Advances in Role and Reference Grammar. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Van Valin (ed.), 1–164. Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the mechanisms of language acquisition. In Michael Tomasello (ed.), The New Psychology of Language: Cognitive and Functional Approaches to Language Structure, 221–49. Mahwah, NJ: LEA [available on RRG website, https://rrg.caset. buffalo.edu/]. Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax– semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin, (eds.), Empirical Issues in Formal Syntax and Semantics 2, 373–389. The Hague: Thesus [available on RRG website, https://rrg.caset.buffalo.edu/]. Van Valin, Robert D., Jr. 2001. The acquisition of complex sentences: A case study in the role of theory in the study of language development. Chicago Linguistic Society Proceedings 36(2): 511–531. Van Valin, Robert D., Jr. 2002. The development of subject–auxiliary inversion in English wh-questions: An alternative analysis. Journal of Child Language 29: 161–175. Van Valin, Robert D., Jr. 2003. Some remarks on the bekommen-‘passive’ in German. Invited paper presented at the 9th German Linguistics Association Conference, University at Buffalo (SUNY). Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen, 62–82. Tübingen: Narr [available on RRG website, https://rrg.caset.buffalo. edu/]. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2006. Semantic macroroles and language processing. In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie and Angela D. Friederici (eds.), Semantic Role Universals and Argument Linking: Theoretical, Typological, and Psycholinguistic Perspectives, 263–301. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic categories in RRG. In Van Valin (ed.), 161–178. Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and controllers. In Guerrero Valenzuela, Ibáñez and Belloro (eds.), 45–68. Van Valin, Robert D., Jr. 2012a. Lexical representation, co-composition, and linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon, Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in Generative Lexicon Theory, 67–107. Dordrecht: Springer.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

Van Valin, Robert D., Jr. 2012b. Some issues in the linking between syntax and semantics in relative clauses. In Bernard Comrie and Zarina Estrada Fernández (eds.), A Typological Overview of Relative Clauses in Languages of the Americas, 47–64. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory. In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan Timberlake (eds.), Language Typology and Historical Contingency. In honor of Johanna Nichols, 91–124. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2016. An overview of information structure in three Amazonian languages. In M. M. Jocelyne Fernandez-Vest and Robert D. Van Valin, Jr. (eds.), Information Structuring of Spoken Language from a CrossLinguistic Perspective, 77–92. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. 2018a. Some issues regarding (active) accomplishments. In Rolf Kailuweit, Lisa Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 71–94. Freiburg: FRIAS. Van Valin, Robert D., Jr. 2018b. Dative case and oblique subjects. In Jóhanna Barðdal, Steven Carey, Thórhallur Eythórson and Na’ama Pat-El (eds.), NonCanonically Case-Marked Subjects: The Reykjavík-Eyjafjallajökull Papers, 115–131. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics– Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon Tyne: Cambridge Scholars Publishing. Van Valin, Robert D., Jr. 2021. Cosubordination. In Van Valin (ed.), 241–283. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and David P. Wilkins. 1993. Predicting syntactic structure from semantic representations: remember in English and Mparntwe Arrernte. In Van Valin (ed.), 499–534. Van Valin, Robert D., Jr. and David P. Wilkins. 1996. The case for ‘effector’: Case roles, agents and agency revisited. In M. Shibatani and S. Thompson (eds.), Grammatical Constructions, 289–322. Oxford: Oxford University Press. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Walton, Charles. 1986. Sama Verbal Semantics: Classification, Derivation and Inflection. Manila: Linguistic Society of the Philippines. Watters, James K. 1993. An investigation of Turkish clause linkage. In Van Valin, (ed.), 535–560. Winther-Nielsen, Nicolai. 1995. A Functional Discourse Grammar of Joshua: A Computer-Assisted Rhetorical Structure Analysis. Stockholm: Almqvist & Wiksell. Winther-Nielsen, Nicolai. 2021. Why Eve shouldn’t eat the snake: An intelligent answer from corpus-driven information structure and reference tracking in Biblical Hebrew. In Van Valin (ed.), 285–307. Yang, Byong-seon. 1994. Morphosyntactic Phenomena of Korean in Role and Reference Grammar: Psych-Verb Constructions, Inflectional Verb Morphemes, Complex

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

169

170

ROBERT D. VAN VALIN, JR.

Sentences, and Relative Clauses. PhD dissertation, University at Buffalo (SUNY). Published by Hankuk Publishers, Seoul, 1994. Zaenen, Annie, Joan Maling and Hölskuldur Thráinsson. 1985. Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3: 441–483.

Notes 1 In the tree diagrams modifiers are related to the units modified via arrows. Kallmeyer and Osswald (Chapter 20) propose an alternative formalization of the layered structure of the clause which does not use arrows to represent adjunct modifiers. I will continue to use the traditional RRG representations. 2 Bohnemeyer and Van Valin (2017) show that the core is the unit of the LSC which expresses a single event; it has the macro-event property, which the nucleus and the clause lack. See Section 1.3.4.4. 3 For explications of these notions, see Bentley (Chapter 11). 4 The auxiliary did is not attached to anything in the tree in Figure 1.4. Its status will be addressed in Section 1.2.3. With respect to the PrCS and resumptive elements, the claim that PrCS XPs never co-occur with a resumptive element holds true for English and many other languages. However, in head-marking languages (see Section 1.2.7.2) which have a PrCS, e.g. Tzotzil (Aissen 1987), an argument in the PrCS is crossreferenced on the verb, which could be viewed as a type of resumptive element; what would be impossible would be for there to be an independent pronoun in the clause referring to the PrCS RP. Spanish presents an interesting case as well. Dative arguments are obligatorily realized by a dative clitic on the verb and optionally also by a dativemarked noun or pronoun (see (58) in Section 1.2.7.2); in wh-questions in which the wh-expression is dative, such as (58d), it occurs in the PrCS and is still cross-referenced by the dative clitic in the core. Hence the issue of resumption and the PrCS is more complex than it appears to be in English. 5 It’s been claimed by a number of linguists, most notably Kayne (1994) in his theory of the antisymmetry of syntax, that all displacement is to the beginning of the sentence. These Dhivehi examples are a counterexample to this claim. 6 I am grateful to colleagues from Iran (Dr Mozhgan Neisani, Dr Zahra Ghane’) for bringing this problem to my attention. See Winther-Nielsen (2021) for the same problem regarding the analysis of Biblical Hebrew. The post-detached position (PoDP) involves an intonation break and resumptive cataphoric pronoun for arguments, e.g. What did shei eat after lunch, Maryi?

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

7 ‘DetP’ will be used for ‘determiner phrase’ rather than the usual ‘DP’, because ‘DP’ has already been used to stand for ‘detached position’ in PrDP and PoDP. 8 See Section 1.2.7.2 for an explanation as to why the independent RP is not core-internal. 9 In Van Valin (2005) ‘RP’ stands for ‘referential phrase’, but this term is problematic for various reasons (see Peterson, Chapter 2 of this volume). In Van Valin (2008) it was replaced by ‘reference phrase’ and defined as a potentially referring expression, thereby taking into account the fact that RPs can be non-referential in certain contexts. 10 RP, MP and nucleus are universal categories, while PP is not, since there are languages which lack adpositions altogether, e.g. Dyirbal. 11 See Peterson (this volume, Chapter 2) for an in-depth discussion of this topic. 12 Peterson (Chapter 2) argues that Kharia, a South Munda language (India), makes no categorial distinctions among content morphemes at all. 13 The existence of attributive modifiers in Lakhota has been controversial since Boas and Deloria (1941). See Ullrich (2018) for a detailed analysis of modification in Lakhota. 14 Bohnemeyer (2019) presents an alternative conception of operators and their relationship to the LSC based on a different theory of tense and aspect from the one assumed traditionally in RRG. 15 The formalization of RRG proposed in Kallmeyer and Osswald (Chapter 20) does away with the operator projection and represents both constituents and operators in the same tree structure. I will continue to use the traditional representations. 16 Lakhota lacks tense marking, so this sentence can be interpreted as present or past, depending on context. 17 This is not the complete array; see Ullrich (2011: 820–822) for a full list. 18 See Peterson (Chapter 2) for more discussion and exemplification of this claim. 19 See O’Connor (2008) for the outlines of a prosodic projection for the LSC in which the intonational properties of different sentence types and constructions can be represented. 20 Additional examples of attachment to more than one projection will be given later in this section and in Section 3. 21 See Toratani (Chapter 9) for an in-depth discussion of adverbs and Ibáñez Cerda (Chapter 10) for a detailed discussion of adpositional phrases. 22 If the adjunct PP were treated as a parenthetical, then Max drank, after the argument with Mary, an entire bottle of wine would be acceptable, but it is highly marked and not the same structure as (30b). 23 For detailed discussion of the ordering of constraints on adverbs, see Toratani (Chapter 9). 24 Unlike Japanese, German is not consistently left-branching; it has many more prepositions than postpositions, and in complex sentences

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

171

172

ROBERT D. VAN VALIN, JR.

25

26

27

28 29 30

31

32 33

34

35 36

dependent units follow the main clause (default), as in English and unlike Japanese, where they precede it. Ross (1967) proposed treating these phrases as NPs, rather than PPs, since in a language like German, for example, a case-marked phrase headed by a noun would be analysed as an NP, not a PP. This is problematic, however, because it fails to distinguish direct core arguments from oblique core arguments, a distinction relevant to important grammatical phenomena. The CSP node in the RP applies to languages like Tagalog, Japanese and Korean which have case markers that are not bound morphemes but are not adpositions, either. In Korean and Japanese, for example, the distinction between case marker and postposition is reflected in the ability of case particles to be omitted and the inability of postpositions to be dropped. As noted in Van Valin (2013: 93), the recognition of the distinctiveness of head-marking languages goes back at least as far as Wilhelm von Humboldt (1836). The -A signifies that the final vowel undergoes ablaut in certain environments. See Ullrich (2011: 754). There are varieties of Latin American Spanish which allow (57c); see Belloro (2007). Foley (2010) and Bickel (2010) reject the notion of cosubordination as a valid linkage type. Van Valin (2021) shows that their arguments are based on an outmoded version of the LSC and make assumptions about cosubordination which are incorrect. In fact, a reanalysis of their data supports the validity of cosubordination as a nexus relation. The term ‘subject’ is in single quotes because it has no theoretical status in RRG. This construction, usually called conjunction reduction, contains a controller–pivot (the missing RP) relationship; see Sections 1.7.2 and 12.4 for further discussion of conjunction reduction. See LaPolla (Chapter 5) for a presentation of the RRG approach to grammatical relations. See París (Chapter 15) for an analysis of it-clefts. These discussions assume the previous version of the IRH presented in Van Valin (2005), but the issues relating to the many-to-one mapping between the two sides of the IRH are not affected by the revisions proposed here. Let occurs in the passive voice in some limited combinations, such as let go meaning ‘fired’, e.g. He was let go after showing up at work drunk. Note there are no comparable combinations with come, e.g. *He was let come to the party. See Van Valin (1990b), Van Valin and LaPolla (1997, §7.5), and Van Valin (2005, §§5.3, 7.5), for detailed discussions of reflexivization in English. The representations are meant as ‘generic’ junctures, like those in (59), because the issue of nexus will be addressed separately below.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

37 ‘DO’ indicates that the argument is an agent with an activity predicate, and ‘do′’ is an unspecified activity predicate used in accomplishment representations which do not specify the nature of the causing activity, e.g. in ‘The cat killed the rat’, the cat’s killing action is unspecified. 38 See Osswald (2021) for discussion of Dowty’s own revisions of the decompositional system. 39 The term ‘accomplishment’ has been used in a number of different senses by Vendler and Dowty and in RRG; see Van Valin (2018a) for detailed discussion. 40 The contrast with respect to inherent endpoints is rendered in English by simple past vs. past progressive, but it is not an inflectional opposition. See Van Valin (2018a) for examples from other languages which do not involve inflectional distinctions. Also, the addition of processes to the system was proposed in Van Valin (2005). 41 Because stumble into the room is what Talmy (2000) has termed a ‘satelliteframed’ expression and enter the room stumbling is a ‘verb-framed’ expression, they are composed differently. Stumble is a manner of motion verb, and so its LS is do′ (x, [move.manner′ (x)]), and the remainder of the LS is supplied by the PP into the room (the satellite). In the verb-framed enter the room stumbling, the motion predicate is unspecified, and the verb contributes the path and goal, with the manner component added, as in (81a2ii). In Van Valin (2018a) it was argued that an expression like stumble into the room is a nuclear juncture, and this suggests that enter the room stumbling could be one as well. 42 This also implies that the LS for activity verbs which potentially take an incremental theme or path would have a different LS from those that do not, e.g. shine (do′ (x, [shine′ (x)])), vs. run (do′ (x, [run′ (x)]) ^ PROC covering.path.distance′ (x, (y))). Note that run’s LS would be different from stumble’s, because Bill stumbled doesn’t necessarily involve him covering any distance, but Bill ran does. 43 Two points need to be kept in mind. First, the elements of the decomposition which are in boldfaceþprime, e.g. do′, or in all caps, e.g. CAUSE, INGR, are part of a semantic metalanguage and are not English, despite the obvious similarities. Second, when representing the LSs of sentences from other languages, only the lexical items from the language which function as arguments should appear in LSs. In a sentence like (13a) from Lakhota, for example, the LS would be do′ (wíŋyaŋ kiŋ hená, [dance′ (wíŋyaŋ kiŋ hená)]) (plus operators) ‘Those women dance’, not eˇ chúŋ′ (wíŋyaŋ kiŋ hená, [waˇ chí′ (wíŋyaŋ kiŋ hená)]. See Van Valin (2005: 45–46), esp. footnote 9. 44 Three things need to be noted with respect to causation. First, the causatives under discussion do not include so-called ‘internal causatives’ associated with verbs like (intransitive) rot or bloom. Causation involves a causer and a causee, and the single argument of these verbs is neither, hence they cannot be analysed as causative in the same sense as the

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

173

174

ROBERT D. VAN VALIN, JR.

45 46

47

48 49

50

examples in (91g). These verbs can be better described as involving ‘inner force’, a notion expressed in e.g. Lakhota grammar by the instrumental prefix na-, e.g. -pˇ hópA ‘to be popped or burst open’ þ na- ¼ napˇ hópA ‘to burst with explosive force from within, burst or pop open by itself or from inner force, to explode; erupt (as a volcano)’ (Ullrich 2011: 485, 386) (see Bentley 2022 for verbs of inner predisposition). Second, representing causation by CAUSE is a gross oversimplification, since it fails to take into account the different types of causality in the IRH found in (81a, b, e). A first step toward a more sophisticated approach to causation can be seen in Van Hooste’s (2018) proposal for incorporating the theory of force dynamics proposed in Talmy (2000) into RRG. Third, it is well known that in some languages (e.g. Mandarin, Chief 2007; Japanese, Ikegami 1985; Korean, Park 1993) causatives don’t necessarily entail that the causation has a result state, which yields apparently contradictory sentences translated as ‘I killed him but he didn’t die’, ‘I broke it but it didn’t break’. Chief (2007) shows that the predicates in such examples are scalar in nature, unlike their English counterparts, which means they involve non-quantized changes of state of the melt variety (not the widen variety). In order to express a quantized change of state with a specific result state, the result state must be explicitly realized in some way, in which case there is a result state, and the resulting sentences are contradictory like their English translations. See Van Valin (2005: 42), fn. 5. For detailed discussion, see Van Valin and LaPolla (1997), §4.7.3, Van Valin (2005), §2.3. The difference between (94a2ii), e.g. Bill entered the room stumbling, and (94b4iii), e.g. Bill sat reading the newspaper, lies in the fact that the stumbling is an integral part of the motion, as can be seen from the fact that it’s possible to integrate stumble into the motion predication, i.e. Bill stumbled into the room. Accordingly, the motion cannot be separated out as a separate action; in other words, no stumbling, no motion. Sitting is not part of the action of reading, on the other hand, as one can read sitting, standing or lying, hence the stance verb represents an independent action from the reading. Russian has a lexical verb of possession, imet’, which is subject to restrictions on its use and is not a colloquial expression of possession. This contrasts with Croatian imati ‘have’, which would be used in sentences like (95). (R. Matasovi´ c, personal communication). The Russian verb imeti ‘have, possess’ can be used in existential constructions. (R. Matasovi´ c, personal communication). Semantic macroroles are an original RRG construct; see Van Valin (1999) for a historical overview and a comparison with analogous ideas in other approaches. All of the x-arguments with activity predicates are subtypes of the basic notion of effector, the doer of the action, which may or may not be animate or volitional.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

51 An exception is found in the present-tense series of Georgian verbs, which have a nominative [A]–dative [U] pattern with transitive verbs (Van Valin 1990a). 52 In some languages, e.g. Lakhota, Yaqui (Uto-Aztecan), Belhare (SinoTibetan), the primary pattern with three-argument verbs is (103a) with the recipient as the undergoer, rather than (103a′) as in German, Croatian and many other languages (see Dryer 1986). Guerrero and Van Valin (2004) proposed a revision to the AUH to account for this: a verb may follow Principle A (lowest-ranking argument), Principle B (second highest ranking), or Principle C (both A and B are possible). Most threeplace verbs in English follow C, but there are some that are restricted to A, e.g. put, donate. A possible example of an English verb which follows only B is envy. Yaqui has all three possibilities, with B the predominant pattern (see Section 1.6.4.2). German signals the different options via derivational morphology, e.g. schenken ‘give as a gift’(¼ (100c)), laden ‘load’(¼(100b)) (A) vs. beschenken (¼(100c’)), beladen (¼(100b’))(B). 53 The reason seem′ (x, y) is [MR 0] is that neither of the arguments can appear as a direct core argument in the core headed by seem. The x-argument can only be realized in a to-PP, and the y-argument can only appear as an extraposed that-clause, e.g. seem′ (Bill, [like′ (Mary, pizza)]) —> It seems to Bill that Mary likes pizza, *Bill seems that Mary likes pizza. Note that in Mary seems to like pizza, Mary is not a semantic argument of seem but rather is one of like. This construction will be discussed in Section1.7. 54 O’Connor’s (2008) prosodic projection makes possible the explicit representation of the prosodic properties of utterances. This is very relevant to information structure, but it is a separate projection from the information structure projection, because there are languages which appear not to use prosody in marking information structural contrasts, e.g. Karitiâna (see C. Everett 2006, 2008; Van Valin 2016). In addition, prosody may interact directly with the operator projection, e.g. marking IF contrasts, and with the constituent projection, e.g. marking grouping of constituents. 55 The ‘proi’ is for expository purposes only and would not be part of an RRG analysis of conjunction reduction. See Section 1.7.2; Chapter 12, Section 12.4. 56 The qualification ‘highest-ranking macrorole core argument’ is necessary, because in a passive construction, the actor, which is the highest-ranking macrorole, appears as an adjunct in the core-level periphery, which leaves the undergoer as the highest-ranking macrorole core argument and therefore the controller of person agreement and the controller in conjunction reduction. See Matasovi´ c (2017) for a presentation of an RRG-based theory of agreement. 57 RRG is thus a type of construction grammar, albeit one very different from ‘mainstream’ Construction Grammar, e.g. Sign-Based Construction

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

175

176

ROBERT D. VAN VALIN, JR.

58

59

60

61

62

63

64

Grammar (Sag 2012), which is very close to HPSG, or Cognitive Construction Grammar (Goldberg 2009; Langacker 2009). What unites the mainstream approaches is the assumption that constructions are stored pairings of form and meaning and therefore that everything from a bound affix, to a word, to a phrase, to a passive are constructions, and consequently the lexicon–grammar distinction is a false dichotomy, with lexical items just being simpler form–meaning pairings than grammatical constructions. One inevitable consequence of this assumption is that all constructions are language-specific, and accordingly it is extremely difficult, if not impossible, to capture any cross-linguistic generalizations. RRG does not make this assumption: constructions in RRG are not stored form–meaning pairings but rather compilations of language-specific applications of the general principles of the theory. Numerous examples will be given in the text. Hence cross-linguistic generalizations are readily captured in the RRG variant of construction grammar. Nolan and Diedrichsen (2013) introduce a more elaborate representation for CSs, but the crucial points distinguishing the RRG notion of constructions from mainstream Construction Grammar still hold. See also Diedrichsen (2011). The undergoer can be omitted in (121a′, b′), in which case the remaining RP is interpreted as the actor. See Van Valin (2005: 117, fn. 16), for discussion of this aspect of the construction. In Chapter 19 a way is proposed to handle languages like German and Croatian using a new kind of syntactic template which has both fixed and unordered features. This example is adapted from Toratani (2002). The LS for taberu ‘eat’ is the activity LS, not the full active accomplishment LS that the translation calls for; this does not affect the points at hand. This leads to a second motivation for not assigning the argument introduced by -(r)are a macrorole. By the AUH in (99) it should be an actor, which seems odd for the argument of an LS emphasizing how the referent is affected by some event/state of affairs. If it were undergoer, on the other hand, the semantics would be appropriate, but then a sentence like (125b) would have two undergoers (Taroo, piza) in a single core, which the theory doesn’t allow. The best solution is simply not to assign a macrorole to the argument in question, as the selection of this argument as PSA is independently accounted for in the linking system. Not given, but it would have the same basic information as the German and Japanese ‘plain’ passive CSs, mutatis mutandis, for English. See Van Valin (2005: 132) for the English passive CS. In Yaqui accusative case marking and plural marking are in complementary distribution; hence no ACC marking is possible in this sentence with a plural undergoer.

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Principles of Role and Reference Grammar

65 More specifically, (143a) has been called ‘raising to subject’ or ‘matrixcoding as PSA’, and (143b) has been known as ‘raising to object’, ‘exceptional case marking’, or ‘matrix-coding as non-PSA’. 66 While teach and forget can take that-clauses, these LSs cannot be realized as that-clauses due to the missing arguments in the embedded LSs; they must be realized as non-subordinate core junctures. Moreover, if operators were represented, the embedded LSs would also lack tense operators. 67 Another instance of argument identity is reflexivization, which will not be discussed in detail in this chapter. See Bentley (2006), González Vergara (2006, 2009, 2011), Van Valin (1990a, b), Van Valin and LaPolla (1997, §7.5), Van Valin (2005, §§5.3, 7.5). 68 There are two simplifications in this LS. First, bring would have a complex LS involving motion and possession, but this is not relevant for the issue at hand. Second, ‘PURP’ is the abbreviation for the purposive interclausal relation in (94d12). 69 It should be noted that in Van Valin and LaPolla (1997: 521) the same structure as in (155b) is posited for conjunction reduction. This analysis is abandoned in Van Valin (2005), and the structure in (155c) is proposed for this construction in Van Valin (2005: 229).

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

177

https://doi.org/10.1017/9781316418086.003 Published online by Cambridge University Press

Part Two

Topics in RRG: Simple Sentences

https://doi.org/10.1017/9781316418086.004 Published online by Cambridge University Press

https://doi.org/10.1017/9781316418086.004 Published online by Cambridge University Press

2 Lexical and Grammatical Categories in RRG John M. Peterson

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ABLE ACT ADJP AOR C:TEL CONT EMPH

ability marker active voice adjectival phrase aorist culminative telic V2 continuous V2 emphatic

OBLIG OPT POL PRESUM PROX.HEARER PSBL QUAL

FAC HON HS HUM IF INT ISC

REAL REM RP SH TAM THEM TNS

MID

factive honorific hearsay human illocutionary force interrogative intentional state construction middle voice

N NHUM

neuter, neutral non-human

VIC

2.1

V2

obligation optative politeness presumptive evidential proximate to hearer possibility qualitative predication (
‘be’) realis remote reference phrase subject honorific tense, aspect, mood theme tense unit denoting aspect/ Aktionsart verbal inflectional clitic

Introduction

The present chapter deals with the status of lexical and grammatical categories in RRG, including parts-of-speech systems and grammatical

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

182

JOHN M. PETERSON

categories such as tense, aspect, mood (TAM) and negation, which are referred to in RRG as ‘operators’.* With respect to lexical categories, RRG differs from many other theories of language which assume a small, closed set of universal parts of speech such as ‘noun’, ‘verb’ ‘adjective’, etc., and which consider these to be the heads of corresponding syntactic categories, namely NP, VP and AdjP. By contrast, RRG does not assume that such categories are universal, in line with an increasing amount of typologically oriented research questioning the universality of these categories. Instead, RRG assumes functionally motivated, non-endocentric syntactic categories, such as nucleus or NUC (containing the predicate), referential phrases or RPs, and modifying phrases or MPs. Although these syntactic slots are typically realized by verbs, nouns and adjectives/adverbs, respectively, this is not required by the theory. Many languages, including English, allow for non-verbal predicates, non-adjectival modifying phrases, etc., while others show little or no evidence for lexical categories such as ‘noun’ and ‘verb’ or allow for clausal referents, modifiers and predicates. And even those languages which do possess traditional parts of speech often show so-called ‘category squishes’, in which assignment to a particular category is at best an arbitrary decision. Hence, categories such as NUC, RP and MP are not universally linked to particular lexical categories per se, although individual languages will generally have language-specific restrictions as to which elements may occupy these slots. The present chapter also provides a description of the grammatical categories referred to in RRG as operators, that is, those categories which ‘ground’ a clause, core or nucleus and which are closely linked to finiteness, a topic which has been receiving increasing attention in typological research in recent years. These categories include tense, aspect, deontic mood (‘modality’ in RRG terms), epistemic mood (‘status’), evidentials, etc. These units all have scope over a particular level of the sentence structure (nucleus, core, clause) and play an important role in the determination of nexus relations in complex sentences within the theory. This also includes categories which are primarily concerned with questions of reference, such as definiteness and deixis, which ground the RP.

2.2

Lexical and Syntactic Categories in RRG

2.2.1 General Issues A basic tenet of descriptive linguistics is that linguistic categories are only valid to the extent that they are empirically justified, and the view that there is no such thing as a universal grammatical category is gaining acceptance.1 This holds for all categories, including lexical categories such * I would like to thank Utz Maas for his comments on an earlier version of the present study. Needless to say, any remaining errors and misconceptions are my own.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

as noun, verb and adjective (to name just three) as well as the corresponding endocentric syntactic categories, NP, AdjP and VP, even though these categories – or at least the lexical categories – are generally considered universal. Although there are considerable differences of opinion with respect to the ‘correct’ approach to this topic, most researchers would probably agree with an approach which views lexical categories as language-specific categories based on feature bundles, in which ‘prototypical’ members of a particular class possess all of the features potentially associated with the particular class.2 For example, prototypical nouns in English would denote persons, places or things, mark for the plural (usually with -s), be compatible with definite and indefinite articles, etc., while less prototypical nouns, for example, may not be compatible with plural marking or the definite article. Problematic for this approach, however, is determining which features are more important than others, if indeed a hierarchy can be established at all, as well as how many features are ‘enough’ for an item to be assigned to a particular category. We will return to this topic in the following pages. As Himmelmann (2008: 260) notes, there is little doubt that ‘the inventory of function words is highly language-specific’. Rather, ‘[w]henever there is a controversy regarding the number and kind of syntactic categories in a given language, it pertains to the linguistic classification of content words’. Himmelmann assumes two grammatical levels for the discussion of parts of speech: (i) ‘the level of terminal syntactic categories where lexical items are categorised according to their phrase-structural properties’, and (ii) ‘the level of lexical categories proper where lexical items are categorised according to those grammatical features which are not directly relevant for phrase structure’ (Himmelmann, 2008: 263). To illustrate these two levels,3 Himmelmann proposes a hypothetical example in a language L, in which the most basic phrasal categories always consist of an overt function word X and an overt content word Y, with virtually all content words being compatible with all function words. This can be schematically portrayed as in (1). Here the status of the function word X distinguishes the phrase from other phrase types, hence ‘XP’. (1)

[XP [X] [Y]]

These two levels, that is, lexical and syntactic, need not correspond to one another directly, although this will often be the case, as, for example, with Ns as heads of NPs, Adjs as heads of AdjPs and Vs as heads of VPs in many languages. Himmelmann (2008: 264) notes five logical possibilities with respect to these two levels and their correspondence to one another, given in Table 2.1. With respect to category I, both levels possess distinct categories but these do not directly correlate in Ia (‘6¼’), while they do in Ib (‘¼’). Returning to our example above, if, following Himmelmann, we imagine that L has a further restriction such that not all content words Y behave similarly, as some Ys take plural suffixes, others take plural prefixes, while

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

183

184

JOHN M. PETERSON

Table 2.1 Possible alignments between lexical and syntactic categories (Himmelmann, 2008: 264) Lexical categories Ia Ib II III IV

distinct distinct indistinct distinct indistinct

Terminal syntactic categories 6¼ ¼

distinct distinct distinct indistinct indistinct

the members of a third class do not allow pluralization, then we would have case Ia, with distinct lexical categories based on the compatibility of Y with plural markers and also distinct terminal syntactic categories, but with no direct correspondence between the two levels, as the compatibility of plural affixes does not correlate with the compatibility of a particular Y with a particular X. To these two descriptive levels we can add a third level, that of function, here primarily reference, predication and attribution. This can involve a direct correlation between terminal syntactic cateogies and functional categories, that is, XP1 ¼ reference, XP2 ¼ predication, etc., or there can be no direct correlation, so that XP1, XP2, etc. can both be used in reference, predication and modification. At one extreme we would then have endocentric categories specialized for certain sentence-level functions, for example N ¼ NP ¼ reference, V ¼ VP ¼ predication, etc., and at the other extreme – at least putatively – no correlations whatsoever, although it is questionable whether such a language is possible. Such mismatches are not only a theoretical possibility; many researchers have claimed that the languages they are describing either do not in fact possess a certain category such as ‘N’ or ‘NP’, or that the respective language can at least best be described without recourse to such categories. Even in languages such as English, which possess relatively clearly defined lexical classes such as noun, verb, and adjective, it has long been known that there is no one-to-one relationship between part of speech and function. Consider the examples in (2a–g), from Van Valin (2008: 165). (2)

Chris will [nuc [pred see]] the movie. Chris [nuc is [pred a very good detective]]. Pat [nuc is [pred exceedingly tall]]. Pat [nuc is [pred in the house]]. Chris [nuc [pred wiped]] the table [nuc [pred squeaky clean]]. f. Pat [nuc [pred pushed]] the table [nuc [pred out the door]]. g. Chris [nuc was [pred elected]] [nuc [pred president of the club]]

a. b. c. d. e.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

PRED in PRED in PRED in PRED in PRED in PRED in PRED in PRED in PRED in PRED in

NUC ¼ V NUC ¼ NP NUC ¼ ADJP NUC ¼ PP NUC1 ¼ V NUC2 ¼ ADJP NUC1 ¼ V NUC2 ¼ PP NUC1 ¼ V NUC2 ¼ NP

Lexical and Grammatical Categories in RRG

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

RP

NUC

RP

RP

PRED

NUC AUX

PRED

V Chris will

see

NP the movie.

Chris

is

a very good detective

SENTENCE CLAUSE CORE RP

NUC AUX

PRED PP

Pat

is

in the house

Figure 2.1 Constituent projections for (2a), (2b) and (2d) (adapted from Van Valin 2008: 166)

As the examples in (2) show, the predicate of a nucleus in English need not be verbal. While the predicate is verbal in (2a), in (2b) it is an NP, in (2c) an AdjP and in (2d) a PP. Figure 2.1 presents the constituent projections for (2a) (from this author), and (2b) and (2d) (adapted from Van Valin, 2008: 166).4 In English, as in many other languages, there is a general requirement that clauses in which the predicate is not verbal must contain a kind of ‘auxiliary’ such as the copula is in order for the clause to be grammatical. As a very good detective in (2b), exceedingly tall in (2c) and in the house in (2d) are not verbs, the copula is is required; however, it is important to stress two things here. First, that while is is a verb, it is not the predicate; this function is filled by the NP in (2b), by the AdjP in (2c), and by the PP in (2d). Second, this is a language-specific criterion which is not found in all languages or which is only found with certain interpretations. For example, one quite common situation is found in Russian, where a clause whose predicate is non-verbal does not require a copula when there is a present or atemporal interpretation, as in (3). (3)

Russian eto stol this table ‘This [is a] table.’

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

185

186

JOHN M. PETERSON

In other languages, the element which is often considered a copula may not even be verbal, as in Maltese (Semitic, Malta), where the ‘copula’ is in fact a pronoun agreeing with the subject with respect to person and number (and gender in the third-person singular) with a present or atemporal interpretation, as in (4). (4)

Maltese (Ambros 1998: 69) Dan il-professur this.m.sg def-professor ‘This professor is Maltese.’

huwa 3m.sg

Malti. Maltese.m.sg

As Van Valin (2008: 166) notes, non-verbal predicates do not pose a problem in RRG, since the clause is not an endocentric category: ‘Indeed, there is no head at all. The nucleus cannot be considered the “head” of the core or the clause, because it is not a lexical category, on the one hand, and is often phrasal, on the other. The notion of “head” is of no relevance to the layered structure of the clause.’ Instead, all universal categories assumed in RRG are semantically motivated, so that in RRG a clause minimally consists of a core – which minimally consists of a nucleus which contains the predicate – and possibly one or more arguments. Instead of positing universal lexical categories which are also the heads of corresponding phrasal categories – cross-linguistically an even more problematic assumption – mainstream RRG has assumed since Van Valin (2008) that the core contains a predicate, the functional/semantic base of the nucleus, and possibly one or more referential phrases, referred to as RPs. Just as the predicate need not necessarily be verbal, neither does an RP have to be nominal. In addition, RRG assumes that these units may be modified by modifying phrases or MPs, which can (but need not) be specialized so as to modify only an RP, as with prototypical adjectives, or only predicative elements, as with prototypical adverbs, although again this is not necessary, as there are many languages with either no or only very few adjectives and/ or adverbs, where these functions are realized by other categories, such as nouns or verbs. Not all NPs are RPs and not all adjectives or adverbs are MPs. For example, expletive or ‘dummy’ subjects like those in (5a) are not referential and hence not RPs: these are pronouns and therefore NPs, but they are not referential and hence not RPs. Similarly, while a very good detective in example (2b) is an NP, repeated here as (5b), it is not referential and therefore also not an RP. (5)

a. It looks like it’s going to rain today. b. Chris is a very good detective.

The same holds for the relationship between adjectives / adjective phrases and the MP. For example, while the AdjP very good in (5b) is a modifier in the NP a very good detective, and hence an MP, the AdjP very good in (6) is not a modifier, and therefore not an MP.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

(6)

That detective is very good.

In languages which do not have the usual parts of speech such as noun, adjective or verb, these units will of course belong to different, languagespecific categories. For example, as we will see in Section 2.2.3.1 on Kharia, translation equivalents of nouns, adjectives and verbal stems all belong to the open class of contentive lexemes of that language, whereas NPs and finite verbs in English correspond most closely to the Kharia-specific units ‘Case-syntagma’ and ‘TAM/Person-syntagma’, respectively. Again, these structural units must be carefully distinguished from the functionally motivated units RP, MP and predicate. As mentioned above, modifying phrases or MPs in many languages have as their default value either adjectives or adverbs, depending on what is being modified, that is, a noun (the tall tree) or a verb (the mouse ran quickly into the closet). These appear in the peripheries of RPs and clauses (Van Valin 2008: 172).5 As Van Valin notes, MPs also have a layered structure, for two reasons. First, many languages allow modifier phrases with an adjectival nucleus to take a core argument, such as the German der auf seinen Sohn stolze Vater (lit. ‘the of his son proud father’). Secondly, the modifiers can themselves be modified, ‘which means that they must have a periphery to house the modifying MP, e.g. the very quickly extinguished fire’ (Van Valin 2008: 172). In addition to adjectives and adverbs, many languages, including English, also allow other, much more complex categories to serve as modifiers. Consider the following examples, from the same source. (7)

a. The Charles and Di syndrome is no longer relevant. b. The God is dead philosophers are mostly dead. c. My grandson likes to give me the who’s the boss now, silly old grandpa wink frequently.

In (7a) the MP contains a conjoined RP in its nucleus, that in (7b) contains a clausal nucleus, and the nucleus of the MP in (7c) contains a constituent question as well as a vocative in its post-detached position [PoDP]. Figure 2.2 gives the constituent projection of (7c) (PrCS ¼ pre-core slot). Even the seemingly inconspicuous category of pre-/postpositional phrases is not without problems. As Van Valin (2008: 170) notes, these are normally thought of as endocentric projections of a (pre- or postpositional) lexical head in predicative adpositional phrases, such as the underlined element in (8), which has a prepositional nucleus consisting of the preposition in, which functions as a predicate and thus licenses the object the library. (8)

Chris saw Pat in the library.

The status of this unit in (8) is markedly different from that of the underlined units in (9), from the same source, referred to as non-predicative PPs as

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

187

188

JOHN M. PETERSON

RP CORER PERIPHERY

NUCR

MP

N

COREM NUCM SENTENCE CLAUSE

PoDP

PrCS

CORE

PERIPHERY

RP

NUC

MP

AUX PRED RP

the who’ s the boss

RP

CORER

COREM NUCM

PERIPHERY

ADV

MP

now,

MP

silly old

NUCR N grandpa

wink

Figure 2.2 Structure of RP with MP containing sentential nucleus (adapted from Van Valin 2008: 173)

these units do not predicate but are licensed by the verb and therefore do not license an object. (9)

a. Chris showed the photo to Pat. b. Chris stole the photo from Pat. c. Chris presented Pat with the photo.

As the underlined units in (9) are non-predicative, they do not contain a PRED and hence neither a NUC p (nucleus of the PP) nor a CORE p . Rather, they consist merely of a pre- or postpositon and an RP. Figure 2.3 contrasts the internal structure of these two PP types (see also Chapter 10 for adpositional phrases in RRG). With respect to these two PP types, Van Valin (2008: 171) notes that predicative PPs most closely correspond to the X-bar notion of an endocentric category, as the nucleus here is always a P and always a head. He also notes that a modifier like right in right under the table can be handled in terms of a modifier in the periphery within the predicative PP, while with a non-predicative PP there is no prepositional nucleus, the only nucleus being that of the RP. As such, they can arguably be analysed as ‘exocentric PPs,

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

a.

PP

b.

PP COREP

P

RP

NUCP

RP

CORER

PRED

CORER

NUCR

P

NUCR

N to

the girl

N in

the library

Figure 2.3 Non-predicative (a) and predicative (b) prepositional phrases in English (Van Valin 2008: 171)

i.e. PPs without a prepositional nucleus. Hence there is no layered structure, and this predicts that the admittedly limited set of PP-internal modifiers should not occur with a non-predicative PP, which seems to be the case: *Chris showed the photo right to Pat/*Chris stole the photo right from Pat/*Chris presented Pat right with the photo.’

2.2.2 Gradience in Lexical Categories Any definition of lexical classes in terms of feature bundles will eventually run into problems as there will always be at least some content morphemes which show many but not all of the ‘prototypical’ features associated with a certain lexical category. For example, although most nouns in English and many other languages can appear in the plural, this is problematic with others, such as peace. Similarly, while most nouns in German are compatible with definite and indefinite articles, such as der/ein Mann ‘the/a man’, others, such as Verlaub ‘leave; permission’ are compatible with neither, in this case being restricted to occurrence in the fixed (prepositional) phrase mit Verlaub with the preposition mit ‘with’ and the meaning ‘with respect; if you will pardon my saying so’. Although Verlaub is listed in many dictionaries with the definite article der (masculine, singular), it is in fact probably never found with this article in actual speech, although it is considered a noun for other reasons, for example its compatibility with the preposition mit, which presupposes an NP (which is assumed to have a nominal head!) and its perceived inherent (masculine) gender, as inherent gender is restricted to nouns in German. Similar problems are encountered with what Ross (1972) refers to as ‘category squishes’ or sequences of forms whose feature bundles gradually change so that it is often difficult to decide which category they should belong to. Van Valin (2008: 174) cites the examples in (10).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

189

190

JOHN M. PETERSON

(10) a. b. c. c′. c′′.

The savant quickly proved the theorem That the savant quickly proved the theorem the savant/*the quickly proving the theorem the savant having quickly proved the theorem (I am aware of ) the theorem having quickly been proved by the savant. d. the savant’s/*the quickly proving the theorem d′. the savant’s having quickly proved the theorem d′′. the theorem’s quickly being proved by the savant e. the savant’s/the quick proving of the theorem e′. *the savant’s quick having proved of the theorem e′′. the quick proving of the theorem by the savant e′′′. *the theorem’s quick proving by the savant f. the savant’s/the quick proof of the theorem f′. the quick proof of the theorem by the savant f′′. the savant’s/the quick proof

(astonished the professors) (astonished the professors)

(astonished the professors)

(astonished the professors)

(astonished the professors) (astonished the professors)

Although proved in example (10a) is clearly verbal and proof in the (10f) examples is clearly nominal, the status of the forms ending in -ing is much less clear. While some instances of proving, such as those in (10c) and (10d), take a direct object, suggesting that this form is a verb, the forms in (10e) and (10e′′) require an ‘object’ marked by the preposition of, suggesting that this is a noun. Furthermore, the ‘subject’ NP in (10c) appears as a bare NP, while in (10d) – as in the more ‘nominal’ form in (10e) or the clearly nominal form proof in the (10f) examples – it appears in the genitive. These and other factors such as adjectival modification, optional complements, tense, aspect and voice are summarized in tabular form in Van Valin (2008: 175).6 As we see in (10), even in languages such as English, which otherwise seem to have quite clear lexical classes, it is often not feasible to assume that each and every word form unambiguously belongs to a particular lexical category. RRG thus assumes that it is the individual features which are relevant and not the labels themselves: What is relevant to the grammar is the features themselves, not any category labels that might be overlaid on them. So ‘noun’ is just a useful descriptive label for a certain pattern of lexical item distributions in a language, just like ‘subject’ is a useful descriptive label for a certain consistent pattern of restrictive neutralizations in a language. Analogous to grammatical relations, some languages have well-defined and sharply differentiated lexical categories, while others do not. (Van Valin 2008: 176) In the following two subsections, we discuss concrete examples from two other languages, Kharia (Munda, India) and Wari’ (Chapakuran, Brazil and Bolivia), where assuming clear-cut lexical classes and strictly endocentric syntactic categories is even more difficult, and we will show how these ‘problematic’ elements are in fact dealt with straightforwardly in RRG.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

191

Lexicon Open class (contentive morphemes, may be used in all functions)

Closed class Proforms / Deictics (referential and predicative function, used attributively only if marked by the genitive)

Grammatical Morphemes (never used referentially, predicatively, or attributively)

Figure 2.4 The Kharia lexicon (Peterson, 2011a: 78)

2.2.3

‘Problematic’ Categories

2.2.3.1 Parts of Speech in Kharia In the South Munda language Kharia, spoken in eastern-central India, Peterson (e.g. 2011a, b, 2013) argues that the familiar lexical categories of ‘noun’, ‘verb’, ‘adjective’ etc. are not only unnecessary for an adequate grammatical description of the language, assuming their presence in fact considerably complicates the discussion. Peterson divides the Kharia lexicon into two major classes: the vast majority of content morphemes, that is, morphemes denoting referents, attributes and events/states, may be used freely in referential, attributive and predicative functions with no overt derivational marking, ‘light verbs’, etc. This class is open as new morphemes (e.g. modern technology and loanwords in general) are constantly being added to its ranks. There is also a closed class which can be further divided into two subclasses: ‘Proforms/ Deictics’, which may be used freely in referential and predicative function, but may only be used attributively if marked for the genitive, and ‘grammatical morphemes’, which may never be used referentially, predicatively or attributively. This is illustrated in Figure 2.4. (11)–(12) present two examples demonstrating the flexibility of content morphemes. (11)

Kharia ((11a) from Peterson (2011a: 76); (11b) adapted from Malhotra (1982: 136)) a. lebu ɖel¼ki. man come¼mid.pst ‘The/a man came.’ ro ɖel¼ki. b. bhagwan lebu¼ki God man¼mid.pst and come¼mid.pst ‘God became man [¼ Jesus] and came [to earth].’

(12)

Kharia (Peterson 2013: 133) [In a play about me and you, in which both of us will be taking part.] ro am¼ga iɲ¼na¼m.’ ‘umboʔ. ‘naʈak¼te iɲ¼ga ho¼kaɽ¼na¼iɲ play¼obl 1sg¼foc that¼sg.hum¼mid.irr¼1sg and 2sg¼foc 1sg¼mid.irr¼2sg no

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

192

JOHN M. PETERSON

am¼na um¼iɲ pal¼e. ɖirekʈar seŋ¼gaʔ iɲ¼te ho¼kaɽ¼oʔ. 2sg¼inf neg¼1sg be.able¼act.irr director early¼foc 1sg¼obl that¼sg.hum¼act.pst am¼ga am¼na¼m.’ 2sg¼foc 2sg¼mid.irr¼2sg ‘“In the play I will be him and you will be me.” “No. I can’t be you. The director already made me him. You will be you.”’

This process is entirely productive, given a proper context. See examples (13)–(22), from Peterson (2011a: 75–92), which give some idea of the pervasiveness of this flexibility throughout the lexicon. (13)

Interrogatives: i ‘what; which; do what?’

(14)

Indefinites: jaha~ ‘something; some (attribution); do something’

(15)

Quantifiers: moɲ ‘one (referential/attributive); become one’

(16)

Properties: rusuŋ ‘red (one); red (attributive); become red’, maha ‘big (one); big; grow, become big’

(17)

Proper names: aʔghrom ‘Aghrom (name of a town) (referential/attributive); come to be called “Aghrom” (middle voice), name [something] “Aghrom” (active voice)’

(18)

Status and Role: ayo ‘mother; become a mother (middle voice), accept someone as a mother (active voice)’

(19)

Deictics and proforms: iɖaʔ ‘yesterday; become yesterday (middle), turn (e.g. today) into yesterday (active, e.g. with God as subject)’

(20)

Physical objects and animate entities: kaɖoŋ ‘fish; become a fish (middle), turn into a fish (active)’

(21)

Locative: tobluŋ ‘top, rise (middle), raise (active)’

(22)

Activities: siloʔ ‘plowing (n.); plowed; plow’

In addition to this precategoriality, what appear to be entire NPs can also denote an event. Consider examples (23)–(24), both of which have the same propositional content. (23)

Kharia (Peterson 2013: 134) ho rochoʔb¼te col¼ki¼ɲ that side¼obl go¼mid.pst¼1sg ‘I went to that side’

(24)

Kharia (Peterson 2013: 134) ho rochoʔb¼ki¼ɲ that side¼mid.pst¼1sg ‘I went to that side’ (lit.: ‘I that-side-d’)

Example (24) shows that an analysis of Kharia as possessing an N/V distinction is problematic. Even if we were to treat rochoʔb ‘side’ in (24) as a (presumably zero-derived) verb, we would then have a verb modified by the demonstrative ho ‘that’. As (25)–(26) show, the construction in (24), with https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

an apparent NP as the semantic base of the predicate, is entirely productive and can also contain both quantifiers and genitive attributes, in addition to demonstratives. This is especially problematic for assuming ‘rampant zero conversion’ (Evans and Osada 2005), as this would entail productive zero derivation of a verb (i.e. a lexical stem), from a full-fledged NP (i.e. from a complex syntactic unit). (25)

Kharia (Peterson 2013: 134) ubar rochoʔb¼ki¼ɲ two side¼mid.pst¼1sg ‘I moved to both sides (i.e. this way and then that).’

(26)

Kharia (Peterson 2013: 135) a. oʔ¼yaʔ teloŋ¼ki. house¼gen roof¼mid.pst ‘The house’s roof was thatched.’ b. oʔ¼yaʔ teloŋ¼oʔ¼ki. house¼gen roof¼act.pst¼pl ‘They thatched the house’s roof.’

On the basis of data such as these, Peterson (2013) assumes two structurally defined categories, the Case-syntagma and the TAM/Person-syntagma, both of which have the same potential structure for the semantic base and which differ only with respect to their functional marking, namely case or TAM/voice and person/number/honorific marking, motivated by similar structures assumed for Tongan by Broschart (1997). The structure of the semantic base is given in (27). None of these elements is obligatory, as long as some non-enclitic unit is present. The Kleene star (*) denotes that potentially any number of content morphemes (lex) is possible, including zero. (27)

(lex¼gen) (dem) (quant (clf )) (lex¼gen) (lex*) (¼poss) (¼num)

The maximal potential structure of the Case-syntagma is given in (28), where X is equal to the structure of the semantic base in (27). (28)

X¼case

Case in (28) refers to postpositions such as buŋ ‘with; ins’, seŋ ‘before’, etc., as well as the two cases given in (29), so that any postposition, ¼te, or the lack of any overt marking signals that the unit is a Case-syntagma. (29)

Case: Direct (zero marking)7  the case of subjects and indefinite direct objects; Oblique (marked by ¼te)  marks definite direct objects, ‘indirect objects’ and adverbials

The genitive is not a case in this sense: whereas the direct and oblique cases are relevant at the clause level to mark the relation of a constituent to the predicate (roughly: subject/non-subject), the genitive is only relevant within the semantic base and serves to integrate one semantic base into a larger semantic base. The oblique case also cannot appear in a TAM/Personsyntagma, as (30) shows, whereas the genitive can (31).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

193

194

JOHN M. PETERSON

(30)

Kharia (Peterson 2013: 139) *sahar¼te¼ki¼ɲ. city¼obl¼mid.pst¼1sg ‘I went to the city.’

(31)

Kharia (Peterson 2013: 139) ayo¼yaʔ¼yoʔ mother¼gen¼act.pst.3sg ‘he or she made [it] mother’s’ (lit. ‘he or she mother’s-ed [it]’)

The structure of the TAM/Person-syntagma is given in (32), where X again refers to the semantic base given in (27). (V2) or ‘vector verbs’, as they are often referred to in South Asian studies, refers to phonological and morphosyntactic words which mark Aktionsart or the passive. voice refers to basic voice, that is, the active and middle (Klaiman 1991).8 (32)

X (V2*) (¼prf )¼tam/voice¼person/num/hon

Examples (33)–(36) illustrate this structure with a number of TAM/Personsyntagmas in Kharia. The semantic base is underlined. (33)

Kharia (Peterson 2013: 142) col¼ki¼may go¼mid.pst¼3pl ‘they went’

(34)

Kharia (Peterson 2013: 142) goʔɖ¼ki boksel¼nom sister.in.law¼2poss c:tel¼mid.pst ‘she became your sister-in-law’ (¼ ‘she your sister-in-law-ed’)

(35)

Kharia (Peterson 2013: 142) ho¼jeʔ u¼jeʔ¼ki that¼sg.nhum this¼sg.nhum¼pl ‘that became these’ (¼ ‘that these-d’)

(36)

Kharia (AK, 1: 69)9 ~ ro tama am¼pe u naw kuʈum¼te¼ga and now 2¼2pl this nine family¼obl¼foc sadi biha kersoŋ¼na¼pe . . . marry marry marry¼mid.irr¼2pl ‘And now you will marry in only these nine families . . .’

goʔɖ¼ki c:tel¼mid.pst

Examples (37)–(38) show a Case-syntagma and a TAM/Person-syntagma with the same semantic base (underlined), differing only with respect to their functional marking. Their basic structures are illustrated in Figure 2.5.10 (37)

Kharia (elicited) bides¼aʔ lebu¼ki¼yaʔ rupraŋ¼te abroad¼gen person¼pl¼gen appearance¼obl ‘the appearance of foreigners’ (e.g. as the object of the predicate)

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

CASE-SYNTAGMA SEMANTIC BASE

CASE

bidesa lebukiya rupra

=te

TAM/PERSON-SYNTAGMA SEMANTIC BASE

bidesa lebukiya rupra

TAM/PERSON-MARKING =ki=may.

Figure 2.5 TAM/Person- and Case-syntagmas with the same semantic base

(38)

Kharia (Peterson 2013: 157) bides¼aʔ lebu¼ki¼yaʔ rupraŋ¼ki¼may. abroad¼gen person¼pl¼gen appearance¼mid.pst¼3pl ‘They took on the appearance of foreigners (e.g. by living abroad so long).’

These two syntactic units are defined purely structurally, and both can appear in attributive, predicative and referential functions, as examples (39)–(42) from Peterson (2013: 137) show. Case-syntagma in attributive function (39)

Kharia koloŋ daru kuda millet bread Tree ‘a millet bread tree’ (name of a children’s story) (Note: kuda modifies koloŋ and the two together modify daru.)11

TAM/Person-syntagma in attributive function (40)

Kharia (Peterson 2011a: 74) yo¼yoʔj lebu col¼ki. see¼act.pst.1sg man go¼mid.pst ‘The man I saw left.’

Case-syntagma in predicative function (41)

Kharia (AK 1: 57) . . . ro u¼ga ho jinis¼aʔ and this¼foc that animal¼gen ‘. . . and this [is] that animal’s meat.’

komaŋ. meat

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

195

196

JOHN M. PETERSON

TAM/Person-syntagma in referential function (42)

Kharia (MT, 1: 180) tomliŋ kunɖab aw¼ki behind qual¼mid.pst milk laʔ¼ki¼may ina no ipfv¼mid.pst¼3pl because

khaɽiya Kharia

gam ɖom¼na say pass¼inf

ɖel¼ki¼may. u¼ki tomliŋ uʔɖ¼ga this¼pl milk drink¼foc come¼mid.pst¼3pl ‘[Those who] were in the rear (¼ literally: ‘they were behind’) were called ‘Milk Kharia’ because they came drinking milk.’

Finally, an analysis of the forms given in the preceding pages as consisting of nominals and ‘hidden verbs’ is also not tenable. As noted above, it is not possible in Kharia to combine TAM/Person marking with case marking or with postpositions, so that a construction such as the one in (43), which ends in a postposition, is not grammatical if directly followed by TAM/Personmarking, as (44) shows.12 (43)

Kharia (Peterson 2013: 145) jughay duʔkho buŋ ikuʔɖ very much sorrow ins ‘very sad’, literally: ‘with very much sorrow’

(44)

Kharia (Peterson 2013: 145) jughay duʔkho buŋ goʔɖ¼ki. *ikuʔɖ very much sorrow ins c:tel¼mid.pst ‘(He) became very sad’ (¼ ‘with very much sorrow’)

However, if the semantic part of this unit is re-ordered as in (45), with two constituents instead of just the one in (43), the unit which is not marked by a postposition may serve as the semantic base of a TAM/Person-syntagma. This is a common predicate type in Kharia (Peterson 2011a: 220). (45)

Kharia (RD, 1: 18) ikuʔɖ jughay gupa lebu duʔkho buŋ watch person sorrow ins very much ‘The shepherd (¼ watch person) became very sad.’

goʔɖ¼ki. c:tel¼mid.pst

At issue here is that the unit ending in the postposition may not be directly followed by a TAM marker, whereas the unit which does not end in a postposition/case marker can. Furthermore, if a ‘light verb’ such as hoy ‘become’ (borrowed from Kharia’s Indo-Aryan neighbour Sadri) is inserted between buŋ ‘ins’ and goʔɖ¼ki in (44), this results in a grammatical predicate, consisting of a Case-syntagma and a marker of qualitative predication (
copula).13 (46)

Kharia (Peterson 2011a: 111) jughay duʔkho buŋ ikuʔɖ very much sorrow ins ‘(He) became very sad.’

hoy become

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

goʔɖ¼ki. c:tel¼mid.pst

Lexical and Grammatical Categories in RRG

Thus ‘hidden verbs’ in Kharia would have a very different distribution from the ‘light verb’ hoy ‘become’ or in fact any other supposed verb. They could only be assumed to occur after ‘predicate nominals’ which are not case-marked (e.g. for the structure in (34), i.e. those units we consider to be the semantic base of either a TAM/Person- or a Case-syntagma), but not in structures such as (44), so that they would also not be able to occur where the overt ‘light verb’ hoy does (compare (44) with (46)). Thus, hidden verbs do not simplify the analysis but rather complicate it considerably, as they would not only be non-overt but would also have a unique distribution. Example (47) presents a simplified example of an intransitive sentence in Kharia and Figure 2.6 presents its constituent projection in the present analysis. Kharia (adapted from Roy and Roy 1937: 180–181) u¼kiyar taŋ el¼aʔ sori¼ga ɲog¼e uɖ¼e¼kiyar. this¼du now 1pl.excl¼gen com¼foc eat¼act.irr drink¼act.irr¼du ‘They two will now eat and drink with us.’

(47)

RRG can thus easily account for analyses of individual languages which do not have supposedly universal categories such as ‘noun’, ‘adjective’ and/or ‘verb’, without needing any empty categories or hidden verbs. Finally, there is a rather marginal type of TAM/Person-syntagma in Kharia, quotatives, in which the semantic base consists of an entire sentence, as in example (48).

SENTENCE PERIPHERY

CLAUSE

RP

PERIPHERY

CORE

CASESYNTAGMA

CORER

TAM/PERSONSYNTAGMA MP

PP

NUC

NUC

RP

CORE P CASESYNTAGMA NUCR

COREM NUCM

u=kiyar

RP

ta

NUCP PREDP

PRO el=a

P sori=ga

Figure 2.6 The constituent projection of (47)

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

PRO og=e

u =e

=kiyar.

197

198

JOHN M. PETERSON

(48)

Kharia (Kerkeʈʈ¯ a 1990: 31) iɖib tunboʔ ‘kersoŋ¼e la! kersoŋe la!’ loʔ¼na¼kiyar. night daytime marry¼act.irr voc rep cont¼mid.irr¼du ‘Day and night they both will keep on [say]ing “Marry! Marry!”’

Although Peterson (2013) offers no final analysis of this form, as it is unique in his corpus, the next section, which deals with very similar predicates in the Chapakuran language Wari’ in which these structures appear to be much more common, offers a possible solution to analysing predicates of this type in Kharia.

2.2.3.2 The ‘Intentional State Constructions’ in Wari’ A similar construction to the rather marginal Kharia quotative predicate discussed in the previous section is the much more productive ‘intentional state construction’ (ISC) in Wari’, a member of the Chapakuran family of Brazil and Bolivia, to which we now turn. As Everett (2008: 383) notes, in Wari’, a predicate-initial language (VOS), in verb-initial sentences the verb is immediately followed by the VIC or ‘verbal inflectional clitic’, which marks for tense, voice, person, number and gender and agrees with both subject and object. See examples (49)–(50), where the VIC is underlined. (49)

(50)

Wari’ (Everett 2008: 383) xirim te Quep na-in do 3sg.real.pst/prs-3n house father.1sg ‘My father made a house long ago.’

pane rem.pst

ta. emph

Wari’ (Everett 2008: 384) wao’. Ten ta weave pass.3sg type of basket ‘Baskets are woven.’

In sentences in which the verb is not the first sentence constituent, for example with constituent questions, this sentence-initial unit is followed by a different class of clitics which mark tense and which agree with the gender of the item in sentence-initial position, regardless of its function in the sentence, as in (51). (51)

Wari’ (Everett 2008: 385) Ma’ co that.prox.hearer m/f.real.pst/prs ‘Who is speaking?’

tomi’ speak

na? 3sg.real.pst/prs

As Everett (2008: 386) writes, ‘Many Amazonian languages report on others’ thoughts, character, reactions, and other results of intentional states by means of quotatives, i.e. literally putting words in people’s mouths.’ Wari’ also makes use of quotatives for these purposes, but their use is apparently much more common in that language, having also spread to other uses. What is interesting is that in Wari’, similar to the Kharia

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

199

example in (48), there is no morpheme denoting ‘say’. (52) provides an example of this construction in Wari’, where the indices i, j and k show the respective congruence relationships. (52)

Wari’ (Everett 2008: 386) Ma’ co that.prox.hearer m/f.real.pst/prs

mao go.sg

na -ini 3sg.real.pst/prs-3n

Guajarái naj -namk ‘oro Guajará (Brazilian city) 3sg.real.pst/prs-3pl.f collective ‘“Who went to Guajará?” (said) the chief to the women.’

narimak’ woman

taramaxiconj. chief

Everett (2008: 391–392) notes a number of similarities between the ISC predicator and simple verbs, suggesting that the clausal semantic base of the ISC occupies the same functional slot as verbs otherwise do, namely the predicate, although it consists of an entire clause. These include:14 • the ISC predicate appears in the clausal position otherwise occupied exclusively by the verb • the last syllable of the ISC predicate is stressed, as with the verb in other sentence types. This suggests that the ISC predicate is considered a single unit • the predicate of an ISC may be modified just like a verb • the predicate of an ISC may undergo compounding, like all verbs. If the ISC predicate is considered the predicate, just like verbs in verb-initial sentences, the following generalizations may be made for constituent order in Wari’ (Everett 2008: 392): • Wari’ sentences always begin with a verb, ISC predicate, or preverbal mood marker • the VIC appears directly after the predicate, whether this is clausal or a simple verb • tense appears in sentence-second position. Everett (2008: 399) argues that Wari’ inflectional clitics – that is, those which follow the predicate of the clause – follow the NUC in an RRG analysis. As such, they follow the ISC predicate as well, which then occupies the NUC slot. Figure 2.7, adapted from Everett (2008: 398),15 shows the structural analysis which he assumes for example (52), where the entire ‘quote’ is viewed as a nucleus.16 The structure in Figure 2.7 captures the fact that the nucleus of the clause, which contains the predicate, consists of a form which, despite its role as the predicate, also has many of the trappings of a clause, as it consists of a precore unit and a core, which itself contains a nucleus (predicate) and two arguments. Nevertheless, it is clearly the predicate of the larger clause. This then, together with the two RPs narima’ and taramaxicon, forms the core of the higher-level clause.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

200

JOHN M. PETERSON

SENTENCE CLAUSE CORE NUC

RP

RP

narima’

taramaxicon

PRED CLAUSE PreCore

CORE NUC

NP

PRO RP

PRED V

Ma’

co mao na-in Guajará nanam ‘oro

Figure 2.7 Constituent projection of (52) (adapted from Everett 2008: 398)

2.3

Grammatical Categories: The Operators

A great deal of research has traditionally gone into modelling an appropriate representation of the argument structure of predicates – both in RRG and other syntactic theories. As discussed in Chapter 6 of the present volume, in RRG this information is stored in the lexicon in the form of logical structures, which contain information on the basic meaning of the predicate, its Aktionsart class(es), and the number of arguments and their thematic relations. The predicate and its arguments provide us with what can be termed the ‘sentence base’ (see Maas 2004), that is, the basic propositional information, but without any information relating to the specifics of the utterance (i.e. is the information a command, a request, a statement or a question? Does it refer to a past, present, future, iterative or hypothetical event/state? Are the referents identifiable, etc.?) Information of this type is referred to as ‘sentence modality’ or ‘semantic finiteness’ in Maas (2004), to distinguish it from the overt marking of this information, which he refers to as ‘morphological finiteness’. This information is also often referred to collectively as the ‘grounding’ or ‘anchoring’ of the sentence. This can be portrayed as in Figure 2.8. This grounding corresponds closely to Tomasello’s (2008) three kinds of ‘syntax’ , that is, ‘simple’, ‘serious’ and ‘fancy syntax’, which accompany what he views as the three basic evolutionary foundations of human cooperative communication:

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

Sentence Sentence base Predicate A1, A2, A3, …

Sentence modality Speech act type (command / request, declarative, question, …)

Time Questions of reference … of referential state / identifiable event …

Figure 2.8 The sentence base and sentence modality

Thus, since requesting prototypically involves only you and me in the here and now and the action I want you to perform, combinations of natural gestures and/or linguistic conventions require no real syntactic marking but only a kind of ‘simple syntax’ in a grammar of requesting . . . But when we produce utterances designed to inform others of things helpfully, this often involves all kinds of events and participants displaced in time and space, and this creates functional pressure for doing such things as marking participant roles and speech act functions with ‘serious syntax’ in a grammar of informing. Finally, when we want to share with others in the narrative mode about a complex series of events with multiple participants playing different roles in different events, we need even more complex syntactic devices to relate the events to one another and to track the participant across them, which leads to the conventionalization of ‘fancy syntax’ in a grammar of sharing and narrative. (Tomasello 2008: 244–245).17 In RRG, this information is divided into two broad categories: the first group we will refer to as ‘proposition-grounding’, as these categories ground the event or state with respect to time, modality, speech act, etc. The second group we will refer to as ‘referent-grounding’, as it refers to referential identifiability.

2.3.1 Proposition-Grounding Operators We begin with a discussion of propositional grounding, which is expressed in individual languages by language-specific grammatical categories, such as present tense, past tense, hearsay, mirative, etc. The grounding information expressed by these language-specific categories is divided into three different groups according to which level of the layered structure of the clause (LSC) they ground, the nucleus, the core or the clause. RRG assumes that individual languages will differ as to which of these grounding concepts are expressed through grammatical categories, and there is no requirement that every language have all of the possible operators. In fact, most languages do not have grammatical categories expressing all of these semantic categories,

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

201

202

JOHN M. PETERSON

Table 2.2 Operators in the layered structure of the clause, from Van Valin (2005: 9) Nuclear operators: Aspect Negation Directionals (only those modifying the orientation of an action or event without reference to the participants) Core operators: Directionals (only those expressing the orientation or motion of one participant with reference to another participant or to the speaker) Event quantification Modality (root modals, e.g. ability, permission, obligation) Internal (narrow scope) negation Clause operators: Status (epistemic modals, external negation) Tense Evidentials Illocutionary force (IF)

although it is assumed that all languages will have grammatical means to express at least two of these categories – negation and illocutionary force (Van Valin 2005: 9). These operators are summarized in Table 2.2. Nuclear operators such as aspect or directionals modify the state/event itself without reference to the participants, while the core-level operators modify the relation between the arguments and the event, most notably directionals, (deontic) modality (‘be able’, ‘must’, etc.), narrow scope negation, and event quantification. Finally, operators which have scope over the entire clause include ‘status’ (epistemic modality and external negation), as well as tense, evidentials and illocutionary force. Operators are represented by an operator projection, given below the respective language example. This projection is joined to the constituent projection through the nucleus. Figures 2.9 and 2.10 illustrate this for English and Japanese. Note that in the left-hand example in Figure 2.9 the information for the aspect operator comes from two different positions: the auxiliary be preceding leaving and the suffix -ing on this form. In fact, this information can potentially appear anywhere in the sentence. The only claim that RRG makes with respect to the placement of operators is that ‘the ordering of the morphemes expressing operators with respect to the verb indicates their relative scopes. That is, taking the nucleus as the reference point, the morphemes realizing nuclear operators should be closer to the nucleus than those expressing core operators, and those manifesting clausal operators should be outside of those signaling nuclear and core operators’ (Van Valin 2005: 11). All evidence so far suggests that this is true of language in general. Although there may be language-specific variation of the ordering of the operators within a particular level, cross-linguistic data for operator marking appears to substantiate this claim for those markers for different levels which appear together on the same side of the nucleus: see examples (53)–(57) (from Van Valin 2005: 10).18

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

RP

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

NUC

RP

PRED

NUC AUX

PRED

V Will

PP

they have to be leav-ing?

ASP MOD TNS

Pat

in the library.

V

PP

NUC

NUC

CORE

CORE

CLAUSE

IF

was

TNS

CLAUSE

IF

CLAUSE CLAUSE

SENTENCE

SENTENCE

Figure 2.9 Constituent and operator projections for two English sentences (adapted from Van Valin 2005: 14)19

SENTENCE CLAUSE PERIPHERY

CORE

RP

RP

NUC PRED

PP Taroo ga NOM

V

Kazue no uti GEN

de hon o

house in

‘Did Taroo read a/the book at Kazue’s house?’

book

yon-da ka?

ACC

read-PST

Q

V NUC CORE CLAUSE CLAUSE

TNS IF

SENTENCE Figure 2.10 Constituent and operator projections in a Japanese sentence (adapted from Van Valin 2005: 14)

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

203

204

JOHN M. PETERSON

(53)

(54)

Kewa (Papua-New Guinea) (Van Valin 2005: 10) a. Íra-pa-niaa-ru. cook-prf-down-1sg.pst ‘I burned it downward (as a hill).’ (V-ASP-DIR-TNS) b. Íra-pa-saa-ru. cook-prf-up-1sg.pst ‘I burned it upward (as a hill).’

(V-ASP-DIR-TNS)

Turkish (Van Valin 2005: 10) a. Gel-iyor-du-m. come-prog-pst-1sg ‘I was coming.’

(V-ASP-TNS)

b. Gel-emi-yebil-ir-im. come-able.neg-psbl-aor-1sg ‘I may be unable to come.’

(V-MOD-STA-TNS)

(55)

Korean (Van Valin 2005: 10) An-tul-li-wu-(si)-lswuiss-cianh-ass-keyss-up-nita. neg-hear-caus-pass-(sh)-able-neg-pst-presum-pol-decl ‘(I) guess that (he) [HON] might not be heard.’ (NEG-V-MOD-NEG-TNS-EVID-IF)

(56)

English (Van Valin 2005: 10) a. He may be leaving soon. b. She was able to see them. c. Will they have to be leaving?

(IF/TNS-STA-ASP-V)20 (IF/TNS-MOD-V) (IF/TNS-MOD-ASP-V)

Tiwi (Van Valin 2005: 10) ŋ -ru-untiŋ-apa. ¯¯ 1sg-pst-prog-eat ‘I was eating.’

(TNS-ASP-V)

(57)

e

Cross-linguistic data on the position of operators (or markers of finiteness in general)21 within the sentence also suggests that these operators can theoretically appear at any position in the sentence and are not restricted to any particular position (e.g. adjacent to the verb, sentencefinal, or sentence-second (i.e. the Wackernagel position)), and it is only their respective order when they appear together on the same side of the predicate which is restricted. For example, in his cross-linguistic study of finiteness, Maas (2004) suggests the typology of morphological finiteness given in Table 2.3. Types I, II and III denote simple predicates which consist of a single predicating unit (generally a simple verb), while Types Ia, IIa, IIb and IIc are complex predicate types which consist of at least two parts (whether verbal or not) which together express a single predicate. Types I and Ia are typical of isolating languages. Here, neither the simple predicate, consisting of a single element (usually a verb) (Type I, cf. (58)a), nor either of the two or more parts of the complex predicate (Type Ia) is morphologically marked for ‘finite’ categories (cf. (58)b).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

Table 2.3 Predicate types with respect to ‘finiteness’ marking Type

Scheme

I Ia II IIa

Eþ Eþ Eþ Eþ

IIb

E þ P (Modifying-finite þ Modified)

IIc III

E þ P (Modifying-finite þ Modified-finite) (E þ P)-finite

P P (Modifying þ Modified) P-finite P (Modifying þ Modified-finite)

No (verbal) morphology No (verbal) morphology Morphologically finite predicate Modified element (¼ ‘lexical verb’) is finite, modifying (or ‘situating’) element is invariable (e.g. ‘particle’) Modifying element is finite, modified element is non-finite (¼ auxiliaryparticiple type) Both elements are finite Movable affixes

(Maas 2004: 379, modified and reprinted with kind permission by de Gruyter) E refers to the complement of a predicate, which may be either an argument or an adjunct; P denotes the predicate; þ refers to a loose concatenation of words and - to a narrow morpheme concatenation.

(58)

Mandarin Chinese a q` u a. Type I: t¯ 3sg go ‘s/he goes’ a y` ao q` u b. Type Ia: t¯ 3sg want/fut go ‘s/he wants to / will go’

In Type II, typical of fusional and agglutinating languages, the simple predicate is marked for all finite categories (Type II), while for the complex predicates in Types IIa–c either the modified element (i.e. the lexical verb), the modifying or situating element (e.g. the auxiliary) or both is/are marked for all finiteness categories. Note that only complex predicates of Type IIb correspond to the traditional notion of auxiliaries. Also, as the Maltese examples in (59b–c) and (61a–b) show, languages can belong to more than one category. (59)

Maltese (Semitic, Neo-Arabic, Malta) ji-kteb a. Type II 3sg.m-write.ipfv fut ‘he writes’ ji-kteb b. Type IIa sa fut 3sg.m-write.ipfv ‘he will write’ c. qed ji-kteb prog 3sg.m-write.ipfv ‘he is writing’

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

205

206

JOHN M. PETERSON

(60)

Type IIb: English a. She has written. b. I have gone.

(61)

Type IIc: Maltese a. ir-rid im-mur 1sg-want.ipfv 1sg-go.ipfv ‘I want to go’ (literally: ‘I-want I-go’) b. kon-t mur-t cop.pst-1sg go.pfv-1sg ‘I had gone’ (literally: ‘I-was I-went’)

Finally, in Type III at least some of the language-specific finiteness categories are marked elsewhere in the clause and not together with the other finiteness categories. For example, in (62), from Paez, a Chibcha language (possibly a language isolate) spoken in Columbia, all finiteness markers attach to the predicate in the default case, as in example (62a), with the form uʔxuetstxu ‘I went’, where the marker -txu ‘fac.1sg’ directly attaches to the verb form. In (62b), on the other hand, it attaches (in its phonologically determined alternative form -tx) to the nominal element nengasu ‘through Belcazar’ to mark this element as focused, while the verb form uʔxuets is now ‘partially finite’ (see the discussion in Maas 2004: 378). (62) Type III: Paez a. xuʔna yesterday ‘Yesterday I

(Maas 2004: 378) juʔ nava ʧamb-na uʔx-ue-ts-txu village-to go-ipfv-prog-fac.1sg them but went to the village, but the car was broken.’

kar:o car

suw-ku. broken-fac.3sg

uʔx-ue-ts juʔ saʔ tja-xu paʔ jaʔ-tx. b. nenga-su-tx Belcazar-through-fac.1sg go-ipfv-prog them and dem-from come already-fac.1sg ‘I passed Belcazar and am already back from there.’

As noted above, RRG does not posit a privileged position for such marking, and markers of these categories can appear almost anywhere in the sentence. This allows us to account for such cross-linguistic variability in a straightforward way, without having to resort to movement or other processes to derive the actual word order in a particular language. In fact, the only claim that RRG makes is that the ordering of these morphemes with respect to the predicate – when they appear on the same side of this unit – corresponds to the relative scopal properties of the individual operators. And in the case of the Paez example in (62) this is unproblematic, as the ‘mobile’ markers are clausal operators which appear in the final position when attached to the verb in (62a), after aspectual marking and therefore conforming to this constraint. The position of this marking in (62b) directly preceding the predicate is also not an exception to this rule, as here the marking of this clausal operator is not on the same side of the verbal root as the other operators and hence this constraint does not apply here.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

SENTENCE CLAUSE CORE PRED TAM/PERSON-syntagma NUC NUC NUC

RP

le fly

Ø 3SG

col kan =ki Ø go CONT=MID.PST DECL

‘It [= the parrot] continued flying away.’

NUC NUC NUC

ASP

CORE CLAUSE CLAUSE

TNS IF

SENTENCE Figure 2.11 Constituent and operator projections in Kharia

Finally, as discussed in Chapter 13 of this volume, the scopal properties of these operators play a pivotal role in the discussion of nexus and juncture relations in RRG. Consider Figure 2.11, from Kharia (own data), which provides an example of nuclear cosubordination, as the scope of kan, a marker of aspect/Aktionsart, has obligatory scope over both nuclei, leŋ ‘fly’ and col ‘go’. Operators are represented formally in the semantic representation of the sentence in italicized caps within angled brackets indicating their scope within the logical structure. The overall pattern is given in (63), from Van Valin (2005: 50), with arbitrarily chosen values. Note that each of the individual operators refers to a grammatical category within a specific language and as such its value is subject to the system of that language, for example, with respect to tense, which could be pst/npst in one language, pst/prs/ fut in another, or fut/nfut in yet another. (63)

〈IF DEC 〈EVID HS 〈TNS PST 〈STA IRR 〈NEG Ø 〈MOD OBLIG 〈EVQ SG 〈DIR Ø 〈ASP PERF 〈LS〉〉〉〉〉〉〉〉〉〉

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

207

208

JOHN M. PETERSON

(64), also from Van Valin (2005: 50), provides an example of the formal representation of the operators for the English question Has Kim been crying?. (64)

〈IF INT 〈TNS PRS 〈ASP PERF PROG 〈do′ (Kim, [cry′ (Kim)])〉〉〉〉

2.3.2 Reference-Grounding Operators As noted in Section 2.3, the RP also partakes in the semantic/pragmatic grounding of the clause, for instance with respect to definiteness. Table 2.4 summarizes those operators which are assumed for the RP in RRG. Table 2.4 is still somewhat tentative as it is based on the operators which RRG assumed for the NP before the notion of the RP was introduced in Van Valin (2005: 28) and further justified in Van Valin (2008). We thus tentatively assume that the same operators hold for the RP as for the NP, as we are dealing with the same underlying function here, namely the semantic/ pragmatic grounding of referential units, although this topic requires further study. As with the sentence, RPs also possess an operator projection. Figure 2.12 provides an example of such a projection for an RP in English. Similar to the position of sentence-grounding operators discussed in the previous section, there are no general requirements in RRG that the individual operators appear adjacent to any particular element(s) of the RP, only language-specific rules regulating this, although it is expected that operators appearing on the same side of the nucleus will be arranged according to their scope, with nuclear operators appearing closer to the nucleus, followed by core operators. To give an example of the positional variability of these operators, consider the different orders in the Dyirbal examples in (65). Dyirbal is spoken in northern Australia and has exceptionally free word order. Note that all three variants in (65) are grammatical and have the same meaning. Note also the position of the determiners with respect to the nouns which these modify. Table 2.4 Operators in the layered structure of the RP (adapted from Van Valin 2005: 24) NuclearR operators: Nominal aspect (count–mass distinction, classifiers in classifier languages) CoreR operators: Number Quantification (quantifiers) Negation RP operators: Definiteness Deixis

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

RP CORER

PERIPHERYR ADJ the

three big

NUCR N bridge-s N NUCR CORER

QNT DEF

NUM

CORER RP

Figure 2.12 Layered structure of the RP in English (adapted from Van Valin 2005: 25)

(65)

Dyirbal (Van Valin 2005: 28) a. Bayi bargan baŋul yaɽa-ŋgu d̨ urga-ɲu gambi-ɽa. det.abs wallaby.abs det.erg man-erg spear-tns mountains-loc ‘The man speared the wallaby in the mountains.’ b. Baŋul gambi-ɽa yaɽa-ŋgu bayi d̨ urga-ɲu bargan. det.erg mountains-loc man.erg det.abs spear-TNS wallaby.abs c. Yaɽa-ŋgu d̨ urga-ɲu gambi-ɽa bargan baŋul bayi. man.erg spear-TNS mountains-loc wallaby.abs det.erg det.abs

The operator projection of the various orders in (65) is straightforward and is given in Figure 2.13, which presents the projection of the structures in (65b). Similar to the proposition-grounding operators in the semantic representation of the sentence, the reference-grounding operators are represented in the semantic representation of the RP. An example is given in (66), from Van Valin (2005: 52), assuming that these are the same for the RP as those previously assumed for the NP. The conventions used here (i.e. angle brackets denoting scope, italicized small caps, etc.) are the same as those used to represent the proposition-grounding operators. 9 is the existential operator, roughly meaning ‘there is (at least one)’. (66)

〈 DEIC PROX 〈 DEF þ 〈 NEG Ø 〈QNT 9 〈 NUM SG 〈 nasp count 〈N〉〉〉〉〉〉〉

The logical structure for a phrase like the scarf would then be represented as in (67), also from Van Valin (2005: 52). (67)

〈DEF þ 〈NEG Ø 〈QNT 9 〈NUM SG 〈NASP COUNT 〈(scarf )〉〉〉〉〉〉

Although in principle every RP in a sentence has such a semantic representation, for practical reasons it is seldom feasible to include all of this

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

209

210

JOHN M. PETERSON

SENTENCE CLAUSE PERIPHERY

RPLOC RPERG

CORER

PRODEM

CORE RPERG

NUC

RPABS

CORER

PRED

CORER

NUCR

NUCR N

RPABS PRODEM

N

Ba ul

gambi- a

DET.ERG

mountains-LOC man-ERG

ya a- gu

V

N

bayi

durga- u

bargan.

DET.ABS

spear-TNS

wallaby.ABS

N

N

V

N

NUCR

NUCR

NUC

NUCR

CORER

CORER

CORE

RP DEIC

NUCR

CLAUSE RP

TNS CORE R

SENTENCE DEIC

RP

Figure 2.13 The structure of the Dyirbal sentence in (65b) (adapted from Van Valin 2005: 29)

information in the full semantic representation of the sentence, hence it is generally omitted. An example of a full structure is given in (68) for the simple intransitive question Has the tall man been crying? (Van Valin 2005: 53). The underlining of man in the semantic representation of the RP denotes that it is the head of the RP.22 (68)

〈IF INT 〈TNS PRES 〈ASP PERF PROG 〈do′ (x, [cry′ (x)] 〉〉〉〉 〈DEF + 〈QNT 9 〈NUM SG 〈NASP COUNT 〈be′ (man (x), [tall′])〉〉〉〉〉 -----

2.4

Questions for Future Research

Rauh (2010: 378–383) raises a number of important questions with respect to lexical and syntactic categories in RRG which need to be addressed and to which we now turn. These questions can be divided into two broad categories. The first are questions concerning the status of lexical and syntactic categories and how these relate to one another in RRG. For example, Rauh (2010: 380) raises a question with respect to examples (2b–d) above, repeated here in (69).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

(69)

b. Chris [nuc is [pred a very good detective]]. c. Pat [nuc is [pred exceedingly tall]]. d. Pat [nuc is [pred in the house]].

PRED in NUC ¼ NP PRED in NUC ¼ ADJP PRED in NUC ¼ PP

As Rauh notes, is in these examples is not viewed as the predicate in RRG, but rather the predicate is the NP, AdjP and PP, respectively. As such, the question arises as to how is is introduced into the sentence. The usual answer would be to assume that English, like many languages, requires a ‘dummy verb’ such as the copula when the predicate is not verbal. However, as Rauh rightly notes, if labels such as noun and verb are merely useful descriptive labels for certain language-specific feature-bundle patterns and not universally valid categories, this is not an option. As has been argued in the preceding pages, in RRG there is no need to specify that predicates which are traditionally viewed as consisting of NPs, AdjPs, PPs, etc., require a ‘dummy verb’ – in English or elsewhere. Recall from Section 2.2 that it is the features which are relevant to the grammar, not these convenient categorial labels (Van Valin 2008: 176). It would therefore be more correct to say that a copula is required with stative predication in English when the distributional properties of the content morpheme of the predicate are not compatible with TAM/person marking (*Pat (exceedingly) talls.). No reference is necessary to categories such as noun, adjective, verb, etc. Rather, the theory recognizes that individual content morphemes can have language-specific distributional properties, which can only be determined through tests and not through ‘logical’ conjecture, based, for example, on whether they denote an action, object, etc. These language-specific restrictions include the (non-)compatibility of a particular content morpheme with TAM or case markers, determiners, etc. However, as we saw in Section 2.3.1 for Kharia, this is not equivalent to saying that these content morphemes actually are nouns or verbs. Also, in view of the problems with ‘category squishes’ described in Section 2.2, where only some of the supposedly defining features of a particular category are present, direct reference to specific features as opposed to (at best more-or-less predictable) feature bundles provides us with a much more exact description of the grammatical operations involved. Another area mentioned by Rauh, qualia structure (Pustejovsky 1995), is of a very different nature and raises more serious questions, which we can only touch upon here. Qualia structure is one of four interpretive levels, the other three being argument structure, event structure and the interaction of semantic levels. This level consists of four basic roles (Pustejovsky 1995: 76): • • • •

CONSTITUTIVE: the relation between an object and its constituent parts FORMAL: that which distinguishes the object within a larger domain TELIC: its purpose and function AGENTIVE: factors involved in the origin or ‘bringing it about’.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

211

212

JOHN M. PETERSON

Rauh refers to qualia structure on two occasions with respect to open questions in RRG, highlighting what is in her view the importance of lexical categories in constructing the logical structure of the sentence from the lexicon. For example, ‘Lexically, the NPs should be represented by items which are described on the basis of qualia structures in the lexicon, since such a description identifies them as Ns and indicates their compatibility with those peripheries and operators that are N-specific rather than CLAUSE-specific’ (Rauh 2010: 380). Also, ‘To begin with, in order to construct the semantic representation of predicates and their arguments, a specification of nouns (items with a qualia structure) is required’ (Rauh 2010: 382). As just noted, we believe that there is no need to refer to lexical categories in describing the compatibility which Rauh is referring to, if such restrictions are described in terms of individual features and not lexical categories. The status of qualia structure, however, requires some discussion. To begin with, the presence of qualia structure is not restricted to ‘nouns’, since verbs also have qualia structure (see e.g. §6.2.5 in Pustejovsky 1995). Nor is event structure restricted to ‘verbs’, since nominalizations or nouns such as war in English also have event structures (see e.g. Pustejovsky, 1995, §§8.4 and 8.5). Of course, the structure and type of information contained at these levels depends among other things on whether an action or an object is being described, but it is worth recalling here that saying that a content morpheme denotes an action is not equivalent to saying that it is a ‘verb’ with a predictable set of distributional (and other) features, and to my knowledge, the lexical structures given in Pustejovsky (1995) do not contain any direct reference to such lexical categories, although they do, of course, contain information relating to semantic features, such as ‘physobj’, ‘process’, ‘state’, etc. This issue requires further study, and it is likely that we will have to reevaluate a number of assumptions with respect to semantic structure in order to be able to do away with lexical categories entirely. Furthermore, languages such as Kharia, discussed in Section 2.3.1, where virtually all content morphemes can be used predicatively, referentially and attributively, will undoubtedly present us with new challenges with respect to structuring the lexicon. However, as lexical categories do not play a very prominent role in Pustejovsky’s (1995) formalism, this does not appear to be an insurmountable task, although Rauh (2010: 378–383) is certainly correct in noting that this will require assuming a much richer lexicon in RRG, containing considerably more language-specific information on individual morphemes than is currently the case. At any rate, languages such as Kharia with its almost total lack of distributional restrictions on content morphemes, or Wari’ with its clausal predicates in the highly productive intentional state constructions, discussed in Section 2.3.2, as well as ‘category squishes’ even in languages where we have come to accept the presence of clear-cut lexical categories as self-evident, force us to accept that these

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

highly intuitive lexical categories cannot play a role in a theory of language with claims to universality. In sum, while much work remains to be done, it seems clear that traditional ‘categorial thinking’ cannot serve as the basis for a universal theory of language and is in fact at the heart of many descriptive problems which have plagued linguistic theory for decades. The view taken here is that these problems can only be overcome by the wholesale rejection of such categories in language theory, granting them nothing more than the status of convenient labels in informal discussion.

References Ambros, Arne A. 1998. Bon˙ gornu, kif int? Einführung in die maltesische Sprache. Wiesbaden: Reichert Verlag. Bisang, Walter. 2001. Finite vs. non finite languages. In Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, Vol. 2, 1400–1413. Berlin: Mouton de Gruyter. Broschart, Jürgen. 1997. Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1(2): 123–165. Croft, William. 2000. Parts of speech as language universals and as languageparticular categories. In Petra M. Vogel and Bernard Comrie (eds.), Approaches to the Typology of Word Classes (Empirical Approaches to Language Typology, 23), 65–102. Berlin: Mouton de Gruyter. Croft, William. 2005. Word classes, parts of speech, and syntactic argumentation. Linguistic Typology 9(3): 431–441. Evans, Nicholas and Toshiki Osada. 2005. Mundari: The myth of a language without word classes. Linguistic Typology 9(3): 351–390. Everett, Daniel L. 2008. Wari’ intentional state constructions. In Van Valin, Jr. (ed.), 381–410. H˚akansson, Gisela and Jennie Westander. 2013. Communication in Humans and Other Animals (Advances in Interaction Studies, 4). Amsterdam: John Benjamins. Haspelmath, Martin. 2007. Preestablished categories don’t exist: Consequences for language description and typology. Linguistic Typology 11(1): 119–132. Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1): 31–80. Hengeveld, Kees and Jan Rijkhoff. 2005. Mundari as a flexible language. Linguistic Typology 9(3): 406–431. Himmelmann, Nikolaus P. 2008. Lexical categories and voice in Tagalog. In Peter K. Austin and Simon Musgrave (eds.), Voice and Grammatical Relations in Austronesian Languages (Studies in Constraint-Based Lexicalism), 247–293. Stanford, CA: CSLI Publications.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

213

214

JOHN M. PETERSON

Hockett, Charles F. 1960. The origin of speech. Scientific American 203: 88–111. [Reprinted in Wang, William S.-Y. 1982. Human Communication: Language and Its Psychobiological Bases. Scientific American: 4–12]. Kerkeʈʈa, K. P. 1990. Jujhair ɖ¯ ãɽ (khaɽiy¯ a n¯ aʈak). Ranchi: Janj¯ at¯ıya Bh¯ as¯ a ˙ Ak¯adem¯ı, Bih¯ ar Sark¯ ar. Klaiman, M. H. 1991. Grammatical Voice (Cambridge Studies in Linguistics, 59). Cambridge: Cambridge University Press. Luuk, Erkki. 2010. Nouns, verbs and flexibles: Implications for typologies of word classes. Language Sciences 32(3): 349–365. Maas, Utz. 2004. ‘Finite’ and ‘nonfinite’ from a typological perspective. Linguistics 42(2): 359–385. Malhotra, V. (1982). The Structure of Kharia: a Study of Linguistic Typology and Language Change. Unpublished PhD dissertation. New Delhi: Jawaharlal Nehru University. Nikolaeva, Irina (ed.). 2007. Finiteness. Theoretical and Empirical Foundations. Oxford: Oxford University Press. Peterson, John. 2005. There’s a grain of truth in every ‘myth’, or, Why the discussion of lexical classes in Mundari isn’t quite over yet. Linguistic Typology 9(3): 391–405. Peterson, John. 2011a. Kharia. A South Munda language (Brill’s Studies in the Languages of South and Southwest Asia, 1). Leiden: Brill. Peterson, John. 2011b. Aspects of Kharia grammar – A Role and Reference Grammar (RRG) approach. In Rajendra Singh and Ghanshyam Sharma (eds.), Annual Review of South Asian Languages and Linguistics, 81–124. Berlin: Mouton de Gruyter. Peterson, John. 2013. Parts of speech in Kharia: a formal account. In Rijkhoff and van Lier (eds.), 131–168. Pustejovsky, James J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Rauh, Gisa. 2010. Syntactic Categories: Their Identification and Description in Linguistic Theories (Oxford Surveys in Syntax and Morphology, 7). Oxford: Oxford University Press. Rijkhoff, Jan and Eva van Lier. 2013. Flexible Word Classes: Typological Studies of Underspecified Parts of Speech. Oxford: Oxford University Press. Ross, John Robert. 1972. The category squish: Endstation Hauptwort. In Papers from the Eighth Regional Meeting. Chicago Linguistic Society. April 14–16, 1972, 316–328. Roy, Sarat Chandra and Ramesh Chandra Roy. 1937. The Kh¯ ari¯ as. Ranchi: Man ˙ in India. Tomasello, Michael. 2008. Origins of Human Communication (The Jean Nicod Lectures). Cambridge, MA: MIT Press. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic categories in Role and Reference Grammar. In Van Valin (ed.), 161–178.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface (Studies in Language Companion Series 105). Amsterdam: John Benjamins.

Notes 1 See, for example, Haspelmath (2007). 2 A notable exception to this is Croft (e.g. 2000, 2005), who approaches the topic from a cross-linguistic perspective and hence begins with semantics instead of language-specific morphosyntactic categories, assuming prototypical nouns (¼ objects), verbs (¼ actions) and adjectives (¼ attributes) and mapping language-specific forms/constructions against these. As what is at issue for our discussion at this point is the necessity (or rather, the lack thereof ) for assuming universally valid categories for language-specific feature bundles, we follow here the approach laid out in Himmelmann (2008). Further approaches to the topic abound in the literature; to cite just a few of the more recent typological works on this topic: Evans and Osada (2005) and the responses to this article (Peterson 2005; Croft 2005; Hengeveld and Rijkhoff 2005; Luuk 2010; and the many contributions in Rijkhoff and van Lier 2013). 3 In the following, we will assume that there is a clear distinction between ‘word level’ and ‘syntactic level’, a prerequisite for this analysis, although it is well known (see e.g. Haspelmath 2011) that the boundary between the two levels is often not as clear as generally assumed. 4 The operator projections have been omitted here as these are not discussed until Section 2.3. 5 For an example of this, see the three peripheries in Figure 2.2. 6 See Maas (2004) for similar data from German (p. 368) and Portuguese (p. 374). 7 Zero marking is restricted to the Case-syntagma, as the TAM/Personsyntagma is always overtly marked as such. Thus, the lack of any marking is unambiguous as it is only found with Case-syntagmas in the function of subject or indefinite direct object. 8 As the passive is marked by a V2, not by an obligatory fusional TAM/Voice marker, and always occurs together with the middle voice, it has a different status than the active and middle, which can be considered ‘basic’. 9 Sources given with this format are from this author’s own corpus. The letters are the speaker’s initials. The first number refers to the text from this speaker and the second number, after the colon, refers to the line of this text. 10 The triangles in Figure 5 indicate that there is further internal structure which we cannot deal with here but which does not affect the discussion at hand. For a fuller discussion, see Peterson (2013).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

215

216

JOHN M. PETERSON

11 In Peterson (2011b, §4.6.1) it is shown that kuda koloŋ daru is not a compound. 12 The following discussion and the corresponding examples are from Peterson (2013: 145–146). 13 This discussion and the accompanying examples have been taken from Peterson (2011a: 111). 14 For reasons of space, Everett’s argumentation has been somewhat simplified here. 15 Here and in the following, minor changes will be made to analyses and figures, such as the use of ‘RP’ as opposed to ‘NP’, as the concept of the RP was not fully justified until Van Valin (2008). Furthermore, minor changes with respect to glosses, etc., have silently been made and will not be commented on in the following figures. 16 The representation of the operators has been left out of Figure 2.7, as these are not dealt with until Section 2.3. 17 Of these three functions, the last two, i.e. helpful informing and the narrative mode, seem to be species-specific and are not found even with our closest evolutionary relatives, the non-human primates. For example, Tomasello (2008: 38) cites studies on ‘linguistic apes’, i.e. human-raised apes which could communicate through signs, in which 96–98% of all signed productions were imperatives, with the remaining 2–4% having no clear functional interpretation but with no clear cases of helpful information or narratives. Similarly, the data cited in H˚akansson and Westander (2013) contain no clear cases of non-human animal communication where helpful information is being offered (other than (largely involuntary) alarm signals) or narratives are being told. Hockett (1960 [1982]: 6) already notes that his design-feature of ‘displacement’ or ‘being able to talk about things that are remote in space or time (or both) from where the talking goes on’ is found not only in human communication but also in bee-dancing. The referential communication of bee-dancing, however, arguably does not correspond to what Tomasello (2008) refers to as ‘inform[ing] others of things helpfully’ (and certainly not to the ‘narrative mode’) but presumably to a prompt or command to other members of the hive to go to the place being described. With that, the last two functions in Tomasello’s (2008) analysis appear to be unique to humans. 18 For the original references, see the respective examples in Van Valin (2005: 10). 19 ‘[. . .] in English, illocutionary force is indicated by the position of the tense marker in the main clause: interrogative by core-initial tense, declarative by core-internal tense, and imperative by no tense.’ (Van Valin, 2005: 10, fn. 2). 20 See note 19.

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

Lexical and Grammatical Categories in RRG

21 Although the notion of finiteness is closely related to operators in RRG, the two are not entirely identical, as, for example, RRG does not view markers for honorificity (politeness, deferentiality) as operators, although these are generally viewed as markers of finiteness in languages where these are found, e.g. Korean or in many South Asian languages. For further discussion of finiteness from a typological perspective, see Maas (2004), Bisang (2001) and the studies in Nikolaeva (2007). 22 On the use of this notation, see Van Valin (2005: 52).

https://doi.org/10.1017/9781316418086.005 Published online by Cambridge University Press

217

3 A Conceptually Oriented Approach to Semantic Composition in RRG Ricardo Mairal Usón and Pamela Faber 3.1

Introduction

This chapter discusses semantic representation in Role and Reference Grammar (RRG) and how it can be enriched with a more conceptual or ontologically oriented approach.* An ontologically oriented model of semantic representation would provide a framework for the formalization of meaning construction. In fact, Van Valin and Mairal Usón (2014) argue that if the RRG linking system is given a more conceptual orientation, this can substantially enhance its semantic representations. Frameworks such as Wierzbicka’s (1996) Natural Semantic Metalanguage and Pustejovsky’s (1995) Generative Lexicon Theory have moved or are moving in this direction. A case in point is the Brandeis Semantic Ontology (BSO) (Havasi et al. 2007; Pustejovsky et al. 2006), a large lexicon ontology and lexical database. The BSO depends on the Generative Lexicon (GL), a theory of linguistic semantics that focuses on the distributed nature of compositionality in natural language. This is an issue that RRG will also have to address. Previous RRG research on the linking algorithm has primarily targeted the syntactic level (layered structure of the clause) and paid somewhat less attention to the semantic level (logical structure). However, semantic decomposition is an important issue in RRG because its goal is to draw empirically valid generalizations for as wide a range of languages as possible. A linguistic theory that aspires to universal applicability should base meaning construction and semantic representation on a set of basic concepts or near-primitives, which can be found in a wide range of languages. Although there have been various proposals for a finer-grained representation of states and activities (Van Valin and Wilkins 1993; Van Valin and

* This research was funded by the Spanish Ministry of Economy and Competitiveness: grants FFI2011–29798-C02–01 and FFI2014–53788-C3–1-P.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

LaPolla 1997) and of active accomplishments (Van Valin, 2005, 2018; cf. Bentley 2019), these issues still need to be explored in greater depth. It goes without saying that the most basic function of language is to convey meaning, and this involves considerably more than syntactic structures. At some level, an enriched lexical representation system would have to account for the encyclopedic knowledge that speakers possess, which enables them to choose the predicate that best represents the meaning they wish to convey. For example, native speakers of English are intuitively aware of the difference between related verbs such as those referring to loss of possession (e.g. steal, purloin, pilfer, filch, swipe) or those referring to the emission of a loud sound (e.g. scream, shriek, screech, bawl). However, the ability of language users to differentiate between related predicate meanings not only entails the consideration of single events (e.g. drink) as compared to others in the same lexical domain (tipple, sip), but also an awareness of the configuration of subevents designated by the same verb within a larger ‘frame’ (Fillmore 1977, 1982). Frame Semantics (Fillmore 2006) studies how linguistic forms evoke frame knowledge, and how the frames thus activated can be integrated into an understanding of the passages that contain these forms. Frames are crucial elements to consider in the representation of lexical meaning. For example, in certain cases, frames override constraints on the events that a single predicate can encode. According to Croft (1991, 2012), the only way in which two subevents can co-exist in the same verb is if they are causally related (e.g. smash, which encodes the action of smashing as well as its result). However, this is not necessarily true. Goldberg (2010: 44–49) states that the verb double-cross designates an event of betrayal following a state or event of understood cooperation. The betrayal is not caused by the state of trust, nor does the betrayal cause the state of trust. Instead the state of trust is part of the background frame that is presupposed in order for the profiled or asserted act to count as double-crossing. This one subevent is profiled, while the background frame presupposes one or more subevents without a causal relation between them. Frame knowledge is thus essential to understanding the meaning of the verb. A considerable part of a speaker’s language competence is the ability to choose the right predicate and to select the most suitable arguments to fill the slots activated by it in the communicative context. An effective lexical representation system should be capable of accounting for differences in lexical meaning within the context of a semantic domain and frame. The first step in designing a representation for this purpose would presumably involve the specification of a core set of basic concepts that serves as a foundation for meaning representations. This is the only way to obtain a formal representation with sufficient explanatory adequacy to account for those aspects of meaning that transcend the code itself and are part of the shared cultural knowledge and frames of a given speaker community.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

219

220

RICARDO MAIRAL USÓN AND PAMELA FABER

Indeed, RRG acknowledges the need for a more fine-grained system of semantic representation. In fact, Van Valin and LaPolla (1997: 114) admit that ‘there is as yet no adequate decompositional representation for the primitive state and activity predicates which are the building blocks of the system’. Despite the various proposals made to improve and enhance the decompositional representation in RRG, none of them has as yet been entirely satisfactory. The rest of this chapter is organized as follows. Section 3.2 gives an overview of semantic representation in RRG for all grammatical categories and discusses semantic decomposition. Section 3.3 explores how logical structures can be enhanced by conceptual information, more specifically by the ontology in FunGramKB. Section 3.4 lists the most relevant conclusions that can be derived from this chapter.

3.2

Semantic Representation in RRG

Since Foley and Van Valin (1984), the core component of RRG has been the bidirectional linking algorithm that captures how syntactic and semantic representations are mapped onto each other. This linking algorithm comprises a semantic level and a syntactic level, which model the communication act between a speaker and an addressee. The semantic phase of the linking algorithm begins in the lexicon, where a meaning representation is constructed from the information stored there. This representation is the input for the syntactic phase, which involves the assignment of syntactic functions and morphosyntactic properties. The syntactic phase, which proceeds in the opposite direction, starts with an utterance. After the application of a syntactic parser, the morphosyntactic properties of the input sentence are represented by means of the layered structure of the clause (LSC). This syntactic representation is a combination of both universal and language-specific distinctions. The interpretation of the sentence stems from mapping rules that link the syntactic representation to the semantic representation.

3.2.1 Semantic Representation of Predicates The classification of predicates in RRG is an enriched version of Vendler’s (1967) theory of Aktionsart. Van Valin (2005) augments the four basic classes (state, activity, achievement and accomplishment) with one more class, namely, semelfactives. He also distinguishes causatives as a parameter that crosscuts these classes.1 Each of these classes is represented by a logical structure (see Chapter 1). States and activities are primitives. In fact, they are the definiens or genus in the representation of achievements, semelfactives and accomplishments. For example, a state predicate, such as see is represented by see′ (x, y). The representation is composed of a primitive and two argument variables.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

Activity predicates are expressed with do′,2 which is a characteristic of this class of predicate, along with a primitive designating the activity carried out. Activity predicates can have one argument (do′ (x, [hiccup′ (x)])), or two arguments (do′ (x, [write′ (x, y)])). As previously mentioned, achievements, semelfactives, accomplishments and active accomplishments are based on states and activities in combination with operators. More specifically, an achievement comprises a state or activity predicate combined with INGR,3 indicating the sudden onset of the eventuality. For instance, the vase cracked is represented as INGR cracked′ (vase). Semelfactive predicates have a SEML operator and can be either activities or states. The activity designated by these predicates is punctual but does not cause a change of state or lead to a result. Thus, the janitor hiccupped would be SEML do′ (janitor, [hiccup′ (janitor)])). A semelfactive predicate can also be a state, for example flicker (SEML shine′ (x, y)). Accomplishments are represented by means of BECOME, which signals a change of state that reflects some sort of internal transition or transformation, for example ossify: BECOME hardened′ (x). In Van Valin (2005: 44, 2018), BECOME is broken down into PROC, the operator which indicates the duration of the change, and INGR, the operator which indicates the change of state itself. Active accomplishments are a complex structure, composed of an activity and an achievement, which reflects the telic nature acquired by the predicate in conjunction with a noun phrase or an adpositional phrase. For example, the witch brewed a potion is represented as follows: do′ (witch [brew′ (witch, potion)]) & INGR exist′ (potion). This formulation also includes the ampersand (&), which means ‘and then’. Finally, each of these predicate classes also has a causative version.4 Although verbs are extremely important, language competence also involves other categories. The RRG treatment of adjectives, adverbs and nouns is explained in Van Valin (2005: 48–51) and Van Valin and LaPolla (1997: 154–171). When adjectives function as predicates, they are treated as states in RRG logical structures (Van Valin 2005: 48). (1)

The chandelier is broken broken′ (chandelier).

Adjectives can also occur as part of attributive predications. Note that be′ is only found in the representation of attributive predications, not in the representation of the resulting state; moreover, be′ is not the representation of the copula, but the operator of a state, comparable to the operator do′. The operator be′ thus figures in these representations as the principal characteristic of attributive logical structures: (2)

a. Laura is wealthy. be′ (Laura, [wealthy′]) b. John is afraid of birds. feel′ (John, [afraid′ (birds)])

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

221

222

RICARDO MAIRAL USÓN AND PAMELA FABER

In regard to adverbs, Van Valin (2005: 49) explains that they are represented as one-place predicates and can modify different parts of a logical structure. For example, the argument of a time adverb can be the entire logical structure. In (3), the entire event of Lucas walking and then beginning to be at the zoo takes place within the time frame designated by tomorrow. (3)

Tomorrow, Lucas will walk to the zoo. tomorrow′ (do′ (Lucas, [walk′ (Lucas) ]) & INGR be-at′ (zoo, Lucas))

Generally speaking, manner adverbs modify activity predicates, whereas pace adverbs can modify any type of durational predicate. In (4), the activity of stealing, performed by Bugsy, is both an activity and durational predicate. As a result, it can be carried out in a certain way (furtively) and at a certain speed (fast): (4)

Bugsy furtively stole the jewels fast. furtive′ (do′ (Bugsy, [steal′, (Bugsy, jewels))]) CAUSE [fast′ (BECOME stolen′ (jewels))]

Aspectual adverbs modify basic state or activity predicates. In (5), there is an evident change of state since the water that was in the teapot is no longer in a liquid state because it has been transformed into vapour. (5)

The water in the teapot totally evaporated. BECOME (total′ (evaporated′ (water)))

In (6), a sustained activity is performed by the juror, who does not stop moving throughout the time period of the stated event (trial). (6)

The juror fidgeted continuously. continuous′ (do′ (juror, [fidget′ (juror)]))

Finally, Van Valin (2005) describes how predicative prepositions are represented and underlines the fact that they are also state predicates. As shown in (7), the logical structure be-in′ is the highest predicate since the grassland is the place where the kangaroo is jumping. As such, it has two arguments, namely, the grassland and the logical structure for jump. (7)

The kangaroo jumped in the grassland. be-in′ (grassland, [do′ (kangaroo, [jump′ (kangaroo)])])

This contrasts with the logical structure of the active accomplishment, the kangaroo jumped to the grassland, in which the predicative preposition (PP) indicates the final location of the kangaroo and not the location of the jumping event.

3.2.2 Semantic Representation of Nouns Nouns as a grammatical and semantic category have never been a focus in RRG since their syntax is less complex. Up until now, RRG has proposed a semantic representation based on the qualia analysis in Pustejovsky (1995).

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

223

Although, unlike verbs, nouns (e.g. cat and table) lack a logical structure, they have semantic properties in the form of qualia roles, which parameterize their meaning. According to Pustejovsky et al. (2006: 3), nouns can be described in terms of the following: • formal role: the basic type distinguishing the meaning of a word • agentive role: the factors involved in the object’s origins or ‘coming into being’ • telic role: the purpose or function of the object if there is one • constitutive role: the relation between an object and its constituent parts. Since the formal role refers to the type_of relation and the constitutive role refers to the part_of relation, this is still another indicator of an implicit typology of concepts underlying the semantic description of nouns. Just as important is the fact that the agentive role identifies a set of events linked to the object concept, whereas the telic role refers to an event description, which is associated with that object as its function (Pustejovsky et al. 2006: 333). According to Pustejovsky (2001: 5), qualia provide functional tags for words, which are linked to the network of concepts in Figure 3.1. Within the context of this concept lattice or network, each entity is described in terms of its qualia structure. For example, Figure 3.2 shows the representation of the qualia structure of violin, which is classified as a functional entity or artefact that has been created for a purpose. The formal role (F) indicates that violin is a type of musical instrument. Its agentive role (A) shows that a violin is the result of a building process. The telic role (T) reflects that the purpose of a violin is to produce music; and the constitutive role (C) indicates the parts of the violin, such as the strings.

T

Entity

Event

Quality

Natural Functional Complex Natural Functional Complex Natural Functional Complex Physical Abstract Direct Purpose coffee

knife

book

die

eat

Count Mass Info

Figure 3.1 Tripartite concept lattice (Pustejovsky 2001)

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

read

red

heavy dangerous good rising frightened

224

RICARDO MAIRAL USÓN AND PAMELA FABER

Figure 3.2 Qualia structure of violin (Pustejovsky and Batiukova 2019: 162)

The variables w, x, y and z refer to the different entities involved in events typically associated with a violin. The violin itself is assigned the variable x and is given a basic type in F; the agent building the violin is y (‘y builds x’), as expressed in A. The person using the violin in T to produce the musical sound on the violin is z (‘z produces sound on x’); and in C, the strings are identified as w (‘w of x’) (Pustejovsky and Batiukova 2019: 162). This overview of the RRG treatment of lexical units shows that words belonging to all grammatical categories would benefit from a system of semantic decomposition, which would make their meaning explicit and highlight the relations between words in the same category as well as with those in different categories. For example, such a system would show that predicates describing emotional states could be further decomposed in terms of feel, or those predicates that express possession could be decomposed in terms of have. Artefact entities such as table, gun and bread would be assigned to conceptual categories of furniture, weapon and food, respectively. Humans would also have different types of social roles (aunt, neighbour, friend, etc.) and professional roles (lawyer, zookeeper, hairdresser, etc.). Despite the fact that in RRG there is still no clearly defined system that indicates which words should be treated as primitives and which should be defined in terms of other more basic lexical units, this could be achieved by specifying the meaning representations of lexemes and linking them to an underlying ontological system.

3.2.3 Semantic Decomposition in RRG Although the use of English words to designate potential primitives is a notational convenience (Van Valin 2005: 46), the criteria used to select these primitives are somewhat less clear. Even though lexical primitives may have conceptual correlates, they still lack an ontology. There is also the need for research that will provide evidence of the scope of semantic primitives or at least give some indication of their potential cross-linguistic validity. This would doubtlessly contribute to clarifying the lexical–conceptual interface and how it works. In the RRG lexical representation system, another issue is why certain predicates undergo semantic decomposition whereas others do not. For example, in the lexical representations for love (8) and snore (9), the same

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

predicates (love′ and snore′) are used in their own definitions. Nonetheless, love, which is a verb of feeling, could be defined in terms of feel, whereas snore is a type of sound emission. (8)

love

love′ (x, y)

(9)

snore

do′ (x, [snore′ (x)])

The same also occurs in (10) and (11) except for the fact that the BECOME operator introduces the resulting state.5 BECOME soft′ (x)

(10)

soften

(11)

scatter [do′ (x, Ø)] CAUSE [BECOME scattered′ (y)]

These examples are in direct contrast to the representations in (12–14) in which each predicate is semantically decomposed. More specifically, learn is defined as to come to know, receive is to come to have, and show is to cause to see. In this causative sense, x does something that causes y to come to see z. (12)

learn BECOME know′ (x, y)

(13)

receive BECOME have′ (x, y)

(14)

show [do′ (x, Ø)] CAUSE [BECOME see′ (y, z)]

According to Van Valin (1993) and Van Valin and Mairal Usón (2014), it is necessary to improve the lexical representation system so as to obtain more consistent semantic decompositions. However, part of the problem lies in the fact that logical structures, as they were initially conceived, only focus on capturing grammatically relevant aspects of word meaning and neglect contextual meaning.

3.3

Towards a Conceptual Approach

Semantic decomposition of any kind is problematic because there is no simple solution for the atomization of meaning. Nevertheless, semantic features or attributes tend to pop up in different guises within a wide variety of approaches, such as Wierzbicka’s Natural Semantic Metalanguage (NSM) (see Goddard and Wierzbicka 2002), Mel’cuk’s Meaning Text Theory (e.g. Mel’cuk 1989, 2012), Jackendoff’s (1983, 1990, 1997) Conceptual Semantics, or Ontological Semantics (see Nirenburg and Raskin 2004). This is also something that Role and Reference Grammar has to deal with. Even though semantic decomposition may at first appear to be a difficult enterprise, the intuition persists that smaller meaning units must exist at some level to encode conceptual content. This means organizing the lexicon in lexical classes or domains. Each lexical domain has a basically hierarchical organization with other types of relations linking each predicate to others in a network structure. Factorization determines where the chain of the decompositional system ends.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

225

226

RICARDO MAIRAL USÓN AND PAMELA FABER

More specifically, conceptual information is encoded in a predicate definition, which to a certain extent is conceived as a frame with slots, having certain selection restrictions and default values. In this respect, each verb in the lexicon can be said to activate its own scenario, which determines its semantic participants. Designating a set of undefinables for a semantic metalanguage means the adoption of a system of lexical decomposition which can possibly function cross-linguistically. In this regard, each lexical domain has a set of functions that act on the superordinate term to generate more specific hyponyms and codify the most relevant subdomains. Ideally, the metalanguage established through semantic decomposition could also be used for conceptual representations. A more ontologically driven representation can thus provide a solution to the problems arising from the use in lexical representations of an unstructured inventory of semantic primes that are created as needed. Meaning definitions within the lexicon can and should point to the position of a concept within a network. This is in line with Pustejovsky (1995: 6) when he writes the meanings of words should somehow reflect the deeper conceptual structures in the cognitive system, and the domain it operates in. This is tantamount to stating that the semantics of natural language should be the image of nonlinguistic conceptual organizing principles, whatever their structure. Such a conceptual configuration can be organized onomasiologically (in meaning areas) rather than semasiologically (in alphabetical order). The definitions in an ideal dictionary would thus reflect conceptual categories, as codified in the genus or superordinate term of each definition. The features or properties that distinguish concepts from others in the same category would be linguistically represented by the clauses that modify the genus. Definitions would be coherent not only on a microstructural level, but also on a macrostructural one (Mairal Usón and Faber 2007). According to Levy (2003), it is only through the study of the usage of terms in a public language that we can have an independent way of fixing the contents of people’s concepts. This coincides with Dummett’s (1991) Priority Thesis, which takes language first and concepts second. Since dictionaries are the codification of conceptual content in public language, they are valid texts for the extraction of conceptual information regarding meaning parameters, arguments, semantic roles, semantic restrictions, etc. It is well known that categories have a basically hierarchical organization, given that hierarchies are central to cognition (Jackendoff 1997: 16). Within such hierarchies, concepts are related both vertically and horizontally by different types of conceptual relation. When this organization is applied to the structure of a semantic or conceptual domain, the resulting structure is

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

an ontology (in the artificial intelligence sense rather than the philosophical sense), defined by Gruber (1976) as an explicit specification of a conceptualization. One crucial property of a conceptual system is that no concept can be described without an account of its relationships to others (Lamb 1998: 147). When such representations are anchored to a well-designed conceptual ontology or network that relates word senses, the primitives in a conceptual lexical structure are no longer regarded as predicates but rather as conceptual units taken directly from an ontology.6 The ontology that can be used to specify the concepts in conceptual logical structures is FunGramKB, a multilingual online environment for the semiautomatic construction of a multipurpose lexico-conceptual knowledge base for natural language processing (NLP) systems (Periñan-Pascual and Arcas-Tuñez 2007, 2010; Mairal Usón and Periñan-Pascual 2009).7 The structure of FunGramKB has two levels: a linguistic level and a conceptual level. The three components of the linguistic level are the lexicon, grammaticon and morphicon, which store the lexical, grammatical and morphological information for each word. In a parallel way, the conceptual level is also composed of three modules: the ontology (concept hierarchy), the cognicon (scripts encoding procedural knowledge) and the onomasticon (proper names of entities and events). In the same way as other lexically based ontologies, the main premise of FunGramKB is that the conceptual representations of objects, attributes and events are mapped onto language in some significant way. The assumption is that the structure of the lexicon is based on a core set of undefinables or conceptual invariants and that these invariants can be extracted from lexicographic resources that document our shared knowledge of the world. Whereas the lexicon houses language-specific syntactic and morphological information, the language-neutral ontology captures the meanings and interrelationships of concepts that are not explicitly stated in the lexicon. The first step in building the ontology was the specification of an inventory of basic concepts that are assumed to be lexicalized in a wide range of languages. The ontology acquisition methodology is based on the extraction of type_of hierarchies from dictionary definitions (Martín-Mingorance 1984, 1990, 1995; Hirst 2009; Amsler 1980, 1981). Whereas the genus designates the superordinate concept of the defined word, the differentiating features are the properties that make the concept different from other members of the same conceptual category. The meaning of a word is thus an access point to a concept or conceptual structure of some kind. Pustejovsky (2001: 5) also claims that his qualia are linked to a network of concepts (though this relational structure is never specified). Consequently, all problems are transferred to the ontology since each word sense is represented as a pointer to some concept or category within the ontology. The FunGramKB ontology has the following levels. At the highest level are meta-concepts (e.g. #EVENT). At the next level are basic concepts

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

227

228

RICARDO MAIRAL USÓN AND PAMELA FABER

(e.g. þHUMAN_00), which are used to define other basic concepts and terminal concepts. At still another level are terminal concepts (e.g. $METEORITE_00), which are more specific and not used in the meaning postulates of other concepts. The final level of subconcepts (e.g. PASTEURIZE), has the same thematic frame as a terminal concept (i.e. $STERILIZE_00), but specifies one of its arguments to a greater degree. (In the case of PASTEURIZE, the second argument is restricted to milk products.) The concepts at the meta-conceptual, basic, terminal, and subconcept levels are preceded by the symbols, #, þ, $, and , respectively. Except for the subconcept level, each also has a numerical code, (_00, _01, _002, etc.). The methodology used to extract concepts consists of four phases: (1) conceptualization, (2) hierarchization, (3) remodelling, and (4) refinement.8 The source was the Longman Dictionary of Contemporary English (Procter 1978). All basic and terminal concepts in FunGramKB are described in terms of semantic properties represented in thematic frames and meaning postulates. Each event is assigned to one thematic frame. A thematic frame contains the number and type of participants (with selection restrictions) involved in the event or quality. Although not all participants are lexicalized, they are implicit to the understanding of the concept. The second type of conceptual schema is a meaning postulate, which is a set of one or more logically connected predications (e1, e2, etc.) with the generic features of concepts. Only one thematic frame and one meaning postulate are assigned to each concept. Meaning postulates are encoded in COREL (Conceptual Representation Language) (Periñán-Pascual and Mairal Usón 2010). This machine-readable language defines all of the units. It can also be used to obtain conceptual relations by applying inheritance and inference mechanisms to the meaning postulates (Periñán-Pascual and Arcas-Túnez 2005). In each representation, the meaning postulates provide the basis for natural semantic decomposition since each lexeme is linked to a concept.

3.3.1 Conceptual Logical Structures Adopting this conceptual approach would significantly modify RRG in that logical structures (LSs) would become conceptual logical structures (CLSs). Although CLSs resemble LSs in that they maintain the Aktionsart distinctions, the two differ because in CLSs, predicates become concepts. This type of ontologically based representation has the format in Table 3.1. As can be observed, in Table 1, the lexical primitives are replaced by concepts (), which provide a more granular semantic decomposition and represent any conceptual unit, whether basic or terminal. One of the advantages of this approach is that each lexical unit is linked to a conceptual unit and is enriched by the inheritance of the conceptual properties (i.e. thematic frame and meaning postulate) of the related conceptual unit. As widely discussed in Van Valin and Mairal Usón (2014), CLSs arise from the

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

Table 3.1 Inventory of conceptual logical structures Predicate class

Conceptual logical structure

State Activity Achievement

(x) or (x, y) do′ (x, [ (x) or (x, y)]) INGR (x) or (x, y), or INGR do′ (x, [ (x) o (x, y)]) SEML (x) or (x, y) SEML do′ (x, [ (x) or (x, y)]) BECOME (x) or (x, y), or BECOME do′ (x, [ (x) or (x, y)]) do′ (x, [ (x, (y))]) & BECOME (z, x) o (y) α CAUSE β, where α and β are logical structures of any type

Semelfactive Accomplishment Active accomplishment Causatives

interaction of information in the ontology and the lexicon. In order to ascertain how the ontology and the lexicon actually interact and arrive at what has come to be known as the lexical–conceptual interface, it is necessary to review what type of information is present in both the ontology and the lexicon. The choice of English as the conceptual language for lexical representation is not a problem since FunGramKB is based on the premise that at a certain level, all languages contain a set of core concepts that lie at the heart of communication. Since the ontology contains non-language-specific information, the language used to encode concepts is irrelevant (in fact, numbers could also have been used). Each concept derives its meaning from its position in the ontology as well as from its relations with other concepts. In other words, a concept is defined by its conceptual route. This is exemplified in the representation of $STERILIZE _00 in Figure 3.3. The ontology in Figure 3.3 represents $STERILIZE_00 as a material event that is a type of transformation process, represented by the meta-concept #TRANSFORMATION. At the more subordinate levels are the basic concepts, þCHANGE_00 and þCLEAN_01. As shown, þCHANGE_00 is a type of transformation, whereas þCLEAN_01 is a type of change in that an entity that is dirty is changed, so that the dirt is removed and it becomes clean. In Figure 3.3, $STERILIZE_00 is a daughter concept of þCLEAN_01. The cleaning process in $STERILIZE_00 involves killing bacteria (or something perceived to have bacteria). As such, $STERILIZE_00 is a terminal concept with various predicates bound to it, namely, fumigate, sanitize, disinfect, pasteurize, decontaminate and sterilize. Predicate selection is constrained by the context. World knowledge (not syntax) pertaining to the entity with bacteria, the sterilization instrument or method used, and the end result pursued guide the speaker’s choice of predicate. Although the meaning distinctions encoded in the predicates do not have an impact on syntactic form, this type of semantic–pragmatic information should be included in the RRG lexicon since it is a major component of lexical competence.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

229

230

RICARDO MAIRAL USÓN AND PAMELA FABER

Figure 3.3 FunGramKB Ontological Hierarchy for $STERILIZE_00

Moreover, the fact that lexical entries are linked to conceptual units and thus have different conceptual routes is a way of resolving polysemy. CLSs facilitate the selection of the most suitable concept in the case of predicates with more than one meaning. In this sense, each meaning of a predicate is linked to a different concept. The result is a robust enumerative lexicon with fine-grained meaning postulates, capable of generating a complex conceptual network. It is thus no longer necessary to condense all the meanings of a lexical unit into a single infra-specified representation as proposed by Generative Lexicon Theory. In CLSs, differences in meaning are not marked by the syntactic behaviour of a lexeme but rather are specified in its meaning. Accordingly, nuances of meaning resulting from differences in syntactic constructions do not affect the conceptual representation or the selection of the conceptual unit. This method of dealing with polysemy is exemplified by fix. For instance, one of the various meanings of fix is to repair something that is broken or not working properly (15). (15)

The plumber fixed the leaky tap.

Another meaning of fix is to prepare a meal or drinks. (16)

The chef fixed a dish of stewed artichokes.

FunGramKB links each of these senses to a different concept: þREPAIR_00 and þCOOK_00, respectively. Consequently, depending on the sense, fix has one of the following formats: (17)

a. do′ (x, [þREPAIR_00 (x, y)]) b. do′ (x, [þCOOK_00 (x, y) & become (y)]

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

The selection of (17a) or (17b) depends on the basic conceptual event since cooking is a creation event whereas repairing is a change event. Accordingly, þREPAIR_00 and þCOOK_00 differ in their conceptual routes and positions in the ontology. In the case of þCOOK_00, the conceptual path is #MATERIAL >> þDO_00 >> þCREATE_00 >> þCOOK. Its meaning postulate is the following: (18)

þ(e1: þCREATE_00 (x1) Theme (x2) Referent (f1: þHUMAN_00) Beneficiary (f2) Instrument (f3: þFOOD_00)Means (f4: (e2: þHEAT_00 (x1) Theme (x3: f3) Referent))Manner)

In contrast, the conceptual path of þREPAIR_00 is: #MATERIAL >> #TRANSFORMATION >> þCHANGE_00 >> þREPAIR_00. However, the change resulting from a þREPAIR_00 event is of a different type since it involves changing an entity that was damaged in the past so that it can function again. (19)

þ(e1: þCHANGE_00 (x1)Theme (x2)Referent (f1: (e2: past þDAMAGE_00 (x3)Theme (x2)Referent))Reason (f2: (e3: pos þOPERATE_00 (x4)Theme (x2)Referent))Result)

The existence of a previously specified ontology provides a more specific conceptual context that allows users to access concepts from a structured inventory. There is thus no need to create ad hoc primitives on demand since it is the ontology that determines the concept unit in the representation.

3.3.2 Case Studies: pasteurize, obsolete and invoice This section provides examples of the CLSs of verbs, adjectives and nouns.9 As previously mentioned, each lexical unit in the lexicon is linked to a conceptual unit in the ontology. For example, PASTEURIZE is a subconcept linked to the terminal concept, $STERLILIZE_00 and as such, it is marked with a dash (). Its conceptual route is #EVENT > #MATERIAL > #TRANSFORMATION > þCHANGE_00 > þCLEAN_01 > $STERILIZE_00. It does not appear in the basic hierarchical structure of the ontology since it shares the basic thematic frame and meaning of $STERILIZE_00. The difference lies in the fact that in PASTEURIZE, there is a specification of the second argument (milk). Words in different languages can thus be linked to PASTEURIZE, such as pasteurize (English), pasteurizar (Spanish), pastörize etmek (Turkish), 巴氏滅菌法 (Chinese). Example (20) shows a partial representation of the CLS for the predicate pasteurize. (20)

pasteurize: do′ (x [PASTEURIZE (x, y)])

In (20), pasteurize is represented as an activity predicate that is the lexical designation of the conceptual unit PASTEURIZE. However, this predicate can be decomposed further. Figure 3.4 shows the conceptual entry for

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

231

232

RICARDO MAIRAL USÓN AND PAMELA FABER

Figure 3.4 FunGramKB conceptual entry for PASTEURIZE

PASTEURIZE, which consists of a thematic frame and a meaning postulate. Thematic frames and meaning postulates are two complementary types of conceptual schema used to encode the meaning of the concepts in the ontology. The thematic frame for PASTEURIZE is inherited from the terminal concept $STERILIZE_00, which involves #TRANSFORMATION. In the thematic frame of #TRANSFORMATION concepts, there are two participants: (i) a Theme Entity that transforms another entity and (ii) a Referent Entity that is transformed by another entity. However, the selection restrictions for PASTEURIZE specify that the theme is human and the referent is restricted to milk. In Figure 3.4, the meaning postulate of pasteurize is composed of two events, the second of which is the purpose of the other. In the second event (e2), the theme (x1) kills a referent (x3), in this case, bacteria. Since PASTEURIZE is a type of $STERILIZE_00, which entails a cleaning event as well as a killing event, there is a semantic decomposition of this predicate into more basic units. In fact, it is possible to reach a semantic representation in COREL composed of semantic primitives by applying stepwise conceptual decomposition (Periñán-Pascual and Arcas-Túnez 2005), where the defining term in a meaning postulate can become the defined term in another. This process is based upon Dik’s (1978) stepwise lexical decomposition as applied in Faber and Mairal Usón (1999). However, instead of lexical items, the results of this process are regarded as conceptual units. This type of conceptual representation can also be used to represent qualities designated by adjectives such as obsolete. The ontological hierarchy in Figure 3.5 shows that obsolete is linked to the meta-concept of #QUALITY, and that it exists as a social perception #SOCIAL. An entity can only be

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

Figure 3.5 FunGramKB Ontological Hierarchy for concepts linked to $OBSOLETE_00

Figure 3.6 FunGramKB conceptual entry for $OBSOLETE_00

perceived as þUSELESS_00 (!obsolete)) within a social context in which something better has been created to replace it. In the same way as with PASTEURIZE, words from different languages can also be linked to $OBSOLETE_00 (e.g. Engl. obsolete, Sp. obsoleto, Fin. vanhentunut, Hung. elavult). The CLS for these predicates has the following format: (21)

$OBSOLETE_00 (x)

This CLS for obsolete consists of the conceptual unit $OBSOLETE_00, which, in turn, is composed of a thematic frame and a meaning postulate (Figure 3.6). Since it is a state, the thematic frame only has one participant (x1), which is a previously useful entity that is now socially perceived as useless. Obsolete is a predicate that can be applied to a wide variety of objects, such as typewriter. As is well known, a typewriter is a machine with keys that are

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

233

234

RICARDO MAIRAL USÓN AND PAMELA FABER

pressed to print letters of the alphabet onto paper. Since the 1990s, personal computers more efficiently perform the functions of typewriters. Thus, understanding why typewriters have become obsolete entails elements of world knowledge, such as an awareness of past and present time frames as well as of the affordances of typewriters in comparison to those of more modern machines that can be used for the same purpose. The meaning postulate in Figure 3.6 describes the three events that occur when an entity, such as a typewriter, is described as obsolete: (22)

(e1: þBE_01 (typewriter)Theme (x2: $OBSOLETE_00) Attribute) The typewriter is obsolete.

(23)

(e2: þBE_01 (typewriter)Theme (x2: þUSELESS_00) Attribute) The typewriter is useless (in the present).

(24)

(e3: þBE_01 (typewriter)Theme (x2: þUSEFUL_00) Attribute) The typewriter was useful (in the past).

As reflected in (22–24), $OBSOLETE_00 is a more specific type of þUSELESS_00 since it presupposes a time sequence or period during which the (once useful) entity has become useless. The representation in (25) portrays þUSELESS_00 as the negation of þUSEFUL_00. (25)

*(e1: þBE_01 (x1)Theme (x2: þUSELESS_00)Attribute) þ(e2: n þBE_01 (x1)Theme (x3: þUSEFUL_00)Attribute)

The conceptual entry for þUSEFUL_00 in (26) states that something is useful when it helps a human to do something. (26)

þ(e1: þBE_01 (x1)Theme (x2: þUSEFUL_00)Attribute) þ(e2: pos þHELP_00 (x1)Theme (x3: (e3: þDO_00 (x4: þHUMAN_00)Theme (x5)Referent))Referent)

All of this information is inherited by $OBSOLETE_00. When specifically applied to a typewriter, what is also activated is the list of affordances provided, which other more modern machines now perform more efficiently. Finally, nouns, such as invoice, are also represented in terms of a conceptual logical structure. From an ontological perspective, þINVOICE_00 is a basic concept (see Figure 3.7). It is a member of the meta-conceptual category of #PHYSICAL_OBJECT and more specifically, that of #SELF_CONNECTED_OBJECT. Its conceptual path is þINFORMATION_OBJECT_00! þWRITING_00! þDOCUMENT_00. An invoice is a type of information object, which is informative because of the writing in/on it, and which has the form of a document. Again, lexemes in different languages, such as invoice (English), factura (Spanish), faktura (Norwegian), ‫( ﻓﺎﺗﻮﺭﺓ‬Arabic), can be linked to this conceptual entry, which has the following CLS: (27)

þINVOICE_00 (x)

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

Figure 3.7 FunGramKB Ontological Hierarchy for þINVOICE_00

Figure 3.8 FunGramKB conceptual entry for þINVOICE_00

As a noun, invoice does not have a thematic frame. However, this does not mean that it lacks a relational structure. As a commercial document that itemizes a transaction between a buyer and a seller, invoice indirectly refers to a complex knowledge configuration, which speakers have to access in order to understand the term. This information should be included in the meaning postulate of the conceptual entry. Accordingly, the conceptual entry for þINVOICE_00 in Figure 3.8 is composed of a meaning postulate with two main events, which state what type of entity an invoice is (e1) and the context in which it is produced (e2).

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

235

236

RICARDO MAIRAL USÓN AND PAMELA FABER

This second event contains three subevents (e3, e4, e5). Examples (28–31) show each event and subevent within the corresponding frame, which encodes the purpose of an invoice (f1) and the possible reasons for issuing it (f2, f3). (28)

(e1: þBE_00 (x1: þINVOICE_00)Theme (x2: þDOCUMENT_00) Referent) [An invoice is a document.]

(29)

(e2: þGIVE_00 (x3: þHUMAN_00)Agent (x1)Theme (x3) Origin (x4: þHUMAN_00) Goal) [A human (x3) gives an invoice to another human (x4)]

(30)

(f1: (e3: þPAY_00 (x4) Agent (x5)Theme (x6) Origin (x3) Goal))Purpose [The purpose of an invoice is for x4 to pay x3.]

(31)

(f2: (e4: past þSELL_00 (x3)Agent (x7)Theme (x8)Origin (x4)Goal))Reason ^ (f3: (e5: past þWORK_01 (x3)Theme (f4: x4) Goal))Reason [The reason for issuing an invoice is because x3 sold something to x4 in the past or because x3 worked for x4 in the past.]

As reflected in (28–31), an invoice activates knowledge that is linked to a given sociocultural context. This shows that language does not exist in a void but is closely related to the culture and society in which it is used. Evidently, invoices would have little or no meaning for a tribe in the Amazon rainforest. In contrast, invoices are crucial in a sociocultural environment that contemplates commercial transactions involving the emission of documents that notify a receiver that payment is due for goods or services of some type. When this kind of encyclopedic knowledge is included in a conceptual entry, it provides users with the information necessary to understand the concept. As can be observed, within the context of an underlying ontology, a relatively simple representation, such as þINVOICE_00 (x), do′ (x [ PASTEURIZE (x, y)]), or $OBSOLETE_00 (x), allows users to access a rich variety of encyclopedic information regarding objects, processes and attributes. This is possible because of the thematic frames, meaning postulates and inference mechanisms stored in the knowledge base. In sum, the only difference from current LSs is that CLSs are made of conceptual units and not lexical words. CLSs maximize informativeness and minimize redundancy without losing the simplicity and elegance of the formalism in LSs. A lexical entry is now defined within the framework of a CLS, which is able to connect to a huge knowledge base from where it is feasible to retrieve information via a reasoning engine, which in turn uses two major functions: inheritance and inference (see Periñán-Pascual and ArcasTúnez 2010). By using artificial intelligence techniques, an RRG lexical representation in terms of a CLS can be successively augmented by information coming from different modules of the knowledge base (i.e. the ontology, the cognicon and the onomasticon).10 Finally, a CLS is further enriched by the inclusion of the operators that provide a fully-fledged semantic representation of an input sentence (Van Valin and Mairal Usón 2014).

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

3.4

Conclusion

RRG acknowledges the need for a more detailed theory of semantic representation in order to justify semantic decomposition as well as the primitive states and activities upon which it is based. This requires the formulation of a representation that is not limited to syntactically relevant factors. Such a representation system would be able to specify which lexeme would serve as the generic term in the definition of a concept and would allow users to effectively retrieve the semantic properties of each predicate in the lexicon. RRG requires an ontological semantic theory, which is the natural evolution of previous efforts to improve lexical representations, such as lexical templates with internal and external variables, based on lexical functions and qualia. For this reason, this chapter has provided a detailed description of a conceptually oriented representation system that accounts for the nonpropositional dimension of meaning, and which allows users to access contextual or encyclopedic meaning. In addition, it argues that adopting an ontological approach in the form of CLSs instead of standard LSs in RRG has numerous advantages. For example, CLSs can be automatically transduced to a COREL scheme which can be used to obtain inference operations with a reasoning engine. Consequently, the addition of CLSs to the RRG system substantially enriches its semantic representations.

References Amsler, Robert A. 1980. The Structure of the Merriam-Webster Pocket Dictionary. PhD dissertation, University of Texas at Austin. Amsler, Robert A. 1981. A taxonomy for English nouns and verbs. In Proceedings of the 19th Annual Meeting of the Association for Computational Linguistics, 133–138. Morristown, NJ: Association for Computational Linguistics. Bentley, Delia. 2019. The logical structure of verbs of quantized and nonquantized change. International RRG Conference, 19–21 August 2019. University at Buffalo (SUNY). Butler, Christopher S. 2012. An ontological approach to the representational lexicon in Functional Discourse Grammar. Language Sciences 34: 619–634. Cortés, Francisco, Carlos González and Rocío Jiménez. 2012. Las clases léxicas. Revisión de la tipología de predicados verbales. In Ricardo Mairal Usón, Lilián Guerrero, and Carlos González (eds.), El funcionalismo en la teoría lingüística. La Gramática del Papel y la Referencia. Introducción, avances y aplicaciones, 59–84. Madrid: Akal. Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press. Croft, William. 2012. Verbs: Aspects and Causal Structure. Oxford: Oxford University Press.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

237

238

RICARDO MAIRAL USÓN AND PAMELA FABER

Dik, Simon. 1978. Stepwise Lexical Decomposition. Lisse: de Ridder. Dummett, Michael. 1991. Frege and Other Philosophers. Oxford: Oxford University Press. Faber, Pamela and Ricardo Mairal Usón. 1999. Constructing a Lexicon of English Verbs. Berlin: Mouton de Gruyter. Fillmore, Charles J. 1977. Scenes-and-frames semantics. In Antonio Zampolli (ed.), Fundamental Studies in Computer Science, 55–88. Dordrecht: North Holland. Fillmore, Charles J. 1982. Frame Semantics. In Linguistic Society of Korea (ed.), Linguistics in the Morning Calm, 111–137. Seoul, South Korea: Hanshin Publishing. Fillmore, Charles J. 2006. Frame Semantics. In Dirk Geeraerts (ed.), Cognitive Linguistics: Basic Readings, 373–400. Berlin: Mouton de Gruyter. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Goddard, Charles and Anna Wierzbicka. 2002. Meaning and Universal Grammar: Theory and Empirical Findings. Amsterdam: John Benjamins. Goldberg, Adele. 2010. Verbs, constructions and semantic frames. In Malka Rappaport Hovav, Edit Doron and Ivy Sichel (eds.), Lexical Semantics, Syntax, and Event Structure, 39–58. Oxford: Oxford University Press. Gruber, Jeffrey S. 1976. Lexical Structures in Syntax and Semantics. New York: North-Holland. Havasi, Catherine, James Pustejovsky and Anna Rumshisky. 2007. An evaluation of the Brandeis semantic ontology. Proceedings of the Fourth International Workshop on Generative Approaches to the Lexicon. Paris, France. Hirst, Graeme. 2009. Ontology and the lexicon. In Steffen Staab and Rudi Studer (eds.), Handbook on Ontologies, 269–292. Berlin: Springer. Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, MA: MIT Press. Jackendoff, Ray. 1990. Semantic Structures. Cambridge, MA: MIT Press. Jackendoff, Ray. 1997. The Architecture of the Language Faculty. Cambridge MA: MIT Press. Lamb, Sydney M. 1998. Pathways of the Brain: The Neurocognitive Basis of Language. Amsterdam: John Benjamins. Levy, Jacob T. 2003. Language rights, literacy, and the modern state. In William Kymlicka and Alan Patten (eds.), Language Rights and Political Theory, 230–249. Oxford: Oxford University Press. Mairal Usón, Ricardo and Pamela Faber. 2007. Lexical templates within a functional cognitive theory of meaning. Annual Review of Cognitive Linguistics 5: 137–172. Mairal Usón, Ricardo and Carlos Periñán Pascual. 2009. The anatomy of the lexicon component within the framework of a conceptual knowledge base. Revista Española de Lingüística Aplicada 22: 217–244. Martín Mingorance, Leocadio. 1984. Lexical fields and stepwise lexical decomposition in a contrastive English-Spanish verb valency dictionary. In Reinhardt Hartmann (ed.), LEX’eter ’83 Proceedings. Papers from the

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

International Conference on Lexicography at Exeter, 226–236. Tübingen: Max Niemeyer. Martín Mingorance, Leocadio. 1990. Functional Grammar and Lexematics. In Jerzy Tomaszczyk and Barbara Lewandowska-Tomaszczyk (eds.), Meaning and Lexicography, 227–253. Amsterdam: John Benjamins. Martín Mingorance, Leocadio. 1995. Lexical logic and structural semantics: methodological underpinnings in the structuring of a lexical database for natural language processing. In Ulrich Hoinkes (ed.), Panorama der Lexikalischen Semantik, 461–474. Tübingen: Gunter Narr. Mel’cuk, Igor. 1989. Semantic primitives from the viewpoint of the Meaning–Text Linguistic Theory. Quaderni di Semantica 10(1): 65–102. Mel’cuk, Igor. 2012. Semantics: From Meaning to Text. Amsterdam: John Benjamins. Nirenburg, Sergei and Victor Raskin. 2004. Ontological Semantics. Cambridge, MA: MIT Press. Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2005. Microconceptualknowledge spreading in FunGramKB. In 9th IASTED International Conference on Artificial Intelligence and Soft Computing, 239–244. Anaheim-CalgaryZurich: ACTA. Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2007. Cognitive modules of an NLP knowledge base for language understanding. Procesamiento del Lenguaje Natural 39: 197–204. Periñán Pascual, Carlos and Francisco Arcas-Túnez. 2010. The architecture of FunGramKB. In Proceedings of the 7th International Conference on Language Resources and Evaluation, pp. 2667–2674. Marrakech, Morocco: LREC. Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2010. La gramática de COREL: un lenguaje de representación conceptual. Onomazein 21: 11–45. Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2011. The Coherent Methodology in FunGramKB. Onomazein 24: 13–33. Periñán-Pascual, Carlos and Ricardo Mairal Usón. 2012. La dimensión computacional de la RRG: la estructura lógica conceptual y su aplicación en el procesamiento del lenguaje natural. In Carlos González Vergara, Lilián Guerrero Valenzuela and Ricardo Mairal Usón (eds.), El funcionalismo en la teoría lingüística: la Gramática del Papel y la Referencia: Introducción, avances y aplicaciones, 333–348. Madrid: Akal. Procter, Paul. 1978. Longman Dictionary of Contemporary English. Harlow, UK: Longman. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Pustejovsky, James. 2001. Type construction and the logic of concepts. In Pierrette Bouillon and Federica Busa (eds.), The Language of Word Meaning, 91–123. Cambridge: Cambridge University Press. Pustejovsky, James and Olga Batiukova. 2019. The Lexicon. Cambridge: Cambridge University Press. Pustejovsky, James, Catherine Havasi, Jessica Littman, Anna Rumshisky and Marc Verhagen. 2006. Towards a generative lexical resource: The Brandeis

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

239

240

RICARDO MAIRAL USÓN AND PAMELA FABER

semantic ontology. Proceedings of the Fifth Language Resources and Evaluation Conference, LREC 2006, 1702–1705. www.cs.brandeis.edu/arum/publications/lrec-bso.pdf. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Amsterdam Studies in Theory and History of Linguistic Science Series 4: 1–1. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2018. Some issues regarding (active) Accomplishments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 71–93. Freiburg; Universität Freiburg. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Ricardo Mairal Usón. 2014. Interfacing the lexicon and an ontology in a linking algorithm. In María Ángeles Gómez, Francisco Ruiz de Mendoza and Francisco Gonzálvez-García (eds.), Theory and Practice in Functional-Cognitive Space, 205–228. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. and David P. Wilkins. 1993. Predicting syntax from semantics. In Robert D. Van Valin Jr., (ed.), Advances in Role and Reference Grammar, 499–534. Amsterdam: John Benjamins. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Wierzbicka, Anna. 1996. Semantics: Primes and Universals. Oxford: Oxford University Press.

Notes 1 See Cortés, González and Jiménez (2012) for an in-depth explanation of the tests used to differentiate one predicate class from another. 2 RRG distinguishes between the primitive do′ and the operator DO in order to represent the prototypical agency in verbs such as assassinate, perpetrate, promise, etc., which require a human subject capable of volition: *The cockroach assassinated the vice-president. 3 INGR stands for ‘ingressive’. 4 See Cortés, González and Jiménez (2012) for an explanation of how to determine the Aktionsart of a predicate. 5 BECOME, which, as previously noted, is broken down into PROC INGR, is used here for convenience. 6 Butler (2012) argues that a truly functional approach should provide a rich account not only of lexical phenomena but also of conceptualization. 7 FunGramKB and related publications can be accessed at www.fungramkb.com.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

A Conceptual Approach to Semantic Composition

8 See Periñan-Pascual and Mairal Usón (2011) for an in-depth description of the COHERENT methodology. 9 Predicative prepositions also have their corresponding conceptual logical structures, which are obtained from the sub-ontology that stores their corresponding conceptual units. 10 Periñán-Pascual and Mairal Usón (2012) have shown the applications of a CLS in the area of machine translation and cross-linguistic information retrieval.

https://doi.org/10.1017/9781316418086.006 Published online by Cambridge University Press

241

4 Semantic Macroroles Rolf Kailuweit

4.1

Introduction

Semantic macroroles are a major contribution of Role and Reference Grammar (RRG) to linguistic theory. Foley and Van Valin (1984) introduced two levels of semantic roles: an open list of specific thematic relations and two generalized semantic roles, or macroroles, actor and undergoer. Specific thematic relations were arranged in a graded Aktionsart-based continuum. Further work in RRG (Van Valin 1993, 2005, 2010; Van Valin and LaPolla 1997) slightly modified the initial formalization. Dowty (1991) took up the basic insights of Foley and Van Valin (1984), introducing the concept of proto-roles (Kailuweit 2004). It should be acknowledged that Dowty (1991) is currently the most prominent approach to generalized semantic roles1 (Rappaport Hovav and Levin 2015). However, approaches in typology (Haspelmath 2008; Bickel 2011) as well as in cognitive semantics (Jackendoff 2007) have built upon the RRG concept of macroroles to develop specific semantic generalizations. This chapter will deal with the theoretical foundations of macrorole assignment in RRG. In the first part, the status of thematic relations in RRG will be discussed in the context of a short historical review of Fillmore’s (1968) deep cases and Gruber’s (1965 [1976]) theta-roles, as well as of the discussion on generalized semantic roles in the 1980s and early 1990s. In the second part, special attention will be paid to the role of Aktionsart, semantic and syntactic transitivity, and causativity in macrorole assignment. Although the focus of the chapter is on standard RRG, the third part of the chapter discusses alternative approaches, concerning the number of macroroles that should be postulated and the features that are pertinent to macrorole assignment.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

4.2

Thematic Relations

In RRG, semantic roles, called thematic relations, are understood as a function of decomposed logical structures, following Gruber ([1965] 1976) and Jackendoff (1972), in contrast with Fillmore’s (1968) lists of case roles. Thematic relations are fundamental to making the concept of valency fruitful for syntactic description. Tesni`ere ([1959] 1965), along with Hockett (1958), one of the founders of the concept of valency, did not distinguish clearly between syntactic and semantic functions in the definition of the up to three actants a predicate could take. Hence, Tesni`ere ([1959] 1965) was not able to account for the passive construction in a coherent way.2 As was pointed out by Ruwet (1972), this is only possible if the levels of syntactic functions and thematic relations are clearly separated. Thematic relations, such as agent and patient, refer to the semantics of the predicate’s arguments, which remain stable, while the syntactic realization of the arguments varies between the active and the passive construction. (1)

Peter ate an apple. [Peter ¼ agent, apple ¼ patient]

(2)

The apple was eaten by Peter. [Peter ¼ agent, apple ¼ patient]

Thematic relations are not just isolated representations of the individual arguments that are taken by a predicate. They are sets of labels, one for each argument, with each label specifying the relation of that argument in the event in comparison to the other arguments. However, research in line with the initial approaches of Gruber ([1965] 1976) and Fillmore (1968) failed to determine the criteria to establish a coherent set of thematic relations. Thematic relations such as agent, patient or theme, etc. were supposed to be systematic generalizations across verb-specific roles such as runner (for run′), killer and killed (for kill′), possessor and possession (for have′), judger and judgement (for consider′), etc. Nonetheless, the number of thematic relations varied considerably in the literature, ranging from six (agent, theme, location, source, path, goal) in rigorous localist approaches to a few dozen in other approaches (see Rauh 1988; Dowty 1991; Fillmore 2003; Rappaport Hovav and Levin 1988; see Rappaport Hovav and Levin 2015: 600 for an overview). Fillmore introduced several nonlocalist roles – deep structure cases in his terminology – which have been widely discussed in the literature: instrument, force, experiencer, beneficiary, etc. However, while natural forces are participants of very specific events, experiencers are generalizations at a more abstract level, referring to arguments of verbs of emotion, perception and cognition. The difficulties in elaborating a coherent inventory of thematic relations led some authors to do away with thematic roles as categories of semantic description.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

243

244

ROLF KAILUWEIT

Rappaport Hovav and Levin (1988), Grimshaw (1990) and Tenny (1994) advocated an aspectual approach. According to these authors, the aspectual properties of a predicate alone determine the syntactic realization of the arguments. Telicity was deemed to be a pertinent criterion: an argument affected by a telic process has to be realized as an object at some level of representation.3 However, Rappaport Hovav and Levin (2015: 606) pointed out that change-of-state verbs, the most stable class in terms of their linking properties, are aspectually heterogeneous, including telic or variably telic verbs (degree achievements according to Dowty (1979), e.g. flatten versus widen).4 Moreover, no aspectual criterion is available to predict the linking of two-place state predicates such as those in (3) and (4). (3)

Peter abstained from alcohol and smoking.

(4)

Mary believed in justice.

Hence, not only aspectual properties, but also non-aspectual properties such as sensation, perception, volition and control, play an important role in argument linking. According to François (1997), aspectual as well as participantoriented properties (sensation, perception, control and causation5) are indispensable for argument linking. Macroroles in RRG as instantiations of generalized semantic roles are a construct that allows such an integration.

4.3

Macroroles as Generalized Semantic Roles

Foley and Van Valin (1984) considered macroroles to be generalizations about thematic relations. After developing logical structures (LS) as a formalism for decompositional representation of verb semantics,6 they based macrorole assignment on both LS and thematic relations (ibid., p. 53). As Van Valin (1999) pointed out, later versions of RRG denied thematic relations theoretical status, arguing that thematic relations are redundant for lexical representations.7 Thematic relation labels of the kind we find added to LS in Table 4.1 are only maintained for convenience as mnemonics (Van Valin and LaPolla 1997: 128). In RRG the relevant generalizations for macrorole assignment do not operate directly over thematic relations but over argument positions in logical structures representing activities and states, as can be seen in Table 4.1.8 Importantly, the criteria applied in macrorole assignment are not entirely aspectual. This becomes evident if we compare the two leftmost positions of the Actor-Undergoer Hierarchy (AUH) illustrated in Table 4.2. Both positions imply an activity component (DO (x,. . .)) or (do′ (x,. . .)). The argument of DO is a controlling human argument: for example, the first argument of murder. In contrast, the first argument of kill need not have the referential value þhuman. A natural force (e.g. hurricane) or an abstract concept or state of affairs (e.g. disease, drinking so much) can be the first argument of (do′ (x,. . .)).9

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

Table 4.1 Logical structures of states and activities and thematic relations (see Van Valin and LaPolla 1997: 115) I. STATE VERBS A. Single argument 1. State or condition 2. Existence B. Two arguments 1. Pure location 2. Perception 3. Cognition 4. Desire 5. Propositional att. 6. Possession 7. Internal experience 8. Emotion 9. Attrib./Identific. II. ACTIVITY VERBS A. Single argument 1. Unspecified action 2. Motion 3. Static motion 4. Light emission 5. Sound emission B. One or two arguments 1. Performance 2. Consumption 3. Creation 4. Directed perception 5. Use

broken′ (x) exist′ (x)

x ¼ PATIENT x ¼ ENTITY

be-LOC′ (x, y) hear′ (x, y) know′ (x, y) want′ (x, y) consider′ (x, y) have′ (x, y) feel′ (x, y) love′ (x, y) be′ (x, y)

x ¼ LOCATION , y ¼ THEME x ¼ PERCEIVER , y ¼ STIMULUS x ¼ COGNIZER , y ¼ CONTENT x ¼ WANTER , y ¼ DESIRE x ¼ JUDGER , y ¼ JUDGEMENT x ¼ POSSESSOR , y ¼ POSSESSED x ¼ EXPERIENCER , y ¼ SENSATION x ¼ EMOTER , y ¼ TARGET x ¼ ATTRIBUTANT , y ¼ ATTRIBUTE

do′ (x, do′ (x, do′ (x, do′ (x, do′ (x,

Ø) [walk′ (x)]) [spin′ (x)]) [shine′ (x)]) [gurgle′ (x)])

x ¼ EFFECTOR x ¼ MOVER x ¼ ST - MOVER x ¼ L - EMITTER x ¼ S - EMITTER

do′ (x, do′ (x, do′ (x, do′ (x, do′ (x,

[sing′ (x, (y))] [eat′ (x, (y))]) [write′ (x, (y))]) [see′ (x, (y))]) [use′ (x, y)])

x x x x x

¼ PERFORMER , y ¼ PERFORMANCE ¼ CONSUMER , y ¼ CONSUMED ¼ CREATOR , y ¼ CREATION ¼ OBSERVER , y ¼ STIMULUS ¼ USER , y ¼ IMPLEMENT

The other three positions are related to a stative component in the LS (pred′ (x, y) or pred′ (x)). RRG distinguishes between one-place and twoplace states. The argument of a one-place state is not only an entity that exists, but also and most prototypically an entity undergoing a change of state, being destroyed or killed, or coming into existence. Hence, the aspectual properties of accomplishments and achievements, which macrorole assignment does not directly refer to, are indirectly represented insofar as (pred′ (x)) prototypically formalizes the last component of the LS of achievements and accomplishments.10 Note that not all oneplace states are represented in the lexicon as monovalent verbs. Most typically, one-place states appear as part of the semantic representation of causation. Causative verbs show a complex semantic representation with both a cause and an effect component (αCAUSEβ). Causativity itself is not represented in the AUH. As Van Valin (2005: 57) points out: ‘the role assignments of the causative logical structure are those of the constituent activity and other logical structures; no new roles are added. The effect component (. . .CAUSEβ) prototypically consists of a resultative one-place state. As an effect of a breaking event, for example, an entity becomes broken.’

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

245

246

ROLF KAILUWEIT

Table 4.2 Actor-Undergoer Hierarchy (see Van Valin and LaPolla 1997: 127, 146; Van Valin 2005: 61) A CTOR

U NDERGOER

argument of DO AGENT

1. argument of do′ (x,. . .)

1. argument of pred′ (x, y)

2. argument of pred′ (x, y)

argument of pred′ (x)

EFFECTOR

LOCATION

THEME

PATIENT

MOVER

PERCEIVER

STIMULUS

ENTITY

ST - MOVER

COGNIZER

CONTENT

L - EMITTOR

WANTER

DESIRE

S - EMITTOR

JUDGER

JUDGEMENT

PERFORMER

POSSESSOR

POSSESSED

CONSUMER

EXPERIENCER

SENSATION

CREATOR

EMOTER

TARGET

SPEAKER

ATTRIBUTANT

ATTRIBUTE

OBSERVER

PERFORMANCE

USER

CONSUMED CREATION LOCUS IMPLEMENT

¼ increasing markedness of realization of argument as macrorole

Two-place states (pred′ (x, y)) formalize the semantics of bivalent stative verbs of possession or emotion such as own or like. In line with localist approaches (Gruber [1965] 1976; Jackendoff 1972), the x-argument of (pred′ x, y) could be thought of informally as a specific instance of a location. The y-argument of (pred′ x, y) informally corresponds to specific instances of a theme in terms of Gruber ([1965] 1976). To sum up, the AUH represents a continuum of argument positions in different subparts of lexical representations. The leftmost argument is the prototypical candidate for actor, the rightmost argument the prototypical candidate for undergoer.11 By default, the number of macroroles follows from the LS. For lexical representations with no arguments, for example rain′, no macrorole will be assigned. One-place intransitive activities, for example do′ (x, [shine′ (x)]), will take an actor. One-place intransitive nonactivities, for example exist′ (x), will take an undergoer. Macroroles always correspond to direct core arguments, although not every direct core argument bears a macrorole.12 By default, transitive predicates with two or three arguments take two macroroles. The default macrorole assignment principles are the following. (5)

Default Macrorole Assignment Principles a. Number: the number of macroroles a verb takes is less than or equal to the number of arguments in its LS: 1. If a verb has two or more arguments in its LS, it will take two macroroles. 2. If a verb has one argument in its LS, it will take one macrorole.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

b. Nature: for verbs which take one macrorole: 1. If the verb has an activity predicate in its LS, the macrorole is actor. 2. If the verb has no activity predicate in its LS, the macrorole is undergoer. (Van Valin 2013: 78) (6)

a. The hunter [A(ctor)] killed the bear [U(ndergoer)] b. The child [A] saw her friends [U] c. The girl [A] was running d. The stars [A] were shining e. The ghost [U] did not exist

[do′ (x, Ø]CAUSE[BECOME dead′ (y)].13 do′ (x, [see′ (x, y)]) do′ (x, [run′ (x)]) do′ (x, [shine′ (x)]) exist′ (x)

However, there are important cases where the number of macroroles is not entirely predictable from the LS. The LS have′ (x, y) corresponds to twoplace predicates of possession independently of whether these predicates are syntactically transitive (the own-type, e.g. German besitzen) or intransitive (the belong to-type, e.g. German gehören). The same holds for two-place predicates of liking. The LS like′ (x, y) corresponds to transitive predicates (English like, French aimer bien, German mögen, etc.) as well as intransitive predicates (French plaire, German gefallen, etc.). In such cases, macrorole intransitivity has to be marked in the lexicon. (7)

a. German gefallen b. German gehören

like′ (x, y) [MR1] have′ (x, y) [MR1]

For transitive verbs, the default macrorole assignment rules predict the number of macroroles but do not stipulate which argument takes the actor macrorole and which argument takes the undergoer macrorole. The default choice follows from the consideration that ‘given the LS for a transitive verb, the leftmost argument in it will be the actor and the rightmost will be the undergoer’ (Van Valin 2013: 78). It follows from the examples in Van Valin (2013: 78) that this also holds for causatives with two LS components (cause and effect) linked together with a CAUSE operator. Starting from the left with the αCAUSE component of the LS, in example (8) the actor macrorole is assigned to the x-argument (dog). Undergoer assignment will start from the right of the CAUSEβ component. The rightmost accessible argument will be the y-argument (boy) that will be selected for undergoer. (8)

The dog scared the boy. [do′ (x ¼ dog, Ø] CAUSE [feel′ (y ¼ boy, afraid′])]).

Note that if the undergoer were always the rightmost argument, predicates represented by a causative structure with a two-place state in the CAUSEβ component should always select the second argument of pred′ (x, y) as undergoer. (9)

give [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]

(10)

load [do′ (x, Ø)] CAUSE [BECOME be-on′ (y, z)]

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

247

248

ROLF KAILUWEIT

However, some three-place predicates of transfer (give), putting (load, spray), or removal (drain, empty) show variable undergoer choice. Depending on the language, the verb subclass, and, in the last analysis, the individual lexeme,14 some predicates allow the first argument of the two-place state in the CAUSEβ component (pred′ (x,. . .)) to take the undergoer macrorole (e.g. Mary gave Peter a book. Max loaded the truck with hay). In standard RRG, this pattern was considered to be a marked undergoer choice. The German applicative construction provides a theory-independent argument for the markedness of the pred′ (x,. . .)-as-undergoer construction. While the morphological unmarked laden (‘load’) realizes pred′ (. . .,y)) as undergoer, the pred′ (x,. . .)-as-undergoer construction requires additional marking with the morpheme be-. (11)

a. Max lädt die Kisten auf den Laster Max load.3sg the box.f.pl.acc on the truck ‘Max loads the boxes on the truck.’ b. Max belädt / *lädt den Laster mit Kisten Max be.load3sg load3sg the truck.acc with box.f.pl.dat ‘Max loads the truck with boxes.’

Van Valin (2005: 126) acknowledged that from a strictly typological viewpoint, the markedness hypothesis has to be abandoned due to the fact that in primary object languages15 the pred′ (x,. . .)-argument seems to be the unmarked choice for undergoer (see Guerrero and Van Valin 2004; Diedrichsen 2008). While most languages follow principle A in (12), some languages follow principle B. (12)

Principles for Undergoer-choice (Van Valin 2005: 126) 1. Principle A: lowest-ranking argument in LS (default) 2. Principle B: second-highest-ranking argument in LS

Obviously, one could also apply these principles to different constructions in one and the same language, in which case the pred′ (x,. . .)-as-undergoer construction of English give or load would no longer be an instance of marked undergoer choice but of principle B in a language that generally follows principle A. In Foley and Van Valin (1984: 29), macroroles were defined in purely semantic terms, actor as ‘the argument which expresses the participant which performs, effects, instigates or controls the situation denoted by the predicate’, and undergoer as ‘the argument which expresses the participant which [. . .] is affected by it in some way’. However, macrorole (in)transitivity at the lexical level and variable undergoer choice show that macrorole assignment is not only determined by position in LS but also by the semantic properties which determine whether a core argument is realized as a direct core argument in the morphosyntax.16 Indirect core arguments cannot be undergoers. From the RRG point of view, it is because of the semantic entailments that they get from verbs that core arguments are realized as direct or indirect.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

If undergoer selection is flexible, at least for some predicates, then one might be tempted to assume that macrorole assignment parallels the basic claim of the so-called (neo-)constructionist approach. In her syntactic predicate-based approach Borer (1994, 2005) claims that semantic interpretation derives mostly from the syntactic realization of arguments. There is evidence that, at least from the hearer’s perspective, the syntactic configuration entwined with the undergoer choice determines meaning (see Rappaport Hovav and Levin 2015: 615–616 for relevant discussion). However, in the neo-constructionist perspective, the meaning is assigned by a syntactic substructure, in RRG the meaning derives from the lexically determined and in some cases flexible assignment of undergoer, that is, a semantic category playing a role in syntax (Van Valin 2004: 74–78). Let us consider again the case of the locative alternation. The semantics of the two constructions with load are not equivalent. According to Anderson (1971), there is always a holistic interpretation of the direct object (i.e. the undergoer). All of the hay is loaded onto the truck when the pred′ (. . .,y)argument is chosen as undergoer and appears in the direct object position, and the truck may or may not be full. When the pred′ (x,. . .)-argument is chosen as undergoer and as direct object, the truck is completely filled, without any implication as to whether all of the hay is loaded or not (cf. Van Valin 2007). Kailuweit (2008: 350) described the semantic effects of marked undergoer choice in the following way: if the x-argument of a transitive predicate pred′ (x, y) is exceptionally chosen as undergoer, its interpretation does not rely on its inherent semantic properties. It is the undergoer function that contributes certain prototypical semantic values to the argument, that is, at least one of the following properties: ‘causally affected’, ‘change of state’, or ‘incremental theme’. In psycholinguistic terms, one could argue that the linking from syntax to semantics reveals that the ‘wrong’ argument in the LS has been chosen for undergoer. This leads to cognitive dissonances that can only be resolved by referring to the semantics prototypically associated with the undergoer. Hence, the pred′ (x,. . .)-argument is not just conceived as a location, but as an argument primarily affected in the event. The truck the hay is loaded onto is interpreted as being completely filled, which is not the case when the pred′ (. . .,y)-argument, the hay, is chosen as undergoer. To sum up, RRG macroroles are neither mere generalizations of thematic relations, nor is their assignment entirely motivated by aspectual (Aktionsart) differences. While the basic Aktionsart opposition between activities and states plays a crucial role, other aspects are pertinent, too. Not only do the proto-agent properties ‘volition’, ‘sensation’ and ‘control’17 help to distinguish a prototypical agent (argument of DO) from a mere effector (first argument of do′ (x,. . .)), but they also implicitly motivate the order of the arguments in two-place states. Proto-agent properties characterize a perceiver, cognizer, wanter, judger, possessor, or experiencer argument and justify placing it at a higher level of activity (pred′ (x,. . .)) in

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

249

250

ROLF KAILUWEIT

comparison with a stimulus, content, desire, judgement, possessed, or sensation argument (pred′ (. . .,y)).18 The processing rule, which processes logical structures from left to right (or starting from the element that comes first in time), is a syntactic device to cope with causal chains such as (agent > instrument). Since RRG presupposes identical LSs for transitive and intransitive predicates of possession or liking, macrorole transitivity is affected by the semantic properties which result in the morphosyntactic realization of core arguments as direct or indirect. Finally, to our mind, variable undergoer choice adds an element to the projectionist RRG approach that shows interesting parallels with the neo-constructionist approach. The undergoer function, although semantic in nature, can override the semantics associated with the argument position of the LS of a given entry in the lexicon. However, variable undergoer choice is not a putative structural option for any three-place predicate, but a semantic property determined by the lexicon.

4.4

Putative Challenges to Macrorole Assignment

In this section, we deal with some types of predicates that can be said to be challenging for the RRG theory of macrorole assignment.

4.4.1 Causativity When Foley and Van Valin (1984) first proposed the AUH, they presupposed – in line with Dowty (1979) – a logical link between causative constructions and accomplishments. In addition, they considered the argument of αCAUSE to be an argument of DO. As the argument of αCAUSE is necessarily the most active argument in the hierarchy, no problem arose with macrorole assignment. (13)

Joan broke the glass [DO (Joan)] CAUSE [BECOME broken′ (glass)] (Foley and Van Valin 1984: 52)

However, later research in RRG (Van Valin and LaPolla 1997: 97, 106–109; Van Valin 2018) has shown that equating causative constructions with accomplishments was erroneous. On the one hand, there are accomplishments that are not causative, like the one in (14a), which contrasts with its causative counterpart in (14b). (14)

a. The ice melted (accomplishment) b. Hot water melted the ice (causative accomplishment)

On the other hand, non-causative states, activities and achievements can also have corresponding causative constructions. (15)

a. The boy is afraid (state) b. The ball bounced around the room (activity) c. The balloon popped (achievement)

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

(16)

a. Bill’s owning a gun frightened Martha (causative state) b. The girl bounced the ball around the room (causative activity) c. The cat popped the balloon (causative achievement)

The representation of causativity did not pose formal macrorole assignment and linkage problems for Foley and Van Valin (1984). Perhaps for this reason, the interaction of the AUH and causative constructions has not been discussed in detail in later versions of RRG. Some putatively problematic cases for macrorole assignment that arise from the isolated application of the AUH for the individual predicates in causative constructions can easily be ruled out if macrorole assignment begins with the actor and simply occurs in the causal chain from left to right. These cases are discussed below. We will start with causative activities. Van Valin and LaPolla (1997: 107) propose the LS [do′ (girl, Ø)] CAUSE [do′ (ball, [bounce′ (ball)])] for (16b). Both components of this LS consist of activities. Hence, the two arguments (girl and ball) occupy the same position in the hierarchy, but the processing rule assures that the actor is assigned to girl. Being the rightmost argument, ball in the CAUSEβ part takes the undergoer macrorole, although it is an argument of do′ (x,. . .). Obviously, this a coherent interpretation from the semantic point of view, since the CAUSEβ component is the effect of the causing event. This is also true for activities embedded into causative chains (Van Valin and LaPolla 1997: 121). The AUH explains why knife could eventually become actor in an instrument-subject construction (cf. 17b). (17)

a. Tom is cutting the bread with a knife [do′ (Tom, [use′ (Tom, knife)])] CAUSE [[do′ (knife, [cut′ (knife, bread)])] CAUSE [BECOME cut′ (bread)]] b. The knife cuts the bread [do′ (knife, [cut′ (knife, bread)])]

A proposal to make the possible undergoer assignment to arguments of do′ (x,. . .) more explicit would be to mark this position as accessible for undergoer choice in the AUH (Kailuweit 2012: 110) (see Table 4.3). In (16a), an example of a causative state, macrorole assignment does not follow straightforwardly from the AUH. Van Valin and LaPolla (1997: 107) Actor

Undergoer

argument of 1. argument 1. argument 2. argument argument of pred′ (x) DO of do′ of pred′ of pred′ (x,…) (x, y) (x, y) = increasing markedness of realization of argument as macrorole Figure 4.1 Actor-Undergoer Hierarchy (modified)

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

251

252

ROLF KAILUWEIT

propose the LS [have′ (Bill, gun)] CAUSE [feel′ (Martha, [afraid′])] for (16a). Both compounds of CAUSE consist of a two-place state predicate (have′(x ,y) and feel′ (x, y)). For the β part the y-argument is a predicate [afraid′]. The two x-arguments of these predicates (Bill and Martha) are both instances of the same middle position of the AUH. If we follow the rule of processing from left to right, Bill will be the actor. However, this is inaccurate from a semantic point of view. It is not Bill, but his possession of a gun that is responsible for causing Martha’s fear. Hence, the actor of frighten is not Bill, but the whole αCAUSE compound. It may seem strange to think of a state of affairs as an actor, but otherwise frighten would cease to be a macrorole transitive predicate in (16a).

4.4.2 Three-Place Predicates According to Van Valin (2007), almost all three-place predicates are causative constructions. We have noted that these predicates give rise to problems of variable undergoer choice. The overall LS of these predicates would be [do′ (x, Ø)] CAUSE [BECOME pred′ (y, z)]. The actor is represented in the αCAUSE part and the CAUSEβ part hosts the remaining two arguments that are accessible for undergoer assignment. As we have already seen, the conditions of undergoer assignment vary across languages and even across individual predicates. Nonetheless, there are a few (but highly frequent) threeplace predicates that do not seem to be causative. In the remainder of this section, I will deal with talk (to somebody about something)19 as a candidate for a possible non-causative three-place activity and with envy (somebody (for) something) as a candidate for a possible a non-causative three-place state. In Van Valin and LaPolla (1997: 116–118) we find a detailed description of English verbs of saying, including verbs of talking. The authors propose a general LS for non-causative verbs and an LS for causative verbs of telling. (18)

talk: [do′ (x, [express.(α).to.(β).in.language.(γ)′ (x, y)])] tell: [do′ (x, [express.(α).to.(β).in.language.(γ)′ (x, y)])] CAUSE [BECOME aware.of′ (y, z)], where y ¼ β, z ¼ α

Talk is listed as a verb that selects an α component and a β component (talk to Pat about Sandy) or a γ component (talk Cajun). Since it is the absence of causativity that distinguishes verbs of talking from verbs of telling, the causative LS in (18) cannot describe the three-place use of talk. In a later chapter concerned with coreference, Van Valin and LaPolla (1997: 401) propose an LS for talk, when they discuss the coreference of the reflexive in Bob talked to Susan about herself. (19)

[do′ (Bob, [express.(α).to.(β).in.language.(γ)′ (Bob, Susani)])] ᴧ [about′ (herselfi)]

The binding facts seem to be the reason why the β component appears as the y-argument of the two-place state and the α component as the argument of the preposition co-referring with the y-argument. In a footnote Van Valin

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

and LaPolla (1997: 668) remark that about herself is an argument PP and not an argument-adjunct, because the PP is ‘the optional realization of the α variable’. The fact that the preposition seems to be semantically empty and commutable (talk about, talk of as well as German sprechen über, sprechen von) corroborates this view. However, it is precisely for this reason that it seems problematic to use an argument-adjunct construction to represent the third argument of talk.20 Staudinger et al. (2008) proposed an alternative analysis. RRG allows for several classes of two-place activities: do′ (x, [sing′ (x,(y))]); do′ (x, [see′ (x,(y))]); do′ (x, [tap′ (x,(y))]), etc. (Van Valin and LaPolla 1997: 115).21 The second argument represents an instance of pred′ (x, y) (ibid.: 127), but the predicate embedded in activities is not a state. Hence, RRG could theoretically allow for three-place activities with an argument in each of the three medium positions of the AUH: do′ (x), pred′ (x, . . .), and pred′ (. . ., y). Staudinger et al. (2008) thus suggested the following LS for French parler, the counterpart of English talk. (20)

parler `a qn de qc ¼ do′ (x, [talk′ (x, y, z)])

Note that according to the formalism used in RRG, the predicate talk′ (x,(y),(z)) is not a state, similarly to sing′ (x, (y)) in do′ (x, [sing′ (x, (y))]). As indicated before, activities do not consist of states embedded under a do′ operator. Hence, in an RRG description of talk, one could associate the two arguments y and z with the positions pred′ (x, . . .) and pred′ (. . ., y) of the AUH, as RRG associates the y-argument of sing′ with the pred′ (. . ., y) position. However, this solution would not be available for three-place noncausative states. Verbs of envy are possible candidates for this class (Kailuweit 2012: 118). In line with two-place stative verbs of emotion such as love or hate, envy takes an emoter argument, the person who experiences envy, and a target argument, the envied person. In addition, envy possesses a third argument expressing the object or circumstance the target argument is envied for.22 For the English verb envy, this third argument could be realized as a direct core argument (21a) or as an oblique (21b). (21)

a. Many people envied the shareholders their large profits. b. Max envied Paul for his wealth. c. Max envied him.

The fact that the third argument, the object of envy, can also be marked by for (21b) may suggests that it is an adjunct, which gets incorporated into the core as a direct argument, by analogy with recipient beneficiaries, such as Max baked a cake for Paul vs. Max baked Paul a cake. Thus, a possible LS for (21b) could be because.of′ ([have′ (Paul, wealth)], [envy′ (Max, Paul)]). However, in (21a) and (21b) the target, the envied person, is the undergoer, while the for-complement becomes the undergoer in the recipient-beneficiary construction Max baked a cake for Paul ! Max baked Paul a cake.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

253

254

ROLF KAILUWEIT

French envier (‘envy’) shows variable undergoer choice. The target can be realized either as an undergoer (22a) or as non-macrorole direct core argument in the dative case (22b). As in English (21c), the object of envy argument is optional if the target argument is realized as undergoer. (22)

a. Beaucoup d’hommes vous envient pour Madame Linné Many of men you.2pl.acc envy.3pl for Mrs Linné ‘Many men envy you for Mrs Linné’ (FRANTEXT: HERMARY-VIEILLE, C.) b. . . .cette belle figure creuse [. . .] il la lui enviait this pretty face haggard he 3f.sg.acc 3m.sg.dat envied ‘It was this pretty haggard face that he envied him for.’ (FRANTEXT: ETCHERELLI)

The same case variation as in French can be found in German. In addition, the contrast between the two constructions is realized with the help of the applicative construction with the prefix be-. German neiden (‘envy’) realizes the target argument as a non-macrorole direct core argument in the dative case and the object of envy as the undergoer. German beneiden realizes the target as the undergoer and the object of envy as a prepositional object: jemanden um etwas beneiden (‘to envy somebody for something’). Hence, the object of envy seems to be the unmarked choice for undergoer in German. Admitting that envy is a three-place state would raise the problem of assigning an LS that provides information about macrorole assignment. One might consider envy′ (x, y, (z)) as a possible LS for envy. However, the three argument positions in this LS cannot correspond to the three stative argument positions in the AUH. As far as the three positions referring to states are concerned – first argument of pred′ (x, y), second argument of pred′ (x, y) and argument pred′ (x) – they represent two basically different types of state. On the one hand, there are two-place states (pred′ (x, y)), for example verbs of emotion (love′ (x, y)) or possession (have′ (x, y)), with a less affected second argument (a theme in Gruber’s terms). On the other hand, there are one-place states (pred′ (x)) with an affected prototypical undergoer argument, for example resultative verbs of dying (BECOME dead ′ (x)) or destruction (BECOME broken′ (x)). Hence, the structures pred′ (x) and pred′ (x, y) logically exclude each other and cannot be added to each other. Note that verbs of envy are not the only candidates for non-causative three-place states. A more language-specific case is the French verb en vouloir `a quelqu’un de quelque chose ‘to be angry with someone about something’.23 The argument denoting the person who experiences anger is realized as the subject, and hence it takes a macrorole. In contrast to verbs of envy, en vouloir is M-intransitive. Therefore, the two other arguments, the person the anger is directed at, and the circumstance that the anger is related to, are

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

realized as a direct core argument in the dative case and as an oblique object respectively. In comparison with verbs of envy, en vouloir raises an additional problem: the nature of the macrorole. RRG describes non-episodic verbs of anger as instances of verbs of internal experience with the LS feel′ (x, y). The x-argument corresponds to the experiencer, and the y-argument is a predicate of sensation having one internal argument that is not accessible for macrorole assignment. Therefore, the predicates of internal sensation are by default M-intransitive. As intransitive states, they select an undergoer. The undergoer macrorole is assigned to the x-argument, the experiencer. (23)

Episodic verbs of emotion (Van Valin and LaPolla 1997: 156, 402) Pat is angry at Kelly. [feel′ (Pat ¼ experiencer ! 1. arg. of pred′ (x, y) ! Undergoer ! PSA, [angry.at′ (Kelly)] ¼ sensation ! 2. arg. of. pred′ (x, y))]

A possible LS for three-place en vouloir would be feel′ (x, angry.at′(y).about′ (z)) with two internal arguments of the sensation predicate. However, the feel′ (experiencer, sensation) analysis is neither convincing for the French three-place en vouloir nor for its two-place German or Spanish counterparts zürnen and rabiar ‘to be angry at’. While German zürnen realizes the argument denoting the person the anger is directed at as a dative, Spanish rabiar expresses this with a preposition (contra, lit. ‘against’). Independently of these syntactic differences, this argument is obligatory for French en vouloir, German zürnen and Spanish rabiar. To consider an obligatory argument inaccessible for macrorole assignment seems to be problematic. In addition, if the argument denoting the person the anger is directed at were embedded into the sensation predicate, ‘being angry at James’ and ‘being angry at Paul’ would be two different sensations. Kailuweit (2005b, 2013) proposes that French en vouloir, German zürnen and Spanish rabiar are atypical activities rather than states. They denote an emotional behaviour motivated by anger and directed at a person held responsible for the annoying situation.24 Therefore, the macrorole is an actor and not an undergoer. An LS that parallels the one proposed for verbs of saying (Staudinger et al. 2008) would be able to deal with macrorole assignment and linking.

4.5

Alternative Approaches to Macroroles

The fact that languages such as German or Russian possess direct core arguments in the dative case, which behave differently from other nonmacrorole arguments in syntax, led to intense discussion of the necessity for a third macrorole. The German passive construction with bekommen (‘get’) seems to be a candidate for illustrating the theoretical contribution that a third macrorole could make. Diedrichsen (2008) discusses the following data.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

255

256

ROLF KAILUWEIT

(24) LS: [do′ (Eltern, Ø)] CAUSE [BECOME have′ (ich, Computer)] a. Active voice Meine Eltern haben mir diesen Computer geschenkt my parents have.3pl 1sg.dat dem.acc computer given ‘My parents gave me this computer.’ b. Passive voice Dieser Computer ist mir von meinen Eltern geschenkt worden dem.nom computer be.3sg 1sg.dat by my parents given become ‘This computer was given to me by my parents.’ c. Bekommen-passive voice Ich habe diesen Computer von meinen Eltern geschenkt bekommen I.nom have.1sg dem.acc computer by my parents given received ‘I got/received this computer, given by my parents’, ‘I was given this computer by my parents.’ bekam von der Polizei den (25) a. Er 3m.sg.nom get/receive.3sg.pst by def.f.sg.dat police def.m.sg.acc Führerschein entzogen. driving licence.sg revoke.pst ‘He got/received the driving licence revoked by the police.’ b. *Er bekam den Führerschein. Er wurde von der Polizei entzogen. ‘He received the driving licence. It was revoked by the police.’

In addition to the passive construction where the privileged syntactic argument (PSA) is the argument marked by accusative in the active voice (24b), German allows the dative-coded recipient argument of many ditransitive verbs to become a PSA in the passive (24c). PSA assignment to the accusative argument of the active construction requires the werden-passive construction, while PSA assignment to the dative argument of the active construction is realized by the bekommen-passive construction. Diedrichsen (2008) analyses the bekommen-passive construction in (24c) as a nuclear juncture, as Bentley (2006: 39) does for the corresponding Italian construction with avere ‘have’. However, German bekommen is further grammaticalized as shown in (25), where it loses its physical reception meaning. Hence, according to Diedrichsen (2008), the PSA-argument in (25) cannot be considered the actor of a complex predicate bekommen (‘receive’) þ entziehen (‘revoke’). Since PSA choice is restricted to macrorole arguments, Diedrichsen (2008) considers Lehmann et al.’s (2004) proposal of introducing a third macrorole, called ‘indirectus’. Like the proto-recipient introduced by Primus (1999) into the proto-role-approach,25 the indirectus controls and affects the undergoer, but is controlled and affected by the actor. Lehmann et al. (2004: 19) define the indirectus as a generalization about the thematic relations (microroles) recipient, beneficiary, experiencer, emittent (the participant of a transfer event that gives something away) and sympatheticus (the secondary affected participant with a three-place predicate).

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

However, the notion of indirectus is restricted to the third argument of three-place predicates. Hence, from a syntax-to-semantics viewpoint, generalizations concerning the semantics of dative arguments cannot be captured.26 A third macrorole would be a qualitatively different concept in comparison with actor and undergoer (Van Valin 2004). While the two established macroroles are universally valid for the description of morphosyntactic phenomena, the third macrorole is not. As Lehmann et al. (2004) themselves recognize, some languages do not permit three core arguments. From a semantics-to-syntax viewpoint, there is no consistent treatment of the arguments that are candidates for the indirectus. German, Russian and Dyirbal assign dative to the recipient, beneficiary, experiencer, emittent, or sympatheticus of three-place predicates, while English or Jakaltek marks this argument with an adposition (Van Valin 2004: 81). In English, the assumption of a third macrorole would single out the to-PP as a special type of prepositional object. This is not justified on morphosyntactic grounds. As far as the German bekommen-passive is concerned, Diedrichsen (2008) rejects Lehmann et al.’s approach, putting forward two alternative solutions. First, she considers the bekommen-passive as a PSA-selection mechanism whose function is to select a highly topical dative undergoer as PSA. This would imply accepting datives as undergoers, which is not foreseen in standard RRG. Alternatively, an approach more in line with standard RRG would account for the bekommen construction as a lexical passive in that undergoer status is assigned to a dative recipient that is the non-macrorole argument of the ditransitive active. From the viewpoint of typology, Bickel (2011) and Bickel et al. (2014) refer explicitly to the concept of macroroles to develop a descriptive toolkit for the different alignment types of ditransitive constructions. Bickel et al. distinguish between different sets of macroroles (or proto-roles) for oneplace, two-place and three-place predicates. Building on a tradition in typology which started with Comrie (1978) and Dixon (1979), they distinguish between S as the argument of an intransitive verb, A as the most actor-like argument in a transitive verb, and O or P (in Bickel et al. 2014)27 as the not most actor-like argument in a transitive verb. In addition, they label the two non-actor-like arguments of ditransitives as G (for goal) and X or T (for theme). The generalized semantic roles G28 and X/T stand for the most goal-like and for the other (not most goal-like, not most actor-like) argument respectively. While the only semantic motivation for S is the fact that the predicate takes only one argument, S does not serve to describe split intransitivity. A and O/P seem to correspond to actor and undergoer in RRG. However, the generalized semantic roles are assigned to any two-place predicate, even to predicates lacking a referential second direct core argument. In addition, neither G nor X, the two less active arguments of threeplace predicates, have to be direct core arguments. In typology, the function of the generalized semantic roles is to serve as a tertium comparationis for

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

257

258

ROLF KAILUWEIT

Figure 4.2 Four types of languages, accusative and ergative alignment (Haspelmath 2008)

different forms of alignment. For example, marking S and A in the same way characterizes accusative alignment. While grouping O/P and X/T together corresponds to indirective alignment, secundative alignment is characterized by marking O/P and G/R identically (Bickel 2011). Haspelmath (2008) pursues a very similar approach, which he includes directly into the RRG framework. He proposes four macroroles: A ¼ actor, U ¼ undergoer of monotransitive sentences, R ¼ macro-recipient, corresponding to Primus’s (1999) proto-recipient and the aforementioned indirectus of Lehmann et al. (2004), and finally T ¼ (macro-)theme (see Figure 4.2). These macroroles are defined in purely semantic terms. They account for the parallels between monotransitive and ditransitive alignments in a straightforward way without needing to refer to marked or unmarked undergoer choice, as in standard RRG, or to parameterize undergoer selection for the indirective–secundative contrast, as in Guerrero and Van Valin (2004). Although U, T, R and A seem to correspond to the four rightmost positions of the AUH, it is unclear how their semantic values could be determined in a precise way. Haspelmath (2011) himself criticizes the approach of Bickel et al. (2014), but his critique refers indirectly to his own RRG approach, too (Haspelmath, p.c., 30 January 2013). His point is that Bickel’s generalized semantic roles are universal semantic categories aimed at the description of individual languages. However, being not only syntactically undetermined but also semantically highly heterogeneous, they do not serve to draw typological generalizations. Indeed, it remains unclear in both approaches what agent-like, goal-like or theme-like mean beyond purely intuitive notions.29

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

Table 4.3 Features and values for activity clusters (from Kailuweit 2013) þc 4

c 2

c 0

þm 2

m 1

m 0

r 0

r 2

þr 4

In line with Comrie’s approach, Haspelmath (2011) points out that S, A, P/O/ U, T/X, and R/G should only be comparative concepts referring to the syntactic relations associated in a given language with verbs selecting a prototypical agent and patient (A:P), verbs of transfer (A:R:T), and one-place verbs of uncontrolled change of state such as verbs of dying (S). This means that not every two-place predicate assigns an active and a passive generalized semantic role. Only predicates that code the opposition of the arguments, like prototypical verbs with an agent and a patient, do: for example verbs like English see, like, own or enter follow the pattern of break or kill, but appeal to and belong to do not. This is in line with standard RRG. However, Haspelmath’s (2011) comparative concepts do not apply to all predicates in every language and do not formulate hypotheses about universal semantic structure. Hence, they differ considerably from RRG macroroles. Kailuweit (2013, 2018) maintains the two macroroles of standard RRG, devising, however, a feature-based Activity Hierarchy, which departs from the RRG AUH. Following Rozwadowska (1988), Kailuweit (2013) takes three features into account: causative and/or control [c], mental (sentient) [m], and resultative (change of state) [r]. In line with Reinhart (2002), these features can assume one of the three values þ,  or . In addition, the features are weighted (see Table 4.3). The feature [c] is a strong actor feature, [m] is a weak actor feature, and [r] is a strong undergoer feature. The presence of a strong feature, indicated with [þ], will duplicate the value of the presence of a weak feature. If an argument is underspecified for one feature or has an intermediate degree of that feature [], the value will be half of the [þ] value. The combination [þcþmr] represents the prototypical actor with the value 4þ2þ0 ¼ 6, while the prototypical undergoer corresponds to the combination [cmþr] 0þ04 ¼ 4. Nine intermediate combinations are mathematically possible. Hence, the system distinguishes eleven possible degrees of activity. The attribution of the values follows from a detailed semantic analysis of events and participants, evaluating aspectual properties as well as causativity, sentience, responsibility and control. In the following, sketch-like descriptions will be given to account for the non-causative three-place predicates talk and envy as well as for some twoplace verbs of emotion. Some of the examples discussed above are repeated for convenience. (26)

Bob talked to Pat about Sandy

(27)

a. Many people envied the shareholders their large profits b. Max envied Paul for his rewarding task

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

259

260

ROLF KAILUWEIT

(28)

a. Bill’s owning a gun frightened Martha b. The mother feared for her children c. The mother feared her children

As far as talk is concerned, the speaker controls their activity to a certain degree, but not entirely (mimics and gesture, unforeseen effects of the speech, slips of the tongue, etc.), hence, the value for [c] is . The addressee does not control the event but takes part in it as a sentient being. The topic of conversation neither controls the event nor is it affected by it. Hence, the speaker is the most active argument assuming the actor macrorole. The addressee is neither the most active nor the most passive argument. As with other arguments exhibiting an intermediate degree of activity, it is realized with the preposition to in English (and the dative in the case of French parler ‘talk’). The topic of conversation is the most passive argument of the event, although it assumes rather a high degree of activity (0 in comparison with 4 of the prototypical patient). This may explain the realization with a preposition (about/of in English, sur/de in French). Verbs of envy denote non-episodic emotions with three participants. In episodic emotions, prototypically the target, for example the person or the state of affairs the emotion is directed at, is at the same time the causer of the emotion. The experiencer is affected, undergoing a change of emotional state in a given situation. This is the case of frighten in (28a) selecting a target that functions as an inanimate causer. By contrast, emoters of non-episodic emotions express a subjective judgement that they are held responsible for, because they control it to a certain degree [c]. Some verbs of emotion show another argument, helping to identify the target. In (28b), for example, the children are in danger, but they are not the danger themselves as in (28c). Hence, the children represent a point of reference to find out about the danger as the target of the mother’s fear. With envy, the target and the point of reference are realized. The envy is directed to a person being the target. The third argument specifies why this person has become the target of envy. The attribution of [m] accounts for the fact that the point of reference is a human being (prototypically, but not necessarily with fear for) or a property associated with a human being (in the case of envy). For English envy, the emoter as the most active argument represents the actor, the target as the most passive argument the undergoer. The point of reference of a middle degree of activity appears as the non-macrorole argument.30 Further research will determine whether the Activity Hierarchy can describe the same number and even more predicate classes in as coherent a way as the LS-based AUH does.31 Perhaps LSs and activity values can also be combined in the lexical entries of the predicates to enable a fine-grained semantic description of the linking-relevant properties.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

Table 4.4 Activity Hierarchy of arguments of selected predicates [þcþmr] [þcmr] [cþmr]

6 4 4

AGENT - CAUSER

[cþmr] [cmr] [cmr] [cþmþr] [cmþr]

2 1 0 -2 -4

ADDRESSEE

4.6

INANIMATE CAUSER SPEAKER

/

EMOTER

POINT OF REFERENCE OF EMOTIONS TARGET OF EMOTIONS

/

TOPIC OF CONVERSATION

AFFECTED EXPERIENCER PATIENT

(undergoing a change of state)

murder (x,. . .) frighten (x,. . .) talk (x,. . .), envy (x,. . .), fear (x,. . .), fear for (x,. . .) talk (. . .,y,. . .) fear for (. . .,y), envy (. . .,y,. . .) envy (. . .,z), talk (. . .,z) frighten (. . .,y) murder (. . .,y)

Conclusion

The macroroles of RRG constitute a powerful tool to address the challenges of argument realization. Specifically, the theory of macrorole assignment captures the key insights of the approaches called ‘aspectual’, ‘causal’ and ‘neo-constructionist’ by Rappaport Hovav and Levin (2015). In the vast majority of cases, RRG makes correct predictions about argument linking by using predicate decompositions and identifying the relevant facets of the predicate’s meaning. Individual facets may be more effectively captured by other approaches, which are mainly, or exclusively, ‘featurebased’, ‘decompositional’, ‘aspectual’, ‘causal’ or ‘neo-constructionist’. Nevertheless, macrorole assignment in RRG currently seem to be the most complete approach to capture meaning generalizations relevant to the syntactic realization of arguments.

References Abreu Gomes, Cristina. 2003. Dative alternation in Brazilian Portuguese: Typology and constraints. Language Design 5: 67–78. Anderson, Stephen. 1971. On the role of deep structure in semantic interpretation. Foundations of Language 6: 197–219. Bellosta von Colbe, Valeriano. 2004. Das indirekte Objekt als syntaktisches Argument ohne Makrorolle. In Kailuweit and Hummel (eds.), 183–204. Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: De Gruyter. Bickel, Balthasar. 2011. Grammatical relations typology. In Jae Jung Song (ed.), The Oxford Handbook of Linguistic Typology, 399–444. Oxford: Oxford University Press. Bickel, Balthasar, Taras Zakharko, Lennart Bierkandt and Alena WitzlackMakarevich. 2014. Semantic role clustering: An empirical assessment of semantic role types in non-default case assignment. Studies in Language 38(3): 485–511. Borer, Hagit. 1994. On the projection of arguments. In Elena Benedicto and Jeffrey Runner (eds.), Functional Projections, 19–47. Amherst: GSLA.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

261

262

ROLF KAILUWEIT

Borer, Hagit. 2005. Structuring Sense II: The Normal Course of Events. Oxford: Oxford University Press. Comrie, Bernard. 1978. Ergativity. In Winfred P. Lehmann (ed.), Syntactic Typology: Studies in the Phenomenology of Language, 329–394. Austin: University of Texas Press. Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in a RRG-perspective. In Kailuweit et al. (eds.), 87–145. Dixon, Robert M. W. 1979. Ergativity. Language 55: 59–138. Dowty, David. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel. Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619. Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative. Language 62: 808–845. Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert Harms (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart and Winston. Fillmore, Charles J. 2003. Valency and semantic roles: The concept of deep structure case. In Vilmos Ágel, Ludwig M. Eichinger, Hans Werner Eroms, Peter Hellwig, Hans Jürgen Heringer and Henning Lobin (eds.), Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung / Dependency and Valency: An International Handbook of Contemporary Research 1, 457–475. Berlin: Walter de Gruyter. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. François, Jacques. 1997. La place de l’aspect et de la participation dans les classements conceptuels des prédications verbales. In Jacques François and Guy Denhi`ere (eds.), Sémantique linguistique et psychologie cognitive. Aspects théoriques et expérimentaux, 119–156. Grenoble: Presse Universitaire. Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA: MIT Press. Gruber, Jeffrey S. 1965 [1976]. Studies in Lexical Relations. PhD dissertation, MIT. Published as: Lexical Structures in Syntax and Semantics. Amsterdam: North-Holland. Guerrero, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of primary object Languages. International Journal of American Linguistics 70(3): 290–319. Haspelmath, Martin. 2008. Ditransitive constructions: Towards a new Role and Reference Grammar account? In Van Valin (ed.), 75–100. Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for alignment typology. Linguistic Typology 15: 535–689. Hockett, Charles. 1958. A Course in Modern Linguistics. New York: Macmillan. Ibáñez Cerda, Sergio. 2008. ‘Saying’ verbs in Spanish. Deepening the lexical semantics description. In Van Valin (ed.), 3–21. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

Jackendoff, Ray. 2007. Language, Consciousness, Culture: Essays on Mental Structure (Jean Nicod Lectures). Cambridge, MA: MIT Press. Kailuweit, Rolf. 2004. Protorollen und Makrorollen. In Kailuweit and Hummel (eds.), 84–104. Kailuweit, Rolf. 2005a. Lokativalternanz bei transitiven Verben – Englisch, Französisch, Spanisch und Deutsch im Vergleich. In Barbara Wotjak and Christian Schmitt (eds.), Beiträge zum romanisch-deutschen und innerromanischen Sprachvergleich, Vol. 2, 183–196. Bonn: Romanistischer Verlag. Kailuweit, Rolf. 2005b. Linking: Syntax und Semantik französischer und italienischer Gefühlsverben. Tübingen: Niemeyer (Linguistische Arbeiten). Kailuweit, Rolf. 2008. A RRG description of locative alternation verbs in English, French, German and Italian. In Kailuweit, Wiemer, Staudinger and Matasovi´ c (eds.), 328–355. Kailuweit, Rolf. 2012. Macropapeles: entre semántica y sintaxis. In Carlos González Vergara, Lilián Guerrero and Ricardo Mairal (eds.), El funcionalismo en la teoría lingüística: La Gramática del Papel y la Referencia, 103–123. Tres Cantos: Ediciones Akal. Kailuweit, Rolf. 2013. Radical Role and Reference Grammar (RRRG): A sketch for remodelling the syntax–semantics interface. In Brian Nolan and Elke Diedrichsen (eds.), Linking Constructions into Functional Linguistics: The Role of Constructions in Grammar, 103–41. Amsterdam: John Benjamins. Kailuweit, Rolf. 2018. Activity Hierarchy and argument realization in (R)RRG. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 189–213. Freiburg: Albert-LudwigsUniversität, Universitätsbibliothek. Kailuweit, Rolf and Martin Hummel (eds.). 2004. Semantische Rollen. Tübingen: Narr. Kailuweit, Rolf, Björn Wiemer, Eva Staudinger and Ranko Matasovi´ c (eds.). 2008. New Applications of Role and Reference Grammar. Newcastle upon Tyne: Cambridge Scholars Publishing. Kibrik, Alexander E. 1985. Toward a typology of ergativity. In Johanna Nichols and Anthony Woodbury (eds.), Grammar Inside and Outside the Clause, 268–323. Cambridge: Cambridge University Press. Kibrik, Alexander E. 1997. Beyond subjects and objects: Towards a comprehensive relational typology. Linguistic Typology 1: 279–346. Langacker, R. W. 1990. Concept, Image, and Symbol. Berlin: Mouton de Gruyter. Langacker, R. W. 1991. Foundations of Cognitive Grammar. 2: Descriptive Application. Stanford, CA: Stanford University Press. Lehmann, Christian, Yong-Min Shin and Elisabeth Verhoeven. 2004. Direkte und indirekte Partizipation. Zur Typologie der sprachlichen Repräsentation konzeptueller Relationen. Munich: Lincom Studies in Language Typology. Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity at the Syntax– Lexical Semantics Interface. Cambridge, MA: MIT Press.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

263

264

ROLF KAILUWEIT

Nissenbaum, Helen Fay. 1985. Emotion and Focus. Stanford, CA: CSLI Publications. Pinker, Steven. 1989. Learnability and Cognition: The acquisition of Argument Structure. Cambridge, MA: MIT Press. Primus, Beatrice. 1999. Case and Thematic Roles: Ergative, Accusative, Active. Tübingen: Niemeyer. Rappaport Hovav, Malka and Beth Levin. 1988. What to do with theta roles. In Wilkins (ed.), 7–36. Rappaport Hovav, Malka and Beth Levin. 2015. The syntax–semantics interface. In Shalom Lappin and Chris Fox (eds.), The Handbook of Contemporary Semantic Theory (2nd ed.), 593–624. Oxford: Wiley-Blackwell. Rauh, Gisa. 1988. Tiefenkasus, thematische Relationen und Thetarollen. Tübingen: Narr. Reinhart, Tanya. 2002. The Theta system: An overview. Theoretical Linguistics 28(3): 229–290. Rozwadowska, Bozena. 1988. Thematic restrictions on derived nominals. In Wilkins (ed.), 147–165. Ruwet, Nicolas. 1972. Théorie syntaxique et syntaxe du français. Paris: Seuil. Staudinger, Eva, Matthias Hartung and Rolf Kailuweit. 2008. Linking syntax to semantics: template selection and PP-Attachment ambiguities. In Kailuweit, Wiemer, Staudinger and Matasovi´ c (eds.), 389–413. Tenny, Carol. 1994. Aspectual Roles and the Syntax–Semantics Interface. Dordrecht: Kluwer. Tesni`ere, Lucien. [1959] 1965. Eléments de syntaxe structurale. Paris: Klincksiek. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Robert D. Van Valin Jr. (ed.), Advances in Role and Reference Grammar, 1–164. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax– semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin (eds.), Empirical Issues in Formal Syntax and Semantics, Vol. 2, 373–389. The Hague: Thesus. Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen, 62–82. Tübingen: Narr. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D, Jr. 2006. Semantic macroroles and language processing. In I. Bornkessel et al. (eds.), Semantic Role Universals and Argument Linking: Theoretical, Typological and Psycholinguistic Perspectives, 263–302. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. 2007. The Role and Reference Grammar analysis of three-place predicates. Suvremena Lingvistika 63(1): 31–64. Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface. Amsterdam: John Benjamins.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

Van Valin, Robert D., Jr. 2010. Role and Reference Grammar as a framework for linguistic analysis. In Bernd Heine and Heiko Narrog (eds.), The Oxford Handbook of Linguistic Analysis, 703–738. Oxford: Oxford University Press. Van Valin, Robert D., Jr. 2013. Lexical representation, co-composition, and linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon, Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in Generative Lexicon Theory, 67–107. Dordrecht: Springer. Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplishments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 5–26. Freiburg: Freiburg University Library. Van Valin, Robert D., Jr. and Randy R. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Van Valin, Robert. D., Jr. and David Wilkins 1996. The case for ‘effector’: Case roles, agents, and agency revisited. In Masayoshi Shibatani and Sarah. A. Thompson (eds.), Grammatical Constructions: Their Form and Meaning, 289–321. Oxford: Oxford University Press. Wilkins, Wendy (ed.). 1988. Thematic Relations, Vol. 21: Syntax and Semantics. New York: Academic Press.

Notes 1 The expression ‘generalized semantic roles’ has been established as a neutral term to refer to the RRG macroroles, proto-roles (Dowty 1991), Kibrik’s (1985, 1997) hyperroles, and Langacker’s (1990, 1991) role archetypes, etc. (see Van Valin 1999; Kailuweit 2004; Rappaport Hovav and Levin 2015). 2 Tesni`ere ([1959] 1965) calls the prepositional phrase of the passive construction a ‘counter-agent’, but he fails to capture the semantics of the subject of the passive construction. 3 According to Levin and Rappaport Hovav (1995), in the case of intransitive verbs, telicity and directed change are criteria for unaccusativity. 4 ‘The telicity of the event depends on whether the patient of the event shows a bounded or unbounded change in the gradable property associated with the verb’ (Rappaport Hovav and Levin 2015: 607). 5 Following the prevailing view at that time that all accomplishments are causative and that all causatives are accomplishments (Dowty 1979: 186), Grimshaw (1990) considered causativity an aspectual property. However, later work in RRG (Van Valin and LaPolla 1997: 97) has shown that causativity is independent of Aktionsart (see Section 4.4.1). 6 Logical structures in RRG are based on Dowty’s (1979) formalism. As will become clear when causativity is discussed, the details of Aktionsartdriven verb classification were later developed in RRG in ways which diverge from Dowty (1979).

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

265

266

ROLF KAILUWEIT

7 However, even in later versions of RRG the concept of macrorole is justified by the assumption that each macrorole ‘subsumes a number of specific argument types (thematic relations). The generalized agenttype role will be termed actor and the generalized patient-type role will be called undergoer’ (Van Valin and LaPolla 1997: 141; see also Van Valin 2006: 60). 8 ‘Actor and undergoer are generalizations across classes of specific argument positions in logical structure’ (Van Valin and LaPolla 1997: 142). 9 For mnemotechnical purposes, the first argument of kill is not named agent but effector. The latter is a generalized label for the (do′ (x,. . .)) position (Van Valin and Wilkins 1996). 10 There are isolated cases of non-causative achievements and accomplishments with a do′-component in their LS. INGR or BECOME plus activity predicate describe an onset of an action, e.g. Russian zaplakat’ ‘burst out crying’ (INGR do′ (x, [cry′ (x)])) vs. zagovorit’ ‘start talking’ (BECOME do′ (x, [talk′(x)])) (Van Valin and LaPolla 1997: 104). 11 Van Valin and LaPolla (1997: 143) expresses this idea referring to thematic relations and not to the positions on the hierarchy they stand for: ‘Actor and undergoer are generalized semantic roles whose prototypes are the thematic relations AGENT and PATIENT, respectively’ (see also Van Valin 2004: 63; Van Valin 2005: 60). 12 This is not only the case with three-place constructions such as to give somebody something, but also with two-place activities such as to eat fish. See Van Valin and LaPolla for one such case, Italian mangiare (‘eat’) (1997: 148–150). 13 In this example, the symbol Ø refers to a unspecific action that can cause death. 14 The so-called dative alternation does not appear in German or in the standard varieties of Romance languages (but see Abreu Gomes (2003) for colloquial Brazilian Portuguese). Kailuweit (2005a) showed that Pinker’s (1989) semantic subclass approach to locative alternation does not hold for German or Romance languages. Some of the English subclasses that, according to Pinker, allow for locative alternation do not have any instances in French, Spanish or German, while other classes that do not permit alternation in English do so in German, French or Spanish. However, in every alternating class in these languages, we also find verbs that do not alternate. Hence, there is no uniform relation between Pinker’s semantic classes of locative verbs and locative alternation (see Kailuweit 2005a, 2008). 15 In a primary object language the recipient argument of a ditransitive verb – the y-argument of BECOME be.at’(y,z) in the LS of these verbs – is treated like the patient of a monotransitive verb. The theme argument – the z-argument of BECOME be.at’(y,z) – is marked in a distinct way (Dryer 1986).

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

Semantic Macroroles

16 Van Valin (2004: 74–78) pointed out that macroroles are basically semantic categories that play a role in syntax. Syntax is considered nonautonomous in RRG. Therefore, syntactic (in)transitivity is at least partly semantically motivated. 17 Primus (1999: 36) substitutes Dowty’s (1991) proto-agent property ‘volition’ by ‘control’ to describe cases that exclude ‘volition’ in the strict sense, for example responsibility or the mere capacity to start or stop an event. In RRG, too, responsibility is referred to, at least informally, to define the properties of the actor: ‘the actor is the entity to which responsibility for the action or event is attributed’ (Van Valin and LaPolla 1997: 143). 18 ‘Seeing, thinking, liking, believing, etc., involve some kind of internal activity (mental, emotional or perceptual) on the part of the participant, whereas being seen, being thought about, being liked or being believed does not require any action or effort of any kind on the part of the participant. Hence the participant denoted by the first argument is more active and hence more agent-like than the participant referred to by the second argument, and, accordingly, the first argument is closer to the agent end of the hierarchy than the second argument’ (Van Valin 2005: 58). 19 The existence of non-causative three-place predicates is widely ignored in the literature. Haspelmath (2008: 14) refers at least incidentally to talk as a verb ‘lacking an undergoer’, adding that ‘intuitively it is strange to claim that ditransitive constructions are intransitive’. 20 Ibáñez Cerda (2008: 6) criticizes this representation, highlighting the fact that the α variable could be realized as a direct object with verbs of saying. 21 Two other classes of two-place activities are verbs of consumption and creation with an unspecified object: eat pizza, write letters. Note that the objects of these predicates are not affected undergoers. 22 ‘Envy defines three central roles, that of the envier, the envied, and a feature or possession of the one envied, over which he is envied’ (Nissenbaum 1985: 108). 23 The clitic en is a desemantized part of the lexical realization of the predicate that could be interpreted as a generic placeholder for the state of affairs (the correlate) the anger is about. Curiously, the de-complement that can normally be substituted by the clitic en appears alongside the clitic in the sentence. It is obviously not a fourth argument, but a coreferential concretization of the correlate. 24 Spanish rabiar is somewhat different. It does not only accept a person as the second argument but also an object. Hence, it corresponds to English rage against. 25 For a discussion of Primus’s approach, and its compatibility with the RRG framework, see Bellosta von Colbe (2004). 26 Bellosta von Colbe (2004) uses the same argument to criticize Primus’s proto-recipient approach.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

267

268

ROLF KAILUWEIT

27 Bickel et al. (2014) use P (for patient) instead of O and T (for theme) instead of X. See Haspelmath (2011) for an overview on the history of denominations for generalized roles in typology. 28 Haspelmath (2011) and other authors refer to G as R (for recipient). 29 Bickel et al. (2014) refer to entailments in line with Dowty (1991), but Dowty’s approach allowing 1,024 theoretically possible combinations of proto-role features turns out to be a rather inaccurate tool to determine the activity degree of an argument in comparison with the AUH. 30 Note that French envier and German neiden/beneiden allow for flexible undergoer selection, i.e. a construction selecting the second passive argument as undergoer (secondary alignment in a language of indirective alignment). Kailuweit’s (2013) approach does not account for this alternation. 31 Kailuweit (2018) discusses further classes of predicates, especially verbs of directed motion such as climb, enter or reach.

https://doi.org/10.1017/9781316418086.007 Published online by Cambridge University Press

5 Grammatical Relations Randy J. LaPolla

5.1

Introduction

This chapter discusses one of the major advances of Role and Reference Grammar (RRG) relative to other theories of grammar: the view of grammatical relations as construction-based, and so not only language-specific but construction-specific, rather than being global categories of the whole language and found in every language. We also discuss the RRG conception of the function of grammatical relations in referent tracking, which was one of the major insights that led to the development (and naming1) of RRG. These two insights have influenced the development of ideas outside RRG.

5.2

Background

Before the mid 1970s, a common assumption among linguists was that there is a global category in all languages called ‘subject’ as well as other grammatical relations that we can talk about, and most theories assumed some conception of grammatical relations, though there was much disagreement about and no universal notion of ‘subject’, the grammatical relation discussed the most (Platt 1971; Van Valin 1977, 1981; Foley and Van Valin 1977, 1984; Keenan 1976; Gary and Keenan 1977; Comrie 1981). The word ‘subject’ derives from a Latin translation (subiectum) of Greek hypokeímenon ‘the underlying thing’, a concept that began with Aristotle’s theory of truth, where Aristotle defined ‘subject’ as the entity that can have a predication about it, that is, what the proposition is about, the topic about which a predication is made. Aristotle did not have a separate term for grammatical subject. This led to centuries of debate about the nature of subject (see Seuren 1998, pp. 120–133, for an overview). Attempts were made to distinguish grammatical subject from psychological subject (e.g. von der Gabelentz 1869: 378), the latter essentially topic, and what was called ‘theme’ in the Prague School

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

270

RANDY J. LAPOLLA

terminology. A third term, ‘logical subject’ (often now seen as agent), was sometimes used, but different scholars associated it either with grammatical subject or with psychological subject (particularly in logic). Bloomfield (1914: 61) used the term ‘subject’ to refer to topics and also to heads of phrases. Starting with Van Valin (1977, 1981) and Foley and Van Valin (1977, 1984), there were challenges to the notion of ‘subject’ and other grammatical relations (‘direct object’, ‘indirect object’) as global categories within a single language, and as valid categories cross-linguistically (see also Dryer 1997). Currently there are three major positions on this question: (1) grammatical relations are global within a language and universal cross-linguistically, and just need to be identified in different languages (the rationalist/generativist/Chomskyan tradition); (2) grammatical relations exist, but are not necessarily global and not universal, and so need to be defined in each language in terms of the constructions that manifest them, if there are any (most empiricist/typological/explanatory approaches); and (3) there are no grammatical relations, only part–whole relations within constructions (Radical Construction Grammar; Croft 2001, 2013). Marantz (1982, 1984) has argued that grammatical relations should not be seen as primitives or tied to semantic roles. For example, ‘subject’, as a grammatical category, is not simply a particular semantic role, such as agent (see also Jespersen 1909–1949, vol. III, 11.1). ‘Subject’ is also not simply topic; it must have grammatical properties beyond just being what the clause is about. Empiricist linguists would generally agree with this position.

