Case, Typology and Grammar (Typological Studies in Language)

CASE, TYPOLOGY AND GRAMMAR TYPOLOGICAL STUDIES IN LANGUAGE (TSL) A companion series to the journal “STUDIES IN LANGUA...

Author: Anna Siewierska | Jae Jung Song

59 downloads 2250 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

CASE, TYPOLOGY AND GRAMMAR

TYPOLOGICAL STUDIES IN LANGUAGE (TSL) A companion series to the journal “STUDIES IN LANGUAGE” Honorary Editor: Joseph H. Greenberg General Editor: Michael Noonan Assistant Editors: Spike Gildea, Suzanne Kemmer

Editorial Board: Wallace Chafe (Santa Barbara) Bernard Comrie (Los Angeles) R.M.W. Dixon (Canberra) Matthew Dryer (Buffalo) John Haiman (St Paul) Kenneth Hale (Cambridge, Mass.) Bernd Heine (Köln) Paul Hopper (Pittsburgh) Andrej Kibrik (Moscow)

Ronald Langacker (San Diego) Charles Li (Santa Barbara) Andrew Pawley (Canberra) Doris Payne (Oregon) Frans Plank (Konstanz) Jerrold Sadock (Chicago) Dan Slobin (Berkeley) Sandra Thompson (Santa Barbara)

Volumes in this series will be functionally and typologically oriented, covering specific topics in language by collecting together data from a wide variety of languages and language typologies. The orientation of the volumes will be substantive rather than formal, with the aim of investigating universals of human language via as broadly defined a data base as possible, leaning toward cross-linguistic, diachronic, developmental and live-discourse data. The series is, in spirit as well as in fact, a continuation of the tradition initiated by C. Li (Word Order and Word Order Change, Subject and Topic, Mechanisms for Syntactic Change) and continued by T. Givón (Discourse and Syntax) and P. Hopper (Tense-Aspect: Between Semantics and Pragmatics).

Volume 38

Anna Siewierska and Jae Jung Song (eds) Case, Typology and Grammar

CASE, TYPOLOGY AND GRAMMAR IN HONOR OF BARRY J. BLAKE Edited by

ANNA SIEWIERSKA Lancaster University

JAE JUNG SONG University of Otago

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Case, typology, and grammar : in honor of Barry J. Blake / edited by Anna Siewierska, Jae Jung Song. p. cm. -- (Typological studies in language, ISSN 0167-7373; v. 38) “Bibliography of the published works of Barry J. Blake”: p. Includes bibliographical references and index. 1. Grammar, Comparative and general. 2. Grammar, Comparative and general--Case. 3. Typology (Linguistics) I. Blake, Barry J. II. Siewierska, Anna. III. Song, Jae Jung, 1958- . IV. Series. P201.C37 1998 415--dc21 98-17515 ISBN 90 272 2937 6 (Eur.) / 1 55619 651 2 (US) (Hb; alk. paper) CIP © Copyright 1998 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

Table of Contents

List of Abbreviations Preface Bibliography of the published works of Barry J. Blake On the semantic frames of be and possessive have. Keith Allan

vii ix xii 1

Crow is sitting chasing them: grammaticalization and the verb ‘to sit’ in the Mantharta languages, Western Australia. Peter Austin

19

Factors of typology in language acquisition: some examples from Warlpiri. Edith L. Bavin

37

Markedness and iconicity: some questions Byron W. Bender

57

Throw the baby from the window a cookie: English and Pennsylvania German in contact Kate Burridge

71

The great Daghestan case coax Bernard Comrie and Maria Polinsky

95

Iwaidja mutations and its origins Nicholas Evans

115

Functional control with and without structure- sharing Richard Hudson

151

The verbal suﬃx -ngany in Warrwa William McGregor

171

From event sequence to grammar: serial verb constructions in Kalam Andrew Pawley and Jonathan Lane

201

Passive-to-ergative vs inverse-to-ergative Anna Siewierska

229

vi

contents

Benefactive marking in Oceanic languages: from possessive classiﬁers to benefactive markers Jae Jung Song

247

Ergativity, transitivity, and clitic coreference in four Western Austronesian languages: a Lexicase analysis Stanley Starosta

277

A discourse explanation for the cross-linguistic diﬀerences in the grammar of incorporation and negation. Sandra A. Thompson