5.3

The RRG View of Grammatical Relations

The RRG view of grammatical relations is of the second type mentioned above, though it does not accept the traditional typology of grammatical relations as ‘subject’, ‘direct object’ and ‘indirect object’. Grammatical relations are seen as construction-specific conventionalized patterns where the construction limits the possible interpretations of the role of a particular participant in the action described in the clause (see Van Valin and LaPolla 1997: 242–316; Van Valin 2005; LaPolla 2006). It is the identification of the semantic and/or pragmatic role of the referent in an event or state of affairs that is relevant to the concept of grammatical relations. There are other types of conventionalized constraints on referent identification in some languages, such as the gender or noun class markers in many languages, and the sortal classifiers of Thai and Chinese, but while they do participate in referent tracking (see Van Valin 1987 for a typology of referent tracking systems), as these are not relational and do not constrain the interpretation of the role of the referent in the event, they are not considered relevant to grammatical relations.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

Grammatical relations may be conventionalized associations of the position of a referring expression in the clause with some semantic role or macrorole, such as in English, where an immediate preverbal reference to some referent in a clause with an active transitive verb will constrain the interpretation to one in which that referent is seen as the Actor2 of the action denoted by the verb, and an immediate postverbal reference to some referent in the same clause will constrain the interpretation to one in which that referent is seen as the Undergoer of the action (e.g. given the expression Bob hit Bill, the conventions of English usage constrain the interpretation to one in which Bob must be understood as the one doing the hitting and Bill must be understood as the one being hit). This is construction-specific and language-specific, that is, not all constructions in English work that way and not all languages have the same constraints on interpretation. Grammatical relations may be conventionalized associations of marking on nouns or pronouns with particular semantic roles, such as in English, where Nominative case marking of a pronoun in an active transitive clause constrains the interpretation to one in which the referent of the pronoun is seen as the Actor of the action denoted by the verb, and Accusative case marking of a pronoun constrains the interpretation to one in which the referent of the pronoun is seen as the patient or recipient of the action, such as in HeNominative took meAccusative to the station. In the case of Modern English, the case marking is largely redundant, in that grammatical relations are also marked by word order, and in fact case marking has largely come to be determined by word order, but this was not the case in older varieties of English, which did not use word order consistently to mark grammatical relations, and in many other languages, such as Dyirbal (see below). In these languages the case marking is very important not only for tracking the roles of referents in discourse, but also for tracking the different constituent parts of phrases, as they do not necessarily appear together in the clause. Grammatical relations may also be conventionalized assumptions that referring expressions in two clauses both refer to the same referent, such as in English, where there is a conventionalized assumption of coreference in conjoined clauses such that a referring expression representing a particular role in one of the clauses and a particular role implied in the other clause must be understood as coreferential (e.g. in Jim picked up the newspaper and threw it, there is a forced assumption that the referent of Jim is the same referent as the implied Actor of the second clause, the one that threw the newspaper).3 Many other possible ways of constraining this particular functional domain exist as well. See Section 5.3.3 for more discussion. Each of these conventionalized forms or constructions has the function of limiting the possible interpretation(s) of the role of a referent referred to or implied in an utterance, to aid in the interpretation of the identity and role of the referent. Although traditionally these different constructions have

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

271

272

RANDY J. LAPOLLA

been seen as part of one grammatical category (e.g. ‘subject’), they are not one category, but instead are individual ways of constraining the interpretation of who is doing what to whom, and languages differ in terms of whether or not they constrain this functional domain at all, and if they do constrain it, they differ in terms of which particular roles are identified, and the particular mechanisms used to constrain the interpretation.

5.3.1 Privileged Syntactic Argument: Controller and Pivot As the phenomenon we are talking about is construction-specific, and there are many different types of restricted neutralizations, the term ‘subject’ is not appropriate, and instead we use the term ‘privileged syntactic argument’ (PSA) for an argument that is the controller or pivot of a restricted neutralization of semantic roles for grammatical purposes, generally referent tracking (i.e. keeping track of who is doing what in discourse). No other grammatical relations are recognized in this approach; the characteristics that have traditionally been associated with ‘direct object’, such as taking accusative case, being able to appear as the privileged syntactic argument in a passive clause or being the target of applicative constructions, are seen as properties of the macrorole Undergoer, while those associated with ‘indirect object’ are seen as properties of the non-macrorole direct core argument (see the discussion of (5) below). The terms ‘controller’ and ‘pivot’ (first used in Heath 1975) refer to different types of PSA, as in the examples in (1), the English conjoined clause coreference construction, where the immediately preverbal (core-initial) referring expression in the first clause is the controller, while the implied argument of the second clause is the pivot of the construction (marked by ‘(pivot)’, where the implied argument would be if it appeared in an independent clause), regardless of whether either clause is an active or passive construction. (1)

a. Bobcontroller handed Jim the money and (pivot) left. [A,S] b. Jimcontroller was handed the money by Bob and (pivot) left. [S,S] c. The moneycontroller was handed to Jim but (pivot) not seen again after that. [S,S] d. Jimcontroller took the money and (pivot) was seen later buying a new car. [A,S] e. Jimcontroller took the money and (pivot) thanked Bob for it. [A,A] f. Jimcontroller smiled and (pivot) took the money. [S,A]

In this construction, there is obligatory coreference between the controller, whether it is the Actor (A) of an active transitive clause or single direct argument of an intransitive clause (S)4 or the Undergoer of a passive clause (S), and the pivot, again whether it is A or S. The construction then aids in the inference of who is doing what. The restricted neutralization we find in this particular construction is [A,S], but as shown in the examples, the correspondence can be [A,A] or [S,S] as well, as what is important is that

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

the arguments participating in the coreference are grammatically either A or S; the Undergoer of an active transitive clause does not participate in this coreference. The Undergoer argument cannot simply be left to implication when using this structure, and so a different construction must be used to allow the Undergoer to participate in the coreference of the conjoined clause coreference construction. The construction used is the English passive construction, as in (1b, c ,d), as it is an intransitive clause in which the Undergoer is the PSA. Using the passive construction in this conjoined clause coreference construction allows the Undergoer argument (regardless of what specific semantic role it has) to participate in the coreference construction by casting it as an S. The English passive construction allows limited variable access to the syntactic controller and syntactic pivot positions. Without that construction the Undergoer would not be able to appear as the PSA. Some languages manifest constructions with a particular restricted neutralization but do not have constructions that allow variable access to the PSA (i.e. they do not have alternative voice constructions such as passive; e.g. Enga (Papua New Guinea), Warlpiri (Australia), Lakhota (North America); see Van Valin and LaPolla 1997, pp. 274–285 for details). These constructions are said to have an invariable syntactic controller and invariable syntactic pivot. Some languages have one or a few constructions (e.g. the English passive construction or the Dyirbal antipassive construction – see the next section) that allow limited variable access to the PSA. These constructions are said to have a variable syntactic controller and a variable syntactic pivot. And some languages, such as Tagalog (see Section 5.3.2), have many constructions allowing a range of semantic roles access to the PSA. These constructions have a semantic controller with no neutralization. As we will see, in Tagalog there is no neutralization we can call S, nor even neutralization of a single grammatical category of Actor or Undergoer in terms of the marking on the predicate, as different types of actor and undergoer (e.g. with different degrees of intention, agentivity, transitivity, affectedness) are marked differently on the predicate depending on the nature of the action and the Topic of the clause.5 In other constructions (e.g. the Reflexive Construction) there is a generalized Actor, and it is an invariable semantic controller. For ease of discussion we will generalize across the different types of actor, patient and location in Tagalog by glossing the various relevant forms as ‘Actor-Topic’, ‘Patient-Topic’ or ‘LocationTopic’, respectively.

5.3.2 Alignment The particular PSAs in the conjoined clause coreference construction discussed here are found in English, but many other languages, even closely related ones, do not manifest PSAs in conjoined clause coreference constructions, and so the determination of the relevant argument of the second

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

273

274

RANDY J. LAPOLLA

clause would be left to inference unconstrained by the grammar, and undergoers in active clauses could be left to inference as easily as A and S. It is also the case that some languages manifest constructions for conjoined clause coreference with particular PSAs, but the restricted neutralization is [S,U] rather than [S,A].6 Dyirbal, a Pama-Nyungan language spoken in northeastern Australia (Dixon 1972, 1980), is a well-known example of a language that has grammaticalized an [S,U] restricted neutralization for conjoined clause coreference. That is, the arguments participating in the coreference must be U and/or S, but A does not participate in the coreference unless it is cast as an S in the Dyirbal antipassive construction. Consider the examples in (2) (adapted from Dixon 1980: 462). (2)

a. balan guda buŋa-n baŋgul yara-ŋgu bura-n 3sgf.abs dog.abs descend-pst 3sgm.erg man-erg see-pst The dogcontroller went downhill and the man saw (pivot). b. bayi yara buŋa-n buralŋanyu bagun guda-gu 3sgm.abs man.abs descend-pst see:pst:antip 3sgm.abs dog-dat The mancontroller went downhill and (pivot) saw the dog. c. balan guda baŋgul yara-ŋgu bura-n buŋa-n. 3sgf.abs dog.abs 3sgm.erg man-erg see-pst descend-pst The man saw the dogcontroller and (pivot) went downhill.

These examples parallel the English examples in (1), but the interpretation of the implied argument in the second clause of (2a) is obligatorily coreferential with the Undergoer in the first clause, that is, it must be that the dog is the controller and pivot (i.e. is the one that went downhill). In order to have coreference that involves an A argument, the Dyirbal antipassive construction must be used. This construction is an intransitive construction with the verb marked with the antipassive marker ŋa, and has the Actor as the single direct argument in the absolutive case and the Undergoer in the dative case. Because Dyirbal has this antipassive construction, there is variable access to the controller and pivot positions, and so the controller is a variable syntactic controller and the PSA is a variable syntactic PSA, though they manifest a different restricted neutralization from the corresponding English construction. These grammaticalized constraints on interpretation we have been looking at force a particular interpretation of an utterance in both English and Dyirbal, but as the restricted neutralizations are different, the interpretations are different. For example, if The man saw the dog and went downhill is said in English, the interpretation has to be that the man went downhill; but if the corresponding structure is used in Dyirbal, as in (2c), the meaning has to be that the dog went downhill. We see here that the construction must be taken as a whole, as it is the total construction that influences the interpretation, and is not simply the sum of the individual words. Where there is a choice of argument for PSA, the RRG theory of PSA selection posits a default choice, depending on the Privileged Syntactic

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

Argument Hierarchy (3) and the privileged syntactic argument selection principles (4): actor for PSA in so-called nominative-accusative alignment, and Undergoer in ergative-absolutive alignment. (3)

Privileged Syntactic Argument Selection Hierarchy arg. of DO > 1st arg. of do′ > 1st arg. of pred′ (x, y) > 2nd arg. of pred′ (x, y) > arg. of pred′ (x)

(4)

Privileged syntactic argument selection principles a. Syntactically accusative constructions: highest-ranking macrorole is default choice. b. Syntactically ergative constructions: lowest-ranking macrorole is default choice.

Yet it isn’t the case that all languages necessarily have such a default choice. As Foley and Van Valin argued (1984, §4.3) Tagalog has many different constructions for having different semantic roles as PSA, yet none is a default choice. In all these languages where we have seen a choice of PSA, the choice of pivot is determined by pragmatic factors, such as the identifiability or topicality of the referent involved (the default is used when there is no difference in terms of the pragmatic factors), and so we refer to these as pragmatic PSAs, as opposed to those based strictly on semantic factors, which we call semantic PSAs. While the conjoined cross-clause coreference construction in English manifests an [S,A] restricted neutralization, it is not the case that all constructions in English manifest a restricted neutralization, and it is not the case that all of the constructions that do manifest a PSA in English manifest the same restricted neutralization. For example, in the following example of the English purposive construction the controller and pivot of the construction are the Undergoers of the two clauses, not the Actor or S of the clauses: (5)

He left this formcontroller for you to sign (pivot).

As in the construction in (1), the controller determines the reference of the implied argument in the second clause, but in this case the referent referred to appears as the traditional direct object (the Undergoer in RRG) of the first clause, and is understood as the traditional direct object (the Undergoer in RRG) of the second clause as well. This shows that the concept of PSA is not the same as the traditional concept of ‘subject’. Aside from the possibility of different restricted neutralizations, there is also the possibility of unrestricted neutralization. Some languages, such as Riau Indonesian (Gil 1994) and Mandarin Chinese (LaPolla 1993), have not conventionalized any constraints on referent role identification of the type associated with grammatical relations in any constructions in the language (though they may have conventionalized other types of constraints on interpretation). This means that in all the constructions we have looked

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

275

276

RANDY J. LAPOLLA

at interpretation is based entirely on inference from context and is not constrained by the structure. In other languages, only some constructions will have unrestricted neutralization. It is entirely language-specific and construction-specific. In English, for example, the relative clause construction manifests neutralization, but no restriction, that is, any argument of the modifying clause can appear as the head of this construction, as shown in (6): (6)

a. b. c. d. e. f.

the girl who[A] sang the song the girl who[U] the police saved the girl who[S] just came in the girl to whom[non-macrorole core argument] the award was given the car in which[peripheral argument: location] the man was held up the car out of which[peripheral argument: source] the radio was stolen

In Tagalog, a Malayo-Polynesian language spoken in the Philippines, on the other hand, there is a restricted neutralization for the Tagalog relative clause construction: the head of the construction must be the grammatical Topic of the modifying clause, and so it is a syntactic pivot for that construction. To explain this we will first give some background on the structure of the Tagalog main clause. Tagalog is a consistently focus-initial language and so in general the predicate appears in initial position and the Topic appears (when it is not a pronoun and not in focus) at the end of the clause, and the information structure is Comment-Topic rather than Topic-Comment. The predicate in most cases marks aspect, realis/irrealis, and the semantic role of the Topic of the clause, and so the Topic is the controller of the semantic role marking (it is a semantic and not syntactic controller because the marking differs with each semantic role, i.e. there is semantic restriction but no neutralization relative to each type of marking). Topic here is actually a grammatical status, as it is an argument singled out for special morphological treatment, as well as a pragmatic status, as it is what the clause is about. Generally, almost any referent associated with the situation in some way, whether core or peripheral argument semantically or even a very indirectly affected referent, can be the Topic of the clause, though usually (but not always) it is one that is identifiable to the hearer, and in many cases the predicate takes a form to reflect the semantic role of that referent, and the marking of the other references in the clause is usually also different.7 These different constructions are not passive or antipassive, but simply different ways of profiling the event (Foley and Van Valin 1984, §4.3), similar to the choice of the A Construction vs. the O Construction in Jawarawa (Dixon 2000, 2004), depending on what is considered to be what the clause or discourse is about, but with more choices for Topic in Tagalog. The representation of the Topic argument (if it appears in the clause and is not a pronoun) generally takes either a marker of specificity (ang or si with proper names of single referents) or a topic form of demonstrative pronoun

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

(most commonly ’yung) to mark it as the Topic. The set of pronouns, which appear as second-position clitics, also distinguishes Topical from nonTopical referents, with the latter appearing as possessive forms or dative forms. In (7) there is a short natural conversation to exemplify this feature (from my own fieldwork; see also Schachter (2008: 337–338) for sets of constructed parallel examples with the same arguments but with different choices of topic, and Schachter and Otanes (1972, Ch. 5) for many of the major constructions used for marking different semantic roles).8 (7)

1. Jirehel: Madali lang kasi’ng gumawa ng salsa eh. ma-dali lang kasi ang gawa ng salsa eh stat-easy just because spec make poss sauce emphatic ‘Making sauce is easy.’ 2. Wendy: Oo, madali lang. oo ma-dali lang yes stat-easy just ‘Yes, it’s easy. 3. Gawin mo lang ketsap, gawa-in mo lang ketsap do-pt:irr 2sg.poss just ketchup You just make it with ketchup, 4. tapos lagyan mo ng tomatoes, tapos lagay-an mo ng tomatoes finish put-lt 2sg.poss poss tomatoes then you add tomatoes, 5. lagyan mo ng salt and pepper to taste, tapos na. lagay-an mo ng salt and pepper to taste tapos na put-lt 2sg.poss poss salt and pepper to taste finish cs add salt and pepper to taste; then, it’s done.’

In (7), the first speaker, Jirehel, refers to the making of sauce using a form (gumawa) where the infix -um- marks it as Actor-Topic (it happens that in this utterance the speaker has made the relevant clause the Topic of a higher clause, but the phenomenon is the same), then the second speaker, Wendy, uses the same root in line 3 of the example, but in the irrealis Patient-Topic construction, to profile the event from the point of view with the sauce as the Topic. She then follows this in lines 4 and 5 with two tokens of the root lagay ‘put’ in the Location-Topic construction, to keep the sauce as the Topic, but with the sauce now having the semantic role of the location where the tomatoes and salt and pepper are to be added. Notice how there is no overt reference to the Topic in any of Wendy’s utterances (e.g .’yung salsa ‘the sauce’ could have been added to the end of each of Wendy’s utterances in lines 3–5, but it wasn’t), yet we can tell what is being talked about (what is the pragmatic and grammatical Topic) because of the marking on the predicate. We now can return to the issue of grammatical relations, that is, restricted neutralizations. The choice of different roles as Tagalog Topic

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

277

278

RANDY J. LAPOLLA

affects the particular form of the construction and the behaviour of the Topic in this and other constructions, such as the Tagalog relative clause construction, where the head of the construction generally must be the Topic of the modifying predicate (i.e. the form of the predicate must be the same as if it were a main clause with that referent as Topic). For example, if we were to recast the clauses in (7) as relative clause constructions (and with realis predicates), we would get the forms in (8): (8)

a. lalaking gumawa ng salsa lalake ¼ng gawa man lnk make b. salsang ginawa ng ketsap salsa ¼ng gawa ng sauce lnk poss c. salsang nilagyan ng tomatoes salsa ¼ng in-lagay-an ng sauce lnk realis-put-lt poss

‘man who made salsa’ ng salsa poss sauce ‘salsa made with ketchup’ ketsap ketchup ‘salsa in which tomatoes were put’ tomatoes tomatoes

In Tagalog, then, for this construction there is a clear restricted neutralization: the head of the construction must be the grammatical Topic of the predicate that modifies it, unlike in English, where there is neutralization but no restriction on what argument can be the head of an English relative clause construction. We saw that in Tagalog the controller of the semantic role marking on the predicate (the Topic) is a semantic controller. English agreement, on the other hand, manifests a syntactic controller. That is, the agreement is not with a particular semantic role like in Tagalog, but manifests a restricted neutralization, where the agreement is with the Actor or S of the clause, regardless of its semantic or pragmatic role. We can see this from the examples in (9). (9)

a. I am helping Bill with his homework b. Bill is being held up by the activities c. d. e. f.

Bill is smiling Bill is falling into a trap There are many people in the park There are people grilling meat there

[agreement with [agreement with passive S] [agreement with [agreement with [agreement with [agreement with

A] Undergoer as S Actor] S Undergoer] non-topical S] non-topical A]

Acehnese (Durie 1985, 1987), an Austronesian language spoken in Indonesia, manifests a system of pronominal agreement on the verb with Actormarking prefixes and Undergoer-marking suffixes on the verb. The marking is restricted to Actor and Undergoer, respectively, and there is no neutralization of Actor and Undergoer, even with intransitive constructions. See, for example, the following (from Durie 1987: 366, 369): (10)

a. (Gopnyan) geu-mat (3sg) 3-hold ‘(S)he holds me.’

lôn 1sg

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

b. (Lôn) lôn-mat-geuh (1sg) 1sg -hold-3 ‘I hold him/her.’

Grammatical Relations c. Geu-jak (gopnyan) 3-go (3sg) ‘(S)he goes.’

d. Lôn rhët(-lôn). 1sg fall(-1sg) ‘I fall.’

From these examples we can see that the prefixes are used for the Actors of transitive constructions (10a, b) and the single arguments of intransitive constructions where the action is voluntary and so the argument is an Actor (so-called SA), as in (10c), and the suffixes are used for the Undergoers of transitive constructions and the single argument of intransitive constructions where the action is involuntary (an Undergoer, so-called SP or SO), as in (10d).9

5.3.3 Referent Tracking The different choices of grammatical Topic we saw in Tagalog have the same function in discourse as the variable syntactic pivots in English and Dyirbal in terms of allowing for topic chains where an (often unmentioned) referent can remain the topic across clauses even if its semantic role changes, and the structure helps constrain the inference of the identity and role of the relevant referent, as in (7). This type of referent-tracking mechanism is known as a ‘switch function’ (Foley and Van Valin 1984: 321–374; Van Valin 1987; Van Valin and LaPolla 1997, §6.4), as the structure identifies the difference in function of the referent while maintaining it as topic. There are differences among languages, though, not only in the type of restricted neutralization, such as in English vs. Dyirbal, but also in terms of what referents can be an unmarked topic of the clause: if we contrast English, Dyirbal, Tagalog and Chinese, we can see that English and Dyirbal severely limit what can be topic in the unmarked clause structure ([S,A] or [S,U] respectively); Tagalog allows just about anything relevant to be topic, but marks it morphosyntactically, thereby constraining the inference of the referent and its role; Chinese does not manifest any restricted neutralizations (LaPolla 1993) and so there is no grammatical restriction on what can be topic in the unmarked clause structure, though there is no marking of it as topic other than initial word order, and no marking of its role or identity, and so identification of the referent and its role is not aided by the structure (see, for example, the discussion of (12) below). This forms a separate though related typological cline, from most restricted to least restricted in terms of access to topic, aside from the typological cline related to the degree to which the structure constrains the interpretation of the role and identity of the referent being tracked. Yet another type of restricted neutralization with variable PSAs, known as the ‘switch-reference’ pattern, is found in Barai, a language of Papua New Guinea (Olsen 1978, 1981; Van Valin 2009, §4), and in Choctaw, a Muskogean language of North America (Heath 1975, 1977), among others. In this type of system, when clauses are coordinated, the PSAs of the individual clauses (defined differently in each construction and in each language) can be

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

279

280

RANDY J. LAPOLLA

coreferential or not, and there is marking on the predicate to show this (i.e. whether the PSA is the same as in the following clause or different from the following clause), as in the following Choctaw examples (adapted from Heath 1977: 212): (11)

ca:, a. (0i)-0-pi:sa-ˇ (3A)-3P-see-same b. 0-0i-pi:sa-na:, 3A-3P-see-different

0i-iya-h 3A-go-present 0i-iya-h 3A-go-present

‘Hei sees himj and hei goes’ ‘Hei sees himj and hej/k goes’

As can be seen from these examples, even without overt arguments, the affix on the predicate marking whether the PSA in the following clause is the same or different from that of the marked clause constrains the interpretation of who is doing what. The pivots in this sort of system are generally invariable syntactic pivots. We have seen above that there are different kinds of constructions languages can have for constraining the inference of who is doing what, if they have any at all. They may have some constructions that have invariable semantic pivots (i.e. restriction with no neutralization), such as the verbal marking in Tagalog and Acehnese, or invariable syntactic pivots, such as reference across conjoined clauses in Warlpiri and Enga, or variable syntactic pivots, such as reference across clauses in Dyirbal and English, and the relative clause construction in Tagalog. The restricted neutralization found in a construction could treat [S,A] the same in opposition to [U], or it may treat [S,U] the same in opposition to [A]. As this is a construction-specific phenomenon, the same language may have different constructions with different PSAs, as we saw with Tagalog (see also Van Valin 1981 on Jakaltek and Van Valin and LaPolla 1997: 282ff. on Tzutujil), and constructions with different PSAs can sometimes be combined into a single complex structure, such as when a Tagalog reflexive construction (invariable semantic pivot) and any type of Tagalog clause structure (variable syntactic pivot) are combined to form a complex structure. And of course a language may not have any constructions that manifest PSAs, such as Mandarin Chinese and Riau Indonesian.

5.4

How Grammatical Relations Develop

Lexico-grammatical structure becomes ‘grammaticalized’ or ‘lexicalized’ (becomes what we think of as grammar or words) through repeated use of particular patterns to constrain the hearer’s interpretation of the speaker’s intention in a particular way until it becomes conventionalized on the societal level and habitualized on the individual level (LaPolla 2015). Ontogenetically we start with no structure, including no grammatical relations, and in each society different types of constructions will emerge out of the interactions of the speakers, and so each language is

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

unique in terms of what sorts of structures have conventionalized in the language. There are some languages, such as Riau Indonesian (Gil 1994) and Mandarin Chinese (LaPolla 1993), which have not conventionalized constraints on referent identification of the type associated with grammatical relations (though they may have conventionalized other types of constraints on interpretation). What this means is that there are no conventionalized associations which relate position in word order, the marking on the nouns or verb, and so on with particular semantic roles, and so the structures of the language do not force particular interpretations of the role of referents mentioned in discourse. The addressee can still infer a context of interpretation in which the utterance will make sense, but this inferential process is relatively unconstrained compared to a language that has constructions that force a particular interpretation within this functional domain, and so the addressee will have to rely more on the assumptions of real-world semantics (what makes sense given common knowledge about the world) to make sense of the utterance. This does not mean that the inference won’t be influenced by conversational implicatures. There may be common conversational implicatures that can influence the interpretation. For example, as there is a rather strong frequency correlation between topic and actor in Chinese (and many other languages), there is a conversational implicature that an animate topic (the referent referred to by the utterance-initial referring expression) is the actor of the clause. It is simply a conversational implicature because it can be cancelled by the semantics of the referents or the requirements of the context of interpretation, such as in (12): (12)

Xuéshe¯ng yˇıj¯ıng f¯ a-le chéngj¯ı student(s) already distribute-pfv grade ‘The students were already given their grades.’

This expression could potentially be understood as either ‘The students were already given their grades’ or ‘The students already gave out grades (to someone else)’, but it was understood in the context in which it was uttered as ‘The students were already given their grades’ because students normally receive grades, not give them out, and it made more sense in the context. What happens in one type of conventionalization of grammatical relations is that a conversational implicature of ‘actor as topic’ appears so often in discourse that it becomes a conventional implicature, and then becomes so strongly conventionalized that speakers cannot accept any other interpretation.10 Note that all conventionalization (grammaticalization) is of constructions, not individual words, and not globally in the language (Bybee 2003; Himmelmann 2004; Gisborne and Patten 2011), so the RRG assumption that grammatical relations are construction-specific is much more in line with what we know about grammaticalization and with the facts of languages, and so is more empirically sound than a view that posits abstract global grammatical relations in all languages.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

281

282

RANDY J. LAPOLLA

The difference between conversational implicature, conventional implicature, and obligatory marking forcing a particular interpretation is the degree to which speakers are free to use or not use the particular form to constrain the hearer’s inferential process, and also the degree to which the form forces a particular interpretation. Old English did not constrain the identification of the role of a referent with word order, though it did constrain the interpretation of referent role using a complex system of case. Even so, the frequency with which reference to actors preceded the verb in topic position led to a conversational implicature that gradually strengthened as the case-marking system weakened, until we ended up with the current system of Modern English, where word order alone constrains the interpretation of the role of the main referents, and what was originally the primary means of constraining the interpretation of the role of the referent (the case marking) is now non-existent or, in the case of pronouns, is now secondary, often assigned by word order. Conventionalized constraints on the interpretation of coreference across clauses also develop in a similar way. Initially there is no syntactic constraint on cross-clause coreference, and so the interpretation of which referring expressions (including zero) co-refer is completely dependent on inference from real-world semantics. For instance, in the following example from Rawang, a Tibeto-Burman language of northern Myanmar, any of the three coreference patterns given in the three translations would be possible, and which would be correct would depend on the addressee’s inference of which is most likely to be the interpretation intended by the speaker given the addressee’s assumptions about hitting and crying and what is known about the people involved.11 ph u a nɯ̀ ŋɯ̄ a:ʔm`ı ¯ ŋí dɯ́ s ̀ ŋ dip bɯ́ ` ph u dɯ́ -s ̀ ŋ dip bɯ́ -` a nɯ̀ ŋɯ̄ -ap-`ı ¯ ŋ-í Apung-agt Adeu-loc hit pfv-tr.pst ps cry-tmdys-intr.pst (i) ‘Apung hit Adeu and (Apung) cried’ or (ii) ‘Apung hit Adeu and (Adeu) cried’, or (iii) ‘Apung hit Adeu and (someone else) cried’ e

e

e e e

e

e e

(13)

Utterances of this type are somewhat rare, though; more often only one possible actor is mentioned, as in an utterance like John finished eating and left, and so the conversational implicature that the actor is the same in both clauses (and it is only an implicature at first) can become strengthened to the point that it becomes conventionalized as the only possible interpretation, as in English, where a clause such as John put the rock next to the chameleon and turned brown has to mean that John turned brown, even if it makes no sense, unlike in a language where this coreference pattern has not conventionalized (e.g. Chinese) and so it would more likely be interpreted as meaning the chameleon turned brown. Agreement or cross-referencing on the verb develops as an unstressed pronoun is reinforced by a stressed pronoun or full noun phrase often

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

enough for the unstressed pronoun to become cliticized to the verb. We clearly see this process in Angami Naga, a Tibeto-Burman language of Northeast India (Giridhar 1980: 32, 59): the verbal prefixes (1sg ¯ a-, 2sg n̂ -, 3sg puô-) are transparently derived from the free pronouns (1sg ¯ a, 2sg n¯o, 3sg puô), and can be used together with the free pronouns, as in (14a), or with a noun phrase, as in (14b) (see LaPolla 1992a, 1994 for other examples from TibetoBurman languages). a. n¯ o n̂ -d¯ ov¯ı 2sg 2sg-clever ‘You are clever.’

ˇ ˇ

(14)

b. nhıcunyô puô-d¯ ov¯ı boy 3sg-clever ‘(The) boy is clever.’

Relational marking on noun phrases often arises as marking of location, such as a locational noun, is used to constrain the inference of the relationship of some referent to the state of affairs being predicated to a locational sense, such as source, and then gets extended to the marking of other sorts of participants (e.g. agents) through predictable pathways (see the following section for more discussion, and also LaPolla 2004). This can further conventionalize into fully paradigmatic case marking.

5.5

Why Syntactic Relations Develop

We have seen that syntactic relations develop from a form that is repeated over and over again in discourse to the point that it becomes conventionalized as an obligatory part of the language, and thereby forces a particular interpretation where otherwise there would be two or more possible interpretations. But why would speakers repeat a form so often that this would happen? The answer lies in the culture of the speakers of the language, their way of thinking, their value system. For a form to be used often enough for it to become conventionalized, it must constrain the interpretation process of the addressee in a way that is important to the speaker, so important that the speaker is willing to put extra effort into constraining the addressee’s inferential process in that particular way to make it more likely the addressee will ‘get it right’. That is, the speaker wants to make sure the addressee will infer that part of the communicative intention correctly, more so, possibly, than other parts of the intention, and often uses a particular form that they have used successfully before (and/or other people have used successfully before) to constrain the interpretation in the same way, over and over again. (We are creatures of habit and imitation, and although we sometimes innovate, we more often go along with our usual habits and also will imitate others.) In the case of syntactic relations, what must be important to the speakers is that the addressee correctly infer the roles of the major participants. The clearest example of this is the development of relation morphology on the

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

283

284

RANDY J. LAPOLLA

noun phrase of the type agentive, patient, and/or anti-agentive.12 Marking of participant role is, at least initially, marking of semantic role. In many of the languages I’ve looked at (the Tibeto-Burman languages; LaPolla 1995, 2004), there is a clear development of agentive marking through the extension of ablative or instrumental marking to constrain the inference of which participant is the agent. This begins only in contexts where there could be confusion, such as when there are two human referents mentioned in an utterance, and it is optional at that stage. The first speaker to do this would have had the desire to constrain the interpretation of the semantic roles, and in order to do so used a form already in the language (e.g. ablative marking; it is easier to use material already in the language than to create totally new material). Over time, this marking can become obligatory and can also be extended to other sorts of agent-like referents. The motivation for patient or anti-agentive marking is the same, but in the case of these markers, the speakers chose to constrain the interpretation of the role of a non-agent rather than an agent. In some of the older systems this type of marking has developed beyond simple semantic marking, as speakers have used material already in the language (the semantic marking) to constrain the interpretation in new ways. In some cases the pattern that gets conventionalized might not specifically involve extra effort on the part of the speaker but simply reflects the discourse habits of the speakers (which again will reflect the culture of the speakers). For example, in a culture where actors are very often made the topic of conversation, and topics are mentioned in clause-initial position (also a choice that influences the construction of the context of interpretation – see LaPolla 2019), we might see this result over time in the conventionalization of a word order constraint such as that in English. We can see this tendency developing in some Tibeto-Burman languages, such as Qiang (LaPolla with Huang 2003), but it has not yet fully conventionalized. For example, in a Qiang transitive clause construction with two unmarked noun phrases referring to human referents, usually the first one will be understood as referring to the Agent, but pragmatics still controls word order more than semantics, and so if some other referent is more topical than the Agent, the noun phrase referring to the Agent will not appear first. In this kind of situation the Qiang Agentive marking is often used to constrain the addressee’s interpretation of the relative roles. Agent-first is then the default and unmarked situation, and could develop into an obligatory interpretation with more reinforcement through repeated occurrence. The motivation for the development of constraints on a particular functional domain may not originally be part of the native culture but can come though language contact: when people are bilingual in another language that obligatorily constrains the interpretation of some functional domain, such as the marking of source of information (i.e. has obligatory evidential marking), and they use that language often enough for the habit of constraining the evidential sense to become established,

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

they may eventually feel the necessity to constrain the interpretation of source of information when using their own language. They can then borrow the forms of the language that already has it, or use native material for that purpose, and it may then develop into an obligatory category in their own language. This is still repeated action based on the desire to constrain the interpretation in a particular way leading to conventionalization, but in this case the motivation came into the culture of the speakers through the influence (habits) of another culture (see LaPolla 2009). Relevant to grammatical relations, the development of person marking on the verb in some Tibeto-Burman languages seems to be related to language contact (see LaPolla 2001). Although all conventionalization has its origin in repeated actions that have a cultural motivation, it is not always possible to find a direct link between some motivation and the linguistic form post facto, especially if the conventionalization happened in the far-distant past (though see the papers in Enfield 2002 and De Busser and LaPolla 2015) and if there has been considerable phonetic reduction of the forms used in the constructions. We continue to use forms that are no longer transparently motivated just because they are there, and are part of our habits of language use. We can see this in the layering of marking, for example the fossilization and maintenance of the -r plural in children, even though it is not seen as a plural marker by most modern English speakers.13 The motivations for many words used in English today are opaque to modern English speakers, such as why we say dial to make a phone call, but they use the forms anyway. In some cases, sound changes can make what was once transparently motivated opaque. For example, the modern word for ‘crow’ in Mandarin Chinese is w¯ u, which is not transparent, but when we look at the way it would have been pronounced when it was first used (reconstructed as *ʔa), we can see that at that time it was motivated as onomatopoeia. In terms of syntax, we may have less evidence for the motivation of a particular word-order pattern, but in some cases we can see the effect it has and possibly assume that that effect was the motivation. For example, the English pattern of marking mood with word order, that is, putting a different element in initial (Theme) position in different moods,14 may have developed because of a desire to clearly constrain the interpretation of mood.

References Bloomfield, Leonard. 1914. An Introduction to the Study of Language. New York: Henry Holt. (Photostatic reprint 1983 by John Benjamins, Amsterdam, with Forward by Konrad Koerner and Introduction by Joseph Kess.) Bybee, Joan. 2003. Mechanisms of change in grammaticalization: The role of frequency. In Brian D. Joseph and Richard D. Janda (eds.), The Handbook of Historical Linguistics, 602–623. Oxford: Blackwell.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

285

286

RANDY J. LAPOLLA

Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Oxford: Blackwell. Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William. 2013. Radical Construction Grammar. In G. Trousdale and T. Hoffmann (eds.), The Oxford Handbook of Construction Grammar, 211–232. Oxford: Oxford University Press. De Busser, Rik and Randy J. LaPolla (eds.). 2015. Language Structure and Environment: Social, Cultural, and Natural Factors. Amsterdam: John Benjamins. Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. Dixon, R. M. W. 1980. The Languages of Australia. Cambridge: Cambridge University Press. Dixon, R. M. W. 2000. A-constructions and O-constructions in Jarawara. International Journal of American Linguistics 66(1): 22–56. Dixon, R. M. W. 2004. The Jarawara Language of Southern Amazonia. Oxford: Oxford University Press. Dryer, Matthew. 1997. Are grammatical relations universal? In Joan Bybee, John Haiman and Sandra A. Thompson (eds.), Essays on Language Function and Language Type, 115–143. Amsterdam: John Benjamins. Durie, Mark. 1985. A Grammar of Acehnese. Dordrecht: Foris. Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11: 365–399. Enfield, Nicholas. 2002. Ethnosyntax: Explorations in Grammar and Culture. Oxford: Oxford University Press. Foley, William A. and Robert D. Van Valin, Jr. 1977. On the viability of the notion of ‘subject’ in universal grammar. Proceedings of the Berkeley Linguistics Society 3: 293–320. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Gabelentz, H. Georg C. von der. 1869. Ideen zu einer vergleichenden Syntax. Wort- und Satzstellung. Zeitschrift für Völkerpsychologie und Sprachwissenschaft 6: 376–384. Gary, Judith Olmstedt and Edward L. Keenan. 1977. On collapsing grammatical relations in universal grammar. In P. Cole and J. M. Sadock (eds.), Grammatical Relations (Syntax and Semantics 8), 83–120. New York: Academic Press. Gil, David. 1994. The structure of Riau Indonesian. Nordic Journal of Linguistics 17: 179–200. Giridhar, P. P. 1980. Angami Grammar (CIIL Grammar Series 6). Mysore: Central Institute of Indian Languages. Gisborne, Nikolas and Amanda Patten. 2011. Construction Grammar and grammaticalization. In Heiko Narrog and Bernd Heine (eds.), The Oxford Handbook of Grammaticalization, 92–104. Oxford: Oxford University Press.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

Halliday, M. A. K. 1994. An Introduction to Functional Grammar (2nd ed.). London: Arnold. Heath, Jeffrey. 1975. Functional relationships in grammar. Language 51(1): 89–104. Heath, Jeffrey. 1977. Choctaw cases. Proceedings of the Berkeley Linguistics Society 3: 204–213. Himmelmann, Nikolaus P. 2004. Lexicalization and grammaticization: Opposite or orthogonal? In Walter Bisang, Nikolaus P. Himmelmann and Björn Wiemer (eds.), What Makes Grammaticalization? A Look from Its Components and Its Fringes, 21–42. Berlin: Mouton de Gruyter. Hopper, Paul. 1991. On some principles of grammaticization. In Elizabeth C. Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, Vol. 1, 17–36. Amsterdam: John Benjamins. Jespersen, Otto. 1909–1949. A Modern English Grammar on Historical Principles, Vols. I–VII. London: Allen & Unwin. Keenan, Edward L. 1976. Towards a universal definition of ‘subject’. In Charles N. Li (ed.), Subject and Topic, 305–333. London: Academic Press. LaPolla, Randy J. 1992a. On the dating and nature of verb agreement in Tibeto-Burman. Bulletin of the School of Oriental and African Studies 55(2): 298–315. LaPolla, Randy J. 1992b. Anti-ergative marking in Tibeto-Burman. Linguistics of the Tibeto-Burman Area 15(1): 1–9. LaPolla, Randy J. 1993. Arguments against ‘subject’ and ‘direct object’ as viable concepts in Chinese. Bulletin of the Institute of History and Philology 63 (4): 759–813. LaPolla, Randy J. 1994. Parallel grammaticalizations in Tibeto-Burman: Evidence of Sapir’s ‘drift’. Linguistics of the Tibeto-Burman Area 17(1): 61–80. LaPolla, Randy J. 1995. On the utility of the concepts of markedness and prototypes in understanding the development of morphological systems. Bulletin of the Institute of History and Philology 66(4): 1149–1185. LaPolla, Randy J. 2001. The role of migration and language contact in the development of the Sino-Tibetan language family. In R. M. W. Dixon and A. Y. Aikhenvald (eds.), Areal Diffusion and Genetic Inheritance: Case Studies in Language Change, 225–254. Oxford: Oxford University Press. LaPolla, Randy J. 2004. On nominal relational morphology in Tibeto-Burman. In Ying-jin Lin, Fang-min Hsu, Chun-chih Lee, Jackson T.-S. Sun, Hsiu-fang Yang and Dah-an Ho (eds.), Studies on Sino-Tibetan languages: Papers in Honor of Professor Hwang-cherng Gong on his seventieth birthday, 43–74. Taipei: Institute of Linguistics, Academia Sinica. LaPolla, Randy J. 2006. On grammatical relations as constraints on referent identification. In Tasaku Tsunoda and Taro Kageyama (eds.), Voice and Grammatical Relations: Festschrift for Masayoshi Shibatani (Typological Studies in Language), 139–151. Amsterdam: John Benjamins. LaPolla, Randy J. 2009. Causes and effects of substratum, superstratum and adstratum influence, with reference to Tibeto-Burman languages. In

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

287

288

RANDY J. LAPOLLA

Yasuhiko Nagano (ed.), Issues in Tibeto-Burman Historical Linguistics (Senri Ethnological Studies 75), 227–237. Osaka: National Museum of Ethnology. LaPolla, Randy J. 2010. Hierarchical person marking in Rawang. In Dai Zhaoming (ed.), Forty Years of Sino-Tibetan Language Studies: Proceedings of ICSTLL-40, 107–113. Harbin, China: Heilongjiang University Press. LaPolla, Randy J. 2015. On the logical necessity of a cultural connection for all aspects of linguistic structure. In Rik De Busser and Randy J. LaPolla (eds.), Language Structure and Environment: Social, Cultural, and Natural Factors, 33–44. Amsterdam: John Benjamins. LaPolla, Randy J. 2019. Arguments for seeing Theme-Rheme and TopicComment as separate functional structures. In J. R. Martin, Y. Doran and G. Figueredo (eds.), Systemic Functional Language Description: Making Meaning Matter. London: Routledge. LaPolla, Randy J., with Chenglong Huang. 2003. A Grammar of Qiang, with Annotated Texts and Glossary (Mouton Grammar Library 39). Berlin: Mouton de Gruyter. Marantz, Alec P. 1982. Grammatical relations and explanation in linguistics. In Annie Zaenen (ed.), Subjects and Other Subjects: Proceedings of the Harvard Conference on the Representation of Grammatical Relations, 1–24. Bloomington, IN: IULC. Marantz, Alec P. 1984. On the Nature of Grammatical Relations (Linguistic Inquiry Monograph 10). Cambridge, MA: MIT Press. Platt, J. F. 1971. Grammatical Form and Grammatical Meaning: A Tagmemic View of Fillmore’s Deep Structure Concepts. Amsterdam: North-Holland. Olson, Mike. 1978. Switch-reference in Barai. Proceedings of the Berkeley Linguistics Society 4: 140–157. Olson, Mike. 1981. Barai Clause Junctures: Toward a Functional Theory of Interclausal Relations. PhD dissertation, Australian National University. Schachter, Paul, revised by Lawrence A. Reid. 2008. Tagalog. In Bernard Comrie (ed.), The World’s Major Languages (2nd ed.), 833–855. London: Routledge. Schachter, Paul and Fei T. Otanes. 1972. A Tagalog Reference Grammar. Berkeley, CA: University of California Press. Seuren, P. A. M. 1998. Western Linguistics: An Historical Introduction. Oxford: Blackwell. Van Valin, Robert D., Jr. 1977. Ergativity and the universality of subjects. Proceedings of the Chicago Linguistic Society 13: 689–706. Van Valin, Robert D., Jr. 1981. Grammatical relations in ergative languages. Studies in Language 5(3): 361–394. Van Valin, Robert D., Jr. 1987. Aspects of the interaction of syntax and pragmatics: Discourse coreference mechanisms and the typology of grammatical systems. In Jef Verschueren and Marcella Bertuccelli-Papi (eds.), The Pragmatic Perspective: Selected Papers from the 1985 International Pragmatics Conference (Pragmatics and Beyond Companion Series 5), 513–531. Amsterdam: John Benjamins.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and controllers. In Lilián Guerrero Valenzuela, Sergio Ibáñez and Valeria A. Belloro, (eds.), Studies in Role and Reference Grammar, 45–68. Mexico: UNAM. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Notes 1 The name ‘Role and Reference Grammar’ derives from the early focus on the semantic roles and pragmatic functions in discourse referent tracking of grammatical relations. 2 Following best practice in typology, for language-specific (descriptive) categories and constructions I will capitalize the initial letters of the name of the category or construction, but for comparative concepts I will not capitalize the first letter. So, for example ‘Actor’ refers to the language-specific grammatical category manifesting a particular neutralization of semantic roles in the language under discussion, while ‘actor’ refers to the comparative concept of the one who performs an action. As there are no universal or cross-linguistic grammatical categories, descriptive and comparative concepts need to be kept distinct. 3 Note that the identification of the referent of it as the same as that of the newspaper is not due to grammatical relations, but simply to inference; there is nothing in the grammar that obligatorily constrains the interpretation, the way the inference of the relationship between Jim and the thrower of the newspaper is constrained by the grammar. 4 English S is itself a neutralization of semantic macroroles for grammatical purposes that is not found in all languages (see below on Acehnese and Tagalog). Actor and Undergoer are also language-specific restricted neutralizations of semantic roles for grammatical purposes, hence are called macroroles, but are at a lower level than the restricted neutralizations of macroroles we are talking about here. In English and many other languages there is also variable access to Undergoer status when there is both a theme and a recipient or location in the clause, e.g. in the construction I gave the book to Mary, the book is the Undergoer, but in the construction I gave Mary the book, Mary is the Undergoer, and in Load the truck with hay, the truck is the Undergoer, while in the construction Load the hay on the truck, the hay is the Undergoer. The choice of one construction or the other usually depends on the relative topicality of the referents. See Van Valin and LaPolla 1997: 144ff.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

289

290

RANDY J. LAPOLLA

5 For example, in a Tagalog single argument clause, semantically different arguments can be marked differently, e.g. given the root dulas, madulas can be used for ‘slip (unintentionally)’ and dumulas can be used for ‘slide (i.e. slip intentionally)’, and the marking of an intransitive actor can be different from that of a transitive actor, even with the same root, e.g. labas ‘outside’: lumabas ‘come/go out’ vs. maglabas ‘bring/take out’. 6 In the typology literature the [S,A] restricted neutralization is often referred to as ‘nominative-accusative alignment’, as in Latin the A and S take the nominative case in opposition to the U, which takes the accusative case, and the [S,U] restricted neutralization is often referred to as ‘ergative-absolutive alignment’, as in Dyirbal the U and S take the absolutive case in opposition to the A, which takes the ergative case. In some of the literature, languages are talked about as having one of these alignments, but a more empirical approach is to look at the individual constructions, as not all constructions in the language necessarily manifest the relevant alignment, and a single language can have different constructions manifesting different alignments (see Van Valin 1981 on Jakaltek and Van Valin and LaPolla 1997: 282ff. on Tzutujil). 7 An example of an indirectly affected referent as Topic would be Huwag mong ubusan ng gasolina si Ricky [negimp 2sgnt¼lnk finish-lt poss gasoline spec pn] ‘Don’t use up all the gasoline on Ricky’ (using the Location Topic construction – cf. the use of the locative expression ‘on X’ in the English translation for the one affected). 8 Abbreviations used in the Tagalog examples: at Actor-Topic infix or prefix; cs Change of State marker; irr Irrealis; lnk clitic Linker; loc Locative marker; lt Locative-Topic suffix; poss Possessive linker; pt Patient-Topic suffix or infix or prefix; redup Reduplication of initial syllable for marking imperfective and planned actions; spec Specific referent; stat Stative predicate. 9 In some languages there is another type of person marking on the predicate that is not based on semantic role or PSA, as the marking reflects the speaker, and possibly hearer and third-person argument based on a person hierarchy such as 1 < 2 < 3. The predicate may also have marking for when the actor is lower on the hierarchy than the undergoer. This type of system is called hierarchical marking. See LaPolla (2010) for one example. 10 Even as a conversational implicature the default interpretation can be very strong. For example, the implicature that actions occur in the order that they are talked about is quite strong in English, and so the average speaker would say that they got married and had a baby means something different from they had a baby and got married, but the implicature can be cancelled, e.g. by adding but not in that order after either of the two possible orders.

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

Grammatical Relations

11 Abbreviations used in the Rawang example: agt agentive marker; intr. past intransitive past-tense marker; loc locative marker; pfv perfective aspect marker; ps predicate sequence marker (marks non-final clause); tmdys time marker (marks a past action as having occurred within the past few days); tr.past transitive past-tense marker. In the Rawang example in (13) the tones are high á, mid ¯ a, low ` a. All syllables that end in a stop consonant are in the high tone. Open syllables without a tone mark are unstressed. A colon marks non-basic long vowels. In the Angami examples in (14), the tones are mid-level ¯ a and low falling â. 12 Anti-agentive marking differs from patient marking in that it is not marking what role a particular referent has, but what role it does not have: it marks the mention of a human referent (at least human patients and datives, but sometimes possessors as well) as not being agents. See LaPolla (1992b, 2004). 13 George W. Bush famously said ‘Childrens do learn when standards are high and results are measured’ (Reuters, 26 Sep 2007), showing that even the -en plural does not constrain the sense of plurality enough for him. This is an example of layering (Hopper 1991: 22), and we can see that layering occurs when someone feels the existing marking is not constraining the inference of that particular semantic domain enough. 14 In unmarked cases, in declarative mood the subject is in initial position; in polarity interrogatives, the polarity-marked auxiliary occurs in initial position; in question-word questions, the question word appears in initial position; in imperatives the verb appears in initial position. So the addressee’s inference of the mood of the clause (which generally represents how the speaker is interacting with the addressee) is greatly constrained by the first constituent (the Theme), and this influences the interpretation of the rest of the utterance. Theme (as in Theme-Rheme) is important precisely because the inferences drawn after hearing the Theme influence the rest of the interpretation. (See Halliday 1994, Ch. 3; LaPolla 2019.)

https://doi.org/10.1017/9781316418086.008 Published online by Cambridge University Press

291

6 Argument Structure Alternations James K. Watters

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ANTIC ASP CON CONN DIR ELAT

6.1

anticausative aspect marker (general) continuative connective direction, secondary theme elative

MOOD NONFUT PREP PRON PV SP

mood non-future preposition pronominal preverb (Abaza) subject prefix (Bantu)

Introduction

In the most general sense, ‘argument structure’ refers to ‘the specification of and relation between a word’s semantic and syntactic arguments’ (Jackendoff 2002: 134). Specifying the arguments that are semantically determined by a predicate and determining the relationship between those arguments and their syntactic realization has been a key field of research in all current linguistic theories. We can imagine a spectrum of possibilities, from the possibility that all the details of a word’s syntax follow from a correct account of its semantics to the other extreme claiming that the relation between the semantic and syntactic configurations are simply arbitrary. Role and Reference Grammar (RRG) takes a middle course in accounting for the relationship between a verb’s meaning and its syntax: the syntactic argument structure is projected from the verb’s lexical information, along with other predicates in the clause, and is also modified by discourse pragmatics.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

The purpose of this chapter is to give an introductory account of how verbal argument structures and their alternations are handled in RRG.* The literature across linguistic theories on these issues is vast and there is no space to review other models or make detailed comparisons or to discuss argument structures of non-verbal predicates. RRG proposes a logical structure (LS) associated with each verb that is based on a system of lexical decomposition. Unlike the lexical decomposition found in Generative Semantics, or, more recently in Distributed Morphology, the decomposition is largely defined by Aktionsart or lexical aspect (Vendler 1967[1957]). Following work by Dowty (1979), RRG applies semantic tests to determine a verb’s syntactically relevant lexical decomposition. The alternations in the syntactic arrangements of a verb’s arguments complicate attempts to generalize about the verb’s argument structure. In the Chomskyan tradition, such alternations are often accounted for by a derivational process involving movement. RRG, however, is a monostratal theory in which each sentence has a single morphosyntactic representation linked by a set of rules to its semantic representation. It is important to note that the mapping between the semantic and syntactic representation does not involve movement or syntactic derivation of any kind. Argument structure alternations involve different syntactic alignments among arguments projected by the LS and may also involve adjuncts (the elements of the peripheries of the layered structure of the clause). The order of presentation in this chapter is based in part on the distinction between lexical and syntactic aspects of argument structure. After some introductory comments in Section 6.2, Section 6.3 focuses on lexical processes that affect the linking of arguments to macroroles. Section 6.4 looks at syntactic processes that affect the linking of arguments or adjuncts to the privileged syntactic argument (PSA). Section 6.5 discusses two processes, the passive and antipassive, that reflect marked mappings at both levels: they can have a lexical effect on the linking to macroroles and also a syntactic effect, determining which argument functions as the PSA. The examples in this chapter come from a selection of languages from different linguistic families, including some that have been discussed before in the RRG literature. However, most of my fieldwork has been in the Tepehua branch of the Totonac-Tepehua language family (Mexico). Because of this, apart from English examples, illustrations of some of the more complex issues will often be drawn from examples in Tepehua.1 A key concept in RRG is the Completeness Constraint, which places an important restriction on argument structure alternations.

* This chapter has benefited significantly from comments by a peer reviewer and suggestions by Delia Bentley. Of course, any remaining mistakes and obscurities are my own fault.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

293

294

JAMES K. WATTERS

(1)

Completeness Constraint: All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence. (Van Valin 2005: 129–130)

This constraint allows for a bidirectional mapping between the syntactic structure of the clause and the lexical structure. Two points of clarification: (a) Van Valin points out that ‘explicitly specified’ means that a variable or a constant fills the argument position in the logical structure; if it is filled by ‘Ø’, it is unspecified (2005: 130); and (b) note that only referring expressions in the syntax need to be linked to an argument in the semantic representation. These points will be relevant in some of the discussion in this chapter.

6.2

Argument Structure Alternations

Consider the verb give in English or its equivalent in some other language. It evokes a scene that includes a giver, something given, and a person or object that receives what is given. As part of the meaning of the verb, these arguments are included in the lexical entry for give but the arguments may be realized in different ways syntactically. For example, argument structure alternations discussed in the literature on English syntax include examples such as the following. (2)

a. Randy gave the book to Kim. b. Randy gave Kim the book.

This particular kind of alternation, commonly known as the dative alternation, is a lexical alternation, affecting the mapping from the LS to the undergoer. As will be seen in 6.3.2, not all languages have such an alternation. Other alternations in the argument structure of give in English are possible. (3)

a. b. c. d.

The book was given to Kim (by Randy). Kim was given the book (by Randy). Randy gave the book. Randy was always giving.

In each of these examples, the sentence entails that someone gave, something was given, and someone was the recipient or intended recipient (see Williams 2015: 199–202). The argument structure is shaped by the passive constructions in (3a) and (3b). The examples in (3c) and (3d) show that each may have one unspecified argument but, unlike many languages, without any morphology to signal the change. The syntactic arrangement of the verb and its arguments is significantly different in each case.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

The linking of arguments has been a key element of RRG since its inception (Foley and Van Valin 1984). As pointed out by Van Valin and LaPolla (1997: 384, 389–392), the linking between semantic arguments and their position in the syntax has two major phases: 1. mapping the arguments in logical structures onto macroroles, and 2. mapping the macroroles and other arguments onto the syntax. The core arguments are those that are determined by the semantic decomposition of the verb. Adjuncts are non-argument PPs and adverbs, which occur in a periphery. RRG recognizes a third class of argument, the ‘argument-adjuncts’, which will be presented in 6.4. The syntactically relevant semantics of a sentence is represented in the LS (see Chapter 3 of this volume). That semantic representation determines the kind of syntactic template (e.g. tree structure) that the sentence maps onto, following a default principle. Syntactic template selection principle: The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core. (Van Valin 2007: 130)

There are typically language-specific restrictions, as well, such as the English constraint that all cores have a syntactic valence of 1 (i.e. English requires dummy subjects for predicates without any semantic argument, such as rain).

6.3

Lexical Alternations

The standard mapping of a verb’s arguments onto the actor and undergoer macroroles follow the Actor-Undergoer Hierarchy (AUH; Chapter 4). There are, however, processes that are common cross-linguistically that alter the standard argument structure or alignment. These include the two types of constructions discussed in this section: noun incorporation and the ditransitives or dative constructions.

6.3.1 Noun Incorporation A common feature of head-marking languages is noun incorporation. In some languages, such as West Greenlandic, noun incorporation is so productive that it is reported ‘there are infinitely many possible forms involving noun incorporation’ (Sadock 1991: 84). We can start with an example from Nahuatl (Uto-Aztecan) given by Sapir (1911: 260) in one of the first discussions of noun incorporation.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

295

296

JAMES K. WATTERS

(4)

a. ni-c-qua in nacatl 1sbj-3obj-eat the flesh ‘I eat the flesh.’ b. ni-nica-qua 1sbj-flesh-eat ‘I flesh-eat.’

In (4a) the verb is clearly transitive, requiring the direct object pronominal form. The verb in (4b), however, is intransitive and so lacks the direct object marking. Now consider the following examples from Tongan (Polynesian) (5a–b) and Yucatec Maya (6a–b) (from Mithun 1984: 851, 857). (5)

(6)

a. Na'e inu 'a pst drink abs ‘John drank the kava.’ b. Na'e inu kava pst drink kava ‘John kava-drank.’

e conn 'a abs

kavá kava

-'é erg

Sione. John

Sione John

a. t-in-ˇ c’ak-Ø-ah cˇe'. comp-I-chop-it-prf tree ‘I chopped a tree.’ b. cˇ'ak-ˇ ce'-n-ah-en. chop-tree-antip-prf-I(abs) ‘I wood-chopped.’

In both of these alternations, the verb is intransitive when the noun is incorporated, as can be seen by the absolutive case for the actor in both (5b) and (6b). (The intransitivity of the Yucatec Maya form in (6b) is also marked by the antipassive suffix, which is typically present in the derivation of an intransitive verb from a transitive.) The key difference regarding semantic arguments in (4), (5) and (6) is that the unincorporated patient is referential and maps onto the undergoer position in a transitive construction. The incorporated nouns, however, are non-referential and therefore do not correspond to any specific argument in the LS, resulting in intransitive sentences. This kind of noun incorporation, therefore, has a detransitivizing effect. Another type of noun incorporation – perhaps the most common crosslinguistically (McGregor 1997) – is the incorporation of nouns referring to body parts. These constructions typically do not change a transitive verb into an intransitive. Rather than deleting the undergoer macrorole, they display an alternation in the linking to undergoer: in the nonincorporated forms, the body part is the undergoer; in the incorporated forms, the undergoer is the person, the possessor of the body part. In some languages the only nouns that can be incorporated into the verb are body parts.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

Body part incorporation results in part from the frequent recurrence and natural cohesion of many activities affecting parts of the body, e.g. ‘to hand-wash’ or ‘to tooth-brush’. In addition, noun incorporation of body parts allows affected persons to assume a primary case role, such as subject or direct object, rather than merely oblique possessor. (Mithun 1984: 858) The person affected can assume ‘a primary case role’ in these constructions because the transitivity of the verb is unchanged, with the possessor of the body part linking to the undergoer macrorole, as the ʻaffected personʼ.2 Mithun (1984: 857) provides the following example from Tupinamba (Tupí). (7)

a. s-oβá a-yos-éy his-face I-it-wash ‘I washed his face.’ b. a-s-oβá-éy I-him-face-wash ‘I face-washed him.’

Unlike the previous examples of noun incorporation, the verb continues to be transitive – both forms in (7) have an actor and undergoer. Both sentences have the same LS but differ in regard to undergoer selection. (8)

a. wash′ (a-, [have.as.part′ (s-, oβá)]) ¼ (7a) b. wash′ (a-, [have.as.part′ (s-, oβá)]) ¼ (7b)

In both examples, the first person singular prefix a- maps onto the actor macrorole. In (8a) ‘his face’, as patient, is the unmarked choice for undergoer. The incorporated form in (8b) requires the marked choice for undergoer and the possessor maps onto the undergoer macrorole. One other kind of noun incorporation needs to be briefly discussed here. At the beginning of this section, it was mentioned that Greenlandic Eskimo is known to have extremely productive noun incorporation (Sadock 1991). A language with similar constructions is Southern Tiwa (Kiowa-Tanoan) and both languages are discussed in Rosen (1989). Consider the following examples from Allen et al. (1984: 297). (9)

a. Wisi bi-musa-tuwi-ban two 1sg:B-cat-buy-pst 3 ‘I bought two cats.’ b. Yedi ibi-musa-tuwi-ban those cats-B:B-buy-pst ‘They bought those cats.’

These examples, like many in Allen et al. (1984) and Sadock (1991), are notably different from the noun-incorporation examples discussed previously, in at least two ways. First, the incorporated nouns are modified by a quantifier (9a) and a deictic (9b), and, second, they are clearly referential.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

297

298

JAMES K. WATTERS

Van Valin and LaPolla (1997: 66–68) present an analysis of similar constructions in Greenlandic Eskimo, showing how RRG accommodates such structures with distinct operator and constituent projections in the layered structure of the noun phrase. The significant conclusion is that, unlike the earlier examples discussed in this section, these do not involve the incorporation of a bare noun, but rather full RPs, as shown by the fact that they are referential and by the presence of modifiers.

6.3.2 Ditransitives For our purposes here, I will assume the definition of ditransitives offered by Malchukov et al. (2010: 1): A ditransitive construction is defined here as a construction consisting of a (ditransitive) verb, an agent argument (A), a recipient-like argument (R), and a theme argument (T). As they point out, this excludes some verbs of transfer which do not involve a recipient (such as put) as well as benefactives (discussed in Section 6.4); but it includes verbs with a ‘recipient’ in both a literal and an extended sense. These are verbs that in English manifest the well-known dative alternation. In RRG, ditransitives and other verbs with three arguments require a complex representation with more than one predicate. The common LS for a ditransitive involves two states of affairs joined by CAUSE. (10)

[do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]

Following the AUH, the first argument of do′, x, is the actor, while the theme, the second argument of have′, z, is the undergoer. Over the last fifty years, there have been many studies of the English dative alternation that try to account for the semantic similarities and differences between pairs of sentences such as those in (2) and (11). (11)

a. Chris sent the money to Sam. b. Chris sent Sam the money.

It should be noted that dative alternation is absent from many languages. Consider the following examples from Alacatlazala Mixtec, an Otomanguean language of Mexico (from Zylstra 1991: 13) and from French. (12)

tásh¯ı i t¯ ut¯ u nd¯ aha s¯ıhí ñá (con)give I paper hand mother her ‘I’m giving the paper to her mother.’

(13)

s¯ık¯ o ñá noní noo i (compl)sell she corn face my ‘She’s selling corn to me.’

(14)

Jean a donné le livre ` a Marie John has given the book to Mary ‘John gave the book to Mary.’

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

(15)

*Jean a donné Marie le livre. John has given Mary the book ‘John gave Mary the book.’

Unlike English, Mixtec and French do not have dative alternation. In Mixtec languages the indirect object requires a body-part relational noun, functioning as a preposition (p.c. Carol Zylstra). If the indirect object follows the verb the result is ungrammatical. Likewise in French, a construction similar to the English ‘dative movement’ is ungrammatical. In both languages, the direct object behaves like the direct object in a simple transitive clause and the indirect object must be preceded by a preposition (or a relational noun, in the case of Mixtec). These exemplify what Dryer (1986: 815) calls a ‘direct object’ language: the undergoer of the simple transitive continues to manifest the same morphosyntactic properties in the ditransitive construction (the ‘direct object’ is still the ‘direct object’). Such ‘direct object’ languages require the unmarked undergoer choice predicted by the AUH in ditransitives, choosing theme or patient as undergoer rather than the recipient. However, some languages require a marked undergoer choice in ditransitives: the recipient, not the theme, is consistently linked to the undergoer macrorole. Dryer (1986: 815) calls these ‘primary object languages’. Peterson (2007: 144) provides the example of Hakha Lai, a Tibeto-Burman language. (16)

a. vok na-Ø-hmuʔ pig 2sg.sbj-3sg.obj-see ‘You saw the pig.’ b. na-ka-hmuʔ 2sg.sbj-1sg.obj-see ‘You saw me.’ c. vok na-ka-peek pig 2sg.sbj-1sg.obj-give ‘you gave me the pig.’

The verb agrees with the actor and the undergoer in the simple transitive clauses (16a) and (16b) and with the recipient in the ditransitive clause (16c). This agreement pattern in the ditransitive is obligatory – the alternation found in English dative movement is absent. (For further discussion of Tibeto-Burman examples, see Van Valin and LaPolla 1997: 272–273.) In ‘primary object languages’, unlike the Mixtec and French examples, the recipient is obligatorily linked to the undergoer macrorole. In Van Valin and LaPolla (1997: 387) this marked choice for undergoer is attributed to animacy – the animate indirect object is chosen over the inanimate direct object for undergoer position (see, however, Guerrero and Van Valin 2004 for further discussion). Some languages, such as Mixtec and Tepehua, have no syntactically ditransitive verb roots. Although Tepehua has no verb roots with three direct core arguments, the LS for predicates such as ma:laqatʃa: ‘send’ (17a)

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

299

300

JAMES K. WATTERS

and stʼa: ‘sell’ (18), like their counterparts in other languages, require three semantic arguments. Consider the examples in (17) and (18). (17)

a. ma:laqatʃa:-ɬ send-pfv ‘s/he sent it’ b. ma:laqatʃa:-ni-ɬ send-dat-pfv ‘s/he sent it to him/her’

(18)

a. ʃtaq-ɬi give-pfv ‘s/he gave it.’ b. ʃtaq-ni-ɬ give-dat-pfv ‘s/he gave it to him/her’

Totonac-Tepehua languages have only one (or two) adpositions, with applicatives fulfilling a preposition-like function (Watters 2019). Rather than employing a preposition to refer to the recipient, Totonac-Tepehua languages use the suffix -ni, signalling a marked undergoer choice. The applicative -ni licenses a third argument. In (17) and (18), both the transitive and the ditransitive forms have the basic LS given in (10). The only difference is that the transitive constructions in (17a) and (18a) do not specify the y-argument (the recipient) in the LS. (19)

a. [do′ (x, Ø)] CAUSE [BECOME have′ (Ø, z)] (¼ (17a), (18a)) b. [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)] (¼ (17b), (18b))

Following the Completeness Constraint (1) the unspecified recipient in (19a) is absent from the syntax; in (19b) the recipient is specified in the LS, so must also be syntactically present. The recipient argument cannot appear in a clause unless the applicative suffix, -ni, occurs on the verb. In the case of the verbs in (17a) and (18a), the theme is linked to the undergoer position and what would correspond to the indirect object is unexpressed. In both cases the simple transitive forms of such verbs are typically used in contexts where the recipient must be invoked from the context, as a definite null complement (Fillmore 1986). Semantically, the recipient is required to complete the predication. This is one way in which an argument – in this case the unexpressed recipient – contrasts with adjuncts. Adjuncts modify the predication rather than complete it, and, when syntactically absent, the interpretation of the utterance does not require they be invoked by the listener. The key evidence that the argument of the applicative -ni (the recipient) is linked to the undergoer macrorole is the effect of the antipassive suffix, -nVn (20b) (see Section 6.5.2), which deletes the undergoer.4 With a simple transitive verb in Tepehua, the direct object is syntactically obligatory. Consider the example sentences in (20). If no object noun phrase appears in the

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

clause, it is still transitive, requiring a translation with a definite third singular pronoun (20a). (20)

a. kin-kuku st’a:-y 1poss-uncle sell-ipfv ‘My uncle sells it.’ b. kin-kuku st’a:-nan 1poss-uncle sell-antip ‘My uncle sells.’ c. kin-kuku ki-st’a:-ni-y laqtʃ’iti 1pos-uncle 1obj-sell-dat-ipfv clothes ‘My uncle sells me clothes.’ d. kin-kuku st’a:-ni-nin laqtʃ’iti 1poss-uncle sell-dat-antip clothes ‘My uncle sells clothes’

The translation of (20b) shows that the antipassive suffix -nVn in Tlachichilco Tepehua results in deletion of the patient undergoer. In (20c), the presence of the applicative suffix -ni, marks the recipient as undergoer. This is evidenced by the fact that it is the recipient that is deleted in (20d), when the antipassive suffix is present. In sum, RRG posits only two macroroles even though predicates may have more than two arguments in their logical structure. Ditransitive constructions display different approaches to the syntactic position of the recipient argument. In some languages, the theme is the undergoer, following the default choice on the AUH. In other languages, the recipient is regularly the marked choice for undergoer, reflecting its prominence as the typically more animate non-actor argument. Finally, other languages allow variable linking to the undergoer macrorole.

6.3.3 Other Three-Place Predicates and Lexical Applicatives While cross-linguistic studies such as Malchukov et al. (2010) limit the notion of ditransitive to three-place predicates in which the third argument is some kind of recipient, there are, of course, other kinds of three-place predicates. Common examples in English include the following. (21)

a. Chris put the jar on the table. b. Kim sprayed insect repellent under the bed.

The constructions in (21) involve a predicate with an indirect core argument of location. The predicates put and spray both have the LS [do′ (x, Ø)] CAUSE [BECOME be-loc′ (y, z)] and ‘in the actual semantic representation of a sentence be-loc′ would be replaced by the LS of a preposition’ that corresponds to the specific example (e.g. on, under, in, etc.) (Van Valin and LaPolla 1997: 160). There are at least two ways that the English dative constructions discussed in 6.3.2 are different from three-place predicates like these. First, these

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

301

302

JAMES K. WATTERS

three-place predicates do not display similar argument alternations that affect undergoer assignment. Second, they involve predicative prepositions rather than the non-predicative prepositions found in ditransitives. There are other classes of three-place verbs in English and other European languages that do not involve a dative argument but do display alternating argument structure. One well-known set of such verbs is the spray/load class (Fillmore 1968: 48; Levin 1993: 50–51). (22)

a. He sprayed the paint on the wall. b. He sprayed the wall with the paint.

It has often been noted that this alternation reflects a semantic distinction in that the NP or RP immediately following the verb shows a higher level of affectedness: as Fillmore notes (1968: 48, fn.49), the sentence in (22b) implies the entire wall got painted but (22a) does not. Furthermore, in the corresponding passive sentences, the semantic distinction is maintained, showing that the difference in affectedness is due to undergoer status, not to the syntactic position of direct object. (23)

a. The paint was sprayed on the wall. b. The wall was sprayed with the paint.

This alternation is a modulation of argument structure and, like the dative alternation, reflects different linkings to the undergoer macrorole. The semantic macrorole of undergoer, not the syntactic position of direct object, ‘represents the non-instigating, affected participant in a state of affairs’ (Van Valin 2005: 61–62). Thus, considering the two non-actor arguments in each of the sentences in (22) and (23), the undergoer is the more affected one (see Van Valin 2005: 113–114).

6.3.4 Causative Lexical causative constructions built on intransitive verbs are generally straightforward when it comes to argument structure: the causee is linked to undergoer in both the intransitive and transitive constructions; in the transitive construction, the causer links to actor. (24)

a. The door opened. b. [BECOME open′ (door)]

(25)

a. Chris opened the door. b. [do′ (Chris, Ø)] CAUSE [BECOME open′ (door)]

However, in languages that allow a causative built on a transitive verb base, the result is a predicate with three core arguments. The causer will be linked to the actor macrorole, but, as in the case of ditransitives, the question arises about which of the remaining two arguments will be linked to the undergoer position.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

West Coast Bajau, one of the Sama-Bajaw languages of the Western Malayo-Polynesian branch of Austronesian, as reported by Miller (2007: 303), has a productive morphological causative prefix pe-. (26)

a. Togor bana tiang pagar e. upright very post fence dem ‘The fence post stands very straight.’ b. Boi pe-togor Mali tiang comp caus-upright Mali post ‘Mali erected the fence post.’

pagar fence

e. dem

Miller (2007: 302–308) includes various examples of causatives with pe- on statives, inchoatives, manner, and activity verbs. He reports that few transitive verb roots take the pe- causative, but when they do, ‘the causee is the new undergoer’ (2007: 306). This is also the case in Tepehua, but in a more indirect fashion. The causee appears in the clause due to the presence of the same applicative -ni, presented in 6.3.2, which links an indirect object to the undergoer (the vowel of -ni is lengthened as part of the causative formation rule). (27)

transitive

causative of transitive

ʔah-ya

‘s/he digs it’

maːʔah-ni:-y

‘s/he makes her/him/it dig it’

ʃʔoq-ya

‘s/he unties it’

ma:ʃʔoq-niː-y

‘s/he makes him/her/it untie it’

In clear contrast to the pattern in the West Coast Bajau and Tepehua examples is causative formation on a transitive verb in which the undergoer of the base verb retains the syntactic marking and behaviour of undergoer. The causee then appears as a peripheral or indirect core argument. An example of this can be seen in the Turkish sentences in (28) (from Underhill 1976: 346): (28)

a. Yusuf Yusuf ‘Yusuf b. Yusuf Yusuf ‘Yusuf

di¸s-in-i çek-ti tooth-3poss-acc pull-pst pulled his tooth.’ doktor-a di¸s-in-i doctor-dat tooth-3poss-acc had the doctor pull his tooth.’

c ̣ek-tir-di pull-caus-pst

In both the basic transitive as well as the causative construction, the direct object, dis, ‘tooth’ is marked as accusative and, in the causative, the causee ˙ must be marked as dative. Like other lexical constructions in this section, morphological causatives affect the level of mapping from predicate argument structure onto the actor and undergoer macroroles. For causatives formed on a transitive verb base, two non-actor core arguments are present in the LS, and languages differ in regard to which of the two links to the undergoer macrorole.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

303

304

JAMES K. WATTERS

6.3.5 Anticausative The anticausative construction has been described as the inverse of the causative: instead of adding a causer and causing event to the LS, the anticausative is a derived intransitive in which the undergoer is the PSA. Haspelmath and Müller-Bardey (2001) call the anticausative ‘the most radical agent-removing category’ and include examples such as the following from Hungarian and Turkish. (29)

a. András-t három tárgy-ból elvág-t-ák András-acc three subject-elat fail-pst-3pl ‘They failed András in three subjects.’ b. András három tárgy-ból elvág-ód-ott. András three subject-elat fail-antic-pst(3sg) ‘András failed in three subjects.’

(30)

a. Anne-m kapı-yı aç-tı. mother-1sg door-acc open-pst(3sg) ‘My mother opened the door.’ b. Kapı aç-ıl-dı. door open-antic-pst(3sg) ‘The door opened.’

In each of these examples, the verb root is causative with actor and undergoer and the intransitive is derived by a morphological operation. The RRG literature discusses two kinds of anticausatives. The alternations in (29) and (30) display an anticausative that is similar to the ‘middle’ construction in traditional grammar. In these constructions, ‘the function of the morphological markers is to cancel part of the logical structure’ (Van Valin 2005: 46) – the anticausative removes the causer and causing event from an otherwise causative verb. Typically, the anticausative alternation applies to a minor subclass of verbs. This is the case in Tepehua languages, which have a small group of transitive verb roots that have derived intransitive forms marked by the inchoative prefix, ta-, as in the following examples. (31)

Base verb (causative)

Gloss

Derived form (anticausative)

Gloss

laːqaːɬi-y

‘x breaks it down’

ta-laːqaːɬi-y

‘it breaks down’

tʃeʔe-y

‘x shatters it’

ta-tʃeʔe-y

‘it shatters’

ʔeʃ-a

‘x tears it’

ta-ʔeʃ-a

‘it tears’

teʔe-y

‘x cracks it’

ta-teʔe-y

‘it cracks’

The obvious similarity between the passive and this anticausative construction is that both remove the actor as a core argument. However, there is a key difference between them: the passive has an implicit or understood actor and causing event but this anticausative construction does not. In RRG this means that, unlike the passive, in which the actor is implicit but not a

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

core argument, in these anticausatives the actor and causing event are missing altogether. (32)

causative achievement/accomplishment ! achievement: [do′ (x, Ø)] CAUSE [BECOME/INGR pred′ (y)] ! BECOME/INGR pred′ (y)

This kind of anticausative converts a causative achievement or causative accomplishment with two macroroles into an intransitive achievement or accomplishment with only the undergoer macrorole. A second kind of anticausative maintains the causing activity in the LS but the agent of the activity is unspecified. This kind of anticausative is found in Romance languages and involves a reflexive construction. The analysis is presented in Van Valin and LaPolla (1997: 408–414), based on Centineo (1996), who gives evidence for the presence of the activity predicate in the LS: the occurrence of manner adverbs – which only occur with activities – and an implicit agent (see also Bentley 2006: 126–136). Consider these Spanish examples from González Vergara (2009: 366–374). (33)

a. Pedro ensució la camisa. Pedro stained.3sg the shirt ‘Pedro stained the shirt.’ b. La camisa se ensució. the shirt refl stained.3sg ‘The shirt got dirty.’

González Vergara shows that (33b), like the Italian examples in Centineo (1996), has an implicit causing agent and can occur with manner adverbs. Thus, unlike the anticausatives accounted for by (32), these maintain the causing activity in the LS. (34)

Causative achievement/accomplishment ) achievement: [do′ (x, Ø)] CAUSE [BECOME/INGR dirty′ (y)] ) [do′ (Ø, Ø)] CAUSE [BECOME/INGR dirty′ (y)]

Both of the rules in (32) and (34) change an event with two specified arguments into an achievement or accomplishment with one undergoer but they have significantly different logical structures.

6.3.6

Null Complements and the Activity–Active Accomplishment Alternation Van Valin (2012: 69) discusses the following examples of the relation between argument structure and telicity. (35)

a. b.

Sandy wrote (poetry) for an hour/*in an hour. Atelic Sandy wrote the poem in an hour. Telic

(36)

a. b.

Chris drank (beer) for an hour/*in an hour. Atelic Chris drank the beer in an hour. Telic

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

305

306

JAMES K. WATTERS

To account for this common alternation between atelic and telic uses of some verbs, ‘not just verbs but in fact whole verb phrases must be taken into account to distinguish activities from accomplishments’ (Dowty 1979: 60–62). The RRG analysis of the examples in (35) and (36) requires distinct logical structures for the atelic and telic readings of the verbs. The objects in (35a) and (36a) are non-referring NPs but those in (35b) and (36b) are RPs. As a result, there is a difference in macrorole status, since activities, including multiple-argument activity verbs (the atelic examples), never have an undergoer macrorole. A common explanation for the data in (35) and (36), then, is that the alternation is due to the change in the referential status of the object, that is, whether it is an RP or NP. Van Valin (2012: 69–71) shows that the activity–active accomplishment alternation in other languages often is not due to the inferred status of the object, but is marked on the verb. Changing the aspect marking on the verb can result in a change from activity to active accomplishment (‘He was eating everything for two hours,’ vs. ‘He ate everything in two hours.’) The following lexical rules account for the alternations, showing the change in the LS (see also Van Valin and LaPolla 1997: 180 and, for a revised LS of active accomplishments which has no consequences for the current discussion, Van Valin 2018 and Bentley 2019). (37)

a. Activity [motion] ) Active Accomplishment: do′ (x, [pred′ (x)]) ) do′ (x, [pred′ (x)]) & INGR be-LOC′ (y, x) b. Activity [consumption] ) Active Accomplishment: do′ (x, [pred′ (x, y)]) ) do′ (x, [pred′ (x, y)]) & INGR consumed′ (y) c. Activity [creation] ) Active Accomplishment: do′ (x, [pred′ (x, y)]) ) do′ (x, [pred′ (x, y)]) & INGR exist (y)

These lexical rules reflect the productivity of an alternation between two LSs involving the same base verb. To show that this alternation cannot be attributed simply to the presence of a definite direct object, Van Valin (2012: 70) presents examples from Georgian (Holisky 1981), in which the preverb da- imposes a telic reading of the event. (38)

a. K’ac-i (c’eril-s) c’er-s xuti man-nom (letter-dat) write.prs-3sg five ‘The man is writing (letters) for five hours.’ b. K’ac-i c’eril-s da-c’er-s at man-nom letter-dat pv-write.prs-3sg ten ‘The man will write the letter in ten minutes’

saati. hours c’ut-ši. minutes-in

The kind of alternation accounted for in the rule in (37) is not marked morphologically in English but is in languages such as Georgian and it is marked by aspectual markers in other languages such as Russian. The distinct logical structures of activities and their corresponding active accomplishments have both semantic and syntactic consequences.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

6.4

Syntactic Alternations

This section offers a brief survey of benefactives and constructions which involve an adjunct. These constructions involve both adjunct prepositions in the periphery and argument-adjunct prepositions which ‘introduce an argument into the clause and share it with the logical structure of the core, rather than taking the logical structure of the core as an argument’ (Van Valin and LaPolla 1997: 159). These categories of prepositions also characterize adpositions and many applicatives cross-linguistically.

6.4.1 Benefactive Though often treated as a subtype of the dative construction, a benefactive typically has a different relation to the event referred to by the verb. In the case of ditransitives, that is, constructions in which the third argument in some sense ‘receives’ the second one, the verb itself introduces an LS that includes the three arguments. This is not the case with a benefactive construction. A benefactive construction involves an argument external to the event. Unlike the semantic representation of an adjunct, the semantic representation of a benefactive shares an argument with the LS of the verb. In RRG, the benefactive involves purpose, and can be represented by the semantic representation given for ‘purposive for’ in English in Van Valin and LaPolla (1997: 383) (based on Jolly 1991). (39)

Semantic representation of purposive for: want′ (x, LS2 ^ DO (x,[LS1 . . .CAUSE. . .LS2])

Like adjunct adpositional phrases, the benefactive has the argument it licenses as its first argument and embeds the LS of the core as its second argument. However, the benefactive, unlike an adjunct, shares an argument with the core. (41) is a simplified analysis of the benefactive example (40). (40)

ki-makaː-ni-ɬ ʔaqa-tawn tʃaqaʔ 1sg.obj-make-dat-pfv clf-one house ‘He made me a house.’

(41)

[want′ (x, LS2 ^ DO (x,[make′ (x, house)]) CAUSE [have′ (I, house)]

This benefactive has some similarities to a ditransitive. However, unlike ditransitives, constructions that include CAUSE and a recipient, benefactives more generally describe a situation of affectedness. Thus, many languages use the benefactive construction to express malefactives, as well as with intransitive activities: (42)

ʔik-maqniː-ni-ka-ɬ ki-ʃʔoy 1sbj-kill-dat-pas-pfv 1poss-dog ‘(Someone) killed my dog on me.’

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

307

308

JAMES K. WATTERS

(43)

ʔik-miɬpaː-ni-ya-n 1sbj-sing-dat-fut-2obj ‘I will sing for you.’

The benefactive marks the presence of an argument-adjunct that is not part of the LS associated with the bare verb but, unlike a simple adjunct, it shares an argument with that LS. Therefore, though it is an example of argument structure alternation at the syntactic level, it is a lexical process. Indeed, the Tepehua applicative -ni that marks the presence of the recipient in ditransitives and the experiencer in benefactives is never used to mark the presence of an adjunct in a clause.

6.4.2 Syntactic Applicatives Applicative constructions in head-marking languages correspond in large part to prepositional phrases in dependent-marking languages. This has led some to analyse applicatives as ‘preposition incorporation’, following Baker (1988: 229–304). As already shown, the Tepehua applicative dative suffix -ni results in the indirect object being linked to the undergoer position. This is also true of the applicative constructions in the following examples from Chichewa (Bantu), as reported in Baker (1988: 247–248). ˇ

(44)

a. kalulu a-na-gul-ir-a mbidzi nsapato hare sp-pst-buy-for-asp zebras shoes ‘The hare bought shoes for the zebras.’ b. mbidzi zi-na-gul-ir-idw-a nsapato (ndi kalu!u) zebras sp-pst-buy-for-pas-asp shoes by hare ‘The zebras were bought shoes by the hare.’

Baker points out that in this construction, the applicative -ir has resulted in the benefactive argument usurping properties associated with the direct object: immediate postverbal position, object pro-drop, and (in this example) subject of the passive (compare 44a and 44b). In RRG terms, this is evidence that in the Chichewa applicative construction, the benefactive is the undergoer. As we saw in 6.4.1, this is not surprising, as the benefactive is typically an example of an argument-adjunct rather than a simple adjunct. However, some applicatives allow adjuncts to appear as syntactic arguments of the verb. As reported in Peterson (2007: 18, 19, 22), Hakha Lai, a Tibeto-Burman language, has several optional applicative constructions that allow an adjunct to appear as a verbal argument. ˇ

(45)

a. ka-law ʔan-ka-thloʔ-pii 1sg.poss-field 3pl.sbj-1sg.obj-weed-com ‘They weeded my field (together) with me.’ b. tiilooŋ khaa tivaa kan-Ø-tan-naak boat top river 1pl.sbj-3sg.obj-cross-ins ‘We used the boat to cross the river.’

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

The Hakha Lai examples in (45) have paraphrases in (46): (46)

a. kay-maʔ¼hee ka-law ʔan-thlaw 1sg-pron¼com 1sg.poss-field 3pl.sbj-weed ‘They weeded my field together with me.’ b. tiilooŋ¼ʔin tivaa (khaa) kan-tan boat¼ins river top 1pl.sbj-cross ‘We used the boat to cross the river.’

Tepehua has three applicative prefixes that allow adjuncts to appear as arguments of the verb: comitative tʼaː, instrumental puː, and direction ɬi. (47)

a. kin-ta-tʼaː-ʔa-ɬ 1obj-3pl.sbj-com-go-pfv ‘They went with me.’ huːki b. puː-mi-ɬ ins-come-pfv horse ‘S/he came by horse.’ c. waː yuːtʃa ɬiː-stʼaː-ɬ foc 3pron dir-sell-pfv ‘S/he sold it for that (price) / sold it for that (reason).’

The instrumental in (47b) has a paraphrase with a Tepehua preposition but paraphrases of the applicative constructions in (47a) and (47c) require prepositions borrowed from Spanish. (The paraphrase of the comitative requires kun from Spanish con and the price reading of ɬiː- requires por.) The two key questions in an RRG account of applicative constructions built on a transitive verb base involve the lexical level (48a) and the syntactic level (48b). (48)

a. Does the argument of the base verb or the argument of the applicative link to the undergoer macrorole? b. Is there a restriction regarding which of the non-actor arguments may occur as PSA in a passive construction?

The question in (48b) is touched on in Section 6.5. Data regarding (48a) is provided by Peterson (2007) regarding applicatives in Bakusu (Bantu) and Hakha Lai (Tibeto-Burman). In Bakusu, the instrumental applicative does not manifest standard object properties and a first hypothesis would be that the instrument does not link to undergoer position though it is a syntactic argument of the derived verb. In Hakha Lai, in six of the seven applicative constructions the applied object apparently links to undergoer position, taking on standard object properties. The one exception, again, is the instrumental, a feature that Peterson suggests is due to the fact that instrumentals are typically inanimate, and, therefore, less salient in reported events. In Tlachichilco Tepehua, only the argument of the dative applicative, -ni, is regularly linked to the undergoer macrorole. The other three

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

309

310

JAMES K. WATTERS

applicatives, ɬiː-, puː-, tʼaː-, most often simply allow adjuncts to appear as syntactic arguments of the verb – an important feature since only direct arguments of the verb can be questioned or relativized in Tepehua. However, the linking of the PSA in a passive construction is not limited to the undergoer: any direct syntactic argument of the verb can be the PSA.

6.5

Voice

The RRG account of voice alternations – passive and antipassive – distinguishes two areas of argument linking. RRG’s universal formulation of voice oppositions refers to both the syntactic and lexical dimensions of voice modulation. (49)

General characterization of basic voice constructions in RRG5 a. PSA modulation voice: permits an argument other than the default argument . . . to function as the privileged syntactic argument. b. Argument modulation voice: gives non-canonical realization to a macrorole argument. (Van Valin 2005: 116)

In RRG, the general cross-linguistic characterization of passive and antipassive voice involves both levels of mapping: the lexical assignment of macrorole status and the syntactic determination of PSA. Van Valin and LaPolla (1997) provide the English passive and the Dyirbal (a Pama-Nyugan language spoken in north-eastern Australia) antipassive as prototypical constructions. (50)

a. English passive construction b. Dyirbal -ŋay antipassive construction

Though these are given as prototypical examples, it should be noted that there are languages with passive or antipassive constructions in which modulation occurs on only one of the two levels mentioned in (49). As a result of argument modulation, in a passive construction, the actor macrorole is removed from the core – it is either deleted or moved to the periphery. This typically results in some other argument occurring as the PSA. In the case of the antipassive, the undergoer is similarly removed, typically making an otherwise transitive verb into an intransitive.

6.5.1 Passive According to Keenan and Dryer (2007: 328–329), the following passives are ‘basic passives’ ((51) is their example, (52) is Tlachichilco Tepehua). (51)

John was slapped.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

(52)

Pa:laqsti-saː-ka-ɬ ni Juan cheek-hit-pass-pfv art John ‘John was slapped.’

This is considered to be a ‘basic’ passive because passives like this ‘are the most widespread across the world’s languages’. They have the following characteristics: (i) there is no agent phrase (in Tepehua, as in many other languages, the passive does not allow the actor to appear in the clause at all); (ii) the verb is a transitive verb that is passivized; (iii) in its nonpassivized form, the verb expresses an action with an agent subject and patient object. By these criteria, then, English passives with an explicit actor as well as impersonal passives (passives of intransitives) are not basic passives. The contrast between the two Tepehua forms in (53) illustrates the use of the passive with intransitive verbs, as do the examples from German and Turkish in (54a) and (54b), respectively. (53)

a. Tapaːtsaː-kan maɬkuyuː abril y mayu work-pass(ipfv) month April and May ‘It is worked (people work) the months of April and May.’ b. ʔantʃa ʔalin-kan there exist-pass(ipfv) ‘It is existed there./Something is there.’

(54)

a. Es wird hier getanzt it is here danced. ‘Dancing takes place here.’ b. E˘ glen-il-di. have.fun-pass-pst ‘Fun was had.’

The existence of impersonal passives shows the importance of distinguishing the two aspects of passive in (49). The forms in (53) and (54) involve the argument modulation of (49b), in that the actor is absent from the clause. However, the PSA modulation is irrelevant in these constructions, as there is no PSA. Even very closely related languages can have passive constructions that differ in either the PSA modulation or the argument modulation. In Tlachichilco Tepehua, first- and second-person undergoers (but not third plural) are marked as PSA in the passive construction (55a). In Pisaflores Tepehua, while second-person undergoers must be marked as PSA, first-person undergoers more commonly retain the undergoer marking (55b). (55)

a. k-laqtsʼin-kan-a:-w 1sbj-see-pass-ipfv-1pl ‘We are seen.’ b. kin-ta-laqts’in-kan-a:-n 1obj-3plsbj-see-pass-ipfv-2obj ‘We are seen.’

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

311

312

JAMES K. WATTERS

In the Tlachichilco Tepehua example, (55a), the actor is removed (argument modulation) and the undergoer is the PSA, marked by the ‘subject’ form (PSA modulation). In the Pisaflores Tepehua example (55b), the actor is removed (argument modulation) but the undergoer is still marked as ‘object’, not as PSA (there is no PSA modulation). Besides the variation regarding the presence or absence of PSA modulation, there is notable cross-linguistic variation regarding what arguments can be linked to the PSA. Yaqui (Uto-Aztecan) is an example that only allows a macrorole to appear as PSA, so in the following passive forms, only the undergoer can function as PSA (from Guerrero and Van Valin 2004: 299–300). (56)

a. U chu’u-W ki’i-wa-k. the dog-nom bite-pass-pst.pfv ‘The dog was bitten.’ b. Jamut-ta-u nooka-wa-k. woman-acc-dir talk-pass-pst.pfv ‘Someone talked to the woman’ / *‘The woman was talked to.’

In the passive construction in (56a) the undergoer assumes PSA status; in the passive in (56b), however, the woman is marked by the directional postposition, -u, and is not linked to the undergoer position. As a result, the PSA modulation of the passive does not apply, only the argument modulation, removing the actor. However, some languages allow a wide variety of arguments to link to the PSA position. Van Valin (2005: 121) gives examples from Kinyarwanda (Bantu), in which applicatives can mark several semantic roles. Tepehua is another example. The Totonac-Tepehua languages have pragmatically determined word order and no case marking on the NPs. It is not possible to determine, for a transitive clause in isolation with two third-person singular participants, which is the PSA. The passive and antipassive play a major role in tracking referents. In (57), the instrumental and comitative appear as syntactic arguments of the verb due to the applicatives though they do not have macrorole status. However, in the passive they can map onto the PSA. (57)

a. waː yuːtʃa puː-tʃʼan-nan-kan foc that ins-sow-antip-pass(ipfv) ‘That’s what it’s planted with (instrumental).’ b. ʔik-tʼaː-tʃiwin-ka-ɬ 1sbj-com-speak-pass-pfv ‘I was spoken with (comitative).’

In Tepehua, the discourse-pragmatics directly influence the selection of the PSA, and the freedom for different arguments to function as PSA serves a key pragmatic function: maintaining topics and tracking referents (see Watters 2017).

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

6.5.2 Antipassive The term ‘antipassive’ refers to a change in syntactic alignment of arguments that has traditionally been used to describe a detransitivizing form in ergative languages, such as in the following examples from Dyirbal (from Dixon 1994: 161, 170). (58)

a. yabu banaga-nyu mother.abs return-nonfut ‘Mother returned.’ b. ŋuma yabu-ŋgu bura-n father.abs mother-erg see-nonfut ‘Mother saw father.’ c. yabui [bural-ŋa-ŋu ŋuma-gu] banaga-nyu mother.abs see-antip-rel.abs father-dat return-nonfut ‘Mother, who saw father, was returning.’

The examples in (58a) and (58b) exemplify the standard ergative-absolutive marking for actor in an intransitive clause and in a transitive clause. The absolutive is the PSA in Dyirbal and only the PSA can be relativized. As a result, unlike nominative-accusative languages, for the actor to be relativized, it must be absolutive. The sentence in (58c) shows how this can be done. The antipassive suffix on the verb in the relative clause results in an intransitive construction, the undergoer is no longer a core argument and the actor, as with any intransitive, is in the absolutive, allowing the formation of the relative clause. Much of the discussion in the RRG literature regarding antipassives involves syntactically ergative languages such as Dyirbal. In those cases, the antipassive construction allows the actor to function as PSA. Some linguists use the term antipassive only for ergative languages. However, I am following the perspective articulated by Polinsky (2013): Some researchers insist on the link between the antipassive and ergativity . . ., while others propose that the antipassive is not limited to ergative languages . . . The transitive/antipassive alternation is simply more visible in an ergative language, where it typically involves a change in subject case marking from ergative to absolutive. Considering the two aspects of voice in RRG (49), the same can be said for antipassives as for passives: in some languages the relevant construction only affects the argument modulation and does not directly affect the PSA modulation. In fact, many ergative languages are only ergative in their morphology and do not have the syntactic ergativity found in a language like Dyirbal. In a non-ergative language like Tepehua, the antipassive only involves the level of argument modulation (49b). It occurs on a transitive verb base, marking the absence of the undergoer, whether it is the undergoer of the verb root or the marked undergoer, that is, the argument of the dative

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

313

314

JAMES K. WATTERS

suffix -ni. But it never marks the absence of the applicative argument associated with one of the applicative prefixes (comitative, instrumental, directional). (59)

a. puː-stʼaː-na-ɬ ki-muːral ins-sell-antip-pfv 1poss-bag ‘S/he sold using my bag.’ b. José tʼaː-stʼaː-na-ɬ Kwan José com-sell-antip-pfv Kwan ‘José sold with Kwan.’

These examples involve the antipassive construction, which marks the absence of the undergoer. However, the arguments of the instrumental and comitative applicatives are present. This supports the analysis in 6.4.2 that Tepehua applicatives other than the dative -ni, do not link to undergoer.

6.6

Conclusion

Alternations in argument structure at the lexical and syntactic levels have important functions. Some lexical processes such as causatives and anticausatives are often more limited in productivity and serve to enrich the lexicon. Others, such as noun incorporation and ditransitive alternations, and some applicatives, clearly have semantic effects regarding which argument is the most affected or patient-like (i.e. the undergoer). The activity–active accomplishment alternation involves a change in transitivity and is tied to a difference in the telic or non-telic nature of the event. Syntactic processes discussed in this chapter add syntactic arguments to the verb or affect the linking to the PSA. Voice alternations, such as passive and antipassive, involve mappings at both the lexical and syntactic levels. Apart from simply removing an argument from the core because that argument is not salient at the moment, these constructions often play an important role in tracking referents or maintaining the discourse topic. RRG provides a framework that has the heuristic value of requiring one to determine the logical structure associated with a verb and the linking of arguments in the logical structure to the macroroles and from the macroroles to PSA and syntactic positions in the clause. The alternations discussed in this chapter are captured by capitalizing on the distinction between these levels of linguistic analysis.

References Allen, Barbara J., Donna B. Gardiner and Donald G. Frantz. 1984. Noun incorporation in Southern Tiwa. Indonesian Journal of Applied Linguistics 50: 292–311.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Bentley, Delia. 2006. Split Intransitivity in Italian. Berlin: Mouton de Gruyter. Bentley, Delia. 2019. The Logical Structure of Verbs of Quantized and Non-Quantized Change. Paper presented at the International RRG Conference, 19–21 August 2019, University at Buffalo (SUNY). Centineo, Giulia. 1996. The distribution of si in Italian intransitive/inchoative pairs. In Mandy Simmons and Theresa Galloway (eds.), Proceedings of SALT V, 54–71. Ithaca, NY: Cornell University Press. Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: D. Reidel. Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative. Language 62: 808–845. Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert T. Harms (eds.), Universals in Linguistic Theory, 1–91. New York: Holt, Rinehart and Winston. Fillmore, Charles J. 1986. Pragmatically controlled zero anaphora. In Vassiliki Nikiforidou, Mary VanClay, Mary Niepokuj and Deborah Feder (eds.), Twelfth Annual Meeting of the Berkeley Linguistics Society, 95–107. Berkeley, CA: Berkeley Linguistics Society. Foley, William A. and Robert D. Van Valin. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. González Vergara, Carlos. 2009. One rule to rule them all: Logical structures for Spanish non-reflexive se sentences. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 361–380. Mexico: Universidad Nacional Autónoma de México. Guerrero, Lilián and Robert D. Van Valin. 2004. Yaqui and the analysis of primary object languages. International Journal of American Linguistics 70(3): 290–319. Haspelmath, Martin and Thomas Müller-Bardey. 2001. Valency change. In G. Booij, C. Lehmann and J. Mugdan (eds.), Morphology. An International Handbook on Inflection and Word Formation, 207. Berlin: Walter de Gruyter. www.eva.mpg.de/fileadmin/content_files/staff/haspelmt/pdf/2005val.pdf. Holisky, Dee A. 1981. Aspect theory and Georgian aspect. In B. Comrie and M. Polinsky (eds.), Causatives and Transitivity, 87–120. Amsterdam: John Benjamins. Jackendoff, Ray S. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Jolly, Julia. 1991. Prepositional Analysis within the Framework of Role and Reference Grammar. New York: Peter Lang. Keenan, Edward L. and Matthew S. Dryer. 2007. Passive in the world’s languages. In Timothy Shopen (ed.), Language Typology and Syntactic Description, Vol. 1: Clause Structure, 325–361. Cambridge: Cambridge University Press.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

315

316

JAMES K. WATTERS

Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago: University of Chicago Press. Malchukov, Andre, Martin Haspelmath and Bernard Comrie. 2010. Studies in ditransitive constructions: A typological overview. In Andre Malchukov, Martin Haspelmath and Bernard Comrie (eds.), Studies in Ditransitive Constructions. A Comparative Handbook, 1–64. Berlin: De Gruyter, Mouton. McGregor, William. 1997. Grammatical structures in noun incorporation. In Anne-Marie Simon-Vandenbergen, Kristin Davidse and Dirk Noël (eds.), Reconnecting Language: Morphology and Syntax in Functional Perspectives. Current Issues in Linguistic Theory, 141–180. Amsterdam: John Benjamins. Miller, Mark. 2007. A Grammar of West Coast Bajau. PhD dissertation, (Arlington), University of Texas. Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60: 847–894. Perlmutter, David M. and Paul M. Postal. 1977. Toward a universal characterization of passive. In K. Whistler, R. D. Van Valin et al. (eds.), Proceedings of the Annual Meeting of the Berkeley Linguistics Society, 3, 394–417. Berkeley, CA: Berkeley Linguistics Society. Peterson, David A. 2007. Applicative Constructions. Oxford: Oxford University Press. Polinsky, Maria. 2013. Antipassive constructions. In Matthew S. Dryer and Martin Haspelmath (eds.), The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. http://wals .info/chapter/108. Rosen, Sarah Thomas. 1989. Two types of noun incorporation: A lexical analysis. Language 65: 294–317. Sadock, Jerrold M. 1991. Autolexical Syntax: A Theory of Parallel Grammatical Representations. Chicago: University of Chicago Press. Sapir, Edward. 1911. The problem of noun incorporation in American languages. Language 13(2): 250–282. Underhill, Robert. 1976. Turkish Grammar. Cambridge: MIT Press. Van Valin, Robert D. Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D. Jr. 2007. The Role and Reference Grammar analysis of three-place predicates. Suvremena Lingvistika 33.1(63): 31–64. https://rrg.caset .buffalo.edu/rrg/vanvalin_papers/RRG-Analysis_Three-Place_Pred.pdf. Van Valin, Robert D. Jr. 2012. Lexical representation, co-composition, and linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon, Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in Generative Lexicon Theory: Text, Speech and Language Technology, Vol. 46, 67–107. Dordrecht: Springer. Van Valin, Robert D. Jr. 2018. Some issues regarding (active) accomplishments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 71–93. Freiburg: Freiburg Institute for Advanced Studies, Albert-Ludwigs Universität Freiburg.

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

Argument Structure Alternations

Van Valin, Robert D. Jr and Randy LaPolla. 1997. Syntax: Structure, Meaning, and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1967[1957]. Linguistics in Philosophy. Ithaca, NY, Cornell University Press. (Previously published in Vendler, Zeno. 1957. Verbs and times. Philosophical Review 56: 143–160.) Watters, James K. 2017. Tlachichilco Tepehua: Semantics and function of verb valency change. In Álvaro González and Ía Navarro (eds.), Verb Valency Changes: Theoretical and Typological Perspectives, Typological Studies in Language 120, 165–192. Amsterdam: John Benjamins. Watters, James K. 2019. La preposición en tepehua y construcciones semejantes. In Lilián Guerrero (ed.), Adposiciones y elementos de su tipo en lenguas de América. Estudios sobre Lenguas Americanas 9, 315–344. Mexico: Universidad Nacional Autónoma de México, Instituto de Investigaciones Filológicas. Williams, Alexander. 2015. Arguments in Syntax and Semantics. Cambridge: Cambridge University Press. Zylstra, Carol F. 1991. A syntactic sketch of Alacatlazala Mixtec. In Bradley, C. Henry and Barbara E. Hollenbach (eds.), Studies in the Syntax of Mixtecan Languages, Vol. 3, 1–178. Dallas: Summer Institute of Linguistics and the University of Texas at Arlington.

Notes 1 When citing research done by others, I have tried to keep the same morpheme glosses that they have used. 2 For discussion of body-part incorporation with an intransitive verb, resulting in a change of undergoer assignment (‘possessor raising’), see Van Valin (2005: 145–146). 3 The gloss B refers to one of three gender classes, in this case, one that refers to either an animate plural or an inanimate singular (Allen et al. 1984: 293, fn. 5) 4 The passive or ‘unspecified subject’ construction in Tepehua does not provide evidence of undergoer status. In a Tepehua passive construction, a non-undergoer can map onto the PSA (see 6.4.2). 5 These parallel the ‘two universals of passivization’ presented by Perlmutter and Postal: (i) ‘A direct object of an active clause is the (superficial) subject of the “corresponding” passive,’ and (ii ‘The subject of an active clause is neither the (superficial) subject nor the (superficial) direct object of the “corresponding” passive’ (1977: 399).

https://doi.org/10.1017/9781316418086.009 Published online by Cambridge University Press

317

7 Case Assignment Wataru Nakamura

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ACT ADE AV BV C CC DCA DMAP ELA GER ICR ILL INE IO

7.1

actor adessive actor voice benefactive voice catalyst particle Completeness Constraint domain of case assignment default macrorole assignment principles elative gerund instrumental case rule illative inessive indirect object

LS MP MR NUC OT PRT PSA PURP

Logical Structure modifier phrase macroroe nuclear Optimality Theory partitive privileged syntactic argument purpose

PV QS REC.PFV RP UND

patient voice qualia structure recent perfective reference phrase undergoer

Introduction

The aim of this chapter is twofold: to explicate the theory of case assignment in Role and Reference Grammar (RRG; Van Valin 1991, 1993, 2005, 2009; Van Valin and LaPolla 1997) and to outline its Optimality-Theoretic (OT) implementation (Nakamura 1999a, 1999b) and its extension to instrumental case assignment and case syncretism (Nakamura 2011, 2021). Case in RRG constitutes part of the linking system in which verbal arguments are realized by case/cross-referencing markers, agreement markers,

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

and/or word-order positions, but it plays no direct syntactic role in RRG, unlike in Government and Binding (GB)/Minimalism. Another distinguishing feature of the RRG theory of case assignment is that it ties the core cases (i.e. those that mark A, O and S arguments) to macrorole status (instead of identifying semantic roles of arguments they mark or assigning them to either particular grammatical relations or structural positions) and treats dative case on a par with those core cases by defining it as the default case for non-macrorole core arguments (Van Valin 1991; cf. Silverstein 1980/ 1993). What is notable about the macrorole-dependent theory of case assignment in RRG is that it makes no reference to any phrase-structural positions and therefore is well equipped to handle case-marking systems in both configurational and non-configurational languages. There are five major features of the RRG theory of case assignment as outlined in Van Valin and LaPolla (1997) and Van Valin (1993, 2005). First, RRG assigns the core cases (i.e. nominative, accusative, absolutive and ergative) in terms of the ranking of macroroles instead of tying them to grammatical relations or structural positions (cf. Yip et al. 1987; Marantz 1991; Baker 2015). Second, unlike GB/Minimalism, RRG does not assume that case assignment takes place in all languages whether or not they have any overt case markers; it does not distinguish between syntactic cases and morphological cases and assigns case markers directly to NPs (see Section 7.5 for an alternative proposal). Third, RRG adopts the macrorole-based definition of dative case and treats it on a par with the core cases. Fourth, RRG treats verbal cross-reference and nominal case systems in a unified way due to their functional equivalence. Finally, RRG parameterizes the domain of case assignment (DCA) into the core and clause and allows the set of case assignment rules to apply in each core independently or to all of the cores in a clause jointly. Taken together, these features distinguish RRG from the other syntactic frameworks with respect to case assignment. This chapter is organized as follows. Section 7.2 provides a brief summary of the RRG linking system, highlighting those aspects of it that are relevant to case assignment. Section 7.3 presents the RRG account of accusative, ergative, and active(-stative) (or split-S) case systems and summarizes the RRG account of oblique case assignment with a particular focus on dative and instrumental case. Section 7.4 outlines the OT implementation of the RRG theory of case assignment (Nakamura 1999a, 1999b) and its extension to instrumental case (Nakamura 2021). Section 7.5 introduces the distinction between syntactic and morphological cases into RRG to develop the OT-RRG account of case syncretism. Section 7.6 concludes the chapter.

7.2

Linking Theory in RRG

7.2.1 Syntactic Structure RRG is a monostratal theory that posits only a single syntactic representation for a sentence that consists of two projections, the constituent structure https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

319

320

WATARU NAKAMURA

CLAUSE CORE

RP

NUC

PERIPHERY

RP

MP

PRED

V John

saw

ADV Tom

yesterday

Figure 7.1 The constituent structure of the simple clause in English (RP ¼ reference phrase; MP ¼ modifier phrase)

projection and the operator projection. The former projection represents the clause structure with three nested layers, the nucleus (the predicate), the core (the predicate and its arguments), and the clause (the core and any peripheral elements that modify the core) (as illustrated in Figure 7.1), while the latter consists of auxiliary elements (e.g. aspect, tense, root/epistemic modal, status, evidential, speech act) that are hierarchically and topologically ordered according to the layer they modify.

7.2.2 Semantic Representation and Syntactic Function The above syntactic representation is coupled with the semantic representation of the clause based on the following decompositional representations of predicates (termed logical structures (LS)), adapted from Vendler (1967) and Dowty (1979) (Van Valin 2005: 42–49). (1)

Decompositional representations for Aktionsart classes a. State predicate′ (x) or (x, y) b. Activity do′ (x, [predicate′ (x) or (x, y)]), or c. Achievement INGR predicate′ (x) or (x, y), or INGR do′ (x, [predicate′ (x) or (x, y)]) d. Semelfactive SEML predicate′ (x) or (x, y), or SEML do′ (x, [predicate′ (x) or (x, y)]) e. Accomplishment BECOME predicate′ (x) or (x, y), or BECOME do′ (x, [predicate′ (x) or (x, y)]) f. Active accomplishment do′ (x, [predicate1′ (x) or (x, y)]) & INGR predicate2′ (z, x) or (y) g. Causative α CAUSE β, where α and β are logical structures of any type

(2a)–(2g) illustrate the above decompositional representations. (2)

a. State The cup is shattered. Carl is in the library.

shattered′ (cup) be-in′ (library, Carl)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

b. Activity The children cried. Kim ate fish. c. Achievement The windows shattered. The balloon popped. d. Semelfactive Dana glimpsed the dog. Mary coughed. e. Accomplishment The snow melted. Chris learned French. f. Active accomplishment John ran to the park. Kim ate the fish. g. Causative The dog scared the boy. Max melted the ice. Felix bounced the ball.

do′ (children, [cry′ (children)]) do′ (Kim, [eat′ (Kim, fish)]) INGR shattered′ (windows) INGR popped′ (balloon) SEML see′ (Dana, dog) SEML do′ (Mary, [cough′ (Mary)]) BECOME melted′ (snow) BECOME know′ (Chris, French) do′ (John, [run′ (John)]) & INGR be-at′ (park, John) do′ (Kim, [eat′ (Kim, fish)]) & INGR consumed′ (fish) [do′ (dog, ø)] CAUSE [feel′ (boy, [afraid′])] [do′ (Max, ø)] CAUSE [BECOME melted′ (ice)] [do′ (Felix, ø)] CAUSE [do′ (ball, [bounce′ (ball)])]

A key component of the RRG linking system is the two-tiered system of semantic roles. The first tier is thematic relations such as effector, experiencer, theme and patient. They are defined in terms of argument positions in the decompositional representations. The second tier comprises two semantic macroroles (MR), actor (ACT) and undergoer (UND). These are generalized semantic roles that subsume a number of LS arguments for morphosyntactic purposes and correspond to the two primary arguments of a transitive verb.1 The number and nature of macroroles that a verb takes is determined by the default macrorole assignment principles (DMAP) in (3). (3)

Default Macrorole Assignment Principles a. Number: the number of macroroles which a verb takes is less than or equal to the number of arguments in its LS: 1. If a verb has two or more arguments in its LS, it will take two macroroles. 2. If a verb has one argument in its LS, it will take one macrorole. b. Nature: for verbs which take one macrorole: 1. If the verb has an activity predicate in its LS, the macrorole is actor. 2. If the verb has no activity predicate in its LS, the macrorole is undergoer.

(3a) determines the number of macroroles a verb takes, while (3b) determines which macrorole (actor or undergoer) it is when the verb receives only one macrorole. When the number of macroroles does not follow from the DMAP, it has to be specified in the lexical entry of the verb by a feature [MRα] (where α represents the number of macroroles). The relationship between LS argument slots (or thematic relations serving as mnemonics for them) and macroroles is captured by the Actor-Undergoer Hierarchy (AUH) in Figure 7.2.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

321

322

WATARU NAKAMURA

Actor

Undergoer

Arg. of DO Agent

Arg. of do′ (x,...) Effector

1st Arg. of 2nd Arg. of Arg. of state pred′ (x, y) pred′ (x, y) pred′ (x) Locative Theme Patient Experiencer = increasing markedness of realization of LS argument as macrorole]

[

Figure 7.2 Actor-Undergoer Hierarchy (adapted from Van Valin and LaPolla 1997: 146)

The AUH states that given the LS of a multi-argument verb, the leftmost argument will be the actor and the rightmost argument will be the undergoer. (4a)–(4f) illustrate the default macrorole assignment. (4)

a. b. c. d. e. f.

John [Effector, ACT] killed Bill [Patient, UND]. Bill [Patient, UND] was killed by John [Effector, ACT]. John [Effector, ACT] gave a book [Theme, UND] to Bill [Recipient, Non-MR]. John [Experiencer, ACT] knew the student [Theme, UND]. John [Effector, ACT] ran to the park. John [Patient, UND] disappeared suddenly.

It is important to note that marked assignments to undergoer are possible, as illustrated by the locative alternation in (5a,b), in which the locative argument (the truck) may be chosen as undergoer in violation of the AUH. (5)

a. John loaded hay on the truck. b. John loaded the truck with hay.

The observation that the locative argument in (5b) (the truck) is construed as being fully loaded motivates its choice as undergoer.2 Finally, the privileged syntactic argument (PSA) selection is based on the PSA Selection Hierarchy in (6), the PSA selection principles in (7), and restrictions in (8) (Van Valin 2005: 100). (6)

PSA Selection Hierarchy arg. of DO > 1st arg. of do′ > 1st arg. of pred′ (x, y) > 2nd arg. of pred′ (x, y) > arg. of pred′ (x)

(7)

Privileged syntactic argument selection principles a. Accusative constructions: highest-ranking direct core argument in terms of the PSA Selection Hierarchy (default) b. Ergative constructions: lowest-ranking direct core argument in terms of the PSA Selection Hierarchy (default)

(8)

Restrictions on PSA in terms of macrorole status a. Languages in which only macrorole arguments can be PSA (e.g. German, Sama, Dyirbal, Jakaltek) b. Languages in which non-macrorole direct core arguments can be PSA (e.g. Icelandic, Georgian, Japanese, Kinyarwanda)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Since an actor argument always outranks an undergoer argument in the PSA Selection Hierarchy, (7a) groups A and S arguments against O arguments, while (7b) groups O and S arguments against A arguments.

7.3

The Theory of Case Assignment

7.3.1 Regular Cases This subsection is a summary of the RRG theory of case assignment outlined in Van Valin and LaPolla (1997) and Van Valin (2005). (9) and (10) are the sets of case assignment rules for accusative and ergative case systems, respectively. They refer crucially to the PSA Selection Hierarchy in (6), and not to grammatical relations or phrase structural positions. (9)

Case assignment rules (Accusative) a. Assign nominative case to the highest-ranking macrorole argument. b. Assign accusative case to the other macrorole argument. c. Assign dative case to non-macrorole core arguments (default).

(10)

Case assignment rules (Ergative) a. Assign absolutive case to the lowest-ranking macrorole argument. b. Assign ergative case to the other macrorole argument. c. Assign dative case to non-macrorole core arguments (default).

‘The highest-ranking macrorole argument’ in (9a) conflates A and S arguments, while ‘the lowest-ranking macrorole argument’ in (10a) conflates O and S arguments. What is noteworthy about (9) and (10) is that they define all the core cases (i.e. nominative, absolutive, accusative and ergative) and dative case with reference to (non-)macrorole status and, in so doing, treat dative as one of the regular cases along with the core cases.3 Defining dative as the default case for non-macrorole core arguments accounts for why source arguments of ditransitive verbs may receive dative case as well as ablative case in many languages and leaves room for its being overridden by other oblique cases that denote a more specific semantic content in terms of LS configuration.4 For example, (11a) is the ablative preposition assignment rule for English, which applies to the uses of from in (11b). (11)

a. MR: Non-MR LS: the first argument in the following LS configuration ‘. . . BECOME/INGR NOT have′/be-LOC′ (x, y)’ b. 1. John came from Chicago. 2. John received the watch from Mary.

(11a) is more specific than the macrorole-based definition of dative case in (9c) and (10c), in that it refers to a particular LS argument in addition to its non-macrorole status. This accounts for why the ablative preposition is used to mark source arguments as illustrated in (11b).5

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

323

324

WATARU NAKAMURA

The following shows how (9) and (10), respectively, work in Icelandic and Warlpiri and how the macrorole-based account of case assignment extends to active(-stative) case systems. First, let us consider the Icelandic data given in (12). (12)

Icelandic (Andrews 1990: 188; Van Valin 1991: 172) a. Stelpurnar hláu. the.girls.nom laughed ‘The girls laughed.’ b. Lögreglan tók Siggu fasta. the.police.nom took Sigga.acc fast.acc ‘The police arrested Sigga.’ c. Stelpan sýndi stráknum myndavélina. the.girl.nom showed the.boy.dat the.camera.acc ‘The girl showed the boy the camera.’ d. Ég tel lögregluna hafa tekið Siggu I.nom believe the.police.acc have taken Sigga.acc fasta. fast.acc ‘I believe the police to have arrested Sigga.’

(12a)–(12c) illustrate canonical intransitive, transitive and ditransitive constructions, while (12d) illustrates matrix-coding (‘raising’) constructions in which the PSA of the dependent core shows up in the matrix core. The case assignment in (12a)–(12d) proceeds as follows. The actor arguments in (12a)– (12c) correspond to ‘the highest-ranking macrorole argument’ in (9a) and receive nominative case. The undergoer arguments in (12b,c) correspond to ‘the other macrorole argument’ in (9b) and take accusative case. Likewise, the experiencer argument in the matrix core of (12d) receives nominative case from (9a), while both the matrix-coded argument lögreglan ‘the police’ and the patient argument Sigga in the dependent core correspond to ‘the other macrorole argument’ in (9b) and receive accusative case. Next, let us consider how to account for the DAT-NOM and NOM-DAT-DAT case frames in (13a)–(13c), in which the macrorole assignment does not follow straightforwardly from the DMAP in (3). (13)

Icelandic (Andrews 1990: 210; Van Valin 1991: 174, 175) a. Stráknum líkar slíkir bílar. the.boy.dat likes such cars.nom ‘The boy likes such cars.’ b. Henni hefur alltaf þótt Ólafur leiðinlegur. her.dat has always thought Olaf.nom boring.nom ‘She has always considered Olaf boring.’ c. Ég skilaði henni peningunum. I.nom returned her.dat the.money.dat ‘I returned her the money.’

The DMAP predicts that all of the multiple-argument verbs in (13a)–(13c) take a pair of actor and undergoer. However, this prediction is not borne out; contrary to (3a1), all of (13a)–(13c) contain only one macrorole argument.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

In contrast to the earlier analyses of quirky case in Icelandic (e.g. Andrews 1982, 1990; Yip et al. 1987), RRG derives the irregular (quirky) case marking illustrated in (13a)–(13c) from the irregularity in their macrorole transitivity. Specifically, Van Valin (1991) posits that þykja ‘think’, líka ‘like’, and skila ‘return’ in (13) are lexically prespecified for having only one macrorole ([MR1]) despite having two arguments in their LS. Let us see how this lexical pre-specification allows us to treat dative as one of the regular cases (Van Valin 1991; cf. Narasimhan 1998; Nakamura 1999b, 2008). First, suppose that the two-place state verb líka ‘like’ receives only one macrorole because of its lexical specification. (3b) dictates that the only macrorole is an undergoer, since the verb has no activity predicate in its LS. The AUH requires the theme argument slíkir bílar ‘such cars’ to become an undergoer. This leads the experiencer argument to receive a nonmacrorole status. (14)

Macrorole assignment of the verb líka ‘like’ ([MR1]) MR Non-MR Undergoer

LS:

like′ (boy,

such cars)

The case assignment rules in (9) apply to the combination of the nonmacrorole and undergoer arguments and yield the DAT-NOM case frame in (13a). The same account holds for the DAT-NOM case frame in (13b). Likewise, suppose that skila ‘return’ in (13c) is lexically specified for having only one macrorole ([MR1]). (3b) requires that the only macrorole that the verb takes is an actor, since it has an activity predicate do′ in its LS in (15). (15)

LS of the verb skila ‘return’ [do′ (x, ø)] CAUSE [INGR have′ (y, z) ^ NOT have′ (x, z) ] [MR1]

The effector argument is an actor, but the remaining non-PSA arguments have no choice but to receive a non-macrorole status, as shown in (16): (16)

Macrorole assignment of the verb skila ‘return’ ([MR1]) MR Actor Non-MR Non-MR

LS:

[do′ (I, ø)] CAUSE [INGR have′ (she,

money) ^ NOT have′ (I, money)]

This macrorole assignment accounts for why the non-PSA arguments in (13c) receive dative case. Lexical specification of the number of macroroles that þykja ‘think’, líka ‘like’, and skila ‘return’ involve allows RRG to treat dative not as an example of quirky cases, but as one of the regular cases. This is one of the major points of contrast between RRG and the other syntactic theories with respect to case assignment. (9) also accounts for the contrast between (17) and (18) with respect to case preservation under passivization.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

325

326

WATARU NAKAMURA

(17)

(18)

Icelandic (Van Valin 1991: 150, 151) a. Lögreglan tók Siggu fasta. the.police.nom took Sigga.acc fast.acc ‘The police arrested Sigga.’ b. Sigga var tekin föst af Sigga.nom was taken fast.nom by ‘Sigga was arrested by the police.’ Icelandic (Van Valin 1991: 152) a. Ég hjálpaði honum. I.nom helped him.dat ‘I helped him.’ b. Honum var hjálpaði (af him.dat was helped (by ‘He was helped (by me).’

lögreglunni. the.police.dat

mér). me.dat)

The patient argument Sigga receives nominative case in (17b), since it corresponds to ‘the highest-ranking macrorole argument’ in (9a). The reason for the preservation of dative case under passivization in (18b) is that the non-actor argument in (18a) retains its non-macrorole status after it undergoes passivization in (18b).6 Icelandic provides us with an opportunity to illustrate the effect of an important typological parameter concerning the DCA (i.e. domain of case assignment). As an initial motivation for the DCA parameter, let us consider (19). (19)

Icelandic (Van Valin and LaPolla 1997: 578) Jón telur mér hafa alltaf ϸótt John.nom believes me.dat have always thought leiðinlegur. boring.nom ‘John believes me to have always considered Olaf boring.’

Ólafur Olaf.nom

What is intriguing about (19) is that the theme argument of þykja ‘think’ in the dependent core receives nominative case. This Icelandic example casts doubt on the assumption shared by major syntactic theories that nominative case assignment may occur only in finite clauses and suggests the necessity of making it an option for a language to allow (9) or (10) to apply in each core within a clause independently. Van Valin and LaPolla (1997: 575–581) propose to derive the nominative case assignment in the dependent core in (19) from the DCA parameterization. The fact that the theme argument in the dependent core receives nominative case suggests that the DCA for Icelandic is the core, in contrast to languages such as German, which allows no nominative-marked argument to occur in any dependent core.7 (20) is a summary of the above discussion. (20)

DCA Parameterization a. DCA¼Clause (e.g. English, German) b. DCA¼Core (e.g. Icelandic, Japanese)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

The assumption that the DCA for Icelandic is the core explains the DATNOM case frame of the dependent verb in (19). First, let us assume that the dependent verb in (19) is irregular with respect to macrorole transitivity and that the two-place verb þykja ‘think’ has the feature [MR1] in its lexical entry. (3b) requires the only macrorole to be an undergoer, since þykja ‘think’ has no activity predicate in its LS. The AUH requires the theme argument to be an undergoer, which leads the experiencer argument of þykja to become a non-macrorole core argument. On the other hand, the experiencer argument of the matrix verb telja ‘believe’ is an actor and counts as ‘the highest-ranking macrorole argument’ in (9a). (21) shows how the macrorole assignment proceeds in the matrix and dependent core of (19). (21)

Macrorole assignment in (19) Matrix core [Jón telja ACT Dependent core

mér] Non-MR [(mér) (Non-MR)

þykja

Ólafur] UND

Applying the set of case assignment rules in (9) to the dependent core in (21) accounts for the nominative marking of the theme argument (Ólafur) in (19). In contrast to (9), which derives accusative case systems, (10) accounts for ergative case systems. (22a)–(22c) come from Warlpiri (Pama-Nyungan). (22)

Warlpiri (Hale 1983: 6, 13) a. Kurdu ka parnka-mi. child.abs aux run-npst ‘The child is running.’ b. Kurdu kapi wanti-mi. child.abs aux fall-npst ‘The child will fall.’ c. Ngarrka-ngku ka wawirri man-erg aux kangaroo.abs ‘The man is spearing the kangaroo.’

panti-rni spear-npst

Warlpiri is a double-marking language in which the agreement clitic system operates on a nominative-accusative basis, while independent pronouns and lexical NPs are case-marked according to an ergative-absolutive pattern. The Warlpiri case-marking pattern falls out from (10), which assigns absolutive case to O and S arguments, while assigning ergative case to A arguments.8 Finally, RRG extends the macrorole-based account of accusative and ergative case systems to active(-stative) cross-reference/case systems, illustrated by (23) and (24).9 (23)

Acehnese (Austronesian: Durie 1985, 1987) a. (Gopnyan) geu¼mat lôn (3sg) 3¼hold me ‘(S)he holds me.’

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

327

328

WATARU NAKAMURA

b. Geu¼jak (gopnyan) 3¼go (3sg) ‘(S)he goes.’ (Actor) c. Lôn ehët(¼lôn) 1sg fall(¼1sg) ‘I fall.’ (Undergoer) (24)

Western dialect of Basque (Aldai 2009: 785, 786) a. Peru-k sagarr-a-ø jan d-u-ø. Peru-erg apple-det-abs eaten 3sg.abs-aux-3sg.erg ‘Peru has eaten the apple.’ b. Peru-k dantzatu d-u-ø. Peru-erg danced 3sg.abs-aux-3sg.erg ‘Peru has danced.’ (Actor) c. Peru erori d-a. Peru.abs fallen 3sg.abs-aux ‘Peru has fallen.’ (Undergoer)

Acehnese is a head-marking language with clitics on the verb stem that cross-reference arguments and distinguishes between actor and undergoer arguments in terms of where their cross-referencing clitics occur: obligatory proclitics index actor arguments, while optional enclitics index undergoer arguments. In contrast to Acehnese, Basque is a double-marking language in which up to three arguments may be crossreferenced on the auxiliary verb, while independent pronouns and lexical NPs receive case. Aldai (2008, 2009) classifies dialects of Basque into three major types (Western, Eastern and Central) and states that the Western dialect marks actor and undergoer arguments with ergative and absolutive case, respectively, while the Eastern dialect operates on an ergativeabsolutive basis.10 It is important to recall from Section 7.1 that RRG views verbal crossreference systems as being functionally equivalent to nominal case systems (Van Valin 2013): the same set of case assignment rules applies to both crossreference and case systems that exhibit the split-S pattern, whether it is realized by linear morphological slots of cross-referencing affixes/clitics or nominal case affixes/clitics. Their unified account enables us to handle active(-stative) case systems (most of which are head-marking) on a par with accusative and ergative ones. (25) and (26) are the sets of case assignment rules for the two types of active(-stative) cross-referencing/case systems, which apply to linear morphological positions of verbal cross-referencing clitics (e.g. Acehnese) and nominal case affixes (e.g. Western dialect of Basque). (25)

Case assignment rules (Accusative-active) a. Assign nominative case to the actor argument. b. Assign accusative case to the undergoer argument. c. Assign dative case to non-macrorole core arguments (default).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

(26)

Case assignment rules (Ergative-active) a. Assign ergative case to the actor argument. b. Assign absolutive case to the undergoer argument. c. Assign dative case to non-macrorole core arguments (default).

To summarize, the four sets of macrorole-based case assignment rules in (9) (accusative), (10) (ergative), (25) (accusative-active) and (26) (ergativeactive), together with the DCA parameter in (20), constitute the core part of the RRG theory of case assignment.

7.3.2 Non-Dative Oblique Cases We saw in the last subsection that RRG defines dative as the default case of non-macrorole core arguments and treats it as one of the regular cases along with the core cases. This raises the question of how RRG handles oblique cases (or adpositions) other than dative case. RRG defines non-dative oblique cases in terms of the LS configuration in which the referent of the noun they mark occurs and its (non-)macrorole status (Jolly 1991, 1993; Van Valin 1993, 2005; Van Valin and LaPolla 1997). As a case in point, let us consider how RRG defines instrumental case, with a focus on the instrument, implement, locatum and comitative uses of the English instrumental preposition with. (27)

Instrument with a. John cut the meat with a knife. [do′ (John, [use′ (John, knife)])] CAUSE [[do′ (knife, [cut′ (knife, meat)])] CAUSE [BECOME cut′ (meat)]] Implement with b. Chris wrote the letter with a pen. do′ (Chris, [write′ (Chris, letter) ^ use′ (Chris, pen)]) & INGR exist′ (letter) Locatum with c. John loaded the truck with hay. [do′ (John, Ø)] CAUSE [BECOME be-on′ (truck, hay)] (UND¼locative) Comitative with d. John walked to the concert with Pat. (John and Pat went to the concert.) [do′ (John ^ Pat, [walk′ (John ^ Pat)])] & INGR be-at′ (concert, John ^ Pat) e. Pat served wine with cheese to the guest. (Pat served wine and cheese to the guest.) [do′ (Pat, Ø)] & [BECOME have′ (guest, wine ^ cheese)]

Two points are worth making about (27). First, the implement with is distinct from the instrument with, in that the instrument argument is part of a verb’s causal chain, while the implement argument is not. Second, with exhibits an instrumental/comitative syncretism. Jolly (1991, 1993) makes an important observation that the instrument with marks a potential actor that fails to function as actor, while the locatum with marks a potential undergoer that fails to serve as undergoer and that the comitative with marks one of the co-participants that would

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

329

330

WATARU NAKAMURA

otherwise appear as an actor or undergoer but does not. (28) is the unified definition of the instrument, locatum and comitative uses of with.11 (28)

Instrumental case rule (ICR) (Van Valin 2005: 110) Assign instrumental case to non-MR b argument if, given two arguments, a and b, in a logical structure, with (i) both as possible candidates for a particular macrorole and (ii) a is equal or higher (to the left of b) on the AUH, b is not selected as that macrorole.

The basic idea is that all of the instrument, locatum and comitative uses of with are non-predicative: they are associated with the outcome of certain linking options (as described in (28)), in contrast to locative prepositions (as illustrated in (29)), which are linked to a specific LS configuration. (29)

a. John was running in the park. LS: be-in′ (park, [do′ (John, [run′ (John)])]) b. John walked to the station. LS: do′ (John, [walk′ (John)]) & INGR be-at′ (station, John)

The two prepositions in (29) are associated with their LSs: in introduces the LS ‘be-in′ (. . .)’, a two-place predicate that takes the entire event (‘[do′ (John, [run′ (John)])]’) as the second argument and situates it in a location designated by the first argument, while to introduces the LS ‘INGR be-at′ (. . .)’, whose first argument designates the goal of the movement. Three remarks are in order about the ICR in (28). First, (28) does not extend to the implement with illustrated in (27b). The fact that the two-place activity verb write has no ready-made argument slot for the implement NP a pen disqualifies it as a possible candidate for an actor and puts the implement use of with outside the scope of (28). Second, (28) interacts with the dative case assignment rule in (9c) in an intricate way. For example, (27c) (repeated below) involves a marked undergoer assignment to the locative argument, which forces the theme argument hay to become a non-macrorole.12 (27)

c. John loaded the truck with hay.

The problem is that both (9c) and (28) may apply to hay. Van Valin (2005) argues that the ICR in (28) overrides (9c) and accounts for why hay is marked by the instrumental preposition under the crucial assumption that both instrumental with-phrases (as illustrated in (27a)) and locatum with-phrases (as illustrated in (27c)) constitute a subset of non-macrorole core arguments and that (28) applies to them.13 Finally, (28) applies to nominal cases and adpositions that nullify the morphological distinction between instrumental and comitative, but it is important to note that their syncretism is not common outside the Indo-European languages (Stolz et al. 2006). The fact that the majority of languages outside the Indo-European family distinguish the two cases (as illustrated in (30)) suggests an alternative strategy to analyse the non-comitative and comitative uses of with separately and then

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

to derive the instrumental-comitative syncretism (instead of deriving the two uses of with from a single rule). (30)

Japanese a. John-ga katana-de take-o kit-ta. John-nom sword-ins bamboo-acc cut-pst ‘John cut the bamboo with a sword.’ (Instrumental) b. John-ga Tom-to/*de eiga-o mi-ta. John-nom Tom-COM/ins movie-acc watch-pst ‘John watched the movie with Tom.’ (Comitative) c. John-ga kome-o niku-to/*de maze-ta. John-nom rice-acc meat-com/ins mix-pst ‘John mixed rice with meat.’ (Comitative)

To summarize, RRG divides oblique cases into predicative and nonpredicative oblique cases: the former (e.g. ablative, allative, locative) are associated with their LSs, while the latter (e.g. dative, instrumental) are either assigned to a non-macrorole core argument (only when no other case rule may apply) or are associated with the linking pattern in (28). Van Valin (1993, 2005) and Van Valin and LaPolla (1997) lay a foundation for the decompositional account of oblique case assignment, but they leave it as an open question how to provide a unified account of non-dative oblique cases/adpositions, most notably the instrumental case/adposition.

7.4

An OT Implementation

7.4.1 Regular Case Assignment This section outlines an OT implementation of the RRG theory of case assignment and its extension to instrumental case assignment (Nakamura 1999a, 1999b, 2015, 2021).14 OT is a constraint-based formalism with an emphasis on constraint interaction that views a grammar as a function that maps each input to its correct structural description (Prince and Smolensky 2004). Three fundamental principles of OT are relevant here. First, all constraints are violable. The only requirement for a candidate to be optimal is that it is the minimal violator in the given candidate set. Second, a grammar resolves conflicts among constraints by ranking them in a strict dominance hierarchy, in which each constraint has absolute priority over all the lower-ranking constraints. Third, re-ranking of a set of individually simple constraints yields typological variation. Nakamura (1999a) proposes the set of OT constraints for the regular case assignment in (31). (31)

Case assignment constraints a. Some argument receives nominative case. b. Undergoer arguments receive accusative case. c. Actor arguments receive ergative case. d. Non-macrorole core arguments receive dative case.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

331

332

WATARU NAKAMURA

(31a)–(31d) constitute a dominance hierarchy that receives as an input a pair of the two-tiered semantic representations of a predicate and outputs its case frame. Unlike (9a,b) and (10a,b), (31a)–(31c) make no reference to the ranking of actor and undergoer. Five remarks are in order about (31a)–(31d). First, (31a) comes originally from Jakobson (1936/1984), who analyses nominative as the default case with no semantic content, and is required by languages such as Japanese where every core has to have at least one nominative-marked argument. Second, the relative ranking of (31a,d) determines whether or not a language allows preservation of dative (or some other oblique) case under passivization: when (31d) outranks (31a), non-macrorole core arguments cannot receive nominative case (as illustrated by (18)), while when (31a) outranks (31d), no case preservation under passivization is allowed.15 The latter situation is illustrated by Japanese examples given in (32). (32)

a. John-ga Tom-ni John-nom Tom-dat ‘John bumped into Tom.’ b. Tom-ga/*ni John-ni Tom-nom/dat John-dat ‘Tom was bumped into by John.’

butsukat-ta. bump.into-pst butsuka-rare-ta. bump.into-pass-pst

The dative-marked argument in (32a) bears nominative case under passivization (as shown in (32b). Third, (31a) groups nominative case in accusative systems and absolutive case in ergative systems as a single category in spite of their distributional difference. Fourth, (31b,c) are derived from the hypothesis that accusative and ergative case, respectively, mark undergoer and actor arguments exclusively. Finally, various rankings of (31a)–(31c) yield the accusative, ergative, and two types of active(-stative) case systems, as shown in (33).16 (33)

Accusative case system a. (31d) >> (31a) >> (31b) (>> (31c)) b. (31a) >> (31d) >> (31b) (>> (31c)) Ergative case system c. (31d) >> (31a) >> (31c) (>> (31b)) Accusative-active case system d. (31d) >> (31b) >> (31a) (>> (31c)) Ergative-active case system e. (31d) >> (31c) >> (31a) (>> (31b))

e.g. Icelandic, German e.g. Japanese, French e.g. Warlpiri e.g. Acehnese e.g. Western dialect of Basque

Let us see how the constraint ranking in (33a) derives the NOM-ACC and DAT-NOM case frames in Icelandic. Tableau 7.1 shows how (33a) outputs the NOM-ACC case frame illustrated in (12b).17 (12)

b. Lögreglan tók Siggu the.police.nom took Sigga.acc ‘The police arrested Sigga.’

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

fasta. fast.acc

Case Assignment

Tableau 7.1 Transitive constructions in Icelandic (31d)

(31a)

NOM-NOM

(31b)

(31c)

*!

*

☞NOM-ACC

*

DAT-NOM

*!

ACC-ACC

*!

* *

Tableau 7.2 ‘Dative-subject’ constructions in Icelandic (31d) NOM-NOM

*!

NOM-ACC

*!

(31a)

(31c)

*

*

*

*

*

☞DAT-NOM DAT-ACC

(31b)

*!

*

The input to the constraint hierarchy consists of an actor and undergoer argument. The first and third candidates violate (31b), while the fourth candidate violates (31a). In contrast, the second candidate violates the lowest-ranking constraint alone and emerges as the winner. The winner may violate constraints in OT, as long as the other candidates violate any higher-ranking constraint(s). Tableau 7.2 shows how the DAT-NOM case frame in (13a) is derived. (13)

a. Stráknum líkar slíkir the.boy.dat likes such ‘The boy likes such cars.’

bílar. cars.nom

The first and second candidates violate the top-ranking constraint (31d), since the non-macrorole core argument receives nominative case. The fourth candidate violates (31a), since it contains no nominative argument. The above consideration leaves the third candidate as the winner. The third candidate violates (31b), but it fares better than the other candidates, since the other candidates violate either (31d) or (31a) (which is ranked higher than (31b)). An analogous account holds for ergative and active(-stative) case systems. Tableau 7.3 shows how the ERG-NOM case frame is derived in (22c). (22)

c. Ngarrka-ngku ka wawirri man-erg aux kangaroo.nom ‘The man is spearing the kangaroo.’

panti-rni spear-npst

The difference between Tableau 7.1 and Tableau 7.3 comes down to the relative ranking of (31b) and (31c): when (31b) outranks (31c), an accusative case system emerges, while when (31c) outranks (31b), an ergative case system emerges.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

333

334

WATARU NAKAMURA

Tableau 7.3 Transitive constructions in Warlpiri (31d)

(31a)

(31c)

(31b)

NOM-NOM

*!

*

NOM-ACC

*!

☞ERG-NOM ERG-ACC

* *!

This suggests that the traditional distinction between nominative and absolutive is a by-product of the relative ranking of (31b) and (31c) and that there is no need to postulate absolutive case in addition to nominative case. Finally, the constraint rankings in (33d) and (33e) yield accusative-active and ergative-active case systems, respectively. When (31b) outranks (31a) as in (33d), undergoer arguments always receive accusative marking, while actor arguments receive nominative marking. In contrast, when (31c) outranks (31a) as in (33e), actor arguments receive ergative case, while undergoer arguments receive nominative case. To summarize this subsection, the OT reformulation of the case assignment rules in (9), (10), (25) and (26) in terms of the case assignment constraints in (31) accommodates the major case frames of simple clauses in accusative, ergative, and two types of active case systems and derives the typological variation of case systems from re-ranking of the individually simple constraints that make no reference to the relative ranking of actor and undergoer.

7.4.2 Instrumental Case Assignment Nakamura (2015, 2021) extends the domain of regular case assignment to instrumental case within the OT-RRG framework outlined in the previous subsection (cf. Van Valin and LaPolla 1997; Van Valin 2005, 2009). This extension is inspired by Jakobson (1936/1984), who decomposes eight cases in Russian in terms of three privative semantic features ([peripheral], [directional] and [quantified]) (as partially shown in Table 7.1 below) and defines instrumental case as having the [peripheral] feature alone (which covers oblique core arguments and adjuncts put together), while defining dative case as having the [peripheral] and [directional] feature. These featural definitions suggest that dative case marks oblique arguments whose referents are at the receiving end of an action (e.g. recipient arguments of ditransitive verbs), while instrumental case serves as the default case for adjuncts. Three problems remain to be solved before proposing an OT-RRG account of instrumental case assignment. First, instrumental case marks not only a wide range of adjuncts as illustrated in (34a)–(34e), but also non-macrorole core arguments that involve a violation of the AUH, as illustrated in (34f).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Table 7.1 Jakobson’s (1936/1984) featural definitions of nominative, accusative, dative and instrumental cases Peripheral Nominative Accusative Dative Instrumental

(34)

þ þ

Directional

Quantified

þ þ

Russian (Kilby 1977: 75; Wierzbicka 1980: 23; Janda 1993: 147, 156) a. Ivan napisal pisʹmo ruˇ ckoj. Ivan.nom wrote letter.acc pen.ins ‘Ivan wrote the letter with a pen.’ b. Ivan pil vino litrami. Ivan.nom drank wine.acc liters.ins ‘Ivan drank wine by the litre.’ c. Petr kivnul golovoj. Peter.nom nodded head.ins ‘Peter nodded his head.’ (lit. ‘Peter nodded with the head.’) d. Monax dolžen svjazatʹ usta svoi molˇ caniem. monk.nom must tie.up lips.acc own.acc silence.ins ‘A monk must seal his lips with silence.’ e. Lošadʹ soseda luˇ cše moej horse.nom neighbour.gen better mine.gen i siloj i krasotoj and strength.ins and beauty.ins ‘The neighbour’s horse is better than mine both in strength and in appearance.’ f. Oni gruzili baržu drovami. they.nom loaded barge.acc firewood.ins ‘They loaded the barge with firewood.’

In order to treat instrumental case on a par with the regular cases (i.e. nominative, dative, accusative and ergative), it is imperative to derive all the argument and adjunct uses of instrumental case from a single constraint associated with a unitary meaning. Second, the question of how the instrumental case assignment interacts with the dative case assignment needs to be addressed. What is at stake here is how to explain why hay in (34f) receives instrumental marking despite being qualified to receive dative case because of its status as a non-macrorole core argument. This is a clear indication that it is not enough to define instrumental as the default case for adjuncts. Finally, it remains to be shown how to derive the wide range of interpretations an instrumental-marked noun may have from its unitary meaning. In order to solve the first problem, Nakamura (2015, 2021) defines instrumental case as the default case for everything other than macrorole arguments and proposes (35b) as a constraint that represents the unitary

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

335

336

WATARU NAKAMURA

meaning of instrumental case. Nakamura goes on to revise (31d) (as shown in (35a)) in such a way that it does not apply to non-macrorole core arguments as in (34f) that involve a violation of the AUH (in this case, failure to assign undergoer status to the lowest-ranking argument) and proposes to rank (35a) higher than (35b) universally.18 (35)

Dative and instrumental case assignment a. Non-macrorole core arguments that do not involve a violation of the AUH receive dative case (the underlined part is added to (31d)). b. Case-bearing elements other than macrorole (core) arguments receive instrumental case.

(35a,b) and their ranking recast the Jakobsonian definition of instrumental case in terms of the three-layered clausal structure and semantic macroroles and account for why non-macrorole core arguments such as the one in (34f) (which are beyond the scope of the decompositional definition of instrumental case given in Table 7.1) receive instrumental case.19 (35b) accommodates a wide range of instrumental-marked adjuncts in Russian (and, by extension, other languages) whose interpretations are not determined by predicate semantics, but the highly underdetermined meaning of instrumental case raises the question of how to interpret each of its adjunct uses appropriately. This third problem requires serious consideration, since (35b) alone leaves an instrumental-marked noun unlinked to any argument position in the LS and provides no clue as to the specific role it plays in the clause.20 This amounts to a violation of the Completeness Constraint (CC), a very general principle in RRG that governs the linking between the syntactic and semantic representations. (36)

Completeness Constraint (Van Valin 2005: 129–130) All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence.

Jakobson (1936/1984) suggests that particular meanings of instrumental case (e.g. instrument, implement, unit, manner, path, cause) arise through its interaction with context, but he leaves it open what the context consists of and how its contextual interpretation is derived. There is no space in this chapter to provide a full account of how the various meanings of instrumental case are derived in context (see Nakamura (2021) for detailed discussion), but a few illustrations can be given. First, we may follow Van Valin (2012) in deriving the meaning of implement as in (27b) (and, by extension, (34a)) from the qualia structure (QS) (Pustejovsky 1995) of the instrumental-marked noun in (37a).21

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

(27)

b. Chris wrote the letter with a pen.

(37)

a. Formal: physical-object′ (a), stationery′ (a) Telic: do′ (b, [write′ (b, c) ^ use′ (b, a)]) Agentive: artifact′ (a) Constitutive: . . . b. do′ (b, [write′ (b, c)]) & INGR exist′ (c) c. do′ (Chris, [write′ (Chris, letter) ^ use′ (Chris, pen)]) & INGR exist′ (letter)

Invoking the QS of pen (more specifically, its telic quale) for interpreting the prepositional phrase (and the whole sentence that contains it) in (27e) leads to introducing a new argument position for its complement noun (pen) and thereby prevents a violation of the CC. Van Valin (2012) derives the LS of (27e) (given in (37c)) by merging the LS in the telic quale in (37a) with the LS of the active accomplishment use of the activity verb write in (37b). This operation is termed co-composition (Pustejovsky 1995), an operation that derives the meaning of a phrase compositionally from the head (in this case, the preposition) and its argument when the latter affects the meaning of the phrase (and the clause that contains it) beyond its role as an argument of the head. An analogous account holds for (34a). Second, Nakamura (2021) takes the above QS-based account of (27e) as the first step toward identifying contextually appropriate interpretations of instrumental-marked nouns and extends it to other uses of instrumentalmarked nouns, four of which are given in (34b)–(34e). Let us begin with (34b), which involves a measure unit by which to quantify an indefinite amount of some object, while engaging in some activity involving it (in this case, drinking wine). We may take the instrumental-marked noun litrami ‘litres’ as an adjunct, since it is not subcategorized by the verb. The underspecified meaning of instrumental case requires the unit noun litr ‘litre’ to be linked to an argument position of an event involved by its QS. I propose that the unit interpretation of litrami is derived from its telic qualia given in (38a): litr ‘litre’ refers to a unit that is used to measure an amount of liquid. (38)

a. Qualia structure of litr ‘litre’ TELIC: do′ (x, [use′ (x, y)]) PURP do′ (x, [measure′ (x, z)]) b. LS of pitʹ: do′ (x, [drink′ (x, z)]) c. LS of (34b): do′ (Ivan, [drink′ (Ivan, wine)] ^ [use′ (Ivan, litre) PURP measure′ (Ivan, wine)])

The CC requires us to interpret litrami ‘litres’ as a participant of the event involved by its telic quale. Merging the LS of the activity verb pitʹ ‘drink’ in (34b) with the LS in the telic quale of litr in (38a) yields the LS of (34b) in (38c). Third, the instrumental-marked noun in (34c) fulfils two functions simultaneously. First, the body-part noun golovoj ‘head’ serves as an adjunct to the main predicate and restricts the verb’s meaning by identifying the body part that is directly involved in the event rather than designating a target of

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

337

338

WATARU NAKAMURA

Peter’s action (Wierzbicka 1980). Second, the body-part noun forms a possessive relation with the subject argument: its constitutive quale (‘have.as.part′ (x:human, y)’) allows the hearer to associate Petr and golovoj with the possessor (‘x’) and possessum (‘y’), respectively. Another point to note in this connection is that (34c) encodes the possessive relation in terms of predication (and not in terms of reference, as illustrated in (39)) despite the fact that the possessive predicate remains covert. (39)

golova head.nom ‘Peter’s head’

Petra Peter.gen

(40)

a. Main predicate: b. Covert possessive predicate:

SEML do′ (Peter, [nod′ (Peter)]) have.as.part′ (Peter, head)

The main predicate kivnutʹ ‘nod’ and the covert possessive predicate share the subject argument Petr, but these predicates do not form a nuclear juncture, since Petr participates in the two events in (40a,b) independently.22 This suggests that (34c) embodies a non-subordinate core juncture, in which each of the predicates has its own core. The macrorole assignment proceeds in the two cores independently. The main predicate kivnutʹ ‘nod’ is a single-macrorole verb. Petr functions as undergoer when it counts as the first argument of the covert possessive predicate in (40b) (since it has no activity predicate do′ in its LS), while Petr serves as actor when it counts as the effector argument of the main predicate. The possessum argument golova ‘head’ becomes a non-macrorole core argument and receives instrumental case from (35b), since the possessor argument Petr bears the undergoer status in violation of the AUH.23 In contrast to (34a)–(34c), the instrumental-marked noun in (34d) (molˇcanie ‘silence’) has no (default) value for its constitutive, telic or agentive quale: it is not conventionally associated with any particular material, purpose, function or origin. This yields the consequence that we need to understand the function of molˇcanie with no reference to its QS. Since molˇcanie is neither an argument of the predicate nor part of the predicate, we have to conclude by a process of elimination that this instrumental-marked noun serves as a modifier: it modifies the manner of the action denoted by the verb svjazatʹ ‘tie up.’ Finally, let us consider how the meaning of domain restriction is derived in (34e). The first point to note here is that both sila ‘strength’ and krasota ‘beauty, appearance’ are part of the formal qualia of lošadʹ ‘horse.’ The fact that these attributes constitute part of the attributes of a horse explains why the two instrumental-marked abstract nouns serve to relativize the evaluation of the neighbour’s horse by the speaker and receive the interpretation of domain restriction. The fact that the two abstract nouns are not subcategorized by luˇcše suggests that they function as a modifier of the predicate and receive instrumental case from (35b) as dominated by (35a) (since they are not syntactic arguments).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

To summarize, this section has shown that the OT reformulation of the case assignment rules in (9) (accusative), (10) (ergative), (25) (accusativeactive) and (26) (ergative-active) clarify what these case systems share and where they diverge and has provided a brief summary of how to extend the OT-RRG account of the regular cases to instrumental case.

7.5

An OT-RRG Account of Case Syncretism

One of the major principles of the RRG theory of case assignment (Van Valin 1993, 2005; Van Valin and LaPolla 1997) is that it does not distinguish syntactic cases from morphological cases, under the assumption that each case receives its unique morphological realization (see Section 7.1). However, this assumption is called into question by case syncretism phenomena, where a single case morpheme realizes more than one syntactic case. This section outlines the OTRRG account of case syncretism (based mainly on Nakamura (2011)), with a particular focus on those instances of syncretism in which a single case morpheme realizes more than one regular case other than instrumental case (i.e. nominative, accusative, ergative, dative and genitive). The rest of this section is organized as follows. First, I will introduce the case hierarchy (CH) (Silverstein 1980/1993; cf. Blake 2001; Caha 2013) and will summarize what the CH has to say about the typological variation of case syncretism among the regular cases. Second, I will illustrate the representative types of syncretism predicted by the CH. Finally, I will show how and to what extent these instances of case syncretism are derivable in terms of OT.

7.5.1 The Case Hierarchy The CH comprises two parts: the upper part represents implicational relations among propositional and adnominal case morphemes, while the lower part represents implicational relations among adverbial and propositional case morphemes. (41)

Case Hierarchy (Silverstein 1976, 1980/1993)24 a. Propositional/Adnominal: Nom/Abs: Dat1 ⟵ {Acc, Erg} ⟵ Gen b. Adverbial/Propositional: Dat2 ⟵ Instr, Loc. . .

(41a) states that if a language has a distinct morpheme for genitive, it has a distinct morpheme for accusative and/or ergative, that if a language has a distinct morpheme for accusative and/or ergative, it has a distinct morpheme for dative and nominative/absolutive, while (41b) states that if a language has a distinct morpheme for representing instrumental and/or locative, it has a distinct morpheme for dative. Five remarks are in order about (41). First, (41a) states that nominative/ absolutive and dative constitute the minimal case-marking system and that

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

339

340

WATARU NAKAMURA

three/four/five-way case-marking systems arise as a gradual elaboration of this fundamental contrast.25 Second, (41a) treats nominative and absolutive as different manifestations of the same case (Section 7.4.1). Third, some languages (e.g. Estonian) have no case morpheme termed ‘dative’, but they have some other oblique case morpheme that behaves like it (Matsumura 1994, 1996; see Section 7.5.2). Fourth, a genitive case expresses a possessive relation as distinguished from an attributive modification. This means that attributive markers are not analysed as genitive.26 Finally, (41a) and split accusativity and ergativity determine the possible range of case syncretism among the regular cases. (42) is a list of the patterns of case syncretism predicted by (41a). (42)

Variation of case syncretism a. Gen¼Erg¼Acc¼Dat, Nom b. Gen¼Erg¼Dat, Nom c. Gen¼Acc¼Dat, Nom d. Gen¼Dat, Acc, Nom e. Gen¼Dat, Erg, Nom f. Gen¼Acc, Ade(Dat), Nom g. Gen¼Erg, Dat, Nom h. Gen¼Erg, Acc¼Dat, Nom i. (Gen), Erg, Acc¼Dat, Nom j. Acc¼Dat, Nom

(e.g. Yagnob) (e.g. Kabardian) (e.g. Palauan) (e.g. Old Persian) (e.g. Djaru) (e.g. Estonian) (e.g. Inuktitut) (e.g. Tagalog) (e.g. Hindi) (e.g. Spanish)

7.5.2 Patterns of Case Syncretism (42a)–(42c) are two-way case-marking systems illustrated by Yagnob (Iranian), Kabardian (Northwest Caucasian), and Palauan (Austronesian). They exhibit an extensive syncretism that applies to all non-nominative case morphemes available. (43a)–(43f) come from Kabardian. Kabardian (Colarusso 1992: 167; Smith 1996: 108, 111, 113) š -r f ́ z -m j r jtáhs. a. ɬ’ -m ˙ man-obl horse-nom woman-obl (nom.3).io.act.gave ‘The man gave the horse to the woman.’ š -r j -w h’áhs. b. ɬ’ -m ˙ man-obl horse-nom (nom.3)-act-killed ‘The man killed the horse.’ mabáhna. c. ha-r žás -m ˙ ˘ dog-nom night-obl (nom.3)-bark ‘The dog barks at night.’ f ́ z -m náxra nax’’ ́ zs. d. ɬ’ -r ˙˙ man-nom woman-obl older (nom.3)-is ‘The man is older than the woman.’ jahh. e. máz -m ˘ forest-obl act.(nom.3).carry ‘They carry it to the forest.’ f. ha-m Ø-y -pa-r ˘ dog-obl 3-poss-nose-nom ‘the dog’s nose’ e e

e e

e

e e

(43)

e

e

e

e

e

e e

e

e

e

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

The nominative case suffix -r marks S and O arguments, while the oblique case suffix -m marks A arguments, a wide range of oblique core arguments and adjuncts, and adnominal possessors. In contrast to the nominative case suffix, which may appear only once per clause, the oblique case suffix may appear multiple times within a clause. The fact that Kabardian distinguishes A arguments from S and O arguments in terms of case marking suggests that its case system (as well as its cross-referencing system) involves an ergative alignment. In contrast to Kabardian, Palauan involves an accusative case alignment. (44a)–(44f) show that it uses the oblique preposition er to mark nonmacrorole core arguments, human and/or specific-and-singular O arguments in imperfective clauses, adjuncts and adnominal possessors, while leaving S and A arguments unmarked.27 (44)

Palauan (Josephs 1975: 235; Georgopoulos 1991: 26, 27, 29) a. ak-mo er a katsudo. r.1sg-go obl movies ‘I am going to the movies.’ b. ng-kiltmekl-ii a ulaol a Peter. r.3sg-clean-3sg floor Peter ‘Peter cleaned the floor.’ c. ng-diak ku-nguiu er a hong. neg irr.1sg-read obl book ‘I am not reading the book.’ d. ng-mo er a ngebard er a klukuk. r.3sg-go obl west obl tomorrow ‘She is going to America tomorrow.’ e. A Romana a omeka er a rengalek Romana feed obl children er a kukau. obl taro ‘Romana is feeding the children the taro.’ f. ak-uleldanges er a resensei er ngak r.1sg-honor.ipfv obl teachers obl me ‘I respected my teachers.’

Tables 7.2(a,b) represent the extensive case syncretism in Kabardian and Palauan. Yagnob (Iranian) exhibits an even broader range of syncretism than Kabardian and Palauan: non-macrorole core arguments, A arguments in ergative constructions, definite O arguments in accusative constructions, and adnominal possessors receive the same oblique case marker -i (Comrie 1981: 169–170). (42d)–(42i) illustrate three-way case-marking systems with a genitivedative, genitive-ergative, genitive-accusative, or dative-accusative syncretism. Let us begin with the genitive-dative syncretism in (42d,e). (45a)–(45c) come from Djaru (Pama-Nyungan).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

341

342

WATARU NAKAMURA

Table 7.2(a) Case syncretism in Kabardian Case

Case marking

NOM

-r

ERG GEN DAT INSTR, LOC . . .

-m (oblique)

Table 7.2(b) Case syncretism in Palauan Case

Case marking

NOM

ø

ACC GEN DAT INSTR, LOC . . .

er (oblique)

(45)

Djaru (Tsunoda 1981: 110, 115, 124) a. ŋaniŋa guɲar guɳga 1sg.dat dog dead ‘My dog died last night.’ b. ŋumbir-u ŋa-la maŋari woman-erg child-3sg.dat food ‘A woman gives food to a child.’ c. ŋaᶁu-ŋgu ŋa-ɳa-ŋgu 1sg-erg c-1sg.nom-3sg.dat ɲunuŋa. 2sg.dat ‘I obtained meat for you.’ (or ‘I took/stole

ɲir-a stay-pst

waɽulu-la. night-loc

jambagina-wu child-dat

juŋ-an. give-prs

guju meat

man-i take/get-pst

your meat.’)

(45a,b) indicate that the same case morpheme is used to mark recipient arguments of ditransitive verbs and adnominal possessors. This genitivedative syncretism renders (45c) ambiguous and allows the second-person dative pronoun in (45c) to be construed as a benefactive or possessor (Tsunoda 1981: 110). Likewise, Old Persian (Indo-Iranian) exhibits the same syncretism in its accusative case system (Benvenuto and Pompeo 2015). Romanian (Romance) provides another illustration of genitive-dative syncretism. (46)

Romanian (Cornilescu 2000: 91; Hill 2013: 140) a. cartea profesorului book.def teacher.def.obl ‘the teacher’s book’ ~ deloc bine profesoarei. b. Copiii n-au raspuns children neg-have responded at.all well teacher.def.obl ‘Children have not responded at all well to the teacher.’

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

c. Ion li-a admirat Ion him-has admired ‘Ion admired his friend.’

pe acc

prietenuli friend.def

lui. his

Table 7.3 Case syncretism in Romanian Case marking Case

Inflection

NOM ACC

Nominative

GEN DAT

Oblique

Preposition

pe

Romanian has a two-way inflectional case-marking system for nouns, while using a differential object marker pe usually accompanied by clitic doubling (as illustrated in (46c)) to mark those undergoer arguments of transitive verbs that are relatively high on the definiteness and animacy scales (see Mardale (2008) for discussion of how the two scales are combined). Table 7.3 is a summary of the hybrid case-marking system of Romanian.28 The Romanian case-marking system involves split-accusativity triggered by definiteness and animacy in addition to the genitive-dative syncretism. The genitive-dative syncretism is also attested in Macedonian, Bulgarian, Albanian and Modern Greek and has been recognized as one of the Balkan Sprachbund properties (Tomi´ c 2004: 12–15). Halkomelem (Salish) is a double-marking language that exhibits a two-way case-marking system for nouns, while possessing a split-ergative agreement system:29 it involves a genitive-dative syncretism (using the same oblique case suffix to mark oblique nouns and proper-noun adnominal possessors) and makes no formal distinction among A, O and S arguments, all of which receive nominative (‘straight’) case (Gerdts 1988). The genitive-accusative syncretism is attested in Northern Sámi, Estonian and Finnish (Finno-Ugric). As a point of departure, let us consider the case-marking system in Northern Sámi. The genitive/accusative case morpheme in Northern Sámi marks (i) complement nouns of postpositions and prepositions, (ii) subject arguments of non-finite verbs, and (iii) nouns that designate a standard of comparison in comparative constructions, in addition to adnominal possessors and undergoer arguments of transitive verbs.30 (47)

Northern Sámi (Valijärvi and Kahn 2017: 52, 53) a. Máret orru vánhemiid luhtte. Máret lives parents.gen.pl with ‘Máret lives with her parents.’ b. Son ođii Máhte humadettiin. 3sg.nom slept Máhtte.gen.sg talk.ger ‘S/he slept while Máhtte was talking.’

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

343

344

WATARU NAKAMURA

c. Biila lea skohtera car.nom.sg is snowmobile.gen.sg ‘The car is newer than the snowmobile.’

ođđaseabbo. newer

The genitive-accusative syncretism holds across the board in Northern Sámi. The semantic and syntactic diversity of genitive/accusative-marked nouns indicates that the genitive/accusative case is assigned to nouns that would remain caseless otherwise. Estonian represents a more complex situation than Northern Sámi, in that the genitive-accusative syncretism holds only in the singular, and the genitive/accusative case alternates with the partitive case when it occurs on undergoer arguments of transitive verbs. When the genitive/accusative case alternates with the partitive case, it serves as an indicator of aspectual boundedness, as demonstrated by the contrast between (48b) and (48c).31 (48d) indicates that when transitive constructions with plural undergoer arguments denote a telic event, they receive the nominative case (instead of the genitive/accusative case). (48)

Estonian (Miljan 2008: 2, 177) a. maja aknad house.gen.sg window.nom.pl ‘windows of the house’ b. Raul ehitas suvila. Raul.nom build.pst.3sg cottage.gen.sg ‘Raul built a cottage.’ (completed) c. Raul ehitas suvila-t. Raul.nom build.pst.3sg cottage-prt.sg ‘Raul was building a cottage.’ (incomplete) d. Kass sõi hiire/hiired cat.nom.sg eat.pst.3sg mouse.gen.sg/mouse.nom.pl ‘The cat ate the mouse/the mice.’

ära. up

Table 7.4 shows how undergoer arguments of transitive verbs are casemarked in Estonian. Furthermore, Finnish syncretizes genitive and accusative in the singular declension, exhibits an aspectually conditioned case alternation on undergoer arguments of transitive verbs (which is analogous to the one shown in Table 7.4), and allows the genitive/accusative case to mark a wide variety of nouns, including complement nouns of many postpositions and subject Table 7.4 Case marking of O arguments in Estonian

O arguments

Number

Telicity

Case marking

singular

bounded unbounded

Genitive/Accusative Partitive

plural

bounded unbounded

Nominative Partitive

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

arguments of non-finite verbs, but Finnish is in contrast to Estonian, in that it has distinct accusative case forms for first-person pronouns, second-person pronouns, and third-person animate pronouns and animate interrogative pronouns. A genitive-ergative syncretism is attested in Tagalog (Austronesian), Inuktitut (Eskimo-Aleut) and Mayan languages. (49a)–(49c) come from Inuktitut. (49)

Inuktitut (Johns 1992: 58, 59, 69) a. arna-up angut kuni-ga-a. woman-rel man.abs kiss-pass.ptcp-3sg/3sg ‘John is stabbing/stabbed the seal.’ b. angut ani-juq. man.abs go.out-intr.ptcp.3sg ‘The man went out.’ c. anguti-up qimmi-a man-rel dog-3sg ‘the man’s dog’

The case morpheme that marks adnominal possessors and actor arguments of transitive verbs in the Eskimo languages has been termed ‘relative case’ in Eskimo linguistics. This and similar instances of the genitive-ergative syncretism have repeatedly been argued to have arisen from (a reanalysis of ) nominalized constructions (Trask 1979; Bricker 1981; Starosta et al. 1982; Johns 1992; Kaufman 2009; Coon 2009, 2013). The consensus in the literature cited here is that the syntactic parallelism between the possessive and transitive constructions goes a long way toward accounting for the genitive-ergative syncretism. Another example of the genitive-ergative syncretism comes from Tagalog, which displays a symmetrical voice system with verbs bearing voice morphology to indicate the semantic role of a nominative-marked argument, as illustrated by (50a)–(50h). (50)

Tagalog (Kroeger 1993: 13, 14, 23, 32, 50; Sabbagh 2016: 658) a. Pinutol ng¼magsasaka ang¼sungay ng¼kalabaw. pfv.pv.cut gen¼farmer nom¼horn gen¼buffalo ‘The farmer cut off the buffalo’s horn.’ (Patient voice) b. Ibinigay lahat ng¼mga¼guro sa¼mga¼bata pfv.pv.give all gen¼pl¼teacher dat¼pl¼child ang¼pera. nom¼money ‘The teachers gave all the money to the children.’ (Patient voice) c. Bumili ang¼lalake ng¼isda sa¼tindahan. pfv.av.buy nom¼man gen¼fish dat¼store ‘The man bought fish at the store.’ (Actor voice) d. Binili ng¼lalake ang¼isda sa¼tindahan. pfv.pv.buy gen¼man nom¼fish dat¼store ‘The man bought the fish at the store.’ (Patient voice)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

345

346

WATARU NAKAMURA

e. Ibinili ng¼lalake ng¼isda ang¼bata. pfv.bv.buy gen¼man gen¼fish nom¼child ‘The man bought fish for the child.’ (Benefactive voice) f. Tanging pumansin kay¼Elias si¼Maria Clara. only pfv.av.notice dat¼Elias nom¼Maria Clara ‘Only Maria Clara noticed Elias.’ (Actor voice) g. Kaaalis pa lamang ni¼Pedro nang rec.pfv.leave yet only gen¼Pedro adv dumating ako. pfv.av.arrive 1sg.nom ‘Pedro had just left when I arrived.’ (Voiceless) h. Kapangunguha pa lamang ng¼bata ng¼mga¼mangga. rec.pfv.gather yet only gen¼child gen¼pl¼mango ‘The child has just gathered some/the mangoes.’ (Voiceless)

Four comments are in order about (50a)–(50h). First, (50a,c,d,e) show that the genitive case marker ng/ni encodes not only adnominal possessors, but also transitive actor arguments in non-actor (i.e. patient, benefactive, locative or instrumental) voice constructions and transitive undergoer arguments in non-patient voice constructions. Second, (50c) and (50d) show that nominative-marked undergoer arguments of transitive verbs in non-actor voice constructions are construed as specific (or generic), while genitive-marked undergoer arguments of transitive verbs in actor voice constructions are typically construed as non-specific.32 Third, Tagalog requires undergoer arguments of transitive verbs to receive dative case when they are pronouns or proper names that refer to animate entities (as illustrated in (50f)).33 We may interpret the oblique marking of Elias in (50f) as an instance of differential object marking, under the assumption that sa syncretizes dative and accusative. Finally, Tagalog has a voiceless construction with no overt dependency between the verb inflected for recent-perfective aspect and one of its arguments: the verb bears no voice morphology and neither of its arguments receives nominative marking (Kroeger 1993: 50).34 (50g,h) show that single arguments of intransitive verbs and two major arguments of transitive verbs receive genitive case in recent-perfective constructions.35 This suggests that ng/ni serves as the default case marker of core arguments (cf. Foley and Van Valin 1984: 389). Finally, the accusative-dative syncretism is found in a number of IndoAryan languages (e.g. Hindi, Punjabi, Kashmiri) and Romance languages (e.g. Spanish, Catalan, Southern dialects of Italian). For example, Hindi uses the case clitic -ko to encode animate and/or specific undergoer arguments of transitive verbs and recipient arguments of ditransitive verbs. (51)

Hindi (Mohanan 1994: 59, 80) a. ilaa-ne haar-ko Ila-erg necklace-acc ‘Ila lifted the/*a necklace.’

uthaayaa. _ lift.pfv

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment b. ilaa-ne bacce-ko/*baccaa uthaayaa. _ Ila-erg child-acc/child.nom lift.pfv ‘Ila lifted a/the child.’ c. niinaa-ne bacce-ko kitaab dii. Nina-erg child-dat book.nom give.pfv ‘Nina gave the child a book.’

Likewise, Spanish uses the preposition a to mark not only recipient arguments of ditransitive verbs and goal arguments of intransitive motion verbs, but also undergoer arguments of transitive verbs that are typically both human and specific. (52)

Spanish (Zagona 2001: 13, 14) a. En el mercado vi *(a) los at the market saw-1sg (acc) the ‘At the market (I) saw the neighbours.’ b. En el escritorio vi (*a) los on the desk saw-1sg (acc) the ‘On the desk (I) saw the papers.’ mandó un paquete c. Juan lei Juan dat sent-3sg a package ‘Juan sent a package to José.’

vecinos. neighbours papeles. papers a Joséi. dat José

Since Spanish has no genitive case morpheme in both nouns and pronouns, it exhibits the two-way case-marking system (consisting of nominative and dative) for nouns and local (first and second) person pronouns and the threeway case-marking system (consisting of nominative, dative and accusative) for third-person pronouns.36

‘Dative’ Case in Northern Sámi, Estonian, Finnish and ˇ Karelian Tolmaci Before considering how to derive the variation of case syncretism, a digression is in order about whether or not the case-marking systems in Northern Sámi, Estonian, Finnish and Tolmaˇ ci Karelian fall within the scope of the CH, since they seem to have no dedicated ‘dative’ case morpheme despite having an accusative or genitive/accusative case morpheme. Let us begin with Northern Sámi. 7.5.3

(53)

Northern Sámi (Aikio 2009; Valijärvi and Kahn 2017: 54–56) a. Bussá cˇohkká stuolus. cat.nom sit.prs.3sg chair.loc ‘The cat is sitting on a chair’ (Static location) b. Bussá njuike stuolus láhttái. cat.nom jump.prs.3sg chair.loc floor.ill ‘The cat jumps from the chair onto the floor.’ (Source) c. Nieiddas lea ođđa irgi. girl.loc be.prs.3sg new boyfriend.nom ‘The girl has a new boyfriend.’ (Possessor)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

347

348

WATARU NAKAMURA

d. Mun lean ožžon e-boastta 1sg.nom be.prs.1sg receive.pst.ptcp e-mail.gen Ristenis. Risten.loc ‘I have received an e-mail from Risten.’ (Sender) e. Mun adden beatnagii dávtti. 1sg.nom give.pst.1sg dog.ill bone.gen ‘I gave a bone to the dog.’ (Receiver) f. Máhtes oaivi bávˇ ccˇasta. Matthew.loc head.nom ache.prs.3sg ‘Matthew has a headache.’ (Experiencer) g. Máhtes dállu bulii. Matthew.loc house.nom burn.pst.3sg ‘Matthew’s house burned down on him.’ (Experiencer) h. Máhtes gahˇ cai lássa láhttái. Matthew.loc fall.pst.3sg glass.nom floor.ill ‘Matthew dropped a glass on the floor.’ (Low degree of agentivity)

(53a)–(53e) show that the locative case conflates the distinction between location and source in both the concrete local and possessive domains, while the illative case marks goal arguments of motion verbs and recipient arguments of ditransitive verbs. (53f,g) show that the locative case may mark animate experiencer arguments and demonstrate that the locative case has detached itself from the concrete locative system and has become able to mark a non-spatial (as well as spatial) participant of an event.37 (53h) illustrates a further step taken by the locative case to move out of the physical local domain, since it construes the action taken by Matthew as involuntary or beyond his control (Aikio 2009).38 The fact that the locative case conflates the location/source distinction and covers the dative domain (Seržant 2015) except for recipient arguments and has been grammaticalized to be able to form a paradigmatic contrast with the nominative case and to denote the low degree of agentivity suggests that the locative case in Northern Sámi is abstracted from concrete locative domains to represent pure obliqueness, which the dative case in other languages is supposed to represent.39 Second, the adessive case (one of the external local cases) in Estonian encodes possessive and other non-spatial relations much more frequently than spatial ones (Matsumura 1994); it marks possessor arguments, experiencer arguments, and causee arguments of causative verbs, and ‘logical subjects’ of impersonal verbs, as illustrated by (54a)–(54i).40 (54)

Estonian (Matsumura 1994: 225, 230, 231; Lindström and Vihman 2017: 2, 10) a. Raamat on laua-l. book.nom be.prs.3sg desk-ade ‘The book is on the desk.’ (Location) b. Mu-l on uus auto. 1sg-ade be.prs.3sg new.nom car.nom ‘I have a new car.’ (Possessor)

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

c. Ta-l sur-i laps. 3sg-ade die-pst.3sg child.nom ‘Her child died.’ (Deprived possessor) d. Ta-l on soov kodus olla. 3sg-ade be.prs.3sg wish.nom at.home be.inf ‘She has a wish to be at home.’ (Possessor with an infinitival complement) e. Mu-l on tema-st kahju. 1sg-ade be.prs.3sg s/he-ela sorry ‘I feel sorry for him/her.’ (Experiencer) f. Mu-l/lle on vaja uut arvuti-t. 1sg-ade/all be.prs.3sg need new.prt computer-prt ‘I need a new computer.’ (Modal experience with a nominal complement) g. Mu-l on vaja koju minna. 1sg-ade be.prs.3sg need home.ill go.inf ‘I need to go home.’ (Modal experiencer with an infinitival complement) h. Jaan lase-b ta-l vene keele-s Jaan.nom let-prs.3sg 3sg-ade Russian language-ine vasta-ta. answer-inf ‘Jaan makes/lets her answer in Russian.’ (Causee) i. Ta-l õnnestu-s ülikooli astu-da. 3sg-ade succeed-pst.3sg university.ill enter-inf ‘She succeeded in entering the university.’

The fact that the adessive case not only covers part of the dative domain (i.e. possessor and experiencer arguments) but also causee arguments of transitive verbs and ‘logical subjects’ of impersonal verbs (e.g. (54i)) leads Matsumura (1994) to define the adessive as a functional equivalent to dative (termed ‘adessive-dative’). The allative case also marks some experiencer arguments as well as recipient arguments of ditransitive verbs, as illustrated in (55a,b). (55)

Estonian (Lindström and Vihman 2017: 10) a. Ta and-i-s mu-lle medali 3sg.nom give-pst-3sg 1sg-all medal.gen ‘S/he gave me a medal.’ (Recipient) b. Mu-lle meeldi-b su uus müts. 1sg-all like-3sg 2sg.gen new.nom hat.nom ‘I like your new hat.’ (Experiencer)

However, the crucial difference between the adessive and allative is that experiencer arguments require or prefer the adessive marking when accompanied by an infinitival complement (as illustrated by (54g)) (Lindström and Vihman 2017).41 (54h,i) represents a further departure from the locative use in (54a), since the adessive case in (54h,i) only marks the highest-ranking direct core argument (‘logical subject’) of the embedded infinitival verb. We may take these facts as an indication that the adessive case in Estonian has been grammaticalized to be able to behave as

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

349

350

WATARU NAKAMURA

the ‘pure’ oblique marker abstracted from concrete locative relations (Matsumura 1994). The wide range of functions encoded by the adessive case in Estonian suggests that, like the locative case in Northern Sámi, the adessive case is more prone to grammaticalization than the allative case and hence behaves as the least-marked oblique case morpheme in Estonian. Third, Finnish partitions the dative domain in such a way that possessor and experiencer arguments receive the genitive or adessive (or some other local) case, while recipient arguments of ditransitive verbs receive the allative case, as illustrated by (56a)–(56j).42 (56)

Finnish (Sulkala and Karalainen 1992: 295; Huumo 1996: 81, 82; Karlsson 1999: 96; Kiparsky 2001: 356; Metslang and Erelt 2006: 263; Mazzitelli 2017: 28) a. Poja-lla/*-n ol-i kirje. boy-ade/gen be-pst.3sg letter.nom ‘The boy had a letter.’ (Possessor) b. Poja-lta/*-n katos-i kolikko. boy-abl/gen disappear-pst.3sg coin.nom ‘The boy lost a coin.’ (Deprived possessor) c. Minu-n/lla ol-i nälkä. 1sg-gen/ade be-pst.3sg hunger.nom ‘I was hungry’ (Experiencer) d. Minu-n/lla on ikävä hän-tä. 1sg-gen/ade be.prs.3sg sorry.nom he-prt ‘I feel sorry for him.’ (Experiencer) e. Mies-ten on pakko poistua. man-gen.pl be.prs.3sg obligation.nom leave.inf ‘The men have to leave.’ (Modal experiencer) f. Mauno-n ol-i hauska pääs-tä kotiin. Mauno-gen be-pst.3sg pleasant get-inf home.ill ‘It was nice for Mauno to get home.’ g. Vireni-n onnistui voittaa. Viren-gen succeed.pst.3sg win.inf ‘Viren succeeded in winning.’ h. Anno-i-n sinu-n/Mati-n näh-dä karhu-n. let-pst-1sg 2sg-gen/Matti-gen see-inf bear-gen ‘I let you/Matti see a/the bear.’ i. Pauli kirjoituttaa Harri-lla kirje-en. Pauli.nom write.caus.prs.3sg Harri-ade letter-gen ‘Pauli makes Harri write a letter.’ (Causee) j. Anno-i-n hei-lle karhu-n. give-pst-1sg 3pl-all bear-gen ‘I gave them a bear.’ (Recipient)

The uses of the genitive case in (56c)–(56h) (termed ‘dative-genitive’ in the literature) are distinguished from those uses of it that mark adnominal possessors and undergoer arguments of transitive verbs. Comparing (56a)– (56i) with (54a)–(54i) reveals that the genitive case in predicative possessive

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

ˇ Karelian Table 7.5 Dative domain in Northern Sámi, Estonian, Finnish and Tolmaci

Possessor Experiencer Causee Recipient

Northern Sámi

Estonian

Finnish

Tolmaˇci Karelian

LOC LOC

ADE ADE/ALL ADE ALL

ADE, ABL GEN/ADE GEN/ADE ALL

ADE ADE

ILL

ADE

constructions has been replaced by the external local cases (including the adessive case) and those uses of the genitive case that mark experiencer arguments are competing with the adessive case (as in (56c,d)). (56h,i) show that causees of transitive verbs receive genitive or adessive case, depending on whether they occur in periphrastic or morphological causative constructions.43 What is peculiar about Finnish is that the adessive case has been replacing the ‘dative-genitive’ case, in contrast to Northern Sámi and Estonian, which, respectively, use the locative and adessive case as some kind of ‘pure’ oblique marker.44 Finally, Tolmaˇ ci Karelian, a dialect of Karelian (Finnic), uses the adessive case to mark not only possessor and experiencer arguments but also recipient arguments of ditransitive verbs, due to the loss of the allative case morpheme (which formerly marked goal and recipient arguments) (Oranen 2019). Oranen chooses to retain the traditional term ‘adessive’ (instead of ‘dative’) for etymological and historical reasons, but notes that the adessive case in Tolmaˇ ci Karelian is comparable to the dative case in Russian.45 To summarize, Northern Sámi, Estonian and Finnish use a different case morpheme for recipient arguments of ditransitive verbs than the one used for the rest of the dative domain (in contrast to Tolmaˇ ci Karelian, which uses the adessive case for the entire dative domain), but the fact that the locative case (Northern Sámi), the adessive case (Estonian) and the genitive or adessive case (Finnish) behave, to varying degrees, as a ‘pure’ oblique marker abstracted from concrete locative domains in a paradigmatic contrast with the core cases suggests that the CH holds true even in languages with no dedicated ‘dative’ case morpheme. Table 7.5 summarizes the data from the four languages that are discussed in this subsection.

7.5.4 Deriving Case Syncretism In order to derive the patterns of case syncretism in (42), Nakamura (2011) introduces the distinction between syntactic and morphological cases into RRG and derives their multiple correspondence from an OT-style constraint hierarchy. There are three steps in deriving the patterns of case syncretism in (42). The first step is to recast the case assignment constraints in (31) and another

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

351

352

WATARU NAKAMURA

one that assigns genitive case to adnominal possessors as constraints for syntactic case assignment. The second step is to propose two types of constraints that determine the correspondence between syntactic cases and morphological cases: markedness and faithfulness constraints. The markedness constraints are derived from the CH in (41a). (41)

Case Hierarchy (Silverstein 1976, 1980/1993) a. Propositional/Adnominal: Nom/Abs: Dat1 ⟵ {Acc, Erg} ⟵ Gen

(41a) ranks the propositional and adnominal case morphemes according to their morphological markedness. ‘{Acc, Erg}’ indicates that either of the two case morphemes may be missing and that there is no inherent markedness relation between the two. This means that ‘*Erg’ may outrank ‘*Acc’ or the other way around when both of them are available. From (41a), we may derive a fixed hierarchy of markedness constraints in (57a), which compete against two faithfulness constraints in (57b,c), ‘MAX [Case]’ and ‘IDENT [Case]’ (‘*Acc, *Erg’ in (57)–(59) indicates that ‘*Acc’ and ‘*Erg’ are equally ranked).46 (57)

a. *Gen >> {*Erg, *Acc} >> *Dat b. MAX [Case] c. IDENT [Case] (i.e. IDENT [Gen, Acc, Dat] & IDENT [Gen, Erg, Dat])

‘MAX [Case]’ requires each case feature value in the input (syntactic case) to be realized by some case morpheme (morphological case) in the output, while ‘IDENT [Case]’ requires the case feature values in the input not to be different from the case value in the output. It is important to keep in mind that ‘IDENT [Case]’ amounts to a local conjunction of ‘IDENT [Gen, Acc, Dat]’ and ‘IDENT [Gen, Acc, Dat],’ which means that ‘IDENT [Case]’ is satisfied when a distinct accusative or ergative case morpheme is available in the presence of a distinct genitive case morpheme.47 This is derived from (41a), which requires a distinct accusative and/or ergative case morpheme in the presence of a distinct genitive case morpheme.48 The third step is to derive the patterns of case syncretism in (42) from the interaction of the markedness and faithfulness constraints in (57). Let us begin with the massive syncretism in Kabardian, a Northwest Caucasian language that involves an ergative case alignment. (58) is the constraint ranking for Kabardian. (58)

*Gen >> {*Acc, *Erg} >> MAX [Case] >> IDENT [Case] >> *Dat

(58) ensures that the genitive and ergative case in the input are morphologically realized by the dative case. This explains why actor arguments of transitive verbs, adnominal possessors, and a wide range of oblique arguments and adjuncts bear the same morphological case marking. Tableau 7.4 shows that the constraint ranking in (58) outputs the dative case morpheme in Kabardian when it receives the syntactic ergative or genitive case as input.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Tableau 7.4 Case syncretism in Kabardian

*Gen GEN

*Acc

*!

ERG ☞DAT NOM

ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈ

*Erg

MAX [Case]

IDENT [Case]

*Dat

*

*

*Erg

*Dat

*!

*!

Tableau 7.5 Case syncretism in Old Persian

*Gen GEN

MAX [Case]

IDENT [Case]

*Acc

*!

ACC

*

☞DAT

*

NOM

*!

*!

ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈˈ ˈ

*

Re-ranking of ‘MAX [Case]’ and ‘IDENT [Case]’ as shown in (59) yields the genitive-dative syncretism, illustrated by Old Persian. (59)

*Gen >> MAX [Case] >> IDENT [Case] >> {*Acc, *Erg} >> *Dat

The constraint ranking in (59) yields an accusative or ergative case system in which the dative and genitive cases receive the same morphological marking. Tableau 7.5 shows that the genitive case in the input is mapped to the dative case morpheme in Old Persian. The hierarchized markedness constraints in (57a) require genitive to turn into dative when ‘*Gen’ outranks the two faithful constraints, since dative is the least-marked non-nominative case morpheme. This raises the question of how to handle the genitive-ergative and genitive-accusative syncretism, since no matter how the constraints in (57a)–(57c) may be ranked, their ranking cannot output ergative or accusative when it receives genitive as input. For the purpose of illustration, let us examine the Tagalog case-marking system, which displays both the accusative-dative and genitive-ergative syncretism. The first point to note is that the accusative-dative syncretism is derived from the following constraint ranking. (60)

MAX [Case] >> IDENT [Case] >> *Gen >> *Acc >> *Erg >> *Dat

Given that ‘IDENT [Case]’ requires a distinct accusative and/or ergative case morpheme in the presence of a distinct genitive case morpheme and that faithful realization of accusative incurs a more serious violation than that of ergative in (60), (60) leads accusative to be syncretized with dative, while requiring ergative to be realized faithfully, as shown in Table 7.6.49 The next question to ask is how to derive the genitive-ergative syncretism. It is important to note that no matter how the constraints in (57) may be

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

353

354

WATARU NAKAMURA

Table 7.6 Syntax–morphology mapping in Tagalog Syntax

GEN

Morphology

Gen

ERG

ACC

Erg

DAT

NOM

Dat

Nom

sa/kay

ang/si

----------------------------------------------------------------------------------

Case morpheme

ng/ni

ranked, it is impossible to turn the genitive into the ergative. The reason is, simply, that the dative case always beats the ergative case. This leaves us with no choice but to postulate a case morpheme that underspecifies its case value, so that it may be compatible with the genitive or ergative.50 The above discussion suggests that ng/ni, sa/kay and ang/si constitute a case morpheme with no inherent case value, a dative case morpheme and a nominative case morpheme, respectively. The major piece of evidence for analysing ng/ni as having no inherent case value comes from the fact that they mark not only A, O and S arguments in voiceless constructions but also some instrumental nouns, as illustrated by (61a,b) (Kroeger 1993: 45). (61)

Tagalog (Kroeger 1993: 45) a. Binalutan niya ng¼papel ang¼libro. pfv.wrap.lv 3sg.gen gen¼paper nom¼book ‘He covered the book with paper.’ b. Dadalhin ko ng¼sipit ang¼isda fut.bring.pv 1sg.gen gen¼chopsticks nom¼fish sa¼mesa. dat¼table ‘I’ll take the fish to the table with chopsticks.’

The fact that ng/ni-marked nouns include, but are not restricted to, actor arguments of transitive verbs and adnominal possessors demonstrates that ng/ni lack any inherent case value and that ng/ni functions as the default (elsewhere) case morpheme.51 An analogous account can be given of the genitive-accusative syncretism in Northern Sámi, which allows its genitive/accusative case morpheme to mark complement nouns of both postpositions and prepositions, subject arguments of non-finite clauses, and nouns that represent a standard of comparison in comparative constructions (see Section 7.5.3) in addition to undergoer arguments of transitive verbs and adnominal possessors. (62)

Northern Sámi (Valijärvi and Kahn 2017: a. Oahppit lohket student.nom.pl read.prs.3pl ‘Students read a lot of books.’ b. Beatnaga namma dog.gen.sg name.nom.pl ˇ ‘The dog’s name is Cáhppe.’

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

51, 52) ollu a.lot lea is

girjjiid. book.gen.pl ˇ Cáhppe. ˇ Cáhppe

Case Assignment

c. Máret orru vánhemiid Máret lives parents.gen.pl ‘Máret lives with her parents.’ (¼(47a)) d. Son ođii Máhte 3sg.nom slept Máhtte.gen.sg ‘S/he slept while Máhtte was talking.’ (¼(47b)) e. Biila lea skohtera car.nom.sg is snowmobile.gen.sg ‘The car is newer than the snowmobile.’ (¼(47c))

luhtte. with humadettiin. talk.ger ođđaseabbo. newer

(62a,b) and (47a)–(47c) (repeated above as (62c)–(62e)) show that the genitive/ accusative case morpheme in Northern Sámi behaves rather as the default case morpheme and hence has no inherent case value of its own.52 The genitive-accusative syncretism also obtains in Estonian. The genitive/ accusative case morpheme in Estonian not only marks adnominal possessors and some undergoer arguments of transitive verbs (see Section 7.5.2), but also modifiers of adjectives, complement nouns of most postpositions and some prepositions, and measure adverbials that are singular (Ehala 1994: 180–181; Miljan 2008: 176). The latter two uses are illustrated in (63a)–(63d). (63)

Estonian (Miljan and Cann 2013: 351, 363) a. poisi kõrvale boy.gen.sg beside.all ‘(to) beside the boy’ b. Ta viibis Londonis nädala/nädalaid. 3sg.nom stay.pst.3sg London.ine week.gen.sg/prt.pl ‘S/he stayed in London for a week/for weeks.’ c. Raul ehitas suvila. Raul.nom build.pst.3sg cottage.gen.sg ‘Raul built a cottage.’ (¼(48b)) (completed) d. Raul ehitas suvila-t. Raul.nom build.pst.3sg cottage-prt.sg ‘Raul was building a cottage.’ (¼(48c)) (incomplete)

The case alternation in (63b) shows that measure adverbials that are plural receive partitive case, while those that are singular receive genitive case.53 I follow Miljan and Cann (2013) in defining the genitive/accusative case in Estonian as involving ‘non-subject’ dependency on some head (i.e. a verb, noun or adposition) and deriving its contextual interpretation as found in (63b) and (48b) (repeated as (63c), which is in contrast with (63d)) (quantitative specification and aspectual boundedness) inferentially from its paradigmatic contrast with the partitive case (which Miljan and Cann (2013) analyse as carrying the meaning of ‘part of’; cf. Brattico 2011; Metslang 2014; Norris 2018). Their analysis of the genitive/accusative case morpheme in Estonian as being ‘structural’ amounts to defining it as having no case value. To summarize, this subsection has shown that there are two types of case syncretism, one that involves markedness reduction and is derived from

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

355

356

WATARU NAKAMURA

ranking the constraints in (57) (i.e. the genitive-dative, genitive-ergative/ accusative-dative, and accusative-dative syncretism) and the other that cannot be derived from any ranking of the constraints in (57) (i.e. the genitive-ergative and the genitive-accusative syncretism) and necessitates introduction of the default case morpheme with no case value of its own. It has used the Tagalog, Northern Sámi and Estonian examples to illustrate how the latter type of syncretism is derived by introducing the default case morphemes that are assigned to any NP that would be caseless otherwise.

7.6

Conclusion

This chapter has outlined the RRG theory of case assignment in its original formulation (Van Valin 1991, 1993, 2005, 2009; Van Valin and LaPolla 1997), its OT implementation (Nakamura 1999a, 1999b) and its extension to instrumental case assignment and case syncretism (Nakamura 2011, 2015, 2021). The original RRG case theory is a version of dependent case theory that is dependent not on phrase-structural asymmetry but on the ranking of actor and undergoer, while its OT implementation not only defines nominative and absolutive (conflated as the any-argument case despite their distributional difference), accusative and ergative case with no appeal to the ranking of macroroles, but also derives the typological variation of case systems from re-ranking of the four simple constraints. Assuming either version of the RRG case theory, the chapter has also outlined two attempts to further extend its scope: (i) providing the new macrorole-based definition of instrumental case and deriving the diverse interpretations of instrumental-marked nouns from the interaction of its monosemous meaning with various contextual information rather than from the alleged polysemy of instrumental case (Wierzbicka 1980; Janda 1993; Narrog and Ito 2007) and (ii) deriving the typological variation of case syncretism among dative, accusative, ergative and genitive from the competition of the markedness constraints derived from the Case Hierarchy (Silverstein 1976, 1980/1993) and the faithfulness constraints and the default case morphemes with no case value (Nakamura 2002, 2011).

References Agnihotri, Rama Kant. 2007. Hindi: An Essential Grammar. London: Routledge. Aikio, Ante. 2009. The Structure of North Saami. Course handout, Department of Linguistics, University of Utah. Aldai, Gontzal. 2008. From ergative case marking to semantic case marking: The case of historical Basque. In Mark Donohue and Søren Wichmann (eds.), The Typology of Semantic Alignment, 197–218. Oxford: Oxford University Press.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Aldai, Gontzal. 2009. Is Basque morphologically ergative? Western Basque vs. Eastern Basque. Studies in Language 33: 783–831. Andrews, Avery D. 1982. The representation of case in Modern Icelandic. In Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, 427–503. Cambridge, MA: MIT Press. Andrews, Avery D. 1990. Case structure and control in Modern Icelandic. In Joan Maling and Annie Zaenen (eds.), Syntax and Semantics 24: Modern Icelandic Syntax, 187–234. New York: Academic Press. Baker, Mark. 2015. Case: Its Principles and Its Parameters. Cambridge: Cambridge University Press. Benvenuto, Maria Carmela and Flavia Pompeo. 2015. The Old Persian genitive: A study of a syncretic case. In Anna Krasnowolska and Renata RusekKowalska (eds.), Studies on the Iranian World I: Before Islam, 13–29. Krakow: Jagiellonian University Press. Berlin, Brent and Paul Kay. 1969. Basic Color Terms: Their Universality and Evolution. Berkeley, CA: University of California Press. Blake, Barry J. 2001. Case. Cambridge: Cambridge University Press. Blansitt, Edward L., Jr. 1988. Datives and allatives. In Michael Hammond, Edith A. Moravcsik and Jessica Wirth (eds.), Studies in Syntactic Typology, 173–191. Amsterdam: John Benjamins. Brattico, Pauli. 2011. Case assignment, case concord, and the quantificational case construction. Lingua 121: 1042–1066. Bricker, Victoria. 1981. The source of the ergative split in Yucatec Maya. Journal of Mayan Linguistics 2: 83–127. Caha, Pavel. 2013. Explaining the structure of case paradigms by the mechanisms of Nanosyntax: The Classical Armenian nominal declension. Natural Language and Linguistic Theory 31: 1015–1066. Calabrese, Andreas. 1995. A constraint-based theory of phonological markedness and simplification procedures. Linguistic Inquiry 26: 373–463. Colarusso, John. 1992. A Grammar of the Kabardian Language. Calgary: University of Calgary Press. Comrie, Bernard. 1981. The Languages of the Soviet Union. Cambridge: Cambridge University Press. Coon, Jessica. 2009. Comments on Austronesian nominalism: A Mayan perspective. Theoretical Linguistics 35: 73–93. Coon, Jessica. 2013. Aspects of Split Ergativity. Oxford: Oxford University Press. Cornilescu, Alexandra. 2000. Notes on the interpretation of the prepositional accusative in Romanian. Bucharest Working Papers in Linguistics 2: 91–106. Croft, William. 1991. Syntactic Categories and Grammatical Relations: The Cognitive Organization of Information. Chicago: University of Chicago Press. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67: 547–619.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

357

358

WATARU NAKAMURA

Durie, Mark. 1985. A Grammar of Acehnese on the Basis of a Dialect of North Aceh. Dordrecht: Foris Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11: 365–399. Ehala, Martin. 1994. Russian influence and the change in progress in the Estonian adpositional system. Linguistica Uralica 3: 177–193. Farrell, Patrick. 2009. The preposition with in Role and Reference Grammar. In Lilián Guerrero, Sergio Ibáñez and Valeria A. Belloro( eds.), Studies in Role and Reference Grammar, 179–202. Mexico City: UNAM Press. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Fowler, George. 1996. Oblique passivization in Russian. The Slavic and East European Journal 40: 519–545. Georgopoulos, Carol. 1991. Syntactic Variables: Resumptive Pronouns and A’ Binding in Palauan. Dordrecht: Kluwer Academic Publishers. Gerdts, Donna B. 1988. Object and Absolutive in Halkomelem Salish. New York: Garland. Gönczöl-Davies, Ramona. 2008. Romanian: An Essential Grammar. London: Routledge. Hale, Ken. 1983. Warlpiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1: 5–47. Hill, Virginia. 2013. The direct object marker in Romanian: A historical perspective. Australian Journal of Linguistics 33: 140–151. de Hoop, Helen and Andrej L. Malchukov. 2008. Case-marking strategies. Linguistic Inquiry 39: 565–587. Huumo, Tuomas. 1996. Domain shifts and the grammaticalization of case: A case study of the Finnish adessive. Folia Linguistica Historica 30: 73–96. Inaba, Nobufumi and Rogier Blokland. 2001. Allative, genitive, and partitive: On the dative in Old Finnish. Congressus Nonus Internationalis Fenno-Ugristarum 7, Pars IV: 421–431. Jakobson, Roman. 1936/1984. Beitrage zur allgemeinen Kasuslehre: Gesamtbedeutungen der russischen Kasus. Travaux du Cercle Linguistique de Prague 6: 240–288. Reprinted in Linda R. Waugh and Morris Halle (eds.), 1984. Russian and Slavic Grammar: Studies 1931–1981, 59–103. Berlin: Mouton. Janda, Laura A. 1993. The Geography of Case Semantics: The Czech Dative and the Russian Instrumental. Berlin: Mouton de Gruyter. Johns, Alana. 1992. Deriving ergativity. Linguistic Inquiry 23: 57–87. Jolly, Julia A. 1991. Prepositional Analysis within the Framework of Role and Reference Grammar. New York: Peter Lang. Jolly, Julia A. 1993. Preposition assignment in English. In Robert D. Van Valin, Jr. (ed.), Advances in Role and Reference Grammar, 275–310. Amsterdam: John Benjamins. Josephs, Lewis S. 1975. Palauan Reference Grammar. Honolulu: University Press of Hawaii.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Karlsson, Fred. 1999. Finnish: An Essential Grammar. London: Routledge. Kaufman, Daniel. 2009. Austronesian nominalism and its consequences: A Tagalog case study. Theoretical Linguistics 35: 1–49. Kay, Paul and Luisa Maffi. 1999. Color appearance and the emergence and evolution of basic color lexicons. American Anthropologist 101: 743–760. Kilby, David A. 1977. Deep and Superficial Cases in Russian. Munich: Kubon and Sagner. Kiparsky, Paul. 2001. Structural case in Finnish. Lingua 111: 315–376. Koenig, Jean-Pierre, Gail Mauner, Breton Bienvenue and Kathy Conklin. 2008. What with? The anatomy of a (proto)-role. Journal of Semantics 25: 175–220. Koivisto, Vesa. 2017. The essive in Karelian. In Casper de Groot (ed.), Uralic Essive and the Expression of Impermanent State, 161–184. Amsterdam: John Benjamins. Korhonen, Mikko. 1991[1996]. Remarks on the structure and history of the Uralic case system. Suomalais-Ugrilaisen Seuran Aikakauskirja 83: 163–180. Reprinted in Tapani Salminen (ed.), 1996. Typological and Historical Studies in Language by Mikko Korhonen, 219–242. Helsinki: Suomalais-Ugrilainen Seura. Kroeger, Paul. 1993. Phrase Structure and Grammatical Relations in Tagalog. Stanford, CA: CSLI Publications. Latrouite, Anja. 2011. Voice and Case in Tagalog: The Coding of Prominence and Orientation. Unpublished PhD dissertation, Heinrich Heine University Düsseldorf. Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization. Cambridge: Cambridge University Press. Lindström, Liina and Virve-Anneli Vihman. 2017. Who needs it? Variation in experiencer marking in Estonian ‘need’-constructions. Journal of Linguistics 53: 1–34. McCarthy, John J. 2008. Doing Optimality Theory: Applying Theory to Data. Oxford: Blackwell. Maling, Joan. 1993. Of nominative and accusative: The hierarchical assignment of grammatical case in Finnish. In Anders Holmberg and Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 49–74. Berlin: Mouton de Gruyter. Marantz, Alec. 1991. Case and licensing. Proceedings of the Eastern States Conference on Linguistics 8: 234–253. Mardale, Alexandru. 2008. Microvariation within differential object marking: Data from Romance. Revue Roumaine de Linguistique 53: 449–467. Matsumura, Kazuto. 1994. Is the Estonian adessive really a local case? Journal of Asian and African Studies 46/47: 223–235. Matsumura, Kazuto. 1996. The dative use of the adessive case in Estonian: A corpus-based study. In Kazuto Matsumura and Tooru Hayashi (eds.), The Dative and Related Phenomena, 31–79. Tokyo: Hituzi Shobo. Mazzitelli, Lidia Federica. 2017. Predicative possession in the languages of the Circum-Baltic area. Folia Linguistica 51: 1–60.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

359

360

WATARU NAKAMURA

Metslang, Helena. 2014. Partitive noun phrases in the Estonian core argument system. In Silvia Luraghi and Tuomas Huumo (eds.), Partitive Cases and Related Categories, 177–256. Berlin: Mouton de Gruyter. Metslang, Helle and Matti Erelt. 2006. Estonian clause patterns: From FinnoUgric to Standard Average European. Linguistica Uralica 42: 254–266. Miljan, Merilin. 2008. Grammatical Case in Estonian. Unpublished PhD dissertation, University of Edinburgh. Miljan, Merilin and Ronnie Cann. 2013. Rethinking case marking and case alternation in Estonian. Nordic Journal of Linguistics 36: 333–379. Mohanan, Tara. 1994. Argument Structure in Hindi. Stanford, CA: CSLI Publications. Nakamura, Wataru. 1999a. Functional Optimality Theory: Evidence from split case systems. In Michael Darnell et al. (eds.), Functionalism and Formalism in Linguistics, Vol. 2: Case Studies, 253–276. Amsterdam: John Benjamins. Nakamura, Wataru. 1999b. An Optimality-Theoretic account of the Japanese case system. Studies in Language 23: 607–660. Nakamura, Wataru. 2002. Deriving morphological cases: Markedness considerations. In Reinhard Rapp (ed.), Sprachwissenschaft auf dem Weg in das dritte Jahrtausend. Akten des 34. Linguistischen Kolloquiums in Germersheim 1999, Teil 1, 151–158. Frankfurt am Main: Peter Lang. Nakamura, Wataru. 2008. Fluid transitivity and generalized semantic roles. In Robert D. Van Valin, Jr. (ed.), Investigations of the Syntax–Semantics– Pragmatics Interface, 101–116. Amsterdam: John Benjamins. Nakamura, Wataru. 2011. Case syncretism in typological perspective: An RRG-OT account. In Wataru Nakamura (ed.), New Perspectives in Role and Reference Grammar, 35–63. Newcastle upon Tyne: Cambridge Scholars Publishing. Nakamura, Wataru. 2015. A neo-Jakobsonian account of default oblique cases: Instrumental vs. dative. Paper read at the 2015 International Conference on Role and Reference Grammar, Heinrich Heine University Düsseldorf. Nakamura, Wataru. 2016. A two-tiered theory of case features: The case of the Hindi case(-marking) system. Paper presented in the 49th Annual Meeting of the Societas Linguistica Europaea, University of Naples Federico II. Nakamura, Wataru. 2021. A neo-Jakobsonian account of default oblique cases: Instrumental vs. dative. In Robert D. Van Valin, Jr. (ed.), Challenges in the Analysis of the Syntax–Semantics–Pragmatics Interface. Newcastle upon Tyne: Cambridge Scholars Publishing. Narasimhan, Bhuvana. 1998. A lexical-semantic explanation for ‘quirky’ case marking in Hindi. Studia Linguistica 52: 48–76. Narrog, Heiko and Shinya Ito. 2007. Re-constructing semantic maps: The comitative-instrumental area. Sprachtypologie and Universalienforschung 60: 273–292.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Nelson, Diane. 1998. Grammatical Case Assignment in Finnish. New York: Garland. Nelson, Diane. 2003. Case and adverbials in Inari Saami and Finnish. Nordlyd 31: 708–722. Nevins, Andrew. 2007. The representation of third person and its consequences for person-case effects. Natural Language and Linguistic Theory 25: 273–313. Norris, Mark. 2018. Unmarked case in Estonian nominals. Natural Language and Linguistic Theory 36: 523–562. Oniga, Renato. 2014. Latin: A Linguistic Introduction. Oxford: Oxford University Press. Oranen, Nicklas. 2019. On the use of the adessive case in Tver Karelian. Lähivõrdlusi 29: 204–227. Preminger, Omer. 2012. The absence of an implicit object in unergatives: New and old evidence from Basque. Lingua 122: 278–288. Prince, Alan and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction in Generative Grammar. Oxford: Basil Blackwell. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Rezac, Milan, Pablo Albizu and Ricardo Etxepare. 2014. The structural ergative of Basque and the theory of case. Natural Language and Linguistic Theory 32: 1273–1330. Rissman, Lilia. 2010. Instrumental with, locatum with, and the argument/ adjunct distinction. LSA Annual Meeting Extended Abstracts 2010. Sabbagh, Joseph. 2016 Specificity and objecthood in Tagalog. Journal of Linguistics 52: 639–688. Sadakane, Kumi and Masatoshi Koizumi. 1995. On the nature of the ‘dative’ particle ni in Japanese. Linguistics 33: 5–33. Schultze-Berndt, Eva and Nikolaus P. Himmelmann. 2004. Depictive secondary predicates in cross-linguistic perspective. Linguistic Typology 8: 59–131. Seržant, Ilja A. 2015. Dative experiencer constructions as a Circum-Baltic isogloss. In Peter Arkadiev, Axel Holvoet and Björn Wiemer (eds.), Contemporary Approaches to Baltic Linguistics, 325–348. Berlin: Mouton de Gruyter. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon (ed.), Grammatical Categories in Australian Languages, 112–171. Canberra: Australian Institute of Aboriginal Studies. Silverstein, Michael. 1980/1993. Of nominatives and datives: Universal Grammar from the bottom up. Unpublished manuscript. Reprinted in Robert D. Van Valin, Jr. (ed.), 1993. Advances in Role and Reference Grammar, 465–498. Amsterdam: John Benjamins. Smith, Henry. 1996. Restrictiveness in Case Theory. Cambridge: Cambridge University Press. Smolensky, Paul. 1995. On the Internal Structure of the Constraint Component of UG. Presented at University of California, Los Angeles.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

361

362

WATARU NAKAMURA

Starosta, Stanley, Andrew K. Pawley and Lawrence A. Reid. 1982. The evolution of focus in Austronesian. In Amram Halim, Lois Carrington and S. A. Wurm (eds.), Papers from the Third International Conference on Austronesian Linguistics, Vol. 2: Tracking the Travelers, 145–170. Stolz, Thomas, Cornelia Stroh and Aina Urdze. 2006. On Comitatives and Related Categories: A Typological Study with Special Focus on Languages of Europe. Berlin: Mouton de Gruyter. Sulkala, Helena and Merja Karalainen. 1992. Finnish. London: Routledge. Tamm, Anne. 2007. Perfectivity, telicity, and Estonian verbs. Nordic Journal of Linguistics 30: 229–255. Tomi´ c, Olga Mišeska. 2004. The Balkan Sprachbund properties: An introduction. In Olga Mišeska Tomi´ c (ed.), Balkan Syntax and Semantics, 1–55. Amsterdam: John Benjamins. Trask, Robert L. 1979. On the origins of ergativity. In Frans Plank (ed.), Ergativity: Toward a Theory of Grammatical Relations, 385–404. London: Academic Press. Tsunoda, Tasaku. 1981. The Djaru Language of Kimberley, Western Australia. Canberra: Pacific Linguistics. Valijärvi, Riita-Liisa and Lily Kahn. 2017. North Sámi: An Essential Grammar. London: Routledge. Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and grammatical relations. Natural Language and Linguistic Theory 9: 145–194. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Robert D. Van Valin, Jr. (ed.), Advances in Role and Reference Grammar, 1–164. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2009. Case in role and Reference Grammar. In Andrej Malchukov and Andrew Spencer (eds.), The Oxford Handbook of Case, 102–120. Oxford: Oxford University Press. Van Valin, Robert D., Jr. 2012. Lexical representation, co-composition, and linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon, Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in Generative Lexicon Theory, 67–107. Dordrecht: Springer. Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory. In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan Timberlake (eds.), Language Typology and Historical Contingency, 91–123. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplishments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 71–93. Freiburg: Albert-LudwigsUniversität, Universitätsbibliothek. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning, and Function. Cambridge: Cambridge University Press.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Wierzbicka, Anna. 1980. The Case for Surface Case. Ann Arbor, MI: Karoma Publishers. Wiltschko, Martina. 2006. On ‘ergativity’ in Halkomelem Salish. In Alana Johns, Diane Massam and Juvenal Ndayiragije (eds.), Ergativity, 197–227. Dordrecht: Springer. Yip, Moira, Joan Maling and Ray Jackendoff. 1987. Case in tiers. Language 63: 217–250. Zagona, Karen. 2001. The Syntax of Spanish. Cambridge: Cambridge University Press.

Notes 1 See Levin and Rappaport Hovav (2005: 51–68) for a critical comparison of macroroles with proto-roles (Dowty 1991). 2 The underlying assumption is that the relative degree of affectedness determines which non-actor argument serves as undergoer. See Van Valin (2005: 109–115) for further discussion. 3 Van Valin (1991, 2005) attributes the idea of dative being the default case to Silverstein (1980/1993). 4 For example, Japanese allows source arguments of ditransitive verbs such as morau ‘receive’ and kariru ‘borrow’ to be marked by the dative (as well as the ablative) case particle. 5 See Jolly (1991: 115–119) for an LS-based account of some other nonlocative uses of from (e.g. John suffered from arthritis). 6 German behaves like Icelandic with respect to case preservation under passivization, while Japanese allows no such case preservation. See Section 7.4.1 for an OT reformulation of the case assignment rules in (9) and how it accounts for the contrast between Icelandic/German and Japanese with respect to case preservation. 7 Predicate adjectives that modify covert controllee arguments of dependent cores in Icelandic control constructions may receive nominative case or the case of the controller argument in the matrix core (Andrews 1982: 450–456). This fact shows that the controllee argument may receive nominative case and agree in case with predicate adjectives that modify it. We may take this agreement fact as another piece of evidence that the DCA for Icelandic is the core. 8 Matrix-coding constructions are also attested in ergative and active languages. For example, Basque requires actor arguments of transitive verbs in non-finite gerundive complements of perception verbs to receive absolutive case (Rezac et al. 2014).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

363

364

WATARU NAKAMURA

9 The auxiliary verb in (24b) (d-u-ø) contains the absolutive affix d-, which refers to no clausal argument. The verb dantzatu is a denominal verb that incorporates dantza ‘dance’ and the absolutive affix refers to this incorporated noun. See Preminger (2012) for related discussion. 10 Aldai’s argument is based on the case marking of S arguments of lexically simple unergative verbs. See Aldai (2009: 797–800) for discussion of compound unergative predicates formed with egin ‘do, make’ and a few other light verbs. The fact that Basque has an ergative dialect in addition to an active dialect justifies analysing the actor and undergoer markers in the Western dialect as ergative and absolutive, respectively. 11 See Farrell (2009) for an exception to the ICR. 12 It is questionable whether (28) is in a subset relation with (9c). Rissman (2010) compares the locatum with-phrases with the instrumental withphrases and shows that the former are syntactic arguments, while the latter are adjuncts. See also Koenig et al. (2008) for related discussion of the semantic and syntactic status of implement with-phrases. 13 Van Valin and LaPolla (1997: 381–382) attempt to extend the ICR to the uses of with that mark manner adverbials in English (e.g. with enthusiasm, with difficulty). 14 See de Hoop and Malchukov (2008) for an alternative OT-based case theory. 15 Russian allows some instrumental and genitive-marked arguments (but not any dative-marked arguments) to receive nominative case under passivization (Fowler 1996). One may be tempted to interpret this fact as an indication that Russian also belongs to (33b) (under the assumption that non-subject genitive/instrumental-marked arguments are not undergoers), but this move would leave it unexplained why dativemarked arguments may not undergo passivization in Russian. 16 It remains an open question whether or not there are any other nonaccusative case systems that do not allow case preservation under passivization. 17 An asterisk mark (*) indicates a violation incurred by each candidate for a given constraint. The asterisk followed by the exclamation mark indicates a fatal violation and the candidate that incurs that violation is eliminated from further consideration. The shaded cells indicate that they exert no influence on the selection of the optimal output(s). 18 What is termed ‘instrumental’ in (35b) includes the ablative case in Latin, which combines three formerly distinct cases, instrumental, locative and ablative proper (Oniga 2014: 254). 19 See Van Valin (2005: 111–112) for more relevant data from Croatian and Dyirbal. 20 The instrumental adposition/case has no LS of its own and therefore has no argument slot for a noun it marks (Van Valin and LaPolla 1997; Van Valin 2005). 21 The notion of QS was incorporated into RRG by Van Valin and LaPolla (1997).

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

22 The fact that the possessive relation between Petr and his body part is permanent and holds outside the event of nodding indicates that the main verb kivnutʹ ‘nod’ and covert possessive predicate do not form a complex nucleus. 23 (34c) is comparable to secondary predicate constructions, which describe the (resulting) state of a clausal participant and involve a nonsubordinate core juncture (Schultze-Berndt and Himmelmann 2004). They share the same juncture–nexus type, but the difference lies in the fact that the possessum noun in (34c) specifies which part of the possessor is directly involved in the event instead of representing a temporary state of the possessor. 24 We may compare the two types of dative in (41a,b) to the two types of the dative case particle ni in Japanese (Sadakane and Koizumi 1995): ni as a (non-predicative) case marker and ni as a predicative postposition. See also Blansitt (1988), which may be taken as an attempt to elaborate (41b) by establishing implicational relations among dative, allative and locative. 25 Silverstein (1980/1993) draws an explicit analogy between the development of case-marking systems and that of the colour-term systems (Berlin and Kay 1969; Kay and Maffi 1999). 26 The genitive case clitic in Hindi may case-mark subjects of a small number of intransitive verbs (Mohanan 1994) despite its adjective-like declension (it agrees in gender, number and case with a possessum noun it occurs with) (Agnihotri 2007: 187–189). The chimeric status of this genitive clitic is indicated in (42i) by putting ‘Gen’ in parentheses. 27 The morpheme a in (44) is an NP marker and is not glossed (Georgopoulos 1991: 237). 28 Personal pronouns preserve a four-way case distinction (i.e. nominative, accusative, dative and genitive) (Gönczöl-Davies 2008). We can derive the lack of accusative marking on some transitive undergoer arguments from the definiteness/animacy-based split accusativity. 29 Third-person agreement markers in matrix indicative clauses alone operate on an ergative-absolutive basis (Wiltschko 2006). 30 For simplicity, in this paper, I gloss the genitive/accusative case morphemes in the examples taken from Northern Sámi, Estonian and Finnish as ‘genitive.’ 31 See Tamm (2007) for a detailed semantic account of the alternation between the partitive and accusative/nominative case on undergoer arguments of transitive verbs. See also Miljan and Cann (2013: 365–372) for how the notion of aspectual boundedness arises. 32 The genitive case marker turns out to be neutral to the distinction between specific and non-specific, since it may mark specific common nouns and pronouns in voiceless constructions, as illustrated by (50g,h). 33 See Latrouite (2011: Ch. 5) and Sabbagh (2016: 656–660) for further discussion of the clause-internal distribution of ng/ni and the case alternation between ng/ni and sa/kay.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

365

366

WATARU NAKAMURA

34 Kroeger (1993: 50) notes that gerundive constructions also bear no voice marking on the verb stem and allow only what he terms ‘default case assignment’. 35 Recent-perfective constructions are a type of impersonal (or ‘pivotless’) construction. See Kroeger (1993: 48–51) for a list of impersonal constructions in Tagalog. 36 An ergative-dative syncretism is hardly attested despite the fact that it is predicted to be possible by the CH. The cross-linguistic paucity of the ergative-dative syncretism requires a principled explanation. Nakamura (2011) attributes its near unavailability to an independent semantic constraint to the effect that the farther one of the co-arguments of a verb is away from the other one on the causal chain, the less likely they are to receive the same morphological marking. The essential idea is that the CH works in tandem with this semantic constraint (adapted from Croft (1991: Ch. 5)) to constrain the possible patterns of case syncretism. 37 Aikio (2009) notes that Máhtes in (53g) does not serve as an adnominal possessor. 38 Some verbs subcategorize for a locative-marked argument that resists any immediate spatial interpretation (e.g. heaitit ‘stop’, dolkat ‘get sick of’, beroštit ‘care about’). The same holds for verbs with an illative-marked argument (Valijärvi and Kahn 2017: 54, 56). 39 (53h) is a crucial example, since it shows that the locative case can be treated on a par with the nominative case for the purpose of indicating the relative degree of agentivity. 40 The external local cases (allative, adessive and ablative) have a much stronger tendency to mark non-spatial relations than the corresponding internal local cases (illative, inessive and elative) in Estonian. As a case in point, Matsumura (1994) compares the adessive case with the inessive case in terms of the frequencies of three semantic types, (i) place, (ii) time, and (iii) possessor, in two text corpora and finds that the third type accounts for more than 40% of all the adessive nouns, while the first type that refers to a physical space accounts for less than 8%. In contrast, the first type accounts for almost 80% of all the inessive nouns, while no inessive noun of the third type (possessor) is found. 41 See Matsumura (1996: 40–51) and Lindström and Vihman (2017: 9–11) for more relevant examples and further discussion of the relation between the adessive and allative in Estonian. 42 The allative, genitive and partitive cases mark recipient arguments of ditransitive verbs in Old Finnish (Inaba and Blockland 2001). 43 The fact that the genitive marking of sinu-n/Mati-n in (56h) is retained under passivization and negation (Kiparsky 2001: 356–357) indicates that they are non-macrorole core arguments. 44 Korhonen (1991/1996) notes that the adnominal uses of the genitive case were historically derived from the Proto-Uralic -n lative case by way of its dative (‘dative-genitive’) uses. See also Huumo (1996: 74–75, 83–86) for

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

Case Assignment

45

46

47

48 49

50

51 52

53

the reason the functions borne by the -n lative case have been taken over by the external local cases. Oranen (2019: 211–219) lists the functions borne by the adessive case in Tolmaˇ ci Karelian and compares its adessive case with the Finnish and Estonian counterparts. See also Koivisto (2017: 162–164) for other syncretic patterns (the adessive-ablative syncretism and allative-adessiveablative syncretism) displayed by a few other dialects of Karelian. (57a)–(57c) make no reference to the nominative case value, but it is not representationally underspecified. I follow the spirit of Calabrese (1995) in parameterizing rules/constraints with respect to whether they may refer to (i) all, (ii) contrastive, or (iii) marked feature values and assume that all constraints related to case assignment may refer to contrastive case values. It follows, then, that the nominative case value remains invisible to constraints when it is not contrastive in the case-marking system of a particular language (cf. Nevins 2007: 285–287). Unlike ‘IDENT [Case]’, ‘MAX [Case]’ involves no local conjunction (as shown in (57c)), since it is not sensitive to the distinctness/identity of case morphemes and only ensures that a syntactic non-nominative case is realized by any non-nominative case morpheme. See McCarthy (2008: 214–219) for a formulation and discussion of local conjunction (due originally to Smolensky (1995)). The distinction between ng and ni is that the former marks common nouns, while the latter marks personal names. The same contrast obtains in sa/kay and ang/si. The case-marking system in Hindi is derived from the same constraint ranking as (60) (Nakamura 2016). The difference between Tagalog and Hindi comes down to the availability of a case morpheme that neutralizes the ergative-genitive distinction. This formulation accords well with Foley and Van Valin’s (1984: 389) definition of ng/ni as the default case marker for non-pivot core arguments. Like Northern Sámi, Estonian uses both postpositions and prepositions, but the latter are a relatively new development and are limited in number. Specifically, the vast majority of postpositions in Estonian assign genitive case to their complement nouns. In contrast, the case assignment of prepositions is not dominated by any single case, but the genitive case outnumbers the other cases (Ehala 1994: 180–181). Measure adverbials as illustrated in (63b) exhibit the same case-marking pattern as undergoer arguments of transitive verbs. See Maling (1993), Nelson (1998, 2003) and Kiparsky (2001) for further illustration and discussion of case-marked adverbials (like the one in (63b)) in the Finnic and Sámi languages. See Van Valin (2018) for how to represent analogous measure adverbials in terms of LS.

https://doi.org/10.1017/9781316418086.010 Published online by Cambridge University Press

367

8 Morphology in RRG The Layered Structure of the Word, Inflection and Derivation Francisco J. Cortés-Rodríguez

8.1

Introduction

This chapter offers an overview of the place of inflectional and derivational morphology in Role and Reference Grammar (RRG). A number of researchers have devoted their efforts to outlining the key topics of a functional theory of morphology within this model. Martín Arista (2008: 126) summarizes the different approaches that would give support to such a theory. He suggests that this would draw on structural-functional models such as those of Dik (1997), Van Valin and LaPolla (1997) and Van Valin (2005). Specific relevant elements in the determination of the theory are the layering proposals for grammatical structures such as the clause (Foley and Van Valin 1984; Hengeveld 1989) or the phrase (Rijkhoff 2002) and the manner in which they can be extended to the word domain. Word Syntax approaches – stemming from Marchand (1969) and continuing in syntacticist theories of morphology such as Selkirk (1982) and Baker (2003) – are also influential. To these, it is necessary to add the studies that view morphology from a lexical-semantic perspective, as do Martín Mingorance (1998 [1985]), Lieber (2004) and Štekauer (2005a, 2005b), as they are crucial for the approach to word formation put forward in this chapter. Nevertheless, despite its centrality within a theory of grammar, there are still some issues to address for a fully developed morphological theory within RRG. In order to provide a description of the state of morphological research within RRG, we will follow a widely accepted distinction between two approaches in morphological studies, analytic and synthetic, as described by Aronoff and Fudeman (2011: 12). The former deals with the description of the internal constituency of words: analytical descriptions must state which units are considered to be the components of morphologically complex structures; units such as morphemes, lexemes, affixes, etc. will come into play here.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

The synthetic approach has to do with the mechanisms that are invoked by morphological models to account for the production of morphologically complex words. From a constructivist perspective, analytical morphology takes the perspective of the decoder, whereas when a speaker produces a new complex word they make use of the synthetic machinery. From the point of view of theory building and description, the analytical approach must come first. Only after a morphological theory has established what the building blocks of words are will it be in a position to give an account of the manner in which such pieces are put to work to create complex lexical units. In line with this, our description of morphological theory within RRG will also start with an overview of the proposals offered within the model to provide a sound description of the internal constituents of words. This is the main goal of Section 8.2 in this chapter. Based on the programmatic contribution by Everett (2002) and the developments into a fully articulated model in Martín Arista (2008, 2009, 2011), we will spell out the details of the layered structure of the word (henceforth LSW). Some alternative views on specific aspects of Martín Arista’s proposal (such as those expressed in Cortés-Rodríguez and Sosa 2012; Nolan 2011; Diedrichsen 2011) will also be considered. The synthetic aspects of morphological processes are considered in Sections 8.3 and 8.4. From a synthetic viewpoint, attention is paid to issues such as whether there are specific morphological rules to generate complex words or, on the contrary, they are the result of other types of processes, as are syntactic or phonological operations. Consequently, in synthetic descriptions, one must take into account whether morphology is a separate component of grammar or just a set of phenomena assimilated to other modules. The functional orientation of RRG has led researchers interested in this area to place the emphasis on the external motivation of morphological phenomena. In this regard, there is widespread consensus among RRG morphologists that lexis, syntax, semantics and pragmatics motivate morphology.1 It is at this point, however, that inflectional and derivational morphology take different paths, inflectional morphology aligning itself with syntactic processes while word-formation processes are treated as essentially lexicalsemantic phenomena. Following Stump’s (2001) four-way classification of morphological theories, inflectional morphology in RRG follows an inferential-realizational approach. In opposition to lexical theories of inflection, in which inflectional affixes have lexical entries specifying their morphosyntactic properties, an inferential theory associates morphosyntactic properties with morphological rules, which relate a given inflected wordform to its root. Furthermore, instead of being incremental, the RRG approach to inflection is realizational: the association of a given root or lexeme (e.g. work) to a set of properties like ‘3rd person singular’, ‘present tense’ and ‘indicative mood’ licenses the attachment of the affix -s (cf. Stump 2001: 1–3).2 Derivational morphology lies outside this typology, as it is, in

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

369

370

FRANCISCO J. CORTÉS-RODRÍGUEZ

essence, a lexical phenomenon involving lexical units and the properties encoded in their corresponding lexical entries. In accordance with the distinction between inflectional and derivational morphology, Section 8.3 will be devoted to the manner in which inflectional phenomena are dealt with within the overall apparatus of RRG, while in Section 8.4 we will turn to word-formation processes. After the description of how inflection is encoded in the LSW, Section 8.3 offers an explanatory account of the factors that motivate inflectional marking. The functional orientation of RRG will play a significant role in this account. A description based on this functional standpoint presupposes a view of morphology distributed throughout the different components of the grammatical model. Additionally, the typological commitment of RRG requires paying close attention to the role of inflectional processes not only in dependent-marking languages, but also in head-marking languages, since the interface between inflectional morphology and syntax is much tighter in head-marking than in dependent-marking structures. Section 8.4 will shift the focus to the domain of morphology as a lexical phenomenon. Proposals explaining word formation as a lexicological process involving the interaction of lexical semantics and morphology within RRG will be discussed. The approach to derivational morphology within RRG can be said to be markedly motivated by semantics. Word formation is tightly connected to lexical semantics, and lexical-semantic representations, which are at the core of RRG, become crucial for the development of a theory of derivation. The inclusion of Qualia Theory in semantic representations provides RRG with a very robust system to account for word-formation phenomena, in contrast to other approaches. Some conclusions are drawn in the final section.

8.2

Foundations of RRG Morphology: The Layered Structure of the Word

This section will establish the set of common assumptions that support the different contributions for the development of morphology within RRG from an analytical viewpoint. The description will proceed as follows. Section 8.2.1 lists the basic underpinnings of the morphological theory outlined in this chapter. Section 8.2.2 outlines the layered structure of the word, the analytical tool responsible for grammatical analyses within the word domain in RRG. Finally, Section 8.2.3 sketches the state of research on operators in the LSW.

8.2.1 Theoretical Underpinnings The goal of this section is to lay the foundations of our functional theory of morphology. The notions upon which we will base the description of the internal makeup of complex units are the following:3

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

(a) There are two types of lexical morphemes (or lexemes): free and bound lexemes. Free lexemes belong to the major lexical categories, whereas bound lexemes are derivational affixes. Both free and bound lexemes will be semantically represented by means of their corresponding lexical entry. Both are also grouped within the lexicon into lexical classes defined by their similarity of meaning. The difference between these types of lexeme lies only in their distributional behaviour.4 (b) It follows from what is outlined in (a) that affixal derivation and compounding are essentially the same type of phenomenon. Despite this, in languages like Spanish or English, there are some differences between these two types of process. Whereas the grammatical makeup of affixal formations is quite predictable, since derivative affixes usually specify both the type of base they are attached to and the type of lexical category that outputs from its combination, ‘compounding is only barely syntactic’ (Jackendoff 2009: 14), since the elements of a compound are rather opaque, in that they do not provide this kind of specification. This is especially clear in Spanish compounds, which can have as their nucleus either the rightmost (as in puti-club ‘whorehouse’, lit. ‘hooker-club’) or the leftmost constituent (as in hombre-anuncio ‘sandwich-board man’, lit. ‘manadvertisement’). (c) Inflectional affixes are treated differently, as they are considered to be morphemes with no lexical status (i.e. they are not lexemes) and therefore they are not stored in the lexicon. Apart from Nolan (2011: 74–75), for whom inflectional affixes are allocated to a ‘morpheme store’, other researchers working in RRG morphology (Everett 2002; Martín Arista 2008; Boutin 2011; Van Valin 2013) argue for an inferential realizational approach to inflection (cf. Spencer 1998, 2004; Stump 2001): phonological operations/rules will be responsible for providing the morphosyntactic representation of a given inflected form.5 (d) Whereas derivation and compounding are unified, inflection and word formation are regarded as disparate phenomena from a functional point of view, since they serve different purposes within the overall structure of our grammatical model. Despite this, they must be treated together at some point, as they contribute to the formal structure of the word, which is the maximal unit in morphology. The main theoretical construct that handles the overall internal makeup of the word, and which consequently accounts for the structural distribution of inflectional and derivational morphology, is the layered structure of the word (LSW). Parallel to clauses and phrases, the internal structure of words is based on layering. The first outline of the LSW can be found in Everett (2002), and it becomes a fully articulated model in Martín Arista’s studies (2008, 2009, 2011).

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

371

372

FRANCISCO J. CORTÉS-RODRÍGUEZ

8.2.2 The Layered Structure of the Word In a fashion similar to the layering analysis for clauses and phrases in RRG, the LSW distinguishes three layers: the nucleus, the core and the word. Figure 8.1 represents the LSW of a predicate of the category α as described by Martín Arista (2009: 91). WORD CORE NUCLEUS PREDICATE NUC OPERATORS CORE OPERATORS WORD OPERATORS

NUCLEUS CORE WORD

Figure 8.1 The layered structure of the word (Martín Arista 2009: 91)

The LSW consists of the constituent projection for lexical constituents (free and bound lexemes) and the operator projection for inflectional exponents.6 Figure 8.2 shows the LSW of the Present-Day English genitive form friends’, with the operators of number and case encoded in the operator projection. A typical feature of the LSW of a simplex word is that it shows no functions in the constituent projection. Compare this with Figure 8.3, which analyses the Old English complex word bell-ring-estre (bell-ring-er.nom. sg ‘bell ringer’; Martín Arista 2008: 129). In this case, the lexical constituents bell and -estre occupy the syntactic positions of first and second argument in the constituent projection.7 Because of this, the LSW of compound and derived words takes the additional layer of the Complex word, which has scope over the word layer (Martín Arista 2011: 397). WORDN COREN NUCN friends' NUC CORE

Number

WORD Figure 8.2 The LSW of friends’

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Case

Morphology in RRG

COMPLEX WORDN COREN ARGN

NUCLEUSV

ARGN

WORDN

WORDN

COREN

COREN

NUCN

NUCN

bell

ring

estre.NOM.SG

NUC CORE

COMPLEX WORD

Number

Case

Figure 8.3 The LSW of the OE complex word bellringestre ‘bell ringer’ (Martín Arista 2008: 129)

Figure 8.4 analyses the constituent projection of the recursive formation takal-takal mananyi (‘give out a repeated knocking’) from PitjantjatjaraYankunytjatjara (Australia) (adapted from Martín Arista 2009: 110). As is the case in the layered structure of the clause (LSC) and the layered structure of the phrase (LSP), operators from outer layers have scope over the inner layers. There is a further logical basis for considering inflectional phenomena as operators in the LSW, since this is consonant with the way in which they are analysed in the other syntactic domains in RRG, the LSC and the LSP.

8.2.3 Operators in the LSW Even though there is as yet no detailed description of the types of operators that modify each of the different layers within the LSW, some general guidelines can be offered. Drawing on data from Old and Present-Day English, Martín Arista (2009: 93) takes the morphological status of the bases as the central criterion for the distribution of operators: affixes that cannot combine with derived bases encode nuclear lexical operators, whereas those that attach to derived bases represent core lexical operators. Inflection is also added to this as a ‘G[rammatical]-WORDop[erator]’ (Martín Arista 2008: 135). Thus, inflectional categories will at least appear distributed between

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

373

374

FRANCISCO J. CORTÉS-RODRÍGUEZ

COMPLEX WORDV COREV ARGN

NUCLEUSV

WORDN

PREDICATEV

COREN

NUCN PREDN COMPLEX WORDN

takal-takal

mananyi

Figure 8.4 The LSW of a recursive complex word (adapted from Martín Arista 2009: 110)

the core and the word lexical operators, as we see above in Figure 8.2 for friends’. From the analysis offered in Figure 8.3 it seems clear that among the nominal inflectional features, number (and probably gender) are core operators, whereas case has scope over the word node.8 Nuclear operators, on the other hand, will encode those features that are not subject to percolation into higher structures (i.e. syntax-blind features). Thus, good candidates to be classified as nuclear operators are aspectualizers in languages where lexical aspectual features are regularly marked by morphological means, as is typical of Slavic languages.9 Other possible nuclear operators are some types of lexical distributive processes in Amele (Papua New Guinea) which involve a reduplication process, as in the following clauses (data from Roberts 2015: 19–20):10 (1)

Tob-i ton-i ascend-pred descend-pred ‘They ascend and descend.’

egi-na. 3pl.sbj-prs

(2)

L-i l-i h-u go-pred go-pred come-pred ‘He comes and goes.’

h-u come-pred

ena. 3pl.sbj-prs

Roberts (2015: 19) explains that this type of reduplication takes place within the verbal nucleus, in contrast with other types of reduplication processes

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

that take place outside the nucleus in higher layers, as shown by the different distribution of grammatical markers.11 (3)

Mel f-ec-eb f-ec-eb boy see-ds-3sg.su see-ds-3sg.su ‘The boys are looking at each other.’

egi-na. 3pl.sbj-prs

In (3) each verb has a different subject (DS) marker and 3SG subject agreement. This is followed by verb inflection which includes 3PL agreement with the matrix clause subject, mel ‘boys’, as well as present tense. Table 8.1 shows a classification of the operators mentioned in the previous description; it must be stressed that it is not an exhaustive classification and that this is an issue that still needs further research. Table 8.1 Operators in the LSW (a partial classification) Layer

Operator

Bases

Examples

Nucleus

Aspectualizers (verb) Distributives (verb) Number (Noun) Gender (Noun) Case (Noun)

Underived stems/words

na-pisat (Russ. ‘write up’) to- teran (OE ‘tear to pieces’) book-s vecin-a (Sp. neighbour-fem) John’s

Core Word

Derived Derived

Diedrichsen (2011) draws attention to certain features of inflection which must be considered. She mentions that every lexical category has its own distinct set of inflectional categories. For instance, in German, nouns are inflected for number, case and gender, whereas verbs inflect for tense, modality and also number and person. Furthermore, the typology of inflectional categories and their distribution among word classes show great variability across languages. For instance, whereas definiteness is a grammatical category realized only on the determiner in West Germanic languages (Diedrichsen 2011: 5), it is an inflectional category marked on the noun in Scandinavian languages (Nübling 2008: 46, quoted also in Diedrichsen 2011: 5). Other languages are much less rigid concerning the distribution of inflectional features across word classes, as illustrated by the following examples from Bella Coola (Western Canada; data from Pavey 2010: 86–87). (4)

nuyamł-Ø ti-?immlllk¯ı-tx sing-3sg.sbj prox-boy-prox ‘the boy is singing’

(5)

?immllk¯ı-Ø ti-nus?¯ ulχ-tx boy-3sg.sbj prox-thief-prox ‘the thief is a boy’

(6)

sx-Ø ti-nus?¯ ulχ-tx bad-3sg.sbj prox-thief-prox ‘the thief is bad’

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

375

376

FRANCISCO J. CORTÉS-RODRÍGUEZ

(7)

?imlk-Ø ti-nuyamł-tx man-3sg.sbj prox-sing-prox ‘the one who is singing is a man’

(8)

‘ nus?¯ ulχ-Ø ti-gs-tx thief-3sg.sbj prox-ill-prox ‘The one who is ill is a thief’

In Bella Coola, the attachment of person, number and case affixes, on the one hand, and proximal morphemes, on the other, is flexible with regard to class form, and is instead conditioned by the function of words in a given sentence (Van Valin 2008). This type of phenomenon seems to run counter to the establishment of a typologically valid arrangement of inflectional operators in the LSW (see Aronoff and Fudeman 2011: 201–202 for further arguments on this); this is an issue which calls for further research.

8.3

The Functional Motivation of Inflectional Morphology: Morphological and Syntactic Inflection

One of the main tenets of a functional model of language is the teleological orientation which holds for every language phenomenon. RRG is strongly committed to this functional prerequisite and always seeks external motivation for grammatical phenomena.12 From a synthetic perspective, this implies that the word-internal elements described in the previous section as components of the LSW need an external motivation, which is tantamount to stating that morphological processes are visible to syntax, semantics and pragmatics (Martín Arista 2011: 395).13 In essence, this section seeks to provide an answer to the following issue posited in Van Valin (2013: 96): How is it that syntax can target elements inside a word? In order to provide an answer to this question, it is necessary to take into account that the degree and modes of interaction between the LSW and higher structures in the model is different in dependent-marking and headmarking languages. Head-marked structures are especially challenging since a single phonological word can at the same time be a clause, as shown in the following example from St’át’imcets (Salishan, Canada) (Pavey 2010: 80; from Roberts 1999: 278–279): (9)

tsún-tsi-lhkan tell-2sg.obj-1sg.sbj ‘I told you’

From the perspective of a morphological theory in RRG, it is essential to explain the relationship that holds between the internal structure of the word and the internal structure of its corresponding clause. In this section we aim to illustrate the interaction of the morphosyntactic features encoded in the LSW with those in the layered structures of phrases and clauses in both types of languages. As will be described, feature

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

percolation across the LSW and the other grammatical domains is significantly different if we compare head- vs. dependent-marking structures. Whereas in the latter type of language, percolation is restricted to operators across different structures (see Section 8.3.1), in head-marking structures, morphological features percolate up to the constituent projection in the LSC (cf. Section 8.3.2).

8.3.1 Dependent-Marking Inflectional Morphology Inflectional categories are grammatical markers in the structure of the word to signal its semantic and/or pragmatic import in the construction of higher order structures, namely phrases and clauses. Furthermore, the (complex) word domain is parsimoniously integrated into higher grammatical domains (clauses and phrases) by the inheritance of relational morphological features (i.e. inflection operators). As pointed out in Martín Arista (2008: 137): ‘Inflection at Word or Complex Word Level is tantamount to inflection at the Nucleus level of the Phrase. Consequently, the node Word or Complex Word represents the limit of percolation of morphological features as well as the limit of inheritance of inflectional features from the NP.’ Figure 8.5 illustrates this. The set of semantic markings that may be realized morphologically are those encoded in the operator projections of both the LSC and the LSP. Thus, the morphologically complex Spanish verb cant-aba-n (sing-ipfv.pst-3pl, ‘they were singing’) will trigger percolation of the aspect and tense features when acting as a predicate in a clause like Los niños cantaban villancicos (‘the children were singing carols’). In Figure 8.6 the square brackets delimit the LSW of the word cant-aba-n (sing-ipfv.pst-3pl, ‘they were singing’), which occupies the position of predicate in the LSC of the sentence. There are four operators in the LSW, encoded by two affixes, -aba, which signals both imperfective aspect and past tense, and the suffix -n which marks both third person and plural number. Note that not all four inflectional features are subject to percolation: the -n suffix signals the person and number features that mark agreement of the verb with the subject, but since they are not categories that can be matched to the set of operators available in the LSC, where the verb acts as nucleus, they are not inherited by the clause structure. On the other hand, the imperfective and past operators are inherited by the operator projection in the LSC of this sentence, as indicated by the dotted lines in the figure.

8.3.2 Inflectional Morphology in Head-Marking Structures The analyses presented in Figures 8.5 and 8.6 illustrate the behaviour of inflectional features as encoded in the LSW in dependent-marking languages when integrated into higher syntactic structures. In

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

377

378

FRANCISCO J. CORTÉS-RODRÍGUEZ

RP

CORE LSP (Constituent Projection) NUC WN = NOUN COREW NUCW LSW W

INFL

percolation

NUCW COREW

WN = NOUN NUC LSP (Operator Projection)

Number

CORE RP

Gender

RP

Case

Figure 8.5 Feature percolation in RPs (adapted from Martín Arista 2008: 136)

head-marking languages the integration of the structural domains of the word and the clause is tighter: not only may feature percolation of the operator projection from the LSW to the operator projection of the LSC come into play, but some features of the LSW also occupy positions in the constituent projection of the LSC. Van Valin (2013) offers an analysis of morphologically complex words in head-marking languages based on Everett’s (2002) proposal for the LSW. As mentioned earlier, Everett’s model postulates only a constituent projection in which inflectional affixes are daughters of the lexical core, while derivation takes place within the nucleus. As a consequence, the nucleus is opaque to syntax, but the core is accessible to it. The morphological structure of the Lakhota (Siouan, North

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

SENTENCE CLAUSE CORE

RP

NUC

RP

WV = VERB COREW NUCW Los niños

cant

aba

n

villancicos

NUCW LSW COREW

imperf

WV =VERB

past

WV =VERB

person

WV =VERB NUC

number aspect

CORE CLAUSE

tense Feature Percolation LSW-LSC =

Figure 8.6 Feature percolation in the LSC

America) word Ø-wiˇchá-wa-k’u (inan-3pl.anim.u-1sg.a-give, ‘I gave it to them’) would then be as shown in Figure 8.7. The fact that in Figure 8.7 the morphological markers are part of the constituent projection illustrates the difference between dependent- and head-marking languages with respect to the status of inflection. In headmarking languages, the constituent projection in the LSW will house those inflectional features that instantiate the arguments in the semantic

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

379

380

FRANCISCO J. CORTÉS-RODRÍGUEZ

V

COREW

ARG

ARG

[3SG.NMR] [3PL.ANIM.U]

ARG

NUCW

[1SG.A]

k’u

Figure 8.7 The LSW of Lakhota verb wiˇchá-wa-k’u (Van Valin 2013: 113) (NMR ¼ non-macrorole, ANIM ¼ animate, U ¼ undergoer, A ¼ actor)

SENTENCE

CLAUSE CORE

V

COREV ARG

ARG

[3SG.NMR] [3PL.ANIM.U]

ARG

NUCV

[1SG.A]

k’u

PSA: NOM ACTOR

ACC UNDERGOER

[do′(1SG, Æ)] CAUSE [BECOME have′ (3PL.ANIM) (3SG)] Figure 8.8 The LSC of Lakhota sentence wiˇchá-wa-k’u (Van Valin 2013: 115) (NMR ¼ non-macrorole, ANIM ¼ animate, U ¼ undergoer, A ¼ actor, PSA ¼ privileged syntactic argument)

representation of the core of the clause. In fact, the structure of the lexical unit in the LSW will provide the structure of the core of the clause in the LSC; Van Valin (2013: 115) describes that the two structures are coextensive, since the structure of the COREw provides the structure of the core of the clause. Note that the analysis of the COREw structure in Figures 8.7 and 8.8 is almost identical to the structure of the core in the equivalent English clause ‘I gave it to them’. This proposal thus manages to provide comparable

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

structures for comparable sentences among typologically different languages while at the same time coping successfully with the differences among such structures. In fact, the disparity between head-marking and dependent-marking structures stems from the fact that while inflectional features are consistently located in the operator projection of the LSC in the case of dependent-marked constructions, they can appear distributed between both the constituent and the operator projections in head-marked structures. The analysis of head-marked constructions can also sit comfortably within the model by Martín Arista (2008, 2009, 2011), as is shown in Figure 8.9, which illustrates the percolation of morphological markers from SENTENCE CLAUSE CORE RP

NUC

RP

WV =VERB COREW NUCW LSW

Cant

aba

n

villancicos

NUCW COREW WV =VERB

imperf past

WV =VERB

person: 3

WV =VERB NUC

number: PL aspect

CORE CLAUSE

tense

Figure 8.9 The LSC of Spanish sentence cantaban villancicos

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

381

382

FRANCISCO J. CORTÉS-RODRÍGUEZ

the LSW to both the operator and the constituent projections in the LSC of the Spanish clause cant-aba-n villancicos (sing-ipfv.pst-3pl carols, ‘they were singing carols’). Even though Spanish is mainly a dependent-marking language, it shows subject marking on the verb by means of person and number inflectional affixes. This makes unnecessary the presence of an independent subject constituent, as happens in this sentence, thus allowing head-marked structures. In the absence of an overt subject argument (e.g. los niños ‘the children’ in Figure 8.6), the suffix -n is mapped from the LSW of the verb into the constituent projection of the LSC; this is indicated by the dotted line in the analysis. The other morphological features in the LSW of cantaba-n (sing-ipfv.pst-3pl), namely imperfective and past as encoded in the suffix -aba, percolate into the operator projection of the LSC (cf. also Figure 8.6).

8.4

Word Formation

This section deals with the other basic domain of morphology, namely the creation of new lexical units. It is important to note that more often than not grammatical models tend to focus their interest on the semantic representation of clauses and phrases, but they usually leave aside the question of how to account for the semantic representation of other grammatical structures such as compound and derived words. Contrary to this tendency, the model outlined in this section offers a design for the semantic representation of such units which complies with the following issues: firstly, it captures the semantic variability of word-formation patterns. The introduction of a two-tier enriched representation of lexical structures, which includes not only the event features encoded in logical structures, but also encyclopedic knowledge as captured by Qualia Theory, places RRG in an outstanding position when compared with other proposals of derivational morphology. Section 8.4.1 deals with these aspects and Section 8.4.2 offers some illustrative analyses of the most relevant affixal units in Spanish and English. Secondly, such a format of semantic representation will allow us to account for the grammatical relations that hold between the components of a derived lexeme within its corresponding LSW. This will be the aim of Section 8.4.3. These two aspects reveal a conception of derivational morphology as a double-sided phenomenon: as a lexical process, word-formation involves the creation of a new semantic structure in the wrapper of a lexical unit; at the same time, the constituents of a derivationally complex lexical item are interrelated by different types of grammatical relations, that is, word formation is also a syntactic phenomenon within the word domain.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

8.4.1

Word Formation as a Lexical Process: Lexical Representations and the Derived Lexicon As has already been mentioned, there is a great portion of morphology that is lexical in nature. Word formation unquestionably falls into the realm of lexical semantics, though the vast majority of models tend to overlook this aspect. Levin and Rappaport Hovav (2001: 248–249) offer a brief but revealing description of the neglect of the relationship between lexical semantics and morphology in generative approaches, especially after Lieber (1980). Even in those cases in which attention is paid to argument structure operations in morphological theories (such as, for example, in Baker 1988; Bresnan 1982; Rosen 1989 and Williams 1981, among many others), such operations are essentially syntactic, not semantic. They also cite a fundamental reason for this neglect, namely that these models often lack a comprehensive theory of lexical-semantic representations, a fundamental framework for a sound study of such a feature of morphology. A notable exception is the work of Lieber (2004), which can be considered a major contribution to the lexical semantics of morphology. She contends that a precondition for the study of derivational morphology is the existence of a solid theory of lexical-semantic representation. She also explains that such a theory must be decompositional and based in semantic primitives or atoms ‘of the right “grain size” to allow us to talk of the meanings of complex words’ (Lieber 2004: 4); furthermore, semantic representations must facilitate the analysis of lexical semantics and not only of the semantic properties of higher syntactic structures (such as phrases or clauses). Lastly, it must also allow us to describe the meanings of complex words in the same terms as the meanings of simplex lexemes. RRG can also claim to be exceptional with regard to this issue: lexical-semantic representations lie at the heart of the model, and morphosyntactic phenomena are always externally motivated by the semantic representations assembled from the lexical-semantic structure of its components. There have been many important contributions to the enhancement of RRG lexical representations in the primary lexicon,14 and in the derived lexicon (in the works of Cortés-Rodríguez 2006a, 2006b, 2009; Cortés-Rodríguez and Sosa 2008, 2012, also Nolan 2011). From all these studies it is possible to obtain a detailed picture of the overall configuration of the lexicon in RRG and, specifically, of the structure of lexical semantic representations. As was stated in Section 8.2, the lexicon consists of lexical morphemes of two types: free lexical morphemes or lexemes (which would belong to the traditionally named open word classes) and bound lexemes (derivational affixes). They differ only in their distributional freedom of occurrence in discourse. Both the primary and the derived lexicons are organized internally in lexical groupings of semantic and syntactic affinity, which can be termed lexical classes (see Faber and Mairal Usón 1999 for an overview of the hierarchical organization of the English verbal

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

383

384

FRANCISCO J. CORTÉS-RODRÍGUEZ

lexicon). The lexical entries for both types of lexeme include a formal semantic representation in the format of lexical templates. The latest version of lexical templates for free lexemes (Cortés-Rodríguez 2009; Mairal Usón, Periñan-Pascual and Pérez Cabello de Alba 2012; Ruiz de Mendoza 2013) takes as a backbone for lexical representation the Aktionsart properties of predicates as formally encoded in the logical structures of RRG and enriches them by integrating Pustejovsky’s (1995) Qualia Theory together with the set of lexical functions from the explanatory and combinatorial lexicology (Mel’cuk 1989; Mel’cuk and Wanner 1996; Alonso and Tutin 1996). Despite their complexity, lexical templates are fully fledged repositories of the semantic features associated with a lexeme, either free or bound. By way of example, the lexical template for the Spanish verb captar ‘fathom’ would have the following format (Cortés-Rodríguez and Sosa 2012: 22): (10) captar: eventstr: qualiastr:

know′ (x, y) {Qf: manner : MagnObstr think′ (x, y) Qt: Culm know′ (x, y )}

The format of a lexical entry now consists of two basic components: the event and the qualia structure. In this case, since captar belongs to the class of Cognition verbs, the event structure is a state logical structure which takes know′ as a primitive and has two arguments (x, y), following the format of the logical structures of RRG. The second part of the lexical template includes its qualia structure. Qualia Theory is borrowed from Pustejovsky’s (1995, 2001) semantic model. Van Valin and LaPolla (1997: 184–186) already incorporated qualia as part of the semantic representation of nouns. Our proposal takes it beyond this and aims at including it as a part of any lexical semantic representation. What qualia tell us about a concept is the set of semantic constraints by which we understand a word when embedded in the language, which is also the purpose of our lexical templates. As Pustejovsky and Batiukova (2019: 162) state: The core notion of qualia structure is that there are four aspects that make up our knowledge of a word: the class of entities it denotes (the formal role or quale), how the denoted entity was created (the agentive role), the intended function of this entity (the telic role), and the internal makeup of this entity (the constitutive role). They offer as an example the qualia structure of the noun violin. (11)

violin (x) QS ¼ Formal ¼ musical_instrument (x) Agentive ¼ build (y, x) Telic ¼ produce_music_on (z, x) Constitutive ¼ strings_of (w, x)

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

Turning to the representation of Spanish captar in (10), the template encodes two qualia characterizations: the formal quale Qf encodes a subevent in which a semantic function manner is combined with the lexical functions ‘Magn’ (‘intensity’) and Obstr (‘difficulty’); the combination of these functions in turn modify the primitive think′. Thus, the formal quale describes the degree of difficulty in carrying out the process of thinking implicit in the meaning of captar. The telic quale as encoded in Qt: Culm know′ (x, y) depicts the section of the event that expresses the final aim of the event, to reach knowledge or to understand. As can be seen, one of the goals of the qualia structure section is to represent the semantic attributes by means of which a lexeme is semantically distinguished within the larger set of units that belong to its lexical class. These two components of lexical entries, event structure and qualia, are similar to the so-called ‘skeleton’ and ‘body’ of lexical representations in Lieber’s (2004, 2009) proposal. However, we believe that the lexical entries of RRG have some advantages over Lieber’s system: the event structure captures in essence the semantic features encoded in the logical structures of RRG. Logical structures are formalized representations of the Aktionsart features of a given unit and are solid typologically based representations. Even though Lieber’s skeleton is meant to include a set of universal semantic features, no such set is described, and she defines only those that are syntactically relevant for English (Lieber 2009: 83). Similarly, the body, or set of semantic/pragmatic features which form encyclopedic knowledge, is to a certain extent similar to qualia features, though Lieber allows for some ‘unsystematicity’, insofar as a part of the ‘body’ (the ‘fat’; Lieber 2009: 83) varies from one individual to another. As regards the other part of the ‘body’ (the ‘muscle’), which is shared by a community, again a coherent and systematic description of what it consists of is still lacking. In comparison, qualia theory is restrictively based on the Aristotelian modes of explanation or aitia, meant to give a principled account of how humans understand the world. As described in Pustejovsky and Batiukova (2019: 162), qualia structure is ‘a relational system whose parameters allow us to decompose word meaning in a principled way, accounting for what other concepts and words it can be associated with in different contexts, based on its meaning’. Furthermore, qualia features and event structure descriptions are formally represented by means of the same metalanguage, which enhances the interaction between all elements in semantic representations of lexical and syntactic structures in RRG.

8.4.2 Some Illustrative Analyses In a similar way to free lexical units, affixes are also clustered in affixal classes, and their behaviour as members of these classes is also parallel to

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

385

386

FRANCISCO J. CORTÉS-RODRÍGUEZ

that exhibited by words in the primary lexicon. Some units are more central or higher in the structure whereas other affixes are peripheral, and some are even located in between two lexical classes. In this section a brief description of some of the most relevant affixal classes in Spanish and English will be offered (for a more detailed account, see Cortés-Rodríguez 2006a, 2006b, Cortés-Rodríguez and Sosa 2008, 2012). More specifically, an account of the semantic aspects of one important group of nominalizations is given first, followed by an explanation of the intricate meaning relationships between the components of derived causative verbal formations.

8.4.2.1 Macrorole Nominalizations Let us consider how the semantic content of the Spanish nominalizing affixes -ero (e.g. camión ‘lorry’ > camionero ‘lorry-driver’, guerra ‘war’ > guerrero ‘warrior’, café ‘coffee’> cafetera ‘coffee pot’), -or (e.g. afilar ‘to sharpen’ > afilador ‘sharpener’, calentar ‘to heat up’ > calentador ‘heater’, despertar ‘to wake up’ > despertador ‘alarm clock’, poseer ‘to own’ > poseedor ‘owner’) [-a/-e/-ie]nte (e.g. humillar ‘to humiliate’ > humillante ‘humiliating’, galopar ‘to gallop’ > galopante ‘rapidly advancing / out of control’, desinfectar ‘to disinfect’ > desinfectante ‘disinfectant’), -ista (e.g. novela ‘novel’> novelista ‘novelist’, ébano ‘ebony’ > ebanista ‘cabinetmaker’, izquierda ‘left’ > izquierdista ‘leftist’), [-a/-i]do (e.g. embravecer ‘to become rough’ > embravecido ‘furious/rough’, errar ‘to err’ > errado ‘mistaken/wrong’) and -ario (e.g. arrendar ‘to rent’ > arrendatario ‘tenant’, concesión ‘licence/concession’ > concesionario ‘licensee/franchisee’) and also English -er, -ist, [-a/e]-nt, -ee is represented in this proposal: (12)

φiN: [(xi. . ., [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], where x ¼ macrorole φ ¼ lexical unit LT ¼ lexical template φBASE ¼ base lexeme i ¼ denotational co-indexation

This formula reads as follows: Any derived nominal lexeme φiN belonging to this class denotes a participant entity xi of one of the events encoded in the qualia structure (Qualia:…xi) that forms part of the lexical template of the base lexeme LT…φBASE; hence the derived lexeme and that entity are coindexed by a superscript i. Another way to express this is to describe these processes as nominalizations oriented towards an argument in the semantic representation of the lexical unit that functions as the base of the derivation. There is an additional restriction on this entity: it must have macrorole status. Therefore, the structure in (12) represents the ‘event structure’ of the derivational class of concrete nominalizing affixes, or concrete processual substances/things/essences in Lieber’s (2004: 36) terminology. We propose

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

labelling this derivational class as ‘macrorole nominalizations’ since they encompass both actor and undergoer nominalizations. Actor nominalizations include derived instruments, agents, experiencers, locations, etc., and undergoer formations are prototypically formed in Spanish by means of (-a/-i)do and -ario and in English by -ee and -ed. The two corresponding templates are the following. (13)

φiN: [LT (xi. . ., [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ actor E.g. escritor ‘writer’, poseedor ‘owner’, panfletista ‘pamphleteer’, desodorante ‘deodorant’, lechero ‘milkman’.

(14)

φiN: [LT (. . .xi, [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ undergoer E.g. arrendatario ‘tenant’, fideicomisario ‘trustee’, fallecido ‘deceased’

The labels ‘actor and undergoer nominalizations’ explain the wide scope of these types of derivational processes. The semantics of the templates cannot be associated with a specific semantic function such as ‘agent’. Even though the most prototypical formations correspond to agent nominals, like English writer, runner, violinist, etc., there are many other formations where the notion of ‘agenthood’ is absent (cf. formations like believer, owner, lover, to mention just a few). The term ‘actor’ indicates that all the formations are nominalizations of the macrorole actor, as defined within RRG (see Chapter 4). This, in turn, justifies the superscript i which coindexes the lexical variable for the derived word (φN) with the participant that would receive that macrorole function. That is, they mark the nominals as oriented towards one entity (xi) involved in one of the subevents that form part of the state of affairs depicted by the base word. The variable LT (‘lexical template’) indicates that the event in which this entity participates can be of any kind, a state (pred′), an activity (do′), or any other logical structure. The following representation is to be understood as a subspecification of the actor template, and corresponds to the traditionally labelled ‘agent nominals’, which in RRG terms should be described as ‘effector nominalizations’ (Van Valin and Wilkins 1996). (15)

φiN: [do′ (xi, [φBASE])], x ¼ actor. E.g. driver, runner, smoker.

This structure expresses the semantic content of the most prototypical nominalizations within the class: the derived words corresponding to this construction describe the effector involved in the event expressed by the semantics of the base word. There are two co-indexing possibilities expressed in the above representation depending on whether the formation is deverbal (φV) or not (φ[-V] ). In the case of deverbal effector nouns, co-indexation is usually quite straightforward: the verbal bases typically encode an event that is dynamic, and therefore the meaning of effectorhood derives from the semantic function of its first argument. This is the case of hunter.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

387

388

FRANCISCO J. CORTÉS-RODRÍGUEZ

(16)

hunter [X-er] N: [ (xi , [huntv ])], x = actor

huntV: eventstr: [do′ (x, e1)]

OUTPUT hunter iN : argstr: x: animate qualiastr: {…QA: e1 [do′ (x i , [hunt′(x,y)])]}

The semantic structure of the Spanish nominal creyente ‘believer’ in (17) illustrates the fact that these kinds of derived actor nominals do not exclusively refer to agent arguments. Let us recall that the actor macrorole is assigned to the leftmost argument in a logical structure with two arguments, irrespective of the type of event encoded (Van Valin 2005: 60–67). (17) creyente ‘believer’ [X(-a/-e/-ie)nte] N: [(xi , [(creerV)])], x = actor

creerV: eventstr: e1[pred′ (x,y)]

OUTPUT creyenteiN: argstr: x: human qualiastr: {…QA: e1[believe' (xi,y)]}

The semantic interpretation of denominal actor lexemes – such as pianista ‘pianist’, cabrero ‘goatherd’, banquero ‘banker’, pensionista ‘pensioner’ – likewise arises from co-indexation with one argument in one of the events depicted in the qualia structure of the base noun, as is the case of violinista, where semantic composition is based on the event encoded as the telic quale of violin. (18)

violinista ‘violinist’ [violinN + istaiN] iN: [DO (xi ACTOR , [do′ (xi, [play′ (xi, violinN)])])] pianoN:

typestr = (y: artifact-lcp) argstr: y: phys_object qualiastr: {Qformal: musical.instrument (y) Qconst.: pegs_strings_bow_....of (w,y) Qagent.: e1[do′ (w, [create′ (z,y)])] Qtelic: e1[do′ (xi, [play′(xi, y)])]}

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

The representation in (18) also reveals that there are interesting semantic differences among the members of this subclass. The affix -ista is more restrictive than other members among the actor nominalizing affixes in terms of: (a) the types of bases (it prefers nouns and adjectives, in contradistinction with the more deverbal -or and -nte); and (b) its meaning, as it always involves a degree of volitionality not necessarily present in the other affixes. Thus, its LT would be more specific or hyponymic. (19)

[φBASE: N/adj þ istaiN] iN : [DO (xi, [do′ (xi, [e2: LT . . .φBASE: (Qualia: . . .LT (xi…)])])]

(19) is a good instance of the kind of information that is encoded in an affixal LT. It includes a morphological frame with the constituent makeup of the derived words: [φBASE:N/adj þ istiN] iN (cf. Nolan 2011). This frame includes information of the categories of both the base [φBASE:N/Adj] and the complex word (. . .] iN). The semantic specificity of -ista is expressed through the agent operator ([DO (xi,) (see Van Valin and LaPolla 1997: 118–120), which is understood as an overlay of the effector thematic role. Note that the specific semantic content of this formation is spelled out by the joint effect of the selection of one quale from the LT of the base word, and co-indexation of the affix with one of the arguments involved in the subevent of such a quale.15 Macrorole nominalizations are probably the most significant class to look at if we are to fully appreciate how relevant co-indexation is for the construction of the meaning of complex words. However, there is another mechanism that is to be used in this meaning-construction process: qualia selection. Qualia selection is essential to understand the semantics of derived causative verbs.

8.4.2.2 Derived Causative Verbs One of the great difficulties of a lexical-semantic approach to word formation is the rampant polysemy that most affixes and composition patterns exhibit. Jackendoff (2009: 117) explains that a compound like box car can have a plethora of paraphrases (car that carries boxes, that looks like a box, that is used as a box, etc.) However, this does not mean that complex words are ambiguous or vague; Jackendoff describes them as semantically ‘promiscuous’, which means that it is possible for a complex word to take up all those meanings. A similar approach is defended in the works on word formation within Pustejovsky’s Generative Lexicon framework (Batiukova 2008; Johnston and Busa 1999). This proposal follows the same path in considering that complex words are prima facie semantically and grammatically underspecified structures. However, this does not mean that there are no limits to compositionality. All possible interpretations of a complex word must be anchored – directly or indirectly – in the richer semantic information encoded in the LTs of its components, and all word-formation processes involve the activation of mechanisms to select the meaning or network of possible meanings of the complex word. In this regard, qualia selection from LTs plays a fundamental role.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

389

390

FRANCISCO J. CORTÉS-RODRÍGUEZ

The derivational class of causative verbal lexemes will be of help to provide examples of this phenomenon. The basic lexical template for this class is as follows. (20)

[

BASE

i

+

V ] V:

[do′ (x, Ø )] CAUSE [ e2 : (LT : … base

BASE ...

)]

= lexical unit LT = lexical template BASE = base lexeme i = denotational co-indexation

The formula in (20) encodes a complex semantic structure in which there is a causal bond between two subevents, the induced one corresponding to a state of affairs in which the base word is involved or affected more or less directly. Typically, causative formations are category-changing processes, the bases being nouns or adjectives. Some conversion processes (e.g. conciencia ‘awareness’ > concienciar ‘raise awareness’, completo ‘complete’ > completar ‘to complete’, alegre ‘happy’ > alegrar ‘gladden’) and the affixes a- (e.g. feo ‘ugly’ > afear ‘spoil, make ugly’, largo ‘long’ > alargar ‘lengthen’), -iz- (e.g. suave ‘soft’ > suavizar ‘soften’, colonia ‘colony’ > colonizar ‘colonize’), -ific- (e.g. sólido ‘solid’ > solidificar ‘solidify’, puro ‘pure’ > purificar ‘purify’), and en- (e.g. ancho ‘wide’ > ensanchar ‘widen’, lata ‘can’ > enlatar ‘to can’, jaula ‘cage’ > enjaular ‘encage’) are members of this derivational class in Spanish. The semantic interpretation of every complex word that is an output of causativization is determined by the selective binding or exploitation of one quale of the LT of the base, expressed in (20) by φBASE (cf. CortésRodríguez 2006a; Batiukova 2008). To be more precise, the quale that is selected for binding will endow the caused subevent in (20) ([e2: (LT: . . .φBASE. . . )]) with a specific content. For instance, a formation like Spanish enlatar ‘to can’ has a causative-locative interpretation, which is promoted by the selection of the formal quale (Qf: container′ (xi, y)) of the noun lata ‘can’.16 Consequently, the abstract operator LT corresponding to the caused subevent would receive a specific locative interpretation (BECOME be-in′ (xi, y)]); coindexation (represented by the arrow in (21)) specifies further that the base entity designates a locus (Cortés-Rodríguez and Sosa 2012: 27). (21)

enlatar (en ‘cause’ + lata ‘can’ [en- +

lataNi ]V:

enlatar ‘to can’)

do′ (z, Ø) CAUSE [e2= [BECOME be-in′ (lataiN , y)]]

TYPESTR = (x:

lataN

QUALIASTR=

artifact-lcp)

… QF: container′ (xi,y) QC: metallic′ (x) QA: e1[do′ (z, Ø)]/ artifact′ (x y) QT: e2: [BECOMEbe-in′(xi, y)]

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

On the other hand, deadjectival forms like Spanish suavizar ‘soften’, ensanchar ‘widen’, solidificar ‘solidify’, etc. select the formal quale of the base and the meaning of the complex word will be that of a prototypical causativeresultative verb (‘cause become adj’), as shown in (22) (Cortés-Rodríguez and Sosa 2012: 28). (22)

suavizar (suave ‘soft’ + izar ‘cause’ suavizar ‘to soften’) [suaveAdj i + izar V ]V : do′ (z, Ø ) CAUSE [e 2 = [BECOME suave′ (y)]] suaveAdj i : {…QF: be′(y, suave′i ) …}

As can be seen, co-indexation works hand-in-hand with qualia selection: its function is to bind one element from the base template with another from the affixal template, or the modifying lexeme in the case of compounds.17

8.4.3

Word Formation as a Grammatical Process: The Interface between Lexical Templates and the LSW We are now in a position to consider how the semantic structures underlying word-formation products of the kind outlined previously can find a syntactic counterpart in a syntactic theory of the word. Martín Arista’s (2008, 2009, 2011) LSW as described in Section 8.2 offers a number of significant advantages for this task: it will enable us to design a semanticsto-syntax interface between the semantic descriptions proposed here and the layered structure of the corresponding complex word. In fact, only a few adjustments are needed in order to achieve this goal. The most important of these would be to reassess the grammatical status of some constituents in the LSW, as proposed in the original model. A good example for this would be the LSW of bookseller, analysed in Martín Arista (2009: 92) as an exocentric formation (Figure 8.10). This analysis, which takes as nucleus the verbal base, seems to be based on the fact that the elements of word syntax are a kind of carbon-copy of what they could be at clause level; this is what seems to be stated in the following quotation (Martín Arista 2009: 94): The definition of Word functions is based on Clause functions, which requires an indirect association with a clausal expression: incomer expresses a First Argument whereas income expresses a Second Argument, incomer, outflow and inflow express an ArgumentAdjunct, etc. However, if we want to do justice to the lexical-semantic dimension of word formation, we must provide a semantic motivation for the functional constituents of the LSW, which in turn would be more in line with what is done in the other RRG grammatical domains (Van Valin and LaPolla 1997: 68):

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

391

392

FRANCISCO J. CORTÉS-RODRÍGUEZ

The elements functioning as nuclei of PPs and clauses are predicates in the semantic representation, while nominal nuclei are designated within the semantic representation of the NP. Thus the head of a phrase is a function of its semantics: an NP is headed by a nominal nucleus, a PP by an adpositional nucleus, and a clause by a predicating nucleus. COMPLEX WORDN COREN ARG

NUCV

ARGN

WORD

PREDICATEV

WORDN

CORE

CORE

NUC

NUCN

book

sell

er

Figure 8.10 Exocentric analysis of compound bookseller (Martín Arista 2009: 92)

The logical follow-on from this statement is that if a word has a referential semantic function, its nucleus (or head) should be the element that confers such a value. We can borrow some criteria from Štekauer (2005b: 225–226) to select the nucleus element within the LSW of a complex word: (i) Hyponymy: the complex word is a semantic specification of the head; storyteller, for example, is a hyponym of teller, which in turn is a hyponym of -er. (ii) Subcategorization: heads impose subcategorization restrictions; -ist selects noun bases, -en has a strong preference for monosyllabic bases with final plosives, as in red > redden, short > shorten. (iii) Distributional and categorial equivalence: the head (or nucleus) specifies the lexical class of the derived word. For example, -er formations are consistently denominal; on the other hand, locative prefixes (pre-field or sub-way) are not heads as they do not determine the class form of the derived word. This last criterion draws a key parallel between the nature of the head or nucleus of clauses and phrases, on the one hand, and complex words, on the other. If we take these conditions into account, the layered structure of bookseller could be as shown in Figure 8.11 (Cortés-Rodríguez and Sosa 2012: 36). The structure captures the endocentric character of the bound lexeme -er in the derivation of seller and the recursive compositional process with book. Figure 8.12 shows the structure of non-endocentric complex words in

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

COMPLEX WORDN COREN NUCLEUSN WORDN

PREDICATEN COMPLEX WORDN

CORE N

COREN

NUCN PREDICATEN

NUCN WORDV

PREDICATEN

COREV NUCV PREDICATEV book

sell

er

Figure 8.11 Endocentric analysis of compound bookseller (Cortés-Rodríguez and Sosa 2012: 36)

Spanish, which includes compounds like lavaplatos ‘dishwasher’, sacacorchos ‘corkscrew’, pagafantas (lit. ‘Fanta payer’) ‘conned cuckold’ and conversion formations like destino ‘destiny’, ‘end use ’> destinar ‘assign’, amigo ‘friend’ > amigar ‘make/become friends’, deseo ‘desire’ > desear ‘to desire’. Coindexation in all these cases is revealing: none of the components of the morphological template is co-indexed with the lexical variable of the complex word [lavaN/adj þ platosNα] iNβ, in sharp contrast with endocentric formations as in pianista [pianoN þ istaiN] iN. Thus, it may be more appropriate to describe these complex words as ‘acentric’, since they do not have a lexically saturated nucleus (Cortés-Rodríguez and Sosa 2012: 37).

8.5

Conclusion

The aim of this chapter was to present an overview of morphology in RRG. As stated at the outset, despite the obvious recognition of the centrality of morphological structure, there is as yet no fully articulated theory of the structure of the word domain and the way it interacts with

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

393

394

FRANCISCO J. CORTÉS-RODRÍGUEZ

COMPLEX WORDN COREN WORDV

WORDN

COREV

COREN

NUCV

NUCN

PREDV

PREDN

lava

platos

NUCN

[lavaN/ADJ + platosNα] iNβ : [do′ (xi , [ washV-BASE ]), x= Actor washV :

EVENTSTR: e1 < e2 / e1 CAUSE e2 QFORMAL : (e2: [do′ (xi , [wash'(xi, platos)]) QAGENT: (e1: [do′ (y, [use′ (y, xi)])])

Figure 8.12 ‘Acentric’ Spanish compound lavaplatos (‘dishwasher’) (Cortés-Rodríguez and Sosa 2012: 37)

the structure of higher grammatical domains such as phrases and clauses. However, there are some significant contributions leading to the development of such a morphological theory. The first challenge, the establishment of descriptive tools for the structure of the word, was the topic of Section 8.2, in which we offered the guidelines for a layered structure of the word. Given the explanatory character of the model and its functional (or better, communication-and-cognition) orientation, Sections 8.3 and 8.4 were dedicated to accounting for the functional motivation of inflection and word formation, respectively. In Section 8.3 we offered a description of the interface of inflection within and outside the word. This has in turn revealed the potential of the LSW for the explanation of the interaction between syntactic and semantic structures, closely tied to the notion of ‘operator percolation’ between the LSW and the LS of higher grammatical structures. However, the intricate relation between word structure and clause structure goes beyond feature percolation in the case of head-marking languages, as we also pointed out in this section. Section 8.4 was devoted to the explanation of word formation within RRG. This involves identifying the nature of the connections that hold

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

between morphology, on the one hand, and lexical semantics and syntax, on the other. As a lexicological phenomenon, derivational affixes and processes form part of the lexicon in RRG and – very much like free lexemes – are endowed with a semantic representation in the format of a lexical template. The interaction of affixal and word templates is accounted for by means of two explanatory devices: co-indexation and qualia specification. The section closed with a brief overview of how the semantic structure underlying a derivationally complex word also finds its syntactic counterpart in the LSW, as described in Section 8.2, with some slight adjustments to the proposal put forward by Martín Arista (2008, 2009, 2011). In our opinion, taken together, the proposals outlined in this chapter, which draw on several contributions both from inside and outside RRG, constitute an explanatory framework for the study of morphological phenomena within this model.

References Ackema, P. and A. Neeleman. 2003. Syntactic atomicity. Journal of Comparative Germanic Linguistics 6: 93–128. Alonso Ramos, M. and A. Tutin. 1996. A classification and description of lexical functions for the analysis of their combinations. In L. Wanner (ed.), Lexical Functions in Lexicography and Natural Language Processing, 146–167. Amsterdam: John Benjamins. Anderson, S. 1992. A-Morphous Morphology. Cambridge: Cambridge University Press. Aronoff, M. and K. Fudeman. 2011. What Is Morphology? (2nd ed.). Oxford: Wiley-Blackwell. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago, IL: University of Chicago Press. Baker, M. 2003. Lexical Categories. Cambridge: Cambridge University Press. Batiukova, O. 2008. Morfología: del léxico a la sintaxis oracional. In Actas del VIII Congreso de Lingüística General. Universidad Autónoma de Madrid. www.lllf.uam.es/clg8/actas/index.html. Bentley, D. 2018. Monotonicity in word formation: The case of ItaloRomance result-state adjectives. Transactions of the Philological Society 116 (3): 285–319. Bentley, D. 2019. The logical structure of verbs of quantized and nonquantized change. Paper delivered at the 2019 International Conference on Role and Reference Grammar. The State University of New York, University at Buffalo (SUNY). Bosque, I. 2012. On the lexical integrity hypothesis and its (in)accurate predictions. Iberia: An International Journal of Theoretical Linguistics 4(1): 140–173.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

395

396

FRANCISCO J. CORTÉS-RODRÍGUEZ

Boutin, M. 2011. Towards a realizational approach to morphology in RRG. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 2–34. Newcastle upon Tyne: Cambridge Scholars Publishing. Bresnan, J. (ed.). 1982. The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press. Bybee, J. 1985. Morphology: A Study of the Relation between Meaning and Form. Amsterdam: John Benjamins. Cortés-Rodríguez, F. J. 2006a. Derivational morphology in Role and Reference Grammar: A new proposal. RESLA: Revista Española de Lingüística Aplicada 19: 41–66. Cortés-Rodríguez, F. J. 2006b. Negative affixation within the Lexical Grammar Model. RÆL: Revista Electrónica de Lingüística Aplicada 5: 27–56. Cortés-Rodríguez, F. J. 2009. The inchoative construction: Semantic representation and unification constraints. In C. S. Butler and J. Martín Arista (eds.), Deconstructing Constructions, 247–270. Amsterdam: John Benjamins. Cortés-Rodríguez, F. J. and E. Sosa. 2008. The morphology–semantics interface in word-formation. Revista Canaria de Estudios Ingleses 57: 91–108. Cortés-Rodríguez, F. J. and E. Sosa. 2012. La morfología derivativa. In R. Mairal Usón, L. Guerrero and C. González (eds.), El funcionalismo en la teoría lingüística: La Gramática del Papel y la Referencia, 19–42. Madrid: Akal. Di Sciullo, A. M. and E. Williams. 1987. On the Definition of Word. Cambridge, MA: MIT Press. Diedrichsen, E. 2011. The layered structure of the German word: An RRG approach to inflectional morphology. Ms. Paper delivered at the 2011 International Course and Conference on Role and Reference Grammar. The Pontificia Universidad Católica de Chile, Santiago, Chile. Diedrichsen, E. and B. Nolan. 2011. The syllable in the LSW: A proposal for an RRG morphophonological interface. Ms. paper delivered at the 2011 International Course and Conference on Role and Reference Grammar. The Pontificia Universidad Católica de Chile, Santiago, Chile. Dik, S. C. 1997. The Theory of Functional Grammar. 2 vols. Berlin: Mouton de Gruyter. Everett, D. 2002. Towards an RRG theory of morphology. Lecture given at the 2002 International Conference on Role and Reference Grammar, Universidad de La Rioja, Logroño, Spain. Faber, P. and R. Mairal Usón. 1999. Constructing a Lexicon of English Verbs. Berlin: Mouton de Gruyter. Foley, W. A. and R. D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Hengeveld, K. 1989. Layers and operators in Functional Grammar. Journal of Linguistics 25: 127–157. Jackendoff, R. 2009. Compounding in the Parallel Architecture and conceptual semantics. In R. Lieber and P. Štekauer (eds.), The Oxford Handbook of Compounding, 105–128. Oxford: Oxford University Press.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

Johnston, M. and F. Busa. 1999. Qualia structure and the compositional interpretation of compounds. In E. Viegas (ed.), Breadth and Depth of Semantic Lexicons, 167–187. Dordrecht: Kluwer Academic. Lapointe, S. 1981. A lexical analysis of the English auxiliary verb system. In T. Hoekstra, H. van der Hulst and M. Moortgat (eds.), Lexical Grammar, 215–254. Dordrecht: Foris. Levin, B. and Rappaport Hovav, M 2001. Morphology and lexical semantics. In A. Spencer and A. M. Zwicky (eds.), The Handbook of Morphology, pp. 248–271. Oxford: Blackwell. Lieber, R. 1980. On the Organization of the Lexicon. PhD dissertation, MIT (published by Garland Press, New York, 1994). Lieber, R. 2004. Morphology and Lexical Semantics. Cambridge: Cambridge University Press. Lieber, R. 2009. A lexical semantic approach to compounding. In R. Lieber and P. Štekauer (eds.), The Oxford Handbook of Compounding, 78–104. Oxford: Oxford University Press. Mairal Usón, R. and Cortés-Rodríguez, F. J. 2000–2001. Semantic packaging and syntactic projections in word formation processes: The case of agent nominalizations. RESLA (Revista Española de Lingüística Aplicada) 14: 271–294. Mairal Usón, R. and F. J. Ruiz de Mendoza. 2008. New challenges for lexical representation within the Lexical-Constructional Model. Revista Canaria de Estudios Ingleses 57: 137–158. Mairal Usón, R. and F. J. Ruiz de Mendoza. 2009. Levels of description and explanation in meaning construction. In C. S. Butler and J. Martín Arista (eds.), Deconstructing Constructions, 153–198. Amsterdam: John Benjamins. Mairal Usón, R., C. Periñán-Pascual and M. B. Pérez Cabello de Alba. 2012. La representación léxica: Hacia un enfoque ontológico. In R. Mairal Usón, L. Guerrero and C. González (eds.), El funcionalismo en la teoría lingüística: La Gramática del Papel y la Referencia, 85–102. Madrid: Akal. Marchand, H. 1969. The Categories and Types of Present-Day English Word Formation, 2nd ed. Munich: Beck. Martín Arista, J. 2008. Unification and separation in a functional theory of morphology. In R. D. Van Valin, Jr. (ed.), Investigations of the Syntax– Semantics–Pragmatics Interface, 85–115. Amsterdam: John Benjamins. Martín Arista, J. 2009. A typology of morphological constructions, In C. S. Butler and J. Martín Arista (eds.), Deconstructing Constructions, 85–116. Amsterdam: John Benjamins Martín Arista, J. 2011. Projections and constructions in functional morphology: The case of Old English HR¯ EOW. Language and Linguistics 12(2): 393–425. Martín Arista, J. 2012. La morfología flexiva. In R. Mairal Usón, L. Guerrero and C. González (eds.), El funcionalismo en la teoría lingüística: La Gramática del Papel y la Referencia, 43–58. Madrid: Akal.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

397

398

FRANCISCO J. CORTÉS-RODRÍGUEZ

Martín Mingorance, L. 1998 [1985]. Bases metodológicas para un estudio contrastivo del léxico derivado. In A. Marín Rubiales (ed.), El Modelo Lexemático Funcional, 61–82. Granada: Editorial Universidad de Granada. Mel’cuk, I. 1989. Semantic primitives from the viewpoint of the MeaningText Linguistic Theory. Quaderni di Semantica 10(1): 65–102. Mel’cuk, I. and L. Wanner. 1996. Lexical functions and lexical inheritance for emotion lexemes in German. In L. Wanner (ed.), Recent Trends in MeaningText Theory, 209–227. Amsterdam: John Benjamins. Möllemann, R. 2016. Implications of German Word Formation Processes for a Role and Reference Grammar Approach to Morphology. MSc, Heinrich Heine University Düsseldorf. www.semanticscholar.org/paper/Implications-of-Germanword-formation-processes-for-Mollemann/ fcc1cf700f876136c0300269f5391671155c3be1. Nolan, B. 2011. Meaning construction and grammatical inflection in the Layered Structure of the Modern Irish word: An RRG account of morphological constructions. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 64–101. Newcastle upon Tyne: Cambridge Scholars Publishing. Nübling, D. 2008. Historische Sprachwissenschaft des Deutschen (2nd ed.). Tübingen: Narr. Pavey, E. L. 2010. The Structure of Language: An Introduction to Grammatical Analysis. Cambridge: Cambridge University Press. Periñán-Pascual, C. and R. Mairal Usón. 2009. Bringing Role and Reference Grammar to natural language understanding. Procesamiento del Lenguaje Natural 43: 265–273. Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Pustejovsky, J. et al. 2001. Advances in Generative Lexicon Theory. Dordrecht: Springer. Pustejovsky, J. and O. Baitukova. 2019. The Lexicon. Cambridge: Cambridge University Press. Rijkhoff, J. 2002. The Noun Phrase. Oxford: Oxford University Press. Roberts, J. R. 2015. Distributives in Amele. A Role and Reference Grammar analysis. SIL Electronic Working Papers 2015–001. www.sil.org/system/files/ reapdata/16/20/84/162084215824434557688405164453437430340/ silewp2015_001.pdf. Roberts, T. 1999. Grammatical relations and ergativity in Sta’át’imcets (Lillooet Salish). International Journal of American Linguistics 65: 275–302. Rosen, S. T. 1989. Two types of noun incorporation: A lexical analysis. Language 65: 294–317. Ruiz de Mendoza, F. J. 2013. Meaning construction, meaning interpretation and formal expression in the Lexical Constructional Model. In B. Nolan and E. Diedrichsen (eds.), Linking Constructions into Functional Linguistics, 231–270. Amsterdam: John Benjamins. Ruiz de Mendoza, F. J. and R. Mairal Usón. 2007a. High-level metaphor and metonymy in meaning construction. In G. Radden, K. Köpcke, T. Berg and

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

P. Siemund (eds.), Aspects of Meaning Construction, 33–51. Amsterdam: John Benjamins. Ruiz de Mendoza, F. J. and R. Mairal Usón. 2007b. Levels of semantic representation: Where lexicon and grammar meet. Interlingüística 17: 26–47. Selkirk, E. O. 1982. The Syntax of Words. Cambridge, MA: MIT Press. Spencer, A. 1998. Morphophonological operations. In A. Spencer and A. M. Zwicky (eds.), The Handbook of Morphology, 123–143. Oxford: Blackwell. Spencer, A. 2004. Morphology: An overview of central concepts. In L. Sadler and A. Spencer (eds.), Projecting Morphology, 67–109. Stanford, CA: CSLI Publications. Štekauer, P. 2005a. Meaning Predictability in Word Formation. Amsterdam: John Benjamins. Štekauer, P. 2005b. Onomasiological approach to word-formation. In P. Štekauer and R. Lieber (eds.), Handbook of Word-Formation, 207–232. Dordrecht: Springer. Stump, G. 2001. Inflectional Morphology: A Theory of Paradigm Structure. Cambridge: Cambridge University Press. Van Valin, R. D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface: An Introduction to Role and Reference Grammar. Cambridge: Cambridge University Press. Van Valin, R. D., Jr. 2008. RPs and the nature of lexical and syntactic categories in Role and Reference Grammar. In R. D. Van Valin, Jr. (ed.), Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178. Amsterdam: John Benjamins. Van Valin, R. D., Jr. 2013. Head-marking languages and linguistic theory. In B. Bickel, L. A. Grenoble, D. A. Peterson and A. Timberlake (eds.), Language Typology and Historical Contingency, 91–124. Amsterdam: John Benjamins. Van Valin, R. D., Jr. and D. P. Wilkins. 1996. The case for ‘effector’: Case roles, agents, and agency revisited. In M. Shibatani and S. Thompson (eds.), Grammatical Constructions, 289–322. Oxford: Clarendon Press. Van Valin, R. D., Jr. and R. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Williams, E. 1981. On the notions ‘lexically related’ and ‘head of a word’. Linguistic Inquiry 12(2): 245–274.

Notes 1 We may even include phonology here, though this remains a very underdeveloped area. In Diedrichsen and Nolan (2011) there is an initial contribution in this field within RRG, and Möllemann (2016: 30–34) provides a brief description of the basic conditions necessary for a morphology– phonology interface in RRG.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

399

400

FRANCISCO J. CORTÉS-RODRÍGUEZ

2 In incremental theories it is assumed that lexical units acquire morphosyntactic features only after inflectional morphs are acquired; by way of example, works has the features ‘3rd person singular, present, indicative’ only after the suffix -s is attached to it. 3 See Mairal and Cortés-Rodríguez 2000–2001, Everett 2002, Cortés-Rodríguez 2006a, Martín Arista 2008, 2009 and Nolan 2011, for more details on all these resources. 4 Van Valin (2005: 161) offers a similar overview of the lexicon, when he states that ‘[i]t is necessary to think of the lexicon as having at least two parts, one the traditional storehouse of words and morphemes, and the second a “workshop” where lexical rules and other lexical processes create new lexical forms which would not otherwise be stored’. 5 There are some arguments against morpheme-based approaches to inflection and the lexicon among which the following are worth mentioning, albeit briefly. A morpheme-based approach considers morphemes as a linear sequence of phonemes which are attached to a base. However, morphosyntactic information can be encoded by other means such as tone, stress or nasalization across languages. Further problems for concatenative approaches occur when there is no one-to-one correspondence between meaning and form. While morpheme-based approaches assume that a morphological property has one exponent per word, realizational theories have no such requirement and are thus seemingly closer to the different possibilities found across languages. A case in point is portmanteau affixes like the Finnish nominal -t, which bears two concurrent features, plural number and nominative case (as in talot ‘houses’). In a linear item-andarrangement approach it would be necessary to subcategorize the nominative plural form as exceptional so that the grammar would not first select the plural suffix and then the case suffix in accordance with the usual morphotactics of the language. In a realizational approach -t is the cumulative exponent of two simultaneous features triggered by the corresponding realizational rule(s) (see Boutin 2011: 5–7). Again, realizational rules will account for this phenomenon in the morphology–phonology interface. 6 It is precisely here that the main difference between Martín Arista’s (2009, 2011) description of the LSW and Everett’s (2002) initial proposal lies: whereas Everett (2002) places inflection in the constituent projection of the word as daughters of the core layer (see Van Valin 2013: 112), Martín Arista (2009: 90) proposes instead to treat inflection as part of the operator projection, in which inflectional features are attached at the different layers within the LSW. 7 Their status as arguments is expressed by the nodes ARG in Figure 8.3. Although ARG nodes have been dispensed with in the LSC, Martín Arista (2011) makes use of them in the analyses of morphologically complex words. We have also opted to keep them in order to facilitate the interpretation for those readers who may not be familiar with syntactic analyses within the word.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

Morphology in RRG

8 Diedrichsen (2011: 5–10) offers a similar strategy for the distribution of operators in the LSW of German nouns and verbs. Following Bybee’s (1985) principle of relevance, she correlates the degree of formal fusion of an inflectional category and its bases with the semantic influence of such a category. According to this, the more relevant a category, the closer its degree of fusion or attachment with its base. Number, for example, has a strong impact on nouns, as it involves a multiplication of the concept it expresses. Because of this, stem modulation processes such as umlaut for number marking in German nouns (Lamm ‘lamb’/ Lämm-er ‘lamb-s’; Buch ‘book’/ Büch-er ‘book-s’) will be interpreted as a realization of a nuclear operator; on the other hand, number is less relevant within the verbal inflectional categories and concomitantly it will be encoded in the outer word layer. 9 Compare Russian atelic verbs like pisat’ ‘to write’ and pit’ to drink’, which are morphologically simple words, with telic na-pisat’ ‘to write (up)’, podpisat’ ‘to sign’ or vypit’ ‘to drink (up)’, all prefixed. Van Valin and LaPolla (1997: 90–112) offer a plethora of examples of this type of process from a wide array of languages. 10 See Roberts (2015) for a detailed description of the system of distributives in this language. 11 In the glosses of example (3) ds stands for ‘different subject following’ and su for ‘subject’. 12 See Van Valin and LaPolla (1997: 3–7) for a detailed description of (theory and language) internal and external explanations. 13 This view goes against the Lexical Integrity Hypothesis (see Lapointe 1981; Di Sciullo and Williams 1987; Anderson 1992, etc.), which stipulates that morphology is not visible to the rules of syntax. For arguments in favour of and against this hypothesis, see Ackema and Neeleman (2003) and Bosque (2012), respectively. In clear contrast to formalist modular approaches, in RRG all elements of the theory interact with one another, although their interaction is subject to various constraints. If one assumes Everett’s (2002) version of the LSW, as Van Valin (2013) does, only inflectional features of the COREw are visible to syntax, whereas the internal structure of the NUCLEUSw is not visible to syntax. So in nouns like cats and doorstops the syntax can ‘see’ the plural inflection, as it is relevant for agreement phenomena, but it cannot see that cats has a single element in its NUCLEUSw since the plural suffix -s is an argument in the COREw, but doorstops has a complex nucleus composed of two compounded NUCLEIw. In Martin Arista’s (2008, 2009, 2011) version of the LSW, in which inflectional features are treated as operators, syntax-blind features are analysed as NUCLEUSw operators. On the other hand, COREw and WORDw operators are visible to syntax and can access the LSC and the LSW, as shown in Figures 8.5, 8.6 and 8.9. 14 Among them are the works of Mairal Usón and Ruiz de Mendoza (2008, 2009), Mairal Usón, Periñan-Pascual and Pérez Cabello de Alba (2012),

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

401

402

FRANCISCO J. CORTÉS-RODRÍGUEZ

Ruiz de Mendoza (2013), Ruiz de Mendoza and Mairal (2007a, 2007b), Periñán-Pascual and Mairal Usón (2009) and Bentley (2018, 2019). 15 Möllemann (2016: 12–15) proposes a means to establish the meaning of a morpheme: it must be the semantic material left after removing the host word’s contribution to the semantics of the derived word. Since semantic information is encoded as event and qualia structure, the implication is that an affixal entry will have an incomplete part in the qualia set, to be filled once it is combined with a host lexeme. 16 Other formations of this type are acorralar (corral ‘pen, corral’ > a.corral. inf ‘to corner, to corral’), entronizar (trono ‘throne’ > en.tron(o).iz.inf ‘enthrone’), encarcelar (cárcel ‘jail’ >en.carcel.inf ‘to jail’), alunizar (luna ‘moon’ > a.lun(a).iz.inf ‘to land on the moon’). 17 See Johnston and Busa (1999) for a very similar analysis to NþN compounds within the framework of the Generative Lexicon theory.

https://doi.org/10.1017/9781316418086.011 Published online by Cambridge University Press

9 Adverbs, Mimetics and Ideophones Kiyoko Toratani

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ASP F IDEO IF L LSC

9.1

aspect factitive ideophone illocutionary force linker layered structure of clause

MI NUC P PP RP TNS

mimetic nuclear particle pre-/post-positional phrase reference phrase tense

Introduction

The syntax of adverbs has triggered much controversy both within and across theories (see, for example, Alexiadou 1997; Cinque 1999; Engels 2012; Ernst 2002, 2020).* The central concern includes the ability to offer a principled account of (i) which positions an adverb can occupy within a syntactic representation and (ii) what determines the ordering of multiple adverbs in sentence structure. Reflecting on recent developments in the generative approach to adverbial syntax, Ernst notes the critical consensus that ‘the semantics of individual adverbs is an important determinant – perhaps the main determinant of their ordering’ (2014: 108). While Ernst is considering generative syntax, the essence of his

* I am grateful to Robert Van Valin, Jr. and to a reviewer for their valuable comments; they greatly improved the quality of the paper. I thank James Watters for clarifying points on the Tepehua ideophones, and Janis Nuckolls for confirming the point on the Quechua data. Thanks also go to Elizabeth Thompson for her editorial suggestions. The remaining shortcomings are, of course, solely my responsibility.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

404

KIYOKO TORATANI

reflection has close affinity with how Role and Reference Grammar (RRG) analyses adverbs: it too considers semantics a critical component in theorizing the syntax of adverbs. This affinity may not be accidental, as a series of Ernst’s own work (e.g. 2002), characterized by him as a ‘scopal approach’ (Ernst 2014), is rooted in Jackendoff (1972), and RRG draws on insights from that same work. Assuming that adverbs can be defined as ‘modifiers of constituents other than nouns’ (Schachter 1985: 20), the first part of this chapter offers a foundational sketch of the RRG approach to adverbs using English data, drawing on Van Valin (2005) and Van Valin and LaPolla (1997), with a focus on their subsets, including aspectual adverb (completely), manner adverb (clumsily), temporal adverb (yesterday), evidential adverb (evidently), and other clausal adverbs. It is pointed out that adverbs modify distinct layers of the clause – the nucleus, the core, and the clause – and when the sentence contains multiple adverbs, their appearance corresponds to the order of the layers they modify in terms of the distance from the verb. The second part of the chapter moves to an exemplification of how the RRG approach can be used to analyse the ideophonic adverb, which is typically a type of manner adverb (Schachter 1985: 21).1 While the category status of ideophones remains disputed,2 it has frequently been noted that some function as adverbs, appearing in the same syntactic position as adverbs of prosaic words (Beck 2008: 41; Bobuafor 2013: 351; Ibarretxe-Antuñano 2006: 19; Nuckolls 1996: 72, among others). For instance, observing an example like (1), wherein the ideophone expresses a manner of running, Schaefer notes: ‘Emai IA [i.e. ideophonic adverb] forms assume the canonical syntactic position of adverbs. They occupy [the] postverbal position’ (2001: 341). (1)

Emai (Nigeria’s Edoid group; Schaefer 2001: 347) ómohe láí nyényényé óli the man run-f ideo (with a dash) ‘The man dashed off’/‘the man ran off with a dash’

Despite the pervasive presence of ideophones in the languages of the world (Voeltz and Kilian-Hatz 2001), virtually no work has analysed ideophonic adverbs from a cross-linguistic perspective applying a specific syntactic theory to the data. This chapter represents the first step toward closing the gap by presenting an initial RRG view of the syntax of ideophonic adverbs. The chapter is organized as follows. Section 9.2 lays out theoretical assumptions in RRG’s treatment of adverbs (Van Valin 2005; Van Valin and LaPolla 1997), explaining the syntactic behaviour and properties of adverbs. This is followed by a brief precis of the semantic representation of adverbs. Section 9.3 turns to characteristics of ideophonic adverbs. Section 9.4 contains concluding remarks.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

405

Adverbs, Mimetics and Ideophones

9.2

Theoretical Assumptions

9.2.1

Structural Representation of Clause Structure

9.2.1.1 Layered Structure of the Clause (LSC) RRG’s approach to adverbs is grounded in the conception of the layered structure of the clause (LSC). Figure 9.1 previews an example of the representation of the sentence, Leslie has evidently been slowly immersing herself completely in the new language. Figure 9.1 consists of two parts. The top part of the representation, called the ‘constituent projection’ of the LSC, is centred on three primary syntactic units: the clause (the outermost layer), the core (the mid layer), and the nucleus (the innermost layer). The clause contains the core, which, in turn, contains the nucleus and the arguments of the predicate, categorized as the ‘reference phrases’ (Van Valin 2008). The nucleus contains the predicate, usually a verb or an adjective. The peripheries contain a modifier phrase (MP) (Van Valin 2008: 172) whose nucleus is usually an adverb or an adjective. Each layer of the clause can be modified by a periphery, as indicated by an arrow in the figure. In Figure 9.1, SENTENCE CLAUSE

PERIPHERY PERIPHERY RP

MP

MP

COREM

COREM

NUCM

NUCM

ADV

ADV

CORE NUC

PERIPHERY

PRED RP

MP COREM

PP COREP

NUCP

ADV

CORER

PRED

NUCM

V

RP

PERIPHERYR P

Leslie has evidently been slowly immersing herself completely in

MP

ASP

N

IF

NUCR

ASP

CORER

NUC CORE

TNS

N

the new language.

V NUC

NUCR

CLAUSE CLAUSE SENTENCE

Figure 9.1 Example of LSC (modified from Figure 1.13, Van Valin 2005: 22)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

DEF

RP

406

KIYOKO TORATANI

evidently modifies the clause, slowly modifies the core, and completely modifies the nucleus. A brief note on the term ‘periphery’ is in order, as it is sometimes confused with a concept in prototype theory (e.g. Lakoff 1987), whereby membership in a category is characterized as ‘central’ (i.e. prototype) vs. ‘peripheral’ (non-prototypical), or given a configurational region of ‘central’ (i.e. core) vs. ‘peripheral’ (i.e. margin). As Figure 9.1 makes clear, in RRG, ‘periphery’ means neither a ‘non-prototypical member’ nor a ‘margin’ but refers to a structural unit containing a modifier such as an adverb (Van Valin 2005: 4). The bottom part of the representation in Figure 9.1 is called the ‘operator projection’. Operators refer to ‘grammatical categories . . . [that] modify the clause and its parts’ (Van Valin and LaPolla 1997: 40). They are posited to have scope over a specific layer of the clause: The nuclear operators have scope over the nucleus; they modify the action, event or state itself without reference to the participants. Core operators modify the relation between a core argument, normally the actor, and the action . . . Clausal operators, as the name implies, modify the clause as a whole. (Van Valin 2005: 8–9) Again, the arrows in the figure indicate the modificational relations. Incidentally, Figure 9.1 contains only three grammatical categories (ASP, TNS, IF (i.e. aspect, tense, illocutionary force)), but RRG posits nine operator categories: (i) aspect, (ii) negation, (iii) directional, (iv) event quantification, (v) modality, (vi) status,3 (vii) tense, (viii) evidentials, and (ix) illocutionary force (Van Valin 2005: 12). Each modifies a designated layer (or layers) of the clause, as represented in Figure 9.2: for instance, aspect is a nuclear operator. Some of the operator categories become relevant when we discuss the syntactic characteristics of ideophones.

NUC

Aspect Negation Directionals

CORE

Directionals Event quantification Modality Negation

CLAUSE

Status Tense Evidentials Illocutionary Force

SENTENCE Figure 9.2 Operator projection (adapted from Van Valin 2005: 12)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

9.2.1.2 Adverb Ordering Although adverbs are not operators (Van Valin 2005: 19),4 like operators, they are assumed to modify a specific layer of the clause, as exemplified in (2). (2)

Layer Nucleus: Core: Clause:

Examples of adverbs modifying the specific layer (English) aspectual: completely, continuously pace: quickly; manner: gently; temporal: yesterday epistemic: probably; evidential: evidently; evaluative: unfortunately

The distribution of these adverbs is semantically motivated. Nuclear adverbs modify the predicate in the nucleus. For instance, aspectual adverbs such as completely are nuclear adverbs, because they add information about the aspectual unfolding of the event denoted by the predicate. Core adverbs modify the semantic content of the event denoted by the predicate and the core argument(s). For instance, ‘subject-oriented’ adverbs such as clumsily are core adverbs, as they add information on how the event participant performs the action. Lastly, clause adverbs modify the entire proposition expressed by the clause. Evaluative adverbs such as unfortunately, a type of ‘speaker-oriented’ adverb, are clausal, as they are concerned with the speaker’s evaluation of the propositional content.5 A frequently noted aspect of adverbs in English is that they are not free to occur anywhere within the sentence to yield a given reading. Drawing on Jackendoff (1972) and McConnell-Ginet (1982), Van Valin (2005: 20) discusses the case of manner adverbs, which ‘interact in an important way with the tense operator’: that is, ‘those [adverbs] which occur before the tense operator can be construed as clausal modifiers, while those occurring after tense cannot be’ (Van Valin 2005: 20). This is illustrated in (3). (3)

a. b. c. d. e.

Ruth hid the cash cleverly. Cleverly, Ruth hid the cash. The manner in which Ruth hid the cash was clever. (¼3a) The fact that Ruth hid the cash was clever. (¼3b) Ruth cleverly hid the cash. (Van Valin 2005: 20)

(3a) has the adverb cleverly after tense. It can have the manner reading given in (3c) but lacks the reading of a clausal modifier given in (3d). By contrast, (3b) has the adverb before tense, in which case, the reading as a clausal modifier (3d) is available but the manner reading (3c) is absent. Meanwhile, (3e) locates the adverb immediately before the verb, generating both readings. If a sentence contains multiple adverbs of different types, their distribution is constrained by the ordering of the layers to which they are related. That is, ‘adverbs related to more outer operators occur outside of adverbs related to more inner operators’ (Van Valin 2005: 20). In the case of English, with its ability to place adverbs either preverbally or postverbally, the order of adverbs is affected by their position relative to the verb (cf. Ernst 2002: 154, 2014: 114–115).

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

407

408

KIYOKO TORATANI

First, consider the case of the core adverb occurring preverbally. (4) shows the predicted order of three types of adverbs, and (5) shows different patterns of a sentence containing the three adverbs in different positions. (4)

Order of adverbs when the core adverb precedes the verb: clausal > core > nuclear e.g. evidently [evidential: clausal] > slowly [pace: core] > completely [aspectual: nuclear]

(5)

a. b. c. * d. * e. * f. * g. *

Evidently, Leslie has slowly been completely immersing herself in the new language. Leslie has evidently been slowly immersing herself completely in the new language. Evidently, Leslie has completely been slowly immersing herself in the new language. Slowly, Leslie has evidently been completely immersing herself in the new language. Slowly, Leslie has completely been evidently immersing herself in the new language. Completely, Leslie has evidently been slowly immersing herself in the new language. Completely, Leslie has slowly been evidently immersing herself in the new language. Van Valin (2005: 20)

In (5a), the three adverbs appear preverbally, whereas in (5b), the clausal adverb evidently and the core adverb slowly occur preverbally, but the nuclear adverb completely is postverbal. The adverbs in (5a)–(5b) observe the predicted order, rendering the sentences grammatical. In contrast, (5c)–(5g) are judged ungrammatical, as the adverbs disobey the predicted order given in (4): for example, in (5c), when the nuclear adverb completely precedes the core adverb slowly, the sentence is infelicitous; in (5d), when the core adverb slowly precedes the clausal adverb evidently, the sentence is similarly infelicitous. A similar situation arises when the core adverb occurs postverbally. (6) shows the predicted order, and (7) gives examples with adverbs occurring in different positions. (6)

Order of adverbs when the core adverb follows the verb: nuclear < core < clausal e.g. completely [aspectual: nuclear] < slowly [pace: core] < evidently [evidential: clausal]

(7)

a.

Leslie has been immersing herself completely in the new language slowly evidently. b. Leslie has been completely immersing herself slowly in the new language evidently. c. * Leslie has been immersing herself slowly in the new language completely evidently.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones d. * Leslie has been immersing evidently slowly. e. * Leslie has been immersing completely slowly. f. * Leslie has been immersing evidently completely. g. * Leslie has been immersing slowly completely.

herself completely in the new language herself evidently in the new language herself slowly in the new language herself evidently in the new language Van Valin (2005: 20–21)

Among the different patterns, (7a) and (7b) are felicitous, wherein the three adverbs appear as nuclear, core, and clausal adverbs, in that order, toward the end of the sentence (although the nuclear adverb appears before the verb in (7b)). In contrast, the sentences in (7c)–(7g) are infelicitous; they do not observe the order in (6). For instance, in (7c), the nuclear adverb completely occurs between the core adverb slowly and the clausal adverb evidently, and in (7d) the core adverb slowly appears after the clausal adverb evidently. These examples show that the ordering constraints work in both directions, substantiating the argument that ‘the scope constraints require that the nuclear adverb be closer to the verb than the core adverb, and likewise for the core adverb with respect to the clausal adverb’ (Van Valin 2005: 21). These relationships are readily depicted in the LSC. Figure 9.3 shows the

SENTENCE PERIPHERY

CLAUSE CORE

PERIPHERY MP

PERIPHERY

NUC

RP

RP

COREP

MP

RP

NUCP

PRED

COREM

V

NUCM ADV P

MP

V ASP

IF

N

COREM

NUCR

NUC

CORER DEF

RP

CLAUSE CLAUSE SENTENCE

Figure 9.3 Structure of (7a) (modified from Figure 1.14, Van Valin 2005: 22)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

ADV

NUCM ADV

language slowly evidently. N

ASP

CORE TNS

COREM

CORER NUCM PRED PERIPHERYR NUCR

Leslie has been immersing herself completely in the new NUC

MP

PP

409

410

KIYOKO TORATANI

constituent and operator projection of (7a) (see Figure 9.1 for the syntactic representation of (5b)). Although the examples discussed to this point are English, the fundamental point about adverb order is expected to be cross-linguistically valid. That is, regardless of which language is under investigation, the relative order of adverbs should observe the order of the layers they modify: for example, adverbs modifying the entire clause should appear away from the predicate, and adverbs modifying the predicate should appear in closer proximity to the predicate. It is worth pointing out that temporal adverbs such as tomorrow (in their default position) are core modifiers, not clausal modifiers, although the meaning of temporal adverbs is related to tense, and tense is a clausal operator (Van Valin 2005: 19; Van Valin and LaPolla 1997: 32, 426–428). This point is elaborated by Bohnemeyer and Van Valin (2017), who note that ‘verbal cores are inherently constituents that describe (sub)events and that each core . . . has its own time-positional modifier’ (2017: 159). For instance, in (8), on Monday modifies the event of persuading, whereas on Friday modifies the event of visiting. (8)

Tom persuaded Sally on Monday to visit her sister on Friday. (Bohnemeyer and Van Valin 2017: 159)

It should be emphasized that the occurrence of temporal adverbs in the periphery is a default case; they can also occur in the ‘pre-detached position’, which is ‘outside of the clause but within the sentence’ (Van Valin and LaPolla 1997: 36). For example, consider (9). (9)

a. Sam decided to leave tomorrow. b. Yesterday, John did not show the book to Mary. c. *Tomorrow, Sam decided to leave. (Van Valin and LaPolla 1997: 170)

In (9a), tomorrow appears in the default position, whereas in (9b), yesterday appears in the pre-detached position. When a temporal adverb appears in the latter position, it is no longer contained in the periphery and no longer modifies the core: ‘If a temporal adverb occurs in the left [i.e. pre]-detached position, then it is a clausal modifier’ (Van Valin and LaPolla 1997: 170). This point can be illustrated by comparing (9a) and (9c). If a temporal adverb always remains a core modifier regardless of its position within the sentence, tomorrow in (9a) should be able to be fronted, yielding the meaning of (9a). However, this is not the case. As indicated by the unacceptability of (9c), if the temporal adverb tomorrow appears in the pre-detached position, the ensuing clausal reading implies that the event of deciding will take place tomorrow. However, this turns out to be incompatible with the information of tense, which is past. Now that we have noted the modificational unit of temporal adverbs, it may be worth asking about possible constraints when multiple adverbs

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

modify the same layer of the clause. In fact, when discussing examples like (10), Van Valin (p.c.) suggests semantic-based constraints may be at work; more specifically, if three kinds of core adverbs co-occur (as in (10): a pace adverb quickly, a manner adverb carefully, and a temporal adverb yesterday), ordering constraints appear. (10)

a. b. c. d. e. f. g. h. j.

John carefully opened the box quickly yesterday. ?? John carefully opened the box yesterday quickly. John quickly opened the box carefully yesterday. ?? John quickly opened the box yesterday carefully. * John yesterday opened the box carefully/quickly. John opened the box carefully and quickly yesterday. * John opened the box carefully and yesterday. * John opened the box quickly and yesterday. Yesterday John carefully opened the box quickly.

First, with respect to position, the pace adverb and the manner adverb seem to be used interchangeably ((10a) vs. (10c)). Second, the pace adverb and the manner adverb can be conjoined but neither can be conjoined with the temporal adverb ((10f) vs. (10g/10h)). Third, in the postverbal position, the temporal adverb should follow the manner adverb ((10c) vs. (10d)), and preverbally, the temporal adverb should precede the manner adverb (10j). The differences in acceptability in (10) strongly suggest some semantic layering, at least within the core; however, a closer investigation is called for, with more data.

9.2.2 Semantic Representation of Adverbs Adverbs are represented by logical structures of ‘one-place predicates which take a logical structure or subpart of a logical structure as their argument, following the approach by Jackendoff (1972) and others’ (Van Valin and LaPolla 1997: 162; cf. Van Valin 2005: 49). Which part of the logical structure an adverb takes as its argument depends on the meaning of the adverb, given its syntactic position within the sentence. A few examples are given in (11). (11)

a. Yesterday, Chris ran to the park. a′. yesterday′ (do′ (Chris, [run′ (Chris)]) & INGR be-at′ (park, Chris)) b. Pat elegantly closed the door slowly. b′. [elegant′ (do′ (Pat, Ø))] CAUSE [slow′ (BECOME closed′ (door))] c. The ice melted completely/The ice completely melted. c′. BECOME complete′ (melted′ (ice)) (Van Valin 2005: 49)

Temporal adverbs like yesterday in (11a) take the entire logical structure as their argument (Van Valin 2005: 49). Manner adverbs like elegantly in (11b) take the activity component as their argument (indicating the elegant

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

411

412

KIYOKO TORATANI

manner in which Pat executed the action of closing), whereas the pace adverb slowly in (11b) takes the accomplishment component as its argument (indicating the slowness of the door’s closing event). Finally, in (11c), the aspectual adverb completely takes the basic state component melted′ (x) as its argument. It is left for future work to evaluate whether this uniform treatment of adverbs, that is, positing a one-place predicate as their semantic representation, can fully account for the characteristics of adverbs across the board, either within a given language or cross-linguistically.

9.3

Syntax of Ideophonic Adverbs

As noted in Section 9.1, some ideophonic adverbs can express manner (Schachter 1985: 21). This leads to the question of whether they pattern in the same fashion as manner adverbs like clumsily: that is, whether ideophonic adverbs are also core or clausal modifiers, as discussed in Section 9.2. In answering this question, we first provide preliminary information about Japanese ideophones, called mimetics, as the ensuing discussion largely relies on Japanese data (Section 9.3.1). Second, we use syntactically relevant semantic criteria to classify ideophones into core and nuclear adverbs (Section 9.3.2). Third, we examine whether the classification holds if the ideophone appears in a given syntactic environment (Section 9.3.3). Finally, we turn to a case where ideophones are not contained in the periphery, drawing on data from a Totonac-Tepehua language (Watters 2013) (Section 9.3.4).

9.3.1 Japanese Mimetic Adverbs Japanese mimetic adverbs can be classed into three major forms, which we will call ‘singleton’, ‘reduplicated’, and ‘ri-suffixed’, exemplified in (12). (12)

singleton:

reduplicated:

ri-suffixed:

kon botat een kurukuru batabata betabeta sukkari yukkuri pittari

‘a knock’ ‘a drip’ ‘a short crying voice’ ‘spinning’ ‘continuously falling’ ‘sticky’ ‘entirely’ ‘slowly’ ‘tightly’

Aspect is argued to be part of the lexical meaning of the first two forms (Akita 2009; Toratani 2005). Singletons typically express an event/state lasting a short time, such as a semelfactive event, that is, a single-staged event with no outcome (Smith 1997: 29) (e.g. kon ‘sound of a knock’), an

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

event with a culmination point (e.g. bokit ‘a snap; a manner of a sudden breakage of an object’), or an atelic event lasting a short time (e.g. een ‘a short-lasting crying voice’). They are repeatable: for example, botat ‘a drip (of liquid)’ can be repeated, as in botat botat ‘drip, drip’, or botat botat botat ‘drip, drip, drip’ and so forth, with a phonological break in between. Reduplicated forms such as kurukuru ‘manner of continuous spinning’ have a fully reduplicated structure with no word-internal phonological break. They can express the iteration of a cyclic (e.g. kurukuru ‘iteration of a spinning of an object’) or telic event (e.g. batabata ‘continuously falling’), or the continuation of a dynamic (e.g. ziroziro ‘continuation of one’s staring event’) or static condition (e.g. betabeta ‘continuation of a sticky bodily sensation’). Ri-suffixed mimetics (ending in the form -ri) typically express a non-eventspecific condition, such as sikkari ‘firmly’ or yukkuri ‘slowly’, and are ‘deideophonized’ (cf. Dingemanse 2017), in the sense that they lack the expressivity and the aspectual character present in many singletons and reduplicated forms. When mimetic adverbs occur in a sentence, some forms are marked by to ‘quotative (elsewhere)’ (to-marked) or not marked by to (Ø-marked), depending on the morphological shape of the mimetics, as exemplified in (13). (13)

a. Zyaguti-kara sizuku-ga potot-to (*Ø) oti-ta. faucet-from drop-nom mimetic-p drip-pst ‘A drop (of water) dripped from the faucet potot.’ b. Zyaguti-kara sizuku-ga potot potot-to (*Ø) oti-ta. faucet-from drop-nom mimetic-p drip-pst ‘(Two) drops (of water) dripped from the faucet potot potot.’ c. Zyaguti-kara sizuku-ga potopoto-to/Ø oti-ta. faucet-from drop-nom mimetic-p drip-pst ‘Drops (of water) dripped continuously from the faucet.’ d. Zyaguti-kara sizuku-ga yukkuri-to oti-ta. faucet-from drop-nom mimetic-p drip-pst ‘Drops (of water) dripped slowly from the faucet.’

On the one hand, singletons are obligatorily marked by to, regardless of how many times the form is instantiated: once (13a), twice (13b), and so forth. On the other hand, reduplicated forms (13c) and ri-suffixed forms (13d) can be marked by to or Ø, yielding no truth-conditional differences between the alternative markings (Hamano 1998, among others; see Akita and Usuki 2016 for a discussion of ‘optionality’ of to-marking). As (13) suggests, the most frequently occupied position of mimetics of any morphological shape is immediately preverbal (cf. Toratani 2017: 36–41). However, because Japanese, an OV language, has a flexible word order, clause-internal phrases, including mimetics, can scramble rather freely as long as the verb comes at the end of the clause.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

413

414

KIYOKO TORATANI

9.3.2 Syntactically Relevant Semantic Criteria Certain grammatical categories and the concept of hyponymy provide a semantic diagnostic for determining which layer of the clause the ideophone potentially modifies. 9.3.2.1 Grammatical Categories As Section 9.2 notes, operators modify a specific layer of the clause (see Figure 9.2). If an ideophone expresses the same concept as an operator, it is reasonable to posit that it modifies the same layer of the clause. The question is whether an ideophone can express any concept equivalent to that expressed by an operator. The answer is yes, at least in the following two grammatical categories. The first category is event quantification, which expresses multiple actions of the verb and is classed as a core operator. As (14) shows, some Japanese mimetics express the concept of event quantification. (14)

a. Hito-ga heya-ni hait-te-ki-ta. person-nom room-dat enter-l-come-pst ‘A person came into the room.’ b. Hito-ga zorozoro heya-ni hait-te-ki-ta. person-nom mimetic room-dat enter-l-come-pst ‘People came into the room in line one after another.’

(14a) represents the base sentence with no mimetic. The noun hito ‘person/ people’ has an ambiguous reading, in that it can be singular or plural, as nouns are not obligatorily marked for number in Japanese; yet the default interpretation of the sentence is a single event of coming in. In contrast, (14b)’s mimetic zorozoro implies multiple events of coming in, thereby depicting the motion of a crowd of people lining up and moving forward. Therefore, this mimetic is analysed as a core modifier, as it expresses event quantification. Other languages have ideophones lexically encoded with the notion of mass, as exemplified below. ˇ

a. Chichewa (Bantu; Kulemeka 1993: 224) ‘gather in a mass’ u:nji: b. Luwo (Western Nilotic, South Sudan; Storch 2014: 44) ‘a lot of people/cattle moving together’ m ̀ rm ̀ r c. Siwu (Ghana-Togo Mountain; Dingemanse 2011: 48) ɣ`e`e`e ‘animals swarming in great numbers’ c

(15)

c

The sentence from Pastaza Quechua in (16) makes an analogous point. It contains dzhawww, a variant of dzawn, meaning an ‘action, process, or event that involves a clustering together of individual agents, such as people, birds, bats, or insects’ (Nuckolls 1996: 148). (16)

Pastaza Quechua (Ecuador; Nuckolls 1996: 149) Putan dzhawww hatari-ra! fly ideo rise up-pst ‘The flies rose up dzhawww!’

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

According to Nuckolls, ‘dzawn describes the way flies that had gathered on a molting snake rose up in a swarm when a person approached’ (1996: 149). The fact that putan ‘fly’, glossed in the singular, is interpreted as plural in (16) suggests the ideophone plays a role in assigning the mass interpretation. If this is the case, the ideophone in (16) is a core modifier. The second grammatical concept applicable to ideophones is aspect. As noted in (2), aspectual adverbs are assumed to be nuclear modifiers (Van Valin 2005: 49). Examples are continuously and completely. The assumption is that ‘aspectual adverbs modify the basic state or activity predicate’ (Van Valin 2005: 49): e.g. melt completely: BECOME complete′ (melted′ (x)). If ideophonic adverbs can express a concept similar to completely and continuously and modify the basic state or activity predicate, it seems reasonable to posit them as nuclear modifiers. Some languages seem to have them, or at least the English glosses suggest this possibility (e.g. wic ‘eat all up’ (Luwo, Storch 2014: 47), psí:tí: ‘completely finished’ (Chichewa, Kulemeka 1993: 252), tdip ‘manner of covering completely’ (Didinga, a Southwest Surmic language, Sudan, de Jong 2001: 133)). In Japanese, no mimetics precisely express the meaning of completely and continuously. However, as discussed in Section 9.3.3.2, some mimetics affect the interpretation of the lexical aspect of the verb when they occur adjacent to the verb and, as such, seem to qualify as possible nuclear modifiers. ˇ

9.3.2.2 Hyponymy The second semantic criterion to determine which layer of the clause ideophones modify is hyponymy. It has long been noted that some Japanese mimetics co-occur with a limited set of verbs (Hirose 1981). For instance, tekuteku ‘plodding’ or yotiyoti ‘toddling’ typically co-occur with aruku- ‘walk’. In such cases, the mimetic–verb relationship can be characterized in terms of hyponymy (Toratani 2007: 325–327), where the mimetic is considered the hyponym and the verb is the hyperonym of the mimetic, as tekuteku ‘plodding’ and yotiyoti ‘toddling’ both express a kind of a walking event.6 In other languages as well, some ideophones collocate or co-occur with a limited number of verbs/adjectives (Childs 1994: 188; Creissels 2001: 78). For instance, in Didinga (a Southwest Surmic language in Sudan), the ideophone ɪðaatʃ ɪðaatʃ ‘manner of swallowing easily’ co-occurs with kú ‘swallow’ (de Jong 2001: 136), while in Emai (Nigeria’s Edoid group), the ideophone ghóighói ‘glistering-ly’ co-occurs with jín ‘shine’ (Schaefer 2001: 351), with the former apparently the hyponym of the latter in each pair. The hyponymy relationship highlights various ways in which ideophones and core adverbs modify the predicate’s event. Regular core adverbs can cooccur with a wide variety of predicates (clumsily walk/dance/write/sit etc.), as they refer to a general property common to a group of predicates, such as dynamism, conveying how the event participant performs the action without changing the meaning of the action itself; meanwhile, when ideophones co-occur with a limited set of predicates, they change the reading of the

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

415

416

KIYOKO TORATANI

action from a more or less neutral manner (e.g. ‘walking’) to a much more detailed one (e.g. ‘toddling’). To interpret this difference in terms of the layer of the clause the adverbs modify, we could say the semantic function of ideophones closely parallels that of nuclear adverbs, in that their functions are in line with the nuclear operators’ function of ‘modify[ing] the action, event or state itself without reference to the participants’ (Van Valin 2005: 8–9). This parallelism suggests ideophones that co-occur with the verbs/ adjectives of their hyperonymous category are nuclear adverbs, unless they refer to multiple event participants, in which case, the ideophones are core modifiers.7 The preceding discussion may give the impression that all ideophones are hyponyms (cf. Watson 2001: 393). This is not the case. To draw again on Japanese data, while some mimetics are clearly hyponyms (candidates for nuclear adverbs) as exemplified in (17), others are non-hyponyms (candidates for core adverbs).8 a. b. c. d.

Hyponyms yotiyoti musyamusya turuturu guruguru

a. b. c. d.

Non-hyponyms yukkuri sot syonbori bonyari

(17)

(18)

(candidates for nuclear adverbs) ‘toddle-toddle’ ‘munch-munch’ ‘slippery-slippery’ ‘spinning round and round’ (candidates for core adverbs) ‘slowly’ ‘gently’ ‘dispiritedly’ ‘absentmindedly’

Non-hyponyms in (18) include a pace adverb, yukkuri ‘slowly’, and several ‘subject-oriented’ adverbs. The latter category includes general manner adverbs such as sot ‘gently’ and a subset of ‘psychomimes’ (Martin 1975: 1025) (those expressing a psychological state) such as bonyari ‘absentmindedly’ (18d). An entailment test can be used to distinguish hyponyms from non-hyponyms. If the utterance of an ideophone entails the occurrence of a specific set of events, it can be judged a hyponym; otherwise, it is a nonhyponym. For instance, the utterance of the hyponymous mimetic guruguru ‘spin-spin’ entails that something continuously rotates and is readily associated with a verb such as mawaru ‘turn’, but the utterance of the nonhyponymous mimetic yukkuri ‘slowly’ does not entail a specific event and thus disallows the instantaneous association available with the hyponymous mimetic. Other languages seem to have non-hyponymous types of ideophones as well. For instance, Patent (1998: 196–197) discusses the case of Lai Chin (SinoTibetan), noting that while typical ideophones have elaborate meanings, some have very general meanings and can co-occur with a number of different verbs, such as tsiam-maam ‘with effort/forcefully’ and leŋmaŋ ‘always’.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

To sum up, we make four observations, expressed in (19), albeit with a proviso: as all observations are based on meaning, they may not apply if an overriding syntactic factor comes into play. (19)

Observation 1: Observation 2: Observation 3:

Observation 4:

An ideophone that expresses event quantification is a core adverb. An ideophone that affects the aspectual interpretation of the basic state or activity predicate is a nuclear adverb. A hyponymous ideophone that co-occurs with a clause-mate verb/adjective from its hyperonym category without making reference to a plural number of event participants is a nuclear adverb. A non-hyponymous ideophone is a core adverb.

These observations imply that ideophones are not clausal modifiers. This possibility can be confirmed with Japanese data. As noted in Section 9.2.1.2, typical clausal adverbs express the speaker’s evaluation or judgement of a propositional content (e.g. unfortunately, probably). Mimetics do not seem to convey these concepts. Furthermore, like mimetics, the large majority of ideophones express sound, manner, and the state of an entity, detailing the actions and states expressed by the clause-mate verbs/adjectives, but they do not seem to convey any concepts paralleling those conveyed by clausal adverbs. This suggests that ideophones are not clausal adverbs cross-linguistically as far as the meaning is concerned, but this suggestion requires validation.9

9.3.3 Syntactic Positions of Ideophonic Adverbs As the observations in (19) are based on meaning, the next question is whether a syntactic unit modified by an ideophone can be maintained when it actually appears in a sentence. To answer this question, we begin by considering whether core adverbs can occur as clausal adverbs, just as English manner adverbs such as clumsily can be core or clausal modifiers (Section 9.3.3.1). Next, we examine whether nuclear adverbs can occur as core adverbs (Section 9.3.3.2). 9.3.3.1 Possibility of a Core Adverb Occurring as a Clausal Adverb The first question is whether ideophonic adverbs expressing manner can yield an alternative evaluative (clausal) reading when they change position within a sentence, just like English manner adverbs such as clumsily. (20) shows this type of alternation is unavailable to Japanese mimetics. It is assumed that the mimetic batan ‘a bang’ is a core adverb, as it is not a hyponym of sime- ‘close’. Rather, it refers to the sound of the object (door), reenacting the event of the door’s having been shut. (20)

a. Taroo-ga to-o batan-to Taro-nom door-acc mimetic-p ‘Taro closed the door with a bang.’

sime-ta. close-pst

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

417

418

KIYOKO TORATANI

b. Taroo-ga batan-to to-o Taro-nom mimetic-p door-acc ‘Taro closed the door with a bang.’ c. Batan-to Taroo-ga to-o mimetic-p Taro-nom door-acc ‘With a bang Taro closed the door.’

sime-ta. close-pst sime-ta. close-pst (adapted from Toratani 2007: 11–12)

Example (20) contains the mimetic batan ‘(sound of ) a bang’ in three different preverbal positions. As the identical gloss (with a bang) indicates, it yields only one reading, that of manner, irrespective of where the mimetic occurs within the sentence. This unambiguity of reading is supported by the paraphrasability of the meaning of the mimetic into a manner reading (21a) but not into an evaluative (i.e. clausal adverb) reading (21b). (21)

a. Taroo-no to-no sime-kata-wa batan-to-dat-ta. Taro-gen door-gen close-way-top mimetic-p-cop-pst ‘The way Taro closed the door was with a bang.’ b. *Taroo-ga to-o sime-ta-no-wa batan-to-dat-ta. Taro-nom door-acc closed-pst-nmlz-top mimetic-p-cop-pst ‘The fact that Taro closed the door was with a bang.’ (adapted from Toratani 2007: 12)

In other words, the mimetic remains a core modifier and cannot be clausal, even if it changes syntactic position. This leads to the following two predictions. First, the mimetic should be able to co-occur with a clausal adverb: more specifically, because the mimetic and the clausal adverb modify distinct layers of the clause, they will not cause a semantic clash, unlike two clausal adverbs as in evidently and probably in *Evidently, John probably left (Jackendoff 1972: 87). Second, the mimetic should observe the order constraint with respect to the clausal adverb. That is, the clausal adverb should occur first, followed by the mimetic. (22) and (23) illustrate these points (the mimetic gatyan ‘sound of a clank’ is a core modifier as it is not a hyponym of otosu ‘drop’). (22)

a. Bukiyooni-mo clumsy-p ‘Clumsily (he) b. Bukiyooni-mo clumsy-p ‘Clumsily (he)

kabin-o gatyan-to otosi-te.simat-ta. vase-acc mimetic-p drop-asp-pst dropped the vase with a clank.’ gatyan-to kabin-o otosi-te.simat-ta. mimetic -p vase-acc drop-asp-pst dropped the vase with a clank.’

(23)

?? Gatyan-to bukiyooni-mo kabin-o otosi-te.simat-ta. mimetic-p clumsy-p vase-acc drop-asp-pst ‘With a clank clumsily (he) dropped the vase.’ (adapted from Toratani 2007: 13–14)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

The first point is illustrated in (22) by the ability of the mimetic gatyan ‘sound of a clank’ to co-occur with the clausal adverb bukiyooni-mo ‘clumsily’.10 The second point (word order) is also illustrated in (22) by the expected word order: the clausal adverb bukiyooni-mo ‘clumsily’ precedes the mimetic (core/nuclear adverb). This is further supported in (23) by the mimetic’s inability to cross the clausal adverb. To reiterate, unlike English manner adverbs like clumsily, mimetic adverbs do not yield a clausal reading, even if they change syntactic position. Though verification is necessary, this seems to apply to ideophones across languages.

9.3.3.2 Possibility of a Nuclear Adverb Occurring as a Core Adverb Next, we examine whether ideophones that are semantically determined as nuclear adverbs (those expressing aspect and/or all hyponymous ideophones co-occurring with their clause-mate predicates from their hyperonymous categories) are always nuclear adverbs, regardless of their position within the sentence. To explore this question, we use Japanese data, drawing on Tsujimura and Deguchi’s (2007) observation that the acceptability of the sentence with a for/in-phrase (24a) is affected if a mimetic is added (24b). (24)

a. Mizu-o gohunkan/gohun-de nonda. water-acc for/in 5 minutes drank ‘I drank water for/in five minutes.’ b. Mizu-o gohunkan/*?gohun-de gokugoku nonda. water-acc for/in 5 minutes mimetic drank ‘I drank water (repeatedly) for/*?in five minutes.’ (Tsujimura and Deguchi 2007: 345)

Example (24a) is a simple sentence containing a transitive verb nomu ‘drink’ and a for/in-phrase, diagnostics for atelicity and telicity, respectively. The verb can co-occur with either phrase, making it either atelic or telic. In contrast, in (24b), the verb must be interpreted as atelic when the mimetic gokugoku ‘gulp-gulp’ is present, as indicated by the acceptability of the forphrase but the unacceptability of the in-phrase. Tsujimura and Deguchi note that ‘the sense of repetition associated with reduplicated mimetics . . . affects the telicity of the sentences in which they occur’ (Tsujimura and Deguchi 2007: 344). In RRG terms, this sensitivity of the mimetic to the verbal aspect implies that the mimetic is a nuclear adverb. This, however, does not mean mimetics always function as nuclear adverbs. Observing the pattern in (24), discussed in Tsujimura and Deguchi (2007: 344), Toratani (2007) points out that the mimetic compatible with the atelicity reading of the verb appears in a particular environment. First, the mimetic gokugoku is Ø-marked. Second, it occurs in the immediately preverbal position. Third, the verb comes from the mimetic’s hyperonym category. Of these, if the first two conditions are altered, the mimetic no longer participates in specifying the verb’s aspect, as illustrated in (25), in

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

419

420

KIYOKO TORATANI

which the to-marked version of the mimetic moved out of the immediately preverbal position. Kodomo-ga gokugoku-to mizu-o gohunkan/gohun-de non-da. child-nom mimetic-p water-acc for/in 5 minutes drink-pst ‘The child, in a gulping manner, drank the water in/for five minutes.’ (Toratani 2007: 333)

(25)

The mimetic gokugoku in (25) is reduplicated, expressing aspectual continuity. Furthermore, it co-occurs with a verb of its hyperonymous category, thus satisfying the semantic criterion for nuclear adverbs. However, it does not affect the aspectual reading of the verb; thus, satisfying the semantic criterion alone (Observations 2 and 3 in (19)) is insufficient to claim that the mimetic is a nuclear adverb. The mimetic must satisfy the morphosyntactic condition: it must appear in the immediately preverbal position with no elements intervening between it and the verb. Although a more thorough investigation is necessary, those mimetics and ideophones affecting the aspectual readings of their clause-mate predicates can be posited to be nuclear adverbs; otherwise they are core adverbs. In light of this point, the semantic-based observations in (19) can be revised as (26), which is a working hypothesis for ideophones that co-occur with the predicates of their hyperonymous categories (without making reference to a plural number of event participants). (26) The ideophone that specifies the aspect of the predicate is a nuclear adverb; all other ideophones are core adverbs.

Ideophonic nuclear adverbs likely occur adjacent to the predicate belonging to the hyperonymous category of the ideophone, but this requires confirmation in individual languages. Figure 9.4 represents a nuclear adverb, and Figure 9.5 shows a core adverb (the internal layers of RPs, PPs and MPs are simplified; the operator projections are omitted).

SENTENCE CLAUSE CORE

PERIPHERY RP

PP PERIPHERY MP MI-ADV

NUC PRED V

Mizu-o gohunkan gokugoku non-da. water-ACC for.5.min MIMETIC drink-PST ‘The child gulped water for five minutes.’ Figure 9.4 The mimetic as a nuclear adverb (adapted from Toratani 2007: 333)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

SENTENCE CLAUSE CORE

PERIPHERY RP

MP

RP

PP

NUC PRED

MI-ADV V Kodomo-ga gokugoku-to mizu-o gohunkan/gohun-de non-da. child-NOM MIMETIC-P water-ACC for/in 5 minutes drink-PST ‘The child, in a gulping manner, drank the water in/for five minutes.’ Figure 9.5 The mimetic as a core adverb (modified from Toratani 2007: 334)

As Figures 9.4 and 9.5 show, when they are at clause-internal positions, mimetics are contained in the periphery and modify a specific layer of the clause.

9.3.3.3 Nuclear-Internal Modifier The foregoing discussion suggests ideophonic adverbs modify the nucleus and the core, unlike the clumsy-type manner adverbs, which can modify the core and the clause. Recent RRG work by Watters (2013) suggests ideophones can fall into another type, not found in the pattern of ideophonic adverbs discussed so far. Example (27) shows how ideophones are used in a Totonac-Tepehua language. Noting that nothing can intervene between the ideophone and the verb, Watters (2013) proposes a schematic structure (28) capturing the syntactic tightness of ideophone and verb. (27)

Tlachichilco Tepehua (Totonacan; Watters 2013: 32) a. sk’uli ʔu-y ki-makaː itchily eat-ipfv 1poss-hand ‘it itches my hand’ b. spuy tsuku-y xkaːn in.drops be-ipfv water ‘it’s sprinkling’ c. ʃtay ʔan-Ø circularly go-ipfv ‘it rotates’

(28) [[X]ADV:IDEOPH [Y]VERB]NUC

Interestingly, in these examples, the ideophones co-occur with a very particular verb; unlike the case of Japanese mimetics, however, the verbs do not belong to the hyperonymous category of ideophones, such as ‘scratch’ for ‘itchily’ and ‘turn’ for ‘circularly’. According to Watters (p.c.), the language has ‘a much more limited verb vocabulary compared to English’ and must, therefore, recruit semantically lighter verbs, such as ‘be’ and ‘go’, to partner

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

421

422

KIYOKO TORATANI

with ideophones, with the ideophones assuming the more substantial semantic load.11 This implies reconceptualization is necessary to characterize the structural position of ideophones in (27) since, unlike the subsumption relation noted for Japanese mimetics, the verb can no longer subsume the semantics of the ideophone. Absence of the subsumption relation in (27) indicates the ideophone becomes a critical component in the expression of a given meaning. As Watters (p.c.) notes, the ideophones in (27) are, in theory, adjuncts, in that the sentence remains grammatical even without them, but the omission of the ideophones drastically changes the meaning, suggesting they are required to portray the intended event type. How should (28) be represented in terms of the LSC? There seem to be three possibilities, as shown in Figure 9.6, using (27c) as an example. First, Figure 9.6(c) models according to the representation of nuclear adverbs of prosaic words, in which the adverbs are adjuncts. Since the ideophone in (27c) is the required part of the sentence to convey the intended meaning, this cannot be the correct representation. Next, Figure 9.6(b) indicates that the ideophone and the verb are each dominated by a PRED node of their own, thus implying each element expresses its own meaning: that is, something ‘(moves) circularly’, and it ‘goes’. This cannot be the correct representation either, because the ideophone and the verb are semantically dependent on each other to express the intended meaning of ‘it rotates’. This semantic dependency of the ideophone and the verb suggests they form a complex predicate, which, in turn, implies that Figure 9.6(a) is the correct representation, with the ideophone and the verb dominated by the same PRED node in the nucleus. The representation is also consistent with the understanding that nothing can intervene between the ideophone and the verb. To sum up, the Tepehua ideophones in (27) lend themselves to a category distinct from the ideophonic adverbs discussed in the previous subsections. They enter into a particular semantic relationship with the verbs. The verbs with which they co-occur are not as ‘light’ as DO/SAY verbs, often said to (a)

SENTENCE

(b)

(c)

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CLAUSE

CORE

CORE

CORE

NUC

NUC

PRED IDEO-ADV

PRED V

tay ancircularly go-IPFV ‘it rotates’

PERIPHERY PRED

IDEO-ADV

V

tay circularly ‘it rotates’

ango-IPFV

Figure 9.6 Three possible representations for (27c)

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

MP IDEO-ADV

NUC PRED V

tay ancircularly go-IPFV ‘it rotates’

Adverbs, Mimetics and Ideophones

be required by ideophones to function as a predicate (Childs 1994: 187) (e.g. nikoniko-suru smilingly-do ‘smile’, an example from Japanese); nor are they as ‘heavy’ as verbs in the ideophones’ hyperonym category, the pattern commonly observed in Japanese mimetics (e.g. nikoniko warau [smilingly laugh] ‘smile’). In other words, the Tepehua ideophone–verb semantic relationship constitutes an intermediary type, a category thus far undiscussed.

9.4

Concluding Remarks

This chapter outlines the RRG approach to adverbs, paying particular attention to ideophonic adverbs. It has introduced relevant RRG assumptions on adverbs: (i) that adverbs contained in the periphery and represented in the ‘constituent projection’ of the layered structure of the clause may modify all three layers of the clause, that is, the nucleus, the core, and the clause; (ii) that the order of adverbs modifying a distinct layer of the clause observes the order of the layers of the clause with respect to the position of the verb, as the nuclear adverb is constrained to occur closer to the verb than the core adverb, which, in turn, is constrained to occur closer to the verb than the clausal adverb. The RRG approach to adverbs remains rather preliminary (cf. CortésRodríguez and Rodríguez-Juárez 2019), dealing only with basic phenomena. More analyses are certainly required, for instance, to identify the ordering constraints among the adverbs modifying the same layer, considering their motivations, or to substantiate how the logical structure of ideophonic adverbs can be represented (for example, does this call for a one-place predicate just like adverbs of prosaic words?) to capture the unique semantics of ideophones.

References Akita, Kimi. 2009. A Grammar of Sound-Symbolic Words in Japanese: Theoretical Approaches to Iconic and Lexical Properties of Mimetics. Unpublished PhD dissertation, Kobe University. Akita, Kimi and Takeshi Usuki. 2016. A constructional account of the ‘optional’ quotative marking on Japanese mimetics. Journal of Linguistics 52: 245–275. Alexiadou, Artemis. 1997. Adverb Placement: A Case Study in Antisymmetric Syntax. Amsterdam: John Benjamins. Beck, David. 2008. Ideophones, adverbs, and predicate qualification in Upper Necaxa Totonac. International Journal of American Linguistics 74: 1–46. Bobuafor, Mercy. 2013. A Grammar of Tafi. Unpublished PhD dissertation, Leiden University.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

423

424

KIYOKO TORATANI

Bohnemeyer, Jürgen and Robert D. Van Valin, Jr. 2017. The macro-event property and the layered structure of the clause. Studies in Language 41 (1): 142–297. Childs, Tucker G. 1994. African ideophones. In Leanne Hinton, Johanna Nichols and John J. Ohala (eds.), Sound Symbolism, 178–204. Cambridge: Cambridge University Press. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Cortés-Rodríguez, Francisco, J. and Carolina Rodríguez-Juárez. 2019. The syntactic parsing of ASD-STE100 adverbials in ARTEMIS. Revista De Lingüística y Lenguas Aplicadas 14: 59–79. Creissels, Denis. 2001. Setsuwana ideophones as uninflected predicative lexemes. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 76–85. Amsterdam: John Benjamins. Croft, William and D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press. De Jong, Nicky. 2001. The ideophone in Didinga. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 121–138. Amsterdam: John Benjamins. Dingemanse, Mark. 2011. The Meaning and Use of Ideophones in Siwu. Unpublished PhD dissertation, Radboud University Nijmegen. Dingemanse, Mark. 2017. Expressiveness and system integration: On the typology of ideophones, with special reference to Siwu, STUF – Language Typology and Universals 70(2): 363–385. Dingemanse, Mark. 2019. ‘Ideophone’ as a comparative concept. In K. Akita and P. Pardeshi (eds.), Ideophones, Mimetics, and Expressives, 13–33. Amsterdam: John Benjamins. Doke, Clement Martyn. 1935. Bantu Linguistic Terminology. London: Longmans, Green, and Co. Engels, Eva. 2012. Optimizing Adverb Positions. Amsterdam: John Benjamins. Ernst, Thomas. 2002. The Syntax of Adjuncts. Cambridge: Cambridge University Press. Ernst, Thomas. 2014. The syntax of adverbs. In A. Carnie, D. Siddiqi and Y. Sato (eds.), The Routledge Handbook of Syntax, 108–130. New York: Routledge. Ernst, Thomas. 2020. The syntax of adverbials. Annual Review of Linguistics 6(1): 89–109. Hamano, Shoko. 1998. The Sound-Symbolic System of Japanese. Stanford, CA: CSLI Publications. Hirose, Masayoshi. 1981. Japanese and English Contrastive Lexicography: the Role of Japanese ‘Mimetic Adverbs’. Unpublished PhD dissertation, University of California, Berkeley. Ibarretxe-Antuñano, Iraide. 2006. Sound Symbolism and Motion in Basque. Munich: Lincom Europa. Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

Kulemeka, Andrew Tilimbe Clement. 1993. The Status of the Ideophone in Chichewa. Unpublished PhD dissertation, Indiana University. Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press. Martin, Samuel. 1975. A Reference Grammar of Japanese. New Haven, CT: Yale University Press. McCawley, James D. 1968. The Phonological Component of a Grammar of Japanese. The Hague: Mouton. McConnell-Ginet, Sally. 1982. Adverbs and logical form. Language 58: 144–184. Moshi, Lioba. 1993. Ideophones in KiVunjo-Chaga. Journal of Linguistic Anthropology 3(2): 185–216. Newman, Paul. 1968. Ideophones from a syntactic point of view. The Journal of West African Languages 5: 107–117. Nuckolls, Janis B. 1996. Sounds Like Life: Sound-Symbolic Grammar, Performance and Cognition in Pastaza Quechua. New York: Oxford University Press. Patent, Jason. 1998. A willy-nilly look at Lai ideophones. Linguistics of the Tibeto-Burman Area 21(1): 155–200. Reiter, Sabine. 2011. Ideophones in Awetí. Unpublished PhD dissertation, University of Kiel. Rivero, María-Luisa. 1992. Adverb incorporation and the syntax of adverbs in modern Greek. Linguistics and Philosophy 15: 289–331. Sawada, Harumi. 1978. A contrastive study of Japanese and English sentence adverbials: From the viewpoint of speech act theory [in Japanese]. Gengo Kenkyu 74: 1–36. Schachter, Paul. 1985. Parts-of-speech systems. In T. Shopen (ed.), Language Typology and Syntactic Description, Vol. I: Clause structure, 3–61. Cambridge: Cambridge University Press. Schaefer, Ronald P. 2001. Ideophonic adverbs and manner gaps in Emai. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 339–354. Amsterdam: John Benjamins. Smith, Carlota. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Kluwer Academic Publishers. Storch, Anne. 2014. A Grammar of Luwo: An Anthropological Approach. Amsterdam: John Benjamins. Toratani, Kiyoko. 2005. A cognitive approach to mimetic aspect in Japanese. Proceedings of the Thirty-first Annual Meeting of the Berkeley Linguistics Society, 335–346. Toratani, Kiyoko. 2007. An RRG analysis of manner adverbial mimetics. Language and Linguistics 8(1): 311–342. Toratani, Kiyoko. 2017. The position of to/Ø-marked mimetics in Japanese sentence structure. In N. Iwasaki, P. Sells and K. Akita (eds.), The Grammar of Japanese Mimetics: Perspectives from Structure, Acquisition and Translation (Routledge Studies in East Asian Linguistics), 35–72. London: Routledge.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

425

426

KIYOKO TORATANI

Tsujimura, Natsuko and Masanori Deguchi. 2007. Semantic integration of mimetics in Japanese. Communication and Linguistics Studies 39(1): The Main Session: Papers from the Thirty-Ninth Annual Meeting of the Chicago Linguistic Society, 339–353. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic categories in Role and Reference Grammar. In Robert D. Van Valin., Jr. (ed.), Investigations of the Syntax–Semantics–Pragmatics Interface, 161–178. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Voeltz, Erhard F. K. and Christa Kilian-Hatz (eds.). 2001. Ideophones. Amsterdam: John Benjamins. Watson, Richard L. 2001. A comparison of some Southeast Asian ideophones with some African ideophones. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 385–405. Amsterdam: John Benjamins. Watters, James K. 2013. Transitivity, constructions, and the projection of argument structure in RRG. In B. Nolan and E. Diedrichsen (eds.), Linking Constructions into Functional Linguistics, 23–40. Amsterdam: John Benjamins.

Notes 1 Ideophones refer to a ‘vivid representation of an idea in sound’ (Doke 1935: 118), or ‘[a] member of an open lexical class of marked words that depict sensory imagery’ (Dingemanse 2019: 16). Although the term ‘ideophone’ is so commonly used for African and Amerindian languages that it has become a byword, this chapter uses the term ‘mimetics’ to refer to Japanese, following the tradition of the literature of Japanese linguistics. 2 While some authors contend ideophones are a subset of regular lexical categories available in the language, such as adverb and verb (e.g. Moshi 1993; Newman 1968), others say they constitute a lexical category of their own (e.g. Kulemeka 1993; Reiter 2011). I follow McCawley’s (1968) insights on Japanese mimetics, separating the Japanese lexicon into four strata (native, Sino-Japanese, loans from other languages, and mimetics). That is, I consider ideophones to constitute a lexical stratum of the language whose linguistic materials are recruited to form a unique word group or category, but they do not constitute a grammatical category, ‘ideophone’, per se, on a par with the major categories of verb, noun, adjective and adverb. 3 Status includes clausal negation and epistemic modality. Modality, listed under core operators in Figure 9.2, includes root modals such as those expressing ability and obligation.

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

Adverbs, Mimetics and Ideophones

4 Adverbs are a lexical rather than a grammatical (functional) category, to which operators refer (cf. Van Valin 2005: 26). 5 The terms ‘subject-oriented’ and ‘speaker-oriented’ are from Jackendoff (1972). 6 The term ‘hyperonym’ may be spelled ‘hypernym’. The former is used here following Croft and Cruse (2004). 7 It is worth noting that the rice-beating ideophone in Kisi, pim pim ‘rice beaten by two or more people’, can co-occur only with cuu ‘to beat or pound (usually rice) in a mortar with a pestle’ (Childs 1994: 189), with the former apparently the hyponym of the latter. But since the ideophone affects the interpretation of the number of event participants, it must be considered a core modifier. 8 The distinction is not always clear-cut. As a basic rule of thumb, it can be considered a non-hyponym if the mimetic or ideophone can modify a general activity meaning ‘do something’: yukkuri nanika-o suru ‘do something slowly’ vs. *musyamusya nanika-o suru ‘do something munch-munch’. 9 This refers only to cases where the ideophone is contained in the periphery, modifying a given layer of the clause, excluding cases where the ideophone occurs in the pre-detached position, when it functions as a clausal adverb. 10 When the adverb is accompanied by the focus particle mo as in bukiyoonimo ‘clumsily’, this yields only the clausal (evaluative) reading (cf. Sawada 1978). 11 A reviewer of the chapter noted the Tepehua examples in (28) are reminiscent of the case of adverb incorporation in Modern Greek (Rivero 1992).

https://doi.org/10.1017/9781316418086.012 Published online by Cambridge University Press

427

10 Adposition Assignment and Adpositional Phrase Types in RRG Sergio Ibáñez Cerda 10.1

Introduction

This chapter has a twofold aim: to discuss the RRG treatment of adpositional phrase (AP) types and adpositional assignment (Foley and Van Valin 1984; Van Valin 1991, 1993, 2005; Van Valin and LaPolla 1997; Jolly 1993) and to propose a new typology of adpositional phrase types, following Ibáñez Cerda (2009, 2011). Although we only discuss data from English and Spanish, the insights and analyses presented here can be broadly applied to other languages. The RRG treatment of adpositional systems relies on three principal assumptions. First, the function of adpositions is comparable to that of morphological case, in that adpositional assignment may depend on the semantic role of the argument. In this sense, adpositions can be considered to be analytic case forms (see Chapter 7). Second, RRG assumes that the assignment of adpositions is not idiosyncratic and need not be postulated as part of the lexical entries of verbs. Instead, adpositional assignment follows systematic rules, which are applied in the linking and, rather than being based on grammatical relations or phrase structure positions, depend on the semantic content of the logical structure (LS) of the predicate and on the semantic/syntactic distinction between direct and oblique core arguments. The third assumption is the distinction, based on Bresnan (1982), between predicative and non-predicative adpositions. Non-predicative adpositions mark verbal arguments: they do not license these arguments and they do not add any substantive semantic information to the clause. They are a function of the semantics of the predicate and, thus, they are freemorphemic case markers assigned by the predicate (e.g. to in Pat gave a book to Peter). Contrastingly, predicative adpositions are predicates, in that they contribute substantive semantic information to the clause in which they occur, both in terms of their own meaning and of the meaning of the argument that they license. A typical example of a predicative adposition

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

is one that introduces a peripheral locative – or setting – complement (e.g. in in Julia had dinner in the garden). Importantly, the two types of adposition are closely related. In fact, both functions, predicative and non-predicative, can be, and usually are, played by the same forms. Jolly (1993) takes the predicative adposition to be basic and to be stored in the lexicon. Generally, adpositions will have a case-marking function when their LS is a portion of the LS of the predicate that licenses the argument they mark. For example, English predicative in (as in John had dinner in the park) is closely related to non-predicative in, which marks the locative argument in John inserted the key in the lock. The LS of predicative in is be-in′ (z, y), while the LS of insert is [[do′ (x, Ø)] CAUSE [INGR be-in′ (z, y)]]. Thus, a segment of this last LS is the same as the LS of the preposition, and this is what is behind the marking of an argument of insert with in. As will be pointed out in Section 10.5, the distinction between predicative and non-predicative adpositions, along with the semantic distinction between the notions of verbal argument and free adjunct, and the syntactic difference between those arguments or adjuncts that can display core features – such as the possibility of functioning as controllers or pivots – and those that cannot, will allow us to outline a system of seven adpositional types. The chapter is organized as follows: Section 10.2 summarizes the tenets of RRG which are key for an understanding of adpositions within the framework. Section 10.3 is an introduction to case and adpositions in RRG. The current RRG account of adpositional assignment is discussed in Section 10.4. We then propose our own classification of adpositional phrases (Section 10.5). Finally, in Section 10.6, we draw some brief conclusions.

10.2

Relevant Tenets of RRG

As is stated in Van Valin (2005: 4), RRG bases its analysis of clause structure on two universal semantic distinctions: the one between the predicate and non-predicating elements and, on the other hand, among the nonpredicating elements, the distinction between arguments and non-arguments (see Figure 10.1). The distinction between arguments and non-arguments is based on the idea that the states of affairs represented by the predicates inherently determine the number and type of participants involved in them. The nature of the situation conditions the presence of the participants that are

predicate

Non-arguments

arguments Figure 10.1 Universal oppositions underlying clause structure (Van Valin 2005: 4)

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

429

430

SERGIO IBÁÑEZ CERDA

CLAUSE PERIPHERY

CORE NUCLEUS

Figure 10.2 Components of the layered structure of the clause (Van Valin 2005: 4)

needed in order to make possible the state of affairs. In this view, the participants inherently required by a predicate are the arguments, while those that are not are the non-arguments or adjuncts. Based on these semantic oppositions, RRG proposes the layered structure of the clause (LSC), which consists of the nucleus, the core and the periphery, as represented in Figure 10.2. Similarly to Lexical Functional Grammar (Bresnan 2001) and Functional Grammar (Dik 1989), RRG assumes that syntax and semantics work in parallel. There are no derivations, or any deep levels of representation, but just one single level with two parallel representations.1 There are natural correspondences between syntax and semantics. To begin with, the semantic predicate corresponds to the nucleus of the clause in syntax. Following an iconicity principle (Haiman 1980), arguments, along with the nucleus, belong to the core, whereas the non-arguments are peripheral elements of the clause. Depending on their morphosyntactic properties, arguments can be direct or oblique. The direct core arguments are marked by direct morphological case: nominative and accusative in nominative-accusative systems, and absolutive and ergative in absolutive-ergative systems. Peripheral non-arguments are adjuncts. Against this backdrop, a clause like John gave a book to Mary in the library can be represented as in Figure 10.3. CLAUSE CORE PERIPHERY John gave a book to Mary in the library

NU NUCLEUS Figure 10.3 Layered structure of John gave a book to Mary in the library

Using the notions just outlined, we can say that John and a book are direct core arguments, while to Mary is an oblique core argument. The three of them are participants semantically required by the verb to give. As for the library, it is an adjunct. Table 10.1 summarizes the relation between the units of the syntactic and semantic representations. A third type of clausal component, originally proposed by Jolly (1993), is that of argument-adjuncts in the core. This complement type corresponds to APs that code a verbal argument but, at the same time, are introduced by a predicative preposition.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

Table 10.1 Semantic units underlying the syntactic units of the layered structure of the clause (Van Valin 2005: 5) S EMANTIC

S YNTACTIC

UNIT

Predicate Argument in semantic representation of predicate Non-argument Predicate þ Arguments Predicate þ Arguments þ Non-arguments

UNIT

Nucleus Core argument Periphery Core Clause (¼ Core þ Periphery)

The semantic representation of the clause is built upon the semantic representation of the verb or the predicating element; it starts from six basic Aktionsart types (see (1)) and their causative counterparts (2). (1)

a. b. c. d. e. f.

State: Achievement: Semelfactive: Accomplishment: Activity: Active accomplishment:

The boy is afraid. The balloon popped. The pencil tapped on the table. The ice melted. The soldiers marched in the park. The soldiers marched to the park.

(2)

a. b. c. d. e.

Causative Causative Causative Causative Causative

The dog frightens/scares the boy. The cat popped the balloon. The teacher tapped the pencil on the table. The hot water melted the ice. The sergeant marched the soldiers in the park. The sergeant marched the soldiers to the park.

state: achievement: semelfactive: accomplishment: activity:

f. Causative active accomplishment:

RRG uses a modified version of the representational scheme proposed in Dowty (1979) to capture these classes (see Table 10.2).

Table 10.2 Lexical representations for Aktionsart classes Aktionsart class

Logical structure

STATE ACTIVITY ACHIEVEMENT

predicate′ (x) or (x, y) do′ (x, [predicate′ (x) or (x, y)]) INGR predicate′ (x) or (x, y), or INGR do′ (x, [predicate′ (x) or (x, y)]) SEML predicate′ (x) or (x, y) SEML do′ (x, [predicate′ (x) or (x, y)]) BECOME predicate′ (x) or (x, y), or BECOME do′ (x, [predicate′ (x) or (x, y)]) do′ (x, [predicate1′ (x, (y))]) & INGR predicate2′ (z, x) or (y) α CAUSE β, where α, β are logical structures of any type

SEMELFACTIVE ACCOMPLISHMENT ACTIVE ACCOMPLISHMENT CAUSATIVE

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

431

432

SERGIO IBÁÑEZ CERDA

An assumption which distinguishes RRG from all other syntactic frameworks is that it posits two types of semantic role. The traditional thematic relations of agent, patient, theme, instrument, etc. are only mnemonics for five structural positions in the LS of predicates. In addition to these five positions, RRG postulates the two semantic macroroles actor and undergoer, which are generalizations across thematic relations. Actor is a generalization across agent, experiencer, instrument, and other roles, while undergoer is a generalization subsuming patient, theme, recipient, and other roles. Agent is the prototype for actor, and patient is the prototype for undergoer. The two macroroles are the two primary arguments of a transitive predication, and either one of them can be the single argument of an intransitive verb. The logical structure of the predicate determines which macroroles it takes. If it takes two macroroles, then they must be actor and undergoer. For verbs which have a single macrorole, the default choice follows directly from the logical structure of the verb: if the verb has an activity predicate in its logical structure, the macrorole will be actor; otherwise, it will be undergoer. The default macrorole assignment principles are summarized in (3) (Van Valin 2005: 63). Default Macrorole Assignment Principles a. Number: the number of macroroles a verb takes is less than or equal to the number of arguments in its logical structure. 1. If a verb has two or more arguments in its logical structure, it will take two macroroles; 2. If a verb has one argument in its logical structure, it will take one macrorole. b. Nature: for verbs which take one macrorole, 1. If the verb has an activity predicate in its logical structure, the macrorole is actor. 2. If the verb has no activity predicate in its logical structure, the macrorole is undergoer.

(3)

Although most verbs follow these defaults assignments, there are exceptions (e.g. intransitive two-place predicates such as locative or psych verbs). These must be specified in the lexical entries of the relevant verbs in terms of a simple feature [MR α], with values [MR 0], [MR 1] and [MR 2]. Finally, the relation between macroroles and logical structure argument positions is captured in the Actor-Undergoer Hierarchy (AUH) in Figure 10.4 (Van Valin 2005: 61). ACTOR

Arg. of DO [

UNDERGOER

1st arg. of do′ (x,...

1st arg. of 2nd arg. of pred′ (x, y) pred′ (x, y)

Arg. of pred′ (x)

= increasing markedness of realization of argument as macrorole]

Figure 10.4 Actor-Undergoer Hierarchy

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

Given the logical structure of a transitive verb, the leftmost argument will by default be the actor and the rightmost argument will be the undergoer. Marked assignments to undergoer are possible, typically with three-place predicates, where there can be two arguments competing for a macrorole function, as in (4a–c). (4)

a. [do′ (Pat, Ø)] CAUSE [BECOME have′ (Chris, book)] b. Pat [actor] gave the book [undergoer] to Chris. c. Pat [actor] gave Chris [undergoer] the book.

Example (4c) illustrates the dative shift alternation, whereby the first argument of the two-place state predicate, not the second, is assigned the macrorole undergoer.

10.3

Adpositions in RRG

10.3.1 Case and Adpositions ‘Case is a system of marking dependent nouns for the type of relationship they bear to their heads’ (Blake 2001: 1). Its primary function is to establish and to differentiate, in semantic and grammatical terms, the role or function of verbal arguments with respect to the event denoted by the verb. RRG differs from other syntactic theories, where case marking and agreement are invariably tied to grammatical relations. Since notions like subject and direct object have no place in this framework (see Chapter 5), neither case marking nor agreement can be based on them. Rather, case-marking rules make crucial reference to macroroles and direct core argument status, in addition to the content of the LS in the lexical entry of the verb, as we will show in Section 10.4.2 for oblique case adpositions. Grammatical aspect, modality and negation can also play a role in case marking. For example, in Slavic languages, the direct object of negated clauses is marked by genitive instead of accusative case. In addition, in Spanish and other languages, the animacy or inherent lexical content of a reference phrase (RP) (Silverstein 1976, 1981, 1993) also plays a role in case marking. For example, dative appears instead of (al)lative or genitive case in many constructions to mark nominals whose referent is animate. RRG posits a set of rules for direct or grammatical case assignment (see Chapter 7). Grammatical or direct cases are not normally marked by adpositions and can have an agreement cross-reference on the verbal head: nominative, accusative and dative in accusative languages, and absolutive, ergative and dative in ergative languages. However, in some languages, direct cases can be instantiated by adpositions, like the a preposition for dative case in Spanish, as in Román dio una flor a Tere ‘Román gave a flower to Tere’. Besides the morphological case forms, many languages have a complementary system of adpositions. In addition, in some languages, such as Latin and other Indo-European languages, adpositions can govern specific case

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

433

434

SERGIO IBÁÑEZ CERDA

forms, although this type of case marking goes beyond the scope of this chapter. In RRG, the notion of case marking is extended to cover the function of the cross-referencing pronominal affixes which appear on the verbal base in head-marking languages (Nichols 1986, Van Valin 2005: 16–17). These affixes perform a twofold function: on the one hand, they indicate the relation between the arguments and the verb, and on the other hand, they are arguments in syntax. Only the grammatical or direct cases are marked on the head verb; other semantic relations, if marked, will be marked through adpositions or morphological case on the dependents.

10.3.2

Difference between Predicative and Non-Predicative Adpositions Building upon Bresnan (1982), RRG posits two types of adposition: predicative and non-predicative. Predicative adpositions function as predicates in that they introduce a participant which is not licensed by the verb. This participant is their object, and they are the head of the phrase, which is an adjunct. Non-predicative prepositions, on the other hand, mark an argument licensed by the verb. Thus, they are a function of the semantics of the verb; they themselves do not add any semantic information to the clause. They are like free-morphemic case markers assigned by the predicate. In John gave a book to Mary in the library (see Figure 10.3), to Mary is a non-predicative prepositional phrase (PP) functioning as a core argument, while in the library is a predicative PP functioning as an adjunct. The two types of adposition are represented differently in the layered structure of the clause and hence in the constituent projection. Since nonpredicative adpositions mark arguments of the verb, the phrases in which they occur are APs in the core; the adposition is treated as a case marker, and it is not the head of the phrase. Predicative adpositions, on the other hand, have their own LS and are the head of the phrase in which they occur. The nominal they introduce is their object or argument. The structure of a nonpredicative or argument AP and a predicative AP is given in Figure 10.5 and Figure 10.6, respectively. As mentioned before, there is a third function that an adposition can have: it can mark an argument of the verb, while at the same time contributing its semantics to the clause. This type of PP typically, but not exclusively, appears with locative verbs like those that denote a change of place (e.g. put). Such verbs require a goal argument but the choice of the locative adposition is not completely determined by the verb, as the speaker can choose among some very closely related adpositions, which semantically elaborate on the type of locative relation between the theme and the goal, as shown in (5). (5)

Kim put the book {in / on / next to / behind / on top of / under} the box.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

PP COREP PP NUCP

RP

PRED P

RP P

to

Pat

Figure 10.5 English non-predicative PP

in

the library

Figure 10.6 English predicative PP

In this example, the different prepositions contribute an important component of meaning, unlike the preposition to with a verb like give or show, and therefore they must be considered to be predicative prepositions. As such, they have the structure of a predicative AP (see Figure 10.5), although they occur in the core, as they mark a verbal argument. In RRG terms, they are ‘argument-adjuncts’ in the core. In Section 10.5 we propose a typology of adpositional phrases in terms of their different semantic and syntactic status.

10.4

Adpositional Phrase Types and Adposition Assignment in RRG

In this section, we discuss the RRG treatment of the three types of adpositional phrase introduced above: predicative (10.4.1), non-predicative (10.4.2) and argument-adjunct (10.4.3), drawing on Jolly (1993), Van Valin and LaPolla (1997) and Van Valin (2005).

10.4.1 Predicative or Adjunct Adpositional Phrases As was mentioned, predicative prepositions introduce a participant which is not an argument in the LS of the main verb, that is, an adjunct. They are treated as primitives, in that they have their own lexical entry in the lexicon, and the syntactic phrases of which they are heads have the status of modifiers in the periphery. There are three main types of predicative adposition: (a) locative and temporal setting adpositions (e.g. in the park in (6a) below); (b) sequential adpositions (before, after and during/while); and (c) causative adpositions (e.g. English for). In the standard RRG treatment, predicative adpositions are taken to be two-place predicates which take the

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

435

436

SERGIO IBÁÑEZ CERDA

LS of the nuclear verb of the clause as one of their arguments. This is illustrated in (6a–b). (6)

a. b. c. d.

Chris ran in the park. be-in′ (park, [do′ (Chris, [run′ (Chris)])]) Chris ran to the park. (do′ (Chris, [run′ (Chris)]) & INGR be-at′ (park, Chris))

In (6a), Chris’s running takes place in the park, and therefore the logical structure of the predicative preposition in is the highest predicate in the logical structure; it takes the park and the logical structure for run as its two arguments. This contrasts with the logical structure of a clause like (6c) Chris ran to the park, where the PP expresses the location of the referent of Chris, not the location of the event of running, and in this example to the park is an argument-adjunct PP (see Section 10.4.2). As pointed out in Ibáñez Cerda (2009), predicative or setting PPs are like adverbs, in that they can modify different parts of a verbal LS, and not necessarily all of it, as shown in (7) for Spanish. (7)

Juan cortó el pastel sobre la tabla de madera. John cut the cake on the board of wood ‘John cut the cake on the wooden board.’

Here the PP sobre la tabla de madera ‘on the wooden board’ only refers to the place where the change of state takes place; more precisely, the wooden board is the place where the event of ‘dividing the cake into pieces’ happens. Certainly, Juan is doing something that involves the wooden board, but, clearly, he is not located on the board. Consequently, the semantic scope of the PP is not over the core but only part of it. In this way, we can consider PPs like the one in (7) as partial modifiers of the core. A possible representation for (7) is thus shown in (8). (8)

[do′ (Juan, Ø)] CAUSE [be-on′ (tabla, [BECOME cut′ (pastel)])]2

In contrast to the LS in (6b), here the adverbial predicate be on′ only has scope over the subevent of change of state, leaving the activity subevent out. Apart from the case of (7), the fact that adjunct PPs can modify structural elements other than the whole core can be seen in a sentence with an added instrumental complement (9). (9)

Juan cortó el pastel con un cuchillo sobre la tabla de madera. John cut the cake with a knife on the board of wood ‘John cut the cake with a knife on the wooden board.’

Not only is the change of state under the scope of the locative PP, but part of the action performed by the effector falls inside its scope: John’s knife manipulation and, more clearly, the contact of this instrument with the cake, takes place on the wooden board. This can be represented as follows.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

(10) [do′ (Juan, use′ [Juan, cuchillo])] CAUSE [[be-on′ (tabla, [do′ (cuchillo, [cut′ (cuchillo, pastel)])])] CAUSE [BECOME cut′ (pastel)]]

Here, only one of the activity subevents, the one that has the instrument as an effector, is under the scope of be-on′, while the more external one, the one where Juan is the effector, falls out of the scope of the preposition. In sum, following Ibáñez Cerda (2009), we have argued that the PPs in (7) and (9) behave differently from the one in (6a): the whole core of the clause is one of the arguments of the PP in (6a), but this is not the case with (7) and (9). Nevertheless, none of these PPs codify verbal arguments, they are all headed by a predicative preposition, and they are all modifiers in the periphery. In this sense, they represent different cases of adjunct complements introduced by predicative PPs.

10.4.2

Non-Predicative Adposition Assignment and Oblique Core Arguments Non-predicative adpositions mark a verbal argument: specifically, in RRG terms, an oblique core argument. As was mentioned, these adpositions are not idiosyncratically listed in the lexical entries of verbs, but rather assigned in terms of systematic rules (Foley and Van Valin 1984; Jolly 1993; Van Valin and LaPolla 1997). The basic rules governing the assignment of to, from and with in English are given here. (11)

a. Assign to to non-MR y-argument in logical structure segment: BECOME/INGR pred′ (y, z) b. Assign from to non-MR y-argument in logical structure segment: BECOME/INGR NOT pred′ (y, z) c. Assign with to non-MR b argument if, given two arguments, a and b, in a logical structure with (i) both as possible candidates for a particular macrorole and (ii) a being equal or higher (to the left of b) on the AUH, b is not selected as that macrorole.

We provide some examples of the assignment of the English preposition to in (12a–c). Then, in (13) we illustrate the relevant LS segments. (12)

a. Sally gave/sent/handed the box to Pat. b. Sally showed the box to Pat. c. Sally taught basketweaving to Pat.

(13)

a. . . .BECOME have′ (Pat, box) b. . . .BECOME see′ (Pat, box) c. . . .BECOME know′ (Pat, basketweaving)

¼ give, hand, (send) ¼ show ¼ teach

In all these examples, the RP marked by to is the first argument of a twoplace predicate embedded under a BECOME operator, and it is also a nonmacrorole core argument. The state predicate can be a possession (12a, 13a), perception (12b, 13b) or cognition verb (12c, 13c), as well as a locative predicate (as with send in 12a). Therefore, the argument marked by to can

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

437

438

SERGIO IBÁÑEZ CERDA

be a possessor, a perceiver, a cognizer or a location. Although to marks different types of semantic argument, it always appears in the same kind of LS environment, precisely the one foreseen in the rule in (11a): BECOME/ INGR pred′ (y, z) A similar analysis can be posited for from, which appears in the examples in (14a–c), the relevant LS segments being given in (15a–c). (14)

a. Sandy took/stole/bought the keys from Kim. b. Pat drained the water from the pool. c. Kim escaped from the burning house.

(15)

a. . . . BECOME NOT have′ (Kim, keys) b. . . . BECOME NOT be-in′ (pool, water) c. . . . BECOME NOT be-in′ (burning house, Kim)

¼ take, steal, buy ¼ drain ¼ escape

In each of the LS segments in (15a–c), from marks the first argument of the two-place state predicate, which is a non-macrorole core argument. The difference between these segments and those in (13a–c) is the presence of NOT. This difference in content between to and from was first proposed by Gruber (1965). Again, like to, from does not mark a single thematic relation, but rather it is assigned in a particular LS context – BECOME/INGR NOT pred′ (y, z), and the rule (11b) rightly predicts its appearance. The rule (11c) for preposition with applies in the single-prime examples in (16a′, b′), where the theme argument z is not selected as undergoer and is instead marked by with. In contrast, the non-prime examples (16a, b) obey rule (11a), which requires the marking with to. (16)

a. a′. a′′. b. b′. b′′.

Sally presented the flowers [z] to Kim [y]. Sally presented Kim [y] with the flowers [z]. [do′ (Sally, Ø)] CAUSE [BECOME have′ (Kim, flowers)] Max loaded the olives [z] into his minivan [y]. Max loaded his minivan [y] with the olives [z]. [do′ (Max, Ø)] CAUSE [BECOME be-in′ (minivan, olives)]

Thus, with is assigned in contexts where two arguments can be selected for the undergoer status, but the marked option – in terms of (11a) – is the one that prevails. There are two further contexts where with is assigned: in the marking of an instrument (17a) and in a comitative PP (17b). (17)

a. Tom cut the bread with the knife a′. The knife cut the bread a′′. [do′ (Tom, [use′ (Tom, knife)])] CAUSE [[do′ (knife, Ø)] CAUSE [BECOME cut′ (bread)]] b. Sandy went to the store with Kim b′. Sandy and Kim went to the store b′′. [do′ (Sandy/Kim, [move′ (Sandy/Kim)]) ^ PROC cover.path.distance′ (Sandy/ Kim)] & INGR be-LOC′ (store, Sandy/Kim)

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

In (17a), following the AUH, it is the leftmost argument, Tom, which is assigned the macrorole actor, and as such it is projected as subject. The other argument, the instrument, the knife, is marked by the preposition with. However, when the effector is not projected, the instrument can be selected as actor and then projected in the subject function, as can be seen in (17a′). A similar alternation is found with the comitative. The LS in (17b′′) has two co-agent arguments, Sandy and Kim, and each one can be assigned the macrorole actor and be projected as subject. In (17b) Sandy is projected as subject and, applying the rule (11c), Kim is marked by the preposition with; but in (17b′) both referents are arguments as subject, through a single complex RP. Summing up, English with appears in contexts where two arguments compete for a macrorole status, actor or undergoer, and one is not selected as such, the context foreseen by the rule in (11c). In languages with morphological case (see Chapter 7), dative is usually assigned to mark the recipient of transfer verbs, like those in (12a–c). This marking obeys what is considered in RRG (Van Valin and LaPolla 1997) to be the default rule for the assignment of oblique cases. (18)

Assign dative case to non-macrorole arguments (default).

Thus, in a Croatian example like (19), unuc- ‘grandsons’, the highestranking argument, is selected as actor; cvijet- ‘flowers’, the lowest-ranking argument, is selected as undergoer, and bak- ‘grandmother’, the nonmacrorole core argument, receives dative case. (19)

Unuc-i su bak-i darova-l-i cvije′c-e. grandson-m.pl.nom be.3pl grandmother-f.sg.dat give-pst-pl flower-m.pl.acc ‘The grandsons gave flowers to [their] grandmother.’

As seen above, languages which do not have morphological case marking on nouns use adpositions to mark oblique case. While English uses the preposition to, Spanish uses the preposition a, which is the same form as appears with the goal argument of motion verbs. The difference between the goal preposition and the dative one is that the complement marked by dative a is replaceable or commutable for the dative clitic pronoun le/les; indeed, it usually appears reduplicated or cross-referenced by this clitic. This same preposition a, cross-referenced by the pronoun le/les, also appears with M-intransitive verbs, like the psychological ones gustar ‘to like’ and encantar ‘to love’, as in A María le gusta el chocolate ‘Mary likes chocolate’. These are two-place predicates with only one macrorole argument, which is what the M-intransitive specification means. Since they are also state predicates – like′ (María, chocolate) – their only macrorole argument must be an undergoer and, in terms of the AUH, the second argument outranks the first for the undergoer status. The undergoer receives nominative case, while the

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

439

440

SERGIO IBÁÑEZ CERDA

non-macrorole argument is assigned dative case, which is the default oblique case. Besides psych verbs, there is, however, another large class of M-intransitive predicates in Spanish, whose non-macrorole argument is marked by the preposition de ‘of’, which is the form which also marks the genitive and the ablative. These are verbs like carecer ‘to lack’, constar ‘to consist’, desconfiar ‘to distrust’, abusar ‘to abuse’ and prescindir ‘to dispense’, which, according to the literature (Alarcos Llorach 1972; Martínez García 1986; Hernández Alonso 1990; Cano Aguilar 1999; among others), govern their preposition, which basically means that they have their specific prepositional mark idiosyncratically specified in their lexical entry. Given that the presence of the preposition de is systematic with the majority of these verbs, one can consider them as M-intransitive and, then, postulate the application of a general rule for the assignment of the genitive preposition. This would need, however, an extra specification U¼x in the verbal entries, to signal that a marked choice of undergoer must take place, given that in these cases the first argument is selected as undergoer, contrary to the specifications of the AUH. In this way, in an example like (20a), it is María which gets the macrorole status and the nominative case, and oportunidades is marked by the genitive case. (20)

a. María carece de oportunidades. Mary lacks gen opportunities ‘Mary lacks opportunities.’ b. lack′ (María, oportunidades) c. lack′ (x, y) U¼x

In Spanish, then, there are two clearly identifiable classes of Mintransitive verbs: one marked by dative case and another one marked by genitive case. Thus, the Spanish non-macrorole assignment rule can be reformulated as (21). (21)

a. Assign dative case to non-macrorole arguments (default). b. Assign genitive case to non-macrorole arguments of verbs with the specification U¼x

The importance of the genitive adposition as a non-semantic or grammaticalized mark for non-macrorole arguments in Spanish is additionally attested when considering its ‘competing’ role for marking arguments that are not selected for macrorole status, in contexts where those arguments could or should have been picked up as such. These are contexts where, in English, the rule for with would apply. This is the case with the oblique argument of the transfer (16a′) and locative (16b′) alternation construction. Although some constructions of this type, where a nondefault assignment has taken place, can be marked by the preposition con, the equivalent of with (see, e.g. 22e′, f′), the genitive preposition de appears in all of them.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

(22)

a. La oficina de prensa informó la noticiaal público. The office of press informed the notice dat.def.art public ‘The press office gave the news to the public.’ a′. La oficina informó al público de /*con la noticia. The office informed acc.def.art public gen/*ins the notice ‘The press office informed the public the notice.’ b. Elia le perdonó su mala educación a Marcelo. Elia dat forgave his bad education dat Marcelo ‘Elia forgave his bad manners to Marcelo.’ b′. Elia perdonó a Marcelo de /*con su mala educación. Elia forgave acc Marcelo gen/*ins his bad education ‘Elia forgave Marcelo his bad manners.’ c. Ramón denunció el robo de Martha. Ramón reported the robbery of Martha ‘Ramón reported Martha’s robbery’. c′. Ramón denunció a Martha de /*con robo. Ramón reported acc Martha gen/*ins robbery ‘Ramón reported Martha for the robbery’. d. Ludmila vació todo el líquido de la botella. Ludmila emptied all the liquid from the bottle ‘Ludmila emptied all the liquid from the bottle.’ d′. Ludmila vació la botella de /*con todo su líquido. Ludmila emptied the bottle gen/*ins all its liquid ‘Ludmila emptied the bottle of all its liquid.’ e. Leonardo cargó las manzanas en el camión. Leonardo loaded the apples in the truck ‘Leonardo loaded the apples into the truck.’ e′. Leonardo cargó el camión de/ con manzanas. Leonardo loaded the truck gen/ins apples ‘Leonardo loaded the truck with the apples.’ f. Fernando roció cerveza en todo el cuarto. Fernando sprayed beer in all the room ‘Fernando sprayed beer all over the room.’ f ′. Fernando roció el cuarto de /con cerveza. Fernando sprayed the room gen/ins beer ‘Fernando sprayed the room with beer.’

As can be seen from the prime examples in (22), de ‘of’, is the preposition that, in effect, can appear in all contexts of non-default projection of a theme, when the competition for the undergoer function is between asymmetrical arguments, that is, two arguments which do not have the same status in LS. The domain of con ‘with’ is instead much narrower. One way of analysing these facts is to consider de to be the basic preposition for marking non-prototypical argument projections in Spanish. In this analysis, Spanish de – including the one with genitive function – is an empty preposition which marks asymmetric relations in non-default cases: (i) lexically asymmetrical relations between the two arguments of M-intransitive verbs with the U¼x specification; (ii) non-default coding of asymmetrical

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

441

442

SERGIO IBÁÑEZ CERDA

arguments; and (iii) the asymmetrical relations between a noun and its modifiers. We can thus posit the rule in (23) for the assignment of the de preposition to verbal arguments in Spanish. (23)

Assign de to non-macrorole arguments in non-default projections.

The dative preposition a can still be the default preposition for nonmacrorole arguments, whereas con ‘with’ is the preposition for: (i) the nonactor co-agent in comitative constructions (e.g. Lola fue al cine con Domingo ‘Lola went to the movies with Domingo’); (ii) the non-actor instrument of cut and break verbs; and (c) the non-default coding of arguments with locative and spray verbs.3 Note that the instrumental case rule in (11c) can still partially apply for the assignment of the preposition con in Spanish. (24)

Assign con to non-MR b argument if, given two arguments, a and b, in a logical structure with (i) both as possible candidates for a particular macrorole and (ii) a being equal or higher (to the left of b) on the AUH, b is not selected as that macrorole.

This rule effectively covers the cases where con competes with de for the marking of theme arguments not selected as undergoer (locative and spray verbs) and cases where de does not appear: for example, those where there are two arguments competing for the actor macrorole (comitative and instrument cases). De, on the other hand, appears in marked cases where a theme argument is not selected as undergoer. Thus, the rule in (23) must be revised as in (25). (25)

Assign de to non-macrorole arguments in non-default undergoer selections.

Although both (24) and (25) apply with locative and spray verbs, elsewhere they have their own niche of functionality.

10.4.3 Argument-Adjuncts Currently, two kinds of argument-adjunct adpositional phrases are recognized in RRG: (a) those that are headed by an adposition which has semantic content but nonetheless marks a verbal argument, and (b) those phrases headed by a predicative adposition which introduces a participant not licensed by the verb (i.e. an adjunct) but does not take the whole LS of the verb as an argument, sharing its argument with the verbal LS. Both types are core phrases in syntactic terms. A typical example of the first type of argument-adjunct is the variable preposition which codes the goal argument of motion and change-of-place verbs. Spanish examples are provided in (26) (see also the English example in (5)).

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

(26)

a. Juan fue a / hacia / hasta / adentro de la tienda. John went to / towards / as far as / inside of the house. ‘John went to / towards /as far as / inside the house.’ b. Juan puso el libro en /sobre / dentro de / debajo de la caja. John put the book in-on /over / inside of / under of the box ‘John put the book in-on /over / inside / under the box.’

Ibáñez Cerda (2009) points out that these PPs have different status depending on the preposition that introduces them. In Spanish, if a goal argument is coded by a PP introduced by the preposition a ‘to’, with intransitive motion verbs, and by en ‘in/on’, with change-of-place verbs, it should be analysed as an oblique core argument. Indeed, these prepositions are canonical with the said verb classes. A corpus-based study (Ibáñez Cerda 2005) shows that the goal PP of intransitive motion verbs was coded with a in over 90 per cent of cases. Similarly, the PP of the change-of-place verbs strongly tended to be introduced by en. These prepositions are thus assigned systematically as the unmarked prepositions in the following contexts: a appears when the LS of a predicate has a BECOME pred′ (z, y) segment, which is the case of motion verbs, and en shows up in the structural environment characterized by the presence of INGR pred′ (z, y), which characterizes ‘putting’ verbs. By contrast, when the goal arguments are introduced by other prepositions that have more semantic content, such as dentro ‘inside’, hacia ‘towards’, etc., they should be considered as argument-adjuncts in the core, because although these kinds of prepositions are clearly predicative, the possibility of using them for introducing the goal argument of change-ofplace verbs is not entirely free, as is the case of the locative PPs that do have the status of adjuncts. Poner ‘to put’, the predicate that serves as nucleus in the sentences exemplified in (26b), is the hypernym in the domain of change-of-place verbs. As such, it has a very general and abstract locative meaning; it does not inherently specify much about its goal, and that is why it can be used with goals introduced by almost any locative preposition. This is not the case, however, with other verbs whose inherent semantics blocks the use of certain prepositions for coding their goals, as can be seen in (27). (27)

??Juan metió el libro fuera de la caja. John put.inside the book out of the box ‘John put inside the book out of the box.’ b. ??Juan sacó el libro dentro de la caja. John took.out the book inside of the box ‘John took the book out inside the box.’ c. ?? Juan encerró su perro hacia su casa. John locked his dog towards his house ‘John locked his dog towards his house.’ a.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

443

444

SERGIO IBÁÑEZ CERDA

d. ?? Juan hospedó a Pedro fuera de su casa. John lodged acc Peter out of his house ‘John lodged Peter out of his house.’ e. ??Juan clavó el clavo detrás de la pared. John nailed the nail behind of the wall ‘John nailed the nail behind the wall.’ f. ??Juan sumergió la cabeza debajo de la fuente. John submerged the head under of the fountain ‘John submerged his head under the fountain.’

The examples in (27) indicate that the preposition is determined by the predicate it appears with, and hence, it is a function of that predicate, although it can in turn contribute a portion of meaning to the argument. Importantly, the PPs with non-canonical prepositions behave as core arguments. They control pivots in coordinated clauses, which in RRG is a standard diagnostic for core participants. Thus, the most straightforward reading of (28a) is one where it is the cabin that looks fine. Similarly, (28b) would normally be taken to mean that the fishbowl looks good. el bosque, Juan i llegó hasta la vieja cabañaj (28) a. Paseando por Walking through the forest John arrived to.limit the old cabin veía bien. y aún i/j se and still refl looked fine ‘Walking in the forest, John came across the old cabin and (it) looked fine.’ nuevaj y bien. b. Juan i puso un pezh dentro de la pecera *i/ h/j se ve John put a fish inside of the fishbowl new and refl looks good ‘John put a fish inside the new fishbowl and (it) looks good.’

The LS representation of clauses with these types of argument-adjunct phrase is not different from those with a goal oblique core argument (cf. (13)). The argumental value of the goal is guaranteed by the BECOME pred′ (z, y) segment of the LS of the verb. What changes is the LS of the preposition which fills the pred′ function and instantiates the semantic value which that form has in the lexicon. The second type of argument-adjunct is best exemplified by benefactive complements, like for Sandy in (29). (29)

Robyn baked a cake for Sandy.

Here, for Sandy is an argument-adjunct because the LS of the predicative preposition that heads it, for, has an argument that is also an argument of the main predicate, as can be seen in the representation in (30), proposed by Van Valin and LaPolla (1997: 383). (30)

[[do′ (Robin, Ø)] CAUSE [BECOME baked′ (cake)]] PURP [BECOME have′ (Sandy, cake)]

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

In (30), the PP for Sandy is represented by the segment introduced by the PURP operator. This introduces a substructure with two arguments: Sandy, the beneficiary, is the argument of the preposition; the other argument, cake, is shared with the LS of the verb. The sharing of an argument with the verb differentiates this kind of PP from the adjunct ones, and it is because of this sharing that they are argument-adjuncts in the core. We propose here, however, that there are differences between the two types of argument-adjunct introduced above. The goal of motion and change-of-place verbs, which can be introduced by variable prepositions, are arguments of the verb and, as shown above, can function as controllers of pivots. Thus, they are proper arguments-adjunct in the core. The benefactive, instead, can hardly be said to be a notion that is inherently linked, in semantic terms, to a particular state of affairs. Although it usually appears in clauses where the main verb is an accomplishment predicate (Jolly 1993), beyond this, there is no other clue for predicting its appearance with particular semantic classes of predicates. Moreover, benefactives cannot control pivots in coordinated clauses, as is suggested by the Spanish example in (31). As noted above, this behaviour is a standard test for identifying arguments that belong to the core. (31)

Juani hizo un traje para Pedro j y lo todo un día. i /??j usó John made a suit for Peter and acc.3sg wore all one day ‘John made a suit for Peter and (he) wore it one complete day.’

Even though the suit was made for Pedro, the most natural reading of this example is that John was the one who wore it for a day (perhaps to check how it looked on him before giving it to Pedro). These facts suggest that, in marked contrast with the goal arguments of motion and change-of-place verbs, benefactives introduced by para are not core arguments. This suggestion is further supported by the fact that Spanish has an alternative mechanism for making core arguments out of benefactives; this is the dative or indirect object construction, where the beneficiary is introduced by the preposition a, and in which it can be doubled by the clitic pronoun le. We refer to Ibáñez Cerda (2009) for more in-depth discussion on the different behaviour of benefactives, on the one hand, and datives and recipients, on the other. Another independent criterion to determine that the PPs introduced by para in Spanish do not belong to the core is that there are no non-predicative uses of this preposition. The Spanish counterparts of the English verbs to long and to hope, which appear with non-predicative for, are mostly transitive (e.g. espero las buenas nuevas ‘I’m waiting/hoping for good news’). When they are used intransitively, they are coded with the preposition por, and not with para (e.g. espero por la buena nueva). In sum, PPs introduced by para in Spanish are always predicative and do not behave as core arguments: they code a participant that is not part of the verb semantics. Nevertheless, they are not like the adjunct PPs analysed in

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

445

446

SERGIO IBÁÑEZ CERDA

Section 10.4.1, because they share one argument with the LS of the verb and, although they are introduced by a predicative preposition, they do not take the whole clause as one of their arguments. Our proposal is that they are argument-adjuncts in the periphery. As pointed out by Jolly (1993) and Van Valin and LaPolla (1997), the preposition for is a good example of an adposition which can have the three main functions discussed above: it can be a non-predicative case marker in an oblique argument PP (32a), it can function as an argument-adjunct, as in (29), which is repeated in (32b) for convenience, and it can appear as a predicative preposition heading an adjunct phrase (32c). (32)

a. Lucy longs for a diamond ring. b. Robyn baked a cake for Sandy. c. Rita sang for the students.

As a way of unifying these three uses, Jolly (1993) proposes this extended semantic representation, which is abbreviated as PURP [. . .] in the logical structure in (30). (33)

want′ (x, LS) ^ DO (x, [LS1. . .CAUSE. . .LS2])

This can be paraphrased as ‘x wants some state of affairs (described in LS2) to obtain, and intentionally does LS1, in order to make LS2 happen’. The DO implies that the action in LS1 cannot be non-volitional. In the case-marking function of the preposition, only the first part of (33) gets projected. Lucy longs for a diamond ring can be also paraphrased as ‘Lucy longs to have a diamond ring’, so for really stands for a reduced proposition. Thus, the LS of (32a) is want′ (Lucy, [have′ (Lucy, diamond ring)]), a representation that exactly matches the first part of (33). As seen above, the argument-adjunct use implies the projection of the whole LS in (33). The complete representation of Robyn baked a cake for Sandy is thus as follows. (34)

[want′ (Robyn, [BECOME have′ (Sandy, cake)])] ^ [[DO (Robyn, [do′ (Robyn, Ø)] CAUSE [BECOME baked′ (cake)]] CAUSE [BECOME have′ (Sandy, cake)])]

The representation of the predicative use of for in (32c) is very similar to that in (34), except that there is no sharing of arguments between LS1 and LS2. The benefactive the students, the sole argument of LS2, is introduced as a new argument in LS1. (35)

10.5

[want′ (Rita, [BECOME entertained′ (the students)])] ^ [[DO (Rita, [do′ (Rita, [sing′ (Rita)])])] CAUSE [BECOME entertained′ (the students)]]

A New Typology of Adpositional Phrases in RRG

The current RRG three-way distinction between oblique core arguments, peripheral adjuncts and argument-adjuncts in the core has been very

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

helpful in overcoming the limitations of the traditional binary distinction between arguments and adjuncts. However, on closer examination, it appears that adpositional phrases can play several different functions, in terms of their relation to the clause they appear in. In fact, in Section 10.4.3, we argued that the argument-adjunct status can be ascribed to prepositional phrases with both core and peripheral behaviour, giving rise to a new distinction between argument-adjuncts in the core (goals of motion and change-of-place verbs) and argument-adjuncts in the periphery (benefactives), a distinction that accounts for the different syntactic behaviour of these kinds of adpositional phrases. Following this line of argument and drawing on the RRG distinction between semantic arguments and adjuncts, syntactic core and periphery, and predicative and non-predicative prepositions, Ibáñez Cerda (2011) proposes a principled system of eight logical types of adpositional phrases (APs). Each type is defined in terms of a combination of features relating to the three structural – semantic, syntactic and categorical – dimensions. The values for each dimension can be defined as follows: (a) in consideration of the semantic nature of the participant they code, APs can be [þ/ argument]; (b) in terms of their syntactic status, they can be [þ/ core]; and (c) with respect to the categorical status of their preposition, they can be [þ/ predicative]. The combination of features gives us the following set of possibilities. 1. Oblique Core Argument (þ) Argument (semantic level) (þ) Core (syntactic level) () Predicative preposition (intra-syntagmatic or categorical level) 2. Peripheral Adjunct () Argument () Core (þ) Predicative preposition 3. Argument-Adjunct in the Core (þ) Argument (þ) Core (þ) Predicative preposition 4. Argument in the Periphery (þ) Argument () Core () Predicative preposition 5. Argument-Adjunct in the Periphery (þ) Argument () Core (þ) Predicative preposition 6. Adjunct in the Core () Argument (þ) Core (þ) Predicative preposition

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

447

448

SERGIO IBÁÑEZ CERDA

7. Adjunct-Argument in the Core4 () Argument (þ) Core () Predicative preposition 8. Not labelled () Argument () Core () Predicative preposition Of the eight logical possibilities, only the last type, that is, the one defined by the negative value of all the features, is ruled out by functional principles: there is no way, it seems, in which a semantic adjunct, with no core privileges, can be introduced by a non-predicative preposition. In the following section, examples from English and Spanish of each of the other seven types are provided.

10.5.1

Oblique Core Argument

(þ) Argument (þ) Core () Predicative preposition These are canonical oblique core arguments in RRG terms: semantic arguments of the predicate in the nucleus of the clause, which are coded as APs introduced by a non-predicative adposition. This is the case with the recipient argument of transfer verbs in English (36a) and of the goal of motion and change-of-place verbs (36b–c), when they are introduced by canonical prepositions, like a and en in Spanish, respectively, as these prepositions are predictable from specific positions in the LS of those predicates. (36)

10.5.2

a. Tony gave the book to Peter. b. Juan fue a la tienda. John went to the store ‘John went to the store.’ c. Juan puso el libro en la caja. John put the book in the box ‘John put the book in the box.’

Peripheral Adjuncts

() Argument () Core (þ) Predicative preposition These are canonical peripheral clause participants. They are not semantically required by the predicate in the nucleus of the clause, and thus they are

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

adjuncts. They cannot function as controllers, and hence they must be outside of the core, that is, in the periphery. Finally, their preposition is predicative and, as such, it licenses the participant as its argument. The most straightforward examples of this type of PP are the temporal and locative adjuncts that function as settings of the state of affairs denoted by the predicate. (37)

10.5.3

a. John baked a cake after work. b. John baked a cake in the kitchen.

Argument-Adjunct in the Core

(þ) Argument (þ) Core (þ) Predicative preposition This specification of features implies that the participant is an argument, that it is in the core, and that its preposition is predicative, and this is in fact what differentiates this type of PP from the standard oblique core argument. As seen in Section 10.4.3, in Spanish this type of adpositional phrase is exemplified by the goal PPs of intransitive motion and change-of-place verbs, but only when they are introduced by non-canonical prepositions. These add a semantic specification to the referent of the verbal argument, and that is why they are predicative, although both the argument and the preposition variability are lexical features of the verbal item. As seen before, argumentadjuncts in the core can control pivots in coordinate and subordinate clauses, and hence they behave as core phrases. (38)

10.5.4

a. Juan fue hacia /hasta / adentro de la tienda. John went towards /as far as inside of the store ‘John went towards /as far as / inside the store.’ b. Juan puso el libro sobre / dentro de / debajo de la caja. John put the book on / inside of / under of the box ‘John put the book on / inside / under the box.’

Argument in the Periphery

(þ) Argument () Controller () Predicative preposition As expected from the iconic principle in the semantic–syntactic correlation, semantic arguments of the verb are canonically in the core, that is, they ‘naturally’ have access to some syntactic privileges that set them apart from canonical adjuncts, which in principle do not have those privileges and are peripheral. Nevertheless, there are cases in which a semantic argument can be deprived of all its possible syntactic privileges, and when this happens, it belongs to the core-level periphery. A well-established case of this type is the

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

449

450

SERGIO IBÁÑEZ CERDA

agent PP of the passive construction in English, in which the agent loses all its syntactic privileges and, as a result, it cannot be projected as a core argument. In fact, there are languages in which only the direct arguments are in the core; that is, all oblique arguments seem not to have control or pivot functions or any other type of syntactic privilege (Bickel 2003). Other possible examples of this type are those APs which code participants that are semantically required by the verb, for instance arguments, like en francés in (39a) and con un gesto in (39b), but in an alternative construction can be projected as subject or as direct object, as can be seen in (39c) and (39d). This behaviour indicates that they are indeed arguments. (39)

a. Pedro habló con María (en francés). Pedro talked with Mary (in French) ‘Pedro talked to María in French.’ b. Mauricio le declaró su amor a Tere (con un gesto). Mauricio dat.3sg declared his love dat Tere (with a gesture) ‘Mauricio declared his love to Tere with a gesture.’ c. Pedro habla francés perfectamente. Pedro speaks French perfectly ‘Pedro speaks perfect French.’ d. Su gesto lo declaró todo. His gesture acc.3sg declared all ‘His gesture declared it all.’

Importantly, when these participants are coded as APs, they are clearly optional and peripheral, so they have the status of arguments in the periphery.

10.5.5

Argument-Adjunct in the Periphery

(þ) Argument () Core (þ) Predicative preposition This type of AP is exemplified by the goal complement of Spanish motion verbs which inherently imply a source, like salir ‘to go out’ and partir ‘to leave’. (40)

a. Lolai salió a la callej y i/j estaba sucia. Lola went.out to the street and was dirty ‘Lola went out to the street and she/it was dirty.’ para la cabañaj y estaba sucia. b. Marisai salió i/*j Marisa went.out for the cabin and was dirty ‘Marisa went out to the cabin and she was dirty.’ bosquej y c. Ramóni partió al i/j estaba sucio. Ramón left to.art.def wood and was dirty ‘Ramón left for the wood and he/it was dirty.’ d. Toñoi partió para el bosquej y i /*j estaba sucio. Toño left for the wood and was dirty ‘Toño left for the wood and he was dirty.’

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

As proposed in Ibáñez Cerda (2005), a goal complement may appear with these verbs, although it is not inherently implied by them, because this participant belongs to an enhanced movement frame, which permits those verbs to be frequently coded with this argument. When the goal is introduced by the canonical preposition a, it behaves as a core argument, as can be seen from the fact that it can control a pivot in a coordinate clause as in (40a, c). When the goal of salir and partir is coded with a non-canonical preposition, like para in (40b, d), it cannot function as a controller, and this signals that it is not in the core. One can say that in these examples, the goal is an argument, but it is not syntactically focalized. Hence, in this case, the goal is an argument-adjunct in the periphery.

10.5.6

Adjuncts in the Core

() Argument (þ) Core (þ) Predicative preposition This AP type projects an adjunct, that is, a participant that is not required by the predicate in the nucleus of the clause. Nevertheless, the (þ) core feature indicates that, contrary to what is expected of adjuncts, this kind of AP exhibits some important syntactic properties which give it core status. This is the case with some Spanish manner, temporal and locative PPs, exemplified in (41a–c). (41)

a. La mujer viste con elegancia. The woman dresses with elegance ‘The woman dresses with elegance.’ b. María actuó en el momento adecuado. María acted at the momento right ‘María acted at the right moment.’ c. El puente fue construido en el lado este de la ciudad. the bridge was built in the side east of the city ‘The bridge was built in the east side of the city.’ d. *La mujer viste. the woman dresses e. *María actuó. María acted f. ??El puente fue construido. the bridge was built

The PPs in (41a–c) are obligatory for the grammaticality of these clauses, as can be seen from the comparison with their counterparts in (41d–f), which are pragmatically odd and need some addition to become acceptable (e.g. Finalmente, el puete fue construido ‘In the end the bridge was built’). Nevertheless, the participants projected through them are not semantically required by their respective nucleus predicates, so they are adjuncts in the core.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

451

452

SERGIO IBÁÑEZ CERDA

10.5.7

Adjunct-Arguments in the Core

() Argument (þ) Core () Predicative preposition This kind of AP is found in one of the two dative constructions of Spanish. As has been pointed out by Gutiérrez (1978) and Demonte (1994), among others, in Spanish there are two types of dative construction: one is formed with PPs which code recipient participants that are semantically required by the predicate in the nucleus of the clause, and these are proper core arguments (cf. 42a–b); the other dative construction exhibits PPs which code participants that are not inherent arguments of the verbs they appear with (see 42c–d). (42)

a. Susana (le) dio un regalo a María. Susana (dat.3sg) gave a gift dat María ‘Susana gave a gift to María.’ b. Dulce (le) ofreció un vaso de vino a Julio. Dulce (dat.3sg) offered a glass of wine dat Julio ‘Dulce offered a glass of wine to Julio.’ c. Mario le pintó la casa a Rosa. Mario dat.3sg painted the house dat Rosa ‘Mario painted Rosa’s house (for her).’ d. Ramiro le trabajó dos años al Sr. Rodríguez. Ramiro dat.3sg worked two years dat.def.art Mr. Rodríguez ‘Ramiro worked for Mr. Rodríguez during two years.’ e. *Mario pintó la casa a Rosa. Mario painted the house dat Rosa ‘Mario painted Rosa’s house (for her).’ f. *Ramiro trabajó dos años al Sr. Rodríguez. Ramiro worked two years dat.def.art Mr. Rodríguez ‘Ramiro worked for Mr. Rodríguez during two years.’

As can be seen from the contrast between (42c–d) and (42e–f), the PP of (42c–d) obligatorily co-occurs with the clitic le, whereas this is not the case with (42a–b). However, both types of dative PP can control pivots in nonfinite subordinate clauses. (43)

Pedroj para _ j leer. a. Juani dio un libro a John gave a book dat Peter for read ‘John gave a book to Peter to read.’ hizo un traje a Pedroj para _ b. Juan i le John dat.3sg made a suit dat Peter for ‘John made Peter a suit to wear at the party.’

j

usar en la fiesta. wear at the party

Therefore, we conclude that the second type of dative AP is an adjunctargument in the core.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

10.6

Conclusion

This chapter introduced the standard RRG theory of adposition functions and adposition assignment (Foley and Van Valin 1984; Van Valin 1991, 1993, 2005; Van Valin and LaPolla 1997; Jolly 1993). Starting from the three-way classification of AP types that is standard in RRG, and capitalizing on the distinction between semantic argument and adjunct, syntactic core and periphery, and the categorical distinction between predicative and non-predicative preposition, we then discussed the principled system of adpositional phrase types which was first proposed in Ibáñez Cerda (2011).

References Alarcos Llorach, Emilio. 1972. Estudios de gramática funcional del español. Madrid: Gredos. Bickel, Balthasar. 2003. Clause Linkage Typology. Lecture series delivered at the 2003 International Role and Reference Grammar Conference, UNESP, Sao Jose do Rio Preto, Brazil. Blake, Barry J. 2001. Case. Cambridge: Cambridge University Press. Bresnan, Joan. 1982. The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press. Bresnan, Joan. 2001. Lexical-Functional Syntax, Malden, MA: Blackwell. Cano Aguilar, Rafael. 1999. Los complementos de régimen verbal. In I. Bosque and V. Demonte (eds.), Gramática descriptiva de la lengua española, Vol. 2, 1807–1854. Madrid: Espasa. Demonte, Violeta. 1994. La ditransitividad en español: Léxico y sintaxis. In V. Demonte (ed.), Gramática del español, 431–470. Mexico: Colegio de México. Dik, Simon C. 1989. The Theory of Functional Grammar, Part 1. Dordrecht: Foris. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Gruber, J. S. 1965. Studies in Lexical Relations. PhD dissertation, MIT. (Published with revisions as Lexical Structures in Syntax and Semantics, Amsterdam: North-Holland, 1976.) Gutiérrez, Salvador. 1978. Sobre los dativos superfluos. Archivum XXVII– XXVIII: 415–452. Haiman, John. 1980. The iconicity of grammar: Isomorphism and motivation. Language 56: 515–540. Hernández Alonso, César. 1990. En torno al suplemento. Anuario de Letras XXVIII: 5–25. Ibáñez Cerda, Sergio. 2005. Los verbos de movimiento intransitivos del español. Una aproximación léxico-sintáctica. Mexico: ENAH-UNAM.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

453

454

SERGIO IBÁÑEZ CERDA

Ibáñez Cerda, Sergio. 2009. Prepositional phrases in RRG: A case of study from Spanish. In L. Guerrero, S. Ibáñez and V. Belloro (eds.), Studies in Role and Reference Grammar, 469–490. Mexico: UNAM. Ibáñez Cerda, Sergio. 2011. PP types in RRG: A top-down approach to their classification. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 200–217. Newcastle upon Tyne : Cambridge Scholars Publishing. Ibáñez Cerda, Sergio. 2021. Two-themes constructions and preposition assignment in Spanish. In Robert D. Van Valin, Jr. (ed.), Challenges at the Syntax–Semantics–Pragmatics Interface: a Role and Reference Grammar Perspective, 189–211. Newcastle upon Tyne: Cambridge Scholars Publishing. Jolly, Julia. 1993. Preposition assignment in English. In Van Valin (ed.), 275–310. Martínez García, Hortensia. 1986. El suplemento en español. Madrid: Gredos. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62: 56–119. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon (ed.), Grammatical Categories in Australian Languages, 112–171. Canberra: Australian Institute of Aboriginal Studies. Silverstein, Michael. 1981. Case marking and the nature of language. Australian Journal of Linguistics 1: 227–246. Silverstein, Michael. 1993. Of nominatives and datives: Universal Grammar from the bottom up. In Van Valin (ed.), 465–498. Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and grammatical relations. Natural Language and Linguistic Theory 9: 145–194. Van Valin, Robert D., Jr. (ed.). 1993. Advances in Role and Reference Grammar. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr., and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Notes 1 RRG also posits a third parallel representation for information structure (see Chapter 11), which is not relevant to our discussion. 2 Since Van Valin (2005: 44) the operator BECOME can be decomposed into PROC & INGR, although for simplicity we will not adopt this notation here. 3 The preposition con also appears in semantically determined LS contexts, marking the non-undergoer theme of two-theme verbs, like conectar ‘to connect’, reunir ‘to gather together’ and comparar ‘to compare’, among many others (Ibáñez Cerda 2021).

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

Adposition Assignment and Adpositional Phrase Types

4 Almost all of the proposed labels for the different PP types have been used before in RRG studies, but they also arise directly from the combination of features used in this work. For example, the label Argument-Adjunct in the Core comes from the fact that the PP has the features (þargument), (þcore) and (þpredicative preposition). This last feature gives the label the ‘adjunct’ part of the name, as it is a standard characteristic of adjuncts to be formed with a predicative preposition. Nonetheless, the adjunct part is attached to the ‘argument’ part of the name, because coding a participant which is an argument is the main feature of this type of PP. I am now introducing the label ‘Adjunct-Argument in the Core’ by means of the same procedure: the combination of features is (argument), (þcore) and (predicative preposition). That is, the main feature is that the participant coded in this type of PP is a semantic adjunct, so the first part of the proposed name is ‘adjunct’. The ‘argument’ part of the name comes from the fact that it is a standard characteristic of arguments to have nonpredicative prepositions. Finally, the ‘core’ part indicates that the PP has some syntactic privileges.

https://doi.org/10.1017/9781316418086.013 Published online by Cambridge University Press

455

11 The RRG Approach to Information Structure Delia Bentley

11.1

Introduction

In this chapter we introduce the theoretical constructs adopted by RRG in the treatment of information structure, and we address the question of where information structure fits in the architecture of a grammar.* Different frameworks give different answers to this question. RRG is a parallel architecture theory (Jackendoff 2002), which aims to describe and explain cross-linguistic variation in the interaction of semantics, syntax and discourse pragmatics. To achieve this objective, RRG posits a direct linking between semantics and syntax. Discourse constitutes an independent component of grammar whose role is pervasive in the linking. A correlate of this key aspect of the RRG approach to information structure is that, while some syntactic positions are motivated in pragmatic terms, and it is recognized that the left portion of the clause has pragmatic prominence in discourseconfigurational languages, there is no universal association of specific syntactic positions or projections with specific discourse roles. Put differently, the discourse properties and functions of individual syntactic constituents are not contingent upon the placement in – or the displacement to – particular syntactic positions. Rather, syntax and discourse are independent, though crucially interfacing, components of grammar. In addition, discourse-related meaning (for example, the distinction between the information that has already been given and the new information that is provided with the utterance) is not only expressed syntactically, but also in prosody, morphology and even in lexical choices, as is the case with the selection of verbs with particular types of argument structure. With respect to the constructs that are key in the treatment of information structure, RRG follows Lambrecht (1994) in drawing a distinction

* I am grateful to Mitsuaki Shimojo, with whom I discussed several of the issues dealt with in the chapter.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

between the notions of topic and focus, which are defined in terms of the relations established between, and within, pragmatically structured propositions and, on the other hand, the status of discourse referents in the minds of the speech act participants. A key role in the RRG treatment of information structure is also played by the distinction between presupposition and assertion. The presupposition of an utterance is the information that is shared by speaker and hearer prior to a sentence being uttered. The assertion is the information that is known to the hearer as a result of the sentence being uttered. Pragmatic relations and states will be dealt with in Section 11.2. Then, in Section 11.3, we introduce the positions on the layered structure of the clause that are motivated in terms of the encoding of pragmatic relations. The contrast between presupposition and assertion is reflected in focus structure (Lambrecht 1994), that is, the conventional associations of focus meanings with sentence forms. In accordance with the goal of typological adequacy (Van Valin and LaPolla 1997: 8), these associations are assumed to vary across languages. Focus structure is discussed in Section 11.4, where we also introduce the distinction between the potential domain of focus in the syntax of a given language and the actual domain of focus in an utterance of that language. We then provide examples of cross-linguistic variation in focus structure that depends on the variation in the potential domain of focus as well as on the respective roles of prosody and syntax in the expression of discourse-related meaning (Section 11.5). In Section 11.6 we discuss the pervasive role of discourse in linguistic production and processing, illustrating it with reference to various steps in the semantics–syntax linking. Lastly, we draw some conclusions (Section 11.7).1

11.2

Pragmatic States and Pragmatic Relations

RRG adopts Lambrecht’s (1994: 49) distinction between non-relational and relational constructs in information structure.2 Thus, on the one hand, it is concerned with the status of the discourse referents in the minds of the speech participants: whether they are already established or new from the perspective of the hearer or both interlocutors, and, if they are new, whether they can be individuated or, alternatively, related to previously introduced referents. On the other hand, RRG takes information to be organized relationally and studies the relation between the presupposition, that is, the information that can be taken for granted when an utterance is produced, and the assertion, that is, the information which is known to the hearer as a result of the utterance being produced. Although the two kinds of relational and non-relational construct must be kept separate, and, indeed, research in RRG suggests that they tend to be relevant to different steps in the linking (Section 11.6), in due course we will highlight some alignment tendencies

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

457

458

DELIA BENTLEY

between the cognitive states of discourse referents and the packaging of information in the utterance (Section 11.2.2). In the remainder of this section, we discuss the non-relational and relational information structure constructs in turn.

11.2.1 The Pragmatic Status of Discourse Referents When a discourse referent is introduced for the first time, it is new from the perspective of the speech participants or, simply, that of the hearer.3 Following Prince (1981), RRG differentiates between two types of new referent, called brand-new anchored and unanchored. The former type of referent, but not the latter, is explicitly related to referents that have already been established, or can be identified, in discourse. Examples of referents with these two kinds of cognitive status are provided by the noun phrases in italics in (1a–b). (1)

a. I saw a student outside your door. b. I saw a student from the linguistics department outside your door.

Not only is the referent of the italicized expression introduced as new in both (1a) and (1b), as testified, in English, by the use of an indefinite article, but in neither case can it be individuated by the speech participants. Thus, the pragmatic state of this referent is unidentifiable in both cases, although in (1b) the new referent is explicitly related to the known referent the linguistics department. Contrastingly, in (1a), there is no such correlation.4 After they have been introduced, referents are normally treated as identifiable, that is, as referents which can be individuated by the hearer. In English, this may involve marking with a definiteness operator, like the in (2a) and that in (2b), although we should note that cross-linguistically definiteness does not necessarily signal identifiability and, vice versa, identifiability does not necessarily correlate with definiteness. (2)

a. The student I saw outside your door asked me if I knew where you were. b. That student asked me if I knew where you were.

Observe that the individuals referred to by the italicized expressions in (2a) and (2b) are only identifiable insofar as this is not their first mention, and thus a linguistic representation – or a file – has been created for them in discourse (Lambrecht 1994: 77, see also Chafe 1976). Proper identifiability involves the possibility for the hearer to individuate a single individual or entity, or a single set of individuals or entities, that can be designated with the linguistic expression chosen by the speaker. This case is illustrated in (3). (3)

The head of department is a semanticist.

Although many individuals can be designated with the expression head of department, in this case the use of the definite article signals the speaker’s

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

expectation that the hearer will be able to single out the sole individual that is being referred to with this expression in the given utterance. The notion of frame or schema (Fillmore 1982; Lambrecht 1994: 90) can be relevant to the treatment of a referent as an identifiable one. A frame or schema is a system of associated concepts that are related in such a way that to understand one of them one has to understand the whole system. These systems of concepts justify the treatment of certain referents as identifiable, because, when one of them is introduced, all the associated ones become available. For example, if one is familiar with the concept of an academic department, one will know that departments have heads. Once the notion of department has been introduced, then the concept of the department’s head becomes available. Referents like those in the italicized expressions in (2a–b), which are to some extent cognitively available with reference to the linguistic context or (co-)text, are said to be textually accessible (recall that the intended context of 2a–b is 1a–b). In some cases, these are referents that have been introduced at an earlier stage in discourse and have not been mentioned for a while. Other forms of cognitive availability are inferential or situational accessibility, namely the property of an entity or an individual to be individuated in the physical context, as is the case with (4b) or by means of some relation with an element in the physical or linguistic context. The latter case is illustrated in (5), where the referent of the noun phrase his head of department may not be known to the hearer, but this referent is nonetheless accessible by virtue of its relation to another referent (the latter is encoded by the third-person pronoun his). (4)

a. «Which of these two classrooms do you teach in?» b. «The large one.»

(5)

His head of department is a semanticist.

Accessible referents are not in the current focus of attention prior to the sentence being uttered (Van Valin and LaPolla 1997: 200). Identifiable referents that are in the hearer’s long-term memory, but neither in the short-term memory nor in the current focus of consciousness, are called inactive, whereas referents that are in the hearer’s current focus of consciousness are activated or active.5 An active referent is encoded by them in (6). (6)

Remember John and Pete, the identical twins from high school? I saw them in the lecture today.

Active referents can be deactivated and downgraded to a state of semiactivity or mere accessibility due to the attentional and short-term memory limitations of the interlocutors. In Figure 11.1 we provide a graphic representation of the pragmatic states introduced above. The different strategies exhibited by languages for the coding of referents, for example the marking with definiteness or indefiniteness

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

459

460

DELIA BENTLEY

Referential

identifiable

unidentifiable

active

accessible

inactive

textually

situationally inferentially

anchored

unanchored

Figure 11.1 The cognitive states of referents in discourse (Van Valin and LaPolla 1997: 201)

operators, correspond to different degrees of cognitive accessibility, reflecting the speaker’s assumptions on the addressee’s knowledge and attention at the time of the utterance, and ensuring that the least amount of processing effort is made by the hearer (Van Valin and LaPolla 1997: 201). An important background assumption is, thus, that the speaker makes their utterance congruent with their assumption of the interlocutor’s mental knowledge and disposition. Zero marking normally indicates activation, while pronominal marking tends to correspond to activation or accessibility, and so on. It should nonetheless be borne in mind that, although such tendencies are recurrent cross-linguistically, languages differ in their coding of the pragmatic states of discourse referents (Gundel et al. 1993), and such cross-linguistic differences are within the domain of investigation of RRG. To give but one example, the importance of the notion of cognitive accessibility is evidenced by Belloro’s (2004, 2015) RRG account of object doubling in Argentinian Spanish dialects. (7)

Argentinian Spanish (Belloro 2015: 8–9) a. Le gusta el cine a Juan. him please.3sg the cinema to John ‘John likes cinema’ b. Lo vi a él. him see.pst.1sg acc him ‘I saw him.’

Belloro points out that the object doubled by a clitic is not normally indefinite in the dialects that she investigates. This fact would be puzzling, should the key property of the doubled object be specificity, as was claimed in earlier literature. Specifics can, in fact, be formally indefinite. While a semantic analysis in terms of specificity is challenged by the evidence, an information structure account in terms of identifiability would also seem to be problematic on empirical adequacy grounds, since it is not the case that

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

all definite objects are doubled. The failure of some definite objects to be doubled is captured with reference to the gradience of identifiability (see Figure 11.1). Belloro (2015: 39) argues that the doubled objects encode identifiable referents that are not active, but rather accessible. This claim explains the facts exhaustively, predicting that the construction will not exhibit indefinites, since indefinites encode unidentifiable referents or specifics which are unidentifiable from the perspective of the hearer. On the other hand, it can host generics, in which case the discourse interlocutors identify a class (Lambrecht 1994: 82, 88). An example of doubling of a generic object is given here. (8)

Argentinian Spanish (Belloro 2015: 33) [Nuestro voseo] los divierte mucho a los peruanos our voseo them amuses much acc the Peruvians ‘Our voseo amuses Peruvians a lot.’

Object doubling in Argentinian Spanish thus turns out to be a strategy to mark identifiable referents which may have temporarily fallen out of the current focus of the hearer’s attention, and Belloro’s (2015) analysis of object doubling in terms of the cognitive accessibility of the referent of the object has explanatory power. Other pragmatic states have been the object of research in RRG, for example saliency, or persistence in discourse (Shimojo 2009), although due to space limitations, we will not offer an account of such notions here. In the next section, we turn to the pragmatic relations between units of information that make the flow of information possible.

11.2.2 The Pragmatic Relations between Information Units The construal and processing of information is not only analysed in terms of the status of the discourse referents in the minds of the speech participants, but also in terms of the relations between units of information. An information unit (Van Valin 2005: 77) typically corresponds to the information content of a single syntactic constituent, although it can also refer to the content of a unit that is larger or smaller than a syntactic constituent. Following Gundel (1988) and Lambrecht (1986, 1994, 2000), RRG defines topic as what the speaker wants to request information about, or increase the addressee’s knowledge of, or get the addressee to act with respect to (Van Valin 2005: 68). The definition of topic is, therefore, inherently relational, in that it makes reference to the information unit about which new information is being requested or conveyed in the utterance. Thus, the information unit in italics in (9) is not a topic by virtue of its information content, but rather it can be considered to be a topic because the utterance increases the addressee’s knowledge about it. (9)

The head of department has resigned.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

461

462

DELIA BENTLEY

The topic is often presupposed, insofar as its status as a matter of interest or concern in the utterance is part of the pragmatic presupposition. As was mentioned in Section 11.1, the presupposition of an utterance is the set of relevant propositions, or the information, which is shared by speaker and hearer prior to the utterance.6 The topicality of the information unit in italics in (9) is clear in the context of questions like ‘What did the head of department do?’, since in this context, it is explicit that the head of department is at issue. Even in the absence of such a question, however, it can be claimed that the unit the head of department is topical in (9). Indeed, in light of recent scholarship (Cruschina 2012, 2015; Bentley et al. 2016, building on Reinhart 1981; Frascarelli and Hinterhölzl 2007; Frascarelli 2017, among others) we propose that a distinction ought to be drawn in RRG between aboutness and referential topics. Aboutness topics are referentially new because they are introduced with the utterance, but they are topics in the relational sense of what the utterance increases the addressee’s knowledge about. Referential topics, on the other hand, are shared and referentially old, and, hence, they are typically encoded by anaphoric expressions. In terms of this distinction, the italicized information unit in (9) can be said to be an aboutness topic in an out-of-the-blue context, since it is the subject matter of the sentence. By contrast, the utterance in (10b) has a referential topic, whose status as the current matter of interest is part of the presupposition. (10)

a. «What did the head of department do?» b. «She resigned.»

Referential topics can – but need not – be detached in extra-clausal positions, with the consequent language- and construction-specific requirement for a clause-internal resumptive pronoun (cf. 11a–b). There is no extraclausal detachment with aboutness topics (cf. 9). (11)

a. As for the head of department, she resigned. b. As for the head of department, I saw her in the staff room.

The distinction between aboutness and referential topic highlights the conceptual difference between what the speaker wants to increase the addressee’s knowledge of, or to request information about, and the presupposed, shared, common ground, which the speaker relies upon, and which both interlocutors know to be a current matter of interest. To be sure, a referential topic can be both what the sentence is about and what speaker and hearer know to be at issue. The examples in (10b) and (11a–b) are illustrations of this, in that their referential topics are also what the sentence increases the addressee’s knowledge of. According to an anonymous reviewer, the fact that a referential topic can also be what a particular utterance is about suggests that the notion of referential topic ought to be subsumed within that of aboutness topic. While acknowledging that a referential topic can have the aboutness function in an utterance, we

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

believe, however, that the two notions ought to be kept apart, in that only referential topics are established in previous context. In fact, in textual sequences, when a new aboutness topic is introduced, a previous aboutness topic can take the role of a referential topic, which, however, will not play a role in the topic-comment articulation of the utterance. The aboutness topic of (12a), the proposal, turns into a referential topic in (12b), where the new aboutness topic the teaching assistants is introduced. (12)

a. «The proposal was advanced unanimously by the students.» b. «That’s right, but the teaching assistants were sceptical about it. In particular, they felt that their role was misrepresented . . .»

This is an important reason for distinguishing referential and aboutness topics. In Section 11.4, we shall point out some other advantages of this distinction, which emerge in the analysis of focus structure. In this context, we should stress that both kinds of topic are relational, insofar as a topic is never identified in terms of its information content, but rather by virtue of its relation to the information provided in the sentence or in the broader text or discourse. We should also mention in passing that since an utterance like (9) (The head of department has resigned) presupposes knowledge about academic departments having a head, it could be argued that the referent head of department is part of the presupposition regardless of whether the sentence is uttered in an out-of-the-blue context. We propose, however, that the kind of knowledge mentioned here pertains to the notion of frame or schema (see Section 11.2.1) and that this notion is relevant to the pragmatic states of discourse referents rather than to their relations. Finally, the concept of frame or schema is not to be confused with that of frame-setting topic, which is an information unit that specifies the spatio-temporal coordinates within which the eventuality described in the following utterance holds true, as is the case with After the meeting in After the meeting, we all went for a drink. In RRG, frame-setting topics are normally assumed to occupy the extra-clausal pre-detached position (see Chapter 1, Section 1.6.1 and Figure 1.34).7 The focus of an utterance is defined in RRG as what is asserted, in a declarative sentence, and what is questioned, in an interrogative one (Van Valin 2005: 69). Let us consider the following conversational exchange. (13)

a. «Who has resigned?» b. «The head of department.»

The question in (13a) presupposes the proposition that someone has resigned. The information conveyed by the reply in (13b), in turn, is not the content of the information unit the head of department as such, but rather the content of this unit in relation to the presupposed proposition that someone has resigned. Only the information unit requested in the question and that provided in the answer are focal in the respective

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

463

464

DELIA BENTLEY

Most acceptable

Active Accessible Inactive Brand-new anchored Brand-new unanchored

Least acceptable

Figure 11.2 The Topic Acceptability Scale (Van Valin and LaPolla 1997: 204)

utterances, but they are so with respect to the whole proposition of which they are components.8 Accordingly, focus is defined with reference to the pragmatic relation between two components of a proposition, the one being presupposed, the other being the new contribution of the assertion, or the contribution sought with a question. Thus, the notion of focus is relational, on a par with that of topic. In the last analysis, the notions of topic and focus are to be defined in terms of the speaker’s assessment of the pragmatic relations between the components of a proposition in a given discourse context. These notions will be discussed further in subsequent sections, after we have introduced other aspects of the RRG approach to information structure. Before we conclude this section, we return to the issue of the alignment of cognitive states with pragmatic relations. Underlying the correlation between these two types of pragmatic notion is the following principle: the more accessible the topical information unit is, the less effort is required to process an utterance. Accordingly, accessibility strongly aligns with topicality. On the basis of this principle, a Topic Acceptability Scale was proposed by Lambrecht (1994: 165) and a slightly revised version was then adopted in RRG (see Figure 11.2). Recall that in the discussion of the cognitive states of discourse referents, we pointed out that different strategies are adopted for the encoding of different degrees of accessibility, thus ensuring that the least amount of processing effort is made by the hearer. Zero marking tends to express activation, while pronominal marking tends to correspond to activation or accessibility, and so on. If we combine these coding strategies, ordered in terms of the cognitive states that they express cross-linguistically, with the Topic Acceptability Scale, we obtain the hierarchy shown in Figure 11.3, which expresses the likelihood of coding of topic and focus by means of the strategies that mark arguments in terms of their relative accessibility.

−

Figure 11.3 Coding of referents as topic and focus (Van Valin and LaPolla 1997: 205)

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

11.3

Information Structure and the Layered Structure of the Clause

Before we discuss how presupposed and asserted information is packaged in sentence structure, we briefly introduce the positions on the layered structure of the clause that tend to be associated with discourse functions. These are the pre-detached position (PrDP), which hosts referential and framesetting topics, and the post-detached position (PoDP), which hosts afterthoughts or referential topics.9 Contrastingly, the pre- and post-core slots are filled by foci, although there can be restrictions on the kinds of foci that they admit. In English, the pre-core slot hosts preverbal wh-words (14a) and fronted constituents (14b). (14)

a. What did you read? b. This book I have not read.