307

Applicative constructions in Warrungu of Australia Tasaku Tsunoda

341

Language Index Subject Index

375 377

Abbreviations A abl abs acc actr agr agt aff all anim anti aor apl app appr art asp aux ben caus clc clf cltc cmp coll com cont cor cop ctr dat def dem det dim dir dis

transitive agent ablative case absolutive case accusative actor agreement agent aﬀective allative case animate antipassive aorist animal plural applicative apprehensional article aspect auxiliary benefactive causative modal clitic classiﬁer clitic completive collective comitative case continuous correspondent copula contrast dative deﬁnte demonstrative determiner diminutive direct/direction distal

ds du emph erg ess exc f foc fut gen hab imp impf impr inc inchoat ind indef iness inf inst intent interr intr inv io ipfv irls iv loc m mbnt min mis mns neg

diﬀerent subject dual emphatic ergative (case) essive case exclusive feminine gender focus future genitive case, possessed habitual imperative imperfective impersonal inclusive inchoative indicative indeﬁnite inessive inﬁnitive instrumental intentive interrogative intransitive stem forming suﬃx inverse indirect object imperfective irrealis inﬂecting verb locative masculine gender ambient minal number miscellaneous gender means negative element

viii N nom NP npst nr nrls nt num O obl obj obv pass pat pauc perf perl pfv pl poss pp PP ppd pple prev prep prior pro prog proh prox prs prt

abbreviations nominal, noun nominative noun phrase nonpast nominalizer nonpast realis neuter numeral object oblique object obviative passive patient paucal perfect perlative case perfective plural possessive past perfective prepositional/ postpositional phrase postpositive determinant participle preverb preposition prior action pronoun progressive prohibition proximate present particle

pst pstevid purp Q r rec ref rfl rls S sbj sembl sg spec SS sub subj t ta th tr trns U usit v ve vert VP 1 2 3

past past evidential purposive question particle, question word recipient reciprocal referential reﬂexive realis subject/ intransitive subject subject semblative singular speciﬁc same subject subordinator subjunctive tense tense-aspect marker thematic suﬃx transitive stem forming suﬃx transitive undergoer usitative verbal, verb vegetable gender vertical verb phrase 1st person 2nd person 3rd person

Preface The 15 papers in this volume have been written in honour of Barry John Blake, Foundation Professor of Linguistics at La Trobe University, Melbourne, Australia to commemorate his 60th birthday. The title of the volume and the contents of the contributions reﬂect the three themes that have played a dominant role in Barry’s linguistic work. Born in Ascot Vale, Victoria, Australia in 1937, Barry received his BA (honours) in English Language and Latin in 1958 from the University of Melbourne. In 1960 —after a one-year stint as a secondary school teacher of English, Latin and Ancient History—he went to work for the Department of Defence, initially as a linguist and later as a language training oﬃcer. In 1966 he accepted a position as Australian Institute of Aboriginal Studies research fellow at Monash University (in the ﬁrst linguistic department of Australia), which took him to western Queensland to work on three languages: Kalkatungu, Pitta-Pitta and Yalarnnga. He obtained MA and Ph.D degrees from Monash University in 1968 and 1975, respectively. He taught at Monash University from 1970 till 1987. In recognition of his research achievements, in 1987 he became a fellow of the Australian Academy of Sciences. In 1988 he took up the Foundation Chair in Linguistics at La Trobe University. Barry’s research for the last thirty years has been devoted to the study of Australian Aboriginal languages. His numerous publications include: descriptive grammars of individual languages (Kalkatungu and Pitta-Pitta), pan-Australian surveys of aspects of clause structure, theory speciﬁc analyses of particular phenomena in individual languages and across-languages, works on the genetic classiﬁcation of Australian languages and more popular accounts of the history and nature of Australian languages. Barry’s research on language classiﬁcation has put to rest the long standing debate among Australianists as to the existence of a genetic as opposed to merely typological distinction between the Pama–Nyungan and non-Pama–Nyungan languages. By demonstrating that the verb pronominal preﬁx forms in a range of nonPama–Nyungan languages could have evolved from a single proto-language, he has provided strong evidence for the genetic unity of the non-Pama–Nyungan languages, on the one hand, and the Pama-Nyungan languages, on the other. This work on pronouns resulted in the establishing of a set of proto-pronouns for the non-PamaNyungan languages in addition to the proto-Australian pronouns reconstructed earlier by R.M.W. Dixon. The implications of there being two separate proto-languages in Australia are, needless to say, highly signiﬁcant and will undoubtedly constitute an imporatant topic of future research on Australian languages. While Barry has been concerned with diverse facets of the structure of Australian