The pre-core slot can also, however, host topics in languages with a V2 constraint on word order (Diedrichsen 2008). As for the post-core slot, it has been identified as the position of secondary foci that non-canonically occur in postverbal position in Japanese, which is a verb-final language (Shimojo 1995). The pre-detached position can iterate, thus allowing the utterance to have several topics, while the pre- and post-core slots cannot be repeated. The core-internal positions and the peripheries of the various layers of the clause can also host constituents with particular discourse roles. For example, aboutness topics occur in the core-internal immediately prenuclear position in SV languages. However, these positions are not pragmatically motivated. In other words, they are not defined in terms of the discourse roles that they host or in terms of the constructions in which they are involved and which have pragmatic conditions on their occurrence. In Figure 11.4, we indicate with grey shading, and lack thereof, the positions that are pragmatically motivated and those that are not, respectively. With respect to the issue of the relation between discourse and syntax, RRG differs fundamentally from the syntactocentric approaches to linguistic theory. On the one hand, it contrasts with Chomsky’s (1995) Minimalist Program, which, at least in its original formulation, denied any status to discourse in the syntactic computation, strictly associating discourse with interface effects. On the other hand, it diverges from Cartography (Rizzi 1997, 2006), where movement operations targeting pragmatic positions start

PrDP Pre-Core Slot topic

focus (and topic)

Core Nuc Core Post-Core Slot PoDP focus

topic, afterthoughts

Figure 11.4 Pragmatically motivated (shaded) positions in the layered structure of the clause

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

465

466

DELIA BENTLEY

from a thematic position and land in a position associated with interpretive, scope or discourse effects. In RRG, arguments acquire their thematic roles by virtue of their position in logical structure and not in syntax. Accordingly, the occurrence of an information unit in a pragmatically motivated position does not involve movement but rather the selection of a syntactic template, from the syntactic inventory of the given language, which has the positions needed to spell out the discourse functions of the given utterance. Importantly, the pragmatically motivated positions of the layered structure of the clause are not claimed to be universal (Van Valin 2005: 8). Therefore, while some languages have the possibility of expressing discourse roles in extracore syntax, by selecting a template with one or more of the positions highlighted in grey in Figure 11.4, others solely spell out topic and focus with morphological and prosodic clues within the core (see Sections 11.5, 11.6 and Chapter 12 of this volume for further discussion). No universal one-to-one relation is assumed between discourse roles and functional projections in RRG.

11.4

Focus Structure

The cross-linguistic variation in how presupposed and asserted information is packaged in sentence structure is a concern of RRG. Following Lambrecht (1994: 221–238), the conventional associations of information meanings with sentence forms are referred to as focus structure. Three principal focus structure types are distinguished; two of these involve the interplay of presupposed and asserted information within the clause, whereas the third one does not. Predicate focus is considered to be the universally unmarked type of focus structure. It expresses the type of proposition normally called categorical judgement or statement, which was first identified by the philosophers Brentano and Marty (see also Kuno 1972 and Sasse 1987, among others). This type of proposition asserts or denies that all or some of the members of a set that is defined by a category are members of another set defined by another category (e.g. Some politicians are honest). In pragmatic terms, a predicate-focus structure is construed as a comment about an information unit which is part of the presupposition. The assertion, in turn, is an aboutness relation between the focal predicate phrase and the presupposed topic. The examples in (15a–d), where small caps indicate prosodic prominence within the asserted predicate, are from Lambrecht (1994: 223). (15)

Question: «What happened to your car?» Answer: a. «My car / it broke down.» b. Italian «(La mia macchina) si `e rotta.» the my car refl be3sg break.ptcp

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure c. French «(Ma voiture) my car d. Japanese «(Kuruma-wa) car-top

elle est en panne.» it be3sg in breakdown koshoo-shi-ta.» breakdown-do-pst

The topic of predicate-focus structures is typically treated as the subject of the sentence. It has to be overt, albeit normally pronominal, in English and French, but not in Italian or Japanese. In Japanese, a -wa-marked noun phrase only occurs as the topic of a predicate focus structure if it is contrastive (Shimojo 2010: 323 and Section 11.4). In predicate-focus structures of languages without a grammaticalized subject, for example Mandarin Chinese (LaPolla 1990, 1993, 1995 and see Chapter 5), the topic is not treated as a subject. (16)

Mandarin (Van Valin and LaPolla 1997: 207, from Li and Thompson 1976) N`ei xie¯ sh` u, sh` ushe¯n d` a. that few tree trunk big ‘Those trees, the trunks are big.’

Sentence structures which involve a topic vs. comment opposition but no direct syntactic relation between the topic and the comment are also found in languages with a grammatical relation subject, although these are usually pragmatically and syntactically more complex than the examples in (15). An example is given here. (17)

Students, you have to remind them of the deadlines for coursework submission every second week.

The topic of the utterance in (17) is a detached information unit in the predetached position (see Figure 11.4), whereas the rest of the sentence is a syntactic core (you have to remind them of the deadlines for coursework submission) with an additional peripheral adverbial (every second week), both being within the domain of the assertion. As should be clear from the above, a defining feature of predicate focus is the availability of the subject matter at issue from the presupposition. Another defining feature of predicate focus can be explained with respect to an important distinction made by RRG, that is the contrast between the potential focus domain, which is the syntactic domain in the sentence in which focus can occur in a given language, and the actual focus domain, which is the syntactic constituent or constituents of a given sentence that is or are in focus. In light of this distinction, predicate focus is characterized by the limitation of the actual focus domain to the syntactic core minus the subject-topic. This is shown in the speech act projection in (15a), where the continuous line indicates the actual focus domain, while the dotted line indicates the potential focus domain in English.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

467

468

DELIA BENTLEY

(15)

a.

[ Clause [ Core It [ Nuc broke down]]].

Speech Act Projection

Sentence focus differs from predicate focus insofar as the matter at issue is not introduced previously in discourse, the whole proposition is asserted, and the whole sentence is in the actual focus domain. Given that they do not have a presupposed topic, sentence-focus structures occur out of the blue. Lambrecht (1994: 223), the source of the following examples, also claims that they reply to the question what happened? (18)

(18)

Question: «What happened?» Answer: a. «My car broke down.» b. Italian `e «Mi si rotta poss.dat.1sg refl be3sg break.ptcp c. French «J’ai ma voiture qui est I have.1sg my car rel be.3sg d. Japanese «Kuruma-ga koshoo-shi-ta.» car-nom break.down-do-pst

a.

la macchina.» the car en panne.» in breakdown

[ Clause [ Core My car [Nuc broke down]]].

Speech Act Projection

The fact that the argument is in focus is reflected in the prosody of the English construction in (18a) and in the morphosyntax of the constructions in (18b–d). In both (18b) and (18c), the focal argument occurs postverbally. Since French poses strict constraints on verb–subject order, the focal argument is not encoded as a subject but rather as the undergoer of avoir ‘have’. Verb–subject order can also be found in English, although a locative phrase or dummy there is required preverbally (There arrived three buses at the station) and there are verb-class restrictions on this construction (Levin and Rappaport Hovav 1995: 215–277). In Japanese, the focal argument is marked with the nominative marker ga. The same focus meanings can thus be expressed by different prosodic, morphological and syntactic structures in different languages. Given that, in sentence focus, information is not structured as a contrast between presupposed and non-presupposed information, this type of focus structure expresses thetic statements, that is, unstructured statements which describe a situation or an event, rather than making a statement about a previously introduced individual or entity. Following Lambrecht

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

(1994), these sentences have traditionally been analysed in RRG as topicless structures. However, scholarship on information structure increasingly suggests that all utterances have a topic. Erteschik-Shir (1997) claims that a predication is a function that maps a topic to a proposition, and that a truthvalue is assigned to the proposition in terms of the truth-value of the predicate with respect to its topic. Utterances that do not have an overt topic are argued by Erteschik-Shir (1997: 8, 26–29) to be assessed in terms of a stage topic, which defines the spatial and/or temporal parameters of the situation or event. The adoption of the contrast between referential and aboutness topics allows us to characterize properly the information structure of out-of-the-blue statements, capturing the insight that all utterances have a topic. Thus, if uttered out of the blue, the sentence in (9) (The head of department has resigned) can be said to be a sentence-focus structure, which comprises an aboutness topic (the head of department) that has not been introduced previously in discourse. This sentence-focus structure differs from predicate focus, in that its topic is not part of the presupposition or, put differently, it is not a referential topic, and, thus, it is within the actual focus domain. In this analysis the topic-comment articulation of the utterance does not correspond to its partition into a background contrasted with the actual focus domain, and we shall return to this point below. As for the stage topic of a seemingly topicless sentence-focus utterance, in RRG it can simply be retrieved from discourse and it need not be assigned a syntactic position.10 The third type of focus structure is called narrow focus (Van Valin 2005: 71). It has a single information unit – corresponding to a constituent or part thereof – in the actual focus domain, whereas the remainder of the sentence is part of the presupposition. The focal information unit can be an argument, an adjunct, or the verb. The examples in (19) are drawn from Lambrecht (1994: 223) (though the Italian examples in 19b have been slightly modified). (19)

Question: «I heard your motorbike broke down?» Answer: a. «My car broke down / It was my car that broke down.» b. Italian `e «Mi si rotta la macchina.» poss.dat.1sg refl be3sg break.ptcp the car ` `e «E la macchina che mi si rotta.» be.3sg the car that poss.dat.1sg refl be3sg broken `e «La macchina mi si rotta.» The car poss.dat.1sg refl be3sg broken c. French «C’est ma voiture qui est en panne.» it be.3sg my car that be.3sg in breakdown d. Japanese «Kuruma-ga koshoo-shi-ta.» car-nom break.down-do-pst

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

469

470

DELIA BENTLEY

(19)

a.

[ Clause [ Core My

CAR [ Nuc

broke down]]].

Speech Act Projection

Lambrecht refers to this focus-structure type as argument focus to distinguish it clearly from predicate focus. Indeed, it could be argued that, when the verb phrase is in focus, narrow focus cannot be distinguished from predicate focus. This would be misleading, however, since, in predicate focus, the domain of focus consists of the whole core minus the subjecttopic, and this definition subsumes structures with two coordinated or cosubordinated cores (cf. 20). In narrow focus, instead, a single constituent or part thereof is in focus (cf. 21).11 (20)

a. «What did you do this morning?» b. « [Core I [[ Nuc WORKED] [ Per IN THE LIBRARY]] [AND THEN [ Nuc LEFT ]]].»

Speech Act Projection (21)

a. «When did you buy this house?» b. « Actually, [ Core I [ Nuc RENT] it].»

Speech Act Projection

At this juncture it should be noted that the distinction of actual focus domain (AFD) and background (non-AFD) does not always correspond to the topic–comment opposition, as was the case with (15a). Thus, in the reply to the question in (22a), the topic is he, since the utterance is about the referent of this pronoun and its antecedent John, whereas the verb bought is both part of the background (the non-AFD) and of the comment on the topic John (see Van Valin and Latrouite in Chapter 12 of this volume for further discussion of this point). (22)

a. «What did John buy?» b. «[ Topic He] [ Comment bought A CAR].»

Although bought a car is the comment on the topic he, only a car is in the actual focus domain, since it is in narrow focus.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

A distinction has been drawn between marked and unmarked narrow focus (Van Valin and LaPolla 1997: 209–210). The latter type of narrow focus occurs in an unmarked focus position in the clause, such positions being clearly identifiable in many languages. The two English sentences below, for instance, both illustrate unmarked narrow focus, in that the final position in the core is the unmarked focus position in this language (cf. 23a), the sole exception being wh-words, which are focal and occur by default in the precore slot (cf. 23b). (23)

a. [Core He gave the book to Pat]. b. [Pre-Core S Who] [Core did he give the book to]?

However, since the potential focus domain extends to the whole clause, in English, and focus is primarily marked prosodically, it can also fall in marked positions, as is shown in (24). (24)

a. [Core He gave the book to Pat]. b. [Core He gave the book to Pat]. c. [Core He gave the book to Pat].

When narrow focus falls in a marked position, it is contrastive or corrective, in that it identifies a new selection out of a finite set of alternatives. Thus, (24a–c) involve contrastive narrow focus in the respective contexts listed in (25a–c). (25)

a. She gave the book to Pat. b. He sold the book to Pat. c. He gave the pen to Pat.

It should be noted, however, that narrow focus can be contrastive even when it occurs in an unmarked position. Contrastiveness as a relational information structure property that selects from alternatives is orthogonal to the distinction between presupposition and assertion, in that the alternatives can be stated or predicted as part of the presupposition. Thus, not only can foci be contrastive, but topics can, too. (26)

a. «I heard Mary and Jane came yesterday.» b. «Jane came.»

To the extent that its status as a matter of interest or concern in the utterance has been introduced in (26a), the contrastive information unit in (26b) (Jane) is presupposed. The new information introduced in the assertion in (26b) is the relation between this information unit and the proposition introduced in the previous utterance, as well as the notion that the same relation does not hold true of the other referent introduced previously (Mary). Interestingly, some languages allow the same marking of contrastive topics and foci. This is the case with Japanese, where contrastive topics

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

471

472

DELIA BENTLEY

and foci can both be marked with -wa, although by default contrastive foci are marked with -ga, and contrastive topics with -wa (see Shimojo 2009, 2010, 2011 for extensive discussion). To capture the topic–focus ambivalence in topic marking in Japanese, Shimojo (2011) subsumes Erteschik-Shir’s (1997, 2007) notion of subordinate f(ocus) structure within the RRG framework. A subordinate f-structure is a complex f-structure involving embedding. Contrastive information units can be topics or foci because there is a contextually available set, which is topical, whereas the contrastive information units that are selected from this set are foci. A structure with contrastive -wa is thus represented as follows by Shimojo (2011: 275). (27)

[{xfoc, y}top]-watop [predicate]foc

In terms of sentence structure, Shimojo (2010, 2011) proposes that the -wamarked constituent figures outside the domain of focus in syntax, specifically in the pre-detached position.12

11.5

The Interplay of Focus Structure and Syntax

Since typological variation is a key concern of RRG, attention has been paid to the cross-linguistic variation in the interplay of prosody and syntax in the expression of information structure (Van Valin 1999). In some languages, the three types of focus structure discussed in Section 11.4 are primarily encoded in syntax by means of word order, whereas in others they are mainly encoded prosodically. In this second type of language, focus structure tends to be syntactically flexible, insofar as the potential focus domain is not restricted to a particular portion of the clause. Van Valin (1999) discusses this kind of typological variation in terms of the relative rigidity or flexibility of focus structure and syntax across languages. French, Toba Batak (a Western Austronesian language spoken in Indonesia) and other languages have rigid syntax and rigid focus structure. This means that while the potential focus domain (see Section 11.4) does not extend to the whole clause, focus cannot be freely marked by the syntactic position of the focal unit either, since the syntactic position of arguments is fixed to various degrees. In French, the actor of an active clause occurs in the core-internal immediately pre-nuclear position. In the spoken registers, it is usually omitted, or detached outside the clause, while a coreferent clitic pronoun is spelled out preverbally. We shall not dwell here on the status of this pronoun in morphosyntax, namely whether it must be assumed to occur in a pre-nuclear position or, rather, under the nuclear node (see Belloro 2004, 2015 for the position of core clitic arguments in RRG). The point that is relevant in the current discussion is that the actor of the active voice cannot normally occur in postnuclear position in French. Using terminology adopted in other frameworks, French disallows free subject inversion (though see Lahousse 2011

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

and Lahousse and Lamiroy 2012 for further discussion). Focus, in turn, is avoided in the core-internal immediately pre-nuclear position and is marked post-nuclearly by default.13 The potential focus domain in the French clause is thus as shown in (28). (28)

French [PrDP Marie] [ Clause [ Core elle [ Nuc aime] les bonbons]] ] Mary she love.3 SG the sweets

Speech Act Projection ‘Mary loves sweets.’

Wh-words, which are focal, can occur preverbally, though in the pre-core slot, rather than in a core-internal position. However, they can also occur in situ, thus taking a post-nuclear position (cf. 29a–b). Otherwise, they are clefted (cf. 30a). (29)

French a. Vous avez vu qui? you.2pl have.2pl seen who ‘Who have you seen?’ `? b. Vous allez ou you.2pl go.2pl where ‘Where are you going?’

Given that the extent of the potential focus domain rules out focal subjects in situ, while the syntax of French normally bans post-nuclear subjects, the violation of these pragmatic and syntactic restrictions is avoided by clefting (cf. 30b) or by using subjectless VS constructions (cf. 31). (30)

(31)

French a. C’est qui qui it be.3sg who who ‘Who has done this?’ b. C’est Marie qui it be.3sg Mary who ‘It’s Mary who did this.’

a fait ça? have.3sg done this a fait ça. have.3sg done this

French Il s’ est produit des probl`emes. it refl be.3sg produced some problems ‘There occurred some problems.’

In Toba Batak, which is a language with rigid VOS order and immediately post-nuclear focus, the verb must carry a voice prefix when the undergoer is topical and the actor is not (Van Valin 1999). This suffix indicates that the undergoer has privileged syntactic argument status: the syntactic

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

473

474

DELIA BENTLEY

constraints on the position of O and S are satisfied, as is the constraint on the position of focus. The postverbal actor receives a non-specific reading in the postverbal position. (32)

Toba Batak (Van Valin 1999) Di-jaha guru buku i. pass-read teacher book the ‘The book is read by a teacher.’/‘A teacher is reading the book.’

The case of flexible syntax, which adapts to focus structure, is illustrated by many Indo-European languages, for example Italian, and by Bantu, for example Sesotho and Setswana (Demuth 1989, 1990). Whereas in both groups of languages the potential focus domain is restricted to the nucleus and the following constituents in the clause, as is the case with French (cf. 28), in these languages the syntax allows VS order to mark a subject as focal. (33)

Italian [ Core [ Nuc

Sono emersi] tanti problemi ]. be.3 PL emerged many problems

Speech Act Projection ‘There emerged many problems.’

An important difference between Italian and Sesotho or Setswana is that, in the two Bantu languages, there is an absolute constraint against prenuclear foci, including wh-words. Thus, while Italian does have a prenuclear focal domain, which is the pre-core slot (cf. 34), the two Bantu languages do not. (34)

Italian [PrCS Chi] [Core [Nuc ha fatto] questo]? who have.3sg done this ‘Who did this?’

In Setswana, passivization or clefting of an actor is necessary to avoid a breach of grammaticality (i.e. a subject wh-word in a pre-nuclear position). (35)

Setswana (Van Valin 1999) *Mang o-pheh-ile lijo? who sbj-cook-prf food ‘Who cooked the food?’

(36)

Setswana (Van Valin 1999) Lijo li-pheh-li-o-e ke mang. food sbj-cook-prf-pass-mood by who ‘Who cooked the food?’ (Lit. ‘The food was cooked by who?’)

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

(37)

Setswana (Van Valin 1999) Ke mang ea o-f-ile-ng ntja? cop who rel obj-give-prf-rel dog ‘Who gave you the dog?’ (Lit. ‘It is who that gave you the dog?’)

To return to Italian, not only does the pre-core slot host wh-words, but it also admits contrastive narrow foci. This type of focus fronting is shown in (38b). (38)

Italian a. «Cosa ha comprato Giorgio? what have.3sg bought George ‘What did George buy? A car?’ b. «La bicicletta, Giorgio ha the bike George have.3sg ‘A bike George bought, not a car.’14

La the

macchina?» car

comprato, non la macchina.» bought neg the car

The fact that, in (38b), the focal argument co-occurs with the actor in prenuclear position indicates that the position of the contrastive focus is coreexternal, specifically, the pre-core slot (see Bentley 2008 for a fuller discussion of this point). Observe that the preverbal actor must be assumed to occur coreinternally, specifically in the core-initial position. Indeed, RRG rules out the assumption that a topic can follow the focus in the pre-core positions, that is, those positions which, in other frameworks, are referred to as the left periphery of the clause (Van Valin 2008: xx in disagreement with Rizzi 1997). Languages with flexible syntax and flexible focus structure extend the potential focus domain to the whole clause. Van Valin (1999) provides relevant examples from Slavic, which are reported here. (39)

Russian (Van Valin 1999) ˇ a. «Cto sluˇ ciloc’?» what happened ‘What happened?’ b. «Mašina slomalac’.» car broke.down c. «Slomalac’ mašina.» broke.down car ‘[My] car broke down’

The fact that both SV and VS order are allowed in sentence focus sets Russian apart from the languages discussed previously, suggesting that preverbal focus is not constrained in Russian in the same way as it is in Italian. It should be noted, however, that the topic–focus order is adhered to in declarative clauses with focus on the postverbal argument, while wh-words occur in the pre-core slot (Van Valin 1999, based on Comrie 1979, 1984). (40)

Russian (Van Valin 1999) a. «Kogo zašˇ cišˇ cajet Maksim-Ø?» who.acc defends Maxim-nom ‘Who(m) does Maxim defend?’

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

475

476

DELIA BENTLEY

b. «Maksim-Ø zašˇ cišˇ cajet Maxim-nom defends ‘Maxim defends Viktor.’

Viktor-a.» Viktor-acc

Therefore, it can be hypothesized that, in Russian, core-internal pre-nuclear focus is confined to sentence focus. Further evidence that languages where the potential focus domain extends to the whole clause may nonetheless exhibit restrictions on the type of focus associated with particular sentence forms is provided by Sicilian, which is cognate with Italian. Like Italian, Sicilian places contrastive foci and wh-words in the pre-core slot (Bentley 2008, 2010). (41)

Sicilian (Bentley 2008) a. «Chi ci accattau Pippinu a so niputi? A machina?» what to.him buy.pst.3sg Joseph to his nephew the car ‘What did Joseph buy for this nephew? A car?’ b. «A bicicretta, Pippinu ci accattau.» the bike Joseph to.him buy.pst.3sg ‘Joseph bought him a bike.’

Unlike Italian (cf. 42a), Sicilian also admits non-contrastive pre-nuclear focus. In this case, the focal unit must be assumed to figure core internally, as suggested by its scarce compatibility with another core-internal prenuclear constituent (cf. 43b–c). (42)

Italian a. «Che ha comprato Giuseppe a suo nipote?» what has buy.ptcp Joseph to his nephew ‘What did Joseph buy for this nephew?’ b. «*La bicicletta ha comprato.» the bike has buy.pst.3sg ‘He bought him a bike.’

(43)

Sicilian (Bentley 2008) a. «Chi ci accattau Pippinu a so niputi?» what to.him buy.pst.3sg Joseph to his nephew ‘What did Joseph buy for this nephew?’ b. «A bicicretta (*?Pippinu) ci accattau.» the bike Joseph to.him buy.pst.3sg ‘Joseph bought him a bike.’ c. «A bicicretta (*?a so niputi) ci accattau.» the bike to his nephew dcl buy.pst.3sg ‘He bought a bike for his nephew.’

Data like (42)–(43) differentiate Sicilian from Italian, indicating that there is no ban on core-internal pre-nuclear focus in Sicilian, and that the latter language is more flexible than the former in terms of its focus structure. Sicilian pre-nuclear focus, however, is not entirely free, but rather has affective and emphatic connotations, which are absent from post-nuclear focus (Sornicola 1983; Leone 1995; Cruschina 2006). In particular, preverbal

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

focus in (43b–c) suggests that the response being provided should be expected by the hearer. In light of its special relevance effects, Sicilian core-initial focus is not readily found in sentence focus.15 To sum up, while the extent of the potential focus domain allows languages like Russian and Sicilian to express focus in any position in the clause, their syntactic flexibility allows them to associate specific word orders with specific types of focus structure or even to associate specific syntactic positions with specific types of focus (as is the case with contrastive focus in the pre-core slot and non-contrastive focus in the core-initial position in Sicilian). Therefore, in these languages, the flexibility of focus structure amounts to the extent of the potential focus domain alone and not to the lack of conventional associations between focus meanings and sentence forms. By contrast with Russian and Sicilian, English has flexible focus structure but rigid syntax. Thus, while its syntax normally adheres to SVO order, the prosody is tasked with the expression of focus structure. In (44) to (46), we indicate the domain of focus in predicate focus, sentence focus and narrow argument focus. (44)

a. b.

«What happened to your car?» «(My car)/It BROKE DOWN.»

Speech Act Projection (predicate focus) (45)

a. b.

«What happened?» «MY CAR BROKE DOWN.»

Speech Act Projection (sentence focus) (46)

a. b.

«I hear your bike broke down.» «MY CAR broke down.»

Speech Act Projection (narrow focus)

Narrow focus can fall in any position in the English clause (cf. 46b, 47a–d) and is marked prosodically. (47)

a. b. c. d.

[Core He gave the book to Pat]. [Core He gave the book to Pat]. [Core He gave the book to Pat]. [Core He gave the book to Pat].

Adopting the Autosegmental-Metrical/Tones and Break Indices approach to intonation (Pierrehumbert 1980), O’Connor (2008) has proposed a

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

477

478

DELIA BENTLEY

prosodic projection to represent the prosodic encoding of focus structure in RRG. An important aspect of this proposal is that different languages are said to avail themselves of different prosodic templates, comparable to the syntactic templates which constitute the syntactic inventories of different languages. Although O’Connor (2008) begins to establish the principles which govern the interaction of the prosodic projection with the speech act projection in English, to our knowledge his insights have not yet been extended to the study of the prosodic encoding of information structure in other languages. Significantly, even languages with rigid syntax can deviate from their default word order. For example, English allows fronting, in structures like This book I did not read, and VS order, in there sentences and locative inversion (There arrived three buses at the station) (see Section 11.4). In these cases, information structure is not spelled out by prosody alone, but rather by the choice of a particular constructional schema (Van Valin 2005: 131–135) in semantics–syntax linking. It is to the role of discourse in the linking that we now turn.

11.6

The Role of Discourse in the Linking

In RRG it is assumed that discourse operates at all stages of the semantics– syntax linking, starting from the choice of a predicate for the clause (see Chapter 12 for fuller discussion). Consider lexical alternations like the one between actor and undergoer experiencer verbs, for example fear and frighten. The availability of a topical stimulus as the subject matter at issue may trigger the choice of frighten, which has a stimulus as its actor and subject in the active voice (actor and subject align with topic crosslinguistically). (48)

a. «Did you hear about the funding cuts?» b. «Yes, it frightened everyone at the meeting.»

Conversely, the availability of an experiencer in the presupposition may trigger the choice of fear, since this verb takes an experiencer as its actor and subject in the active voice. (49)

a. «Do you like this resolution?» b. «Actually, I fear its possible consequences.»

Once the logical structure of the predicate is retrieved from the lexicon, its argument positions are filled with the core arguments, which are coded in accordance with their pragmatic states and their roles as part of the presupposition or the assertion (see Figures 11.2 and 11.3). Thus, in English, the activated topical subject of a predicate-focus structure would most likely be an unstressed pronoun, while in Italian and in the languages that allow phonologically null subjects it would most likely not occur at all.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

Phonologically null arguments can figure in logical structure with their person and number features, when these features are spelled out in the verb inflection. (50)

Italian Parl-o. speak-1sg ‘I speak.’ do′ (1sg act , [speak′ (1sg act )])

Alternatively, in the absence of an overt RP or of verb inflection, phonologically null arguments were previously assumed to be expressed by pro, which is a silent variable that receives its value from discourse in the linking (Van Valin 2005: 171–174). However, the current assumption in RRG is that there is no silent pronoun in logical structure, consistently with the RRG ban on null pronouns in syntax, and that the discourse antecedents of the unfilled positions in logical structure can satisfy the valence requirements of the verb. We return to this point at the end of the section. The next step in the linking is macrorole assignment, which is not affected by the discourse status of the arguments. Contrastingly, the latter can play a key role in their morphosyntactic coding with case. We provide here an example from Kaluli, an ergative language spoken in Papua New Guinea (Van Valin 2015). In Kaluli, the predominant word order is OV, and, as is the case with many languages, there is a strong association of actor with topic. Ergative case is normally only marked on the topical actor when the undergoer is animate. Otherwise, both actor and undergoer take absolutive case (51). Pronominal arguments are marked by syntactic position rather than case (52).16 (51)

Kaluli (Van Valin 2015) a. Abi-y` o siabulu-w` o m`enigab. Abi-abs sweet.potato-abs eat.3.tns ‘Abi is about to eat a sweet potato.’ b. Abi-y`e Suela-y` o sandab. Abi-erg Suela-abs hit.3.tns ‘Abi hits Suela.’

(52)

Kaluli (Van Valin 2015) a. E ne sandab. 3sg 1sg hit.3.tns ‘He/she hits me.’ b. Ne e sond` ol. 1sg 3sg hit.1.tns ‘I hit him/her.’

However, when the actor is focal and the undergoer is not, the former is marked by contrastive ergative morphology on pronominal actors and by ergative case on non-pronominal ones, regardless of the animacy of the undergoer.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

479

480

DELIA BENTLEY

(53)

Kaluli (Van Valin 2015) a. Nodo-w` o niba di` ol. one.side- abs 1sg.contr take.1.tns ‘I (not you) take one side.’ b. Nodo-w` o Suela-w`e diab. one.side- abs Suela-erg take.3.tns ‘SUELA takes one side.’

This complex split-ergative system is captured in RRG in terms of the effect of discourse on case marking in the linking. Voice alternations and the choice of privileged syntactic argument are also affected by discourse. Thus, although accusative alignment normally involves the selection of the actor as the subject, this default choice may be overridden because of the role of actor and undergoer in the presupposition and assertion. In particular, the passive, which maps the undergoer to subject, may be chosen when this macrorole is topical and the actor is not. An example is provided in (54), where the salience of an argument as the matter at issue in the first utterance results in the marked subject assignment in the second one. (54)

a. «The students’ proposal has many advantages.» b. «Yes, I know. It will be discussed at tomorrow’s staff meeting.»

We pointed out above that passivization can also be a strategy to avoid preverbal foci, as is the case with the Setswana data in (36). The next step in the linking is the choice of a syntactic template from the syntactic inventory. This is affected by discourse in several ways. To begin with, passivization results in the failure of overt realization of the actor or its occurrence outside the syntactic core. In such cases, the syntactic template chosen for the clause does not have positions for all the arguments within the core, but can host the actor in a periphery of the core (cf. 55b–c vs. 55a). (55)

a. [Core The staff [Nuc will discuss] the proposal] b. [Core The proposal [Nuc will be discussed]] c. [Core The proposal [Nuc will be discussed]] [Per by the staff].

Secondly, we pointed out above that referential topics, afterthoughts, whwords and fronted foci occur in extra-clause or extra-core positions, in which case the template selected for the sentence must include these positions. (56)

French (cf. 27) a. [PrDP Marie] [Clause [Core elle [Nuc aime] les bonbons]]] Mary she loves the sweets ‘Mary, she loves sweets.’ b. Italian (cf. 37a) [[PrCS La bicicletta] [Core Giorgio [Nuc ha comprato,]]] [PoDP non la macchina.] the bike George has bought neg the car ‘A bike George bought, not a car.’

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

Finally, in languages with flexible syntax, narrow focus and sentence focus may involve marked word order patterns, for example VS, in languages with default SV order (Section 11.5). Again, this is reflected in the choice of a syntactic template for the clause, as is the case with the Russian example of sentence focus previously given in (39c) and repeated below for convenience. (57)

Russian [Core [Nuc Slomalac’] mašina]. broke.down car ‘[My] car broke down’

Prosody also plays a role in the linking, spelling out focus and providing clues on the domains of presupposition and assertion in the processing of the sentence. As was mentioned in Section 11.5, it has been proposed that the role of prosody in information structure ought to be represented in a prosodic projection (O’Connor 2008). Although in this section we have discussed the role of discourse in the linking from semantics to syntax, which pertains to language production, we should note that discourse also plays a key role in the linking from syntax to semantics, which pertains to language comprehension. By way of conclusion of the section, we will briefly touch upon this direction in the linking with reference to zero anaphora. Zero anaphora is puzzling because it would seem to challenge the Completeness Constraint (Van Valin and LaPolla 1997: 325), which states that all the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in that sentence. In Mandarin Chinese and other languages, a presupposed argument can be entirely silent (Van Valin 2005: 174). Differently from the silent privileged syntactic argument of many pro-drop languages (see the Italian example in (50)), the silent argument under discussion here has no pronominal or affixal expression on the verb. Another puzzling case of zero anaphora is that of the Japanese verb-less numeral quantifier construction, where an argument bears accusative case, but there is no overt case assigner in clause structure. In this case, syntax lacks a verb rather than an argument. RRG captures these patterns of zero anaphora by linking the semantic representation directly with discourse, drawing upon Kamp and Reyle’s (1993) Discourse Representation Theory. Proper discussion of the application of this theory to RRG is provided in Chapter 12, where Van Valin and Latrouite propose an Extended Completeness Constraint, which states that completeness can be satisfied by discourse representation structures in combination with arguments that are expressed overtly in syntax. Here we note that discourse representation keeps track of presupposed and asserted information, which can be retrieved directly in the semantics of the clause, without being expressed in syntax. Thus, the silent arguments of Mandarin Chinese can be filled into the semantic representation in syntax–semantics linking because they can be drawn from discourse representation, as can the

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

481

482

DELIA BENTLEY

predicator in the case of the Japanese verb-less numeral quantifier construction (for a detailed RRG treatment of this construction, we refer to Shimojo 2008). Importantly, this kind of direct linking is assumed to be bidirectional and to apply to semantics to syntax linking as well. Thus, the argument positions are linked directly with the discourse representation for zero anaphora in semantics–syntax linking, and, in this kind of linking, the predicator is linked directly to discourse in the Japanese verbless construction mentioned above.

11.7

Conclusion

Some of the most original and significant results of research in RRG suggest that an important way in which languages differ from each other is in terms of the role of discourse in the linking of semantics with syntax and syntax with semantics. RRG offers an approach to information structure which is flexible enough to capture this variation, while also being sufficiently constrained to make important generalizations regarding the expression of pragmatic states and pragmatic relations, and their interface with prosody, morphology and sentence structure across languages. In this chapter we have introduced the constructs adopted in the RRG approach to information structure and we have discussed the role that this framework gives to information structure in the architecture of grammar.

References Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third Person Clitic Clusters in Spanish. MA thesis, University at Buffalo (SUNY). Belloro, Valeria. 2015. To the Right of the Verb: An Investigation of Clitic Doubling and Right Dislocation in Three Spanish Dialects. Newcastle upon Tyne: Cambridge Scholars Publishing. Bentley, Delia. 2008. The interplay of focus structure and syntax. Evidence from two sister languages. In Van Valin, Jr. (ed.), 263–284. Bentley, Delia. 2010. Focus fronting in the layered structure of the clause. In Nakamura (ed.), 4–28. https://rrg.caset.buffalo.edu/rrg/ProceedingsofRRG2009_ 02.pdf. Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence from Italo-Romance. Linguistics 56: 1245–1301. Bentley, Delia, Francesco Maria Ciconte, Silvio Cruschina and Michael Ramsammy. 2016. Micro-variation in information structure: Existential constructions in Italo-Romance. In M. M. Jocelyne Fernandez-Vest and Robert D. Van Valin, Jr. (eds.), Information Structure and Spoken Language in a CrossLinguistic Perspective, 95–120. Amsterdam: John Benjamins.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

Chafe, Wallace. 1976. Givenness, contrastiveness, definiteness, subjects, topics, and point of view. In Charles Li (ed.), Subject and Topic, 27–55. New York: Academic Press. Chafe, Wallace. 1987. Cognitive constraints on information flow. In Russell Tomlin (ed.), Coherence and Grounding in Discourse, 21–52. Amsterdam: John Benjamins. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Comrie, Bernard. 1979. Russian. In Timothy Shopen (ed.), Languages and Their Status, 91–150. Cambridge, MA: Winthrop. Comrie, Bernard. 1984. Russian. In W. S. Chisolm, Jr. (ed.), Interrogativity, 7–46. Amsterdam: John Benjamins. Cruschina, Silvio. 2006. Informational focus in Sicilian and the left periphery. In Mara Frascarelli (ed.), Phases of Interpretation (Studies in Generative Grammar 91), 363–385. Berlin: Mouton de Gruyter. Cruschina, Silvio. 2012. Discourse-Related Features and Functional Projections. New York: Oxford University Press. Cruschina, Silvio. 2015. Focus Structure. In Delia Bentley, Francesco Maria Ciconte and Silvio Cruschina (eds.), Existentials and Locatives in Romance Dialects of Italy, 43–98. Oxford: Oxford University Press. De Cia, Simone. 2019. The Syntactic-Pragmatic Interface in North-Eastern Italian Dialects: Consequences for the Geometry of the Left Periphery. PhD dissertation, University of Manchester. Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho passive. Language 65: 56–80. Demuth, Katherine. 1990. Subject, topic and the Sesotho passive. Journal of Child Language 17: 67–84. Diedrichsen, Elke. 2008. Where is the precore slot? Mapping the layered structure of the clause and German sentence typology. In Van Valin, Jr. (ed.), 203–224. Erteschik-Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax-Discourse Interface. Oxford: Oxford University Press. Fillmore, Charles. 1982. Frame semantics. In The Linguistic Society of Korea (eds.), Linguistics in the Morning Calm, 111–137. Seoul: Hanshin. Frascarelli, Mara. 2017. Dislocations and framings. In Andreas Dufter and Elisabeth Stark (eds.), Manual of Romance Morphosyntax and Syntax (Manuals of Romance Linguistics 17), 472–501. Berlin: Mouton de Gruyter. Frascarelli, Mara and Roland Hinterhölzl. 2007. Types of topics in German and Italian. In Susanne Winkler and Kerstin Schwabe (eds.), On Information Structure, Meaning and Form, 87–116. Amsterdam: John Benjamins. Gundel, Jeanette K. 1988. Universals of topic-comment structure. In Michael Hammond, Edith Moravcsik and Jessica Wirth (eds.), Studies in Syntactic Typology, 209–239. Amsterdam: John Benjamins.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

483

484

DELIA BENTLEY

Gundel, Jeanette K., Nancy Hedberg and Ron Zacharski. 1993. Cognitive status and the form of referring expressions in discourse. Language 69(2): 274–307. Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA: Kluwer. Krifka, Manfred. 2007. Basic notions of information structure. In Caroline Féry, Gisbert Fanselow and Manfred Krifka (eds.), The Notions of Information Structure, 13–55. Potsdam: Universitätsverlag. Kuno, Susumo. 1972. Functional sentence perspective: A case study from Japanese and English. Linguistic Inquiry 3: 269–320. Lahousse, Karen. 2011. Quand passent les cicognes: Le sujet nominal postverbal en français moderne. Paris: Presses Universitaires de Vincennes. Lahousse, Karen and Béatrice Lamiroy. 2012. Word order in French, Spanish and Italian: A grammaticalization account. Folia Linguistica 46: 1–29. Lambrecht, Knud. 1986. Topic, Focus and the Grammar of Spoken French. PhD dissertation, University of California, Berkeley. Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representation of Discourse Referents. Cambridge: Cambridge University Press. Lambrecht, Knud. 2000. When subjects behave like objects: An analysis of the merging of S and O in sentence focus constructions across languages. Studies in Language 24: 611–682. LaPolla, Randy J. 1990. Grammatical Relations in Chinese: Synchronic and Diachronic Considerations. PhD dissertation, University of California, Berkeley. LaPolla, Randy J. 1993. Arguments against ‘subject’ and ‘direct object’ as viable concepts in Chinese. Bulletin of the Institute of History and Philology 63 (4): 759–813. LaPolla, Randy J. 1995. Pragmatic relations and word order in Chinese. In Pamela Downing and Michael Noonan (eds.), Word Order in Discourse, 297–329. Amsterdam: John Benjamins. Leone, Alfonso. 1995. Profilo di sintassi siciliana. Palermo: Centro di Studi Filologici e Linguistici Siciliani. Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax– Lexical Semantics Interface. Cambridge, MA: MIT Press. Li, Charles N. and Sandra Thompson. 1976. Subject and topic: A new typology of language. In Charles Li (ed.), Subject and Topic, 459–489. New York: Academic Press. Nakamura, Wataru (ed.). 2010. Proceedings of the 10th International Conference on Role and Reference Grammar (RRG 2009). https://rrg.caset.buffalo.edu/rrg/Pro ceedingsofRRG2009_02.pdf. O’Connor, Rob. 2008. A prosodic projection for Role and Reference Grammar. In Van Valin, Jr. (ed.), 227–244.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

Pierrehumbert, J. 1980. The Phonology and Phonetics of English Intonation. PhD Dissertation, MIT (published in 1988 by Indiana University Linguistics Club). Prince, Ellen F. 1981. Toward a taxonomy of given-new information. In Peter Cole (ed.), Radical Pragmatics, 223–255. New York: Academic Press. Prince, Ellen. 1992. The ZPG letter: Subjects, definiteness, and informationstatus. In William C. Mann and Sandra A. Thompson (eds.), Discourse Description: Diverse Linguistic Analyses of a Fund-Raising Text, 295–324. Amsterdam: John Benjamins. Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27: 53–94. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer. Rizzi, Luigi. 2006. On the form of chains: Criterial positions and ECP effects. In Lisa Lai-Shen Cheng and Norbert Corver (eds.), Wh-Movement. Moving On, 97–133. Cambridge, MA: MIT Press. Sasse, Hans-Ürgen. 1987. The thetic-categorical distinction revisited. Linguistics 25: 511–580. Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: Wa and ga, and Word Order Flexibility. PhD dissertation, University at Buffalo (SUNY). Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less numeral quantifier construction in Japanese. In Van Valin, Jr. (ed.), 285–304. Shimojo, Mitsuaki. 2009. Focus structure and beyond: Discourse-pragmatics in Role and Reference Grammar. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 111–141. Mexico: Universidad Nacional Autónoma de México. Shimojo, Mitsuaki. 2010. The left periphery and focus structure in Japanese. In Wataru Nakamura (ed.), Proceedings of the 10th International Conference on Role and Reference Grammar (RRG 2009), 315–235. www.acsu.buffalo.edu/ ~rrgpage/rrg.html. Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japanese. In Wataru Nakamura (ed.), New Perspectives in Role and Reference Grammar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing. Sornicola, Rosanna. 1983. Relazioni d’ordine e segmentazione della frase in italiano: Per una teoria della sintassi affettiva. In Paola Beninc` a, Michele Cortelazzo, Aldo Prosdocimi, Laura Vanelli and Alberto Zamboni (eds.), Scritti linguistici in onore di Giovan Battista Pellegrini, 561–577. Pisa: Pacini. Stempel, Wolf-Dieter. 1981. L’amour elle appelle ça – L’amour tu connais pas. In Christian Rohrer (ed.), Logos Semantikos. Studia linguistica in honorem Eugenio Coseriu, Vol. 4: Grammatik, 351–367. Berlin: Mouton De Gruyter. Tortora, Christina. 1997. The Syntax and Semantics of the Weak Locative. PhD dissertation, University of Delaware.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

485

486

DELIA BENTLEY

Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure and syntax. In Ekatarina Raxilina and Yakov G. Testelec (eds.), Typology and Linguistic Theory: From Description to Explanation, 511–524. Moscow: Languages of Russian Culture. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2015. An overview of information structure in three Amazonian Languages. In M. M. Jocelyne Fernandez-Vest and Robert D. Van Valin, Jr. (eds.), Information Structuring of Spoken Language from a Cross-linguistic Perspective, 77–92. Berlin: De Gruyter Mouton. Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Notes 1 The formal representation of discourse in RRG will not be touched upon, as this will be dealt with in Chapter 12 by Latrouite and Van Valin. The abbreviations used in the glosses of the examples are drawn from the Leipzig Glossing Rules (www.eva.mpg.de/lingua/resources/glossing-rules .php). See note 16 for an exception. 2 Lambrecht (1994: 49) discusses the distinction between ‘the pragmatic states of the denotata of individual sentence constituents in the minds of the speech participants’, and ‘the pragmatic relations established between these denotata and the propositions in which they play the roles of predicates and arguments’. 3 For brevity, we do not make explicit reference to the written language, although it should be noted that a referent that is introduced for the first time in a written text is new from the perspective of the reader. 4 As they are used here, the terms unidentifiable and identifiable correspond with Prince’s (1992) concepts of hearer-new and hearer-old, respectively. 5 The terms accessible, active and inactive are from Chafe (1987). 6 In Lambrecht’s (1994: 52) words, the presupposition is ‘the set of propositions lexico-grammatically evoked in an utterance which the speaker assumes the hearer already knows or is ready to take for granted at the time the sentence is uttered’. 7 Some scholars compare frame setters to contrastive topics, insofar as frame setters specify that the eventuality being described is only valid within a particular domain or frame and may not be valid for other domains (Krifka 2007). Assuming that this comparison stands typological scrutiny, from the RRG perspective it does not necessarily follow that

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

The RRG Approach to Information Structure

8

9

10

11

12

13 14

15 16

frame setters and contrastive topics should take the same syntactic position, although some evidence in support of this assumption is reported in Frascarelli (2017: 486). This notion of focus draws upon Lambrecht’s (1994: 213) claim that focus is ‘the semantic component of a pragmatically structured proposition whereby the assertion differs from the presupposition’. The former names Left- and Right-Detached Position reflected a bias in favour of left-to-right writing systems, differently from the new terminology, which is based on order of occurrence in speech. In the Gallo-Italian dialects of Northern Italy, the topic of sentence-focus constructions with verbs of specific lexical classes can be expressed overtly by a locative clitic in subject clitic position (Tortora 1997, and, for an RRG account, Bentley 2018). The question arises of whether this locative clitic should take a syntactic position. This question, however, goes beyond the scope of the present chapter. It should also be noted that the VP does not have any status in the layered structure of the clause, because not all languages provide evidence for it. Therefore, focus on the core minus the subject-topic cannot be defined as focus on the VP constituent in RRG. Further evidence that contrastive topics and foci can be marked in the same way in individual languages has been found in Romance dialects spoken in the north-east of Italy (De Cia 2019). See Stempel (1981) for some deviations from this strong tendency of French. The definite marker in the RP can be said to be generic and it is not a marker of specificity or uniqueness. Thus, this example could be translated as ‘now George is a bike owner, not a car owner’. One could also find an indefinite RP in this example (Una bicicletta Giorgio ha comprato, non una macchina ‘A car George has bought, not a car’), with the same focus structure as discussed above. Core-initial focus occurs in out-of-the blue ascriptive structures, where the focus is on a non-verbal predicate (Sic. Stanca sugnu, lit. tired I-am). The abbreviations contr and tns in the glosses of examples (51)–(53) stand for ‘contrastive’ and ‘tense’, respectively.

https://doi.org/10.1017/9781316418086.014 Published online by Cambridge University Press

487

12 Information Structure and Argument Linking Anja Latrouite and Robert D. Van Valin, Jr.

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: A AFD AV COMM DCA DRS

CLM GCG IGC

actor actual focus domain actor voice comment direct core argument discourse representation structure Discourse Representation Theory clause-linkage marker general common ground immediate common ground

IF IS IU LS

illocutionary force information structure information unit logical structure

RLS RP U UV

DRT

12.1

LS NUC OCA PFD PoCS PoDP

logical structure nucleus oblique core argument potential focus domain post-core slot post-detached position

PrCS

pre-core slot

PrDP PRED PSA

pre-detached position predicator privileged syntactic argument realis reference phrase undergoer undergoer voice

Introduction

The RRG approach to information structure (IS) is laid out by Bentley (Chapter 11), and the linking between syntax and semantics is explicated by Van Valin (Chapter 1), Watters (Chapter 6), Guerrero (Chapter 14) and París (Chapter 15). These discussions did not emphasize the role of IS in linking, and it is to this topic that we turn in this chapter. We begin by

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

489

presenting the representation of IS in the layered structure of the clause (LSC; Section 12.2), and then we will show how context can be represented using a version of Discourse Representation Theory (Kamp and Reyle 1993; von Heusinger 1999) and the different focus types derived from these representations (Section 12.3). The next section illustrates the importance of IS for the analysis of grammatical phenomena (Section 12.4), and this is followed by an analysis of conjunction reduction, which presented problems for an RRG analysis in purely syntactic terms (Section 12.5). Finally, the interaction of IS and the linking algorithm will be explored (Section 12.6).

12.2

The Representation of Information Structure

Bentley introduced the distinction between the potential and actual focus domains and the notion of information units, which are very important parts of the RRG account of IS, along with representation of them in the LSC. The LSC discussed so far has two projections, the constituent projection and the operator projection. The theory posits two more, one for IS and one for prosody (O’Connor 2008). They are treated as distinct projections, because not all languages use prosody to signal IS (Van Valin 2016), even though many, if not most, do. The essential components of the focus structure projection as presented in Van Valin (2005) are: (i) basic information units (IU), (ii) the potential focus domain (PFD), and (iii) the actual focus domain (AFD). These are illustrated in Figure 12.1. The opposition between the PFD and the AFD is central to the RRG analysis of IS. The PFD is a function of the grammar. In English the PFD in simple clauses encompasses the entire clause, as can be readily seen in

SENTENCE CLAUSE CORE RP

NUC

RP

PP

PRED Actual focus domain

Potential focus domain

V Chris

presented

a child

with some flowers

IU

IU

IU

IU

SPEECH ACT

Figure 12.1 Components of the focus structure projection (Van Valin 2005: 77)

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Basic information units

490

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Figure 12.2 (Unmarked) Narrow focus in English

Figures 12.1 and 12.2. In Italian, on the other hand, it excludes the prenuclear core-initial position, as Bentley discusses in some detail (see also Bentley 2008). The AFD is a function of the context in which the sentence occurs; it is the focus of the utterance. It corresponds to the ‘focus domain’ of Lambrecht. There is a very important IS distinction which is not represented here, namely topic vs. comment. According to both Gundel (1988) and Lambrecht (1994), the topic referent is part of the pragmatic presupposition, which is part of the immediate common ground (ICG) (Berio et al. 2017; Krifka and Musan 2012). The ICG consists of the discourse context, which includes the pragmatic presuppositions, plus the immediate social and physical environment in which the speech event occurs. World knowledge is contained in the general common ground (GCG). In Figure 12.1, there is only a single IU outside of the AFD, namely, the RP Chris, which is interpreted as the topic expression which refers to the topic referent. This is not always the case, however, as it is possible for there to be more than one IU outside the AFD, as illustrated in Figure 12.2. The utterance represented in Figure 12.2 could be the answer to the question What did John buy?, which establishes the referent of John and the action of buying as part of the pragmatic presupposition in the ICG, and the AFD is a new car. The part of the PFD which does not include the AFD is termed the background.1 The individual referred to by the RP John is presumably the topic, but this is not indicated in the representation. Balogh (2021) proposes to augment the focus structure representation with notations indicating topic and comment. Following her proposal, the revised representation would be as in Figure 12.3. Since it now codes both focus– background and topic–comment, she labels it the ‘information structure projection’.2

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Figure 12.3 Revised IS-projection

Figure 12.4 The constituent, operator and information structure projections of a sentence

In Figure 12.3 he is indicated as the topic expression, and a new car is presented as the (narrow) focus. The verb bought is part of both the comment and the background. It is possible to represent all three projections – constituent, operator and information structure – in a single diagram, as shown in Figure 12.4. Each projection represents a type of structural information which the grammar may refer to, and the grammar can refer to the content of more than one projection in the analysis of particular phenomena. For example, in Van Valin (2005: 80–81) it is argued that one of the sources of VPs in languages like English is the interaction of the information structure and constituent

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

491

492

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

projections: in predicate focus, as in Figure 12.1, the AFD demarcates a VPlike grouping, which is a crucial part of conjunction reduction constructions (topic-chaining), whereas in ‘VP’-ellipsis and -fronting, there is narrow focus on the privileged syntactic argument (PSA) of the clause, leaving the remainder as a VP-like grouping. In this case the VP-like grouping corresponds to the background, whereas in conjunction reduction (topic-chaining) it corresponds to the comment. (See (4) and (5) in Section 12.3 for more discussion.) The IS-projection is not the only source of VPs, but it is an important one.

12.3

The Representation of the Discourse Context

The focus constructions identified by Lambrecht (predicate focus, sentence focus, and narrow focus) reflect the context in which the utterance occurs. It is important, therefore, to have representations of the discourse context, in order to justify the information structure of the utterance. It will also turn out that the discourse representations will play a role in the linking in certain constructions in some languages. For the representation of the discourse context Van Valin (2005) and Shimojo (2008) employ the version of Discourse Representation Theory (DRT) (Kamp and Reyle 1993) proposed in von Heusinger (1999). In this system, referring expressions and their coreference relations are represented in discourse representation structures (DRS), as in Figure 12.5. The first sentence introduces the referring expression John, which is represented by the variable ‘v’, and the proposition expressed is ‘v stood up’.3 The second sentence carries over the information in the first DRS (which is dark grey rather than black due to its being less salient in the ICG), since that is the context in which it is uttered, and it introduces two new referring expressions, he and the door, which are assigned the variables ‘w’ and ‘x’, John stood up.

He opened the door.

Mary kissed him.

v

w, x

y, z

John (v) v stood up

John (v) v stood up he (w) door (x) w=v w opened x

John (v) v stood up door (x) w=v w opened x Mary (y) him (z) z=w y kissed z

Figure 12.5 A simple example of DRSs

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

respectively. Since ‘w’ is a pronoun, its reference must be specified, and the fact that it refers to ‘v’ is stated, and then the proposition ‘w opened x’ is given. The final sentence is uttered in the context created by the first two sentences, so the information from them is carried over, and the information from the first DRS is light grey and that from the second is dark grey, indicating their relative salience in the ICG. It introduces two more referring expressions, Mary (assigned ‘y’) and him (assigned ‘z’). Him being a pronoun, its reference must be specified as ‘z ¼ w’, and the proposition in it is ‘y kissed z’. How are focus structures related to the DRSs? In Lambrecht’s approach the focus is defined as the assertion minus the presupposition. Consider the following question–answer pairs, in which the focus is in small caps. (1)

a. What did Bill do? b. He bought a new car.

(2)

a. What did Bill buy? b. He bought a new car.

The (b) sentence is structurally the same in both pairs, but the focus is different, due to the different presuppositions in the ICG established by the questions. This can be represented as shown in Figures 12.6 and 12.7.

Figure 12.6 Derivation of predicate focus

Figure 12.7 Derivation of (unmarked) narrow focus

Central to the derivations is the introduction of two meta-variables, P ‘unspecified predicate’ and X ‘unspecified individual or entity’. P is

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

493

494

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

typically realized as do P ‘do something’ for an action, happen P ‘something happened’ for an event, or be P ‘be something’ for certain kinds of states. In Figure 12.6, the presupposition is a function of the question in (1a), and the assertion is in (1b). Since the focus is the assertion minus the presupposition, it is bought a new car, with only the PSA he corresponding to an element in the ICG. In (2), on the other hand, both Bill and buy are part of the presupposition, leaving only a new car as the focus. Representing the presupposition and the assertion separately amounts to breaking up the non-initial DRSs in Figure 12.5 into the previous utterances(s) and the current utterance. What role can the DRSs play in the analysis of grammatical phenomena, beyond explicating question–answer pairs like those in (1) and (2)? They turn out to have a surprising number of applications, as will be shown in the remainder of this chapter. To begin with, consider discourse-driven zero anaphora in languages like Mandarin, Thai and Japanese, in light of RRG’s rejection of phonologically null lexical elements in the syntax. The standard way of dealing with them is to posit phonologically null pronouns (pros) occupying argument positions in familiar tree structures, and the syntax can manipulate the null pronouns exactly like the overt pronouns found in familiar languages. Such an analysis was possible in earlier versions of RRG (e.g. Van Valin and LaPolla 1997: 231–232) but not in contemporary RRG.4 The key idea is that in languages like these three, discourse antecedents can directly satisfy the valence requirements of verbs; in RRG terms, the Completeness Constraint in these languages can be satisfied by overt RPs in the syntax and discourse referents in DRSs. This obviates the need to posit null pronominals, consistent with the RRG ban on null lexical elements in the syntax. In order to illustrate how this works in practice, we will present the analysis of the Japanese verb-less numeral classifier construction in Shimojo (2008). The construction is given in (3b); (3a) is a simple clause with a transitive verb. (3)

a. (Taro-ga) ringo-o ni-ko katta. Taro-nom apple-acc 2-clf buy.pst ‘(Taro) bought two apples.’ b. (Oyatu-wa,) ringo-o ni-ko da. snack-top apple-acc 2-clf be.prs ‘(As for snack,) [Taro got/will eat/etc.] two apples.’ [Literally: ‘As for snack, two apples [acc] is’]

In the first example, accusative case is assigned to the undergoer of the transitive verb kau ‘buy’, which is the normal situation in Japanese. In the second example, however, the RP ringo-o ni ko ‘two apples’ carries accusative case (-o), but there is no transitive verb in the sentence; the only verbal element is the copula, da, which expresses tense and which, importantly, does not assign accusative case. The presence of tense indicates that this is a well-formed clause, not a sentence fragment. The copula is a particular

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

feature of this construction, and without the numeral classifier the clause with the copula and the accusative marking on the undergoer is ungrammatical, unless it is an emphatic utterance to focalize ringo ‘apple’ (M. Shimojo, pers. comm.). The central question this construction poses is, ‘what licenses the accusative case on the RP in the absence of a transitive verb?’ A second question concerns the interpretation of the actor in the sentence. It is a general principle of Japanese that continuing, highly activated elements of the ICG, especially topical actors (PSAs), are not normally expressed overtly. Hence representing the activated constituents is crucial for explaining this construction. Shimojo did an internet search for instances of this construction and looked at the first hundred examples. In eighty-nine of them the actor and the verb were recoverable from the previous utterance; in eight of them the interpretation was predictable from the preceding context (e.g. ‘I just went to Starbucks. Two tall lattes-acc is,’ where the most likely meaning is ‘bought’ or ‘drank’). The final three cases were online classified ads, where the only verb that fits the context is ‘sell’. Thus, the interpretation of the verb-less numeral classifier construction depends on either the immediately preceding utterance or contextually relevant sociocultural background knowledge (the GCG). Shimojo’s RRG analysis based on representing context via DRSs works for the eighty-nine discourse-context-dependent examples. The preliminary step in the analysis in terms of linking from semantics to syntax is to represent the activated information in the ICG; in this example, it is ‘Taro got a snack’.5 The first two steps in the linking from semantics to syntax are to get the logical structure (LS) of the predicate from the lexicon along with the referring expressions that fill the argument positions, based on the message to be expressed, first, and assign the semantic macroroles, second. This yields Figure 12.8b. The next step is to select the PSA and assign case. This yields Figure 12.8c. This looks like a standard linking with a transitive verb in any accusative language. But now the general principle that highly activated continuing elements in the ICG are not normally expressed overtly in Japanese comes into play. The next step involves the selection of the syntactic templates from the syntactic inventory. The basic principle states that the number of argument slots in the core template corresponds to the number of specified argument positions in the semantic representation, as a default, and it also assumes a slot for the predicate in the nucleus. Based on these representations, it appears that the syntactic template should have two slots for core arguments, but because the actor-PSA and the verb are part of the presupposition, they don’t figure in the determination of the core template. The only part which is not part of the presupposition is ‘two apples’, and so this results in the selection of a core template with only one argument slot (Figure 12.8d). The final step is linking the accusative undergoer ‘two apples’ to the argument position and inserting the copula in the nucleus to carry the

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

495

496

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Figure 12.8a Representation of pragmatic presupposition for (3b)

Figure 12.8b Steps 1 and 2 of the semantics-to-syntax linking in (3b)

Figure 12.8c Step 3 of the semantics-to-syntax linking in (3b)

Figure 12.8d Step 4 of the semantics-to-syntax linking in (3b)

tense (not represented in these simplified semantic representations). The actor RP Taro and the verb ‘GET’ do not appear overtly but are represented solely in the DRS, as indicated by the curly brackets. This yields the assertion in (3b) depicted in the ‘assertion DRS’ in Figure 12.8e. The Completeness Constraint is satisfied by the presupposition DRS and the syntactic representation. We may refer to the version of the constraint as found in Japanese, Mandarin and Thai as the ‘Extended Completeness Constraint’.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Figure 12.8e Summary of linking from semantics to syntax in Japanese verb-less numeral quantifier construction in (3b)

Figure 12.9 Summary of linking from syntax and discourse to semantics in Japanese verb-less numeral quantifier construction

The presupposition DRS plays an equally important role in the linking from syntax (and discourse) to semantics. The entire linking is summarized in Figure 12.9. The first step is the parser outputting a labelled syntactic tree. The second step is the gleaning of the representation for indicators of the function of the elements. In this example there is only one RP plus the copula in the tree. The RP carries accusative case and so is determined to be an undergoer. Accusative undergoers are associated with M-transitive verbs, which leads to the expectation that there should be a nominative actor and an M-transitive verb in the sentence, but there is neither a nominative RP nor an M-transitive verb in the clause. However, in the presupposition DRS there is an M-transitive verb with an actor argument which would occur as a nominative RP, and so the conclusion is that the xargument in the DRS, Taro, is the actor which pairs with the overt undergoer, ringo-o ni-ko. The next step is to go to the lexicon and retrieve the

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

497

498

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

logical structure of the verb identified in the previous step. But there is no predicating verb in the sentence, and accordingly it is necessary to refer again to the background DRS for it. The logical structure for ‘GET’ is called up and its two arguments are assigned actor and undergoer macroroles. The final step is the association of the arguments in the DRS and the syntax with the arguments in the logical structure: Taro is the x-argument and the actor of ‘GET’, and the a argument is the actor, hence Taro links to the a argument; similarly, ringo-o ni-ko is an undergoer, and the b argument is the undergoer in the logical structure of ‘GET’, therefore ringo-o ni-ko links to the b argument. The resulting interpretation is given below the assertion DRS. Here again the Extended Completeness Constraint is satisfied by arguments from a DRS and the syntax. This construction is particularly interesting, because it requires context not only for the recovery of an argument, which is the usual situation with discourse-driven zero anaphora, but also for the interpretation of the predicate (verb). This is a phenomenon which is not limited to languages like Japanese. English provides an example with ‘VP’-ellipsis across speakers, as exemplified in (4b, b′). (4)

a. Kim is eating an ice cream cone, and Pat is, too. a′. Kim is eating an ice cream cone, and so is Pat. b. Speaker 1: Kim is eating an ice cream cone. Speaker 2: Pat is, too. b′. Speaker 2: So is Pat.

In Van Valin (2005: 231–235) the linking in (4a, a′) was presented in order to show how RRG handles ‘VP’ phenomena despite not positing a VP node in the constituent projection of the clause. Key to the analysis is the projection of the semantic representation of the first clause onto the second clause in the sentence, and this is possible because the two clauses are part of a single sentence produced by a single speaker. This is not the case in (4b, b′), however, as the two clauses are produced by two separate speakers in separate utterances. Yet the interpretation is the same regardless of whether the clauses are part of the same utterance or two utterances. How can the interpretation be accounted for in the multi-speaker situation? The answer, as with the Japanese verb-less nominal classifier construction, involves DRSs. Instead of copying the semantic representation of the first clause resulting from the syntax-to-semantics linking (with the PSA replaced by a variable) onto the second clause, the result of the linking constitutes the assertion made by the utterance, which is represented by a DRS, and it becomes part of the ICG and functions as the pragmatic presupposition for the second assertion. Because it is necessary for the construction, the DRS contains the tense–aspect values of the clause. The proposition and operators contained in this DRS are projected as the semantic representation of the second utterance, and the single RP in the second utterance is

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Figure 12.10 Linking from DRS in cross-speaker ‘VP’-ellipsis in English

linked as the PSA of the second clause. This can be summarized in Figure 12.10. Thus, here again there is significant propositional information recovered from context, which is represented by a DRS. There is another interesting feature of this construction, which is not very common in English, namely the option of having ‘RP AUX [, too]’ or ‘[So] AUX RP’, as in both pairs of examples in (4). What influences the choice? A clue can be found in ‘VP’-fronting, which shows a similar alternation. (5)

a. a′. b. b′.

I expected to find someone mowing the yard, and mowing the yard was Bill. #I expected to find someone mowing the yard, and mowing the yard Bill was. I expected to find Bill mowing the yard, and mowing the yard he was. #I expected to find Bill mowing the yard, and mowing the yard was he.

In the (5a) examples, the first clause does not introduce a specific referent for the actor of mow, and when the referent is introduced, it is more felicitous to have it after the finite auxiliary verb rather than before it; what we have here is in effect ‘focal subject inversion’, rather like what is found in Italian and discussed by Bentley in Chapter 11. When, on the other hand, the first clause introduces a specific referent, as in the (5b) examples, the pronoun referring

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

499

500

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

back to it is strongly preferred in pre-auxiliary position. Thus, when the PSA is focal, it occurs after the finite auxiliary, and when it is topical, it occurs before it. This pattern seems applicable to ‘VP’-ellipsis in (4). When the participant functioning as the PSA of the second clause or utterance has been mentioned before, the ‘Pat is, too’ order is more appropriate, and the PSA, Pat, is a contrastive topic in relation to the PSA of the first utterance or clause, in this case, Kim. When, on the other hand, the PSA referent in the second clause or utterance has not been mentioned before and is therefore new and focal, the ‘so is Pat’ order is more appropriate. Thus, these two constructions offer two of the rare examples of IS-motivated ‘subject inversion’ in English, something normally associated with Italian or Spanish.

12.4

Conjunction Reduction

The opposite of ‘VP’-ellipsis is conjunction reduction: in the former the PSA is focal and is the only core argument in the linked clause, as in (4a), whereas in conjunction reduction the PSA is topical and is missing from the linked clause, for example, Bill is eating an ice cream cone and then will drink a beer. As noted in Section 1.7.2 in Chapter 1, the PSAs of these two (or more) clauses are coreferential, but there is no pronominal element in the linked clauses that can be interpreted as coreferential with the PSA of the first clause. It was suggested there that DRSs are the key to the analysis of conjunction reduction, just as they are to the analysis of ‘VP’-ellipsis. The crucial fact about this construction is that the controller–pivot relationship is restricted to the controller being the first RP in the core of the first clause, and the pivots being realized by the absence of a pre-nuclear RP in the core of the linked clauses. This means the controller is the PSA of the initial clause, and the pivots are the PSA of each of the linked clauses. Hence, in the constructional schema for conjunction reduction in Figure 12.11 a crucial constraint is the restriction of coreference to the PSAs in adjacent clauses. The analysis of Mary bought a new car and drove it to Buffalo is given in Figure 12.12. PSA coreference is specified in the DRS, based on the constructional schema, and the linking algorithm (given here in a reduced form) identifies the PSAs in both clauses. The coreference statement ‘x ¼ v’ satisfies the constraint, since both x and v are the PSAs of their respective clauses. The PSA of the second clause is not lexically filled but is coreferential with the PSA of the first clause, thereby avoiding a Completeness Constraint violation. What happens when the PSAs are not coreferential? This is illustrated in Figure 12.13. In this example, the PSA of clause 1, Mary (v), is not coreferential with the PSA of clause 2, which is coreferential with a new car (y ¼ w), resulting in an

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Figure 12.11 Constructional schema for English conjunction reduction

Figure 12.12 Analysis of conjunction reduction in English

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

501

502

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Figure 12.13 Ungrammatical conjunction reduction in English

ungrammatical sentence. This was termed ‘grammatically constrained’ coreference in Section 1.7.2 in Chapter 1, and it contrasts with ‘grammatically unconstrained coreference’ as exemplified there by the data from Archi (Kibrik 1979). Thus, the RRG analysis of conjunction reduction of the kind found in English, German, Icelandic and many other languages relies on the coreference statements in the DRSs and the grammatical constraints from the constructional schema.

12.5

Information Structure and Linking between Semantics and Syntax

In the remainder of this chapter we will explore the interaction of the linking algorithm with IS, starting with the linking from semantics to syntax, building on the introduction in Section 1.6. It will be shown that IS can potentially interact with every step of the semantics-to-syntax linking. Some of these interactions are found in all languages, for example the effect of the activation status of a referent on the way the expression referring to it is coded, while others, such as the effect of IS on PSA selection, range from non-existent to crucial in different languages.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

The major steps of the semantics-to-syntax linking algorithm are summarized in (6). (6)

Linking algorithm: semantics to syntax (summary) 1. Construct the semantic representation of the sentence, based on the LS of the predicator. 2. Determine the actor and undergoer assignments, following the ActorUndergoer Hierarchy. 3. Determine the morphosyntactic coding of the arguments a. Select the PSA, based on the PSA Selection Hierarchy and associated principles. b. Assign the arguments the appropriate case markers and/or adpositions. c. Assign the agreement marking to the main or auxiliary verb, as appropriate. 4. Select the syntactic template(s) for the sentence following the template selection principles. 5. Assign arguments and adjuncts to positions in the syntactic representation of the sentence.

12.5.1 Step 1 According to the model of a speaker proposed in Levelt (1989), the first step in the production of an utterance is the formulation of an idea or thought to be communicated, and the initial fully linguistic step is to convert it into a semantic representation, which is step 1 in (6). This is primarily semantically driven, but IS can have an influence in a number of ways. First, certain focus constructions favour verbs of certain types. A good example is sentence focus constructions, which have robust contextual constraints on their occurrence (cf. Bentley, Chapter 11). Lambrecht (1987), along with Kuno (1972a), observed that ‘the predicates most commonly permitted in [sentence focus] sentences involve “presenting verbs”, i.e. intransitive verbs expressing appearance or disappearance of some referent in the internal or external discourse setting, or the beginning or end of some state involving the referent’(1987: 373). A second example comes from languages without productive voice systems. It was discussed in Chapter 5 on grammatical relations that IS can influence PSA selection in languages with voice operations (see also Section 12.5.3). It doesn’t follow, however, that IS is completely lacking in influence in languages without voice. Languages may possess verbs which present a state of affairs from different perspectives, as illustrated in (7). (7)

a. a′. b. b′. c. c′.

Bill gave a book to Sally. Sally received/took a book from Bill. Covid-19 killed Bill’s cousin. Bill’s cousin died from Covid-19. Sally led Bill through the park to the beach. Bill followed Sally through the park to the beach.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

503

504

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

These pairs of sentences have different verbs which reflect different PSA choices from among the participants in the state of affairs, and the selection of a verb with its PSA reflects the salience of the referent of the PSA in the discourse. In a context in which the referent of Bill is the topic and the referent of Sally is newly mentioned, (7a, c′) would be more natural than (7a′, c). Similarly, if the topic is the pandemic, then (7b) is appropriate, whereas if Bill’s cousin is being talked about, then (7b′) is perhaps more natural. In a language without voice, the primary way to vary the PSA in terms of the different topicality statuses of the participants is through selecting a verb which describes the same state of affairs as another verb but with a different PSA, as in (7).6 A third, and very important, case of IS influence concerns the coding of referring expressions. In (7a), for example, the referent of the giver of the book is coded as a proper name, Bill. There are, however, a wide variety of options available to a speaker, depending on the status of the referent in the ICG: someone, a man, this guy, this/that man, the man, the man I met at the party yesterday, the jerk/fool/idiot/genius, Bill, Mr Smith, he, that one, . . . All of these different referring expressions can be used to denote the individual ‘Bill Smith’, and the choice is a function of a number of factors, such as how recently he has been referred to, how familiar the speaker believes the interlocutor is with him. (See Bentley’s chapter on IS, Section 11.2, for a detailed introduction to these distinctions.) In some languages, especially head-marking languages, coding options include bound morphological pronouns and anaphors, and clitics. The inventory of coding possibilities varies from language to language, but in every language IS factors influence the selection in actual utterances.

12.5.2 Step 2 The second step in (6) is macrorole assignment. In terms of the AUH, the actor is the leftmost specified argument in the LS, and the default choice of undergoer is the rightmost specified argument in the LS. This was shown in the discussion of macrorole choice in Chapters 1 and 4. With respect to actor selection, the only variation is whether the leftmost argument will be specified, that is, morphosyntactically instantiated in the clause, or left unspecified, as in an ‘agent-less’ passive, for example. In some cases this may be motivated by the activation status of the referent of the leftmost argument, as we saw in the Japanese example in (3b) in which highly activated PSAreferents need not be explicitly mentioned again. In sentences involving causal chains with two effectors, for example Sam opened the chest with the skeleton key, the instigating animate effector can be left unspecified and the intermediate inanimate effector, the instrument, can be selected as the actor, yielding The skeleton key opened the chest.7 In both of these versions the actor is the leftmost specified argument in the LS of the verb.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

As pointed out in Chapters 1 and 4, undergoer selection is not absolute; in some languages, for some verbs with three or more arguments, it is possible to assign undergoer to an argument other than the rightmost. English is one of the most permissive languages in this regard, as variable undergoer selection is possible with most three-argument verbs. The most common alternations are given in (8). (8)

a. Dative shift: Sam handed the letter to Alice vs. Sam handed Alice the letter. b. Transfer alternation: Max presented the prize to Sally vs. Max presented Sally with the prize. c. Locative alternation: Mary sprinkled sugar on the cookies vs. Mary sprinkled the cookies with sugar. d. Removal alternation: Larry drained the water from the pool vs. Larry drained the pool of its water.

In many cases, there is a semantic difference as a result of the different undergoer choices, namely affectedness, as illustrated in (9). (9)

a. Dative shift: Mary taught French to the students (but they didn’t learn a word of it). Mary taught the students French (#but they didn’t learn a word of it). b. Locative alternation: The workmen loaded the boxes onto the truck. (all the boxes loaded, truck may or may not be filled up; default interpretation) The workmen loaded the truck with the boxes. (truck is full, all the boxes may or may not be loaded; default interpretation)

In the dative shift examples the affectedness difference concerns whether the students learned anything from Mary’s teaching: when the students is not the undergoer, there is no implication one way or the other that they learned anything, whereas when the students is the undergoer, it implies that they did learn something, hence the oddity of the denial in the second clause. The contrast in the locative alternation has been much discussed over the past fifty years and needs no further comment. When the alternation is semantically motivated, there is no direct role for IS to play in it. In many cases, however, there is no semantic contrast, as in (8a), and in such cases, IS can influence the choice of undergoer in English, due to the difference in word order. In a language like German or Russian, if the recipient is more topical than the theme, the order of the dative and accusative RPs can simply be reversed. This is not an option that is available to English speakers, as *Sam handed to Alice the letter or *Max presented with the prize Sally is ungrammatical. Since the undergoer immediately follows the verb in active voice cores, changing the argument selected as undergoer changes the word order. Thus, the non-default undergoer choice better reflects the topic-precedes-focus word order tendency, as the following examples show. (‘>’ means ‘is preferred in this context over’)

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

505

506

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

(10)

a. Q: What happened with the letter? A: Sam handed it to Alice > Sam handed Alice the letter/*it. b. Q: Why is Alice upset? A: Sam handed her the letter > Sam handed the letter to her.

(11)

a. Q: What did Mary do with the sugar? A: She sprinkled it on the cookies > She sprinkled the cookies with it. b. Q: What did Mary do to the cookies? A: She sprinkled them with sugar > She sprinkled sugar on them.

Each of the questions establishes the referent of either the recipient or the theme as the topic in this limited context, and the more natural response to it is the one in which the topic expression precedes the newly mentioned referent. Thus, it is reasonable to conclude that there are cases in English where IS can influence undergoer selection. It remains to be seen if there is any evidence that this happens in other languages. German and Croatian, for example, have limited variable undergoer selection, and it is doubtful that it could have the same IS motivation as in English, since, as noted earlier, simply reversing the order of dative and accusative RPs achieves the same effect.

12.5.3 Step 3 The third step in the semantics-to-syntax linking is the assignment of morphosyntactic functions and properties. The former concerns primarily PSA selection and the latter agreement and case assignment. The influence of IS on PSA selection in some languages was discussed in the chapters on grammatical relations (5) and IS (11), and we will present a particularly striking example of it. Ever since Schachter (1976), Tagalog has been seen as a language in which IS plays a role in PSA selection. While this is often discussed in terms of the definiteness or specificity of the undergoer being crucial for PSA and voice selection (see LaPolla Chapter 5), there are additional factors at work, as Latrouite (2011) shows. For example, causative change-of-state predicates (e.g. tatay ‘kill’) strongly favour undergoer voice, whereas activity predicates favour actor voice. Latrouite call this ‘event structural prominence’, and it interacts with IS prominence in the determination of voice and PSA selection. A third factor is whether the actor or undergoer argument is displaced to the beginning of the utterance. Latrouite and Van Valin (2021), building on Schachter and Otanes (1972), survey displacement constructions in Tagalog, and the two that are most relevant to this discussion are ang-inversion and ay-inversion; the former is illustrated in (12) with the verb sulat ‘write’.8 (12)

a. Sulat ng liham kay Lisa ang babae. write gen letter dat Lisa nom woman ‘The woman wrote a letter to Lisa.’

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

b. Sulat ng babae kay Lisa ang liham. write gen woman dat Lisa nom letter ‘A/The woman wrote the letter to Lisa.’ c. Ang babae ang sulat ng liham kay Lisa. nom woman nom write gen letter dat Lisa ‘It was the woman who wrote a letter to Lisa.’ d. Ang liham ang sulat ng babae kay Lisa. nom letter nom write gen woman dat Lisa ‘It was the letter that a/the woman wrote to Lisa.’

(12a, b) represent the basic word order in a simple clause in Tagalog; (a) illustrates actor voice, (b) undergoer voice. The nucleus is the first element in the clause, and the default position for the ang-marked PSA is core-final. The ang-inversion construction in (c) and (d) is usually translated by an English it-cleft, which indicates that it should be interpreted as narrow focus (see Latrouite 2021). In these examples the inverted RP is the nominative RP congruent with the voice of the predicator in the nucleus, that is, in (c) the verb is actor voice, and the displaced RP is the actor, while in (d) the verb is undergoer voice, and the inverted RP is the undergoer. However, this correlation does not always hold: Latrouite and Riester (2018) found that there were cases where the actor was inverted but the verb was undergoer voice, as exemplified in (13a). It is, however, not possible to invert the undergoer in actor voice, as the ungrammaticality of (13b) shows. (13)

a. Ang babae ang sulat kay Lisa ang liham. nom woman nom write dat Lisa nom letter ‘It was the woman who wrote the letter to Lisa.’ b. *Ang liham ang sulat kay Lisa ang babae. nom letter nom write dat Lisa nom woman ‘It was the letter that the woman wrote to Lisa.’

What explains this asymmetry? Latrouite and Riester argue that actor and undergoer arguments have default IS values, with the actor being the default topic and the undergoer being the default focus. With a topical actor and an undergoer as part of the focus, IS does not require special marking, and referentiality and verb semantics influence voice selection. If the IS mapping is divergent, for example if the actor is topical (unmarked) and the undergoer is topical (marked), we get undergoer voice to signal this divergence. If the undergoer is a contrastive displaced topic (marked) and the actor is topic (unmarked), we will get undergoer voice. If the actor is focal (marked) and the undergoer is focal (not marked), then actor voice and displacement (in the case of contrastive focus) tends to be used for this divergence. These default correlations are not surprising, since actors are the default choice for PSA in the vast majority of languages (see Van Valin and LaPolla 1997, §6.5), and the default focus structure, predicate focus, has the PSA as

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

507

508

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

topic and the undergoer as focal. What is surprising is that IS can override morphosyntactic constraints: focal actors are strongly preferred in displaced (inverted) positions, as in (13a); note that the undergoer is the PSA and therefore topical, which is the non-default IS function of undergoers. Hence a displaced focal actor with a topical undergoer PSA and undergoer voice reflects the marked IS-functions of the arguments. Why is (13b) impossible? Two factors may be at work. First, the ang-marked lowest-ranking argument of the verb is normally non-referential, which makes it a poor candidate for displacement. Moreover, ang-marked arguments are always referential. Hence the occurrence of ang-marking on the lowest-ranking argument is incompatible with actor voice. In (12d), on the other hand, undergoer voice signals that the lowest-ranking argument is referential, and therefore angmarking is appropriate. Thus one of the functions of the construction in (12d) is to express a focal referential undergoer. This is a complex and interesting example of the interplay of semantic macroroles and IS functions with PSA and voice selection. Agreement between the verbal complex in the nucleus and one or more core arguments and case assignment shows an interesting asymmetry: there appear to be many more examples of IS-influenced case assignment than of IS-influenced agreement. The main examples of IS-influenced finite verb or auxiliary agreement concern sentence focus constructions in languages with restrictions on the PFD in simple clauses. It was pointed out in Chapter 11 that in Italian the PFD in simple clauses excludes the pre-nuclear position in the core; the AFD is restricted either to the precore slot (PrCS) or to the nucleus plus post-nuclear constituents. Hence no core-internal preverbal focus is permitted, and in the case of sentence focus, the focal ‘subject’ must occur after the nucleus. It was also mentioned that in the southern Bantu languages Setswana and Sesotho (Demuth 1989), the restriction is even stricter: no pre-nuclear focal material is permitted at all. In sentence focus the focal argument must be postnuclear, and the predicate in the nucleus fails to show agreement. The contrast between predicate focus and sentence focus is clear in (14), from Demuth (1989). (14)

Setswana a. Monna o-fihl-il-e. man 3sg-arrive-pfv-mood ‘The/*a man arrived.’ b. Ho-filh-il-e monna. loc-arrive-pfv-mood man ‘There arrived a man,’ or ‘A man arrived.’

Monna ‘man’ must be interpreted as topical, in accordance with its prenuclear position in (14a), which is predicate focus. In the sentence-focus construction in (14b), on the other hand, monna ‘man’ is focal and must occur after the nucleus, and there is no agreement with it; the agreement

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

prefix in (a) is replaced by a locative prefix. Thus, in Setswana there is a complex interaction among IS, word order and agreement, with the result that in clauses with intransitive predicates there is agreement with the single argument only when it topical and pre-nuclear. Italian, by contrast, does show agreement with a post-nuclear argument in sentence-focus clauses. Bentley (2018) investigates the interaction of word order (RP-V vs. V-RP) and agreement in nine Italo-Romance varieties and finds variation in whether the finite verb or auxiliary shows agreement in VS constructions. In some, agreement is obligatory, like in Italian, in some it does not occur, like in Setswana, and in some it is optional. What we have yet to find is an example in which the PFD includes the entire clause, as in English, and agreement with the PSA depends on whether the pre-nuclear RP is within the AFD or not. This is represented in (15). (The AFD is represented by underlining.) (15)

a. [CLAUSE [CORE RP [nucleus V/AUX[þAGR] . . . ]]] b. [CLAUSE [CORE RP [nucleus V/AUX[AGR] . . . ]]]

In many languages there is agreement or cross-reference for more than just the PSA with multi-argument verbs. In some languages, Lakhota (Siouan), for example, undergoers of M-transitive verbs are coded on the verb (in the third person primarily by a zero morpheme) regardless of whether there is an overt undergoer RP or not and whether the argument is topical or focal. In many Bantu languages, for instance Chichewa (Bresnan and Mchombo 1987), Setswana (Demuth and Johnson 1989), Kikuyu (Kihara 2017), on the other hand, the so-called ‘object marker’ is in complementary distribution with an overt undergoer RP. When the referent of the undergoer is focal, it must be expressed by a full RP immediately following the verb; when it is not focal, it must be realized by a bound pronominal on the verb, and an overt undergoer RP is not possible in the same clause but if present must appear in a detached position. Thus, there are clearly IS constraints on ‘object’ agreement or cross-reference in these languages. Finally, we turn to case assignment. A wonderful example of IS-influenced case assignment from Kaluli (Papua New Guinea, Schieffelin 1985) was given in Section 11.6. Korean exhibits two kinds of IS-influenced case marking, which are called ‘case spreading’ and ‘case stacking’ (Park 1995; Han 1999). Nominative case spreading is exemplified in (16) (from Park 1995), and accusative case spreading and stacking are illustrated in (17) (from Han 1999). ˇ

(16)

a. Thoyoil-ey kongcang-eyse pwul-i na-ass-ta. Basic form Saturday-loc factory-loc fire-nom break.out-pst-decl ‘Fire broke out in the factory on Saturday.’ b. Thoyoil-i kongcang-i pwul-i na-ass-ta. Case spreading Saturday-nom factory-nom fire-nom break.out-pst-decl ‘Fire broke out in the factory on Saturday.’

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

509

510

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

The contrast in (16) illustrates nominative case spreading with an intransitive verb. The default form in (a) is an appropriate answer to the question ‘What happened at the factory on Saturday?’ The AFD encompasses pwuli naassta ‘fire broke out’, with thoyoil-ey ‘on Saturday’ and kongcang-eyse ‘in/at the factory’ being in the background within the PFD. Sentence (b), on the other hand, would be an appropriate answer to ‘What happened?’, which means that it is a sentence-focus construction in which the AFD includes the entire clause. Thus it appears that the nominative case marks the AFD in this construction. This correctly predicts that if the question were ‘What happened on Saturday?’, then only kongchang ‘factory’ and pwul ‘fire’ would be nominative, and that if it were ‘What happened at the factory?’, then only thoyoil ‘Saturday and pwul would be nominative. With transitive verbs it is the accusative case that is stacked or spread, and the two forms have different semantic and pragmatic consequences. (17)

a. Chelswu-ka Yenghi-eykey kkot-ul cwu-ess-ta. Basic form Chelswu-nom Yenghi-dat flower-acc give-pst-decl ‘Chelswu gave a flower to Yenghi.’ b. Chelswu-ka Yenghi-lul kkot-ul cwu-ess-ta. Case spreading Chelswu-nom Yenghi-acc flower-acc give-pst-decl ‘Chelswu gave Yenghi a flower.’ c. Chelswu-ka Yenghi-eykey-lul kkot-ul cwu-ess-ta. Case stacking Chelswu-nom Yenghi-dat-acc flower-acc give-pst-decl ‘Chelswu gave a flower to Yenghi.’

The basic form has default interpretations, both semantic and pragmatic. The default interpretation is that the transfer of the flower to Yenghi was completed, but this is not the only possible reading; it is also possible to say without contradiction ‘but she did not receive it’. In terms of IS, the unmarked focus position in Korean, as in other SOV languages, is the immediate preverbal position (Kim 1988), and there are two ways to interpret the AFD in predicate focus in (17a): it may not include Yenghi-eykey ‘Yenghi-dat’ (minimal AFD), or it may include it (extended AFD). In (17b) the accusative case has replaced the dative on the recipient Yenghi; it has ‘spread’ from the undergoer to the non-macrorole recipient. When the recipient is in the accusative case, it is impossible to add ‘but she did not receive it’; it is a contradiction, because the accusative case signals that the transfer was successfully completed. The accusative case canonically marks the undergoer in Korean, and the undergoer codes the most affected participant in the event, as we saw in (9). Assigning the recipient accusative case signals that its referent is affected by the action, which in the context of a transfer event means that it got the item transferred.9 In (17c) the accusative case marker has been added to the dativemarked recipient Yenghi-eykey, resulting in an RP with two case markers. What does the accusative signal? It does not have the same

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

interpretation as the accusative in (b); it is possible to add ‘but she didn’t receive it’ to (c) without contradiction. Rather, it signals that the dative recipient is in the AFD; in other words, it marks the extended AFD. Thus it has a clear IS-related function. What evidence is there that the stacked accusative is a kind of focus marker? One piece of evidence comes from wh-questions. In Korean the wh-word must appear in the preverbal focus position, as in (18a); it is an instance of narrow focus. Stacking the accusative case on the dative RP signals that this RP is in the AFD. Combining case stacking on the recipient with an undergoer wh-question generates a contradiction between the narrow focus of the wh-question and the expanded AFD coded by the case stacking, and the result in (b) is ungrammatical. (18)

a. Chelswu-ka Yenghi-eykey mwuet-ul cwu-ess-ni? Chelswu-nom Yenghi-dat what-acc give-pst-q ‘What did Chelswu give to Yenghi?’ b. *Chelswu-ka Yenghi-eykey-lul mwuet-ul cwu-ess-ni? Chelswu-nom Yenghi-dat-acc what-acc give-pst-q ‘What did Chelswu give to Yenghi?’

Here we have a clash of IS factors as the explanation for the ungrammaticality of (18b), which supports the analysis of the stacked accusative as focusrelated. Han (1999) presents an extensive discussion of the full range of casesubstitution and case-stacking phenomena and formulates case-marking rules to account for them; they supplement the basic RRG case-marking rules for accusative languages introduced in Chapters 1 and 7.10

12.5.4 Step 4 The next step in the linking is selection of the syntactic template for the utterance. The basic principles were introduced in Chapter 1, and they made primary reference to the number of specified arguments in the LSs in the semantic representation. The only reference to IS is indirect, mentioning displacement of arguments to the PrCS/PoCS (post-core slot) and the occurrence of an argument-modulation voice construction, both of which can be motivated by IS factors. The basic core syntactic templates, as introduced in Chapter 1, are maximally unspecified and would not contain any IS features. While this would be true for languages like English, German, and all SOV and VSO languages, where the PFD includes the whole clause, for SVO languages like French, Italian, Setswana and Sesotho, in which the PFD is constrained and excludes the core-initial, prenuclear position, this could be indicated in the syntactic templates, since it is a general constraint in the grammar of these languages. This is shown in Figure 12.14, which represents the constraint that a pre-nuclear core argument is necessarily outside of the PFD. This constraint would be a feature of all core templates in the language.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

511

512

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Figure 12.14 Restricted PFD in some SVO languages

Figure 12.15 Syntactic templates for English wh-question and locative-inversion constructions

There are syntactic templates that are specialized for specific IS properties, and they would have the AFD specified on them, since it is a feature of the construction. Examples from English include narrow focus on a wh-XP in the PrCS and the locative inversion construction (e.g. Into the room walked a tall stranger), which is always sentence focus. They are illustrated in Figure 12.15, from Van Valin (2005). In a language with a PrCS which always expresses narrow focus, regardless of whether the XP in it is a wh-expression or a plain XP, then there would be no [þWH] feature, and the extent of the PFD would be left unspecified. Italian is interesting, in that it combines the restriction in Figure 12.14 with the wh-template in Figure 12.15, yielding a two-part PFD which includes the PrCS, on the one hand, and the nucleus and post-nuclear constituents, on the other, but excludes the pre-nuclear core-initial position (Bentley 2008). This can be seen in Figure 12.16, which is the structure of (38b) in Chapter 11. It means ‘A bike George bought, not a car.’ (The grey shading indicates the discontinuous PFD.)

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Figure 12.16 Split PFD in Italian

The PoDP RP la macchina ‘a/the car’ is outside the scope of the illocutionary force operator of the first clause, due to its being in a detached position. It is therefore necessarily part of the presupposed ICG.

12.5.5 Step 5 The final step involves mapping the elements in the sentence, as developed in steps 1–3, into the syntactic templates determined in step 4, to yield the actual form of the sentence, morphophonological processes aside. A big part of this is linearization of the constituents in the sentence, something that IS has at least some influence on in every language and a great deal of influence in many. Numerous examples of the influence on the linearization of constituents, including displacement to the PrCS, for example, are discussed in the earlier chapter on IS. No further examples need to be introduced here. One very important area of the interaction of IS and linking concerns extraction phenomena and constraints thereon, which are discussed by Shimojo in Chapter 16. IS plays a central role in explaining constraints on extraction phenomena, and this not only makes possible a more principled and motivated account of cross-linguistic variation in this domain, but it also forms the basis of an empirically testable account of the acquisition of these constraints by children. See Van Valin (1994, 1998, 2008) and Van Valin and LaPolla (1997, Epilog).

12.6

Linking from Syntax to Semantics

The ultimate purpose of IS is to facilitate the assimilation by the interlocutor(s) of the information which the speaker wishes to communicate

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

513

514

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

to them assuming the ICG. Hearers arrive at the IS of an utterance via a complex set of considerations and calculations. In terms of the file card metaphor for representing referents cognitively, the file card for the referent of the topic expression is the one onto which the information in the comment should be entered, with special attention paid to the information in the AFD, which may require the creation of new file cards for newly introduced referents. There are many aspects of the interpretive process summarized in step 1 of the syntax-to-semantics linking algorithm to which IS does not contribute. For example, as we have seen, IS can affect PSA selection in some instances, and this in turn affects voice selection, if the language has voice. As we have suggested, this is the speaker’s perspective on the linking. From the hearer’s perspective, on the other hand, the IS influence on PSA selection will normally have a positive effect on information assimilation by keeping PSAhood aligned with topicality, in that the default for the topic expression is that it functions as the PSA and in most language occurs clause-initially. Deviations from the expected correlation of PSAhood, position and topicality are signalled through special prosody or special morphosyntactic constructions such as displacement of a contrastive RP to the PrCS in some languages, use of a morphological contrastive focus marker in others. In Section 12.5.3 it was shown that in Tagalog there are default discourse statuses for actor and undergoer arguments with a multi-argument verb, namely, topical for actors and focal for undergoers. Deviations from this are signalled overtly: focal actors are obligatorily preposed, while topical undergoers must be the PSA with undergoer voice on the verb. So, from a hearer’s perspective, the interplay of position, voice and IS status gives important clues to the interpretation of arguments. Three important phenomena at the syntax–semantics interface which crucially involve IS for their interpretation are determining the scope of negation, determining the scope of quantifiers, especially in sentences with multiple quantifiers, and the resolution of intrasentential pronominalization. Negation is the only operator that can function on all three layers of clause structure, and one might reasonably expect that languages would have three distinct negative morphemes, one for nuclear negation, one for core negation, and one for clausal negation, in order to avoid scope ambiguities. One might, therefore, be disappointed to find out that this rational pattern rarely occurs in human languages, and far more common is the English pattern with a single primary negative morpheme, not in English, which occurs in a restricted range of positions, and which can be interpreted as having nuclear, core or clausal scope. IS provides the means to disambiguate the scope of negation. As argued in §5.5 in Van Valin and LaPolla (1997), the scope of negation in a clause is the AFD. This is illustrated in (19) with the sentence Lucy didn’t buy a new car.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

(19)

a. Lucy didn’t buy a new car. Sally did. b. Lucy didn’t buy a new car. She leased one. c. Lucy didn’t buy a new car. She bought a used pickup. d. Lucy didn’t buy a new car. She leased a new pickup. e. Lucy didn’t buy a new car. Nothing happened.

Marked narrow focus Marked narrow focus Unmarked narrow focus Predicate focus Sentence focus

Each of the possible interpretations of the scope of not corresponds to one of the focus structures proposed by Lambrecht and used in RRG, and is signalled primarily by intonation. The interpretation of quantifier scope, especially in sentences with multiple quantifiers such as Every girl kissed a boy, has been the topic of much discussion in the generative era, and explanations have been offered in purely syntactic terms, requiring the postulation of covert movement, as well as in formal semantic terms. However, Sgall et al. (1986), Kuno (1991), Van Valin and LaPolla (1997: §5.5), Kuno et al. (1999) and Van Valin (2005: §3.6) have argued that IS plays a crucial role in the interpretation of these sentences. The basic principle is: everything else being equal, the more topical quantifier has wider scope than the less topical quantifier. In a nutshell, ‘topic Q > focal Q’. For a sentence like Every girl kissed a boy, the default (unmarked) interpretation is every > a, that is, every girl kissed a different boy, which correlates with predicate focus, the default focus structure, in which every girl is the topic expression and a boy is in the AFD. The nondefault (marked) interpretation, a > every, that is, there is a boy such that every girl kissed him, is possible if the focus structure is marked narrow focus on the PSA, leaving a boy to be the topic expression. This account makes a number of significant and correct predictions about multiple quantifier sentences cross-linguistically, and data from Italian, Japanese, Mandarin and Toba Batak (Austronesian) are analysed in Van Valin (2005). An interesting interaction of focus structure markedness and linking markedness has not previously been discussed, to our knowledge. Consider Table 12.1 regarding (i) Every girl kissed a boy and (ii) A boy was kissed by every girl. The default (unmarked) reading for these sentences has the quantifier in the PSA RP interpreted as having wide scope, as in (a) and (d). To get the other reading requires a marked focus structure, but it appears that these readings are not equally easy to get. Native speakers seem to have a more difficult time

Table 12.1 Interaction of focus structure markedness with linking markedness Sentence

Interpretation

(i)

a. b. c. d.

(ii)

every > a a > every every > a a > every

marked voice

marked focus structure

marked marked þmarked þmarked

marked þmarked þmarked marked

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

515

516

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

getting the reading in (c) than in (b). Why should this be the case? A comparison of (a) and (c) suggests the following: (c) is difficult because it combines the marked (non-default) voice (passive) with marked narrow focus on the PSA, a doubly marked sentence, which yields the same interpretation as the maximally unmarked form, (a). In processing terms, the hearer must process two non-basic patterns to arrive at the same interpretation that the two basic patterns would yield, and it is difficult to justify the processing load. Hence the preference for (a) over (b) with (i) is weaker than the preference for (d) over (c) with (ii). This is an interesting interaction of IS with syntax, namely PSA and voice selection, with semantic consequences. The final area is intrasentential pronominalization. Following the pioneering work on IS and pronoun resolution by Kuno (1972a, b, 1975), Bickerton (1975) and Bolinger (1979), Van Valin (1990) and Van Valin and LaPolla (1997: §5.6) proposed an analysis of sentence-internal pronominalization based on the following principle. (20)

Principle governing intrasentential pronominalization (Van Valin and LaPolla 1997: 227): Coreference is possible between a lexical [R]P and a pronoun within the same sentence if and only if a. the lexical [R]P is outside of the actual focus domain, and b. if the pronoun is in a syntactic argument position and precedes the lexical [R]P, there is a clause boundary between the pronoun and the lexical [R]P.

Both of these conditions involve IS. The first one is illustrated in (21). (21)

a. Sami saw his i/j sister. b. Sam i saw his*i/j sister.

In (21a) Sam is the topic expression, and his sister is in the AFD, and accordingly, Sam can be interpreted as being coreferential with his. In (b), on the other hand, Sam is in the AFD and therefore cannot be interpreted as coreferential with his. The second principle in many instances involves a contrast between the pre-detached position (PrDP) and the PrCS, and the phrases in them differ in terms of their IS function. (22)

a. a′. b. b′.

In Billi’s neighbourhood, hei is a big deal. In hisi neighbourhood, Billi is a big deal. *In Billi’s living room hei put the new plant. In hisi living room Billi put the new plant.

PrDP PrDP PrCS PrCS

In the example in (22a, a′) there is a PP containing a possessive RP in the PrDP, and in (22a) the possessor is a lexical RP and is coreferential with the pronominal PSA of the following clause, whereas in (a′) the possessor in the PrDP is pronominal and the PSA of the following clause is a lexical RP. If both are predicate focus, then coreference is possible in both, because the lexical RP is not in the AFD, and in the ‘backwards’ coreference case, (a′), condition (20b) is met, because the pronoun is not in an argument position; there is also a clause boundary between them. In (22b, b′) there is again a PP

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Influence on choice of template

Syntactic inventory

LSC syntactic template

focus structure

PSA [+prag. infl] (in some languages)

Cognitive model of context

s

Linking ipant partic f o s algorithm tatu urse s Disco

Referent-1: Activated Referent-2: Accessible Referent-3: Inactive etc.

cal [do′ (x, …)] CAUSE [BECOME pred′ (y, z)] lexi n ce o es c uen Infl choi

Lexicon Figure 12.17 The pervasive role of discourse-pragmatics in grammar

containing a possessive RP, but the PP is in the PrCS, not the PrDP, and this means the possessor in the RP is in the AFD. Consequently, (b) violates condition (20a), and coreference is impossible. In (b′), on the other hand, condition (20a) is met, because the AFD is in the PrCS, and condition (20b) does not apply because the pronoun is not in an argument position. These three phenomena all involve IS and the interpretation of sentences, which is consonant with the idea that the point of IS is to facilitate the assimilation of information.

12.7

Conclusion

RRG began as a theory of grammatical relations back in the late 1970s, and one of the fundamental claims made then was that the interaction of discourse-pragmatics and semantic roles in the constitution of grammatical relations is variably grammaticalized across languages, and these differences in grammaticalization underly important typological differences among languages. While in the ensuing decades the scope of the investigations carried out in RRG expanded to more and more phenomena, the fundamental insight that grammatical systems crucially involve discourse-pragmatics continues to be at the heart of RRG analyses, and this chapter, along with those by Bentley, LaPolla and Shimojo, as well as the grammatical sketches in Part V, illustrate this. The idea that discourse-pragmatics permeates grammar is summed up in Figure 12.17, from Van Valin (2005: 182).

References Balogh, Kata. 2021. Additive particle uses in Hungarian: A Role and Reference Grammar account. Studies in Language 45: 428–469.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

517

518

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Bentley, Delia. 2008. The interplay of focus structure and syntax: Evidence from two sister languages. In Van Valin (ed.), 263–284. Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence from Italo–Romance. Linguistics 56: 1245–1301. Berio Leda, Anja Latrouite, Robert D. Van Valin, Jr. and Gottfried Vosgerau. 2017. Immediate and general common ground. In Patrick Brézillon, Roy Turner and Carlo Penco (eds.), Modeling and Using Context, 1–14. Heidelberg: Springer International Publishing. Bickerton, Derek. 1975. Some assertions about presuppositions and pronominalization. Communication and Linguistics Studies 11, Parasession on Functionalism, 580–609. Bolinger, Dwight. 1979. Pronouns in discourse. In T. Givón (ed.), Syntax and Semantics, Vol. XII: Discourse and Syntax, 289–310. New York: Academic Press. Bresnan, Joan and Sam A. Mchombo. 1987. Topic, pronoun and agreement in Chichewa. Language 63: 741–782. Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho passive. Language 65: 56–80. Demuth, Katherine and Mark Johnson.1989. Interaction between discourse functions and agreement in Setswana. Journal of African Languages and Linguistics 11: 21–35. Filchenko, Andrey. 2007. Aspect [sic] of the Grammar of Eastern Khanty. PhD dissertation, Tomsk State Pedagogical University. Gundel, Jeanette. 1988. Universals of topic-comment structure. In M. Hammond, E. Moravcsik and J. Wirth (eds.), Studies in Linguistic Typology, 209–239. Amsterdam: John Benjamins. Gundel, Jeanette. 2012. Pragmatics and information structure. In Keith Allan and Kasia Jaszczolt (eds.), The Cambridge Handbook of Pragmatics, 585–598. Cambridge: Cambridge University Press. Han, Jeonghan. 1999. On Grammatical Coding of Information Structure in Korean: A Role and Reference Grammar Account. PhD dissertation, University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/]. Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA: Kluwer. Kibrik, Alexander E. 1979. Canonical ergativity and Daghestan languages. In Frans Planck (ed.), Ergativity, 61–78. London: Academic Press. ~ u ~ (Kikuyu) Complex Sentences: A Role and Kihara, C. Patrick. 2017. Aspects of G~ıkuy Reference Grammar analysis. PhD dissertation, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/]. Kim, Alan H. O. 1988. Preverbal focusing and type XXIII languages. In M. Hammond, E. Moravcsik and J. Wirth (eds.), Studies in Syntactic Typology, 147–172. Amsterdam: John Benjamins. Koshkaryova, N. 2000. Sentences with ditransitive verbs in Khanty. Paper presented at the 2000 RRG Conference, Dubrovnik, Croatia. ˇ

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Krifka, Manfred and Renate Musan (eds.). 2012. The Expression of Information Structure. (The Expression of Cognitive Categories 5). Berlin: Mouton de Gruyter. Kuno, Susumu. 1972a. Functional Sentence Perspective: A case study from Japanese and English. Linguistic Inquiry 3: 269–320. Kuno, Susumu. 1972b. Pronominalization, reflexivization, and direct discourse. Linguistic Inquiry 3: 161–196. Kuno, Susumu. 1975. Three perspectives in the functional approach to syntax. CLS Parasession on Functionalism, 276–336. Kuno, Susumu. 1991. Remarks on quantifier scope. In H. Nakajima (ed.), Current English Linguistics in Japan, 261–287. Berlin: Mouton de Gruyter. Kuno, S., K. Takami and Y. Wu. 1999. Quantifier scope in English, Chinese and Japanese. Language 75: 63–111. Lambrecht, Knud. 1987. Sentence focus, information structure, and the thetic-categorial distinction. Berkeley Linguistic Society 13: 366–382. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: Cambridge University Press. Latrouite, Anja. 2011. Case and Voice in Tagalog. PhD dissertation, Heinrich Heine University Düsseldorf [available on RRG website, https://rrg.caset. buffalo.edu/]. Latrouite, Anja. 2021. Specification predication: Unexpectedness and cleft constructions in Tagalog. Faits de Langues 52(1): 227–254. doi: https://doi .org/10.1163/19589514–05201011. Latrouite, Anja and Arndt Riester. 2018. The role of information structure for morphosyntactic choices in Tagalog. In Sonja Riesberg and Asako Shiohara (eds.), Perspectives on Information Structure in Austronesian Languages. Studies in Diversity Linguistics, 247–284. Berlin: Language Science Press. Latrouite, Anja and Robert D. Van Valin, Jr. 2021. An RRG account of aspects of the information structure-syntax interface in Tagalog. In Van Valin (ed.), 257–283. Levelt, Willem. 1989. Speaking. Cambridge, MA: MIT Press. Nuhn, Patrick. 2021. Ay-Inversion in Tagalog: Information Structure and Morphosyntax of an Austronesian Language. Düsseldorf: Düsseldorf University Press. O’Connor, Rob. 2008. A prosodic projection for Role and Reference Grammar. In Van Valin (ed.), 227–244. Park, Ki-seong. 1995. The Semantics and Pragmatics of Case Marking in Korean: A Role and Reference Grammar Account. PhD dissertation, University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/]. Planck, Frans. 1995. Double Case: Agreement by Suffixaufnahme. New York: Oxford University Press. Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic, actor-topic, or none of the above. In Charles Li (ed.), Subject and Topic, 491–518. New York: Academic Press.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

519

520

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA: University of California Press. Schieffelin, Bambi. 1985. The acquisition of Kaluli. In Dan I. Slobin (ed.), The Cross-Linguistic Study of Language Acquisition, 525–593. Hillsdale, NJ: Lawrence Erlbaum. Sgall, Petr, Eva Hajicova and Jarmila Panevova. 1986. The Meaning of the Sentence in Its Semantic and Pragmatic Aspects, ed. by Jacob L. Mey. Dordrecht: D. Reidel. Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less numeral quantifier construction in Japanese. In Van Valin (ed.), 285–304. Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics Interface. Düsseldorf: Düsseldorf University Press. Van Hooste, Koen. 2021. A cross-linguistic survey of the instrument-subject alternation. In Van Valin (ed.), 169–188. Van Valin, Robert D., Jr. 1990. Functionalism, anaphora and syntax. (Review article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219. Van Valin, Robert D., Jr. 1994. Extraction restrictions, competing theories and the argument from the poverty of the stimulus. In Susan Lima, Roberta L. Corrigan and Gregory K. Iverson (eds.), The Reality of Linguistic Rules, 243–259. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the mechanisms of language acquisition. In Michael Tomasello (ed.), The New Psychology of Language: Cognitive and Functional Approaches to Language Structure, 221–249. Mahwah, NJ: Lawrence Erlbaum [available on RRG website, https://rrg.caset.buffalo.edu/]. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2008. Some remarks on Universal Grammar. In J. ˝zçali¸skan and K. Nakamura Guo, E. Lieven, S. Ervin-Tripp, N. Budwig, S. o (eds.), Crosslinguistic Approaches to the Psychology of Language: Research in the Tradition of Dan Isaac Slobin, 311–320. New York: Psychology Press. Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots and controllers. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 45–68. Mexico: UNAM. Van Valin, Robert D., Jr. 2016. An overview of information structure in three Amazonian languages. In M. M. Jocelyn Fernandez-Vest and Robert D. Van Valin, Jr. (eds.), Information Structuring of Spoken Language from a Crosslinguistic Perspective, 77–92. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics– Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon Tyne: Cambridge Scholars Publishing.

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Information Structure and Argument Linking

Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. von Heusinger, Klaus. 1999. Intonation and Information Structure. Habilitationsschrift, University of Konstanz.

Notes 1 The background refers to the non-focal part of the utterance, namely the two IUs outside of the AFD in Figure 12.2. It may or may not be part of the pragmatic presupposition, although it usually is. 2 Focus–background and topic–comment represent two logically distinct types of givenness: relational givenness in the former and referential givenness in the latter (Gundel 2012). 3 The propositional representations are simplified for the sake of the discussion; in a more technical presentation they would be represented by the logical structures introduced in Chapters 1, 3 and 4. 4 For a pro-less analysis of the Mandarin phenomena discussed in Van Valin and LaPolla (1997) mentioned earlier, see Van Valin (2005: 174–175). 5 Japanese verbs of giving and receiving are very complex with respect to the social relationships and social statuses of the giver and the receiver. None of that is relevant to this example, and therefore the verb of receiving will be represented by an abstract ‘GET’ predicate. 6 See Van Valin (2009) for discussion of two languages (Liangshan Nuosu [Lolo-Burmese] and Barai [Papuan]) in which IS plays an important role in PSA selection, despite the lack of formal voice constructions in them. 7 This construction is often referred to as the ‘instrument subject alternation’. It is not possible in all languages, and in languages which have it, it is highly constrained. See Van Hooste (2018, 2021) for a detailed discussion of this construction. 8 See Nuhn (2021) for an in-depth RRG analysis of ay-inversion. 9 With intransitive verbs of motion, e.g. kan- ‘go’ as in Chelswu-ka san-ey kaness-ta [Ch-nom mountain-loc go-pst-dec] ‘Chelswu went to(wards) the mountain’, assigning the goal argument locative case means that the moving participant may or may not have made it to the goal, as the translation indicates. It is possible, however, to replace the locative with the accusative case, Chelswu-ka san-lul kan-ess-ta, in which case the sentence means that the moving participant did accomplish reaching the goal, i.e. ‘Chelswu went to(*wards) the mountain’. Obviously, this replacement of the locative by the accusative cannot be considered case ‘spreading’, since there is no other accusative RP in the clause. Rather, it is better to speak of ‘case substitution’, which applies equally to this example and to (17b).

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

521

522

ANJA LATROUITE AND ROBERT D. VAN VALIN, JR.

10 Case stacking, also known as Suffixaufnahme, occurs in a variety of languages (see Planck 1995), and the most common variety involves the possessor in a possessive construction bearing both genitive case and the case assigned to the possessed, the head of the construction. These do not have the semantic or IS effects of the Korean phenomena. A language which appears to have case stacking which seems to be at least in part motivated by IS factors is Eastern Khanty (Uralic; Koshkaryova 2000; but see Filchenko 2007 for an alternative view).

https://doi.org/10.1017/9781316418086.015 Published online by Cambridge University Press

Part Three

Topics in RRG: Complex Sentences

https://doi.org/10.1017/9781316418086.016 Published online by Cambridge University Press

https://doi.org/10.1017/9781316418086.016 Published online by Cambridge University Press

13 The Structure and Semantics of Complex Sentences Toshio Ohori 13.1

Introduction

Most, if not all, grammatical theories have some framework for handling complex sentences. The RRG theory of complex sentences, a.k.a. the theory of clause linkage, was first systematically laid out in Foley and Van Valin (1984, henceforth FVV), building on Olson’s work (1981). Since then, significant revisions have been made, which one may find in Van Valin and LaPolla (1997, henceforth VVLP) and in Van Valin (2005). The RRG theory of complex sentences is unique in that, on the one hand, it is based on the layered representation of clause structure, and, on the other hand, it goes beyond the traditional coordination–subordination dichotomy, so that the theory provides a comprehensive framework which captures both typological diversity and various facets of diachronic development. In this chapter, we review the RRG theory of complex sentences drawing upon a wide variety of languages. Due attention will be paid to some of the recent developments both inside and outside the RRG framework. In the remainder of this section, the RRG theory of sentence structure will briefly be reviewed as a backdrop for the later discussion. Section 13.2 will discuss the structural aspects of clause linkage, and Section 13.3 its semantic aspects. Section 13.4 will provide a sample analysis of complex sentences within RRG, focusing on the interaction of linkage types, operators and reference tracking. In Section 13.5 we offer some concluding remarks.

13.1.1 The RRG Theory of Sentence Structure RRG belongs to functional linguistics (cf. chapter 1 of VVLP) insofar as it pays equal attention to both structure and meaning of the sentence. Within this paradigm, it is assumed that syntactic structures are

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

526

TOSHIO OHORI

motivated in the sense that they have semantic/pragmatic correlates, and the way such correlations hold can be formulated within a relatively narrow range of variation. In RRG, one basic unit of grammatical analysis is the clause, which is characterized by a three-layered endocentric structure. The innermost layer, the nucleus, is typically realized by the verbal predicate which encapsulates the skeletal structure of an event. The nucleus is then complemented by obligatory arguments, resulting in the core. Finally, the core plus optional adjuncts (called periphery in RRG) make up the clause. The clause in itself may or may not constitute a unit of pragmatic force such as assertion and question. When it does, it is identified as a sentence. Otherwise, a clause remains just a clause whose pragmatic force is determined in relation to some other clause which does have such a force. Example (1) is an illustration of the structure of a simple clause from German. As for the operator projection (simplified here), which is represented in Figure 13.1, see (10). (1)

German (Indo-European) Man sollte natürlich alle Ghibli Filme gesehen haben. person should naturally all Ghibli films seen have ‘One should have naturally seen all Ghibli films.’

The layered model of the clause is drastically different from the model of clause structure in many other grammatical theories, which posit the phrasal category VP. In RRG, there is no room for this category. This way

Figure 13.1 The layered structure of the clause

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

of looking at the clause structure is shared by Functional Discourse Grammar (for example, Hengeveld and Mackenzie 2008), and is preceded, though remotely, by field-oriented approaches such as tagmemics. In addition, RRG postulates special structural positions outside the core. One is the pre-core slot (PrCS), typically reserved for a constituent bearing a special, marked, informational status, such as the target of content questions (for example, wh-questions in English) and the pre-core focus. Another is the dislocated position on either side of the clause periphery, which used to be called left-detached position (LDP) or right-detached position (RDP), and is now called pre-detached position (PrDP) and post-detached position (PoDP). This gives a piece of presupposed information, often characterized as a sentence topic.

13.2

Structural Types of Clause Linkage

13.2.1 Juncture: The Layer of Linkage The first parameter for the characterization of complex sentences is the layer of the clause on which the linkage takes place. This is called juncture. The second parameter, called nexus, is concerned with the nature of the linkage, namely the type of dependency that holds between the linked units. This will be discussed in 13.2.2. A maximally broad concept of complex sentences would include linking at any of the three layers of clause structure. We will not exclude, therefore, complex predicates such as Japanese nomi-hosu ‘drink-dry’ > ‘drink up; empty’. Such a broad construal of the term is justified on the grounds that by providing a comprehensive framework for classifying a wide variety of constructions, we may be able to characterize in a coherent manner the process of category change from canonical cases of complex sentences to less canonical ones. Hence, it would be more appropriate to adopt the term complex structures instead of complex sentences. In what follows, we use complex structures as a generic term, and limit complex sentences to refer to clause linkage proper. In tree diagrams, linkage markers are shown as LM for shorthand. First, clause juncture, defined as the linkage at the clausal layer, is illustrated by the following examples. In this section, we shall limit ourselves to coordination for simplicity’s sake. Other types of nexus relations on all three layers will be illustrated in 13.2.2. (2)

Maori (Austronesian; Bauer 1993: 125. ‘>’ indicates the possessor–possessee relation ) Me haere atu (raanei) ia ki toou kaainga, me haere aux move away or 3sg to 2sg.gen>sg home aux move mai raanei koe ki konei hither or 2sg to here ‘He could come to your house or you could come here.’

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

527

528

TOSHIO OHORI

(3)

Iraqw (Afro-Asiatic; Mous 1993: 298) i ‘ayto’ó-r weeriis nee aníng a tlax-úum sbj.3 maize-f sell.3sg.m.prs and 1sg obj.f buy-dur.1sg ‘He is selling maize and I am buying it.’

(4)

Sonora Yaqui (Ute-Aztecan; Dedrick and Casad 1999: 363) ′aháa n-ᡠcai ′ó′olaa tú’i-si n¼ám yi′í-pea táa nee aha my-father ancient good-adv I¼them dance-desiderative but I káa-tú′ii-m wók-ek neg-good-pl foot-have ‘Aha, my revered Father, how I wish I could dance to that music, but I just don’t have good legs.’

Languages may vary with respect to the morphological strategy they adopt for marking coordination. While all coordination markers in (2)–(4) (raanei in (2), nee in (3) and táa in (4)) are words, a clitic or a coordinating particle may also be used. Zero-marking is also not infrequent. In the Maori sentence (2), raanei can be omitted, and in that case the resulting sentence would allow either a conjunctive or a disjunctive interpretation, depending on contextual factors. Generally speaking, the coordinating linkage is of a ‘balanced’ type (Croft 2001) and neither of the linked units is formally reduced or de-ranked. Admittedly, coordinate constructions may have an asymmetric interpretation such as temporal sequence and cause–result, but this is usually taken to be an outcome of inference based on frame-semantic knowledge and contextual information. Schematically, clause coordination is represented in Figure 13.2. Here, there is no obligatory sharing of an argument, and the linked clauses may describe separate events, each with its own Actor-Undergoer assignment. Next, core juncture is the linking of cores, that is, the predicate plus obligatory arguments, and it involves partial sharing of argument

Figure 13.2 Clausal coordination in Maori

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

structure. The shared argument acts as something like a ‘hub’ for joining two units. (5)

French (Indo-European) Je laisserai Jean manger I will.let John eat ‘I’ll let Jean eat the cakes.’

les the

gâteaux. cakes

(6)

Thai (Tai-Kadai; Diller 2006: 172) bo’risa:t1 nam sin4kha:3 khaw3 company lead product enter ‘The company imports products.’

(7)

Turkish (Turkic; Watters 1993: 550) Odam-da yat-mı¸s uyu-ya-mı-yor-du-m. room-loc lie-aux sleep-aux-neg-prog-pst-1sg ‘I was lying in my room unable to go to sleep.’

Core juncture consists of mainly two subtypes depending on the mode of argument linking. Examples (5)–(6) represent the first type, where the shared argument in each sentence (Jean in (5) and in sin4kha:3 in (6)) serves two different functions, namely Undergoer and Actor, in the linked cores. Example (7) represents another type, where the shared argument plays the role of Actor in both of the linked cores. In those theories that posit VP, (7) would be analysed as VP conjoining. The schematic representation of the structure of (5) is shown in Figure 13.3. While the macrorole of the shared core argument is different between (5)–(6) and (7), the fundamental structure is the same, and syntax– semantics mapping is handled by the independent linking algorithm in combination with the logical structure of the predicate. Besides the sharing of an argument, core juncture (in the present case core coordination)

Figure 13.3 Core coordination in French

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

529

530

TOSHIO OHORI

involves obligatory sharing of a periphery, which is a natural consequence of the notion of periphery as an element external to the core. Third, nuclear juncture is formed on the nucleus layer. Here too, a relatively simple example of nucleus coordination is given in (8). (8)

Tukang Besi (Austronesian; Donohue 1999: 182) No-helo’a-ako te ana-no te 3realis-cook-do.for core child-3poss core ‘She cooked the vegetables for her children.’

roukau vegetables

In this example, the linked nuclei, helo’a ako ‘cook-do.for’ act like a simple, unitary predicate, and all arguments are shared. The nucleus-level coordination may be represented as in Figure 13.4. The formation of a unitary predicate is a prominent characteristic of the nucleus juncture. Compare (8) with (9). (9)

Tukang Besi (Donohue 1999: 182) No-helo’a te roukau ako te 3realis-cook core vegetables do.for core ‘She cooked the vegetables for her children’

ana-no child-3poss

While the same predicates occur in (8) and (9), the linkage structures are different. In the former, two core arguments (marked by the non-nominative core article te) follow the unitary predicate which acts like a ditransitive verb. In (9), which is an instance of core juncture, each predicate takes its own Undergoer core argument while sharing the Actor. In the linguistic literature, the term serial verb construction (SVC) is often used inconsistently, but the RRG typology of complex structures enables us to distinguish explicitly between the linkage at the core level and that at the nucleus level. In addition to the layered structure of the clause, RRG has other levels of representation, namely the operator projection, information structure,

Figure 13.4 Nuclear coordination in Tukang Besi

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

and logical structure. These systems interact with clause linkage in various ways. Here, we shall look at the projection of operators in complex structures. According to Van Valin (2005: 9), the following operators can be identified. (10)

Nucleus: Aspect Negation Directionals (only those modifying orientation of action or event without reference to participants) Core: Directionals (only those expressing the orientation or motion of one participant with reference to another participant or to the speaker) Event quantification Modality (root modals) Internal negation Clause: Status (epistemic modals, external negation) Tense Evidentials Illocutionary force

When they occur in a coordinate linkage, operators that belong to the layer higher than the layer where the linkage occurs necessarily take both linked units within their scope. For example, clausal operators such as tense are always shared by the linked cores (and, naturally, by the linked nuclei, as can be seen in (8) and (9)). Core operators, for example root modality, are always shared by the linked nuclei. An illustration of this interaction in the Turkish example (7) is given in Figure 13.5. In linkages other than coordination, different situations hold, which we will discuss in the next section.

13.2.2 Nexus: The Type of Dependency Traditionally, only two types of structural relationship have been identified in order to characterize the linking of predicative units, namely coordination and subordination. Coordination involves no dependency; two units are juxtaposed with or without an overt marker for the coordinate relation. Subordination involves embedding: a structural unit (clause, core or nucleus, in our terms) serves as a constituent dominated by some higher unit (we shall elaborate on this definition below). Let us start with the linking of clauses. We have already seen instances of clausal coordination (2)–(4), where there is no structural dependency. Also, all operators save the outermost one (i.e. illocutionary force) are specified independently for each linked unit, whether they are explicitly marked or not. In contrast, subordination involves embedding, but here one terminological clarification is in order. When we say that a certain sentence exemplifies clausal subordination, the term ‘clause’ is applied to the

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

531

532

TOSHIO OHORI

Figure 13.5 Core coordination in Turkish with operator projection

subordinated unit. For example, a clause may occupy an argument position within a core, as in (11), but we do not call it core subordination. When we refer to an instance of X-to-Y linkage (in this case clause-to-core), we call it by the layer on which X is formed. Hence (11) is an instance of clausal subordination, in which the clause achka chaya-mu-sha-n-ta ‘so many arrived’ occupies a core argument position of the predicate musya-shka-:-chu ‘I did not know’, functioning as a complement clause. (11)

Huallaga Quechua (Quechuan; Weber 1989: 289) Mana musya-shka-:-chu achka chaya-mu-sha-n-ta. not know-prf-1-neg many arrive-afar-nmlz-3-obj ‘I did not know that so many arrived.’

There are two other types of clausal subordination. A clause may occur in an adverbial periphery modifying a clause, as in (12) (i.e. clause-to-periphery or ad-clausal subordination), or it may serve as a constituent modifying a noun, as in (13) (i.e. clause-to-noun subordination) forming a relative clause. (12)

Huallaga Quechua (Weber 1983: 112) noqa eskuyla-chaw ka-yka-sha-:-pita aywa-ku-ra-: Tingo Maria-pa I school-loc be-ipfv-nmlz-1-abl go-refl-pst-1 Tingo Maria-gen ‘From (the circumstance of ) being in school (i.e. after it, from that place, and being tired of being in school) I went to the Tingo Maria area.’

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

(13)

Huallaga Quechua (Weber 1989: 280) Maqa-shu-q runa sha-yka:-mu-n. hit-2obj-nmlz man come-ipfv-afar-3 ‘The man who hit you is coming.’

In Huallaga Quechua, subordination is typically marked by a nominalizer (called a ‘substantivizing’ marker by Weber and glossed as nmlz in the above example), which is -sha in (11)–(12) and -q in (13). In (12), the clause noqa eskuyla-chaw ka-yka-sha-:-pita ‘From (the circumstance of ) being in school’ bears the nominalizer sha, which in turn is followed by the ablative case suffix pita, forming an adverbial periphery. In (13), the clause maqa-shu-q ‘(someone) who hit you’ bears the agentive nominalizer, and it modifies the noun runa ‘man’, forming a larger RP. Figures 13.6–13.8 are structural representations of the said three types of clausal subordination. Subordination can also be formed on core and nuclear layers. Core-to-core subordination is illustrated by (14). (14)

David regretted Amy’s losing the race. (VVLP: 455)

Here, the constituent Amy’s losing the race is embedded in the argument position of another core, David regretted. The subordinated core is in participial form and thus devoid of tense marking, which is a clause-level operator. Next, example (15) illustrates nucleus-to-nucleus subordination, where the second nucleus tέ ‘stay’ adds an aspectual meaning (namely progressive) to the first nucleus g` ará ‘write’. (15)

Khwe (Khoisan; Pavey 2010: 234) a kg` ará-ná X`amá thám̀ ` 3sg.m letter obj write-link ‘He is writing a letter.’

tέ-ὲ-t`e. stay-link-prs

Figure 13.6 Clausal subordination (clause-to-core/complement clause) in Huallaga Quechua

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

533

534

TOSHIO OHORI

Figure 13.7 Clausal subordination (clause-to-periphery/adverbial clause) in Huallaga Quechua

Figure 13.8 Clausal subordination (clause-to-noun/relative clause) in Huallaga Quechua

Since the nucleus does not have an argument position by definition, the subordinate relation is that of adjunction, not embedding. In (15), the second nucleus tέ ‘stay’ can be seen as an operator by itself, given its aspectual meaning. This type of nuclear juncture is dubbed ‘operator construction’ by Hasegawa (1996) for analogous cases in Japanese. The constituent structures of core subordination (core-to-core embedding) and nuclear subordination (nucleus-to-nucleus adjoining) are shown in Figures 13.9 and 13.10 respectively.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

Figure 13.9 Core subordination in English

Figure 13.10 Nuclear subordination in Khwe (Pavey 2010: 235)

Besides coordination and subordination, we may identify yet another type of structure which does not fall under either category. Consider (16), which is also a linking on the clausal layer. (16)

Udihe (Tungustic; Nikolaeva and Tolskaya 2001: 790) Eke bi-lisi-ni ’ana-wa aisi-zeŋe-i time be-cvb-3sg boat-acc mend-fut-1sg ‘If I have time (if there is time), I will mend the boat.’

Here, the nexus type is distinguished from both coordination and subordination by the following properties. First, the first clause, bi-lisi-ni ‘(there) is time’ is devoid of clause-level operator marking, while the second clause aisi-zeŋe-i ‘I will mend the boat’ is marked for tense (namely future). The form which indicates linkage, namely lisi, is a subtype of non-finite ending,

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

535

536

TOSHIO OHORI

that is, it belongs to a conditional converb class. Second, the first clause is not governed by any overt subordinating marker, nor does the linkage involve embedding as in the case of subordination. Linkage like that in (16) often results in a relatively long clause chain, a characteristic shared by coordination but not by subordination. Third, the non-finite verb form indicates the change of subject (notice that the first clause has the syntactic subject eke ‘time’ and the second clause first-person singular), which is not an available morphological option for the finite clause in this language. In order to capture this type of nexus relation, that is, linkage involving operator dependency but not embedding, RRG adopts the term cosubordination, after the work by Olson (1981) on Barai. In addition to these general properties, Roberts (1988) argues, based on evidence from Amele, that cosubordination is differentiated from subordination on the following grounds: (i) a subordinate clause can occur either initially or finally with respect to the main clause, which is not possible for a cosubordinate clause; (ii) the way pronominal reference works is different in cosubordinate and subordinate clauses; and (iii) co-occurrence restrictions on conjunctive particles are different between subordinate and cosubordinate clauses. Admittedly, cosubordination is a highly schematic category and languages may exhibit different concomitant properties within this broad definition. The schematic structure of clausal cosubordination (16) is represented in Figure 13.11. The operator projection is also given in order to visualize clause-level operator dependency. It is true that not all cosubordinate constructions allow clause chaining and switch-reference marking, though they are common among Papuan languages. However, the introduction of the third type of nexus relation enables us to analyse clause linkage phenomena in a more comprehensive and precise manner. Cosubordination on the core layer is exemplified by (17). This is known as a control construction, where the ‘missing subject’ of try is linked to the ‘higher subject’ John. (17)

John must try to wash the car.

This sentence entails that John must wash the car. Hence the second core wash the car is operator-dependent on the first core for deontic modality expressed by must (VVLP: 460). Compare (17) with core coordination, exemplified by (18). (18)

John must tell Bill to wash the car.

From this sentence, it does not follow that Bill must wash the car, that is, the second core wash the car is not operator-dependent on the first core. In other words, the root modality operator must has scope only over the first core. Finally, in nuclear cosubordination, two linked nuclei form a unitary predicate like nuclear coordination. Compare (19) with its core-level

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

Figure 13.11 Clausal cosubordination in Udihe with operator projection

counterpart (5). Unlike in core juncture, where the causee Jean occurs as the ‘hub’ for both linked cores, here this argument is expressed as an oblique constituent marked by the preposition ` a (see also the Tukang Besi examples (8) and (9)). (19)

French Je ferai manger les gateaux ` a Jean. I will.make eat the cakes to Jean ‘I will make Jean eat the cakes.’

It is not always easy to distinguish between coordination and cosubordination at the nuclear level, because there are not many types of operators that apply to this layer. One candidate, however, is aspect. The Japanese example (20) illustrates the point. (20)

Japanese (isolate) Keezai-ga sakamiti-o korogari-oti-tei-ru. economy-nom downhill-acc roll-fall-prog-prs ‘The economy is plummeting (lit. roll-falling the downhill).’

In this example, two nuclei, namely korogari- ‘roll’ and oti- ‘fall’, are linked. The aspect marker -tei- occurs next to the second unit, but it modifies both nuclei. (20) entails that the economy is (metaphorically) rolling-and-falling

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

537

538

TOSHIO OHORI

Figure 13.12 Core cosubordination in English

Figure 13.13 Nuclear cosubordination in French

downhill, and in this sense the first nucleus is operator-dependent on the second nucleus. The structural representations for core cosubordination and nuclear cosubordination are given in Figures 13.12 and 13.13. The three nexus types which we have identified so far are characterized by the distinctive feature matrix specifying the kind of dependency involved in the linkage. (21)

coordination subordination cosubordination

[distributional, operator] [þdistributional, operator] [þdistributional, þoperator]

Coordination does not involve either kind of dependency. When two predicative constituents are coordinated, either can stand by itself. Subordination involves distributional dependency only; a subordinate constituent

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

can take its own operator that applies to the layer of linkage. Cosubordination involves both types of dependency, and, in this sense, it can be seen as the most tightly integrated nexus relation. It is generally known that subordination can be divided into two types, namely structures that involve embedding in an argument position and those that do not (i.e. adjunction). This matrix can therefore be further elaborated by introducing the feature [argument] to capture these two subtypes of subordination. Since it is a reasonable assumption that the argument position is more tightly integrated into the clause structure than the adjunct position, we may arrange the two subtypes of subordination in the following way: (22)

coordination subordination cosubordination

[distributional, [þdistributional, [þdistributional, [þdistributional,

operator] operator, argument] operator, þargument] þoperator]

A general summary of linkage types will be given in Section 13.3.

13.2.3 Elaborations on the Basic Matrix Since the typology of clause linkage was initially introduced in RRG, some important revisions have been made. We shall review two of them, mainly drawing upon Van Valin (2005) and Pavey (2010). The first revision concerns juncture on the sentence level, that is, the combining of sentences having independent illocutionary force. RRG draws a clear distinction between clauses and sentences, so this issue is more than a terminological matter. In this domain, only coordination and subordination exist. Sentence cosubordination is non-existent, because the defining feature of this type of nexus is operator dependency, and it is simply contradictory for a sentence, defined as a unit with full specification of operators, to have operator dependence. Sentence coordination is exemplified by (23) from English, where two sentences, one an imperative and the other a declarative, are combined. In English orthography, this is a single ‘sentence’, but from a theoretical point of view, it is a mini-discourse consisting of two sentences (note that (23) can be rewritten as Be quiet! Or I’ll call the police without relevant change in meaning). The structure of (23) is illustrated in Figure 13.14. One could say that be quiet is not really an imperative but a conditional protasis, meaning ‘if you do not become quiet’, but what matters in this analysis is the scope of illocutionary force, whereas the conditional meaning is derived from the entire construction, which should be specified separately. (23)

Be quiet, or I’ll call the police.

As for sentence subordination, there are multiple possibilities. Parenthetical clauses are one candidate (another type of sentence subordination will be introduced below when we discuss the information structure of the

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

539

540

TOSHIO OHORI

Figure 13.14 Sentential coordination in English

Figure 13.15 Sentential subordination (parenthetical) in English (Kaltenböck et al. 2011: 849, orig. from Burton-Roberts 2005: 180)

sentence). (24) is an instance of sentence-to-sentence subordination and a structural representation is also provided in Figure 13.15. (24)

The main point – why not have a seat? – is outlined in the middle paragraph.

The second recent revision in the typology of clause linkage concerns the elaboration of the layered structure of the clause (see Bickel 1993; Van Valin 2005). In addition to the initial formulation by FVV, RRG has accommodated the left-detached position (LDP), now called the pre-detached position (PrDP), and the pre-core slot (PrCS) in order to capture the information structure of the sentence. Let us see how they interact with clause linkage in turn. Generally speaking, the PrDP is reserved for providing a background assumption, and is separated from the main clause by a prosodic break.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

A sentence is allowed to have only one PrDP, so its presence counts as a criterion for sentencehood. The PrDP position is often occupied by a PP expressing a spatio-temporal setting, but it can also host a predicative unit, as in the French example (25). (25)

French (Van Valin 2005: 192) Moi, quand j’étais jeune, on parlait seulement en français. I when I was young one spoke only in French ‘Me, when I was young, we only spoke in French.’

Here, the subordinate unit, moi, quand j’étais jeune ‘me, when I was young’, has ‘its own left-detached expression, which makes it a sentence, and this sentence is then in the left-detached position of the matrix sentence’ (Van Valin 2005: 192). In this sense, (25) illustrates another type of sentence subordination, namely sentence-to-PrDP subordination. Next, the PrCS is distinguished from the PrDP in that it is a subpart of the clause which is reserved for a marked or narrow focus and is not followed by a prosodic break. The PrCS is typically occupied by an RP, but a clause may also occur in this position (the sentence cannot occur in the PrCS by definition). Unlike in the case of sentential juncture, a clause which occurs in the PrCS is devoid of illocutionary force. The contrast between sentence-to-PrDP subordination and clause-to-PrCS subordination may be illustrated by the following pair of sentences from Japanese. (26)

Japanese Sainoo-ga ar-u-nda-kara ganbar-e. talent-nom have-prs-cop-because work.hard-imp ‘Because/since (you) have a talent, work hard.’

(27)

Japanese Sainoo-ga ar-u-kara koso seekoo-si-tei-ru. talent-nom have-prs-because foc success-do-res-prs ‘Precisely because/??since (X) has a talent (s/he) has made a success.’

In (26), the first predicative unit ends with the assertive predication marker nda. The kara-marked unit expressing reason gives a presupposed piece of information, and though the main clause is in the imperative form, this presupposition remains unchallenged. The fact that the scope of the imperative is limited to the second predicative unit supports the analysis of (26) as an instance of sentence-to-PrDP subordination. In contrast, in (27) the linkage marker kara is followed by the focus marker koso, indicating that sainooga ar-u ‘(X) has a talent’ occupies the PrCS, and hence (27) is an instance of clause-to-PrCS subordination. Without the focus marker koso, a default analysis of this sentence would be clause-to-periphery (¼ad-clausal) subordination. As expected, inserting the assertive predication marker -nda before kara in (27) results in a highly unnatural sentence because of the clash in information structure, -nda strongly preferring to occur in the PrDP and koso in the PrCS.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

541

542

TOSHIO OHORI

Figure 13.16 Sentential subordination (PrDP) in Japanese

Figure 13.17 Clausal subordination (PrCS) in Japanese

(28)

??Sainoo-ga ar-u-nda-kara koso seekoo-si-tei-ru. talent-nom have-prs-cop-because foc success-do-res-prs

The structures of (26) and (27) would look like Figures 13.16 and 13.17. So far, we have identified six types of subordination involving either a clause or a sentence, namely clause-to-core (11), clause-to-periphery (¼adclausal) (12), clause-to-noun (13), sentence-to-sentence (24), sentence-to-PrDP (25)–(26), and clause-to-PrCS (27). In addition, Van Valin (2005) introduces another kind of clause-to-periphery subordination, following the revision of RRG to admit a periphery modifying the core, as in the following example (Van Valin 2005: 194).

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

Figure 13.18 Clausal subordination (clause-to-core periphery/ad-core) in English

(29)

Kim saw Pat after she arrived at the party.

Here the clause after she arrived at the party is analysed as modifying the core (as shown in Figure 3.18). A semantic motivation for analysing this example in this way is that the specification of time is more relevant to the propositional content of the clause compared to the semantic relations expressed by markers such as although, because, and if, all of which have more epistemic orientations. To keep this type of clause-to-periphery subordination distinct from the structure like (12), the term ad-core subordination is introduced. Examining to what extent this subdivision of clause-to-periphery subordination is viable in the light of typological data will be a topic for future investigation. Besides the introduction of sentence-level junctures and the addition of PrDP and PrCS as hosting positions for linkage, there have been debates on the status of cosubordination. Most notably, Bickel (2010) and Foley (2010) independently advance arguments against the notion of cosubordination. Due to the limitation of space, we will not examine the validity of their arguments here, but see Van Valin (2021) for a counter-argument.

13.3

The Semantics of Clause Linkage and Form–Meaning Correlations

The linkage types we have identified thus far can be listed as follows, in order of stronger to weaker structural integration. For the sake of simplicity, cosubordination is placed above subordination for each juncture type, but the order between them is rather tentative compared to the relative order of other relations.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

543

544

TOSHIO OHORI

(30)

Stronger integration nuclear cosubordination nuclear subordination nuclear coordination core cosubordination core subordination core coordination clause cosubordination clause subordination clause-to-core clause-to-noun clause-to-periphery (ad-core) clause-to-periphery (ad-clausal) clause-to-PrCS clause coordination sentence subordination sentence-to-PrDP sentence-to-sentence sentence coordination Weaker integration

Before we discuss the semantics of complex constructions, we shall introduce one conceptual clarification. The linkage types given in (30) are not grammatical constructions in Construction Grammar terms (Fillmore et al. 1988; Goldberg 1995). Rather, they are language-independent structural schemas abstracted from existing sentence types in a wide variety of languages. In RRG terms, constructions are instead defined with reference to specific instantiations of structural schemas in a particular language and the semantic information associated with them. One of the important insights of RRG, and functional linguistics more broadly, is that there is a systematic correspondence between form and meaning (sometimes discussed under the term iconicity; cf. Givón 1980, 1985; Haiman 1985). In this particular case, the generalization is that the relative strength of unit integration on the structural side correlates with the strength of integration on the semantic side. Then, the question is how we might represent the hierarchical arrangement of semantic relations. Silverstein (1976, 1993) proposed the following list as ‘logical relation of clauses’ (Silverstein 1993: 481). (31)

possessive habitual actor habitual agent relative clause (making definite reference) purposive complement desire complement indirect discourse complement temporal adverbial clause if–then

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences disjunction conjunction clause sequence (sequitur) clause sequence (non-sequitur)

Silverstein’s idea was incorporated into FVV, and since then the list of semantic relations, called interclausal semantic relations, has been updated. Ohori (2001, 2005) proposed that the list of semantic relations should be made more systematic in a way that derives certain relations from other more primitive concepts. He introduced a feature system based on a small number of conceptual distinctions, for example, [þ/ control], [þ/ action coherence], and [þ/ temporal proximity], which capture stages of conceptual elaboration. This idea was partly adopted by Van Valin (2005: 206–207), where the following list of semantic relations is proposed. (32)

Stronger integration Causative [1] Phase Modifying subevents: Manner Motion Position Means Psych-action Purposive Jussive Causative [2] Direct perception Indirect perception Propositional attitude Cognition Indirect discourse Direct discourse Circumstances Reason Conditional Concessive Temporal Simultaneous states of affairs Sequential states of affairs Temporally unordered states of affairs Weaker integration

When the two hierarchies (30) and (32) are aligned, we obtain what is called the Interclausal Relations Hierarchy (IRH). Naturally, languages differ with respect to the types of commonly employed linkage types, and the range of semantic relations that constructions of a given linkage type serve to express also differs from language to language. Hence the typological generalization RRG makes about the form–meaning correspondence in the realm of

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

545

546

TOSHIO OHORI

Figure 13.19 Form–meaning correlation in complex structures

complex construction is: when a construction that belongs to a certain syntactic linkage type is associated with some semantic relation, constructions that belong to linkage types on the higher positions on the hierarchy are associated with semantic relations that are also higher on the hierarchy. That is, the correspondence between the two hierarchies will look like Figure 13.19a, but not like 13.19b. Let us substantiate the claim of the IRH by first looking at the expression of causal relations. Given two states of affairs X and Y, the relation X CAUSE Y may hold in a variety of ways along several conceptual parameters. For example, Causative [1] entails X’s direct manipulation and the accomplishment of the resulting state Y, while Causative [2] and Jussive do not have these properties. The former type of causative relation tends to be encoded either morphologically or by nuclear juncture (if not by lexical verbs), but the latter types of relation tend to be encoded by either core or clause juncture, that is, by periphrastic causatives (cf. Shibatani’s 1972 classical discussion of kill and cause to die). This point is clearly illustrated by the following examples from Cora. (33) and (34) are examples of morphological and periphrastic causatives respectively. (33)

Cora (Uto-Aztecan; Vásquez Soto 2002: 225) í̵ Joel ácipoʔu pú wa-ta-ráʔaraʔi-te. det Joel butterfly sbj.3sg compl-prf-fly-caus ‘As for Joel, he made the butterfly fly.’

(34)

Cora (Vásquez Soto 2002: 232) í̵ Juan ru-yáuh pu wa-taʔáih ti wáka-si det John poss.refl.3sg-son sbj.3sg compl-send sbj.comp.3sg cow-pl wa-náwaʔa-n compl-steal-irr ‘As for John, he sent his son to steal the cattle.’

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

In example (33), the causative is formed by the addition of the causative suffix -te to the verb stem ráʔaraʔi ‘fly’. Vásquez Soto reports that in Cora morphological causatives mostly, if not solely, involve non-agentive causees. Therefore (33) is acceptable only if the causee is taken as an inanimate entity and causation involves direct manipulation. In contrast, in (34) which is an instance of periphrastic causative (more specifically clause-to-core subordination), taʔáih ‘send’ functions as the causative verb and the caused event appears as a subordinate clause marked by the pronominal clitic ti. Here the causee has volitional control over his act, unlike in (33) (in fact, the causativization of agentive intransitive is highly restricted for both types of causatives in Cora, but we do not discuss this issue here). An accompanying semantic difference between (33) and (34) is temporal proximity between cause and result. In (33), Joel’s acting on the butterfly immediately causes its flying, while in (34) Juan’s acting on his son makes the latter’s stealing happen after some temporal interval. Another set of examples illustrating the IRH comes from the expression of temporal relations. In Japanese, the morpheme nagara expresses simultaneity. One of its common uses is the expression of manner, namely modification of subevent, as in (35). (35)

Japanese Kare-wa kutibiru-o yugame-nagara He-top lip-acc twist-while ‘He smiled while twisting his lips.’

warat-ta smile-pst

Structurally, this is an instance of core cosubordination. Hence actor coreference is obligatory in this construction. However, nagara can also be used in clause-level linkage, namely clausal cosubordination, as in (36), with no coreference constraint. (36)

Japanese Keehoo-ga nat-tei-nagara mina okunai-ni i-ta. alarm-nom sound-prog-while everyone indoor-dat stay-pst ‘While the alarm kept ringing, everyone stayed indoors.’

In this sentence, nagara also connects two states of affairs that occur simultaneously. But unlike (35), the nagara-marked clause does not constitute a subevent of the main clause, mina okunai-ni ita ‘everyone stayed indoors’. What is more, (36) allows a concessive meaning besides simultaneity, as one can imagine from the English gloss. Thus the relative strength of unit integration on the structural side (namely core vs. clause cosubordination) correlates with that on the semantic side (namely manner subevent vs. concessive relation). While the discussion in this section has been brief, evidence that will reinforce the IRH can be garnered from a greater variety of languages. More importantly, we may add a further dimension to the IRH by taking into consideration more specific morphosyntactic features. In fact, one of

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

547

548

TOSHIO OHORI

Silverstein (1993)’s original insights was that the logical relation of clauses (31) exhibits correlation with a wide range of grammatical phenomena, namely: (i) probability of normal forms, (ii) probability of nominalization, (iii) degree of formal distinctness from unlinked clause, (iv) markedness of connection, (iv) probability of antipassivization, and (v) suspension of agent hierarchy. Similar threads of discussion have been advanced by, among others, Givón (1980), Haiman and Thompson (1984), and Lehmann (1988). (37) is a summary of morphosyntactic features that correlate with the tightness of interclausal relations, adapted from Ohori (2005). (37)

Stronger integration (

Weaker integration ) Nominals Same subject Different subject Not realized (e.g. ‘equi-deleted’) Realized Constrained case marking Normal case marking Verbals Reduced inflection Elaborate inflection Grammaticalized With full lexical content Voice alternation suspended Voice alternation at liberty Operators Shared Not shared Dependent Not dependent Others No explicit signal Explicit signal Word order fixed Free word order

We may now conceptualize the relative strength of unit integration in complex structures in a triadic form: a set of schematic linkage types as in (30), a set of semantic relations as in (32), and a set of morphosyntactic features as in (37), the last of which being more directly applicable to specific constructions from individual languages.

13.4

How It All Works: Old Japanese ‘Switch-Reference’ and Beyond

In this section, we shall see how the RRG theory of clause linkage serves to enhance our understanding of specific grammatical phenomena, focusing on Old Japanese (hence OJ) ‘switch-reference’ (Ohori 1992, 1994, 2002; see also McAuley 2002). It will be shown that what appears to be a switch-reference system in OJ is in fact epiphenomenal, and the apparent switch-reference function derives from the typology of clause linkage. Put differently, reference tracking is a concomitant feature of the clause-linkage structure, along with other important grammatical features.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

Let us start with Akiba (1977, 1978)’s study, where the uses of conjunctive particles in the texts from tenth-century Japanese are closely examined. Rejecting the idea that these particles should be characterized on semantic grounds (for example, in terms of condition, cause, adversity, etc.), Akiba asserts, ‘they are better characterizable in terms of the switch-reference function’ (1977: 611). In fact, according to Akiba’s calculation based on Taketori Monogatari, in 94 per cent of 536 occurrences of the linkage by te the subject is retained across clauses, while in 92 per cent of 120 occurrences of the linkage by ba there was switching of the subject. The following are representative examples of linkage by te and ba. Because these markers are semantically underspecified, they are glossed simply as link. Schematic representations of the sentence structures are also provided. (38)

Old Japanese (Taketori: 18) ito kasikoku tabakari-te Naniha-ni misokani mo-te-ide-nu very cunningly plan-link Naniha-dat secretly carry-link-come-prf ‘(hei) planned very cunningly, and (hei) came carrying (it) to Naniha secretly’; NOT ‘(hei) planned very cunningly, and (hej) came carrying (it) to Naniha secretly.’

(38′)

[(Si) Adv Vi]-te [(Si) (O) PP Adv Vt]

(39)

Old Japanese (Taketori: 41) orosokanaruyauni ihi-kere-ba kokoronomamani-mo e-seme-zu outright speak-prf-link arbitrarily-particle prefix-force-neg ‘(shei) spoke outright, and (shej) could not arbitrarily force (heri) (to marry)’; NOT ‘(shei) spoke outright, and (shei) could not arbitrarily force (herj) (to marry).’

(39′)

[(Si) Adv Vi]-ba [(Sj) (Oi) Adv Vt]

Note that in both examples, the argument structure of the first linked unit is intransitive and that of the second linked unit is transitive, and all arguments are phonologically unrealized. Hence the only formal clue to differentiate the working of reference tracking in these examples is the choice of conjunctive particle. From these considerations, it appears to make sense to analyse the two conjunctive particles, te and ba, as switch-reference markers, the former coding the same-subject (SS) relation, and the latter the different-subject (DS) relation. However, a closer analysis reveals that te and ba differ in several respects other than reference tracking, which cannot be explained if we assume that switch-reference is the defining property of these particles. The primary evidence for this interpretation comes from the interaction of operator scope and the layer of linkage. In the RRG conception of sentence structure, as we have seen in 13.2.1 (10), each layer of linkage has its own operator specification. What is crucial about te and ba linkages is that they have different possibilities for the occurrence of operators, in this case the modal auxiliary. First, consider the following example involving te-linkage. Here an epistemic auxiliary meru occurs in the final unit of the linkage structure.

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

549

550

TOSHIO OHORI

(40)

Old Japanese (Taketori: 37) Tubame ko uma-mu-to suru toki-ha, o-wo sasage-te swallow child bear-aux-comp do time-top tail-acc raise-link nana tabi meguri-te nan, umi-otosu-meru seven time go.around-link particle bear-drop-AUX ‘A swallow, when (it) bears a child, may raise its tail and go around seven times, and then gives birth (to a child).’

In the following example, the modal auxiliary besi, translated here as ‘would’ with an epistemic meaning, also occurs in the final unit of the linkage structure. (41)

Old Japanese (Taketori: 32) Mosi saihahini kami-no tasuke ara-ba, minami-no umi-ni if fortunately god-gen help be-link south-gen sea-dat huka-re-ohasi-nu-besi blow-pass-go-prf- AUX ‘if fortunately there is God’s help, (you) would be blown and go to the south sea.’

In (40), meru ‘may’ has scope over all the units linked by te, namely o-wo sasage ‘raise (its) tail’, nana tabi meguri ‘go around seven times’ and umi-otosu ‘gives birth’. In contrast, in (41) the auxiliary besi has scope only over the clause where it occurs, namely minami-no umi-ni huka-re-ohasi-nu-besi ‘(you) would be blown to the south sea’. The difference between the two examples can be represented schematically as follows, the underlined part indicating the scope of modal auxiliary: (40′)

(41′)

te-linkage [unit 1]-te [unit 2]

epistemic modality

ba-linkage [unit 1]-ba [unit 2]

epistemic modality

This difference in the working of operators is systematic and consistent in OJ. Given that epistemic modality is characterized as a clause-level operator, a plausible analysis of these linkage constructions is that (38) and (40) involving te-linkage are instances of core juncture, while (39) and (41) involving ba-linkage are instances of clause juncture. That is, the structure [unit 1] -te [unit 2] together make up a single clause modified by the epistemic operator. In contrast, for the structure [unit 1]-ba [unit 2], both linked units are clauses and only the latter is within the scope of the epistemic operator. The implication of this analysis for the issue of reference tracking is clear. Since core juncture involves argument sharing as one of its defining features, the retention of subject (more precisely PSA) in te-linkage is a natural consequence of the typology of clause linkage. The analysis of te-linkage as core juncture is further supported by observations of the behaviour of other operators. In the following example, -mu is

https://doi.org/10.1017/9781316418086.017 Published online by Cambridge University Press

The Structure and Semantics of Complex Sentences

analysed as an auxiliary of volition, and it only modifies the second unit of linkage. A schematic representation of operator scope is also given. (42)

(42′)

Old Japanese (Taketori: 20) oni-noyau-naru mono ide-ki-te korosa-mu-to devil-like-cop thing appear-come-link kill-AUX -comp ‘A devil-like monster came up, and (it) meant to kill (me).’

si-ki do-prf

te-linkage [unit 1]-te [unit 2] simultaneous events > sequential events > unspecified b. Causal hierarchy: physical > verbal > underspecified[non-defeasible] > inferred[defeasible] c. Participant’s mental disposition: Intention > internal/direct experience > mental experience: commitment > mental experience: reasoning > non-mental experience: report d. Necessarily shared participant [NSP]: Yes > No e. Referential control hierarchy: main actor > main undergoer > another main core argument > there is no control relation

The juncture–nexus relations and the semantic interclausal relations form the Interclausal Relations Hierarchy (Figure 14.1). The syntactic combinations are ranked in terms of syntactic tightness. The linkage types at the bottom are combinations of whole clauses constituting sentences (e.g. Tyron talked to Tanisha, and Yolanda chatted with Kareem), where two independent units are temporally unordered. As one goes up the hierarchy, the linked units lose more and more features of an independent clause until they are reduced to a bare nucleus or predicate (e.g. nuclear cosubordination, as in Max seemed tired). That is, linkage types at the bottom Strongest Nuclear co-subordination Nuclear subordination daughter peripheral

Nuclear coordination Core co-subordination Core subordination daughter peripheral

Core coordination Clause co-subordination Clause subordination daughter peripheral

Clause coordination Sentential subordination Sentential coordination Weakest Syntactic relations

Closest Causatives [1] Phase Manner Motion Position Means Psych-action Purposive Jussive Causatives [2] Direct perception Indirect perception Propositional attitude Cognition Indirect discourse Direct discourse Circumstances Reason Conditionals Concessive Simultaneous actions Sequential actions Situation-situation: unspecified Loosest Semantic relations

Figure 14.1 Interclausal Relations Hierarchy (Van Valin 2005: 209)

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

are less tight than linkage types at the top end. In the middle portion, in a non-subordinate core juncture, the two cores obligatorily share one core argument; in a subordinate core juncture, the linked core serves as a syntactic argument of the matrix core. The semantic relations are also organized according to the degree of semantic cohesion between the two units, that is, the degree to which they express facets of a single action or discrete actions or events. There is an iconic principle governing the interaction of the two hierarchies: the closer the semantic relation between two propositions, the stronger the syntactic link joining them. That is, the semantic relations at the top end of the hierarchy should be realized by the linkage categories at the top as well, and the relations at the bottom of the hierarchy should be realized by the linkage categories at the bottom of the syntactic side. Moreover, while there is often more than one syntactic realization of a particular semantic relation, the tightest syntactic linkage realizing it should be tighter than the tightest syntactic linkage realizing looser semantic relations. The syntactic relations and the semantic relations are linked together following a set of mechanisms or linking algorithm.

14.3

The Linking Algorithm

Within RRG, the algorithm system links the syntactic, semantic and pragmatic components in a sentence, and it includes universal as well as language-specific components (Van Valin 2005: 129). The system is bidirectional, that is, it links the semantic representation to the syntactic representation (language production), and the syntactic representation to the semantic representation (language comprehension). The algorithm for linking semantics to syntax for simple sentences is summarized in (8); see also Chapter 1 of this handbook.4 (8)

Linking algorithm: semantics ! syntax (Van Valin 2005: 225–226) 1. Construct the semantic representation of the sentence, based on the logical structure of the predicator. 2. Determine the actor and undergoer assignment following the ActorUndergoer Hierarchy 3. Determine the morphosyntactic coding of the arguments a. Select the privileged syntactic argument (PSA), based on the PSA Selection Hierarchy (a′) and the relevant principles of accessibility (a′′) a′. PSA Selection Hierarchy: Arg of DO > 1st arg of do′ >1st arg of pred′ (x, y) > 2nd arg of pred′ (x, y) > arg of pred′ (x) a′′. Accessibility to PSA principles a. Accusative constructions: highest-ranking direct core argument in (a′) (default) b. Ergative constructions: lowest-ranking direct core argument in (a′) (default) c. Restrictions on PSA in terms of macrorole status:

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

565

566

LILIÁN GUERRERO

1. Languages in which only macrorole argument can be PSA (e.g. German, Italian, Dyirbal, Jakaltek, Sama, Yaqui) 2. Languages in which non-macrorole arguments can be PSA (e.g. Icelandic, Georgina, Japanese, Korean, Kinyarwanda) d. Restrictions in terms of argument coding: 1. Languages with case-sensitive PSAs (e.g. English, German, Nepali, Maithili) 2. Languages with case-insensitive PSAs (e.g. Belhare, Tibetan) b. Assign the arguments the appropriate case markers and/or adpositions following the case-marking rules and the postpositional rules for the language. c. Assign the agreement marking to the main or auxiliary verb, as appropriate. 4. Select the syntactic template(s) for the sentences: a. Syntactic template selection principle: The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument position in the semantic representation of the core. b. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre-/post-core slot reduces the number of core slots by 1 (may override b1). 5. Assign arguments to positions in the syntactic representation of the sentence. a. Assign the [WH] argument(s) to the appropriate position(s) in the clause. b. If there is a [þWH] argument of a logical structure: 1. assign it to the normal position of a non-wh-argument with the same function, or 2. assign it to the pre-core slot or post-core slot, or 3. assign it to a position within the potential focus domain of the clause (default ¼ the unmarked focus position). c. A non-wh-argument may be assigned to the pre-core or post-core slot, subject to focus structure restrictions (optional). d. Assign the [WH] argument(s) of the logical structure(s) other than that of the predicator in the nucleus to 1. a periphery (default) a. if the representation is pred′ (RP/LS, LSmain ) where pred′ is an adpositional predicate, then assign the P þ RP/Core/Clausal to the peripherycore . b. if the representation is LSmain pred′/connective′ RP/LS, then assign the P þ RP or CLM þ Clause to the peripheryclause . 2. the pre-core or post-core slot, or 3. the left- or right-detached position.

The algorithm for linking syntax-to-semantics for simple sentences is presented in (9). (9)

Linking algorithm: syntax ! semantics (Van Valin 2005: 226–227) 1. Determine the macrorole(s) and the other core argument(s) in the clause based on the logical structure (LS) of the predicator.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

a. If the verb is intransitive, then assign the PSA either to the macrorole or direct core argument status, depending upon the language (language-specific) b. If the verb is transitive and the language lacks voice oppositions, determine the macroroles from case marking and/or word order (language-specific) c. If the language has a voice opposition, determine the voice of a transitive verb (language-specific) 1. If the construction is syntactically accusative: a. if it is the unmarked voice, the PSA is the actor b. if it is passive, the PSA is not the actor of the predicate in the nucleus 1. the actor may appear as a direct core argument (languagespecific); or 2. the actor may appear in the peripherycore marked by an adposition or oblique case (language-specific); or 3. if there is no actor in the core or the periphery, then replace the variable representing the highest ranking argument in the LS with ‘Ø’. 2. If the construction is syntactically ergative: a. if it is the unmarked voice, the privileged syntactic argument is undergoer. b. if it is antipassive, the privileged syntactic argument is actor: 1. the undergoer may appear as a direct core argument or as an oblique element (language-specific); 2. if there is no undergoer in the core or the peripherycore , then replace the variable representing the lowest ranking argument in the LS with ‘Ø’. 3. Assign macrorole status to the other direct core argument, if it is not dative or in an oblique case (language-specific). d. If the language is head-marking and there are independent RPs in the clause, associate each RP with a bound argument marker (languagespecific). 2. Retrieve from the lexicon the LS of the predicate in the nucleus of the clause and execute step 2 from (8) with respect to it, subject to the following provisos: a. If the language allows variable undergoer selection and if there is more than one choice for undergoer, do not assign undergoer to an argument in the LS. b. Determine the linking of the non-macrorole core argument: 1. If there is a two-place state predicate in the LS and if the nonmacrorole core argument is marked by a locative adposition or dative or a locative-type case, then link it with the first argument position in the state predicate in the LS and link the other non-actor core argument (if there is one) to the second argument position in the state predicate, or 2. If there is a two-place state predicate in the LS and if the non-macrorole core argument is not marked by a locative adposition or dative or a locative-type case, then link it with the second argument position in the state predicate and link the other non-actor core argument (if there is one) to the first argument position in the state predicate.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

567

568

LILIÁN GUERRERO

3. Otherwise, link the animate RP with the first argument position in the state predicate in the LS. 3. Link the arguments determined in step 1 with the arguments determined in step 2 until all core arguments are linked. 4. If there is a predicative adpositional adjunct, then retrieve its LS from the lexicon, insert the LS of the core as the second argument in the LS and the object of the adposition as the first argument. 5. If there is an element in the pre- or post-core slot (language-specific): a. Assign it the remaining unlinked argument position in the semantic representation of the sentence. b. And if there are no unlinked argument positions in the sentence, then treat the wh-word like a predicative preposition and follow the procedure in step 4, linking the wh-word to the first argument position in the LS.

Most complex sentences follow the linking system developed for simple clauses. For instance, the units involving clausal junctures behave largely like independent clauses. Consider the following English constructions (Van Valin 2005: 228–229). (10)

a. Dana jogged through the park, and Kim waved to him b. Kim worked on the assignment in the morning and __i/*j will finish it in the afternoon

In the conjunction construction in (10a), there are two unordered states of affairs, each with its own argument structure. Each unit is linked independently of the other, as if each unit were a simple sentence on its own, and it follows these steps from the semantics-to-syntax linking in (8): construct the semantic representation (step 1), assign the actor and the undergoer (step 2), and determine the morphosyntactic coding of the arguments for each unit (step 3). After selecting the syntactic templates for each unit (step 4), select a sentence template that involves two coordinated units; the conjunction is attached to the second clausal template. Finally, assign arguments to positions in the syntactic representation of the sentence for each unit; the peripheral elements are linked last, after all of the core-internal arguments are dealt with (step 5). This is a typical case of clausal coordination since the two units are independent of each other in terms of TAM values and illocutionary force (IF). The construction in (10b) is structurally very similar to (10a), except that there is a missing argument in the second unit (co-indexed with the main actor Kim) and the IF must be shared across all conjuncts. A sentence like *Did Kim work on the assignment in the morning and will finish it in the afternoon? is ruled out because only the first conjunct is questioned. This yields clausal cosubordination.5 In English, only highly topical elements can receive zero coding, and the PSAs in (10b), which are pragmatic controllers and pivots, are topics (Van Valin 2015: 229). As in (10a), the clauses in (10b) link separately

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

of each other, but the construction as a whole imposes a coreferential constraint when linking the non-initial conjunct. In Yaqui, coreferential PSAs in conjunction constructions marked by into ‘and’ (11a) can be omitted or be expressed by a full (inepo ‘I’) or reduced nominative pronoun (ne ‘I’) in the second component. However, topical elements usually receive zero marking. In the fragment in (11b), the controller ume chapayecam ‘the Chapayecas’ is an argument of the initial clause. There is a missing argument in the five non-initial clauses co-indexed to the pragmatic controller. Note that there is no CLM linking the non-initial clauses, and only the last verb takes TAM operators. Then, conjunction constructions in Yaqui yield clausal cosubordination, because the linked units show operator dependency as well as argument dependency. (11) Yaqui a. Inepo teopo-u bwite-k into (ne) Peo-ta bicha-k 1sg.nom church-dir run.sg.pfv clm 1sg.nom Peter-acc see-pfv ‘I ran to the church, and then (I) saw Peter.’ _i maska-ta emo u’ura, b. [U-me chapayeca-mi yi’i-su-k-o], the-pl chapayeca-pl dance-cmpl-pfv-clm mask-acc self take.off emo koba-t _i teopo bicha-po _i im am¼jajawa, church towards-loc here 3pl.acc¼red.leave self head-loc payum-mea papatta, juya-u-bicha _i sasaka-k cloth-ins.pl red.cover hill-dir-towards red.go.pl-pfv ‘After the Chapayecasi danced, they took off the masks, _i/*j put them in front of the church, _i/*j covered their heads with cloths, and _i/*j went towards the hill.’

The constructional schema in Table 14.1, adapted from the constructional schema for English ‘conjunction reduction’ (Van Valin 2005: 231), captures the essential features of ‘topic-chain’ clauses in Yaqui and some other languages. Note that the semantic representation of each noninitial clause contains a lexically unfilled, obligatory co-indexed argument. Thus, there is a special linking requirement that involves highly topical PSAs as pivots and obligatory coreference with the initial pragmatic controller. The next section examines how the linking algorithm works for adverbial clauses. I show that some adverbial relations can be expressed by adsubordination, but others are coded by non-subordinated linkages. RRG assumes that the semantic relations of the adverbial subordinate clause to the unit it modifies are the same as those of a peripheral PP modifying the core or clause. For example, since a PP like after the party expresses the temporal setting of the event expressed in the core, it then follows that a clause like after she arrived at the party also occupies the peripherycore . Reason clauses, such as Chris was angry because of Pat’s insults, provide the reason, cause or condition of the main clause as a whole; as a result, they occupy the

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

569

570

LILIÁN GUERRERO

Table 14.1 Constructional schema for Yaqui ‘topic chains’ C ONSTRUCTION : Reduction constructions in Yaqui S YNTAX : Juncture: clausal Nexus: cosubordination Construction type: juxtaposition [C L A U S E [ C O R E RPi [N U C . . .]. . .]. . .]1, [C L A U S E [ C O R E -i [N U C . . .]. . .]. . .]2, [C L A U S E [ C O R E -i [N U C . . .]. . .]. . .]n Unit template(s): step 4 in (8) PSA: clause 1: variable syntactic controller ¼ pragmatic controller clause 1 þ n: variable syntactic pivot ¼ pragmatic pivot Linking: controlled argument in 1 þ n ¼ pragmatic pivot (steps 3a, c2) M ORPHOLOGY : CLM: none S EMANTICS : Sequence of events sharing a common primary topical participant P RAGMATICS : Illocutionary force: shared across all conjuncts Focus structure: predicate focus in all conjuncts

peripheryclause . Hence, ad-subordination mostly follows the basic linking algorithm in (8) and (9); the real challenge comes from non-subordinate core junctures with their obligatory sharing of a core argument (Van Valin 2005: 225).

14.4

The Linking Algorithm in Adverbial (Adjoined) Clauses

In the analysis that follows, adverbial relations are organized into five groups: concessive and conditional clauses (14.4.1), reason clauses (14.4.2), temporal clauses (14.4.3), purpose clauses (14.4.4), and manner and means clauses (14.4.5). Adverbial relations may be established at different levels of the clause structure; the relationships at the bottom of the semantic hierarchy usually modify the clause, temporal and purpose clauses may modify the core, and manner and means can usually modify the nucleus.

14.4.1 Concessive and Conditional Clauses Concessive and conditional are the loosest semantic relations in terms of the semantic hierarchies presented in (7). Concessive clauses express a relationship of general incompatibility and counter-expectation between two situations; in (12a), the content of the main clause holds unexpectedly, given the content of the linked clause (Van Valin 2005: 207). In conditional clauses, the protasis (the if-clause) describes a condition for the realization of the situation expressed in the apodosis (the main clause); it can refer to a present, past or future situation (12b), to an imagined situation that might

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

hypothetically happen, as in If he gets the job, we’ll all celebrate, or that did not happen, as in If you had been at the party, you would have seen her new boyfriend (Hetterle 2015: 49). In the English constructions in (12), each clause links separately, but the complete construction encodes two semantically associated states of affairs. Each association is formally captured in the (simplified) logical structure of the construction by different connectors or CLMs with lexical content: IN . SPITE . OF ′ for concessive (12a) and  for conditional (12b). English concessive and conditional clauses (Van Valin 2005: 207) a. Bill made it to work, even though it was snowing heavily. a′. [Bill made it to work] IN . SPITE . OF ′ [it was snowing heavily] b. If it rains, we won’t be able to have a picnic. b′. [If it rains]  [we won’t be able to have a picnic]

(12)

These adverbial constructions do not entail operator dependency; they can even have different IF values, as in if it rains, are we still having a picnic? They do not demand argument sharing either. When the two PSAs are coreferential, like in Were Fred to leave now, he would look like a fool, the linked unit overtly expresses all its core arguments. The ordering distribution of main and adverbial units may also vary (e.g. Even though it was snowing heavily, Bill made it to work for (12a)). These key properties of the constructions follow from there being ad-subordination at the clause level. Because of their semantic association to the main unit, conditional and concessive clauses occur in the peripheryclause . For the purposes of linking, ad-clausal subordination is captured by step 5d1b in the semantics-to-syntax algorithms in (8). Thus, the main clause and the linked clause are linked independently of each other, just as if each were a simple sentence on its own. When assigning arguments (and clauses) to positions in the syntactic representation of the sentence, step 5d1b assigns the relevant CLM þ clause to the peripheryclause . A simplified syntactic representation of ad-clausal subordination in English is presented in Figure 14.2.

SENTENCE PERIPHERY

CLAUSE CORE

CLM

CLAUSE CORE

‘If

it rains,

we won’t be able to have a picnic’ in (12b)

Figure 14.2 Conditional ad-clausal subordination in English

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

571

572

LILIÁN GUERRERO

14.4.2 Reason Clauses Reason clauses express a cause or reason that brings about the situation expressed in the main clause. The logical structure for the reason clause in (13a) includes the BECAUSE ′ connector. (13)

English reason clauses (Van Valin 2005: 207) a. Kim berated Pat because she kissed Chris. a′. [Kim berated Pat] BECAUSE ′ [she kissed Chris]

Some languages have several types of reason clauses. For instance, Cavineña distinguishes cause clauses from reason clauses (Guillaume 2008). The ¼ra clause in (14a) expresses the cause of the main event, whereas the ¼tibu clause in (14b) gives the reason for the occurrence of the main event. Unlike cause clauses, the linked verb in reason clauses can take the full range of verbal morphology, and there are no coreference restrictions. (14)

Cavineña (Pano-Tacanan; Guillaume 2008: 719, 731) a. E-tsaka uje-da [ju-ya aje¼ra] npf-leg painful-asf be-impfv walk¼cause ‘My legs hurt from walking (so much).’ b. Aama! [mi-ra¼dya iye-wa¼tibu] duju-kwe not.exist 2sg-erg¼foc kill-pfv¼reason take-imp.sg ‘No! Since you killed it (a caiman), you take it (not us)!’

In the interclausal semantic relations proposed by Van Valin (2005: 206–207), reason shows a closer semantic association when compared to conditional and concessive because of its causal meaning (i.e. the causal hierarchy in (7b)). However, languages may vary the syntactic linkages used to express these relationships. For instance, in Cristofaro’s study (2003: 229), reason outranks condition in the lack of tense, aspect and mood distinction, as well as the use of case marking/adpositions on the linked verb, but the two are ranked together in terms of the coding of participants. Hetterle (2015: 143) comments that in Chinese, preposed y¯ınw`ei cause clauses and rúgu˘o conditional clauses are under the scope of main clause interrogation, while su¯ırán concession clauses and j`ırán reason clauses are not; this suggests that cause and conditional clauses show a tighter syntactic linkage when compared to concessive and reason clauses.6 Diessel and Hetterle (2011) also suggest that reason clauses in English, German and Japanese function as independent assertions that are only loosely combined with the associated main clause. Reason clauses, but not conditional clauses, are often intonationally separated from the semantically related clause (even in their canonical position). Therefore, the degree of syntactic integration of the linked unit that expresses concessive, condition and reason may vary across languages. This is the case in Yaqui. In Yaqui, reason and concessive clauses are structurally different from conditional clauses. First, concessive and reason clauses are the only clause types that make use of initial CLMs with lexical meaning: ella’apo

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

573

‘although, even though’ (15a), and kiali’ikun ‘because, so that’ (15b). Second, the description of the event in the adverbial unit is fully independent in terms of TAM, negation and IF. Third, concessive and reason clauses can attract restricted discourse markers that occur on the left edge in independent clauses and can have their own pre- or post-detached phrases (PrDP, PoDP). Finally, the dependent PSA takes nominative case; nominative PSAs are completely banned in syntactically subordinate clauses in Yaqui (Guerrero 2006, 2017). (15) Yaqui concessive and reason clauses a. Bempo Guayma-meu saja-ne [ella’apo empo im tawa-ne] 1pl.nom Guaymas-dir.pl go.pl-pot clm 2sg.nom here stay-pot ‘We will go to Guaymas although you stay here.’ tei tu’isi emo jala’eka-n, b. Inepo into in jaboii¼tua, 1sg.nom and 1sg.gen grandpa¼int 1pl.nom good self friend.be-ipfv [kiali’ikun ne au¼waate into ne a-et ai ¼ baisae-Ø clm 1sg.nom 3sg.obl¼miss-prs and 1sg.nom 3sg-loc 3sg.acc¼thank-prs si’ime-tai ] everything-acc ‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for it, for everything.’ (Buitimea; regreso: 37)

As for conditional clauses, three syntactic structures have been identified so far, all of which take the underspecified CLM -o. In addition to -o, one structure takes the Spanish loanword si ‘if’ (16a), and another takes the special verb marker -tek ‘conditional’ (16b); the last example (16c) does not take any extra marking. The linked verb is usually unmarked for TAM values or is marked by the perfective suffix -k only, hence there are some operator restrictions. Additionally, the dependent PSA empo ‘you’ in (16a–b) takes the form of a nominative pronoun, but in (16c) it occurs in its accusative form, enchi ‘you’. (16) Yaqui conditional clauses mam su’utoja-o] a. [Si empoi clm 2sg.nom arm.pl.acc release-clm che’a juni nee a¼kotta-tua-ne empoi 2sg.nom more even 1sg.acc 3sg.acc¼break-cause-pot (Guerrero; Lalo: 57) ‘If you release the arms, you will make me break it [the elbow] even more.’ b. [Empo teopo-u lionok-bae-tek-o], empo emo jin-tua-ne 2sg.nom church-dir pray-want-cond-clm 2sg.nom self cover-cause-pot ‘If you want to pray in the church, you have to cover yourself.’ c. Inepo enchi ania-ne [enchi junuen ’ea-o] 1sg.nom 2sg.acc help-pot 2sg.acc thus think-clm ‘I will help you if you want (me to).’ (Dedrick and Casad 1999: 393)

Therefore, reason and concessive clauses in Yaqui link together two independent units in terms of an initial CLM, operator and argument coding; in (15b), there are two complete sentences, each with its own detached position: there is a topical element in the PrDP of the first component, as for me and my

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

574

LILIÁN GUERRERO

TEXT SENTENCE PrDP

CLM

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

CLM

CLAUSE CORE

RP

Inepo into in jaboii=tua, tei tu’isi emo jala’eka-n, kiali’ikun ne au=waate into ne a-et ai=baisae-Ø si’ime-tai

‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for it, for everything’ in (15b) Figure 14.3 Simplified structure of reason relations as sentential coordination in Yaqui

grandpa, and there is an anti-topic element in the PoDP of the second unit, everything (sentential coordination). In contrast, conditional clauses show two key features signalling structural dependency: the presence of a final, bound and underspecified CLM, and operator dependency; (16c) also satisfies the need for an accusative dependent PSA (ad-clausal subordination). A simplified syntactic representation of (15b) is offered in Figure 14.3. In order to capture the semantic association of ad-core (14a) and adclausal (12, 13, 14b, 16) adjuncts, Van Valin (2005: 229) revised step 4 in the syntax-to-semantic algorithm in (9). The relevant modifications are in boldface in (17). Step 4b allows us to represent concessive, conditional and reason PPs or clauses in the periphery. Sentential coordination for the Yaqui constructions in (15) does not require any modification to the linking algorithm. (17) Revision of step 4 in the syntax-to-semantics linking algorithm (Van Valin 2005: 229) 4. If there is an adjunct phrase in a periphery, a. and if it is in the peripherycore , then retrieve the logical structure of the predicative adposition from the lexicon, insert the logical structure of the core as the second argument in the logical structure and the object of the adposition as the first argument, b. and if it is in the peripheryC L A U S E , then link the adjunct PP or clause logical structure to the matrix logical structure via the semantic representation of the predicative adposition or CLM.

14.4.3 Temporal Relations Next on the semantic continuum are temporal relations. Temporal expressions provide the time specification for the state of affairs expressed in a sentence. They do so indirectly and in relation to another moment, situation or event (Declerck 1997). Typological studies have shown that temporal clauses tend to show a higher degree of integration with the main clause

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

than conditional and reason clauses. For instance, temporal relations (but not reason relations) involve the predetermination of time and aspect values of the linked state of affairs, and might often include reduced verb forms (Cristofaro 2003: 173). Across languages, temporal relations can be expressed by different syntactic means (adverbs, adpositional phrases, finite and nonfinite clauses), and can appear to the left or to the right of the core they modify (i.e. as modifiers in the periphery of the core). The semantic association between the two units can be captured by CLMs with lexical content. For instance, ‘before’ expresses posterior relations with respect to the main event, ‘after’ introduces anterior relations, and ‘while’ introduces co-temporal events. In Spanish, the prepositions antes (de) ‘before’ and después (de) ‘after’ can take phrases, cores or clauses as objects (Guerrero et al. 2017). In (18a–c), there is a subjectless, non-finite infinitival unit. As mentioned before, infinitival verbs are a type of non-subordinate core that depends on the main verb in terms of TAM operator. However, the infinitival unit in temporal clauses can be independently negated from the main clause, as in Sam se durmió después de no cenar ‘Sam fell asleep after not having dinner’. Commonly, the PSA of the main core (Sam) controls the referential identity of the missing argument (pivot) within the linked core, but in (18c) each core has its own PSA. This means that argument sharing is optional. Hence, since the two cores can be independently negated, and they can but do not necessarily share an argument, (18a–c) involve core coordination. There is an additional sequential construction. In (18d), the linked unit is introduced by the complex CLM después de que ‘after that’ and the two units are independent of each other in terms of TAM operators at the level of the clause and argument coding. Therefore, (18d) takes a modifying clause at the periphery of the core, hence ad-core subordination. In Spanish, the unmarked position for temporal adverbs, phrases and clauses is postverbal, though they may also appear at the beginning, as in antes de acostarse, Sam besó a Pat for (18a). The simplified logical structure for sequential constructions is the same as that of a peripheral PP modifying a core and is valid for the two juncture–nexus. (18)

Spanish temporal clauses a. Sami besó a Patj [_i antes de acostarse] ‘Sam kissed Pat before lying down.’ a′. be-before′ ([INGR lie down′ (xi)], [do′ Sami, [kiss′ (Sami, Pat)]]) b. Sami se durmió [después de__i besar a Pat] ‘Sam fell asleep after kissing Pat.’ b′. be-after′ ([do′ (xi, [kiss′ (xi, Pat)])], [do′ (Sami, [sleep′ (Sami)]]) c. Pat vio la película [después de dormirse Sam] ‘Pat saw the movie after Sam fell asleep.’ c′. be-after′ ([do′ (Sam, [sleep′ (Sam)])], [see′ (Pat, movie)]) d . Pat vio la película [después de que Sam se durmió] ‘Pat saw the movie after Sam fell asleep.’ d′. be-after′ ([do′ (Sam, [sleep′ (Sam)])], [see′ (Pat, movie)])

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

575

576

LILIÁN GUERRERO

SENTENCE CLAUSE CORE

PERIPHERY PP COREp NUCp

Pat vio la película

CLAUSE

P

después de que Sam se durmió ‘Pat saw the movie after Sam fell asleep’ in (18d) Figure 14.4 Sequential temporal relations as ad-core subordination in Spanish

Peripheral subordination at the core level does not require any revision of the linking algorithm, as it is captured by the steps in (5d1a) in the semantics-to-syntax linking (8): the predicative preposition takes either a core or clausal unit and, as a whole, is attached to the periphery of the core. When linking syntax to semantics, step 4a yields the representation for spatial and temporal circumstances, based on the predicative adposition. A simplified syntactic representation for the sequential construction in (18d) is presented in Figure 14.4. This is another example of asymmetrical linkage since a linked clause is contained within a sub-clausal unit. Core coordination junctures do require revision of the linking algorithm, especially when the main actor controls the pivot in the linked unit, as in (18a–b). In order to capture the fact that there is a syntactic argument slot missing in the linked unit, the template selection principles in the semantics-to-syntax linking (8) were revised, and a new principle (step 4b) was added for non-subordinate core junctures. Step (4b1) specifies that the linked unit is a semantic argument of the matrix verb, as in Chris tried to see Pat. However, temporal clauses do not fulfil a semantic slot in the logical structure of the main clause. The nature of Spanish temporal infinitival cores in (18a–b) can be captured by step (4b2). (19)

Modifications to the semantics-to-syntax linking (Van Valin 2005: 262–263) 4. Syntactic template selection principle b. Universal qualification of the principle in (a): The number of slots in a core is reduced by 1 if: 1. it is the matrix core in a non-subordinate core juncture in which the linked core is a semantic argument of the matrix verb, and/or7 2. it is the linked core in a non-subordinate core juncture.

The statement in step 4b does not specify which syntactic slot is missing, since that is a construction-specific feature. The syntax-to-semantic linking needs some modifications too. The relevant adjustments apply after the

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

logical structure of the clause has been retrieved (step 2), and the linking of macrorole and non-macrorole core argument has taken place (step 3). Again, step 4b can capture the fact that there is a syntactic argument missing from the linked core which must be interpreted as being the same as one of the syntactic arguments from other positions. The logical structures for (18a) and (18b) contain a lexically unfilled co-indexed argument x, which is controlled by the PSA of the main clause (Sam). (20)

Modifications to the syntax-to-semantic linking (Van Valin 2005: 280–281): 4. In non-subordinate core junctures, one of the arguments of the matrix core must be linked to an argument position in the embedded logical structure: a. if the matrix predicate is a control verb, this follows the theory of obligatory control: causative and jussive verbs have undergoer control, while all other (M-)transitive verbs have actor control; otherwise, b. if the matrix predicate is not a control verb, then link the unlinked syntactic argument in the matrix core to the logical structure argument position of the pivot of the linked core.

That is, all temporal clauses in (18) semantically modify the core of the main unit, but only the linked unit in (18d) occupies a peripheral position; the non-finite linked units in (18a–c) yield a non-subordinate core juncture. A simplified representation for core coordination expressing sequential relations in Spanish is presented in Figure 14.5. Depending on their role in discourse, temporal clauses can also occupy the pre- or post-detached position, with or without a prominent pause between the two units (Guerrero et al. 2017). In a sentence like Ricardo S fue expulsado en 1906, después de pasar dos años removiendo el caso de la muerte de DG ‘Ricardo S was expelled in 1906, after spending two years investigating the death of DG’, there is a non-finite core, and the missing argument is coindexed with the main actor. In this construction, there are two temporal expressions, the PP ‘in 1906’, and the ad-clause introduced by the preposition después de ‘after’. Because there is another core modifier, I suggest that the after-clause does not modify the core, but is linked to the sentence directly in the post-detached position (PoDP). Notice that there is another SENTENCE CLAUSE CORE

CLM

CORE

Pat vio la película después de besar a Pat ‘Sam fell asleep after kissing Pat’ in (18b) Figure 14.5 Sequential temporal relations as core coordination in Spanish

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

577

578

LILIÁN GUERRERO

SENTENCE CLAUSE CORE

PoDP PERIPHERY

PP

PP

CORE

COREp Ricardo S fue expulsado NUCp

RP

CORE

PRED p

PRED p en

NUCp

1906,

después de pasar dos años removiendo el caso de la muerte de DG

‘Ricardo S was expelled in 1906, after spending two years investigating the death of DG’ Figure 14.6 Simplified structure of sentential subordination in Spanish

temporal PP modifying the core. The representation of this construction is presented in Figure 14.6. According to the linking algorithm in (8), the temporal PP is captured by the steps in 5d1a, while the temporal clause is placed at the PoDP following the steps in 5d3. Not all temporal clauses are introduced by predicative prepositions. General when-clauses are particularly interesting because the CLM does not specify the temporal sequence between the main and the adverbial units. For English, it has been said that when-clauses can express any temporal relation, although they strongly encode temporal overlap ‘events even though the exact extent of the overlapping is subject to variation’ (Cristofaro 2003: 159; Diessel 2008: 474). The same has been said for Spanish constructions introduced by cuando ‘when’ as in (21). In corpus analysis, cuandoclauses turn out to be the most productive temporal clause type (GerardoTavira 2018). The examples in (21a–b) are from Conti (2012), and (21c) is from Gerardo-Tavira (2018). (21)

a. _i llámame [cuando _i llegues] ‘Call me when you arrive.’ b. [Cuando _i llegó a Tepic], _i quedó fascinado con el verdor del paisaje ‘When he arrived at Tepic, he was fascinated by the green landscape.’ c. [Cuando _iþj éramos novios], _i mej llegaba a visitar de leva traslapada ‘When we were dating, he would visit me wearing his finest clothes.’

There is an assumed correlation between the use of a comma (as a correlate of an intonational break) and the extra-clausal position of the temporal adjunct (Van Valin 2005: 192–193). Conti (2012) recognizes this correlation and proposes that the clausal juncture in (21a) occupies the periphery of the clause, while that in (21b–c), with a comma after the linked unit, is outside the clause, and is linked directly to the sentence in the PrDP. Since cuando/ when are not predicative prepositions, but CLMs with lexical content, one

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

may wonder what the logical structure of the constructions in (21) looks like. The Interclausal Semantic Relation Hierarchy (Van Valin 2005: 206–207) recognizes temporal circumstances as the spatial or temporal parameters of an event (e.g. Kim saw Pat after she arrived at the party) and temporally ordered states of affairs. There are simultaneous states of affairs, where one state of affairs is temporally coterminous with another, like in (22a), as well as sequential states of affairs, in which one state of affairs follows another temporally, with or without any temporal overlap, as in (22b). Note that the linkage markers do not have lexical meaning but resemble some sort of conjunction/connector. That is, the semantic association between the units cannot be captured by the CLMs, but instead depends on the pragmatic context. (22)

a. a′. b. b′.

Max was dancing, and at the same time Susan played the piano. [do′ (Max, [dance′ (Max)])] ^ [do′ (Susan, [play′ (Susan, piano)])] Juan finished reading the newspaper, and then Carlos walked into the room. [do′ (Juan, [read′ (Juan, paper)])] & [do′ (Carlos, [walk′ (Carlos) & INGR be-at′ (room, Carlos)])]

14.4.4 Purpose Relations Next in the semantic continuum are purpose relations, in which one action is done with the intent of realizing another state of affairs. Unlike reason, conditional and temporal relations, purposive situations show a strong preference for argument sharing (recall the Kokama examples in (6)). First Cutrer (1993), and then Van Valin (2009), proposed two types of purposive constructions in English based on the lexical manifestation of the controllee in post-nuclear position: ‘pure’ purpose clauses that exhibit an obligatory missing syntactic argument in the linked unit (23), and rationale clauses, in which there is not a missing argument, but a co-indexed pronoun (24). Note that the controllee must be associated with the immediately post-nuclear argument (undergoer) in the infinitival core. The pre-nuclear argument (actor or PSA) may be the same (23a–b, 24a–b) or different (23c, 24c). The data comes from Cutrer (1993). (23)

a. Johni caught a fishj [_i to eat _j for dinner] b. Johni gave a tapej to Maryk [_k to listen to_j] c. Johni built a chestj for his sisterk [_k to put her clothes in_j]

(24)

a. Johni bought Maryj a bookk (in order) [ _i to please herj] b. Johni sang the childrenj a lullaby (in order) [ _i to calm themj] c. John bought the turkeyi (in order) for his wifej [ _j to cook iti]

In Cutrer’s analysis (1993: 177), purpose clauses are a type of core juncture, and therefore, there must be an argument shared between the two cores. It is possible to have two control relations, as in (23b), but only one is obligatory: the postverbal gap must be obligatorily controlled by the main theme.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

579

580

LILIÁN GUERRERO

In opposition to this, rationale clauses are considered a type of clause juncture, and hence there are no obligatorily shared arguments between the two clause units; instead, there is a post-nuclear pronoun when transitive, as in (24). Unlike the linked undergoer, the actor gap has a nonobligatory control relation, since it may be filled or unfilled. In (23a–b) and (24a–b), the main actor controls the identity of the missing agent in the linked unit; in (23c) and (24c) the controller is a non-macrorole oblique core argument. Thus, despite the semantic similarities of the two constructions, their syntactic properties are different. However, the position and the lexical coding of the controllee are subject to language-specific conditions. In order to avoid structural assumptions in particular languages, it is highly recommended to describe the function of the controller and controllee in terms of actor/agent (e.g. pre-nuclear in English), and undergoer/theme, recipient (e.g. post-nuclear in English), and the lexical coding of the controllee in terms of structural and inherent control (Guerrero 2012, 2013). For instance, there are three purposive constructions in Yaqui (Guerrero 2011, 2013, 2017): the first type (25a) lacks CLM; the second type (25b) is marked by the general CLM -kai, and the third type (25c–d) is marked by the postposition -betchi’ibo ‘for/in order to’. The linked core in (25a–b) fully depends on the main core for operator values and cannot be independently negated. In contrast, the linked core in (25b–c) can be unmarked or marked by the potential suffix -po, and it can be negated. Furthermore, all purposive constructions demand argument sharing, but obligatory shared arguments make use of a special lexical coding. Thus, there is structural control (gap/missing syntactic argument) when the controller is the main actor (25a–c), and inherent control (co-indexed pronoun) when the controller is the main undergoer (25d). Since Yaqui is a head-final language, both the controller and the controllee must be pre-nuclear.8 (25) Yaqui purpose clauses a’abo _i ji’i-bwa-se-k a. Peo-Øi Peter-nom here something-eat-move.purp.sg-pfv ‘Peter came here to eat.’ a′. [do′ (Peter, [MOTION′ (Peter)])] ^ [do′ (Peter, [eat′ (Peter, something)])] nei ino piaroa-k b. [Kafe-ta _i ji’i-pea-ka] coffee-acc drink-want-clm 1sg.nom 1.self borrow.money-pfv ‘I borrowed some money in order to drink coffee.’ b′. want′ (I, [do′ (I, [drink′ (I, coffee)])] ^ DO [do′ (I, [borrow′ (I, money)])] e CAUSE [do′ (I, [drink′ (I, coffee)])] juya-u siika [mas-ta _i me’e-betchi’ibo] c. U o’ou-Øi det man-nom hills-dir go.sg.pfv deer-acc kill.sg-clm ‘The man went to the hills in order to kill the deer.’ c′. want′ (man, [do′ (man, Ø) CAUSE [BECOME dead′ (deer)]) ^ DO [do′ (man, [move′ (man)]) & INGR be-at′ (hills, man)]] [u-ka o’ou-taj kaa ai_me’e-ne-betchi’ibo] d. U maso-Øi bwite-k det deer-nom run.sg-pfv det-acc man-acc neg 3sg.acc¼kill.sg-pot-clm ‘The deer ran quickly in order for the man not to kill it.’

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

d′. [do′ (deer, [run′ (deer)] PURP [do′ (man, Ø) CAUSE [NOT BECOME dead′ (deer)]]] yoi-taj kaba’i-mkreuwa-k [_j amk¼wiria-ne-betchi’ibo] e. Min-Øi Fermín-nom foreigner-acc horse-pl lend-pfv 3pl.acc¼feed-pot-clm ‘Fermín lent the foreigner the horses in order for him to feed them.’ e′. [do′ (Fermín, [lent′ (Fermín, horses)] CAUSE [BECOME have′ (foreigner, horses)]] PURP [do′ (foreigner, [feed′ (foreigner, horses)])]

Notice that Yaqui does not distinguish between purpose (25a–d) and rationale clauses (25e) based on an obligatory syntactic gap, since a referential controlled undergoer must be overt. In the last example, the two cores share both the theme and the recipient, which are lexically expressed as core arguments of the main unit; in the linked unit, the recipient yoi ‘foreigner’ controls the actor PSA of the linked verb ‘feed’, hence there is structural control (missing syntactic argument), while kaba’im ‘horses’ controls the undergoer of ‘feed’, and so there is inherent control (co-index pronoun). Since there are unfilled/filled syntactic arguments in the linked core which need to be obligatorily controlled by a core argument in the main unit, Yaqui purpose clauses behave as a kind of control construction (Guerrero 2017).9 As a matter of fact, the special lexical coding of the controllee is exactly the same in purpose clauses and constructions taking actor and undergoer control verbs. In Yaqui, actor control verbs such as modal, phasal and desiderative constructions demand a cover controllee in the linked unit (26a), whereas undergoer control verbs like causative and jussive constructions demand an overt controllee, as depicted in (26b). In purpose, causative, and jussive constructions, a missing argument coindexed with the main theme would be ungrammatical. (26) Yaqui inherent actor and undergoer control verbs kaa _i yi’i-pea-Ø. a. Lupe-Øi Lupe-nom neg dance-int-prs ‘Lupe does not want to dance.’ Goyo-taj sawe-k [tekil-ta aj¼tekipanoa-ne-’u] b. Empoi 2sg.nom Goyo-acc order-pfv land-acc 3sg.acc¼work-pot-clm ‘You ordered Goyo to work the land.’

Comanche, another Uto-Aztecan language, also shows noteworthy controller–controllee patterns. According to McDaniels (2014), in rationale clauses either an object gap (27a) or a subject gap may occur (27b), or neither (27c), but not both. Thus, the lexical manifestation of the controllee varies in rationale clauses (it can be overt or covert), and this variation does not depend on its function inside the linked unit. Purpose clauses show the same controller–controllee patterns, but the linked unit is marked by the accusative suffix -a (i.e. a nominalized purpose clause). In (27d) there is a missing syntactic argument serving as the theme of the dependent verb, which is obligatorily controlled by the main theme. According to McDaniels (2014: 79), Comanche purpose and rationale constructions consist of subordinated clauses at the core level.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

581

582

LILIÁN GUERRERO

(27) Comanche (Northern Uto-Aztecan; McDaniels 2014: 75, 73, 77, 83) n¯ii t¯im¯i-i [n¯ii _j t¯ihka-tui] a. Yuhu-noko-p¯i-aj fat-bake-res-acc 1p(s.clitic) trade-pfv 1p eat-unr ‘I bought frybreadj in order for me to eat _ j’. peekwij kw¯ih¯i-n [uj b. S-u-t¯i-sei _i t¯ihka-tui] def-3p-nom-dm fish catch-pfv 3p(o.clitic) eat-unr ‘Hei caught a fish (in order) _ i to eat it’. u-hkupa uj t¯i¯iki [u-hkaj p¯ii napinai-tui] c. S-u-t¯i-sei def-3p-nom-dm 3p-inside 3p(o. clitic) put 3p-acc 3p.co save/select/stash-unr ‘Hei put it in there (in order) for himi to set it aside’. t¯i-poo-p¯i-aj [p¯ii yaa-t¯i d. S-u-t¯i-sei _j niha-tui-a] def-3p-nom-dm unsp.o-make.mark-res-acc 3p.co.gen read-unr-acc take-ipfv ‘He is taking a booki to read _i’.

The co-indexed pronouns in purpose clauses marked by -betchi’ibo in Yaqui and the rationale clauses in Comanche do not require any modification to the linking algorithm, but the obligatory control relation observed in purpose clauses in (25a–c) mainly follows the modifications introduced in the semantics-to-syntax algorithms in (19) for non-subordinate core junctures, but some revisions are needed for the syntax-to-semantics linking (20). Typological studies have found that the argument structure in the linked unit is not logically predetermined by the meaning of purpose, ‘although . . . typical purposive situations are bound to their matrix event via agent-binding’ (Schmidtke-Bode 2009: 50). Schmidtke-Bode’s work on purpose clauses reveals significant trends for coreference in general, and same-subjects in particular. The second most common pattern involves the main theme. In Yaqui purposive constructions, coreference is not an option but a must. Therefore, in order to capture this complexity, additional information may be needed with respect to the semantics of the clause, and the fact that the lexical manifestation of the controllee is subject to language-specific restrictions. As is, the step in 4 in (20) establishes that there is a controller–controllee relation among an argument in each core inside a non-subordinate core juncture; step 4a is relevant for those constructions including a control verb as a matrix verb; step 4b tries to capture constructions without coreferential control. However, Yaqui purpose clauses do not have a control verb, and yet the selection of the controller follows the theory of obligatory control. I therefore suggest adding the specification in the new step 4c, including the referential control hierarchy introduced in (7e). (28)

Revision of the syntax-to-semantics linking algorithm (Guerrero 2013: 17) 4. In non-subordinate core junctures, one of the arguments of the matrix core must be linked to an argument position in the embedded logical structure: a. If the matrix predicate is a control verb, then follow the theory of obligatory control constructions: causative and jussive verbs have undergoer control, while all other (M-)transitive verbs have actor control, or

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

b. If the matrix predicate is not a control verb, then link the unlinked syntactic argument in the matrix core to the logical structure argument position of the pivot of the linked core, otherwise c. If the non-subordinate core juncture expresses a purpose relation, and 1. if the construction demands structural control, then the linked argument must be unfilled (e.g. syntactic controllee in English, Lakhota, Dyirbal); otherwise, 2. if the construction demands inherent control, then the linked argument can be filled by a bound pronoun, clitic or agreementinflection, (e.g. semantic controllee in Acehnese, Chuj, Yaqui, Toqabaquita) 3. the identity of the controllee may follow the referential control hierarchy: the referential identity of an obligatory controlled argument is determined by the main actor > the main undergoer > another main core argument > there is no control relation.

Instead of ‘pure’ syntactic control constructions, the steps in 4c try to capture the referential controller–controllee patterns found in purpose and rationale clauses across languages (see Guerrero 2012, 2013). The controllee can be covert and show structural control, as depicted by the actor gap in (23), (24), (25a–c), (26a), and (27b), and the undergoer gap in (23), (27a, d); or it can be overt and yield inherent control, as in the co-indexed pronoun for the undergoer in (24), (25d–e), (26b), (27c–d). Step 4c3 seeks to mirror the most common coreferential patterns observed cross-linguistically, as well as Cutrer’s (1993) description. In sum, purposive constructions may be expressed by different syntactic linkages in particular languages. Simplified representations for core cosubordination (25b) and core coordination (25c) in Yaqui are presented in Figure 14.7 and Figure 14.8, respectively. The constructional schema in Table 14.2 tries to capture the syntactic and semantic properties of the Comanche rationale clause in (27a). Here, there are two referential control relations. There is inherent control (co-indexed pronoun) involving the main PSA and the linked actor; and there is structural control (gap) involving the main and the linked theme. SENTENCE CLAUSE CORE CORE

CLM

CORE

nei ino piaroa-k [Kafe-ta _i ji’i-pea-ka] ‘I borrowed some money in order to drink coffee’ in (25b) Figure 14.7 Simplified structure of purposive core cosubordination in Yaqui

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

583

584

LILIÁN GUERRERO

SENTENCE CLAUSE CORE

CORE

CLM

[mas-ta _i me’e-betchi’ibo] U o’ou-Øi juya-u siika ‘The man went to the hills in order to kill the deer’ in (25c) Figure 14.8 Simplified structure of purposive core coordination in Yaqui

Table 14.2 Constructional schema for ‘I bought frybread in order for me to eat’ in (27a) C ONSTRUCTION : ‘Rationale’ construction in Comanche S YNTAX : Juncture: core Nexus: coordination Construction type: adverbial non-subordination [C L A U S E [ RP [C O R E 1 . . .]] PURP [ RP [C O R E 2 . . .] (RP)]]] Unit template(s): default PSA: the controller is the matrix actor in CORE 1 the controllee is the linked actor in CORE 2 Linking: the number of core slots is maintained, i.e. inherent control (4c2) Non-PSA: the controller is the matrix undergoer in CORE 1 the controllee is the linked undergoer in CORE 2 Linking: the number of core slots is reduced by 1, i.e. structural control (4c1) M ORPHOLOGY : Verb: non-finite PSA Controllee: filled Non-PSA Controllee: unfilled CLM: none S EMANTICS : One action is done with the intent of realizing another state of affairs The identity of the controllee follows the referential control hierarchy (4c3) P RAGMATICS : Illocutionary force: unspecified Focus structure: unspecified

14.4.5 Manner and Means Relations Finally, the semantic relations at the left end of the semantic scale denote modifying subevents such as manner, means, motion and position. Most studies on adverbial subordination do not mention manner/means (Ford 1993; Cristofaro 2003; Givón 2001), while some only consider manner/similarity relations (Dixon and Aikhenvald 2009). Others include both and highlight the verb form and an overt CLM (Hengeveld 1998; Van Valin 2005); and some others group manner, time and location together (Thompson et al. 2007). In English, manner and means relations are usually coded by the -ing verb form; an example is given in (29a). In the Spanish example in (29b), taken from Conti (2012), there is a gerund verb form ending in -ando.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

(29)

a. John cut the bread by slicing it b. El pájaro volaba imitando el aleteo de su madre ‘The bird flew imitating the wingbeats of its mother.’

In Yaqui, these adverbial relations are coded by the general CLM -kai. In the constructions in (30), -kai introduces the manner in which a motion event is carried out (30a), and the means by which an action is carried out (30b). The example in (30c) expresses some sort of positive/negative circumstance around the main action. The entire clause describes a complex event in which the main actor is engaged. Because -kai is a general CLM, the actual semantic relation between the two units is recovered from the discourse context, the nature of the events, and world knowledge (e.g. two ongoing events involving the actor; a functional answer for why). In the logical representations, ^ represents two simultaneous states of affairs. Note that there is an unfilled position in the second component co-indexed with the main PSA. (30) Yaqui manner and means clauses tebae-koko-ka] te ama jooka a. [junama’a _i si over.there int hungry-die.pl-clm 1pl.nom there live-prs ‘We live over there, starving to death.’ (Silva; HVH: 213) a′. [live′ (1pli, there)] ^ [starved′ (xi, there)] ousi tomi-yo’o-k [ _i yeewe-kai] b. Into¼nei dm¼1sg.nom a.lot money-win-pfv play-clm ‘And then, I won a lot of money by playing.’ (Buitimea; chapayeca: 83) b′. [do′ (1sgi, [won′ (1sgi, money)])] ^ [do′ (1sg, [play′ (xi)])] [_i ji’i-bwa-ka] to’o-ne c. [Kat]¼e’ei neg.imp¼2sg.nom something-eat-clm lie-pot ‘You will go to bed without eating.’ (Silva; zorrillo: 13) c′. [do′ (2sg, [sleep′ (2sg)])] ^ [NEG do′ (xi, [eat′ (xi, Ø)])]

Yaqui -kai clauses can be considered equivalent to an imperfective participial -ing clause (Givón 2001). However, -kai clauses are more restrictive than English -ing forms in terms of control. While in the English construction in (31a) the antecedent of the missing PSA of ‘coming’ can be ‘she’ or ‘he’, the Yaqui clause in (31b) rules out the possibility of the main object as the controller. The possibility for ing-infinitival clauses with accusative/genitive subjects (31c) is also disallowed. (31) a. She saw him coming out of the library Joan-taj bicha-k [teopo-po _i yeu¼sim-kai] b. Inepoi 1sg.nom John-acc see-pfv church-loc out¼go.sg-clm ‘I saw John leaving the library.’ c. Na’a tajkai-reo-Øi¼bea [_/*ai tajkai-m nenenka-ka] aui ania-n dem tortilla-nmlz¼dm 3sg tortilla-pl red.sell-clm self help-ipfv ‘As for this tortilla maker, she helped herself by selling tortillas.’

The linked unit marked by -kai is not syntactically subordinated to the main event, but adds a second predication to it. Several major features define

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

585

586

LILIÁN GUERRERO

manner and means kai-clauses in Yaqui. The linked event can denote a state as well as an activity predicative (manner is not limited to motion events, for instance), and the two events are co-temporal (simultaneous). Additionally, manner and means clauses are usually reduced units (intransitive verbs or transitive verbs without referential core arguments), the verb must be unmarked for TAM affixes and negation, and must generally occur at the beginning or the end of the sentence without a pause. More importantly, they are subjectless clauses. The main PSA (the actor) must control the identity of the missing syntactic argument in the linked unit. In sum, this clause-linkage type yields nuclear or core cosubordination depending on the valence of the linked verb. In a nuclear juncture, the two nuclei combine to form a complex nucleus with a single set of core arguments (30a–b). Figure 14.9 gives the syntactic representation of nuclear cosubordination in (30b).10 In a core juncture, the two cores share one but not all arguments, as in (30c) and (31b–c). Figure 14.10 gives the syntactic representation of core cosubordination in (31c). The actual linking of this construction is the same as the one in rationale/purpose clauses, in which there is a missing syntactic argument in the linked unit co-indexed with the main PSA, and the linked verb is unmarked for TAM suffixes.

SENTENCE CLAUSE CORE RP

NUCLEUS NUC

NUC

CLM

[ _i yeewe-kai] Into=nei ousi tomi-yo’o-k ‘And then, I won a lot of money by playing’ in (30b) Figure 14.9 Nuclear cosubordination in Yaqui

SENTENCE CLAUSE CORE PrCs CORE

CLM

CORE

RP Na’a tajkai-reo-Øi=bea [_i tajkai-m nenenka-ka] aui ania-n ‘As for this tortilla maker, she helped herself by selling tortillas’ in (31c) Figure 14.10 Core cosubordination in Yaqui

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

14.5

Conclusion

The main goal of this chapter was to examine how the linking algorithm works in adverbial (adjoined) clauses. I have argued that the structural linkages which express adverbial relations across languages are heterogeneous: some relations make use of looser syntactic combinations, while others are encoded by tighter syntactic linkages. Furthermore, the same semantic relation can show more than one syntactic linkage with different degrees of semantic and syntactic integration. The theory of clause union in RRG, based on juncture–nexus relations, conveniently accounts for these structural complexities. On one hand, adverbial relations can be coded by ad-subordination, as well as those relations expressed by coordination and cosubordination. On the other hand, only ad-subordinated linkages occupy a peripheral position at the level of the core or the clause, whereas nonsubordinated linkages demand a different syntactic representation.

References Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In K. Ebert (ed.), Studies in Clause Linkage (Papers from the First Köln-Zürich Workshop). Arbeiten des Seminars für Allgemeine Sprachwissenschaft (no. 12), 23–55. Zurich: University of Zürich. Bickel, Balthasar. 2003. Clause Linkage Typology. Lecture series delivered at the 2003 International Role and Reference Grammar Conference, UNESP, Brazil. Bickel, Balthasar. 2010. Capturing particulars and universals in clause linkage: A multivariate analysis. In Bril (ed.), 51–104. Bril, Isabelle. 2010. Clause Linking and Clause Hierarchy: Syntax and Pragmatics. Amsterdam: John Benjamins. Conti, Carmen. 2012. Subordinación periférica y subordinación dependiente: clasificación estructural de la subordinación adverbial en español. In R. Mairal Usón, L. Guerrero and C. González-Vergara, El funcionalismo en la teoría lingüística: La Gramática del Papel y la Referencia, 269–286. Spain: AKAL. Conti, Carmen. 2021. Cosubordinación en español. Bern: Peter Lang. Cristofaro, S. 2003. Subordination. New York: Oxford University Press. Cutrer, Michelle. 1993. Semantic and syntactic factors of control. In Robert Van Valin (ed.), Advances in Role and Reference Grammar, 167–196. Amsterdam: John Benjamins. Declerck, Renaat. 1997. When-Clauses and Temporal Structure. London: Routledge. Dedrick, John M. and Eugene H. Casad. 1999. Sonora Yaqui Language Structures. Tucson, AZ: University of Arizona Press. Diessel, Holger. 2008. Iconicity of sequence: A corpus-based analysis of the positioning of temporal adverbial clauses in English. Cognitive Linguistics 19(3): 465–490.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

587

588

LILIÁN GUERRERO

Diessel, Holger. 2013. Adverbial subordination. In S. Luraghi and C. Parodi (eds.), Bloomsbury Companion to Syntax, 341–354. London: Continuum. Diessel, Holger and Katja Hetterle. 2011. Causal clauses: A cross-linguistic investigation of their structure, meaning and use. In P. Siemund (ed.), Linguistic Universals and Language Variation, 23–45. Berlin: Mouton. Dixon, R. M. W. and Alexandra Aikhenvald. 2009. The Semantics of Clause Linking. Oxford: Oxford University Press. Foley, William A. 2010. Clause linkage and nexus in Papuan languages. In Bril (ed.), 51–104. Ford, Cecilia. 1993. Grammar in Interaction: Adverbial Clauses in American English Conversations. Cambridge: Cambridge University Press. Gast, Volker and Holger Diessel. 2012. Clause Linkage in Cross-Linguistic Perspective: Data-Driven Approaches to Cross-Clausal Syntax. Berlin: Mouton. Gerardo-Tavira, Rebeca. 2018. Oraciones subordinadas temporales: orden, iconicidad y relaciones entre eventos. PhD dissertation, Universidad Nacional Autónoma de México. Givón, Talmy. 2001. Syntax: An Introduction. Amsterdam: John Benjamins. Guerrero, Lilián. 2006. The Structure and Function on Yaqui Complementation. PhD thesis, University at Buffalo (SUNY). Guerrero, Lilián. 2009. On the semantic dimension of complementation. In L. Guerrero, S. Ibáñez and V. Belloro. Studies in Role and Reference Grammar, 319–343. Mexico: UNAM. Guerrero, Lilián. 2011. Clause linkage and purpose clauses in Southern UtoAztecan Languages. In W. Nakamura (ed.), New Perspectives in RRG, 217–245. Newcastle upon Tyne: Cambridge Scholars Publishing. Guerrero, Lilián. 2012. Más sobre controladores y pivotes: el caso de las cláusulas de propósito. In R. Mairal Usón, L. Guerrero and C. GonzálezVergara (eds.), El funcionalismo en la teoría lingüística: la Gramática del Papel y la Referencia, 307–329. Madrid: AKAL. Guerrero, Lilián. 2013. Controller–controllee relations in purposive constructions: A construction based account. In B. Nolan and E. Diedrichsen (eds.), Linking Constructions into Functional Linguistics: The Role of Constructions in Grammar, 1–22. Amsterdam, John Benjamins. Guerrero, Lilián. 2017. On purpose, causal and reason clauses in Yaqui. International Journal of American Linguistics 83: 679–718. Guerrero, Lilián. 2019. Yaqui clauses and the Interclausal Relations Hierarchy. In R. Kailuweit, L. Künkel and E. Staudinger (eds.), Applying and Expanding Role and Reference Grammar. Germany: NIHIN. Guerrero, Lilián, V. Belloro and Carmen Conti. 2017. Motivaciones en conflicto en la posición de adjuntos temporales de secuencia. Onomázein 36 (June): 98–121. Guillaume, Antoine. 2008. A Grammar of Cavineña. Berlin: Mouton. Hengeveld, Kees. 1998. Adverbial clauses in the languages of Europe. In J. Van der Auwera (ed.), Adverbial Constructions in the Languages of Europe 335–420. Berlin: Mouton. Hetterle, Katja. 2015. Adverbial Clauses in Cross-Linguistic Perspective. Berlin: Mouton.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

Linking Syntax and Semantics in Adverbial (Adjoined) Clauses

Lehmann, Christian. 1988. Towards a typology of clause linkage. In J. Haiman and S. A. Thompson (eds.), Clause Combining in Grammar and Discourse, 181–225. Amsterdam: John Benjamins. Mati´ c, Dejan, Rik van Gijn and Robert D. Van Valin, Jr. 2014. Information structure and reference tracking in complex sentences. In R. van Gijn, J. Hammond, D. Mati´ c, S. van Putten and A. V. Galucio, Information Structure and Reference Tracking in Complex Sentences, 1–44. Amsterdam: John Benjamins. Matthiessen, Christian M. I. M. and Sandra A. Thompson. 1988. The structure of discourse and ‘subordination’. In J. Haiman and S. A. Thompson (eds.), Clause Combining in Grammar and Discourse, 275329. Amsterdam: John Benjamins. McDaniels, Todd. 2014. Rationale and purposive clauses in Comanche. International Journal of American Linguistics (80): 69–97. Pérez Quintero, María Jesús. 2002. Adverbial Subordination in English: A Functional Approach. Amsterdam: Rodopi. Schmidtke-Bode, K. 2009. A Typology of Purpose Clauses. Amsterdam: John Benjamins. Thompson, Sandra, Robert Longacre and Shin J. Hwang. 2007. Adverbial clauses. In T. Shopen (ed.), Language Typology and Syntactic Description, Vol. 2, 237–300. Cambridge: Cambridge University Press. Vallejos, Rosa. 2014. Reference constraints and information structure management in Kokama purpose clauses: A typological novelty? International Journal of American Linguistics 80(1): 39–67. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantic Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2007. Recent developments in the Role and Reference Grammar theory of clause linkage. Language and Linguistics 8(1): 71–93. Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and controllers. In L. Guerrero, S. Ibáñez and V. Belloro (eds.), Studies in Role and Reference Grammar, 45–68. México: UNAM. Van Valin, Robert D., Jr. 2021. Cosubordination. In R. Van Valin (ed.), Challenges at the Syntax–Semantics–Pragmatics Interface: A Role and Reference Grammar Perspective, 241–254. Newcastle upon Tyne: Cambridge Scholars Publishing. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wash, Suzanne. 2001. Adverbial Clauses in Barbareño Chumash Narrative Discourse. PhD dissertation, University of California, Santa Barbara.