x

preface

languages, as epitomized in his 1987 book Australian Aboriginal Grammar (London: Croom Helm), the theme which has most consistently reoccurred throughout his work is case. Given the rich concentration of inﬂectional case languages in Australia, this is not altogether surprising. As is well known, most Australian languages have rich inﬂectional case systems, though some, only for the non-core relations. Moreover, unlike in other parts of the world, the majority of the languages with inﬂectional case marking in Australia, display ergative or split ergative case marking, a phenomenon which has aroused not only much theoretical interest but also considerable controversy. The issues that have featured most prominently in Barry’s discussions of case are: the identiﬁcation of cases, particularly the use of distributional as opposed to functional criteria, the relationship between synthetic and analytic exponents of case, the order of the emergence and decay of morphological cases, the relationship between case marking and alternative forms of marking via cross-referencing pronouns on the verb and word order, and the nature, origin and distribution of ergative as opposed to accusative case marking systems. His interest in the distribution of ergativity in particular has more recently taken him beyond his home continent into the Philippines and Indonesia, and led him to ‘discover’ the existence of ergativity in languages of these regions, thereby demonstrating that ergativity is much more widely attested than previously thought. Many of Barry’s insights on these topics, which appeared in his books and articles published in the 70s and 80s, are summarised in Case (1994) written for the Cambridge Textbooks in Linguistics series. Inherently intertwined with Barry Blake’s work on case is the second central theme of his research, linguistic typology. His Language Typology: Cross-linguistic Studies in Syntax (1981), written with Graham Mallinson, together with Bernard Comrie’s Language Universals and Linguistic Typology were the ﬁrst textbooks in the ﬁeld of language typology. Both books are widely regarded as having had a considerable impact on developing interest in the ﬁeld of typology and inspiring new work in this area. A particularly valuable feature of Barry’s textbook is the inclusion of a range of statistical data, which enables the reader to form a better understanding of typological methodology and provides a sound basis for evaluating the relevance of the typological correlations discussed. Another laudable characteristic of this book is the insistence on viewing typological generalizations not as sole ends of typological research but rather as essential means to formulating external explanations for the distributuion of cross-linguistic data. Despite the passage of time, most of the topics covered in the book continue to be relevant and the explanations oﬀered for the presented typological generalizations have lost little of their appeal. While the analyses of cross-linguistic data in Language Typology are presented in an essentially model-neutral framework, incorporating general functional princi-

preface

xi

ples, much of Barry’s typological work has been coupled, in a symbiotic manner, with his interest in diﬀerent models of grammar. Unlike many typologists, who eschew formal models of grammar altogether or restrict their attention to only one, Barry has constantly sought to bring the insights of language typology into the realm of formal syntactic theory, and vice versa. He has developed analyses of various syntactic phenomena, particularly grammatical relations and grammatical relations changing operations such as passives, antipassives, and applicative constructions in a number of grammatical frameworks including: Case Grammar, Relational Grammar, Lexical Functional Grammar, Role and Reference Grammar, Lexicase and Word Grammar. He has also elaborated, independently of Kenneth Hale, a nonconﬁgurational analysis of the clause structure of Kalkutungu. Barry’s attitude to grammatical models is reﬂected most admirably in Relational Grammar (1990, London: Routledge) in which he not only presents a highly lucid account of the basic tenets of the theory, but also greatly informs the RG analyses of clause structure by bringing his expertise in a wide range of languages to bear on them. The above three themes of Barry’s research, case, linguistic typology and grammar cover much of the ground of modern theoretical linguistics. It is no coincidence therefore that the contributions to this volume cover a suitably wide range of language phenomena and several diﬀerent theoretical persuasions, and thus reﬂect his research themes closely. The editors sought to limit the contributors to this volume to three diﬀerent categories: Barry’s Ph.D students, Barry’s colleagues at Monash and La Trobe universities, and ﬁnally scholars who have shared or exchanged their ideas about, and admiration for, language with Barry. Some of the contributors fall into more than one category. Needless to say, there are many students, colleagues, and friends of his who may have been left out, and the editors express their abject apologies, and implore their understanding. It is hoped that there will be more opportunities for all in the future. Perhaps it is ﬁtting to close this preface by reﬂecting on one important feature of language which the editors years ago learned about from Barry as his students. He not only showed them that language is an object of great sophistication. But he also taught them to appreciate it as an object of beauty. The editors would like to invite the reader to not only view the following chapters as contributions to the discipline, but also appreciate them in this spirit, just as Barry will. Lancaster Melbourne September 1997

A.S. J.J.S.