Notes 1 Previous works dealing with adverbial clauses in RRG include Bickel (1993, 2003, 2010), Conti (2012), Van Valin (2007, 2009), and Guerrero (2012, 2013, 2019). Some recent cross-linguistic studies dealing with

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

589

590

LILIÁN GUERRERO

2

3 4 5

6 7

8

9 10

complex constructions have also emphasized unique aspects of the juncture–nexus theory (Schmidtke-Bode 2009; Bril 2010; Gast and Diessel 2012; Mati´ c et al. 2014; Hetterle 2015); see also Conti (2021) for a discussion on cosubordination and other linkages on Spanish. In the examples, coreferential and shared arguments are co-indexed, and ‘_’ signals an implicit argument; in non-English examples, the linked unit is in brackets. In While washing the car, Dana saw Chris, Dana may be understood as the controller of the missing argument. See Van Valin (2005) for a detailed description of the theory of linking in simple and complex constructions. Although operator sharing is an important feature, ‘there appear to be cases of cosubordination in which operator sharing is possible but not obligatory’ (see Bickel 2003, 2010; Van Valin 2007: 80; Foley 2010). For instance, Bickel (2010: 53) found that in Belhare, Nepali, and several other languages of South Asia, the scope of interrogative markers is indeterminate: depending on the context, the sentence may be interpreted as having conjunct or disjunct scope. In some cases, these sentences can be read as conjunctions or temporal relations. Van Valin (2021) revisits Foley’s and Bickel’s concerns on the variable scope of the IF operator and explains how the problematic examples do not question the validity of cosubordination as a nexus relation, but they do highlight the language-specific constraints at the clausal and sentence levels. See also Van Valin (2005: 282–284) for discussion of IF operators in reason, conditional and concessive clauses. In the final reformulation, Van Valin uses only ‘and’, meaning that both steps must be satisfied. If ‘or’ is also included in the linking, nonsubordinate core junctures in peripheral positions can be neatly captured, without additional modifications. The logical structure in (25a′) tries to capture the notion of motion-cumpurpose as modifying subevents. In Van Valin (2005), the formal representation of a purposive relation is want’ (xi, [LS2. . . x. . .]) ^ DO (xi, [LS1. . . xi . . .] e CAUSE [LS2. . . xi. . .]). In (25b), this logical structure captures the idea that a volitional action is performed with the intention of another event, and the fact that the main actor is directly involved in the intended event (same-subject clauses). Since the betchi’ibo-clauses in Yaqui are not implicative, the logical structure in (25c) does not contain the CAUSE component. PURP in (25d–e) is a simplified version of the logical structures in (25b–c). For details on the analysis of control constructions in RRG, see §7.3.1 in Van Valin (2005). This is a highly marked nuclear juncture; see Van Valin and LaPolla (1997: 671, fn. 19) for another example found in French involving perception verbs.

https://doi.org/10.1017/9781316418086.018 Published online by Cambridge University Press

15 Cleft Sentences and Relative Clauses Luis París

15.1

Introduction

Role and Reference Grammar (RRG) holds the view that a sentence is not just a concatenation of words into a sequence but the incarnation of a set of formal templates that are stored in syntax.* Syntactic templates have two effects that are crucial for the understanding of the nature of linguistic symbols. First, they are a key factor in determining the pattern followed by the influx of information contributed by each word into the new symbol (i.e. the phrase). Second, the phrase is an enclosure where certain relations occur among its constituents that are shielded from some external influences. Cleft sentences and relative clauses make this encapsulation of the phrase evident. Sentence (1) illustrates a typical instance of a relative clause structure (hereafter, RC). (1)

The student who failed the test quit the program.

This sentence is an instance of an externally headed relative clause (EHRC), which will be analysed in Section 15.2. It contains a relative clause (‘who failed the test’) that modifies the head noun (i.e. ‘student’) of the referential phrase (RP) – in which the relative clause is embedded. This head noun is the antecedent that fixes the reference of the relative pronoun (‘who’) in the relative clause and, in doing so, it might also demand agreement on nominal features like person, gender and number depending on the morphology of the language. The head noun functions in (1) as the privileged syntactic argument (PSA) of the main clause; however, relative clauses can modify any non-PSA direct core argument and even adjuncts modulo Keenan and Comrie’s (1977) Accessibility Hierarchy.1 In the same sense, the co-indexed wh-form can have the function of any direct core argument or it can even be an adjunct in the * This paper has benefited greatly from the careful reading, insightful questions and several suggestions by Delia Bentley, one of the editors of this Handbook. Of course, I am responsible for all remaining errors.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

592

LUIS PARÍS

relative clause (though see Section 15.2 for some relevant restrictions in Malagasy, a Western Austronesian language spoken in Madagascar). Therefore, the shared participant is not required to play any specific syntactic or semantic role in either the matrix clause or the relative clause. This has important consequences for the syntax of RCs. The relative clause is safeguarded from any requirement or influence from the matrix clause by the RP in which it is embedded. Operators like tense, grammatical aspect or negation affect each clause (matrix and relative) independently. All the formal properties imposed on the relative pronoun come from the head noun – and are of a referential nature (person, gender and number) – or from the matrix verb of the relative clause. These properties are captured by an analysis of relative clauses as RP nuclear modifiers (Pavey 2004; Van Valin 2005: 220–223; 260–266; Van Valin and LaPolla 1997: 497–499). This means that RCs, just like English attributive adjectives, are elements in the periphery that describe either a property of the referent or an event in which it participates. This semantic dependency is reflected in the syntax and further justifies the nuclear periphery analysis: for those languages with a rich nominal morphology, the noun imposes agreement relations on the attributive element. The semantics of RCs displays properties that illustrate the insufficiency of truth conditional semantics to account for linguistic meaning. The meaning of the relative clause and the head noun in (1) can be expressed by the following simple sentence. (2)

The student failed the test (and quit the program).

Thus, the information about the world is the same in both cases. However, the meanings are quite different because the composition of the respective semantic representations is different. This will be captured later in this chapter (see the representation in (10)) by embedding the logical structure (LS) of the RC as an argument of the attributive LS, which is consistent with the RC syntax, in that the relative clause is a modifier of the RP nucleus. In addition to EHRCs, Section 15.2 will also discuss Internally Headed Relative Clauses (IHRCs), free relatives and non-restrictive relatives. In IHRC, the head noun is inside the embedded clause, which we know is embedded due to the presence of a complementizer. The so-called ‘free relatives’ are not really RCs, since they lack a head noun as well as a relative pronoun in many languages. Finally, in non-restrictive RCs, the embedded clause is separated from the head noun by a pause, and it does not fulfil the same pragmatic function characteristic of relative clauses, namely, the identification of the referent of the head noun. This chapter also discusses cleft sentences, specifically, it-clefts as exemplified by (3), which will be the focus of Section 15.3. (3)

It was John who asked the question.

The components of it-clefts are a matrix clause with a ‘dummy’ ‘it’ pronoun, a copula clause and a clefted noun phrase (John in (3)), which is the

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

antecedent that fixes the reference of either a relative pronoun or, if this is absent, an argument or adjunct slot in the embedded clause, as in (4). (4)

It was in 1996 Mary had her baby!

In (3) the subordinated clause is introduced by a relative pronoun but it can also be subordinated by a complementizer, as in (5). (5)

It was in 1996 that Mary had her baby!

The analysis offered in this chapter adopts a perspective that highlights the interfaces that constitute the internal structure of clefts. From a truthconditional perspective, sentence (3) carries the same content as a simple sentence like John asked the question. In fact, (3) describes a single event whose description is contributed by the LS of the embedded verb (i.e. ask). In this sense, the embedded clause provides the bulk of the semantic content carried by the whole sentence, whereas the matrix clause contains almost no truth-conditional information. This is the reason why this chapter introduces the notion of clefts as a structure that materializes an ‘inverted’ linking. By ‘inverted linking’ I mean that the stronger semantic clause, or the clause that provides the meaning for the whole sentence, is syntactically dependent on a semantically weaker matrix clause (see Section 15.3). This linking is not accidentally associated with a particular information structure. In clefts, matrix clauses carry the pragmatic information weight, since they contain the narrow focus of the sentence, namely, the clefted noun (i.e. John in (3)). Further, this information is also the reference value of the relative pronoun in the dependent clause. Therefore, the matrix clause is semantically weaker, yet informatively stronger, than the dependent clause, so that the syntax of clefts is clearly driven by pragmatics. This conclusion is consistent with the view held in RRG on the depth of the insertion of communicative needs in grammatical structure. Information structure digs down into the core of the grammatical engine. There are strong similarities between RCs and clefts. The most salient one is that both constructions involve an embedded clause that can be introduced by a relative pronoun, which might be absent in some contexts. In addition, both the embedded clause in clefts and those in RCs are presupposed. These resemblances motivate some proposals in which cleft clauses are analysed as relative clauses. In contrast, the view pursued in this chapter – in particular in Section 15.4 – focuses rather on the differences between the two types of embedded clause. One of the properties of relative clauses as defined here is that they are modifiers of the nominal nucleus of a noun phrase. There is clear evidence that this is not the case with clefts. For instance, as will be discussed in due course, in RCs, quantifiers in the matrix clause range over the reference of the noun as restricted by the relative clause, while in clefts they have scope over the meaning of the antecedent noun without any restriction imposed by the embedded clause. This

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

593

594

LUIS PARÍS

evidence leads to a representation of clefts in terms of a clause-linkage relation in which the embedded clause modifies the core of the matrix clause. Section 15.4 will also explore fundamental pragmatic differences between the two constructions. The embedded clause in an RC is a presupposition that helps the listener to identify an otherwise unidentifiable participant. In contrast, the activation status of the referent of the clefted noun in clefts is irrelevant; what is crucial here instead is that this noun is the focal element. The chapter is organized as follows. Section 15.2 presents an analysis of RCs: their syntax, semantics and syntax–semantics linking. Section 15.3 introduces cleft sentences and subsection 15.3.1 the relevant portion of the linking algorithm that accounts specifically for clefts. Section 15.4 offers a contrastive analysis of clefts and RCs. The chapter ends with Section 15.5, which presents a short conclusion.

15.2

Relative Clauses

The ingredients of a typical English RC like (1), repeated in (6a) for ease of exposition, are a clause introduced by a relative pronoun (i.e. who) co-indexed with a head noun outside the RC and preceding it (i.e. student) and, finally, a matrix clause, which the head noun belongs to. The RC is embedded within the RP headed by the antecedent noun – student – as can be seen, for example, by the fact that a proform can replace the entire RP (see 6b). (6)

a. The student who failed the test quit the program. b. She quit the program.

The events described by the RC – the failing event – and by the matrix clause – the quitting event – might not be connected beside the fact that they share the same participant. A central relation within the RP is that of obligatory co-indexation between the head noun and the relative pronoun. As long as it belongs to the RP, there may be intervening material – all within the RP that contains the relative clause – between the head noun and the relative pronoun, as in (7). (7)

The student of the teacher who called you yesterday is sick.

However, the two nouns have the right features to function as antecedent of the pronoun – they are both [þhuman] – and the interpretation is ambiguous. Clearly, the preferred interpretation takes the closer antecedent (i.e. teacher). In addition, if there is more available intervening material between the head noun and the relative pronoun, the chances of acceptability of the sentence are proportionally lowered. The fact that there is no restriction on the function that the head noun can have in the matrix clause is illustrated below. It can be a direct core argument like in (6a) or it can even be an adverbial as in (8).

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses (8)

I saw Mary at the bar where you met her.

In this case the relativizer is not a pronoun but a wh-form with an adverbial function in the RC; in other languages, it could be any pro-adverbial form.2 In English as well as many other languages there are no restrictions on the function of the co-indexed proform in the relative clause, as suggested by (8). However, some languages pose specific requirements. As noted by Keenan (1976), Malagasy allows only the co-indexed element to be the subject of the relative clause (9). (9)

a. Na-hita ny vehivavy [(izay) nan-asa ny zaza] Rakoto. pfv.atv-see det woman comp pfv.atv-wash det child Rakoto ‘Rakoto saw the woman that washed the child.’3 *‘Rakoto saw the woman that the child washed.’ b. Na-hita ny zaza (izay) nan-asa ny vehivavy Rakoto. pfv.atv-see det child comp pfv.atv-wash det woman Rakoto ‘Rakoto saw the child that washed the woman.’ *‘Rakoto saw the child that the woman washed.’ c. Na-hita ny zaza (izay) sas-an’ ny vehivavy Rakoto. pfv.atv-see det child comp wash-pass det woman Rakoto ‘Rakoto saw the child that was washed by the woman.’

RCs require a matrix clause with an available antecedent noun that heads an RP that contains an embedded clause via a proform co-indexed with the antecedent noun. The lexical category of the antecedent is fixed: it can only be a noun. This is easily relatable to the communicative function generally attributed to (restrictive) RCs, namely that they help identify the referent of the antecedent noun by relating it to some already known event. The syntactic representation of the EHRC in (1) is given in Figure 15.1. Just like any adjectival modifier, the relative clause is in the Periphery of the nominal nucleus of the Referential Phrase. The semantic representation expresses the contribution of the RC as an attributive modifier (i.e. be′ (. . .)) of the head noun. This predication is inserted into the matrix LS slot that corresponds to the first argument, that is, the first argument of quit′, which should be the position occupied by the head noun. The participant referred to by the head noun student is shared with the predicative LS (be′(student, . . .). In short, (10) represents that the semantics of the RC fills in the argument slot in the main LS that corresponds to the head noun. In particular, this is not a co-indexation relation, but a position that belongs to two LSs (the one that corresponds to quit′ and the one that corresponds to be′). The slot of this shared participant is shown in (10) by a thick broken underline (Van Valin 2012). In addition, this position is co-indexed with the argument position filled with the wh-word in the relative clause LS.4 The superscript ‘¬i’ refers to the activation status of the participant; in particular, the head noun of the RC is unidentifiable, which mean that it is neither active or accessible nor

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

595

596

LUIS PARÍS

SENTENCE CLAUSE CORE RP

NUC

CORE

PRED

NUCN

RP

PERIPHERY CLAUSE PrCS

CORE

RP

NUC

RP

PRED The student who

failed the test

quit

the program

Figure 15.1 Externally headed relative clauses

inactive, it is unidentifiable and either anchored or non-anchored (Chafe 1987; Pavey 2004: 5). (10)

INGR quit′ ([be′ (student1¬i, [INGR fail′ (who1, test)])], program) ----------

The regular syntax-to-semantics linking algorithm applies to relate Figure 15.1 to (10) but, in addition, we need the following further restrictions (Van Valin 2012: 57). (11)

a. Retrieve from the lexicon an attributive LS (i.e. be′ (x, [pred′])) and substitute the LS of the verb in the relative clause (i.e. [INGR fail′ (x, y)]) for the second argument of the attributive LS (i.e. [pred′]). b. Co-index the first argument in the attributive LS with either the unlinked argument position in the relative clause LS, or, if there is a relative pronoun, to the argument position linked to the relative pronoun. c. Insert the attributive LS into the argument position in the matrix LS occupied by the head noun, replacing the variable in the first argument position in the attributive LS with the head noun.

There is still the question whether this structure complies with the Completeness Constraint in cases like (12), where there is no overt relative pronoun. The absence of an overt relative pronoun is possible in all cases except in those where the pronoun is linked to the argument in the LS that should be the PSA/subject. (12)

The student I saw in the library was not screaming.

In this case, the undergoer of see is not realized within the embedded clause. This requires the reduction of a slot in the verb core, although the

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

structural conditions for this reduction are entirely different from those that characterize control structures (Van Valin 2005: 255). Control involves a non-subordination nexus at a core juncture, in which the relevant core is a semantic argument of the matrix core. Quite differently, the relative clause in RC is a subordinated optional modifier of a noun nucleus. The answer to this problem is to resort to the formulation of the Completeness Constraint in Van Valin (2005: 233). Under this principle, the argument of an LS that is not syntactically realized within the clause can be satisfied anywhere within the sentence that contains that clause. In this way Completeness is respected. Still, as is the case with control structures, we need to determine the specific structural conditions that license this slot reduction in RCs. The following is an attempt to capture the relevant restrictions. (13)

The syntactic slots of the core of a finite clause may be reduced by 1 if the argument to which the missing slot should have been linked is fulfilled by a variable co-indexed with an antecedent noun that is the nucleus modified by the phrase containing the empty slot.

In this way we identify the specific condition that allows the optional reduction of an argument slot in some English EHRCs. Internally headed relative clauses are characterized by the head noun being part of the relative clause. In particular, the head noun is expressed in a position within the relative clause, while it is omitted in the matrix clause. This type of RC is found in Bambara, an SOV language. (14)

Bambara (Mande, Africa; Bird 1968, cited in Van Valin 2012) [Ne ye so min ye] tye ye san. 1sg pst horse rel see man pst buy ‘The man bought the horse that I saw.’

The relevant question is how the hearer recovers the semantic function of the co-indexed participant in the matrix LS. Upon hearing (14), a transitive syntactic template is retrieved because of the verb buy. There is a relativizer which indicates the presence of an embedded clause headed by the verb see, which contains two arguments: ne (‘I’) and so (‘horse’). The word order of the language (SOV) indicates that the first one is the Actor (and, hence, the PSA) while the second is the Undergoer and, thus, the non-PSA macrorole argument. In turn, the matrix clause nucleus is buy, which requires two arguments, although there is only one: tye (‘man’). The fact that the verb buy selects human actors is the only cue that indicates, first, that man is the Actor and, second, that there is a missing Undergoer argument, which has to be found in the relative clause. There are two candidates, one is I and the other is horse. World knowledge indicates that people are less likely to be sold than horses and this might motivate the choice of horse in this case. In other cases, the structure may be disambiguated on the basis of contextual

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

597

598

LUIS PARÍS

SENTENCE CLAUSE CORE RP

RP

NUC

CLAUSE CORE RP

RP

Neye

NUC

so min ye

tye

ye

san.

Actor Undergoer Actor LEXICON

see′(x, y)

Undergoer [do′ (v, Ø)] CAUSE [BECOME have′ (v, w)]

(15a) [be′ (x, [pred′])]

LEXICON

[be′ (xi , [see′ (ne, soi)])] co-indexing (15b) RC LS substitution (15c)

[do′ (tye, Ø)] CAUSE [BECOME have′ (tye, [be′ (xi, [see′ (ne, soi)])])] Figure 15.2 Internally headed relative clauses (from Van Valin 2012)

clues. Taking into account the preceding considerations, the rules governing linking from syntax to semantics in IHRCs are the following (Van Valin 2012: 60): (15)

a. Retrieve from the lexicon an attributive LS and substitute the LS of the verb in the relative clause for the second argument. b. Co-index the first argument in the attributive LS with the argument in the relative clause LS identified as the head noun. c. Insert the attributive LS into the open argument position in the matrix LS.

The constituent structure of this construction, shown in Figure 15.2, is extremely interesting. One of the salient features is that it allows the antecedent of a co-indexation relation to be more embedded than the coindexed variable. Typically, the antecedent precedes the co-indexed element in the semantic representation and the antecedent is either equally or less syntactically embedded than the co-indexed element. However, this is not the case with IHRCs. In IHRC, the Completeness Constraint is satisfied since the argument linked to a reduced slot is satisfied somewhere else within the sentence. However, it is worth pointing out the structural difference between

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

reducing an argument slot in the matrix core and reducing it in an embedded clause. In particular, binding a variable with an argument nested more deeply in the LS is not at all frequent in languages like English, if possible at all, but this is exactly the case with [be′ (xi, [see′ (ne, soi)])], where the lexical element so is an argument of an LS contained within another LS with a variable bound to the embedded lexical argument. This relative clause is not a modifier of an RP, but rather the relative clause and the co-indexed element are arguments of the same clause. Another issue raised by IHRCs concerns co-indexation. Typically, obligatory co-indexation like the one exhibited in RCs is governed by structural restrictions. In the case of obligatory reflexives, the antecedent must hold a semantic role that is higher on the macrorole hierarchy than the one held by the co-indexed element. Ultimately, this principle is motivated by the semantic role hierarchy, and therefore, it is independently motivated (Van Valin 2005: 61). Control structures are another example: the controller is an argument of the matrix verb and the controllee is an argument in the LS embedded in the matrix verb LS. In EHRCs the antecedent is higher than the proform both syntactically and semantically. Syntactically, the antecedent is an argument of the matrix clause that contains the clause linked to the LS with the co-indexed proform. Semantically, the higher status of the antecedent does not need to match a higher ranking on the Actor-Undergoer Hierarchy, since they are mapped to macroroles within different LSs. (16)

The dog bit the boy that killed the spider.

The antecedent boy is an Undergoer in the matrix clause. Yet, the co-indexed relative pronoun is an Actor in the embedded clause. Therefore, the higher status of the antecedent bears on the LS that contains it in relation to the LS that includes the proform. The Bambara IHRC is quite different in this regard. The antecedent is within a clause that is at the same syntactic level as the reduced slot. In turn, the semantics contains the bounded variable in the LS of the dependent verb, which is nested within the attributive structure that contains the antecedent as an argument. This contrast – plus the fact that the relative clause precedes the matrix clause – strongly suggests that the co-indexing relation in Bambara IHRCs is driven by pragmatics and world knowledge, as we argued above.5 As for non-restrictive relative clauses, these are clearly different from restrictive ones, in that there is a pause separating the relative clause from its antecedent. In addition, the attribution does not serve the purpose of helping the hearer to identify the referent. Indeed, as can be seen in (17), the antecedent can be a proper noun. Consequently, the information it provides does not need to be presupposed. These properties are well taken care of by the analysis in Figure 15.3, in which the relative clause is a peripheral modifier of the RP (Van Valin 2005: 222).

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

599

600

LUIS PARÍS

RP

Periphery CLAUSE

N

PrCs

CORE

RP

NUC

RP CORE NUC

Chris, who

loves

soccer

Figure 15.3 Non-restrictive relative clauses

(17)

Chris, who loves soccer, flew to Barcelona.

Finally, we should briefly analyse the so-called ‘headless relative clauses’ or ‘free relative clauses’ as exemplified in (18) and (19). The embedded clauses in those sentences (what John said and los que rompieron el libro ‘those who tore the book’) are introduced by wh-words (what and los que) which satisfy argument positions in the matrix LS. The wh-words are in turn modified by an embedded clause. (18)

What John said surprised Sandy.

(19)

Los que rompieron el libro corrieron a sus casas those.m.pl that tear.pst.3pl the-m.sg book run. pst.3pl to their house.pl ‘Those who tore the book apart ran to their houses.’

From a purely semantic standpoint, these examples are like relative clauses, since the argument position in the matrix clause is filled in by the entity referred by the wh-word. This entity is in turn modified by the LS of the embedded verb. In particular, it is not the case that the LS of the embedded clause is an argument of the matrix LS, but rather it is part of the information that fills in that argument position. This meaning is captured by (20), an LS that is structurally identical to the one assigned to the relative clauses previously analysed. There is a crucial difference, though. In relative clauses the underlined argument is filled in by a full noun that is the head of the relative clause and functions as the antecedent of the relative pronoun. In contrast, in (20) the position is occupied by a pronoun and the typical slot of the relative pronoun in the attributive LS is filled by a variable co-indexed with the pronoun. It is clear from (20) that the entity that caused the surprise is the referent of what, and the LS that corresponds to the embedded clause helps to restrict the possible referents of that pronoun, which is just what plain restrictive relative clauses do. In addition, the description in (20) derives the fundamental property of the pronoun in (18) and (19), namely that they do not

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

have antecedents and are thus free to any (contextually appropriate) interpretation. (20)

[do′(whati [be′[do′ (John, say′(John, xi)])]) CAUSE [BECOME surprised′(Sandy)]] -------

Is there any formal correlate of this semantics? The most prominent one is that the nominative inflexion on the main verb in (19) is plural and is being controlled by the pronoun los que ‘those that’. Crucially, the inflexion is not determined by the whole embedded clause los que rompieron el libro but only by the pronoun. In addition, the nominative case of the pronoun is also determined by the matrix clause. In a sentence where the pronoun plays a non-PSA role in the matrix clause, it has to be expressed with an accusative/dative marker (i.e. a ‘to’) like in Pedro reemplazó a los que llegaron tarde ‘Peter replaced those who arrived late’. Had the embedded clause determined the shape of the pronoun, it would have been just los que (nominative) not a los que (accusative). This is also true in cases where the pronoun has an Actor role in the matrix clause and an Undergoer role in the embedded clause (for example, El que invitaste llegó tarde ‘The one you invited arrived late’). In such a case the pronoun should be realized as a nominative form consistent with its meaning in the matrix clause.6 In contrast, the relative pronoun in plain relatives takes the shape that is consistent with the semantics it takes from the embedded clause. For example, in El niño al que le regalé el juguete sonrió ‘The boy whom I gave the toy smiled’: the pronoun expresses the Beneficiary of the verb regalar ‘to give’ and the Actor of the main verb sonreír ‘to smile’, but its formal realization is consistent only with its Beneficiary role in the relative clause.7 The generative tradition assumes either a ‘matching’ or a ‘raising’ analysis of free relatives which subsumes them within plain relative clauses (Cinque 2015). Both analyses assume the presence of an empty position outside the relative clause (the position of the head/antecedent), which is the landing site for the pronoun of free relatives. One of the problems of subsuming the free relatives to plain relatives is that they do not behave in the same way in the so-called ‘extraction’ phenomena. The antecedent of the plain relative El discurso que diste sorprendió a Sandy ‘The talk you gave surprised Sandy’ can be replaced by a wh-word as in ¿Qué que diste sorprendió a Sandy? ‘What that you gave surprised Sandy?’. However, we cannot ask for the identity of the agent of the matrix clause in (19) as in *¿Quiénes rompieron el libro corrieron a sus casas? ‘*Who tore the book apart ran to their houses?’. The sentences in (18) or (19) cannot be reduced to regular relative clauses by claiming an empty position for the antecedent that is filled in by the pronoun coming from the relative clause. They need a different and specific treatment. The evidence presented so far suggests, albeit not conclusively, that the pronouns in (18) and (19) are not relative pronouns. They are more properly analysed as arguments of their matrix clauses, as in Figure 15.4. We can call

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

601

602

LUIS PARÍS

SENTENCE CLAUSE CORE RP

Periphery

NUC

CLAUSE

Periphery

V

PP

CORE NUC

RP

Los que rompieron el libro corrieron a sus casas ‘Those who tore the book ran to their houses’ Figure 15.4 Pronounless relative clauses

the embedded clause ‘pronounless relative clause’ in order, first, to emphasize the opposition with a headless relative clause analysis and, second, to point to the exceptional nature of this ‘relative clause’. In sum, this is a construction of its own that cannot be understood by just applying an old label to it. The linking algorithm has to incorporate a statement that guarantees that the interpretation of the non-syntactically realized argument of the verb in the embedded clause is driven by the co-indexation with the pronoun in the matrix clause.

15.3

Cleft Sentences

Cleft sentences are as complex as they are interesting and, similarly to RCs, they involve a matrix and a subordinate clause. There are different subtypes of cleft, but in this chapter the focus will be on ‘it-clefts’, as represented by sentence (21). This example includes a cleft clause (who blamed the bus driver) introduced by a wh-pronoun that obtains its reference from the antecedent noun (teacher) in the matrix clause, also called ‘clefted noun’. This matrix clause consists of a copula with dummy it. (21)

It was the teacher who blamed the bus driver.

The wh-form that introduces the cleft clause can fulfil any direct core argument position. In fact, it can even express adjuncts, as can be seen in (22). (22)

It was in Buffalo where/that Shelly met her husband.

Clefts have properties that illustrate the complexity of the grammatical interfaces in natural languages. The same semantic information carried by (21) and (22) can be expressed by the simple sentences in (23) and, respectively, (24) without any loss in the description of the relevant events.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses (23)

The teacher blamed the bus driver.

(24)

Shelly met her husband in Buffalo.

The unmarked expression of a proposition is a simple sentence, but clefts involve a bi-clausal structure with an unequal distribution of the content. The semantic weight of (23) and (24) is conveyed by their respective verb LSs. Yet, in clefts like (21) or (22), these content verbs are expressed in embedded clauses. In some sense, this is an ‘inverted’ linking, in that the semantic asymmetry by which one clause is richer in content than the other is not reflected by the syntactic asymmetry. The dependent clause expresses the semantically richer content, while the matrix clause conveys a weaker meaning. This ‘inverted’ syntax–semantics interface is pragmatically driven. It will be shown later that, while being semantically weaker, the matrix clause expresses the element that bears the focus of the sentence. Furthermore, the syntactic realization of the clefted noun in the matrix clause is determined by the semantic role this participant plays in the embedded cleft clause, as clearly shown by (25). In particular, the PP to the clerk needed to introduce this participant into the matrix clause is motivated by the syntax–semantics interface of the embedded clause. This participant is the third argument –Beneficiary – of a three-place predicate (give), therefore, it is a non-macrorole argument that can thus be expressed by a to-PP. There is no motivation for the formal properties of this argument (i.e. it being a prepositional phrase) in the matrix clause. (25)

It was to the clerk that I gave my application.

Clefts are a primary example of a syntactic structure that is not entirely semantically motivated. Indeed, we can speak metaphorically of a syntax– semantics interface that is twisted for the expression of a particular kind of information structure. This is a point that might contribute to our understanding of linguistic systems. At every moment speakers are faced with the daunting task of capturing new situations and this means that every context requires to some degree novel linguistic responses. The lexicon and the rules that regulate the syntax–semantics interface are highly structured and, yet, clefts show that they are flexible enough to accommodate to the demands of a particular speech situation. We can think of the communicative situation as the encounter of two vectors coming from opposite directions in a sort of ‘pincer effect’: a structuring force that holds the pieces together and the need to adapt to the particular demands of a context. Clefts show that the supposedly more rigid part of the system, syntax, has built-in templates specifically designed to serve information structure, one of the subsystems that deals with adjustment to the dynamics of contexts. There is no part of grammar that is immune to the demands of communication. Clefts allow speakers to highlight the particular insertion of the content contributed by an isolated element – the clefted constituent – into the

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

603

604

LUIS PARÍS

incomplete content of the cleft clause. The single proposition has been divided into an open proposition – a representation that comes with an open slot or variable – expressed by the clefted clause and a matrix clause that provides the value for that variable. The cleft clause is part of the presupposition and, hence, outside the potential focus domain in a strong sense: neither its internal constituents taken individually nor the entire clause as a unit can be the focus. Example (21), repeated in (26a) for convenience, cannot be a felicitous answer to the question What happened?. The potential focus domain only has scope over the matrix clause as confirmed by focus sensitive-particles. The negative operator in (26b) ranges only over the matrix clause and, consequently, on the co-indexation of the teacher as the Actor of the embedded clause. (26)

a. It was the teacher who blamed the bus driver b. It wasn’t the teacher who blamed the bus driver.

Further, a yes/no question can only ask for the clefted RP. Was it the teacher who blamed the bus driver? can be answered by (26a) whereas Was it the bus driver who was blamed by the teacher cannot. The appropriate context for an itcleft sentence is a contrastive one; it had been asserted that somebody else other than the teacher (for example, a student) had accused the bus driver. This contrastive focus typically conveys an exhaustive reading of the clefted RP; namely, it was the teacher and nobody else who accused the bus driver (Kiss 1998). Languages like French have constraints preventing PSAs – subjects – to be part of the actual focus domain in a sentence focus structure (SF). Since French also has rigid SVO word order, preverbal elements cannot normally be in focus. In consequence, French might use clefts as in (27) or presentational clefts as in (28) as pragmatically felicitous in a context that demands SF (Lambrecht 1994: 226, 2000: 653; Van Valin 1999: 519). (27)

J’ai ma voiture qui est en panne. I have.prs.1sg my car that be.prs.3sg in breakdown ‘My car broke down.’

(28)

Y a Jean qui a téléphoné. there have.pst.3sg Jean who have. pst.3sg call.ptcp ‘Jean called.’

These sentences can be a felicitous answer to a What happened? type of question. The rather fixed French word order and the pragmatic restriction of subjects as topics motivates the use of clefts. In RRG the presence of a bi-clausal structure like the it-cleft raises the issue of nexus and juncture linkage. ‘Nexus’ refers to the nature of the dependency between the two clauses, specifically, whether one entirely depends on the other (subordination) or they are independent of each other (coordination) or, finally, they are co-dependent (cosubordination). In addition, the theory of ‘juncture’ individuates the layer in which the

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

combination occurs: nuclear, core or clausal (see Chapter 13). Since clefts contain an obligatory co-indexation relation among members of different clauses, the corresponding juncture is at the core layer. Regarding nexus, we can rule out daughter subordination since the subordinate clause does not fill an argument slot in the matrix clause and the co-indexation relation might involve adjuncts in the embedded clause. Furthermore, the embedded verb inflexion agrees with the wh-form and the clefted element in the matrix clause does not control agreement. This can be seen in Spanish where the antecedent noun is a first-person pronoun while the embedded verb agrees with the third person that corresponds to the embedded pronoun (Pavey 2004: 29). (29)

Fui yo el que chocó el auto. be.pst.1sg I the.m.sg that crash.pst.3sg the.m.sg car ‘It was me who crashed the car.’

This should be another piece of evidence that supports a core peripheral subordination in a core juncture analysis. In the RRG account, the nucleus of a clause with a copula is the attributive expression, in this case, the RP. The embedded clause is a modifier of a core with a nominal nucleus. In this way the similarity between it-clefts and RCs is also captured (see Figure 15.5). The embedded clause in it-clefts is a peripheral modifier at the core layer and, hence, cannot be part of the focus. There are some exceptions to this, but they are other subtypes of clefts. For instance, there-clefts like the one in (30) might be appropriate answers to What happened? (30)

There was one man that kept interrupting. (Huddleston 1984: 489; Pavey 2004: 64)

This is an existential cleft with a lightly informative antecedent noun; the cleft clause might be part of the focus (Bentley et al. 2015). Another example is offered by it-clefts with ‘this’ or ‘that’ in place of the ‘it’ pronoun, like (31). The information in the matrix clause retrieves already known information SENTENCE CLAUSE

RP

CORE

PERIPHERY

NUC

CLAUSE

PRED AUX

PrCS

RP

It was the teacher who

CORE NUC blamed

RP the bus driver.

Figure 15.5 Cleft sentence

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

605

606

LUIS PARÍS

so that the assertion falls directly on the identity relation between the copula and the cleft sentence, which is the focus of the assertion.8 (31)

a. That’s the reason I don’t want to go to Miami! b. Yeah. Wasn’t that somewhere in Southern Florida where they thought those people got AIDS from bug bites. . . (Hedberg 2000: 902; Pavey 2004: 32)

Spanish does not have It-clefts but it does have a construction that is structurally similar to the English one. It places the focus on the cleft constituent in a clause with a copula followed by an embedded clause introduced by a relative pronoun. (32)

Fueron los niños los que pidieron leche. Be.pst.3pl the.pl child. pl the.pl that ask.pst.3pl milk ‘It was the children who asked for milk.’

A difference with English is that in the Spanish cleft, the clefted noun might be the subject of the matrix clause – as in (32) – since it controls the agreement with the verb inflexion (i.e. the copula). Hence, the absence of an ‘it’ pronoun cannot be attributed to the Spanish ‘pro-drop’ status. Despite the variety of relative clauses existing in Spanish, here the relative pronoun has to be of a certain kind – that is, lo que (‘the.m.sg that’), la que (‘the.f.sg that’), los que (‘the.pl that’) and so on. As in English, the morphosyntactic realization of the antecedent noun is determined by the meaning of the embedded verb. In (33) the subject is impersonal and the antecedent is the nominal element of a non-predicative PP that marks indirect objects. (33)

Fue a los niños be.pst.3sg to the.pl child.m.pl que les dio la that them give.pst.3sg the.f.sg ‘It was to the children to whom he gave the milk.’

a to leche. milk

los the.m.pl

Furthermore, if the co-indexed argument plays the role of an adjunct in the cleft clause, it should be preceded by the relevant predicative preposition, as shown in (34). (34)

Fue en el baño donde encontré la llave. Be.pst.3sg in the.m.sg bathroom where find.pst.1sg the.f.sg key ‘It was in the bathroom where I found the key.’

In spite of being syntactically embedded, the verb LS of the cleft clause determines the morphosyntactic realization of the matrix participant. This is a key property that justifies a different analysis of RCs and clefts. The syntax–semantics linking in clefts displays properties of an ‘inverted’ interface. Why is it, then, that the cleft clause is not represented as the matrix clause? The presence of a subordinator – either a relative pronoun or a complementizer – is the crucial piece of evidence for the embedding of the cleft clause. In addition, an antecedent that controls the reference of a

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

pronominal expression typically occupies the syntactic dominant position, which in this case is the matrix clause.

15.3.1 Linking Cleft Sentences I shall follow Pavey (2004: 215) in assuming that the LS that corresponds to itclefts is based on the be′ predicate structure that expresses the meaning of clauses with a copula, as shown in (35). This predicate captures the specificational nature of the predication, one of the meanings that can be expressed by the copula in English and Spanish, the one that corresponds to clefts. In particular, the specificational meaning highlights the ‘exhaustive reading’ in the sense that it was the teacher and nobody else who blamed the bus driver.9 (35)

be′([INGR blame′(who1, bus driver)], teacher1) ----------

There is an important assumption underlying (35) which is that LSs are not merely registering propositional contents (interpreted either as a cognitive state or as truth-conditional content). In fact, if this were the case, this LS should also correspond to The teacher blamed the bus driver since, as I pointed out earlier, this sentence and the one in (21), (26a) have the same propositional value (see also Lambrecht 1994: 22). On the contrary, it is assumed that LSs should also be sensitive to the way symbols are combined which, in addition, highlights the systematicity of the syntax–semantics–pragmatics interface. The salient property of clefts rests on the satisfaction of communicative needs. The LSs of clefts should be consistent with the fact that a speaker is interchanging information with an addressee who is in a different cognitive state. It-clefts are nested in a communicative context where one interlocutor had asserted that an individual ‘x’ was the Actor of a blaming event. The speaker utters the itcleft to contradict such a statement and offer a different identity for that participant.10 The syntax-to-semantics linking proceeds as follows. The clause with the copula demands that we retrieve a specificational structure and the LS of the verb in the embedded clause. Since it has no recoverable referential value, the argument slots of the specificational LS are filled in by the LS of the embedded clause and the full RP. This RP is the specificational element; hence, it has to fill in the second argument position. Then, we proceed to coindex the wh-pronoun in the embedded verb LS with the specificational RP. If there is a complementizer, then, the co-indexation involves the unspecified argument slot of the embedded verb LS. Finally, the bus driver is an RP in postverbal position. The Undergoer is, in turn, linked to the lowest-ranked argument in the verb LS. The highest-ranked argument is linked to the remaining wh-word or, if there is no available wh-word, it remains syntactically unexpressed.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

607

608

LUIS PARÍS

15.4

Contrasting Clefts and Relative Clauses

There is a sense in which the central meanings of it-clefts and RCs are equivalent in terms of truth conditions. Specifically, in the sentence The student that failed the test quit the program, the truth-conditional content of the RC can also be expressed by a simple sentence (The student failed the test) and this exact content can be expressed by an it-cleft (It was the student who failed the test). The semantic difference in truth-conditional terms is reduced to the minimum; it-clefts add to the interpretation of the relevant noun the further assumption that the student was the only member who performed the action (exhaustive reading). The bulk of the semantic contrast lies in the way the semantic representation is assembled and these different assemblages are the encodings of different communicative functions. Speakers use RCs to provide further information that will help hearers to identify the referent of a head noun. It-clefts, instead, highlight the identity of one participant with a particular role in one event. RRG captures this divergence through an independently motivated distinction. It is assumed that the be′ predicate includes different kinds of predications (Van Valin 2005: 48), even though they might all be expressed via a copula verb in a particular language. Two of them are relevant here; one is attributive predication and the other is specification (Declerck 1988: 47; Pavey 2004: 29), represented in (36) and (37), respectively. The former states a predication relation between a predicate and a referential expression. It subsumes attributive adjectives and, thus, the relation between relative clauses and their head nouns. In contrast, the specification relation coindexes two referential expressions; one might be a constant (i.e. Chris) and the other is a definite description (the winner), a representation that allows us to identify a unique individual in the world. (36)

a. Pat is tall a′. be′(Pat, [tall′])

(37)

a. Chris is the winner a′. be′(Chris, [the winner])

For the purpose of representing it-clefts, the definite description of structures like (37a′) is replaced by the clefted noun. The LS that captures the meaning of the sentence is thus shown in (38). (38)

be′([INGR blame′ (who1, bus driver)], teacher1). ----------

It is the head noun that provides the specifying information and, hence, it plays the role of the predicative element. Indeed, it is predicated of the embedded logical structure (i.e. ‘. . . blame′ . . .’) which constitutes the meaning of the embedded clause. The element that receives this predication is the co-indexed argument in the embedded verb LS (i.e. the wh-form). In contrast, the matrix verb of the relative clause in RC provides the LS for the whole sentence, as can be seen in (10), which is repeated below in (39).

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

The predicative structure (i.e. be′) that corresponds to the semantic relation between the head noun and the relative clause is inserted into one of the matrix LS argument slots (i.e. the first argument of quit′).The participant that fills in a semantic slot is literally shared with the predicative LS. This is not a typical co-indexation relation but a unique slot that belongs to two LSs, which is shown with a thick broken underline, as was pointed out above. In addition, this position is co-indexed with the argument slot filled by the whword in the relative clause LS. (39)

INGR quit′([be′ (student1, [INGR pass′(who1, test)])], program) ----------

The site of the syntactic contrast resides in the linkage of the embedded clause with the matrix clause. It-clefts involve a direct relation between two clauses, while the relative clause in RC acts as a modifier of the head noun and it is embedded into an RP in RC. This difference is attested, for example, by the different scope of the universal quantifier in (40) and (41) (Davidse 2000: 1114, cited in Pavey 2004: 193). (40)

It was all the passengers who had committed the murder.

(41)

All the students who attended will receive a bonus point.

In (40) the quantifier has scope over only the clefted noun, and the coindexed pronoun in the cleft clause inherits this interpretation. The full RP denotes the set of ‘all passengers’ and all of them committed the murder. In other words, the cleft clause does not affect the interpretation of the noun phrase in the matrix clause. By contrast, in (41), the quantifier has scope over the RP the students, which has already been restricted by the relative clause. Therefore, the relevant set does not include all the (contextually possible) students but only the subset of them that attended. Indeed, it is implicated that some students did not attend the class. The relative clause restricts the set of the noun to a subset and it is only to this subset that the universal quantifier applies. These properties can be derived naturally by representing relative clauses within the RP whereas cleft sentences are outside the RP, as shown in Figure 15.6. With respect to information structure, neither the relative clause in RC nor the cleft clause in it-clefts are part of the assertion, but are instead presupposed. They both carry a proposition with a variable, namely, an open proposition. However, these embedded clauses differ in subtle ways. The relative clause helps the interlocutor to identify the reference of the head RP, which is necessarily a non-identifiable participant. In contrast, the

Det

Noun embedded clause It-clefts

Det

Noun embedded clause Relative Clause

Figure 15.6 The contrast between cleft sentences and relative clauses

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

609

610

LUIS PARÍS

reference of the clefted noun could be identifiable or not, but it receives an exhaustive reading and, crucially, its pragmatic function as focus is to saturate the open proposition introduced by the embedded sentence. The cleft clause provides the event in which this entity was involved. Understanding focus in Lambrecht’s sense, the embedded clause in clefts carries information that is necessarily part of the focus. In the example we have been analysing, this is the semantic role ‘Actor’ in the blaming event of the teacher (clefted noun). The teacher is the focal element but the focus is not only the identity of the teacher but also its role (Actor) in the event described by the embedded clause. Let’s call the kind of open propositions carried by embedded clauses in clefts ‘operative presuppositions’ since they carry information that is necessarily in focus. The embedded clause in RCs also carries an open proposition, but it does not function in the same way in information structure. Its information helps in the identification of a participant that might not be in focus at all. In sum, RCs are related to the issue of identifying a referent whereas clefts are centred on focus structure and, consequently, their respective embedded open propositions have different functions: one is an ‘operative presupposition’, whereas the other is just a plain presupposition. Their syntactic templates are consistent with their different informative roles since the relative clause is embedded into the RP, whereas the cleft clause is a modifier of the matrix core and it does not play any role within the RP. In some sense, any sentence with a pronominal form is associated with an open proposition. The pragmatic notion of open proposition that is relevant here is that of an open proposition that is specified by the focus. The embedded clauses of both constructions are open propositions, but only the cleft clause is an open proposition in a pragmatically relevant way, namely, it is the presupposition specified by the element in focus. One condition that makes it possible for the relative clause in RC to be presupposed, even though it carries specifying information, is to be found in its syntactic layout. While the cleft clause is outside the clefted RP, the relative clause is a modifier within the RP in RC. The embedding of the relative clause in the RP leaves it out of the potential focus domain in a weak sense, that is, it can be the focus as a whole, but its internal constituents cannot be in focus on their own. This is shown in (42a), where the wh-form targets the whole RP in (42b). (42)

a. Which student quit the program? b. The one that failed the test.

Another contrasting behaviour related to the inherent pragmatic functions of the constructions is that the head noun in it-clefts can be a proper noun whereas this is not the case with RCs, as shown in (43) and (44), respectively. (43)

It was Peter who sold the house.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

(44)

*Peter who sold the house came to see you.

In sum, the relative clauses in RC can be the focus as a whole because they are nuclear modifiers of the RP, which might be within the potential focus domain of the sentence. If the relative clause modifies a noun head that is outside the potential focus domain – for example, an RP within an adverbial clause in a clause juncture – the relative clause loses the possibility of being part of the focus.

15.5

Conclusion

The study of RCs and clefts is enlightening in different ways and at different levels. From a very abstract perspective, RC is a primary example of the need for phrases in any grammatical theory. It is a structure – not just a mere concatenation– that results from a meaningful combination of symbols. The relative clause in RC is semantically and syntactically independent from the matrix clause because it is embedded within a noun phrase. The RP is an enclosure that fosters relations among the symbols inside, preventing them from having certain relations with symbols outside. In addition, RC reveals the structural richness of human languages. As defined here, relative clauses are embedded into noun phrases as peripheral modifiers of the nucleus. It turns out that different subordinations are possible even in a single language, through a relative pronoun or no subordinator at all. Furthermore, a typological perspective instructs us about different structural conditions for the antecedent–pronoun coindexation relation. In the typologically more frequent externally headed relative clause (Van Valin 2005; Comrie 1998), the antecedent noun is in the matrix clause whereas the referentially dependent form or argument slot belongs to the dependent clause. In internally headed relative clauses the situation is the opposite, in that the full noun is part of the dependent clause whereas the argument slot that takes its value from that noun is in the matrix clause. Clefts too illustrate very general properties of linguistic systems. The perspective I have taken here to describe clefts bears on the question: how deep does information structure carve into the grammatical system? In other words, the issue is whether a system that is designed to satisfy communicative needs operates at the grammatical core. A functionalist linguist would predict that information structure goes all the way down to the centre of the grammatical system, and this is precisely what it-clefts show. The matrix clause does not carry the semantically rich content of the sentence, which is in fact contributed by the embedded clause. This embedded clause determines even the morphosyntactic realization of the attributive expression (typically a noun phrase but also a prepositional phrase) in the matrix clause. This is what, in this chapter, has been called ‘inverted’ linking. However, the matrix clause conveys the focal element of the

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

611

612

LUIS PARÍS

assertion, that is, it provides the salient information structure content. In short, information structure reverses the dependency relation that would be predicted from mere semantic considerations. The contrast between clefts and RCs highlights how two structures that share striking similarities can be at the same time extremely different. Both clefts and RCs contain an embedded clause that can be introduced by a relative pronoun. Both of them hold a co-indexation relation between an element in the matrix clause and another in the embedded clause. However, the relative clause in RC is a modifier of the antecedent noun, whereas clefts involve an interclausal relation such that the cleft clause is dependent on the matrix clause; it is in the periphery of the matrix core. This analysis explains among other facts the different scope of quantifiers in the matrix clause. Their different structural shape is systematically tied to their different communicative function. The function of RCs revolves around the identity of the participant expressed by the head noun. The listener has to identify it with the help of the information carried by the embedded clause, information that qualifies as a presupposition. The function of clefts is centred on information structure. The matrix clause conveys a participant whose identity, together with the role it plays in the event introduced by the embedded clause, constitutes the focus of the sentence. In this sense, the embedded clauses in RCs and clefts both carry open propositions but the one in RC counts as a plain presupposition while the one in clefts is an ‘operative presupposition’, which means that part of it is in focus.

References Bentley, Delia, Francesco Maria Ciconte and Silvio Cruschina. 2015. Existentials and Locatives in Romance Dialects of Italy. Oxford: Oxford University Press. Bird, Charles. 1968. Relative clauses in Bambara. Journal of West African Languages 5: 35–47. Chafe, Wallace. 1987. Cognitive constraints on information flow. In Robert Tomlin (ed.), Coherence and Grounding in Discourse (Typological Studies in Language 11), 21–52. Amsterdam: John Benjamins. Cinque, Guglielmo. 2015. Three phenomena discriminating between ‘raising’ and ‘matching’ relative clauses. Semantics–Syntax Interface 2(1): 1–26. Citko, Barbara. 2004. On headed, headless and light-headed relatives. Natural Language and Linguistic Theory 22(1): 95–126. Comrie, Bernard. 1998. Rethinking the typology of relative clauses. Language Design 1: 59–86. Davidse, Kristin. 2000. A constructional approach to clefts. Linguistics 38(6): 1101–1131. Declerck, Renaat. 1988. Studies on Copular Sentences, Clefts and Pseudo-Clefts. Berlin: De Gruyter Mouton.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

Hedberg, Nancy. 2000. The referential status of clefts. Language 76(2): 891–920. Huddleston, Rodney. 1984. Introduction to the Grammar of English. Cambridge: Cambridge University Press. Keenan, Edward L. 1976. Remarkable subjects in Malagasy. In Charles Li (ed.), Subject and Topic, 247–301. New York: Academic Press. Keenan, Edward L. and Bernard Comrie. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8: 63–99. Kiss, Katalin. 1998. Identificational focus versus information focus. Language 74(2): 245–273. Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representations of Discourse Referents. Cambridge: Cambridge University Press. Lambrecht, Knud. 2000. When subjects behave like objects: An analysis of the merging of S and O in sentence focus constructions across languages. Studies in Language 24(3): 611–682. Pavey, Emma L. 2004. The English It-Cleft Construction: A Role and Reference Grammar Analysis. Unpublished dissertation, University of Sussex. Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure and syntax. In Ekatarina Raxilina and Jakov Testelec (eds.), Typology and the Theory of Language: From Description to Explanation, 511–524. Moscow: Languages of Russian Culture. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2012. Some issues in the linking between syntax and semantics in relative clauses. In Bernard Comrie and Zarina EstradaFernández (eds.), Relative Clauses in Languages of the Americas: A Typological Overview, Typological Studies in Language 102, 47–64. Amsterdam: John Benjamins. Van Valin, Robert and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Notes 1 Keenan and Comrie (1977) propose a universal hierarchy of RP functions where the highest-ranking position represents the one most accessible to license a relative clause. (i)

Subject > DO > IO > OBL > GEN(itive) > O(blique)COMP

The hierarchy should also be interpreted implicationally, namely, that every language that accepts relative clauses for an RP down the hierarchy, must also license RCs for higher-ranked positions in the hierarchy.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

613

614

LUIS PARÍS

2 I do not call ‘where’ an outright pronoun because its function in the RC is adverbial, even if its antecedent might be unequivocally a noun (‘I know the bar where you met Sally’). 3 I use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php) plus the following addition: ATV stands for active voice. 4 The only alternative semantic representation that I can think of would be one in which the two LSs are coordinated and their shared argument coindexed. (i)

INGR quit′ (student1, program) ^ INGR fail′ (student1, test)

This representation is truth-conditionally equivalent to (10), but is less influenced by the syntactic realization, and, crucially, it does not represent the attributive relation. 5 It might be appealing to attempt an analysis of the Bambara IHRC that assigns a different semantics to it so that we can maintain the pattern of a hierarchical relation between antecedent and co-indexed element. In particular, an LS that obviates the attributive relation and simply relates the meaning of the two clauses through coordination, as in (i). (i)

[see′ (ne, soi)] ^ [do′ (tye, Ø)] CAUSE [BECOME have′ (tye, xi)]

However, as was mentioned earlier, I contend that relative clauses are semantically different from their pseudo-paraphrases in which coordination replaces the RC. 6 Realizing the pronoun with an accusative shape turns the sentence ungrammatical (pret ¼ preterite). (i)

*Al que invitaste llegó To.the that invite.pret arrive.pret ‘The one to whom you invited came late.’

tarde. late

7 Citko (2004) labels the fact that this pronoun might have a case consistent with one clause (the matrix) but not with the other (the embedded clause) as a ‘non-matching’ property. She claims that ‘non-matching’ pronouns are characteristic of ‘light-headed relative clauses’ and that sentences like the one in (19) fall under this label. Her analysis of light-headed relatives is couched in the Minimalist framework (she claims there is an empty node being occupied by the raised pronoun). I prefer to maintain the specific label ‘pronounless relative clauses’ in order to highlight the differences between the two proposals. 8 Certain construction types might fall under the overarching notion of clefts but they may instantiate a different information structure pattern. In clefts, the embedded clause is not the focal element. However, pseudocleft sentences involving a presentational clause of the type in (i) contain a thetic or sentence-focus structure. The embedded clause is, thus, within focus (Lambrecht 2000, Bentley et al. 2015). (i)

There was a man that came to my house.

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

Cleft Sentences and Relative Clauses

9 The other possible interpretations of the copula are ‘predicative’, ‘identificational’ and ‘equative’ (Van Valin 2005: 48). 10 Certainly, ‘blame’ is ambiguous. The interpretation that is targeted here is based on example (21), (26a) and it has the sense of a speech act verb, an act that lasts an instant even if it has lasting consequences. Under this interpretation, it cannot be an Accomplishment or an Activity (The teacher blamed the bus driver *in/??for a month). Another interpretation of this verb targets a mental state that can last for a possibly unlimited period of time. This interpretation is especially suitable when used in the present (John blames his parents for his academic failure).

https://doi.org/10.1017/9781316418086.019 Published online by Cambridge University Press

615

16 Extraction Restrictions in Complex Sentences Mitsuaki Shimojo

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ASP DES EVI FP IU PFD

aspect desiderative evidential sentence-final particle information unit potential focus domain

16.1

Introduction

PrCS PrDP PoCS PoDP RP

pre-core clot pre-detached position post-core slot post-detached position reference phrase

For nearly five decades, extraction restrictions, also known as island constraints, have been known to be an issue that every syntactic theory should address, but they have remained elusive.* The elusiveness is rooted in the intricate nature of the phenomena, which are not only language- and construction-specific but also related to non-structural grounds such as lexical-semantic and pragmatic factors, which influence extractability. This chapter lays out the Role and Reference Grammar (RRG) approach to extraction restrictions (Van Valin 1996; Van Valin and LaPolla 1997) and shows how cross-linguistic and language-internal (i.e. cross-constructional) variations are captured by the theory. In regard to this purpose, this chapter also outlines a case study from Japanese, which is radically different from English-type languages due to its relatively restriction-free characteristics. Japanese exhibits restrictions in both

* I thank Delia Bentley and Robert D. Van Valin, Jr. for their valuable comments on earlier versions of the chapter.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

617

in-situ and ex-situ constructions, and therefore, serves as a useful ground for the RRG applications. Since Ross’s (1967) discovery of the phenomena, syntactic constraints on extraction, including those given in (1), have drawn much attention and there has been further reformulation of the original proposal, beginning with Chomsky (1973). In the course of the evolution of the Chomskyan theory of syntax, island constraints have been subsumed under the general principle of subjacency, which is stated in (2). (1)

(2)

a. b. c. d. e. f. g. h.

Jon believes [NPthe rumour [Sthat Mary lost her phone]]. *What does Jon believe [NPthe rumour [Sthat Mary lost __ ]? Jon talked to [NPthe neighbour [Swho bought the house recently]]. *What did Jon talk to [NPthe neighbour [Swho bought __ recently]]? Jon ate a sandwich after [SMary made an omelette]. *What did [SJon eat a sandwich after [SMay made __ ]]? [SThat Jon ate the sandwich] was obvious. *What was [S[Sthat Jon ate __ ] obvious]?

(complex NP) (complex NP) (adjunct) (subject)

Subjacency condition (Chomsky 1973) Extraction cannot cross more than two bounding nodes (NP, S) in a single movement.

In this formulation, it has been assumed that these constraints are purely structural with no reference to a semantic, pragmatic, or other basis. Therefore, they have been taken as evidence of autonomous syntax and ultimately of a theory of universal grammar. While there has been much theory-internal debate with respect to the status of the subjacency condition (see Yoshimura 1992 and references cited therein), there have also been proposals to account for the constraints, external to the syntactocentric theories, such as semantic and pragmatic approaches (Erteschik-Shir 1973; Erteschik-Shir and Lappin 1979; Haig 1996; Kuno 1987; Kuno and Takami 1993; Shimojo 2002; Van Valin 1996; Van Valin and LaPolla 1997), cognitive and performance-based proposals (Deane 1991, 1992; Hofmeister 2007; Kluender 1990; Kluender and Kutas 1993), and parsing and expectation-based claims (Chaves 2013). In short, the subjacency accounts are problematic because the constraints are sensitive to a range of factors beyond pure syntactic grounds. Furthermore, for a number of theories, the subjacency account is problematic since it assumes movement, which is a theory-internal assumption, and therefore incompatible with monostratal and non-transformational theories such as RRG. In addition, there is difficulty with a movement-based approach in accounting for constraints which do not involve movement without added theory-internal stipulations. Lakhota is a language that does not favour the subjacency account. Van Valin (1996: 36–37) points out that Lakhota blocks wh-questions formed within a complex NP despite its wh-in-situ characteristics. A sentence containing a word such as táku or tuwá and the question

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

618

MITSUAKI SHIMOJO

marker he in Lakhota is ambiguous between wh-question and yes/no question interpretations because the words are ambiguous between wh-word and indefinite-specific pronoun readings (3a). However, the ambiguity is resolved with complex reference phrases (RPs) and adverbial clauses. As shown in (3b), the wh-question reading is unacceptable, and táku in the adverbial clause can be interpreted only as ‘something’; thus, this is analogous to the subjacency constraint observed in English. (3)

[Tuwá thaló ki manú] iyúkˇ cą he? who/someone meat the steal think q ‘Who does he think stole the meat?’ or ‘Does he think someone stole the meat?’ b. [Wiˇ cháša ki táku yúte] eˇ chúhą, tha-wíˇ cu man the *what/something eat while his-wife ki mní ikíˇ cíˇ cu he? the water get.for q ‘While the man was eating something, did his wife get him water?’ *‘What did his wife get him water, while the man was eating?’ (Intended wh-question: ‘Did his wife get him water, while the man was eating what?’) (Van Valin 2005: 277)

a.

In order to account for the restriction in (3b), the movement-based account requires a movement at abstract levels so that subjacency applies to languages such as Lakhota (Chomsky 1986); however, such an account lacks a theory-external appeal. Furthermore, since Lakhota-type languages, which have wh-in situ, exhibit extraction constraints, movement or filler-gap dependency is not an essential feature to account for the restrictions. For this reason, Lakhota-type languages also pose a challenge to any approach which claims that processing difficulties or parsing preferences are rooted in filler-gap dependency as an underlying principle. Lastly, there are languages which do not obey subjacency yet exhibit construction-specific restrictions and/or constraints under certain semantic and pragmatic conditions. These languages include Chinese (Jin 2013), Danish (Erteschik-Shir and Lappin 1979; Jensen 2001), Korean (Hong 2003), Swedish (Allwood 1982) and Japanese (Haig 1996; Shimojo 2002). Japanese will be discussed at length later.

16.2

Role and Reference Grammar Accounts

RRG combines structural and pragmatic grounds to capture both languageinternal and cross-linguistic variations in extraction restrictions.1 The relevant pragmatic property is represented in the focus structure projection of a sentence, which indicates the domain of the sentence that represents or may represent the focus, that is, pragmatic assertion in the sense of Lambrecht (1994: 52). There are cross-linguistic variations as to how rigid or flexible a language is with respect to the placement of focus (Van Valin 1999). In Italian, for example, a preverbal non-contrastive focal RP is not allowed, and therefore, a focal subject RP needs to be postverbal (i.e. word

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

ordering is relatively flexible). English exhibits the opposite pattern with a relatively rigid SVO ordering and flexible placement of focus within the sentence; either the subject or the object can be focal in the same word order (see Chapter 11 in this volume). With respect to focus structure for complex sentences, the cross-linguistic concern is to what extent the focus domain is extended into subordinate clauses. Some languages such as Polish restrict the focus domain to matrix clauses only, and in other languages such as English, the focus domain includes certain types of subordinate clauses (Van Valin 1996). In RRG, language-internal variation in terms of the focus domain is described structurally as stated in (4). (4)

The potential focus domain in complex sentences The potential focus domain extends into a subordinate clause if and only if the subordinate clause is a direct daughter of (a direct daughter of ) the clause node which is modified by the illocutionary force operator (Van Valin 2005: 275).

For example, an object complement clause is a direct daughter of a clause node; therefore, the potential focus domain (PFD), where a focal element may appear, extends into the subordinate clause. In contrast, this is not the case with an adverbial subordinate clause, which is in the periphery to the core, and only the subordinate clause as a whole, which is a constituent of the matrix clause, may function as an information unit. Figures 16.1 and 16.2 show the contrast between the two types of subordination with respect to the scope of the PFD, which is represented by the dotted line. SENTENCE CLAUSE CORE RP

NUC

RP

PRED

CLM

CLAUSE

V Kim

told

CORE Pat that

RP

NUC

PERIPHERY PP

PRED

ADV

V she will arrive IU

IU

IU

[IU

IU

at the party

late.

IU

IU]

IU

SPEECH ACT

Figure 16.1 Potential focus domain in clausal (daughter) subordination (Van Valin 2005: 214)

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

619

620

MITSUAKI SHIMOJO

SENTENCE CLAUSE CORE RP

Pat

NUC

PERIPHERY RP

PP

PRED

NUC

CLAUSE

V

P

CORE

saw

Kim

RP

NUC

PP

PRED V after IU

IU

she

arrived at the party

IU

IU

SPEECH ACT

Figure 16.2 Potential focus domain in ad-core subordination (Van Valin 2005: 216)

In both types of subordinate clause shown in the figures, the subordinate clause as a whole can be an information unit (IU), which can be focal, as in the case of an answer to ‘what did Kim tell Pat?’ and ‘When did Pat see Kim?’. However, the ad-core subordinate clause in Figure 16.2 does not allow internal constituents to be distinct information units because they cannot be the focus individually. This is shown in the question–answer pairs in (5). The unacceptable answer in (c) shows that the internal constituents of the subordinate clause are outside of the PFD. (5)

Q: Did Pat see Kim after shei arrived at the party? A: a. No, Sally. b. No, before. c. *No, shei left.2

The cross-linguistic and language-internal observations in terms of focus domains correlate with extraction restrictions. First, languages like Polish, which restrict the focus domain to matrix clauses, prohibit extraction out of tensed subordinate clauses; therefore, extraction is a matrix phenomenon in these languages. On the other hand, English-type languages allow extraction out of a subordinate clause within the PFD. Thus, extraction is possible out of an object complement clause (e.g. where did John tell Mary that he will arrive __?), but not out of an adverbial subordinate clause (e.g. *where did John see Mary after he arrived __?). In other words, an extraction site must be within the PFD and this is captured by the principle stated in (6).

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences (6)

General restriction on extraction constructions (Van Valin 1996: 54) The displaced element (or the in situ question word in a language like Lakhota) must be linked to an argument position in the semantic representation of a clause within the PFD of the illocutionary force (IF) operator.

From a cross-linguistic perspective, extractability can be captured in terms of the scale given in (7). Languages like Polish allow extraction only from a matrix clause and this is the most inflexible type. Languages like English and Lakhota are more flexible, allowing extraction from certain types of subordinate clause. The scale would be complete with another type which allows extraction out of a greater range of subordinate clauses. Japanese is of this type and the following section discusses how the RRG principle given in (6) applies to Japanese. (7)

16.3

Range of extractability: cross-linguistic variations Matrix clause only Subordinate clauses: direct daughter of clause only More subordinate clauses

Polish . . . English, Lakhota . . . Danish, Japanese, Swedish . . .

Application of the RRG Analysis to Japanese

The cross-linguistic variations outlined thus far raise a question about the validity of the RRG principles in a language in which extraction is allowed more freely, and therefore, this section discusses extraction restrictions in Japanese. Studies on extraction constraints have been centred on English-type languages, in which extraction constraints are more or less structurally predictable. In relatively island-free languages, restrictions are typically described on a semantic and pragmatic basis, and Japanese is no exception (see the references cited for these languages in Section 16.1).

16.3.1 Possible Extraction in Complex Sentences The possible range of extraction in Japanese is exemplified by the following three major extraction constructions: wh-question, relativization and topicalization. (8)

Wh-question formation in a relative clause seefu-wa [dare-ga sekkee-shita] doroon-o koohyoo-shimashita ka government-top who-nom design-do.pst drone-acc release-do.pst q ‘The government announced the drone who designed?’

(9)

Relativization out of a relative clause [seefu-ga [ __ sekkee-shita] doroon-o koohyoo-shita] enjinia government-nom design-do.pst drone-acc release-do.pst engineer ‘The engineer who the government announced (s/he) designed the drone.’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

621

622

MITSUAKI SHIMOJO

(10)

Topicalization out of a relative clause sono enjinia-wa [seefu-ga [ __ sekkee-shita] that engineer-top government-nom design-do.pst doroon-o koohyoo-shita drone-acc release-do.pst ‘The engineer, the government announced (s/he) designed the drone.’

Furthermore, the extensive range of possible extraction is exemplified by the following wh-questions formed in different types of subordinate clause: a noun complement, a sentential subject and an adjunct clause (relativization and topicalization out of these subordinate types are also possible; see Shimojo 2002). (11)

Wh-question formation in a noun complement jon-wa [dare-ga doroon-o sekkee-shita] toiu joohoo-o Jon-top who-nom drone-acc design-do.pst comp information-acc utagatte-imasu ka doubt-asp q ‘Jon doubts the information that who designed the drone?’

(12)

Wh-question formation in a sentential subject [dare-ga doroon-o sekkee-shita koto]-ga yosoogai deshita ka who-nom drone-acc design-do.pst nmlz-nom unexpected cop.pst q ‘It was unexpected that who designed the drone?’

(13)

Wh-question formation in an adjunct clause [paatii-ni dare-ga kita toki] jon-ga deteikimashita ka party-dat who-nom come.pst when Jon-nom leave.pst q ‘Jon left when who came to the party?’

16.3.2 Focus Domain in Complex Sentences The focus-based RRG principle discussed in Section 16.2 predicts the association between the focus domain and a possible extraction site. This prediction is borne out in Japanese because the potential focus domain covers the individual constituents of a subordinate clause regardless of the principle given in (4). This is demonstrated by the following tests. Only the asserted part of an utterance can be interpreted as being negated. If the constituent can be negated in a conversational exchange, then it is a possible focus (i.e. in the potential focus domain). The example in (14) shows the negation test applied to a relative clause. The possible negation of the information expressed by the relative clause shows that the information may be interpreted as part of the assertion. (14)

A: seefu-ga [doroon-o sekkee-shita] enjinia-o koohyoo-shita government-nom drone-acc design-do.pst engineer-acc release-do.pst ‘The government announced the engineer who designed the drone.’ B: iya, misairu da no missile cop ‘No, a missile.’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

As Van Valin and LaPolla (1997: 626) point out, in English, an adjunct whquestion is ambiguous in a complex sentence in which the PFD extends over the subordinate clause, as shown in (15a), because ‘when’ can be interpreted as modifying either the matrix clause or the object complement. However, ambiguity does not arise in a subordinate clause which is outside a PFD, as in (15b). (15)

a. Wheni/j did Skinner say __i that Krycek would be at the missile silo __j? b. Wheni/*j did Skully interview the witness __i who saw the alien spacecraft in the silo __j?

In contrast, the corresponding Japanese sentences are both ambiguous, as shown in (16), which indicates that both the object complement and the relative clause may be part of the assertion (i.e. within a PFD) in Japanese.3 (16)

a. itsu kuroichekku-ga kakunooko-ni iru-to sukinaa-ga itta? when Krycek-nom silo-loc exist-quot Skinner-nom say.pst ‘Wheni/j did Skinner say __i that Krycek would be at the missile silo __j?’ b. itsu kakunooko-de uchuusen-o mita mokugekisha-to when silo-loc spacecraft-acc see.pst witness-com sukarii-ga menkai-shita? Skully-nom interview-do.pst ‘Wheni/j did Skully interview the witness __i who saw the alien spacecraft in the silo __j?’

On the one hand, the PFD excludes the pre-detached position (PrDP) and the post-detached position (PoDP), which contain a sentence-initial topic and a sentence-final (i.e. post-predicative) topic respectively. On the other hand, the focus domain includes the pre-core slot (PrCS), which contains a sentence-initial narrow-focus (i.e. a narrow-focus subject or a pre-posed narrow-focus), and the post-core slot (PoCS), which contains a postpredicative focus. Figure 16.3 shows the layered structure of the clause with respect to the PFD. Japanese uses post-nominal markings that correlate with the information structuring of the sentence. The topic of a sentence, if any, is outside the PFD SENTENCE PrDP

CLAUSE PrCS

CORE RP

RP

PoDP PoCS

NUC PRED

SPEECH ACT

Figure 16.3 Layered structure of the clause and potential focus domain

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

623

624

MITSUAKI SHIMOJO

and it may be overtly marked with the topic marker wa. In contrast, if a nominative argument is marked with the nominative marker, it is within the PFD, whether it is narrow focus (PrCS) or part of a broad-focus (RP). The direct evidence for a detached position being outside of the PFD comes from the observation that a wh-question word cannot be topicalized, as shown in (17).4 (17)

dare-ga/*wa sono uchuusen-o who-nom/*top that spacecraft-acc ‘Who saw the spacecraft?’

mimashita ka see.pst q

Also, as shown in (18), it is slightly more awkward to negate information expressed by an element which is part of the topic than to negate information expressed by a non-topic (cf. (14)).5 (18)

A: [doroon-o sekkee-shita] sono enjinia-wa seefu-ni drone-acc design-do.pst that engineer-top government-dat koohyoo-sare-ta release-do.pass-pst ‘The engineer who designed the drone was publicized by the government.’ B: ?iya, misairu da no missile cop ‘No, a missile.’

The functional principle given in (6) predicts that an extraction site is not permitted within a topic phrase, which is outside the PFD, and this is indeed the case, as shown in (19). The sentence with the topicalized complex RP is not felicitous when no particular presupposition for the sentence is present (for example, when the sentence is given out of the blue). (19)

[dare-ga shuppan-shita] hon-ga/*wa who-nom publication-do.pst book-nom /*top ‘Do books that who published sell well?’

yoku ureru? well sell

This is also the case with relativization and topicalization out of a topicalized complex RP, as shown in (20) and (21) respectively. (20)

[[ __ shuppan-shita] hon-ga/*wa yoku ureru] publication-do.pst book-nom/*top well sell ‘The actress who books that (she) publishes sell well.’

joyuu actress

(21)

sono joyuu-wa [ __ shuppan-shita] hon-ga/*wa that actress-top publication-do.pst book-nom/*top ‘The actress, books that (she) publishes sell well.’

yoku ureru well sell

However, the topicalization of an extraction site is possible only when the sentence is intended to be overtly contrastive. In other words, the topic marking in the examples above would be possible if the question in (19) could be interpreted as contrastive with respect to books which someone

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

publishes as opposed to books which others publish, and if those in (20) and (21) can be interpreted as contrastive with respect to books as opposed to other things. In fact, the imposed contrastiveness associated with an extraction site further supports the principle in (6). Erteschik-Shir (2007) claims that contrast is contextually constrained to occur only if a contrast set is available, and because contrast is represented by singling out a subset of a whole to separate it from the remaining subset, the singled-out subset represents a subordinate focus, which is embedded in the matrix topic. An example is given in (22). ‘John’ is part of the presupposition for (22A); however, it is a newly singled-out member of the previously given set (hence it is focal). In the formal notation, a contrastive topic represents the subordinate focus structure, as shown on the right-hand side. In other words, a (matrix) topic which contains a focus simultaneously represents a contrast, and this corroborates the principle in (6) because a topic must be contrastive (i.e. focal) in order to contain an extraction site. (22)

Q: Tell me about your brothers John and Bill. A: JOHN is the smart one. Contrastive topic: [{Johnfoc, Bill}top ] top

A topic which allows an extraction site within cannot be in the detached position, which is outside the PFD; it must be within the PFD of the sentence as it contains a focus element, being placed within the clause, either coreexternally in the pre-core (or post-core) slot or core-internally.6

16.3.3

Extraction Restrictions in Complex Sentences

16.3.3.1 Relativization and Topicalization The possible range of extraction in Japanese has been outlined thus far, and now discussions of extraction restrictions are in order. Despite the high degree of extractability, the language is not free from restrictions (Kuno 1973; Haig 1996; Shimojo 2002). Examples of unacceptable topicalization and relativization are given in (23) and (24) respectively. (23)

*sono kuruma-wa keesatsu-ga [__ nusunda] otoko-o that car-top police-nom steal.pst man-acc ‘The car, the police arrested the man who stole (it).’

taiho-shita arrest-do.pst

(24)

*[keesatsu-ga [__ nusunda] otoko-o taiho-shita] kuruma police-nom steal.pst man-acc arrest-do.pst car ‘a car which the police arrested the man who stole (it).’

These cases of unacceptable extraction are not predictable by the focusbased principle in (6) since relative clauses in Japanese are within a PFD, as discussed earlier. The observed unacceptability is due to the aboutness condition that is not met in these cases. The condition stated in RRG terms is given in (25).

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

625

626

MITSUAKI SHIMOJO

(25) Pragmatic-aboutness condition on topicalization and relativization The sentence fragment following a topical element in the pre-core slot or a restrictive relative clause must be pragmatically interpretable as being about the pre-core slot element or the head noun. (Van Valin and LaPolla 1997: 627)

The constraints are observed when it is difficult to obtain the interpretation that the sentence or the relative clause is about the displaced entity. In other words, when a subordinate element is relativized or topicalized, the ‘intervening’ matrix clause as well as the subordinate clause (i.e. the extraction site) must be relevant to the displaced topic or the relative clause head. This general principle, which relates to the notion of relevance, is valid for extraction from an RP in English (Kuno 1987). (26)

a. Who did John write a book about? b. ?Who did John destroy a book about? c. ?Who did John lose a book about?

The examples in (26b, c) show degraded extractability. In these cases, it is difficult to obtain the interpretation that the sentence is about the displaced element. As Chaves (2013: 308) puts it, while the action of writing is immediately relevant to the book’s topic, destroying or losing the book is not. In other words, in these questionable cases, the displaced element is not relevant to the assertion made by the matrix clause. In RRG, this aboutness principle relates to the principle of focus domain given in (6) because the verbs in (26b, c) are informationally distinctive and therefore naturally draw the focus, preventing the object RP from being the actual focus (Van Valin 2005: 288–289). The aboutness condition is further demonstrated by the contrast between the two Japanese examples in (27). (27)

a. *[keesatsu-ga [ __ nusunda] otoko-o taiho-shita] kuruma (cf. (23)) police-nom steal.pst man-acc arrest-do.pst car ‘a car which the police arrested the man who stole (it)’ b. [[ __ kite-iru] yoohuku-ga yogorete-iru] shinshi wear-asp clothes-nom get.dirty-asp gentleman ‘a gentleman who the clothes that (he) is wearing are dirty’ (Kuno 1973: 239)

(27a) is unacceptable because the stolen car is not immediately relevant to the action of arresting per se (the ‘car’ in this case is more naturally interpreted as a police car used for the arrest, the vehicle that is directly relevant to the action of arresting). In (27b), on the other hand, the state of being dirty is immediately relevant to the person who is wearing the dirty clothes. Hence, this relativization is perfectly acceptable.7 Because Japanese allows the so-called gapless relative clauses, in which the head noun is relatable to the relative clause only semantically or pragmatically (Matsumoto 1997), the truncation diagnostic given in (28) serves as a useful means to see how relatable a head noun is to the matrix clause.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences (28)

The truncation test for aboutness (Shimojo 2002: 77) Relativization and topicalization out of embedded clauses are possible if the relative or topic construction is acceptable even without the embedded clause in which the extracted NP functions.

In essence, if the displaced RP is relevant to the intervening matrix clause, the intended relevance should be inferable or at least should not be contradictory even if the clause containing the extraction site is omitted. The truncated versions of (27a, b) are given in (29a, b) respectively. (29)

a. #[keesatsu-ga otoko-o taiho-shita] kuruma police-nom man-acc arrest-do.pst car ‘a car which the police arrested the man (who stole (it))’ b. [yoohuku-ga yogorete-iru] shinshi clothes-nom get.dirty-asp gentleman ‘a gentleman who the clothes (that (he) is wearing) are dirty’

The possible interpretation of (29a), the truncated version, is ‘a car which the police used in arresting the man’, and this reading contrasts with the intended reading, which is shown by the translation. On the other hand, the possible reading of (29b) is consistent with the intended reading, and the observed contrast between the two truncated versions corresponds with the contrast in acceptability between the two original versions in (27). Inoue (1976) claimed a subject bias in extraction, and Hasegawa (1985: 292) proposed as a structural principle that the relativization or topicalization out of a relative clause is allowed if the displaced phrase is the subject of the (lower) relativized clause and the head of the relative clause is the subject of the higher clause. Although it has since been shown that this is problematic as a principle (see Shimojo 2002: 74–5 and the references cited therein), the said subject bias, if taken as a tendency, is not so inconsistent with what the pragmatic-aboutness condition predicts. Assuming a subject represents the target of predication, the displaced element functions as (part of ) the target of predication if the displaced element is the subject of the lower relative clause which is in turn the subject of the higher clause. The example of an acceptable extraction given in (27b) represents this structure. However, it should be noted that subjecthood does not warrant required aboutness because aboutness is also affected by lexical choice. This point is demonstrated by the relativization shown in (30). While this example represents the optimal structure per Hasegawa (1985), the verb choice for the inner relative clause affects the extractability. (30)

[[ __ kaita/*yonda] hon-ga eegaka-sare-ta] write.pst/read.pst book-nom cinematization-do.pass-pst ‘Jon who a book which (he) wrote/*read was made into a movie’

jon Jon

Truncation of this example results in [hon-ga eegakasareta] jon ‘Jon who a book was made into a movie’. Possible interpretations of this truncation are Jon as the author, editor, translator, etc. of the book (as translated as ‘Jon whose

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

627

628

MITSUAKI SHIMOJO

book was made into a movie’), not a mere reader, and this contradicts the intended reading of [[ __ *yonda] hon-ga eegakasareta] jon.

16.3.3.2 Postposing Sentence elements are also displaced through the so-called postposing construction. While Japanese is considered a rigid verb-final language, it allows sentence elements to be in post-predicative positions, typically in spoken Japanese (Shimojo 1995).8 An example is shown in (31b), in which the nominative argument appears in the postverbal position, in contrast with (31a), which represents the canonical ordering of arguments (the postposed elements are underlined in the examples). (31)

a. keesatsu-ga otoko-o taiho-shita yo police-nom man-acc arrest-do.pst fp b. __ otoko-o taiho-shita yo keesatsu-ga man-acc arrest-do.pst fp police-nom ‘The police arrested a man.’

As shown in Figure 16.3, there are two positions for post-predicative elements in Japanese, PoCS and PoDP. While both are structurally marked positions in the verb-final language, they represent distinct focus-structure properties. The PoDP is a post-predicative topic position, the post-predicative counterpart of PrDP, and for this reason, the preceding discussion of topicalization applies here. For the PoDP-type (topic) postposing, the preceding part of the sentence including the matrix clause must be about the displaced element. The examples in (32) show the correspondence between acceptable relativization (a), topicalization (b) and PoDP-type postposing (c). (32)

a. [[ __ kite-iru] yoohuku-ga yogorete-iru] shinshi wear-asp clothes-nom get.dirty-asp gentleman ‘a gentleman who the clothes that (he) is wearing are dirty’ b. ano shinshi-wa [ __ kite-iru] yoohuku-ga yogorete-iru that gentleman-top wear-asp clothes-nom get.dirty-asp ‘That gentleman, the clothes that (he) is wearing are dirty.’ c. [ __ kite-iru] yoohuku-ga yogorete-iru yo ano shinshi-wa wear-asp clothes-nom get.dirty-asp fp that gentleman-top ‘That gentleman, the clothes that (he) is wearing are dirty.’

On the other hand, because a postverbal element in PoCS does not represent a topic, the aboutness condition is not a requirement for the PoCS-type (focus) postposing.9 For example, the postposing in (33b) is acceptable despite the unacceptability of the corresponding relativization. (33)

a. *[keesatsu-ga [ __ nusunda] otoko-o taiho-shita] kuruma (¼27a) police-nom steal.pst man-acc arrest-do.pst car ‘a car which the police arrested the man who stole (it)’ b. keesatsu-ga [ __ nusunda] otoko-o taiho-shita yo kuruma-o police-nom steal.pst man-acc arrest-do.pst fp car-acc ‘The police arrested the man who stole a car.’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

629

However, it has been observed that this type of postposing is sensitive to the focus structure of a sentence. Simon (1989) gives the examples in (34) to point out that transparency defined by genericness correlates with the gradient acceptability. The postposing is acceptable with a matrix copula (34a), and it is least acceptable with a specific action verb (34c). (34)

a. kore-wa [__kinoo mottekita] wain da this-top yesterday bring.pst wine cop ‘This is wine which Mari brought yesterday.’ b. ?* [__ kinoo mottekita] wain-ga aru yesterday bring.pst wine-nom exist ‘There is wine which Mari brought yesterday.’ c. *[__ kinoo mottekita] wain-o nomu yesterday bring.pst wine-acc drink ‘I’ll drink the wine which Mari brought yesterday.’

yo fp yo fp yo fp

mari-ga Mari-nom mari-ga Mari-nom mari-ga Mari-nom (Simon 1989: 166)

Also, the acceptability of postposing is affected by the particular focus placement determined by the preceding context. The postposing in (35B) is acceptable in isolation or in a context such as (35A). The relative clause in (35B) represents the actual focus and the postposed argument functions within the actual focus domain. (35)

A: jon-wa dare-ni aitagatte-ru no? John-top who-dat meet.des-asp fp ‘Who does John want to meet?’ hitoi-ni aitagatte-ru n da B: [__i __j kaita] write.pst person-dat meet.des-asp nmlz cop ‘(He) wants to meet the person who WROTE THE BOOK.’

yo fp

ano that

hon-oj book-acc

However, in the context given in (36A), the relative clause is outside the actual focus domain, and the matrix (bridging) clause represents the narrow focus, which clearly impairs its postposability.10 (36)

A: John-wa [ano hon-o kaita] hito-ni atta no? John-top that book-acc write.pst person-dat meet.pst fp ‘Did John meet the person who wrote the book?’ hitoi-ni aitagatte-ru n da yo ano B: *iya, [__i __j kaita] no write.pst person-dat meet.des-asp nmlz cop fp that ‘No, (he) WANTS TO MEET the person who wrote the book.’

The preceding observations point to the functional requirement in PoCStype postposing that the displaced entity in the PoCS must function in the part of the utterance which represents the assertion.11 This requirement predicts that if the matrix clause represents a narrow focus, as in (36B), the postposing is unacceptable. This basic requirement for displacement of nontopic entities also applies to wh-questions, which are discussed in the following section. Overall, postposing exhibits ambivalent properties. On one hand, PoDP-type postposing is analogous to relativization and topicalization because the displaced entity functions as a topic of the entire

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

hon-oj book-acc

630

MITSUAKI SHIMOJO

sentence. The PoCS-type postposing, on the other hand, is similar to whquestions because the displaced element (or the in-situ question word) must be associated with the focus of the sentence. In this regard, Takita’s (2014: 139) example given in (37) is worth a brief discussion. Takita states that the postposing is unacceptable if the postposed argument ‘that book’ maintains the accusative marking, but the sentence becomes acceptable if the case marker is absent (i.e. if it is a ‘bare-topic’).12 (37)

[taroo-ga __ suteta kara] hanako-ga totemo okotte-iru yo Taro-nom discard.pst because Hanako-nom very get.angry-asp fp ano hon-{*o/Ø} that book-acc/zero.particle ‘Because Taro discarded that book, Hanako is very angry.’

In RRG terms, the observation above is captured as follows. The postposing with the zero marking for the postverbal argument is the PoDP type because the postverbal argument is readily interpreted as a topic. The postverbal RP is referential, and the lack of case marking is characteristic of a topic. Also, the ‘book’ can be taken as the entity which the whole sentence is relevant to, that is, that Hanako is very angry about the book which was discarded. Yet, the overt accusative marking of the postverbal argument de-topicalizes the argument and therefore shifts it to the PoCS; hence, the postposability is degraded by the intervening matrix clause, which represents assertion.

16.3.3.3 Wh-questions While wh-questions in complex sentences in Japanese are generally acceptable, some restrictions have been observed. Consider the examples in (38), taken from Haig (1979: 90–91, grammaticality judgement original). (38)

a. ?* [dare-ga hanako-ni kureta] inu-ga shinde-shimatta? who-nom Hanako-dat give.pst dog-nom die-asp.pst ‘The dog that who gave to Hanako died?’ b. ?* kono ko-wa [ __ hanako-ni __ kureta] inu-ga shinde-shimatta this child-top Hanako-dat give.pst dog-nom die-asp.pst ‘This child, the dog which (he) gave Hanako died.’

The wh-question formed out of the relative clause in (38a) is unacceptable. Given the unacceptability of the corresponding topicalization in (38b), Haig claims that the questioned noun phrases raise an issue that essentially relates restrictions on wh-questions to the aboutness condition discussed earlier. Kuno (1987: 23) has proposed the Topichood Condition for Extraction, which subsumes wh-questions in English such as (26b, c). However, restrictions observed in Japanese wh-questions are of a different nature from those in relativization and topicalization, which are conditioned on aboutness. Consider the examples in (39), taken from Hasegawa (1981: 281).

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences (39)

a. *sono yoohuku-wa [ __ kite-ita] shinshi-ga yukuehumee da that clothes-top wear-asp.pst gentleman-nom missing cop ‘The clothes, the gentleman who was wearing (them) is missing.’ b. [ __ nani-o kite-ita] shinshi-ga yukuehumee-na no? what-acc wear-asp.pst gentleman-nom missing-cop fp ‘The gentleman who was wearing what is missing?’

The unacceptable extraction in (39a), and unacceptable truncation yoohukuwa shinshi-ga yukuehumeeda ‘the clothes, the gentleman (who was wearing (them)) is missing’, shows that the topicalized element ‘clothes’ is not immediately relevant to the gentleman’s being missing. Despite this, the corresponding wh-question is acceptable (39b), and this undermines the claim which relates wh-words to aboutness. In this regard, Van Valin and LaPolla (1997: 627–629) argue that the function of a displaced element in a wh-question (i.e. the wh-word) is focus, and therefore, they must function in the PFD. This claim is valid for wh-questions in Japanese despite the wh-in situ characteristics. What influences extractability in Japanese wh-questions are lexical-semantic factors which interact with the PFD. As discussed thus far, the default range of potential focus domain is language-specific. The structural constraint given earlier in (5) represents the default range of PFD in languages like English, while in Japanese, the PFD excludes only detached positions. Yet, in both languages, the PFD interacts with lexical-semantic factors, as Van Valin and LaPolla (1997: 630) state: lexical semantic factors may also influence the potential focus domain, both in terms of preventing a position in the potential focus domain from being the actual focus domain and of overriding the principle [in (5)] and permitting the actual focus domain to be in structural configurations where it would otherwise be impossible. An object complement is part of the PFD by default since it is a direct daughter of the clause node. However, it has long been observed, as shown in (40a), that extraction is blocked by so-called non-bridge verbs, which denote a particular manner, such as ‘murmur’, ‘whisper’ and ‘shout’, unlike bridge verbs such as ‘say’, ‘think’ and ‘believe’ (Erteschik-Shir 1973; Müller 1995; Takami 1992). In RRG terms, these semantically highlighted verbs shift the focus and reduce the PFD away from the complement clause, hence they block the wh-question (Van Valin 1996: 50). The same argument applies to the unacceptable extraction when a bridge verb is used with a manner adverb (Kennedy 1989), as shown in (40b). The manner adverb causes the same focus shifting, preventing the object complement from being the actual focus domain. (40)

a. What did Jon say/*murmur/*whisper that Mary had bought? b. *What did Jon say abruptly that Mary stole?

Analogous lexical-semantic factors are responsible for restrictions on whquestions in Japanese, as seen in (38a). However, wh-question formation is

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

631

MITSUAKI SHIMOJO

632

affected only in a certain type of relative clause. First, consider Hasegawa’s (1989) proposal cited in (41), with respect to different readings of a whquestion formed in a relative clause. (41)

Two possible readings of a wh-question formed in a relative clause (i) Narrow reading: what is questioned (i.e. the wh-phrase) is what is to be identified by the question. (ii) Broad reading: what is questioned is NOT what is to be identified by the question. What is to be identified is represented by the relative clause head noun.

According to Hasegawa, the interpretation depends on two separate factors: the referability of the head noun and the inherent topic-worthiness of the wh-expression. The former bears immediate relevance on the present discussion. Now consider the contrast between (42) and (43). (42)

Q: [dare-ga kaita] shoosetsu-ga yoku uremasu ka who-nom write.pst novel-nom well sell q ‘Novels that who writes sell well?’ A: murakami-haruki desu Murakami.Haruki cop ‘(It’s) Haruki Murakami.’ A′: [murakami-haruki-ga kaita] shoosetsu desu Murakami.Haruki-nom write.pst novel cop ‘(It’s) the novel which Haruki Murakami wrote.’

(43) Q:

[dono kyooju-ga suisenshite-iru] hito-ga saiyoo-sare-soo desu ka which prof.-nom recommend-asp person-nom employment-do.pass-evi cop q ‘A person that which professor has recommended is likely to be hired?’ A: *suzuki-kyooju desu Suzuki-prof. cop ‘(It’s) Prof. Suzuki.’ A′: [suzuki-kyooju-ga suisenshite-iru] hito desu Suzuki-prof.-nom recommend-asp person cop ‘(It’s) the person that Prof. Suzuki has recommended.’ (Nishigauchi 1986: 74)

Wh-questions as in (42) have a narrow reading, because the head of the relative clause shoosetsu ‘novel’ is non-referential and, therefore, what is to be identified by the question is who it is that sells their novels, rather than the particular novels which the writer publishes. On the other hand, whquestions as in (43) have a broad reading, because the head noun hito ‘person’ is referential, that is, what is to be identified by the question is who the candidate is, and this identification of the person is done by way of identifying the recommender of the candidate. Hence, Hasegawa argues that the short answer, which supplies just the value for the wh-word, is awkward in the context of (43). As expected, for wh-questions such as (43) to be felicitous, there must be a shared assumption in the context that there is a set of candidates and each of them is recommended by a professor. The purpose of the wh-question is to identify a particular candidate via

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

identifying a particular candidate–recommender pair. Wh-questions as in (43) are thus virtually synonymous with ‘which’-questions such as ‘which candidate is most likely to get the position?’ The distinction between narrow reading and broad reading sheds light on why constraints are observed in some wh-questions. A narrow reading contains a single focus of question, since what is questioned (i.e. the wh-phrase) is what is to be identified by the question. On the other hand, a broad reading represents two foci because what is to be identified by the question (i.e. the relative clause head noun) represents a focus of the question, and this focus element is additional to the canonical focus of the question (i.e. the wh-phrase). In other words, wh-questions which represent a broad reading are functionally complex, and it is this type of wh-question that is subject to restrictions due to lexical-semantic factors. Haig’s example given earlier in (38a) is a broad-reading question with the referential head noun ‘dog’, which represents what is to be identified by the question; thus, for this question, [dare-ga kureta] inu ‘the dog that who gave’ is virtually synonymous with ‘which dog’. However, in (38a), there is an additional referential noun ‘Hanako’ placed between the foci of the question, which would otherwise be contiguous. This shifts the focus away from the foci of the question, and for this reason, this is analogous to the English examples in (40), where the actual focus is shifted away from the extraction site to a semantically highlighted element. This argument is supported by the observation that the wh-question in (38a) becomes perfectly acceptable if the recipient of the giving is not overtly expressed, as shown in (44). (44)

[dare-ga kureta] inu-ga shinde-shimatta? who-nom give.pst dog-nom die-asp.pst ‘The dog that who gave (Hanako, you, etc.) died?’

The claim above is supported further by the following data. As shown in (45), broad-reading wh-questions become increasingly awkward, if there is gradient acceptability for a speaker, as more referential expressions intervene between the foci of the questions. The degraded acceptability correlates with the degree of lexical-semantic highlighting on the intervening elements, which shifts the focus away from the foci of the question [dare-ga okutta] hon ‘the book that who sent’ (¼ ‘which book’). (45)

a. [dare-ga okutta] hon-ga nakunatta no? who-nom send.pst book-nom become.lost.pst fp ‘The book that who sent had been lost?’ b. ?[dare-ga taroo-ni okutta] hon-ga nakunatta no? who-nom Taro-dat send.pst book-nom become.lost.pst fp ‘The book that who sent to Taro had been lost?’ c. *[dare-ga yuubinkyoku-de taroo-ni okutta] hon-ga nakunatta no? who-nom post.office-loc Taro-dat send.pst book-nom become.lost.pst fp ‘The book that who sent to Taro at a post office had been lost?’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

633

634

MITSUAKI SHIMOJO

On the other hand, if the wh-expression is kept adjacent to the head of the relative clause, the other focus of the question, the restriction is ameliorated even if there are other referential expressions in the relative clause. In (46b, c), the foci of the question represent a contiguous focus unit; therefore, it is functionally equivalent to a single focus of a question. okutta] hon-ga nakunatta no? (46) a. [dare-ga who-nom send.pst book-nom become.lost.pst fp ‘The book that who sent had been lost?’ b. [jon-ga dare-ni okutta] hon-ga nakunatta no? Jon-nom who-dat send.pst book-nom become.lost.pst fp ‘The book that John sent to whom had been lost?’ c. [jon-ga sono yuubinkyoku-de dare-ni okutta] hon-ga nakunatta no? Jon-nom that post.office-loc who-dat send.pst book-nom become.lost.pst fp ‘The book that John sent to whom at the post office had been lost?’

In the case of narrow-reading wh-questions, on the other hand, acceptability is not affected by additional referential phrases in the relative clause. Examples are given in (47b, c). The wh-question is not blocked in these cases because the relative head noun does not represent a focus of the question (i.e. what is to be identified by the question) and does not form a focus unit of the question with the wh-expression. dasu] hon-ga yoku ureru no? (47) a. [dare-ga who-nom publish book-nom well sell fp ‘Books that who publishes sell well?’ b. [dare-ga amerika-de dasu] hon-ga yoku ureru no? who-nom America-loc publish book-nom well sell fp ‘Books that who publishes in America sell well?’ c. [dare-ga amerika-de jerii-paaneru-to dasu] hon-ga yoku ureru no? who-nom America-loc Jerry.Pournelle-com publish book-nom well sell fp ‘Books that who publishes with Jerry Pournelle in America sell well?’

Also, the preceding discussion is relevant to multiple wh-questions, as shown in (48). Multiple wh-questions are functionally similar to broadreading wh-questions because both types of question contain double foci. In both cases, the question becomes constrained if there are intervening referential phrases, as is the case in (48b, c). (48)

a. dare-ga nani-o okutta no? who-nom what-acc send.pst fp ‘Who sent what?’ b. ?dare-ga taroo-ni nani-o okutta no? who-nom Taro-dat what-acc send.pst fp ‘Who sent Taro what?’ c. *dare-ga yuubinkyoku-de taroo-ni nani-o okutta no? who-nom post.office-loc Taro-dat what-acc send.pst fp ‘Who sent Taro what at the post office?’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

635

Again, if the foci of a question are kept contiguous as shown in (49b, c), these questions are considerably more acceptable, if not perfectly acceptable. (49)

a. dare-ga nani-o okutta no? who-nom what-acc send.pst fp ‘Who sent what?’ b. taroo-ni dare-ga nani-o okutta no? Taro-dat who-nom what-acc send.pst fp ‘Who sent Taro what?’ c. yuubinkyoku-de taroo-ni dare-ga nani-o okutta no? post.office-loc Taro-dat who-nom what-acc send.pst fp ‘Who sent Taro what at the post office?’

To summarize, whereas wh-questions in complex sentences in Japanese are widely acceptable, restrictions are observed in wh-questions that are functionally complex with double foci. The underlying principle is the required association of the assertion part of an utterance and a displaced entity (i.e. a wh-expression in situ in the case of Japanese wh-questions); this means that the displaced entity (or a wh-expression in situ) must function in the focus of the utterance. Before concluding, we consider a constraint on ‘why’ questions. It has been pointed out that the wh-adverbial why exhibits some peculiar characteristics across languages (Jin 2019). Likewise, it has been observed that naze in Japanese exhibits wh-island effects unlike the other wh-words (Lasnik and Saito 1984; Richards 2000; Fujii and Takita 2007). For example, while naze in a complement clause is acceptable, as shown in (50a), a complex NP and an adverbial clause are unacceptable with naze in (50b, c). (50)

a. jon-wa [taroo-ga naze kinoo sono hon-o katta tte] itta no? John-top Taro-nom why yesterday that book-acc buy.pst comp say.pst fp ‘Did John say that Taro bought the book yesterday why?’ b. *jon-wa [taroo-ga naze kinoo katta] hon-o karita no? John-top Taro-nom why yesterday buy.pst book-acc borrow.pst fp ‘Did John borrow the book which Taro bought yesterday why?’ c. *[taroo-ga naze kinoo sono hon-o yonda atode] shiken-ga Taro-nom why yesterday that book-acc read.pst after exam-nom umaku itta no? well go.pst fp ‘Did the exam go well after Taro read the book yesterday why?’

On the other hand, the other wh-adverbials such as dooyatte ‘how’ and nominal wh-adverbials such as {donna/dooyuu} riyuu-de ‘for what reason’ are not subject to the restriction, as shown in (51).13 (51)

a. jon-wa [taroo-ga kinoo dooyatte tsukutta] ryoori-o tabeta no? John-top Taro-nom yesterday how make.pst dish-acc eat.pst fp ‘Did John eat the dish which Taro made how yesterday?’

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

636

MITSUAKI SHIMOJO

b. [taroo-ga kinoo donna riyuu-de kaetta kara] Taro-nom yesterday what.kind.of reason-by go.home.pst because jon-ga okotta no? John-nom get.upset.pst fp ‘Did John get upset because Taro went home yesterday for what reason?’

The contrast in acceptability between (50) and (51) suggests that the observed restriction is due to the property of the particular wh-word why. Van Valin (2002: 169) points out the adsentential nature of why and states that the answers to who, what and how involve some change in the original sentence (which is also the case with the nominal wh-adverbial for what reason in (51b)), but the answer to a why question is normally a because-clause, which is adjoined to the unchanged original sentence. Why by itself can be a felicitous question when it follows a preceding statement, and the answer to the question does not change or elaborate the preceding statement, as shown in (52). (52)

A1: B: A2:

John was upset. Why? John was upset because Taro was late.

As Van Valin (2005: 284) explains, the interclausal ‘reason’ relation, [LS1] BECAUSE′ [LS2], contains two independent logical structures, which, however, are linked by a subordinating conjunction. Therefore, there is only one illocutionary operator for the two logical structures. For this reason, only one of the clauses can be the potential focus domain at a time.14 This is exemplified by (52A2), in which ‘Taro was late’ is in the potential focus domain and ‘John was upset’ represents the presupposition. In a why question, the wh-word, which corresponds with [LS2] of the interclausal semantic relation above, must be in the potential focus domain since it is the focus of the question. This means that the clause representing the other logical structure must be outside the potential focus domain, necessarily representing presupposition.15 Then, why do we observe the extraction restriction on questions such as those in (50b, c)? There is a simple Gricean explanation for this. Since the non-wh-part of a why question must be presupposed, only the presupposition that is relevant to the point of the question must be expressed, because of the maxim of relevance. In (50b), for example, the presupposition ‘Taro bought the book yesterday’ is relevant to the why question since it represents one of the logical structures in the ‘reason’ relation, but ‘John borrowed the book’ is not relevant because it is not part of the semantic relation. Inclusion of the irrelevant proposition in the question must draw the focus for the Gricean reason, which shifts the potential focus domain away from the wh-expression. This is essentially the same Gricean explanation which applies to the unacceptable questions in (40) discussed earlier. With the account above, the acceptable why question formed in an object complement such as (50a) is expected because the embedded clause is a

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

direct daughter of the matrix clause node (see the principle in (6)). In other words, what is expressed by the embedded clause is required by and hence directly relevant to the matrix clause. To summarize, the fundamental functional requirement that the wh-expression must function in the part of an utterance which represents the assertion is a valid principle for naze ‘why’ questions in Japanese as well.

16.4

Conclusion

This chapter has focused on extraction restrictions in complex sentences, which are typically discussed under the assumption that such restrictions are structurally definable. This is largely due to the tradition of subjacencybased accounts, long associated with extraction restrictions, as well as English-type languages being used as the cardinal ground for analysis. As discussed throughout this chapter, RRG combines structural and functional constraints to capture both cross-linguistic and language-internal variation in extraction restrictions. The fundamental functional requirement is the association of the assertion part of an utterance and a displaced entity (or a wh-expression in situ) such that the displaced entity (or a wh-expression in situ) must function in the part of the utterance which represents the assertion. Obviously, the three extraction constructions discussed in this chapter, wh-question formation, topicalization, and relativization, represent different functions. As Van Valin and LaPolla (1997: 627) put it, the wh-element of a wh-question represents a focus, the topic in a topicalized sentence represents either a topic or a focus, and the head noun of a relative clause represents a topic, hence the construction-specific variations. Topicalization in English such as ‘that book I wouldn’t buy’ contains the preposed element in the pre-core slot, hence, it is focal. In Japanese, a topic in the detached position is topical, and a contrastive topic, which is placed within the clause, whether core-internally or externally, is focal, as discussed in Section 16.3.2. Despite these different functions, however, they share the underlying property that they must function in the assertion part of an utterance (i.e. within the PFD). While there is an obvious connection between a wh-element (which is focal) and assertion, relativization and topicalization share the property that ‘the clause in which the displaced NP functions is always about the referent of the NP’ (Van Valin and LaPolla 1997: 627). In other words, such a clause functions as predication about the displaced element, and ‘predication and “assertion about” are fundamentally related notions’ (p. 629). The analysis of Japanese shows that the RRG principle which relates extraction to assertion is valid for the language, despite its radical difference from other languages with respect to the possible range of PFD. This offers an important implication for the universality of the principle, particularly

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

637

638

MITSUAKI SHIMOJO

because it is assumed (Van Valin 1996, 1998, 2005; Van Valin and LaPolla 1997) that the constraints are ultimately derivable from Grice’s (1975) cooperative principle and the maxim of quantity, which are the general principles of rational human behaviour (hence, a challenge to the subjacency principle as evidence of autonomous syntax, and ultimately to a theory of universal grammar). Van Valin (1998) uses the following from Kempson (1975: 190) to illustrate how the Gricean explanation is rooted in the claim that the focus domain is the domain for extraction (i.e. questioning and predication).16 The speaker believes the hearer knows (and knows that the speaker knows) a certain body of propositions (i.e. that there is a pragmatic universe of discourse) and in making a certain utterance . . . he believes that the hearer, knowing the conventions of the language and hence the conditions for the truth of the proposition in question, will recognize a subset of those conditions as being part of that pragmatic universe of discourse and hence neither assertible, deniable or queriable (without violating the quantity maxim), and a second mutually exclusive subset of the conditions as being outside the pragmatic universe of discourse. This latter set, he will interpret as being asserted, denied, commanded or queried. In RRG, the cross-linguistic variations in extraction are captured in terms of how deep into the sentence structure assertion may be represented. By default, a matrix clause is always within the PFD; thus, it is a possible extraction site across languages. In some languages, the PFD may be extended to include those subordinate clauses which are tightly connected to the matrix clause, as a direct daughter of a clause. As in the case of Japanese, the PFD may be extended further to include other subordinate clauses; yet, even in Japanese, detached topic positions are excluded from the PFD. This exclusion is important corroboration of the RRG principle since a topic does not represent assertion. In Kempson’s (1975) terms, a topic represents part of the pragmatic universe of discourse which is neither assertible, deniable nor queriable without violating the quantity maxim. Extraction is allowed in a topic only if the topic is focal, hence contrastive, and this conforms to the Gricean explanation. Lastly, the RRG principle is corroborated further by the observed lexical-semantic influence on wh-question formation in Japanese, which again relates extractability to the Gricean explanation.

References Allwood, Jens S. 1982. The complex NP constraints in Swedish. In E. Engdahl and E. Ejerhed (eds.), Readings on Unbounded Dependencies in Scandinavian Languages, 15–32. Stockholm: Almqvist & Wiksell.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In K. H. Ebert (ed.), Studies in Clause Linkage, 23–55. Papers from the First KölnZürich Workshop. Zürich. Chaves, Rui P. 2013. An expectation-based account of subject islands and parasitism. Journal of Linguistics 49: 285–327. Chomsky, Noam. 1973. Conditions on transformations. In S. R. Anderson and P. Kiparski (eds.), A Festschrift for Morris Halle, 232–285. New York: Holt, Rinehart, and Winston. Chomsky, Noam. 1986. Knowledge of Language. New York: Praeger. Deane, Paul. 1991. Limits to attention: A cognitive theory of island phenomena. Cognitive Linguistics 2: 1–63. Deane, Paul. 1992. Grammar in Mind and Brain: Explorations in Cognitive Syntax. Berlin: Mouton de Gruyter. Erteschik-Shir, Nomi. 1973. On the Nature of Island Constraints. PhD dissertation, Massachusetts Institute of Technology. Erteschik-Shir, Nomi. 2007. Information Structure: The Syntax–Discourse Interface. Oxford: Oxford University Press. Erteschik-Shir, Nomi and Shalom Lappin. 1979. Dominance and the functional explanation of island phenomena. Theoretical Linguistics 6: 41–86. Fujii, Tomohiro and Kensuke Takita. 2007. Wh-adverbials in-situ, their island-(in)sensitivity and the role of demonstratives in wh-in-situ licensing. Nanzan Linguistics: Special Issue 3(1): 107–126. Grice, H. P. 1975. Logic and conversation. In P. Cole and J. L. Morgan (eds.), Syntax and Semantics 3: Speech Acts, 41–58. New York: Academic Press. Haig, John. 1979. What relative clauses are about. Papers in Linguistics 12: 57–109. Haig, John. 1996. Subjacency and Japanese grammar: A functional account. Studies in Language 20: 53–92. Hasegawa, Nobuko. 1981. A Lexical Interpretive Theory with Emphasis on the Role of Subject. PhD dissertation, University of Washington. Hasegawa, Nobuko. 1985. On the so-called ‘zero pronouns’ in Japanese. The Linguistic Review 4: 289–341. Hasegawa, Yoko. 1989. Questioning vs. identifying: A functionalist analysis of the [a candidate that which professor recommended was hired?] construction in Japanese. In K. Hall, M. Meacham and R. Shapiro (eds.), Proceedings of the 15th Annual Meeting of the Berkeley Linguistics Society, 138–149. Berkeley, CA: Berkeley Linguistics Society. Hofmeister, Philip. 2007. Retrievability and gradience in filler-gap dependencies. In M. Elliott, J. Kirby, O. Sawada, E. Staraki and S. Yoon (eds.), Proceedings from the Main Session of the Forty-Third Annual Meeting of the Chicago Linguistic Society: Vol. 43–1, 109–123. Chicago, IL: Chicago Linguistic Society. Hong, Sun-ho. 2003. On island constraints in Korean. In G. K. Iverson and S.-C. Ahn (eds.), Explorations in Korean Language and Linguistics, 107–125. Seoul: Hankook Publishing.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

639

640

MITSUAKI SHIMOJO

Inoue, Kazuko. 1976. Henkei bunpoo to nihongo (Transformational Grammar and Japanese). Tokyo: Taishukan. Jensen, Anne. 2001. Sentence intertwining in Danish. In E. EngbergPedersen and P. Harder (eds.), Ikonicitet og struktur, 23–39. Preprint from Netværk for Funktionel Lingvistik, Department of English, University of Copenhagen. Jin, Dawei. 2013. Information structure constraints and complex NP islands in Chinese. In Stefan Muller (ed.), Proceedings of the 20th International Conference on Head-Driven Phrase Structure Grammar, 110–120. Stanford, CA: CSLI Publications. Jin, Dawei. 2019. A semantic account of quantifier-induced intervention effects in Chinese why-questions. Linguistics and Philosophy 43: 345–387. Kempson, Ruth M. 1975. Presupposition and the Delimitation of Semantics. Cambridge: Cambridge University Press. Kennedy, Becky. 1989. On Bridging. Unpublished manuscript, Harvard University. Kluender, Robert. 1990. A neurophysiological investigation of wh-islands. In K. Hall, J.-P. Koenig, M. Meacham, S. Reinman and L. A. Sutton (eds.), Proceedings of the 16th Annual Meeting of the Berkeley Linguistics Society, 187–204. Berkeley, CA: Berkeley Linguistics Society. Kluender, Robert and Marta Kutas. 1993. Subjacency as a processing phenomenon. Language and Cognitive Processes 8(4): 573–633. Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge, MA: MIT Press. Kuno, Susumu. 1987. Functional Syntax: Anaphora, Discourse and Empathy. Chicago, IL: University of Chicago Press. Kuno, Susumu and Ken-ichi Takami. 1993. Grammar and Discourse Principles: Functional Syntax and GB Theory. Chicago, IL: University of Chicago Press. Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus, and the Mental Representations of Discourse Referents. Cambridge: Cambridge University Press. Lasnik, Howard and Mamoru Saito. 1984. On the nature of proper government. Linguistic Inquiry 15: 235–289. Matsumoto, Yoshiko. 1997. Noun-Modifying Constructions in Japanese: A Frame Semantic Approach. Amsterdam: John Benjamins. Miyagawa, Shigeru. 1987. Wa and the wh phrase. In J. Hinds, S. K. Maynard and S. Iwasaki (eds.), Perspectives on Topicalization: The Case of Japanese wa, 185–217. Amsterdam: John Benjamins. Müller, Gereon. 1995. A-Bar Syntax: A Study in Movement Types. Berlin: De Gruyter Mouton. Nakagawa, Natsuko, Yoshihiko Asao and Naonori Nagaya. 2008. Information structure and intonation of right-dislocation sentences in Japanese. Kyoto University Linguistic Research 27: 1–22.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

Nishigauchi, Taisuke. 1986. Quantification in Syntax. PhD dissertation, University of Massachusetts, Amherst. Richards, Norvin. 2000. An island effect in Japanese. Journal of East Asian Linguistics 9: 187–205. Ross, John Robert. 1967. Constraints on Variables In Syntax. PhD dissertation, Massachusetts Institute of Technology. Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: Wa and ga, and Word Order Flexibility. PhD dissertation, University at Buffalo (SUNY). Shimojo, Mitsuaki. 2002. Functional theories of island phenomena: The case of Japanese. Studies in Language 26: 67–123. Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japanese. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing. Simon, Mutsuko E. 1989. An Analysis of the Postposing Construction in Japanese. PhD dissertation, University of Michigan. Takami, Ken-ichi. 1992. Preposition Stranding: From Syntactic to Functional Analyses. Berlin: Mouton de Gruyter. Takita, Kensuke. 2014. Pseudo-right dislocation, the bare-topic construction, and hanging topic constructions. Lingua 140: 137–157. Tomioka, Satoshi. 2009. Why questions, presuppositions, and intervention effects. Journal of East Asian Linguistics 18: 253–271. Van Valin, Robert D., Jr. 1996. Toward a functionalist account of so-called extraction constraints. In B. Devriendt, L. Goossens and J. van der Auwera (eds.), Complex Structures: A Functionalist Perspective, 29–60. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the mechanisms of language acquisition. In M. Tomasello (ed.), The New Psychology of Language: Cognitive and Functional Approaches to Language Structure, 221–249. Mahwah, NJ: Lawrence Erlbaum. Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure and syntax. In E. Raxilina and Y. G. Testelec (eds.), Typology and Linguistics Theory: From Description to Explanation, 511–524. Moscow: Languages of Russian Culture. Van Valin, Robert D., Jr. 2002. The development of subject–auxiliary inversion in English wh-questions: An alternative analysis. Journal of Child Language 29: 161–175. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Yoshimura, Kyoko. 1992. LF subjacency condition in Japanese. Coyote Papers: Working Papers in Linguistics 8: 143–162.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

641

642

MITSUAKI SHIMOJO

Notes 1 The term ‘extraction’ and related terms such as ‘wh-question formation’, ‘relativization’, ‘topicalization’ and ‘displaced element’ are used in the present chapter only for expository purposes. Likewise, a gap in examples is used only to show the place in which the displaced element functions. RRG is monostratal and does not posit any movement operation to move an element from one syntactic position to another. 2 Van Valin and LaPolla (1997: 487) note that the (c) response becomes acceptable if the entire clause is repeated, as in ‘no, after she left’. This is predicted by the RRG claim because the entire subordinate clause functions as an information unit in this case. 3 However, ambiguity depends on the linear position of the wh-word. If it intervenes with the subordinate elements, it is interpreted as modifying the subordinate clause only (e.g. [kuroichekku-ga itsu kakunooko-ni iru-to] sukinaa-ga itta?), and if it appears after the subordinate clause, it modifies the matrix clause only (e.g. [kuroichekku-ga kakunooko-ni iru-to] itsu sukinaaga itta?). 4 It has been claimed by Miyagawa (1987: 188) that the use of wa with a whphrase is acceptable only under the following specific conditions which represent explicit contrastiveness: an identifiable set of referents is presupposed in the immediate conversational context and every member of this set is exhaustively represented in the wh-wa question. 5 The negation given in (18B) is only mildly awkward, and this is predicted by the contrastive property of the wa-marked topic, which contains a focal element, as discussed later in this section. 6 See Shimojo (2011: 281–285) for discussion of marked use of the topic marker, which involves a mismatch between the inherent focus structure of the topic marker and the contextually determined focus structure of an utterance. Such a mismatch yields a particular implicature due to focus shifting caused by the inherent property of the topic marking. 7 The corresponding topicalization of (27b) is also perfectly acceptable, which contrasts with the unacceptable topicalization counterpart of (27a), which was given in (23). 8 The postposing considered here is the so-called non-pause-type postposing, in which the postverbal unit is prosodically part of the preceding unit, produced with a single intonation contour and without an intervening pause after the predicate. Prosodically separate postverbal units are best considered an afterthought and are hence structurally separate from the preceding unit (see Simon 1989 and Nakagawa et al. 2008 for discussions of the two types of postposing). 9 For the PoCS-type postposing, the sentence may also be about the displaced element, as in [ __ kiteiru] yoohuku-ga yogoreteiru yo ano shinshi-ga ‘That gentleman, the clothes that (he) is wearing are dirty’, in which the nominative argument ‘that gentleman’ in the PoCS represents a focus.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Extraction Restrictions in Complex Sentences

10 The postposed argument ‘book’ in this example is outside the actual focus domain but it is in the PoCS. The ‘book’ is not a topic of the sentence since the matrix clause ‘X wants to meet the person’ is not about the book, and for this reason, the ‘book’ is not topicalizable with the topic marker wa in this sentence. 11 See Shimojo (2002: §4.2) for detailed discussion in this regard. 12 Takita (2014) calls the construction with a postverbal bare-topic pseudo right dislocation and claims that bare topics are base-generated in this construction, unlike standard right dislocation, which is generated by ‘repetition and deletion’. 13 Ambiguous wh-expressions such as nande ‘by/with what’ (nominal adverbial) and ‘why’ (adverbial) are island-sensitive only for the adverbial interpretation (Fujii and Takita 2007). 14 See Bickel (1993: 33) for a related discussion about adsentential subordination. 15 The presupposition requirement for a why question is also claimed by Tomioka (2009: 265). 16 Van Valin (1998) shows that this analysis has important implications for language acquisition. He shows that there is data relevant to the constraints in the input to children and therefore that they can be learned. In essence, the critical construct of the RRG analysis is information structure, which derives ultimately from the Gricean principle; therefore, an account of the acquisition of extraction constraints does not have to appeal to a theory of universal grammar.

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

643

https://doi.org/10.1017/9781316418086.020 Published online by Cambridge University Press

Part Four

Applications of RRG

https://doi.org/10.1017/9781316418086.021 Published online by Cambridge University Press

https://doi.org/10.1017/9781316418086.021 Published online by Cambridge University Press

17 Role and Reference Grammar and Diachronic Syntax Ranko Matasovic´ 17.1

Introduction

This chapter will briefly review the work on diachronic syntax and morphosyntax in Role and Reference Grammar (RRG) and show how the tools of language description developed by this theory can be used to account for several aspects of language change. Although RRG was designed primarily as a synchronic syntactic theory, it has nevertheless been successfully applied in description, and sometimes explanation, of a number of diachronic syntactic processes (for a survey, see Wiemer 2008a). It is unsurprising that most diachronic work in RRG was devoted to Indo-European languages, as their history is generally better known than the histories of most non-IndoEuropean languages, but there are exceptions to this claim. Among diachronic studies dealing with non-Indo-European languages in the RRG framework we can mention Ohori’s dissertation (1992) and article on the development of Old Japanese clause linkage (1994) and Wilkins’ account of the diachronic development of verbs expressing associated motion in Mparntwe Arrernte, an Aboriginal language of Australia (1991). Ardis Eschenberg’s dissertation on the article system of Omaha (2005) also contains important insights into its diachronic development. Within the IndoEuropean family, especially well represented are diachronic studies of contemporary languages such as German (Diedrichsen 2008), French (Kailuweit 2008), English (Nicolle 2008) and Croatian (Matasovi´ c 2012), but aspects of the histories of dead languages such as Latin have also been analysed in the RRG framework (Cennamo 2001). Since RRG is not a theory that lays much emphasis on the intuition of the ideal speaker/hearer as the source of evidence for the grammaticality of sentences, and invented ‘laboratory’ examples do not play a great role in the argumentation about the syntactic structures of a language, it is easy to analyse dead languages with limited corpora within the RRG framework. RRG was therefore successfully used in the analysis of some aspects of syntax

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

648

RANKO MATASOVIC´

of such languages as Ancient Greek (Rupnik-Matasovi´ c 2008), Old English (González Orta 2002) and Latin (Michaelis 1993). Although these studies are chiefly synchronic and descriptive rather than diachronic, they are also relevant for the development of historical syntax in the RRG framework. It is clear that a reliable and careful description of the syntax of the early stages of individual languages is a prerequisite for an account of the syntactic developments that occurred between those early stages and their presentday descendants. The rest of this chapter is organized as follows: in Section 17.2 we discuss a set of parameters by which different theories of diachronic syntax can be compared and evaluated. We show how RRG answers the fundamental questions about the nature and causes of syntactic change, and argue that it is not a mere tool of linguistic description, but a theory that makes falsifiable empirical predictions. In Section 7.3 we discuss how syntactic change is represented in the RRG framework and how different elements of syntactic structure are affected by diachronic developments. We also show how empirically attested types of syntactic change can be analysed in RRG and argue that this theory allows for a more fine-grained analysis of diachronic processes than theoretically neutral accounts of diachronic syntax. Section 7.4 (Conclusion) summarizes the main points made in the chapter and proposes a few directions for future research.

17.2

The Goals of Diachronic Syntax: RRG Compared to Other Theories

A theory of diachronic syntax should be judged by the same scientific standards as any other scientific theory, that is, it should conform to the criteria of simplicity (or economy), empirical adequacy and predictive and explanatory power. Different theories can thus be compared, assuming it is agreed that, other things being equal, a theory is better than its rival if it is as simple as possible and empirically adequate (not contradicted by evidence) and if it also has more predictive and explanatory power than its rival(s). To be empirically adequate can mean two different things for a theory of diachronic syntax, as it needs to be both typologically adequate (able to analyse the structure of typologically different languages without imposing any preconceived features on them) and diachronically adequate (able to adequately represent which structures change and exactly how this happens in attested or reconstructed diachronic processes). To describe a syntactic change means to establish a set of rules leading from the syntactic representation at stage A of a language L to another syntactic representation at a subsequent stage B of the language L. It is clear that both syntactic representations at stages A and B should be adequately described, so synchronic

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

empirical adequacy is a prerequisite of diachronic empirical adequacy, and RRG was specifically designed to be as typologically adequate as possible in the analysis of languages of different types (Van Valin 2005). The predictive and explanatory power of a theory reflects itself in its ability to answer a number of empirically testable questions, such as the following (Matasovi´ c 2002): (i) What exactly changes when syntactic structures change? (ii) Are there impossible syntactic changes, or are some changes just more probable than others (and why)? (iii) What is the relationship between syntactic, phonological, morphological, semantic and pragmatic changes? (iv) What is the relationship between syntactic change and language learning? Is change somehow tied to imperfect language learning, or are there any other mechanisms of change? (v) What types of syntactic change should be distinguished? Are there changes that are more ‘fundamental’ than others? Let us briefly sketch how RRG addresses these questions in turn. i. The exact ‘locus’ of syntactic change directly depends on the architecture and ontology of the syntactic theory. A change that involves a simple rearrangement of elements in one theory can involve a radical restructuring of different representations (or ‘projections’) in a different theory. For example, in the history of French, the adjunct pas, which originally meant ‘step’, was reanalysed as a negation operator. The sentence Il ne voit pas la fille (he-neg.-sees-neg.-art.-girl) originally meant ‘He does not see the girl at all’, where pas was the equivalent of the English adverbial expression at all. After the reanalysis (during the Middle French period), the element pas lost its adverbial properties and was reduced to the status of a purely grammatical morpheme, essentially a negation always used in conjunction with the preverbal negation marker ne. In the Chomskyan frameworks, we would represent this change as a modification of labels on the constituent tree, but in RRG a lexical element which was represented in the periphery of a clause was reanalysed as an operator, which should be represented on a different ‘projection’ (i.e. the operator projection rather than the constituent projection). ii. Chomskyan accounts of historical syntax (e.g. Lightfoot 1979, 2001; Roberts 2007) explicitly deny the possibility of syntactic changes that would affect the structure of Universal Grammar (UG), which is conceived as being innate, syntax-driven and autonomous (independent of other modules of grammar). RRG, as a theory that denies (or rather, does not deem it necessary to posit) innate Universal Grammar makes no such assumptions, which means that there are no a priori impossible syntactic changes. However, some changes are rather unlikely if they contradict certain functional principles that affect the structure of grammars of human languages. This is another way in which RRG (and other functional theories of syntax) differs from the generative

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

649

650

RANKO MATASOVIC´

approaches, which often explicitly claim that every change not affecting the UG is equally probable (Kroch 2001: 726). Thus, for example, in RRG it would be natural to claim that demonstratives are more likely to turn into definite articles than into aspect markers on verbs, both because of their similar function (establishing reference and marking definite reference) and because they have similar positions in the layered structure of the reference phrase (RP). Such a claim amounts to a statistically verifiable prediction which can be checked by looking at the history of various languages in which the diachronic path from demonstratives to definite articles can be observed. For example, in the Romance languages the articles developed from Latin demonstratives ille and iste, in Ancient Greek the Homeric demonstrative ho became an article by the Classical period, and similar developments can be observed in the history of a number of Balkan languages (e.g. Bulgarian and Albanian) and Armenian (for other examples, see Matasovi´ c 2002: 59). The development from demonstratives to aspect markers on verbs is, to the best of our knowledge, unattested.1 A theory that cannot or does not make such verifiable predictions is clearly weaker with respect to its predictive power than RRG. iii. Syntactic theories that presuppose the autonomy of syntax would tend to predict that syntactic change is generally independent of changes in other domains of grammar. RRG, as a theory that explicitly denies that syntax is autonomous (in the generativist sense), sees syntactic change in close interdependence with changes in phonology, morphology, semantics and pragmatics. For example, in a number of languages it can be observed that the emphatic pronoun, used for emphasizing (or contrasting) the reference of a personal pronoun, becomes the reflexive marker, used to establish obligatory coreference in the clause or discourse. This change, which occurred in the history of the English reflexive (him-self etc.) as well as in Irish (féin), Ancient Greek (he-autón) and other languages, is motivated by the semantic and pragmatic proximity between emphasizing that a referent controls the action (rather than someone else) and asserting that a referent acts upon himself/ herself (rather than upon someone else). Since the emphasizing pronoun, by virtue of its meaning and pragmatic use, has to have an antecedent, it can also be used to establish coreference between itself and its antecedent. Thus, in many languages the emphasizing and reflexive markers are homophonous, as in the English sentences John saw himself in the mirror (reflexive himself) and John closed the mirror himself (emphasizing). The reverse process (from emphasizing to reflexive) is unattested, as far as we know, since the use of a reflexive marker to emphasize that a referent controls the action would not be pragmatically or semantically motivated. iv. Generativist theories of syntactic change tend to assume that change essentially takes place during language learning and is a consequence of

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

imperfect acquisition of the mother tongue. On the other hand, functionalist theories, including RRG, make no predictions as to the connection between language learning and syntactic change. In principle, it is possible that syntactic change occurs during adulthood, and that it is caused by several other factors (besides imperfect learning of the first language), for example interference with other languages in bilingual social groups, imitation of linguistically creative and socially prestigious individuals, etc. v. A lot of work in generativist diachronic syntax has been devoted to showing that syntactic changes are S-curved by nature, proceeding gradually until a setting of ‘parameters’ is affected, when they become abrupt and far-reaching, basically changing the typological profile of a language within a generation. Thus, while the parametric approaches to diachronic syntax, coupled with the minimalist assumption of an underlyingly universal head-first clause structure, have to predict that the more common diachronic change will generally be from the OV order to the ‘more economical’ VO order (Roberts 2007), RRG does not make any such assumptions, and both of the attested changes – from OV to VO and from VO to OV – are equally possible and ‘natural’ (see later in this section). RRG does not posit any fundamental parameters from the setting of which other features of a language’s syntax would follow, but it accepts the insights of language typology that certain syntactic features tend to cluster together in languages, at least statistically, and that tendencies towards typological consistency can often be discerned in the history of languages. There are usually functional explanations for clustering of syntactic features in languages of a certain type, and diachronic developments toward typological consistency can therefore often be viewed as being motivated by a general principle of economy, or ease of processing. Thus, it has been shown that Greenbergian wordorder universals can be explained by a general principle to the effect that languages tend to be consistent in their branching direction, such that non-lexical syntactic categories tend to consistently branch either to the left or to the right – this is Dryer’s ‘branching direction theory’ (1992). In diachronic terms, this will mean that languages will tend to change their syntactic structures in such a way as to increase consistency in the direction of their branching categories. For example, the order of the relative clause and the head noun it modifies will tend to be harmonic with the order of the verb and its object, so that in languages with OV order relative clauses will usually precede their head nouns, while in VO languages they will follow them, since both relative clauses and objects (as potentially complex RPs) are branching categories. It is important to note that Dryer was able to show that not all of Greenberg’s word-order universals really hold cross-linguistically, if a sufficiently representative language database is constructed and areal and genetic biases are excluded. For example, the position of adverbs with

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

651

652

RANKO MATASOVIC´

respect to verbs is not correlated with other pairings of syntactic constituents, and this is because both adverbs and verbs are non-branching lexical categories. A theory that clearly distinguishes lexical and potentially branching syntactic categories, such as RRG (Van Valin and LaPolla 1997: 70–71), is clearly easier to square with Dryer’s results concerning word-order correlations, and it is also better at predicting those types of syntactic changes that are leading towards a more consistent word order, which is – according to the branching direction theory – easier for processing. There is no general consensus as to what constitutes explanation in diachronic syntax, but in principle, every theory is expected to show how the impossibility, or statistical rarity, of certain syntactic changes follows from its general principles. A theory that assumes that VP is a universal type of syntactic constituent, as most varieties of generativism do, will find it difficult to explain why the VP does not behave as a syntactic constituent in diachronic processes, that is, why we seldom find that a language of the SOV type changed to OVS, or vice versa, while changes from VSO to SOV are rather common (e.g. in Ethiopian Semitic languages), just like changes from SOV to SVO (e.g. in Romance and Germanic languages). A theory like RRG that explicitly denies the cross-linguistic universality of VP as a constituent does not face this problem, since it does not predict that VP (or V and O as single constituents) will behave as a unit in diachronic changes. This does not mean that RRG would exclude the possibility of a change from SOV to OVS, since there may exist diachronic paths of word-order change that involve simultaneous movements of more than one constituent, but such changes are likely to be much less common, cross-linguistically, than straightforward changes involving just one of the non-controversial constituents such as the verb or one of its arguments. However, a theory that makes absolutely no predictions about possible and probable diachronic change is less of a theory and more like a set of devices for describing historical processes. It must be emphasized, therefore, that RRG does make some predictions about the diachronic probability of syntactic changes, and these predictions either follow from its architecture and ontology or they can be precisely formulated using the conceptual tools developed by the theory. Let us look at a few examples. RRG claims that the order of operators in the operator projection is universal, and predicted by the Natural Serialization Principle (NSP, cf. Foley and Van Valin 1984: §5.3; Van Valin and LaPolla 1997: 51; Van Valin 2005: 12). NSP predicts that the order of operators in languages is highly restricted, so that operators of broader scope cannot be closer to the nucleus (the verbal root) than operators with narrower scope. This means, for example, that there is no language in which a sentence operator such as illocutionary force is closer to the root than a nuclear operator, such as the morpheme

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

expressing the aspect of the verb. In a language that has only suffixes, this will necessarily mean that the suffix for aspect is closer to the root than the suffix expressing illocutionary force, for example the question marker. Now a consequence of the universality of the NSP is that no syntactic change is possible that would violate it, which means that, for example, if a language already has suffixes expressing illocutionary force – or some other clause- or sentence-level operators – no new suffixes expressing operators with a narrower scope, such as aspect, can develop. Thus, in Indo-European languages that had developed tense markers expressed by suffixes, new aspectual markers could not be grammaticalized as suffixes, but rather developed as prefixes, for example in Old Irish, where the simple preterite car-ais (love3sg.pret.) ‘he loved’ is opposed to the new perfect ro-car-Ø (prf.-love-3sg.). In other languages, the development was more complex, for example in Slavic, where full aspectual oppositions developed only after the new prefixed forms (e.g. Croatian po-gledati ‘take a look’) came into opposition with nonprefixed (gledati ‘watch’) and both prefixed and suffixed forms (po-gleda-avati ‘watch repeatedly’), but in no Indo-European language can we observe a development of new suffixes or postverbal particles expressing aspectual meanings, as such morphemes would necessarily follow inflexional morphemes expressing tense, thus violating the NSP (for details see Matasovi´ c 2002: 63–64 and Wiemer 2008a for some corrections). The NSP plays a role in another generalization that has been suggested concerning the frequency of grammaticalization patterns. It has been argued that operators with narrower scope tend to grammaticalize as operators with broader scope, but not vice versa (Foley and Van Valin 1984: 216–217; Matasovi´ c 2008). This means that, for example, core operators such as morphemes expressing deontic modality will develop into clausal operators such as those expressing status (epistemic modality), but that the reverse pattern would be rare or unattested. An example of such a pattern is the change of the English modal must from deontic (The letter must arrive next week) to epistemic (The letter must be in the mail). Similarly, nuclear operators such as those expressing aspect regularly grammaticalize as operators with scope over the clause or sentence, such as markers of evidentiality. An example of this pattern is the development of the participle of the auxiliary ima ‘have’ in Macedonian, which was originally just the marker of perfective aspect, but it has developed into an evidentiality marker expressing the ‘hearsay’ meaning, for example in the sentence Ti si imal kupeno kniga (you be.3sg.pret have.ptcp bought.ptcp book) ‘You bought a book (someone told me)’. This empirical generalization, which certainly requires further elaboration and empirical corroboration, is not directly implied by any principles of RRG, but the NSP, which is one of the theory’s fundamental principles, allowed its precise formulation. In theories which do not posit the NSP or its equivalent, it would not be possible either to formulate this generalization or to check it empirically in a systematic manner. However,

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

653

654

RANKO MATASOVIC´

just as any other hypothesis formulated in the framework of a scientific theory, it could eventually be refuted, if enough counterexamples are discovered. In a similar vein, it has been argued (Wiemer 2008b) that, at least in IndoEuropean languages, the diachronic development seems to be towards a tightening of the correlation between case assignment and privileged syntactic argument (PSA) selection (see Chapter 7). In early Indo-European languages, including Germanic, Greek and Balto-Slavic, we often find verbs which do not take the accusative as the case of their lowest macrorole argument, but in which that argument can still become PSA in some constructions, for example the passive. In later stages of these languages, the lowest macrorole arguments are always marked with the accusative. Likewise, in early Indo-European languages we often find non-nominative marked PSAs, for example in Old English, which allowed non-nominative PSAs (Roberts 1995: 176f.).2 (1)

ac gode ne licode na heora geleafleast ac asende But God.dat.sg neg like.pret neg their faithlessness.nom but send.pret Him to fyr of heafonum. them.acc to fire of heaven.dat.pl ‘But Godi didn’t like their faithlessness, but Øi sent them fire from heaven.’

In (1), the dative-marked RP (gode) is the privileged syntactic argument of the coordinative construction, as the coreferent RP was deleted in the second coordinated clause. In Modern English, on the other hand, the (pronominal) experiencer argument of the first verb would have to be in the nominative case (he rather than *him). Actually, only Icelandic preserves the pattern attested in Old English, which probably goes back to ProtoGermanic. It remains to be seen whether such a trend towards tighter correlation between case marking and PSA selection can be found in the history of other language families for which we have sufficient evidence, but RRG at least allows us to formulate such a hypothesis very precisely. On a more general level, it has been claimed that PSAs (or pivots, as they used to be called in earlier versions of RRG) generally tend to develop from two sources: either from the semantic notion of macroroles (Actor and Undergoer), or from the pragmatic notion of topic. The first scenario has been assumed for languages such as English (and presumably many more IndoEuropean languages), while the other development probably took place in the Philippine languages such as Tagalog (Foley and Van Valin 1984: 134–148), in which PSAs still have some topic-like features (for example, only PSAs can be relativized, and heads of the relative clauses may be seen as their topics, i.e. as what the relative clauses are ‘about’). If this is correct, in languages such as English, semantically motivated features such as case marking, which had correlated closely with semantic roles, became gradually de-semanticized and acquired purely syntactic functions. In languages of the Philippine type, on the other hand, pragmatically

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

motivated topic and/or focus markers gradually lost their pragmatic functions and became markers of PSAs. It remains to be seen whether this general diachronic typology of PSA development can be extended to other languages and language families. There may well be other directional and universal patterns of diachronic development that are precisely formulable in the RRG framework, but the current level of our knowledge is insufficient to make strong hypotheses. For example, in a number of ancient or archaic Indo-European languages the reflexive pronouns derived from the Proto-Indo-European root *s(w)e- can be used in the embedded clause to refer back to its antecedent in the main clause. This is the case with Latin se in (2). (2)

Pompeiusi a me petivit ut sei-cum et apud sei Pompey from me ask.3sg.ind that with.himself and at himself essem cottidie. be.1sg.sbjv daily ‘Pompey asked me to be with him and at his house, daily.’ (Gildersleeve and Lodge 1992: 325)

In terms of RRG’s layered structure of the clause we can say that the domain of reflexivization in such languages is the clause, since the reflexive marker is within the same (superordinate) clause, although it is not the argument of the same verb as its antecedent (i.e. it is not in the same core). On the other hand, in most modern Indo-European languages (including the Romance languages, which developed from Latin), the antecedent of reflexivization and the reflexive marker (derived from *s(w)e-) must be in the same core,3 which means that the domain of reflexivization is the core, not the clause. For example, the reflex of Latin se in French, the clitic se, cannot be used in the subordinate clause as coreferent with the subject of the main clause.4 It appears, then, that the archaic use of the reflexive marker found in Latin was changed in a way that led to the narrowing of the domain of reflexivization in many modern Indo-European languages (with the exception of Icelandic, which allows for ‘long-distance reflexivization’ but seems to be archaic in this regard, cf. Van Valin and LaPolla 1997: 613 and Wackernagel 2009: 515 for ‘long-distance reflexivization’ in Old Icelandic). It would be interesting to know if there are attested instances of diachronic developments of reflexive markers by which the domain of reflexivization in a language was broadened, rather than narrowed. This is clearly a worthwhile topic for future research.

17.3

Representing Syntactic Change in the RRG Framework

Since any aspect of the syntax and the levels of representation that are mapped to it in the linking are subject to change, representing diachronic

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

655

656

RANKO MATASOVIC´

processes involves showing how they affect different elements of syntactic representation and the linking algorithms. In RRG terms this means that diachronic changes can affect: • • • • •

the the the the the

inventory of syntactic templates inventory and arrangement of operators focus structure projection (the potential focus domain) logical structure of verbs linking algorithms (e.g. the case-assignment principles).

Representing changes in all the affected levels of representation gives us a more fine-grained picture of historical processes than is usual in theoryneutral accounts of diachronic syntax. To give an example, what would in pre-theoretical terms be described as change in clause alignment from ergative to nominative-accusative in northern dialects of Tabassaran (Harris and Campbell 1995: 249–250), in RRG involves specifying how this change affected all of the elements above. Tabassaran was originally an ergative language, just like the other languages of its family (NE Caucasian), with subjects of transitives in the ergative case and objects and intransitive subjects in the absolutive case. However, at some stage, topics were cliticized on verbs, and, as subjects were the most common topics, this cliticization became obligatory for subjects. At that time, the clitics still distinguished the ergative and the absolutive case of the subject. Southern dialects of Tabassaran still preserve this stage (note that the first-person singular pronoun has the same form in the absolutive and the ergative). (3)

uzu gak’wler urgura-za I.erg firewood.abs burn-I.erg ‘I burn firewood’

(4)

uzu urgura-zu I.abs burn-I.abs ‘I burn, am on fire’

In RRG terms, we would say that at this stage the following changes occurred: (1) The inventory of syntactic templates was enriched with a construction in which the topic was obligatorily copied in postverbal position. The full pronouns were preserved in the pre-detached position in such constructions (the default position for topics). (2) The focus structure was affected, as the topical element now had to be postverbal. (3) By a subsequent change, the focus structure was affected again, as the clitics lost topicality and became pure agreement markers expressing the lowest macrorole argument of the verb (the absolutive argument) and the highest argument of the verb (the ergative argument). This change affected the syntactic templates of the language, as the clitics now became integrated into the core.

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

Now, in the northern dialects, the verb-agreement system became entirely nominative-accusative, as they lost the distinction between the two types of verbal subject clitics. The form -za, which had originally been the first-person singular ergative, was generalized (the case marking was also affected, but this is a different issue). (5)

izu bisnu-za džaq’a I.erg catch-I bird.abs ‘I caught a bird’

(6)

izu t’irxnu-za I.abs fly-I ‘I flew’

In RRG terms, this means that the linking algorithm (case assignment for clitics) was affected at this stage, in that the rule that applies in them states that the highest macrorole (clitic) argument received the ‘nominative’ case, just as in Latin or any other nominative-accusative language, and unlike in Tabassaran’s closest relatives. This is only a step from a further change in which the clitics became fully integrated into verbal morphology as person endings agreeing with the ‘subject’ (the highest-ranking macrorole argument) of the verb. In this system – which is perhaps how contemporary Northern Tabassaran should be analysed – the language presumably acquired the head-marking feature by which the verbal suffixes, rather than free pronouns, should be interpreted as arguments, and this again affected the inventory of syntactic templates in the language. Similar developments can be observed in other languages, in which pronominal clitics, originally used in pragmatically marked constructions to mark direct and indirect objects, became agreement markers on verbs. This is what happened in several Balkan languages with ‘clitic doubling’, including Macedonian and Bulgarian, and a similar process can be observed in a number of Spanish dialects (especially in Argentinian Spanish, see Belloro 2004). Although in those languages the change from clitics in pragmatically marked constructions to agreement markers on verbs did not lead to alignment change, the initial steps of the process were very similar to what happened in Tabassaran, as the syntactic templates and focus structure were similarly affected. Thus, the RRG approach allows us to clearly see the similarities and differences in historical processes occurring in very different languages. It is a reasonable prediction that complex changes affecting several aspects of syntax will be cross-linguistically much rarer than simple changes affecting only a single syntactic structure or representation. This would mean that, for example, clause-alignment changes will be much rarer, cross-linguistically, than simple word-order changes, which affect only the syntactic templates of a language and not the other aspects of its syntax. However, other factors having little to do with pure syntax have to be taken

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

657

658

RANKO MATASOVIC´

into account, for example the discovery of areal linguistics that both word order and clause alignment are areally stable features of language, which are therefore unlikely to change except in some historically extraordinary circumstances (for example mass migrations of speakers of a language to a different language area). Thus, the Indian Subcontinent has apparently been characterized as a language area where the SOV languages have predominated for at least two thousand years and possibly much longer (Vedic Sanskrit, the ancestor of all Indo-Aryan languages, was SOV, and ProtoDravidian was also probably SOV). However, Romani, which is spoken by people who emigrated from Northern India roughly a thousand years ago, is SVO, like Armenian, Greek and most other languages it was affected by on the path from India to Europe. Synchronic generalizations about universal, or quasi-universal aspects of syntactic structure always have their bearing upon hypotheses about possible and probable patterns of change. Thus, it has been noted that the universal aspects of clause structure (the nucleus, core, periphery and clause) are all semantically motivated, while the non-universal aspects (the detached phrases, the extra-core slots) are pragmatically motivated (in the sense that they are associated with constructions that have strong pragmatic restrictions on their occurrence, Van Valin 2005: 8). It should be noted that these are empirical findings, not something postulated by the theory, but they have led to the hypothesis that semantically motivated features of language are diachronically stable, while pragmatically motivated features are rather unstable and more likely to change in the history of a language. Although this hypothesis appears to be motivated by the theoretical architecture of RRG, it has not been proved or subjected to systematic research beyond a rather impressionistic overview of the literature on grammaticalization (Matasovi´ c 2002). A theory of diachronic syntax must also be able to capture different types of syntactic change. Instances of syntactic change can generally be classified as one of three types: reanalysis, extension, and borrowing (Harris and Campbell 1995). Reanalysis is a process that involves potentially ambiguous syntactic structures which receive new structural representation and are parsed differently by language speakers before and after the change in question. Extension is the process of generalization of rules or constructions from one domain to others, and borrowing involves the adoption of constructions from one language into another. All three of these types of change can be represented in RRG, and their representation in the RRG framework allows the researcher to make a more fine-grained analysis than is usual in general accounts of historical syntax. A case of syntactic reanalysis followed by extension is presented in the RRG framework by Diedrichsen (2008). She shows how in colloquial German the nuclear juncture involving the verb bekommen ‘get’ and a number of ditransitive verbs of transfer was reanalysed as a passive complex verb form. An example of such a passive construction is (7):

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

659

Role and Reference Grammar and Diachronic Syntax (7)

Ich habe diesen Computer von meinen Eltern I have this.acc.sg Computer from my.dat.pl parent.dat.pl bekommen. got.ptcp ‘I was presented with this computer by my parents.’

Subsequently, this construction was extended to some other verbs with a meaning similar to bekommen (e.g. kriegen and erhalten), so that in the contemporary colloquial language it is possible to form a passive with these auxiliaries as well. Moreover, the construction was extended to a few monotransitive verbs (e.g. schneiden ‘cut’), so that a sentence such as (8) is now acceptable. (8)

Er bekommt die Haare von ihr geschnitten. he get.3sg.prs art.acc.pl hair.acc.pl from she.dat cut.ptcp ‘He gets his hair cut by her.’

Diedrichsen’s analysis allows us to see not only how reanalysis and extension work on the syntactic templates of the affected constructions, but also how such syntactic changes slowly spread through the vocabulary of a language, in a way that is affected by the lexical meanings of individual items. Borrowing of syntactic patterns can also be represented within the RRG framework. Its existence and, indeed, widespread occurrence, do not present a theoretical problem, since RRG does not make any theoretical assumptions about when and how syntactic change takes place. As far as the theory is concerned, syntactic change can occur both during language acquisition, as a consequence of imperfect learning, and in adulthood, either as a result of interference in the minds of bilingual speakers (syntactic borrowing), or for some other reason. Here, again, the theory allows us to represent different aspects of syntactic borrowing: firstly, borrowing can involve the copying of specific syntactic templates as abstract patterns, for example with respect to word order. This was the case when the SOV word order was borrowed in Ethiopian Semitic languages such as Amharic from neighbouring Cushitic languages such as Somali, or when the inversion rule in the formation of yes/no questions was borrowed into French from Germanic (Frankish), cf. French Vient-il? (comes-he) ‘Is he coming?’ with inversion, which is also found in Dutch and German, but not in most Romance languages. Secondly, syntactic borrowing can involve the borrowing of individual lexical items that has consequences for syntax, for example when a verb such as German schmecken ‘to taste’ was borrowed into dialectal Croatian (šmekati). Like German schmecken, which takes the dative experiencer and the nominative source (e.g. EsNOM schmeckt mirDAT gut ‘It tastes good to me’), Croatian dialectal šmekati has a quirky case frame (ToNOM miDAT dobro šmeka ‘It tastes good to me’). It remains to be seen whether such items can be borrowed if the recipient language does not already have verbs with such

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

geschenkt presented

660

RANKO MATASOVIC´

logical structures and case frames (Croatian has the inherited verb sviđati se ‘to like’, which has essentially the same syntax as the borrowed šmekati). This also applies to instances where concrete forms are not borrowed but the syntactic behaviour of individual lexical items is influenced by the syntax of their translational equivalents in another language (syntactic calques). For example, it is tempting to see the case frame of Croatian darivati ‘present (imperfective)’ and darovati ‘present (perfective)’ as a syntactic calque of Latin donare ‘present’, since both verbs show variable undergoer selection, allowing both the theme argument and the recipient to occur in the accusative (which is the default case of undergoers). This means that two different thematic roles (with different positions on the Actor-Undergoer Hierarchy) can be undergoers (Van Valin 2005: 111). The Croatian verb darivati can take either accusative of the recipient and instrumental of the theme (darivati djecuACC kolaˇcimaINS ‘to present children with cakes’), or the dative of the recipient and the accusative of the theme (darivati djeciDAT kolaˇceACC ‘to present cakes to children’), similarly to its Latin equivalent which takes either dative and accusative (donare alicuiDAT aliquidACC ‘to present something to someone’) or accusative and ablative (donare aliquemACC aliquareABL ‘to present someone with something’; Latin ablative is the usual equivalent of the Slavic instrumental case). However, Croatian darivati has etymologically cognate verbs in other Slavic languages, including Russian (darivat’), which also have two pairs of case frames parallel to Croatian, which makes the hypothesis that those case frames are inherited from Proto-Slavic more likely than the hypothesis that they were independently borrowed from Latin. Inheritance seems even more likely in light of the fact that verbs with such case frames are otherwise almost unattested in Slavic: in Croatian, the only other example is nuditi ‘offer’, which can also be construed with both the accusative and the dative (nuditi djeciDAT kolaˇceACC ‘to offer the cakes to children’) and the accusative and the instrumental (nuditi djecuACC kolaˇcimaINS ‘to offer children with cakes’). In the case of nuditi, a plausible explanation of its exceptional case frame and variable undergoer selection properties would be that it is the result of extension from darivati, which is inherited. The Latin verb meaning ‘to offer’ (offerre) could not have been the model for nuditi, as it can be construed only with the accusative and the dative (offerre aliquidACC alicuiDAT ‘to offer something to someone’). It seems fair to say that most lexical borrowing involves syntactic adaptation to pre-existing patterns: for example, one does not expect to find cases where verbs with ergative case frames are borrowed into a language with nominative-accusative morphosyntax, or verbs showing person agreement borrowed into a language with no person agreement. However, there are documented cases of borrowing of lexical items that either introduce new syntactic patterns into a language or contradict pre-existing ones. For example, Tagalog, a genderless language, has borrowed adjectives from Spanish, and those borrowed adjectives must agree in gender with their

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

head nouns in the RP (Schachter and Otanes 1972: 166–168; Matasovi´ c 2014); the same rule holds for adjectives borrowed from Sanskrit into Marathi. Conversely, Croatian, in which gender agreement in the RP is obligatory, has borrowed a number of adjectives from English that show no gender agreement (e.g. super ‘great, super’ and cool). Although borrowing is easily accounted for and represented in RRG, it is unclear whether there are constraints on grammatical borrowing in this theory, and it remains to be seen if the theory offers any clues to such potential constraints. For example, the borrowability of elements represented on the operator projection would need to be investigated. Research into language contact has long shown that grammatical morphemes – including inflexional elements – can be borrowed, though usually on a smaller scale than lexical morphemes, but it is unclear whether there is a difference of borrowability of different classes of grammatical morphemes. Since RRG explicitly distinguishes two classes of grammatical elements with respect to their syntactic representation, mapping some grammatical morphemes in the constituent projection (e.g. causative markers, or person markers of verbs), and others on the operator projection (e.g. aspect, tense and illocutionary force markers), it would be interesting to investigate whether these two classes differ with respect to borrowability. Again, it seems that borrowing syntactic templates – which may be influenced by pragmatic factors – is much more common than borrowing of elements representable on the operator projection, which is strongly semantically motivated. However, besides intuitions based on careful reading of literature on language contact, there is at present little systematic research to corroborate such a claim. Finally, in this chapter we have discussed the implications of RRG for a theory of diachronic syntax, but RRG is actually more a general theory of linguistic structure than a simple syntactic theory, since the interface of syntax, semantics and pragmatics plays a central role in its architecture and explanatory goals. Therefore, everything that has been claimed here about historical syntax also has consequences for historical semantics and pragmatics, and our discussion has shown a number of such consequences. Moreover, the boundaries between morphology and syntax are set somewhat differently in RRG in comparison to other theories, so diachronic RRG is also in part a theory of historical morphology. It has been established that most bound grammatical morphemes were in origin independent words with syntactic functions (‘Today’s morphology is yesterday’s syntax’, Givón 1971: 413), and this means that the change of their grammatical status will often affect how they are represented in RRG. Thus, many Classical Greek directional verbal prefixes (represented as nuclear operators in the operator projection) were still independent adverbs in Homeric Greek, where they would be represented as elements of the periphery in the constituent projection. For example, in Homeric Greek ep`ı ‘over, above’ is still separated from the verb teín¯o ‘extend’, as in (9):

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

661

662

RANKO MATASOVIC´

(9)

Ep`ı n` yks tétatai deilo~ısi above night.nom.sg extend.3sg.prf.middle poor.dat.pl ‘The night has extended itself over the weak mortals.’

broto~ısin. mortal.dat.pl

However, by the Classical period (5th century BCE) we find it only as the prefix in epiteín¯o ‘extend over, cover’ (Lindemann and Färber 2003: 47). On the other hand, not all diachronic developments that involve a free morpheme becoming bound need affect the way those morphemes are represented in RRG. For example, the clitic pronouns of Literary French, such as the je ‘I’ (from Latin ego), have largely become bound person-marking prefixes in colloquial French, but in both stages of its development je would probably be represented as an argument in the constituent projection. Only if the grammaticalization process is taken a step further, and je becomes a pure personagreement marking on the verb, while the former topicalized stressed pronoun (moi) changes its status and becomes the pragmatically unmarked subject pronoun (like English I or German ich), would the way the morpheme je is represented change, as moi would be linked to the argument position in the constituent projection (rather than being represented as the topicalized element in the pre-detached position, as in (10)). (10)

17.4

Moi, je parle français. I 1sg speak French ‘(As for me,) I speak French.’

Conclusion

It has been argued that Role and Reference Grammar provides useful tools in the investigation of diachronic processes affecting the syntactic structure of languages of various types. The fact that RRG is designed to be descriptively adequate in representing typologically very diverse living languages, and that its adequacy can be tested empirically on languages with unlimited corpora and living native speakers should mean that this theory is also an adequate tool for description of dead languages with limited corpora and no native speakers. Describing two synchronic stages of a single language within a uniform theoretical framework is a prerequisite for the objective analysis of the changes that occurred from one stage to the other, and it has been argued here that RRG is up to that task. We have also claimed that the theory makes interesting predictions about the probability of certain types of diachronic changes that can be tested empirically. Moreover, the architecture of the theory enables the investigator to formulate specific hypotheses about the directionality of changes on particular levels of syntactic structure, and we believe that fruitful applications of RRG lie in proposing and testing such hypotheses, for example with respect to the diachronic widening or narrowing of the scope of operators, the expanding or shrinking domains of certain syntactic processes such as reflexive binding or

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

agreement, and the interaction of the focus structure domains with the domains of purely syntactic processes. As the review of the published diachronic applications of RRG testifies, this theory is also well designed to cope with the description and explanation of cross-linguistically common grammaticalization patterns. Unfortunately, there has not been a lot of work in diachronic syntax within the framework of RRG, but it is to be hoped that historical linguists will realize the potential of this theory in the future.

References Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third-Person Clitic Clusters in Spanish. Unpublished MA thesis, University at Buffalo (SUNY) [available on the RRG website, https://rrg.caset.buffalo.edu/]. Cennamo, Michela. 2001. On the reorganization of voice distinctions and grammatical relations in late Latin. In Claude Moussy (ed.), De Lingva Latina Novae Questiones (Actes du X`eme Colloque International de Linguistique Latine, Paris-S`evres, 19–23 avril 1999), 51–65. Louvain: Peeters. Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in a RRG-perspective. In Kailuweit et al. (eds.), 87–145. Dryer, Matthew S. 1992. The Greenbergian word order correlations. Language 68: 81–138. Eschenberg, Ardis. 2005. The Article System of Umonhon (Omaha). PhD dissertation, University at Buffalo (SUNY) [available on the RRG website, https:// rrg.caset.buffalo.edu/]. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Gildersleeve, Basil L. and Gonzalez Lodge. 1992. Latin Grammar. Edinburgh: Nelson. Givón, Talmy. 1971. Historical syntax and synchronic morphology: An archaeologist’s field trip. Chicago Linguistic Society 7(1): 394–415. Gonzalez Orta, Maria M. 2002. Lexical templates and syntactic variation: The syntax–semantics interface of the Old English speech verb secgan. In Ricardo Mairal Usón and Miguel J. Pérez Quintero (eds.), New Perspectives on Argument Structure in Functional Grammar, 291–302. Amsterdam: John Benjamins. Harris, Alice C. and Lyle Campbell. 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Kailuweit, Rolf. 2008. Some remarks on RRG and grammaticalization: French verbal periphrases. In Kailuweit et al. (eds.), 69–86. Kailuweit, Rolf et al. (eds.). 2008. New Applications of Role and Reference Grammar. Newcastle upon Tyne: Cambridge Scholars Publishing. Kroch, Anthony S. 2001. Syntactic change. In Mark Baltin and Chris Collins (eds.), Handbook of Contemporary Syntactic Theory, 699–729. Oxford: Oxford University Press.

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

663

664

RANKO MATASOVIC´

Lightfoot, David W. 1979. Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, David W. 2001. How to Set Parameters. Arguments from Language Change. Cambridge, MA: MIT Press. Lindemann, Hans and Hans Färber. 2003. Griechische Grammatik. Teil II: Satzlehre, Dialektgrammatik und Metrik. Heidelberg: Winter. Matasovi´ c, Ranko. 2002. On representing syntactic change: Towards a theory of diachronic syntax. Suvremena lingvistika 53–54: 57–72. http:// hrcak.srce.hr/16335. Matasovi´ c, Ranko. 2008. Patterns of grammaticalization and the layered structure of the clause. In Kailuweit et al. (eds.), 45–57. Matasovi´ c, Ranko. 2012. Las construcciones reflexivas del croata desde una perspectiva histórica. In Ricardo Mairal Usón, Lilián Guerrero and Carlos González Vergara (eds.), El funcionalismo en la teoría lingüística: la Gramática del Papel y la Referencia, 171–186. Madrid: Akal. Matasovi´ c, Ranko. 2014. Adnominal and verbal agreement: Areal distribution and typological correlations. Linguistic Typology 18(2): 171–214. Michaelis, Laura. 1993. On deviant case marking in Latin. In Robert D. Van Valin (ed.), Advances in Role and Reference Grammar, 311–373. Amsterdam: John Benjamins. Nicolle, Steve. 2008. Scope and the functions of be going to. In Kailuweit et al. (eds.), 58–68. Ohori, Toshio. 1992. Diachrony in Clause Linkage and Related Issues. PhD thesis, University at Buffalo (SUNY) [available on the RRG website, https://rrg.caset.buffalo.edu/]. Ohori, Toshio. 1994. Diachrony of clause linkage. In William Pagliuca (ed.), Perspectives on Grammaticalization, 135–149. Amsterdam: John Benjamins. Roberts, Ian. 2007. Diachronic Syntax. Oxford: Oxford University Press. Roberts, Linda. 1995. Pivots, voice and macroroles: From Germanic to universal grammar. Australian Journal of Linguistics 15(2): 157–214. Rupnik-Matasovi´ c, Maja. 2008. Case assignment in Classical Greek. In Kailuweit et al. (eds.), 146–156. Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA: University of California Press. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and LaPolla, Randy. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wackernagel, Jakob. 2009. Lectures on Syntax (edited and translated by David Langslow). Oxford: Oxford University Press. Wiemer, Björn. 2008a. Applications of RRG in diachronic syntax: Overview and open questions. In Kailuweit et al. (eds.), 2–44. Wiemer, Björn. 2008b. Changing relations between PSA-selection, macroroles and case assignment: Insights from the diachrony of Slavic, Baltic and other Indo-European languages. In Kailuweit et al. (eds.), 157–203.

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

Role and Reference Grammar and Diachronic Syntax

Wilkins, David. 1991. The semantics, pragmatics and diachronic development of ‘associated motion’ in Mparntwe Arrernte. Buffalo Papers in Linguistics 91(1): 207–257.

Notes 1 This does not mean that such a development is entirely impossible, only that it is very improbable. Rare patterns of grammaticalization occasionally lead to unexpected homonymy of grammatical morphemes. In Omaha (a Siouan language) there is a set of articles which are also used as auxiliary verbs and evidential markers, and Eschenberg (2005: 176–206) convincingly shows how original copulas expressing orientation of their subject were reanalysed as classificatory articles. For example, Webaxu khe. Zi. (pencil be.horizontal yellow) ‘The pencil is positioned horizontally. It is yellow’ was reanalysed as Webaxu-khe zi (pencil-definite yellow) ‘The pencil is yellow’ (Eschenberg 2005: 183). 2 We use the Leipzig abbreviations (https://www.eva.mpg.de/lingua/resources/ glossing-rules.php) with the following addition: pret ¼ preterite. 3 Or the simple clause including the core and the pre-core slot (Van Valin 2005: 167). 4 I.e. one cannot say something like *Jeani pense que Pierre sei frappe (lit. Jean thinks that Pierre himself is.hitting) ‘Jeani thinks that Pierre is hitting himselfi’. French works exactly like English in this regard, but unlike Latin.

https://doi.org/10.1017/9781316418086.022 Published online by Cambridge University Press

665

18 Functional Acquisition Processes in Child Language Links to Role and Reference Grammar Richard M. Weist

18.1

Introduction

This chapter reviews a functional theory of language acquisition based on the proposal that children utilize their understanding of cognitive and communicative principles to construct a grammar that integrates semantic and pragmatic notions. The chapter explores child language data that are relevant to such issues as how layered clause structure, operator projection, predicate structure and grammatical relations are acquired within this communication-and-cognition framework. Specifically, the chapter explains how layered clause structure is founded on the cognitive notions of predicate and argument. The cross-linguistic research on the child’s emerging tense-aspect-modality system is related to issues of predicate structure and operator scope. An evaluation of the semantic and syntactic properties of grammatical relations is presented based on the acquisition data. The chapter presents a brief description of some of the basic assumptions of the functional theory of Role and Reference Grammar (RRG, Van Valin 2005). In the process of showing how the language acquisition data map to the RRG framework, the chapter includes contrasts with alternative theories, such as autonomous syntax theory. From the perspective of conceptual development, the infant-toddler is viewed as a relatively proficient information processor with the capacity to discover fundamental linguistic relationships in the spirit of the theory of Operating Principles (Slobin 1985). The chapter begins with a presentation of related views of the semantic structure of child language during the early multiple word phases of acquisition (Section 18.2). This is followed by a minimal presentation of the clause structure and operator scope proposed within the RRG framework as well as the basic properties of predicate structure. With a focus on Turkish, Japanese and Polish, the next section (i.e. 18.4) examines the role of predicate structure in the acquisition of tense and aspect. The case for

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

wide-scope processing of temporal, aspectual and modal information is presented in Section 18.5, then Section 18.6 shows the importance of restricted neutralization in the acquisition of grammatical relations, and the concept of linking rules is related to child language data. Finally, the acquisition of wh-questions is considered from different theoretical perspectives in Section 18.7.

18.2

Semantic Relations in the Early Multiple Word Stage

The three research projects reviewed in this section share the following properties: (1) they base their arguments on child language data, (2) they propose that children understand predicate-argument structure that incorporates semantic relations, (3) they do not find evidence to support a Chomsky-type theory of the acquisition process (see, for example, Chomsky 1965 or Pinker 1984), and (4) they share with Van Valin (1993: 2), ‘the conviction that grammatical structure can only be understood with reference to its semantic and communicative function’. However, none of the investigators credit the toddler with the capacity to process inflectional morphology (cf. the research on tense, aspect, and modality described in Section 18.4). Limited-Scope Formulae. In his seminal study of eleven children (20 to 26 months old) encompassing five different languages, Braine (1976: 4) presented the argument ‘that the first productive structures are formulae of limited scope for realizing specific kinds of meanings’. According to Braine’s observations, a limited-scope formula was ‘a rule that maps elements of a semantic representation into positions in the surface structure’ (p. 69). Five of the ten patterns summarized by Braine (pp. 56 and 57) are as follows, with a semantic value linked to the position pattern: (1) identification, ‘it/that’ þ X; (2) disappearance, ‘allgone’ þ X; (3) negation, ‘no’ þ X; (4) actor–action relations, Kendall swim, and (5) requests, ‘want/have-it’ þ X. Braine argued that ‘ALL early learning can be interpreted as a learning of position patterns’ (p. 66) and that the child language data do not support theories that attribute grammatical structure such as S ! NP þ VP and VP ! V þ NP to the language learner.1 In Section 18.6, I will return to an attempt to explain the emergence of wh-question formation in English with a Braine-type concept of ‘lexically-specific formulae’ (Rowland and Pine 2000: 157). ‘Case’ (or Semantic Function) Grammar. Bowerman (1973) investigated the early stages of word combinations in two children learning Finnish, Seppo and Rina, and one child learning English, Kendall. Citing Chomsky (1965), Bowerman summarized that ‘the constituent structure assigned to children’s utterances by transformational generative grammar is largely gratuitous’ (p. 222) and specifically, ‘sentence-subjects tend initially to be restricted largely to the semantic function ‘agent’ with a hand full of exceptions for some children’ (p. 189). As an alternative, Bowerman

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

667

668

RICHARD M. WEIST

proposed a ‘case’ grammar based on the semantic functions agent (i.e. the instigator of action) and object (i.e. the entity affected by action or state identified) (see Fillmore 1968). In fact, Kendall and Seppo produced exceptions to the ‘agent-as-subject’ regularity, for example, Kendall’s pillow fell and thread break and Seppo’s kissa putto ‘cat falls’, hirri pelä(sty) ‘mouse is-afraid’, and torni kaatuu ‘tower falls-down’. Furthermore, the agent/subject is not always animate, for example, helikopteri lentää ‘helicopter flies’. These ‘exceptions’ may preview the children’s understanding of more abstract grammatical functions. The data indicate that children know more than ‘limitedscope’ mapping rules (as Braine claimed), but the children were not yet credited with the capacity to process morphosyntactic information (cf. Toivainen 1980 for Finnish). Usage-Based Grammar and Verb-Island Constructions. Tomasello (1992) focused on the manner in which verbs emerged in the corpus of his daughter Travis from a cognitive-cultural perspective. According to Tomasello (1992: 23), ‘young children’s early verbs and relational terms are individual islands of organization in an otherwise unorganized grammatical system’, and further, ‘In the early stages, the child learns about arguments and syntactic marking on a verb-by-verb basis, and ordering patterns and morphological markers learned for one verb do not immediately generalize to other verbs.’ In contrast to the supposition that language emerges in the child guided by innate universal grammar, Tomasello (2003: 21–31) proposed that language is acquired by utilizing the principles of ‘intentionreading’ and ‘pattern-finding’. According to the theory, children learn a set of verb–argument constructions having a set of ‘slots’ to be filled with specific semantic relations, for example, the schema ‘X made (this) Y’ was realized as, Maria made this duck and Linda made ice. Before the age of 3, Tomasello (2000: 71) found no motivation for productive verb categories or inflectional morphemes. Hence, the verb in the construction [X singing] would not be classified as an ‘activity’ verb, and the verb in [X fall-down] would not be an ‘achievement’ verb. Further, neither the -ed in spilled Weezer milk (Travis, 1;9) nor the -ing in Weezer drinking the eggs (Travis 1;7) were judged to be productive morphemes coding tense and aspect because there was very little evidence for contrast at this phase of development (cf. a rare contrast: came off Grover (1;9) versus smoke coming out the chimney (1;11) (see Tomasello 1992, Appendix). Contrary to Tomasello’s predictions, crosslinguistic findings reveal the early influence of semantic relations on verb categories and the rapid emergence of contrasts in the finite morphology (see Tables 18.4 and 18.11). Theories of Early Combinations and RRG. In his paper relating functional linguistic theory to language acquisition, Van Valin (1991) presented an argument for the relevance of RRG to child language research. Van Valin began with the distinction between the ‘adaptationist’ and the ‘constructionist’ views of language acquisition. Given the adaptationist view, the child is innately programmed with a set of universal principles, and the

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

child sets parameters to conform to the target language (see, for example, Pinker 1984, Radford 1990, or Sano and Hyams 1994). Alternatively, according to the constructionist view, children learn language by utilizing cognitive structures such as linguistic information processing principles (see, for example, Slobin’s (1985) theory of ‘Operating Principles’). In their analysis of single and early multiple word utterances, child language researchers have observed the critical role that semantic and pragmatic principles play in their data. The most fundamental principles of acquisition that are derived from the child language data are also theoretical primitives in the RRG theory. Thus, the RRG theoretical framework has the potential to provide important insights into the question of how language is acquired. This chapter will explore RRG’s potential in this regard.

18.3

Clause Structure, Operators and Predicate Structure in RRG

Clause Structure and Operators. The components of clause structure are related to the scope of operators in Table 18.1 (cf. Van Valin, 2005: 12, Fig. 1.4). Within the layered structure of the clause (LSC), the semantic units motivate syntactic units at each layer of the structure, and grammatical categories called ‘operators’ modify the clause structure at different layers. In other words, the syntactic unit at a specified layer can be said to be within the scope of a set of operators. Operators that modify the nucleus have a narrow scope and those modifying the clause a wide scope. When I turn to the child language data, I will consider the hypothesis that there is a correlation between operator scope and the order of acquisition. Semantic Structure of Predicates. Initially motivated by the work of Vendler (1967) and Dowty (1979), Van Valin’s thinking about predicate structure has been refined from the four basic categories found in Foley and Van Valin (1984: 39, Table 2) through Van Valin and LaPolla (1997: 109, Table 3.4 ) to Van Valin (2005: 45, Table 2.3). Table 18.2 contains the lexical representations for the Aktionsart classes that are basic to the system and are most frequently cited in language acquisition research. Predicates in the four basic categories have the following properties: (1) States; static, (2) Activities; dynamic and atelic, (3) Accomplishments; dynamic and telic, and (4) Achievements; dynamic, telic, and punctual. Predicates that contain Table 18.1 Layered structure of the clause (LSC) and operator scope in RRG Semantic units

Syntactic units

RRG operators in layers with scope

Predicate Predicate þ Argument(s) (Arg.) Predicate þ Arg. þ Non-Arg. ¼ Proposition Proposition þ DP elements

Nucleus Core Clause ¼ Core þ Periphery Sentence

Directional, Negation (Neg.), Aspect Deontic modality, Directional, Neg. Illocutionary Force (IF), Evidentials, Tense, Status, Neg. None

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

669

670

RICHARD M. WEIST

Table 18.2 Lexical representations for four basic Aktionsart classes Aktionsart classes

Examples

Logical structure

State Activity Achievement Accomplishment

to be, know, want cry, sing, run around break, fall, find learn Polish, melt, make toys

predicate′ (x) or (x, y) do′ (x, [predicate′ (x) or (x, y)]) INGR predicate′ (x) or (x, y) BECOME predicate′ (x) or (x, y)

INGR (or instantaneous change) or BECOME (or change over time) in their logical structure have the telic property. The course of language acquisition is shaped by the interaction of the semantic structure of predicates with the layered clause structure and its operators. Cross-linguistic research has revealed a pervasive interaction of lexical and grammatical aspects. The next few sections of this chapter will probe the child language data that are relevant to this interaction revealing links to the RRG theoretical framework. Syntactic Realization of Logical Structure. Thematic relations are the semantic roles/functions of the arguments in logical structure (see Van Valin 2005: 58, Fig. 2.3). The linking system to syntax involves two semantic macroroles, Actor and Undergoer. These macroroles are generalizations across argument types: Actor across agent-type and Undergoer across patient-type thematic relations. The agent argument of DO is the least marked choice for Actor, followed by the first argument of do′ (x), while it is the most marked choice for Undergoer, while the patient/entity argument of state pred′ (x) is the least marked choice for Undergoer, followed by the second argument of pred′ (x, y), and the most marked choice for Actor. The Actor-Undergoer Hierarchy specifies the mapping of thematically linked arguments to macroroles, and, depending on language typology, mapping continues to the privileged syntactic argument (PSA), which is the RRG equivalent to syntactic subject (Van Valin, 2005: 127, Fig. 5.1). In a nominative-accusative language like English, Polish or Japanese, the PSA will be in the nominative case.

18.4

Predicate Structure and the Acquisition of Operators

Cross-linguistic research focusing on the acquisition of the finite morphology of the tense-aspect-modality (TAM) system of child language has consistently discovered a link between TAM morphology and categories of Aktionsart as follows: (1) past tense – perfective aspect – telic Aktionsart (e.g. achievements) and (2) non-past – imperfective aspect – atelic Aktionsart (e.g. activities) (see, for example, Weist 1986, 2014a, b). The relationship between the TAM domain of finite morphology and elements of predicate structure is particularly theoretically revealing, as the evidence has been found at a very early phase of language acquisition, and it extends across a

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

large variety of languages. From the RRG perspective, the acquisition data provide insights into the relationship between the predicate that shapes clause structure and the operators that modify levels of clause structure. Predicate Structure and Operators in Turkish. I begin with a study by Aksu-Koç (1998) on child Turkish because the investigation involved the following four qualities: (1) it was sensitive to the emergence of TAM morphology in a very young child (Deniz 1;3 to 1;11), (2) it included child-directedspeech, (3) the classification of Aktionsart was empirically motivated, and (4) the results revealed an unambiguous link to Aktionsart. Four multifunctional affixes of Turkish were analysed, each having a value of tense, aspect and modality (i.e. specifically evidential). The affixes with their meanings and examples are as follows: (1) -dI past, perfective, and direct experience, gel-dI ‘he came’; (2) -Iyor present and progressive, gel-Iyor ‘he is coming’; (3) -Ir habitual-generic and possibility, gel-Ir ‘he comes’; and (4) -mI¸s past/present, resultative, and indirect experience, gel-mI¸s ‘he has come evidently’. A set of tests motivated by Dowty (1979) and organized for child language research by Shirai and Andersen (1995) were used to categorize Deniz’s verbs into categories of Aktionsart (i.e. state, activity, achievement and accomplishment). Aksu-Koç (1998: 263 and 264) identified four stages of development as follows: 1. 1;5.9 to 1;9.9: emergence and restricted use of -dI with achievement verbs (e.g. bul- ‘find’) 2. 1;7.3 to 1;7.8: flexible use of -dI and emergence and restricted use of -Iyor with activity (e.g. ye- ‘eat’) and state verbs (e.g. yat- ‘lie:down’) 3. 1;7.23 to 1;8.14: flexible use of -dI and -Iyor and emergence and restricted use of -Ir and -mI¸s 4. 1;8.27 to 1;10.19: flexible use of all four forms. Throughout these stages, the distribution of the target morphemes in Deniz’s corpus closely resembled the distribution in her mother’s childdirected speech. In her earlier research, Aksu-Koç (1988) studied the acquisition of the inflectional morphology of Turkish in three children: ES (1;9–1;11), YK (1;11–2;1) and SÖ (2;2–2;5). A ‘restrictive stage’ was not found, and the children were in the ‘flexible use’ phase during the earliest sample. For example, in early samples, the three children inflected activity verbs with -dI (i.e. past, perfective, direct) one-third less frequently than verbs involving a change-of-state. Predicate Structure and Operators in Japanese. Shirai (1993) investigated the relationship between tense-aspect morphology and Aktionsart categories in the corpus of Sumihare (Noji 1976, 1977, cited by Shirai). The research focused on three tense-aspect morphemes: -ta (past), -te i- (durative), and -ru (non-past). Regarding the interaction of tense-aspect morphology and Aktionsart in Japanese, the durative/continuative marker -te i- has an action-inprogress meaning with activity and accomplishment verbs and a resultative state meaning with achievement verbs. Within the Sumihare corpus, Shirai

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

671

672

RICHARD M. WEIST

Table 18.3 The percentage of verb tokens produced with -ta and with -te i- by Sumihare for two predicate categories and three age periods -ta

-te i- (-teiru, -teru, -toru)

Aktionsart categories

1;6

1;7

1;8

2;1

2;2-1

2;2-2

Activity Achievement

11.1 73.3

7.9 65.8

28.3 58.7

81.8 18.2

52.6 47.4

47.9 49.3

(1993: 200 and 201) analysed the emergence of -ta from 0;11 to 1;8 and -te ifrom 1;11 to 2;2. Table 18.3 contains the percentage of verb tokens produced with -ta and with -te i- by the child Sumihare during three particularly salient acquisition periods. For -te i-, the interval of 2;2 was split in half due to the rapid development at age 2;2. The results show the relationship between elements of predicate structure and tense-aspect morphology. At an early age, the relatively high percentage of verb tokens were found in achievements with -ta (i.e. ‘past/perfective’) and activities with -te iþru (i.e. the actionin-progress meaning). As the developmental process rapidly unfolded, the likelihood of activity verbs with -ta increased, and the likelihood of achievement verbs with -te i- (i.e. the resultative state meaning) increased. As was found with Deniz learning Turkish, the link between the do′ (x) predicate structure and present-progressive meaning and between INGR pred′ (x) structure and past-perfective meaning was found in the Japanese child Sumihare. However, like Aksu-Koç’s (1988) Turkish findings, there was no ‘restrictive’ stage since Sumihare inflected activity verbs as well as achievement verbs with -ta. Shirai (1998) continued his investigation of the relationship between tense-aspect morphology (i.e. -ta, -te i-, and -ru) and categories of Aktionsart with three more Japanese children: Aki (1;5 to 3;0) (Miyata 1995, cited by Shirai), Yocchan (1;11 to 2;2) (Clancy 1985), and Taachan (1;10 to 2;2) (Kokuritsu, Kokugo and Kenkyujo 1982, cited by Shirai). When the tense-aspect morphemes became productive, each child had a relatively high frequency of verb types in one of the Aktionsart categories. For Aki, achievement -ta verb types were the most frequent at 2;1 and remained highly frequent to 2;7, and activity-te i- verb types emerged as the most frequent at 2;4 and remained so to 2;7, supporting Shirai’s (1993) investigation of Sumihare. Aki’s mother’s child-directed speech was also analysed, and in general, the relationship between tense-aspect morphemes and Aktionsart categories supported Aki’s acquisition pattern. However, the ‘expected’ pattern was not complete for Yocchan and Taachan. Following expectations, at productive usage, the achievement category was closely linked to past -ta, but this was not the case for durative -te i-. Hence, the expectations of Aksu-Koç’s ‘restricted stage’ are not always realized. Predicate Structure, Operators and Finite Morphology in Polish. Weist et al. (2004) utilized the corpora of six children learning Polish (1;3– to

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

4;11) and six children learning English (1;2 to 4;11) from the CHILDES archives (MacWhinney and Snow 1985) to investigate the emergence of tense-aspect morphology within the RRG theoretical framework. The method of investigation was ‘predicate tracking’. Employing Dowty-type tests, the process began by assigning verbs to four basic Aktionsart categories (i.e. state, activity, accomplishment and achievement). Following Aktionsart classification, dynamic predicates were partitioned into telic (i.e. accomplishment and achievement) and atelic (i.e. activity) sets (see Table 18.2). Polish has perfective and imperfective aspect where imperfective is the least marked form, and English has progressive and nonprogressive aspect where the non-progressive (or ‘simple’) form is the least marked. In Polish, there is an interaction of tense and aspect such that non-past perfective verbs have future-tense meaning, while non-past imperfective verbs have present-tense meaning. Imperfective future has a periphrastic form. English has quasi-modal (i.e. going-to/will) future forms. With two values of aspect and three values of tense, there might be six inflectional forms to evaluate. However, present tense in perfective aspect does not exist in Polish, and the children learning English in this study did not use the future tense progressive aspect form. In order to balance the cross-linguistic comparison, we reduced the statistical analyses to four tense-aspect forms in each language: 1. Polish: (a) perfective past and non-past (future meaning) (b) imperfective past and non-past (present meaning) 2. English: (a) non-progressive past and (going-to / will) future (b) progressive past and present. The average age of the acquisition of the tense-aspect morphology for atelic and telic predicates in Polish and English is presented in Table 18.4. The values in bold print refer to the telic data. Each value is calculated by summing over the set of atelic and telic predicates for each child, and then summing over the six children in the language group. Thus, Table 18.4 represents a relatively highly productive phase in the acquisition process. The Polish data reveal an exceptionally clear interaction between the property of telicity within predicate structure and grammatical aspect. If the predicate is telic, perfective forms are acquired first, and in contrast, if the predicate in atelic, the imperfective forms are acquired first. The underlying form of predicate structure shapes the acquisition pattern. This lexical– grammatical aspect interaction is also found in the English corpora, but not as strongly. From the perspective of the RRG theoretical framework, it is obvious that an adequate theory of language acquisition must include an understanding of predicate structure (see Table 18.2), and predicate structure will have to have a primitive component within clause structure (i.e. the nucleus for RRG; see Table 18.1).

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

673

674

RICHARD M. WEIST

Table 18.4 The average age of the emergence of tense-aspect forms for (atelic/telic) predicates (a) Polish

Aspect

Tense

Perfective

Imperfective

Past Non-past (Present) Non-past (Future)

3;4 / 2;2 Non-existent 2;9 / 2;3

2;7 / 3;1 2;1 / 2;8 * {2;5 / 3;0}

* {} The future imperfective data were scored but could not be compared to English.

(b) English

Aspect

Tense

Non-progressive

Progressive

Past Non-past (Present) Non-past (Future)

2;11 / 2;7 Excluded 2;11 / 2;8

3;2 / 3;5 2;10 / 2;10 *{2;4 / 2;7} Not observed

* {} ‘Present’ progressive forms were scored with and {without} the auxiliary.

The Weist et al. (2004) predicate tracking study evaluated three dependent variables. In addition to the age of emergence (see Table 18.4), the likelihood that a tense-aspect form would occur and contrasts in tenseaspect forms were also investigated. The likelihood data complement the emergence data; for example, a form that emerged early was highly likely. Contrast is an important indicator of acquisition. Children acquiring a relatively inflected language like Polish do not hear bare stems and neither do they produce them. A tense and/or aspect morpheme may emerge as a ‘frozen’ form, and contrast provides evidence for functional status. Three types of contrast were evaluated for all predicates (i.e. telic and atelic combined) (1) tense only (e.g. past perfective / future perfective), (2) aspect alone (e.g. past perfective / past imperfective), and (3) tense-aspect (e.g. past perfective / present imperfective). The contrast findings are presented in Table 18.5. Contrary to conventional wisdom, there was no evidence to support the claim that aspect is acquired before tense. Furthermore, there was considerable evidence that tense morphology codes deictic relations. Weist et al. (2009) conducted a second predicate tracking study in order to investigate the impact of predicate structure on the acquisition of agreement and tense morphology. The data for this study were found in the corpora of six children learning Polish (1;0 to 4;11) and six children learning English (1;2 to 4;11) taken from a variety of sources (see, for example, the Kraków project, Smoczy´ nska 1985). The set of predicates being tracked were either stative or dynamic, with the dynamic set further classified as atelic versus telic. The telic subset included predicates with

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

Table 18.5 The average age of the emergence of tense-aspect, tense, and aspect contrasts summing over values of lexical aspect Language

Tense-Aspect

Tense

Aspect

Polish English

2;10 3;0

2;4 3;0

2;11 3;3

Table 18.6 The percentage of agreement and tense contrasts for the three predicate types in Polish and English Language

Polish

Aspect

PFV

Type

Agreement

Stative Atelic Telic

Impossible 18 96

Type

Tense

Stative Atelic Telic

Impossible 6 97

English IPFV

NPROG

PROG

100 100 46

100 58 81

Non-standard 80 75

100 100 34

100 95 98

Non-standard 65 35

BECOME/INGR in their logical structure. The single dependent measure was contrast, and in one analysis, the contrast findings were evaluated with grammatical aspect included, as shown in Table 18.6. Once again, the impact of predicate structure as well as language typology is evident. In both languages, contrasts in finite morphology are likely to be found in dynamic predicates when bounded grammatical aspect is combined with the semantic property of telicity. Regarding typology, the difference in the behaviour of stative predicates is the most obvious. In Polish, the perfective form of a stative predicate has inchoative meaning, making it telic (e.g. lubi´c/ polubi´c ‘to like / to get to like’). In English, progressive aspect codes the property ongoing, which is absent in the logical structure of a stative predicate. The influence of language typology is also revealed in the acquisition of dynamic predicates due to the fact that the marked aspectual form in Polish is linked to external perspective (i.e. perfective (PFV)), and in English the marked form is linked to internal perspective (i.e. progressive (PROG)). In Polish, for example, the perfective member of an atelic aspectual pair has delimited meaning, for instance, for pobawi´c/bawi´c ‘to play’ the perfective pobawi´c means ‘to play for a while’ (see Weist et al. 2004: 50 for examples). The delimited semantics may require additional learning (see also Stoll 1998 for Russian). Weist et al. (2009: 1335, Table 7) also found the average age and range of the initial agreement and tense contrasts overall, that is, without

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

675

676

RICHARD M. WEIST

distinguishing values of grammatical aspect. In general, there was considerable overlap in the range of the acquisition of agreement and tense morphology, that is, there was no evidence for a sequence (cf. Meisel 1994: 90, where UG derives an agreement then tense acquisition sequence). Regarding typology, agreement and tense morphology were acquired more rapidly in the Polish children. Weist et al. (2009: 1329–1331) outlined the typological differences between the English and Polish finite morphological systems, and they show how information processing is facilitated within the Polish system. Within the scope of Operating Principles theory, Slobin (2001: 441) referred to this process as ‘typological bootstrapping’. In summary, language acquisition theories based on limited-scope formula, or slot-and-fill lexical concepts, or variations on these themes, have no prospects (i.e. theoretical mechanisms) to explain the emergence of contrasts in finite morphology which are paired with communicative competence. Utilizing their capacity to process linguistic information, children gain access to agreement and tense-aspect distinctions at a relatively early phase in development. The evidence within discourse contexts demonstrates the children’s understanding of the relevant contrast (e.g. 2nd versus 1st person and past versus future tense).

18.5

Clause Structure and Operator Scope: Two Hypotheses

‘Layer-emergence’ Hypothesis. The relationship between the layers of clause structure and the scope of operators within the RRG framework is outlined in Table 18.1. Applying this structural relationship to the language acquisition process, it might be argued that operators modifying ‘simple’ layers of clause structure should be acquired before operators modifying ‘complex’ layers (see Van Valin 1991). The argument went as follows: ‘Hence the status of an operator as nucleus, core, or causal, which is a function of its inherent semantic complexity, is the source of the increasing “cognitive complexity” that Weist and Stephany claim to be the explanation of the order of acquisition’ (Van Valin 1991: 18).2 Since aspect modifies the nucleus and tense modifies the clause, aspect morphology should emerge in child language before tense morphology: the child should code temporal contour before temporal relations between speech time and event time. I will refer to this expectation as the ‘layer-emergence’ hypothesis. The early research on aspect and tense in child language appeared to support this prediction; see, for example, Aksu-Koç (1988), Antinucci and Miller (1976) and Bronckart and Sinclair (1973). These investigators observed the emergence of finite morphology with the potential to code both temporal contour (i.e. completed) and temporal relations (i.e. event time prior to speech time) (e.g. -dI (past-perfective-direct) for Turkish). Regarding the child language data, it was claimed that past-tense morphemes were never found with the following properties: (1) inflecting activity verbs, (2) combined with

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

imperfective aspect, and (3) making reference to remote prior events. With deference to Piaget’s theory of conceptual development (Piaget 1954, 1971), the child language researchers ruled out the deictic function of past-tense morphology, for example, ‘the child lacks an abstract conception of time that would allow him to construct, between any two events X and Y, the relation Event Y precedes Event X, even when no other type of relation holds between event X and Y’ (Antinucci and Miller 1976: 184). Combining longitudinal observations with experimental findings, Weist et al. (1984) discovered that none of the three claims about the emergence of past-tense morphemes were correct for child Polish. During spontaneous caregiver– child interactions and within controlled elicitation conditions, 2-year-old Polish children produced activity verbs inflected for imperfective aspect in past tense referring to moderately remote prior events. In fact, the Polish findings are supported by the Turkish (Aksu-Koç 1988) and Japanese (Shirai 1998) data, revealing a low frequency of activity predicates inflected for past tense (see Table 18.3 for Japanese). Regarding the Piagetian conceptual limitation argument, decades of research with infants and toddlers has demonstrated that young children have the capacity to remember past experiences (see, for example, Bauer 1996, 2007 and Rovee-Collier 1997). The idea that the concept of aspect is functional in child language prior to the concept of tense may also be influenced by a methodological artefact. Child language data have often been collected in a ‘here-and-now’ restricted environment; see, for example, Aksu-Koç (1998: 261) ‘During the sessions Deniz and her mother played with toys, read, looked at pictures in books, and talked.’ In contrast, Bowerman (1981: 1 and 2) took ‘copious notes’ on her daughter’s utterances throughout the day, and she found that, ‘Spontaneous reference to both past and future (and nonactual) events was well established from the one-word period.’ Some of Bowerman’s observations of Christy’s temporal utterances are found here in Table 18.7. Christy’s capacity to talk about the ‘not-here-and-now’ is quite consistent with the developmental research on memory processes. In other words, children who can think about prior and subsequent events can also talk about these events if given the opportunity. Table 18.7 shows that when Christy was producing uninflected two- and three-word combinations, she was locating event time prior-to, at, and subsequent-to speech time. As the inflectional morphology emerged, her use of tense matched the utterance context. Regarding past time utterances, the Aktionsart of her verbs was atelic as well as telic. Relating to the ‘layer-emergence’ hypothesis, while wider-scope operators, such as tense, incur greater linguistic complexity, the more recent developmental findings indicate that young children have the capacity to process this information (see also Table 18.5). ‘Operator-Scope’ Hypothesis: Temporal Domain. Within the RRG framework, the concept of scope defines the relationship between operators and layers of clause structure. The ‘operator-scope’ hypothesis predicts that children understand scope relations as the TAM morphology emerges.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

677

678

RICHARD M. WEIST

Table 18.7 Melissa Bowerman’s observations of her daughter Christy’s temporal utterances Tense-aspect morphology absent Time

Utterance

Context

Past Present Future

(1;5). Write Sissy. (1;6) Kiddicar ride. (1;9) Pie open later.

She notices where her sitter had written on her hand. Said while playing with the kiddicar. She anticipates taking an unopened pie home. Tense-aspect morphology present

Past Past Past Present Future

Activity (2;0) I cried. Achievement (2;0) I didn’t fall down water. Emily did. (2;0) Balloon was outside. (1;10) Mommy Ø coming now. (2;1) She’s won’t get wet.

Remembering a walk earlier in the day. Remembering an earlier incident in a wading pool. Yesterday, her balloon was outside. Said as her mother arrives at the door. Said holding a doll while sitting on the potty.

Regarding the temporal domain, as children acquire the concept of grammatical aspect, we expect to find evidence for a link to the concept of nucleus, and as the concept of tense emerges, the relationship between the concept of tense and the clause layer of structure will be found. Regarding the link between the aspect operator and the nucleus layer, the evidence for the link is decisive (see Table 18.4). The predicate tracking research shows that predicates with the property of telicity in their semantic structure, for example achievements, will emerge initially in perfective aspect in Polish and non-progressive in English, and predicates lacking this property (i.e. activities) will emerge first in the imperfective aspect in Polish and progressive in English. This kind of operator to nuclear layer relationship has been found cross-linguistically. When tense morphology emerges, is there additional evidence that children understand a link to clause structure? The argument can be made that this link can be found in the relationship between tense and temporal adverbs. According to Van Valin and LaPolla (1997: 169), ‘adverbs are like operators, in that they modify different layers of the clause’. More specifically, temporal adverbs are viewed as core modifiers. Thus, if temporal adverbs emerge in child language during the same phase of acquisition as tense, we will have evidence for a link to layers of clause structure. Weist and Buczowska (1987) studied the emergence of time-related adverbs in Polish. Considering the data from three children in the age range 2;4 to 3;2, Weist and Buczowska analysed utterances containing adverbs within their discourse context. Table 18.8 contains the age of the initial occurrence of each of the nine adverbs in at least one of the three children. The set of adverbs was partitioned into three categories: immediate, cyclic, and remote. Cross-linguistic research has shown that when deictic adverbs specifying a daily cycle emerge, they are likely to enter into mismatches with tense. In Wawrzon’s sentence Jutro miałem ponton a ju˙z si˛e zepsuł ‘Tomorrow,

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

Table 18.8 The initial occurrence of the temporal adverbs in three sets of adverbs Cyclic*

Immediate

Remote

Adverb

Age

Adverb

Age

Adverb

Age

ju˙z ‘already’ teraz ‘now’ zaraz ‘soon’

2;4

wczoraj ‘yesterday’ dzisiaj ‘today’ jutro ‘tomorrow’

2;7

dawno ‘long ago’ kiedysˊ ‘in the past’ pó´zniej ‘later’

2;9

2;4 2;5

2;5 2;7

2;11 2;9

* The initial occurrence of these adverbs always involved at least one tense-adverb deictic contradiction.

(I) had (a) pontoon and already it broke’, the adverb jutro is pointing to the future, and the tense (-ł-) is pointing to the past. From the perspective of cognitive development, Friedman (1978: 281) argued that such incongruities signalled a deficiency in the child’s early tense system. However, to the contrary, the Polish data showed that children initially coordinate the ‘immediate’ temporal adverbs (ju˙z ‘already’, teraz ‘now’, and zaraz ‘soon’) with tense. The tense–adverb incongruities emerge at a later phase of acquisition when children are learning the meaning of the ‘cyclic’ adverbs. The child’s earlier understanding of tense remains intact. While a direct comparison of the emergence of temporal adverbs with tenses was beyond the scope of the Weist and Buczowska study, the Polish data cited in Table 18.4 indicate a close proximity. Relating to the ‘operator scope’ hypothesis, the Polish data support the claim that children demonstrate an understanding of different layers of structure when tense-aspect morphology is acquired.3 The aspectual adverb ju˙z ‘already’ modifies the nuclear layer and the temporal adverbs teraz ‘now’ and zaraz ‘soon’ (not to mention the cyclic and remote adverbs) modify the core layer. ‘Operator-Scope’ versus ‘Layer-Emergence’ Hypothesis: Modal Domain. In the temporal domain, the ‘layer-emergence’ argument proposes that morphemes with the potential to code deictic relations (i.e. tense) in the adult’s language code temporal contour relations (i.e. aspect) in the child’s language. Within the RRG framework, this argument is motivated by the fact that aspect modifies a more primitive layer of clause structure than tense. There is a companion argument in the modal domain (Van Valin 1991: 17, Fig. 3). Since the operator ‘modality’ (covering deontic modality) is a core operator and status (including epistemic modality) is a clause operator, along with evidentials, deontic modality should be acquired before epistemic or evidential modality. Considering data from children learning Greek and English, Stephany (1986) presented the argument that ‘Epistemic modal meanings develop later than deontic ones in language acquisition’ (p. 393) and furthermore, the reason for this acquisition sequence can be found in the Piagetian view of cognitive development. According to Stephany (1986: 393), ‘Studies of

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

679

680

RICHARD M. WEIST

Table 18.9 Sentence examples in the acquisition of existential meaning in Korean Age

Morpheme / knowledge

Child’s initials and context

Sentence

1;9

-TA new, direct

TJ just finished a visual search

1;11

-E assimilated

PL overhears a conversation about her recent haircut

2;2

-CI shared

PL recalls having seen a monster on TV

2;4

-TAY indirect

TJ talks to mother about a TV talking bird

Chaca-ss-TA Find-PAST -TA ‘(I) found (it)’ Polami meli ippukey hay-ss-E Hair prettily do-PAST -E ‘Had hair done, it was pretty’ Kweymwul Tibi-ey-na iss-CI? Monster TV-LOC -only exist-CI? ‘Monsters exist only on TV?’ Sayka ppai ha-n-TAY Bird bye do-PRESENT -TAY ‘Bird says “bye”’

cognitive development have shown that the notion of possibility as a distinct form of reality develops in Piaget’s pre-operational stage (from about 2 or 3 to about 7 or 8).’ Just as Piaget’s claims about memory processes during the sensorimotor stage were deficient, so too were his claims about the child’s conceptual capacity during the ‘pre-operational’ stage (see, for example, Mandler 1990; Newcombe and Huttenlocher 2003; Spelke 1991). Choi’s (1991, 1995) investigation of verb-final suffixes in Korean is relevant to the operator-scope hypothesis. However, the child Korean findings do not map directly to the Stephany (1986) and Van Valin (1991) proposed deontic to epistemic acquisition sequence. Choi evaluated the acquisition of desire and intention modal suffixes (cf. deontic, obligation and permission (Van Valin, 2005: 9: Table 1.2)) and evidential modals. While the prototypical deontic to epistemic sequence can’t be evaluated, the very early function of the evidential modals (i.e. clausal operators like tense) can be evaluated. Choi analysed the emergence of four evidential morphemes (-TA, -E, -CI and -TAY) and two other morphemes (-LAY desire and -KKEY intention)4 in the corpora of three children between the ages of 1;8 and 2;11. The evidential morphemes have the following meanings: (1) -TA: new knowledge acquired through direct experience and being assimilated, (2) -E: previously assimilated knowledge, (3) -CI: previously established, certain, and shared knowledge, and (4) -TAY: knowledge learned from an indirect source. Table 18.9 contains sentence examples together with the utterance context revealing the nature of the knowledge. The evidential morphemes became productive prior to the desire and intention morphemes. The Korean data demonstrate the robust nature of the infant-toddler’s information processing capacity and their ability to understand the semantic complexity at different layers of clause structure.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

18.6

Grammatical Relations

Restricted Neutralization. Does child language contain grammatical relations, and if so, what kind? From the RRG perspective, it depends on the language. While semantic roles are universal, syntactic functions are not, making it very implausible that they are innate. According to Van Valin and LaPolla (1997: 274), ‘grammatical relations exist only where there is a restricted neutralization of semantic or pragmatic relations for syntactic purposes’. Languages can be classified based on the nature of their restricted neutralization possibilities, as summarized in Table 18.10. For typological distinctions, S refers to the single argument of an intransitive verb, A refers to the Actor of a transitive verb, U refers to the Undergoer of a transitive verb, and d-S refers to, ‘the single NP of an intransitive verb derived via voice from a transitive verb’ (Van Valin and LaPolla 1997: 268–269). Returning to Section 18.2, in her investigation of the acquisition of Finnish and English, Bowerman (1973) did not think that there was enough evidence for restricted neutralization to justify or motivate the concept of subject. While Kendall produced utterances like pillow fall, thread break, the utterances of this form were infrequent and viewed as part of the ‘hand full of exceptions’ to the ‘agent-as-subject’ rule. According to Tomasello (1992, 2003), item-based schemas included one or more ‘slots’ and a verb. The slots were filled by semantic functions (see also Section 18.2). Once again, the only ‘grammatical’ relations were semantic in nature, supporting the assertion that semantic roles are universal. Following Van Valin’s argument, we need to find evidence for restricted neutralization in the child language data to make the case for syntactic relations. Hence, Bowerman, Tomasello, and Braine (as well) would compare child language to Acehnese and not to one of the target languages, for example, English. As discussed in Section 18.4, Weist et al. (2009) used the predicate tracking methodology to discover how finite morphology emerges in children learning Polish and English in the age range from 1;0 to 4;11. By tracking the emergence of this morphology within a set of predicates within each individual child, it is possible to determine how the concept of restricted neutralization develops. Within each language, agreement contrasts emerge at a similar age across predicate types (see Table 18.11). In general, the Table 18.10 Typology of restricted neutralization of semantic roles (Van Valin and LaPolla 1997: 269)

Language

Intransitive verbs

Transitive verbs

Grammatical relations

‘Subject’ classification

Acehnese Warlpiri English Dyirbal

no yes yes yes

no no yes yes

No Yes Yes Yes

Semantic [S,A] [S,A,d-S] [S,U,d-S]

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

681

682

RICHARD M. WEIST

Table 18.11 Average age and range of initial person agreement contrasts in English and Polish English

Polish

Predicate types

Age

Range

Age

Range

Stative Atelic Telic

3;2 3;1 3;1

2;7–3;8 2;5–3;9 2;4–3;4

2;1 2;4 2;4

1;11–2;3 2;0–2;7 2;1–2;6

acquisition process is more rapid in Polish, the more highly inflected language. Limiting the analysis to Polish, we investigated the children’s capacity to transfer their knowledge of agreement from one tense to a second tense. We calculated the number of months it would take a child to make the transfer (i.e. lag time), and we found that for some of their predicates, all of the children transferred their capacity to form contrasts in agreement from one tense to another within one month (i.e. zero lag). By approximately 2;4, restricted neutralization is a property of child Polish, but exactly how do we classify child Polish at this phase of development? At 2;4, it appears that child Polish should be classified as an [S,A] type like Warlpiri at least. However, like English, Polish has passive voice, and the Weist, et al. (2009) study did not evaluate this possibility. This possibility was the focus of a paper by Weist (1990). Weist (1990) investigated all of the facets of restricted neutralization in child Polish, including the concept of voice. The research included the corpora of four Polish children with ages ranging from 1;7 to 1;11 for two children and 2;1 to 2;6 for the other two children. The children’s TAM system was productive with contrasts in tense, aspect and mood. Transitive verbs were sometimes observed with a direct object, and intransitive verbs never were. The children contrasted nominative with accusative case, and their verbs agreed with the privileged argument in person, number and gender. During the period from 1;6 to 2;0, the children produced intransitive verbs with a theme or patient as the core argument (i.e. with the macrorole Undergoer) (e.g. spasˊ´c ‘to fall down’, przyjsˊ´c ‘to come/arrive’, and uderzy´c si˛e ‘to get hit’), and they produced intransitive verbs with an agent core argument (i.e. with the Actor macrorole) (e.g. ˊspiewa´c ‘to sing’, płwa´c ‘to swim’, and bawi´c si˛e ‘to play’). Regarding the pseudo-reflexive particle si˛e (refl), the children followed the rules for si˛e placement as follows: (1) immediately after the verb: Marta (1;8) kr˛eci si˛e ‘(it) turns’, and (2) optionally, after the first stressed element in the sentence: Marta (1;10) co si˛e kr˛eci? ‘what is turning?’ versus Marta (1;9) czemu kr˛eca si˛e kóleczka? ‘why are the little wheels turning?’. In Polish (child as well as adult), the transitive/intransitive contrast is distinctive, as the following observation demonstrates: Kubusˊ (2;4) Otwiea-m. Tak si˛e otwiera-Ø ‘(I) am opening (1sg) (the box). (It) opens (3sg) like this’. Here it is clear that Kubusˊ has shifted from the transitive verb otwiera´c to the intransitive otwiera´c si˛e. The transitive verb otwiera´c is inflected for first-person

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

singular, the pronoun is dropped, and the direct object is omitted. The particle si˛e with the intransitive verb otwiera´c si˛e has been moved to the position after the word tak, and the verb is inflected for third singular agreeing with ‘it/box’. In summary, there is clear evidence children have the capacity to neutralize the semantic roles Undergoer and Actor for syntactic purposes. However, these observations demonstrate neutralization that is restricted to the intransitive context. Polish children produce passive participle verb forms in adjectival constructions (e.g. Bartosz (1;11) Zepsute, to zepsute jest ‘Broken, (it) is broken’), but they do not produce passive voice constructions (see Weist 1990: 1338–1341). Hence, child Polish has the properties of an [S,A] language (like Warlpiri) and not the fluent [S,A, d-S] language (like adult Polish). Linking Rules. Does child language contain grammatical relations, and if so, what kind? While there are variations on these themes, there have been two major approaches to answering this question. One approach begins with the investigation of child language data and derives a hypothesis about grammatical relations from what children produce and comprehend (see, for example, Brown 1973; Bowerman 1973). The second approach begins with a linguistic theory and makes predictions about what should be found in the child language data that correspond to that theory (see, for example, Pinker 1984; Radford 1990). Invariably, the investigators that begin with the child language data arrive at the conclusion that grammatical relations are learned, and those that begin with a linguistic theory (for example, some version of Chomsky’s thinking) assume at the outset that grammatical relations are innate. A revealing demonstration of these two alternatives concerns Pinker’s concept of innate linking rules, and Bowerman’s observations of language acquisition in her daughters Christy and Eva. According to Pinker (1984), children are innately equipped with linking rules that initiate the process of mapping thematic roles onto syntactic functions. A hierarchy of thematic roles and a hierarchy of syntactic functions were proposed to establish the linking process. The hierarchy for thematic roles contains agent, theme/patient, and location/goal/source, and the synchronized hierarchy for syntactic functions contains subject, object and oblique object, in that order. Direct links between thematic roles and syntactic functions (e.g. agent to subject and patient to object) are ‘canonical’ (i.e. the innate default), and the theory predicts that learning will be facilitated. The set of thematic relations is defined by verb-argument structure. For an activity predicate like to cry, the logical structure contains a single agent core argument, and the agent is canonically linked to subject. An achievement predicate like to fall has a single theme core argument, and since there is no agent in the predicate structure, the theme is canonically linked to subject. Non-canonical mapping occurs when the default links are ‘crossed’; for example, the predicate to have contains the core arguments of location and theme, where the location role becomes subject and not the higher-ranking role of theme. For verbs taking three arguments such as to give, the canonical

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

683

684

RICHARD M. WEIST

Table 18.12 The acquisition sequence for canonical and non-canonical constructions Two-argument verbs Child

Age

Canonical theme ¼ subject

Non-canonical location/goal/source ¼ subject

Christy Eva

1;9 1;8

Baby fall down (on) me Necklace stay (in) purse

Daddy have cake Ernie got spoon

Three-argument verbs Child

Age

Canonical: Agent – verb – theme – goal

Non-canonical: Agent – verb – goal – theme

Christy Eva

2;0 1;10

I gave a nipple Mommy Christy bring that for (¼to) me

I give Mommy a bottle Linda gave my (¼me) choochoo

mapping is as follows: agent to subject, theme to direct object, and beneficiary to oblique object. The acquisition of sentences with the theme linked to the oblique object should be delayed. Bowerman (1990) tested Pinker’s hypothesis with observations of her daughters Christy and Eva, and some critical comparisons are presented here in Table 18.12. When one observes the child language data, evidence for innate linking rules is absent: Pinker’s ‘non-canonical’ constructions emerge at the same time as his ‘canonical’ constructions.5 In addition to anomalies found in the acquisition of English (i.e. the child language data), there are other intractable predictions. Children learning Acehnese will be at a loss to determine what to do with their innate potential facing a target language void of syntactic relations, and children learning ergative languages, such as Dyirbal, will experience genetic interference as the target language links patient to subject and not agent (see Van Valin 1992 on ergative languages and implications for language acquisition).

18.7

Acquiring wh-questions

Rowland and Pine (2000) reviewed previous research on the acquisition of the subject–auxiliary errors in the acquisition of wh-questions. They pointed out that there is a phase when children produce un-inverted whquestions (e.g. What you did say?) and inverted yes/no questions (e.g. Does the kitty stand up?) at the same time as they are beginning to produce correctly inverted wh-questions (e.g. What did you doed?). The authors evaluate two movement-rule accounts of the acquisition phenomena which they judged to be lacking. Their alternative theory is based on the idea that children have ‘a distributional learning mechanism that learns and reproduces lexically-specific formulae’ (p. 157) and, more specifically, ‘lexically-specific wh-word þ auxiliary combinations’ (p. 157). Rowland and Pine (2000: 163)

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

view their acquisition theory as a variation on Braine’s (1976) theory of limited-scope formulae (see Section 18.2). According to the theory, children learn ‘wh-word þ auxiliary markers’ which are highly frequent in childdirected speech. In other words, there was no explanation for the underlying dynamics that might foster an un-inverted wh-question. Rowland and Pine analysed the corpus of Adam (see Brown (1973) with a focus on the ‘uninversion period’, i.e. mean length of utterance (MLU) 3.24–4.10), and they examined maternal utterances from a subset of transcripts prior to this period. As auxiliaries entered Adam’s corpus, he began to produce both inverted and un-inverted wh-questions, and during the ‘un-inverted period’, the relative number of un-inverted versus inverted questions increased and then decreased, terminating in a higher percentage of inverted wh-questions. The authors categorized every wh þ aux combination as inverted or un-inverted, and they counted the frequency of occurrence in the maternal speech sample, all well-formed. Some of Adam’s inverted combinations (e.g. what do, where did) were frequent in the input, but others were not found in the input (e.g. what was, which does). To make predictions worse, some of Adam’s un-inverted combinations (e.g. why did/didn’t/don’t) were, in fact, present in the input. No explanation was given for theoretical ‘misses’ (i.e. zero frequency inverted, what was, and relatively high frequency uninverted, why don’t). Van Valin (2002: 165) demonstrated that a general syntactic principle combined with an understanding of information processing can predict the acquisition of wh-question formation in Adam’s corpus. The principle (the IF principle) is as follows: ‘Illocutionary force in English is signaled by the position of the tense-bearing morpheme: a) In declarative utterances, tense appears core-internally, and b) In interrogative utterances, tense appears core-initially.’ In order to apply such a principle, the child needs to be able to process tense (see Tables 18.4 and 18.5). The auxiliaries is, does and has reveal their finite properties when they enter into the following contrasts: (1) am/are/is versus was/were, (2) do/does versus did, and (3) have/has versus had. The tensed properties of the modals can, could, shall, will, may and might are not as salient. This combination of the IF principle and morphological processing brought Van Valin to the following prediction: ‘In wh-questions, children will initially place only those auxiliaries which are explicitly tensed in core-initial position.’ Hence, we can predict which wh-questions will be inverted (i.e. is, are, was, do, does, did, have, has and had) and which ones will be un-inverted (i.e. can, could, may, might, shall, should and will). Furthermore, since the negative auxiliaries end in n’t, which does not signal tense, negative auxiliaries will not be inverted in the initial phase of wh-question acquisition. Van Valin’s predictions were based on the assumption that children process wide-scope operators and integrate that information into the clause structure. The predictions based on this assumption were supported by the data from Brown’s Adam provided by Rowland and Pine (2000). Van Valin (1998) penetrated further into the acquisition of wh-questions showing how such

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

685

686

RICHARD M. WEIST

issues as the emergence of subject and object wh-question and constraints on long-distance wh-questions can be understood within the RRG framework without the need to propose an autonomous, parameterized Language Acquisition Device or Universal Grammar.

18.8

Conclusion

In his paper linking RRG to language acquisition, Van Valin (1991: 9) presented two ‘perspectives’: (1) ‘Syntax is . . . relatively motivated by semantic, pragmatic, and cognitive concerns’ and (2) ‘the child actually learns language and CONSTRUCTS a grammar during the process of language acquisition’. Regarding how the construction process functions, Van Valin cited Slobin’s (1985) theory of Operating Principles. According to Slobin (1985: 1158), children are innately equipped with a ‘language-making capacity [LMC]’ and Slobin proposed that ‘LMC must begin life with some initial procedures for perceiving, storing, and analyzing linguistic experience, and for making use of capacities and accumulating knowledge for producing and interpreting utterances’. The theory of Operating Principles has been criticized by child language researchers, for example Bowerman (1985: 1281), and linguists, for example Pinker (1989: 463), since Slobin’s information processing theory lacks a model of grammar. This chapter was designed to investigate the potential that RRG might be able to fill this void. Semantic functions such as agent and theme/patient are operational in the earliest phase of language acquisition, when single-word utterances dominate the child’s corpus. The child’s capacity to process the finite morphology of the target language becomes apparent as multiplemorpheme utterances emerge, especially in highly inflected languages like Turkish and Polish. As soon as there is evidence for the acquisition of the elements of a tense-aspect-modality system, the semantic structure of the predicate influences the pattern of acquisition. Hence, on the one hand, children are utilizing their innate information processing capacity (which might be referred to as their LMC), and on the other hand, this capacity to process information is combined with at least some knowledge of predicate structure (see Tables 18.3 and 18.4). While these two components of linguistic knowledge are necessary to explain the acquisition data, they are not sufficient. Children integrate this knowledge within a concept of clause structure (see Table 18.1). Potential explanations of language acquisition based exclusively on principles of information processing (e.g. Slobin 1985), or with categories of predicate structure (e.g. Shirai 1998) will fall short if they fail to explain how integration into clause structure is accomplished. From the perspective of RRG, clause structure is created from the semantic principle of predicate-argument structure, and it contains levels of semantic complexity. The ‘operator-scope’ hypothesis presented here predicts that children identify the scope of operators as the concepts of

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

tense, aspect, and modality are being acquired. The predicate tracking data (e.g. Weist et al. 2004) showed that Polish children acquire tense (i.e. a clause operator) and aspect (i.e. a nuclear operator) during the same phase of acquisition. Similarly, Choi’s study of the emergence of evidential morphemes (i.e. clause operators) found early productivity of clause operators in the modal domain of the TAM system. These facts require that children are constructing their understanding of clause structure simultaneously with their decoding of the elements of the tense-aspect-modality system. Van Valin’s (2002) study of the acquisition of subject–auxiliary inversion revealed how the integration of the tense operator within clause structure explains the emergence of wh-questions in child language data. From the viewpoint of cognitive development, the child language data do not support the idea that the infant-toddler is somehow limited in their capacity to process complex relations and construct meaningful representations in the manner imagined by some early child language researchers such as Braine (1976) or more recent researchers such as Rowland and Pine (2000, 2003). In contrast, the child language data reviewed here are consistent with Spelke’s ‘core knowledge systems’ theory of cognitive development, where the core knowledge systems are thought of as ‘mechanisms for representing and reasoning about particular kinds of ecologically important entities and events’ (Spelke 2000: 1233). In retrospect, Slobin’s (1985) set of Operating Principles might be thought of as the core knowledge system for language.

References Aksu-Koç, Ayhan. 1988. The Acquisition of Aspect and Modality: The Case of Past Reference in Turkish. Cambridge: Cambridge University Press. Aksu-Koç, Ayhan. 1998. The role of input vs. universal predispositions in the emergence of the tense-aspect morphology: Evidence from Turkish. First Language 18: 255–280. Antinucci, Francesco and Ruth Miller. 1976. How children talk about what happened. Journal of Child Language 3: 167–190. Bauer, Patricia. J. 1996. What do infants recall of their lives? Memory for specific events by one- to two-year-olds. American Psychologist 51: 29–41. Bauer, Patricia. J. 2007. Remembering the Times of Our Lives: Memory in Infancy and Beyond. Mahwah, NJ: Lawrence Erlbaum. Bowerman, Melissa. 1973. Early Syntactic Development. Cambridge: Cambridge University Press. Bowerman, Melissa. 1981. Notes for the Nijmegen Workshop. MPI, Nijmegen, Netherlands. Bowerman, Melissa. 1985. What shapes children’s grammars? In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 2, 1257–1319. Hillsdale, NJ: Lawrence Erlbaum.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

687

688

RICHARD M. WEIST

Bowerman, Melissa. 1990. Mapping semantic roles onto syntactic functions: Are children helped by innate linking rules? Linguistics 28: 1253–1289. Bronckart, Jean-Paul and Hermina Sinclair. 1973. Time, tense, and aspect. Cognition 2: 107–130. Braine, Martin D. S. 1976. Children’s first word combinations. Monographs of the Society for Research in Child Development. Braine, Martin D. S. 1992. What sort of innate structure is needed to ‘bootstrap’ into syntax? Cognition 45: 77–100. Brown, Roger. 1973. A First Language. Cambridge, MA: Harvard University Press. Choi, Soonja. 1991. Early acquisition of epistemic meanings of Korean: A study of sentence-ending suffixes in the spontaneous speech of three children. First Language 11: 93–120. Choi, Soonja. 1995. The development of epistemic sentence-ending modal forms and functions in Korean children. In Joan Bybee and Suzanne Fleichman (eds.), Modality in Grammar and Discourse, 165–205. Amsterdam: John Benjamins. Chomsky, Noam. 1965. Aspects of a Theory of Syntax. Cambridge, MA: MIT Press. Clancy, Patricia. 1985. The acquisition of Japanese. In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 1: The data, 373–524. Hillsdale, NJ: Lawrence Erlbaum. Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Fillmore, Charles. 1968. The case for case. In Emmon Bach and Robert Harms, (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart and Winston. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Friedman, William J. 1978. Development of time concepts in children. In H. W. Ruse and Lewis P. Lipsitt (eds.), Advances in Child Development and Behavior, Vol. 12, 267–298. New York: Academic Press. Liang, Lu Yao, Dandan Wu and Hui Li. 2019. Chinese preschoolers’ acquisition of temporal adverbs indicating past, present, and future: a corpusbased study. Journal of Child Language 46: 760–784. Lieven, Elena V. M., Julian M. Pine and Gillian Baldwin. 1997. Lexically-based learning and early grammatical development. Journal of Child Language 24: 189–219. MacWhinney, Brian and Catherine Snow. 1985. The child language data system (CHILDES). Journal of Child Language 1: 271–296. Mandler, Jean M. 1990. A new perspective on cognitive development in infancy. American Scientist 78: 236–243. Meisel, Jürgen M. 1994. Getting FAT: Finiteness, agreement, and tense in early grammars. In Jürgen M. Meisel (ed.), Bilingual First Language Acquisition: French and German Development, 89–127. Amsterdam: John Benjamins. Newcombe, Nora and Janelle Huttenlocher. 2003. Making Space: The Development of Spatial Representation and Reasoning. Cambridge, MA: MIT Press.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

Palmer, Frank R. 2001. Mood and Modality. Cambridge: Cambridge University Press. Piaget, Jean. 1954. The Construction of Reality in the Child. New York: Basic Books. Piaget, Jean. 1971. The Child’s Conception of Time. New York: Ballantine Books. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge, MA: Harvard University Press. Pinker, Steven. 1989. Review of Slobin, Cross-linguistic evidence for language making capacity, and Bowerman, What shapes children’s grammar? Journal of Child Language 16: 456–463. Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Blackwell. Rovee-Collier, Carolyn K. 1997. Dissociations in infant memory: Rethinking the development of implicit and explicit memory. Psychological Review 104: 467–498. Rowland, Caroline F. and Julian M. Pine. 2000. Subject-auxiliary inversion errors and wh-question acquisition: ‘what children do know?’ Journal of Child Language 27: 157–182. Rowland, Caroline F. and Julian M. Pine. 2003. The development of inversion in wh-questions: A reply to Van Valin. Journal of Child Language 30: 197–212. Sano, Tetsuya and Nina Hyams. 1994. Agreement, finiteness, and the development of null agreements. Proceedings of NELS 24: 543–558. Shirai, Yasuhiro. 1993. Inherent aspect and the acquisition of tense-aspect morphology in Japanese. In Heizo Nakajima and Yukio Otsu (eds.), Argument Structure: Its Syntax and Acquisition, 185–211. Tokyo: Kaitakusha. Shirai, Yasuhiro. 1998. The emergence of tense-aspect morphology in Japanese: Universal predisposition. First Language 18: 281–310. Shirai, Yasuhiro and Roger W. Andersen. 1995. The acquisition of tense/ aspect morphology: A prototype account. Language 71: 743–762. Slobin, Dan I. 1985. Crosslinguistic evidence for the language-making capacity. In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 2, 1175–1256. Hillsdale, NJ: Lawrence Erlbaum. Slobin, Dan I. 2001. Form–function relations: How do children find out what they are? In Melissa Bowerman and Stephen C. Levinson (eds.), Language Acquisition and Conceptual Development, 406–449. Cambridge: Cambridge University Press. Smoczy´ nska, Magdalena. 1985. The acquisition of Polish. In Dan I. Slobin (ed.), The Cross-Linguistic Study of Language Acquisition, Vol. 1, 595–686. Hillsdale, NJ: Lawrence Erlbaum. Stephany, Ursula. 1986. Modality. In Paul Fletcher and Michael Garman (eds.), Language Acquisition, 375–400. Cambridge: Cambridge University Press. Stoll, Sabine. 1998. The role of lexical Aktionsart for the acquisition of Russian aspect. First Language 18: 351–378.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

689

690

RICHARD M. WEIST

Spelke, Elizabeth S. 1991. Physical knowledge in infancy: Reflections on Piaget’s theory. In Susan Carey and Rochel Gelman (eds.), The Epigenesis of Mind: Essays on Biology and Cognition, 133–170. Hillsdale, NJ: Lawrence Erlbaum. Spelke, Elizabeth S. 2000. Core knowledge. American Psychologist 55: 1233–1243. Toivainen, Jorma. 1980. Inflectional Affixes Used by Finnish-Speaking Children Aged 1–3 Years. Helsinki: Suomalaisen Kirjallisuuden Seura. Tomasello, Michael. 1992. First Verbs: A Case Study of Early Grammatical Development. Cambridge: Cambridge University Press. Tomasello, Michael. 2000. First steps toward a usage-based theory of language acquisition. Cognitive Linguistics 11: 61–82. Tomasello, Michael. 2003. Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge, MA: Harvard University Press. Van Valin, Robert D., Jr. 1991. Functionalist linguistic theory and language acquisition. First Language 11: 7–40. Van Valin, Robert D., Jr. 1992. An overview of ergative phenomena and their implications for language acquisition. In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 3, 15–38. Hillsdale, NJ: Lawrence Erlbaum. Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In Robert Van Valin (ed.), Advances in Role and Reference Grammar, 1–164. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the mechanisms of language acquisition. In Michael Tomasello (ed.), The New Psychology of Language: Cognitive and Functional Approaches to Language Structure, 221–249. Mahwah, NJ: Lawrence Erlbaum. Van Valin, Robert D., Jr. 2002. The development of subject–auxiliary inversion in English wh-questions: An alternative analysis. Journal of Child Language 29: 161–175. Van Valin, Robert D., Jr. 2005. The Syntax–Semantics–Pragmatics Interface: An Introduction to Role and Reference Grammar. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning, and Function. Cambridge: Cambridge University Press. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Weist, Richard M. 1986. Tense and aspect. In Paul Fletcher and Michael Garman (eds.), Language Acquisition, 356–374. Cambridge: Cambridge University Press. Weist, Richard M. 1990. Neutralization and the concept of subject in child Polish. Linguistics 11: 1332–1348. Weist, Richard, M. 2002. The first language acquisition of tense and aspect. In Rafael Salaberry and Yasuhiro Shirai (eds.), The L2 Acquisition of TenseAspect Morphology, 25–78. Amsterdam: John Benjamins.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

Functional Acquisition Processes in Child Language

Weist, Richard, M. 2009. One-to-one mapping of temporal and spatial relations. In J. Guo, E. Lieven, N. Budwig, S. Ervin-Tripp, K. Nakamura and S. Özçali¸skan (eds.), Crosslinguistic Approaches to the Psychology of Language: Research in the Tradition of Dan Isaac Slobin, 69–80. Hillsdale, NJ: Lawrence Erlbaum. Weist, Richard M. 2014a. Developing temporal systems. Psychology of Language and Communication 18(2): 126–142. Weist, Richard M. 2014b. Future temporal reference in child language. In Philippe De Brabanter, Mikhail Kissine and Saghie Sharifzadeh (eds.), Future Times: Future Tenses, 87–113. Oxford: Oxford University Press. Weist, Richard M. and Ewa Buczowska. 1987. The emergence of temporal adverbs in child Polish. First Language: 7, 217–229. Weist, Richard M., Aleksandra Pawlak and Jenell Carapella. 2004. Syntactic– semantic interface in the acquisition of verb morphology. Journal of Child Language 31: 31–60. Weist, Richard M., Aleksandra Pawlak and Karen Hoffman. 2009. Finiteness systems and lexical aspect in child Polish and English. Linguistics 47(6): 1321–1350. Weist, Richard M., Hanna Wysocka, Katarzyna Witkowska-Stadnik, Ewa Buczowska and Emilia Konieczna. 1984. The defective tense hypothesis: On the emergence of tense and aspect in child Polish. Journal of Child Language 11: 347–374.

Notes 1 Influenced by Braine’s (1976) ideas regarding positional learning, Lieven et al. (1997) found further evidence for the argument that children learning English base their early utterances on ‘low-scope, idiosyncratic formulae’ (p. 189). However, admittedly, the authors were unable to see how the knowledge of these formulae motivated the child’s innovative development into morphosyntactic information processing. Ironically, Braine (1992) also proposed an acquisition model having the fundamental concepts of RRG with the potential to understand further language development (see Van Valin and LaPolla 1997: 643). 2 Regarding cognitive complexity in development, Weist and Stephany held opposing views. Like Antinucci and Miller (1976) and others, Stephany accepted the Piagetian theory of cognitive development which imagined serious limitations in the child’s thought processes. In particular, the potential for conceptual time travel into the past and the future was theoretically unavailable, making tense in child language deficient, lacking deictic relations. Weist et al. (1984) and elsewhere (for example, Weist 2014b) made the opposite argument. 3 Liang et al. (2019) investigated the acquisition of temporal adverbs in Mandarin. The youngest group of children in their study was 2;6. While

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

691

692

RICHARD M. WEIST

their repertoire was limited, the 2-year-old children used temporal adverbs to locate events in the past, present and future. Regarding the layered structure of the clause (see 18.1), this constitutes evidence for core layer modification in a tenseless language. 4 Palmer (2001: 13) classifies desire and intention outside the deontic modal domain. Choi (1995: 179, Table 2) included a comparison of the early acquisition of sentence-ending suffixes with evidential meanings (e.g. -ta and -e) with the later acquisition of modal auxiliaries having deontic meanings (e.g. -ya tway (obligation) and -su isse (ability)). 5 Concerning the acquisition of tense and aspect, Weist (2002) reviewed mismatches between the acquisition predictions derived from the Principles and Parameter framework and the child language data, for example, the ‘principle of economy of derivation’, Sano and Hyams (1994: 551). In contrast, Weist (2009) showed how Slobin’s (1985, 2001) information processing theory predicts ‘typological bootstrapping’ within a comparison of the acquisition of temporal and spatial systems in Polish and Finnish children.

https://doi.org/10.1017/9781316418086.023 Published online by Cambridge University Press

19 Grammatical Aspects of Language Processing in the Brain A Role and Reference Grammar Perspective Robert D. Van Valin, Jr. Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: ∴ A aSTG BA CLM IF

therefore actor anterior superior temporal gyrus Brodmann’s area clause-linkage marker illocutionary force

INGR LH LS LTI MR NMR NP NUC

ingressive left hemisphere logical structure linking template inventory macrorole non-macrorole noun phrase nucleus

19.1

Introduction

PP prepositional phrase PPT past participle PrCS pre-core clot PrDP pre-detached position PRED predicator PSA privileged syntactic argument RH right hemisphere RP reference phrase RRG Role and Reference Grammar TNS tense U undergoer V verb VP verb phrase XP unspecified phrase type

As noted in Chapter 1, Role and Reference Grammar (RRG) strives to be a component of a model of the communicative competence of a native speaker of a human language, and following Kaplan and Bresnan (1982), it is incumbent upon theories making such a claim to be implementable in

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

694

ROBERT D. VAN VALIN, JR.

testable models, psycholinguistic or computational, of language processing.* Nolan (Chapter 21) illustrates various computational implementations of RRG. Van Valin (2006) attempted to make the RRG linking system compatible with the results of psycholinguistic investigations of sentence processing. With respect to production, the RRG linking system was found to fit well with a psycholinguistic model of the speaker. With respect to comprehension, the key idea was the precompiling of aspects of the RRG linking system in semantic and syntactic representations in order to facilitate it. This chapter will expand the discussion into the realm of cognitive neuroscience, specifically the organization of the language processing system in the brain, and the analysis to be presented builds on the 2006 proposal. The discussion begins with comprehension and is oriented around offering a solution to a long-standing mystery about language and the brain. It concerns language in the right hemisphere of split-brain patients. This phenomenon raises fundamental questions about the way language is represented and processed in the brain. The mystery is presented in Section 19.2, followed in Section 19.3 by an overview of ‘dual-stream’ models of language processing. These models lead to an important question concerning Brodmann’s area 22 in the left temporal lobe, the answer to which involves the proposals put forth in Van Valin (2006), which is the topic of Section 19.4. The following section presents the RRG account of the relevant right-hemisphere linguistic abilities of split-brain patients and its implications for the analysis of certain types of aphasia. The focus in Section 19.6 then turns to language production, and there is a brief discussion of the implications of the RRG processing model developed for comprehension for production. The main points are summarized in the final section.

19.2

Language in Split-Brain Patients

As a last-ditch effort to control serious epileptic seizures, in the 1940s surgeons began to sever the corpus callosum, the thick band of fibres which links the two hemispheres of the brain and is the conduit for information travelling between the hemispheres; this procedure is called ‘commissurotomy’. This isolated the hemispheres from each other, cutting off the flow of information between them. Starting in the 1960s, researchers began the detailed investigation of the cognitive and behavioural consequences of this surgery.1 One of the obvious areas of interest was language,

*

I would like to thank Delia Bentley, Evelina Fedorenko, David Kemmerer, Jean-Pierre Koenig and Anja Latrouite for helpful discussion and comments on earlier drafts. Some of this research was supported by the German Research Foundation (DFG) through Cooperative Research Center 991 ‘The structure of representations in language, cognition and science’ at the Heinrich Heine University Düsseldorf.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

since the neural structures underlying the grammatical aspects of it are known to be located in the left hemisphere, and the standard view was that the primary areas were Broca’s area in the left inferior frontal lobe (syntax) and Wernicke’s area in the left temporal lobe (semantics) (see Figure 19.1). It was expected that the isolated right hemisphere would not have any linguistic capabilities, and beyond single word recognition, none were found initially (Gazzaniga and Hillyard 1971). ‘In general, what was meant by “language capacity” in these patients was the ability to understand written or spoken words’(Gazzaniga 1983: 526). This ability to understand individual words showed that lexical storage was bilateral, with each hemisphere potentially having a lexicon of some kind (see also Zaidel 1983). It should be noted that not all split-brain patients showed this kind of right-hemisphere linguistic ability. In the 1980s, Gazzaniga and his colleagues began to test split-brain patients with the same tasks that were used in the study of what was then called ‘agrammatic aphasia’. The motivation for this is laid out in Baynes and Gazzaniga (1988: 120): Schwartz, Saffran, and Marin (1980) used a two-choice-picture-pointing paradigm to test the ability of agrammatic aphasics to understand semantically reversible active and passive sentences (e.g. the boy hit the girl; the boy was hit by the girl). Agrammatic aphasics were unable to choose correctly the picture that matched the sentence they heard. In other words, they could not use the syntactic constraints in the sentences to determine who hit whom. In contrast, Linebarger, Schwartz, and Saffran (1983) found that aphasics who were unable to interpret correct active and passive sentences were able to judge accurately whether sentences were grammatical. The Linebarger et al. result is important because it suggests that even though a patient may not be able to use a grammatical constraint in a comprehension task, he or she may still know something about grammar and be able to use this knowledge in other tasks, such as a grammaticality judgment task. Baynes and Gazzaniga (1988: 124) report that two split-brain patients show the same linguistic ability out of their right hemispheres that the ‘agrammatic’ aphasics in Linebarger et al. (1983) exhibited: ‘The most striking finding in these two commissurotomized patients with right hemisphere language is the dissociation between the [impaired-RVV] ability to comprehend syntactically constrained sentences and the [intact-RVV] ability to judge their grammaticality. This dissociation can be found in two distinct populations, agrammatic aphasics and commissurotomy patients.’ These results will be discussed in detail in Section 19.5, and an explanation for them will be proposed. How, then, is it possible for an isolated right hemisphere to make grammaticality judgements? If the rules, representations, constraints, etc., that

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

695

696

ROBERT D. VAN VALIN, JR.

constitute a speaker’s grammatical competence are located in the left hemisphere in right-handers, then it should not be possible for right-handed splitbrain patients to show any grammatical competence in their right hemisphere, yet some do. This is the primary phenomenon to be explained. The first step in answering this question is to review some of the contemporary models of language in the brain.

19.3

Models of Language Processing in Contemporary Cognitive Neuroscience

Until the early 1990s the view that the primary language areas of the brain are Broca’s and Wernicke’s areas, which are in the left frontal and temporal lobes, respectively, and which are connected by the arcuate fasciculus, was widely assumed to be basically correct. Broca’s area includes the pars opercularis and the pars triangularis in the inferior frontal gyrus, while Wernicke’s area involves the superior temporal gyrus and the angular gyrus. The locations of these two areas are represented in Figure 19.1. However, evidence that other parts of the brain were directly involved in language processing began to accumulate. Mazoyer et al. (1993) showed that when French speakers were asked to listen to well-formed French sentences, a list of words, and Urdu sentences, there was activation in the anterior temporal lobe only when French sentences were heard, not in any other condition. Of particular relevance for this discussion is the claim put forth in Dronkers et al. (1994) that the anterior part of Brodmann’s area [BA] 222 in the left temporal lobe is involved in syntactic processing based on a study of

Figure 19.1 Language-related areas of the left hemisphere Basic drawing of the labelled brain from: A. Blackburn and C. Hwozdek ‘Labeled Brain’ 2016, openclipart.org, licensed under CCO 1.0: creativecommons.org/publicdomain/zero/1.0/legalcode.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

a large number of aphasics of various types. This was an extremely controversial claim back then, in part because it was incompatible with the traditional model. By the time the full study was published (Dronkers et al. 2004), it was no longer so controversial, as other studies (e.g. Stromswold et al. (1996), Stowe et al. (1998), Friederici et al. (2000), Friederici et al. (2003)) found evidence that the anterior part of the left temporal lobe is involved in morphosyntactic processing. Recognition of the role of the anterior temporal lobe in syntactic comprehension is one of the factors that led to the development of what are called ‘dual-stream models’ of language processing. There are several (e.g. Hickok and Poeppel (2004, 2007), Saur et al. (2008), Friederici (2009, 2012), Bornkessel-Schlesewsky and Schlesewsky (2013) and Blank et al. (2016)), and they are by no means in agreement on all points. But they all agree in positing dorsal and ventral processing streams. The dorsal stream involves the same brain regions as in the traditional model, while the ventral stream goes through anterior BA 22 to connect with Broca’s area (BA 44, 45). In these models, ‘a ventral stream processes speech signals for comprehension, and a dorsal stream maps acoustic speech signals to frontal lobe articulatory networks’ (Hickok and Poeppel 2007: 393). The differences among the models will not be explored here; what is crucial for this discussion is that all of the models mentioned above include the anterior temporal lobe in the ventral stream.3 Note that BA 22 is not just adjacent to the primary auditory cortex (BA 41, 42) but surrounds it, a convenient location for a brain region crucially involved in the comprehension process. What is the exact role that anterior BA 22 plays in the comprehension process? The evidence from Dronkers et al. (1994, 2004) and the studies cited earlier, along with Turken and Dronkers (2011) and a more recent one involving Icelandic aphasics (Magnusdottir et al. 2013) and magnetoencephalography (MEG) studies of language comprehension (Herrmann et al. 2011; Brennan et al. 2012), is that anterior BA 22 plays a role in the early phase of morphosyntactic processing which is typically referred to as ‘structure building’. Friederici summarizes this as follows: Based on neurophysiological data, these phrase structure building processes have been localized in the anterior superior temporal cortex approximately 120–150 ms after word category information is available as an early automatic syntactic process (Herrmann et al. 2011, Friederici et al. 2000, Shtyrov et al. 2003). The involvement of the aSTG [anterior superior temporal gyrus] during syntactic phrase structure building has been confirmed by functional MRI (fMRI) studies using syntactic violation paradigms (Friederici et al. 2003), as well as a natural language listening paradigm (Brennan et al. 2012). It has been proposed that, in the adult brain, these processes can be fast because templates of different phrase structures (e.g. determiner phrase, prepositional

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

697

698

ROBERT D. VAN VALIN, JR.

phrase) represented in the aSTG/STS are available automatically once the phrasal head (e.g. determiner, preposition) is encountered (Bornkessel and Schlesewsky 2006). (Friederici 2012: 263) The notion of ‘structure building’ can be characterized in a number of ways. One way involves rules of the kind proposed in generative grammar (e.g. Merge) whereby syntactic structure is constructed. Another way is through the combination of stored pieces or ‘chunks’ of structure, along the lines of Tree Adjoining Grammar or constructional approaches.4 It will be argued in Section 19.4 that RRG makes a very specific claim about what structure is and how it comes into play in the comprehension process. Thus, anterior BA 22 seems to be involved in what Friederici terms ‘syntactic structure building’, in a sense to be determined, during the very early phase of language comprehension and therefore plays a central role in the process. It will turn out that this ‘structure building’ is not purely syntactic. What is the function of posterior BA 22, which, along with part of the angular gyrus, constitutes what is traditionally known as Wernicke’s area? It is associated with semantic processing and the comprehension of meaning, both lexical and propositional. ‘Semantic processes at the sentential level are more difficult to localize. They seem to involve the anterior temporal lobe, as well as the posterior temporal cortex and angular gyrus (Lau et al. 2008, Obleser et al. 2007)’ (Friederici 2012: 263). The interpretation of sentences involves both anterior and posterior BA 22, and this means that anterior BA 22 is not restricted to syntactic ‘structure building’ but rather is central to the comprehension process in multiple ways. Since the anterior temporal lobe appears to be involved in processing syntactic information and semantic information at least at the sentential level, the function of this neuroanatomical region has been discussed as reflecting general combinatorial processes which are involved in phrase structure building as well as in semantic combinatorics (Hickok and Poeppel 2007, Saur et al. 2008). (Friederici 2012: 263) BA 22 is multifunctional: the anterior part is claimed to ‘build’ syntax and together with the posterior part is involved in the interpretation of sentences, and this is all carried out by cortex with the same cytoarchitectonic properties. This is a profound fact: the same cortical structure is central to the processing of both syntax and semantics,5 and this suggests strongly that syntax and semantics are neither as distinct as many theorists assume, nor are they handled by completely separate neural systems. Indeed, Federenko et al. (2020: 19) argue: ‘Taking all the available data into consideration, it therefore seems that a cognitive architecture whereby syntactic processing is not separable from the processing of individual word meanings is most likely’ [emphasis in original].

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

This leads to a fundamental question: What does BA 22 process, such that it appears to have syntactic properties which the anterior part handles and semantic properties which both parts handle? The answer to this question requires recourse to the proposals put forth in Van Valin (2006), and it leads to the answer to the main question as well.

19.4

RRG as a Processing Model

The RRG linking system maps a semantic representation into a syntactic representation and vice versa, and this is motivated by the fact that speakers go from meaning to form to utterance as part of speech production, while listeners go from utterance to form to meaning as part of the comprehension process. But RRG is not a processing model; it is a model of grammar which has facets that are incompatible with the results of psycholinguistic investigations of sentence processing. For example, the syntax-to-semantics linking algorithm assumes that the parser outputs a complete labelled tree diagram, which the linking algorithm then maps into the semantic representation. However, all of the results from studies of sentence processing show that interpretation is incremental, with comprehenders starting the interpretive process at the very first word encountered (Schlesewsky and Bornkessel 2004). Moreover, it assumes that displaced wh-expressions are linked last after all of the other arguments have been linked, whereas studies show that speakers try to resolve the long-distance dependency as soon as possible (Stowe 1985; Clifton and Frazier 1989; Boland et al. 1995; Traxler and Pickering 1996; Koenig et al. 2003). These abstractions, while appropriate for a model of grammar, must be abandoned in a processing model. There is a further factor to be taken into consideration, namely, the speed of processing. In the previous section it was noted that syntactic structure building starts as little as 120 ms after word recognition, and the subjective impression that speakers have is that interpretation is normally instantaneous. In other words, it seems to language users that they access the meaning directly. Because interpretation occurs simultaneously with parsing, the RRG linking system must be integrated into the parser. One way this could be accomplished is suggested by the approach to sentence comprehension proposed in Townsend and Bever (2001). They postulate a two-phase comprehension process: the first is termed ‘pseudosyntax’, which is statistical and yields an initial assignment of syntactic structure and thematic relations, and the second, which they call ‘real syntax’, is a minimalist derivation to check the results of the first phase. They adduce a large amount of convincing evidence for the initial phase (2001: 188–209), and the focus here will be on ‘pseudosyntax’, as it combines parsing and determining the initial interpretation of a sentence. They characterize it as follows.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

699

700

ROBERT D. VAN VALIN, JR.

Pseudosyntax consists of the immediate initial processes that isolate major phrases, differentiate lexical categories, and assign initial thematic relations. Pseudosyntax involves recognition of function morphemes and lexical categories, which segregate and distinguish phrases and verbs. Assignment of words to syntactic categories and major phrases coincides with the application of frequent sentence patterns that assign these phrases to thematic roles. The sentence patterns that are appropriate for a particular sentence depend on subcategorization properties of verbs. (Townsend and Bever 2001: 187) The ‘frequent sentence patterns’ referred to above are labelled ‘canonical sentence templates’, with the most frequent one being ‘NVN actor action patient’. They marshal a substantial amount of experimental evidence in support of the role of canonical sentence templates in ‘pseudosyntax’ (2001: Ch 7). Hence, ‘pseudosyntax’ is in essence statistically driven templatic parsing, in which the templates contain information about the thematic relations of the XPs. It results in what is called ‘good enough’ comprehension (Ferreira and Patson 2007). RRG makes use of syntactic templates, which were introduced in Chapter 1. The core templates are for the most part bare pieces of the layered structure of the clause which are unspecified with respect to the types of phrases that fill the slots in the template. Consequently, they also have no information regarding the semantic function of the possible phrases that can occur in them. All of that information is supplied by the linking algorithm. The first step in transforming bare syntactic templates into ‘pseudosyntax’ parsing templates (or, more accurately, ‘linking templates’) is to augment them with phrasal category information, and accordingly Townsend and Bever’s default canonical sentence template for English would be RP-NUC-RP. The enhanced RRG templates reflect the fact that, in RRG, syntax is exocentric and the major syntactic categories are not projections of lexical categories. So instead of NP, a projection of the lexical category noun, there is RP, reference phrase, the category of referring expressions which may or may not be headed by a noun, and instead of VP there is NUC, nucleus, the unit of the layered structure of the clause which houses the predicating expression, which may or may not be verbal in nature. Thus lexical category information, which seems so important for the processing of Indo-European languages, is largely irrelevant for languages in other parts of the world (e.g. the Philippines, North America (indigenous languages)). What is relevant to parsing and interpretation in all languages is the function of the word as an argument, predicating element, or modifier. The other crucial aspect of canonical sentence templates is the assignment of thematic relations to phrases in the template. Here again, RRG takes a different tack from Townsend and Bever. Rather than augmenting the

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

templates with thematic relations, the RRG templates would have the semantic macroroles actor (A) and undergoer (U). Hence the default canonical sentence template for English would be RP:A-NUC-RP:U. The advantage of using macroroles can be seen in (1). (1)

a. a0 . b. b0 . c. c0 .

Max broke Mary’s chair. Mary’s chair was broken by Max. Max likes Mary. Mary is liked by Max Max received Mary’s letter. Mary’s letter was received by Max.

Agent-Patient Experiencer-Stimulus Recipient-Theme

If the templates were enhanced with thematic relations, each of these sentences would require a separate template in both the active and passive voices, for a total of six, because the thematic relations are different in each pair of arguments. In terms of macroroles, on the other hand, only one template is needed in each voice, because in both active and passive versions Max is the actor and Mary’s chair/Mary/Mary’s letter is the undergoer. Using macroroles rather than thematic relations reduces the number of canonical sentence templates needed. The most common canonical sentence template for English, RP:A-NUC-RP:U, is given in Figure 19.2. This template has the core argument RPs annotated with semantic macroroles. In addition, the operator projection (lower part) carries empty representations for tense and illocutionary force, the obligatory operators in a finite

Figure 19.2 Simple English transitive linking template

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

701

702

ROBERT D. VAN VALIN, JR.

Figure 19.3 Intransitive template (verbal)

Figure 19.4 Intransitive template (non-verbal)

(main) clause in English; the exact values of these operators are a function of the clause being processed. There is also an optional place for peripheral adjuncts; only the core-level periphery is shown here, but there may be peripheries modifying the nucleus, core or clause. In Figures 19.3 and 19.4, the two basic intransitive templates are given. The first is for verbal nuclei, and the https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

single RP core argument is not annotated, because it can be either actor or undergoer, for instance Bill [actor] laughed (at the joke) vs. The gravy [undergoer] spilled (on the tablecloth). The second is for non-verbal nuclei, such as Sally is a lawyer [NP]/very tall [AdjP]/in the library [PP], all of which require the auxiliary be.6 The next two templates are for three-place predicates in English. The first (Figure 19.5) is for sentences with two direct core arguments (the macrorole arguments) and an oblique core argument or argument-adjunct. (2)

a. a0 . b. c.

Mary presented the first-place trophy to Sam. Mary presented Sam with the first-place trophy. Sam put the first-place trophy in a box /on the shelf. Sam gave the box to his assistant.

Oblique Oblique Oblique Oblique

core argument core argument argument-adjunct core argument

The second (Figure 19.6) is for the so-called ‘double-object’ construction; in RRG terms there are three direct core arguments, an actor, an undergoer and a non-macrorole argument (NMR). When both macrorole arguments are present in an active-voice English core, the actor is the coreinitial RP and the undergoer is the immediately post-nuclear RP. Accordingly, in Sam gave his assistant the box, for example, his assistant is the undergoer, not the box. It should be noted that there are templates for RPs and PPs (see Chapter 1). The final templates to be introduced in this initial discussion are for passive-voice constructions, in Figures 19.7 and 19.8. The template in Figure 19.7 handles transitive verbs, while the one in Figure 19.8 is for three-argument verbs which have an oblique third core argument (e.g. the examples in (2)). Passives of the ‘double-object’ construction would require a template like that in Figure 19.8 except for the non-macrorole core argument being an RP rather than a PP. The actor of the passive verb may optionally occur in a PP headed by by in the core-level periphery, and other adjuncts (e.g. temporal or locative expressions) may also appear in the peripheries. These templates are the first part of the solution to the problems outlined at the beginning of the section; they identify the phrases and their functions and assign semantic macroroles to the two primary arguments in a transitive or ditransitive clause. As should be clear, from an RRG perspective, there is nothing ‘pseudo’ about pseudosyntax. The second part is arriving at the correct interpretation of the sentence, and for this we have to take a look at the lexicon and the nature of lexical entries. In the RRG theory of grammar, it is claimed that all of the morphosyntactic properties of a completely regular verb can be derived from its semantic representation together with the linking algorithm (see e.g. Van Valin and LaPolla 1997: 154ff; Van Valin 2005: 66). However, this involves applying rules, principles and constraints, all of which would slow down a processing system. Hence in a processing model it is better to precompile the relevant information in the lexical entries for verbs, so that when the verb is recognized, all of this information becomes available immediately, as proposed in Van Valin (2006).7

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

703

704

ROBERT D. VAN VALIN, JR.

Figure 19.5 Template for three-place predicates

Figure 19.6 Template for ‘double-object’ construction

The two things that are most important for the rapid interpretation of a template is the assignment of semantic macroroles and non-predicative prepositions. For a simple transitive verb like kill or smash, the logical structures [LSs] would be as in (3).

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Figure 19.7 Passive template for transitive verbs

Figure 19.8 Passive template for three-place predicates

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

705

706

ROBERT D. VAN VALIN, JR.

(3)

a. General: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR pred0 (U: y)] b. kill: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR dead0 (U: y)] c. smash: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR smashed0 (U: y)]

Once the template in Figure 19.2 has been matched and the verb recognized, the basic ‘who did what to whom’ meaning is available by integrating the arguments from the template with the LS. This can be illustrated with The burglar smashed the window. 1. The matching mechanism selects the template in Figure 19.2, yielding [CORE [RP A: the burglar] [NUC smashed ][RP U: the window]] (simplified). 2. The lexical entry for the predicate in the nucleus is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. RP: A the burglar ¼ A: x, ∴ the burglar ¼ x, etc.),8 yielding the interpretation [do0 (A: the burglar, [affect0 (. . .)])] CAUSE [INGR smashed0 (U: the window)]. The assignment of syntactic structure and the semantic interpretation are accomplished rapidly thanks to the annotations shared by the templates and the lexical representations. This is accomplished in BA 22 using both anterior and posterior resources (the ‘sentential combinatorics’ mentioned in the previous section) as well as other regions in the left temporal lobe. What was called ‘early structure building’ in the previous section is, on this view, assembling syntactic templates, that is, selecting the RP templates for the RP slots in the clausal template, and matching the resulting template with the utterance being processed. Things are a bit more interesting when there are more than two arguments. A good example of a three-place verb is the transfer predicate present; its LS would be [do0 (x, [affect0 (x, z)])] CAUSE [INGR have0 (y, z)]. Present, like most three-place verbs in English, permits more than one possible choice of the undergoer macrorole, as shown in (2a, a0 ), and this choice is signalled by the preposition assigned to the non-macrorole argument. Rather than invoking macrorole assignment rules and preposition assignment rules as in an RRG grammar, the lexical entry for present specifies both possibilities, as in (4). (4)

present: a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)] b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: with z)]

Once the template in Figure 19.5 has been matched and the verb recognized, the basic ‘who did what to whom’ meaning is available by integrating the arguments from the template with the LS. For verbs like give that occur in the ‘double-object’ construction (e.g. Sam gave his assistant the box), their LSs would be as in (5). (5)

give:

a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)] b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: z)]

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

The LS in (5a) integrates with the template in Figure 19.5, while the one in (5b) integrates with the template in Figure 19.6. The comprehension of a passive sentence such as The window was smashed by the burglar is just as straightforward. 1. The matching mechanism selects the template in Figure 19.7, yielding [CORE [RP U: the window][NUC was smashed ]][PERIPHERY [PP A: by the burglar]] (simplified). 2. The lexical entry for the predicate in the nucleus is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. PP: A: by the burglar ¼ A: x, ∴ the burglar ¼ x, etc.), yielding the interpretation [do0 (A: the burglar, [affect0 (. . .)])] CAUSE [INGR smashed0 (U: the window)] In the case of three-place predicates (e.g. present), it is the coding of the NMR which is the key to determining which LS is the correct one. (6)

a. The first-place trophy was presented to Sam by Mary. 1. The matching mechanism selects the template in Figure 19.8 yielding [CORE [RP U: the trophy ][NUC was presented ] [PP NMR: to Sam]][PERIPHERY [PP A: by Mary]]. 2. The lexical entry for the predicate in the nucleus in (4a) is selected because of the NMR: to specification and is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. PP: A: by Mary ¼ A: x, ∴ Mary ¼ x, etc.), yielding the interpretation [do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: the trophy)]. b. Sam was presented with the first-place trophy by Mary. 1. The matching mechanism selects the template in Figure 19.8, yielding [CORE [RP U: Sam ][NUC was presented ] [PP NMR: with the trophy ]][PERIPHERY [PP A: by Mary]]. 2. The lexical entry for the predicate in the nucleus in (4b) is selected because of the NMR: with specification and is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. PP: NMR: with the trophy ¼ NMR: z, ∴ the trophy ¼ z, etc.), yielding the interpretation [do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (U: Sam, NMR: the trophy)]

These examples illustrate how the combination of semantic-macroroleaugmented syntactic templates and annotated lexical entries for verbs can rapidly yield the basic interpretation of a sentence (i.e. ‘who did what to whom’). It has been argued (McKoon and Love 2011) that the internal semantic complexity of a verb affects the speed at which it is processed, and this is readily captured in the LS representations. There were two issues mentioned at the beginning of this section that RRG as a processing model must address, namely incremental interpretation and displaced wh-questions. Incremental interpretation can be accounted for in

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

707

708

ROBERT D. VAN VALIN, JR.

Figure 19.9 English subject wh-Q template

terms of parallel processing of statistically weighted templates in conjunction with a beam-search algorithm of the type presented in Jurafsky (1996), which ranks candidate structures and lexical items within a specific range of probability, dropping candidates that fall outside that range as the process moves forward. At the first word in a sentence there are many possible candidate templates, and as each new word is encountered, the number of candidates is reduced until the correct combination of templates is chosen. Displaced wh-questions are handled by the templates in Figures 19.9 and 19.10. A few remarks are in order about these two templates. First, optional peripheries have been omitted. Second, there are no traces or other kinds of phonologically null elements in the trees. Third, the subject wh-Q template is compatible with transitive verbs and three-argument verbs with an oblique third core argument, while the non-subject wh-Q template works for verbs with two or more arguments. Intransitive verbs require whQ versions of the templates in Figures 19.3 and 19.4. Fourth, when the verb is transitive, these templates are active voice; passive voice requires distinct wh-Q templates. Fifth, these apply only to simple sentences; more complex templates would be required for long-distance wh-questions (see Figure 19.11)). An example of a simple wh-question and its analysis is given in (7). (‘~A’ means ‘not the A argument of the active voice transitive or threeplace verb in the main clause’.) (7)

What did Mary give to Sam? 1. The matching mechanism selects the template in Figure 19.10, yielding [PrCS [RP ~A: what ]][CORE {did} [RP A: Mary] [NUC give ] [PP NMR: to Sam]].9

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Figure 19.10 English non-subject wh-Q template

Figure 19.11 Long-distance wh-question linking template

2. The lexical entry for the predicate in the nucleus in (5a) is selected because of the NMR: to specification and is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. RP: A: Mary ¼ A: x, ∴ Mary ¼ x, etc.), yielding the interpretation [do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: what)].

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

709

710

ROBERT D. VAN VALIN, JR.

The ~A wh-expression what is integrated into the LS as the U, since undergoer is a kind of non-actor. An example of a long-distance question template is presented in Figure 19.11. The application of this template to a wh-question is presented in (8).10 (8)

What did Mary say that Sally gave to Sam? 1. The matching mechanism selects the template in Figure 19.11, yielding [Clause [PrCS [RP ~A: what ]][CORE {did} [RP A: Mary] [NUC say ]] [CLM that ] ! [CLAUSE: U [CORE [RP A: Sally] [NUC gave ] [PP NMR: to Sam]]]]. 2. The lexical entries for the predicates in the main clause nucleus, say, and in the subordinate clause nucleus, give, are activated, and the LS form give in (5a) is selected because of the NMR: to specification, and the annotations on the arguments in the template and the LSs guide the integration, yielding the interpretation do0 (A: Mary, [say0 (Mary, U: [[do0 (A: Sally, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: what)]].

This template works for a variety of verbs but not all (see Shimojo, Chapter 16): What did Mary think/believe/expect/anticipate/claim/deny/doubt/assert that Sally gave to Sam? These templates directly address the interpretation of the wh-expression and do not put it off until the rest of the sentence has been processed, as the linking algorithm does. The template in Figure 19.11 introduces templates for complex sentences, and in this case the structure involves subordination (embedding). There are non-embedded complex sentences as well, and sentences like Mary persuaded Sam to wash the dishes are represented by templates like the one in Figure 19.12.

Figure 19.12 Infinitival complement template

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Sentences with this structure (e.g. Mary persuaded/promised Sam to wash the dishes) obligatorily ‘share’ an argument; that is, the infinitival complement is missing an argument, and accordingly one of the core arguments in the first core is interpreted as also filling a role in the second core. In this sentence the actor of wash is missing; when the verb in the first core is persuade, the undergoer Sam is interpreted as the actor of wash, whereas when the verb is promise, the actor Mary is so interpreted. The LS for the persuade version would be [do0 (Mary, [affect0 (Mary, Sam)])] CAUSE [want0 (Sam, [do0 (Sam, [wash0 (Sam, dishes)])])]. Despite multiple occurrences in the LS, the RP Sam appears only once in the sentence as the undergoer of persuade. These last two examples barely scratch the surface of the intricacies of complex sentences, but they illustrate how the system can be applied to them. What happens when breakdowns such as garden paths occur? What happens when an unfamiliar structure is encountered? These are situations in which step-by-step reanalysis and reinterpretation is required. On top of the template-based processing discussed earlier, there is also the possibility of step-by-step processing as well. In this mode, the syntactic representation of the sentence is constructed, and then the steps of the syntax-to-semantics linking algorithm are carried out literally, which permits double-checking of all relevant aspects of the construction. All of the examples of linking with semantically augmented syntactic templates have been from English, which has a relatively rigid word order that is crucial for the interpretation of the core arguments. Can this approach to language comprehension be applied to languages with various degrees of word-order flexibility and which rely on case marking rather than linear order for the interpretation of core arguments? The answer is ‘yes’, and the application to case-marking languages with very flexible word order will be illustrated with an example from Croatian. (9)

darovati ‘give as a gift’ [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: y, U: z)] a. Unuc-i su bak-i darova-l-i cvije´ c-e. grandson-m.pl.nom be.3pl grandmother-f.sg.dat give-pst-pl flower-m.pl.acc ‘The grandsons [x] gave flowers [z] to [their] grandmother [y].’ b. Baki su unci darovali cvije´ ce. c. Cvije´ ce su baki unci darovali. d. Baki su darovali cvije´ ce unci.

The Croatian verb darovati ‘give as a gift’ is a three-argument transfer verb like English give or present. The LS in (9) has the same semantic annotations as in (4a) and (5a), and it represents the meaning of all of the sentences in (9a–d) and the many other possible arrangements of these five words. The only thing that is invariable in terms of word order is the occurrence of the word su as the second word; it is a kind of auxiliary verb which shows person and number agreement with the nominative RP and must always be the second word in the sentence.11

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

711

712

ROBERT D. VAN VALIN, JR.

Interestingly, darovati shows the same undergoer alternation that present does. (10)

a. Unuc-i su bak-u darova-l-i cvije´ c-em. grandson-m.pl.nom be.3pl grandmother-f.sg.acc give-pst-pl flower-m.pl.ins ‘The grandsons [x] gifted [their] grandmother [y] with flowers [z].’ b. Baku su unci darovali cvije´ cem. c. Cvije´ cem su baku unci darovali. d. Baku su darovali cvije´ cem unci.

As in (2a0 ) and the LS in (4b), but not in (9), the y-argument in the LS is the undergoer, and the z-argument is the non-macrorole core argument in the instrumental case, which is analogous to the instrumental preposition with in English. As in (9), all of the examples in (10) have the same meaning. The crucial difference between (9) and (10) is the coding of the non-macrorole argument, dative in (9) and instrumental in (10), and accordingly it must be specified in the LSs for darovati, as in (11), just like in (4) for English present. (11)

darovati ‘give as a gift’ a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: dat y, U: z)] ¼ (9) b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: INST z)] ¼ (10)

It is clear from the examples in (9) and (10) that linear order plays no role in the interpretation of these sentences and that the interpretation of the core arguments is based on case marking. In Van Valin (2006) ‘case association principles’ were used to handle the linking between case forms and macroroles. The grammar of Croatian includes the ones in (12) as well as additional ones for cases which are not found in (9) and (10). (12)

Croatian case association principles a. NOM [Actor > Undergoer] b. ACC [Undergoer] c. DAT [NMR] d. INST [NMR]

The nominative case codes the actor with some intransitive verbs (those with an activity predicate in their LS) and active-voice transitive and ditransitive verbs, and it codes the undergoer with some intransitive predicates (those without an activity predicate in their LS) and with passive-voice transitive and ditransitive verbs. The accusative case marks the undergoer of an active-voice transitive or ditransitive verb. The other two cases mark non-macrorole arguments, dative arguments being direct core arguments, while instrumental ones are oblique core arguments. In RRG, whether an XP is a core argument or a peripheral adjunct is a consequence of the semantic function of the XP and not on its place in the linear string. (Languages may impose linearization restrictions, of course.) This means that the syntactic templates which represent the layered structure of the clause need not have a fixed order, and this is what languages like Croatian require: unordered templates. The only ordering constraint in

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Figure 19.13 Croatian linking template for (9)

the clause concerns the auxiliary, which must be the second word in all clauses. This can be represented as in Figures 19.13 and 19.14. The dotted lines connecting the RPs and the nucleus to the core node indicate that the order of these elements is not fixed but that any of them could occupy any of the positions. The solid line connecting the AUX node to the core node signals, on the other hand, that its position in the core is fixed and it must invariably be in the second position. Information structural considerations like the relative topicality vs. focality of the referents of the RPs strongly influences the order of words in context, as is well known. Croatian not only allows the RPs and nucleus to occur in any order, but it also permits discontinuous constituents, as in (13a) vs. (13b). (13)

a. Naš-a uˇ cionic-a je udobna. our-nom classroom-nom be.3sg comfortable ‘Our classroom is comfortable.’ b. Naš-a je uˇ cionic-a udobna. our-nom be.3sg classroom-nom comfortable ‘Our classroom is comfortable.’ [literally ‘Our is classroom comfortable.’]

Je ‘is’ is like su ‘are’ in (9) and (10), as it must be second in the clause, and second position may be defined as being after the first constituent, as in (13a), or after the first word, as in (13b). (Native speakers prefer (13b).) Determiners are connected to the NUCLEUSR via the operator projection of

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

713

714

ROBERT D. VAN VALIN, JR.

Figure 19.14 Croatian linking template for (10)

the RP, and Croatian, unlike English, does not require strict adjacency among RP constituents, nor is their order fixed. Hence in Croatian linking templates like those in Figures 19.13 and 19.14, not only are the constituents in the clause unordered, but there are few constraints on the RP templates as well.12 These templates apply, accordingly, to all of the word-order variants in (9) and (10), as well as others not exemplified, and other threeargument verbs with a dative or instrumental third argument. The matching process goes as follows for (9c), for example. 1. The matching mechanism selects the template in Figure 19.13, yielding [CORE [RP ACC:U: Cvije´ce] [AUX su ] [RP dat:NMR: baki][RP nom:A: unci] [NUC darovali ]] 2. The lexical entry for the predicate in the nucleus in (11a) is selected because of the NMR: DAT specification and is activated, and the annotations on the arguments in the template and the LS guide the integration (i.e. RP:NOM:A: unci ¼ A:x, ∴ unci ¼ x, etc.), yielding the interpretation [do0 (A: unci, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: baki, U: cvije´ce)] To sum up, the determination of the syntactic structure and the semantic interpretation of an utterance is achieved rapidly due to the annotations on both the templates and the lexical representations of predicates. This is carried out in BA 22 using both anterior and posterior resources, the ‘sentential combinatorics’ mentioned in the previous section, as well as other areas of the left temporal lobe. What was called ‘early structure building’ in

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

the previous section is, on this view, the assembling of syntactic templates, for instance selecting the RP templates for the RP slots in the clausal template, and the matching of the resulting template with the utterance being processed. The integration of the information in the lexical representation of the predicate in the nucleus of the clause yields a basic ‘who did what to whom’ meaning (a ‘good enough’ interpretation). In the case of breakdowns, the system can execute the syntax-to-semantics linking algorithm in a controlled, step-by-step fashion. At the end of Section 19.3, the following question was raised: What does BA 22 process, such that it appears to have syntactic properties which the anterior part handles and semantic properties which both parts handle? The answer should now be clear: semantic-macrorole-augmented syntactic linking templates and the lexical entries for predicates. Given the integration of both syntactic and semantic information in the linking templates and in the lexical entries for predicates, it is in fact difficult to claim that anterior BA 22 is primarily for syntactic processing and posterior BA 22 is for semantic processing. Rather, all of BA 22 is involved both aspects of language comprehension, consonant with the point made by Federenko et al. (2020) mentioned earlier. They are also crucial for the resolution of the mystery stated at the end of Section 19.2: How is it possible for the isolated right hemisphere in a split-brain patient to make grammaticality judgements?

19.5

Language in Split-Brain Patients Revisited

The linguistic capacity of the right hemisphere in split-brain patients is not uniform; as pointed out in Section 19.2, some patients show no linguistic ability at all, some are able to recognize single words, and a small minority are capable of making grammaticality judgements. The latter two groups are both relevant for this discussion, not just the third. The fact that some patients can understand individual words is evidence, following Gazzaniga (1983) and Zaidel (1983), that storage of lexical items in the brain is bilateral in some people. RRG, like many (but not all) theories, postulates that the grammar of a human language has a lexicon containing (minimally) the words and morphemes of the language, and the lexical entries for verbs would contain the LS(s) for the verb (e.g. (3b, c), (4a, b), (11a, b)), including the precompiled information on macrorole assignment and the preposition or case of non-macrorole arguments. What the research on commissurotomized patients has shown is that in some people the lexicon is in both hemispheres, not just the left. RRG grammars also have a repository for the syntactic templates that play a role in the linking from semantics to syntax; it is called the ‘syntactic inventory’. These are the same syntactic templates that are augmented with morphosyntactic and semantic macrorole information to create the linking templates at the heart of the RRG processing

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

715

716

ROBERT D. VAN VALIN, JR.

system developed in the previous section. These linking templates must be stored somewhere, and so it is appropriate to posit the existence of a ‘linking template inventory’ (LTI). The lexicon and the LTI house a language’s lexical and core grammatical resources, respectively, and given that the lexicon is bilaterally instantiated in some speakers,13 it is reasonable to assume that the LTI is as well. The situation prior to commissurotomy is, therefore, as follows for at least some right-handed native speakers. Lexical resources are stored in both hemispheres, and the mechanisms for inflecting and deriving words are located in the left hemisphere. Grammatical resources involved in comprehension are of two kinds. First there are the linking templates, which minimally contain phrase-structural information and the semantic (macro)roles of the arguments in the structure. They may also contain syntactic, semantic and/or information-structural constraints on the construction represented by the template. As an example of these constraints, consider reflexive constructions, as in Bill saw himself in the mirror. They are subject to a variety of constraints governing the choice of the controller (antecedent, i.e. Bill), properties of the reflexive pronoun (himself), and how far away the reflexive pronoun can be from the controller. A rough approximation of the constraints in English, German and Icelandic is given in (14); the semantic role hierarchy is, in a generic, oversimplified form, agent > experiencer > location/goal/source > theme > patient. (14)

Restrictions: a. English

b. Icelandic

c. German

Controller Higher on semantic role hierarchy Highest ranking direct core argument [MR] Older spkrs: NOM only Younger: like Icelandic

Reflexive Agrees with controller person, number, gender Special form only 3 per. No number or gender agreement [sig] Same as Icelandic [sich]

Syntactic domain Simple clause (not across more than one core) Complex clause (can cross more than one core) Same as English

There are both semantic and morphosyntactic constraints here. English is unique among the three languages in allowing more than one possible controller in a clause, as in Bill talked to Sam about himself, where either Bill or Sam can be the one being talked about, something that is unthinkable in the other two languages with sig/sich. Older German speakers have a unique morphologically defined controller, the nominative RP, while Icelanders and younger German speakers have a semantically based constraint. The latter two languages have an invariable reflexive anaphor in terms of person (3rd only), number and gender, in contrast to English. Finally, in the structure represented in Figure 19.12, Bill asked Sally to help himself is possible in Icelandic but not in German or English. These constraints would be part of

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

the linking template for that construction, which would share structural and semantic macrorole properties with non-reflexive constructions (e.g. Bill asked Sally to help Mary).14 Like lexical items, these templates are stored in both hemispheres. The second kind of grammatical resource consists of the mechanisms for selecting and interpreting templates; they are primarily in the left temporal lobe, including BA 22. The interpretation of templates includes the integration of the semantic representations of lexical items with the templates based on the semantic role annotations and other information, as well as the evaluation of the sentence with respect to any constraints on the construction. Thus what may be termed ‘comprehension competence’ is distributed: it is found in the content of the linking templates, in the content of the lexical representations of verbs and other predicates, and in the interpretive mechanisms. After the commissurotomy, what are the linguistic resources in the right hemisphere? At most there would be lexical items in the lexicon and linking templates in the LTI. They are cut off from the left-hemisphere interpretive mechanisms mentioned earlier. Yet a few commissurotomized patients are able to make grammaticality judgements using their right hemisphere. The main grammatical information available in their right hemisphere is in the linking templates, namely, the structure of phrases and clauses, and there is minimal grammatical information in the lexical entries for verbs: while the LSs cannot be interpreted semantically, the number of arguments and the preposition(s) marking oblique arguments can be read off the LSs. Consequently these two sources of information must be the basis for the grammaticality judgements. The issue, then, is how is this grammatical information sufficient for making grammaticality judgements but insufficient for comprehending ‘syntactically constrained sentences’ like reversible passives, as discussed in Section 19.2. The linking templates contain three kinds of information: a representation of phrase structure, the semantic roles of the arguments in the structure, and construction-specific constraints. The semantic role information and the construction-specific constraints (e.g. (14)) cannot be processed in the right hemisphere, and so they are in effect invisible. The two kinds of information express features overlaid on the phrase structure, and they require the sophisticated neural mechanisms found in the left temporal lobe for their interpretation. This accounts for the inability to correctly interpret reversible passives. The default strategy of interpreting the first argument as the actor is a manifestation of the ‘subject preference’ (Wang et al. 2009), which has been found to be operative across languages, and not from an ability to interpret some of the semantic role annotations on the templates but not others. It is primarily the phrase-structure information that makes possible the grammaticality judgements through the mechanism of pattern matching, something the right hemisphere excels at. It is heavily involved in face recognition, for example, and has its own pattern-matching resources independent of the left hemisphere, unlike

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

717

718

ROBERT D. VAN VALIN, JR.

the situation with language. Hence the hypothesis is that split-brain patients who exhibit grammatical behaviour out of their right hemisphere are basing their grammaticality judgements on matching the input sentence with a phrase-structure pattern represented in a linking template. Confronted with sentences like Are you going to the store? and *Are going you to the store?, the patient’s right hemisphere can find a template matching the pattern in the first sentence, but there is no template with the pattern of the second sentence, and therefore it must be judged as ungrammatical. It may seem odd that a passive sentence can be judged grammatical but cannot be interpreted, but this follows from the pattern-matching hypothesis: passive sentences like those in (15a–c) show the same basic pattern (RP-V-PP) as the intransitive sentences in (15a0 –c0 ), namely the one in Figure 19.3. (15)

a. a0 . b. b0 . c. c0 .

Bill was chased by a big dog. Bill was running by a big dog. Donald is being interviewed on Fox News. Donald is binging on Fox News. Sally has been arrested by the police. Sally has been waiting by the police.

The morphological contrast between present and past participles, which is crucial to the interpretation of these sentences, is not represented as a structural feature of the templates in question and is, therefore, difficult to apprehend, apparently. Nevertheless, the passive sentences can be matched with the RP-V-PP template in Figure 19.3 and therefore judged to be grammatical. This account makes a prediction that, if correct, would provide empirical support for it. Baynes and Gazzaniga (1988) report the results of having their commissurotomized patients undertake the same ten tests that Linebarger et al. (1983) gave their ‘agrammatic’ aphasic subjects. The prediction is: subjects’ judgements will be more accurate when the issue is phrasestructural in nature and therefore involves templates and pattern matching than when it involves a non-phrase-structural issue that does not involve the phrase-structural aspects of templates. The ten tests are given in (16) along with a specification as to whether the issue is phrase-structural or not, followed by the scores of the two hemispheres of J.W., one of the patients discussed in Baynes and Gazzaniga (1988), on an index of sensitivity to grammaticality, A0 (see Linebarger et al. 1983: 379). On this index, 1.0 is a perfect score, meaning all examples were correctly identified as grammatical or ungrammatical. (16) a. b. c. d.

Strict subcategorization of verbs Particle movement Subject-aux inversion Empty elements17

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Phrase-structural Phrase-structural Phrase-structural Phrase-structural

16

J.W. LH15 .91 .92 .90 .89

J.W. RH .90 .97 .80 .77

Grammatical Aspects of Language Processing in the Brain

e. f. g. h. i. j.

Tag questions (co-indexing) Left branch condition18 Gapless relative clauses Phrase structure rules Reflexives (co-indexing) Tag questions (aux copying)

Non-structural Phrase-structural Phrase-structural Phrase-structural Non-structural Non-structural Mean accuracy

.92 .94 .92 .75 .85 .92 .93

.77 .92 .89 .90 .65 .74 .84

J.W.’s right hemisphere performed surprisingly well, and the three tests with the biggest difference in performance between the two hemispheres were the three non-phrase-structural tests involving agreement and/or coindexing. Baynes and Gazzaniga (1988) comment on J.W.’s performance: As Fig. 1 [¼ (16)] illustrates, J.W.’s left hemisphere performs well. Although less accurate, his right hemisphere has the most difficulty with the same conditions as the agrammatic aphasics. Inspection of Fig. 1 also indicates that there is a much tighter relation between the pattern generated by J.W.’s right hemisphere and that generated by the aphasics. Moreover, the pattern of his left hemisphere appears to be distinct from that of his right hemisphere. The failure to find any interhemispheric similarity suggests that cross-cueing is not accounting for the good performance of the right hemisphere. (123) The other patient discussed, V.P., had nearly identical performance in both hemispheres. V.P.’s results are more difficult to interpret because of the lack of a convincing difference between her left and right hemispheres. It is especially important to establish a difference in this subject because of the demonstration that some fibers of the splenium remain intact. Because V.P. has previously demonstrated greater right hemispheric syntactic competence than J.W. and because she appears to have some ability to generate speech from her right hemisphere, it would not be surprising that her right hemisphere would do well at this task. Without a greater dissociation of right and left hemisphere performance, little more can be said about V.P.’s performance at this time. (Baynes and Gazzaniga 1988: 124) It seems clear that V.P.’s very similar performance in both hemispheres sheds little light on the prediction being tested, especially given that her two hemispheres are not completely isolated from each other, but the results from J.W.’s right hemisphere strongly suggest that the prediction is correct, because the three tests on which his right hemisphere performed most poorly in comparison with his left hemisphere were precisely the three tests that were not phrase-structural; they involve agreement and coreference, not phrase structure. Thus, the idea that the grammatical competence exhibited by the right hemisphere after commissurotomy derives from the

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

719

720

ROBERT D. VAN VALIN, JR.

interaction of linking templates together with the pattern-matching abilities of the right hemisphere is plausible and provides an explanation for J.W.’s (and possibly V.P.’s) grammatical abilities,19 and this in turn supports the RRG approach to language comprehension laid out in Section 19.4. The question raised at the end of Section 19.2 has been given a plausible, theoretically motivated answer. It might well be objected that this is a rather far-reaching conclusion to be drawn based on data from only one patient, at most two. But J.W. is in fact not unique; there are also the four aphasics in Linebarger et al. (1983), who exhibited the same dissociation as J.W. As mentioned back in Section 19.2, Baynes and Gazzaniga (1988: 124) comment ‘[t]he most striking finding in these two commissurotomized patients with right hemisphere language is the dissociation between the [impaired-RVV] ability to comprehend syntactically constrained sentences and the [intact-RVV] ability to judge their grammaticality. This dissociation can be found in two distinct populations, agrammatic aphasics and commissurotomy patients.’ They were tested on the same materials, and ‘[t]he conditions in which Linebarger et al.’s original four aphasic subjects were consistently weakest were 5[¼(16e)], 9[i], and 10[j].’ (1988: 122–123) Those are precisely the three non-phrase-structural tests and the three that J.W. performed the worst on. Because of this, they conclude: ‘As J.W.’s right hemisphere shows a similar pattern of performance it seems likely that it employs the same mechanism as that of the agrammatic aphasics’ (1988: 124). However, an alternative conclusion would be that the aphasics use the same mechanism as J.W.’s right hemisphere. What are the consequences of the alternative conclusions? They turn out to be of great significance. Linebarger et al. (1983) is cited as a seminal work which demonstrates that aphasics still retain their grammatical knowledge but cannot access it or get it to function in the comprehension process. They can access their grammatical knowledge only enough to make grammaticality judgements. The results themselves, however, and their implications for the status of syntactic knowledge in agrammatism, are quite clear: the comprehension deficit in agrammatism does not reflect loss of the capacity to analyse syntactic structure. We are left now with the important question of why agrammatic aphasics seem unable to make use of this capacity in the comprehension and production of sentences. (Linebarger et al. 1983: 390) Linebarger et al. reject a number of alternative explanations for the dissociation, including template matching: ‘In general, any appeal to “template matching” as an alternative strategy for performing this task seems to us to require templates which are in fact generalizations over structures of considerable syntactic specification’ (1983: 386). The RRG linking templates are just exactly that, as the discussion of them in Section 19.4 made clear. They are syntactically specific, and well motivated empirically and theoretically.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

721

It should be noted that this dissociation between the capacity to comprehend and the ability to make grammaticality judgements has been replicated by numerous scholars in a wide variety of languages, for example Schwartz et al. (1987), Wulfeck (1988), Lukatela et al., (1988) for Serbian, Wulfeck et al. (1991) for Italian, and Lu et al. (2000) for Mandarin Chinese. There are two glaring problems for Linebarger et al.’s account. First, it offers no explanation for how these aphasics make grammaticality judgements and why they are limited to this. Second, it does not apply to J.W. His grammatical knowledge is fully intact in his left hemisphere, and his right hemisphere does not contain any grammatical knowledge of the kind assumed by Linebarger et al. Further, J.W.’s performance on the tests in (16) cannot be attributed to partial, constrained access to his grammatical competence, which is locked away in the left hemisphere with no way to access it from the right hemisphere. If one assumes the analysis given by Linebarger et al., the fact that J.W. performed the same as the aphasics is a striking coincidence, nothing more. Another striking coincidence is found in a comparison of Linebarger et al.’s subjects, J.W.’s results and the results reported in Wulfeck (1988), a replication of the Linebarger et al. study. (17)

Overall mean A0 (sensitivity to grammaticality) Linebarger et al. (1983: 379) Highest Wulfeck (1988: 76) Normal controls: Baynes and Gazzaniga (1988: 123) J.W.’s LH:

.94 .972 .93

All: .90

Lowest: Aphasics: J.W.’s RH:

Linebarger et al.’s subjects have mean A0 values of .94, .91, .91 and .84, which yields an overall value of .90. The best-performing subject is comparable to J.W.’s left hemisphere and near the normal controls in Wulfeck’s study, whereas the worst performer is comparable to the aphasics in Wulfeck’s study and J.W’s right hemisphere. The similarities in the performance of J.W.’s right hemisphere and that of the aphasics in the two studies argue against a solution to the aphasic’s abilities which necessarily excludes J.W. from consideration. What about the alternative conclusion, namely, that the aphasics use the same mechanism as J.W.’s right hemisphere? The explanation given for J.W.’s behaviour applies equally to the aphasics, and it resolves the paradox inherent in Linebarger et al.’s interpretation of their results, namely, why is an aphasic’s grammatical competence accessible for making grammaticality judgements but not other tasks? The answer is that grammatical knowledge in the form of linking templates is preserved, but the processing mechanisms in BA 22 and other parts of the left hemisphere are out of commission. Consequently, the aphasics rely on the intact right hemisphere with its lexicon, LTI and pattern-matching capacity to make grammaticality judgements, just like J.W.’s right hemisphere. This explanation is not limited to fixed-word-order languages like English. For a language like Croatian, the analysis presented here predicts that, just as in English, the argument structure information of verbs will be preserved

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

.84 .821 .84

722

ROBERT D. VAN VALIN, JR.

in the lexical entries in the bilaterally instantiated lexicon, but, unlike in English, they will be formulated in terms of morphological cases (cf. (11)). The linking templates (cf. Figures 19.13, 19.14) are inherently unordered with respect to the positions of the nucleus and the case-marked RP arguments. The grammaticality judgements made by aphasics would involve pattern matching, just as in English, but the matching would be based on the case arrays in the sentence and in the linking template. This crucially depends on sensitivity to morphological case being intact in the aphasics, and this is what Lukatela et al. (1988) found in their study of Serbian aphasics, which concludes ‘the findings of the present study are consistent with other research both on richly inflected languages and on fixed wordorder languages like English . . . A comparison of agrammatics’ performance across tasks shows that subjects who standardly fail in an objectmanipulation task may succeed in a grammaticality judgment task that tests comprehension of the same linguistic structures’ (1988: 13). Thus, the account developed here can explain the Serbian results in terms of the same mechanism that was given for the English results, namely pattern matching with linking templates.

19.6

Implications for Semantics-to-Syntax Linking and Language Production

This chapter has up to now been concerned with language comprehension, which is modelled in an RRG grammar by the syntax-to-semantics linking algorithm. In order to create a plausible processing system, two major changes to the system were proposed. First, the bare syntactic templates of the grammar were augmented with syntactic category information (RP, PP) and semantic macroroles, yielding linking templates. Second, the LSs of verbs and other predicators were augmented with semantic macroroles and the adposition or case carried by oblique core arguments. The result of this is that parsing and syntax-to-semantics linking are collapsed into two steps, as illustrated for a variety of examples in Section 19.4. What implications do these changes have for the linking algorithm for mapping semantics into syntax, which underlies possible models of language production? This question was left open at the end of Van Valin (2006), where it was noted that there were now two sets of syntactic templates in the theory, the bare templates used in linking from semantics to syntax, and the semantically augmented linking templates used in linking from syntax to semantics. There are two sets of LSs as well: the ‘pristine’ LSs used in semantics-tosyntax linking, and the morphosyntactically and macrorole-enhanced LSs used with the linking templates in syntax-to-semantics mapping. There is clearly a great deal of redundancy here that needs to be resolved. The semantics-to-syntax linking algorithm was presented in Chapter 1 and will just be summarized here, using the example Max was interviewed by CNN during the insurrection. (The operator projection is omitted.)

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

(18)

a. Step 1: Construct the semantic representation of the sentence, based on the LS of the main predicator. be-during0 (insurrection, [do0 (CNN, [interview0 (CNN, Max)])]) b. Step 2: Determine the actor and undergoer assignments, following the Actor-Undergoer Hierarchy (AUH).

c. Step 3: Determine the morphosyntactic coding of the arguments (privileged syntactic argument (PSA) selection, case assignment, adposition assignment).

d. Step 4: Select the syntactic template(s) for the sentence following the template selection principles.

e. Step 5: Assign arguments to positions in the syntactic representation of the sentence.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

723

724

ROBERT D. VAN VALIN, JR.

Figure 19.15 Bock and Levelt’s model of grammatical encoding and RRG

There are five steps in the linking algorithm. It was pointed out in Van Valin (2006: 280–282) that these five steps parallel the process of grammatical encoding in the model of a speaker proposed in Bock and Levelt (1994). This parallelism between a psycholinguistic model of speech production and an abstract grammatical theory, as in Figure 19.15, suggests strongly that the RRG semantics-to-syntax linking system has great potential as the basis for a production model. The steps in (18) involve applying rules (‘assign macroroles’, ‘select the PSA’, ‘assign case’, etc.), selecting a bare syntactic template, and inserting lexical items into the template. The content of these rules, etc., is the same information contained in the linking templates and morphosyntactically augmented LSs in the comprehension system, and this is no accident, of course. As an alternative to applying these linking steps sequentially, it is possible to condense the linking into constructing the semantic representation using the augmented LSs with lexical choices for the argument

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Figure 19.16 Condensed semantics-to-syntax linking

positions in them and operators, selecting the passive linking template in Figure 19.7, and integrating the two. This can be represented as in Figure 19.16. The semantic representation is composed of two LSs, do0 (A: x, [interview0 (x, U: y)]) and be-during0 (w, z), the lexical items filling the argument positions in them (the LS for interview fills the z-argument position in the prepositional LS), and the values for the minimally required operators. It is integrated into the passive template from Figure 19.7 using the annotations on the linking template: the undergoer Max occupies the core-initial PSA position which carries the undergoer annotation, the actor CNN appears as the object of the preposition by in the core-level periphery due to its actor annotation, and the adjunct temporal preposition during heads a prepositional phrase which is also in the core-level periphery. The general features of the construction, that is, those that are cross-linguistically valid, as well as the language-specific idiosyncratic features, the verb morphology, for example, are contained in the constructional schema associated with the passive templates. Thus, the five steps in the linking algorithm in (18) have been reduced substantially into a fast production process, analogous to the reduced comprehension process described in Section 19.4. In (18) the AUH, the PSA Selection Hierarchy and the adposition assignment rules, among others, play a role in the grammar version of the linking algorithm. They represent important, linguistically significant generalizations captured in RRG. Are they lost in the condensed linking in Figure 19.16? They are not lost, but their status has changed. The AUH, for example, governs macrorole assignment in step 2 in (18). There is no macrorole assignment per se in the process in Figure 19.16: both the linking template and the LS of the main predicator have the macroroles

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

725

726

ROBERT D. VAN VALIN, JR.

specified in them. The linking regularities captured by the AUH are generalizations across LSs; in other words, the generalizations expressed in the AUH are now meta-generalizations across LSs (as in the meta-grammar of (L) TAG; see Kallmeyer 2010) rather than being a rule constraining the assignment of macroroles. The LSs in (3) express the basic relationship with an Mtransitive verb, while the LSs in (4), (5) and (11) lead to the metageneralization that variable undergoer choice is possible with certain three-place verbs in certain languages. This also holds true for PSA selection, adposition and case assignment, etc. Thus, the constraints and rules as well as the hierarchies that inform them are not invoked as grammatical operations to be applied sequentially, as in (18), but rather are conceptualized as abstract meta-generalizations across the linking templates and the augmented LSs. One of the daunting challenges facing language processing models, both of comprehension and production, is accounting for the incremental nature of processing. This issue was addressed in Section 19.4 for comprehension, where the solution was to posit parallel template processing, whereby the number of competing templates is reduced as the system encounters more and more of the input utterance, until eventually there is a single templatic structure remaining (for unambiguous sentences). With production, on the other hand, the issue is that speakers can start an utterance without having planned it to the end (Bock 1995), and accordingly, the linking in (18) is an idealization which assumes that the entire utterance is planned out in advance. In reality, the semantic representation in Figure 19.16 is compatible with the possibilities in (19), among others. (19)

a. b. c. d. e. f.

CNN interviewed Max during the insurrection During the insurrection, CNN interviewed Max. During the insurrection CNN interviewed Max. Max was interviewed by CNN during the insurrection. During the insurrection, Max was interviewed by CNN. During the insurrection Max was interviewed by CNN.

The speaker has to decide whether the main clause is to be active or passive voice, a decision influenced by the relative topicality of the referents of the RPs CNN and Max, and whether the temporal PP is to occur in its default position in the core-level periphery or in initial position, and if initial, whether it is a frame-setting topic in the pre-detached position or part of the assertion in the pre-core slot. If the speaker begins with during, then (19a, d) are eliminated, but the remaining four are still possible. If they begin with CNN, then only (19a) is possible, and likewise if the utterance starts with Max, only (19d) is possible. There are different commitment points in the sentence; if it starts with the PP, the speaker is not yet committed to the voice of the main clause, whereas if it starts with an RP, the speaker is committed to the form of the main clause and the placement of the PP.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

This example has assumed that the speaker had decided on the predicator, but it is not unusual for there to be competing predicators with different implications in play. An alternative way of describing Max’s encounter with CNN would be with the verb talk (to) as in (20). (The tense has been changed to present perfect in both semantic representations.) (20)

a. Max has talked to CNN during the insurrection. b. ⟨IF DEC ⟨TNS PRES ⟨ASP PERF [be-during0 (insurrection, [do0 (A: Max, [talk0 (Max, NMR to: CNN)])])]⟩⟩⟩

The speaker has options for the semantic representation as well as the syntactic form of the utterance, and for the sake of the discussion, it is assumed that Max is the topic and so the topic expression, the RP Max, will be the subject of the main clause. Consequently, the main clause beginning with Max has. . . can be continued with either been interviewed by CNN or talked to CNN, reflecting two very different syntactic patterns. The speaker can begin a sentence before deciding on a final structure for the utterance. Because structural templates play a crucial role in the processing system, the incremental nature of processing is captured by competition among possible templates which is resolved at critical points during the sentence. For comprehension, the number of potentially relevant templates is reduced word by word as the receiver processes the sentence linearly. The result of this winnowing should ideally be a single structure associated with a single meaning, and multiple structures and meanings create ambiguity. For production, the speaker can begin uttering a sentence without having decided the final form and meaning of it. Here again, there are semantic representations and structural patterns competing, and the choices made at critical points are a function of the speaker’s intention to communicate a particular message. It has become clear that in order to use an abstract competence grammar for language processing, both production and comprehension, it is necessary to make changes to the system. In particular, the rules and constraints that the abstract grammar employs in the mapping between meaning and form and between form and meaning are precompiled in the syntactic templates and LSs for verbs and other predicators. These linking templates and augmented LSs play a central role in both production and comprehension, affording rapid and direct coding in the case of production and what native speakers perceive as nearly instantaneous interpretation with respect to comprehension. There are, accordingly, two versions of RRG: the familiar abstract grammar in Figure 19.17, and the processing system in Figure 19.18. The same rules, principles, constraints, etc. are captured in both versions, albeit differently. In the version in Figure 19.17, they are stated explicitly as part of the linking algorithm, as in (18), whereas in the version in Figure 19.18 they are meta-generalizations over the linking templates and augmented LSs.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

727

728

ROBERT D. VAN VALIN, JR.

Figure 19.17 RRG as an abstract model of grammar

Figure 19.18 RRG as a language processing system

The double-headed arrow in Figure 19.18 signals that the linking is bidirectional, which is expected in RRG, and it is annotated with ‘Winnowing’ and ‘Discourse-Pragmatics’. ‘Winnowing’ was discussed above. ‘Discourse-pragmatics’ indicates that, from an RRG perspective, it is an important factor in the mapping between form and meaning (see Bentley, Chapter 11, Latrouite and Van Valin, Chapter 12). The interaction with morphosyntax has been labelled ‘information structure’ and ‘information packaging’, and the latter term refers to the packaging of the informational content of the utterance by the speaker in order to facilitate its apprehension by interlocutors. This entails that discourse-pragmatics affects the linking to syntax more than the linking to semantics, because the speaker is making choices regarding everything from lexical choices to word order which can be influenced by it. Indeed, in Van Valin and LaPolla (1997: 423–430) it was shown that discoursepragmatics can interact with and influence every step of the semantics-tosyntax linking (see also Sections 12.5–12.7 in Chapter 12). It is not the case that it influences every step in every language; rather, looking across languages one finds the influence at a given step in one or more languages. However, there are no steps in Figure 19.18. In terms of the processing version of RRG, the influence of discourse-pragmatics can be seen in the lexicon and the LTI. In the lexicon, it affects the choice of LS for the predicator, the choice of referring expression(s) to fill the argument slots in the LS(s), and even in some cases macrorole selection. In the LTI, it can influence the choice of

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

construction, whether the utterance needs a PrDS for a frame-setting topic, a PrCS for a contrastive topic or focus, or the order of the words in the sentence, among other things. The importance of discourse-pragmatics for interpretation is rather different. It is not tied to specific steps in the linking, even in the system in Figure 19.17 (Van Valin and LaPolla 1997: 428). It ranges from constraining the interpretation of pronouns within sentences (Van Valin and LaPolla 1997: §5.6), as well as the scope of quantifiers (Van Valin 2005: §3.6) and negation (Van Valin and LaPolla 1997: 219), to supplying missing arguments in discourse (Van Valin 2005: 171–175) and even missing predicators (Shimojo 2008). It can, therefore, filter out certain possible interpretations, as well as supporting other ones, as part of the winnowing.

19.7

Conclusion

This investigation of language processing in the brain from an RRG perspective took as its starting point a puzzling phenomenon: the ability of some split-brain patients to make grammaticality judgements using only their right hemisphere. The explanation for this phenomenon led to a view of grammatical knowledge as being split up among linking templates, lexical entries for verbs and other predicates, and the neural mechanisms in the left temporal lobe which combine and match templates with sentences and integrate the semantic representations in order to capture the basic ‘who did what to whom’ meaning rapidly and efficiently (‘good enough’ interpretation). Moreover, these same constructs showed themselves to be crucial for language production as well. The title of Baynes and Gazzaniga (1988) asks ‘Right hemisphere language: Insight into normal language mechanisms?’, and the answer is clearly affirmative. On the RRG view, then, there is actually no ‘phrase structure building process’ of the kind discussed in Section 19.2; the basic phrase structure configurations already exist in the form of the linking templates, which carry syntactic and semantic information, are combined to form more complex structures, and integrate with the lexical semantic representations of verbs and other predicates, which also carry morphosyntactic information. Hence syntactic processing and semantic interpretation are intimately intertwined, leading to the conclusion of Fedorenko et al. (2020: 1) that ‘lexico-semantic and syntactic processing are deeply interconnected and perhaps not separable’. The RRG conception of grammatical knowledge as involving linking templates turned out to have unexpected implications for the understanding of certain types of aphasia. As discussed, there is a similar puzzling phenomenon in that domain, namely, the dissociation between the ability to make grammaticality judgements and the inability to make use of that same grammatical knowledge in comprehension and production. The

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

729

730

ROBERT D. VAN VALIN, JR.

explanation for the dissociation turned out to involve the same devices as the right hemisphere grammaticality judgements. While there were only a small number of split-brain patients and only a small minority exhibited any linguistic capacity in their right hemisphere, the dissociation in aphasia has been documented in numerous studies involving speakers from a variety of languages. The paper began with a reference to Kaplan and Bresnan’s assertion that there should be a link between abstract competence grammars and processing models. It has been argued that the mechanisms of RRG can be the basis for neurolinguistic models of both language comprehension and production, and the exact nature of the adaptations required by the processing models have been articulated explicitly and summarized in Figures 19.17 and 19.18. A distinctive characteristic of the RRG account is the distributed representation of grammatical knowledge in terms of linking templates and augmented LSs, which, in terms of the dual-stream models, is fully compatible with them being accessible for the dorsal stream (production) as well as the ventral stream (comprehension). RRG thus meets the requirement that a grammatical theory claiming to capture speakers’ linguistic competence should be able to be used in a testable psycholinguistic, or in this case, neurolinguistic, processing model.

References Baynes, Kathleen and Michael S. Gazzaniga. 1988. Right hemisphere language: Insight into normal language mechanisms? In F. Plum (ed.), Language, Communication and the Brain, 117–126. New York: Raven Press. Blank, Idan, Zuzanna Balewski, Kyle Mahowald and Evelina Fedorenko. 2016. Syntactic processing is distributed across the language system. NeuroImage 127: 307–323. Bock, Kay. 1995. Sentence production: From mind to mouth. In J. L. Miller and P. D. Eimas (eds.), Speech, Language and Communication (2nd ed.), 181–216. New York: Academic Press. Bock, Kathryn and Willem Levelt. 1994. Language production: Grammatical encoding. In M. Gernsbacher (ed.), Handbook of Psycholinguistics, 945–984. New York: Academic Press. Boland, Julie, Michael Tanenhaus, Susan Garnsey and Greg Carlson. 1995. Verb argument structure in parsing and interpretation: Evidence from whquestions. Journal of Memory and Language 34: 774–806. Bornkessel, Ina and Matthias Schlesewsky. 2006. The extended Argument Dependency Model: a neurocognitive approach to sentence comprehension across languages. Psychological Review 113: 787–821. Bornkessel-Schlesewsky, Ina and Matthias Schlesewsky. 2013. Reconciling time, space and function: A new dorsal–ventral stream model of sentence comprehension. Brain and Language 125: 60–76.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Brennan, Jonathan, Yuval Nir, Uri Hasson, Rafael Malach, David J. Heeger and Liina Pylkkänen. 2012. Syntactic structure building in the anterior temporal lobe during natural story listening. Brain and Language 120: 163–173. Brodmann, Korbinian. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde. Leipzig: Johann Ambrosius Barth. Clifton, Charles, Jr. and Lynn Frazier. 1989. Comprehending sentences with long-distance dependencies. In Greg Carlson and Michael Tanenhaus (eds.), Linguistic Structure in Language Processing, 273–317. Dordrecht: Kluwer. Davis, Anthony and Jean-Pierre Koenig. 2000. Linking as constraints on word classes in a hierarchical lexicon. Language 76: 56–91. Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B. Redfern and Jeri J. Jaeger. 1994. A reconsideration of the brain areas involved in the disruption of morphosyntactic comprehension. Brain and Language 47(3): 461–463. Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B. Redfern and Jeri J. Jaeger. 2004. Lesion analysis of the brain areas involved in language comprehension. Cognition 92: 145–177. Federenko, Evelina, Idan A. Blank, Mathew Siegelmann and Zachary Mineroff. 2020. Lack of selectivity for syntax relative to word meanings throughout the language network. Cognition 203: 1–19. Ferreira, Fernanda and Nikole D. Patson. 2007. The ‘good enough’ approach to language comprehension. Language and Linguistics Compass 1(1–2): 71–83. Friederici, Angela D. 2009. Pathways to language: Fiber tracts in the human brain. Trends in Cognitive Sciences 13: 175–181. Friederici, Angela D. 2012. The cortical language circuit: From auditory perception to sentence comprehension. Trends in Cognitive Sciences 16: 262–268. Friederici, Angela D., Martin Meyer and D. Yves von Cramon. 2000. Auditory language comprehension: An event-related fMRI study on the processing of syntactic and lexical information. Brain and Language 74: 289–300. Friederici, Angela D., Shirley-Ann Rüschemeyer, Anja Hahne and Christian J. Fiebach. 2003. The role of left inferior frontal and superior temporal cortex in sentence comprehension: Localizing syntactic and semantic processes. Cerebral Cortex 13: 170–177. Gazzaniga, Michael S. 1983. Right hemisphere language after brain bisection: A 20-year perspective. American Psychologist 38: 525–537. Gazzaniga, Michael S. 2000. Cerebral specialization and interhemispheric communication: Does the corpus callosum enable the human condition? Brain 123: 1293–1326. Gazzaniga, Michael S. 2005. Forty-five years of split-brain research and still going strong. Nature Reviews Neuroscience 6: 653–659. Gazzaniga, Michael S. and Steven A. Hillyard. 1971. Language and speech capacity of the right hemisphere. Neuropsychologia 9: 273–280.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

731

732

ROBERT D. VAN VALIN, JR.

Herrmann, Björn, Burkhard Maess, Anja Hahne, Erich Schröger and Angela D. Friederici. 2011. Syntactic and auditory spatial processing in the human temporal cortex: An MEG study. NeuroImage 57: 624–633. Hickok, Gregory and David Poeppel. 2004. Dorsal and ventral streams: A framework for understanding aspects of the functional anatomy of language. Cognition 92: 67–99. Hickok, Gregory and David Poeppel. 2007. The cortical organization of speech perception. Nature Reviews Neuroscience 8: 393–402. Jurafsky, Dan. 1996. A probabilistic model of lexical and syntactic access and disambiguation. Cognitive Science 20: 137–194. Kallmeyer, Laura. 2010. Parsing Beyond Context-Free Grammars. Heidelberg: Springer. Kallmeyer, Laura, Rainer Osswald and Robert D. Van Valin, Jr. 2013. Tree wrapping for Role and Reference Grammar. In G. Morrill and M.-J. Nederhof (eds.), Formal Grammar: Proceedings of the 17th and 18th International Conferences (Lecture Notes in Computer Science 8036), 175–190. Heidelberg: Springer. Kaplan, Ronald and Joan Bresnan. 1982. Lexical-Functional Grammar: A formal system for grammatical representation. In Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, 173–281. Cambridge, MA: MIT Press. Koenig, Jean-Pierre and Gail Mauner. 2003. Arguments for adjuncts. Cognition 89: 67–103. Lau, Ellen F., Colin Phillips and David Poeppel. 2008. A cortical network for semantics: (de)constructing the N400. Nature Review Neuroscience 9: 920–933. Linebarger, Marcia C., Myrna F. Schwartz and Eleanor M. Saffran. 1983. Sensitivity to grammatical structures in so-called agrammatic aphasics. Cognition 13: 361–392. Lu, Ching-Ching, Elizabeth Bates, Ping Li, Ovid Tzeng, Daisy Hung, Chih-Hao Tsai et al. 2000. Judgements of grammaticality in aphasia: The special case of Chinese. Aphasiology 14: 1021–1054. Lukatela, Katerina, Stephen Crain and Donald Shankweiler. 1988. Sensitivity to inflectional morphology in agrammatism: Investigation of a highly inflected language. Brain and Language 33: 1–15. Magnusdottir, S., P. Fillmore, D. B. den Ouden, H. Hjaltason, C. Rorden, O. Kjartansson, et al. 2013. Damage to left anterior temporal cortex predicts impairment of complex syntactic processing: A lesion-symptom mapping study. Human Brain Mapping 34: 2715–2723. Mazoyer, B. M., N. Tzourio, V. Frak, A. Syrota, N. Murayama, O. Levrier, G. Salamon, S. Dehaene, L. Cohen and J. Mehler. 1993. The cortical representation of speech. Journal of Cognitive Neuroscience 5: 467–479. McKoon, Gail and Jessica Love. 2011. Verbs in the lexicon: Why is hitting easier than breaking? Language and Cognition 3(2): 313–330.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

Nolan, Brian and Elke Diedrichsen (eds.). 2013. Linking Constructions into Functional Linguistics: The Role of Constructions in Grammar. Amsterdam: John Benjamins. Obleser, Jonas, Jonas Zimmermann, John Van Meter and Josef P. Rauschecker. 2007. Multiple stages of auditory speech perception reflected in event-related fMRI. Cerebral Cortex 17: 2251–2257. Pylkkänen, Liina. 2019. The neural basis of combinatory syntax and semantics. Science 366: 62–66. Saur, Dorothee, Björn W. Kreher, Susanne Schnell, Dorothee Kümmerer, Philipp Kellmeyer, Magnus-Sebastian Vry, et al. 2008. Ventral and dorsal pathways for language. PNAS USA 105(46): 18035–18040. Schlesewsky, Matthias and Ina Bornkessel. 2004. On incremental interpretation: Degrees of meaning accessed during sentence comprehension. Lingua 114: 1213–1234. Schwartz, Myrna F., Eleanor M. Saffran and Oscar S. Marin. 1980. The word order problem in agrammatism: I. Comprehension. Brain and Language 10: 249–262. Schwartz, Myrna F., Marcia C. Linebarger, Eleanor M. Saffran and David S. Pate. 1987. Syntactic transparency and sentence interpretation in aphasia. Language and Cognitive Processes 2: 85–113. Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less numeral quantifier construction in Japanese. In Robert Van Valin (ed.), Investigations of the Syntax–Semantics–Pragmatics Interface, 285–304. Amsterdam: John Benjamins. Shtyrov, Yury, Friedemann Pulvermüller, Risto Näätänen and Risto J. Ilmoniemi. 2003. Grammar processing outside the focus of attention: An MEG study. Journal of Cognitive Neuroscience 15: 1195–1206. Stowe, L. A. 1985. Parsing wh-constructions: Evidence for on-line gap location. Language and Cognitive Processes 1: 227–245. Stowe, L. A., C. A. J. Broere, A. M. J. Paans, A. A. Wijers, G. Mulder, W. Vaalburg and F. Zwarts. 1998. Localizing cognitive components of a complex task: sentence processing and working memory. Neuroreport 9: 2995–2999. Stromswold, Karin, David Caplan, Nathaniel Alpert and Scott Rauch. 1996. Localization of syntactic comprehension by positron emission tomography. Brain and Language 52: 452–473. Townsend, David J. and Thomas G. Bever. 2001. Sentence Comprehension: The Integration of Habits and Rules. Cambridge, MA: MIT Press. Traxler, M. and M. Pickering. 1996. Plausibility and the processing of unbounded dependencies: An eye-tracking study. Journal of Memory and Language 35: 454–475. Turken, And U. and Nina F. Dronkers. 2011. The neural architecture of the language comprehension network: Converging evidence from lesion and connectivity analyses. Frontiers in Systems Neuroscience 5: 1.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

733

734

ROBERT D. VAN VALIN, JR.

Tyler, L. K. and W. Marslen-Wilson. 2008. Fronto-temporal brain systems supporting spoken language comprehension. Philosophical Transactions of the Royal Society B: Biological Sciences 363: 1037–1054. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2006. Semantic macroroles and language processing. In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie and Angela D. Friederici (eds.), Semantic Role Universals and Argument Linking: Theoretical, Typological, and Psycholinguistic Perspectives, 263–301. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wang, Luming, Matthias Schlesewsky, Balthasar Bickel and Ina BornkesselSchlesewsky. 2009. Exploring the nature of the ‘subject’-preference: Evidence from the online comprehension of simple sentences in Mandarin Chinese. Language and Cognitive Processes 24: 1180–1226. Wulfeck, Beverly B. 1988. Grammaticality judgments and sentence comprehension in agrammatic aphasia. Journal of Speech and Hearing Research 31: 72–81. Wulfeck, Beverly, Elizabeth Bates and Rita Capasso. 1991. A crosslinguistic study of grammaticality judgments in Broca’s aphasia. Brain and Language 41: 311–336. Zaidel, Eran. 1983. On multiple representations of the lexicon in the brain: The case of two hemispheres. In Michael Studdert-Kennedy (ed.), The Psychobiology of Language, 105–122. Cambridge, MA: MIT Press.

Notes 1 See Gazzaniga (2005) for an overview of the history of split-brain research and (2000) for an explication of the methodologies used. 2 Brodmann (1909) analysed the structure of the cortex and grouped areas in the cortex with the same cytoarchitectonic features into what are called ‘Brodmann’s areas’. This means that all of the cortex in a given BA, e.g. BA 22, has the same cytoarchitecture. 3 Some models, e.g. Hickok and Poeppel (2007), Tyler and Marslen-Wilson (2008), posit right-hemisphere involvement in the comprehension process, and these concern semantic and pragmatic aspects of comprehension, not morphosyntactic processing, which they locate exclusively in the left hemisphere. 4 Pylkkänen (2019) proposes a view of language in the brain which crucially involves the left anterior temporal lobe and which treats ‘syntax as knowledge, semantics as process’ (64–65):

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

Grammatical Aspects of Language Processing in the Brain

The neuroscience-of-language field has long assumed that our brains build syntactic structure during language processing. Today it is reasonable to question this assumption. . . . But what if we cannot find evidence that our brains actually build syntactic structure? Syntax may be something that the brain knows rather than does. Perhaps the combinatory steps, which consume energy and make our neurons fire, are all semantic, and syntactic processing amounts to comparing these semantic structures to our stored knowledge of syntax. The knowledge may have the format of generative rules that create structure . . . or may represent the structures themselves. 5 This should not be interpreted to imply that no other brain regions in the left temporal lobe are involved in the comprehension process; the point is that BA 22 is important for it. 6 Be is attached to the AUX node in the constituent projection because it is structurally required for non-verbal nuclei and passive verbs as well, whereas ‘(aux)’ in the templates represents auxiliary verbs which are only part of the expression of operators, e.g. be in the progressive aspect or have in the perfect. Structural be is required even when there are no operators, e.g. Sally wants to be a lawyer/tall. 7 The precompiling of syntactic and semantic information as proposed in Van Valin (2006) and here moves RRG closer to (Sign-based) Construction Grammar and Head-Driven Phrase Structure Grammar (e.g. Davis and Koenig 2000). RRG was already a species of construction grammar in the generic sense, since it recognizes the importance and centrality of the notion of grammatical construction in linguistic theory and description; the realization of this in the theory is rather different from ‘standard’ versions of Construction Grammar, however. 8 This should be read as: in the linking template the RP the burglar is the A, and in the LS the A is the x-argument, and consequently the burglar fills the x-argument position in the LS. The same holds for the window and the y-argument position in the LS. 9 Did is in curly brackets to signify that it is part of the operator projection rather than the constituent projection. 10 The embedded clause can be iterated for additional long-distance whquestions. The long-distance (non-local) nature of the tree in Figure 19.11 is an unusual feature for templates, since in approaches like (L)TAG (Kallmeyer 2010) tree templates are all local, and a non-local tree would be the result of multiple combinatory operations (Kallmeyer et al. 2013). For an explanation of why the embedded clause is a daughter of the CLAUSE node rather than the CORE node in Figure 19.11, see Van Valin 2005: §6.3. The tree in Figure 19.12 also has some non-local properties, albeit different ones from Figure 19.11.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

735

736

ROBERT D. VAN VALIN, JR.

11 This account is somewhat oversimplified, but it is adequate for the purposes of this discussion. 12 There are languages with no adjacency restrictions on RP constituents and no grammatical constraints on linearization, e.g. many Australian Aboriginal languages, and they too would have unordered clausal and phrasal templates, yielding grammatically (but not pragmatically) unconstrained word order. See Van Valin (2005: 28–29) for an example. 13 The presence of word storage in the right hemisphere raises interesting questions for theories like Distributed Morphology, which do not posit a lexicon in the usual sense but rather a repository of abstract roots that must undergo derivation in the syntax before they can surface as words. If this repository were in both hemispheres, the isolated right hemisphere in a split-brain patient would have no access to the syntactic machinery which derives words from the abstract roots. Yet some are able to recognize words, which is unexpected due to the lack of the necessary syntactic machinery. 14 In an RRG grammar these constraints would be expressed in the constructional schema for the construction. We refer to the relevant discussion in Chapter 1 (§§1.6.3–1.6.4) and to Van Valin (2005: Chapters 5, 7) and Nolan and Diedrichsen (2013) for extensive exemplification. 15 All of the values except the ones for mean accuracy are extrapolated from Fig. 1 in Baynes and Gazzaniga (1988). 16 This actually involves both lexical and phrase-structural information, both of which are available in the right hemisphere, as noted above, since the lexical entries for verbs are the same in both hemispheres. 17 These included: gaps associated with wh-questions (subject and object), missing arguments in infinitival but not tensed complements, double gaps. 18 How many birds did you see? vs. *How many did you see birds? 19 It has been suggested that J.W.’s right-hemisphere grammatical ability is the result of an incomplete isolation of the hemispheres or of incomplete left-lateralization of language at the time of the surgery. The former is unlikely, for three reasons. First, Baynes and Gazzaniga report that an MRI showed a complete separation of the hemispheres, second, they also point out that there is no evidence of cross-cueing between the hemispheres, and third, his test results were different from V.P.’s, who did not have a complete isolation of the two hemispheres. With respect to the possibility of incomplete lateralization, it can’t be ruled out a priori, but it predicts neither the pattern of his responses nor the similarity of his responses to the aphasics, unlike the account proposed above.

https://doi.org/10.1017/9781316418086.024 Published online by Cambridge University Press

20 Formalization of RRG Syntax Laura Kallmeyer and Rainer Osswald

Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: AC ACT AM ANIM AOR ASP CFTG CO DP ECS EDL IF LCFRS LM

20.1

actor active accompanied motion animate aorist aspect Context-Free Tree Grammar core detached position extra-core slot extended domain of locality illocutionary force Linear Context-Free Rewriting System linkage marker

MOD NUC OP OPS PERIS PrCS PSBL RP STA TAG TNS TWG UG

modal nucleus operator operator structure periphery structure pre-core slot possible reference phrase status Tree Adjoining Grammar tense Tree Wrapping Grammar undergoer

Introduction

Role and Reference Grammar (RRG) has been developed as a theory of grammar that covers typologically distinct languages and is able to capture the interaction between syntax, semantics and pragmatics.* The design of * The research presented in this chapter was supported by the Collaborative Research Centre 991 ‘The Structure of Representations in Language, Cognition, and Science’ funded by the German Research Foundation (DFG) and the project TreeGraSP funded by the European Research Council (ERC).

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

738

LAURA KALLMEYER AND RAINER OSSWALD

RRG was not driven by specific formal considerations. In particular, there is no formal core that plays a crucial role in RRG, as, for example, the theory of feature structures does in Head-Driven Phrase Structure Grammar (HPSG; Pollard and Sag 1994). The goal of this chapter is to present RRG as a formalized grammatical theory that puts emphasis on mathematical and logical rigour. In particular, it will be shown how the possible universal and language-specific syntactic templates of RRG can be formally specified, and by which formal operations they can be combined into larger syntactic structures. Working typologists who use RRG as their framework for linguistic analysis may not regard a thorough formalization as particularly important. In fact, one of the appeals RRG has for field linguistics is that it does not come with an overly heavy theoretical load but keeps a good balance between a rich and elaborate set of notions and explanatory mechanisms and a semi-formal, intuitive presentation. A formalization can, however, help to identify and eliminate possible gaps and inconsistencies of the theory and, thereby, to improve the theory. Moreover, a formalization can serve as a basis for computational implementations of RRG. While a thorough formalization may not be absolutely necessary for a computational treatment from an engineering perspective (see Chapter 21 on computational implementation and applications of RRG), it can contribute to implementations that give full consideration to the overall architecture of RRG as a theory of grammar.

20.2

The Task of Formalizing RRG Syntax

In RRG, syntactic representations are composed of syntactic templates stored in the syntactic inventory. Figure 20.1a shows a simple example of a syntactic representation; possible candidates for syntactic templates are shown in Figure 20.1b. A formalization first needs to decide on what kind of formal structures to use. In line with the tradition of RRG, tree structures will be employed for this purpose, where nodes can carry additional features

(a)

(b)

SENTENCE

SENTENCE

CLAUSE PrCS

CORE

CLAUSE PrCS

NUC RP

what

RP

did TNS

Kim

CLAUSE

PERIPHERY

CORE

CORE CORE

PRED V

ADV

smash

yesterday

PrCS

RP

RP

CLAUSE

Figure 20.1 Examples of syntactic representation and syntactic templates in RRG

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

NUC PRED V

PERIPHERY ADV

Formalization of RRG Syntax

besides the category labels. The second task is to define the modes by which the syntactic representations are composed from the members of the syntactic inventory (Section 20.3). Next, the templates available in the inventory must be specified, and this should be done in a way that allows us to capture generalizations among them within and across languages (Section 20.4). A further point that needs special treatment is the formalization of the operator projection (Section 20.5). A key component of RRG’s approach to syntactic analysis is the layered structure of the clause: sentences are assumed to have an internal structural layering consisting of clause, core and nucleus. The different layers serve as attachment sites for different types of operators: tense operators attach to the clause, modality to the core, aspect to the nucleus, etc. The core level is also the default attachment site for arguments. In the following, we will refer to the subtree of a syntactic representation consisting of the root and its non-peripheral clause, core, nucleus and predicating descendants as the clausal skeleton of the representation. The syntactic structures in RRG are basically labelled trees, and there are good reasons to use tree structures in a formalization as well. Trees provide the most natural way to analyse syntactic structures since they build on the basic relations of immediate dominance and linear precedence.

20.2.1 Approaches to Formalizing RRG Syntax The formal specification of syntactic structures in RRG is briefly addressed by Van Valin and LaPolla (1997, §2.5), where two approaches are discussed: (i) the specification by immediate dominance and linear precedence rules and (ii) the specification by syntactic tree templates and their combination. As to approach (i), Van Valin and LaPolla (1997: 69f ) propose the following universal immediate dominance rules for the constituent structure of simple sentences:1 (1)

SENTENCE DP CLAUSE ECS PERIPHERY CORE NUC PRED ARG XP

! ! ! ! ! ! ! ! ! !

{(DP)}, CLAUSE XP j ADV {(ECS)}, CORE, (PERIPHERY), {NP*} XP j ADV XP j ADV ARG*, NUC PRED V j XP PRO j XP NP j PP

The commas on the right-hand side of these rules do not indicate any ordering of the subconstituents. The ordering is specified by additional linear precedence rules, which are partly universal and partly languagespecific. For example, English, a verb-medial language, obeys the following linear precedence rules (Van Valin and LaPolla 1997: 71):

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

739

740

LAURA KALLMEYER AND RAINER OSSWALD

(2)

NP > NUC NUC > NP* > PP*

As Van Valin and LaPolla point out, the range of possible syntactic constellations specified by the above rules needs to be further constrained by the linking of syntax to semantics, which includes constraints on the syntactic realization of arguments depending on semantic structures in the lexicon. In addition to the constituent structure rules in (1) and (2), a separate set of immediate dominance and linear precedence rules is needed for specifying the structure of the operator projection. The idea of specifying the constituent structure and the operator structure separately by different context-free grammars was originally proposed by Johnson (1987), based on the observation that the ordering among the operators is systematically correlated with their scope given by their attachment site at the clausal skeleton, whereas the surface order of the operators relative to arguments and adjuncts is much less transparent and often requires crossing branches. The two grammars taken together then constitute a projection grammar, giving rise to a constituent projection and an operator projection. Johnson formally defines a projection grammar as a quadruple

 P ¼ T; I; ðAi Þi2I , ðGi Þi2I , where T is a set of terminal symbols, I is a finite set of ‘projection’ indices and, for every i 2 I, Ai is a subset of T and Gi is a formal grammar with terminal symbols in Ai . A string s over T then belongs to the language generated by P if and only if its ith projection, that is, the concatenation of the elements of s belonging to Ai , is in the language generated by Gi . While it seems reasonable to distinguish between constituent structure and operator structure, Johnson’s proposal has the problem of being purely surface oriented. As a consequence, it does not enforce matching clausal skeletons in the two projections. However, corresponding clausal skeletons in both projections are taken for granted in the syntactic representations of RRG. A further problem arises from the assumption that the operator projection can be represented as a tree, that is, that each operator contributes only to one layer (cf. Section 20.5 for counterexamples). Approach (ii), the second approach discussed by Van Valin and LaPolla, postulates an inventory of elementary syntactic trees that can be combined into more complex syntactic structures. Van Valin and LaPolla (1997: 654, note 34) point out that Tree Adjoining Grammars (TAG, Joshi and Schabes 1997) may provide a way to formalize such tree templates and their composition. Building on this idea, Kallmeyer et al. (2013) (see also Kallmeyer and Osswald 2017; Osswald and Kallmeyer 2018) propose a formalization of RRG as a grammar based on so-called elementary trees and TAGinspired tree composition operations. We will detail this approach in Section 20.3. A slightly different proposal has been suggested by Nolan (2004), who argues for a formalization of RRG that systematically exploits feature-based

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

representations, similar in style to HPSG (Pollard and Sag 1994) and, more recently, Sign-Based Construction Grammar (Sag 2012). Representing constituent structures in this way calls for features, or attributes, by which the subconstituents can be addressed. This can be done either by reconstructing tree structures as feature structures based on formal features such as first and rest, or by employing functional notions like subject, direct-object, etc. However, in RRG, configurational syntactic notions are usually not considered as basic but rather as derived concepts. While the representation of the constituent projection proposed in Nolan (2004, §5.5), which builds on the immediate dominance rules given in (1), is not fully explicit about the attributes involved, it seems that it uses either pure list-oriented attributes (first and rest, or even 1st, 2nd, 3rd, etc.) or attributes specific to the categories of the subconstituents.

20.2.2

The Formal Treatment of Operators and Peripheral Elements The formalization presented in the following assumes a single syntactic tree structure in which operator components are distinguished by a special category OP and a feature structure that characterizes their contribution. For example, the operator did in Figure 20.2 contributes [tense past] to the CLAUSE and the definiteness operator the contributes [def þ] to the RP layer of boy.2 This representation, together with the approach to operator adjunction presented in Section 20.5, turns out to be sufficient for capturing the scope-related ordering among the operators. The operator projection can then be defined as the subtree consisting of the clausal skeleton plus the components with category OP.3 In RRG’s graphical presentations of syntactic structures, peripheral structures and linkage markers are usually attached to the clausal SENTENCE CLAUSE PrCS OP[TENSE past] RP PRO what

CORE

did

NUC

RP OP[DEF +] CORER the

NUCR

V[PRED +] smash

N boy Figure 20.2 Operator marking by features

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

741

742

LAURA KALLMEYER AND RAINER OSSWALD

skeleton by arrows (cf. Figure 20.1a). The formalization proposed here does not make use of separate PERIPHERY nodes but marks peripheral structures by a feature [peri þ], as indicated in Figure 20.3. As with operators, peripheral elements are subject to the iconicity principle that their relative ordering respects the layering of their attachment sites. This aspect will be taken care of in the formalization of the operator projection and the periphery structure described in Sections 20.5 and 20.7. The proposed modes of combining peripheral structures with the clausal skeleton are similar to those used for operators (Section 20.3.2). Linkage markers can be treated similarly. To sum up, the syntactic structures of RRG can be formalized as labelled trees, where node labels can carry additional features. Since the labels can be regarded as feature values, too (e.g. category labels as values of the feature cat), we may assume without restriction of generality that node labels are sets of attribute-value specifications. From this perspective, MP[PERI þ] is short for [cat mp, peri þ]. Introducing features allowed us to get rid of the PERIPHERY nodes, whose only purpose is to mark their daughters as peripheral. By the same line of reasoning, we can eliminate the PRED nodes, whose purpose is to mark their daughters as predicating (cf. Van Valin 2005: 13), by adding [PRED þ] to the respective daughter nodes, as illustrated by the V nodes in Figures 20.2 and 20.3.

SENTENCE CLAUSE CORE RP

MP[PERI +]

NUC

CORER COREM V[PRED +] OP[DEF +] NUCR

NUCM

N

ADV

Kim

deliberately

smashed

MP[PERI +]

RP CORER

COREM

NUCR

NUCM

the

MP[PERI +] N

COREM vase yesterday NUCM A blue

Figure 20.3 Periphery marking by features

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

ADV

Formalization of RRG Syntax

20.3

A Tree Rewriting Formalism for Syntactic Composition in RRG

The standard presentation of RRG gives only an informal description of how syntactic templates are combined to more complex syntactic structures. As mentioned in Section 20.2.1, Van Valin and LaPolla (1997: 654, note 34) suggest that a formal account of the modes of composition may show some similarity to Tree Adjoining Grammars (TAG, Joshi and Schabes 1997). They rightly point out that TAG is a grammar formalism and not a linguistic theory in its own right. TAG per se does not make any commitments about what kind of categories and what kind of syntactic configurations are appropriate for linguistic analysis. There is one caveat, however: the standard adjunction operation of TAG aims at modifying binary branching structures. Since flat syntactic structures are prevalent in RRG, the formalization proposed in the following employs slightly different modes of composition.

20.3.1 Elementary Trees The architecture of RRG assumes an inventory of syntactic templates as elementary building blocks for syntactic representations (cf. Figure 20.1b). These templates often have a more complex structure than just consisting of a root node together with a series of immediate daughters. They can thus capture a wider range of dependencies than standard phrase structure rules like those listed in (1). In TAG, this property is called the extended domain of locality (EDL) represented in elementary trees (Joshi and Schabes 1997: 95f ). In particular, templates can contain a predicative node and slots for all the arguments of that predicate. An example is given in Figure 20.4, which shows the templates that could be used for generating the syntactic structure for (3). (3)

Kim smashed the blue vase.

In this example, the transitive verb template associated with the verbal predicate smash contains not only the NUC node but also the CORE and the two RP argument slots. Note that the lexical elements (marked grey in the figure) are not part of the templates but are added in a separate step. The availability of larger syntactic units allows one to lexicalize parts of the grammar, up to the point of a fully Lexicalized Tree Adjoining Grammar (LTAG), in which every elementary tree is required to have a lexical anchor. The similarities between RRG’s tree templates and (L)TAG’s elementary trees suggest a formalization of RRG syntax along the following lines: A language is syntactically described by a tree rewriting grammar comprising a set of tree templates, lexical elements filling the anchor nodes of these templates, and certain operations for combining them into syntactic

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

743

744

LAURA KALLMEYER AND RAINER OSSWALD

CORE RP

RP

NUC

CLAUSE

SENTENCE

CORE

CLAUSE

RP

CORER

PRED

NUCR

V

DEF

N

smashed

the

Kim

RP MP

PERIPHERY

CORER

COREM

NUCR

NUCM

N

A

vase

blue Figure 20.4 RRG templates

representations of phrases and sentences. More specifically, a grammar consists of a finite set of elementary trees that can be composed inductively via three basic tree composition operations, namely sister adjunction, substitution and wrapping substitution. Elementary trees are defined as labelled ordered trees whose internal nodes are labelled with categories such as CLAUSE, CORE, RP, N, V etc., and whose leaf nodes are labelled either with lexical items or with syntactic categories. The assumption that elementary trees are ordered trees means: (i) every two nodes that do not stand in a dominance relation are ordered by linear precedence; (ii) if a node n1 precedes a node n2 then every node dominated by n1 precedes every node dominated by n2. Note that these requirements exclude structures with crossing branches. Figure 20.5 shows possible elementary trees (including lexical elements) for the syntactic analysis of sentence (3) and illustrates how they combine. The choice of the elementary trees in the example reflects the following general principles: Each lexical predicate comes with its entire layered structure, including argument slots for all its arguments (see the trees for Kim, smashed, blue and vase in Figure 20.5). Periphery elements, operators and linkage markers contribute an adjunct tree. Such a tree cannot fill an argument slot but has to be adjoined, that is, attached, to the clausal skeleton as an additional daughter of some appropriate node. An asterisk at the root node indicates that a tree is an adjunct tree (see the trees for the operator the and the modifier blue in Figure 20.5). In addition, there can also be non-lexicalized elementary trees (see the SENTENCE–CLAUSE tree in Figure 20.5). The formalization of RRG syntax presented here allows for a further level of (de)composition: Elementary trees are specified in a so-called

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

745

Formalization of RRG Syntax SENTENCE CLAUSE

SENTENCE

CLAUSE

CLAUSE

CORE RP

NUC

CORE

RP

RP

CORER

RP

NUC

CORE∗R

NUCR

CORER

V

OPdef

MPperi

N

NUCR

smashed

the

N

COREM

vase

Kim

NUCM

RP

V

CORER

smashed

RP∗

RP CORER

OPdef

NUCR

MPperi

NUCR

N

COREM

N

Kim

NUCM

vase

the

A

A

blue

blue

Figure 20.5 Elementary RRG trees and their composition

V smashed

CLAUSE

RP

CORE∗R

CORE

CORER

MPperi

NUCR

COREM

N

NUCM

RP NUC RP V

N vase

A blue Figure 20.6 Lexical anchoring of elementary tree templates

meta-grammar, a system of tree constraints, which captures generalizations across elementary trees (more on this in Section 20.4.2). Moreover, lexical items are stored in a separate lexicon and enter their elementary trees by a process of lexical anchoring under constraints. In other words, we distinguish unanchored elementary trees from the anchored trees that enter tree composition. The former are called elementary tree templates. Figure 20.6 shows the decomposition of three of the elementary trees from Figure 20.5 into their tree template and the lexical anchor. The place where the anchor has to be inserted is marked with a diamond symbol. Note that introducing elementary trees as ordered trees does not prevent us from defining grammars for languages with free word order. If, for instance, a language allows the arguments of a verb to appear in an arbitrary order, then the verb would have elementary trees for each such ordering at its disposal. What is important here is that elementary trees are not atomic

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

A

746

LAURA KALLMEYER AND RAINER OSSWALD

building blocks but are generated in a modular fashion from classes of tree constraints in the meta-grammar. Since fewer ordering constraints in the meta-grammar correspond to a larger set of elementary tree templates, it follows, loosely speaking and without taking into account morphological markings, that free word order is descriptively more simple in the metagrammar than strict word order.

20.3.2 Simple Substitution and Sister Adjunction The most basic mode of composition for syntactic templates is substitution. The trees for Kim and for vase in Figure 20.5, for instance, are added by substitution to the tree of smashed, filling the two RP argument slots. In the following, a tree with label X is meant to be a tree whose root carries the category label X. A tree β with label X can be substituted for a leaf node labelled with X (the substitution node) of a tree α by ‘identifying’ the root node of β with the substitution node (cf. Figure 20.7a). More generally, if the nodes are labelled by feature structures, the two feature structures must be compatible, and the node of the resulting tree is labelled by the unification of the two feature structures. Each non-terminal leaf node in an elementary tree is a substitution node and must obligatorily be filled by substitution or by the substitution part of wrapping substitution (see 20.3.3). Substitution is the main mode of composition for expanding argument nodes by the syntactic representations of specific argument realizations. In RRG, periphery elements and operators are only indirectly linked to the layered structure, by a PERIPHERY edge or by an edge to their layer in the operator projection. We include them in the layered constituent structure while making sure the information about whether they are operator or periphery elements is preserved. More concretely, the directed edge with label PERIPHERY is replaced by an unlabelled immediate dominance edge between the target layer and the periphery element and a feature [peri þ] on the root of the peripheral structure, often abbreviated as a subscript peri on the node category. In the case of operators, they also attach as a daughter (a)

(b)

Figure 20.7 Schematic sketch of simple substitution (a) and sister adjunction (b)

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

747

to the target node in the constituent tree. Their category (OP) indicates that they are operators and their contribution is indicated within further features attached to the OP node. Their elementary trees are rooted by a node that merges with the target tree. An advantage of this is that one can formulate constraints by means of feature specifications. For instance, one can require that there is only one definiteness operator in an RP. We will see how to use features for this purpose in more detail in Section 20.5. Peripheral structures cannot be added by substitution since they do not attach to leaves but to internal nodes, in general, and the same holds for the operators; cf. Figure 20.5. The mode of composition proposed for these cases is (sister) adjunction (see also Kallmeyer et al. 2013).4 As with substitution, we assume that the templates available for adjunction have a root label which coincides with the label of the target node (cf. Figure 20.5). For convenience, the root of an adjunction tree is marked by an asterisk in the graphical presentations. We call elementary trees with this marking adjunct trees. A further constraint on adjunct trees is that their root node has only a single daughter. The idea behind this is, as explained earlier, that the tree below the root is the actual periphery element while the root node captures more or less the PERIPHERY edge from RRG textbooks. The root label of an adjunction tree specifies the attachment site at the phrasal skeleton. In Figure 20.8, for example, the adverbial completely adjoins at the nucleus while yesterday adjoins at the core.

CLAUSE CORE RP

RP

NUC

Kim

NUC∗

V

MPperi

finished

CORE∗

RP

MPperi RP yesterday the cake

completely CLAUSE CORE RP Kim

RP

NUC MPperi

Vpred

completely finished Figure 20.8 Sister adjunction of periphery elements at different layers

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

MPperi

the cake yesterday

748

LAURA KALLMEYER AND RAINER OSSWALD

Sister adjunction is defined in such a way that the target node must not be a leaf node. Because of their single-daughter property, roots of adjunct trees are excluded as target nodes for sister adjunction, but it is of course possible to adjoin more than one adjunct tree to the same node of the target tree. The adjunction operation consists in merging the root of the adjunct tree with the target node (which amounts to unifying their feature structures) and adding the daughter of the adjunct tree root as a new daughter to the target node. This can take place in any position among the already existing daughters of that target node.5 See Figure 20.7b for a general schematic illustration of sister adjunction. A complication arises when an operator or periphery element targeting a specific layer occurs between elements that are part of a different layer. Examples are given in (4): (4)

a. He ate the apple completely. b. John did not eat the apple.

In (4a), the nucleus, which consists of the verb ate, is followed by an argument, which is part of the core, and after that comes an adverb that is in the nuclear periphery. In (4b), the tense operator did is placed between an argument (part of the core) and the verb (also part of the core). But it should attach at the clausal level. For the moment, we will ignore this complication and come back to it in Sections 20.5 and 20.7.

20.3.3 Wrapping Substitution Control constructions and extraction from complements pose a problem for the modes of composition presented so far. Consider the examples of whextraction in (5). (5)

a. What does John think Kim smashed? b. What does John think Mary claimed Kim smashed?

Clearly, it would not be appropriate to assume a separate complex template for each of these constructions. The syntactic representations are to be composed of basic argument structure templates in a systematic way. There are several options for achieving this goal, depending on the presumed inventory of elementary templates. First we need to decide on the proper syntactic representations of the examples in (5). While sentences of this type are discussed in the context of island constraints in Van Valin and LaPolla (1997: 615) and Van Valin (2005: 273), no structural analysis is provided there. Due to the nature of the embedding constructions, the basic binary pre-core slot pattern [CLAUSE[PrCS . . .][CORE . . .]] shown in Figure 20.1 does not apply to the present case. Figure 20.9 shows two possible alternatives. The analysis in Figure 20.9a assumes a pre-core variant of the clausal subordination pattern [CLAUSE [CORE . . .][CLAUSE . . .]]. The

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

749

Formalization of RRG Syntax (a)

(b)

SENTENCE

SENTENCE

CLAUSE

CLAUSE

PrCS

CORE

CLAUSE

RP

RP NUC

CORE

PrCS

CLAUSE

RP

RP

CORE

CLAUSE

RP NUC

CORE

NUC

RP

what does John think Kim smashed

NUC

what does John think Kim smashed

Figure 20.9 Two possible syntactic representations of wh-extraction from complements

(a)

(b)

SENTENCE

SENTENCE

CLAUSE

CLAUSE CLAUSE

CLAUSE PrCS RP

CORE

PrCS

CLAUSE

RP NUC think

CORE

CLAUSE

CORE

CORE RP

CLAUSE

CLAUSE

think

CORE RP NUC smash

CLAUSE CORE

CLAUSE

RP NUC claim

Figure 20.10 Wh-extraction via simple substitution (a) and wrapping substitution (b)

structure in Figure 20.9b, by contrast, assumes an additional clause node, and the pre-core slot pattern is [CLAUSE[PrCS . . .][CLAUSE . . .]]. In the following, we restrict the discussion to the first option since we regard it as difficult to come up with an independent motivation for the additional clausal node in Figure 20.9b. Figure 20.10 sketches two ways of composing the syntactic representation of example (5b). Figure 20.10a employs substitution only, but at the price of assuming a special elementary template associated with think that has a precore slot in addition to its normal argument slots. Assuming such a template would raise the further problem of providing information about which of the arguments within the embedded clauses is referred to by the referent phrase in the pre-core slot. The templates in Figure 20.10b are more straightforward in this respect since they represent proper argument structure templates in that the pre-core slot is locally connected to the core from which the wh-word is extracted. The dashed edge between the CLAUSE node dominating the pre-core slot and the lower CLAUSE node stands for a dominance relation, that is, a (possibly empty) sequence of immediate

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

NUC smash

RP NUC

RP NUC claim

RP

CLAUSE

CLAUSE

CLAUSE

750

LAURA KALLMEYER AND RAINER OSSWALD

Figure 20.11 Wrapping substitution

dominance edges in the final derived tree. We call such an edge a d-edge. Under this analysis, the long-distance dependency comes about by the insertion of material at the d-edge, that is, between the pre-core slot and the corresponding core.6 More flexible templates and a more complicated tree composition mechanism are needed in this case. The clause node of the smash structure is split in two and its upper part is unified with the upper clause node of the think structure (similar to the adjunction mechanism introduced in the previous section), thereby keeping the pre-core slot on the left side of the structure; the lower part of the split node is substituted at the lower clause node of the claim structure. The mode of composition just described is referred to as wrapping substitution (Kallmeyer et al. 2013; Osswald and Kallmeyer 2018). Wrapping substitution has the general form depicted in Figure 20.11.7 By definition, this operation involves a tree with a d-edge. Such an edge stands for a dominance relation in the final derived tree, that is, it specifies a place in an elementary tree where additional nodes and edges from other elementary trees can be inserted.8 Wrapping consists basically of splitting the tree at the d-edge and wrapping it around a target tree. More specifically, with reference to Figure 20.11, the subtree β rooted at the lower node (labelled X) of the d-edge fills a substitution slot in the target tree α while the upper node of the d-edge (labelled Y) merges with the root of α. Concerning this upper node, all descendants to the left (resp. right) of the d-edge are located to the left (resp. right) of the target tree α in the resulting combined tree, and all nodes of γ dominating the upper node (labelled Y) of the d-edge also dominate the merged node after the wrapping. The example in Figure 20.10b is a case of wrapping with empty γ and δR, (i.e. a wrapping where the upper part only adds nodes to the left of the d-edge but neither to its right nor above).9 It is instructive to compare the different tree composition options of Figure 20.10 with respect to their applicability for linking, especially with respect to template selection. As already explained, the syntactic inventory does not provide a single template for the complex syntactic structure of (5b). Rather, the structure has to be composed from argument structure templates which in turn are selected by the chosen lexical entries. The composition in Figure 20.10a has the disadvantages that the template selected by think is not an argument structure template

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

but has to be stipulated for embedded wh-questions, and that an advanced mechanism for coreference is needed. The composition scheme in Figure 20.10b, by comparison, has the advantage that the pre-core slot can be immediately linked when the template is selected. Here, the underlying assumption is that the syntactic inventory provides argument structure templates for wh-fronting. Note that this assumption differs slightly from RRG’s standard understanding of syntactic templates, which would assume a reduced core template plus a pre-core slot template in this case (Van Valin 2005: 15). Elementary trees in LTAG, by comparison, are commonly assumed to have substitution nodes for all arguments of the lexical head, irrespective whether they are realized within the CORE or outside of it (cf. Section 20.3.4). A further decomposition of elementary trees can be expressed by means of tree descriptions in the meta-grammar (cf. Section 20.4.2).

20.3.4 Extended Domain of Locality As mentioned in Section 20.3.1, an important characteristic of the LTAG formalism is the extended domain of locality (EDL) of elementary trees (Joshi and Schabes 1997: 95f ), which means that elementary trees represent full argument projections and that they can have a complex constituent structure. As explained above, the present formalization of RRG builds on a similar assumption. This is particularly crucial in the case of long-distance dependencies across clausal complements, as in Figure 20.10b. Here, the wh-element in the pre-core slot originates from the same elementary tree as the verb smash that it depends on. In the final derived structure, the two are separated from each other by the intervening structures of claim and think. A possible alternative solution, avoiding this extended domain of locality and using only simple substitution, is to have the information about the whmarked argument percolate to the top of the tree. A constraint-based formalization of this percolation process could roughly work as follows. The core node of the reduced template for smash carries a (set-valued) feature which contains the referential index of the participant not locally realized as well as its wh-marking. A general constraint then ensures that clause and sentence nodes collect the non-realized indices of their subordinate clauses and cores minus the indices that are realized in pre-core slot daughters and the like. This way of bookkeeping for modelling long-distance dependencies is in fact closely related to the use of ‘slash’ or ‘gap’ features in the approaches of Sag and Wasow (1999) and Ginzburg and Sag (2001), among others. To sum up, the approach just discussed can get along with simple substitution at the price of a considerable amount of bookkeeping. By comparison, the approach exemplified in Figure 20.10b allows a fully local argument linking but requires a more complex method of tree

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

751

752

LAURA KALLMEYER AND RAINER OSSWALD

composition. The former strategy is employed in GPSG/HPSG-related frameworks while the latter strategy is characteristic of approaches in the line of LTAG (cf. Kroch 1987).

20.3.5

Formal and Computational Properties of Tree Wrapping Grammars The type of tree rewriting grammar formalism introduced above, with composition operations sister adjunction, substitution and wrapping substitution, is called Tree Wrapping Grammar (TWG, Kallmeyer et al. 2013; Kallmeyer 2016). Note that Kallmeyer et al. (2013) use a slightly different definition of wrapping substitution, which, however, gives rise to more binary trees. We use the TWG version presented in Kallmeyer (2016) and Osswald and Kallmeyer (2018). One can investigate the formal properties of TWGs independently of the concrete shape of elementary trees chosen for RRG. We will not go into detail here but restrict ourselves to pointing out the following (see Kallmeyer 2016). TWGs are more powerful than context-free grammars (CFGs), which is due to the wrapping substitution operation. The expressive power depends in particular on how many d-edges are allowed to stretch across a single node in the final derived tree. In our example in Figure 20.10b, for instance, we have one single d-edge stretching across the roots and cores of the claim and the think elementary trees (see the grey dashed edge in Figure 20.12, which is the d-edge from the elementary tree of smash). A TWG where the maximum number of d-edges stretching across a node is limited to some constant k is called a k-TWG, and Kallmeyer (2016) links k-TWG to simple Context-Free Tree Grammars (CFTG, Kanazawa 2016) of

CLAUSE PrCS

CLAUSE

RP

CORE RP NUC think

CLAUSE CORE RP NUC CLAUSE claim

CORE RP

NUC smash

Figure 20.12 D-edge in final derived tree for Figure 20.10b

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

rank k by showing that for every k-TWG, a simple CFTG of rank k can be constructed that generates the same language. Simple CFTGs of rank k are, in turn, equivalent to well-nested Linear Context-Free Rewriting Systems (LCFRS) of fan-out k þ 1. Consequently, k-TWGs are in particular mildly context-sensitive. The notion of mildly context-sensitive languages and formalisms was introduced in Joshi (1985) in an attempt to characterize the amount of context-sensitivity required for natural languages, where context-sensitivity is used in a formal language sense, in contrast to context-free languages and grammars. Roughly, a mildly context-sensitive grammar formalism is able to generate more than all context-free languages, can account for crossserial dependencies, is polynomially parsable,10 and has the constantgrowth property, which means that, if we order the sentences of a language according to their length, the length grows in a linear way. It is interesting that coming from RRG, a typologically motivated linguistic theory, and developing a tree rewriting formalization of it, we end up within a class of formalisms that seems to support Joshi’s hypothesis that natural languages are mildly context-sensitive.

20.3.6 Crossing Branches and Discontinuous Structures The formal syntactic framework introduced so far presupposes ordered trees (cf. Section 20.3.1). This means that every two nodes stand either in a dominance relation or a linear precedence relation, and if one node precedes another node then all descendants of the first node precede the descendants of the second node. It follows that crossing branches, which are a frequent phenomenon in RRG syntactic analyses, are not supported by the formalism. Crossing branches often occur with peripheral adjuncts at the nucleus or the clause, for example. Because of the single syntactic tree assumption of the present approach (cf. Section 20.2.2), operators are another potential source of crossing branches. The proposed solution in both cases is to attach the respective elements at a lower node of the phrasal skeleton, in order to avoid a crossing of branches in the tree, while keeping track of their scope by means of appropriate features. Details on how to do this will be given in Sections 20.5 and 20.7. Another reason for crossing branches can be discontinuities in the constituents in the clausal skeleton. Discontinuities in the nucleus may occur with predicates that are multi-word expressions (e.g. particle verbs) or complex predicates (nuclear cosubordinations). An example of the latter is given by the English resultative construction shown on the left of Figure 20.13. In such cases, we assume that the discontinuous node can be split into two components, both carrying the same identifier nuc-id as a feature (see the right of Figure 20.13), in order to signal that the two nodes are actually components of the same nucleus.11 Note that the two components would be part of the same elementary tree.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

753

754

LAURA KALLMEYER AND RAINER OSSWALD

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

NUC RP

NUC V[

John

NUC[ RP

+]

painted

RP

NUC AP[

the house

red

NUC V[

+]

John

NUC[

1]

RP

NUC AP[

+]

painted

1]

the house

+]

red

Figure 20.13 Discontinuous complex predicates

20.4

Specifying the Syntactic Inventory

20.4.1 The Structure of Syntactic Templates The syntactic templates are the basic building blocks of syntactic representations. Let us now turn to the question of how to specify the templates of the syntactic inventory in a systematic way. The discussion of the two approaches to wh-fronting in Section 20.3.3 shows that the question of which templates must be available in the inventory is not independent of the modes of composition employed. If extraction structures are fully linked locally, as in Figure 20.10b, then there is no need for a pre-core slot template to be attached separately. As mentioned in Section 20.3.1, the present formalization proposal of RRG syntax assumes that argument structure templates have slots for each of the arguments of the lexical anchor. We call this the full argument projection assumption.12 For example, the templates in Figure 20.14 represent (some of the) alternative realization patterns of transitive verbs in English. They would be used for the different forms of eat in (6) (in the respective order in which they occur in the figure). (6)

a. b. c. d. e. f.

The boys ate the entire cake. The cake was already eaten. The entire cake was eaten by the boys. What did the boys eat today? I planned to serve the cake that the boys just ate. I expected them to eat the cake.

Notice the two templates for the passive (b and c), one in which only the undergoer occurs as an argument, and a second template which includes the realization of the actor by a peripheral by-phrase.13 Standard RRG would probably regard the peripheral by-phrase as a separate template that can be adjoined to the passive core. In the context of the present framework, the

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

755

Formalization of RRG Syntax (a)

(b)

(c)

CLAUSE

CLAUSE

CLAUSE

CORE

CORE

CORE

RP NUC[VOICE act] RP

RP NUC[VOICE pass]

RP NUC[VOICE pass]

AUX V[PRED +]

V[PRED +]

PP [PERI +]

AUX V[PRED +]

P

RP

by

(d)

(e)

CLAUSE

PrCS

CORE

RP[WH +] RP NUC[VOICE act] V[PRED +]

(f)

NUC∗N CLAUSE[PERI +] PrCS

CORE

CORE NUC[VOICE act] RP V[PRED +]

RP[REL +] RP NUC[VOICE act] V[PRED +]

Figure 20.14 Basic transitive predication template for English with variants

question is whether the by-phrase should be added by (sister) adjunction like an adverbial. By listing the two realization patterns for the passive as elementary templates in Figure 20.14, we decide against adjunction in this case. The main disadvantage of modelling the by-phrase by adjunction would lie in the constraints that need to be imposed, since adding a peripheral by-phrase that encodes the actor is restricted to specific constructional circumstances. The optional peripheral by-phrase is part of the description of how passive voice is realized in English. Of course, a mere enumeration of the two passive templates in Figure 20.14 is not satisfactory for a theory of grammar. At some point, the theory should state explicitly that it is the addition of the by-phrase which relates the second template to the first. In our framework, this relation is encoded at the level of template specifications (cf. Section 20.4.2). Template (d) in Figure 20.14 represents an elementary pre-core slot argument structure template. This differs from the template system informally suggested in Van Valin (2005: 15), in which the structure in question is composed of a reduced core template and a pre-core slot clause fragment (cf. Figure 20.1b). Again, we propose that compositions of this type are best modelled at the level of template specifications. Template (e) is used when the predicate is the anchor of a restrictive relative clause where the relative pronoun is the object. In this case, the peripheral tree is an adjunct tree that is intended to be adjoined to the nucleus of the noun phrase that it modifies. The last template (f ) is the one for a subjectless infinitival core, which can be embedded as an argument of category CORE, as in (6f).

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

756

LAURA KALLMEYER AND RAINER OSSWALD

20.4.2 Template Specification by Tree Descriptions In the previous section, we raised the question of how to characterize the (universal and language-specific) syntactic templates in a systematic way. The templates proposed above as elementary are more complex than the fragmentary templates originally assumed in the RRG literature. The key advantage of the formalization approach described here is that the composition of syntactic representations can be reduced to the three modes of composition introduced in Section 20.3, namely simple substitution, sister adjunction and wrapping substitution. Furthermore, linking can be performed locally within elementary trees (cf. Section 20.8). However, this leaves us with the problem of how to describe in which way the templates are built from more elementary components, and how they are related to each other. As mentioned before, it would be rather unsatisfactory if we had to regard the two passive templates in Figure 20.14 as independent units without being able to make explicit the relation between them. The proposed solution is to treat syntactic templates as minimal models of tree descriptions. Relations between templates can then be captured by relating the respective descriptions. For instance, the specification of the passive template with by-PP consists of the specification of the simple passive template plus a specification of the by-PP and constraints on its position. The use of tree descriptions for specifying syntactic templates in a modular way is inspired by the meta-grammar approach of Crabbé and Duchier (2005), where a meta-grammar is basically a system of tree descriptions that defines the syntactic inventory as the set of the associated minimal models. Tree descriptions in this sense consist of dominance and precedence constraints as well as category and feature assignments. Consider the example specifications in Figure 20.15, which are depicted in tree-like core-spine

core-clause

pre-core-slot

pre-nuc-rp

post-nuc-rp

CORE

CLAUSE

CLAUSE

CORE

CORE

NUC

CORE

PrCS ≺ CORE

RP ≺ NUC

NUC ≺ RP

V[PRED +] clause-spine:= core-spine ∧ core-clause

base-transitive := clause-spine ∧ pre-nuc-rp ∧ post-nuc-rp

CLAUSE

CLAUSE

CORE

CORE

NUC

RP ≺ NUC ≺ RP

V[PRED +]

V[PRED +]

Figure 20.15 Example specifications of syntactic fragments

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

diagrams but are to be read as tree descriptions. For instance, the specification with the name pre-core-slot says that there are three distinct nodes n0, n1 and n2, labelled respectively by CLAUSE, PrCS and CORE, where n1 and n2 are daughters of n0 and n1 immediately precedes n2 (expressed by ≺). Figure 20.15 illustrates how the basic transitive template of Figure 20.14 can be defined by a piecewise combination of such specifications. The precore slot template of Figure 20.14 can likewise be defined by conjoining the specifications clause-spine, pre-core-slot, and pre-nuc-rp. In this way, common components of elementary templates are made accessible in the metagrammar.14 Furthermore, syntactic tree descriptions can be linked to descriptions of semantic representations (either predicate-logical formulas or frames), and the meta-grammar can include a constraint-based formulation of RRG’s linking algorithm (cf. Kallmeyer et al. 2016). The syntax– semantics interface will be briefly discussed in Section 20.8.

20.5

Formalizing the Operator Projection

20.5.1 Operators in RRG RRG links the representation of operators to the layered structure of the clause. Recall that the layered structure reflects the distinction between predicates, arguments and non-arguments. The core layer consists of the nucleus, which specifies the verb, and its arguments. The clause layer contains the core as well as extracted arguments. Operators are closed-class grammatical categories such as aspect, modality and tense. Each type of operator is assumed to attach to a specific layer: for instance, tense operators attach to the clause, modality to the core, aspect to the nucleus (see Table 20.1 for the mapping between operator types and corresponding layers). Moreover, the surface order of the operators reflects their attachment site in that the higher the layer an operator is attached to, the further Table 20.1 Operators in the layered structure of the clause (cf. Van Valin 2005: 9) Layer

Operators

Nucleus

Aspect (ASP ) Negation (NEG ) Directionals (event orientation)

Core

Directionals (participant orientation) Event quantification Modality (MOD ) (ability, permission, obligation, etc.) Negation (NEG )

Clause

Status (STA ) (epistemic modals, external negation) Tense (TNS ) Evidentials Illocutionary Force (IF )

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

757

758

LAURA KALLMEYER AND RAINER OSSWALD

away from the nucleus the operator occurs on the surface. The mapping from operators to levels of the layered structure explains (i) the scope behaviour of operators, since structurally higher operators take scope over lower ones, and (ii) surface order constraints for operators; higher operators are further away from the nucleus of the structure. The problem is that the constituent and the operator structure are not completely parallel; one can have structures where an operator belonging to a specific layer is, on the surface level, surrounded by elements belonging to a lower layer in the constituent structure. Examples of a clause-level tense operator occurring within the core are given in (7) and (8) (taken from Van Valin, 2005: 10) for English and Turkish, respectively. (7) (8)

[Mary enter-edTNS the room]CORE. [Gel- emiyebilirim]CORE. come- able.neg MOD - psbl STA - aor TNS - 1sg ‘I may be unable to come.’

In (8), the clause-level status and tense operators occur between the verb and the pronominal affix, which is part of the core. Even though the constituent structure and the operator structure are not fully aligned, they depend on each other. Their hierarchical order is the same and the existence of a layer in the operator projection requires that this layer also exists in the constituent structure. For instance, one can only have clause-level operators if a clause node exists in the constituent structure. While the ordering among the operators is thus systematically correlated with the scope given by their attachment site at the clausal skeleton, the surface order of the operators relative to arguments and adjuncts is much less transparent and would require crossing branches if everything were captured in a single tree with operators attaching where they take scope. For this reason, RRG usually represents the constituent structure and the operator structure as different projections of the clause. The syntactic representation for (7) on the left side of Figure 20.16 illustrates this idea: the upper part gives the constituent projection while the lower part gives the operator projection. If we integrate the operator projection into the clausal skeleton of the constituent structure, we obtain the tree with crossing branches shown on the right of Figure 20.16, due to the above-mentioned mismatches between the two projections.15 Besides crossing branches, additional complications arise from the following two facts: first, a functional element can contribute more than one operator and, second, an operator can be distributed over more than one element in the sentence. An example is (9) where the functional element will contributes both tense and illocutionary force (IF), while aspect (ASP) is jointly contributed by be and -ing. (9)

WillIF,TNS they beASP leav-ingASP?

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE

CLAUSE

CORE RP

NUC

V enter-ed

Mary

CORE

RP

NUC

the room

V

RP

RP V TNS

Mary enter -ed the room

NUC CORE CLAUSE

TNS

Figure 20.16 Constituent structure and operator projection for (7)

CLAUSE CORE RP

CLAUSE NUC V

will they

be

leav

-ing

IF TNS RP

V ASP

NUC

ASP

CORE TNS IF

CORE NUC ASP

will

they

be

V

ASP

leav -ing

CLAUSE CLAUSE

Figure 20.17 Constituent structure and operator projection for (9)

RRG assumes the structure on the left of Figure 20.17 (adapted from Van Valin 2005: 14). If the operator projection is integrated into the constituent structure, we obtain the graph on the right of Figure 20.17, which is not even a tree but only a directed acyclic graph since the node labelled will has two incoming edges. Note, however, that as long as nodes with more than one incoming edge arise only from contributing several operator categories to the same level of the layered structure, the graph can easily be turned into a tree by putting all these categories into a single node with a label (or, rather, a feature structure) that captures the fact that the element is an operator that contributes several categories to the same layer. The node dominating will in Figure 20.17, for instance, could have a category OP with features TNS ¼ fut and IF ¼ int. Cases with a single element contributing two operators that take scope at different layers exist as well. Examples are finite modal verbs as in (10) that provide modality at the core level and tense at the clause. (10)

John mightTNS,MOD win.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

759

760

LAURA KALLMEYER AND RAINER OSSWALD

CLAUSE CORE RP

CLAUSE NUC V

John

might

win V NUC

MOD TNS

CORE

CORE

RP

NUC TNS

John

MOD might

V win

CLAUSE

Figure 20.18 Constituent structure and operator projection for (10)

The corresponding constituent structure and operator projection in RRG is shown on the left of Figure 20.18, and on the right we see the graph that one would obtain when integrating the operator projection into the constituent tree. This example shows that if we integrate the operator projection into the constituent tree while assuming that each operator has a separate node that is the daughter of its syntactic layer, we obtain directed acyclic graphs instead of trees, which might come with considerable complications concerning the formal properties of such a framework. In particular, it is clear that such structures cannot be generated with the TWGs introduced in the previous section. If we choose to disentangle the operator projection to a certain extent from the constituent structure, we have two options. Either we can generate the operator projection tree separated from but to a certain extent in parallel with the constituent structure. This is what Johnson (1987) proposed in his formalization, which uses two different context-free grammars, one for analysing the sequence consisting of the verb plus arguments, and one for the sequence consisting of the verb plus operators (cf. Section 20.2.1). Each takes those parts of the sentence as input (or yield) that corresponds to its set of terminals. This approach makes sure that verb and arguments, on the one hand, and verb and operators, on the other, appear in the right order, independent from each other. But there are some shortcomings: First, the formalization does not guarantee that each layer targeted in the operator projection is actually present in the constituent structure. Secondly, a problem remains with elements that contribute more than one operator, possibly to more than one layer, since the different projection grammars are each assumed to generate trees.

20.5.2 A Feature-Based Implementation of the Operator Projection In the following, we will present the proposal of Kallmeyer and Osswald (2017), which integrates operators according to their surface position into the constituent structure, thereby avoiding crossing branches, while

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

keeping track of the operator projection within the feature structures. The classification of an element as an operator and the contribution of that element to the various layers is captured within the category of its node. More specifically, an operator element has the category label OP and a feature structure that specifies its contribution to the CLAUSE (feature cl), the CORE (feature co) and the NUC (feature nuc) layer. The tree structures corresponding to the sentences (9) and (10) are shown in Figure 20.19. Note that the operator scope information is now entirely captured within the feature structure at the OP node and does not depend on the category of the node to which the OP attaches. In order to obtain trees of the type given in Figure 20.19, we use sister adjunction for adding operators. This leads to spurious ambiguities since operators could in principle adjoin to any of the three layers that they attach to, provided this does not yield crossing branches. In the first tree in Figure 20.19 for instance, we could also attach the clausal operator will lower, at the CORE node, and be could attach higher, also at the CORE node. In the following we assume that among the different combinations of attachment sites that are possible, the preferred one is the one where operators are placed as close as possible to the layer that corresponds to their scope (if it scopes at several layers, the lowest is considered relevant). Aside from this spurious ambiguity issue, if we adjoin operators in an unrestricted way, it leads to undesired overgeneration since we can leave

CLAUSE OP

CL

IF int TENSE fut

will

CORE RP they OP

NUC NUC ASP prog

be

V

OP

leav

NUC ASP prog

-ing

CLAUSE CORE RP John

OP

CL [TENSE pres] CO [MOD epistemic]

might

NUC V win

Figure 20.19 Encoding the operator structure in node features for (9) and (10)

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

761

762

LAURA KALLMEYER AND RAINER OSSWALD

out operators that are obligatory, adjoin several operators of the same type even in cases where this is ungrammatical, and we can generate ungrammatical linear orders of operators, for instance the ones in (11). a. *They be will leaving? b. *John not might win.

(11)

Kallmeyer and Osswald (2017) introduce edge features in order to express the necessary constraints on adjunction. The basic idea is that each edge (or, under a different but equivalent view, each node) should be able to pass information on elements already attached and elements still required to the left and to the right. Once the derivation is completed, the edge information of sister nodes must be compatible with each other. To this end, each node (or, as Kallmeyer and Osswald (2017) put it, each edge) has a left and a right feature structure that is not part of the proper feature structure of the node but interacts with the left and right feature structures of neighbouring nodes. They are graphically depicted in shaded boxes to the left and the right of the node in question (see Figure 20.20). (The feature structure of the node itself will often be omitted or, if needed, given in the middle.) Note that the features nuc, co and cl mentioned above and exemplified in Figure 20.19 are not part of the edge features but are proper node features. Edge features are less about the actual content of nodes but rather about requirements and information that need to be passed around in order to constrain syntactic composition. Let us illustrate this approach by using it to characterize a certain operator as obligatory. In Figure 20.20, we have a verb without tense marking. Therefore, to the left of the NUC node, we signal [tns ]. The features we are using here are mostly binary features (i.e. values are only þ or ), and their intended meaning is that a value  expresses the absence of something while a value þ expresses the presence of or the requirement (presence in the derived tree) for something. The feature [tns þ] on the right

CLAUSE PrCS

TNS +

CORE

RPwh what

TNS 3

CLAUSE∗

TNS +

OPtns

RP

TNS 3

TNS −

the girl

TNS −

does Figure 20.20 Obligatory adjunction of a tense operator

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

NUC V say

Formalization of RRG Syntax

2

• 1 [: : : ]

1

[: : : ] •

unification between neighbouring edges

2

[: : : ] •

•

3

[: : : ] •

[: : : ] •

feature percolation on the left fringe

3

[: : : ]

feature percolation on the right fringe

Figure 20.21 Final feature unifications

of the PrCS signals that to the right of this node, a tense operator has to be present. If no tense operator adjoins, the absence and the requirement of the presence will eventually unify (we will explain later why) and this will lead to a unification failure. If, however, a tense operator adjoins as in Figure 20.20, the requirement is satisfied. The features to the left and to the right of the tense operator tell us that to its right, no other tense operator exists while to its left one can pass the information that there is now a tense operator. (OPtns is short for category OP with feature structure [cl [tense . . .]].) Let us explain now exactly how edge features work (see Kallmeyer and Osswald 2017), in particular how they percolate through the tree (cf. Figure 20.21). As mentioned, nodes can have special left and right feature structures. In the final derived tree, the left feature structure of a node ν unifies with the right feature structure of its immediate sister to the left. Furthermore, the left feature structure of a node ν that does not have a sister to the left unifies with the left feature structure of the mother of ν, provided this mother is not the root node of an elementary tree or the lower node of a d-edge. Similarly, the right feature structure of a node ν that does not have a sister to the right unifies with the right feature structure of the mother of ν, again provided this node is not the root node of an elementary tree or the lower node of a d-edge. In our example, once we have performed the adjunction, we obtain the tree at the top of Figure 20.22, and after the final edge feature unifications, the result is the tree at the bottom. Such edge features can be used to require certain adjunctions and, as is the case in this example, to require them exactly once. The adjunction of the operators needs to be controlled in the following two respects: (i) Adjoining an operator is obligatory if the information conveyed by the operator is required for a sentence to be complete. (ii) The scope-related ordering of the operators must be respected. In our approach, these constraints are both implemented with the help of edge feature structures; we have just seen an example of type (i). We will now introduce features that guarantee (ii) as well. Concretely, we want to achieve that, from the nuclear predicate outwards, one encounters first the NUC operators, then the CORE operators and then the CLAUSE operators. To this end, we assume an edge feature ops (for operator structure) that specifies which operator projection layer(s) have

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

763

764

LAURA KALLMEYER AND RAINER OSSWALD

before final unifications: CLAUSE PrCS

TNS +

TNS +

OPtns

TNS −

does

RPwh

CORE RP

TNS 3

what

TNS −

TNS 3

the girl

NUC V say

after final unifications: CLAUSE PrCS

TNS +

TNS +

RPwh

OPtns does

TNS −

TNS −

TNS −

what

RP

CORE

TNS −

the girl

TNS −

NUC V say

Figure 20.22 Obligatory adjunction of a tense operator: derived tree

been reached so far. Its value is a feature structure with features cl_l, co_l and nuc_l for the three layers, each with possible values þ or . These features are used in such a way as to guarantee that nuclear, core and clausal operators have to appear in this order when moving outwards in the sentence, starting from the nuclear predicate. Figure 20.23 illustrates this mechanism by applying it to the operator structure of the sentence in (12). (12)

John might not win.

For instance, a core operator such as not that adjoins to the left of the predicate has a requirement to the right that the level cl_l have a value  (i.e. has not been reached yet). To the left, it just gives the information co_l þ, which has the effect of disallowing nuclear operators to appear here. The operator might, which contributes not only tense at the clausal level but also modality at the core level, comes with the same requirements to its right. But, in contrast to not, it states to the left of the edge that now cl_l has value þ (as well as co_l). An inverse order (John not might win) is therefore excluded. To the left of might, there cannot be any further core (or nuclear) operators. The derived tree for sentence (10) is shown in Figure 20.24. Recall that in addition to the ops feature, nodes of category OP have features nuc, co and cl that indicate the contribution of the operator. For example, since might contributes tense at the clause level and epistemic modality at the core level, its OP feature structure is [cl[tense pres], co[mod epistemic]] (cf. Figure 20.19). (We left this out of Figures 20.23 and 20.24 for

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE TNS +

TNS 3 OPS 4

RP

CORE

TNS 3 OPS 4

NUC ⎡

John CORE ⎡

⎤

TNS + CL_L + ⎦ OPS CO_L +

⎣

∗

OPmod,tns

V

⎤

TNS − ⎡ ⎤ ⎢ NUC_L − ⎥ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ CL_L −

win

TNS − OPS CL_L −

might

⎡

⎤

TNS − ⎡ ⎤ ⎢ NUC_L − ⎥ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ CL_L −

CORE∗

TNS − OPS [CO_L +]

OPneg

TNS − OPS CL_L −

not Figure 20.23 Keeping track of the operator projection in edge features

the sake of readability.) Since the features of the OP nodes specify which operator projection layer(s) the operator belongs to, we can deterministically map a derived tree to the standard RRG structure in which the constituent structure and the operator projection are separated.

20.6

The Construction of Complex Sentences

20.6.1 Coordinate, Subordinate and Cosubordinate Constructions A crucial assumption of RRG concerning the structure of complex sentences is the distinction between embedded and non-embedded dependent structures. Embedded dependent structures correspond to subordinations. By contrast, non-embedded dependent structures, which are referred to as cosubordination structures, have the basic form [[ ]X [ ]X]X. It is characteristic of this type of construction that operators that apply to category X are realized only once but have scope over both constituents. Cosubordination differs from the coordination of two independent structures in that the latter type of construction has the form [[ ]X [ ]X]Y, where Y is a category one level above X in the layered structure. The general schemas for non-subordinate structures are shown in

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

765

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

CLAUSE TNS +

⎡ TNS 3 OPS 4

RP

TNS 3 OPS 4

CORE

⎤

TNS + CL_L + ⎦ OPS CO_L +

⎣

OPmod,tns

TNS − OPS CL_L −

TNS − OPS [CO_L +]

OPneg

TNS − OPS CL_L −

NUC

⎡

John

⎤ ⎡ ⎤ TNS − TNS − ⎤ ⎤ ⎡ ⎡ ⎢ ⎥ ⎢ ⎥ − − NUC_L NUC_L ⎢ ⎥ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦ CL_L − CL_L −

not

might

CLAUSE ⎡

⎡

⎤

TNS + ⎣ CL_L + ⎦ OPS CO_L +

⎡

RP

⎤

TNS + ⎣ CL_L + ⎦ OPS CO_L +

⎡

⎤

TNS + ⎣ CL_L + ⎦ OPS CO_L +

CORE

⎤ TNS − ⎤ ⎡ ⎥ ⎢ CL_L − ⎥ ⎢ ⎣OPS ⎣CO_L − ⎦⎦ NUC_L −

⎡

⎤

⎡

win

⎡

⎤

TNS + ⎣ CL_L + ⎦ OPS CO_L +

TNS − CL_L − ⎦ OPmod,tns ⎣ OPS CO_L +

⎤

TNS − ⎣ CL_L − ⎦ OPS CO_L +

⎡

⎤ TNS − ⎡ ⎤ ⎢ NUC_L − ⎥ ⎥ OPneg ⎢ ⎣OPS ⎣CO_L − ⎦⎦ CL_L −

⎡

⎤ TNS − ⎡ ⎤ ⎢ NUC_L − ⎥ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ NUC CL_L −

⎡

John

might

not

⎤ TNS − ⎡ ⎤ ⎢ NUC_L − ⎥ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ CL_L −

⎤ ⎡ ⎤ TNS − TNS − ⎡ ⎡ ⎤ ⎤ ⎢ ⎢ ⎥ NUC_L − ⎥ NUC_L − ⎥ ⎢ ⎢ ⎥ ⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦ CL_L − CL_L −

win

Figure 20.24 Derived tree for (10) before (top) and after (bottom) final edge feature unification

⎡

Formalization of RRG Syntax

X X

X

767

Y

… …

X

X

X

… …

X

Figure 20.25 RRG template schemas for non-subordinate nexus types

sister adjunction CLAUSE CORE*

CORE*

CORE

CORE

CORE

CORE

NUC

NUC

NUC

CLAUSE CORE

CORE

CORE CORE NUC

CORE

CORE

CORE

NUC

CLAUSE

CORE

CORE

CORE

CORE

NUC

NUC

NUC

CORE NUC

wrapping substitution Figure 20.26 Two ways of compositionally deriving cosubordination constructions

Figure 20.25 (cf. Van Valin and LaPolla 1997: 507 and Van Valin 2005: 224). However, these structures cannot serve as elementary trees in the sense introduced in Section 20.3 since there are infinitely many of them; that is, they have to be derived by operations in the syntax. Among the tree composition operations described in Section 20.3, both sister adjunction and wrapping substitution allow the derivation of cosubordinate structures, as sketched in Figure 20.26 for the case of core cosubordination. The following two examples of multi-verb constructions show possible use cases for the two kinds of derivations. Both examples describe transitive motion scenarios. The example in (13), taken from Ullrich (2011), illustrates a common pattern for simultaneous event constructions in Lakhota (Siouan) (cf. Ullrich 2018; Osswald and Van Valin 2022). (13)

Yu-slóhaŋ a-wíˇ cha-Ø-ye. by.pulling-slide am-3pl.ug.anim-3sg.ac-go ‘She was dragging them away.’

In Lakhota, arguments are marked by pronominal affixes on the main verb at the end of the clause. The simultaneous event construction in (13) contains a second verb that precedes the main verb and is dependent in that it is morphophonologically reduced and does not carry any pronominal

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

768

LAURA KALLMEYER AND RAINER OSSWALD

CLAUSE CORE[COSUB +]∗ COSUB +

CORE

COSUB +

CORE[COSUB +]

COSUB −

CORE

NUC V

NUC V Ø

Figure 20.27 Core cosubordination via sister adjunction: analysis of (13)

markings. The construction has the properties of a core cosubordination (Ullrich 2018). Since there are no shared arguments to be taken care of syntactically, due to the head marking, and because the construction has the same-subject/same-object constraint, the cosubordinate structure can be derived most naturally by sister adjunction, as illustrated in Figure 20.27. Under this analysis, the main verb selects an elementary tree that already has two core nodes and requires the adjunction of an additional core daughter. This requirement can be captured via a binary edge feature cosub which signals that a structure is a cosubordination structure. In addition, the higher CORE node of a CORE cosubordination can have a feature [cosub þ] while the lower ones are marked [cosub ], which ensures that the adjunction takes place at the upper CORE node. Note that these node features do not interact with the edge features. The Japanese example in (14) (from Croft et al. 2010: 219) shows the use of the te- construction for combining cause, manner and direction of motion. (14)

Watashi wa taru o korogashi-te chikashitsu ni ire-ta. 1sg top barrel acc roll(caus)-te basement loc take.into-pst ‘I rolled the barrel into the basement.’

Assuming that (14) is also an instance of a core cosubordination construction, the derivation seems best described as an application of wrapping substitution (see Figure 20.28). Under this analysis, the first two arguments are introduced by the first verb and are shared by the second verb. The elementary tree chosen for the second verb has, therefore, a reduced number of arguments. Its remaining arguments are ‘controlled’ by the elementary tree of the te -marked first verb, like in a control construction. Multiply embedded control and matrix-coding constructions pose a number of challenges for RRG’s syntactic analysis as presented in Van Valin and LaPolla (1997) and Van Valin (2005). Consider the examples in (15) and their syntactic representations in Figure 20.29.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

769

Formalization of RRG Syntax

(a)

CLAUSE CORE CORE RP

RP

CORE

NUC

watashi wa taru o

RP

korogashi-te

NUC

chikashitsu ni

ire-ta

(b) CLAUSE

RP

CORE [COSUB +]

CORE [COSUB +]

CORE

CORE

RP

CORE

NUC

RP

NUC

korogashi-te

ire-

Figure 20.28 Core cosubordination analysis of (14) derived by wrapping substitution

SENTENCE CLAUSE CORE RP

NUC

CORE RP

CORE

LM NUC RP

LM NUC

RP SENTENCE

Mary expected John

to

ask Kim

to

clean the floor CLAUSE CORE CORE

SENTENCE

CORE

CLAUSE

RP

NUC

CORE LM

NUC

RP

LM NUC

RP

CORE CORE

RP NUC RP John told Kim

John tried CORE

CORE

LM NUC

LM NUC

to

try

to

to

persuade Kim

to

RP

clean the floor

Figure 20.29 Syntactic representations of the examples in (15)

(15)

a. Mary expected John to ask Kim to clean the floor. b. John tried to persuade Kim to clean the floor. c. John told Kim to try to clean the floor.

Structures like the ones in Figure 20.29 show a mismatch between syntax and semantics in the following sense: semantically, the infinitival

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

clean the floor

770

(a)

LAURA KALLMEYER AND RAINER OSSWALD

CLAUSE

CLAUSE CORE

CORE

CORE

CLAUSE CORE

NUC RP

RP NUC RP

CORE

NUC RP

(to) ask

expected

CORE

CORE

RP NUC RP NUC RP NUC RP

(to) clean

(b)

CLAUSE

CLAUSE CLAUSE

CORE CORE

CORE

CORE

CORE

CORE CORE

CORE

NUC

NUC

RP

NUC RP

RP NUC

CORE

CORE

RP

(to) persuade

tried

Figure 20.30 Composition of templates by wrapping substitution for (15a) and (15b)

CLAUSE CORE

CLAUSE

CLAUSE CORE

RP NUC RP expected

PrCS

CORE

RP

NUC (to) ask

CORE

CORE NUC RP

PrCS

CORE

CORE CORE

RP RP NUC RP NUC

NUC RP

(to) clean

Figure 20.31 Composition of the syntactic representation of (16a)

complements are arguments of the respective control predicates, but syntactically, they do not behave like core arguments. Example (15a) is instructive in showing how multiply embedded constructions can give rise to coordinating syntactic configurations. Simple substitution cannot be applied here if we assume that the verbs expect and ask select elementary templates that contain a core daughter of the clause for the infinitival complement. Figure 20.30a sketches a possible solution that uses wrapping substitution, where split nodes carry differing top and bottom categories. Figure 20.30b shows that the same compositional mechanism works for control verbs such as try. which call for a core cosubordination template. Moreover, the treatment of wh-fronting introduced above straightforwardly extends to embedded control and matrix-coding constructions like those in (16). (16)

a. Whom did Mary expect John to ask to clean the floor? b. What did Mary expect John to ask Kim to clean?

For example, the syntactic representation of (16a) can be composed of elementary argument structure templates as illustrated in Figure 20.31. Note that clausal complements as in (17a) also show a mismatch between syntax and semantics in that they do not attach to the core but to the clause.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

(a) CLAUSE CORE

CLAUSE

NUC

RP

John thinks

(b)

CLAUSE CORE RP

Kim smashed the vase

CLAUSE

CLAUSE

PrCS

CLAUSE

OPtns

RPwh

CORE

does

what

RP

RP

NUC

NUC

CORE RP

CLAUSE

NUC

John think

Kim smashed Figure 20.32 Subordination via simple and wrapping substitution for (17a) and (17b), respectively

But in contrast to the cosubordination cases just discussed, multiple embeddings of clausal complements correspond to embeddings on the syntactic side, though at the clause level and not at the core level. For this reason, simple substitution is sufficient for clausal complementation (cf. Figure 20.32a). (17)

a. John thinks (that) Kim smashed the vase. b. What does John think (that) Kim smashed?

However, as already discussed in Section 20.3.3, cases of extraction out of subordinated constituents, like in (17b), require wrapping substitution (cf. Figure 20.32b).

20.6.2 Operators in Complex Sentences Operators in cosubordination constructions that appear in one of the components and that target the cosubordination layer have scope over all components. In the following, we will illustrate how this characteristic property of cosubordination can be formally achieved, going through one example involving sister adjunction and one example involving substitution. The Turkish sentence in (18) (taken from Van Valin, 2005: 201) is an example of a core cosubordination construction (see also Bohnemeyer and Van Valin 2017: 155f ). On the surface, the deontic modal operator -meli (‘should, ought to’) is embedded in the second core, but it takes scope over the entire complex core.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

771

772

LAURA KALLMEYER AND RAINER OSSWALD

CLAUSE CORE[MOD deont] CORE[MOD deont]

CORE[MOD deont]

NUC LM

NUC OP[CO [MOD deont]] PRO

V gid

ip

V

meli

yiz

gör

Figure 20.33 Derived tree for (18)

(18) [[Gid-ip]CORE [gör-meli-yiz]CORE]CORE. go-lm see-mod-1pl ‘We ought to go and see.’

Let us assume that the first core is added to the second core by sister adjunction, similar to the analysis of the Lakhota construction in Figure 20.27. The modal operator in (18) adjoins to the second embedded core node and carries a feature indicating that it is a core operator. The result should be the derived structure in Figure 20.33 with the mod feature shared between all the three CORE nodes involved (cf. Van Valin, 2005: 204).16 In a cosubordination, an operator embedded in one part of the complex structure generally takes scope over the larger category. Accordingly, in all elementary trees for cosubordination configurations, the relevant features (here mod) are shared between the lower and the higher category in question (here the two CORE nodes). This is taken to be a general property of cosubordination structures. Corresponding to this, we assume that when mapping our derived structure to the standard RRG structure, the operator targets the highest corresponding node, as long as there is no higher operator level and no substitution node in between. In the case of Figure 20.33, this is the core of the entire sentence. Figure 20.34 gives the derivation of the tree in Figure 20.33, attaching the tree for gid ip by sister adjunction. Note that this implies that there is a special tree for cores such as gör yiz in this example, providing two CORE nodes and requiring an adjunction at the higher one in order to be completed to a cosubordination structure. This requirement is again expressed by an edge feature cosub. We can adjoin several cores such as gid ip but we have to adjoin at least one in order to switch the feature from  to þ. Besides the edge feature, we also assume (as before, see Figure 20.27) a boolean node feature cosub that expresses whether a node roots a cosubordination structure or not (see also the CORE nodes in Figure 20.34). This enforces an adjunction of the gid tree at the cosubordination CORE node and not at the lower CORE node.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE

COSUB +

CORE

COSUB + MOD 1

CORE

COSUB − MOD 1

∗

NUC LM V

ip

gid

COSUB +

CORE

COSUB + MOD 2

COSUB −

CORE

COSUB − MOD 2

NUC

PRO

V

yiz

gör

CORE

COSUB − MOD deont

∗

OP[CO [MOD deont]] meli Figure 20.34 Derivation for (18)

A CORE cosubordination example that involves wrapping substitution instead of sister adjunction is the English example in (19a) (cf. Van Valin 2005: 203), where we have a core consisting of three embedded core constituents where the first contains the modal operator must. This operator takes scope over the entire large core. By contrast, in (19b) we have a structure consisting of two cores which constitute a clause. That is, we have a core coordination and not a core cosubordination. In this case, the modal embedded in the first core scopes only over this one and not over both cores. (19)

a. [[Kim mustMOD go]CORE [to try]CORE [to wash the car]CORE]CORE b. [[Kim mustMOD ask Pat]CORE [to wash the car]CORE]CLAUSE

Concerning the analysis of (19a), we propose that the different cores are combined with each other via wrapping substitution and not via sister adjunction. The difference, compared to (18) is that the verbs go and try both syntactically select for an infinitival complement clause (i.e. for a CORE argument), which should be expressed via a corresponding substitution node. The derivation is shown in Figure 20.35. The node features contributed by CORE operators (here [mod deont]) are shared between the different CORE nodes that are part of the cosubordination construction. An edge feature for enforcing the adjunction of further CORE sisters in a cosubordination is not necessary since the substitution nodes guarantee that CORE arguments have to be added. The shared operator scope in (19a) is a standard criterion for distinguishing cosubordinate from coordinate constructions. Another diagnostic is the

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

773

774

LAURA KALLMEYER AND RAINER OSSWALD

COSUB + MOD 1

CORE

CORE

COSUB − MOD 1

COSUB − MOD 1

CORE

CLAUSE RP

NUC CORE

Kim

V go

CORE

COSUB − MOD deont

CORE

COSUB − MOD 2

COSUB + MOD 2

CORE

COSUB − MOD 2

∗

CORE

COSUB + MOD 3

CORE

COSUB − MOD 3

to try

OP[CO [MOD deont]]

to wash the car

must Figure 20.35 Derivation for (19a)

independent accessibility of the embedded cores by time-positional adverbials, which are analysed as core-level modifiers (cf. Bohnemeyer and Van Valin 2017). While (19b) does allow independent time-positional modification, as in (20a), this is not an option for (19a): both (20b) and (20c) are excluded. (20)

a. Kim must ask Pat now to wash the car tomorrow b. #Kim must go now to try to wash the car tomorrow c. #Kim must go to try now to wash the car tomorrow

As to operators in subordinated CORE or CLAUSE arguments, since these arguments are added by substitution, their root nodes block edge feature percolation and we can have different operators within the argument and outside. In (21) for instance (adapted from Van Valin 2005: 200), the tense operator will is part of the argument CLAUSE added by substitution, and it does not percolate upwards to the matrix clause. (21)

Kim told Pat that she will arrive late.

The substitution of the clausal argument into the tree anchored by told is shown in Figure 20.36. The operator will in the embedded clause contributes tense at the clausal layer. The corresponding feature [cl [tense fut]] is shared along the clausal skeleton of the embedded sentence, via the ½op 2  features. But it is not transported into the embedding clause, which can have a different tense feature. The edge feature percolation also stops at the substitution node, which means that the tense in the embedded clause cannot satisfy a tense requirement in the embedding clause.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE[OP [CL [TENSE past]] CORE[OP [CL [TENSE past]] RP Kim

CLAUSE[OP 2 ] LM

CORE[OP 2 ]

NUC[OP [CL [TENSE past]] RP

that RP

NUC[OP 2 ]

MPperi

Pat

she

arrive

late

told

CLAUSE

CORE[OP 1 ]∗ ⎡

⎤

TNS + CL_L + ⎦ OPS CO_L +

⎣

OP 1 [CL [TENSE fut]]

TNS −

will Figure 20.36 Derivation for (21)

20.7

Modification and Periphery

Adjuncts in RRG are part of the periphery, which can be seen as a structure similar to the operator projection since each adjunct targets a specific layer and, starting from the nucleus and moving outside, the nuclear adjuncts have to precede the core adjuncts, which in turn have to precede the clausal adjuncts. Kallmeyer and Osswald (2017) briefly mention that this can be modelled in a way similar to the treatment of operators explained above. In the following, we will illustrate with an example how to capture the ordering constraints for periphery elements via edge features in a way analogous to the corresponding word-order constraints for operators. We will restrict ourselves to adverbs but other modifiers would be treated in a similar way. According to Van Valin (2005: 41), adverbs may ‘modify all three layers of the clause; aspectual adverbs like completely and continuously modify the nucleus, pace adverbs like quickly and manner adverbs like carefully modify the core, and epistemic adverbs like probably and evidential adverbs like evidently modify the clause’. Depending on their position, manner adverbs can actually be clausal or core modifiers, see (22), from Van Valin (2005). (22)

775

a. Ruth cleverly hid the cash. b. Ruth hid the cash cleverly. c. Cleverly, Ruth hid the cash.

(ambiguous) (core modifier) (clause modifier)

The core modifier reading signifies that Ruth hid the cash in a clever way while the clausal modification has the meaning that it was clever of Ruth to hide the cash. (22a) is ambiguous between the two readings while (22b)

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

776

LAURA KALLMEYER AND RAINER OSSWALD

seems to have only the ‘in a clever way’ reading and (22c) only the widescope reading (‘it was clever to hide the cash’). As to linear precedence, the main difference between modifiers and operators is that modifiers can often be placed to the left or to the right of the nucleus (see (22)). Moreover, modifiers differ from operators regarding constraints on adjunction: they are rarely obligatory and there can be multiple modifiers targeting the same layer, as in (23). If multiple modification is not possible, this is due to semantic or pragmatic ill-formedness (e.g. conflicting aspectual information). (23)

a. Ruth carefully hid the cash slowly. b. Evidently, Ruth possibly hid the cash.

(several core modifiers) (several clause modifiers)

In order to model the word-order constraints for periphery modifiers, while going through the sentence in a direction away from the nuclear predicate, we have to keep track of the modifier layer that we have already reached. As an example, consider the possible placements of the adverbs evidently (evidential, clausal periphery), slowly (pace adverb, core periphery) and completely (aspectual adverb, nuclear periphery) within the sentence in (24), an example taken from Van Valin (2005: 20). (24)

Leslie has been immersing herself in the new language.

The corresponding syntactic tree (without periphery elements) is given in Figure 20.37. Note that here we choose a version where the aspectual operator been, which scopes at NUC, attaches to CORE. This is what a parser would choose in order to avoid crossing branches while being able to adjoin an adverbial at CORE that is placed between been and immersing. All three adverbs can be placed either to the left or to the right of the nucleus, and combinations of this are also possible (cf. Van Valin 2005: 20). The position of the adverbs has to be such that if more than one is to the left of the verbal nucleus, within this group, clausal adverbs have to precede core adverbs, which in turn have to precede nuclear adverbs, while for the group of adverbs placed to the right of the verb, the opposite order has to be respected (first nuclear, then core and then clausal adverbs). CLAUSE CORE RP

OP

OP

Leslie has been

NUC

RP

PP

V

herself

in the new language

immersing Figure 20.37 Syntactic tree for (24)

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE CORE

RP

Leslie

PERIS 1

OP

PERIS 1

PERIS 2

OP

been

has

ADV

herself in …

V immersing

CORE∗ PERIS [CL_L +]

NUC

PERIS 2

CORE∗ PERIS [CO_L +]

evidently

ADV

PERIS [CL_L −]

slowly

Figure 20.38 Derivation for (25a)

In order to control the order of modifiers with respect to the layered structure, we assume an edge feature peris (periphery structure) similar to the feature ops (operator structure) introduced in Section 20.5. The way this feature controls modifier order is exactly as in the operator case. The operator feature ops has of course to be percolated along the edge features of the periphery trees and, similarly, operator trees have to percolate the peris feature. Figure 20.38 illustrates how the example in (25a) is derived using the peris edge feature for keeping track of the periphery structure level reached. (25)

a. Leslie has evidently been slowly immersing herself in the new language. b. *Leslie has slowly been evidently immersing herself in the new language.

Sentence (25b), by contrast, would be ruled out since the requirement to have peris [cl_l ] when adjoining slowly would not be satisfied. In addition to the edge features, which constrain the order of the periphery elements, the ADV node itself can have a feature that indicates at which layer it takes scope. For instance [scope cl] for evidently and [scope co] for slowly.

20.8

Linking Syntax and Semantics

The main focus of the present chapter is on formalizing the syntactic side of RRG. This section shows briefly how the described tree rewriting approach can be extended to combining syntactic and semantic composition along the lines of Kallmeyer and Osswald (2013) and Kallmeyer et al. (2016). In this approach, the linking between syntactic and semantic components is

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

777

778

LAURA KALLMEYER AND RAINER OSSWALD

CLAUSE CORE[IND e] RP[IND x] NUC[IND e] RP[IND y] Kim

V[IND e]

the vase

smashed

e = [do′(x, Ø)] CAUSE [BECOME smashed′(y)], x = Kim, y = vase ⎤ causation ⎡ ⎤ ⎥ ⎢ activity ⎥ ⎢ ⎥ ⎢ ⎣ ⎦ person ⎥ ⎢ ⎥ ⎢ x ⎥ ⎢ ‘Kim’ ⎡ ⎤⎥ e⎢ ⎥ ⎢ change_of_state ⎥ ⎢ ⎥ ⎢ ⎢ ⎥ y vase ⎢ ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎣ ⎣ smashed_state ⎦⎦ y ⎡

Figure 20.39 Linking syntax and semantics: derived tree

encoded by an interface feature ind(ex), which can occur at nodes in the syntactic tree. The value of this feature is an identifier, or label, which refers to a specific component of the semantic representation. Consider the simple example in (26) and its syntactic representation in the upper part of Figure 20.39. (26)

Kim smashed the vase.

The figure illustrates how the syntactic constituents, represented by nodes in the tree, are linked to components of the associated semantic representation. In fact, the figure shows two alternative semantic representations: a logical structure of the type common in RRG (cf., e.g., Van Valin 2005: 151), and a semantic frame, represented as an attribute value matrix, that captures basically the same information about the event and its participants using types, attributes and values. Possible ways of translating RRG’s logical structures into semantic frames are discussed in Osswald and Van Valin (2014) and Osswald (2021). An advantage of using frames is that constraints on semantic representations can be formalized in terms of types and attributevalue constraints, and that semantic composition comes down to frame unification. In accordance with the general approach introduced in Section 20.3, the syntactic tree in Figure 20.39 is derived by, first, combining elementary tree templates with lexical anchors and, second, combining the resulting lexicalized elementary trees via substitution, sister adjunction

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

CLAUSE CORE[IND e] ⎤ causation ⎥ ⎢ activity ⎥ ⎢ ⎥ ⎢ x ⎥ ⎢ ⎡ ⎤⎥ ⎢ ⎥ e⎢ change_of_state ⎥ ⎢ ⎢ ⎥ ⎥ ⎢ y ⎢ ⎥⎥ ⎢ ⎢ ⎥ ⎥ ⎢ smashed_state ⎦⎦ ⎣ ⎣ y ⎡

RP[IND x] NUC[IND e] RP[IND y] V[IND e] V[IND e ] smashed

Figure 20.40 Anchoring the default transitive template with ‘smashed’

⎡

CLAUSE CORE[IND e] RP[IND x] NUC[IND e] RP[IND y] V[IND e]

⎤ causation ⎢ ⎥ activity ⎢ ⎥ ⎢ ⎥ x ⎢ ⎥ ⎡ ⎤ ⎢ ⎥ ⎢ ⎥ e⎢ change_of_state ⎥ ⎢ ⎥ ⎢ ⎥ y ⎢ ⎥⎥ ⎢ ⎢ ⎥⎥ ⎢ smashed_state ⎦⎦ ⎣ ⎣ y

smashed

x

RP[IND x ]

RP[IND y ]

Kim

the vase

person ‘Kim’

y vase

Figure 20.41 Syntactic composition for (26) after argument linking

or wrapping substitution. For the given example, the grammar would contain a transitive template (an elementary tree) that gets anchored by the lexical item smashed (cf. Figure 20.40). The anchoring step (i.e. the insertion of the smashed tree into the V◊ node) induces a unification of the feature structures on the two V nodes. The linking between semantic participants and syntactic argument slots can then be computed via a constraint-based formulation of the linking algorithm along the lines of Kallmeyer et al. (2016). The result of the lexical anchoring and the argument linking is the elementary tree plus the semantic frame at the top of Figure 20.41. The trees for Kim and the vase and their respective frames can now be added by substitution, which means that the frames labelled x′ and x unify and likewise the frames labelled y′ and y. The

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

779

780

LAURA KALLMEYER AND RAINER OSSWALD

result is the interlinked pair consisting of a syntactic tree and a semantic frame shown in Figure 20.39.

20.9

Conclusion

The formalization of RRG syntax presented in this chapter puts emphasis on drawing a clear line between declarative and procedural elements and on the proper modelling of syntactic compositionality. Leaning on concepts from Tree Adjoining Grammars, we formalized the syntactic dimension of RRG as a tree rewriting grammar consisting of elementary tree templates that are anchored by lexical items and that are combined by three modes of composition: (simple) substitution, (sister) adjunction, and wrapping (substitution). Moreover, elementary tree templates are specified by classes of tree constraints (in the so-called meta-grammar), which allows us to make explicit how the different templates are structurally related to each other. Employing wrapping substitution turns out to be vital for the appropriate modelling of syntactic composition in RRG. Besides allowing the compositional derivation of extraction from complement constructions from elementary argument construction templates (see Section 20.3.3), wrapping substitution is also well suited to deriving coordination and cosubordination chains, including those that can arise in embedded control and matrixcoding constructions (see Section 20.6.1). The proposed formalization does not introduce a separate tree structure for representing the operator projection but treats operators as part of the constituent tree. In order to avoid crossing branches, operators need not be directly attached to the constituent nodes that define their scope-taking behaviour. Rather, this information is encoded in node features associated with the operator tree. An elaborate system of features is also responsible for enforcing the correct surface order of the operators, which reflects their scopal domain (see Section 20.5.2). While the formalization of RRG’s semantic structures and its linking system is beyond the scope of the present chapter, we sketched in Section 20.8 how semantic representations can be compositionally integrated with the formal syntactic framework introduced here. We proposed the use of semantic frames for this purpose, but the described interface between syntax and semantics is open to other formal semantic approaches, including a formalized version of RRG’s logical structures. A complete formalization of RRG as a theory of grammar along the lines of the present chapter requires additional steps, of course. In particular, RRG’s linking algorithm should be fully spelled out as a system of constraints that make reference to the chosen formal syntactic and semantic representations. Another important and non-trivial task is the formal representation of information structure and its integration with the syntactic and semantic representations.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

References Bladier, Tatiana, Andreas van Cranenburgh, Kilian Evang, Laura Kallmeyer, Robin Möllemann and Rainer Osswald. 2018. RRGbank: A Role and Reference Grammar corpus of syntactic structures extracted from the Penn Treebank. In Proceedings of the 17th International Workshop on Treebanks and Linguistic Theory, 5–16. Linköping Electronic Conference Proceedings. Bladier, Tatiana, Laura Kallmeyer, Rainer Osswald and Jakub Waszczuk. 2020. Automatic extraction of tree-wrapping grammars for multiple languages. In Proceedings of the 19th Workshop on Treebanks and Linguistic Theories, 55–61. Association for Computational Linguistics. Bohnemeyer, Jürgen and Robert D. Van Valin, Jr. 2017. The macro-event property and the layered structure of the clause. Studies in Language 41(1): 142–197. Boyd, Adriane. 2007. Discontinuity revisited: An improved conversion to context-free representations. In The Linguistic Annotation Workshop at ACL 2007, 41–44. Prague, Czech Republic. Crabbé, Benoit and Denys Duchier. 2005. Metagrammar redux. In Henning Christiansen, Peter Rossen Skadhauge and Jørgen Villadsen (eds.), Constraint Solving and Language Processing, Lecture Notes in Computer Science 3438, 32–47. Berlin: Springer. Crabbé, Benoit, Denys Duchier, Claire Gardent, Joseph Le Roux and Yannick Parmentier. 2013. XMG: eXtensible MetaGrammar. Computational Linguistics 39(3): 1–66. Croft, William, Jóhanna Barðdal, Willem Hollmann, Violeta Sotirova and Chiaki Taoka. 2010. Revising Talmy’s typological classification of complex events. In Hans C. Boas (ed.), Contrastive Studies in Construction Grammar, 201–236. Amsterdam: John Benjamins. Frank, Robert. 2002. Phrase Structure Composition and Syntactic Dependencies. Cambridge, MA: MIT Press. Ginzburg, Jonathan and Ivan A. Sag. 2001. Interrogative Investigations: The Form, Meaning and Use of English Interrogatives. Stanford, CA: CSLI Publications. Johnson, Mark. 1987. A new approach to clause structure in Role and Reference Grammar. In Davis Working Papers in Linguistics 2, 55–59. Davis, CA: University of California. Joshi, Aravind K. 1985. Tree adjoining grammars: How much contextsensitivity is required to provide reasonable structural descriptions? In D. Dowty, L. Karttunen and A. Zwicky (eds.), Natural Language Parsing, 206–250. Cambridge: Cambridge University Press. Joshi, Aravind K. and Yves Schabes. 1997. Tree-adjoining grammars. In Grzegorz Rozenberg and Arto Salomaa (eds.), Handbook of Formal Languages, Vol. 3: Beyond Words, 69–123. Berlin: Springer. Joshi, Aravind K., Laura Kallmeyer and Maribel Romero. 2008. Flexible composition in LTAG: Quantifier scope and inverse linking. In Harry Bunt and Reinhard Muskens (eds.), Computing Meaning, Vol. 3, 233–256. Berlin: Springer.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

781

782

LAURA KALLMEYER AND RAINER OSSWALD

Kallmeyer, Laura. 1999. Tree Description Grammars and Underspecified Representations. PhD thesis, Universität Tübingen. Technical Report IRCS-99-08 at the Institute for Research in Cognitive Science, Philadelphia. Kallmeyer, Laura. 2016. On the mild context-sensitivity of k-tree wrapping grammar. In Annie Foret, Glyn Morrill, Reinhard Muskens, Rainer Osswald and Sylvain Pogodalla (eds.), Formal Grammar: 20th and 21st International Conferences, FG 2015, Barcelona, Spain, August 2015 (Revised Selected Papers. FG 2016, Bozen, Italy, August 2016, Proceedings, number 9804 in Lecture Notes in Computer Science), 77–93, Berlin: Springer. Kallmeyer, Laura and Rainer Osswald. 2013. Syntax-driven semantic frame composition in lexicalized Tree Adjoining Grammars. Journal of Language Modelling 1(2): 267–330. Kallmeyer, Laura and Rainer Osswald. 2017. Combining predicateargument structure and operator projection: Clause structure in Role and Reference Grammar. In Marco Kuhlmann and Tatjana Scheffler (eds.), Proceedings of the 13th International Workshop on Tree Adjoining Grammars and Related Formalisms (TAGþ13), 61–70. Association for Computational Linguistics. Kallmeyer, Laura, Rainer Osswald and Robert D. Van Valin, Jr. 2013. Tree wrapping for Role and Reference Grammar. In Glyn Morrill and Mark-Jan Nederhof (eds.), Formal Grammar (FG 2012/2013), Lecture Notes in Computer Science 8036, 175–190. Berlin: Springer. Kallmeyer, Laura, Timm Lichte, Rainer Osswald and Simon Petitjean. 2016. Argument linking in LTAG: A constraint-based implementation with XMG. In David Chiang and Alexander Koller (eds.), Proceedings of the 12th International Workshop on Tree Adjoining Grammars and Related Formalisms (TAGþ12), 48–57. Association for Computational Linguistics. Kanazawa, Makoto. 2016. Multidimensional trees and a ChomskySchützenberger-Weir representation theorem for simple context-free tree grammars. Journal of Logic and Computation 26(5): 1469–1516. Kroch, Anthony. 1987. Unbounded dependencies and subjacency in a Tree Adjoining Grammar. In Alexis Manaster-Ramer (ed.), The Mathematics of Language, 134–172. Philadelphia: John Benjamins. Nolan, Brian. 2004. First steps toward a computational RRG. In Proceedings of the International Role and Reference Grammar Conference 2004, 196–223. Institute of Technology Blanchardstown, Dublin. Osswald, Rainer. 2021. Activities, accomplishments and causation. In Robert D. Van Valin, Jr. (ed.), Challenges in the Analysis of the Syntax–Semantics– Pragmatics Interface: A Role and Reference Grammar Perspective, 3–30. Newcastle upon Tyne: Cambridge Scholars Publishing. Osswald, Rainer and Laura Kallmeyer. 2018. Towards a formalization of Role and Reference Grammar. In Rolf Kailuweit, Eva Staudinger and Lisann Künkel (eds.), Applying and Expanding Role and Reference Grammar (NIHIN Studies), 355–378. Freiburg: Albert-Ludwigs-Universität, Universitätsbibliothek.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

Formalization of RRG Syntax

Osswald, Rainer and Robert D. Van Valin, Jr. 2014. FrameNet, frame structure, and the syntax–semantics interface. In Thomas Gamerschlag, Doris Gerland, Rainer Osswald and Wiebke Petersen (eds.), Frames and Concept Types, 125–156. Dordrecht: Springer. Osswald, Rainer and Robert D. Van Valin, Jr. 2022. The description of transitive directed motion in Lakhota (Siouan). In Laure Sarda and Benjamin Fagard (eds.), Neglected Aspects of Motion Events Description: Deixis, Asymmetries, Constructions, 209–233. Amsterdam: John Benjamins. Pollard, Carl J. and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago, IL: University of Chicago Press. Rambow, Owen. 1994. Formal and Computational Aspects of Natural Language Syntax. PhD thesis, University of Pennsylvania. Rambow, Owen, K. Vijay-Shanker and David Weir. 1995. D-tree grammars. In Proceedings of the 33rd Annual Meeting of the Association for Computational Linguistics, 151–158. Association for Computational Linguistics. Rambow, Owen, K. Vijay-Shanker and David Weir. 2001. D-tree substitution grammars. Computational Linguistics 27(1): 87–121. Sag, Ivan. 2012. Sign-Based Construction Grammar: An informal synopsis. In Hans Boas and Ivan Sag (eds.), Sign-Based Construction Grammar, 61–188. Stanford, CA: CSLI Publications. Sag, Ivan A. and Thomas Wasow. 1999. Syntactic Theory: A Formal Introduction. Stanford, CA: CSLI Publications. Ullrich, Jan (ed.) 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN: Lakota Language Consortium. Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Constructions in Lakota. PhD thesis, Heinrich Heine University Düsseldorf. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Notes 1 Round brackets indicate optionality, curly brackets mean that the category in question may not be available in every language. DP stands for ‘detached position’ and ECS for ‘extra-core slot’ (covering pre- and postcore slots). 2 The given analysis is slightly simplified in that the operator did contributes also to other layers than CLAUSE; see Section 20.5 for a more detailed exposition, which also covers operators that are realized as bound morphemes. 3 The possible occurrence of ‘crossing branches’ due to mismatches between the surface position of an operator or a periphery element and its scope will be discussed in Sections 20.5 and 20.7.

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

783

784

LAURA KALLMEYER AND RAINER OSSWALD

4 Sister adjunction has been introduced as a composition operation on socalled d-trees by Rambow et al. (1995). Notice that sister adjunction differs from adjunction in TAG. In fact, the latter operation is more akin to the wrapping operation described in Section 20.3.3. 5 We will see later how edge features can be used to constrain the positions of operators or periphery elements among the daughters of their target nodes. 6 This analysis is inspired by TAG, where the slot for the wh-phrase and its predicate originate from the same elementary tree and the elementary tree of the embedding verb adjoins in between; see Kroch (1987). 7 Wrapping substitution is related to the concept of flexible composition proposed in Joshi et al. (2008), which allows one to interpret TAG adjunction as a wrapping operation. 8 The idea of d-edges or dominance links has been used in various extensions of TAG, for instance in V-TAG (Rambow 1994), Tree Description Grammars (Kallmeyer 1999), and D-Tree-Substitution Grammar (Rambow et al. 2001). 9 Bladier et al. (2020) propose a slight extension of wrapping substitution where in cases of an empty tree γ, the upper node targeted by the operation need not be a root node. 10 This means that for a given grammar, one can implement a parsing algorithm such that for a sentence of length n, the program computes in at most cnm steps whether the sentence belongs to the language, c and m being fixed constants. 11 This transformation of splitting a node with a discontinuous span into several nodes for the different components was also proposed by Boyd (2007), though not with a unique identifier for the split node. 12 A similar constraint on elementary trees is often proposed in TAG-based approaches to linguistic analysis; cf., e.g., Frank (2002). 13 Since in passive constructions, the auxiliary be is structurally required for nucleus formation, it has a representation in the constituent projection as an AUX node; cf. Van Valin (2005: 13, note 3). 14 A computational implementation is provided by the meta-grammar compiler XMG (eXtensible MetaGrammar, Crabbé et al. 2013), which allows specifications of RRG trees including d-edges (xmg.phil.hhu.de). The resulting grammars can then be processed using the TuLiPA parser (github.com/spetitjean/TuLiPA-frames). 15 Such a representation is chosen in the RRGbank (rrgbank.phil.hhu.de) and RRGparbank (rrgparbank.phil.hhu.de), both treebanks of syntactic RRG structures, see Bladier et al. (2018). 16 We slightly simplify here because the mod feature would actually be embedded under features op and co, for instance [op[co[mod deont]]].

https://doi.org/10.1017/9781316418086.025 Published online by Cambridge University Press

21 Computational Implementation and Applications of Role and Reference Grammar Brian Nolan

21.1

Introduction

Functional linguists of all persuasions consider that theories of linguistic structure should be directly relatable to testable theories of language production and comprehension. Computational models assist in this process, and today professional linguists can be found working in many areas of knowledge and skills using strategies and techniques that did not exist a generation ago. This chapter therefore examines the range of computational implementations and applications of Role and Reference Grammar (RRG) and the characteristics of RRG that makes this functional model of language amenable to implementation in software. We examine the elements of RRG treated in software: the architecture of the lexicon, the linking system, the functional philosophy encapsulated within speech acts in a ‘language as communication’ perspective and the logical structures that provide a meta-representation of an utterance. Important also are the layered structure of the clause and noun phrase. Constructions in RRG have also received a computational treatment. RRG has been used successfully to characterize languages as diverse as Irish, Persian, Maori, Arabic and Japanese, along with many accusative and ergative languages, while the languages that have undergone a computational treatment in RRG have included English, Modern Standard Arabic, Spanish, Biblical Hebrew, German, and others. The chapter, then, is a testimonial to the potential of an RRG model of language to be implemented in software, and to be useful in tasks such as machine translation. It includes some discussion of what must be transformed and augmented in an RRG account of a phenomenon for computational purposes, and why computer scientists have come to RRG in

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

786

BRIAN NOLAN

particular for their needs. The RRG model is rapidly evolving to new levels of descriptive, explanatory and computational adequacy. By computational adequacy we mean the degree to which a theoretical model can be formally specified to enable computational modelling as a benchmark for that theory. As will become clear, we find that the RRG model is most suitable as a basis for computational implementation in software and can serve as a rich testing ground for the linguistic theory. Computational models of RRG, by intention linguistically motivated, deal with problems at the interfaces between concept, semantics, lexicon, syntax and morphology, including language processing and grammar. The challenge is to determine how best to design and implement the model and its principled interfaces in software. What makes RRG suitable for implementation in software is that it is a model of grammar that posits a direct mapping between the semantic representation of a sentence and its syntactic representation (Van Valin 2005, 2008). RRG is a monostratal theory, positing only one level of syntactic representation. The actual form of the sentence and its linking algorithm can work in both directions from syntactic representation to semantic representation and vice versa. In RRG, semantic decomposition of predicates and their semantic argument structures are represented as logical structures. The lexicon in RRG takes the position that lexical entries for verbs should contain unique information only, with as much information as possible derived from general lexical rules. RRG makes use of a set of thematic roles organized into a hierarchy in which the highestranking roles are Actor (for the most active participant) and Undergoer (for most affected). Amongst the design challenges is the requirement for a digital model of the lexicon as a persistent store, either as a database, persistent XML tree structure or serializable object. The architecture of the lexicon is then an empirical problem for the software designer. It is clear that while the RRG lexicon is traditionally considered with respect to verbal logical structures, the software typically needs the lexicon to also include nouns, adjectives, prepositions, and so on, in some meaningful way. To deal with many of the problems found in modelling language within the model, constructions also need to be stored as occurrence patterns. Additionally, lexemes need to be meaningfully stored within the lexicon. As grammatical morphemes are very much language-specific, a morpheme store is also needed within the software. All the software and computing applications mentioned in the case studies make extensive use of the bidirectional linking system from morphosyntax to semantics and semantics to morphosyntax. The implementation may follow construction-oriented processing according to some language-specific set of rules translated into computer code. The reasons why many computer scientists and computational linguists use RRG are precisely because of the desire to deploy a functional linguistic

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

model that has a strong theory of the lexicon and a role for constructions. Important also is the idea of projecting from the lexicon in a bidirectional linking system, the layered structure of the word (LSW), the layered structure of the noun phrase (LSNP), and the layered structure of the clause with its sub-model of nexus–juncture relations. The approach of RRG to argument realization is highly principled in a typologically coherent way and this adds considerable value to all language modelling efforts in software. Additionally, the interfaces within RRG that reach from lexicon through semantics via morphosyntax into speech act, information structure, pragmatics and discourse are considered to be highly valuable. The RRG linguistic model strives for high levels of descriptive, explanatory, predictive and typological adequacy (Butler 2009). Through the deployment of the RRG model in computational applications it is shown to meet computational adequacy. In many respects with regard to computational implementation, RRG can be considered as a ‘functional generative’ model with parse and generate ‘directions’ within the bidirectional linking system. It does not, however, express itself as treating the Chomsky-typed languages as formal constructs. It is more typologically sensitive than that. While RRG can be modelled within procedurally described or declarative approaches, it is always as a semantically motivated and lexically projected model of language. It may make use of attribute-value matrices (AVMs) and constructions, depending on the object of study within the computational model. The chapter has the following organIsation. In Section 21.2, we characterize the start of computational work on RRG, concentrating on the syntactic side of the linking system. We look at the use of generalpurpose programming languages to create parsers for English, German, Dyirbal and Biblical Hebrew. In Section 21.3, we give a description of an RRG-based machine translation engine for Modern Standard Arabic that parses a sentence in the native Arabic orthography and translates it into grammatical English. The software employs an interlingua bridge architecture. Section 21.4 outlines current work on intelligent conversational agents and humanoid avatars, where RRG is used as the language engine. In Section 21.5, the FunGramKB project and related toolsets oriented towards the RRG model are discussed. In Section 21.6, the contribution of RRG to linguistic modelling in natural language processing (NLP) applications is discussed.

21.2

The Computational Modelling of RRG

Computational work using RRG has been ongoing since 2004 in a variety of ways (Nolan 2004). Initial concerns with computational treatments of RRG were with exploring the possibilities of expressing the theory in a computationally tractable way as a feature-based unification grammar. This

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

787

788

BRIAN NOLAN

involved creating a hierarchy of types, each of which was structured as an attribute-value matrix with a set of features and associated values. A set of rules was created and lexicon entries were structured to unify with the rules, somewhat along the lines of a phrase structure grammar. Early modelling was achieved using the constraint-based computer programming language Prolog, treating elements of RRG as a definite clause grammar. In this approach, the lexical rules and the lexical entries shared a common formalism. However, it proved not to be feasible to formalize the full RRG model using Prolog, as the RRG theory is not a context-free grammar. It also proved not to be possible to use Prolog to implement the operator projection, the linking system, the rich lexicon, and the full layered structure of the clause as understood in RRG. Later, additional modelling was undertaken using a software toolset, based on Common Lisp, called the Linguistic Knowledge Builder (LKB). The LKB is an opensource grammar engineering environment for creating grammars and lexicons of natural languages, for typed feature structure grammars (Copestake 2002). LKB is typically used for grammars with typed feature structures, such as Head-Driven Phrase Structure Grammar (HPSG), with unification. The type system constrains the allowable structures and provides a way of capturing linguistic generalizations. At that time, the Prolog language and the LKB system were seen to offer several advantages in potentially making computational linguistics and natural language processing accessible to linguists with a very limited background in computer science. While the LKB system was successfully used to build and model small RRG grammars of English, these remained at the level of small-scale experiments. Overall, the LKB proved to be unwieldy for handling the core theoretical elements of RRG, such as the design of the lexicon, and the linking system. Following the limitations found in LKB and the complexity of large-scale Prolog programs, development work on RRGbased computational linguistics progressed to more modern generalpurpose programming languages. These languages also offered the powerful advantage of enabling the creation of graphical user interfaces (GUIs) to generate dynamic visualizations of RRG trees of the layered structure of the clause more easily. They also offered the opportunity to create more robust applications that implement the RRG linguistic model along with more sophisticated data structures and XML. Guest (2008: 435–453) reports on RRG-based software developed for parsing a selection of sentences from Dyirbal (including sentences with discontinuous constituents and free word order), English (fixed word order), and a small sample of Dutch. This parser exploits two features of RRG: first, that syntactic structure is represented in terms of templates instead of rules, and second, that grammatical categories (operators) are represented differently from predicates, arguments and adjuncts. Guest extends a chart parser strategy to enable varying degrees of word-order flexibility and parsing via templates instead of rules. The choice of templates rather than rules was

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

made because templates enable more information to be captured, which in turn makes it easier to extract the meaning from the sentence. From Guest’s (2008: 435) perspective, RRG is a suitable theory for a computational treatment for several reasons. 1. RRG posits constituent, operator, and focus projections, such that the various aspects of a sentence can be dealt with separately. This enables words that modify other words to be removed from the constituent projection and placed in an operator projection, which shows how these words modify the meanings of the words in the constituent projection. Therefore, only the main constituents of a sentence have to be parsed, simplifying the parsing process. 2. Secondly, RRG has a strong bidirectional linking system from syntax to semantics. Grammatical constructs are designed to be both crosslinguistically valid and to make the meaning relatively easy to extract. Conversely, she finds there are some elements of RRG which make it harder to implement. Guest (2008: 436) notes these to include: 1. It is much more difficult to parse with templates than with rules, and RRG templates are particularly hard because, in a parse tree, lines are allowed to cross. The parse trees are not made up of parents and children, and nodes can have modifiers (such as PERIPHERY) attached to them. 2. In RRG, word-order constraints are implicit in the templates. The theory supports modelling of examples from many languages which include fixed and free word order, and mixed varieties in between. Parsing algorithms tend to handle either fixed-word-order or free-wordorder languages. Most parsing algorithms, such as those based on HPSG and context-free grammar, are based on rules. Although templates can easily be reduced to rules, information is lost in going this route. As we mentioned, Guest (2008: 436) describes an approach based on a chart parser that has been modified to handle both templates and varying degrees of word-order flexibility. This parsing algorithm is characterized by Guest as consisting of several stages: 1. Use Toolbox1 to tag the sentences. Toolbox, a semi-interactive tagging program, was chosen by Guest because: (i) A user can define their own tags and (ii) it is easy to ensure tags are correct. Toolbox will ask the user if there is more than one possibility for a word. 2. Strip away the operators. This removes all words that modify other words, based on a correct tagging of head and modifying words. The modifying tags that belong to head words are specified in a file, and the end result is a simplified sentence to be input to the parser.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

789

790

BRIAN NOLAN

3. Parse the simplified sentence using templates. Guest achieves this by collapsing the templates to rules, parsing using a chart parser and then rebuilding the trees at the end. The chart parser has been modified to handle varying degrees of word-order flexibility. 4. Draw the resulting parse tree. The parser has been tested by Guest on example sentences from Dyirbal, English and Dutch. This proof-of-concept parsing method for RRG, using templates, can handle both fixed word order and variable word order without any difficulty. Diedrichsen (2014: 105–142) describes an implementation of a sentence parser for German that implements the linking system of RRG from syntax to semantics (see also Diedrichsen 2008; Van Valin and Diedrichsen 2006), and an implementation of the RRG lexicon in a data structure. This RRG parser for German employs the concept of a construction as a Table 21.1 The German bracket structure construction (based on Diedrichsen 2014: 109) CONSTRUCTION: German bracket structure construction SIGNATURE: VFIN [ARG1. . .ARG2. . .ADJ1. . .PARTICLE] PART.OF.VERBAL.EXPRESSION Space between brackets may be empty (if clause is intransitive) CONSTRAINT: No constraints or restrictions. WORKSPACE: Real-time processing according to the following construction-specific rules SYNTAX: Bracket Structure: Bracket is ALWAYS opened by the finite verb. The right bracket is filled according to the following rules: 1. If VFIN ¼AUX {PART.OF.VERBAL.EXPRESSION ¼ full verb (PSTP)}; 2. If VFIN ¼full verb {PART.OF.VERBAL.EXPRESSION ¼ separable prefix}; 3. If VFIN ¼light verb {PART.OF.VERBAL.EXPRESSION ¼ rest of collocation}; Order of arguments within brackets follows topic comment structure and principle of end weight. The following rules apply: 1. General constraints: pronoun > other, RP > PP 2. Case-based argument ordering constraint: NOM > DAT > ACC (default) 3. If ACC ¼ pronoun, then ACC > DAT (default) MORPHOLOGY: AUX: May be any auxiliary or modal verb VFIN may be any verb. PHONOLOGY: Not specified. SEMANTICS: [þtelic], where telicity may be only invoked for conversational purposes: adds intensity, expression PRAGMATICS: Turn taking: Signal for dimension of Turn Constructional Unit (TCU); right brace marks end of TCU Illocutionary force: not specified Focus structure inside braces: TOP>FOC

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

language-specific structural pattern to guide its operation over the ‘bracket structure’ of the German clause (Table 21.1). The construction is considered to be a grammatical object (Nolan 2013; Diedrichsen 2014: 109). Inside these brackets, the main information of the sentence is placed in an order that reflects the information structure. The software accepts an input sentence in German and parses it into its respective tokens. Each token is checked to determine its lexical category and the morphological marking of each token is unpacked for its feature set. The parse captures the structural representation of the clause and grammatical features like tense in RRG notation. The parser is able to recognize and parse eight construction types of the Standard German clause. Included are conventional sentences using active transitive and intransitive structures with V2 and V1 patterns. The word order is varied such that the direct object may appear in sentence-initial position. The structure is parsed in terms of the layered structure of the clause, which involves a nucleus, a core with arguments, a pre-core slot and a clause layer that subsumes the other elements. The sentence structures are presented within a GUI as a series of embedded frame structures according to the style of attribute-value matrices (AVMs). Specifically, this style of representation has the format of [feature: value] and value can contain nested AVMs such as [feature: [feature: value]]. This parser is the first application able to parse German sentence constructions according to RRG rules and provide an elaborate set of features with each of the tokens from the constructions. German grammar is considered to be notoriously difficult for computational applications, as it has (i) a variety of forms in its lexicon set, due to the (ii) three genders and (iii) its rich morphology, and it also has (iv) great freedom in word order, of which the only regularity is that with declarative clauses, the finite verb has to appear in the second position within the order of constituents. This parser from Diedrichsen, which does not claim full coverage, is a valuable approach to a full computational representation of German, and serves as a robust proof of concept to demonstrate again that the RRG linguistic model is amenable to computational implementation. RRG-related computational work has been carried out on the parsing of Biblical Hebrew (Winther-Nielsen 2008). This work involved the creation of a web-based software application, called the Role-Lexical Module (RLM), to build a lexicon of logical structures and semantic representations for RRG. The goal of the RLM project has been to develop and test new methods for lexical analysis within RRG in the analysis of Biblical Hebrew. Specifically, this software is a computational tool for doing a semantic analysis of Biblical Hebrew verbs in a text and organizing the results in an online lexical database. The RLM tool uses an electronic database for Hebrew, and a linguist can choose any clause from Genesis 1–3, with simple translation glosses, and can then perform various computer-aided analyses. The RLM is able to display the nodes of the RRG syntax trees representing the layered structure of the clause. For computational work on the RLM tool, the RRG theory offered the advantage of a bidirectional linking between syntactic

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

791

792

BRIAN NOLAN

structure and the representation of semantic structure in the lexicon through mapping rules. The RRG theory supports a clear, semi-formal representation of meaning and structure in all languages. The linguist can classify verbs into Aktionsart classes, represent the meaning of clauses following a selection of logical structure, and gradually build an RRG lexicon for Biblical Hebrew. The tool stores the basic lexical meanings of verb classes in the semantic metalanguage of RRG and it offers a lexical representation of meaning in actual linguistic contexts through a WordNet glossing. This tool is used for parsing, lexical representation, and verb class decomposition. The RLM tool offers a WordNet lexical selector, a text display of transliterated and glossed Hebrew, a syntactic parser for RRG, and a logical structure decomposer. Related to this project is the work of Gottschalk (2014), who developed a computable account of RRG based on constructional schemata. Gottschalk’s approach places emphasis on thematic relations within three-place predicates rather than the semantic macroroles of actor and undergoer. In Gottschalk’s view, the standard linking algorithm of RRG has been computationally enhanced for variable undergoer linking in English three-place predicates. An innovation of Gottschalk’s computational work is the use of conceptual graphs to implement logical structures in software. Gottschalk implements conceptual graphs as an approach to the semantic component of RRG. These have been applied within the Biblical Hebrew semantic parser as part of the RLM, as a functional extension. The application of conceptual graphs has the advantage that computational approaches to ancient languages such as Biblical Hebrew can be more easily developed (Gottschalk 2019). In this approach, a linking algorithm from syntax to semantics is reduced to a set of lexical rules which match AVMs defining the layered structure of the clause against an ontology of Biblical Hebrew.