Publications of Barry Blake 1969 The Kalkatungu Language. Canberra: AIAS. [Revised version of MA thesis.] 1970 ‘‘Acoustic phonetics and the study of Aboriginal languages.’’ In D.C. Laycock (ed.) Linguistic Trends in Australia. Canberra: AIAS, 27–38. 1971 ‘‘Jalanga: an outline morphology.’’ Papers on the languages of Australian Aboriginals. Canberra: AIAS 12–27. The Pitta-Pitta Dialects. Monash Linguistic Communications 4. [With B.G. Breen.] ‘‘Jalanga and Kalkatungu: Some comparisons.’’ Papers on the languages of Australian Aboriginals. Canberra: AIAS, 29–33. 1972 Review of D. Crystal: Linguistics. AUMLA 38: 268–9. ‘‘Salvage study in Australian Aboriginal Languages.’’ Monash Linguistic Communications 6: 1–10. 1973 ‘‘Linguistics.’’ Victoria Careers III. 1974 ‘‘The Causative in Kalkatungu.’’ In B.J. Blake (ed.) Papers in Australian Languages. Monash Linguistic Communications 14: 1–21. Review of R.M.W. Dixon: The Dyirbal language of North Queensland. Lingua 34: 286–99. (ed.) Papers in Australian Aboriginal Languages. Monash Linguistic Communications 14. 1975 Review of M. Samuels: Linguistic evolution. Linguistics 164: 77–85.

the published works of barry j. blake

xiii

1976 ‘‘-ku and Kalkatungu.’’ In R.M.W. Dixon (ed.), Grammatical categories in Australian languages. Canberra. AIAS & New Jersey: Humanities Press, 464–67. ‘‘Are Australian Languages syntactically nominative-ergative or nominative accusative?’’ In R.M.W. Dixon (ed.), Grammatical categories in Australian languages. Canberra. AIAS & New Jersey: Humanities Press, 485–94. ‘‘The bivalent suﬃx -ku.’’ In R.M.W. Dixon (ed.), Grammatical categories in Australian languages. Canberra. AIAS & New Jersey: Humanities Press, 421–24. ‘‘Case mechanisms in Kalkatungu.’’ Anthropological Linguistics 18.7, 287–93. ‘‘On ergativity and the notion of subject.’’ Lingua 39: 281–300. 1977 Case Marking in Australian Languages. Canberra: AIAS. [Based on PhD.] 1978 Review of A. Radford: Italian syntax. Babel 14: 32–5. ‘‘From semantic to syntactic antipassive in Kalkatungu.’’ Oceanic Linguistics XVII: 164–9. 1979 ‘‘Degrees of ergativity in Australia.’’ In F. Plank (ed.) Ergativity. London: Academic Press, 291–305. ‘‘Pitta-Pitta.’’ In R.M.W. Dixon and B.J. Blake (eds), Handbook of Australian Languages, vol. 1. Canberra: ANU Press and Amsterdam: John Benjamins, 183–242. Review of J. Anderson: Prolegomena to a theory of grammatical relations. AUMLA 50: 314–5. Review of J. Heath: Linguistic diﬀusion in Arnhem Land. Talanya 6: 114–6. Review of W.C. McCormack & S.A. Wurm (eds.): Approaches to language. Babel 15: 20–2. Review of L. Palmer: Descriptive and comparative linguistics. Babel 15: 46–7. A Kalkatungu Grammar. Canberra: Paciﬁc Linguistics. ‘‘Introduction.’’ Handbook of Australian Languages, vol. 1. Canberra: ANU Press & Amsterdam: John Benjamins, 1–25. [With R.M.W. Dixon.] ‘‘Australian case systems: some typological and historical considerations.’’ In S.A. Wurm (ed.), Australian Linguistics Studies. Canberra: Paciﬁc Linguistics, 323–94. Handbook of Australian Languages, vol. 1. Canberra: ANU Press & Amsterdam: John Benjamins. [Ed. With R.M.W. Dixon.] ‘‘It needs more than Vegemite.’’ Review of the Australian edition of Collins Eng lish Dictionary. Overland 78: 67–9.