21.3

Machine Translation: An RRG Interlingua

Machine translation (MT) is a subfield of computational linguistics that investigates the use of computer software to translate text (or speech) from one natural language to another. In a software application on a larger scale to those already mentioned, the RRG linguistic theory has successfully been used to underpin the language model in a lexically based proof-of-concept machine translation system called UniArab (Nolan and Salem 2010, 2011) that supports fundamental aspects of Arabic, including lexical category, agreement and tenses. UniArab is based on the bidirectional linking algorithm of RRG (syntax to semantics and vice versa). UniArab takes Modern Standard Arabic as input in the native orthography, parses the sentence(s) into a logical metarepresentation based on the fully expanded RRG logical structures and generates perfectly grammatical English output with full agreement and morphological resolution. UniArab utilizes an XML-based implementation of elements

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

of RRG in software. In the analysis of Arabic within the software, the lexical and grammatical properties of the Arabic words are extracted automatically. From the parse, UniArab then creates a computer-based representation for the logical structure of the Arabic sentence(s). In this application, RRG has been used to motivate the computational implementation of the lexicon as an XML data structure and to implement in software the RRG bidirectional linking system to build the parse and generate functions between the syntax–semantic interfaces. Through seven input phases, including the morphological and syntactic unpacking, UniArab extracts the logical structure of an Arabic sentence. Using the XMLbased metadata representing the RRG logical structure, UniArab then accurately generates an equivalent grammatical sentence in the target language of English through four output phases. Nolan and Salem (2011) report on the development of this machine translation system as a rule-based semantically oriented interlingua bridge framework for machine translation of Arabic language processing using the RRG linguistic model. The UniArab system is able to analyse intransitive, transitive and ditransitive Arabic sentences in native Arabic (right to left) orthography and extract their logical structure. Through a detailed study of the Arabic language, an analyser was developed that can successfully process many of the unique features and challenges present in Arabic. This logical structure is then used in the generation phase, where the sentence is translated into another language, in this case, English. The Arabic language is written from right to left and it has complex, language-specific grammar rules and a relatively free word order. These distinguishing features pose a major challenge in processing Arabic text for linguistic analysis. The UniArab framework demonstrates that RRG is a foundation for building multi-language machine translation systems. It is common to categorize the three different MT methodologies as direct, transfer and interlingua. These levels of analysis are illustrated by what has become known as the Vauquois triangle (Vauquois 1968) (see Figure 21.1). The methodologies differ (i) in the linguistic depth of analysis of the source language and (ii) in the degree to which they attempt to reach a languageindependent representation of linguistic meaning of the source and target languages. An interlingua is designed to be a language-independent representation from which translations can be generated to different target languages. According to Dorr (1992: 46), in order to adopt an interlingua approach to machine translation, one must construct a language-independent representation that lends itself readily to the specification of a systematic mapping that operates uniformly across all languages. Dorr describes such a representation based on the lexical conceptual structure of Jackendoff (1990), which abstracts away from syntax sufficiently to enable languageindependent encoding, while retaining enough structure to be sensitive to the requirements for language translation. Machine translation using an interlingua approach typically involves semantic analysis of the source

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

793

794

BRIAN NOLAN

Figure 21.1 The Vauquois triangle (based on Dorr et al. 2006: 384)

language (Dorr et al. 2006: 1). At the interlingua level, a single underlying meta-representation represents the underlying semantics of the source language SL1 text., which will be used to generate a translation into a target language TL1. In principle, an interlingua representation of a sentence contains sufficient information to allow generation in any language; the more target languages there are, the more valuable an interlingua becomes. When there are many source languages (SLn) and many target languages (TLn) for translation, an interlingua bridge is deployed. This has an advantage in facilitating the scaling up to multiple languages. To add a language to an interlingua bridge, one needs to add, for that language, a lexicon, a parse function into a meta-representation, and a generate function from the metarepresentation. To translate from one source into N target languages, one needs (1 þ N) steps using an interlingua. However, to translate pairwise among all the languages, one needs only 2N steps using an interlingua compared to about N2 with transfer – this is a significant reduction in development, design and processing cost. In addition, it is not necessary to consider the properties of any other language during the analysis of the source language or generation of the target language; each analyser and generator can be built independently by a monolingual development team. Each system developer only needs to be familiar with their own language and the interlingua. An interlingua MT system will include one lexicon for each language. An advantage of the interlingua approach is that interlingua representations have the potential to be used by other multilingual NLP applications. More importantly, using the logical structures of RRG as the interlingua meta-representation

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

provides distinct advantages, especially given the strong, justified claims of RRG’s typological adequacy. Another advantage of using the logical structures of RRG as the interlingua meta-representation is that the MT is immediately, and by design, linguistically motivated. UniArab uses the interlingua bridge approach (Figure 21.2) in its architecture that allows for language extensibility within the software. The UniArab MT model (Nolan and Salem 2011: 320) thus has two main stages: inputparse-analysis and output-generate. This interlingua-based MT is done via an intermediate semantic representation, based on RRG logical structures, of the source language text. Each language therefore needs a bridge from that language source to the interlingua logical structure [parse source to LS] and a bridge from the interlingua logical structure back to the language [generate target from LS] (Figure 21.3). This approach allows the software to scale up should additional languages be added beyond the initial source (SL1) and target (TL1) language pairs, and avoid a translation complexity problem between languages. The implementation of the interlingua bridge architecture solves the translation complexity problem as automatic language translation is made from a source language into a kernel

Figure 21.2 MT–RRG-based interlingua bridge approach (from Nolan and Salem 2011: 315)

Figure 21.3 The RRG interlingua for UniArab (from Nolan and Salem 2011: 315)

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

795

796

BRIAN NOLAN

meta-representation (the input parse phase) and generates to a target language from the meta representation (the generate phase). Ambiguity problems for an interlingua in a multilingual system are still likely if one of the languages involved has two or more potential forms for a single given word in one of the other languages. However, a semantically oriented approach to MT can potentially disambiguate more easily than other strategies. For an interlingua to be completely language-neutral, it must represent not the words of one or another of the languages but rather language-independent lexical units. Any distinction that can be expressed lexically in the languages of the system must be represented explicitly in the interlingua representation. In this software, the RRG logical structures are strategically used as the basis of the meta-representation in the interlingua bridge with a lexicon encoded in XML. The UniArab system can generate the target language output through classifying every Arabic word in the input source text and creating a meta-representation of the sentence(s) input as a text in a fully populated RRG-style logical structure, including the various nominals and their associated features of [defþ, mascþ], and so on. Arabic is considered to have six major ‘parts of speech’. These are verbs, nouns, adjectives, proper nouns, demonstratives and adverbs. The major parts of speech in the Arabic language have their own attributes, and the UniArab software has a strong analytical component that can extract all these attributes from the words in a sentence. The UniArab system accepts Arabic as its source language (Figure 21.4). The morphology parser and word tokenizer have a connection to the

Figure 21.4 The conceptual architecture of the UniArab system (from Nolan and Salem 2011: 321)

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

lexicon, which holds all attributes of a word. UniArab was developed with the lexicon encoded in XML, and also stores all data in XML format. The system can understand the part of speech of a word, agreement features, number, gender and the word type. The syntactic parse unpacks the agreement features between elements of the Arabic sentence into a semantic representation (the logical structure) of the state of affairs of the sentence. The structure of the UniArab system and its operational phases are described next. Phase 1. Input of Arabic language sentence: The input to the system consists of one or more sentences in Arabic. UniArab can accept as input one or multiple sentences in a single input. Phase 2. Sentence Tokenizer: Tokenization is the process of demarcating and classifying sections of a string of input characters. In this phase, the system splits the text, which may consist of paragraphs, into sentence tokens. The resulting tokens are then passed to the word tokenizer phase. Phase 3. Word Tokenizer: In this phase, sentences are split into word tokens. The Arabic sentence is read from right to left. Phase 4. Lexicon XML Data-source: The lexicon is constructed as a set of XML documents for each component category of Arabic. The lexicon contains the Arabic lexemes in this XML data source ordered by each lexical category. Phase 5. Morphology Parser: This works directly with both the lexicon and tokenizer. A connection is made to the lexicon data-source of Phase 4. To understand the morphology of each word, the software tokenizes each sentence and determines the word relationships. This phase of the system holds all attributes specific to each word of the source sentence. Phase 6. Syntactic Parser: Determines the structure of the Arabic sentence. At this point, the types and attributes of all the words in the sentence are known. Phase 7. Syntactic linking (RRG): The software develops the link from syntax to semantics out of the phrasal structure created in Phase 6 to create a logical structure for later use in generation into the target language, and also to act as the link in the opposite direction from semantics to syntax. Phase 8. Logical Structure: The creation of logical structure is the most crucial phase. An accurate representation of the logical structure of an Arabic sentence is the primary strength of UniArab. Having described the various interlingua parse phases, we now characterize in more detail the generate phase from the meta-representation (i.e. the populated logical structure) to the target language (Nolan and Salem 2011: 323). The target language generation phases in the UniArab system follows a syntactic realization model based strongly on RRG. Generation takes, as

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

797

798

BRIAN NOLAN

input, the universal logical structure of the input sentence(s), and produces, as output, the grammatically correct morphosyntax of the target language. In the UniArab system, Phases 9 to 12 are for generation of the target language, which, in this case, is English. Phase 9. Semantics to Syntax: Once the software has an input sentence and has produced a structured syntactic representation of it, the grammar can map this structure from a semantic representation. In this phase, the system uses the linking algorithm of RRG to determine Actor and Undergoer assignments, assign the core arguments and assign the predicate in the nucleus. It determines the grammatical ‘subject’ by analysing the agreement marking on the verb and the various nominals. The system hinges on the use of logical structures at the metalevel. Phase 10. Syntax Generation: The generation phase from the interlingua bridge meta-representation to the morphosyntax of a particular target language will, of course, depend on the characteristics of the target language. The generation phase implements the RRG semantics-to-syntax linking system. Phase 11. Generate English Morphology: The system generates English morphology. There are some ‘tricky’ special cases, handled well by the UniArab system, such as adding the copula verb ‘to be’ into the English copula sentence, or changing the source language verb’s tense to an appropriate and grammatically correct tense in the target language, depending on the tense distinctions found within the target language. Also, the word order in the target language is considered and applied correctly. Phase 12. English Sentence Generation: The process of generating an English sentence can be as simple as keeping a list of rules, and these rules can be extended through the life of the MT system. The system applies some operations in English, such as vowel change in the lexical item of English to denote sg vs. pl, for example man vs. men. Sometimes this accompanies affixation: break/broke/broken (¼brokeþen) to denote various tense and aspect distinctions. This proof-of-concept software generates grammatically correct English. This computational work has shown that RRG facilitates the translation process from a specific source human language to other target languages. The advantage of UniArab lies in the deployment of an interlingua architecture which uses a robust functional linguistic model founded on RRG in the machine translation kernel. UniArab is flexible and scalable for multilingual generation. One of the primary strengths of UniArab is the accurate representation of the RRG logical structure of an Arabic sentence. As an interlingua MT application, UniArab has both parse and generate functions (Nolan and Salem 2010, 2011). Currently, UniArab covers a representative broad selection of words and can translate intransitive, transitive and

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

ditransitive clauses, as well as Arabic copular-like nominative clauses. UniArab has been shown to outperform existing machine translators in the processing of simple sentences, confirming that RRG is a capable candidate for interlingua-based machine translation. UniArab can accept as input individual sentences or whole paragraphs of text consisting of multiple sentences. An area of future development with the UniArab model is to (i) address the problem of word sense disambiguation and lexical polysemy, and (ii) treat complex sentences and nexus–juncture relations.

21.4

Language-Aware Conversational Agents and Avatars

In this section we highlight important computational work, using RRG, which underpins the development and implementation of (i) text-based conversational agents and (ii) humanoid avatars for Sign Language.

21.4.1 Speech Acts, Knowledge and Conversational Agents Following on primarily from the work on machine translation described earlier that implemented the RRG bidirectional linking system of RRG in software, recent work has advanced and implemented a model of conversational agents in a computational framework that builds on the notion of speech acts from discourse, as a proof of concept (Panesar 2017, 2019). Intelligent conversational agents are a software paradigm that can be used to exploit the possibilities presented by human knowledge, human language and human/agent spoken and written communications, possibly over massive open distributed systems such as the Internet. In this conversational agent framework, a view of software-based agents is proposed based on Nolan (2014). A conversational agent has both ‘internal’ and ‘external’ models to support the speech acts (Searle 1969), common ground formulation (Kecskes and Zhang 2009) and discourse. The internal model of the agent is concerned with the internal state of the agent, based upon the intersection at any given time of the agent’s internal beliefs, desires and intentions, known as BDI states. The external model of the agent is composed of an interaction model with its world (human agent and other agent). These models support the construction of a conversation-specific emergent common ground. Importantly, the conversational agent also has a language model in software that is related to its interaction model to support the interactive bidirectional communication in human language through speech acts. The linguistic model employed to motivate the design of the language model is RRG (Nolan 2012; Van Valin 2005) with its bidirectional linking system. The agent framework has been implemented (Panesar 2019, 2020) as a model of conversational agents in a computational framework. It builds on the notion of a speech act from discourse within the functional RRG model

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

799

800

BRIAN NOLAN

of grammar, and the construction of a common ground in a discourse workspace to underpin the conversational interaction. The framework connects the intelligent conversational agent paradigm to the RRG model of language, leveraging the notion of constructions as grammatical objects. It builds on work on understanding constructions as grammatical objects within RRG (Nolan 2011, 2012, 2013; Diedrichsen 2011; Nolan and Diedrichsen 2013) and the role of computational approaches to functional grammars (Nolan and Periñán-Pascual 2014). This framework has potential for use in testing hypotheses to support claims of adequacy (Butler 2009) within an RRG approach. It extends our understanding of the grammar–pragmatics interface and the construction of an emergent common ground between interlocutors. The management framework for a language understanding system supporting conversational agents is seen in Figure 21.5, with the applicable sub-models, in particular with the language model based on RRG. Here, the natural language processing system is at the heart of the architecture. To build an intelligent conversational agent in software one needs to represent a number of important aspects of a conversational agent-based system. One needs to capture (i) the set of beliefs that the agent has at any

Figure 21.5 Dialogue management framework for a language understanding system supporting conversational agents and a dynamic common ground

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

given time, (ii) the goals that the agent will try to achieve, (iii) the actions that the agent performs, (iv) the knowledge of the effects of these actions, (v) the environment information the agent has (which may be incomplete or incorrect), (vi) the ongoing discourse interaction that the agent has with other (human) agents and their environment over time, (vii) human language understanding, and (viii) conversation tracking over a discourse. The parameters of the conversational agent are characterized in its (i) personality and emotional status, (ii) awareness of social context, (iii) model of the conversational partner (¼ other conversational actor), (iv) dialogue tracking, (v) agent dialogue-turn tracking, and (vi) linguistic information concerning languagespecific grammar and clause structure and speech acts. In the architecture outlined here, the relationship between agent parameters and behavioural choices made during the parse/generation of natural language conversation is governed by the Agent Dialogue Manager. Typically, this rule-based system consults other elements of the architecture to maintain the BDI state of the conversational agent, its personality and emotional status etc. to control a generalized natural language conversation system and supply the appropriate parameters for conversational behaviour. Importantly, a model of an agent system is distinguished by its active use of social concepts. Panesar (2017, 2019, 2020) designed and implemented this linguistically orientated, conversational software agent framework as a proof of concept. She discusses the relationship between natural language processing and knowledge representation, and connects this with the goals of RRG (Figure 21.6). In this implemented framework, Phase 1 is concerned with

Figure 21.6 Conceptual framework of the conversational software agent (from Panesar 2017: 190)

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

801

802

BRIAN NOLAN

the RRG language model. Phase 2 manages the agent cognitive model and its interfaces with the BDI model, planning model and knowledge model, while Phase 3 handles the agent dialogue model, including dialogue management via the RRG language model. The advantages provided by RRG (Panesar 2019) reside in its fitness-forpurpose for computational implementation and its level of computational adequacy. Panesar implements a computational model of the linking algorithm that utilizes a speech act construction as a grammatical object and the (BDI) sub-model of beliefs, desires and intentions (Rao and Georgeff 1995). This work builds on elements of earlier research on the RRG interlingua-based machine translation engine described earlier. This RRGbased conversational software agent framework has been successfully implemented in software (Panesar 2017, Pokahr et al. 2014) using conceptual graphs and the resource description framework (RDF). The innovation of this research resides in the combination of models and their interoperability. The research provides new insights into the interface between language and knowledge, via an RRG perspective.

21.4.2 Modelling Sign Language in an RRG Avatar Computational modelling using RRG has been applied to a humanoid avatar to motivate an extended architecture for the lexicon to accommodate the requirements of visual gestural sign languages (SL). Murtagh (2019a, 2019b) designed and implemented a signing avatar which used RRG as its linguistic model. Due to the visual gestural nature of Irish Sign Language (ISL), and indeed all sign languages, and the fact that ISL has no written or aural form, to communicate an ISL utterance in computational terms required the implementation of a humanoid avatar capable of movement within threedimensional (3D) space (Figure 21.7). In the creation of the avatar, it was

Figure 21.7 The avatar model (from Murtagh 2019b: 95)

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

necessary to define all of the ISL phonological parameters in computational terms. The lexicon extensions involve an enhancement to qualia structures and qualia theory (Pustejovsky 1991, 1995). For the representation of sign languages, Murtagh posited an articulatory structure level in a framework called Sign–A. This research is work in progress in the development of a full linguistically motivated computational framework for ISL, based on RRG, and involving the avatar. In characterizing a sign language, one needs to refer to the various sign articulators (hands, fingers, eyes, eyebrows, etc.), as these are what are used to articulate various phonemes, morphemes and lexemes of an utterance (Figure 21.8). According to Murtagh (2019a), the visual gestural realization of a word in SL involves the simultaneous and parallel expression of a varied number of manual features (MFs) and non-manual features (NMFs), together in concert over a timeline. The MF phonological parameters are defined as location, orientation, movement and handshape or relative configuration. The NMF phonological parameters are eyebrow movement, movement of the eyes and eyelids, mouth patterns, tongue movement, and blowing of the cheeks, and it also includes head tilting and shoulder movement. This description of the articulators and how they act in concert in an utterance is modelled in this avatar in RRG linguistic terms and represented in the lexicon. The new level of lexical representation, the articulatory structure level, is based on an extension of the qualia theory from the generative lexicon of Pustejovsky (1991, 1995) and caters specifically for computational linguistic phenomena consistent with signed languages, enabling the adequate representation of ISL within the RRG lexicon. To facilitate the development of this RRG-motivated computational framework for ISL, all of the ISL phonological parameters, for both MFs and NMFs, are rigorously described and defined in computational terms, taking into consideration a sign realization timeline, an event duration, and a 3D signing space. Any given ISL phoneme, morpheme or lexeme may be

Figure 21.8 The hand articulator in 3D space with nodes on fingers and thumb (from Murtagh 2019b: 98)

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

803

804

BRIAN NOLAN

realized simultaneously and in parallel along a timeline. The position of the hands in 3D space has consequences at a syntactic and also at a semantic level for ISL. Also, morphemes are articulated at particular points or loci in relation to the signer for pronominal and anaphoric reference. In addition, the order or linear sequence in which these units are realized is significant with regard to the syntax and semantics of the language. Therefore, all the information applicable to the signed lexical items must be accounted for in order to accurately represent the language. The parameters required to represent ISL can be explained more easily if one uses the analogy of instruments playing together in an orchestra, where the various articulators would be represented by the instruments and each articulator will play its own part in producing an overall production or articulation, similar to the instruments playing their parts in an orchestra. This research by Murtagh (2019a, 2019b) fuses a constructional perspective of important phenomena in SLs, in particular ISL, within the RRG context. The innovation within Murtagh’s research is that it leverages the development of an avatar within 3D space, employs RRG as the linguistic model, and provides computational phonological parameters capable of representing and realizing ISL lexemes and their articulatory structure levels within the extended architecture of the RRG lexicon. Timeline and event duration are included, as is a definition of the 3D signing space and the ISL body anchored parameters. This development, called the Sign–A framework by Murtagh, has the potential to be employed as a linguistically motivated application of avatar technology for sign languages, based on implementing RRG in software, and to be used for real-time text-to-sign machine translation.

21.5

Functional Grammar Knowledge Base Environment

21.5.1 The FunGramKB Initiative FunGramKB2 (‘Functional Grammar Knowledge Base’) is an online environment for the semi-automatic construction of a multipurpose lexicoconceptual knowledge base for natural language processing (NLP) systems and natural language understanding. It is multifunctional and multilingual, and designed to support many NLP tasks, including information retrieval and extraction, machine translation, dialogue-based systems, and so on, and with many natural languages (Guerra García and Sacramento Lechado 2014; Felices-Lago and Ureña Gómez-Moreno 2014; Montiel-Ponsoda and Aguado-de-Cea 2014; Sánchez-Cárdenas and Faber 2014). Periñán-Pascual and Arcas-Túnez (2010) describe the FunGramKB knowledge base as comprising three major knowledge levels, consisting of several independent but interrelated modules. At the lexical level, the

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

Lexicon stores morphosyntactic, pragmatic and collocational information about lexical units. The Morphicon helps the system to handle cases of inflectional morphology. At the grammatical level, the Grammaticon stores the constructional schemata which help RRG to construct the semantics-to-syntax linking algorithm. At the conceptual level, the FunGramKB Ontology is presented as a hierarchical catalogue of the concepts that a person has in mind. Semantic knowledge is stored in the form of meaning postulates. It consists of a general-purpose module (Core Ontology) and several domain-specific terminological modules (Satellite Ontologies). The Core Ontology is an IS-A conceptual hierarchy which allows nonmonotonic multiple inheritance. The Core Ontology takes the form of a universal concept taxonomy, where ‘universal’ means that every concept we imagine has, or can have, an appropriate place in the ontology. Importantly, the Core Ontology is linguistically motivated but not languagedependent. The FunGramKB Cognicon stores procedural knowledge by means of scripts (i.e. conceptual schemata in which a sequence of stereotypical actions is organized on the basis of temporal continuity). In the Cognicon, scripts are structured into predications within a linear temporal framework, based on a representation of time as a partially ordered graph where nodes represent events and arcs are tagged with one or more relations of temporal ordering. In FunGramKB, every predication included in a script represents an event E which is treated as an interval consisting of a pair of time points (i, t) (i.e. the start time-point (i) and the end time-point (t)). The FunGramKB Onomasticon stores information about instances of entities and events, and stores two different types of schemata (i.e. snapshots and stories), as instances can be portrayed synchronically or diachronically. The Onomasticon stores information about named entities and events (i.e. instances of concepts), in the form of bio-structures. The population of the Onomasticon is taking place semi-automatically, through the extraction of structured information from Wikipedia, to turn this information into a rich knowledge base of more than 2.6 million entities. This knowledge base is accessible on the Internet. In the FunGramKB architecture (Figure 21.9), every lexical or grammatical module is language-dependent, whereas every conceptual module is shared by all languages. In other words, a computational linguist must develop one lexicon, one morphicon and one grammaticon for English, one lexicon, one morphicon and one grammaticon for Spanish, and so on. Knowledge engineers build just one ontology, one cognicon and one onomasticon to process any language input conceptually. FunGramKB adopts a conceptualist approach as the ontology becomes the pivotal module for the whole architecture. The FunGramKB lexical and grammatical levels are grounded in the functional-cognitive linguistic theories of RRG and the related Lexical Constructional Model (LCM) (Ruiz de Mendoza and Mairal Usón 2008; Mairal Usón and Ruiz de Mendoza 2009), allowing the FunGramKB system to

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

805

806

BRIAN NOLAN

Figure 21.9 The architecture of FunGramKB (source: www.fungramkb.com)

capture syntactic-semantic generalizations which are able to provide both explanations and predictions of language phenomena. They state that, at the FunGramKB conceptual level, the model of a ‘scheme’ is fundamental to the representation of the world knowledge. In the knowledge base, conceptual schemata are classified according to the parameters of prototypicality and temporality. Conceptual representations can store prototypical knowledge or can serve to describe instances of entities or events. A key factor in successful reasoning within an NLP system is that all the knowledge schemata must be represented with the same formal language, so that effective information sharing can take place across all conceptual modules. In FunGramKB, this formal conceptual representation language is called COREL. Because the FunGramKB conceptual modules use COREL as the common internal language for schemata representation, natural language understanding systems will only require one common reasoner. In RRG, the semantic and the syntactic components are mapped via a linking algorithm, which includes a set of rules that account for the syntax– semantics interface. As a result, RRG allows an input text to be represented in terms of a logical structure. In FunGramKB, the RRG logical structure has been enhanced by a new formalism called the Conceptual Logical Structure (CLS) (Periñán-Pascual and Mairal Usón 2009). The benefits are that CLSs are real language-independent representations, since they are made of concepts and not words. One of the consequences of this interlingua approach is that redundancy is minimized while informativeness is maximized. The benefits

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

of interlingua-based systems in multilingual settings, which we have already discussed in relation to UniArab, are also highlighted by PeriñánPascual and Arcas-Túnez (2014), based on an implementation of RRG in the FunGramKB software. The inferential power of the reasoning engine is more robust if predictions are based on cognitive expectations. In order to perform some reasoning with the input, the CLS is transduced into a COREL representation, so that it can be enriched by the knowledge in meaning postulates, scripts, snapshots and stories. According to Ruiz de Mendoza and Mairal Usón (2008) and Mairal Usón and Ruiz de Mendoza (2009), as it is grounded within the RRG framework, the LCM goes beyond the core grammar and incorporates meaning dimensions that have a long tradition in pragmatics and discourse analysis. The LCM recognizes four levels of constructional meaning, and in the Grammaticon, each one of these constructional levels is computationally implemented into a Constructicon. Level 1 (argumental layer) provides the core grammatical properties of lexical items. Level 2 (implicational layer) is concerned with the inferred meaning related to low-level situational cognitive models (or specific scenarios), which give rise to meaning implications of the kind that have been traditionally handled as part of pragmatics through implicature theory. Level 3 (illocutionary layer) deals with traditional illocutionary force, which is considered a matter of high-level situational models (or generic scenarios). Level 4 (discourse layer) addresses the discourse aspects, with particular emphasis on cohesion and coherence phenomena. The FunGramKB Suite provides a set of user-friendly tools to browse, check and edit the knowledge base. Specifically, conceptual, lexical and grammatical modules can be browsed via a GUI, displaying specific feature-value information about their elements. At this point in time, FunGramKB is a mature multipurpose knowledge base for NLP systems. Its goal is the creation of an NLP knowledge base sufficiently robust to help language engineers to design intelligent natural language understanding systems. The FunGramKB conceptual level enables the full integration of semantic, procedural and episodic knowledge by sharing both the knowledge representation language and the reasoning engine. As a result, expectations on the occurrence of typical events in a given situation are based on COREL schemata, a concept-oriented interlingua whose inferential power is greater than the traditional approach to lexical semantics. The FunGramKB lexicogrammatical levels are grounded in RRG to capture syntactic–semantic generalizations which can manage and interpret data. Both the RRG and the related LCM frameworks inspired the construction of the CLS, a lexically driven interlingua through which the system is able to predict a wide range of linguistic phenomena in the language generation process. The CLS serves as the pivot language between the input text and the COREL representation, while COREL serves as the pivot language between the CLS and the automated reasoner.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

807

808

BRIAN NOLAN

21.5.2 FunGramKB and ARTEMIS Another project which aims to provide a fully implemented computational model is ARTEMIS (‘Automatically Representing Text Meaning via an Interlingua-Based System’) within FunGramKB. It exists as an application within the environment of FunGramKB, implemented with a parser for the computational treatment of the syntax and semantics of sentences. ARTEMIS deploys the FunGramKB knowledge base which exploits constructional schemata as a system for meaning representation (Periñán-Pascual and Arcas-Túnez 2004, 2005, 2007, 2008, 2010; Periñán-Pascual and Mairal Usón 2009, 2010, 2011; Periñán-Pascual 2013; Cortés-Rodríguez 2016a, 2016b; Díaz Galán and Fumero Pérez 2016, 2017; Fumero Pérez and Díaz Galán 2017). ARTEMIS is an NLP application, with a syntactic parser, designed primarily for natural language understanding. ARTEMIS is linguistically motivated and adopts the RRG linguistic theory as a foundation (Fumero Pérez and Díaz Galán 2017; Díaz Galán and Fumero Pérez 2017). For ARTEMIS, RRG has a number of features that make it suitable for NLP work. 1. RRG is a model where morphosyntactic structures and grammatical rules are explained in relation to their semantic and communicative functions. 2. RRG is a monostratal theory where the syntactic and semantic components are connected via a bidirectional linking algorithm. 3. RRG is a model that makes strong claims to typological adequacy. 4. RRG has a model of the lexicon connected with the linking system. ARTEMIS is a linguistically grounded application, from both a semantic and syntactic perspective. Theoretically, it draws from two functional linguistic theories, those of RRG and the LCM mentioned earlier. ARTEMIS is able to formalize meaning by representing it as a logical structure, and its functional approach entails that grammar can only be explained through the interaction of syntax and semantics, which is achieved via a bidirectional linking algorithm. The LCM allows ARTEMIS to account for compositional meaning, providing a machine-tractable representation of constructions, the constructional schemata. One important element that binds FunGramKB with RRG and the LCM is the integration of the LCM’s constructional templates and RRG’s lexical representations / logical structures into the knowledge base’s language-independent formalism for text meaning representation, the CLS. The architecture of ARTEMIS has been designed around three main components. The first of these is the grammar development environment (GDE), comprising the set of syntactic, constructional and lexical rules necessary for the parsing of natural language expressions. The GDE integrates two types of construct: a catalogue of attribute-value matrixes (AVMs) to describe grammatical units, and a set of production rules (grammatical, lexical and constructional) to allow it to produce a feature-based grammar. The second component is the CLS constructor, which produces an initial CLS text meaning representation. This is essentially an advancement of RRG’s logical

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

structures. The third component is the COREL scheme builder, which transforms the CLS into the formal FunGramKB representation language (which we mentioned earlier, called COREL), ultimately arriving at an extended COREL scheme. Rules in ARTEMIS can be divided into syntactic rules to account for the generation/recognition of the underlying layered structure of a clause (LSC), constructional rules which guide the embedding of the structure of argument-predicate constructions (L1 constructions in the LCM) and lexical rules to tokenize the abstract features encoded in the LSC by utilizing the information stored in the FunGramKB Lexicon and Ontology. The process involved in understanding natural language with the ARTEMIS tools is summarized in the following workflow: ARTEMIS: [USER] ➔ text ➔ CLS representation ➔ COREL Scheme ➔ [Reasoner]

Whereas RRG is able to connect the lexical entries of verbs with their syntactic realizations via the linking algorithm, it cannot satisfactorily account for constructional meaning. This requirement to cater for constructions motivated some adjustments to the RRG model. ARTEMIS required a constructionist linguistic model, the LCM, to be able to process this type of meaning, since, at sentence level, argument constructions can take precedence over core verbal semantics and may alter the argument structure of the predicate. To account for such a possibility, Periñán-Pascual and Arcas Túnez (2014) proposed and implemented in ARTEMIS a modification of the RRG LSC by adding a new CONSTR-L1 node between the clause and the core nodes. Cortés-Rodríguez and Mairal Usón (2016) further consider that the addition of this new node entails the redefinition of the original RRG precore slot position as a pre-C-L1 position. The rationale for this change is that the pre-core slot may encompass not only those core constituents stated by RRG, fronted and interrogative elements, but also constituents which are triggered by a construction. Another necessary adjustment to the RRG model derives from the fact that ARTEMIS shares some characteristics of unification grammars, in such a way that parsing relies not only on syntactic rules but also on the semantic and grammatical information contained in the AVMs. A consequence of this is that, whereas in RRG abstract grammatical categories such as illocutionary force, aspect or negation are described in the operator projection, in ARTEMIS these values, and the function words associated with them, are represented in the form of feature-bearing matrices, which now belong to the constituent projection. In this approach, every grammatical category in the LSC has to be methodically described by listing the attributes that define it in a feature-bearing matrix. At the same time, these attributes need their own description in another AVM, in such a way that, whenever a category, an attribute or a new part of speech (POS) is introduced, the corresponding AVM must be created and stored in the catalogue of AVMs in the GDE. The attributes that characterize each of the categories in the AVM, which would belong in the RRG operator projection, should also in turn be defined by AVMs.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

809

810

BRIAN NOLAN

One of the key differences between the linguistic models adopted by ARTEMIS and RRG is that, in ARTEMIS, constructional meaning is a defining feature. Constructional meaning in ARTEMIS can be derived from the information contained in the core grammar of the verb available in the Lexicon, in the form of lexical templates, and from the constructional schemata contained in the Grammaticon. This module compiles the description of constructions as AVMs which enumerate the main features and establish the set of constraints that characterize each construction. Following the LCM, the Grammaticon stores four types of construction: argumental (L1), implicational (L2), illocutionary (L3) and discursive (L4). To capture the difference between the information provided by the Lexicon and the Grammaticon, Periñán-Pascual and Arcas-Túnez (2014: 181–187) make a distinction between kernel constructions and other argumental constructions. The kernel constructions are determined by the semantics of the verb, which, depending on the variables in the lexical template stored in the Lexicon, can be: Kernel 0 (zero argument verbs), Kernel 1 (intransitive), Kernel 2 (monotransitive) and Kernel 3 (ditransitive), The second set of constructions are those argumental constructions which cannot be derived from the lexical templates. ARTEMIS is the first system to employ a substantial knowledge base to generate a full rich logical structure for use by NLP applications requiring language comprehension capabilities (Pascual and Arcas-Túnez 2014: 171). The research by the FunGramKB team (Mairal Usón and Cortés-Rodríguez 2017) has been concerned with the development of a natural language processing laboratory using tools developed within the framework of RRG. As a result, a number of computational resources have been built with a focus on different domains. ARTEMIS is one of the computational resources that form part of the research project FUNK Lab, which uses the linguistic theories of RRG and the LCM. The different domains are: 1. ARTEMIS, Automatically Representing Text Meaning via an Interlinguabased System, is a computational resource which allows for the automatic generation of a conceptual logical structure, that is, a fully specified semantic representation of an input text. 2. Navigator, a tool that allows a user to retrieve data from the lexical entries in the English lexicon (e.g. morphosyntactic, pragmatic and collocational information) and from the conceptual entries in the core ontology (e.g. thematic frame, meaning postulate), as developed within the framework of the FunGramKB project. This resource allows the user to navigate through the linguistic (the English lexicon) and the conceptual levels (the ontology). 3. RONDA, RecOgniziNg Domains with IATE, is a tool used to categorize a text or a collection of documents in different specialized domains. 4. CASPER, CAtegory and Sentiment-based Problem FindER, is a resource to analyse micro-texts (e.g. tweets) for the automatic detection of userdefined problems by following a symbolic approach to topic categorization and sentiment analysis.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

5. DAMIEN, DAta MIning Encountered, is a workbench that allows researchers to do text analytics by integrating corpus-based processing with statistical analysis and machine-learning models for data-mining tasks. 6. DEXTER, Discovering and Extracting TERminology, is a tool developed as an online multilingual workbench. It is provided with a suite of tools for (i) the compilation and management of small and medium-sized corpora, (ii) the indexation and retrieval of documents, (iii) the elaboration of queries by means of regular expressions, (iv) the exploration of the corpus, and (v) the identification and extraction of term candidates (i.e. unigrams, bigrams and trigrams). The future direction of the work of the FUNK lab, including ARTEMIS, is to move towards a knowledge-informed intelligent cognitive agent which has RRG as the linguistic model. This research is ongoing.

21.6

The RRG Contribution to NLP

The RRG theoretical framework creates a relationship between syntax and semantics and can account for how semantic representations are mapped into syntactic representations. There are several properties of RRG as a theory that make it well suited to computational implementation. First, the theory has a semi-formal logical formalism. This enables the development and maintenance of software systems that interpret the formalism. Implementing the RRG theory in software requires translating its semiformal logical formalism into code. In particular, this translation of the theory into code poses challenges in the implementation of the lexicon and the bidirectional linking system of RRG. Furthermore, because the linking system in RRG is bidirectional, an implemented grammar can be used for both parsing and generation. As well as its typological adequacy, RRG has a level of computational adequacy which allows for the creation of algorithms that are computationally tractable. This enables RRG grammars to be used in a range of applications, and enables linguists to test hypotheses. The languages that have undergone some level of computational treatment in RRG include English, Irish, Modern Standard Arabic, Irish Sign Language, Biblical Hebrew, German and the Romance languages (Spanish, France, Italian, Romanian). Sign languages (Murtagh 2019a, 2019b), as visual gestural languages, pose interesting problems for both functional models of grammar (Leeson and Nolan 2008) and computational models of these languages. A consequence of this computational work has been the enrichment of the theoretical elements of the RRG theory as a functional model, especially in its semantic and lexical underpinnings, where they connect with concepts, and the building of software applications that demonstrate its viability in natural language processing. This computational work provides compelling evidence that the functional approach of RRG has a

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

811

812

BRIAN NOLAN

positive and crucial role to play in natural language processing. RRG delivers a credible and realistic linguistic model to underpin the kinds of NLP applications discussed in this chapter, and also in applications with a functional typology orientation. Why should a grammar deal with linking from syntax to semantics at all? Should not specifying the possible realizations of a particular semantic representation suffice? The answer of course is ‘no’. Functional linguists of all persuasions refute this using the argument that theories of linguistic structure should be directly relatable to testable theories of language production and comprehension. Computational models assist in this process. Based on our experience to date with computational models of RRG, we can verify that RRG is suitable as a basis for computational implementation in software. Computational models of RRG serve as a rich testing ground for the linguistic theory. Indeed, Butler and Gonzálvez-García (2014: 451) note that RRG is the only functional model in what they call ‘functional-cognitive space’ that has a number of ongoing computational research activities. Butler and Gonzálvez-García note in particular several of the case studies discussed here (Guest 2008; Gottschalk 2012, 2019; Nolan 2004; Nolan and Salem 2010, 2011; Murtagh 2019a, 2019b). In addition, recent work in connection with the RRG-inspired FunGramKB knowledge base has seen important advances in the computational implementation of RRG-based procedures. Butler and Gonzálvez-García consider these studies as representing an important growing trend in RRG (see also Butler and Martín Arista 2009). An important question (Nolan 2016), from a computational RRG perspective is: ‘What can theoretical linguistics do for natural language processing research?’ In order to be considered fit for purpose, certain levels of adequacy are expected from contemporary linguistic models (Butler 2009). These levels span the descriptive, typological, psychological and explanatory adequacy required in a viable model of language. In today’s world we can consider that another level of adequacy is also required, that of computational adequacy. This is a concept well known to those scientists and professionals working within computer science and software development. A computer system and its underlying model must be fit for purpose as specified. It must deliver its results in a coherent, timely and efficient manner with efficient utilization of available resources. In linguistic modelling terms, a linguist is concerned with the model of the computation in the mind of a person who actually computes the link between the concept– semantics–syntax interfaces in an utterance. In computational linguistic terms, however, this model is then a subset of the model of grammar used, and once implemented in software, its systematic behaviour. Specifically, considered as a computational system, does the model work to deliver the correct and expected results optimally and efficiently? An implication of the requirement of computational adequacy for a linguistic model implemented in software is that one must address the appropriate levels of granularity required for software specifications in order to actually model

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

a grammar. This extra level of detail needed to specify the model of grammar such that it is understandable by a computer system puts additional and complex demands on the linguist to reach the correct levels of granularity and formalized precision so that the model can be programmed. For RRG, where syntax is not viewed as autonomous but semantically motivated, the scale of the challenge is increased. However, as a scientific enterprise, it is worthwhile as it has the potential to model the computation of language from lexicon, semantics and morphosyntax. In functionally motivated models of language a grammar–lexicon–construction continuum exists and the interfaces between these need to be rigorously expressed (Nolan and Periñán-Pascual 2014; Nolan and Salem 2011; Nolan 2014). As a way of treating linguistic and computational complexity in pursuit of linguistic realism with a robust cultural awareness of linguistic conventions, this scientific enterprise of creating linguistically motived language software and computational models of grammar is crucial. The combination of computer science with RRG-style linguistics has the potential to create many useful innovations that will cause a phase shift in our expectations (Nolan and Diedrichsen 2019; Kailuweit et al. 2019). As it has become increasingly pervasive in our lives, human language technology is now a central component of computer science and computational linguistics in its treatment of natural language in software. Central to future research and development in this area is the convergence of knowledge of computer science and linguistics and other related disciplines. Crucially, the key ingredients are an understanding of linguistics and how languages work to meet the levels of adequacy needed to be successful and accepted, and to guide computer science in application development and the engineering of solutions of value to us. The functionally oriented RRG theory of linguistics can contribute considerably to NLP research. One way to measure a syntactic theory in practical computational applications is to consider the computational complexity of its implemented parsing and generation algorithms. At its present stage of development, computational work in RRG has tended not to address issues relating to computational complexity at a theoretical or empirical level. To date, the focus of computational work has been on the creation of RRG-based software applications that serve some particular purpose. Consequently, computational work on RRG has not focused theoretically on the science of parsing, concentrating instead on implementing linguistically motivated parsers for particular applications and a variety of languages. One exception to this is found in Kallmeyer et al. (2013), which describes a body of research on the computational formalization of RRG that uses the building blocks provided by the theory of Tree Adjoining Grammar (TAG), with modified operations for combining trees. In this, Kallmeyer et al. (2013: 175) report on a tree rewriting system, called Tree Wrapping Grammar (TWG), that captures the basic tree-composition principles of RRG, with the

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

813

814

BRIAN NOLAN

aim of formalizing the composition of syntactic templates and their compositional mechanisms, while drawing on concepts from TAG theory. The authors provide two operations for syntactic composition, which they call (wrapping) substitution and sister adjunction. The first operation provides a model of argument insertion as well as the construction of long-distance dependencies. The second operation implements adjunction to non-binary trees. Their approach to tree construction and parsing is part of a larger project that aims at a full formalization of RRG. An advantage of such a formalization is that it facilitates a computational treatment of the grammar. The authors define a Cocke–Younger–Kasami (CYK) parsing schema for tree wrapping grammars, which can be employed for RRG parsing. Such a formalization of the grammatical structures of RRG consists of two parts (Kallmeyer et al. 2013: 177): the specification of the elementary trees and the definition of the compositional operations acting on them. Kallmeyer et al. (2013: 183) provide a formal definition of the elementary trees and the composition operations (wrapping) substitution and sister adjunction that operate on them in the form of a TWG. The syntactic inventory of RRG can be characterized as a formal grammar of this type. A finite set of elementary trees is specified via a meta-grammar, including the formulation of RRG’s linking theory and the specification of different syntactic realizations for a particular valence frame. These elementary trees are then combined into larger trees using the TWG. According to the authors, the TWG can generate all context-free languages, and cross-serial dependencies can be described (Kallmeyer et al. 2013: 189). This work contributes to the formalization of RRG, and, clearly, this is a direction for future research. Various projects have developed computational resources on the basis of RRG, in particular the FunGramKB project, which is quite mature at this stage. In this chapter, we have sketched the landscape of computational work in RRG. We have discussed how RRG as a theory supports this computational work, described a number of computational projects that use RRG, and highlighted some implemented software systems using RRG. These applications and their target languages are quite diverse. RRG grammars have, in the main, been used for practical tasks, where the syntactic and semantic analyses provide useful features. In one important NLP task, machine translation uses a grammar to parse and also to generate. We explored ways in which computational work can inform linguistic research and indicated some areas of future research direction. This field is dynamic and we hope this chapter will encourage interested linguists to engage in RRG research in computational linguistics.

References Butler, C. S. 2009. Criteria of adequacy in functional linguistics. Folia Linguistica: Acta Societas Linguistica Europaea 42(1): 1–66.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

Butler, C. S. and F. Gonzálvez-García., F. 2014. Exploring Functional-Cognitive Space (Studies in Language Companion Series 157). Amsterdam: John Benjamins. Butler, C. S. and J. Martín Arista (eds.). 2009. Deconstructing Constructions. (Studies in Language Companion Series 107). Amsterdam: John Benjamins. Copestake, A. 2002. Implementing Typed Feature Structure Grammars (Center for the Study of Language and Information Publication Lecture Notes). Stanford, CA: CSLI publications. Cortés-Rodríguez, F. 2016a. Parsing simple clauses within ARTEMIS: The computational treatment of the layered structure of the clause in Role and Reference Grammar. In Proceedings of the 34th International Conference of AESLA, Alicante, 14–16. Cortés-Rodríguez, F. 2016b. Towards the computational implementation of Role and Reference Grammar: Rules for the syntactic parsing of RRG phrasal constituents. Círculo de Lingüística Aplicada a la Comunicación (CLAC). 65: 75–108. Cortés-Rodríguez, F. and R. Mairal Usón. 2016. Building an RRG computational grammar. Onomazein 34: 86–117. Díaz Galán, A. and M. Fumero Pérez. 2016. Developing parsing rules within ARTEMIS: The case of Do auxiliary insertion. In C. Periñán-Pascual and E. Mestre-Mestre (eds.), Understanding Meaning and Knowledge Representation: From Theoretical and Cognitive Linguistics to Natural Language Processing, 283–302. Newcastle upon Tyne: Cambridge Scholars Publishing. Díaz Galán, A. and Fumero Pérez, M. 2017. ARTEMIS: state of the art and future horizons. Revista de Lenguas para Fines Específicos 23(2): 16–40. https:// ojsspdc.ulpgc.es/ojs/index.php/LFE/article/view/917. Diedrichsen, E. 2008. Where is the PreCore slot? Mapping the layered structure of the clause and German sentence topology. In Van Valin (ed.), 203–224. Diedrichsen, E. 2011. The theoretical importance of constructional schemas in RRG. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 168–199. Newcastle upon Tyne: Cambridge Scholars Publishing. Diedrichsen, E. 2014. A Role and Reference Grammar parser for German. In Nolan and Periñán (eds.), 105–142. Dorr, B. J. 1992. Machine translation divergences: A lexical semantic perspective. Proceedings of the Second Seminar on Computational Lexical Semantics, 169–183. Toulouse, France. Dorr, B. J., E. H. Hovy and L. S. Levin. 2006. Machine translation: Interlingual methods. In Natural Language Processing and Machine Translation Encyclopedia of Language and Linguistics (2nd ed.). http://citeseerx.ist.psu.edu/viewdoc/ summary?doi¼10.1.1.58.532. Felices-Lago, A. and P. Ureña Gómez-Moreno. 2014. FunGramKB term extractor: A tool for building terminological ontologies from specialised corpora. In Nolan and Periñán-Pascual (eds.), 251–270.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

815

816

BRIAN NOLAN

Fumero Pérez, M. and A. Díaz Galán. 2017. The interaction of parsing rules and argument-predicate constructions: Implications for the structure of the Grammaticon in FunGramKB. Revista de Lingüística y Lenguas Aplicadas 12: 33–44. Gottschalk, J. 2012. The persuasive tutor: A BDI teaching agent with Role and Reference Grammar language interface – Sustainable design of a conversational agent for language learning. ITB Journal 13(2): 31–50. https://arrow .tudublin.ie/itbj/vol13/iss2/3. Gottschalk, J. 2014. Three-place predicates in RRG: A computational approach. In Nolan and Periñán-Pascual (eds.), 79–104. Gottschalk, J. 2019. On the application of conceptual graphs in RRG: First steps towards a functional computational processing model. In R. Kailuweit, L. Künkel and E. Staudinger (eds.), Applying and Expanding Role and Reference Grammar, 325–354. Freiburg: NIHIN Studies. https://freidok.unifreiburg.de/data/16830. Guerra García, F. and E. Sacramento Lechado. 2014. Exploring the thematicframe mapping in FunGramKB. In Nolan and Periñán-Pascual (eds.), 233–250. Guest, E. 2008. Parsing for Role and Reference Grammar. In Van Valin (ed.), 435–453. Jackendoff, R. S. 1990. Semantic Structures. Cambridge, MA: MIT Press. Kailuweit R, L. Künkel and E. Staudinger (eds.). 2019. Applying and Expanding Role and Reference Grammar. Freiburg: NIHIN Studies. https://freidok.unifreiburg.de/data/16830. Kallmeyer, L., R. Osswald and R. D. Van Valin, Jr. 2013. Tree wrapping for Role and Reference Grammar. In G. Morrill and M.-J. Nederhof (eds.), Proceedings of the Formal Grammar 17th and 18th International Conferences, 175–190. Heidelberg: Springer. Kecskes, I. and F. Zhang. 2009. Activating, seeking, and creating common ground: A socio-cognitive approach. Pragmatics & Cognition 17:(2): 331–355. Leeson, L. and B. Nolan. 2008. Digital deployment of the Signs of Ireland Corpus in Elearning. In Language Resources and Evaluation LREC2008 – 3rd Workshop on the Representation and Processing of Sign Languages: Construction and Exploitation of Sign Language Corpora, Marrakech, Morocco. Mairal Usón, R. and F. Cortés-Rodríguez. 2017. Automatically representing text meaning via an interlingua-based system (ARTEMIS): A further step towards the computational representation of RRG. Journal of ComputerAssisted Linguistic Research 1: 61–87. Mairal Usón, R. and F. Ruiz de Mendoza. 2009. Levels of description and explanation in meaning construction. In C. S. Butler and J. Martín Arista (eds.), Deconstructing Constructions (Studies in Language Companion Series 107), 153–198. Amsterdam: John Benjamins. Montiel-Ponsoda, E. and G. Aguado-de-Cea. 2014. Applying the lexical constructional model to ontology building. In Nolan and Periñán-Pascual (eds.), 313–338.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

Murtagh, I. 2019a. Motivating the computational phonological parameters of an Irish Sign Language avatar. In Nolan and Diedrichsen (eds.), 323–339. Murtagh, I. 2019b. A Linguistically Motivated Computational Framework for Irish Sign Language. PhD thesis, Trinity College Dublin School of Linguistic Speech & Communication Science. www.tara.tcd.ie/handle/2262/89131. Nolan, B. 2004. First steps towards a computational Role and Reference Grammar. In Proceedings of the International Conference on Role and Reference Grammar, Dublin, July 2004, 196–223. https://rrg.caset.buffalo.edu/rrg/ RRG2004%20Book%20of%20Proceedings.pdf. Nolan, B. 2011. Meaning construction and grammatical inflection in the layered structure of the Irish word: An RRG account of morphological constructions. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 64–103. Newcastle upon Tyne: Cambridge Scholars Publishing. Nolan, B. 2012. The Syntax of Modern Irish: A Functional Account. Sheffield: Equinox. Nolan, B. 2013. Constructions as grammatical objects: A case study of the prepositional ditransitive construction in Modern Irish. In B. Nolan and E. Diedrichsen (eds.), Linking Constructions into Functional Linguistics, (Studies in Language Companion Series 145), 143–178. Amsterdam: John Benjamins. Nolan, B. 2014. Extending a lexicalist functional grammar through speech acts, constructions and conversational software agents. In Nolan and Periñán (eds.), 143–164. Nolan, B. 2016. What can theoretical linguistics do for natural language processing research? In C. Periñán-Pascual and E. Mestre Mestre (eds.). Meaning and Knowledge Representation, 235–248. Newcastle upon Tyne: Cambridge Scholars Publishing. Nolan, B. and E. Diedrichsen (eds.). 2013. Linking Constructions into Functional Linguistics: The Role of Constructions in Grammar (Studies in Language Companion Series 145). Amsterdam: John Benjamins. Nolan, B. and E. Diedrichsen (eds.). 2019. Linguistic Perspectives on the Construction of Meaning and Knowledge: The Linguistic, Pragmatic, Ontological and Computational Dimensions. Newcastle upon Tyne: Cambridge Scholars Publishing. Nolan, B. and C. Periñán-Pascual (eds.). 2014. Language Processing and Grammars: The Role of Functionally Oriented Computational Models (Studies in Language Companion Series 150). Amsterdam: John Benjamins. Nolan, B. and Y. Salem. 2010. UniArab: An RRG Arabic-to-English Machine Translation Software. In W. Nakamura (ed.), Proceedings of the 10th International Conference on Role and Reference Grammar (RRG 2009), 243–270. linguistics.buffalo.edu/people/faculty/vanvalin/rrg/ProceedingsofRRG2009_02.pdf. Nolan, B. and Y. Salem. 2011. UniArab: RRG Arabic-to-English Machine Translation. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 312–346. Newcastle upon Tyne: Cambridge Scholars Publishing. Panesar, K. 2017. A Linguistically Centred Text-Based Conversational Software Agent. Unpublished PhD thesis, Leeds Beckett University. www.researchgate.net/

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

817

818

BRIAN NOLAN

publication/319376849_PhD_Thesis_%27A_linguistically_centred_textbased_conversational_software_agent%27. Panesar, K. 2019. Functional linguistic based motivations for a conversational software agent. In Nolan and Diedrichsen (eds.), 340–371. Panesar, K. 2020. Conversational artificial intelligence: Demystifying statistical vs. linguistic NLP solutions. Journal of Computer-Assisted Linguistic Research 4: 47–79. Periñán-Pascual, C. 2013. Towards a model of constructional meaning for natural language understanding. In Nolan and Diedrichsen (eds.), 205–230. Periñán-Pascual, C. and F. Arcas-Túnez. 2004. Meaning postulates in a lexicoconceptual knowledge base. In Proceedings of the 15th International Workshop on Databases and Expert Systems Applications, 8–42. Los Alamitos, CA: IEEE. Periñán-Pascual, C. and F. Arcas-Túnez. 2005. Microconceptual-knowledge spreading in FunGramKB. In Proceedings of the 9th IASTED International Conference on Artificial Intelligence and Soft Computing, 239–244. Anaheim: ACTA Press. Periñán-Pascual, C. and F. Arcas-Túnez. 2007. Deep semantics in an NLP knowledge base. In Proceedings of the 12th Conference of the Spanish Association for Artificial Intelligence, Universidad de Salamanca, 279–288. Periñán-Pascual, C. and F. Arcas-Túnez. 2008. A cognitive approach to qualities for NLP. Procesamiento del Lenguaje Natural 41: 137–144. Periñán-Pascual, C. and F. Arcas-Túnez. 2010. The architecture of FunGramKB. In Proceedings of the International Conference on Language Resources and Evaluation, LREC 2010, Malta, 2667–2674. www.lrec-conf.org/proceed ings/lrec2010/summaries/284.html. Periñán-Pascual, C. and F. Arcas-Túnez. 2014. The implementation of the FunGramKB CLS Constructor in ARTEMIS. In Nolan and Periñán-Pascual (eds.), 165–196. Periñán-Pascual, C. and R. Mairal Usón. 2009. Bringing Role and Reference Grammar to natural language understanding. Procesamiento del Lenguaje Natural 43: 265–273. Periñán-Pascual, C. and R. Mairal Usón. 2010. Enhancing UniArab with FunGramKB. Procesamiento del Lenguaje Natural 44: 19–26. Periñán-Pascual, C. and R. Mairal Usón. 2011. The COHERENT methodology in FunGramKB. Onomázein 24: 13–33. Pokahr, A., L. Braubach, C. Haubeck and J. Ladiges. 2014. Programming BDI agents with pure Java. In German Conference on Multiagent System Technologies, MATES2014, 216–233. Switzerland: Springer International Publishing. Pustejovsky, J. 1991. The generative lexicon. Computational Linguistics 17(4): 409–441. Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Rao, A. S. and M. P. Georgeff. 1995. BDI agents: From theory to practice. In Proceedings of the 1st International Conference on Multi-Agent Systems ICMAS95, 312–319.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Computational Implementation and Applications

Ruiz de Mendoza, F. and R. Mairal Usón. 2008. Levels of description and constraining factors in meaning construction: an introduction to the Lexical Constructional Model. Folia Linguistica 42(2): 355–400. Sánchez-Cárdenas, B. and P. Faber. 2014. A functional and constructional approach for specialized knowledge resources. In Nolan and Periñán-Pascual (eds.), 297–312. Searle, J. 1969. Speech Acts. Cambridge: Cambridge University Press. Van Valin, R. D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, R. D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–Pragmatics Interface. (Studies in Language Companion Series 105). Amsterdam: John Benjamins. Van Valin, R. D., Jr. and E. Diedrichsen. 2006. A Bonsai Grammar for German. www.academia.edu/26126987/A_Bonsai_Grammar_for_German. Vauquois, B. 1968. A survey of formal grammars and algorithms for recognition and transformation in machine translation. Proceedings of the IFIP Congress-6, 254–260. Winther-Nielsen, N. 2008. A Role-Lexical Module (RLM) for Biblical Hebrew: A mapping tool for RRG and WordNet. In Van Valin (ed.), 455–478.

Notes 1 Available from SIL: https://software.sil.org/toolbox/. 2 www.fungramkb.com.

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

819

https://doi.org/10.1017/9781316418086.026 Published online by Cambridge University Press

Part Five

Grammatical Sketches

https://doi.org/10.1017/9781316418086.027 Published online by Cambridge University Press

https://doi.org/10.1017/9781316418086.027 Published online by Cambridge University Press

22 A Grammatical Sketch of Cheyenne (Plains Algonquian, USA) Avelino Corral Esteban Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: 1 2 3 4 12 A

first singular second singular third (proximate) singular fourth (obviative) singular first person plural inclusive actor

IF II IMPER IN INAN INFR

AN ANA ASP CISL CLM CNJ CON DEIC DIR DITR

animate anaphoric aspect cislocative clause linkage marker Conjunct order conjectural mode deictic directional ditransitive

LS MOD N NMR OBV PoDP PRET PrCS PrDP PSA

ECS

extra-core slot

REP

END EQ EPEN EVID EXO

endophoric event quantification epenthetic evidentiality exophoric

RP TEMP TNS TRNS U

illocutionary force inanimate plural impersonal construction initial stem inanimate inferential or dubitative mode logical structure modality nominal stem non-macrorole obviative post-detached position preterite or narrative mode pre-core slot pre-detached position privileged syntactic argument reportative or attributive mode reference phrase temporal tense translocative undergoer

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

824

AVELINO CORRAL ESTEBAN

FAI FII FTA FTI I

22.1

animate intransitive final stem inanimate intransitive final stem animate transitive final stem inanimate transitive final stem inanimate singular

VAI

animate intransitive verb

VII VTA

inanimate intransitive verb animate transitive verb

VTI

transitive inanimate verb

X

unspecified actor

Introduction

This chapter examines the grammar of Cheyenne or Ts˙ehésen˙estsest˙otse (Plains Algonquian, USA) within the framework of Role and Reference Grammar (henceforth RRG) (Van Valin and LaPolla 1997; Van Valin 2005).* Although this grammatical sketch aims to cover as much ground as possible, for reasons of brevity, it only covers in some depth the interaction of syntax, semantics, and pragmatics in simple sentences. Thus, by way of illustrating the pervasive influence of pragmatics in all areas of Cheyenne grammar, the last part of the paper includes an account of the relation between word order and information structure, and the influence of pragmatics on morphosyntactic coding and macrorole assignment. The organization of the chapter is as follows: after a brief account of the principal morphosyntactic features of Cheyenne, Section 22.2 offers a description of a number of grammatical areas which benefit from analysis in RRG terms. Section 22.3 provides an account of the relationship between information structure and two linguistic mechanisms which are particularly typical of Algonquian languages, namely the proximate/obviation distinction and the direct/inverse system. These serve to mediate between macrorole assignment and morphosyntactic coding and, consequently, compensate for the lack of a rigid word order and a proper case-marking system in Cheyenne.

22.2

Grammatical Sketch

After introducing a number of key features of Cheyenne grammar (22.2.1), in this section we discuss basic clause types (22.2.2), verb valence and transitivity (22.2.3), the marking of core arguments, argument-adjuncts and adjuncts (22.2.4, 22.2.5) and the linking algorithm (22.2.6).

* The data come mainly from my native consultants, supplemented by existing language materials, such as two collections of texts (Leman 1980a and 1987) and a dictionary (Fisher et al. 2006). I wish to express my gratitude to the Cheyenne people, especially my late friend Ralph Redfox, for kindly sharing their knowledge of this language with me. Needless to say, all errors remain my sole responsibility.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

825

22.2.1 Basic Morphosyntactic Properties of Cheyenne Cheyenne is a polysynthetic language which, like other Algonquian languages, is believed to exhibit a pragmatically influenced word order (Leman 1999: 37), which is concerned with the way in which the message is formulated, rather than with the identification of semantic roles or syntactic functions. In accordance with the concept of ‘newsworthiness’ (Mithun 1987), which appears to imply importance and unpredictability, the language tends to place any elements expressing relevant information preverbally and any elements expressing trivial information postverbally. Owing to the pronominal-argument and head-marking nature of Cheyenne, verbs have an extremely complex templatic structure, which not only includes bound forms covering the grammatical information of the core arguments of the predicate, but also derivational stems and particles expressing grammatical categories like aspect, tense, modality, etc., which are treated as operators modifying the different layers of the clause, as shown in Figure 22.1 (modified from Corral Esteban 2017: 310). The relative ordering of the elements within the verbal complex is as follows. The person cross-referenced on the verbal prefix of a main clause always refers to the most pragmatically salient participant, in accordance with the Person Hierarchy 2 > 1 > 3 > 4 > I(inanimate).1 The preverbal operators between the prefix and the verb stem express the realis/irrealis distinction, past or future tense, epistemic modality, event quantification, negation, deontic modality, direction,2 and the different aspectual distinctions. Predicates can be simple or complex, depending on whether they comprise one single stem or more than one stem. The morphemes constituting a complex predicate, which incorporate adverbial, nominal and verbal meanings into the verb, are commonly referred to as initial, medial and final, in accordance with their relative position within the verbal complex (Bloomfield 1946). The theme marker indicates voice or, more accurately in Algonquian terminology, whether the construction is direct or inverse. In broad terms, the direct construction is used when the subject of the transitive clause outranks the object in the Person Hierarchy, and the inverse is used when the object outranks the subject. Following the theme marker – and sometimes merged with it in a portmanteau – a number of suffixes that function as pronominal affixes appear, VERB STRUCTURE - INDEPENDENT ORDER Prefix Salience

Preverb IRR TNS DIR (2) E. MOD EQ *NEG D. MOD ASP DIR (1)

Predicate

Theme marker + Suffix(es)

initial+medial+final Voice *NEG Cross-reference

Postverb EVID IF

Person Number Animacy Prox / Obv

Figure 22.1 Verb structure in Cheyenne (modified from Corral Esteban 2017: 310)

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

826

AVELINO CORRAL ESTEBAN

as they cross-reference the core arguments of the predicate and cover grammatical information about them in terms of person, number, animacy, and proximate/obviative status. The two outermost operators, the clausal operators of evidentiality and illocutionary force, follow the pronominal suffixes. To summarize, in Cheyenne operators realized as bound morphemes occur on both sides of the nucleus. (tns - neg - mod - asp - dir - V - neg - evid - if) É - s - sáa - tón˙eše – éne - e’e-óo’ó’t(a)-˙ o-hé-n˙ ose-Ø 3-pst-neg-be.able.to-stop-up.in-look.fti-3>I.3sg.a.Isg.u-negevid.3sg.a.Isg.u.rep-decl ho’honáevose. mountain ‘He is said to have not been able to stop looking up the mountain.’

(1)

Example (1) and its representation in Figure 22.2 indicate that, in Cheyenne, the order of the morphemes expressing operators with respect to the nucleus reflects their relative scope and, consequently, confirms the validity of the principle governing this ordering (Van Valin and LaPolla 1997: 49–52),

SENTENCE CLAUSE CORE PRO

NUC

PRO

RP

PRED V Éi - s – sáa – tónėše – éne - e´e - óo´ ´t- ói+j - hé -nósei+j - Ø ho´honáevosej DIR ASP MOD NEG TNS

NUC NUC CORE CORE

NEG

CLAUSE CLAUSE CLAUSE

EVID IF

SENTENCE Figure 22.2 Layered structure of the clause with constituent and operator projections

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

since core operators such as internal negation,3 and (deontic) modality occur closer to the verb than two clausal operators like tense and evidentiality, but are positioned further from the verb than are nuclear operators, such as aspect or directionals. Likewise, example (1) shows that the verb in Cheyenne exhibits pronominal affixes, in the form of a prefix such as é- ‘3’, and suffixes, namely -˙o- and -n˙ose, which cross-reference the different arguments in the clause. Thus, all the grammatical relations are coded in the verb, rather than in the reference phrases (RPs), which occur freely in the clause (Corral Esteban 2014: 381). The person cross-referenced by the prefix refers to the most pragmatically salient direct core argument in terms of the Person Hierarchy 2 > 1 > 3 > 4 > I, which gives preference to local participants over animate non-local participants, and to animate non-local participants over inanimate entities. In intransitive constructions, the prefix always cross-references the only argument of the verb. However, in transitive constructions, a choice is made regarding which of the arguments of the verb is more pragmatically salient in terms of the aforementioned ranking. Furthermore, in transitive predications, the Person Hierarchy works closely with a Semantic Function Hierarchy based on animacy and topicality, whereby agent outranks recipient and benefactive, and recipient and benefactive outrank patient and theme. A harmonic alignment between the two hierarchies is said to reflect a direct construction, whereas a disharmonic alignment between them is said to exhibit an inverse construction. (2)

Né-méhót-o. 2-love.vta-2>3.2sg.a.3sg.u 4 ‘You love her.’ love′ (2sg, 3sg.f)

! Direct construction

(3)

Né-méhót-a. 2-love.vta-3>2.3sg.a.2sg.u ‘She loves you.’ love′ (3sg.f, 2sg)

! Inverse construction

In these two transitive constructions, the verb has two semantic arguments – a second-person singular argument and a third-person singular argument. As second person outranks third person in the Person Hierarchy, both constructions select the prefix né- ‘2’, regardless of the semantic function of the argument in each construction. However, despite having the same prefix, each construction exhibits different relations between the verb and its arguments, as indicated by the suffixes -o and -a, which reflect a direct and an inverse construction respectively. In (2), the person cross-referenced by the prefix né- indicating a second-person participant, corresponds to the higher semantic macrorole, namely the actor, so the Person and Semantic Function hierarchies are in proper alignment, and the construction is then marked as direct. In (3), by contrast, the hierarchies are not properly aligned,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

827

828

AVELINO CORRAL ESTEBAN

as the same person prefix né- corresponds to the lower semantic macrorole, that is to say, the undergoer, so the construction is marked as inverse.

22.2.2 Basic Clause Types Cheyenne, like the other Algonquian languages, exhibits three major divisions of verb forms, commonly named ‘orders’ following Bloomfield (1946: 97–103): Independent, Conjunct and Imperative. The Independent order includes all verb forms other than imperatives that can stand alone, and the Conjunct order is used for all dependent verb forms. Hence, while the former generally occurs in main clauses (4), the latter appears in subordinate clauses (5–9). (4)

Hetane é-mane-Ø ma’xemené-mahpe. man 3-drink.vti-3>I.3sg.a.Isg.u apple-water ‘That man drank the apple juice.’ [do′ (hetane, [drink′ (hetane, ma’xemené-mahpe)]) & ingr consumed′ (ma’xemené-mahpe)]

(5)

[. . .]

tsé-x-heše-man˙e-se ma’xemené-mahpe. cnj-pst-thus-drink.vti-3>I.3sg.a.Isg.u.ind apple-water ‘[. . .] that he drank apple juice.’

(6)

[. . .]

tsé-’-éše-man˙e-se cnj-pst-already-drink.vti-3>I.3sg.a.Isg.u.ind ‘[. . .] after he drank apple juice.’

ma’xemené-mahpe. apple-water

(7)

[. . .]

tsé-x-homá’xe-man˙e-se cnj-pst-because-drink.vti-3>I.3sg.a.Isg.u.ind ‘[. . .] because he drank apple juice.’

ma’xemené-mahpe. apple-water

(8)

[. . .]

vé’-man˙e-stse cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv ‘[. . .] if he drank apple juice.’

ma’xemené-mahpe. apple-water

(9)

[. . .]

ma’xemené-mahpe. apple-water

mom˙ oxe-man˙e-stse cnj-drink.vti-3>I.3sg.a.Isg.u.sbjv ‘[. . .] I wish he drank apple juice.’

The main difference between the two verbal orders lies in the function of their prefix: whereas in the Independent order the prefix serves to express the most pragmatically salient person, in the Conjunct order it indicates the grammatical mood, namely realis (e.g. indicative tsé- (5–7)) or irrealis (e.g. conditional vé’- (8), optative mom˙oxe- (9), etc.). The prefix in a dependent clause functions as a clause linkage marker and is usually accompanied by a preverbal particle that identifies the type of subordinate clause (e.g. -hešein complement clauses (5), -éše- in temporal adverbial clauses (6), -homá’xe- in adverbial clauses of reason (7), etc.), which can lead to an instance of either daughter subordination (5) or ad-subordination (6–8).

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

The RRG notion of clause structure is based on two fundamental contrasts: the contrast between predicating and non-predicating elements, and the one between those non-predicating elements that function as arguments and those that do not. Regarding the first contrast, Cheyenne sentences have two major types of predicate, namely verbal (10) and non-verbal, which can be nominal (11 and 12) or adjectival (13).5 (10)

Néhe hetane é-ón˙están-a he’nétoo’o. deic.end.prox.an man 3-open.vti-3>I.1sg.a.Isg.u door ‘This man opened the door.’ [do′ (néhe hetane, Ø)] cause [become opened′ (he’nétoo’o)]

(11)

Tá’tóhe ma’háéso é-ma’heón˙e-hetane-(é)ve-Ø. deic.exo.dist.an old.man 3-holy-man.n-be.fai-3sg.a ‘That old man is a holy man.’ be′ (tá’tóhe ma’háéso, [ma′heón˙ehetane′ ])

(12)

Muhammad Ali é-cassiusclay-h-eve-Ø. Muhammad Ali 3-Cassius.Clay.in-epen-fai-3sg.u ‘Muhammad Ali is Cassius Clay.’ equate′ (Muhammad Ali, Cassius Clay)

(13)

Na-né’ame é-p˙ehév-˙ ahe-o’o 1.poss-parents.pl 3-good.in-fai-3pl.a ‘My parents are good.’ be′ (na-né’ame, [good′ ])

Unlike a verbal predicate, such as ón˙están ‘open’ in (10), the formation of a non-verbal predicate requires an initial nominal or adjectival stem and a final verb stem, which functions as a copula. This is evidenced by ma’heón˙ehetane ‘holy man’ and -eve ‘be’ in the identificational construction in (11), Cassius Clay and -eve in the equational construction in (12), and p˙ehév‘good’ and -˙ahe ‘be’ in the attributive construction (13).6 The final stem, by contrast, does not occur in specificational predications, which are expressed in Cheyenne through a different syntactic construction involving the presence of a relative clause. hetane tsé-ama’én-o (14) Naa’háanéhe deic.ana.dist.an man clm-drive.vti-3>I.3.a-I.u heóve-am˙aho’hest˙ otse. yellow-car ‘That man is the schoolbus driver.’ (lit. ‘That man is the one who drives the yellow car.’) be′(naa’háanéhe hetane, [drive′ (naa’háanéhe hetane, heóve-am˙ aho’hest˙ otse)])

The post-copular element in a specificational construction is realized syntactically through a relative clause, such as tsé-ama’én-o heóve-am˙ aho’hest˙otse ‘who drives the yellow bus’ in (14). This relative clause includes a lexical verb, so it is able to express more easily the second participant of a

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

829

830

AVELINO CORRAL ESTEBAN

specificational predication, as this is the more referential element of the construction, unlike in a predicational sentence where the second participant is less referential or not referential at all.

22.2.3 Verb Valence and Transitivity Verbs in Cheyenne are divided into four types according to the criteria of syntactic transitivity, based on the number of direct core arguments, and animacy: inanimate subject þ intransitive verb (II), animate subject þ intransitive verb (AI), animate subject þ transitive verb þ inanimate object (TI), and animate subject þ transitive verb þ animate object (TA).7 The number of macroroles of a verb corresponds closely to its characterization according to the notion of syntactic transitivity in Cheyenne, as the number of direct core arguments is generally the same as that of macroroles, provided the arguments refer to a particular entity. Thus, single macrorole verbs are usually intransitive: (15)

Peter he-m˙ešemo é-máse-am(e)-˙ehn(e)-óho. Peter 3.poss-grandfather.obv 4-happily-asp.in-walk.fai-4sg.a ‘Peter’s grandfather is walking happily.’ do′ (Peter he-m˙ešemo, [walk′ (Peter he-m˙ešemo)])

The activity verb ¯ehne ‘walk’ occurs in an intransitive construction including an animate participant that is lexically realized by the obviative RP argument Peter he-m˙ešemo ‘Peter’s grandfather’. This argument appears in the logical structure (LS) of the verb and is cross-referenced on the verb through the prefix é- ‘4’ and the pronominal affix óho, which refers to a fourth person / obviative singular animate patient. As this single direct core argument of each verb is referential, the verb can be considered to have one macrorole. An exception to such a correlation between the number of direct core arguments and macroroles in intransitive constructions is provided in Cheyenne, like in other languages, by ‘weather’ verbs. (16)

É-ho’ééto-Ø. I-snow.vii-Isg.a ‘It is snowing. (lit. ‘It snows.’) snow′(Ø)

The direct core argument of an II verb such as ho’ééto ‘snow’ is crossreferenced on the verb by the prefix é- ‘I’ and a null pronominal suffix if their argument is singular, or the suffix -n˙estse if it is plural. This is not actually a semantic argument of the predicate (Van Valin 2008: 8). Therefore, even though these verbs are classified as intransitive in syntactic terms, semantically speaking they are atransitive. There are other intransitive constructions showing a mismatch between the syntactic valency of a verb and its semantic transitivity, owing to the influence of referentiality on Cheyenne.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne (17)

É-h-ma’xe-mané-stove taa’é-va. 3-tns-much.drink.vai-imper night-temp ‘There was a lot of drinking at night.’ do′ (Ø, [drink′ (Ø, Ø)])

This impersonal construction in (17) exhibits an intransitive verb stem: mane ‘drink’. This, along with the presence of the prefix é- ‘3’, implies the presence of an animate argument. However, the occurrence of an impersonal suffix such as -htove/stove/nove in this construction does not allow the addition of a pronominal affix cross-referencing the single argument of the verb, which prevents the prefix from being referential. Thus, the only direct core argument of the verb – only coded in the prefix – is not a macrorole in an impersonal construction, and the verb must be considered atransitive, which shows non-identity between S-transitivity and M-transitivity. The fact that Cheyenne only cross-references referential arguments can also be evidenced in intransitive verbs corresponding to a two-place predicate whose second argument is only implied: (18)

Ná-mané-me. 1-drink.vai-1pl.a ‘We drink.’ do′ (1pl, [drink′ (1pl, Ø)])

The activity verb mane ‘drink’ in (18) has one direct core argument that is coded on both the prefix ná- ‘1’ and the pronominal affix -me and cross-references a first-person plural animate argument, namely the first argument in the LS. However, no reference is made to the second semantic argument of the verb; hence the verb mane ‘drink’ in (18) can be said to take one macrorole, which is assigned to the only direct core argument of the verb. Arguments with generic reference are generally cross-referenced on the verb in Cheyenne. (19)

Vé’hó’e é-ohk˙e-šénetam-ovo xamaevo’˙estan(e)-oho. white.people 3-eq-loathe.vta-3>4.3pl.a.4sg.u Indian.obv ‘White people usually find Indians repulsive.’ (Fisher et al. 2006: 257) loathe′ (vé’hó’e, xamaevo’˙estanoho)

In this example, both the prefix é- ‘3’ and the pronominal affix -ovo crossreference two arguments – a third-person/proximate plural animate argument and a fourth-person/obviative singular argument. However, despite being cross-referenced on the verb, the two arguments are not macroroles since they are non-specific. There is, however, a context where an argument with a generic reference is not coded on the verb in Cheyenne. This occurs when there is a close semantic relationship between the verb and the argument, which would be a typical object of the verb. In such cases, the object RP is incorporated into the verb.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

831

832

AVELINO CORRAL ESTEBAN

(20)

a. É-o’ene-me¯na-Ø. 3-pick.in-berry.fai-3sg.a ‘He is berry picking.’ (Fisher et al. 2006: 215) do′ (3sgM, [pick′ (3sgM, o’ene)]) b. É-o’en-an˙ otse men˙ otse. (3) -pick.vti-3>I.3sg.a.Ipl.u berry.pl ‘He is picking the berries.’ do′ (3sgM, [pick′ (3sgM, men˙ otse)]) & ingr picked′ (men˙ otse)

The difference between activities and their telic version, that is to say active accomplishments, is observed syntactically in Cheyenne. Unlike the active accomplishment transitive verb o’en ‘pick’ in (20b), the activity intransitive verb o’eneme¯na ‘berry-pick’ in (20a) is formed by the initial stem of the verb ‘pick’, o’ene, and a final stem of the noun ‘berry’, mene. Also, while o’en ‘pick’ has two direct core arguments, cross-referenced by the prefix é- ‘3’ and the suffix -an˙otse, which indexes a third-person singular animate agent and a plural inanimate patient, the bipartite verb stem o’ene-me¯na ‘berry-pick’ has only one direct core argument – a third-person singular animate agent, which is coded by the prefix and the null suffix. Additionally, as only the first argument of the activity verb in (20a) is specific, the verb only takes one macrorole. Conversely, the active accomplishment verb in (20b) has two specific arguments and, therefore, takes two macroroles. Except for these examples showing the influence of referentiality, the number of direct core arguments and macroroles also tends to coincide in Cheyenne transitive constructions provided the arguments are referential. Consequently, two macrorole verbs are usually transitive. (21)

Ná-ono’átam-oo’o tsé-héstoo’éšeeh-ae-se. 1-respect.vta-1>3.1sg.a.3pl.u cnj-raise.vta-3>1.3pl.a-1sg.u.ptcp ‘I respect my parents.’ (lit. ‘I respect who raised me.’) respect′ (1sg, tsé-héstoo’éšeehaese)

The two-place verb ono’átam ‘respect’ has two arguments, namely a firstperson singular animate experiencer and a third-person plural animate theme, which are cross-referenced on the verb by the prefix ná- ‘1’, the theme marker -o- and the pronominal affix -o’o. The second argument is also realized by a full lexical RP, namely tsé-héstoo’ éšeehaese ‘my parents’ (lit. ‘those who raised me’). As both direct core arguments are referential, the verb can be said to have two macroroles. Finally, unlike other Algonquian languages (e.g. Blackfoot, Plains Cree or Ojibwa), Cheyenne ditransitive verbs have a very complex system of suffixes (Corral Esteban 2014: 419–422), which vary with respect to the animacy and number of the theme argument. Thus, verbs such as the stems mét ‘give’, véestomev ‘ask for’, nomáhtseh ‘steal’, véstomev ‘promise’, or vóo’seh ‘show’ behave like true ditransitive verbs, since they cross-reference the three core arguments morphologically.8

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne (22)

Né-nomáhts˙eh-aenóvo mo’éhno’h¯ ame. 2-steal.ditr-3>2>4.3pl.a.2sg.u.4.nmr horse.obv ‘They stole the horse from you.’ [do′ (3pl, Ø)] cause [become not have′ (2sg, mo’éhno’h¯ ame) & become have′ (3pl, mo’éhno’h¯ ame)].

The predicate nomáhts˙eh ‘steal’ in (22) shows an instance of causative accomplishment where an activity is carried out by a third-person plural animate actor, cross-referenced by the suffix -aenóvo. This activity leads to a change of possession whereby the former participant manages to appropriate something, which is represented by the fourth-person/obviative participant mo’éhno’h¯ame ‘horse’, which previously belonged to the addressee. These three arguments are coded on the prefix né- ‘2’ and the pronominal suffix -aenóvo. The influence of referentiality on the cross-reference of core arguments in Cheyenne (Corral Esteban 2014: 409–410) can also be observed in ditransitive verbs not including a reference to some of their semantic arguments. In such cases, the arguments are not cross-referenced on the verb morphology, which means that both the syntactic transitivity of the verb and the number of macroroles is reduced: (23)

Ná-mét-onóneo’o n˙estámane 1-give.ditr-1>3>I.1pl.a.3pl.u.Isg.nmr 12.poss.food na-néson˙ehaneo′o. our.children ‘We give our children our food.’ [do′ (1pl, Ø)] cause [become have′ (nanéson˙ehaneo’o, n˙estámane)]

(24)

Ná-méa(’tov)-noneo′o na-néson˙ehaneo′o. 1-give.vta-1>3.1pl.a.3pl.u 12.poss.children ‘We give away our children to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, nanéson˙ehaneo’o)]

(25)

Ná-mé(a’e)-ánóne n˙estámane. 1-give.vti-1>I.1pl.a.Isg.u 12.poss.food ‘We give our food to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, n˙estámane)]

(26)

Ná-méa(‘e)-e¯me. 1-give.vai-1pl.a ‘We give something to somebody.’ [do′ (1pl, Ø)] cause [become have′ (Ø, Ø)]

Although the three-place predicate ‘give’ requires three semantic arguments, in Cheyenne these are only coded on the verb when they refer to specific participants. The pronominal affixes -o-nó-ne-o’o in (23) crossreference the three semantic arguments of the logical structure, namely a first-person plural agent, a third-person plural recipient nanéson˙ehaneo’o ‘our children’, and an inanimate singular theme n˙ahtámáne ‘our food’. However,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

833

834

AVELINO CORRAL ESTEBAN

the pronominal affixes -no-ne-o’o in (24) code only a first-person plural agent and a third-person plural theme nanéson˙ehaneo’o ‘our children’; the pronominal affixes -á-nó-ne in (25) code only a first-person plural agent and the inanimate theme n˙estámane ‘our food’; and finally, the pronominal affix -e¯me in (26) only cross-references the first-person plural agent. The role of referentiality is not only noticed on the pronominal affixes, but also in the verb stem, which adopts different forms depending on the number of macroroles the verb takes. Thus, in Cheyenne, the predicate ‘give’ can be expressed through a ditransitive stem mét, a TA stem méa’tov, a TI stem méa’e, or an AI stem méa’e (23–26). As regards the correlation between the number of direct core arguments and macroroles, in the intransitive verb in (26) the only direct core argument of the verb is a macrorole. The same correlation holds in (24) and (25), as the two transitive verbs have two direct core arguments and both are macroroles. Since only two macroroles are allowed in RRG, in the ditransitive construction in (23) only two of the direct core arguments are macroroles – the first-person plural agent and the recipient nanéson˙ehaneo’o ‘our children’, the other – the theme n˙estámane ‘our food’ – being the non-macrorole core argument. They are all identified through cross-referencing.9 To summarize, except for cases such as constructions including ‘weather’ verbs10 and impersonal constructions, and the problem posed by generic reference, which can be coded on the verb via either pronominal affixes or noun incorporation, M-transitivity closely corresponds to Stransitivity in Cheyenne, due to the effect of referentiality on the coding of arguments and, most importantly, on the form of the syntactic core template: the presence of a non-referential argument in a transitive construction affects not only the pronominal affixes but also the form of the verb stem, since the non-referential argument is not cross-referenced on the verb and is, therefore, not represented within the core. Finally, the Cheyenne examples also show that the number of macroroles that a verb takes is less than or equal to the number of semantic arguments in its logical structure.11 A particularly challenging case is a typical Algonquian transitive construction, commonly referred to as the ‘unspecified actor’ construction (Hockett 1996), as it includes no reference to the agent of the action denoted by the verb. (27)

Ná-éno’eéh-án-e na-vénótse. 1-heal.vta-X>1–1sg.u 1.poss-tepee.loc ‘I was healed at home.’ (lit. ‘Someone healed me in my tepee.’) be-at′(home, [do′(Ø, Ø)] cause [become healed′ (1sg, Ø)]])

The two-place predicate éno’eéh ‘heal’ occurs in a transitive construction including an unspecified argument that corresponds to the first argument in the LS. Bearing in mind that non-referential participants are not coded

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

on the verb in Cheyenne, it is not surprising that the unspecified agent in this construction is not cross-referenced on the verb. This assumption appears to be supported by the fact that, while the verb stem is transitive, the suffixal morphology resembles that of the AI paradigm. Furthermore, the presence of a lexicalized agent in this construction would entail changing the form of the suffixal morphology. However, we might come to a different conclusion if we take into account that this construction uses a different theme marker (e.g. -án or -Ø) depending on whether the specified argument is local or non-local. A possible explanation for this could be that the theme markers in this construction indicate direct and inverse morphology, which means that these bound markers cross-reference an unspecified participant.12 In summary, it is difficult to gauge whether this construction reveals an identity between S-transitivity and M-transitivity or not. Finally, the transitivity of the verb shows an important difference between the unspecified actor construction and the impersonal construction illustrated by example (17), for, unlike the impersonal construction, which is built on an intransitive verb stem and eliminates any cross-reference to the single argument on the verb suffix, the unspecified actor construction involves a transitive verb but AI morphology, which implies that at least one argument is coded on the verb.

22.2.4 Argument Coding on the Verb Only two different types of argument can be distinguished in Cheyenne: direct core arguments and argument-adjuncts, as core arguments can be direct but not oblique in this language. Direct core arguments must be referential to be cross-referenced on the verb. However, if they are thirdperson RPs, they can be morphologically marked or unmarked, since Cheyenne uses a reference-tracking system to establish a clausal disjoint reference between multiple third-person participants. Thus, in situations involving more than one third-person RP, the most pragmatically salient RP (i.e. the proximate) is left unmarked, and any other less pragmatically salient RP is marked with an obviative marker. The proximate/obviative system cannot be considered to be a type of case marking, as the notions of proximate and obviative are not associated with any semantic role in particular, as is illustrated by the following constructions involving two third-person referents. (28)

Náhe ka’˙eškóne é-ohke-vést˙ ahém-óho deic.end.dist.an boy 3-eq-help.vta-3>4.3sg.a.4sg.u néhe ma’háhk˙ese-ho. deic.end.prox.an old.man-obv ‘That boy always helps this old man.’ do′ (náhe ka’˙eškóne, [help′ (náhe ka’˙eškóne, néhe ma’háhk˙eseho)])

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

835

836

AVELINO CORRAL ESTEBAN

(29)

Néhe ma’háhkése é-ohke-vést˙ ahém-áá’e deic.end.dist.an old.man 3-eq-help.vta-4>3.4sg.a.3sg.u náhe ka’˙eškóne-ho. deic.end.prox.an boy-obv ‘That boy always helps this old man.’ / ‘This old man is always helped by that boy.’ do′ (náhe ka’˙eškóneho, [help′ (náhe ka’˙eškóneho, néhe ma’háhkése)])

Despite the fact that these sentences include the same arguments and are semantically equivalent, the arguments show a different proximate/obviative status. Thus, whereas in (28) the agent is the proximate náhe ka’˙eškóne ‘that boy’, and the benefactive is the obviative néhe ma’háhk˙eseho ‘this old man’, in (29), there is a shift with respect to the proximate marking. Now, the agent náhe ka’˙eškóneho ‘that boy’ is obviative and the benefactive néhe ma’háhkése is proximate. A representation of direct core arguments (28) is provided in Figure 22.3 for a subsequent comparison with that of incorporated arguments in applicative constructions in Figure 22.4 and argument-adjuncts and adjuncts in Figure 22.5. As there is no correlation between the proximate/obviative distinction and semantic roles or syntactic functions, it is necessary to take into account the information provided by the nominal morphology and the prefix and pronominal suffixes on the verb. Thus, the difference between these two examples of transitive predication lies in that, while in the direct construction in (28) the proximate argument náhe ka’˙eškóne ‘that boy’ is the actor and the obviative argument néhe ma’háhk˙eseho ‘this old man’ is the benefactive, the opposite occurs in the inverse construction in (29), as the obviative argument náhe ka’˙eškóneho ‘that boy’ is now the actor and the proximate argument néhe ma’háhkése ‘this old man’ is the benefactive.

SENTENCE CLAUSE CORE RP

PRO

NUC

PRO

RP CORER

CORER NUCR

PRED

NUCR

N

V

N

Náhe ka´ škónei

éi- ohk - vést hém - óhoi+j

‘That boy always helps this old man.’ Figure 22.3 Representation of direct core arguments (cf. 28)

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

néhe ma´háhk sehoj

A Grammatical Sketch of Cheyenne

SENTENCE CLAUSE CORE PRO

NUC

PRO

RP RPIP

NUCP PRED

PERIPHERY

PRED

P

V

CLAUSE

CORER

RPPOSS PRO NUCR N

Nái-htse-vést- homó´hem-ohoi+j vého

N he-stónahoj tséhmóheeohtséstovetse

‘I will dance with the chief’s daughter at the meeting.’ Figure 22.4 Representation of an applicative construction with a derived two-place verb

SENTENCE CLAUSE CORE PRO

NUC

PERIPHERY PRO

RP

RP

RP

NUCP

CORER

CORER

CORER

PRED PRED

NUCR

NUCR

NUCR

N

N

N

hoo´hénóva

éše va

P

V

Nái- ést - o´tse - n tsei+j m xe´ stóon tsej ‘I put the books in the bags yesterday.’

Figure 22.5 Representation of a clause including an argument-adjunct and an adjunct

Cheyenne has a number of applicative constructions where the addition of a relative root in the form of an initial stem increases the syntactic valence of the verb and, consequently, allows it to take one more direct core argument. (30)

N˙a-htse-vést-˙ ohomó’hem-óho vého he-stóna-ho 1-fut-with.in-dance.fta-1>4.1sg.a.4sg.a chief 3.poss-daughter-obv tsé-h-móheeohtsé-stove-tse. cnj-pst-meet.vai.imper-ptcp ‘I will dance with the chief’s daughter at the meeting.’ do′ (1sg, [dance′ (1sg)]) ^ do′ (vého hestónaho, [dance′ (vého hestónaho)])

While the predicate ‘dance’ requires only one semantic argument, the complex verb vést˙ohomó’hem ‘dance with’ in (30), formed by an adpositional initial vést- ‘with’ and a transitive verb stem ˙ohomó’hem ‘dance’,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

837

838

AVELINO CORRAL ESTEBAN

cross-references two arguments, namely a first-person singular agent and a third-person singular comitative vého hestónaho ‘the chief’s daughter’, through the prefix ná- ‘1’ and the pronominal affix -o, which also includes a reference to the argument vého hestónaho ‘the chief’s daughter’. Thus, although it does not refer to a participant semantically required by the predicate ‘dance’, the RP vého hestónaho ‘the chief’s daughter’ is treated as an argument of the complex verb vést˙ohomó’hem ‘dance with’, yielding a derived two-place verb. Likewise, unlike argument-adjuncts (see Section 22.2.5), this RP is crossreferenced on the verb, and, unlike adjuncts, it is not marked as an oblique.13 The fact that the cross-reference of the RP vého hestónaho ‘the chief’s daughter’ is subject to the addition of the adpositional initial vést‘with’ makes this argument the closest equivalent to an oblique core argument. The only difference is the fact that the adpositional initial is not determined by the verb. Rather, it is predicative and, consequently, it contributes its own predicate to the clause. It is also of note that the possessor RP in the genitive phrase is represented like an element branching from the RP node that agrees in person with the prefix designating the possessor, namely he- ‘his/her’, which behaves as a core argument and is, consequently, represented as a daughter of the nominal core.

22.2.5 Adjuncts, Argument-Adjuncts and Adpositional Marking Adjuncts in Cheyenne can appear in very different guises, for example: verbal premodifiers (e.g. -ohke- ‘usually’ in (31)), nominal phrases (e.g. méó’ne ‘on the road’ in (31)), free particles (e.g. hó’ótova ‘sometimes’ in (31)), or dependent clauses (e.g. tséhmóheeohtséstovetse ‘at the meeting’ in (30)). They function as optional modifiers because they are never cross-referenced on the verb or incorporated with an adpositional initial into the verb. Furthermore, they may receive oblique (e.g. locative (31), temporal (17) or instrumental (see example in note 13)) case marking through the nominal suffixes -va, -e, or -o, when they are realized through nominal phrases. (31)

Hó’ótóva ná-ohke-ameváen-one méó(’o)-n-e. sometimes (1)-eq-pass.vta-1>3.1pl.a.3sg.u road-epen-loc ‘Sometimes we pass him on the road.’ (Fisher et al. 2006: 11) [do′ (1pl, Ø)] cause [ingr overtaken′ (3sgM)]

The locative adjunct méóne ‘on the road’ is formed by the noun méó’o ‘road’ plus the oblique nominal suffix -e, which denotes location. This locative adjunct is neither cross-referenced on the verb nor incorporated into the verbal complex as a stem-forming element. Despite their reference to one obligatory argument of a verb, argumentadjuncts in Cheyenne are not cross-referenced on the verb, which makes them resemble adjuncts syntactically. Cheyenne also does not distinguish adjuncts from argument-adjuncts in terms of marking, as the same oblique

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

suffixes, -va, -e or -o, are added to both adjuncts and argument-adjuncts regardless of their syntactic function. (32)

Ná-ést-o’tse-n˙ otse m˙ oxe’˙estóon˙ otse hoo’hénó-va 1-in.in-put.fti-1>I.1sg.a.Ipl.u books bags-loc éše-e¯-va. day-epen-temp ‘I put the books in the bags yesterday.’ [do′ (1sg, Ø)] cause [ingr be.in′ (hoo’hénóva, m˙ oxe’˙estóon˙ otse)]

The verb ‘put’ in (32) requires three obligatory arguments – an agent, a patient, and a locative. However, táho’tsé ‘put’ only has two direct core arguments, the agent and the patient, as can be evidenced from the TI stem and the verbal suffix -n˙otse, which codes only two arguments – a first-person singular animate agent and an inanimate plural patient – but does not reflect the presence of any marker representing the location. The location is expressed by an RP, hoo’heno ‘bags’, which is marked as oblique by the nominal suffix -va,14 but is not coded on the verb, which results in its being marked as a peripheral oblique. However, unlike adjuncts, such as éšee¯va ‘yesterday’, argument-adjuncts appear to be complemented by a locative adpositional initial particle such as ést- ‘in’ in (32). This is an adpositional nucleus incorporated into the nucleus of the core functioning as an applicative marker that licenses the third core argument, the argument-adjunct hoo’hénóva ‘the bags’. The stem-forming element reflects the presence of a locative argument and it is this adpositional element, rather than the locative RP, that is obligatory in the sentence. The locative particle is an important component of the meaning of the sentence, and its form is not determined by the verb – the particle ést‘in’ can be replaced by -táh(o’k)- ‘on’, -áhto’- ‘under’, etc. This incorporated particle must therefore be considered to be an adpositional nucleus that functions as an applicative marker. Finally, this construction also shows that not all three-place predicates are realized through a ditransitive construction in Cheyenne.

22.2.6 The Linking Algorithm in Cheyenne The study of the interaction between the various components of grammatical structure generally shows language-specific variation. Bearing this in mind, for an examination of systematic relationships in the semantic content of a predicate, the syntactic behaviour it exhibits, and the pragmatic background that surrounds the act of communication, a bidirectional analysis of the linking system based on example (33) is presented in Figures 22.6 and 22.7, illustrating how the syntax–semantics–pragmatics interface works in Cheyenne. (33)

Na-né’ame ná-met-aenóvo 1.poss.parents 1-give.ditr-3>1>I.3pl.a.1sg.u.Isg.nmr ‘My parents gave me a book.’

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

m˙ oxe’˙estoo’o. book

839

840

AVELINO CORRAL ESTEBAN

From semantics to syntax: SENTENCE

SYNTACTIC INVENTORY

CLAUSE

4

CORE RP

PRO NUC PRO

PRO

RP

PRED V ná- métaenóvo m xe´ stoo´ok (1j) (3PLi>1SGj>ISGk)

Na-né´amei

5 LEXICON

ARG

3

INVERSE: 3>1>I

2

ACTOR

1

PSA: ARG

ARG

UNDERGOER NMR

do′ (3 [na-né´ameiACV], Ø)]CAUSE[BECOME have′ ( [1sg j ACS],

I [m xe´ stoo´okACV ]]

Figure 22.6 Linking from semantics to syntax in Cheyenne ACS ¼ accessible; ACV ¼ activated

SENTENCE

PARSER

CLAUSE

Direction of the Action? - Inverse PSA = Undergoer

CORE

1 RP

PRO NUC PRO

PRO

RP

PRED V ná- métaenóvo m xe stoo´ok (1j) (3PLi>1SGj>ISGk)

Na-né´amei

ACTOR

3

LEXICON

UNDERGOER

ACTOR

NMR

UNDERGOER

2 [do′(x i,Ø)]

CAUSE [BECOME

Figure 22.7 Linking from syntax to semantics in Cheyenne

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

have′ (y j,zk)]

1

A Grammatical Sketch of Cheyenne

This linking procedure from semantics to syntax starts with the construction of the semantic representation of the sentence, which is drawn from the logical structure of the predicate. As is the case with other languages, the selection of the actor and undergoer arguments in monotransitive constructions presents no difficulty in Cheyenne, as the highest-ranking argument in the logical structure is selected as the actor and the lowest-ranking argument is selected as the undergoer. Ditransitive constructions indicate that Cheyenne represents an instance of what Dryer (1986) called a ‘primary object language’, that is to say a language where the only pattern that occurs with three-argument predicates corresponds to the marked selection for undergoer.15 Thus, when a verb has three arguments, the second-highest, rather than the lowest-ranking argument in the logical structure is selected as the undergoer macrorole and, consequently, the lowest-ranking argument is the non-macrorole direct argument. In example (33), therefore, the leftmost argument in the logical structure of na-né’ame ‘my parents’ is selected as the actor, and the second leftmost argument – the participant referring to the speaker – is chosen as the undergoer instead of the expected rightmost argument m˙oxe’˙estoo’o ‘book’, which now becomes the non-macrorole argument. The determination of the morphosyntactic coding of the arguments shows an interesting specific feature of Algonquian languages, as the choice of a specific argument on the prefix shows the existence of a privileged syntactic argument (PSA) in these languages. This PSA involves a pragmatically determined neutralization of semantic roles for syntactic purposes, as the choice of the prefix is pragmatically determined – it is only determined by the Person Hierarchy – and can correspond to different semantic roles depending on the ranking of the participants on the hierarchy (e.g. agent (4), patient (19), experiencer (1), recipient (33), etc.). In a direct transitive construction, the PSA corresponds to the actor, and in an inverse transitive construction the PSA matches the undergoer. Thus, in example (33), bearing in mind that it is an instance of inverse construction, the PSA is the undergoer cross-referenced on the prefix ná- ‘1’. As regards the syntactic representation of the sentence, the syntactic template for a three-place predicate like mét ‘give’ must contain a core including three argument positions, which will be filled by the pronominal arguments cross-referencing the core arguments, namely the prefix ná- ‘1’ and the portmanteau verbal suffix -aenóvo, which cross-references the three semantic arguments of the verb. Finally, the linking of arguments to positions in the syntactic template is also an interesting feature of Cheyenne. The fact that word order in this language is pragmatically determined allows for greater flexibility when it comes to selecting the syntactic template of the sentence because of the positions (e.g. pre-core slot (PrCS) or any extra-core slot (ECS)) to which the independent RPs are assigned.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

841

842

AVELINO CORRAL ESTEBAN

The direction of the linking process from syntax to semantics involves the interpretation of the overt morphosyntactic form of a sentence and the deduction of the semantic functions of the elements in the sentence deriving from such an interpretation, as shown in Figure 22.7. The information provided by the verbal affix -aenóvo reflects the fact that, in this inverse construction, the verb is ditransitive and has three direct core arguments – a third-person plural agent, a first-person singular recipient, and an inanimate singular theme. The prefix also enables us to identify which constituent acts as the PSA, since, according to the ranking of the Person Hierarchy, which favours first person over third and inanimate person, ná- ‘1’ acts as the PSA of the construction. As regards the identification of the semantic macroroles of the core arguments, a ditransitive verb like mét ‘give’ has three possible candidates for only two macroroles. However, taking into account the accurate grammatical information provided by the bound markers ná- ‘1’ and -aenóvo, and considering that Cheyenne invariably shows the marked undergoer selection in ditransitive constructions, we can then select the proximate participant na-né’ame ‘my parents’ as actor (‘x’) and the first-person participant as undergoer (‘y’), whereas the third and last core argument, namely the inanimate participant m˙oxe’˙estoo’o ‘book’, becomes the non-macrorole argument (‘z’). The linking of the core arguments to the corresponding slots in the semantic representation of the sentence may be problematic in Cheyenne due to the fact that the number of pronominal affixes in the verb does not necessarily correspond to the number of direct core arguments. This is mainly because the verbal suffix is generally a portmanteau that cross-references all the direct core arguments of a verb, and the prefix only codes one of these core arguments, which does not allow for a one-to-one correspondence between direct core arguments and semantic arguments. Thus, it is necessary to bear in mind what type of stem the verb presents and what kind of grammatical information is provided by each of the pronominal affixes in order to gauge its number of direct core arguments. It would therefore seem logical to assume that the portmanteau cross-references all the arguments and the prefix crossreferences one of them again, as represented in Figure 22.7. The linking algorithm shows the important role played by pragmatics in the interaction between semantics and syntax in Cheyenne and reveals a series of grammatical properties: the variable word order exhibited by RPs, the pragmatically determined PSA, the cross-referencing of arguments through a portmanteau, and even the lack of correlation between the proximate/obviative status of referents and their semantic function in non-local contexts.

22.3

The Role of Pragmatics in Focus Structure, PSA Selection and Macrorole Assignment

The lack of a proper case-marking system and the use of a pragmatically influenced word order, which is not concerned with the identification of

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

grammatical relations or semantic roles and, even, not completely determined by the notion of presupposition and assertion, make the process of the transfer of information between syntax and semantics and vice versa more difficult than in other languages. Thus, the goal of this section is to discuss the role of pragmatics in clause structure, PSA selection and macrorole assignment by discussing the interaction between word order, the reference-tracking system of obviation, and the direct/inverse system.

22.3.1 Word Order and Focus Structure As discussed in Section 22.2.1, word order in Cheyenne encodes pragmatic factors rather than argument relations. Therefore, the language exhibits a wide range of variation in word order, which appears to be sensitive to the notion of newsworthiness. Specifically, the information is distributed in descending order of importance and unpredictability in the sentence. Importantly, the fact that word order in Cheyenne is not as strict as in other languages, English for example, does not necessarily mean that it is completely free, since it may be constrained by focus structure. Before explaining the fundamentals of Cheyenne word order from the perspective of RRG, it is necessary to examine the way in which the arrangement of discourse functions fits into the RRG model of sentence structure. The arrangement of the pragmatic categories in a specific linear order has a functional explanation. The leftmost category (see Figure 22.8), which includes topic particles and adverbials, equates to the pre-detached position (PrDP) in syntax, whose position is set off from the clause-internal elements by a pause. The next category, a focused constituent occupying the pre-core slot (PrCS) within the clause, corresponds to elements introducing discourse topics expressing a non-contrastive change of discourse topic. Alternatively, the same position can host an element introducing a contrast, regardless of whether this element has been previously mentioned in the discourse (in

SENTENCE CLAUSE PrDP

PoDP CORE PrCS

RP

(RP)

PRO NUC PRO PRED V Topic Part/Adv

Shifting/Information/Contrastive Topic Focus Topic/Focus

NEWSWORTHY

Verbal complex Information/Non-shifting Focus Topic

NON-NEWSWORTHY

Figure 22.8 Information structure-based sentence template

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

Familiar Topic

843

844

AVELINO CORRAL ESTEBAN

which case it is a contrastive topic) or is newly asserted (contrastive focus).16 Unless it is considered newsworthy, the verbal complex follows the focused element, forming the core of the clause and serving as the point of reference for the placement of the other clausal elements within the template. Both new information and easily inferable, already evoked or predictable, information, which serves to recover background information, occupy the postverbal position. Finally, the rightmost element, called a familiar topic in Figure 22.8, can also be used for an afterthought, which is included in order to reiterate its referent by a way of summary or to help to maintain the reference.17 This is also separated from the previous clause-internal constituents by a pause, meaning that it is located in the post-detached position (PoDP). The availability of two different positions for informational focus can be explained with reference to the concept of newsworthiness. Indeed, the positioning of informational focus before or after the verbal complex appears to be related to the significance given by the speaker to this information. However, the fact that word order in Cheyenne is governed by the principle of newsworthiness and that this notion is not entirely comparable to the pragmatic notions of assertion and presupposition that underlie the discourse-pragmatic statuses of focus and topic in RRG, makes it difficult to integrate the information structure of Cheyenne into the layered structure of the clause. Nevertheless, the tendency towards verb-initial word order,18 and the identification of the preverbal position (PrCS) as the most pragmatically marked position within the clause (it is the position occupied by the elements considered more informationally prominent in terms of newsworthiness) are suggestive of a correlation between word order and the different focus types proposed in RRG. Cheyenne appears to use a verb-initial order for predicate focus, the universally unmarked type of focus structure. (34)

Q:

A:

É-tóneto’omenehe-Ø(-he) Richard? 3-what.happen.vai-3sg.u(-if ) Richard ‘What happens to Richard?’ é-sáa-mé’ov-ó-he-ho 3-neg-find.vta-3>4.3sg.a.4sg.u-neg-4sg.u mo’éhno’hāme (Richard). horse.obv Richard ‘Richard/He does not find two horses.’

neše two

The pragmatic presupposition in this type of focus structure includes knowledge of a certain topic, and the assertion expresses a comment about the topic. The most common sentence type in Cheyenne in this context is a comment–topic structure where the predicate mé’ov ‘find’ and the patient neše mo’éhno’h¯ ame ‘two horses’ constitute the focus (marked with small caps), and the agent Richard corresponds to the topic. It is important to remember,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

SENTENCE CLAUSE

PoDP

CORE

RP

PRO

NUC

PRO

RP

PRED V Éi-sáa-méev-ó- he – hoi+j

neše mo´éhno´hamej (Richard)

Figure 22.9 Unmarked predicate-focus structure in Cheyenne

SENTENCE CLAUSE CORE PRO NUC

RP

PRED V é-onén šeotse-Ø na-am ho´héhame

Figure 22.10 Unmarked sentence-focus structure in Cheyenne

however, that it is very difficult to find examples of constructions containing given or familiar topics occupying a postverbal position, so that leaving the topical referent Richard unspecified appears to be more common, unless this previously mentioned information deserves to be reiterated in order to specify or clarify the reference. Figure 22.9 represents the focus structure of (34). The actual focus domain (the heavy solid line in Figures 22.9 to 22.11) in the unmarked type of predicate-focus structure comprises the core verbal constituent, if the verb is intransitive, and, additionally, the postverbal ECS, when there is (are) some overt RP argument(s). Furthermore, as expected, prosodic prominence falls on the focal constituents é-sáaméevóheho and neše mo’éhno’h¯ame ‘two horses’ in decreasing order. If the sentence included the

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

845

846

AVELINO CORRAL ESTEBAN

SENTENCE CLAUSE PrCS RP

CORE PRO NUC PRED V

Na-amaho´héhame é-onén še-otse-Ø

Na-am ho´héhame Figure 22.11 Unmarked narrow-focus structure in Cheyenne

familiar topic Richard, it would be represented as a clause-external topic at the PoDP. In that case, this topical constituent would be out of the potential focus domain (the broken line in Figures 22.9 to 22.11) and those that follow), which is coextensive with the clause in Cheyenne. While it is not so common, the positioning of a topic such as Richard in the pragmatically marked preverbal position – where it would be activated as the discourse topic – would illustrate an instance of a marked SV(O) word order for this focus type. It is of note that, due to its newsworthy nature, this constituent would be given greater prominence than the verbal complex. The second type of focus structure is sentence focus, where no pragmatic presupposition is evoked, and all information is new and, therefore, in focus. (35)

Q:

A:

É-ta-tón˙esóotse-Ø I-trans-what.happen.vii-Isg.u ‘What happened yesterday?’ é-onénėše-otse-Ø I-broken.in-become.fii-Isg.u ‘My car broke down.’

éše-e¯-va? day-epen-temp na-am˙ aho’héhame. 1.poss-car

The most common word order pattern in this situation, which places the verb in clause-initial position and the RP in postverbal position, represents the unmarked structure of this focus type because there is no element occupying the PrCS slot. Figure 22.10 represents the focus structure of (35). As there is no presupposition in this structure, the assertion extends over the entire proposition, as is indicated by the actual focus domain in Figure 22.10. Both the core and the subject RP in the ECS are associated with special intonational realization, although its intensity decreases as it approaches the clause-final position.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

Placing the argument RP na-am˙ aho’héhame ‘my car’ in the pragmatically marked preverbal position would also be acceptable in this focus structure, but the word order would represent a case of marked sentence focus. In this case, the argument RP na-am˙ aho’héhame ‘my car’ would be highlighted by the speaker, possibly due to the fact that it represents the new discourse topic.19 The difference between predicate-focus and sentence-focus structures is therefore both syntactic and prosodic. Unlike sentence focus sentences, predicate focus sentences (34) have at least one topical element, which is preferably only cross-referenced by the pronominal affixes on the verb, but can also occur in the PoDP slot when overtly realized.20 In the latter case, this topical element is not stressed, unlike the postverbal RP in the sentence focus construction (35). The third type of focus structure is narrow focus, where the focus domain is limited to a single constituent. (36)

Q:

A:

É-onén˙eše-otsé-n˙ese ne-no’ka’éesó’hest˙ otse. I-broken.in-become.fai-Isg.u.infr 2.poss-motorbike ‘I heard / It is said that your motorbike broke down.’ na-am˙ aho’héhame é-onén˙eše-otse-Ø. 1.poss-car I-broken.in-become.fii-Isg.u ‘My car broke down.’

The narrow-focus construction generally includes an element in the PrCS, for example na-am˙ aho’héhame ‘my car’ in (36). Unlike English, Cheyenne tends to mark narrow-focus constituents both syntactically and prosodically, with the narrow-focus constituent generally occurring in preverbal position – as the information it provides is normally considered by the speaker to be essential to the hearer – where it receives special prosodic prominence. Figure 22.11 represents the focus structure of (36). The constituent under narrow focus is the only one that is prosodically stressed; hence, the actual focus domain does not generally extend into the core in Cheyenne. Thus, in Cheyenne, the unmarked narrow-focus position for single constituents with focal properties is the PrCS. However, when such focal material occurs in other positions, the marked narrow-focus structure is called for. This situation is rather infrequent, as the narrowfocus constituent provides the most important information and, consequently, the information it provides is generally considered newsworthy, but it might still occur in constructions including a transitive verb and two overt argument RPs where either the verb or either of the RPs could be considered by the speaker to convey more newsworthy information. In that case, the actual focus domain would move elsewhere within the clause. Thus, narrow focus on a newsworthy element is a case of unmarked narrow focus, while narrow focus on a non-newsworthy element is a case of marked narrow focus. To summarize, while in Cheyenne the influence of the discoursepragmatic concept of newsworthiness on the syntactic arrangement of

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

847

848

AVELINO CORRAL ESTEBAN

constituents within a clause is paramount, the word order is not free, as it also appears to encode discourse-pragmatic functions such as focus and topic. Cheyenne selects verb-initial order as its unmarked word order in the predicate- and sentence-focus structures. The default interpretation of elements in the PrCS in these constructions is that of marked focus, for example contrastive narrow focus (but note that contrastive topics can also occur there, see Shimojo 2009, 2010, 2011 and Chapter 11 on information structure). Finally, prosodic prominence is always associated with the placement of focal elements in the PrCS or, in the absence of an RP occupying such a slot, in the core-initial position.21 Thus, assuming that in predicate- and sentence-focus structures the unmarked focus position for RPs in Cheyenne is postverbal and the marked focus position for RPs is preverbal – more specifically in the PrCS – the potential focus domain extends over the entire clause in all types of focus structure, but the actual focus domain is: (1) from the core to the end of the clause in unmarked predicate- and sentence-focus structures; or (2) just the PrCS in the unmarked narrow-focus type. Finally, while the notion of newsworthiness fails to correspond to the dichotomies topic vs. focus or even given vs. new information (Leman 1999), a correlation between newsworthy information and focus emerges from the analysis, as evidenced by the fact that, excluding the variable positioning of informational focus, the preferred constituent order in Cheyenne is one where focal material (informationally prominent or contrastive) precedes the verbal complex, and the topic material (known or presupposed) appears after the verbal complex.22

22.3.2 The Proximate/Obviative System Now that the link between word order and information packaging in Cheyenne has been established, this section turns to an analysis of the proximate/ obviative system. According to Aissen (1997: 709), this marking serves as a reference-tracking system whereby the language is able to differentiate multiple non-local participants in such a way that, in a discourse span involving two or more third-person animate arguments,23 only one of them is unmarked and placed in the foreground (the proximate) and all others are morphologically marked and relegated to the background (the obviative). When a discourse span begins, the most pragmatically salient participant, the proximate, is usually promoted to a preverbal position, and the obviative, which refers to the other participants, tends to occupy a postverbal position. Taking into account the fact that the preverbal position is generally reserved for newsworthy information and receives special prosodic prominence in Cheyenne, it might be logical to assume a certain correlation between the notions of proximate and focus, on the one hand, and those of obviative and topic, on the other. This correspondence is not absolute,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

however, as it is possible to place an obviative argument in clause-initial position even when this position has inherently focal properties, as is the case with an obviative interrogative pronoun, for example. As discussed in Section 22.2.4, there is no correlation either between the distinction between proximate and obviative marking, which differentiates two nonlocal participants in terms of discourse salience, and semantic roles or syntactic functions (cf. 28–29). While it is not completely clear what factor determines the choice between proximate and obviative, discourse salience must be one such factor, in the sense that the proximate is the most prominent participant of the text or the topic of discourse. The first thing a speaker must do before formulating a message is decide which of the third-person referents is the central participant in the text. Subsequently, they will leave this proximate argument, which tends to appear in preverbal position, unmarked, and mark all other third-person animate participants, which generally occupy postverbal positions, as obviative. Indeed, it is very common to find long stretches of narrative text where the same referent, considered by the speaker as the most important participant, is left unmarked as proximate. (37)

1. K˙asovááhe é-h-ne’-éva-ame-vé’ots˙e-hoo’o. young.man 3-pst-cisl-back-continue-go.on.warpath.vai-3sg.a.pret ‘A young man returned from scouting. 2. É-š-k˙ahane-otsé-hoo’o. 3-pst-tired.in.become.fai-3sg.u.pret He was tired. 3. É-x-ho’˙ehahtse-Ø tósa’e n˙ehéóhe o’hé’e. 3-pst-make.fire.vti-3>I.3sg.a.Isg.u somewhere there river.loc He made a fire somewhere there at a river. 4. Tsé-’-e¯e-hoo’˙e-se n˙ehéóhe tóx˙e(ha)-ho’˙esta-va cnj-pst-around-stay.vai-3sg.u.ind there beside-river.loc As he was sitting there beside the fire, É-x-ho’˙ehó’t-aehoono nevá’˙esesto. 3-pst-come.to.vta-4>3.4sg.a.3sg.u.pret someone.obv someone came up to him. 5. “K˙asovááhe, né-ta-néhovan-˙ ahtse-ma”, young.man 2-let-wrestle.vta-recp-12 é-x-het-aehoono. 3-pst-say.vta-4>3.4sg.a.3sg.u.pret “Young man, let’s wrestle!” he said to him. 6. É-h-néhovan-ahtsé-sesto. 3-pst-wrestle.vta-recp-3pl.rep They are said to have wrestled. 7. É-s-ts˙eheta’é(e’tov)’oe-sesto tsé-’-a’enó’ne-otse-Ø 3-pst-face.vta-4>3.4sg.a.3sg.u.rep cnj-pst-dark.in.become.fii-Isg.a.ind He is said to have been pushing him towards the dark.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

849

850

AVELINO CORRAL ESTEBAN

8. Hápó’e, é-s-ts˙eheta’é(e)’(t)ov-ósesto tsé-h-v¯ oho’˙eho’ta-tse. likewise 3-pst-face.vta-3>4.3sg.a.4sg.u.rep cnj-pst-light.vii-Isg.ind Likewise, he is said to have pushed him toward the light. 9. É-’- an˙ a’ham-osesto. 3-pst-throw.down.vta-3>4.3sg.a.4sg.u.rep He is said to have thrown him down. 10. “Né-h¯ o’tah-e”, é-x-het-aesesto [. . .]. 2-beat.vta-2>1.2sg.a.1sg.u 3-pst-say.vta-4>3.4sg.a.3sg.u.rep “You beat me”, he is said to have said to him [. . .].’ (Leman 1980a: 37)

The fact that the proximate/obviative distinction serves to track topic continuity across a discourse span can be observed in (37), an extract from an oral narrative. The main discourse topic, k˙ asovááhe ‘a young man’, is established from the very outset, as it is placed in preverbal position, a pragmatically important position in Cheyenne. Once a referent has been chosen as the main character in the story, it becomes the proximate argument, and all other third-person participants receive obviative marking – which can be observed in the nominal suffix (e.g. nevá’˙esesto [obv] ‘someone’) and in the verbal suffix, (e.g. é-xhetaehoono ‘he [obv] said to him [prox]’). Thus, for example, the argument nevá’˙esesto ‘someone’ (s. 4) is introduced as new information in sentence 4, but, very possibly owing to its indefinite character, is not considered by the speaker to have pragmatic salience. Thus, it is marked as obviative and placed in postverbal position (note that the proximate counterpart would be nevá’˙es˙estse [prox] ‘someone’). This implies that there is no shift in the discourse-pragmatic status of these two referents and, consequently, that the proximate status of the argument k˙ asovááhe ‘a young man’ remains unaltered throughout the stretch of discourse under examination (the obviative form would be k˙ asováaheho [obv] ‘a young man’), as can be observed in sentences 2 and 3, and in the dependent clause in sentence 4. However, it is also possible to find texts where the proximate referent changes very frequently, even from clause to clause. (38)

1. Vé’ho’e naa xaa-vo’˙estane é-’-e¯e-néše-ohtsé-sesto [. . .]. white.man and Indian 3-pst-around-continue-go.vai-3pl.a.rep ‘A white man and an Indian were going together. 2. [. . .] Naa tsé’tóhe xaa-vo’˙estane and deic.prox.exo.an Indian é-’-eše-aahtse’-tótoéše-na-s˙etse. 3-pst-lie-already-lie.with.eyes.open.vai-?-3sg.u.rep [. . .] this Indian is said to have been already lying with his eyes open. 3. Naa vé’hoé é-’-osee-hóhta’˙ ahané-tano-s˙estse. and white.man 3-pst-very-tell.story.vai-want.fai-3sg.a.rep And the white man is said to have really wanted to tell his story.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne 4. É-h-n˙e-hetó-sesto tsé’tóhe xaa-vo’˙estan(e)-óho, 3-pst-ana-tell.vta-3>4.3sg.a.4sg.u deic.prox.exo.an Indian-obv He told this Indian: 5. “ná-ta-ovóe-hósésta na-ováxest˙ otse!” 1-let-first-tell.vti.3>I.3sg.a.Isg.u 1.poss-dream é-x-het-ósesto. 3-pst-tell.vta-3>4.3sg.4.sg.u.rep “Let me first tell my dream!” he is said to have told him. 6. Naa tsé’tóhe xaa-vo’˙estane and deic.prox.exo.an Indian é-h-p˙ehév-áts˙está-n˙ ose. 3-pst-good.in-regard.fti-3>I.3sg.a.Isg.u and this Indian thought well of that. 7. Naa néhe vé’ho’e é-’-as˙est-(h)óhta’hane-s˙estse. and deic.end.prox.an white.man 3-pst-start-tell.story.vai-3sg.a.rep And that white man started to tell his story. 8. [. . .] Naa n˙ehéóhe tsé-s-ta-éš˙e-ho’óhta’˙ ahan˙e-se, and there cnj-pst-trns-already-arrive.in.storytelling.vai-3sg.a.ind [. . .] And when he got to that point in the story, tse’tóhe xaa-vo’˙estane é-h-n˙e-het-ósesto this Indian 3-pst-ana-tell.vta-3>4.3sg.a.4sg.u.rep this Indian said to tsé’tóhe vé’hó’e, [. . .].’ deic.end.prox.an white.man.obv this white man [. . .].’ (Leman 1980a: 40)

In this different story, shifts of proximate/obviative status occur very frequently because the speaker selects two discourse topics, rather than only one as in the previous story. The speaker considers the two referents, vé’ho’e ‘white man’ and xaa-vo’˙estane ‘Indian’, as equally important. This is supported by the first sentence of the story, where both referents appear in preverbal position forming a complex RP. From this moment onwards, whenever the speaker refers separately to either of them in the story, the referent being referred to receives the proximate status and occupies the preverbal position, for example tsé’tóhe xaa-vo’˙estane ‘this Indian’ in sentence 2 and vé’hoé ‘the white man’ in sentence 3. This leads to a rapid series of changes of discourse topic and of the discourse-status of the arguments, so that xaa-vo’˙estane ‘Indian’ is the proximate argument in sentences 2, 6 and 8, whereas vé’ho’e ‘white man’ is the proximate argument in sentences 3 and 7. Each of these two referents also receives obviative marking in one sentence, that is to say xaa-vo’˙estan(e)-óho [obv] ‘Indian’ in sentence 4 and vé’ho’e [obv] ‘white man’ in sentence 8. The fact that they are not considered pragmatically important in these two sentences is supported by their obviative marking – reflected both in the nominal suffixes and the verbal suffixes – and their placement in the postverbal position.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

851

852

AVELINO CORRAL ESTEBAN

The analysis of the proximate/obviative distinction in these two texts shows that the main topic of discourse in Cheyenne tends to be proximate, and that the proximate/obviative status of referents remains unaltered until a new discourse topic (most commonly, a new focal element or an old topical element that is reintroduced into the discourse) is established, generally bringing about a shift in the discourse span and leading to a new proximate/obviative assignment. However, as the proximate and obviative distinction is not invariably tied to macrorole assignment, it cannot by itself help us to determine which macrorole corresponds to each argument. This can be observed, for example, in sentences like 5, 7, 8, 9 and 10 in (37) or 5 in (38), where verbs crossreference two third-person animate arguments with non-overt RPs, whose interpretation may confuse the hearer.

22.3.3 The Direct/Inverse System As seen in Section 22.2.1, the Person Hierarchy and the Semantic Function Hierarchy work together as a single unified system in order to establish the link between syntactic and semantic information, as shown in Figure 22.12 (see also Wolvengrey 2011: 57–63). The interaction between the two hierarchies establishes a correspondence between actors and local participants, on the one hand, and undergoers and non-local participants, on the other, in such a way that a harmonic alignment between the two hierarchies leads to a direct construction, and a disharmonic alignment between the two scales results in an inverse construction. Both the Person Hierarchy and the Semantic Function Hierarchy rank their elements in terms of their inherent pragmatic salience or prominence. However, each hierarchy appears to be based on a different notion of salience or prominence. While the Semantic Function hierarchy is sensitive to the degree of animacy of the participants,24 the Person Hierarchy may be governed by several pragmatic factors, such as animacy or topicality, as it ranks local over non-local participants and animate arguments over inanimate arguments.25 As discussed in Section 22.2.1, the person represented by the prefix in an intransitive sentence always corresponds to the only direct core argument of the verb, so that the suffixal morphology is only concerned with the Speech Act forms Person: 2

>

(Semantic Macrorole: Actor Semantic Function:

1

non-Speech Act forms >

X

> >

3

>

4

>

I

Undergoer)

Agent > Recipient / Benefactive > Patient / Theme higher salience ---------------------------- lower salience

Figure 22.12 Interaction between the Person and Semantic Function hierarchies

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

distinction between singular and plural number (see examples 11 and 13) or obviative marking (see 15). However, in transitive constructions, where more than one participant is involved, the prefix only cross-references the most pragmatically salient argument in terms of the Person Hierarchy. (39)

Né-vóom-˙atse. 2-see.vta-1>2.1.sg.a.2sg.u ‘I saw you.’

(40)

Né-vóom-e. 2-see.vta-2>1.2sg.a.1sg.u ‘You saw me.’

In these examples the prefix né- ‘2’ is selected as the prefix heading the verbal complex, thereby acquiring a special syntactic status, namely the PSA of the construction. Neither the prefix né- ‘2’ nor the pronominal suffix -˙ atse or -e can determine the semantic roles of the arguments of the verb by themselves. The prefix cross-references only the higher-ranking person, but is not indicative of semantic role on its own – it cross-references the undergoer in (39) and the actor in (40) – and the portmanteau formed by the theme marker plus the pronominal affixes tends to provide information about the direction of the construction, and to cross-reference the direct core arguments of the verb, but is sometimes ambiguous, especially in the Conjunct order. An examination of both elements, therefore, becomes necessary in order to link arguments and semantic roles. Thus, the prefix in both examples is a second-person participant; the portmanteaus -˙ atse and -e indicate that (39) and (40) are an inverse and a direct construction respectively, and that these constructions include a first-person singular actor and a firstperson singular undergoer, respectively. The direct/inverse mechanism is especially useful when the point of view changes for pragmatic reasons, and the core arguments are not lexicalized as RPs. (41)

Ná-mó’ot-óneo’o Heév˙ ahetaneo’o 1-invite.vta-1>3.1pl.a.3pl.u Oklahoma.Cheyennes ‘We invited Oklahoma Cheyennes.’ naa ná-vés(e)-e’hanám-aeneo’o. and 1-with.in-eat.fta-3>1.3pl.a.1pl.u ‘. . . and they ate with us.’ (Leman 1980a: 71)

In this example, the pronominal affixes -ae-ne-o’o, which provide unambiguous grammatical information about each of the core arguments of the verb, help us to identify that the clause ná-vés(e)e’hanámaeneo’o is an inverse construction, as it involves a third-person plural animate participant acting on a first-person plural animate participant. To summarize, as the morphology of Cheyenne has no case distinctions, the only way to distinguish actor from undergoer in a transitive construction – see examples 5, 7, 8, 9 and 10 in (37) or 5 in (38) – is through the joint

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

853

854

AVELINO CORRAL ESTEBAN

analysis of the direct/inverse system, the Person and Semantic hierarchies and the proximate/obviative status. The direct/inverse system provides a successful link between the arguments of a predicate and their semantic roles by encoding the interaction between these hierarchies. Thus, this efficient direct/inverse system, supported by the proximate/obviative distinction in non-local environments and without recourse to a fixed syntactically oriented rigid word order, plays a similar role to case marking in other languages in order to link semantic arguments with specific semantic functions in Cheyenne. Finally, once macroroles have been assigned to semantic arguments, the speaker must select an appropriate word-order pattern following the principle of newsworthiness. In accordance with this pragmatic principle, the preverbal position in Cheyenne clauses is considered newsworthy, as it is here that discourse-prominent material is placed, and, by contrast, less discourse-salient information is placed in postverbal position.

22.4

Conclusion

The goal of this grammatical sketch has been to explore sentence structure in Cheyenne from the perspective of RRG through the analysis of a range of grammatical issues serving both to reveal the core components shared by all languages and to highlight those that are specific to Cheyenne grammar. The analysis of Cheyenne grammar in Section 22.2 showed that there is no evidence for the postulation of grammatical relations in addition to semantic predicate–argument relations, save for a pragmatically influenced PSA that is represented by the verbal prefix. It also indicates the hierarchical scope order of operators in Cheyenne, which is broadly defined in accordance with the RRG proposal. Finally, it reveals the fundamental role played by pragmatics, especially in terms of argument coding, macrorole assignment and word order. Section 22.3 argued that, despite the word-order variability displayed by Cheyenne, it is possible to integrate information structure into clause structure and explore the intricate mechanism used by this language to accommodate semantic information into syntactic structure. Cheyenne links arguments and particular semantic roles through an intricate mechanism consisting of two components, which work in combination with the proximate/obviative distinction in contexts involving non-local participants. These components are illustrated by: (1) the correlation between the Person Hierarchy and a Semantic Function Hierarchy and (2) a binary system of verbal direction expressed by a theme marker or direction-marking morpheme on the verb, which is complemented by the grammatical information about the participants provided by the pronominal affixes. The direct/ inverse system works along with the reference-tracking system of obviation,

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

interrelates the Person and Semantic Function Hierarchies, and acts as a mediator between all grammatical components. It is thus responsible for the interplay between syntax, semantics and pragmatics. The efficiency shown by this system in coding the morphosyntactic properties of arguments from the semantic information and in linking morphosyntactic coding to macrorole assignment accounts for the existence of the ‘relatively free’ word order in Cheyenne and the absence of a true case-marking system. All in all, these findings support the assumption of RRG that syntax is underpinned by semantic and pragmatic factors. Furthermore, the analysis of Cheyenne grammar highlights the validity of RRG as a framework for the analysis of highly polysynthetic languages.

References Aissen, Judith. 1997. On the syntax of obviation. Language 73(4): 705–750. Bloomfield, Leonard. 1946. Algonquian. In C. Osgood and H. Hoijer (eds.), Linguistic structures of Native America (Viking Fund Publications in Anthropology 6), 85–129. New York: Viking Fund. Corral Esteban, Avelino. 2014. An analysis of transitivity in Cheyenne. Acta Linguistica Hungarica 61(4): 379–422. Corral Esteban, Avelino. 2017. Multi-verb constructions in Cheyenne. In B. Nolan and E. Diedrichsen (eds.), Argument Realization in Complex Predicates and Complex Events: Verb-Verb Constructions at the Syntax–Semantics Interface (Studies in Language Comparison Series), 305–346. Amsterdam: John Benjamins. Dryer, Matthew S. 1986. Primary object, secondary objects, and antidative. Language 62: 808–845. Firbas, Jan. 1992. Functional Sentence Perspective in Written and Spoken Communication. Cambridge: Cambridge University Press. Fisher, Louise, Wayne Leman, Leroy Pine Sr. and Marie Sanchez. 2006. Cheyenne Dictionary. Lame Deer, MT: Chief Dull Knife College. Hockett, Charles F. 1996. Voice in Algonquian verbs: A suggestion. In John D. Nichols and Arden C. Ogg (eds.), nikotw¯ asik iskw¯ ahte¯m, p¯ askihte¯payih! Studies in Honour of Hans Christoph Wolfart. Winnipeg: Algonquian and Iroquoian Linguistics: 257–261. Leman, Elena. 1999. Word Order of Major Constituents in Cheyenne Narratives. Dallas, TX: Summer Institute of Linguistics. Leman, Wayne. 1980a. Cheyenne Texts: An Introduction to Cheyenne Literature (Occasional Publications in Anthropology, Linguistics Series No. 6). Greeley, CO: Museum of Anthropology, University of Northern Colorado. Leman, Wayne. 1980b. A Reference Grammar of the Cheyenne Language, Vols. 1 and 2 (Occasional Publications in Anthropology, Linguistics Series No. 5). Greeley, CO: Museum of Anthropology, University of Northern Colorado.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

855

856

AVELINO CORRAL ESTEBAN

Leman, Wayne. 1987. Náévâhóó’ôhtséme / We are going back home: Cheyenne History and Stories Told by James Shoulderblade and Others. Memoir 4. Winnipeg: Algonquian and Iroquoian Linguistics. Meeussen, Achilles E. 1962. The independent order in Cheyenne. Orbis 11: 260–288. Mithun, Marianne. 1987. Is basic word order universal? In Russell S. Tomlin (ed.), Coherence and Grounding in Discourse, 281–328. Amsterdam: John Benjamins. Petter, Rodolphe. 1952. Cheyenne Grammar. Newton, KS: Mennonite Publication Office. Russell, Dale W. 1987. Cheyenne Verb Agreement in GPSG. Doctoral dissertation, University of Illinois at Urbana-Champaign. Shimojo, Mitsuaki. 2009. Focus structure and beyond: Discourse-pragmatics in Role and Reference Grammar. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro (eds.), Studies in Role and Reference Grammar, 111–141. Mexico: Universidad Nacional Autónoma de México. Shimojo, Mitsuaki. 2010. The left periphery and focus structure in Japanese. In Wataru Nakamura (ed.), Proceedings of the 10th International Conference on Role and Reference Grammar (RRG 2009), 315–335. Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japanese. In W. Nakamura (ed.), New Perspectives in Role and Reference Grammar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing. Siewierska, Anna (ed.). 1997. Constituent Order in the Languages of Europe. Berlin: Mouton de Gruyter. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics– Pragmatics Interface. (Studies in Language Companion Series 105). Amsterdam: John Benjamins. Van Valin, Robert D. Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wolvengrey, Arok. 2011. Semantic and Pragmatic Functions in Plains Cree Syntax. MA thesis, University of Amsterdam.

Notes 1 The prefix selecting the most pragmatically salient person only occurs in the Independent order (except in the conjectural mode, which is introduced by the prefix mó-). The prefix in the Conjunct order conveys other grammatical meanings (e.g. tsé- ¼ realis, vé’- ¼ irrealis, m˙ ah- ¼ potential, hó’- ¼ iterative, momóxe- ¼ optative, etc.). The independent and the conjunct verbal orders generally occur in main and subordinate clauses, respectively. 2 There are two main types of directional particle: (1) the cislocative neh- and translocative ta-, which serve to express the locational or temporal deixis

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

3

4

5

6

of the action, with the speaker or the speech act itself as the deictic locus; and (2) phoric particles, such as the anaphoric and cataphoric particles n˙eand ts˙e-, which serve to link elements in discourse, namely a place old or a place new in discourse respectively, and appear close to the prefix. There are also other directional initial particles such as he’am- ‘up’, e’e- ‘up’, an (˙ohe)- ‘down’, ést- ‘into’, sé’- ‘into’, hóe- ‘out’, etc., which always occupy the rightmost position preceding the verb. Negation in Cheyenne is commonly expressed through two separate and complementary particles, namely sáa-, a preverbal particle, and -hé, a postverbal particle that normally occurs between the theme marker and the pronominal affixes. For the sake of simplicity and because it is not always possible to separate the theme marker from the pronominal affixes on a verb, both parts will be analysed as a portmanteau suffix throughout the chapter. In order to indicate the direction of constructions including transitive verbs, a notation such as ‘1>3’ will be used, for example to indicate a first-person actor and a third-person undergoer. For intransitive verbs, its only argument will be referred to as ‘1’, for example to indicate that the actor or undergoer is the first person. Adjectives in Cheyenne are always realized as bound particles that attach to verbs (see example 38(6), in section 22.3.2) or nouns (11) and behave like adverbs or adjectives in English, respectively. Likewise, they can also form non-verbal predicates when they are attached as initials to final verb stems in attributive constructions (see 13). By contrast, pronouns (see 37(4)), numerals (see 34), adpositions (see 37(4)), and some adverbs (see 37(3)) are realized as free particles in Cheyenne and, consequently, excluded from the possibility of functioning as predicates. Albeit less frequently, other types of identificational sentence can be expressed in Cheyenne by means of a linking or copular verb (a), the combination of a preverbal particle and a nominal predicate (b), or even a verb-less clause (c). E.g. a.

b.

c.

Maah(e)-ótse éhó’tán˙estse na-m˙ aheó-n-e. arrow.pl I-be.here.vii-Ipl.u 1.poss-house-epen-loc ‘The arrows are in my house.’ be.in′ (na-m˙ ahe¯o ¯’o, maah(e)ótse) John ná-he-vésenéhe(‘tov)-n˙ otse. John 1-have-friend.vta-1>3.1sg.a.3sg.u have.as.friend′ (1sg, John) ‘John is my friend.’ (lit. ‘I have John as friend.’) consider′ (1sg, [be′(John, [na-(ve)séné]) Hé’tóhe máto na’˙estse hóhta’heo’o. deic.exo-prox-inan also one story ‘This is also another story.’ (Leman 1987: 314) be′ (hé’tóhe, [na’˙estse hóhta’heo’o])

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

857

858

AVELINO CORRAL ESTEBAN

7 Each verb type has morphological paradigms showing different theme markers and pronominal affixes. This information is available in a number of sources (Petter 1952; Meeussen 1962; Leman 1980b; Russell 1987; Corral Esteban 2014). 8 Other ditransitive constructions in Cheyenne consist of a verb plus a final verb stem, such as, for example: -’seh, in causative constructions, -omótah, in benefactive constructions, etc. 9 See Section 22.2.6 for more detailed information on undergoer selection in Cheyenne ditransitive constructions. 10 An aspect of ‘weather’ verbs that is difficult to explain is the fact that, unlike the impersonal construction, they do not seem to reflect Mtransitivity in the form of their verb stem, as they do not include any morphological marker indicating that the syntactic valence has been reduced. A possible solution would be to assume that the consideration of the single core argument of ‘weather’ verbs is culture-specific, and that, for example, some cultures understand that there is a non-specific agent carrying out the activity denoted by the predicate. 11 Applicative constructions (30) would be an exception to this assumption, where the addition of a relative root, in the form of an initial, increases the syntactic transitivity and the number of macroroles without altering the argument structure. 12 This construction appears to resemble the English-style passive in the sense that it shows the promotion of the patient and the demotion of the agent, as reflected in the common preverbal position of the former and its cross-reference by the prefix, on the one hand, and the omission of the latter, on the other. However, if we assume that the theme markers -Ø and -án indicate direct and inverse constructions, then direct forms would not be fully passive, since the patient would not be promoted to subject status, even though the actor is obligatorily demoted. 13 Adjuncts can also occur in constructions including a relative root, with the fundamental difference that they are not cross-referenced on the verb: E.g.

M˙ ahaemen˙ otse náve’š˙epén˙ ohan˙ otse ho’honáéva. corn.pl 1-by.means.of-grind.vti-1>I.1sg.a.Ipl.u rock.ins ‘I ground corn with a rock.’

The AI verb pen˙oha ‘grind’ only codes a first-person singular agent and an inanimate plural patient. 14 It is very difficult to come to a decision about the status of the suffix -va. On the one hand, given there is an applicative (incorporated adposition) licensing hoo’hénóva ‘in the bags’ in (32), it does not seem likely that this applied argument may be marked by a postposition. On the other hand, the fact that the suffix -va may occur on a variety of argument-adjuncts and adjuncts with different meanings (e.g. time, location, instrument)

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

A Grammatical Sketch of Cheyenne

15

16

17 18

19

appears to suggest that it is not a case marker in the usual sense. In view of the above, it seems plausible to consider it a kind of generalized oblique marker, with no inherent semantics of its own. The evidence for the marked undergoer choice in ditransitive constructions comes from both coding (indexing and ordering of bound forms) and behavioural properties (reflexivization, reciprocalization, and direct/ inverse marking) (Corral Esteban 2014: 403–415). On the one hand, the ditransitive verbal paradigms resemble their TA more than their TI counterparts, and the pronominal markers cross-referencing the recipient occur closer to the verb than those cross-referencing the theme. On the other hand, the reflexive/reciprocal suffix only permits coreference between actor and recipient, but not between actor and theme, and direct/inverse marking shows a clear secundative bias since the impossibility of constructing ditransitive sentences with inanimate recipients that are prototypically human, or first- or second-person themes that are prototypically non-human, implies that only the person of either the actor or recipient, but not that of theme, can appear as a prefix. I am assuming that focal elements occupy the PrCS rather than the PrDP in Cheyenne because there is no intonation break between most of the preverbal RPs and the verbal complex, which suggests that they are part of the clause. Albeit infrequently, it is also possible to find left-dislocated familiar topics in Cheyenne. This assumption concerning word order is based on my analysis of word order patterns in independent clauses found in thirty-five Cheyenne stories told by different native speakers, which were randomly taken from Leman (1980a). The analysis revealed that the most common sentence type involves no explicit RPs, and the second most common type involves one RP. The analysis of intransitive clauses including an RP shows that the most common word-order pattern is SV (63.48%), which appears to be more related to newsworthiness (new discourse topic: 83.01% vs. continuing topics: 16.99%) than to givenness (new information: 59.47% vs. old information: 40.53%). In transitive clauses including two overt RPs, the most common word-order pattern is SVO (86.67%), followed far behind by OVS (8.89%). In the more common situation displayed by transitive clauses, that is involving an object RP, the most common pattern is VO (66.41%)). This situation also appears to hold in presentational sentences, where all the information is asserted and nothing is taken as presupposed. While Cheyenne appears to follow a verb-initial pattern as its basic, but not obligatory, word order, sentences that include a direct core argument in preverbal position also appear to be very common at the beginning of a story or conversation, because of the speaker’s interest in establishing a discourse topic from the outset.

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

859

860

AVELINO CORRAL ESTEBAN

20 As discussed above, albeit rather infrequently, this topical element could also occur in the PrCS if it is considered newsworthy. 21 Except when a single constituent is foregrounded by the speaker, the verb will be the most pragmatically salient element, as is usual in a sentence-focus construction. 22 This word order appears to disprove the general assumption that, in terms of information structure, it is cross-linguistically more common for topical participants to precede focal participants (Firbas 1992; Siewierska 1997; among others). 23 Owing to language attrition, save for some pronouns, it is not possible to differentiate proximate and obviative inanimate arguments in Cheyenne nowadays. 24 The higher-ranking semantic roles (agent, recipient or benefactive) are linked with animate participants, while the lowest-ranked semantic role (patient) is associated with inanimate participants. 25 Additionally, the specific preference of second person over first person may also be motivated by cultural factors (e.g. the tuistical, rather than ego-focused, nature of the Algonquian culture, politeness conventions, empathy or modesty, among others).

https://doi.org/10.1017/9781316418086.028 Published online by Cambridge University Press

23 A Grammatical Sketch of Yimas (Lower Sepik, Papua New Guinea) William A. Foley Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php) with the following additions and modifications: ASP CLM DEIC DIR DL ELEV FR.DIST HAB INACT MOD N NINCORP NFN NR.DIST NUC NUM

23.1

aspect clause linkage marker deictic directional dual elevational far distal habitual inactivated modality noun incorporated noun non-finite near distal nucleus number

PAST PC PERF PN PRES POT PRO RED RM RX SEQ TNS V VAL I-X

past tense paucal perfective proper noun present tense potential mode bound pronoun ?reduplicative? remote referential expression sequence tense verb valence noun class

Yimas as a Polysynthetic Language

Yimas is a morphologically highly complex polysynthetic language spoken in the Sepik basin region of the northern swampy lowlands of Papua New Guinea and is one of six languages in the Lower Sepik sub-family of the Lower Sepik-Ramu family (see Foley (1986, 2005, 2017a) for comparative studies of this family). It has a typical profile for a polysynthetic language:

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

862

WILLIAM A. FOLEY

it is heavily head marking, and, while the morphological structure of words is very extensively elaborated, the syntactic structure of phrases and clauses is only weakly so. In fact, Yimas almost entirely lacks the familiar syntactic category of phrases, and at the clause level it is highly non-configurational (Hale 1983). The one notable difference of Yimas from most other polysynthetic languages is an extensive system of noun classes, eleven major ones and half a dozen minor ones. For a detailed exposition of Yimas grammar, see Foley (1991).

23.2

Nouns, Verbs and Agreement

As a polysynthetic head-marking language, the morphology of agreement is much in evidence in the language. The language has two major word classes, nouns and verbs. Words corresponding in meaning to adjectives in English or other familiar languages are divided between these two classes, with most belonging to the verb class, though there are three true adjectives. Many grammatical categories distinguish the two major classes of nouns and verbs, but a quite noticeable contrast is in their formal patterns of inflection: nouns (and the adjectives) are inflected for noun class and number by a set of suffixes, while verbs mark the noun class and number of their core argument participants for the most part with prefixes, as in this example. (1)

yura antmaŋkl kpa-ŋkl kla-n-am-(n)tut dog.iii.sg cockatoo.vi.dl big-vi.dl vi.dl.nom-iii.sg.erg-eat-rm.past ‘The dog ate two big (sulphur crested) cockatoos.’

The marking of noun class and number on nouns is largely fusional, so it is difficult to separate the root from the inflection, and generally both noun class and number are portmanteau as well (especially in the singular), and there are morphophonemic rules at work too complex to go into here. However, the nominal agreement suffixes are isolatable when they appear on adjectives which are in concord with nouns they modify, as with -ŋkl vi.dl on the adjective root kpa ‘big’ in (1). Verbs have overt affixes that mark noun class and number of their core arguments, and these work under a very complex split ergative-accusative and direct-inverse alignment that will be explained below, so they are labelled in (1) simply as nominative and ergative. The distinction between the two major classes, noun and verb, is very rigid in Yimas; there is no overlap or flexibility so salient in Austronesian languages. A verb can only be used as a noun when it is overtly derived by a morphological process of nominalization (this will be the focus of the second half of this chapter), while nouns can never be used as verbs. There are no processes of denominal verbalization in the language, nothing equivalent to English I hammered the nail down into the floor, where the noun hammer denoting an instrument is used as verb denoting doing an action

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

using that instrument; the equivalent Yimas expression would have to be something like ‘I hit the nail with a hammer piercing (the floor), going down.’ Verbs can only be used as arguments when they are derived into nouns by nominalization. Unmarked predicates are always verbs. The language does possess a copula (of very complex morphology showing agreement in noun class and number), and this is required whenever a noun functions as a predicate. (2)

23.3

k-n akrŋ akk vi.sg-fr.dist frog.vi.sg cop.vi.sg ‘That’s a frog.’ (said on hearing the croaking)

Clause Structure, Information Structure and Head Marking

The syntactic templates to build syntactic structures in Yimas are very simple, in marked contrast to the elaborate morphological templates for verbs, which given the space limitations, I will not be able to explore in detail here, though here is a summary of the morphological possibilities for finite verbs (non-finite verbs are nominalizations); a number of these slots permit multiple fillers: MOD-PRO-ASP-ADV-ELEV/DIR-VAL-NINCORP-V-VAL-DIR-ASP-TNS-PC-PRO

As this formula demonstrates, much of this word-level complexity has to do with the nesting of operators at various levels (Foley 2017b), but here is a typical example: ka-mpu-pay-ma-takat-ɲa-mpan-m (likely-3pl.erg-firstinside-touch-imp-3pl.dat-vii.sg.nom) ‘let them first apply it (paint) to them inside’. Essentially, below the clause level there is not very much syntactic structure; the language strongly prefers to build up clauses simply by stringing words together, and the ordering of these words within the clause is highly flexible. While Yimas, like languages of many Papuan families, has some typological markings of being right-headed, it is by no means verb-final. Any order of the verb and its arguments and adjuncts is acceptable, except verb-initial variants. Given this, there is no evidence for a pre-core slot or a post-core slot, and surprisingly, even the leftdislocated position seems weakly developed, mainly restricted to vocative uses, if these are to be analysed as such. The right-dislocated position, by contrast, is quite often used, essentially for afterthoughts, that is, when the speaker believes the pronominal affix for a core argument on the verb is insufficient to identify its referent, and this most commonly occurs when the referent is a subject, so it somewhat functions then as a switchreference device when there are multiple animate participants in an ongoing text. As a common tendency, also, locative or temporal expressions tend to occur at the periphery of the clause, the beginning or the

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

863

864

WILLIAM A. FOLEY

end (and again usually the former), but this is only a tendency, and it is not unusual to find them midway in a clause, say, following a noun functioning as a core argument. In any case, Yimas clauses, like those of many Papuan and other purely oral languages, tend to be information-poor, containing few constituents. Clauses containing a verb and two overt core argument nouns are very rare indeed, much less than 1 per cent of all clauses. Where they might occur is at the beginning of a narrative to introduce the characters. Yimas has no specialized presentative constructions for introducing new referents like ‘there was an old woman who lived . . .’, so new nouns are introduced baldly in new mentions at the beginning of a story, and this usually by naming them; ‘Yapalmay and Mampalmay lived with their brother Yampwiŋkawi’, for instance, is the opening line of the legend which explains how men’s penises got shortened. There is no distinct clause type for sentence focus. About half of all Yimas clauses with transitive verbs consist of just the verb with perhaps a peripheral temporal or locative expression (often marked with the oblique suffix -n ~ -nan, the only true nominal case marker in the language), and the other half will have a single overt noun as a core argument, again with perhaps a peripheral adjunct. What determines these patterns is mainly information structure. Nouns whose referents are activated or easily accessible are normally only expressed by the corresponding bound affixes for their noun class and number. To have an overt noun and its corresponding bound affix co-occur in a context in which there is no ambiguity as to their reference makes the referent of the noun highly contrastive. It takes on this contrastive reading from the conflict in interpretation between the overt noun and the bound affix. Overt nouns are used when their referents are inactivated or inaccessible and thereby being introduced into the stream of speech; they are focal in information structure, either part of predicate focus or themselves as narrow focus, as clearly shown in this question-pair. (3)

Q:

A:

wara ipa-n(a)-am-n? what 1pl.nom-pres-eat-pers ‘What are we going to eat?’ numpran ipa-n(a)-am-n pig.iii.sg 1pl.nom-pres-eat-pres *??numpran na-kay-ɲ(a)-am-n pig.iii.sg iii.sg.nom-1pl.erg-pres-eat-pres ‘We’re going to eat pork.’

The combination of the overt noun and bound verbal prefix here is very strange because it forces a contrastive reading where the question does not set a context in which this is felicitous. The overt noun signals an inactivated reading, while the bound verbal prefix, an activated or easily accessible one, two contradictory readings that could only be reconciled in a contrastive focus reading, but this makes little sense in this context. So the two most

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

common clause types for Yimas are those with just a verb, where the referents of all core participants are activated or accessible, or a verb and one overt noun, in which the overt noun has an inactivated or inaccessible referent. This is the basic system, though there are a few wrinkles and complications, particularly with wh-questions (Foley 1991: 430–433). Due to the sensitivity of the bound verbal affixes to information structure, the analysis of the head-marking language Lakhota in Van Valin (2013) needs to be expanded a bit. Van Valin (2013) did not intend his analysis of Lakhota to be applicable to all head-marking languages, but rather focused on the type represented by Lakhota, and in this spirit we offer this analysis of Yimas as a further enrichment of the descriptive typology. For Lakhota, Van Valin argues that the verbal prefixes, which may be null, saturate the argument positions within the core and that any overt nouns are in core-external, but clause-internal positions, essentially sisters of the core node under the clause. Given that no basic bound affix of any person, number or noun class combination in Yimas has a null exponent, it would be extremely undesirable to appeal to null affixes in this language, but without them, to apply Van Valin’s analysis of Lakhota to Yimas would require violating the Completeness Constraint, an even more undesirable result, if we want to satisfy all the lexical requirements of the verb and its subcategorized arguments in the core, as we clearly do in this language with no voice alternations or extractions. So instead, I propose that both bound verbal affixes and overt nouns occur in the core, as long as the nouns are inactivated, that is, they do not co-occur with a bound prefix. Those that do, occur in the same core-external position that Van Valin proposes for Lakhota. All of this is illustrated in the following very typical Yimas sentence and its representation in layered structure and information structure. The only constraint on word order here is that the conjunction kanta ‘but’ occurs preferentially in second position, though this is not rigid, and the verb cannot be initial; otherwise all possibilities are acceptable, though some will be more marked pragmatically. Note particularly the separation by this conjunction of the modifier kamta- ‘empty’, which is actually a nominalized verb in a relative clause (Section 23.7.2), hence the tense marker -k irr, from the noun parwa ‘dock’ that it modifies, but linked to by the noun class and number concordial suffix proper to nouns of this class and number. Note too the fact that the oblique case suffix -n only occurs on the noun parwa ‘dock’, but applies to its linked modifier kamta‘empty’ as well. The bound verbal affix is a circumfix for this verb because it has negative polarity (Foley 1991: 251–263). (4)

parwa-n kanta kamta-k-wa ta-pu-tay-kiak-rm kay dock.ix.sg-obl but empty-irr-ix.sg neg-3-see-irr-3dl.nom canoe.viii.sg ‘But they both didn’t see a canoe at the empty dock.’

In Figure 23.1, which represents the structure of (4), RX stands for referential expression, in order to treat nouns and bound prefixes with a single label

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

865

866

WILLIAM A. FOLEY

SENTENCE

CLM

CLAUSE

PERIPHERY

CORE

RXIX.SG.OBL

NUCN

RXVIII.SG

NUCLEUS

RXNOM

PERIPHERYN PRED RXIX.SG

parwa-n

kanta kamta-k-wa

dock.IX.SG-OBL but

ta-pu-tay-kiak-rm

kay

empty-IRR-IX.SG NEG-3-see-IRR-3DL.NOM canoe.VIII.SG FOCUS V

(INACT)

NUCLEUS

CORE

SPEECH ACT

CLAUSE

TNS

SENTENCE

Figure 23.1 Clause structure with constituent, operator and focus projections

and avoid positing phrases where there are none, as such are very depauperate in Yimas. I have also simplified for purposes of exposition the representation of the relative clause, as these will be discussed in more detail in Section 23.7.2, and treated it like an adjective, as here it behaves exactly as a true adjective like kpa ‘big’ would. The syntactic structure of phrases is extremely simple in Yimas, and what look like noun phrases are more akin to compounds in structure than

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

phrases. Essentially a noun phrase can consist of no more than two constituents, a noun and its preposed modifier, and those modifiers can only be of two kinds, a possessor, which must be a pronominal or a proper name, or one of the three adjectives, and these cannot co-occur. We can say ama-na kay 1sg-poss canoe ‘my canoe’ and kpa kay big canoe ‘big canoe’ in that fixed order of constituents, but a noun phrase like *ama-na kpa kay 1sg-poss big canoe is impossible in Yimas. One would say ama-na kay kpa-y 1sg-poss canoe.viii.sg big-viii.sg ‘my canoe, the big one’, where the adjective kpa ‘big’ has been converted into a noun by the noun class and number concordial suffix -y and, like kamta-k-wa empty-irr-ix.sg ‘empty’ in (4), can be floated away from the noun it modifies and no longer forms a constituent with it, so that this is actually two referential phrases in apposition to each other. Ignoring nominalizations of verbs to be discussed below, referential phrases in Yimas can be made up of at most two constituents, such as ama-na kay or kpa kay, and the vast majority consist of just one, so I will continue to refer to them as referential expressions (RX) in preference to RPs, especially as they are typically at the word (X0) level. The only operator that RXs take is number, at the core level, singular, dual and plural, and adjectives and possessives are peripheral modifiers, so, as in Figure 23.2. RX

RX

COREN

PERIPHERYN

ADJ

kpa big

COREN

NUCLEUSN

NUCLEUSN

N

N

kay

kacmpt

canoe.VIII.SG

canoe.VIII.PL

NUCLEUSN

NUCLEUSN

COREN

NUM

RX

COREN

RX

Figure 23.2 Phrase structure with constituent and operator projections

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

NUM

867

868

WILLIAM A. FOLEY

The only other phrase type in Yimas is the postpositional phrase headed by postpositions like nampan ‘toward, for, because of’ or kantk ‘together with’, which form the predicate and nucleus of the phrase and their core arguments are RXs. Postpositional phrases are always predicative peripheral adjuncts; there are no non-predicative peripheral adjuncts in the language, because, as we shall see next, all arguments subcategorized by a verb must be projected as arguments in the core, either as an independent RX or as a bound verbal one.

23.4

Logical Structures and Macroroles

Yimas, like some other Papuan languages (e.g. Kalam; Pawley 1993), has a restricted lexicon of monomorphemic verbs, only around a couple of hundred verb roots. More complex verbal expressions are constructed by these roots in serial verb constructions in nuclear junctures. Yimas has basic underived intransitive, transitive and ditransitive verbs, although there are only four roots in the last class. There are also lexical processes that derive verbs, a reciprocal prefix that removes a core argument and two causative prefixes, six applicative affixes and possessor raising that add one, but unlike some Bantu languages such as KiHaya (Duranti and Byarushengo 1977), in no case can a verb be derived that exceeds the number of arguments of an underived verb; the maximum number of core arguments of any verb is three. Beyond these processes of decreased and increased transitivity, there are no derivational processes in Yimas that affect a verb and its arguments’ mapping into constituent structure. There are no voice alternations such as passive or antipassive and no alternations in realization such as we find in English between the farmer loaded hay on the truck versus the farmer loaded the truck with hay. Given a verb’s logical structure and the arguments it subcategorizes, there is one and only one mapping into constituent structure that corresponds to it. The case marking of the arguments is very complex and will be discussed in the following section; here I am just concerned with their basic realization as constituents. Consider an intransitive unaccusative verb like mal- ‘die’, as in na-mal 3sg.nom-die ‘he just died’. This is an achievement predicate and so has the following logical structure: INGR died′ (he). As he is an argument in a frame of pred′ (x), it will be assigned the undergoer macrorole by the universal Actor-Undergoer Hierarchy (AUH), and as the sole argument will appear in nominative case as a bound prefix to the verb. Now consider an intransitive unergative verb like iray ‘cry’ as in na-iray 3sg.nom-cry ‘she cried’. This is an activity predicate with the logical structure do′ (she, [cry′ (she)]). As she is an argument in a frame of do′ (x), by the universal AUH it will be assigned the actor macrorole, but again as the sole argument will appear in nominative case as a

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

bound prefix to the verb. Now consider a canonical transitive verb such as warapa- ‘cut with a flat sharp instrument’ as in (5). (5)

kaŋk-ɲan na-ka-warapa-ntut shell.vi.sg-obl 3sg.nom-1sg.erg-cut-rm.past ‘I cut it with a shell.’

The logical structure of this clause, including the oblique nonsubcategorized RX, would be as in (6). (6)

[do′ (I, Ø)] CAUSE [[ do′ (shell, Ø)] CAUSE [INGR cut′ (it)]]

By the universal AUH, it as the argument of pred′ (x) would be assigned the undergoer macrorole and therefore be realized by nominative case in this direct arrangement of arguments (see Section 23.5). But there are two potential actors, I and shell, as both are first arguments of do′ (x, . . .) (Yimas, unlike Tagalog, has no constructions which differentiate grammatically between volitional deliberate actions and those which are not, and there are no verb roots which lexicalize this contrast like see versus watch, so there is no need to postulate an operator DO in logical structures.) Yimas, unlike English, does not permit sentences like the shell cut it; the animate instigator who instigates the causal chain must be mentioned overtly and outranks the instrument. Hence it is the animate instigator, here I, which is assigned the actor macrorole and receives ergative case, realized in the core along with the undergoer, while the instrument is realized as an oblique RX in the periphery of the core. Finally, consider the canonical ditransitive verb ŋa‘give’ as in (7). (7)

trawsistm tma-mpu-ŋa-ŋa-ntut trousers.v.dl v.dl.nom-3pl.erg-1sg.acc-give-rm.past ‘They gave me two pairs of trousers.’

Note that Yimas belongs to the relatively rare type of triple-agreement languages; given the proper information structure configuration, all three arguments subcategorized by the verb are realized through bound pronominal affixes on the verb, and so all three must be regarded as core arguments. Yimas seems to be a mix of a direct object language or a primary object language, but just what is it? There are two non-actor core arguments, the theme and the recipient, accessible to the undergoer macrorole, but which one or is there only one? The relevant principles in the revised AUH given in Van Valin (2005: 126) are Principle A: lowest-ranking argument in argument structure or Principle B: second-highest-ranking argument in argument structure. Consider (8), the logical structure for (7). (8)

[do′ (they, Ø)] CAUSE [BECOME have′ (me, trousers)]

The actor macrorole is straightforward: as they is the first argument of do′ (x, . . .) and thereby is assigned the actor macrorole and realized in ergative

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

869

870

WILLIAM A. FOLEY

case. But the undergoer is more problematic: by Principle A, trousers, the lowest-ranking argument would be the undergoer, but by Principle B, it would be me, the second-highest-ranking argument. Is there any way to determine which or must both be reckoned as undergoers? The answer to this question is not easily apparent. There are no voice alternations to check for accessibility, and the evidence is conflicting, but the bulk seems to favour the claim of two undergoers, as in Lake Bantu languages (Bresnan and Moshi 1990; Duranti and Byarushengo 1977). For example, the case hierarchy to be described in Section 23.5 mandates the following ranking ACC > ERG > NOM. Note in (7) that the referent of trousers is indicated by nominative case and me by accusative. This might establish me as the real undergoer as it outranks trousers in this case hierarchy. But as we shall see, nominative is usually the only obligatory case on a verb if it has any agreement prefixes at all, so it would seem strange to claim less obligatory affixes as marking the undergoer over more obligatory ones; this favours trousers as the undergoer. Yet the nominative case affix always occurs on the periphery of the verb, either right at the beginning or right at the end, in contrast to the more highly salient internal positions closer to the verb stem occupied by the recipient affix. This, then, again may favour the claim of me as the undergoer. On the other hand, recipient arguments cannot be directly relativized like the actor or theme, but require structures that are ambiguous with coordinated clauses; the fact that the actor and theme pattern alike as against the recipient would seem to favour trousers as the undergoer. Yet in possessive-raising constructions like ‘he hit me on the arm’, for which in head-marking languages the affected person is crosslinguistically typically the undergoer, not the body part, the affected person in Yimas is always expressed exactly as the recipient of a ditransitive verb (Foley 1991: 300–303) (for comparable data, see the Muskogean languages, Choctaw (Broadwell 2006), Chickasaw (Munro 1984; Munro and Gordon 1982) and Creek (Martin 2011)). The evidence is murky and inconclusive, but tentatively, I conclude that ditransitive verbs in Yimas take two undergoers, one by Principle A and one by Principle B. But paralleling the way RRG has rejected the notion of a universal grammatical category of subject, although it can serve as a useful descriptive category for some languages such as English, perhaps we should consider abandoning the concept of the macrorole undergoer as a necessary universal category present in the grammar of all languages, although again valuable for many languages. The concept of undergoer does no necessary descriptive work in Yimas, and insisting that we find one raises a host of problems. The lack of verb-based lexical alternations as with English load above also suggests its lack. The behaviour of applicative affixes in the language can be simply described in terms of adding core arguments with no reference to the concept of undergoer, which in fact just complicates their description. But in keeping with the spirit of RRG and this volume, I will use the notion in the remainder of this chapter. The question now advanced is whether

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

Yimas is unique or do other triple-agreement languages pose the same challenge? It appears at least the Muskogean languages do.

23.5

Case Marking: Split Ergativity and Inversion

Case marking in Yimas is very complex, but there is no quirky or semantically based case. All core case marking is determined by the function and relative topicality of the arguments, and all core case is realized by verbal bound pronominal affixes. The only other case marking is the sole -n ~ -nan for non-core constituents illustrated in (4) and (5). Core case marking in Yimas is a complex blend of split ergative-accusative alignments overlaid by a direct-inverse system. The split system separates the local persons, namely the immediate speech act persons, first and second, from the non-local person, the third person. The local persons, first and second, align according to a three-way system, distinguishing the transitive actor in ergative case, from the intransitive subject, actor or undergoer, S in Dixon’s (1979) terms, in nominative case, and from the transitive undergoer in accusative case. The non-local third person distinguishes the transitive actor with ergative case from the intransitive subject and the transitive undergoer, both marked with nominative case. Tables 23.1 and 23.2 present the forms. Note that the second-person and third-person singular ergative prefixes are homophonous. For ditransitive verbs, the theme argument, that which is given or transferred, is indicated by the relevant nominative prefix, and the recipient by the accusative prefixes for the local persons, but by a distinct set of dative suffixes for non-local persons. In the actual marking of verbs with

Table 23.1 Yimas agreement affixes for local persons

DL 1 PL SG DL 2 PL SG

PRONOUN

ERG

NOM

ACC

kapa ipa ama kapwa ipwa mi

ŋkrakaykaŋkrannann-

kapaipaamakapwaipwama-

ŋkrakraŋaŋkulkulnan-

Table 23.2 Yimas agreement affixes for non-local persons

SG DL PC PL

PRONOUN

ERG

NOM

DAT

mn mrm mŋkt mum

nmpŋklmpu-

naimpakrapu-

-nakn -mpn -ŋkt -mpun

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

871

872

WILLIAM A. FOLEY

these affixes, two principles apply, both aspects of the direct-inverse system of the language: first, local persons outrank non-local persons, specifically 1st > 2nd > 3rd, and second, surprisingly, accusative prefixes outrank ergative ones which in turn outrank nominative ones. There are also two purely structural constraints: one, the most highly ranked prefix must appear in the immediately preverbal position, and, two, every verb should have a nominative affix, and that must occur on the left edge of the verb in initial position, unless usurped or demoted by a modal or other inflection to the right edge, as in (4). Let’s see how this system all works. Consider how one would say ‘I hit them’ versus ‘they hit me’ in Yimas. The direct form, ‘I hit them’ (local person acting on non-local person) is straightforward. The local person is the transitive actor and ergative, hence ka- 1sg erg. The nonlocal person is the transitive undergoer, and the case-marking system for the non-local person is ergative-nominative, so the undergoer is realized as nominative, pu- 3pl nom, simultaneously satisfying the requirement for an overt nominative. Because local > non-local and ERG > NOM, the firstperson ergative prefix will occupy the salient immediately preverbal position, and the third-person nominative the left edge. (9)

pu-ka-tpul 3pl.nom-1sg.erg-hit ‘I hit them.’

The inverse form ‘they hit me’ is a little more complicated. The undergoer is a local person, and local persons have accusative forms (e.g. ŋa- 1sg acc). Note that the local person is higher by both local > non-local, and ACC > ERG, so ŋa- 1sg acc must occupy the salient immediately preverbal position. The non-local actor would normally take ergative case, and indeed there is an ergative form for third plural: mpu- 3pl erg. But the expected form *mpuŋa-tpul is ungrammatical because it runs afoul of the requirement for an overt nominative; such a verb lacks a nominative prefix on the left edge. So instead, the third plural pronominal is realized by the corresponding nominative prefix, pu- 3pl nom. (10)

pu-ŋa-tpul 3pl.nom-1sg.acc-hit ‘They hit me.’

Ditransitive verbs are no different. Consider ‘I gave it to them’ versus ‘they gave it to me’, the first, direct, the second, inverse. The direct form is straightforward. The non-local recipient undergoer is realized by a dative suffix, so it doesn’t compete with the interaction of the prefixes where the direct-inverse system holds sway. The actor, the ergative pronominal, is first person and therefore the higher ranked by both local > non-local and ERG > NOM, so it appears in the immediately preverbal position. The theme argument undergoer is inanimate; themes with human or higher animate

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

referents are completely prohibited for ditransitive verbs in Yimas. The affixes in Tables 23.1 and 23.2 only apply to referential expressions with human or very high animate referents, like pigs, dogs or crocodiles. Nouns with such referents belong to one of three noun classes. Nouns with all other referents and particularly all nouns with inanimate referents belong to one of the other seven major or a handful of minor noun classes. These classes also have verbal bound referential expressions, but they usually occur in only one case, nominative, again demonstrating the privilege of the nominative case. For our examples here let us take muraŋ ‘oar’, a noun of noun class VI, as the referent of ‘it’, the theme undergoer; the relevant nominative verbal affix is k- vi.sg.nom. As it is a nominative prefix, it satisfies the requirement of an overt nominative and will appear on the left edge, so the resulting form is as in (11). (11)

k-ka-ŋa-r-mpun vi.sg.nom-1sg.erg-give-perf-3pl.dat ‘I gave it to them.’

The inverse form is more interesting, as local recipient undergoers are realized by accusative pronominals, the most highly ranked. As a firstperson undergoer realized by an accusative pronominal, it ranks highest by both local > non-local and ACC > ERG > NOM, so it must occupy the immediately preverbal position. Now the third-person plural actor and the third-person singular theme undergoer slug it out. Both are third person, so local > non-local will not distinguish them, but ERG > NOM will. The actor argument is ergative, and because the theme undergoer is inanimate, it only has the option of being nominative, so by ERG > NOM, the actor argument is of the next highest rank and occupies the position next closest to the verb after the accusative prefix, while the hapless theme argument satisfies the requirement of an overt nominative on the left edge. (12)

k-mpu-ŋa-ŋa-t vi.sg.nom-3pl.erg-1sg.acc-give-perf ‘They gave it to me.’

Let’s see how we could handle this very complex case system by extending to Yimas case assignment the rules for ergative/accusative constructions provided in Van Valin (2005: 108), reproduced here as (13). (There is a much more detailed and typologically varied description of these in Van Valin and La Polla (1997), but the points made below would still apply.) (13)

Rules for accusative case constructions: a. assign nominative case to the highest-ranking macrorole b. assign accusative case to the other macrorole Rules for ergative case constructions: a. assign absolutive case to the lowest-ranking macrorole b. assign ergative case to the other macrorole

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

873

874

WILLIAM A. FOLEY

It is clear that these will need some modification for Yimas. First, we have no need for an absolutive case, and invoking it would greatly complicate the description. So-called absolutive case is simply nominative, in Yimas and, I would argue, universally. That is why, like nominative in accusative constructions, erstwhile absolutive case constituents are normally formally unmarked and why the nominative is always the privileged case in Yimas regardless of whether it shows up in the accusatively aligned local persons or the ergatively aligned non-local persons. Further, the problem of the lower-ranked macrorole in (13a) is problematic in the case of ditransitive verbs with two undergoers, but let us see how we go. For non-local persons the modification is very straightforward: simply assign nominative case to the lowest argument on the AUH (and dative case to the other one if present) and ergative case to the highest macrorole. The rules for case assignment to the local persons are more complex, because the rules for accusative and ergative case constructions apply simultaneously and partially, that is, only something like the (b) conditions apply. If we were to apply all the conditions, we will get case conflict, a transitive actor will be assigned nominative case by accusative condition (a), but ergative by ergative condition (b). But if we absorb the (a) conditions into and restate the (b) conditions, the system will work: (a) assign accusative case to the lower-ranked macrorole when there are at least two (there is a complication with ditransitive verbs with local recipients; there assign accusative case with the next to highest (or lowest, either will work) argument on the AUH); (b) assign ergative case to the highest-ranked macrorole, again where there are at least two; (c) assign nominative to any core argument that has not received case by (a) or (b). The tree diagram in Figure 23.3 represents the mapping of (11) using these rules from logical structure to constituent structure.

23.6

Pivots or Privileged Syntactic Arguments

There is no evidence in Yimas for a privileged syntactic argument (PSA) such as subject in English or the ang phrase in Tagalog. Different constructions target different constituent types, and no one type emerges as privileged across the wide range of constructions. Most constructions in Yimas simply target the notion of core argument. Yimas, like many Papuan languages, employs clause chaining as a clause linkage device (see Roberts, this volume, Chapter 25). But unlike most, it has no restrictions for this to shared subjects or switchreference tracking and it permits clause chains to be formed on any shared argument without any morphological difference, including none (14b).

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

SENTENCE

CLAUSE

CORE

RX

RX

NUCLEUS

RX

PRED

V

k-

ka-

VI.SG.NOM 1SG.ERG

CASE:

MACROROLES:

LEXICON:

a-rgive-PERF

ERG

DAT

ACTOR

UNDERGOER

-mpun 3PL.DAT

NOM

UNDERGOER

[do′ (I, Ø)] CAUSE [BECOME have′ (them, it (VI.SG))]

Figure 23.3 Linking from semantics to syntax in (11) of Yimas

(14)

a. tmal kray-mpi ya-kay-am-wat sun.v.sg dry-seq V.pl.nom-1pl.erg-eat-hab ‘the sun having dried it, we eat the food’ b. tmal l-ŋka-p(u)-mpi kumpwia sun.v.sg down-go-away-seq flying.fox.viii.pl wa-kay-tay viii.pl.nom-1pl.erg-see ‘The sun having set, then I saw flying foxes.’

amtra food.v.pl mnta then

In (14a) the argument shared between the clauses and elided from the first clause is the undergoer, while in (14b) there are no shared arguments. Some

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

875

876

WILLIAM A. FOLEY

constructions do have an intransitive subject/transitive subject (S/A) pivot, such as genitivization in nominalized non-finite complements. (15)

a. God-na anti papak-t-wal God-poss ground.viii.sg carve-nfn-custom.v.sg ‘God’s making of the world’ b. *anti-ɲa God papk-t-wal ground.viii.sg-poss God carve-nfn-custom.v.sg ‘the world’s making by God’ c. Yakayapa(n)-na am-t-wal pn-poss eat-nfn-custom.v.sg ‘Yakayapan’s (way of ) eating’

In (15a) the actor of a transitive verb papk ‘carve’ has been genitivized in the nominalized non-finite complement, but if we try this with the undergoer in (15b), the result is ungrammatical (though okay in English). However, with the single argument of an intransitive verb (15c), genitivization is again possible. Other constructions, though, work on an intransitive subject/transitive object (S/O) pivot, for example the scope of elevational/directional affixes. Consider these examples. (16)

a. kay i-ɲa-l-ampu-n canoe.viii.sg viii.sg.nom-pres-down-float-pres ‘The canoe is down there.’ b. kay naŋ-l-arm-na-ŋkan-i canoe.viii.sg imp.pl-down-board-imp-pc-viii.sg.nom ‘You all board the canoe down below.’ ‘*You all down below board the canoe.’

Note that in (16a) the scope of the elevational prefix l- is over the sole argument of the intransitive verb ampu- ‘float’, while in (16b) it is over the undergoer of the transitive verb arm- ‘board’ and cannot be construed as modifying the actor, a clear S/O pivot. The only way to say the equivalent of the starred translation would be something like ‘you all are standing/sitting down below and now you all board the canoe’.

23.7

Clause Linkage and Nominalization

23.7.1 Non-Finite Nominalizations Yimas contrasts with many Papuan languages and in particular with those of the Trans New Guinea family like Amele (Roberts, this volume, Chapter 25) in preferring subordination rather than the cosubordination by extensive clause chaining as its favourite clause linkage device. It accomplishes subordination through nominalizations, both non-finite and finite. Yimas completely lacks the category complementizer; all nominalizations are simply complex nouns, and as nouns, all must be marked for noun class and, if relevant, number. The formal template is easiest to see with nonfinite agentive nominalizations, such as irut ampa-r-mprum mat.ix.pl weavenfn-ii.dl ‘two female mat-weavers’; a parallel full finite clause would be: https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

irut imp(a)-ampa-wat mat.ix.pl 3dl.nom-weave-hab ‘they both weave mats.’ Note first of all that the nominalization makes a distinction in noun class, here noun class II, the class for female humans, that the corresponding finite verbal prefix impa- 3dl.nom cannot; this is in keeping with its status as a noun, distinguishing the full range of classes that nouns do. The finite tense marker -wat hab is replaced by the non-finite marker -ru (with a number of allomorphs). All core argument marking by bound verbal affixation beyond the realization of the agent that the entire nominalization refers to is now completely prohibited; the undergoer of the finite clause can be carried over into the nominalization, but it cannot ever be realized by a bound RX: *irut w(a)-ampa-r-mprum mat.ix.pl ix.pl.nom-weave-nfn-ii.dl, *w(a)-ampa-r-mprum ix.pl.nom-weave-nfn-ii.dl. Nor can the undergoer be genitivized as a modifier of the nominalization, as it can in English: *irutɲa ampa-r-mprum mat.ix.pl-poss weave-nfn-ii.dl ‘the two female weavers of mats’. This indicates that the undergoer must remain inside the core of the nominalization to satisfy the Completeness Constraint for the verb. But the whole nominalization can be modified, for example by an adjective: irut ampa-r-mprum yua-mprum mat.ix.pl weave-nfn-ii.dl good-ii.dl ‘two good female mat-weavers’. The concordial pattern here tells us that the head of the nominalization is the noun class and number suffix. The non-finite verb is embedded underneath it as a core, with its subcategorized arguments. Because this is an agentive nominalization, the actor is actually bound by the head of the nominalization, the noun class and number suffix, and therefore prohibited from occurring. This all suggests the structure in Figure 23.4 for this construction. All non-finite nominalizations in Yimas have the structure of Figure 23.4. As we shall see, the structure of finite ones is very similar except that the node of clause is introduced with its operators, modifiers and inflectional possibilities. The nucleus of a non-finite nominalization can be as complex as a finite verb, simply lacking tense and verbal pronominal agreement, and can consist of incorporated adverbials, applicative derivations or verbs in a serial verb construction, that is, verbs linked in a nuclear juncture, as in these examples: kpa-nti-pramuŋ-tu-mat big-adv-sleep-nfn-i.pl ‘men who sleep too much’, taŋkway-cakal-cu-mprum watch.over-feel-nfn-ii.dl ‘the two women looking after (the kids)’, namtamparawt-ɲan api-c-awt foot.ix.dl-obl put.innfn-sg ‘a sock (that which puts inside at the feet)’. The different kinds of non-finite nominalizations in Yimas cover the midrange of the Interclausal Semantic Relations Hierarchy (Van Valin 2005: 208), psych action, jussives, etc. What distinguishes the different types of non-finite nominalizations are the different heads or noun class markers they take (or head noun, as in complements of desire), but they all have the structure of Figure 23.4. When they function as complements, that is, as core arguments of a matrix verb, they can and usually do trigger agreement with that verb, the noun class marker of the nominalization matching the affix on the verb. Beside agentive nominalizations, there are four other types of nominalization that function as complements, exemplified in (17).

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

877

878

WILLIAM A. FOLEY

RX

COREN

CORE

RX

NUCLEUSN

NUCLEUS

RX

PRED

V

irut

ampa-r

mat.IX.PL

N

-mprum

weave-NFN

II.DL

NUCLEUSN

COREN

NUM

RX Figure 23.4 Constituent structure of a non-finite nominalization in Yimas

These are marked by specific noun class marker suffixes utilized only for this function, or in the case of complements of desire by a bound noun wampuŋ, literally ‘heart’. Furthermore, as true complements they function as core arguments of the complement-taking verb of the core of the main clause and are therefore commonly (17a, b, c), though not obligatorily (17d), indicated there by a corresponding bound verbal affix of the same noun class. (17)

a. complement of words or thought (thinking is internal speech) patn wayk-r-mpwi pia-ka-i-c-mpun betelnut.v.sg buy-nfn-talk talk.nom-1sg.erg-tell-perf-3pl.dat ‘I told them to buy betelnut.’

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

b. complement of activity nam wark-t-nti tia-ka-ira-karŋkra-t house enclose-nfn-act act.nom-1SG.erg-appl-tired-perf ‘I’m tired of building houses.’ c. complement of customary or habitual action yaki am-t-wal ntak-na-k tobacco.v.pl eat-nfn-custom.v.sg leave-imp-V.sg.nom ‘Stop smoking!’ d. complement of desire tpuk am-t-wampuŋ kpa-n ama-na-t-n sago.x.sg eat-nfn-heart.v.sg big-v.sg 1sg.nom-pres-feel-pres ‘I really want to eat sago.’ (‘I have a big desire to eat sago’)

Note in (17d) that the adjective kpa ‘big’ modifies the nominalization. Core arguments of non-finite nominalizations may be the pivot of control constructions and when they are, the pivot is strictly S/A; undergoers may never be controlled. Otherwise, genitivization applies, again strictly under an S/A pivot, illustrated above in (15) and in this example. (18)

patn kpuc-t-wal betelnut.v.sg chew-nfn-custom.v.sg ‘their (manner of ) chewing betelnut’

mpu-na-kn 3pl-poss-v.sg

The controllers of the S/A pivot in control constructions can be any of the three core arguments, and what determines which is the Theory of Obligatory Control (Foley and Van Valin 1984: 307–311; Van Valin 2005: 243): (1) causative and jussive verbs have undergoer control and (2) all other transitive verbs have actor control. In control structures in Yimas the nominalized complement always functions as an undergoer, so with simple transitive complements there is no issue: the controller can only be the actor, the only other macrorole, as in (17b, d) above and in the examples below. (19)

a. patn wayk-r-mpwi betelnut.v.sg buy-nfn-talk ‘I forgot to buy betelnut.’ b. yaki am-t-wampuŋ tobacco.v.pl eat-nfn-heart.v.sg ‘He wants to smoke.’

pia-ka-kacapal talk.nom-1sg.erg-forget na-na-t-n 3sg.nom-pres-feel-pres

Ditransitive verbs are more complicated as there are two potential controllers, the actor or the second undergoer. In accord with the Theory of Obligatory Control, if the complement is a causative or jussive one, there is undergoer control, as in (17a) and (20). (20)

impram pay-c-mpwi basket.vii.sg carry-nfn-talk ‘He asked us to carry a basket.’

na-kra-kankantakal 3sg.nom-1pl.acc-ask

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

879

880

WILLIAM A. FOLEY

Note that crucially the jussive or causative semantics is necessary; just because a verb of saying is ditransitive does not mean it will have undergoer control. (21)

kay yamal-c-mpwi pia-mpu-ŋa-taŋ-tmi canoe.viii.sg carve-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk ‘They talked with me about (my, their or someone else’s) building a canoe.’

Here, although the verb is ditransitive through derivation with an applicative, there is no obligatory control, because the semantics is not causative or jussive. Ditransitive verbs that do not have causative or jussive semantics can have, as expected, actor controllers. (22)

tpuk am-t-mpwi pia-mpu-ŋa-taŋkway-cmi sago.x.sg eat-nfn-talk talk.nom-3pl.erg-1sg.acc-appl-talk ‘He told me (while looking at me) about (his) eating sago.’

In addition to functioning as complements, non-finite nominalizations can function as peripheral adjuncts and as modifiers, essentially non-finite relative clauses. Let us look at the latter first as they are simpler. All relative clauses in Yimas, finite and non-finite, are simply juxtaposed in apposition to the noun or referring expression they modify, just as adjectives can be (17d), They are adjoined (Hale 1976) and not embedded, as in this example. (23)

pu-k namat pu-ŋkl-awl-k 3pl-prox person.i.pl 3pl.nom-3pc.erg-get-irr [nampt wark-r-mat] house.pl enclose-nfn-i.pl ‘They few got these people who build houses.’

The structure of (23) is essentially that of a non-finite nominalization nampt wark-r-mat house.pl enclose-nfn-i.pl ‘the ones (masculine) building houses’ as in Figure 23.4 modifying a core argument pu-k namat 3pl-prox man.i.pl ‘these men’, in the same way as kamta-k-wa empty-irr-ix.sg ‘empty’ modifies parwa-n dock.ix.sg-obl ‘at the dock’ in Figure 23.1. Non-finite relative clauses differ from the nominalizations that we have seen thus far only in the noun class marker they take. Instead of the specialized noun class markers for complements in (17), non-finite relative clauses just use the basic adjectival concord set for the various noun classes and numbers of the nouns the modify: awruk [awt yara-t-uŋ] torch.x.sg [fire yara-nfn-x.sg] ‘a torch for picking up fire’, with just some exceptions for those modifying nouns denoting human males or females. Non-finite nominalizations functioning as peripheral adjuncts always occur with the oblique suffix -n ~ -nan. They have a very specialized semantic function, indicating two events which are simultaneous. Such a semantic relation is quite low on the Interclausal Semantic Relations Hierarchy (Van Valin 2005: 208), lower, for example, than circumstances, reasons, conditionals and concessives, but this is inaccurate for Yimas, as all of these latter are expressed by finite nominalizations, while simultaneous events must be

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

expressed with non-finite nominalizations. While oblique non-finite nominalizations have final noun class and number markers as their heads, require -ru and have the basic structure of Figure 23.4, they differ in two crucial respects: first, their actors, if overt (and they are always actors, as I have no examples of states, achievements and accomplishments occurring in this construction), must occur as bare RXs like undergoers and may not be genitivized (24c). In this feature, oblique non-finite nominalizations are more like clauses than the ones previously presented, though formally they are still clearly cores attached to a nominal nucleus which is a noun class and number suffix, which is then inflected for oblique case. And second, the noun class and number suffixes do not show concord with an external RX, but rather mark those features for the actor of the nominalization, as in these examples. (24)

a. [mpa now

irm-kia-r-ŋkt-ɲan] stand-night-nfn-pc-obl paŋkra-na-ma-ŋka-pu-kia-k 1pc.nom-pres-in-go.by.land-away-night-irr ‘Standing now, we few walk inside.’ b. [wark-r-mat-ɲan nam] kumpwi mnta enclose-nfn-i.pl-obl house boy.I.pl then numa-mpu-ntak-t village.nom-3pl.erg-leave-perf ‘While building a house, the boys left the village.’ c. arm nampt ya-mpu-tawɲcak-kia-k water house.pl house.pl.nom-3pl.erg-flood-night-irr [m-um pay-kia-r-mat-ɲan num-un-mat] nr.dist-i.pl lie-night-nfn-i.pl-obl village-obl-i.pl ‘The water flooded the houses, while they, the villagers, slept.’

In (24a, b) the actor of the nominalizations is the same as that of the matrix verb, and being such, it can be elided. But such control is not obligatory, as (24c) demonstrates. There the actor of the nominalization is num-un-mat village-obl-i.pl ‘villagers’, while that of the matrix verb is arm ‘water’, which is always formally plural in Yimas, hence the bound prefix mpu- 3pl.erg. Any core argument can be the controller of an elided actor of an oblique nonfinite nominalization. (25)

pu-kra-ant-t pan-t-mat-nan 3pl.nom-1pl.acc-hear-perf pound.sago-nfn-i.pl-obl ‘They heard us when pounding sago.’ (Either they or we could be pounding sago.)

Finally, these oblique non-finite nominalizations can be used to describe locations. In that case, the noun class and number marker is invariably -a ix. sg, but there is no known synchronic noun of that class in Yimas that the affix could refer to. It seems to be fossilized, but the overall structure is typical of a non-finite nominalization: mawrun tu-r-a-n enemy.i.sg kill-nfn-ix.sg-obl

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

881

882

WILLIAM A. FOLEY

‘a place for killing the enemy, battlefield’, awŋkw-cpaŋ-t-a-n go.down.into. water-bathe-nfn-ix.sg-obl ‘a bathing place’.

23.7.2 Finite Nominalizations: Relative Clauses Finite nominalizations occur with tense suffixes and bound verbal pronominal affixes for core arguments. Finite nominalizations are used in relative clauses and oblique (i.e. adverbial-type) clauses as peripheral adjuncts of the matrix clause, but they may never be used as complements (i.e. core arguments of the matrix verb). All finite nominalizations essentially have the structure of relative clauses, and the formation of relative clauses is somewhat complicated in Yimas, so I can only give a brief overview here. Their formation is complicated as they often involve yet another linear rearrangement of bound pronominal affixes for core arguments, the complexities of which, due to the split case system and the direct-inverse system, are considerable. Relative clauses are adjoined, not embedded, so the noun being modified does not form a constituent with it and, if recoverable from context, can be elided. The relative clause itself is headed by a noun class and number suffix on its verb that is in concord with the noun that it modifies or, if missing, that is recoverable from the context. This bound noun class and number suffix indexes one of the core arguments of the verb, actor or undergoer, that is the relativized noun, and in that sense, relative clauses in Yimas can be seen as internally headed, as the external modified noun, if present, is always resumed in the relative clause by a bound verbal affix according to its noun class and number. That noun class and number suffix, like those in non-finite nominalizations, occurs at the end of the verb of the relative clause and heads it, though not necessarily at the end of the relative clause, as constituents, especially peripheral adjuncts, but not restricted to them, can follow the verb. The noun class and number marker binds the missing relativized noun in the relative clause and provides its referent. The very simplest relative clauses are just of this structure; generally they do not co-occur with a noun and so function as RXs on their own: mal-k-n die-irr-i.sg ‘a corpse’, ŋa-t-ø giveperf-v.sg ‘a gift’, kalc-k-n strengthen-irr-i.sg ‘a strong person’, wa-kia-k-ra go-near-irr-v.pl ‘means/route of going’. Their structure is represented in Figure 23.5. These simple relative clauses function as kinds of patientive nominalizations in opposition to the agentive nominalizations described earlier. (26)

tu-r-awt kill-nfn-i.sg ‘killer’ amp-r-awt kindle-nfn-i.sg ‘fire lighter’

tu-t-Ø kill-perf-i.sg ‘someone killed’ amp-(r)-ra kindle-perf-v.pl ‘firewood’

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

kalc-r-awt strengthen-nfn-i.sg ‘hard worker’ am-t-awt eat-nfn-i.sg ‘eater’

kalc-k-n strengthen-irr-i.sg ‘a strong person’ am-t-ra eat-perf-v.pl ‘food’

Most relative clauses are rather more elaborate than this. The verb of the relative clause (and many are just the verb) is typically suffixed as a whole to the near distal deictic stem m- ‘that’. Further, they can have bound RX

COREN

CLAUSE

NUCLEUSN

CORE

NUCLEUS

RX

PRED

V

TENSE

N

wa-kia-k

-ra

go-NEAR-IRR

V.PL

NUCLEUS

NUCLEUSN

CORE

COREN

CLAUSE

NUM

RX

Figure 23.5 Constituent structure of a simple finite nominalization in Yimas

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

883

884

WILLIAM A. FOLEY

pronominal affixes for their subcategorized core arguments. Consider these examples. (27)

a. namat [m-na-taw-nt-um mnti] person.i.pl nr.dist-pres-sit-pres-i.pl there ‘the people who are sitting there’ b. namat [m-kra-tpul-c-um] person.i.pl nr.dist-1pl.acc-hit-perf-I.pl ‘the people who hit us’ c. namat [m-kay-tpul-c-um] person.i.pl nr.dist-1pl.erg-hit-perf-i.pl ‘the people who we hit’

Example (27a) illustrates a case where the head and relativized argument is the intransitive subject. This single core argument is realized solely by the head suffix, the verb bearing no other bound pronominal bound affix. In (27b) the relativized noun and head is the actor of a transitive verb, realized by the final head suffix, while the undergoer is realized by its proper prefix. Finally, in (27c) the relativized noun and referent of the final head suffix corresponds to the undergoer argument of a transitive verb, while the actor appears now as a prefix in its proper position. All types of core arguments and even peripheral adjuncts are relativizable (though recipient undergoers with difficulty due to morphological complications (see Foley 1991: 417–418)); here are some examples. (28)

a. on actor Elias [m-kra-pay-pra-kia-ntuk-ŋkt-ø mota-nan] pn nr.dist-1pl.acc-carry-toward-near-rm.past-pc-i.sg motor-obl ‘(It was) Elias who brought us few by motor.’ b. on undergoer anti [God m-n-papk-ntuk-i] ground.viii.sg God nr.dist-3sg.erg-carve-rm.past-viii.sg ‘the world which God made’ panmal m-n-tpul-c-(n)ak narmaŋ man.i.sg nr.dist-3sg.erg-hit-perf-i.sg woman.ii.sg ‘the man which the woman hit’ c. on theme undergoer impram [m-(n)-nan-(n)a-ampa-ŋa-nt-m] basket.vii.sg nr.dist-3sg.erg-2pl.acc-pres-weave-appl-pres-vii.sg ‘the basket which she is weaving for you’ d. on peripheral location (again the fossilized suffix -a ix.sg is employed) [maramara m-mpu-t-r-a-n] goods.v.pl nr.dist-3pl.erg-lay.down-perf-ix.sg-obl m-ra ya-kay-ɲa-tacay-kulanaŋ nr.dist-v.pl v.pl.nom-1pl.erg-pres-see (red: tay-)-walk ‘Where they laid out the goods, we walked around looking at them.’ (Literally ‘at the place that they laid out the goods. . .’)

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

e. as possessor [awt m-nanaŋ-pampay-caw-na-ntut- ø] na fire nr.dist-dur-appl-sit-dur-rm.past-i.sg poss kalakn na-n-tay-mpi-yara-k child.i.sg 3sg.nom-3sg.erg-see-seq-pick.up-irr ‘The son of him who was tending the fire found it.’ f. as possessed manm p-ka-tay m-ɲa cult.house.vii.sg vii.sg.nom-1sg.erg-see nr.dist-poss [m-mpu-tkam-r-m] nr.dist-3pl.erg-show-perf-vii.sg ‘I saw the cult house, yours, which they showed.’

Figure 23.6 illustrates the proposed representation for the relative clause of the second example in (28b). What is unusual about Yimas in comparison to more familiar languages is that the entire relative clause, with the exception of the ‘leaked’ core argument narmaŋ ‘woman’, corresponds to a single word, and such words can be quite complex (28a, e). But that is entirely in keeping with the polysynthetic typological profile of the language; the word is its favoured grammatical unit and structures above that are very weakly syntactically developed. This fact should again caution us against building grammatical theories solely on the basis of what is familiar to us, such as constituency: they blind us to other possibilities. The final type of relative clause to consider are those in which the relativized noun is a temporal expression. These are obliquely case-marked like relative clauses on locative adjuncts, and, also like them, the head of these temporal relative clauses is a fossilized noun class and number suffix, of the form -mp vii.sg. But in this case we know the noun to which it refers because it is still synchronically in the language: pucm ‘part, piece, time’. All the semantic relations on the Interclausal Semantic Relations Hierarchy that are expressed typically by adverbial subordinate clauses in more familiar languages, circumstances, reasons, conditionals and hypotheticals and concessives, are expressed by this type of temporal relative clause in Yimas. The differences between them are captured in the choice of tense and mood for both the relative clause and the matrix clause. For example, for temporal circumstances, the verb of the relative clause occurs with one of the past or present tenses or that of the definite future. (29)

[m-mpu-ŋa-na-tay-ɲc-mp-n] nr.dist-3pl.erg-1sg.acc-pres-see-pres-vii.sg-obl pu-ka-apan-kt 3pl.nom-1sg.acc-spear-rm.fut ‘When they see me, I will spear them.’

But conditionals and hypotheticals are marked with mood affixes that mark unreal or as yet unrealized events.

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

885

886

WILLIAM A. FOLEY

RX

COREN

CLAUSE

NUCLEUSN

CORE

NUCLEUS

PRED

RX

m-

n-

NR.DIST 3SG.ERG

TENSE

V

RX

N

RX

tpul-ct l-ctpu

-(n)ak -(n (n)ak

narma narm r

hit-PERF

I.SG

wom woman

NUCLEUS

NUCLEUSN

CORE

COREN

NUM

CLAUSE

EIC DEIC

RX

Figure 23.6 Constituent structure of a relative clause (28b) in Yimas

(30)

[m-mpu-ya-kr-mp-n Wamur-mat] nr.dist-3pl.erg-come-rm.fut-vii.sg-obl Wambramas-i.pl mambayŋki wunt kantk banana.vi.pl sago.grub.v.pl with pay-pra-kt ama-na-ra carry-toward-rm.fut 1sg-poss-v.sg ‘If the Wambramas people will come, they can bring my bananas and sago grubs.’

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

A Grammatical Sketch of Yimas

Counterfactuals are slightly different. They require the potential modality prefix ant-, which expresses that an event is unlikely to happen, to co-occur with a past- or present-tense marker on the verbs of both the relative clause and the matrix clause. As the potential modality marker is a prefix, it triggers rearrangement on verbs of the matrix clause of the normally leftmost bound nominative pronominal prefix to right edge. (31)

[tuŋkuruŋ ant-ka-tay-c-mp-n] eye.vi.sg pot-1sg.erg-see-perf-vii.sg-obl ant-ka-tu-r-(n)ak pot-1sg.erg-kill-perf-iii.sg.nom ‘If I had seen the eye (of the crocodile), I would have killed it.’

Finally, concessives are formed by having the polarity of the relative clause being positive, and that of the matrix clause, negative. (32)

23.8

[kay i-ka-ak-r-mp-n] canoe.viii.sg viii.sg.nom-1sg.erg-push-perf-vii.sg-obl arm-n ta-ka-wul-c-i water-obl neg-1SG.erg-put.down-perf-viii.sg ‘Although I pushed the canoe, I didn’t put it down into the water.’

Conclusion

Yimas is an intriguing language that has much to offer grammatical theory in its unusual grammatical structures due to its complex and rare mix of a polysynthetic verb structure with an elaborate noun-class concordial system. Sadly, it is now rapidly approaching extinction, but fortunately it was documented when it was still in a relatively vibrant state and its full richness able to be appreciated. Without the input of data from languages like Yimas, our grammatical theory will be very impoverished indeed. But many such crucial languages like Yimas are dying before our very eyes. Grammatical theory is a scientific endeavour worthy of pursuit, but unless we document the full range of grammatical possibilities, it will never be complete. We may never know the grammatical riches the human language capacity is able to evolve. And time is rapidly running out to do that. Our discipline needs seriously to look at its priorities.

References Bresnan, Joan and Lioba Moshi. 1990. Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21: 147–185. Broadwell, George Aaron. 2006. A Choctaw Reference Grammar. Lincoln, NE: University of Nebraska Press. Dixon, Robert M. W. 1979. Ergativity. Language 55: 59–138.

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

887

888

WILLIAM A. FOLEY

Duranti, Alessandro and Ernest Byarushengo. 1977. On the notion of ‘direct object’. In Ernest Byarushengo, Alessandro Duranti and Larry Hyman (eds.), Haya Grammatical Structure, 54–71. Los Angeles: Southern California Occasional Papers in Linguistics. Foley, William. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge University Press. Foley, William. 1991. The Yimas Language of New Guinea. Stanford, CA: Stanford University Press. Foley, William. 2005. Linguistic prehistory in the Sepik Ramu basin. In Andrew Pawley, Robert Attenborough, Jack Golson and Robin Hide (eds.), Papuan Pasts: Cultural, Linguistic and Biological Histories of Papuan-Speaking Peoples (Pacific Linguistics 572), 109–144. Canberra: The Australian National University. Foley, William. 2017a. The languages of the Sepik-Ramu basin and environs. In Bill Palmer (ed.), The Languages and Linguistics of the New Guinea Area: A Comprehensive Guide, 177–412. Berlin: de Gruyter. Foley, William. 2017b. Yimas: The profile of a polysynthetic language of New Guinea. In Nicholas Evans, Michael Fortescue and Marianne Mithun (eds.), Handbook of Polysynthesis, 808–829. Oxford: Oxford University Press. Foley, William and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Hale, Kenneth. 1976. The adjoined relative clause in Australia. In Robert Dixon (ed.), Grammatical Categories in Australian Languages, 76–105. Canberra: Australian Institute of Aboriginal Studies. Hale, Kenneth. 1983. Warlbiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1: 5–47. Martin, Jack B. 2011. A Grammar of Creek (Muskogee). Lincoln, NE: University of Nebraska Press. Munro, Pamela. 1984. The syntactic status of object possessor raising in Western Muskogean. Proceedings of the Berkeley Linguistics Society 10: 634–649. Munro, Pamela and Lynn Gordon. 1982. Syntactic relations in Western Muskogean: A typological perspective. Language 58: 81–115. Pawley, Andrew. 1993. A language which defies description by ordinary means. In William Foley (ed.), The Role of Theory in Language Description, 87–130. Berlin: de Gruyter. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory. In Balthasar Bickel, Lenore Grenoble, David Peterson and Alan Timberlake (eds.), Language Typology and Historical Contingency, 91–124. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

https://doi.org/10.1017/9781316418086.029 Published online by Cambridge University Press

24 A Grammatical Sketch of Avatime (Kwa, Niger-Congo, Ghana) Saskia van Putten and Rebecca Defina Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php) with the following additions: C CLM CM HAB INT

noun class clause linkage marker clause marker habitual intentive

24.1

Introduction

IT REC SVM VEN

itive recurrent serial verb marker ventive

In this chapter we give an overview of the grammar of Avatime from a Role and Reference Grammar (RRG) perspective. Avatime belongs to the Kwa branch of the Niger-Congo language family. Within Kwa, it belongs to the Ka-branch of the group of Ghana-Togo Mountain (GTM) languages. It is spoken in the south-east of Ghana, by about 15,000 speakers. As a background to the discussion of the grammar, two properties of Avatime phonology are important to mention. First, Avatime is a tone language. It has three level tones: low (marked ` a), high (unmarked) and extra-high (marked á). The extra-high tone has a limited distribution and occurs mostly on grammatical morphemes and function words. Second, there is advanced tongue root (ATR)-based vowel harmony. Vowels in most affixes and clitics harmonize in their ATR value with the nearest root vowel. þATR vowels are i, e, o, u and ATR vowels are i, ɛ, a, , u. _ _ The information presented in this chapter is based on data collected during several field trips to the Avatime area between 2008 and 2013.1 More elaborate descriptions of the grammar of Avatime can be found in Defina (2016a) and Van Putten (2014a). c

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

890

SASKIA VAN PUTTEN AND REBECCA DEFINA

Part I of this chapter gives an overview of the basic grammatical structures of simple sentences. It describes lexical categories, predicate classes and logical structure, the layered structure of the clause, including the operator projection, referential phrases and adpositional phrases and the linking from semantics to syntax in simple sentences. Finally, it examines focus structure. Part II discusses the syntax and semantics of serial verb constructions and the questions they raise for the Interclausal Relations Hierarchy (Van Valin 2005: 209).

PART I. BASIC GRAMMATICAL STRUCTURES

24.2

Lexical Categories

Avatime has the distinct lexical categories of noun, verb, adjective, ideophone, adverb, preposition and postposition. Nouns are morphologically distinguished by their noun-class prefixes (see Section 24.5.1) and can be modified by adjectives, numerals and determiners. Nouns are used as referring expressions. They cannot function as predicates or modifiers and there is no way to derive other lexical categories from nouns. Verbs are distinguished from other lexical categories by their ability to be marked with a subject prefix and aspect, modality, negation and direction markers (see Section 24.4.2). Verbs are used as predicates. They can be used as referring expressions only when they are nominalized. To fully nominalize a verb, the verb is reduplicated and optionally marked with the noun-class prefix ku-/ku-. There are two constructions where the verb is marked with a _ noun-class prefix without being reduplicated: in non-finite complement constructions, where it is marked with the prefix ku-/ku- and in the focus _ construction (see Section 24.7), where it is marked with the prefix ki-/ki-. In _ these cases, the verb appears to be nominalized to a certain extent, but it cannot function as a full noun in this form. Avatime has a small class of non-derived and non-ideophonic adjectives. These modify nouns and cannot be used by themselves as predicates or referring expressions. They do not share any morphological properties to distinguish them from other lexical categories. The majority of property words are not adjectives, but verbs or ideophones. Verbs which include a property or a state as part of their meaning can be turned into adjectives by reduplication, such as kpakpa ‘dry’ from the verb kpa ‘to dry’. Adjectives can be nominalized by prefixing them with a noun-class prefix. For instance, the ̣ ami ‘small one’ (referring to adjective s`ıs` ̣ ami ‘small’ can be turned into -s`ıs` something of class 1 singular). Avatime also has a small class of non-ideophonic adverbs. Like the adjectives, they do not share any morphological properties, but they are distinguished by their function as modifiers of predicates or clauses. These adverbs cannot be used as predicates, referring expressions or modifiers of nouns. Some examples are nyafɛ ‘maybe’, ` abla ‘now’ and kóko ‘already’. c

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

There is a large class of ideophones that overlaps in function with the classes of adjectives and adverbs. Ideophones are ‘marked words that depict sensory imagery’ (Dingemanse 2012, see Chapter 9 of this volume). They can be used to modify nouns, predicates and clauses, but they cannot be used as predicates or referring expressions. Ideophones are distinguished from other word classes by their marked phonology (e.g. long words, repeated syllables, long final vowels). Some examples are pititi ‘white’, ha~a~a~a~ ___ ‘intensely’ and trátrátrátrá ‘very neat’. Finally, Avatime has small sets of prepositions and postpositions. There are only two prepositions: locative ní and comitative (a)n`ı. There are five postpositions, four of which have evolved from nouns. An example is ese ‘under’, which comes from the noun kese ‘ground’ (for examples of adpositions see Section 24.5.2). As we show in Section 24.5.2, the postpositions are very noun-like in their use and can be analysed as heading referential phrases (RPs). Nevertheless, they are a separate lexical category on both morphological and syntactic grounds. Morphologically, they are distinct from nouns as they have lost their noun-class prefixes. Syntactically, they cannot be modified with nominal modifiers such as definite articles and can only occur in the position following another referential phrase.

24.3

Predicate Classes and Logical Structure

We distinguish four main Aktionsart classes in Avatime: states, activities, semelfactives and accomplishments. The primary distinguishing factor is their interpretation with progressive and perfective aspects. For more detail about Avatime Aktionsart classes, see Defina (2009, 2018). States are interpreted as present in the perfective, as can be seen in the example in (1). They can be marked as progressive, where they are interpreted as persistent states (e.g. ‘the building is still in Vane’). ligba¼lὲ li-lí ̣ ní ̀ ʋan ̀ _ _ c3.sg-building¼def c3.sg.sbj.pfv-be.at loc Vane ‘The building is in Vane.’ Logical Structure: be.at′ ( ̀ ʋan ̀ , ligbalὲ) _

c

c

c

(1)

c

Activities are dynamic events without an inherent endpoint. They are generally interpreted as occurring in the past in the perfective, as in example (2), and as occurring in the present in the progressive (e.g. ‘Afua is eating rice’). (2)

`afua a-tá ki-mimi _ _ _ Afua c1.sg.sbj.pfv-chew c4.sg-rice ‘Afua ate rice.’ Logical Structure: do′ (Afua, [eat′ (Afua, kimimi)]) _ _ _

Accomplishments are state-change events with internal duration. Like states, they are typically interpreted as present in the perfective, as can be

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

891

892

SASKIA VAN PUTTEN AND REBECCA DEFINA

seen in example (3), but they are interpreted like activities in progress with the progressive (e.g. ‘the door is opening’). (3)

`-pópo¼l` o o ɛ-dr` a c2.sg-door¼def c2.sg.sbj.pfv-open ‘The door is open.’ Logical Structure: BECOME open′ (` opópo)

Semelfactives are events that lack duration. Like activities, they are interpreted as past actions in the perfective (e.g. ‘she came’). However, unlike activities, they cannot be modified by duration adverbials such as ‘for 5 minutes’ and ‘quickly’. They can be used with the progressive, in which case they are interpreted as iterative or repeated actions, as in example (4). (4)

o-ne¼e ὲέ-ba c2.sg-mother¼def c1.sg.sbj.prog-come ‘The mother comes regularly.’ Logical Structure: SEML come′ (one)

As yet, no clear diagnostic for the classes achievement and active accomplishment has been found for Avatime. The typical test using the modifiers ‘for X time’ or ‘in X time’ is not applicable as this specific semantic distinction is not consistently made. There is, however, a group of atomic telic predicates where the endpoint is reached in a single step rather than gradually (Caudal and Nicolas 2005), which can be distinguished based on their incompatibility with the adverb petee ‘completely’, as in example (5). (5)

`hol` osófo¼e a-trɛ ní o o (*petee) pastor¼def c1.sg.sbj.pfv-go loc Ho (*completely) ‘The pastor went to Ho (*completely).’ Logical Structure: do′ (osófo, [go′ (osófo)]) & INGR be.at′ (Ho, osófo)

In addition to the Aktionsart predicate classes discussed above, it is worth asi mentioning causative predicates. Some verbs, such as tr ‘put on’ and b` _ ‘show’ are always causative. Other verbs have a non-causative reading when they are used intransitively and a causative reading when they are used transitively, such as dra ‘open’ in example (6) (see also (3) above). c

`-pópo¼l` -d ̀ b` asi¼ɛ a-dra o o _ c1.sg-teacher¼def c1.sg.sbj.pfv-open c2.sg-door¼def ‘The teacher opened the door.’ asi, Ø) CAUSE [BECOME open′(` opópo)] Logical Structure: do′ ( ̀ d ̀ b`

c c

(6)

c c

24.4

Clause Structure and Operators

24.4.1 Clause Structure Canonical constituent order in Avatime is SVO: the privileged syntactic argument (‘subject’; PSA) precedes the verb, and other arguments

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

Sentence

Clause

RP

Periphery

Core

Pro Nuc

RP

Pred V

ɔd ́ zɛɛ

á-

PP

ta áʋanà ní lịgbalɛ mɛ̀

Figure 24.1 The layered structure of the sentence in example (7)

(‘direct/indirect object’) follow the verb. Adverbial phrases follow the postverbal arguments. There is head-marking of the PSA on the verb. A canonical transitive sentence with an adverbial phrase in Avatime can be seen in example (7) and the syntactic structure of this example is shown in Figure 24.1. Following Van Valin (1985) and Van Valin and LaPolla (1997), we analyse the PSA-indexing prefix on the verb as an argument in the core, whereas the RP with which this prefix is coreferential is analysed as occurring outside the core, but inside the clause. This captures the observation that a lexical RP or independent pronoun is not necessary, and example (7) remains grammatical when leaving out the RP ́ dzɛɛ ‘the woman’. c

á-ta á-ʋa¼n` a ní li-gba¼le ́ -dzɛ¼ɛ _ c1.sg-woman¼def c1.sg.sbj.pfv-chew c3.pl-bean¼def loc c3.sg-room¼def mὲ inside ‘The woman ate beans in the room.’

c

(7)

Avatime makes use of the pre-core slot (PrCS) for question formation and focus marking (see also Section 24.7 and Van Putten 2016). The question word or focused RP is placed in the pre-core slot and marked with a final extra-high tone, as shown in examples (8) and (9).

á-ʋa¼ná ́ dzɛ¼ɛ c3.pl-bean¼def:foc c1.sg-woman¼def ‘The woman ate [the beans]FOC.’ c

(9)

a-ta egé ́ -dzɛ¼ɛ what c1.sg-woman¼def c1.sg.sbj.pfv-chew ‘What did the woman eat?’ c

(8)

(elic-QUIS-foc_100714_SO)

a-ta c1.sg.sbj.pfv-chew (elic-QUIS-foc_100714_SO)

The pre-detached position is frequently used to host topical expressions or expressions encoding spatial or temporal background information (see Van Putten 2014b). When an argument that is not the PSA is pre-detached, a

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

893

894

SASKIA VAN PUTTEN AND REBECCA DEFINA

Sentence

PrDP

Clause

PrCS

RP

Core

Pro Nuc Pred RP

RP

kivoe áʋaná ɔd ́ zɛɛ

V

á-

ta

Figure 24.2 A sentence with pre-detached position and pre-core slot (example 11)

resumptive pronoun occurs in its regular postverbal position, as shown in example (10). Example (11) shows a sentence with both focus-marking and a pre-detached element. The syntactic structure of this example is shown in Figure 24.2. Note that the PrCS is not the same structural position as the regular position for PSA RPs, in that it is also within the clause but outside the core. Focused elements always precede the PSA RP. (10)

á-ʋa¼n` a, ́ -dzɛ¼ɛ c3.pl-bean¼def c1.sg-woman¼def ‘The beans, the woman ate them.’

(11)

kivoe á-ʋa¼ná ́ -dzɛ¼ɛ yesterday c3.pl-bean¼def c1.sg-woman¼def ‘Yesterday, the woman ate [the beans]FOC.’

a-ta c1.sg.sbj.pfv-chew

na c3.pl

á-ta c1.sg.sbj.pfv-chew

c

c

24.4.2 Operators In this section, we give an overview of the operators that have scope over the different layers of the layered structure of the clause. The Avatime verb is usually marked with a prefix that indexes the PSA. This prefix (henceforth referred to as the PSA prefix) also encodes aspect or mood. It is optionally followed by other aspect, mood and directionality prefixes. There is no tense marking. An overview of the categories marked on the verb and the order in which they occur can be seen in (12) and an example with all slots occupied can be seen in (13). (12)

PSA þ Perfective/Progressive/Habitual/Potential/Subjunctive þ (Negation) – (Intentive) – (Recurrent) – (Directional) – Root

(13)

w m ́ -tá-zɛ̌ -zɛ-pan`ı ̣ 1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg ‘I will not be going to talk with you.’ c

c

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

(R0811291)

A Grammatical Sketch of Avatime

895

Table 24.1 Paradigms of PSA prefixes (most noun classes omitted)

me-/maki-/ki _ wo-/w mle-/mlɛe-/abe-/bɛ-

mo-/m ku-/kuwo-/w_ mla o-/ ba c

c

c

1s 1p 2s 2p C1.SG C1.PL ...

Set 1 Perfective (most verbs)

c

Person/ number/ noun class

Set 2 - Negative (with extra-high tone) - Subjunctive (some verbs) - Perfective (locative and copula verbs)

Set 3 - Subjunctive (some verbs) - Habitual (with zɛ̌ - prefix) mi-/miki-/ki-_ _ wu-/wumli-/ml_ii-/i- _ _ ibi-/b _

Potential

Progressive

máàkíạ̀ wáàmláàáàbíạ̀

mèé-/mὲέkìí-/kìí-̣ ̣ wèé-/wὲέmlèé-/mlὲέèé-/ὲέbèé-/bὲέ-

There are five paradigms of PSA prefixes, as shown in Table 24.1. For the sake of brevity, the forms for the noun classes other than class 1 singular and plural have been omitted (a full table can be seen in Van Putten 2014a or Defina 2016a). As the table shows, there is no one-to-one correspondence between each paradigm and an aspect/mood category. It is also clear that the prefixes are formally related to each other; it is likely that they have evolved from combinations of a prefix indexing the PSA plus an aspect/mood marker, but they cannot be synchronically analysed as such. We will now briefly discuss the operators one by one. For more detail about these markers and their semantics, see Defina (2018). Given that aspect marking modifies the internal temporal structure of the event itself, the aspect markers are considered to be nuclear-level operators (Van Valin and LaPolla 1997; Van Valin 2005). Avatime has four aspectual categories: perfective, progressive, habitual and recurrent. Their forms are described in (14). (14)

Aspect marking • The perfective is marked with a PSA prefix from set 1 or, in the case of locative and copula verbs, set 2. It is the most frequently occurring verb form in Avatime, some examples of which have already been presented in (7)–(11). The progressive has its own paradigm of PSA prefixes, as shown in • Table 24.1. • The habitual uses a PSA prefix from set 3, combined with the prefix zˇe-/zɛ̌ -, see example (15). • The recurrent is one of the two optional aspect/mood prefixes and indicates repeated or recurring actions and situations. It is marked with the prefix zˇe-/zɛ̌ -,2 which is always combined with one of the obligatory aspect/mood categories, see example (16).

(15)

mὲ mí-zˇe-do maw` uye li-bo¼l`e ní s ̀ lὲ 1sg.sbj.hab-hab-say God c3.sg-word¼def loc church inside ‘I preach the word of God in church.’ (life_100614_WE) c

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

896

SASKIA VAN PUTTEN AND REBECCA DEFINA

(16)

ma-zɛ̌ -w` a a-xwὲ¼na ní cocoa marketing board 1.sg.sbj.pfv-rec-do c3.pl-work¼def loc cocoa marketing board ‘I used to work at the cocoa marketing board.’ (life_AB)

There are two directionals: the itive (‘away’), marked with the prefix ze-/zɛand the ventive (‘towards’), marked with the prefix bá-/bé-, shown in example (17). These forms express orientation or motion away from or towards the deictic centre. This means they are core-level operators (see Van Valin and LaPolla 1997: 45). (17)

a-bá-feke télef` on¼ye c1.sg.sbj.pfv-ven-pick.up telephone¼def ‘He came and picked up the telephone.’

(finsto_100716_DQ)

Another core operator is the intentive, which indicates the intention of the actor to carry out the action expressed by the verb. The intentive is marked with the optional prefix tá- which combines with either the perfective aspect or the subjunctive mood. For an example of the former, see (13). Negation is expressed by a PSA prefix from set 2 with an extra-high tone. An example can be seen in (13). When the progressive is negated, this PSA prefix is followed by the prefix li-/li-. When the habitual is negated, the prefix _ zˇe-/zɛ̌ - is deleted, leaving only its tone and lengthening the vowel of the PSA prefix. Negation marking can be used for both core-level and clauselevel negation. Avatime has three clause-level operators marked on the verb: potential, subjunctive and imperative. Their properties are described in (18). (18)

Clause-level operators marked on the verb • The potential has its own PSA agreement paradigm as shown in Table 24.1. The potential PSA prefix cannot be combined with negation marking. For a negative potential interpretation, the negative perfective intentive is used. • The subjunctive is marked with a subject prefix from set 2 for some verbs and for other verbs with a subject prefix from set 3. An example is shown in (19). When the subjunctive is negated, a prefix ku-/ku- is added while the PSA _ prefix remains the same. The subjunctive has irrealis and illocutionary force interpretations. • The imperative is marked by the absence of a PSA prefix, as in ba ‘Come!’. It is only used for second-person singular commands. For other commands, the subjunctive is used. To negate imperative clauses, the negative subjunctive is used.

(19)

mla-trɛ 2.pl.sbj.sbjv-go ‘You (pl) should go.’ / ‘Go! (to multiple people)’

The final operator to be discussed here is question marking, an illocutionary force operator. Polar questions are marked with a clause-final

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

particle na or, more frequently, with intonation only. Content questions are marked with a question word in the pre-core slot (see example (8) in Section 24.4.1). According to RRG (Van Valin and LaPolla 1997; Van Valin 2005), the order of operators in a given language always follows the same pattern whereby clausal operators occur furthest from the verb, core operators closer than clausal operators, and nuclear operators are the closest to the verb. In Avatime, this prediction is mostly borne out, but there are two exceptions: (i) the (core) directional markers occur closer to the verb than the (nuclear) aspect markers, and (ii) the (core) intentive marker also occurs closer to the verb than the (nuclear) perfective marker. The directional markers have clearly grammaticalized from motion verbs (ba ‘come’ and za ‘pass’). Before grammaticalization, they would have occurred as the first verb in a serial verb construction (SVC), directly followed by the second verb (see Sections 24.9 and 24.10 for more information about serial verb constructions). As all aspect and mood marking in SVCs is on the first verb, it now precedes these directional prefixes. A similar explanation could account for the intentive following the perfective: the intentive táis also likely to be the result of a grammaticalization process from a serial verb construction, probably from the verb trɛ ‘go’ or trá` a ‘come’. Another possible explanation for the intentive following the perfective is the hypothesis that the perfective forms used to be the bare subject agreement markers unmarked for aspect/mood, as is the case in many related languages (Welmers 1973). Through language change, the default PSA prefix came to be used solely for perfective marking and thereby turned into a nuclear-level operator, which happened to occur in a position preceding the intentive marker. Figure 24.3 shows the sentence in example (13), repeated here as (20), with both constituent and operator projections. c

24.5

m ́ -tá-zɛ̌ -zɛ-pan`ı ̣ w 1.sg.sbj.pfv.neg-int-rec-it-talk 2.sg ‘I will not be going to talk with you.’ c

(20)

(R0811291)

Referential and Adpositional Phrases

24.5.1 Referential Phrases Headed by Nouns Like clauses, referential phrases (RPs) have a layered structure. The nucleus of the Avatime RP can be either a noun or a postposition. When it is headed by a postposition it functions as an adpositional phrase, and this type of RP will be discussed in Section 24.5.2. This section will instead discuss RPs headed by nouns. Avatime nouns, like those in many Niger-Congo languages, take part in a noun-class system. Each noun (with the exception of some loanwords)

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

897

898

SASKIA VAN PUTTEN AND REBECCA DEFINA

Sentence

Clause

Core

Nuc

NP

NP

Pred V

mɔ-́ tá-zɛ-̌zɛ-

panì

wɔ

V ASP

Nuc Nuc DIR Core MOD Core STAT Clause ASP

Figure 24.3 Sentence from example (20) with constituent and operator projections

carries a noun-class prefix which indicates its gender and number. Within the referential phrase, there is agreement on numerals, articles and demonstratives (see example (24)). For more detailed information on the noun-class system and the various nominal modifiers, see Schuh (1995), Van Putten (2014a) and Defina (2016a). The constituent order in Avatime RPs is shown in (21). Adjectives are considered to occur in a periphery of the nucleus, following Van Valin (2005), whereas the other nominal modifiers are operators on different layers of the RP. (21)

noun – adjectives – numeral – determiner – particles

The only operator on the nuclear level is number, which is marked on the noun class prefix (e.g. ̀ -kli ‘leg’ vs. `ı ̣-kli ‘legs’). The numerals are operators on _ _ the core level, as they provide quantification. The determiners are operators on the RP level, as they ground the RP in discourse. The class of determiners contains the elements shown in (22), of which only one can occur in a given RP. c

(22)

Determiners • The definite article, which is a monosyllabic enclitic and agrees with the noun class of the head noun.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

• The indefinite article t , prefixed with a noun-class agreement prefix. This article has a specific indefinite interpretation (e.g. ‘some’, ‘a certain’). For a non-specific indefinite interpretation, the determiner slot is unfilled. a and distal demonstrative l ̀ , both prefixed • The proximal demonstrative y` with a noun-class agreement prefix. c

c

The class of particles contains words such as tsyɛ ‘also’, k ‘by contrast’ and k`o ‘only’. Like definiteness and deixis markers, they ground the RP in discourse and are therefore considered to be RP-level operators. Possession in Avatime is mostly indicated by juxtaposition with a possessor– possessed word order. The regular independent pronouns are used as possessor pronouns. For possessed kinship terms, there is a slightly more complex construction in which the possessor pronoun fuses with the noun-class prefix of the possessed kinship term.3 With a nominal possessor, this fused possessor pronoun is also present, and, thus, in these cases there is head-marking of the possessor on the possessed noun. Examples of kinship and non-kinship terms with nominal and pronominal possessors can be seen in (23). c

! yɛ ̀ -m` a-n ‘her town’ ! ye-ne ‘her mother’ ! n` uv ̀ -ɛ li-kuto-lɛ ‘the child’s hat’ ! n` uv ̀ -ɛ ye-ne ‘the child’s mother’ c c

c c

a-n ‘the town’ ̀ -m` o-ne ‘mother’ li-kuto-lɛ ‘the hat’ o-ne ‘mother’ c

c

þ þ þ þ

c

yɛ ‘c1.sg’ yɛ ‘c1.sg’ -n` uv ̀ -ɛ ‘the child’ -n` uv ̀ -ɛ ‘the child’

c

c c

(23)

c c

The possessor occurs in the RP initial position (RPIP), analogous to both the pre-detached position and pre-core slot in the layered structure of the clause (see Van Valin 2005: 26). An example of a complex RP in Avatime can be seen in (24). The constituent and operator projections of this example are shown in Figure 24.4. (24)

bá-dzɛ kpekpe tia-b` a¼a ke-pe¼` a c1.pl.woman short c1.pl-two¼def c6.sg-house¼def ‘The house of the two short women.’

24.5.2 Adpositional Phrases As mentioned in Section 24.2, there are both prepositions and postpositions in Avatime. Both of Avatime’s prepositions – the comitative preposition (a)n`ı ‘and/with’ and the general locative preposition ní – are predicative prepositions. This means they occur in the nucleus of a prepositional phrase (see Figure 24.5 and Chapter 10 of this volume). The five locative postpositions are indicators of what Ameka (1995) calls the search domain, that is, the part of the reference object where something is located. Postpositional phrases are not locative by themselves but behave like noun phrases; they can, for instance, serve as arguments of a verb. Because of this, we analyse the postpositional phrase as an RP with the postposition as its nucleus and the noun expressing the location in an RP-initial position, analogous to possessive constructions. This makes sense from a historical perspective, as most postpositions have evolved from nouns which would have combined

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

899

900

SASKIA VAN PUTTEN AND REBECCA DEFINA

RP

CoreR

RPIP

RP

CoreR NucR

R

N

MP

bá-dzɛ

N

kpekpe tịabà =a

N NUM NucR CoreR RP

NucR

ke-pe=à

N NUM NucR QUANT CoreR DEF RP DEF

Figure 24.4 Complex RP with constituent and operator projections (example 24)

PP

CoreP RP

NucP

RPIP

CoreR

RP

NucR

Pred PREP

ní

POSTP

ɔk̀plɔnɔ ̀

abà

Figure 24.5 Constituent projection of a PP (example 25)

with the locatum in a possessive construction (e.g. ‘the upper surface of the table’ ! ‘on the table’, see also Section 24.2). In locative phrases, a postpositional phrase must be combined with the preposition ní, as shown in example (25). The constituent projection of this

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

example is shown in Figure 24.5. Instead of a postposition, a noun can also be used to indicate the search domain, as in example (26). This follows from the analysis presented above: in this case the two nouns occur in a possessive construction, as the situation with the prepositions would have been before grammaticalization was complete. Thus, example (26) literally translates as ‘located at the back of the car’. (25)

ní ̀ -kpl ¼n ̀ loc c2.sg-table¼def ‘on the table’

(26)

`-hui¼l` ní o o loc c2.sg-car¼def ‘behind the car’

c

c

c

24.6

ab` a on

ke-de¼` a c6.sg-back¼def

Macroroles, Arguments and Linking from Semantics to Syntax

In this section, we discuss how logical structure is linked to syntactic structure, including how semantic macroroles are assigned in Avatime and how the privileged syntactic argument is selected. Macrorole selection in Avatime simple sentences follows the ActorUndergoer Hierarchy (AUH) shown in (27). The highest-ranking argument is selected as the actor. When it comes to the selection of the undergoer, Van Valin (2005) shows that, across languages, there are two possible principles: either the lowest-ranking or the second-lowest-ranking argument is selected as the undergoer. When a verb has only two core arguments, the two principles will always select the same undergoer. However, a difference emerges with three-place predicates. In Avatime three-place predicates, the two non-actor arguments always follow the verb with the ‘indirect object’ preceding the ‘direct object’, as shown in example (28). The logical structure of this sentence is shown in (29) and makes clear that the highest-ranking argument is Kofi, the second-highest is the first-person singular pronoun and the lowest-ranking argument is egumena ‘the cows’. Given that there is no case marking and no indexing of either of the two arguments on the verb, there doesn’t seem to be clear evidence for which of the two is the undergoer and therefore which of the two principles applies. (27)

The Actor-Undergoer Hierarchy (based on Van Valin and LaPolla 1997 Fig. 4.2) ACTOR

UNDERGOER

-----------------------------------------------------> ’ ¼ increasing markedness of realization of argument as macrorole]

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

901

902

SASKIA VAN PUTTEN AND REBECCA DEFINA

(28)

k` ofi a-b` asi mɛ _ Kofi c1.sg.sbj.pfv-show 1.sg ‘Kofi showed me the cows.’

(29)

do′ (Kofi, Ø) CAUSE [see′ (1s, egumen` a)]

e-gume¼n` a c3.pl-cow¼def

After assigning macrorole status, the PSA is selected. In Avatime, the PSA is the argument that is indexed on the verb. Avatime has a nominativeaccusative alignment pattern: if there is only one macrorole argument, this is selected as the PSA, and if there are two macroroles, the actor is selected as the PSA. The PSA is invariable: there is no passive construction in Avatime and non-macrorole PSAs are not allowed. In example (28), the actor, Kofi, is selected as the PSA. After the selection of the PSA, the appropriate syntactic template is selected. The syntactic template for a canonical Avatime sentence has one core slot before the verb for a head-marked argument and up to two core slots for RP arguments after the verb (see Section 24.4.1). In the case of example (28), a syntactic template with two core slots after the verb is chosen, as there are three arguments in the logical structure. The preverbal head-marking slot is linked to the PSA, which is Kofi. This slot can only contain PSA-indexing prefixes and does not allow referential phrases. Therefore, it is filled with a PSA-indexing prefix of noun class 1 singular (class 1 is used for people) and the RP Kofi is placed in the pre-core position (see also Section 24.4.1). The undergoer and non-macrorole arguments are linked to the two postverbal core-internal slots.

24.7

Focus Structure

In addition to the constituent projection and operator projection, RRG posits a third component of grammar, representing the information structure of the sentence: the speech-act projection (see Chapter 11 of this volume). In this section, we briefly discuss focus marking in Avatime. For more detail, see Van Putten (2016). The potential focus domain in Avatime canonical clauses coincides with the entire clause, that is, any element of a clause can potentially be in focus. Focused elements in canonical clauses are not marked as such with intonation; the focus interpretation comes purely from the context. To make clearer which part of the sentence is in focus, a syntactic focus construction can be used (see also Section 24.4.1). This construction is often used when the focus is contrastive. In this construction, a constituent is placed in the pre-core slot and marked with a final extra-high tone. This structure is most frequently used for narrow focus, that is focus on the constituent in the pre-core slot only, as was shown in example (9) in Section 24.4.1. To mark narrow focus on the verb, a copy of the verb root, prefixed with the noun class prefix ki-/ki- is placed in the pre-core slot and _

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

marked with the final extra-high tone, while the inflected verb remains in its normal position. An example is shown in (30), where the focus is on the lexical content of the verb, but the same construction can also be used to mark focus on an aspect, modality or status operator (see also Van Putten 2016). bɛ-tá-h ki-h ́ c4.sg-grind:foc c1.pl.sbj.pfv-int-grind ‘Will they [grind]FOC it (or pound it)?’

l c3.sg c

c

c

(30)

(illness_100616_SO-DS)

The focus construction may also be used for sentence focus or predicate focus (focus on the verb þ argument that is not the PSA or PP). A sentence focus interpretation is possible when the PSA is marked for focus and a predicate focus interpretation is possible when an argument that is not the PSA or a peripheral PP is marked for focus. An example of the latter can be seen in (31). (31)

(One speaker is telling a story in which a boy riding a bicycle meets a girl. The listener asks a clarification question.) ὲέ-kpɛ yɛ tsyɛ, gas ́ c1.sg too bicycle:foc c1.sg.sbj.prog-put (pear_100630_GoD-FB) ‘Is she also [riding a bicycle]FOC ?’ c

For the linking from semantics to syntax, the use of the focus construction means that a syntactic template with a pre-core slot will be chosen (see also Section 24.4.1). If the element to be placed in the pre-core slot is a non-PSA argument, the number of postverbal core slots will be reduced by one. PSA indexing on the verb is obligatory, so the preverbal core slot which contains the verb prefix cross-referencing the nominal PSA argument is always filled, irrespective of whether the PSA occurs in the pre-core slot or in its regular pre-core position.

24.8

Summary

In Part I of this grammatical sketch, we have discussed the grammar of Avatime simple clauses. Avatime is a language with a clear distinction between nouns and verbs, small lexical categories of adjectives, adverbs and adpositions and a large category of ideophones. It has a rather rigid constituent order, in which the PSA precedes the verb and other arguments follow the verb, followed by adverbial phrases. The PSA is indexed on the verb with head-marking. Question words and optionally focused elements occur in the pre-core slot. Referential phrases can have either a noun or a postposition as their nucleus. If a noun heads the RP, it can be followed by a number of modifiers, most of which agree in noun class with the head noun. Possession is indicated by juxtaposition with possessor–possessed word order. When a postposition heads the RP, no modifiers are possible and the postposition is

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

903

904

SASKIA VAN PUTTEN AND REBECCA DEFINA

always preceded by another RP that functions like a possessor. Locative phrases are formed with the locative preposition ní. Most operators of the layered structure of the clause are marked on the verb by the choice of PSA prefix out of several possible paradigms and/or a specialized prefix. The order of operators does not always follow the expected clausal-core-nuclear order. This may be due to the origin of some prefixes as verbs in serial verb constructions and to the change from an unmarked to a marked perfective.

PART II. THE SYNTAX AND SEMANTICS OF SERIAL VERB CONSTRUCTIONS In this part, we discuss the relation between syntax and semantics in serial verb constructions (SVCs). First, we describe the properties of SVCs in Avatime and how they are distinguished into syntactic subtypes (Section 24.9). We then discuss the semantic functions of Avatime SVCs and how they fit within the Interclausal Relations Hierarchy connecting syntactic structures and semantic functions (Section 24.10).

24.9

Serial Verb Constructions

Serial verb constructions are sequences of two or more finite verbs in one clause without overt marking for coordination or subordination. In Avatime, only the first verb is fully inflected. Subsequent verbs are either bare, as in example (32) or marked with a reduced agreement prefix, as in example (33). These reduced prefixes are specific to SVCs (for more information, see Defina 2016b). All verbs share their PSA and may also share another argument, as is the case with example (33). (32)

o-di ŋwὲ c1.sg.sbj.pfv-sit drink ‘S/he sits drinking.’

(33)

a-kpέ a-ŋwya ki-dít c1.sg.sbj.pfv-throw c4.sg-thing:indf svm.c1.sg.pfv-put.in:loc -k` a¼ɛ c1.sg-father¼def ‘She threw something to the father.’ (contrexp-28_120912) c

c

Avatime SVCs can be divided into three subtypes: nuclear, core and sequential, based on a number of morphosyntactic differences. In what follows we briefly describe the properties of the different types of SVC and show how they can be analysed in RRG. For a more elaborate account, see Defina (2016b). The different subtypes can be distinguished by examining their behaviour with operators at different levels. Negation and most aspect and mood

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

markers are fused to some degree with the PSA agreement markers (see Section 24.4.2) and can only be marked on the first verb of an SVC. These markers always have broad scope over the whole SVC. Even the intentive, which is the only morphologically separate mood marker, can only occur with the first verb and scopes over the whole construction. This suggests that the verbs in all Avatime SVCs are linked below the clause level. Aspect and aspectual adverbials are nuclear-level operators and can be used to distinguish a set of nuclear-level SVCs from the rest. Many SVCs allow aspectual adverbials between the verbs with narrow scope over a single verb, as can be seen in example (34). The recurrent aspect can also be used to mark each verb in these SVCs individually, as can be seen in example (35). Neither of these is possible with the nuclear-level SVCs. While aspectual adverbials are occasionally accepted between the verbs of these nuclear SVCs, they always have broad scope over the whole construction and the recurrent aspect can only be marked on the first verb, as can be seen in examples (36) and (37). This indicates that these SVCs are formed through nuclear co-subordination while the other Avatime SVCs are joined at a higher level. (34)

`ı-dru¼l`e ` e-bu agb`el`ı¼ye kóko kí ̣ Kwami c1.sg.pfv-remove cassava¼def c2.pl-mound¼def already give Kwami ‘He already dug cassava mounds for Kwami.’ (But may keep them for himself )

(35)

a.

(36)

ba-dí ̣ koko gu c1.pl.pfv-sit already talk ‘They already sat talking.’

(37)

a. ba-zɛ̌ -dí ̣ ŋwὲ k` u-g` oda c1.pl.pfv- REC -sit drink c6.sg-palmwine ‘They were sitting drinking palmwine.’ b. *ba-dí ̣ zɛ̌ -ŋwὲ k` u-g`oda c1.pl.pfv-sit REC -drink c6.sg-palmwine

` ` m`a-zɛ̌ -dzɛ Ohol` o a-w` a a-xwὲ¼na svm.1.sg.pfv-work c3.pl-job¼def 1.sg.pfv- REC -go Ho ‘I was going (repeatedly) to Ho and working.’ ` ` b. m`a-dzɛ Ohol` o a-zɛ̌ -w` a a-xwὲ¼na 1.sg.pfv-go Ho svm.1.sg.pfv- REC -work c3.pl-job¼def ‘I went to Ho and was working.’ (moved to Ho for some time) ku-nugu¼y` o c5.sg-mouth¼def

Directional prefixes and locational and temporal adverbials are corelevel operators and their scope properties distinguish the final two subtypes. Sequential SVCs can occur with distinct locational and temporal adverbials modifying each verb phrase, as can be seen in example (38). These SVCs also allow the directional prefix to occur on the second verb with narrow scope, as can be seen in example (39). This suggests these SVCs consist of distinct cores with their own peripheries and so we analyse them as core coordinations.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

905

SASKIA VAN PUTTEN AND REBECCA DEFINA

(38)

ma-ts` a tomatoes¼ye ní li-vlɛ¼lὲ _ 1.sg.sbj.pfv-cut tomatoes¼def loc c3.sg-morning¼def ` a-kpɛ ní k`e-zi¼a mὲ ábl` a svm.1.sg.pfv-put loc c6.sg-bowl¼def inside now ‘I cut the tomatoes in the morning and put them in the bowl now.’

(39)

k` a-wɛ¼a ma-k ̀ 1.sg.sbj.pfv-take c6.sg-axe¼def ‘I took the axe to go cut the tree.’ c

906

zɛ-ts` a IT -cut

`-se¼l` o o c2.sg-tree¼def

The other non-nuclear SVCs also allow locational and temporal adverbials between the verbs. However, in these cases there can be only one adverbial and it has broad scope over the whole construction, as in example (40). This suggests these SVCs share a single periphery and are joined via core cosubordination. Nuclear SVCs do not allow locational or temporal adverbials to occur between the verbs, consistent with their nuclear juncture. (40)

` a. m` a-dzɛ Ohol` o kiv` oe 1.sg.sbj-go Ho yesterday ‘I went to Ho for Akosua yesterday.’ ` b. *m` a-dzɛ Ohol` o kiv` oe 1.sg.sbj-go Ho yesterday

kí ̣ give

Akosua Akosua

kí ̣ give

Akosua Akosua

` mon` o o today

The distinguishing properties of these three subtypes of Avatime SVCs are summarized in Table 24.2. Table 24.2 Characteristic properties of Avatime SVC subtypes

Juncture–nexus type Can aspectual adverbials occur Can the recurrent occur Can directionals occur Can locational or temporal adverbials occur

24.10

between verbs? with restricted scope? on subsequent verbs? with restricted scope? on subsequent verbs? with restricted scope? between verbs? with restricted scope?

Nuclear

Core

Sequential

Nuclear cosubordination

Core cosubordination

Core coordination

Marginally No No No Marginally No No No

Yes Yes Yes No Yes No Yes No

Yes Yes Yes Yes Yes Yes Yes Yes

Semantic functions of SVCs and the Interclausal Relations Hierarchy

24.10.1 The Interclausal Relations Hierarchy There are striking tendencies across languages in how semantic functions are expressed by syntactic constructions: closer semantic functions tend to be expressed by syntactic constructions that more closely resemble a single clause and have a stronger link between their elements, while looser

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

semantic relationships tend to be expressed by syntactic constructions with a looser, or weaker, link between their elements. This is known as the Interclausal Relations Hierarchy, which was first proposed by Silverstein (1976), in his study of split case marking in ergative languages, as well as by Givón (1980), in a more direct and detailed proposal. It has since been further extended and solidified within the theory of RRG (Foley and Van Valin 1984; Van Valin 2005). The hierarchy is presented in Figure 24.6. The alignment of semantic functions and syntactic constructions is not one to one. Not all languages make use of all the listed syntactic construction types. Each construction type may express multiple semantic functions and, vice versa, a given semantic function may be expressible through several different constructions. This leads to rather complex instantiations of the interclausal relations hierarchies in individual languages (e.g. Casti 2012; Kockelman 2003) and a full consideration of the Avatime interclausal relations is beyond the current paper. However, Avatime SVCs and related semantic functions present some intriguing cases for this hierarchy, which are worth discussing here. The primary prediction of the Interclausal Relations Hierarchy is that the tightest syntactic construction used to express a semantic function should be tighter than or equal to the tightest syntactic construction used to express a looser semantic function. There are two cases involving SVCs where there is an apparent deviation from this prediction in Avatime. By examining how Avatime SVCs fit within and deviate from the Strongest Nuclear co-subordination Nuclear subordination Daughter Peripheral

Nuclear coordination Core co-subordination Core subordination Daughter Peripheral

Core coordination Clausal co-subordination Clausal subordination Daughter Peripheral

Clausal coordination Sentential subordination Sentential coordination Weakest

Closest Causative [1] Phase Manner Motion Position Means Psych-action Purposive Jussive Causative [2] Direct perception Indirect perception Propositional attitude Cognition Indirect discourse Direct discourse Circumstances Reason Conditional Concessive Simultaneous actions Sequential actions Situation-situation: unspecified Loosest

Figure 24.6 The Interclausal Relations Hierarchy (Van Valin 2005: 209)

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

907

908

SASKIA VAN PUTTEN AND REBECCA DEFINA

predictions of this hierarchy, we can gain a better understanding of the hierarchy itself, Avatime SVCs, and their place within the wider ecology of Avatime grammar.

24.10.2 Semantic Functions of Avatime SVCs In this section, we review the semantic functions that can be expressed using the three types of SVCs discussed in Section 24.9: nuclear cosubordination, core co-subordination and core coordination. We will then consider any alternative means of expressing these semantic functions in Avatime and review the relations between syntactic constructions and semantic functions in terms of the predictions of the Interclausal Relations Hierarchy. The nuclear co-subordination SVCs are used to express modifying subevents such as the position, (example (32)), or manner (example (33)) in which an action is carried out. They are also used to express modification of the path of motion, as in example (41). These three semantic functions cluster neatly together in the Interclausal Relations Hierarchy and together with means form the set of ‘modifying subevents’ functions (Van Valin 2005: 206). Manner can also be expressed using a core SVC, as discussed later. (41)

be-ple e-ku c1.pl.sbj-descend svm-enter ‘They descended into Gbadzeme.’

ní loc

Gb` adzɛmɛ Gbadzeme (Avatime-history_110905_BB_129)

As discussed in Section 24.5, it is also possible to express motion accompanying an action with a directional prefix in a simple clause, as in example (42). b` a-li¼` a a-zɛ-b` asi¼bl _ _ c1.sg.sbj.pfv-it-show¼1.pl.obj c5.pl-palm.tree¼def ‘He went to show us the palm trees.’ (Conv-ablorme_100715_SO-AS) c

(42)

These directional prefix motion functions are a different type of motion semantic function from the one expressed by the nuclear SVCs. The nuclear SVCs further specify a path of motion. The directional prefixes express concomitant motion which is either towards or away from the deictic centre. This latter function is closer to the one described for the motion semantic function in the hierarchy presented by Van Valin (2005: 206). The use of directional prefixes to express motion creates an apparent deviation from the predictions of the Interclausal Relations Hierarchy, since clauses with directional prefixes are closer to simple clauses than the nuclear SVCs used to express the tighter semantic function of manner. This deviation likely arises from the derivation of the directional prefixes from SVCs (see Section 24.4.2). If the directional prefixes were still verbs within nuclear co-subordinate SVCs, these semantic functions would be expressed by equally tight syntactic constructions and there would be no deviation

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

from the hierarchy. It is also notable that there is a semantic split between the functions of the directional prefix and the motion nuclear SVCs where this deviation occurs. It could then be possible to split motion into two functions with accompanying motion positioned as closer than manner, and path of motion remaining where motion is currently (looser than manner). This would resolve the deviation in the hierarchy. The core SVCs are used for adding and introducing or marking arguments. Examples (34) and (40) show the use of this type of SVC to add a beneficiary using the verb kí ‘give’. They can also be used to add instruments, manners or means, as in examples (43)–(45), respectively, all with the verb k ̀ ‘take’. c

(43)

k` a-wɛ¼a y` ai a-k ̀ c1.sg.sbj-take c6.sg-axe¼def break ‘He used an axe to split the tree.’

(44)

a-k ̀ c1.sg.sbj.pfv-take ‘He left in anger.’

(45)

ku-z` o dzi a-k ̀ c1.sg.sbj.pfv-take c5.sg-theft become ‘Through theft he became a rich man.’

c

ku-siyeyome c5.sg-anger

`-se¼l` o o c2.sg-tree¼def

sὲ leave

ohonete rich.person

c c

These SVCs can also be used to mark or introduce theme arguments which are already called for by another verb in the SVC. For instance, in example (46) k ̀ ‘take’ is used to introduce the theme banana. However, as example (47) shows, kpɛ ‘put’ already licenses agent and theme arguments. c

a-k ̀ k ̀ ranti¼ɛ kpɛ ní c1.sg.pfv-take banana¼def put loc ‘S/he put the banana into the basket.’

(47)

a-kpɛ l`ı-k` uto¼l`e c1.sg.pfv-put c3.sg-hat¼def ‘S/he put the hat on.’

k` a-s ¼ya c6.sg-basket¼def c

(46)

mὲ inside

c

c

With the exceptions of manner and means, these functions are not included within the list of interclausal semantic functions. These argument-adding functions are relations between a predicate and its dependents and are typically intraclausal. However, the fact that they can be expressed by a combination of verbs in Avatime suggests they could be added to the hierarchy. They bear clear similarities to the causative [1] function – to directly bring about a state of affairs through an event or action: for example, paint the table red or break the bowl. This function also involves relations between a predicate and its arguments and is sometimes expressed by a single verbal lexeme. Given the overlap with manner and means functions and the general role of these functions, it would make sense for them to be added immediately below means in the hierarchy.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

909

910

SASKIA VAN PUTTEN AND REBECCA DEFINA

The core coordination SVCs are used to express sequential actions, as in examples (35), (38) and (39). These SVCs can always be paraphrased by coordinated clauses, as shown in example (48). (48)

`-kpɛ a. ma-ts` a tomatoes¼ye a ní 1.sg.sbj.pfv-cut tomatoes¼def svm.1.sg.pfv-put loc k`e-zi¼a mὲ c6.sg-bowl¼def inside ‘I cut the tomatoes and put them in the bowl.’ b. ma-ts` a tomatoes¼ye lɛ̌ m` a-kpɛ ní 1.sg.sbj.pfv-cut tomatoes¼def and 1.sg.sbj.pfv-put loc k`e-zi¼a mὲ c6.sg-bowl¼def inside ‘I cut the tomatoes and put them in the bowl.’

This use of core coordination to express sequential action is another instance of a deviation from the predictions of the Interclausal Relations Hierarchy. Sequential actions are among the loosest types of semantic relations. Many tighter semantic relations, such as circumstance and reason, are expressed by the looser syntactic relations of clausal subordination and coordination. For instance, the circumstantial in example (49), where a preposed subordinate clause marked with the clause linkage marker xé indicates the point in time at which the main clause happens. ki-tá-hal`ı ̣ s`ı-wa¼sὲ ̣ p ́ ¼ɛ kù í-ṣ (49) xé _ _ clm 1.pl.sbj.prog-sow finish¼cm 1.pl.sbj.pfv-int-gather c7-weed¼def mὲ petee ní ̀ -ny ¼n ̀ all loc c2.sg-farm¼def inside ‘When we finish sowing, we gather all the weeds from the farm.’ (rice_100613_EN-MM) c

c

c

c

c

It is notable that, like with the previous deviation, a semantic distinction not recognized in the hierarchy plays a role. Whereas all sequential actions can be expressed by clausal coordination, the sequential actions that can be expressed by core coordination SVCs must be performed by the same actor and must constitute a single culturally relevant unit, generally with an overarching goal (see Defina 2016b). If this type of sequential action relation were distinguished as a separate, tighter, semantic relation category in the hierarchy, the deviation would be resolved.

24.11

Summary

This part has surveyed the semantic functions and syntactic properties of SVCs in Avatime, showing how they constitute three distinct juncture– nexus types, see Table 24.3. Each subtype of Avatime SVC poses a question for the Interclausal Relations Hierarchy and its predicted links between syntactic

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

Table 24.3 Semantic functions of SVCs in each subtype Subtype

Semantic functions

Nuclear co-subordination Core co-subordination

Modifying (manner, motion (complex path), and position) Argument adding and theme marking Modifying (manner and means) Sequential actions (same actor, coherent unit)

Core coordination

constructions and semantic functions. The argument-adding and marking functions of the core co-subordination SVCs are not included within the hierarchy, but should arguably be listed below the modifying functions. The grammaticalization of SVCs indicating motion away from or towards the deictic centre has created a deviation from the hierarchy’s predictions, with the tighter directional prefix construction expressing a function that is looser than the manner function expressed by nuclear SVCs. The use of core coordinating SVCs to combine certain types of sequential actions presents another deviation. Both of these deviations relate only to more restricted semantic functions and can be resolved through separating the specific semantic functions expressed with the tighter syntactic construction from the more general motion and sequential action semantic functions. This review of Avatime SVCs within the context of the Interclausal Relations Hierarchy provides a clearer understanding of the internal structure of Avatime SVCs and complex sentences more generally. It also suggests that the list of semantic functions expressed by syntactic linkage is far from final and may need further language-specific or general modifications or additions. This is not an entirely new finding, and the point was made by Van Valin (2005: 211) that the syntactic side of this hierarchy is better understood than the semantic side, which should not be taken as an exhaustive list of semantic functions.

References Ameka, Felix K. 1995. The linguistic construction of space in Ewe. Cognitive Linguistics 6: 139–181. Casti, Francesco. 2012. Testing the Interclausal Relations Hierarchy: Aspectual and Modal Periphrases in Modern Sardinian. PhD dissertation, University of Manchester. Caudal, Patrick and David Nicolas. 2005. Types of degrees and types of event structures. In Claudia Maienborn and Angelika Wöllstein (eds.), Event Arguments: Foundations and Applications, 277–300. Berlin: De Gruyter. Defina, Rebecca. 2009. Aspect and Modality in Avatime. MA thesis, Leiden University.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

911

912

SASKIA VAN PUTTEN AND REBECCA DEFINA

Defina, Rebecca. 2016a. Events in Languages and Thought: The Case of Serial Verb Constructions in Avatime. PhD dissertation, Radboud University, Nijmegen. Defina, Rebecca. 2016b. Serial verb constructions and their subtypes in Avatime. Studies in Language 40(3): 648–680. Defina, Rebecca. 2018. Tense, aspect and mood in Avatime. Afrika und Übersee 92: 65–97. Dingemanse, Mark. 2012. Advances in the cross-linguistic study of ideophones. Language and Linguistics Compass 6(10): 654–672. Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Givón, T. 1980. The binding hierarchy and the typology of complements, Studies in Language 4(3): 333–377. Kockelman, Paul. 2003. The Interclausal Relations Hierarchy in Q’ecqchi’ Maya. International Journal of American Linguistics 69: 25–48. van Putten, Saskia. 2014a. Information Structure in Avatime. PhD dissertation, Radboud University, Nijmegen. van Putten, Saskia. 2014b. Left-dislocation and subordination in Avatime (Kwa). In Rik van Gijn, Jeremy Hammond, Dejan Mati´ c, Saskia van Putten and Ana Vilacy Galucio (eds.), Information Structure and Reference Tracking in Complex Sentences, 71–98. Amsterdam: John Benjamins. van Putten, Saskia. 2016. Discourse functions of focus marking in Avatime. Journal of African Languages and Linguistics 37(1): 91–130. Schuh, Russell G. 1995. Avatime noun classes and concord. Studies in African Linguistics 24(2): 123–149. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon, ed., Grammatical Categories in Australian Languages, 112–171. Canberra: Australian Institute of Aboriginal Studies. Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota clause. In Johanna Nichols and Anthony C. Woodbury (eds.), Grammar Inside and Outside the Clause, 363–413. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantic Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning, and Function. Cambridge: Cambridge University Press. Welmers, William E. 1973. African Language Structures. Berkeley, CA: University of California Press.

Notes 1 All recordings are archived at The Language Archive at the MPI for Psycholinguistics in Nijmegen (https://corpus1.mpi.nl/ds/asv/?1). Examples in this chapter that are taken from this corpus are followed by the filename of the recording.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

A Grammatical Sketch of Avatime

2 This prefix is different from the habitual prefix, despite being homophonous with it. See Defina (2018) for more details. 3 We could call this inalienable possession, but unlike most languages that have an inalienable vs. alienable distinction, this class consists only of kinship terms and does not include body parts.

https://doi.org/10.1017/9781316418086.030 Published online by Cambridge University Press

913

25 A Grammatical Sketch of Amele (Papuan, Papua New Guinea) John R. Roberts Abbreviations We use the Leipzig abbreviations (www.eva.mpg.de/lingua/resources/gloss ing-rules.php), with the following additions: A ARG ASP CLM CONT CORE R DCA DEIC DN DS DUn DV EVQ HORT HABP IF IU INGR IRIT IT IVC LS

actor macrorole argument aspect clause linkage marker continuative aspect RP core direct core argument deictic DCA-non-macrorole different subject DCA-undergoer dependent verb event quantification hortatory habitual past illocutionary force operator information unit ingressive irregular iterative aspect iterative aspect impersonal verb construction logical structure

PrDP PRED PRO PRSP PSA PSD PSR QNT R(EAL) RG REMP RP RPIP RRG SEQ SIM

pre-detached position predicate pronoun prospective tense privileged syntactic argument possessed possessor quantifier realis status regret particle remote past tense reference phrase RP initial position Role and Reference Grammar sequential event simultaneous event

SR SS STA

switch-reference same subject status operator

SVC TNS

serial verb construction tense operator

TODP

today’s past tense

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

MP NEGF NMR NUC NUC R NUM PNG PP PoCS PoDP

modifier phrase negative future tense non-macrorole clause nucleus RP nucleus numeral Papua New Guinea postpositional phrase post-core slot post-detached position

~ & {&} ^ {^}

PrCS

pre-core slot

*

25.1

U YESTP

¼

undergoer macrorole yesterday’s past tense morpheme break infix clitic break reduplication sequential conjunction seq conj in complex sentence overlapping conjunction overlap conj in complex sentence ungrammatical

Introduction

Amele is a Papuan language spoken in Madang Province, Papua New Guinea (PNG). The Amele people inhabit an area of approximately 120 square kilometres between the Gum and Gogol rivers just south of the town of Madang in PNG. The area extends from the coast to about 14 kilometres inland. Amele is the largest of the Gum family of languages (Z’graggen 1975: 13) with a population of approximately 5,300 speakers (Lewis et al. 2014).1 This chapter provides a linguistic description of the Amele language from a Role and Reference Grammar (RRG) perspective. Section 25.2 describes the basic syntax of the language and Section 25.3 examines serial verb constructions and the switch-reference system.*

25.2

Common Topics

Nominative-accusative agreement is suffixed to the verb stem and up to four core arguments can be marked on the verb. There are only two major lexical categories, nouns and verbs, with very little overlap between them. Alternative undergoer selection may be made for ditransitive verbs. There is no passive construction in the language and the only choice for privileged syntactic argument (PSA) is [S, AT]. Focus may be expressed morphologically and by incorporation of modifier elements into the verb word.

25.2.1 Basic Clause Patterns Typologically, Amele has head-last syntax with OV order and postpositions.2 The language is also head-marking, and core arguments of the verb can be * Robert Van Valin provided helpful comments and suggestions in the compilation of this chapter. However, any errors in analysis or content are the responsibility of the author.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

915

916

JOHN R. ROBERTS

PSA (RP) agent experiencer

Adjunct (RP/PP) temporal location

Arg-Adjunct (RP/PP) goal path source instrument benefactive accompaniment …

DCA (RP) patient recipient perceived perceiver addressee possessor …

VERB

Figure 25.1 Basic syntax of the clause in Amele

marked on the verb by cross-reference agreement. PSA agreement has one basic type of morphology and direct core argument (DCA) agreement has another basic type of morphology.3 The argument agreement morphology follows a nominative-accusative pattern. Therefore, the PSA argument agreement morphology is called nominative (nom) and the DCA agreement morphology is called accusative (acc). The basic syntax of the clause is given in Figure 25.1. The PSA is the first element in the clause and the DCA occurs immediately preceding the verb. After the PSA is the slot for any adjunct reference phrase (RP) or postpositional phrase (PP). Following that is the slot where any argument-adjuncts occur. Typically, no more than two or three non-verbal elements are expressed in any one clause. However, up to four arguments can be encoded on the verb. There is also an order in which the different types of argument can be marked on the verb. Up to three arguments can be marked on the verb and the order is given in (1). The rightmost argument is the PSA agreement and this is obligatory on the finite verb. This is the nominative agreement morphology. The DCA-undergoer (DUn, where Un is undergoer) argument agreement attaches directly to the verb stem. This is obligatory for some verbs, optional for other verbs, and not allowed for others. Many verbs also allow optional DCA-non-macrorole (DN) argument agreement to be marked and this requires the applicative (applied object) marker. The form of the DUn and DN marking is the accusative agreement morphology in each case. It is possible to have a maximum of two DCA arguments marked on the verb, either DUn þ DN or DN þ DN. The linear order of argument marking on the verb in (1) is thus a mirror image of the ordering of arguments in the clause in Figure 25.1. (1)

Order of arguments marked on the verb: verb stem DUn.Agr APPLþDN.Agr (APPLþDN.Agr) þPSA.Agr

Verb agreement is formalized as in (2). nom agreement only applies to the finite verb form. acc agreement can apply to both the finite and infinitive verb forms, for example hel-ad-ec [throw-3pl.acc-inf] ‘to throw them’, hel-i-ad-ec [throw-appl-3pl.acc-inf] ‘to throw to them’. See Roberts (1996, 1997c).

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele (2)

917

Verb agreement in Amele: nom agreement cross-references the highest-ranking macrorole argument in the finite verb. acc agreement can cross-reference any other macrorole or non-macrorole argument in the finite or non-finite verb.4

25.2.1.1 M-Intransitive Verbs M-intransitive verbs have only one macrorole core argument. The main types of M-intransitive verbs are stative verbs and motion verbs. There is no copular verb like ‘be’ in Amele. Instead, the language uses several posture verbs in certain stative predications (i.e. bilec ‘to sit (down)’, nijec ‘to lie (down)’, tawec ‘to stand (up)’).5 As illustrated in Table 25.1, these posture verbs have a state function and an activity function. For the state function the single macrorole is undergoer. For the activity function the single macrorole is actor. An example of each function is given in (3). (3)

Posture verb examples: a. Jo jobon gemo taw-ena. house village middle stand-3sg.nom.prs be-village middle′ (stand′ (3sg [jo])) ‘The house stands in the middle of the village.’ b. Dana ben taw-im-ei ma-g-en. man big stand-ss.seq-3sg.nom tell-1pl.acc-3sg.nom.remp do′ (3sg [dana], [stand′ (3sg [dana])]) . . . ‘The headman stood up and spoke to us.’ c. Cam qila gagadic¼ca taw-ei-a. sun today strength¼add stand-3sg.nom-todp today′ (be′ (3sg [cam], [strong′])) ‘The sun is strong today.’

State: posture

Activity: assume posture

State: attributive

There are two basic types of M-intransitive motion verb: those that specify [moveþpath] and those that specify [moveþmanner]. [moveþpath] motion verbs are verb-framed and [moveþmanner] motion verbs are satellite-framed (Talmy 2007). Each type of motion verb is M-intransitive and the single argument is actor in each case. However, there is a syntactic difference between these types of motion verb. The [moveþpath] motion Table 25.1 M-intransitive posture verbs bilec

nijec

tawec

‘sit’ ‘sit (down)’ ‘be’ ‘lie’ ‘lie (down)’ ‘be’ ‘stand’ ‘stand (up)’ ‘be’

sit′ (x) do′ (x [sit′ (x)]) be′ (x) lie′ (x) do′ (x [lie′ (x)]) be′ (x) stand′ (x) do′ (x [stand′ (x)]) be′ (x)

state: posture activity: assume posture state: attributive, identificational, specificational state: posture activity: assume posture state: attributive, identificational, specificational state: posture activity: assume posture state: attributive, identificational, specificational

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

918

JOHN R. ROBERTS

verb allows an allative argument to be expressed either with a ¼ca ‘towards’ PP, as in (4a), or with an applied object marked on the verb, as in (4b). The applied object in (4b) is a non-macrorole direct core argument. The [moveþmanner] motion verb, on the other hand, does not allow an allative argument to be expressed, as illustrated in (5). (4)

Allative argument applicative alternation with [moveþpath] motion verbs: a. Qa uqa¼ca b-ei-a. dog 3sg¼towards come up-3sg.nom-todp [do′ (3sg [qa], [move.upwards.towards.ref.point′ (3sg [qa])]) ^ proc cover.path.distance′ (3sg [qa])] & ingr be-toward′ (3sg, 3sg [qa]) ‘The dog came up to him.’ b. Qa b-i-t-oi-a. dog come up-appl-3sg.acc-3sg.nom-todp [do′ (3sg [qa], [move.upwards.towards.ref.point′ (3sg [qa])]) ^ proc cover.path.distance′ (3sg [qa])] & ingr be-toward′ (3sg, 3sg [qa]) ‘The dog came up to him.’

(5)

Allative argument disallowed with [moveþmanner] motion verbs: a. *Qa uqa¼ca cob-oi-a. dog 3sg¼towards walk-3sg.nom-todp b. *Qa cob-i-t-oi-a. dog walk-appl-3sg.acc-3sg.nom-todp

Other verbs that are M-intransitive include ededec ‘gleam’ do′ (x, [gleam′ (x)]), asalec ‘laugh’ do′ (x, [laugh′ (x)]), busuec ‘fart’ seml fart′ (x), silolec ‘ooze’ do′ (x, [ooze′ (x)]), fojec ‘vomit’ do′ (x, [vomit′ (x)]), tatiec ‘look up’ do′ (x, [look.up′ (x)]), bodoec ‘soften’ become soft′ (x), sanan mec ‘start’ ingr start′ (x). Impersonal verbs also have to be treated as M-intransitive. A typical example of an impersonal verb construction (IVC) is given in (6). There is an optional free pronoun ija ‘1sg’ which corresponds to the subject pronoun in the English translation. However, in the Amele form this pronoun is crossreferenced on the verb with acc morphology. The nom agreement always codes 3sg in an IVC but the reference is unspecified. The agreement is dummy, neutral agreement which is part of the syntactic template for the construction and there is no actor argument with these IVCs. Such IVCs normally express a physiological or psychological experience and the 1sg. acc agreement refers to the experiencer argument. (6)

Impersonal verb construction: (Ija) cucui-t-ei-a. (1sg) fear-1sg.acc-3sg.nom-todp feel′ (1sg, [afraid′]) ‘I am afraid.’

25.2.1.2 M-Transitive Verbs M-transitive verbs have two macrorole core arguments: actor and undergoer. Amele has M-transitive state verbs, such as fec ‘see’ see′ (x, (y)), doc ‘know’

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

know′ (x, (y)), gawec ‘want’ want′ (x, y), meleec ‘believe’ believe′ (x, (y)), cucuiec ‘fear’ fear′ (x, (y)). With some M-transitive non-state verbs, expression of the undergoer argument on the verb is optional, with others this is obligatory. (7) illustrates an active verb, j-ec [eat-inf] ‘to eat’, with optional DUn marking on the verb. In (7a) this verb has the logical structure (LS) do′ (x, [eat′ (x, (y))]) with an optional second predicate argument. In (7b) this verb has the LS do′ (dana, [eat′ (dana, ha)]) with the second predicate argument position filled. This argument is realized in the syntax by the RP ha ‘sugarcane’. Since ha is neither animate nor a count noun, in this context it does not trigger acc agreement on the verb. The predicate is, therefore, an activity with a nonspecific object. In (7c), on the other hand, ho ‘pigs’ is animate and a count noun and therefore triggers acc agreement -ad ‘them’ on the verbs qoc ‘hit’ and jec ‘eat’.6 In this case the object is also specific and ‘eat’ is an active accomplishment. (7)

Active verb with optional DUn marking: a. Dana eu j-egi-na. man that eat-3pl.nom-prs do′ (3pl [dana], [eat′ (3pl [dana], (Ø))]) ‘Those men are eating.’ b. Dana eu ha j-ein. man that sugarcane eat-3pl.nom.remp do′ (3pl [dana], [eat′ (3pl [dana], ha)]) ‘Those men ate sugarcane.’ c. Dana eu age ho a-q-i je-ad-ein. man that 3pl pig 3pl.acc-hit-dv eat-3pl.acc-3pl.nom.remp [do′ (3pl [dana], [eat′ (3pl [dana], 3pl [ho])]) ^ proc consume′ (3pl [dana], 3pl [ho])] & [ingr consumed′ (3pl [ho])] ‘Those men killed the pigs and ate them.’

The M-transitive verb cesul-d-oc [help-3sg.acc-inf] ‘to help him/her’ in (8) has obligatory DUn marking. In (8) the DUn marking -t ‘1sg.acc’ on the verb expresses the obligatory second argument. The 3pl.nom argument is actor and the 1sg.acc argument is undergoer. (8)

Active verb with obligatory DUn marking: Age cesul-t-eig-a. 3pl help-1sg.acc-3pl.nom-todp do′ (3pl, [help′ (3pl, 1sg]) ‘They helped me.’

25.2.1.3 Three-Argument Verbs Three-argument verbs have three core arguments in their LS. The ditransitive verb ihac-d-oc [show-3sg.acc-inf] ‘to show him/her/it’ is illustrated in (9). The LS of this verb is [do′ (x, Ø)] cause [become see′ (y, z)]. In (9a) the perceiver argument 1sg is selected as undergoer and is realized on the verb as 1sg. acc agreement attached directly to the verb stem. Because the perceiver argument has been selected as undergoer, there is no way of encoding the

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

919

920

JOHN R. ROBERTS

perceived argument, ho eu ‘that pig’, on the verb. In (9b) ho eu ‘those pigs’ is selected as undergoer and is realized on the verb as 3pl.acc agreement. In this case, the 1sg.acc perceiver is also marked on the verb as a DN applied object argument. (9c) shows that all three core arguments can be expressed as agreement marking on the verb alone. Without any clarifying RPs in the clause, (9d) is ambiguous between the first LS where 1sg is the perceiver argument and the second LS where 1sg is the perceived argument – although, the first interpretation would be deemed the more likely. (9)

3-argument verb: a. (Uqa) ho eu ihac-t-ei-a. (3sg) pig that show-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, ho)] ‘He showed me that pig.’ b. Ho eu ihac-ad-i-t-ei-a. pig that show-3pl.acc-appl-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, 3pl [ho])] ‘He showed those pigs to me.’ c. Ihac-ad-i-t-ei-a. show-3pl.acc-appl-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, 3pl)] ‘He showed them to me.’ d. Ihac-t-ei-a. show-1sg.acc-3sg.nom-todp [do′ (3sg, Ø)] cause [become see′ (1sg, Ø)] or [do′ (3sg, Ø)] cause [become see′ (Ø, 1sg)] ‘He showed me something (unspecified).’ or ‘He showed me to someone (unspecified).’

25.2.1.4 A Four-Argument Verb There is one verb which allows an additional fourth core argument to be marked. This is ‘give’, as illustrated in (10).7 Here the verb stem is realized as acc morphology which agrees with the recipient. The meaning of ‘give’ is constructional rather than lexical, and so there would be a constructional schema for it. Furthermore, Latrouite and Van Valin (2014) argue that constructional meaning should be represented differently from lexical meaning; the constructional meaning for ‘give’ would be [do′ (x, Ø)] CAUSE [INGR have′ (y, z)]. There is an infinitive form for each person and number (e.g. itec ‘to give me’, ihec ‘to give you (sg)’, utec ‘to give him/her’, etc.) The regular verb agreement morphology attaches to this derived stem. The first argument (recipient) of have′ (x, y) in ‘give’ logical structure is assigned the undergoer macrorole by virtue of being the argument encoded in the verb stem itself. Thus, the recipient argument is always the undergoer and there is no alternative construction where the second argument (theme) of have′ (x, y) is the undergoer. The ‘^ feel.negatively.affected′ (z)’ represents the malefactive ‘on me’ argument. It is the LS proposed for the Japanese adversative passive (Imai 1998; Toratani 2002).

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele (10)

The verb ‘give’: Eeta¼nu ut-ad-i-t-ag-a? what¼for 3sg.acc-3pl.acc-appl-1sg.acc-2sg.nom-todp be-for′ (eeta, [do′ (2sg, Ø)] cause [INGR have′ (3sg, 3pl)] ^ feel.negatively.affected′ (1sg)) ‘Why did you give him them on me?’

25.2.1.5 Non-Verbal Predicates Amele does not have non-verbal predicates as such. With descriptive and equational clauses a posture verb is used where it is necessary to specify a clausal operator category. However, if such specifications are not required then the posture verb is omitted. Examples are given in (11). (11)

Stative non-verbal predicates: a. Mel eu hag¼ca. Attributive boy that sickness¼with ‘That boy is sick.’ a′. Mel eu hag¼ca nij-en. boy that sickness¼with lie-3sg.nom.remp ‘That boy was sick.’ be′ (mel, [sick′]) b. Misag uqa iwal-ad-ec. Identificational Misag 3sg teach-3pl.acc-nmlz ‘Misag is a teacher.’ b′. Misag uqa iwal-ad-ec bil-ol-oi. Misag 3sg teach-3pl.acc-nmlz sit-habp-3sg.nom ‘Misag used to be a teacher.’ be′ (Misag, [a teacher′]) c. Danben age wool-ad-ec. Specificational Danben 3pl surpass-3pl.acc-nmlz ‘Danben (village) are the winners.’ c′. Danben age wool-ad-ec nij-ein. Danben 3pl surpass-3pl.acc-nmlz lie-3pl.nom.remp ‘Danben (village) were the winners.’ be′ (Danben, [wooladec]) d. Mei ija¼na cof-t-ec. Equational father.1sg.psr 1sg¼of supervise-1sg.acc-nmlz ‘My father is my boss.’ d′. Mei ija¼na cof-t-ec bil-en. father.1sg.psr 1sg¼of supervise-1sg.acc-nmlz sit-3sg.nom.remp ‘My father was my boss.’ equate′ ([have.as.orientation.kin′ (1sg, mei-)], coftec)

25.2.2 Lexical Categories Whereas English has four major word categories of verb, noun, adjective and adverb, which can be distinguished on morphological and syntactic grounds, Amele only has two: verbs and nouns. Words that function as

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

921

922

JOHN R. ROBERTS

nominal modifiers (adjectives in English) and verbal modifiers (adverbs in English) cannot be distinguished on morphological or syntactic grounds in Amele from words that function as the head of an RP (see Roberts 1987: 154–156, 158 for details). Therefore they are all categorized as ‘nouns’. In English, many words can belong to multiple lexical categories, such as break (verb and noun), red (noun and adjective), fast (adjective and adverb), round (adjective, adverb, noun, verb, preposition). By contrast, in Amele there is very little overlap between members of the verb class and those of the noun class. Only three instances of overlap have been observed: cad ‘enemy’ (noun) and cadec ‘to fight’ (verb), mele ‘truth’ (noun) and meleec ‘to believe’ (verb), and cucuiec ‘to fear’ (verb) and cucuian ‘his/her fear’ (inalienably possessed noun).

25.2.2.1 Verbs There are six morphosyntactic forms of the verb: regular verb, impersonal verb, reciprocal verb, light verb, serial verb, and dependent switch-reference verb. Serial verbs are described in Section 25.3.1. They are not marked for tense or for nom agreement. Dependent switch-reference verbs are described in Section 25.3.2. They are not marked for tense either. Amele also has an interrogative verb adec ‘to when’, as illustrated in (12). (12)

Interrogative verb: A~ad-eb dur ~when-3sg.nom.ds.sim.irr ‘Whenever will he come?’

uqa 3sg

h-ugi-an? come-3sg.nom-fut

25.2.2.2 Nouns Amele has three morphological classes of nouns: regular nouns with uninflected stems, inalienably possessed nouns with possessor agreement inflection, and deverbal nouns, derived from the infinitive form of a verb. Inalienably possessed nouns are described in Roberts (1987: 171–175, 2015b). They comprise kin terms,8 body-part terms and personal attribute terms. Deverbal nouns are formed from the infinitive form of the verb. The infinitive suffix, -ec/-oc, on the verb functions as a nominalizing suffix on the noun (e.g. cob-oc [walk-inf] ‘to walk’ or [walk-nmlz] ‘a walk’). 25.2.2.3 Postpositions Postpositions are an important minor word class in Amele (Roberts 1987: 160–161). They are clitic words which must attach to a preceding host element and they may not be stranded (Roberts 1991c, 1992, 1996). They can be predicative or non-predicative, and we return to this in Section 25.2.4.2.

25.2.3 The Layered Structure of the Clause In this section syntactic templates and operators are described and illustrated.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

SENTENCE CLAUSE (PERIPHERY) CORE (RP)

(RP/PP) (RP/PP)

(RP)

(RP)

NUC

(ARG)

(ARG)

ARG

DUn

DN

PSA

PRED V Figure 25.2 The layered structure of the clause

SENTENCE PrDP CLAUSE

SENTENCE CLAUSE PoDP

PrDP template

PoDP template

CLAUSE

CLAUSE

CORE

CORE

PrCS template

PoCS template

PrCS

PoCS

Figure 25.3 Optional syntactic structures

25.2.3.1 Syntactic Inventory The basic structure of the clause in Amele is given in Figure 25.2 and Amelespecific clause-internal linear precedence rules are given in (13). Amele has pro-drop verb agreement which functions as the argument (ARG) of the predicate. The only obligatory element in the active clause is the verb. (13)

Amele-specific clause-internal linear precedence rules: a. XP* > CORE (verb final) b. RP (PSA) > RP/PP (Adjunct) > RP/PP (Arg-Adjunct) > RP (DCA) > NUC

Templates for optional syntactic structures are given in Figure 25.3. The pre-detached position (PrDP) is for topical established information and the post-detached position (PoDP) is for additional information, such as clarification. The pre-core slot (PrCS) is for focal new information or ‘heavy’ constructions, such as a nominal modified by a relative clause. The postcore slot (PoCS) is for postposed clausal elements.

PrDP example Example (14) illustrates a clause-external topic caja eu ‘that woman’ in the pre-detached position. There is a resumptive pronoun uqa ‘she’ in the clause for the argument in the PrDP.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

923

924

JOHN R. ROBERTS

(14)

Clause-external topic in PrDP: Caja eu, uqa me qee. woman that 3sg good not ‘That woman, she is no good.’

PoDP example In (15) the RP dana eu ‘that man’ in the PoDP adds clarifying information to the identity of the PSA eu ‘that’ in the clause. This pronoun is resumptive for the argument in the PoDP. (15)

Clarifying information in PoDP: Eu uqa jeje-g¼ca that 3sg voice-3sg.psr¼with ‘He got his voice back, that man.’

m-en, put-3sg.nom.remp

dana man

eu. eu

PrCS examples In the unmarked form, temporal adjuncts (RP or PP) occur after the PSA RP, as in (16a). Alternatively, the temporal adjunct can be placed in the PrCS as focal new information, as in (16b). (16)

Temporal adjunct in PrCS: a. Ija cum ceta gug filfil cabi¼na ceh-ig-an. 1sg yesterday yam kind different garden¼in plant-1sg.nom-yestp ‘I planted different types of yam in the garden yesterday.’ b. Cum ija ceta gug filfil cabi¼na ceh-ig-an. yesterday 1sg yam kind different garden¼in plant-1sg.nom-yestp ‘Yesterday I planted different types of yam in the garden.’

Arguments and adjuncts with a relative clause are typically placed in the PrCS, as shown in (17). (17) Locative adjunct with a relative clause in the PrCS: Cudun dan ben taw-ena eu¼na place fig big stand-3sg.nom.prs that¼at ‘I hid it at the place where the big fig tree is.’

ija 1sg

jahun-d-ug-a. hide-3sg.acc-1sg.nom-todp

PoCS example In (18) there is an instrumental argument-adjunct PP dubin¼na ‘with stalk’ and a goal argument-adjunct PP camac ta¼na ‘into the sago scrapings’ in the final clause. The second argument-adjunct PP is located in the PoCS. (18) Goal argument-adjunct PP in the PoCS: Age wa wet-i dubin¼na basec-d-ogi-na camac ta¼na. 3pl water scoop-dv stalk¼with pour-3sg.acc-3pl.nom-prs sago scrapings¼in ‘They scoop up the water and pour it in with the stalk into the sago scrapings.’

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

25.2.3.2 Operators Illocutionary force Illocutionary force (IF) is the outermost operator and the form of expression of different types of IF in Amele is illustrated in (19)–(23). Statement (19) is expressed by a declarative sentence. Command (20) is expressed by the imperative form of the verb. The imperative is identical in form to the today’s past tense form. A yes/no question (21) is expressed with the sentence-final question particle ¼fo. An information question (22) is expressed by an interrogative word, such as in ‘who’, eeta ‘what’, cel ‘which’. Exhortation (23) is expressed by the hortative form of the verb. (19)

IF: statement Age aluh¼na bel-ein.9 3pl mountain¼to go.nsg-3pl.nom.remp ‘They went to the mountain.’

(20)

IF: command Age aluh¼na bel-eig-a! 2pl mountain¼to go.nsg-2pl.nom-imp ‘Go to the mountain!’

(21)

IF: yes/no question Age aluh¼na bel-ein¼fo? 3pl mountain¼to go.nsg-3pl.nom.remp¼q ‘Did they go to the mountain?’

(22)

IF: information question In aluh¼na nu-i-an? who.sg mountain¼to go-3sg.nom-yestp ‘Who went to the mountain (yesterday)?’

(23)

IF: exhortation Ege aluh¼na bel-ec¼nu! 1pl mountain¼to go.nsg-inf¼hort ‘Let us go to the mountain!’

Evidentials Amele does not have evidentials. Status Amele expresses realis/irrealis status on the switch-reference verb (see Roberts 1990, 1994 for a wider study of this phenomenon in Papuan languages). The simultaneous ds nom agreement has different forms depending on the realis/irrealis status of the clausal operator category marked on the final verb in the linked clauses, traditionally called a clause chain in Papuan linguistics. The clausal operator categories with realis status are: present tense, today’s past tense, yesterday’s past tense, remote past tense, habitual past tense/aspect, negative past tense. The clausal

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

925

926

JOHN R. ROBERTS

operator categories with irrealis status are: future tense, prospective (about to) tense, negative future tense, and imperative, prohibitive, hortative, optative, counterfactual, apprehensive IF categories. An example of realis status is given in (24a) and an example of irrealis status is given in (24b). (24)

a. Marking of realis status: Ho bu~busal-en pig dur ~run out-3sg.nom.ds.sim.r ‘They killed the pig as it ran out.’ b. Marking of irrealis status: Ho bu~busal-eb pig dur ~run out-3sg.nom.ds.sim.irr ‘They will kill the pig as it runs out.’

age q-oig-a. 3pl hit-3pl.nom.todp

age q-oqag-an. 3pl hit-3pl.nom-fut

Tense The tense categories in Amele are metrical. Metrical tenses mark degrees of temporal remoteness from the deictic centre. Amele has a present tense (3a) and three degrees of past tense: today’s past tense (3c), yesterday’s past tense (22), and remote past tense (3b), as well as a negative past tense and a past habitual tense/aspect (11b′).10 The language has a regular future tense (25), a negative future tense (38), and a prospective (about to/intentional) (26) tense.11 The event referred to in the prospective tense is nearer to the deictic centre than the event referred to in the regular future tense. It is therefore metrical. (25)

(26)

Future tense: Age cabi¼na 3pl going¼to ‘They will go.’

bel-oqag-an. go.nsg-3pl.nom-fut

Prospective tense: Age cabi¼na bel-oqag-a 3pl going¼to go.nsg-3pl.nom-prsp ‘They stood about to go.’

bil-i aux-dv

taw-eig-a. stand-3pl.nom-todp

Modality Categories of deontic modality, such as ability, permission and obligation are expressed lexically in Amele (Roberts 1987, 2001). Event quantification Event quantification (Roberts 2015a) is marked on the verb by distributive inflection that is homonymous12 with the acc agreement, -ad ‘plural’ and -al ‘dual’. Some examples are given in (27). In (27a) the motion verb belec ‘to go’ is intransitive and the -ad ‘plural’ marker indicates a multiple event of each woman going her own way. In (27b) and (c) calec ‘to arrive’ is intransitive. The -ad ‘plural’ marker in (27b) indicates multiple events of men arriving, while in (27c) the -al ‘dual’ marker indicates two events of men arriving.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele (27)

Plural and dual event quantification: a. Caja age bud-u bel-ad-ein. woman 3pl disperse-dv go.nsg-distr.pl-3pl.nom.remp ‘The women dispersed in all directions / each to her own place.’ b. Dana age cal-ad-ein. man 3pl arrive-distr.pl -3pl.nom.remp ‘The men all arrived.’ c. Dana ale cal-al-esin. man 3du arrive-distr.du-3du.nom.remp ‘The men both arrived.’

The EVQ marker -ad can also indicate an exclusive action, as in (28). (28)

Ija saen cecelac sum-i-h-ig-a qee¼nu 1sg time long wait-appl-2sg.acc-1sg.nom-todp not¼for ija cu~cul-h-i l-im-ig nu-ad-ig-a¼da. 1sg dur~leave-2sg.acc-dv go-ss.seq-1sg.nom go-excl-1sg.nom-todp¼rg ‘I waited for you for a long time but in vain. So regretfully I left and went without you.’

Aspect There are three types of aspect that can be expressed by reduplicating some part of the verb word: durative aspect, regular and irregular iterative aspect. Durative aspect is marked on the simultaneous ss/ds verb and conveys the idea that there is an extended temporal overlap of events. Without this marking the temporal overlap is punctiliar. Durative aspect is expressed by different kinds of CV~ or V~ reduplication of the verb stem (Roberts 1991a). See (24a, b), (28), for examples of CV~ reduplication and (12) for an example of V~ reduplication. With some verbs, the reduplicative (C)V~ marking of durative aspect applies to the nom agreement suffixation rather than to the verb stem, as illustrated by (29). If there is accusative marking on the verb then the reduplicative marking for durative aspect applies there, as in (30). (29)

Durative aspect marked on nom agreement: Co-Ø a-e~en lips-3sg.psr open-dur ~ 3sg.nom.ds.sim.r ija dunuh meci-d-ug-a. 1sg inside observe-3sg.acc-1sg.nom-todp ‘While he opened his mouth I looked inside.’

(30)

Durative aspect marked on acc agreement: Age eu cunug ihac-te~t-eig 3pl that all show-dur ~ 1sg.acc-3pl.nom.ss.sim sa-t-ein. explain-1sg.acc-3pl.nom-remp ‘As they showed me everything they explained it to me (lit. explained me).’

The meaning of the regular iterative is a repeated, regular action. This aspect is expressed by rightward reduplication of the whole stem if the verb

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

927

928

JOHN R. ROBERTS

does not have an acc marker, otherwise the acc marker is reduplicated either in place of, or in addition to the reduplication of the verb stem (Roberts 1987: 252–256). For some minimal stem verbs, such as l-ec ‘to go’, the reduplicated stem is of the serial verb form (e.g. li~li-ec ‘to go repeatedly’; see Section 25.3.2). An example of regular iterative aspect is given in (31a). (31)

a. Regular iterative: Gow-ec eu fale~fale-ei-a. light-nmlz that flash~it-3sg.nom-todp 〈IF DECL 〈TNS TODP 〈ASP IT seml flash′ (gowec) 〉〉〉 ‘That light flashed repeatedly.’ b. Irregular iterative: Gow-ec eu fale~fule-ei-a. light-nmlz that flash~irit-3sg.nom-todp 〈IF DECL 〈TNS TODP 〈ASP IRIT seml flash′ (gowec) 〉〉〉 ‘That light flashed intermittently.’

The meaning of the irregular iterative is a repeated action that is irregular in some way (i.e. haphazard, spasmodic, intermittent, etc.). This form involves reduplication of the verb stem but with a disharmonic vowel change in the reduplicated formant. There are eight types of disharmonic vowel change possibilities, which are determined by phonological factors.13 An example of irregular iterative is given in (31b). The aspectual notions of continuative and completive can be expressed periphrastically with a serial verb construction (see examples (45) and (46) in section 25.3.1).

25.2.4

The Structure of RPs and PPs

25.2.4.1 RP Structure The various structures of the RP are given in (32). The head noun in the RP can be a regular noun, an inalienably possessed noun or a deverbal noun. The modifier can be a noun, such as dana caub [man white] ‘white man’, dana me [man good] ‘good man’, or a modifier (a small class of words that only function as modifiers), such as dana bahic [man very] ‘real man’, dana qee [man not] ‘not a man’, or a phrase, such as dana mel iwal-ad-ec [man child teach-3pl.acc-nmlz] ‘school teacher’. (32)

Reference phrase structures: a. noun  modifier  (numeric) quantifier  deictic/indefinite article  universal quantifier b. interrogative deictic þ noun c. PP/RP þ noun

A mass noun can be modified by a general quantifier, such as leih ‘some’ or geh ‘much’. A count noun can be modified by a numeric quantifier, such as lecis ‘two’ or cijed ‘three’. Amele has a pental counting system and the cardinal numbers are: osahic/osol ‘one’, lecis ‘two’, cijed ‘three’ wal oso ‘four’,

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

ebum oso ‘one hand (10)’. Ordinal numbers can be formed by adding the nominalizing suffix -doc to the numeral (e.g. osahic-doc ‘first’, lecis-doc ‘second’, cijed-doc ‘third’). There are also a number of relationship terms that can function as ordinal numbers, for instance matu ‘firstborn/first’, milum ‘secondborn/second’, subig ‘lastborn/last’. The quantifier can be followed by a deictic element, such as eu ‘that’ or ceheleg ‘up there’, or the indefinite article oso ‘a/one’. The article can co-occur with a deictic element, as in dana eu oso [man that art] ‘one of those men’. The universal quantifier cunug ‘all’ occurs at the end of the RP, as in jo nag cijed eu cunug [house small three that all] ‘all those three small houses’. With respect to (32a), the head noun can be questioned with an interrogative, such as eeta ‘what’ or in ‘who’, for example Hina ija-in in? [2sg name-2sg. psr who.sg] ‘What (lit. who) is your name?’ The numeric quantifier can be questioned with an interrogative, such as ganic ‘how much/many’, for example jo ganic? [house how many] ‘how many houses?’, saab ganic? [food how much] ‘how much food?’. However, when the deictic is questioned, the interrogative form, such as cel ‘which’, is placed in the RPIP, which is the precore slot in the RP, as in (32b). Thus the interrogative form is cel dana [which man] ‘which man?’. With respect to (32c), the preceding PP/RP expresses the possessor of the noun or a specification of the noun. (33) illustrates the difference in how alienable and inalienable possession is expressed.14 Alienable possession in (33a) is expressed with a possessive PP preceding the possessed noun. The possessive PP functions as an argument of the possessed noun, which is the head of the nucleus of the possessive noun phrase. The possessive PP is therefore non-predicative. Inalienable possession in (33b) is expressed by suffixal agreement in person (first, second or third) and number (singular, dual or plural) with the possessor noun phrase. The possessor need not be expressed by an overt RP or pronoun. Thus the possessor agreement functions as the argument (ARG) of the inalienably possessed head noun.15 Semantically, inalienably possessed nouns are kinship terms, body-part terms or personal attributes. In (33b) cuduni means ‘my personal place’ or ‘the place that belongs to me in some way’, whereas in (33a) ijana cudun simply means ‘my place’ without the connotations of personal ownership. (33)

a. Alienable possession: ija¼na cudun ‘my place’ 1sg¼of place have′ (1sg, cudun) b. Inalienable possession: (ija) cudu-ni ‘my (personal) place’ 1sg place-1sg.psr have.as.attribute′ (1sg, cudu-)

When the modifier follows the head noun it has an attributive function. In (34a) jo us nijec describes a house where people are sleeping. When the

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

929

930

JOHN R. ROBERTS

modifier precedes the head noun it has a specificational function. In (34b) us nijec describes the type of house. Adjectival-cum-nominal modifiers are represented as predicates which take the item in the NUCR as an argument (underlined). (34)

a. Modifier with attributive function: jo us nij-ec ‘sleeping house (i.e. a house that is asleep)’ house sleep lie-nmlz be′ (jo, [asleep′]) b. Modifier with specificational function: us nij-ec jo ‘sleeping house (i.e. a house for sleeping in)’ sleep lie-nmlz house be′ (jo, [PURP sleep′])

The basic structure of the RP is given in Figure 25.4. Most modifying elements in the RP can occur on their own as referring expressions. Some examples are given in (35). Thus, such modifiers are represented in the syntactic structure as well as in the operator projection. (35)

a. Ben/eu/oso h-ona. big/that/one come-3sg.nom.prs ‘The big (man) is coming.’/ ‘That (man) is coming.’/ ‘Someone is coming.’ b. Cijed h-ogi-na. three come-3pl.nom-prs ‘The three (men) are coming.’

The structure of the alienably possessed PP and inalienably possessed RP are given in Figure 25.5. Ija cebinami ‘my brother’ is the inalienably possessed RP with cebinami as the head noun. Caja ‘woman’ is the head noun of the alienably possessed RP and ija cebinamina functions as the possessor PP. It is clear from the reference that ija ‘1sg’ is a woman. RP

MP

CORER

MP

NUCR

MP

N

N

jo house

nag small

NUM

DEM

cijed three

eu that

‘those three small houses’

NUCR CORER RP RP

QNT DEF DEIC

Figure 25.4 The basic structure of the RP in Amele

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

RP CORER PP RP RPIP RP PRO ija 1SG

NUCR P

CORER NUCR ARG N

N

cebina-mi =na caja opp.sex.sibling-1SG.PSR =of woman ′ (have.as.procreation. ′ (1SG, cebina-), caja)

‘my brother’s woman’

Figure 25.5 The structure of the possessive RP in Amele

Amele uses a pronominal copy strategy where an RP is immediately followed by a personal pronoun. The pronoun indicates the person and number of the RP, as illustrated in (36). The pronoun functions as appositive to the RP and since it is not possible for other clausal elements to occur between the RP and the pronoun, it forms a constituent with the RP. (36)

Pronominal copy strategy: Caja eu uqa/ale/age qaj-ei/-esi/-eig-a. woman that 3sg/3du/3pl cry-3sg.nom/3du.nom/3pl.nom-todp ‘That/those (du)/those (pl) woman/women cried.’

25.2.4.2 PP Structure Amele only has postpositional phrases (PPs). They can be predicative or non-predicative. Predicative PPs occur with temporal or locative adjuncts (e.g. Mande¼na [Monday¼on] ‘on Monday’, be-on′ (Mande, x) or jo¼na [house¼in] ‘in the house’ be-in′ (jo, x)), or with goal, path, source, or instrument argument-adjuncts, as described in Section 25.2.7. The possessor-marking postposition ¼na ‘of’, described in Section 25.2.4.1, is non-predicative.

25.2.5 Constraints on A and U Selection With a ditransitive verb like ihacdoc ‘to show him/her’ in (9), either the perceiver x-argument in see′ (x, y) or the perceived argument y can be selected as undergoer. The latter is the unmarked choice. When the perceived argument is selected as undergoer it is coded on the verb with acc marking attached directly to the verb stem and the perceiver argument can also be marked as a non-macrorole applied object. However, if the perceiver is selected as undergoer it is coded on the verb with acc marking attached

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

931

932

JOHN R. ROBERTS

directly to the verb stem and the perceived argument cannot be marked on the verb. The verb ‘give’ is exceptional in that the undergoer argument functions as the stem of the verb. The first argument of have′ (x, y) (recipient) in ‘give’ LS is thus the only choice for undergoer assignment and there is no alternative argument-marking construction where the second argument of have′ (x, y) (theme) is the undergoer. An example of this was given in (10). When the goal argument in an argument-adjunct PP is a person, an applied object construction may be used as an alternative expression, as illustrated in (4). This is also the case for the source adjunct PP argument, and the benefactive adjunct PP argument, illustrated in (37). For the malefactive argument shown in (38) there is only the applicative form and there is no corresponding argument-adjunct PP form. (37)

Benefactive argument-adjunct applied object: a. Uqa age¼nu jo eu ceh-al-ei-a. 3sg 2pl¼for house that plant-3du.acc-3sg.nom-todp ‘He built both those (two) houses for you (plural).’ b. Uqa jo eu ceh-al-i-ad-ei-a. 3sg house that plant-3du.acc-appl-2pl.acc-3sg.nom-todp ‘He built both those (two) houses for you (plural).’ c. [do′ (3sg, [build′ (3sg, 3du [jo])]) ^ proc create′ (3du [jo])] & ingr exist′ (3du [jo]) purp [become have′ (2pl, 3du [jo])]

(38)

Malefactive argument-adjunct: Ene cain salal-i-t-ag-aun. here proh slide-appl-1sg.acc-2sg.nom-negf NOT [be-here′ (2sg, [do′ (2sg [slide′ (2sg)]] ^ feel.negatively.affected′ (1sg)] ‘Don’t slide here and annoy me (coll. don’t slide on me here).’

25.2.6 Postposition Assignment The only non-predicative postposition assigned by rule is ¼na, which is assigned to the possessor in the alienable possession RP (cf. 33a). (39)

Rule assigning ¼na ‘possessor’: Assign ¼na to the x-argument in the RP logical structure segment: have′ (x, y)

25.2.7 Coding of Adjuncts Temporal adjuncts can be an RP, such as cum ‘yesterday’, uqadec ‘tomorrow’, qila ‘today/now’, cel saen ‘what time’, or a PP, such as Mande¼na ‘on Monday’, Ogas¼na ‘in August’. Locative adjuncts can be an RP, such as ene ‘here’, ono ‘there’, ceheleg ‘up there’, ana ‘where’, or a PP, such as jo¼na [house¼in] ‘in the house’, na¼na [tree¼on] ‘on the tree’, eeta¼na [what¼in] ‘in what’.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

25.2.8 PSA Alignment(s) In Amele the only choice for PSA with a transitive verb is [AT] as there is no passive construction in the language. However, IVCs have exceptional PSA verb agreement coding. Although the undergoer is the only DCA in an IVC and therefore the highest-ranking DCA, it is coded as acc, contrary to (2). This is because 3sg.nom is assigned by default in an IVC.

25.2.9 Information Structure The pre-detached position is outside the potential focus domain. The information structure of (14) is illustrated in Figure 25.6. The RP caja eu ‘that woman’ is a clause-external topic in the pre-detached position and, as a topic, it is outside of the potential focus domain of the proposition. Uqa ‘she’ is the resumptive pronoun in the clause. The potential focus domain (dotted line) is the attributive clause uqa me qee ‘she (is) no good’ and the actual focus domain (solid line) is the predication me qee ‘(is) no good’. Similarly, the post-detached position is outside the potential focus domain. The information structure of (15) is illustrated in Figure 25.7. The RP dana eu ‘that man’ is in the post-detached position and clarifies the identity of eu ‘that’ in the main clause. As clarifying information, it is outside of the potential focus domain of the proposition. The potential focus domain is the statement eu uqa jejegca men ‘he got his voice back’. Here eu ‘that (man)’ functions as a resumptive pronoun for dana eu ‘that SENTENCE PrDP

CLAUSE CORE

RP

PRO

NUC

NEG

PRED ‘That woman, she is no good.’ N (14) Caja eu, woman that IU

uqa 3SG IU

me good

qee. not IU

SPEECH ACT Figure 25.6 The PrDP and the potential focus domain

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

933

934

JOHN R. ROBERTS

SENTENCE CLAUSE RP

PoDP

CORE PP

NUC

ARG

RP

PRED PRO (15) Eu that IU

V uqa 3SG

jeje-g=ca m-en, voice-3SG.PSR=with put-3SG.NOM.REMP IU

IU

dana eu. man eu IU

‘He got his voice back, that man.’ SPEECH ACT Figure 25.7 The PoDP and the potential focus domain

man’. The actual focus domain is the predication uqa jejegca men ‘got his voice back’. Amele has a range of particles that can attach to the end of the sentence which qualify the proposition, such as da ‘but, however, nevertheless’, do ‘be encouraged, let’s do it’, fo ‘yes/no question’ fa ‘dubitive question, maybe’, ijom ‘certainly’, le ‘permission granted’, and mo ‘supplication, pleading’ (Roberts 1990). Some of the sentence particles, such as the question particles, fo and fa, can occur either at the end of the sentence or they can be focused on a particular constituent in the sentence. In (40a) the yes/no question particle, fo, occurs at the end of the sentence. It has scope over the whole proposition and therefore expresses sentence focus. In (40b–c) the question particle is focused on a particular sentence constituent and expresses marked narrow focus on that constituent. (40d) illustrates the use of the question word in ‘who’. It is ungrammatical to have the question particle fo at the end of the sentence in this case since the question word expresses narrow focus and the particle at the end of the sentence would express sentence focus. However, as illustrated in (40e), it is possible to focus on the question word with the question particle. This expresses marked narrow focus. (40)

Focus expressed morphologically: a. Ija aide-ni cabi¼na nu-i-a¼fo? [unmarked sentence focus] 1sg wife-1sg.psr garden¼to go-3sg.nom-todp¼q ‘Did my wife go to the garden?’ b. Ija aide-ni¼fo cabi¼na nu-i-a? [marked narrow focus] 1sg wife-1sg.psr¼q garden¼to go-3sg.nom-todp ‘Was it my wife that went to the garden?’

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele c. Ija aide-ni cabi¼na¼fo nu-i-a? [marked narrow focus] 1sg wife-1sg.psr garden¼to¼q go-3sg.nom-todp ‘Was it to the garden that my wife went?’ d. In cabi¼na nu-i-a(*¼fo)? [unmarked narrow focus] who.sg garden¼to go-3sg.nom-todp(¼q) ‘Who went to the garden?’ e. In¼fo cabi¼na nu-i-a? [marked narrow focus] who.sg¼q garden¼to go-3sg.nom-todp ‘Who is it that went to the garden?’

It is possible to incorporate certain modifier constituents into the verb word to express emphatic focus for the purposes of contrast or correction, or closer specification, for instance. For example, the intensifier bahic ‘very, must, really’ can occur either preceding the verb, as in (41a), or be incorporated into the verb between the verb stem and the verb suffixation, as in (41b). Other modifiers which can be focused in this way are the limiters dih ‘just’ and himec ‘only’, and the negators qee ‘not’ and cain ‘don’t’. (41)

Emphatic focus: a. Age Anut bina-n 2pl God fame-3sg.psr ‘Really praise God!’ b. Age Anut bina-n 2pl God fame-3sg.psr ‘REALLY praise God!’

bahic really

sul-eig-a! lift up-2pl.nom-imp

suleig-a! lift up2pl.nom-imp

25.2.10 Linking In this section the semantics-to-syntax linking for some two-place and threeplace predicates is illustrated. Figure 25.8 shows the semantics-to-syntax linking in a transitive sentence with acc (DUn) marking. Here 3pl [dana] is the first argument in the active predicate LS do′ (3pl, [eat′ (3pl [dana], 3pl [ho])]), and is assigned the actor macrorole. 3pl [ho] ‘pigs’ is the second argument in this LS and also the only argument in the stative predicate LS ingr consumed′ (3pl [ho]). This argument is assigned the undergoer macrorole. The actor argument, 3pl [dana], is selected as PSA and coded as nom agreement on the verb. The undergoer argument, 3pl [ho], is coded as acc agreement on the verb in order to indicate the plurality of the animate referent. Figure 25.9 illustrates the semantics-to-syntax linking in an IVC. The 1sg argument in feel′ (1sg, [afraid′]) is assigned the undergoer macrorole according to the Actor-Undergoer Hierarchy (AUH). The IVC is coded as 3sg.nom agreement by default. There is therefore no linking between this argument marking in the syntax and the logical structure. Figure 25.10 illustrates the semantics-to-syntax linking for a three-place predicate, ihacdoc ‘to show him’. The leftmost argument, 3sg ‘he’, is assigned the actor macrorole and the rightmost argument, 3pl [ho] ‘pigs’,

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

935

936

JOHN R. ROBERTS

SENTENCE CLAUSE RP

RP

RP

CORE

RP

NUC ARG ARG PRED ‘Those men … ate the pigs.

N

PRO

(7c) Dana eu age man that 3PL

N

V

ho … je-ad-ein. pig eat-3PL.ACC-3PL.NOM.REMP

PSA: NOM

ACTIVE: 3PL

ACTOR

[

′ (3PL [dana], [

DCA: ACC UNDERGOER

′ (3PL [dana], 3PL [ho])]) … ] & [INGR

′ (3PL [ho])]

Figure 25.8 Semantics-to-syntax linking in a transitive sentence with ACC marking

SENTENCE CLAUSE RP

CORE NUC ARG PRED ‘I am afraid.’

PRO (6)

Ija 1SG

V cucui-t-ei-a. fear-1SG.ACC-3SG.NOM-TODP

IMPERSONAL: 3SG.NOM

PSA: ACC UNDERGOER

feel′ (1SG, [afraid′ ]) Figure 25.9 Semantics-to-syntax linking in an impersonal verb construction

is assigned the undergoer macrorole. The actor is selected as PSA and coded as nom agreement on the verb. The undergoer is coded as a DUn by acc agreement attaching directly to the verb stem. The non-macrorole core argument, 1sg ‘me’, is coded as a DN by the applicative marker and acc agreement marking on the verb.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

SENTENCE CLAUSE RP

CORE NUC ARG ARG ARG PRED ‘He showed those pigs to me.’

N (9b) Ho eu pig that

PSA: NOM

V ihac-ad-i-t-ei-a. show-3PL.ACC-APPL-1SG.ACC-3SG.NOM-TODP

ACTIVE: 3SG

DCA: ACC

DCA: ACC

NMR

UNDERGOER

ACTOR

[

′ (3SG, Ø)] CAUSE [BECOME

′ (1SG, 3PL [ho])]

Figure 25.10 Semantics-to-syntax linking in a ditransitive predicate

25.3

Language-Specific Topics

The language-specific topics discussed in this chapter are serial verb constructions in Section 25.3.1 and switch-reference in Section 25.3.2.

25.3.1 Serial Verb Constructions Kroeger (2004: 229–230) shows that prototypical serial verb constructions (SVCs) have the following syntactic and semantic properties:16 (42)

Characteristic properties of SVCs: a. A prototypical SVC contains two or more morphologically independent verbs within the same clause, neither of which is an auxiliary. b. There are no conjunctions or other overt markers of subordination or coordination separating the two verbs. c. The serial verbs belong to a single intonation contour, with no pause separating them. d. The entire SVC refers to a single (possibly complex) event. e. A true SVC may contain only one specification for tense, aspect, modality, negation, etc., though these features are sometimes redundantly marked on both verbs.17 f. The two verbs in the SVC share at least one semantic argument.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

937

938

JOHN R. ROBERTS

g. Obligatory non-coreference: a true SVC will not contain two overt NPs which refer to the same argument. h. A prototypical SVC contains only one grammatical subject.

SVCs in Amele do not have all of the prototypical properties listed in (42). In an SVC the non-final verb in the series is marked with -i or -u. Verbs with an -ec infinitive form take -i, while verbs with an -oc infinitive form take -u. The -i/-u marking indicates the verb is dependent (dv).18 This contravenes principle (42b) as the dv marker indicates a nexus relationship. The PSAs of the verbs in an SVC typically have the same referent.19 There are two basic types of SVC. In one type, illustrated in (43a), the verbs are in a cosubordinate relationship. In the other type, illustrated in (44a), the verbs are in a superordinate–subordinate relationship. By comparison, the verb manimei in (43b) is fully inflected with switch-reference morphology and is therefore not a serial verb. Similarly, ehimeig in (44b) is not a serial verb. (43)

Cosubordinate SVC: a. Caja uqa ceta man-i j-ei-a. woman 3sg yam roast-dv eat-3sg.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [[do′ (3sg [caja], Ø)] cause [become roasted′ (ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ^ proc consume′ (3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉 ‘The woman roasted and ate yam.’20 b. Caja uqa ceta man-im-ei j-ei-a. woman 3sg yam roast-ss.seq-3sg.nom eat-3sg.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3sg [caja], Ø)] cause [become roasted′ (ceta)] {&} [[do′ (3sg [caja], [eat′ (3sg [caja], ceta) ^ proc consume′ (3sg [caja], ceta)] & [ingr consumed′ (ceta)]])] 〉〉〉 ‘The woman roasted yam and ate it.’

(44)

Subordinate SVC: a. Age ja eh-i n-eig-a. 3pl firewood take-dv come down-3pl.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP 〈DIR COME DOWN [[do′ (3pl, Ø)] cause [do′ (3pl, [move.away.from.ref.point′ (ja)])] 〉〉〉〉 ‘They brought firewood down.’ b. Age ja eh-im-eig n-eig-a. 3pl firewood take-ss.seq-3pl.nom come down-3pl.nom-todp 〈IF DECL 〈STA REAL 〈TNS TODP [do′ (3pl, Ø)] cause [do′ (3pl, [move.away. from.ref.point′ (ja)])] {&} [do′ (3pl, [move.down.to.ref.point′ (3pl)])] 〉〉〉 ‘They brought firewood and came down.’

In (43a) the verbs mani ‘roast’ and jeia ‘she ate’ describe a series of closely related events. Compare (43b) in which manimei is the fully inflected ss.seq form. Here the roasting and eating are interpreted as separate consecutive events and they are separate clauses in the syntactic representation. The arguments caja ‘woman’ and ceta ‘yam’ in (43a) are shared in the LS for the two predicates. In the syntax for (43a) illustrated in Figure 25.11 the argument caja is represented by an RP in the clause and by 3sg.nom

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

SENTENCE CLAUSE RP

RP

N (43a)

CORE

RP

PRO

Caja uqa woman 3SG

CORE

CORE

RP NUC

NUC

PRED

PRED

V

V

N

ceta man-i yam roast-DV

ARG

‘The woman roasted and ate yam.’

j-ei-a. eat-3SG.NOM-TODP

Figure 25.11 SVC with cosubordinate core juncture

agreement on the verb jeia ‘she ate’. The argument ceta is represented as a core RP argument of mani ‘roast’. It is not coded on the verb by acc agreement as it is inanimate and a mass noun. The SVC is a cosubordinate CORE [[CORE] [CORE]] juncture because it is not possible to have temporal or locative clausal adjuncts applying separately to either core. By comparison, in (43b) it would be possible to qualify jeia ‘she ate’ with the temporal adjunct hibna ‘later’ because it is a separate clause. (44a) describes a single event of ‘take down’. The verb n- ‘come down’ is a directional modifier of the verb ehi ‘take’ and is coded in the LS as such. There is no shared argument between the verbs as n- is represented in the LS as a directional operator. (44a) therefore contravenes principle (42f). Compare (44b), where ehimeig is fully inflected for ss.seq. Here the event ehimeig ‘they took’ is linked to the event neiga ‘they came down’ by {&} in the LS. They are separate consecutive events in the LS and are separate clauses in the syntactic structure. In the syntactic structure for (44a) in Figure 25.12 the verb n- is in a subordinate relationship to ehi. This is ad-nuclear subordination. The 3pl.nom argument agreement -eig attaches to n- as the final verb in the series. With respect to the characteristic SVC properties detailed in (42), a particular SVC may not necessarily belong to a single intonation contour (cf. 42c). Compare the sentence in Figure 25.13. The li ‘go’ verb is a directional modifier of ehi ‘take’. The first two cores form one intonational unit and the last core is a separate intonational unit. However, the dv coding marks this as describing a series of linked events which should be interpreted as a unified complex event. The subordinate SVC has a range of modifying functions in addition to expressing directionality. Examples are given in (45)–(49). In each case the modifying verb follows the verb that is modified. (45)–(46) express aspect periphrastically. (45) illustrates how continuative aspect is expressed by a

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

939

940

JOHN R. ROBERTS

SENTENCE CLAUSE RP

CORE RP

NUC

NUC

ARG

PRED ‘They brought the firewood down.’ PRO

N

V

V

(44a) Age ja eh-i 3PL firewood take-DV

n-eig-a. come down-3PL.NOM-TODP

Figure 25.12 SVC with nuclear subordination

SENTENCE CLAUSE RP RP

N

RP

PRO

CORE

CORE

RP NUC

NUC

PRED

PRED

V

V

N

CORE NUC

RP

NUC

ARG

PRED V

N

V

Caja age saab tac-i eh-i l-i bisin ohis m-egi-na. woman 3PL food fill up-DV take-DV go-DV store above put-3PL.NOM.PRS ‘The women fill up (their bilums) with food, take it away and put it in the attic store.’ Figure 25.13 Extended coordinate SVC

posture verb: bilec ‘to sit’, nijec ‘to lie’, tawec ‘to stand’, and (46) illustrates how completive aspect is expressed with the verb hedoc ‘to finish it’. (47)–(49) illustrate some other common forms of this type of subordinate SVC. (45)

Continuative aspect expressed with a posture verb: Age nu-i bil-egi-na. 3pl go-dv sit-3pl.nom-prs ‘They go continuously.’

(46)

Completive aspect expressed with the verb hedoc ‘to finish it’: Age jo ceh-i he-d-oig-a. 3pl house plant-dv finish-3sg.acc-3pl.nom-todp ‘They finished building the house.

(47)

Investigative modification expressed with the verb fec ‘to see’: Uqa wehuc j-i f-ei-a. 3sg soup eat-dv see-3sg.nom-todp ‘She tasted the soup.’

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

(48)

Superlative modification expressed with the verb cuhadoc ‘to excel’: Caja uqa ola-Ø cus-i cuha-d-on. woman 3sg face-3sg.psr scrub-dv excel-3sg.acc-3sg.nom.remp ‘The woman cleaned her face well.’

(49)

Enumerative modification expressed with a numeral:21 Q-u lecis-d-oig-a. hit-dv two-3sg.acc-3pl.nom-todp ‘They hit it twice.’

25.3.2 Switch-Reference Switch-reference (SR) in Amele is judged to be a local syntactic device for monitoring the referentiality of PSA arguments between adjacent clauses as to whether they have identical or non-identical reference (Roberts (2017). Dependent SR verbs occur most commonly in a clause chain.22 A clause chain comprises a string of linked clauses with only the final clause in the chain marked for clausal operator categories, such as tense or illocutionary force. The non-final clauses are dependent on the final clause for clausal operator designation. The operator-dependent clauses are also typically marked with switch-reference morphology. Amele has two basic types of ss/ds morphology. One codes sequential events and the other codes simultaneous overlap of events. The sequential verb is marked with either -im ‘ss.seq’ or -ec/-oc ‘ds. seq’ followed by nom agreement morphology. The simultaneous verb is marked with nom agreement morphology which indicates either ss.sim or ds.sim. The ds.sim morphology is further divided into that which indicates realis status and that which indicates irrealis status of the tense or IF category marked on the final clause in the clause chain.23 The ds.sim verb agrees in status value with the status value of the final clause. ss/ds clauses are most commonly in a cosubordinate relationship with other clauses (see Chapter 13). However, ss/ds clauses can also have a subordinate function. 25.3.2.1 Cosubordinate Switch-Reference Clauses A typical clause chain of four linked clauses in a cosubordinate relationship is illustrated in (50). The verbs neceb and tobocomin are marked for ds.seq and the verb series sumudi bibiligin is marked for ds.sim.r since the tense category of the final verb is realis status. The final verb in the chain is belowan and this verb is marked for yesterday’s past tense and declarative illocutionary force. Sequential events are indicated by ‘{&}’ in the LS and simultaneous events by ‘{^}’. The clauses . . . sumudi bibiligin, neceb, tobocomin and belowan in (50) are in a cosubordinate relationship as the tense, status and IF categories marked on belowan have scope over all the clauses in the chain. Sumudi bibiligin is a continuative SVC. The verb bibiligin modifies sumudi as a continuative event and the durative marking on bi~biligin indicates the ‘waiting’ event overlapped for a period of time with the ‘came down’ (neceb) event.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

941

942

JOHN R. ROBERTS

(50)

Dependent switch-reference verbs: 1. Ija Malolo uqa¼na ka jic anag ono¼nu 1sg Malolo 3sg¼of car road mother there¼for sum-ud-i bi~bil-igin wait-3sg.acc-dv dur ~sit-1sg.nom.ds.sim.r 2. n-ec-eb come down-ds.seq-3sg.nom 3. tob-oc-omin ascend-ds.seq-1sg.nom 4. bel-ow-an. go.nsg-1du.nom-yestp 〈IF DECL 〈STA REAL 〈TNS YESTP 〈ASP CONT 〈ASP DUR be-loc′ (ono, [be-loc′ (jic anag, do′ (1sg, [wait′ (1sg, become be-loc′ (jic anag, [have′ (Malolo, ka))])〉〉 {^} do′ (3sg, [come.down′ (3sg)]) {&} do′ (1sg, [ascend′ (1sg)]) {&} do′ (1du, [move.away.from.ref.point′ (1du)])〉〉〉 ‘While I waited for Malolo’s car there on the main road, he came down, I climbed in and off we (du) went.’

Note that the final clause in a clause chain need not be finite. In (51) the purpose clause ho bubusaleb qoc ‘to kill the pig as it runs out’ is a clause chain with qoc ‘to kill’ as the final clause. The verb bubusaleb in the dependent clause is marked for ds simultaneous irrealis nom agreement morphology as the understood PSA of qoc is ‘those men’, and qoc, being infinitive, has irrealis status. (51) Infinitival final clause: Dana eu ho bu~busal-eb man that pig dur ~run out-3sg.nom.ds.sim.irr ‘Those men came to kill the pig as it runs out.’

q-oc¼nu hit-inf¼for

h-oig-a. come-3pl.nom-todp

A diagram of the syntax of (50) is given in Figure 25.14. The clausal operators of tense, status and IF apply to all the dependent clauses, whereas the core and nuclear operators apply only to a particular verb. The ss/ds markers are treated as clause linkage markers. The rules for how the SR system works are set out in (52) and (53). In terms of Comrie’s (1989) different types of reference-tracking systems, SR in Amele is local and interclausal. (52)

Protocol for selecting switch-reference controlling clause: For a dependent clause in a cosubordinate relationship ({&} or {^}), select the next cosubordinate clause as the controlling clause for switchreference marking.

(53)

Amele switch-reference coreferentiality rule: Check the pivot of the dependent clause against the pivot of the controlling clause for coreferentiality of identity. If the referent is identical, mark ss; if not, mark ds.

With respect to (53), there are complexities in determining referential identity.24 For example, Amele SR is asymmetrical where there is referential

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

943

A Grammatical Sketch of Amele

SENTENCE CLAUSE

CLAUSE RP

CLAUSE CORE

PP

CORE

NUC NUC

V

CLMDS CLAUSE

CORE

CORE

ARGDS NUC

PRED PRO

CLAUSE

CLMDS

V

Ija Malolo uqana ka jic anag ononu sumudi bi~bil-igin

NUC ARG

NUC ARG NUC ARG

PRED

PRED

PRED

V

V

V

n-ec-eb

tob-oc-omin

bel-owan=Ø.

NUC

NUC

NUC

CORE

CORE

CORE

CORE

CLAUSE

CLAUSE

CLAUSE

CLAUSE

NUC NUC

ASP: DUR ASP: CONT

CLAUSE

TNS: YESTP

CLAUSE

STA: REALIS

CLAUSE

IF: DECL

SENTENCE Figure 25.14 Cosubordinate clause chain structure, example (50)

overlap between the pivot referents of the controlling clause and the pivot referents of the dependent clause. Where the pivot referent(s) of the controlling clause is (are) properly included in the set of pivot referents of the dependent clause then ss is marked. However, when the pivot referent(s) of the dependent clause is (are) properly included in the set of pivot referent(s) of the controlling clause then ds must be marked. For example, in (50) 1sg. nom in [3] is properly included in 1du.nom in [4] and in this case ds must be marked. If it was the other way around (‘we (du) climbed in and off I went’) and 1sg.nom was the controlling referent and 1du.nom was the dependent referent then ss would be marked.25 Stirling (1993) suggests that SR in Amele tracks events rather than the main participant(s) in the event. As already mentioned, Amele has asymmetrical

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

944

JOHN R. ROBERTS

ss/ds marking when there is referential overlap between the PSA of the controlling clause and the PSA of the marked clause. When a plural number of participants controls a singular number of participants then ds must be marked. In (54) a multiple (plural) event budu beladeiga with multiple participants controls a singular dependent event age cajimeig with multiple participants. If the SR tracked events, as suggested by Stirling (1993), then it would be expected that a move from a singular event to a multiple event should trigger ds marking. However, ss is marked. This shows that the Amele SR system tracks the coreferentiality of participants across clauses (both plural in this case) rather than the coreferentiality of events. (54) Event quantification and SR marking: Dana caja¼ca age caj-im-eig bud-u bel-ad-eig-a. man woman¼add 3pl arise-ss.seq-3pl.nom disperse-dv go-distr-3pl.nom-todp ‘The people got up and dispersed in all directions (multiple events).’

Further evidence that SR in Amele tracks participants rather than events is provided by the reciprocal verb form, as illustrated in (55). Here a pair of dsmarked verbs functions as the predicate in the nucleus.26 The 3pl.nom agreement at the end of the predicate refers to the whole group of women (caja). The ds-marked verbs express the notion of one womani hitting another womank and this being reciprocated by womank hitting womani. Thus the ds marking refers to individual participants in the overall event of women hitting. (55)

Reciprocal verb and SR marking: Caja q-oc-ob q-oc-ob egi-na. woman hit-ds.seq-3sg.nom hit-ds.seq-3sg.nom 3pl.nom-prs 〈IF DECL 〈STA REAL 〈TNS PRS do′ (3pl [caja], [seml do′ (3sg [caja]i, [hit′ (3sg [caja]i, 3sg [caja]k)]) {&} seml do′ (3sg [caja]k, [hit′ (3sg [caja]k, 3sg [caja]i)])]〉〉〉 ‘The women are hitting each other.’

25.3.2.2 Subordinate Switch-Reference Clauses Switch-reference clauses can also have a subordinate function. The following functions are described below: • core subordinate object clause of perception verbs • layered cosubordination in a clause chain for modification • ad-clausal subordination in a conditional clause An SR clause may function as the object of a perception verb.27 In (56a) the second argument of the perception verb perceive′ is a proposition become arise′ (dedeman) ‘a smell had arisen’. Therefore in the syntax dedeman waseceb occupies the DUn position in the clause between the PSA caja eu ‘that woman’ and the verb don ‘she perceived’. In Roberts (1988a) it is shown that overtly coordinate clauses in Amele, such as a qa ‘but’ clause,

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

SENTENCE CLAUSE RP

CORE CLAUSE RP

Caja eu woman that

N

ARG

CORE NUC

N

NUC

CLMDS

ARG

PRED

PRED

V

V

dedeman was-ec-eb smell arise-DS.SEQ-3SG.NOM

d-on =Ø. perceive-3SG.NOM.REMP

NUC

NUC

CORE CLAUSE CORE CLAUSE

TNS: REMP

CLAUSE

STA: REALIS

CLAUSE

IF: DECL

SENTENCE Figure 25.15 Core subordinate DUn clause, example (56a)

cannot be embedded within another clause. Therefore dedeman waseceb cannot be in a coordinate relationship with don. It is core subordinate, as illustrated in Figure 25.15. Since become arise′ (dedeman) is an accomplishment (ending in a result state) it is interpreted as being in a sequential relationship with the matrix predicate perceive′, that is, the smell arose before the woman perceived it. Therefore, the verb waseceb is marked for ds.seq. Similarly, in (56b) the second argument of the perception predicate see′ is the proposition lie′ (3sg [ma susul] ‘the taro peelings are lying (on the ground)’. Consequently, ma susul eu ninijen occupies the DUn position in the clause between the PSA mala uqa ‘chicken he’ and the verb fen ‘he saw’. Here ma susul eu ninijen is in a core subordinate relationship with fen. Because the proposition lie′ (3sg [ma susul]) is a state it is interpreted as occurring simultaneously with see′. The verb ninijen is therefore marked as ds.sim.r.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

945

946

JOHN R. ROBERTS

(56) Core subordinate DUn clause of perception verbs: a. Caja eu dedeman was-ec-eb d-on. woman that smell arise-ds.seq-3sg.nom perceive-3sg.nom.remp 〈IF DECL 〈STA REAL 〈TNS REMP perceive′ (3sg [caja], [become arise′ (3sg [dedeman])])〉〉〉 ‘That woman perceived that a smell had arisen.’ b. Mala uqa ma susul eu ni~nij-en f-en. chicken 3sg taro peelings that dur ~lie-3sg.nom.ds.sim.r see-3sg.nom.remp 〈IF DECL 〈STA REAL 〈TNS REMP see′ (3sg [mala], [〈ASP DUR lie′ (3sg [ma susul])〉])〉〉〉 ‘Chicken saw those taro peelings lying (there).’

A switch-reference clause can function as a modifier of a following clause. (57) is a sentence containing five clauses. Clauses [1, 3, 5] describe the mainline events, while clauses [2, 4] provide background information. Clause [2] says ‘when the yams ripened . . .’ and clause [4] says ‘when the yams dried . . .’. The secondary nature of clause [2] is indicated by the ss marking on clause [1]. In this case, clause [3] is the controlling clause for the ss marking on clause [1]. Clause [3] is also the controlling clause for the ds marking on clause [2]. Thus clause [2] functions as a temporal modifier to clause [3]. Similarly, the ss marking on clause [3] is controlled by clause [5] and clause [4], marked for ds, functions as a temporal modifier to clause [5]. There are thus layers of cosubordination, which can be schematically represented as in (57a). (57)

Layered cosubordination in a clause chain: a. [Clause-1]SS [[Clause-2]DS Clause-3]SS [[Clause-4]DS Clause-5] 1. Ceta bahim m-i he-d-um-ei yam store put-dv finish-3sg.acc-ss.seq-3sg.nom 2. ceta wal m-ec-eb yam ripe put-ds.seq-3sg.nom 3. ceta eu hun-im-ei yam that bore-ss.seq-3sg.nom 4. gulden h-oc-ob dried come-ds.seq-3sg.nom 5. ceta bahim¼na tac-en. yam store¼in fill-3sg.nom.remp b. 〈IF DECL 〈STA REAL 〈TNS REMP 〈ASP COMPL [do′ (3sg i, [make′ (3sg i, ceta bahim)]) ^ proc create′ (ceta bahim)] & ingr exist′ (ceta bahim)〉 {&} [[ become ripened′ (ceta)] {&} [do′ (3sg i, [dig.up′ (3sg i, ceta)]]] {&}[[become dried′ (ceta)] {&} [do′ (3sg i, Ø)] cause [become be-in′ (ceta bahim, Ø)] cause [ingr full′ (ceta bahim)]])〉〉〉 ‘He finished making the yam store and, when the yams had ripened, he dug up those yams and when they dried he put them in the yam store.’

The syntactic structure of (57) is diagrammed in Figure 25.16. The modifying clauses [2] and [4] are in a cosubordinate relationship to the following matrix clause. The matrix clause is the controlling clause of the ‘subcosubordinate’ clause for the purposes of switch-reference marking.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

947

A Grammatical Sketch of Amele

SENTENCE CLAUSE CLAUSESS

CLAUSESS PERIPHERY

CORE

Ceta bahim mi hedumei

CORE

CLAUSEDS

CORE

CORE

NUC ASP: COMPL

PERIPHERY

CLAUSEDS

ceta wal meceb

NUC

CORE

CLAUSE

ceta eu hunimei gulden hocob ceta bahimna tacen=Ø. NUC

CORE CLAUSE

CLAUSE

CORE

NUC

NUC

CORE CORE

CLAUSE

CLAUSE

CORE CLAUSE

CLAUSE

TNS: REMP

CLAUSE

STA: REAL

CLAUSE

IF: DECL

SENTENCE Figure 25.16 Layered cosubordination in (57)

The PSA in the sub-cosubordinate clause is different to the PSA in the matrix clause in each case, so ds is marked. Note that this produces an anomalous SR marking between clause [1] and the immediately following clause [2], and between clause [3] and the immediately following clause [4]. The verb in clause [1] is marked for ss when the PSA of clause [2] is different to the PSA in clause [1]. The verb in clause [3] is also marked for ss when the PSA of clause [4] is different to the PSA in clause [3]. This shows that the switch-reference marking system takes account of the layers of cosubordination. For clauses signalling the mainline events, such as clauses [1, 3, 5] in (57) and the four clauses in (50) the SR-marking system selects the next cosubordinate clause as the controlling clause for SR marking. For an SR clause indicating background information, such as [2, 4] in (57), it is interpreted being relevant to the following ‘mainline event’ expressing clause, and the SR-marking system selects that clause as the

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

948

JOHN R. ROBERTS

controlling clause for SR marking. A translation of (57) directly reflecting its structure would be ‘He finished making the yam store; the yams ripened, and then he dug up those yams; the yams dried, and then he put them in the yam store’. The protasis (condition clause) in a conditional sentence modifies the apodosis (consequence clause). The condition clause in the protasis can be a switch-reference clause. ss.seq marking for the condition clause is -if/-uf followed by the regular ss.seq nom agreement morphology. For all other ss/ds marking the conjunction fi ‘if’ is cliticized to the switch-reference verb. An example of an ss condition clause is given in (58a) and of a ds condition clause in (58b). In (58a) the ss marking actually indicates the certainty of the consequence. In both cases, the ss/ds condition clause can be postposed after the consequence clause and it is ad-clausal subordination. (58)

Ad-clausal subordination in a conditional clause: a. Qee j-i he-d-uf-eg qaga-h-ig-en. not eat-dv finish-3sg.acc-ss.cond-2sg.nom kill-2sg.acc-1sg.nom-fut ‘If you don’t eat it all (lit. finish) I will kill you.’ b. Ija ja hud-ec-emin¼fi uqa saab man-igi-an. 1sg fire open-ds.seq-1sg.nom¼if 3sg food cook-3sg.nom-fut ‘If I light the fire she will cook the food.’

The fact that switch-reference in Amele occurs in subordinate clauses as well as cosubordinate clauses requires a revision of the protocol for selecting the controlling clause; (52) has therefore been modified as follows. (59)

Protocol for selecting switch-reference controlling clause: 1. For a dependent clause in a cosubordinate relationship, select the next cosubordinate clause as the controlling clause for switch-reference marking. 2. For a dependent clause in a subordinate relationship, select the superordinate matrix clause as the controlling clause for switch-reference marking.

25.3.2.3 Grammatical Functions and Switch-Reference The coreferentiality rule as stated in (53) applies in most cases. However, it does not apply in a straightforward way when the dependent or controlling clause is an IVC. In (60a) cucuiimig ‘I feared something’ is a state verb with two arguments. It is M-transitive. The actor (emoter) argument is marked with 1sg.nom agreement and the undergoer (target) argument is unspecified. Here, cucuiimig functions as the controlling clause for ija cocobig ‘as I walked’ and the dependent clause to busali nuiga ‘I ran away’. With respect to (53), the pivot of ija cocobig is 1sg.nom and is coreferential with the 1sg.nom pivot of cucuiimig. Thus, cocobig is marked ss. Then the 1sg.nom pivot of cucuiimig is coreferential with the 1sg.nom pivot of busali nuiga and cucuiimig is marked ss. Adherence to (53) is straightforward.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele (60)

a. State verb cucuiec fear′ (x, y) ‘to fear something’: Ija co~cob-ig cucui-im-ig 1sg dur ~walk-1sg.nom.ss.sim fear-ss.seq-1sg.nom busal-i nu-ig-a. run away-dv go-1sg.nom-todp ‘As I walked along, I was afraid, and I fled.’ b. Impersonal verb cucuidoc feel′ (x, [afraid′]) ‘to be afraid’: Ija co~cob-ig cucui-t-ec-eb 1sg dur ~walk-1sg.nom.ss.sim fear-1sg.acc-ds.seq-3sg.nom busal-i nu-ig-a. run away-dv go-1sg.nom-todp ‘As I walked along, I became afraid, and I fled.’

Now compare (60b). Here, cucuiteceb ‘I was afraid’ is an IVC.28 In this case, the verb has a 1sg.acc ‘experiencer’ argument and a 3sg.nom ‘empty’ argument. When cucuiteceb functions as the controlling clause to ija cocobig this verb is marked ss. Thus, the 1sg.acc argument in the IVC functions as pivot when it is the controlling clause. However, when cucuiteceb functions as the dependent clause to busali nuiga it is marked ds. This shows that the pivot arguments are not being compared between these clauses. Instead, the controller 3sg.nom argument in the IVC is compared with the pivot in the non-IVC controlling clause. Thus, the coreferentiality rule in (53) needs to be modified to (61). (61) applies to both regular verbs and IVCs. With a regular verb the pivot and controller are one and the same argument so monitoring either function does not select a different argument. However, in an IVC, the pivot and controller are different arguments, so it makes a difference which argument is monitored when the IVC functions as controlling clause or dependent clause. (61)

Amele switch-reference coreferentiality rule: Check the controller of the dependent clause against the pivot of the controlling clause for coreferentiality of identity. If the referent is identical, mark ss; if not, mark ds.

25.3.2.4 Pragmatic Functions of the SR System Often in Amele text, there occur what appear to be ‘anomalous’ ds markings where ds is indicated but the PSA of the marked clause and the controlling clause are the same. These ds markings are, in fact, not anomalous but are indicating a change in discourse theme. The thematic changes are primarily in the area of time, place and possible world setting (real vs. unreal). Thematic changes of time and place are often accompanied by temporal and locative modifier expressions and a ds thematic change of place is most commonly marked on a motion verb. Thematic changes in a possible-world setting are normally a switch from the real world to an intended or proposed action or vice versa, a switch from intended/proposed action to the real world. Examples of this phenomenon are given in Roberts (1987: 303–305, 1988b). However, these are pragmatic extensions of the SR

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

949

950

JOHN R. ROBERTS

system. At core, the switch-reference system in Amele is syntactically motivated. It is a local device for monitoring the referentiality of PSA arguments between adjacent clauses as to whether they have identical or non-identical reference.

References Andersen, T. David and John R. Roberts. 1991. An exception to the hodiernal: non-hodiernal distinction. Studies in Language 15(2): 295–299. Comrie, Bernard. 1989. Some general properties of reference-tracking systems. In Doug Arnold, Martin Atkinson, Jacques Durand, Claire Grover and Louisa Sadler (eds.), Essays on Grammatical Theory and Universal Grammar, 37–51. Cheltenham: Clarendon Press. Hasegawa, Yoko. 1993. A Study of Japanese Clause Linkage: The Connective te in Japanese. Stanford, CAS: CSLI Publications. Imai, Shingo. 1998. Logical Structures and Case Marking in Japanese. Unpublished MA project, University at Buffalo (SUNY). Kroeger, Paul R. 2004. Analyzing Syntax. A Lexical-Functional Approach. Cambridge: Cambridge University Press. Latrouite, Anja and R. D. Van Valin, Jr. 2014. Event existentials in Tagalog: A Role and Reference Grammar account. In I. Wayan Arka and N. L. K Mas Indrawati (eds.), Argument Realisations and Related Constructions in Austronesian Languages: Papers from 12-ICAL, Vol. 2, 161–174. Canberra: Asia-Pacific Linguistics. Lewis, M. Paul, Gary F. Simons and Charles D. Fennig (eds.). 2014. Amele. In Ethnologue: Languages of the World (17th ed.). Dallas, TX: SIL International. www.ethnologue.com/language/aey. Roberts, John R. 1987. Amele. London: Croom Helm. Roberts, John R. 1988a. Amele switch-reference and the Theory of Grammar. Linguistic Inquiry 19(1): 45–63. Roberts, John R. 1988b. Switch-reference in Papuan languages: A syntactic or extrasyntactic device? Australian Journal of Linguistics 8: 75–118. Roberts, John R. 1990. Modality in Amele and other Papuan languages. Journal of Linguistics 26: 363–401. Roberts, John R. 1991a. Reduplication in Amele. In T. Dutton (ed.) Papers in Papuan Linguistics, No. 1, 115–146. Canberra: The Australian National University. Roberts, John R. 1991b. A study of the dialects of Amele. Language and Linguistics in Melanesia 22(1–2): 67–126. Roberts, John R. 1991c. Orthography reform in Amele: Part One. Notes on Literacy 17(4): 1–20. Roberts, John R. 1992. Orthography reform in Amele: Part Two. Notes on Literacy 18(1): 1–32.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

Roberts, John R. 1994. The category ‘irrealis’ in Papuan medial verbs. Notes on Linguistics 67: 5–39. Roberts, John R. 1996. A Government and Binding analysis of the verb in Amele. Language and Linguistics in Melanesia 27(1): 1–66. Roberts, John R. 1997a. Switch-reference in Papua New Guinea: A preliminary survey. In A. Pawley (ed.), Papers in Papuan Linguistics, No. 3, 101–241. Canberra: The Australian National University. Roberts, John R. 1997b. The syntax of discourse structure. Notes on Translation 11(2): 15–34. Roberts, John R. 1997c. GIVE in Amele. In John Newman (ed.), The Linguistics of Giving, 1–33. Amsterdam: John Benjamins. www.sil.org/resources/archives/ 22980. Roberts, John R. 2001. Impersonal constructions in Amele. In Alexandra Y. Aikenvald, R. M. W. Dixon and Masayuki Onishi (eds.), Non-Canonical Marking of Subjects and Objects (Typological Studies in Language 46), 201–250. Amsterdam: John Benjamins. Roberts, John R. 2012. Serial verbs in English: An RRG analysis of catenative verb constructions. Functions of Language 19(2): 201–234. Roberts, John R. 2015a. Distributives in Amele. A Role and Reference Grammar analysis. SIL Electronic Workpapers 2015–001. www.sil.org/resources/ publications/entry/60478. Roberts, John R. 2015b. Inalienable possession in Amele: A Role and Reference Grammar account. SIL Electronic Workpapers 2015–002. www.sil.org/ resources/publications/entry/60480. Roberts, John R. 2017. The typology of switch-reference. In Alexandra Y. Aikhenvald and R. M. W. Dixon (eds.), Handbook of Linguistics Typology, 538–573. Cambridge: Cambridge University Press. Stirling, Lesley. 1993. Switch-Reference and Discourse Representation. Cambridge: Cambridge University Press. Talmy, Leonard. 2007. Lexical typologies. In Timothy Shopen (ed.), Language Typology and Syntactic Description, 66–168. Cambridge: Cambridge University Press. Toratani, Kiyoko. 2002. The Morphosyntax and the Logical Structures of Compound Verbs in Japanese. Unpublished PhD dissertation, University at Buffalo (SUNY). Z’graggen, John A. 1975. The Languages of the Madang District, Papua New Guinea (Pacific Linguistics B-41). Canberra: The Australian National University.

Notes 1 See Roberts (1987) for a full linguistic description of the Amele language from a typological perspective. The dialect of Amele represented in this study is the Haija dialect (Roberts 1991b).

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

951

952

JOHN R. ROBERTS

2 See Roberts (1997b) for the wider consequences of this feature. 3 See Roberts (1996) for an account of PSA agreement morphology and Roberts (2001) for an account of DCA agreement morphology. 4 Non-finite includes infinitive, serial and switch-reference verbs, none of which are marked for tense. 5 In the Amele orthography c is [Ɂ] and q is [g͡ b]. See Roberts (1991c, 1992) for a full account of Amele orthography issues. 6 The form of acc morphology with qoc ‘to hit’ is irregular. 7 See Roberts (1997c) for a full account of ‘give’ in Amele. 8 Kin terms in Amele are divided morphologically into kin orientation terms (have.as.orientation.kin′ (x, y)) and kin procreation terms (have.as. procreation.kin′ (x, y)) (Roberts 2015b). 9 The verb belec ‘to go’ requires a non-singular PSA. 10 Andersen and Roberts (1991) find that in one dialect of Amele the today’s past tense does not occur. There is only present tense and yesterday’s past tense. 11 Prospective tense is expressed by a reduced form of the future tense suffix, -e(n)/-a(n), together with the auxiliary posture verb bil-ec ‘to sit’. 12 They have the same pronunciation and same written form as the corresponding accusative agreement markers. 13 See Roberts (1991a). 14 See Roberts (2015b) for a full account of alienable and inalienable possession in Amele. 15 Kinship terms can also express the plurality of the possessed, e.g. ate-ni-el [daughter-1sg.psr-pl.psd] ‘my daughters’. 16 See Roberts (2012) for an application of these defining characteristics to catenative verbs in English. 17 Note that Kroeger (2004) does not recognize that operators such as tense, aspect, modality, negation have scope over different layers of clause structure. In RRG, instead, aspect is a nuclear operator and, therefore, it is possible in this framework to capture the fact that individual verbs in an SVC can be marked for different types of aspect. 18 This is analogous to the connective -te in Japanese (Hasegawa 1993). 19 There are some serial verb constructions in which this is not the case, though. 20 Bare ‘&’ and ‘^’ are used within the LS of verbs to signal sequential and simultaneous subparts of a single event, whereas ‘{&}’ and ‘{^}’ indicate sequential and simultaneous events in complex sentences. 21 Lecis-doc is the ordinal form of this numeral. See Section 25.2.4.1. 22 Switch-reference in clause chains is common in PNG languages (Roberts 1997a). 23 See the status operator in Section 25.2.3.2 for details of which categories are realis and which are irrealis. 24 Another complexity is how the coreferentiality rule (53) applies to IVCs. See grammatical functions and SR in Section 25.3.2.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

A Grammatical Sketch of Amele

25 For a fuller account of referential overlap in Amele and other Papuan languages, see Roberts (2017). 26 There are two pieces of evidence that qocob qocob functions as the predicate in the nucleus. First, qocob qocob may be substituted for the verb stem in the infinitive form. Cf. q-oc ‘to hit’ with qocob qocob ec ‘to hit each other’. Second, durative aspect (C)V~ reduplication on the verb is marked on the matrix nom agreement in the reciprocal verb form, e.g. q-oc-ob q-ocob e~egin [hit-ds.seq-3sg.nom hit-ds.seq-3sg.nom dur ~3pl.nom.ds. sim.r] ‘while they hit each other’. Cf. (29) where durative (C)V~ reduplication is marked on the nom agreement within the verb word. 27 Contrary to what was stated in Roberts (1988a), nominal clauses in Amele can be marked for ss/ds. 28 The semantic difference between cucuiec and cucuidoc is that with cucuiec the emotion of fear is self-generated, whereas with cucuidoc the fear is a response to an unspecified outside stimulus. Also, with the M-transitive verb, the target argument can be marked on the verb by acc agreement, cucui-ad-ec ‘to fear them’, whereas with the IVC the target argument has to be expressed in a PP, age¼nu cucui-d-oc ‘to be afraid of them’.

https://doi.org/10.1017/9781316418086.031 Published online by Cambridge University Press

953

26 Case and Voice in Amis (Austronesian, Taiwan) Joy J. Wu

Abbreviations The abbreviations and symbols used in the glosses in general follow the Leipzig glossing rules (www.eva.mpg.de/lingua/resources/glossing-rules. php), with the following additions: ABLT ASP AV CN FAC IA LA LNK NEUT

26.1

abilitative aspect marker actor voice common noun factual mood instrumental applicative locative applicative linker neutral voice

NFIN PPN PrCS PrDP PREP RED UV VOL

non-finite personal proper noun pre-core slot pre-detached position preposition reduplication undergoer voice volitative mood

Introduction

This chapter presents a sketch grammar of Amis with a focus on the discussion of the phenomena related to its case marking and voice. Amis is currently spoken by 210,839 speakers in Taiwan.1 It is classified as one of the Eastern Formosan branch languages (Blust 1999) and is reported to consist of five dialects (Tsuchida 1988). The following discussion is based on the data collected from the Central dialect.

26.2

Basic Clause Patterns

Amis is a predicate-initial language. Three types of predicates can be identified: verbal, nominal and prepositional. One-place, two-place and three-place verbal predicates and their logical structures are provided in (1):2

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis (1)

Amis verbal predicates a. Rakat kako i lalan. walk 1s.nom prep road ‘I am walking on the road.’ a′. be-on′ (lalan, [do′ (kako, [walk′ (kako)]) b. Mi-palo Ø-ci Sawmah ci Mayaw-an. av-beat nom-ppn Sawmah ppn Mayaw-dat ‘Sawmah is beating Mayaw.’ Or ‘Sawmah is going to beat Mayaw.’ b′. do′ (Sawmah, [go′ (Sawmah)]) & INGR be-at′ (y, Sawmah) PURP do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) b′′. do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) c. Ma-palo n-i Sawmah Ø-ci Mayaw. uv-beat gen-ppn Sawmah nom-ppn Mayaw ‘Mayaw was beaten by Sawmah.’ c.′ do′ (Sawmah, [beat′ (Sawmah, Mayaw)]) & BECOME beaten′ (Mayaw) d. Pa-nanom kako t-o wawa t-o sayta. caus-water 1s.nom dat-cn child dat-cn soda ‘I gave the child soda to drink.’ (Causative, AV) d′. do′ (kako, Ø ) CAUSE BECOME have.water′ (wawa, sayta)

As shown in (1), verbal predicates are usually affixed with a voice marker (e.g. mi-), followed by various numbers of arguments that are marked by a case marker, be it a nominative case (for the privileged syntactic argument (PSA)), a genitive case (for a non-PSA actor or a possessor), or a dative case (for an oblique core argument or an adjunct). A locative argument or adjunct is marked by the preposition i. Nominal predicates and prepositional predicates are exemplified respectively in (2a) and (2b). (2)

Amis nominal and prepositional predicates a. O singsi cingra. cn teacher 3s.nom ‘He is a teacher.’ a′. be′ (cingra, [teacher′]) b. I loma’ Ø-ci prep house nom-ppn ‘Mayaw is at home.’ b′. be-at′ (loma’, Mayaw)

Mayaw. Mayaw

Unlike verbal predicates, nominal predicates are preceded by a noun class marker3 and prepositional predicates are initiated by a preposition. Special clause types, such as displacement constructions (e.g. 3a) and topic constructions (e.g. 3b), are given below.4

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

955

956

JOY J. WU

(3)

Amis displacement construction and topic construction a. O fafahian a singsi k-o ka-olah-an¼ako cn woman lnk teacher nom-cn nfin-like-la¼1s.gen ‘It is female teachers that I like (better).’ felac i, nga’ay ma-lalo b. O sa-pi-senger t-o cn ia-nfin-soak dat-cn rice top good.for uv-soft o sanay. cn like.this ‘Regarding the reason to soak the rice, it is for softening (it).’

Both of the constructions contain a displaced element that appears before the main predicate, but there is no pause before the element in the displacement construction and the rest of the clause while there is one in the topic construction, which means that the displaced elements in (3a) and (3b) are placed under the pre-core slot (PrCS) and pre-detached position (PrDP) respectively.

26.3

Lexical Categories

The discussion of lexical categories includes the roots and derived words in Amis. As pointed out by Wang (1976), all the root forms in Amis are syntactically nominal.5 In other words, to serve as a verb, root forms, even for roots with inherent verbal meaning, have to be either affixed with a verbal prefix (usually a voice marker6) or placed in the predicate position. The former strategy is more commonly found and examples can be seen in (1); the latter is only found with some state predicates such as miming ‘small’ and tosa ‘two’. Derived verbs with no verbal affixes are termed unaffixed verbs in Wu (2006a). As root forms are syntactically nominal, they can readily appear after a case marker or a preposition (e.g. (4a)) to manifest arguments and adjuncts. However, the derived verbs (e.g. root form plus a voice marker) have to be nominalized7 before they can show up in a nominal position (e.g. (4b–c))8. See Wu (2006a: 70) for some common nominalizing or deverbal strategies in Amis. (4)

Amis nouns and deverbal nouns rakat! a. Na’on-en k-o mind-uv nom-cn walk ‘Good-bye.’ Lit. ‘Mind your walk!’ rakat! b. *Na’on-en k-o mind-uv nom-cn walk rakat-ay! c. Na’on-en k-o mind-uv nom-cn walk-fac ‘Mind the one who is walking!’

The manifestation of predicates was illustrated in Section 26.2; a verbal predicate is expressed by a derived verb (e.g. (1)), a nominal predicate is composed of a noun (derived or base-generated) preceded by a noun class

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

marker (e.g. (2a)), and a prepositional predicate is designated by a preposition plus a noun (e.g. (2b)).

26.4

Macrorole Selection and Argument-Marking

This section discusses the macrorole selection of Amis predicates with different numbers of core arguments and how arguments and nonarguments are marked.

26.4.1

Macrorole Selection of Verbs with Various Numbers of Core Arguments Amis predicates with zero core arguments are found in meteorological or phenomenal verbs such as ma-’orad ‘rain’, and sienaw ‘cold (in terms of weather)’. These verbs can appear by themselves without any co-occurring argument, as illustrated in (5a). As there is no core argument, there is no macrorole for such verbs, and thus, they are macrorole-atransitive (M-atransitive). However, such verbs can also take one core argument, as illustrated in (5b) and become M-intransitive. Their different logical structures are given in (5a′) and (5b′) respectively. (5)

Amis meteorological or phenomenal verbs a. Ma-’orad anini. neut-rain now ‘It is raining today.’ a′. rain′ (Ø) b. Ma-’orad k-o kakarayan. neut-rain nom-cn sky ‘The sky is raining.’ b′. rain′ (kakarayan)

Regarding the assignment of the macrorole for verbs with one core argument (i.e. syntactically intransitive or S-intransitive verbs), the presence or absence of do′ in their logical structure (LS) makes a crucial difference; the presence of do′ makes the only core argument an actor while the absence of do′ makes it an undergoer. In other words, the single argument for intransitive activity verbs such as tireng ‘stand’ and intransitive state verbs such as ma-toni’ ‘soft’ will not be assigned the same macrorole: (6)

Amis verbs with one core argument a. Tireng cingra. body 3s.nom ‘He is standing.’ b. Ma-toni’ k-o-ni a titi. neut-soft nom-cn-this lnk meat ‘This meat is soft.’

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

957

958

JOY J. WU

Although the only argument of the two verbs in (6) ia marked by the same case (i.e. the nominative case), the macrorole of tireng would be actor while that of ma-toni’ would be undergoer. This is where the Role and Reference Grammar (RRG) analysis differs from other studies concerning this issue in Amis (e.g. Liu 1999, Liu 2003, to name just a few). In such research, the single arguments of S-intransitive verbs seem to be treated as the same type and the verbs are labelled as actor-focus or actor-voice verbs. This analysis will be further discussed in the last section of this chapter. When there are two core arguments in the LS of an Amis predicate, the macrorole assignment follows the Actor-Undergoer Hierarchy (AUH) and the default macrorole assignment principles. However, it is also possible for such verbs to have only one macrorole. As illustrated in (1b) and (1c), twoplace verbs can appear with two case-marking patterns: the nominativedative case frame in the actor voice (AV) sentences and the genitivenominative pattern in the undergoer voice (UV) sentences. While the twoplace UV verb follows the AUH and has both an actor and an undergoer, the second argument in their AV counterparts is a non-macrorole core argument, as argued in Wu (2006a). In other words, the two-place AV verbs are Mintransitive. The non-macrorole status of the second argument results either from the linking phase from the argument position in the LS to macrorole or from the voice operation that deprives a core argument of its macrorole status during linking from macrorole to syntactic function. The first case is found when the second argument of an activity verb is non-referential, as exemplified in (7a). The second one is illustrated in (7b). (7)

Amis AV sentences with a two-place predicate a. Mi-nanom cingra (t-o nanom). av-water 3s.nom dat-cn water ‘He is drinking water.’ Or ‘He is going to drink water.’ b. Mi-nanom cingra t-o-ra sayta. av-water 3s.nom dat-cn-that soda ‘He is drinking that soda.’ Or ‘He is going to drink that soda.’

The principal difference between a macrorole second argument (normally a patient) and a non-macrorole one lies in the possibility for this argument to be promoted by the applicative construction; only a non-macrorole argument (or an adjunct) is eligible to appear in such constructions. See Wu (2006a) for the discussion in detail. In addition to the two-place AV verbs, examples illustrating the mismatch between syntactic transitivity and macrorole transitivity is also seen in threeplace predicates. To begin with, the AV construction of a three-place predicate is also deemed M-intransitive. Both of its non-actor arguments are marked by the dative case and, most important of all, both of them can be the promoted argument in the applicative construction, which is a feature for a nonmacrorole argument in the AV construction. Examples are given in (8).

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis (8)

Amis three-place AV predicate and the applicative constructions a. Pa-nanom cingra ci Aki-an t-o-ra sayta. caus-water 3s.nom ppn Aki-dat dat-cn-that soda ‘He gave Aki that soda (to drink).’ (Causative, AV) b. Cima k-o pa-nanom-an nira t-o-ra sayta? who.nom nom-cn caus-water-la 3s.gen dat-cn-that soda ‘Who did he give that soda to drink?’ (causee/recipient as the undergoer) c. O maan k-o pa-nanom-an nira ci Aki-an? cn what nom-cn caus-water-la 3s.gen ppn Aki-dat ‘What did he give to Aki to drink?’ (causand/theme as the undergoer)

As shown in (8), both the recipient argument and the theme argument can be the PSA,9 in particular, the syntactic pivot, of the applicative three-place verb pa-nanom-an ‘give water’. This is a property for a non-macrorole (NMR) argument. In other words, neither of the two non-actor arguments in AV pasentences such as (8a) are macroroles. While the UV three-place predicates are M-transitive, just like the UV twoplace UV predicates, the mismatch between S-transitivity and M-transitivity is still found as there can be at most two macroroles in RRG, while there are three core arguments in a three-place predicate. This means there is competition for macrorole status between the two groups of potential undergoer participants, namely, theme/patient and recipient/beneficiary/source/goal. Hence, two principles of undergoer selection (i.e. Principle A: choosing the lowest-ranking argument in LS and Principle B: choosing the second-highest-ranking argument in LS) have been proposed in RRG under the AUH. As reported in Guerrero Valenzuela and Van Valin (2004), languages tend to exhibit a mixed type that needs both principles to account for their undergoer selection. Amis is also such a language. Wu (2006a, b) has shown that Amis three-place predicates vary in their selection of the undergoer, though Principle B seems to be more commonly employed. The evidence lies in the possibility of having either the theme/patient or the recipient/beneficiary/source/goal argument as the PSA in the UV constructions, and the latter group seems to be favoured by more threeplace predicates.10 Some examples are given below. As one can see, verbs like pa-caliw ‘lend’ can only have the theme argument as the PSA in their UV construction, while verbs like pa-nanom ‘give water’ only allow the recipient argument as the PSA in their UV counterpart. (9)

Some UV three-place predicates a. Ma-pa-cakay n-i Aki k-o foting ci Ofad-an. uv-caus-buy gen-ppn Aki nom-cn fish ppn Ofad-dat ‘Aki sold (other people’s) fish to Ofad.’ a′. [do′ (Aki, Ø)] CAUSE [[do′ (Ofad, Ø)] CAUSE [BECOME NOT have′ (Aki, foting) & BECOME have′ (Ofad, foting)]] b. *Ma-pa-cakay n-i Aki t-o foting Ø-ci Ofad. uv-caus-buy gen-ppn Aki dat-cn fish nom-ppn Ofad

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

959

960

JOY J. WU

c. Ma-pa-nanom¼to n-i ina t-o sayta Ø-ci uv-caus-water¼asp gen-ppn mother dat-cn soda nom-ppn mama. father ‘Mother gave Father soda to drink.’ c′. [do′ (ina, Ø)] CAUSE [BECOME have.water′ (mama, sayta)]11 d. *Ma-pa-nanom¼to n-i ina ci mama-an k-o uv-caus-water¼asp gen-ppn mother ppn father-dat nom-cn sayta. soda

After briefly reviewing how the macroroles are selected for verbs with various numbers of core arguments, now let us see how these arguments and non-arguments are marked in Amis.

26.4.2 Argument-Marking Arguments and non-arguments are marked in Amis through a tri-case system (nominative, genitive, dative) and the preposition i. The nominative case marks the PSA, or the so-called grammatical subject. The genitive case can mark a possessor or an actor in a non-AV sentence (e.g. (1c)), which makes it the equivalent marker of an ergative case. The dative case serves a wide range of functions. It can label a NMR core argument, as seen in the above examples (e.g. (1b)). It can also appear before an oblique core argument or an adjunct. Examples follow.12 (10)

Amis dative case marking an oblique core argument or an adjunct a. Ma-pa-nanom¼to n-o wawa k-o kolong uv-caus-water¼asp gen-cn child nom-cn water.buffalo t-o-ya nanom. dat-cn-that water ‘The child has already fed the water buffalo that water.’ (Theme) n-o wawa k-o a′. Ya nanomi ma-pa-nanom¼to that water uv-caus-water¼asp gen-cn child nom-cn kolong ____i. water.buffalo ‘That water the child has already fed the water buffalo.’ b. Ma-ota’ kako t-o sanek n-o tosiya. neut-vomit 1s.nom dat-cn smell gen-cn car ‘I feel sick from the smell of cars.’ (Reason) c. Ma-tayal kako t-o romi’ami’ad. neut-work 1s.nom dat-cn every.day ‘I work every day.’ (Time) d. Cenger-en¼ako k-o kiladom t-o kohting-ay. color-uv¼1s.gen nom-cn cloth dat-cn black-fac ‘I am going to colour the cloth with the black colour.’ (Instrument)

The dative case marks an oblique core argument in (10a) and various kinds of adjuncts in (10b–d). The oblique status of the theme argument in (10a) is

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

indicated by its displaced version in (10a′), in which it is simply displaced to the sentence-initial position without affecting the rest of the sentence. This is the feature of displacing an oblique reference phrase (RP) in Amis; the displacement of a direct core RP will result in a nominal structure like an equational sentence, as will be seen in the later discussion. An Amis RP that appears clauseinitially, be it a displaced one or not, does not bear any case marking. That is why the demonstrative pronoun before nanom ‘water’ has no case in (10a′). Although both NMR core arguments and adjuncts can be marked by the dative case, there is a crucial difference between them, which lies in the mechanisms to promote their status to PSA. The core arguments can be promoted via plain undergoer voice constructions and applicative constructions: (11)

The strategies for promoting an NMR core argument a. Ma-nanom n-i Aki k-o-ra sayta. uv-water gen-ppn Aki nom-cn-that soda ‘Aki drank that soda.’ (Plain UV) b. Mi-nanom-an n-i Aki k-o sayta. la-water-la gen-ppn Aki nom-cn soda ‘Aki drank the soda.’ (Locative applicative UV) ‘What Aki drank is the soda.’ (Locative applicative UV) c. Olah-en namo ∅-ci Panay. like-uv 2p.gen nom-ppn Panay ‘You have to love Panay.’ (Plain UV) d. Ka-olah-an¼ako ∅-ci Panay. nfin-like-la¼1s.gen nom-ppn Panay ‘Panay is the one I like (most).’ (Locative applicative UV)

The data in (11) illustrates the possibilities for enhancing the status of the second argument of pred′ (the one marked by the dative case) in mi-nanom and ma-olah to become a PSA (i.e. undergoer of a UV verb). Both the plain UV constructions (e.g. (11a) and (11c)) and the applicative UV constructions (e.g. (11b) and (11d)) are applicable here. Note that the number of core arguments in the two predicates has remained the same in the plain UV constructions and the applicative UV constructions.13 Now consider a different case exemplified in (12). (12)

The strategies for promoting an adjunct a. Ma-patay k-o ’oner t-o sapaiyo n-o ’edo. neut-dead nom-cn snake dat-cn medicine gen-cn mouse ‘Snakes may die from the poison for killing mice.’ a′. (BECOME) dead′ (oner) b. Sa-pi-patay n-o mato’asay t-o ’oner k-o ia-nfin-dead gen-cn old.man dat-cn snake nom-cn sapaiyo n-o ’edo. medicine gen-cn mouse ‘The old man killed the snake with the poison for killing mice.’ (Instrument applicative, UV)

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

961

962

JOY J. WU

b′. [do′ (mato’asay, [use′ (mato’asay, sapaiyo no ’edo)])] CAUSE [[do′ (sapaiyo no ’edo, Ø) ]CAUSE [BECOME dead′ (’oner)]] c. Ma-ota’ kako t-o sanek n-o tosiya. neut-vomit 1s.nom dat-cn smell gen-cn car. ‘I feel like vomiting from the smell of cars.’ c′. do′ (kako, [vomit′ (kako, (y))]) d. Sa-ka-ota’ ako k-o sanek n-o tosiya. ia-nfin-vomit 1s.gen nom-cn smell gen-cn car ‘The smell of the car is the reason why I vomit.’ (Instrumental applicative, UV) d′. because.of′ (sanek no tosiya, [do′ (kako, [vomit′ (kako, (y))])

As shown in (12), for a reason/indirect cause adjunct RP marked by the dative case to become a PSA, only the applicative construction can be used. Note that the number of core arguments will change when the applicative constructions are employed, as can be seen from the comparison of the number of arguments in the LS of the non-applicative verb (e.g. (12c′)) and the applicative one (e.g. (12d′)). Another way to make the adjunct in (12a) and (12c) a PSA, in addition to using the applicative UV construction, is to make the adjunct the actor of an AV construction, as illustrated in (13): (13)

The strategies for promoting an adjunct a. Mi-patay k-o sapaiyo n-o ’edo t-o ’oner. av-dead nom-cn medicine gen-cn mouse dat-cn snake ‘The poison for killing mice may kill a snake as well.’ a′. [do′ (sapaiyo no ’edo, Ø)] CAUSE [BECOME dead′ (’oner)] b. Mi-ota’ t-o tamdaw k-o sanek n-o tosiya av-vomit dat-cn person nom-cn smell gen-cn car ‘The smell of cars makes people vomit.’ b′. [do′ (sanek no tosiya, Ø)] CAUSE [BECOME vomit′ (tamdaw)]

The reason/indirect cause adjunct RPs in (12a) and (12c) now become the actors in (13a) and (13b), respectively. As indicated in the logical structures of mi-patay in (13a) and mi-ota’ in (13b), the predicates have become causativized, and there is an effector added to the core of the predicates.14 In other words, the number of core arguments has also been changed. The addition of a core argument is not found in the examples in (11); when the to RPs of mi-nanom and ma-olah become PSAs in the plain or applicative UV constructions, there is no argument addition involved. Hence, the to RP of ma-patay in (12a) and ma-ota’ in (12c) should be analysed differently from the to RPs in mi-nanom in and ma-olah; the former are adjuncts while the latter are NMR core arguments. The last difference between an NMR direct core argument and an adjunct (or even an oblique core argument) is that the status of the former can always be adjusted through the plain voice operation; however, for the latter, it is not always possible. In other words, some to RPs can only be promoted by means of the applicative constructions. For example, the plain voice construction is quite unlikely to be employed to promote the adjunct manifesting temporal expression in (10c), although the

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

applicative form ka-tayal-an ‘place or time for working’ can be used. For some adjuncts that are more likely to be construed as effectors (e.g. an indirect cause like sanek no tosiya ‘smell of the car’ in (12c), they may be promoted to become an actor in AV and UV constructions, as we have seen in (13). Based on the discussion so far, the case assignment rules in Amis can be summarized as (14):15 (14)

Case assignment rules in Amis a. Assign nominative case to the lowest macrorole argument in terms of the PSA Selection Hierarchy. b. Assign genitive case to the other macrorole argument. c. Assign dative case to other direct core argument(s).

The examples in (15) illustrate how the rules in (14) are applied in Amis. (15)

The application of case assignment rules in Amis a. Ma-olah kako ci Panay-an av-like 1s.nom ppn Panay-dat ‘I like Panay,’ a′. like′ (kako, Panay) (Rule(s) applied: (14a) and (14c)) b. Ma-ka-olah¼ako Ø-ci Panay. uv-nfin-like¼1s.gen nom-ppn Panay ‘I love Panay (secretly).’ or ‘Panay was loved by me.’ b′. like′ (ako, Panay). . ..BECOME like′ (ako, Panay) (Rule(s) applied: (14a) and (14b)) c. Ma-stol kako t-o fekeroh. neut-stumble 1s.nom dat-cn rock ‘I stumbled over on the rock.’ c′. stumble′ (kako) (Rule(s) applied: (14a)) d. Ma-stol n-o fekeroh kako. uv-stumble gen-cn rock 1s.nom ‘The rock rolled to me and made me stumble.’ d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)] (Rule(s) applied: (14a) and (14b)) e. Ma-rohem¼to k-o-ra pawli. neut-ripe¼asp nom-cn-that banana ‘The banana has become ripe.’ e′. (INGR/BECOME) ripe′ (pawli) (Rule(s) applied: (14a)) f. Pa-si-fana’ k-o singsi t-o wawa t-o caus-have-knowledge nom-cn teacher dat-cn child dat-cn n-o Amis. gen-cn Amis ‘The teacher is going to teach the children Amis.’ (Causative, AV) f′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)] (Rule(s) applied: (14a) and (14c))

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

963

964

JOY J. WU

In addition to a set of case markers, there is also a preposition i in Amis. This preposition mainly marks arguments with a locative feature (i.e. x in be-loc′ (x, y) or pred-loc′ (x, y)). In addition, it also marks the first argument of the existential verbs ira or awa (i.e. (NOT) exist′ ([pred′ (x, y)])) and possibly the first argument in the embedded logical structure BECOME/INGR pred′ (y, z). Some examples are given in (16): (16)

The functions of the preposition i a. Maro’ kako i Taypak. live 1s.nom prep Taipei ‘I live in Taipei.’ (Neutral voice) a′. live.in′ (Taypak, kako) b. Ira k-o kawas i loma’ nira. exist nom-cn ghost prep house 3s.gen ‘There is ghost in his house.’ (Neutral voice) b′. exist′ ([be-in′ (loma’ nira, kawas)]) c. Pa-nengneng kako t-o-ni~ni t-o/i wawa. caus-see 1s.nom dat-cn-this~red dat-cn /prep child ‘I showed the child this.’ Or, ‘I showed this to the child.’ (Causative, AV) c′. [do′ (kako, Ø)] CAUSE [BECOME see′ (wawa, tonini)]

However, as illustrated in (16), while the first argument of the embedded BECOME/INGR pred′ might be marked in more than one way (e.g. by a dative case or a preposition), the preposition is the only choice for the first argument of pred-loc′ (x, y). Moreover, while the first argument of the embedded BECOME/INGR pred′ can be a possible undergoer and hence a PSA in the UV construction, it is impossible for the first argument of pred-loc′ to be an undergoer, let alone a PSA. This is illustrated by the following contrast between pa-nanom ‘cause to have water’ and pa-teli ‘put’ in (17): (17)

Comparison of the first argument of BECOME/INGR pred′ and pred-loc′ a. Pa-nanom-en k-o sayta t-o nanom! caus-water-uv nom-cn soda dat-cn water ‘Add water to the soda!’ a′. DO (x, [do′ (x, Ø)] CAUSE [BECOME have.water′ (sayta, nanom)] b. Pa-teli kako t-o konga i langa. caus-put 1s.nom dat-cn sweet.potato prep basket ‘I put the sweet potatoes in the basket.’ (Causative, AV) b′. [do′ (kako, Ø)] CAUSE [BECOME be-loc′ (langa, konga)] c. Ma-pa-teli’¼ako k-o konga i langa. uv-caus-put¼1s.gen nom-cn sweet.potato prep basket ‘I put the sweet potato in the basket.’ c′. *Ma-pa-tli’¼ako t-o konga k-o langa. uv-caus-put¼1s.gen dat-cn sweet.potato nom-cn basket

As indicated in (17), the first argument of the embedded be-loc′ (e.g. langa ‘basket’)) cannot be an undergoer in the UV construction. This follows from the claim in RRG that the first argument of be-loc′ or pred-loc′

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

cannot be a macrorole; in other words, two-place locative predicates are always M-intransitive. The following preposition-assignment rules are postulated for Amis: (18)

Preposition-assignment rules for Amis Assign the preposition i to the first argument of . . .pred′ (x, y). . . if it is a nonmacrorole argument: (i) obligatory if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ common noun (ii) optional if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ personal proper noun (iii) optional if pred′ (x, y), pred′ ¼ cognition, possession, and perception

26.5

Grammatical Relations and the Voice System

This section discusses grammatical relations and the voice system in Amis. According to Wu (2006a), Amis only employs a PSA in the formation of relative clauses, a nominal type of displacement constructions and wh-questions. Constructions that might need a grammatical relation in other languages, such as control constructions and reflexivization, can be taken care of by semantic roles in Amis. Hence there are no grammatical relations found in such constructions. As for the voice system, Amis distinguishes two voices: actor and undergoer, and the latter can be further divided into the plain (or non-applicative) UV and the applicative UV. The functions of each voice will be elaborated in the following sub-sections.

26.5.1 Grammatical Relations A relative clause (RC) in Amis is formed by gapping an RP from the modifying clause. The gapped RP is a pivot as it is omitted in the clause. This gapped RP is coreferential with the modified noun that follows the RC. To serve as a head of an RC, its coreferential gapped RP has to be the actor of an AV verb, the undergoer of a plain UV verb, or an applied argument of an applied UV verb in the RC. If the gapped RP does not belong to any of these types, the sentence will be rendered ungrammatical. This is exemplified in (19), in which the gapped RP is indicated by ‘__’ in the RC.16 Hence, there is restricted neutralization of semantic roles on the pivot of an RC in Amis. (19)

Amis relative clauses a. Pivot: Actor of AV verb Ma-patay¼to k-o-ya neut-dead¼asp nom-cn-that Aki-an a wacoi. Aki-dat lnk dog ‘That dog that bit Aki is dead.’

mi-kalat-ay av-bite-fac

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

____i

ci ppn

965

966

JOY J. WU

a′. Pivot: NMR direct core argument of AV verb *Ma-patay¼to k-o-ya mi-kalat-ay k-o waco ____i neut-dead¼asp nom-cn-that av-bite-fac nom-cn dog a tamdawi. lnk person ‘That person that the dog bit is dead.’ b. Pivot: Undergoer of UV verb Tati’ih k-o-ya ma-kaen-ay n-i Aki ____i a bad nom-cn-that uv-eat-fac gen-ppn Aki lnk talii. taro ‘That taro that Aki ate was bad.’ b′. Pivot: (Patient) undergoer of applied UV verb Tati’ih k-o-ya mi-kaen-an n-i Aki ____i a bad nom-cn-that la-eat-la gen-ppn Aki lnk talii. taro ‘That taro that Aki ate was bad.’ c. Pivot: Actor of UV verb *Ma-soso k-o-ya ma-kaen-ay ____i k-o tali neut-fat nom-cn-that uv-eat-fac nom-cn taro a tamdaw. lnk person ‘The person that ate the taro was fat.’ d. Pivot: (Instrument) undergoer of applied UV verb Ma-pitek¼ako k-o sa-pi-cikcik n-i Aki uv-break¼1s.gen nom-cn ia-nfin-cut gen-ppn Aki t-o dateng ____i a po’oti. dat-cn vegetable lnk knife ‘I broke the knife with which Aki cut the vegetable.’ e. Pivot: (Locative) undergoer of applied UV verb Tayra Ø -ci Panay mi-ladom i go nom-ppn Panay neut-fetch.water prep pi-ladom-an n-i Aki ____i a tefoni. nfin-fetch.water-la gen-ppn Aki lnk well ‘Panay went to fetch water at the well where Aki fetched water.’

The same restriction is found in the formation of the displacement construction and wh-questions, both of which involve a displaced nominal element that appears in the clause-initial position. There are two types of such structure. The first type, termed the nominal type, is constructed as an equational sentence in which the displaced RP or the wh-word and the remaining elements of the clause are juxtaposed. This remaining clause is preceded by a nominative case marker, and it is structured like a headless RC. The second type, termed the verbal type, is formed simply by placing an RP or a wh-word at the beginning of the clause. The remaining clause of the verbal type stays structurally unchanged. The wh-word can even appear in situ in the verbal type

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

though it more often appears clause-initially. It is the nominal type that involves a restricted neutralization of semantic roles; that is, its pivot has to be an actor of a gapped AV clause or an undergoer of a gapped UV clause. As for the verbal type, the restricted neutralization is not found. The nominal type is illustrated in (20) while (21), repeated from (10a′), exemplifies the verbal type.17 The displaced RP is underlined and its pivot is indicated by ‘___’ in the clause. (20)

Amis sentences with displaced RPs (the nominal type) a. Pivot: (Patient) undergoer of an applied UV verb O fafahian a singsii k-o ka-olah-an¼ako ___i. cn woman lnk teacher nom-cn nfin-like-la¼1s.gen ‘It is female teachers that I like better.’ b. Pivot: Actor of AV verb Ya wawai k-o mi-pa-nanom-ay ____i t-o that child nom-cn av-caus-water-fac dat-cn kolong. water.buffalo ‘It is that child who fed water to the water buffalos.’ b′. Pivot: Actor of UV verb *Ya wawai k-o ma-pa-nanom-ay ____i k-o that child nom-cn uv-caus-water-fac nom-cn kolong. water.buffalo ‘It is that child who fed water to the water buffalos.’ c. Pivot: (Patient) undergoer of an applied UV verb Ya nanomi k-o mi-pa-nanom-an¼to¼ako ci that water nom-cn la-caus-water-la¼asp¼1s.gen ppn mama-an ____i. father-dat ‘That water is what I gave father to drink.’

(21)

Amis sentences with displaced RPs (the verbal type) Pivot: Oblique core argument of three-place UV verb Ya nanomi ma-pa-nanom¼to n-u wawa that water uv-caus-water¼asp gen-cn child kolong ____i. water.buffalo ‘The child has already fed the water buffalo that water.’

k-o nom-cn

In (20a), the undergoer RP appears in sentence-initial position and there is a gap in the remaining clause that follows the displaced RP. As one can see, there is a case marker ko between the displaced RP and the remaining clause; that is, the clause appears in a nominal position. Furthermore, the verb inside the nominal clause is an applied UV verb. Examining the rest of the examples in (20), we can see that they

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

967

968

JOY J. WU

demonstrate a restricted neutralization of semantic roles, as the pivot in the nominal clause following the displaced element has to be the actor of an AV verb or the undergoer of a UV verb; the latter can be either a plain UV verb or an applied UV verb. The restricted neutralization exemplified in (20) is also observed in the nominal type wh-questions in (22): (22)

Amis wh-questions (the nominal type) a. Pivot: Actor of AV verb Cimai k-o mi-palo-ay ____i t-o wawa? who.nom nom-cn av-beat-fac dat-cn child ‘Who is the one that beat the child?’ a′. Pivot: Actor of UV verb *Cimai k-o ma-palo-ay ____i k-o wawa? who.nom nom-cn uv-beat-fac nom-cn child ‘Who is the one that beat the child? b. Pivot: Undergoer of UV verb O maani k-o ma-kaen-ay n-i Aki ____i? cn what nom-cn uv-eat-fac gen-ppn Aki ‘What is it that Aki ate?’ b′. Pivot: NMR direct core argument of AV verb *O maani k-o kaen-ay Ø-ci Aki ____i? cn what nom-cn eat-fac nom-ppn Aki ‘What did Aki eat?’ c. Pivot: (Patient) undergoer of applied UV verb Cimai k-o ka-olah-an¼iso ____i? who.nom nom-cn nfin-like-la¼2s.gen ‘Who is the one you like?’

The sentences in (22a–a′) exemplify wh-questions concerning an actor of a predicate. As shown in the data, the clause following the interrogative pronoun is preceded by a case marker, which gives the nominal property of the clause. Furthermore, when the interrogative pronoun functions as the actor of the predicate, the verb has to be marked by the AV affix; this pronoun cannot be interpreted as functioning as the actor of a UV verb. When the interrogative pronoun refers to a non-actor in the clause, the verb has to be marked by either the plain UV markers (e.g. (22b)) or the applicative markers (e.g. (22c)). Hence, there is a restricted neutralization of semantic roles.18

26.5.2 Voice Operations and Applicative Constructions Table 26.1 summarizes the voice system in Amis. As shown, Amis distinguishes three voices: neutral, actor and undergoer. Neutral voice is used for marking verbs that are both M-intransitive and S-intransitive as their single macrorole is not specified. The neutral voice will be further explored in Section 26.6. The other two voices will be introduced in more detail here.

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

Table 26.1 Voice markers and the applicative markers in Amis (adapted from Wu 2006b: 289) Voice Neutral voice19 Actor voice Undergoer voice

Plain

Macrorole of the PSA Unspecified Actor Undergoer (unmarked choice)

Applicative

Undergoer (marked choice)

Affixes mi-, ma-, mi-, ma-, ma-, ma-. . ., ma-ka-, -en sa-, -an

Let us begin with the discussion of actor voice. AV verbs always have a nominative-dative case pattern, and they have only one macrorole (i.e. actor). The actor is assigned the nominative case and the NMR argument is marked by the dative case. The voice-marking function of the AV affixes is demonstrated in their co-occurrence with the volitative mood marker -aw, which shows the UV pattern when it attaches to a root form. (23)

-aw volitative construction a. Nanom-aw¼ho¼ako. water-vol¼asp¼1s.gen ‘I will go drink water first.’ (Volitative mood, UV) a′. Mi-nanom-aw¼ho kako. av-water-vol¼asp 1s.nom ‘I will go drink water first.’ (The water is farther away than the one mentioned in (23a).) b. Kaen-aw¼ako k-o dateng. eat-vol¼1s.gen nom-cn vegetable ‘I will try that vegetable.’ (Volitative mood, UV) b′. Kaen-aw k-o wawa t-o sapaiyo. eat-vol nom-cn child dat-cn medicine ‘(I am) afraid that the child will take the medicine.’ c. Olah-aw¼ako kiso? like-vol¼1s.gen 2s.nom ‘May I go to love you?’ (Volitative mood, UV) c′. Ma-olah-aw kako tisonan. av-like-vol 1s.nom 2s.dat ‘I am afraid that I will like you.’

As shown in (23), the suffix -aw manifests an optative reading for the derived verb. Notice that the case-marking pattern for V-aw follows the UV pattern, as the actor is marked by the genitive case. However, when the V-aw forms are affixed with mi-, , and ma-, their case-marking patterns become the AV pattern, as indicated in (23a′), (23b′) and (23c′). This contrast shows the voice-marking function of the AV markers. But what kinds of functions do the AV constructions perform? Clearly, the AV construction has a PSA-modulation function, as it makes a marked choice of PSA in terms of the PSA Selection Hierarchy. Given the fact

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

969

970

JOY J. WU

that Amis displays ergative features in at least the case-marking system and some constructions that involve a PSA such as the relative clause and the nominal type of wh-question, one would expect the lowest-ranking argument to be the unmarked PSA choice. However, in the AV construction, it is the highest-ranking direct core argument that is chosen to be the PSA. What about the argument modulation function? For a two-place predicate, the lowest-ranking direct core argument in the AV sentences should be assigned an undergoer based on the macrorole assignment principles, as such verbs can take at most two macroroles. However, this argument in the AV construction does not surface as a macrorole syntactically, as revealed by its case marking and the fact that its status can be promoted by the applicative construction. Instead, this argument is realized as an NMR core argument in the AV construction. In other words, the lowest-ranking argument of a two-place predicate has been stripped of its macrorole status by the AV operation. However, it is still in the core, as indicated by its behavioural property in serving as a semantic controller in the persuade-type control construction, as discussed in Wu (2006a). Hence, AV constructions also perform an argument-modulation function. This function is even more salient for three-place predicates, as a possible undergoer can also be marked by the preposition in the AV construction in addition to the dative case. Consider the following examples: (24)

Amis three-place predicates a. Pa-caliw Ø-ci Kacaw t-o singsi t-o caus-borrow nom-ppn Kacaw dat-cn teacher dat-cn payso. money ‘Kacaw lent the teacher money.’ (Causative, AV) b. Pa-caliw Ø-ci Kacaw t-o payso i caus-borrow nom-ppn Kacaw dat-cn money prep singsi. teacher ‘Kacaw lent the money to the teacher.’ (Causative, AV) c. Aka pa-caliw-en k-o singsi t-o payso! neg.imp caus-borrow-uv nom-cn teacher dat-cn money ‘Don’t lend the teacher money!’

As shown in (24), the recipient RP singsi can be marked either by the dative case or the preposition in the AV construction. This RP is the second-highestranking argument in the LS of pa-caliw ‘lend’, and it is also a possible undergoer, as indicated in the UV sentence in (24c). The presumed undergoer RP is realized as non-macrorole in the AV construction in (24a), but it is realized as an adjunct in (24c), as the preposition i typically marks a locative RP in the periphery. From the above discussion, we can thus conclude that the actor voice not only modulates the PSA choice but also modulates the semantic status of a core argument by either stripping a macrorole argument of its macrorolehood or realizing a core argument as an oblique element.

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

What about the undergoer voice? As shown in Table 26.1, there are two sets of UV markers: the marked set and the unmarked set. The markers that signify a marked undergoer choice used to be treated as voice markers as well (termed instrumental voice and locative voice) in Amis, but the examples in (25) show that they serve functions other than that of a voice operation. (25)

The applicative UV constructions in Amis a. Aka sa-pi-litek-en k-o-ra caklis neg.imp ia-nfin-chop.tree-uv nom-cn-that axe t-o-ra kilang! dat-cn-that tree ‘Don’t use that axe to chop down the tree!’ n-i Aki t-o takid k-o-ya b. Ma-sa-pi-sanga uv-ia-nfin-make gen-ppn Aki dat-cn bottle nom-cn-that aol. bamboo ‘Aki used that bamboo to make the bottle.’

The sentences in (25) show that when the UV marker -en or ma- and the instrumental applicative marker sa- co-occur in a sentence, only the instrument RP surfaces as the undergoer instead of the patient RP, which would be the default undergoer following the AUH. In other words, the applicative markers indicate a marked undergoer selection. The voice markers and the applicative markers show different operations in the two phases in the RRG linking algorithm. That is to say, the applicative marker affects the linking from argument positions to macroroles, while the voice marker operates at the linking from macroroles to syntactic functions. There are two functions of these applicative markers. First, they may enhance the status of a non-argument such as instrument or location to become a core argument. Second, they can also promote a non-macrorole core argument (e.g. patient in an AV sentence) to become a macrorole. The instrumental applicative construction serves the first function, while the locative applicative construction can perform both functions.20 Based on the discussion so far, the UV pattern should be deemed the default pattern in Amis, which is proven by the fact that it is the unmarked voice of the applicative constructions even when the UV markers do not show up. Although the UV pattern enjoys unmarked status in Amis, there are some predicates that seem to take the AV pattern by default, and for such predicates, UV appears to be the marked pattern. Such predicates can be illustrated by the pa- verbs. Consider the following examples of a pa- verb plus the volitative mood suffix -aw: (26)

pa- verbs suffixed with the volitative marker -aw a. Pa-nanom kako t-o kolong. caus-water 1s.nom dat-cn water.buffalo ‘I feed water buffalos water.’ (Causative, AV)

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

971

972

JOY J. WU

b. Pa-nanom-aw¼ho¼ako k-o kolong. caus-water-vol¼asp¼1s.gen nom-cn water.buffalo ‘I will feed the water buffalo water first.’ (Volitative, UV) c. Mi-pa-nanom-aw¼ho kako t-o kolong av-caus-water-vol¼asp 1s.nom dat-cn water.buffalo ‘I will go to feed water buffalos water first.’

Recall that in the previous discussion, I showed that when a root form is suffixed with -aw, it takes the UV pattern. As we can see in (26a), the papredicates appear with the AV case-marking pattern (i.e. nominative-dative). However, when they are suffixed with -aw, the case-marking pattern becomes the UV pattern. In other words, the pa- predicates behave like a bare root form in the volitative mood construction. When the volitative form pa-nanom-aw is prefixed with mi-, the case pattern becomes the AV pattern again. These examples show that, unlike mi-, pa- does not have a voice-marking function. However, pa- verbs follow the AV pattern by default. To make pa- verbs appear in the UV pattern, the plain UV markers or the applicative forms have to be used. Morphologically, the AV pattern appears to be the default pattern of pa- verbs, while the UV pattern is a marked one. However, syntactically, the UV forms actually turn a marked pattern (i.e. AV) into an unmarked one. This may explain why the UV form pa- . . . -en and the applicative form pa- . . . -an are found much more frequently than the plain pa- forms in Amis.21 As mentioned, AV constructions perform both PSA-modulation and argument-modulation functions. How about UV constructions? Though appearing to be the basic pattern of Amis based on the case marking and the default voice choice of the applicative constructions, they turn out to be the marked voice choice for some predicates that usually appear with the AV pattern by default. For such predicates, their plain UV constructions perform a PSA-modulation function but no argument-modulation function as the macrorole arguments and the NMR core arguments remain unaffected. See Wu (2006a) for more discussion. However, the applicative UV constructions display both functions, as now an adjunct becomes both a macrorole and the PSA. The above discussion shows that both the actor voice and the undergoer voice are deemed basic voice forms.22 Therefore, Amis exhibits a split system in verbal morphology in spite of displaying ergative features in the case-marking system and in some grammatical constructions.

26.6

Special Discussion: An RRG Account for One-Place Predicates in Amis

One-place predicates in Amis usually appear in an unaffixed manner (e.g. tayra ‘go’ and miming ‘small’), or with affixes formally identical with the AV markers such as and ma- (e.g. tangic ‘cry’ and ma-lalok

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

‘diligent’). In the RRG analysis proposed here, the voice marker that appears on one-place predicates is treated as neutral voice as the only macrorole of such verbs is not specified. This analysis is rather different from the previous studies, which seem to assume a unified semantic role for the single argument of intransitive predicates. For example, they either label intransitive predicates as AV verbs (e.g. Liu 1999) or claim that the single argument of intransitive verbs is patient (e.g. Chen 1987). In fact, treating one-place predicates as AV or AF (i.e. agent-focus) verbs has been a fairly common practice in the studies of other Formosan languages as well (e.g. Atayal, as seen in Huang (2000) and Yeh (2015), and Tsou in Zeitoun (1993), to name just a few). However, this unified-semantic-role approach is implausible if we consider the rather different morphosyntactic behaviours of the single argument of the one-place predicates in Amis. In the following, I am going to show how the RRG non-unified macrorole approach for one-place predicates (Van Valin and LaPolla 1997; Van Valin 2005) can account for their peculiar morphosyntactic behaviours in two constructions: the UV -en verbs and the evaluative construction.

26.6.1 The -en Form of One-Place Predicates The first piece of evidence that proves the different macrorole assignments of the sole argument of the one-place predicates comes from their corresponding UV -en form, in which the actor gets the genitive case while the undergoer takes the nominative case. Consider the following examples: (27)

Amis verbs with one core argument a. Tireng cingra. stand 3s.nom ‘He is standing.’ pa-kimad, ta paka-nengneng a′. Tireng-en¼ako stand-uv¼1s.gen caus-speech so.that ablt-see kamo. 2p.nom ‘I will stand up when making a speech so that you can see (me) clearly.’ b. Ma-toni’ k-o-ni a titi. neut-soft nom-cn-this lnk meat ‘This meat is soft.’ k-o-ni a titi. b′. Toni’-en¼ako soft-uv¼1s.gen nom-cn-this lnk meat ‘I will soften this meat.’

Both tireng ‘stand’ and ma-toni’ ‘soft’ are traditionally labelled as AV verbs. When they are suffixed with -en, the only argument in tireng (now tireng-en) is marked by the genitive case. However, the single argument in ma-toni’ (e.g. koni a titi) is still marked by the nominative case in the UV construction toni’-en. The factor affecting the case-marking pattern is the

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

973

974

JOY J. WU

different macroroles assigned to the only arguments of tireng ‘stand’ and ma-toni’ ‘soft’. As the LS of tireng is do′ (x, [stand′ (x)], the xargument will be an actor. But, as there is no do′ in the LS of ma-toni’ (i.e. (BECOME/INGR) soft′ (x)), the x-argument is an undergoer. When the verb is affixed by -en, the agentive UV marker, the actor in tireng-en is marked by the genitive case by default,23 while the undergoer in toni’-en receives the nominative case in this UV -en construction. The above examples indicate the inadequacy of labelling both of the two verbs as AV or AF verbs. Instead, the RRG approach can account for the distinctive nature of the semantic roles played by the single arguments of the two verbs. However, one may run into a problem upon the application of the rules in (14), repeated as (28) below, for intransitive verbs suffixed with the UV marker -en. (28)

Case assignment rules in Amis a. Assign nominative case to the lowest macrorole argument in terms of the PSA Selection Hierarchy. b. Assign genitive case to the other macrorole argument. c. Assign dative case to other direct core argument(s).

As shown in tiring-en in (27a′), the single argument of an -en intransitive verb is always marked by the genitive case. Applying rule (28a) to an -en intransitive verb will yield the wrong case assignment. Therefore, another set of case assignment rules for verbs marked by -en has to be postulated. These rules are stated in (29): (29)

Case assignment rules for verbs marked by -en a. Assign genitive case to the highest-ranking macrorole in terms of the PSA Selection Hierarchy. b. Assign nominative case to the other macrorole argument. c. Assign dative case to other direct core argument(s).

For two-place or three-place -en verbs, all three of the rules in (29) are applicable. But for the one-place -en verbs, only (29a) and (29c) will apply, as there is only one macrorole in such verbs. The examples in (30) illustrate how the rules in (28) and (29) work in Amis. (30)

The application of the revised case assignment rules in Amis a. Ma-olah kako ci Panay-an. av-like 1s.nom ppn Panay-dat ‘I like Panay,’ a′. like′ (kako, Panay) (Rule(s) applied: (28a) and (28c)) b. Ma-ka-olah¼ako Ø-ci Panay. uv-nfin-like¼1s.gen nom-ppn Panay ‘I love Panay (secretly).’ Or, ‘Panay was loved by me.’

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis b′. like′ (ako, Panay). . ..BECOME like′ (ako, Panay) (Rule(s) applied: (28a) and (28b)) c. Ma-stol kako t-o fekeroh. neut-stumble 1s.nom dat-cn rock ‘I stumbled over on the rock.’ c′. stumble′ (kako) (Rule(s) applied: (28a)) d. Ma-stol n-o fekeroh kako. uv-stumble gen-cn rock 1s.nom ‘The rock rolled to me and made me stumble.’ d′. [do′ (fekeroh, Ø)] CAUSE [BECOME stumble′ (kako)] (Rule(s) applied: (28a) and (28b)) e. Ma-rohem¼to k-o-ra pawli. neut-ripe¼asp nom-cn-that banana ‘That banana has become ripe.’ e′. (INGR/BECOME) ripe′ (pawli) (Rule(s) applied: (28a)) f. Rakat-en¼ako. walk-uv¼1s.gen ‘I will walk (to do something.)’ f ′. DO (ako, [walk′ (ako)]) (Rule(s) applied: (29a)) g. Rakat-en¼ako k-o-ni a kayakay. walk- uv¼1s.gen nom-cn-this lnk bridge ‘I will walk past this bridge.’ g′. DO (ako, [walk′ (ako, kayakay)]) & [BECOME walked′ (kayakay)] (Rule(s) applied: (29a) and (29b)) h. Pa-si-fana’ k-o singsi t-o wawa t-o caus-have-knowledge nom-cn teacher dat-cn child dat-cn n-o ’Amis. gen-cn Amis ‘The teacher is going to teach the children Amis.’ (Causative, AV) h′. [do′ (singsi, Ø)] CAUSE [BECOME have.knowledge′ (wawa, no ’Amis)] (Rule(s) applied: (28a) and (28c))

26.6.2 The Evaluative Constructions RRG’s non-unified treatment of the semantic role played by the single argument of the one-place predicates can not only account for the different case-marking patterns of the one-place -en verbs discussed in the previous section but can also offer a better explanation for the evaluative construction, which can render two types of English sentences such as ‘The food tastes good’ or ‘It is difficult to do this job’. Amis examples for the two types are given in (31) and (32) respectively. (31)

Amis evaluative constructions: Type I nengneng-en a. Tada-fangcal a very-good lnk see-uv ‘That picture looks really beautiful.’

k-o-ra nom-cn-that

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

coka. picture

975

976

JOY J. WU

b. ’angerer-ay a kaen-en bitter-fac lnk eat-uv ‘Bitter gourds taste bitter.’ (32)

k-o nom-cn

karokot. bitter.gourd

Amis evaluative constructions: Type II a. Koesit a sowal-en ciira. difficult lnk say-uv 3s.nom ‘It is difficult to talk to him.’ b. Tati’ih a fo’enot-en k-o fanoh n-o howak. bad lnk pull.out-uv nom-cn feather gen-cn duck ‘It is difficult to pull out the feathers of ducks.’ c. Sa-koesit sa a tedal-en k-o-ni saso’ot. so-difficult like.this lnk untie-uv nom-cn-this knot ‘It is so difficult to untie this knot.’

The Amis sentences in (31) and (32) all begin with a state predicate (i.e. the evaluative predicate), which is then followed by the linker a and a UV verb. The difference between the two types of evaluative construction is that the nominative argument is a core argument shared by both predicates in (31) while the one in (32) is an argument of the second verb but not the first one. Here we will limit our discussion to the first type. A similar construction has been found in other Formosan languages such as Seediq (Tsukida 2005) and Atayal (Yeh 2015). Atayal examples are given in (33) (Yeh 2015: 132, original gloss). (33)

Atayal evaluative constructions24 blaq niq-un qu’ mami’ good[av] eat-uv nom rice ‘This (type) of rice tastes good.’

qa’. dem

Yeh analyses the first predicate of such constructions as an AV verb in spite of the fact that it is a state predicate, and she argues that the construction is a fusion of a serial verb construction (SVC) and a commentative complement clause construction (commentative CCC) (Yeh 2015: 132). Her analysis is summarized in Table 26.2 (cited from Yeh 2015: 148). Table 26.2 The relationship between the juxtaposed verbs in the BLAQ evaluative construction, SVCs and commentative CCCs in Squlip Atayal: a comparison (adapted from Yeh 2015: 148) Construction type parameter

SVCs

BLAQ evaluative construction

Commentative CCCs

(A) Manifestation of voice in verb sequences

(a) AV UV

AV UV

(a’) AV AV

(B) Sharing of arguments (C) Sharing of TAM (tense-aspectmodality) information (D) Sharing of polarity value

Obligatory Obligatory

Obligatory Obligatory

Optional Depends

Obligatory

Obligatory

Optional

(b) UV AV

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

(b’) AV UV

Case and Voice in Amis

According to the table, the Atayal evaluative construction shares more features with SVC, but it is argued to also show the feature of the complement clause in its AV–UV sequence. Amis examples analogous to the Atayal SVCs discussed in Yeh (2015) are provided in (34). Similar to those Atayal examples, the second verb in such sentences has to be an AV form or an intransitive verb with an actor (e.g. (34a), but there is no restriction on the voice of the first verb. (34)

Amis SVC examples a. Ma-herek¼to kako a ma-lafi. av-finish¼asp 1s.nom lnk neut-dinner ‘I finished eating dinner.’ k-o fafahian a mi-fihon. b. Ma-rara neut-slow nom-cn woman lnk av-put.on.makeup ‘The woman puts on make-up slowly.’ pa-rakat k-o-ra tosiya. c. Tanam-en¼ako a try-uv¼1s.gen lnk caus-walk nom-cn-that car. ‘I will try to drive that car.’

Wu (2006a) mentions that these constructions all have an actor pivot, either actor of an intransitive verb or that of an AV verb, in the linked core. Like the Atayal examples discussed in Table 26.2, the evaluative construction in Amis also shares the following two features with SVCs: sharing of TAM information and sharing of polarity value. First, in the evaluative construction as well as in the SVC, only the first verb can be marked with TAM information. (35)

Sharing of TAM information in SVCs and the evaluative constructions in Amis a. Mi-sawad¼to kako a mi-mali av-quit¼asp 1s.nom lnk av-ball ‘I have quit playing ball.’ kako a mi-mali¼to b. *mi-sawad¼to av-quit¼asp 1s.nom lnk av-ball¼ asp nengneng-en k-o-ra coka. c. Fangcal¼to a good¼asp lnk see-uv nom-cn-that picture ‘That picture looks good now.’ nengneng-en¼to k-o-ra coka. d. *Fangcal¼to a good¼asp lnk see-uv¼asp nom-cn-that picture

Second, only the first verb in the named constructions can be conjugated with the non-finite markers when following the negator caay ‘not’.25 (36)

The negative sentences of SVCs and the evaluative constructions in Amis a. Caay ka-safon-en nira k-o neg nfin-soap-uv 3s.gen nom-cn ‘He didn’t use soap to wash (his) body.’

tireng body

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

a lnk

mi-ngingoy av-shower

977

978

JOY J. WU

b. *Caay ka-safon-en nira k-o tireng a pi-ngingoy neg nfin-soap-uv 3s.gen nom-cn body lnk nfin-shower nengneng-en kora coka. c. Caay ka-fangcal a neg nfin-good lnk see-uv nom-cn-that picture ‘That picture does not look good.’ ka-nengneng-en kora coka. d. *Caay ka-fangcal a neg nfin-good lnk nfin-see-uv nom-cn-that picture

However, unlike the SVCs in (34), it is the second verb in the evaluative construction that controls the case-marking pattern of the sentence, but not the first verb. As shown in (34c), even though the second verb appears in the AV form, the arguments in the linked core are case-marked with the UV pattern, following the first verb. In other words, the second verb in the SVCs has no voice function. Nevertheless, the arguments of the linked core in the Amis evaluative constructions always follow the casemarking pattern of the UV verb, as seen in the comparison between (37a) and (37b). (37)

The case-marking pattern of the Amis evaluative construction coka. a. Tada-fangcal a nengneng-en¼ako k-o-ra very-good lnk see-uv¼1s.gen nom-cn-that picture ‘I saw the picture and thought that it was very beautiful.’ a nengneng-en k-o-ra coka. b. *Tada-fangcal kako very-good 1s.nom lnk see-uv nom-cn-that picture coka a nengneng-en. c. Tada-fangcal k-o-ra very-good nom-cn-that picture lnk see-uv ‘That picture looks really beautiful.’

Yeh (2015) in particular explores the answer to the question of why the second verb in the evaluative construction has to be marked by the UV form, but not the AV. She proposes that it is the undergoer that is evaluated in the construction, and this pragmatic function accounts for the UV marking of the second verb. However, the RRG approach might offer an even more straightforward answer. Following the RRG analysis, the first verb in (31) will have an undergoer instead of an actor, and the voice sequence in the Amis evaluative construction should be a neutral voice followed by a UV verb. The shared arguments in such sentences play the role of undergoer (undergoer of the intransitive verb and undergoer of a UV verb). Hence, it is quite reasonable to maintain the syntactic status of the undergoer in both cores. If the second verb takes the actor voice, the macrorole status of the undergoer will be removed, as we have seen in the function of the voice discussed earlier. The seeming paradox in Yeh’s study lies in her treatment of the only argument of the first verb as an actor or a unified thematic role, but the RRG perspective can provide an account that resolves the paradox.

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

As for the SVC examples, the shared argument might play different semantic roles in the two linked verbs, but this role is always an actor in the second verb. This may explain why the second verb bears AV marking though the AV does not have any voice function.

References Blust, Robert. 1999. Subgrouping, circularity and extinction: Some issues in Austronesian comparative linguistics. In Elizabeth Zeitoun and Pau Jen-kuei Li (eds.), Selected Papers from the Eighth International Conference on Austronesian Linguistics, 31–94. Taipei: Institute of Linguistics, Academia Sinica. Bril, Isabelle. 2016. Information structure in Northern Amis: A morphosyntactic analysis. Oceanic Linguistics 55(2): 449–479. Chen, Teresa. 1987. Verbal Constructions and Verbal Classification in Nataoran-Amis (Pacific Linguistics C-85). Canberra: The Australian National University. Guerrero Valenzuela, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of primary object languages. International Journal of American Linguistics 70: 290–319. Huang, Lillian. 2000. Verb classification in Mayrinax Atayal. Oceanic Linguistics 39(2): 364–390. Jolly, Julia. 1993. Preposition assignment in English. In Van Valin (ed.), 275–310. Liu, Dorinda Tsai-hsiu. 1999. Cleft Constructions in Amis. MA thesis, National Taiwan University, Taipei. Liu, Emma En-hsin. 2003. Conjunction and Modification in Amis. MA thesis, National Tsing Hua University, Hsinchu. Starosta, Stanley. 1974. Causative verbs in Formosan languages. Oceanic Linguistics 13: 279–369. Tsuchida, Shigeru. 1988. Amis. In Takashi Kamei, Rokuro Kono and Eiichi Chino (eds.), The Sanseido Encyclopedia of Linguistics, Vol. 1: Languages of the World, Part One, 447–449. Tokyo: Sanseido Press. Tsukida, Naomi. 2005. Seediq. In Alexander Adelaar and Nikolaus P. Himmelmann (eds.), The Austronesian Languages of Asia and Madagascar (Routledge Language Family Series), 291–325. London: Routledge. Van Valin, Robert D., Jr. (ed.). 1993. Advances in Role and Reference Grammar. Amsterdam: John Benjamins. Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press. Wang, Samuel Hsu. 1976. The Syllable Structure of Fataan-Amis. MA thesis, National Taiwan Normal University, Taipei.

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

979

980

JOY J. WU

Wu, Joy Jing-lan. 2003. Clausal modifiers in Amis. Concentric: Studies in English Literature and Linguistics 29(2): 59–81. Wu, Joy Jing-lan. 2006a. Verb Classification, Case Marking, and Grammatical Relations in Amis. PhD dissertation, University at Buffalo (SUNY). Wu, Joy Jing-lan. 2006b. The analysis of pa- verbs in Amis. In Henry Y. Chang, Lillian M. Huang and Dah-an Ho (eds.), Streams Converging into an Ocean: Festschrift in Honor of Professor Paul Jen-kuei Li on his 70th Birthday (Language and Linguistics Monograph Series Number W-5). Taipei: Academia Sinica. Yeh, Maya Yu-ting. 2015. An atypical SVC? A study of BLAQ UV qu’ NP construction in Atayal. In Elizabeth Zeitoun, Stacy F. Teng and Joy J. Wu (eds.), New Advances in Formosan Linguistics, 131–155. Canberra: Asia-Pacific Linguistics. Zeitoun, Elizabeth. 1993. A semantic study of Tsou case markers. Bulletin of the Institute of History and Philology 64(4): 969–989.

Notes 1 www.apc.gov.tw/portal/docList.html?CID¼7AD36169AB07E1D0. 2 The phonetic symbols used in the transcription generally follow the IPA system, with the following exceptions: /e/ stands for schwa [ ], /d/ for voiceless lateral [ɬ], /’/ for epiglottis stop /ʡ/, and /ng/ for /ŋ/. The glosses in general follow Wu (2006a), which adopts some of Liu’s (1999) morphemic analyses (e.g. the separation of noun classifiers and case markers) with some slight modification. 3 There are two classes of nouns identified in Amis: personal proper nouns, including people’s names and kinship terms, and common nouns. The noun class marker usually appears with the case marker and forms a case-marking composite, but the case marker does not show up in predicate position. 4 The analysis of the prefixes ka- and pi- as markers of non-finite forms follows Bril (2016), but Bril specifies pi- as a non-finite marker for actor voice verbs (i.e. NFIN.AV). Here, pi- is simply glossed as NFIN. 5 However, Wu employs an ideophone-forming construction and classifies the root forms into five classes: object, state (attribute), state (transient/ result), activity, achievement/semelfactive (Wu 2006a: 156). 6 As argued in Wu (2006a), the voice markers carry both inflectional and derivational function, and the latter function changes not only the lexical categories of the attached roots or stems but also their semantics. These voice markers also have their own decomposed structures. For example, a UV verb does not merely enhance the status of a patient role as it would in an analogous English passive sentence. It also has a e

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Case and Voice in Amis

7 8

9

10 11

12 13

14

15

16

17

different meaning due to the UV marker ma- and the derived verb might become an (active) accomplishment verb. The factual marker -ay also carries a nominalizing function, which roughly means ‘someone who . . .’ or ‘something which . . .’. Exceptions are found in state predicates with a gradable feature (e.g. malalok ‘diligent’); they can readily appear after a case marker in a comparative construction. See Wu (2003, 2006a) for details. Examples (8b) and (8c) exemplify a nominal type of wh-question . Only the PSA can be the missing argument in the nominal clause. This will be further discussed in Section 26.5 on grammatical relations. For a special type of three-place predicate that is prefixed by pa-pi-, Principle B is the only one that is applied. To simplify the discussion, the LS of ma- (active accomplishment, UV) is not represented in the LS of the ma- UV construction of the threeplace predicates. The semantic role of the noun marked by the dative case is indicated in parentheses after the translation. The differences between the plain UV and the applicative UV constructions are subtle. In general, the -en plain UV construction emphasizes the volition of the actor, and it is more likely to refer to a future event; it is also the form that is used in imperative sentences. The ma- plain UV construction highlights the affectedness of the undergoer. The mi- . . . -an applicative UV stresses the identity of the actor. The to RP in (12a) in fact manifests an external causer for the event described by the predicate, though the predicate is non-causative. The same phenomenon is also found in the to RP in (12c). This explains why they can serve as the argument for the mi- counterparts in (13), which carry a causative reading after derivation. Functionally speaking, the to marker here is similar to the English preposition from, which appears to be causative in its predicative roles (Jolly 1993: 293), as in the sentence John died from Malaria. The causative version of this English sentence would be Malaria killed John. This set of rules cannot work for the verbs suffixed by the UV marker -en. We will therefore propose another set of rules for such verbs in the special discussion of Amis one-place predicates in a later section of the chapter. The RC is boldfaced in the examples. The ‘__’ and the coreferencing subscript in the examples is for expository purposes only. It does not refer to a trace or a null copy in the site of the missing RP. RRG does not posit such notions. Choosing one type over the other crucially depends on the status of the displaced RP. If it is a core argument, then the nominal type is preferred or even required for some speakers; if it is an oblique argument or adjunct, then the verbal type is allowed. As only the

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

981

982

JOY J. WU

18

19

20

21 22

23

24

25

nominal type involves a grammatical relation here, we only give details of this type. Notice that the verbal type is also found in the formation of wh-questions and the criterion for choosing the nominal or the verbal one in forming a wh-question is roughly the same as that for a displacement construction. This voice marker is used for intransitive verbs of which the only macrorole is not specified. Such verbs used to be termed actor voice (or agent voice/focus) in most of the previous studies of Amis and other Formosan languages. An interesting feature of these applicative forms is that they can all be used as nouns designating the argument that is affected by them, and some may even be lexicalized. For example, the sa- applicative form can usually refer to an instrument or a reason, while -an applicative form can designate an object that is acted upon (e.g. mi-tilid-an ‘something written’ > mi-tilid ‘write; study’) or a location (e.g. pi-tilid-an ‘school’). This has been pointed out by Starosta (1974) and my investigation confirms this finding. Based on this proposal, I have maintained the terminology of actor voice and undergoer voice in the discussion, instead of using undergoer voice and antipassive voice, or actor voice and passive voice. The nominative argument of the first part of (27a′) is not expressed. It can either be the theme argument (e.g. koni, a nominative demonstrative pronoun meaning ‘this (talk)’) or the recipient argument (e.g. kamo, a 2nd-person plural nominative pronoun meaning ‘you (all)’) of the second core pakimad ‘give a talk’. Judging from the second part of (27a′), kamo seems to be a more natural choice. The omission of this argument in the first part could be to avoid repetition. However, tirengen cannot take a nominative argument by itself without the second core (i.e. *Tirengen ako koni/kamo. is unacceptable.) One can only say ‘Tirengen ako.’, meaning ‘I want to stand.’ For example, when asked to take a seat, one can answer with ‘Tirengen ako.’ The reader should also be aware that the acceptability of (27a′) varies among dialects in Amis. Yeh named this construction BLAQ UV qu’ NP in her work (Yeh 2015). To avoid the terminological confusion between NP and RP in RRG, this construction will simply be named BLAQ Evaluative Construction in this chapter. Amis verbs change their forms after the declarative negator caay ‘not’ and the imperative negator aka. See Wu (2006a) for detailed discussion.

https://doi.org/10.1017/9781316418086.032 Published online by Cambridge University Press

Index

aboutness condition, 625–626, 628 relation, 466 topic, 462–463, 465, 469 absolutive case, 120, 274, 296, 313, 323, 327, 329, 332, 430, 479, 656 case system. See case marking accessibility, pragmatic, 459 accusative case, 119, 142, 271, 323–324, 328, 331, 335, 430, 510, 654, 660, 682, 871 case system, 332, See also case marking Acehnese, 280, 327, 583, 681, 684 acquisition language, 19, 227, 651, 666–667, 686 Operating Principles, 666 activation, 459–460, 464, 502, 504, 594–595, 864 Activity Hierarchy, 259 actor. See macroroles Actor-Undergoer Hierarchy, 108, 244, 251, 295, 321, 432, 503, 565, 599, 670 adjective, 28, 30–31, 33, 47, 51, 101, 183, 186, 191, 221, 232, 390, 415, 862, 866, 890 adjunct. See also adposition adpositional phrase, 430 ideophone, 422 non-phrasal, 44–47 phrasal, 42–44 adposition assignment rules, 435–436 layered structure of, 49, 430 non-predicative, 49–50, 100, 188, 428, 434 postposition, 49, 193, 196, 868, 890–891, 915, 922 predicative, 49, 100, 187, 428, 434–435 adjunct, 101, 434 argument-adjunct, 101, 430, 442 preposition, 49, 51, 189, 434–435, 890–891, 955 semantic representation of, 435–436 adverb, 31, 33, 186–187, 403–404, 890, See also adjunct, non-phrasal acquisition of, 678 adverb ordering, 407 ideophonic, 412 mimetic, 412

semantic representation of, 411 adverbial clause. See juncture–nexus combinations, clausal subordination agent, 9, 22, 107, 243–244 agreement, 63, 66, 72, 116, 119, 153, 278, 282, 299, 327, 375, 377, 508–509, 583, 657, 674, 862, 869, 871, 896, 898, 915–916 Aktionsart, 94, 99, 220, 242, 891 accomplishment, 96, 221, 306 achievement, 95, 221 acquisition of, 669 active accomplishment, 96, 306 activity, 95, 221, 306 causative types, 95, 99, 302 process, 96 semelfactive, 98 state, 95, 220 Alacatlazala Mixtec, 298 Albanian, 343, 650 Algonquian languages, 832 alignment, 258, 273, 656, See also case marking, privileged syntactic argument (PSA), Selection Hierarchy ergative-absolutive, 275 nominative-accusative, 275 split, 871 Amele, 374, 536, 876, 915–950 Amharic, 659 Amis, 954–979 anchoring, pragmatic, 458 Angami Naga, 283 animacy, 141, 143–144, 299, 343, 433, 479, 826–827, 830 anticausative construction, 304 antipassive. See voice aphasia, 718–719 applicative, 142, 248, 272, 300–303, 837–838, 868, 870, 880, 916, 918, 932, 958, 961, 965, 968 appositive, 59 Arabic, 26, 234, 785, 787, 792–793, 796 Arapaho, 832 Archi, 160 argument, 21 core, 22–23 semantic, 22

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press

984

INDEX

argument (cont.) sharing, 73, 550, 558 syntactic, 22 argument structure, 58, 107, 292, 559, 582, 683, 721, See also logical structures alternations, 293–294 argument-adjunct. See adposition, predicative Armenian, 650, 658 aspect, 33, 202, 406, 533, 825, 894, 904, 927 acquisition of, 667 assertion, pragmatic, 457, 466, 493, 618, 629 Atayal, 973, 976 Avatime, 889–911 Bahasa Indonesia, 108 Bakusu, 309 Bambara, 597 Bantu languages, 66, 308–309, 312, 414, 474, 508–509, 868 Barai, 41–42, 148, 279, 536 Barbareño, 562 Basque, 122, 328 Belhare, 566 Bella Coola, 375 Blackfoot, 832 borrowing, 659 branching, 20, 36, 47, 58 direction theory, 651 Broca’s area, 695 Bulgarian, 50, 343, 650, 657 Burmese, 54 case marking, 119, 271, 318, 374, 428, 433, 509, 630, 654, 711, 873–874, 955 accusative system, 323, 327, 657 assignment rules, 323, 328 ergative system, 323 hierarchy, 339 nominative system, 108 syncretism, 339–340 Catalan, 346 Caucasian languages, 153, 160 causation, 7, 89, 245, 250, 302, See also Interclausal Relations Hierarchy, Aktionsart causative construction, 302, 546 Cavineña, 572 Cheyenne, 824 Chichewa, 66–68, 308, 414, 518 Chinese, 231, 270, 279, 281, 572, 618 Mandarin Chinese, 29, 77, 93, 205, 275, 280, 467, 481, 494, 496, 721 Choctaw, 279–280, 870 Chuj, 583 clause layered structure of, 21, 23, 27, 430 non-universal features of, 24 relative, 594, See also relative clause universal features of, 24, 429 clause linkage, 68, 525, 559 asymmetrical, 79 juncture, 69–71, 527 nexus, 71–75, 527 symmetrical, 77 clause linkage marker, 828, 910, 942 cleft sentence, 591, 602 it-cleft, 74, 507, 607 Comanche, 581, 583 comment, 490

complementation, 69, 77, 561, 771, See also subordination complementizer, 592–593, 876, See also clause linkage marker Completeness Constraint, 116, 125, 293, 300, 336, 596–597 extended, 496 complex sentences, 68, 72, 76, 525, 617, See also clause linkage, juncture–nexus combinations concessive clauses, 570 conditional clauses, 570 conjunction reduction, 500, 569, See also topic, topic chain constituent projection, 33, 70, 115, 372, 377 Construction Grammar, 544 constructional schema, 10, 125, 130, 161, 478, 500, 569, 583, 725, 805 contrastiveness, 471 control construction, 64, 151, 156, 158, 536, 581, 748, 879, 970 obligatory, 159, 582 controller. See privileged syntactic argument copula, 39, 185, 221, 602, 798, 829, 863, 895, 917 be, 101 Cora, 546 core, 21 core argument, 5, See also argument, core direct, 81 oblique, 81, 100 Creek, 870 Croatian, 61, 118, 122, 439, 506, 647, 653, 659, 661, 711–714, 721 cross-reference, 64, 72, 319, See also agreement Cushitic languages, 659 Danish, 618, 621 dative alternation, 294, 298, See also variable undergoer dative case, 108, 119, 274, 319, 323, 433, 439–440, 660, 955 dative shift. See dative alternation degree achievement, 97 dependence operator sharing, 71 vs. subordination, 72 dependent-marking languages, 64, 308, 377, 382 derivational morphology, 62, 371 Dhivehi, 26, 122 Didinga, 415 direct/inverse system, 825, 871–872 directionals, 202, 531, 757, 896 Discourse Representation Theory, 481, 489, 492 ditransitive, 298–301, 869–870, 919 Djaru, 340–341 double-marking languages, 68, 328, 343 Dutch, 126, 659, 788 Dyirbal, 27, 31–32, 108, 122, 135, 208, 257, 271, 273–274, 279, 310, 313, 322, 566, 583, 684, 787 Emai, 404, 415 Enga, 273, 280 ergative case, 120, 129, 160, 323, 327, 329, 331, 430, 479, 656–657, 869–872, 960 case system, 327, 332–333, See also case marking

ˇ

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press

Index ergativity, 340, 871 syntactic, 313, See also ergative, case system Estonian, 340, 343–344, 348–349, 351, 355 evaluative construction, 975 event quantification, 202, 406, 414, 417, 757, 825, 926 evidentials, 37, 182, 202, 406, 531, 679 exceptional case marking. See matrix-coding constructions and raising extraction restrictions, 616, 618, 620 Japanese, 621 relativization, 621 topicalization, 621 wh-questions, 621 wh-question linking template, 710 Farsi, 26 Finnish, 343–344, 350–351, 667, 681 focus, 457, 459, 463–464, 605, 610 argument, 470 contrastive, 24, 604, 844 fronting, 475 lexical alternations, 478 marking, 25–26, 39, 465–466, 472, 480, 603 narrow, 469, 477, 593, 847, 864 marked, 471 unmarked, 471 predicate, 466, 478, 844, 864 sentence, 468, 846 focus domain, 622 actual, 467, 489, 508, 934 potential, 467, 472–473, 477, 489, 508, 604, 619, 933 focus structure, 466, 472, 843, 902 complex sentences, 619 projection, 489, 491, 618 subordinate, 472, 625 focus-sensitive elements, 604 Formosan languages, 954, 973, 976 Frame Semantics, 219 Frankish, 659 French, 89, 105, 130, 155, 247, 253–255, 298, 467–468, 480, 529, 537, 541, 604, 649, 659, 662 Middle French, 649 Functional Grammar, 430 FunGramKB, 220, 227–228, 804 gender, 374–375 Generative Lexicon Theory, 102, 218, 230 Georgian, 306, 322, 566 German, 24, 43, 48–49, 51, 61, 68, 108, 118–119, 126, 187, 189, 247–248, 253–255, 257, 326, 505, 566, 659, 716, 787 Germanic languages, 375, 652 gerund, 74, 79, 560, 584 Government and Binding, 19 grammatical relations, 4, 72, 117, 119, 269–270, 272, 827, 854, 965 non-universality of, 4 grammaticalization, 653 Greek, 343, 648, 654, 658 Classical Greek, 661 Homeric Greek, 650, 661 Greenlandic Eskimo, 297 Gum languages, 915 Guugu Yimidhirr, 105

Hakha Lai, 299, 308 Halkomelem, 343 Head-Driven Phrase Structure Grammar, 3, 738, 788 head-marking languages, 63, 295, 308, 370, 376, 378–379, 434, 504, 825, 862, 865, 893, 915 Hebrew, 791 Biblical Hebrew, 785, 787, 791 Hindi, 340, 346 Huallaga Quechua, 532 Hungarian, 135, 304 Icelandic, 24, 68, 108, 118, 140–141, 154, 322, 324–326, 332, 566, 654–655, 697, 716 iconicity, 88, 430, 742 identifiability, 458–459 ideophone, 404, 890, See also adverb illocutionary force (IF), 33, 41, 202, 406, 531, 568, 652, 685, 925 assertion, 35 immediate common ground, 53, 490 incorporation, 22, 295, 308, 834, 915 incremental theme, 97 infinitival complement, 152, 349, 560, 711, 773 inflectional morphology, 62, 371, 377 information structure, 114–116, 456–457, 488 information unit (IU), 115, 461, 489 instrument, 107, 243, 250–251, 387, 432, 437–438, 504, 869, 909, 924, 931, 961, 971 instrumental prefix, 64, 309 instrumental case, 108, 329–330, 334–335, 442, 660, 838 Interclausal Relations Hierarchy, 7, 87–88, 545, 564, 907–908 semantic hierarchy, 83–87, 545, 563, 877 syntactic hierarchy, 82–83, 543, 548 intonation, 26–27, 477, 492, 897, 939 Inuktitut, 340 inverse construction, 827, 852, See also direct/ inverse system Iraqw, 528 Irish, 650, 785 Old Irish, 653 island constraints, 616–617, 748, See also extraction restrictions Italian, 47, 89, 256, 346, 467, 474–476, 478, 480, 512, 566, 618 Jakaltek, 89, 108, 117, 126, 257, 280, 322, 566 Japanese, 26–27, 29, 42, 108, 110, 130, 147, 202, 331–332, 412–414, 417–419, 465, 467, 481, 494–496, 537, 541, 547, 566, 572, 621–626, 671, 677, 768, 785, 920 Old Japanese, 548, 647 juncture, 69–71, 92, 559, See also clause linkage clausal, 70, 527 core, 69, 528 nuclear, 69, 530 sentential, 70 juncture–nexus combinations, 76, 82, 561 clausal coordination, 76, 910 clausal cosubordination, 77 clausal subordination, 910 adverbial, 80, 571 daughter, 79 complex RPs, 92–94 core coordination, 156–157, 529, 905, 908 core cosubordination, 91, 155, 538, 908

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press

985

986

INDEX

juncture–nexus combinations (cont.) core subordination, 73, 79, 88, 533–534, 561 daughter, 150 peripheral ad-core, 150, 543, 575 nuclear coordination, 530 nuclear cosubordination, 77, 88, 207, 538, 586, 908 nuclear subordination, 534, 939 sentential coordination, 539, 561, 574 sentential subordination, 539, 561 Kabardian, 153–154, 340–341, 352 Kaluli, 479, 509 Kashmiri, 346 Kewa, 37, 204 Kharia, 187, 190–198 Khwe, 533 Kikuyu, 66–67, 509 Kinyarwanda, 312, 322, 566 Kokama, 563, 579 Korean, 41–42, 204, 509–510, 566, 618, 680 Lakhota, 22–23, 27–28, 30–32, 34–37, 39, 56, 63, 65, 68–69, 87, 147, 273, 378, 509, 583, 617, 621, 767, 772, 865 Latin, 31, 433, 648, 650, 655, 657, 660 layered structure of the clause. See clause, layered structure of layered structure of the word, 62, 372–373 operators, 373 left-detached position, 26 lexical decomposition, 94–95, See also logical structures lexical entry, 95, 230, 236, 371, 384–385, 428, 703, 707, 715, 786, 788, 809–810 Lexical Functional Grammar, 3, 20 lexical integrity hypothesis, 63 lexical rules, 306, 786, 788, 792, 808 lexicon, 10, 95, 114, 117, 191, 200, 218, 220, 225, 227, 229, 231, 314, 371, 383, 386, 435, 695, 703, 715, 717, 722, 745, 785 Linguistic Knowledge Builder, 788 linking, 777 acquisition of, 683 lexical phase of, 8, 147 morphosyntactic phase of, 8, 147 processing model, 699–700 linking in complex sentences, 150, 568, 576, 582 algorithm, 150 cleft constructions, 607 complex RPs, 596, 598 linking in simple sentences, 116, 841 role of information structure in, 488 semantics-to-syntax, 116, 123, 495, 565, 722, 874 algorithm, 116 syntax-to-semantics, 123, 514, 566, 722 algorithm, 123 logical structures, 95, 99, 200, 243–244, 306, 320, 384 conceptual, 228, 230 Luwo, 414–415 Macedonian, 343, 653, 657 Macro-Event Property, 87, 91 macroroles, 108–109, 242, 244, 271, See also ActorUndergoer Hierarchy actor, 108, 244

default assignment principles, 113, 246 undergoer, 109–110, 244 variable linking, 138, 145, 248, 252, 506, 660, 726 Maithili, 566 Malagasy, 592, 613 Maltese, 186, 205 Mandarin, 105 Maori, 527, 785 Marathi, 661 matrix-coding constructions, 324, 768 mimetics, 412, See also adverb Minimalist Program, 19 modality, 33, 182, 200, 202, 406, 536, 550, 653, 825, 887, 903 acquisition of, 687 mood, 33, 904, See also illocutionary force Mparntwe Arrernte, 647 Muskogean languages, 870–871 Nahuatl, 295 natural language processing, 788, 811 Natural Serialization Principle, 6, 652 negation, 33, 37, 41, 55, 202, 406, 825, 896, 904 Nepali, 566 neutralization restricted, 272, 274–275, 277, 279, 681 unrestricted, 275 nexus, 71–75, 92, 559, See also clause linkage, juncture–nexus combinations coordination, 72, 74–75, 559–560, 765, 773 vs. conjunction, 72 cosubordination, 71–72, 74–75, 536, 560, 753, 765, 944, 946 subordination, 71, 75, 77, 771, 828, 876 ad-subordination, 80, 83, 558, 570, 944, 948 daughter, 79, 83 nominalization, 862, 876–882 nominative case, 119, 129, 271, 323, 326, 328, 331, 430, 657, 682, 868, 871 case system. See case marking Northern Sámi, 343, 347, 351, 354 Norwegian, 234 noun complement, 622 noun phrase, 5, 28, 208, 282, 284, 298, 609, 867, See also reference phrase NP, 28, See also reference phrase nucleus, 21–22 number, 374–375, 867 object, 108, 269 direct, 110 double-object construction, 111 indirect, 110 of predicative PP, 134 primary object language, 248, 299, 841, 869 primary object pattern, 142 oblique constituent, 61 core. See core argument peripheral. See adjunct Ojibwa, 832 Old Church Slavonic, 50 Old English, 50, 282, 372, 648, 654 Omaha, 647 operator, 33, 36, 57, 201, See also reference phrase clausal, 202 core, 202 nuclear, 202

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press

Index ordering, 36, 202, 204 semantic representation of, 207 sharing, 71 operator projection, 34, 38, 40, 55, 202 Optimality Theory, 331

acquisition of, 684 in-situ wh-word, 473, 966 yes/no questions, 58, 618, 934 acquisition of, 684 quirky case, 325

Paez, 206 Palauan, 340–341 Papuan languages, 868, 876, 915 parser, 123, 497, 699, 776, 787, 790 passive. See voice Pastaza Quechua, 414 periphery, 22, 405, 421, 435, 447 Persian, 785 Old Persian, 340, 342, 353 Person Hierarchy, 841 ~ 83 Piraha, Pisaflores Tepehua, 311 Pitjantjatjara-Yankunytjatjara, 373 pivot. See privileged syntactic argument (PSA) Plains Cree, 832 Polish, 619–621, 672–678, 682 polysynthesis, 825, 861 possession, 114, 224, 243, 437, 899, 965 alienable, 929, 932 change of possession verb, 111 existential verb, 105 inalienable, 929 possessive RP, 516, 930 semantic representation, 246 possessor raising, 106, 868 post-core slot, 26, 465, 566, 623, 628–629, 923–924, 956 post-detached position, 26, 187, 465, 577, 623, 923 postposing construction, 628 postposition. See adposition pre-core slot, 24–25, 38, 44, 57, 465, 471, 474, 623, 749, 809, 844, 846, 848, 893, 924, 956 pre-detached position, 26, 410, 463, 465, 467, 623, 893, 923 predicate, 21 preposition, 49, See also adposition presupposition, pragmatic, 457, 462, 490, 493 Principles and Parameters, 3 privileged syntactic argument (PSA), 117–118, 272, 874 controller, 119, 272, 276, 278 pivot, 119, 272, 275 Selection Hierarchy, 118, 275, 322 selection principles, 275 process, 97 processing, language, 696–699 projection grammar, 740 constituent projection, 740 operator projection, 740 speech act projection, 467 pronouns, 61, 186 proximate/obviative system, 826, 835, 848 punctual, 95, 98 Punjabi, 346 purpose clauses, 579

raising, 73, 151, 156, See also matrix-coding constructions Rawang, 282 reanalysis, 658 reason clauses, 572 reciprocal prefix, 868 verb form, 944 reference phrase (RP) core-level periphery, 93 juncture–nexus combinations core coordination, 93 core cosubordination, 93 nuclear coordination, 93 nuclear cosubordination, 93 RP coordination, 92 RP cosubordination, 92 RP subordination, 93, See also relative clause layered structure of, 53 logical structure, 209 operators, 53, 208 classifiers, 54 referent tracking, 270, 272, 279–280, 835 reflexivization, 89, 133 anticausative, 305, 682 diachrony, 650 reflexive binding, 149, 716 role hierarchy, 716 Relational Grammar, 3, 19–20 relative clause, 69, 594, 882 Accessibility Hierarchy, 591 externally headed, 591, 595 free, 600 internally headed, 592, 597 non-restrictive, 599 restrictive, 595 relativization, 625 resultative construction, 70, 104, 753 Rhetorical Structure Theory, 86 Riau Indonesian, 275, 280 right-detached position, 26 Romance languages, 652, 811 Romani, 658 Romanian, 342 Russian, 29, 31, 105, 108, 185, 255, 257, 334, 336, 475–476, 505, 660, 675

Qiang, 37, 284 qualia, 102, 222–224, 384 quantifiers, 55, 208, 297, 928 questions wh-questions, 25, 134, 139, 473, 511, 617, 622, 630, 708

Sama, 126–129, 322, 566 Sama-Bajaw, 303 Sanskrit, 661 Vedic Sanskrit, 658 Seediq, 976 Semitic languages, 652, 659 serial verb constructions, 904, 907, 937 Sesotho, 474, 508 Setswana, 474, 508 Sicilian, 476–477 sign languages, 802 Irish Sign Language, 802 Siwu, 414 Slavic languages, 433, 660 Somali, 659

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press

987

988

INDEX

Southern Tiwa, 297 Spanish, 66–67, 231, 255, 305, 340, 346, 371, 377, 382, 384, 386, 393, 433, 436–437, 439–446, 448–450, 573, 575–576, 578, 605–606, 785, 805 Argentinian Spanish, 460, 657 split brain, 694–696, 715 St’át’imcets, 376 stage topic, 469 Standard Theory, 19 state, change of, 95 status, 37, 182, 202 subjacency, 617 subject, 25, 269–270, 272, 275, 348, 382, 657, 681, See also privileged syntactic argument dummy, 121 within RP, 57 subordinate clause. See juncture–nexus combinations, nexus, subordination adverbial clause, 558 complement clause, 77 relative clause, 594 subordination. See nexus, juncture–nexus combinations nominalization, 876 Swedish, 618, 621 switch-reference, 279, 549, 863, 941, 944, 949 syntactic inventory, 60, 715, 738, 743 Tabassaran, 656–657 Tagalog, 25–26, 28, 30–31, 39, 47–48, 58, 93, 105–106, 117, 275–278, 280, 340, 345, 353, 506, 654 telicity, 221, 305, 344, 419, See also state, change of template, syntactic, 60–61 combination, 61, 743 selection principle, 61, 120 temporal clauses, 577 tense, 33, 37, 201–202 Thai, 270, 494, 496, 529 thematic relations, 9, 107, 242–244 Tibetan, 566 Tibeto-Burman languages, 37, 282–284, 299, 308 Tiwi, 37, 204 Tlachichilco Tepehua, 309–311, 421 Toba Batak, 473, 515 Tolmaˇ ci Karelian, 351 tone language, 889 Tongan, 193, 296 topic, 457, 461–463, 490, 505, 955 aboutness, 462–463, 465, 469 acceptability scale, 464 contrastive, 24, 471

frame-setting, 463, 726 referential, 462–463, 469 topic chain, 570 topicalization, 624–625, 627 Toqabaquita, 583 transitivity, 112–113, 298 acquisition of, 682 M[acrorole] transitivity, 112 S[yntactic] transitivity, 112 Tree Adjoining Grammar, 743, 813 Tree Wrapping Grammar, 752, 813 Tsez, 153 Tsou, 973 Tsova Tush, 110 Tukang Besi, 530 Tupinamba, 297 Turkish, 37, 122, 204, 231, 303–304, 311, 529, 531, 671–672, 677, 686, 758, 771 Tzotzil, 25 Tzutujil, 280 Udihe, 535–536 undergoer. See macroroles UniArab, 792 Ute, 126 Uto-Aztecan languages, 140, 295, 546, 581 voice, 126, 273, 310, 825, 965, 968 acquisition of, 683 actor, 129, 958, 969 antipassive, 310, 313–314 argument modulation, 126, 129, 310 neutral, 968 passive, 126, 310 PSA modulation, 126, 129 undergoer, 129, 958, 971 VP, 23 ellipsis, 24, 498–499 Wari’, 25, 81–82, 190, 198–199 Warlpiri, 83, 273, 280, 324, 327, 334, 682–683 Wernicke’s area, 695 wh-questions. See questions word formation, 383 Yagnob, 340–341 Yaqui, 140–142, 312, 528, 566, 569, 572–574, 580, 583, 585 Yimas, 861–887 Yucatec Maya, 68, 296 zero anaphora, 481

https://doi.org/10.1017/9781316418086.033 Published online by Cambridge University Press