xiv

the published works of barry j. blake 1980

Review of T. Sebeok and J. Umiker-Sebeok (eds): Speaking of apes. Babel 16: 40-l. Review of B. Henisz-Dostert et al.: Machine translation. Babel 16: 46–7. Review of D. Fry: The physics of speech. AUMLA 53: 113. 1981 Review of T. Shopen (ed.): Languages and their status. Australian Journal of Linguistics 1: 138–9. Review of T. Donaldson: Ngiyambaa: the language of the Wangaaybuwan. Australian Journal of Linguistics 1: 117–9. Review of N. Chomsky: Rules and representations. Babel 17: 48–51. Review of B. Comrie: Language universals and linguistic typology. Babel 17: 44–51. Review of P. Austin: A grammar of Diyari. Australian Journal of Linguistics 1: 275–8. Language Typology. Amsterdam: North-Holland. [With G. Mallinson.] Australian Aboriginal Languages. Sydney: Angus and Robertson. [Second edition 1991, Brisbane: QUP.] Handbook of Australian Languages, vol. 2. Canberra: ANU Press/Amstradam: John Benjamins. [Ed. with R.M.W. Dixon.] 1982 Review of F. Merlan: Mangarayi. Lingua 58: 369–86. ‘‘The absolutive: its scope in English and Kalkatungu.’’ In P.J. Hopper and S.A. Thompson (eds), Studies in Transitivity New York: Academic Press, 71–94. [Syntax and Semantics vol. 15.] 1983 ‘‘Structure and word order in Kalkatungu.’’ Australian Journal of Linguistics 3: 143–75. Handbook of Australian Languages, vol. 3. Canberra: ANU Press/Amsterdam: John Benjamins. [Ed. with R.M.W. Dixon.] 1984 ‘‘Problems for possessor ascension: some Australian examples.’’ Linguistics 22: 437–53.

the published works of barry j. blake

xv

1985 Review of J. Heath: Nunggubuyu myths and ethnographic texts; Nunggubuyu dictionary, and functional grammar of Nunggubuyu. Australian Journal of Linguistics 5: 304–10. ‘‘Case markers, case and grammatical relations: an addendum to Goddard.’’ Australian Journal of Linguistics 5: 79–84. ‘‘Referential and conceptual roles.’’ In Festschrift in honour of Arthur Delbridge: a special issue of Beitrage zur Phonetik und Linguistik 48: 29–34. Review of M. McDonald and S.A. Wurm: basic materials in Wankumara (Galali). Oceania 55: 235. ‘‘‘Short a’ in Melbourne English.’’ Journal of the International Phonetic Association 15: 6–20. 1986 ‘‘Unaccusatives: an alternative description.’’ Working Papers in Language and Linguistics 19: 1–15. Review of W. Foley & R. Van Valin jr.: Functional syntax and universal grammar. Australian Journal of Linguistics 6. 1987 Review of T. Shopen (ed.): Language typology and syntactic description. Language 63: 606–19. [With G. Mallinson.] ‘‘Core grammar in Australia: data and lines of development.’’ Lingua 71: 179–202. ‘‘Subordinate verb morphology in western Queensland.’’ In D. C. Laycock and W. Winter (eds), A World of Language: Papers presented to Professor S.A. Wurm on his 65th birthday, Canberra: Paciﬁc Linguistics, 61–8. Australian Aboriginal Grammar. London: Croom Helm. ‘‘English and German: Two Languages Two Thousand Years Apart.’’ Review article based on J. Hawkins: English and German: Unifying the Contrasts. Multilingua 6: 309–23. 1988 ‘‘Redeﬁning Pama-Nyungan: towards a prehistory of Australian languages.’’ Aboriginal Linguistics 1, 1–90. ‘‘Tagalog and the Manila-Mt Isa axis.’’ La Trobe Working Papers in Linguistics 1: 77–90. Review of R. Tomlin: Basic word order—functional principles. Journal of Linguistics 24: 1, 213–23.

xvi

the published works of barry j. blake

Review of M. Ruhlen: A guide to the world’s languages. Journal of Linguistics 261–2. 1989 ‘‘The study of Australian Aboriginal languages.’’ Inaugural address. Melbourne, La Trobe University. ‘‘The loss of Lardie Moonlight.’’ The New Internationalist 191: 15. Review of S. Starosta: The case for Lexicase. Language 65: 614–22. ‘‘Some thoughts on Lexicase.’’ La Trobe Working Papers in Linguistics 2: 1–16. 1990 ‘‘Languages of the Queensland/Northern Territory border: updating the classiﬁcation.’’ In P. Austin, R.M.W. Dixon, T. Dutton and I. White (eds), Language and History: Essays in honour of Luise A. Hercus. ‘‘The signiﬁcance of pronouns in the history of Australian languages.’’ In P. Baldi (ed.), Linguistic Change and Reconstruction Methodology, 435.50, Berlin: Mouton de Gruyter. [Also reproduced in P. Baldi (ed.) 1991. Patterns of change, change of patterns, Berlin: Mouton de Gruyter, 219–34.] Relational Grammar. London: Routledge. Review of J. Haiman and S. A. Thompson: Clause combining in grammar and discourse. Journal of Linguistics 26: 505–9. 1991 ‘‘Woiwurrung: the Melbourne language.’’ In R.M.W. Dixon and B.J. Blake (eds), 30–122. ‘‘Introduction.’’ In R.M.W. Dixon and B.J. Blake (eds), 1–28. [With R.M.W. Dixon.] Review of W. Croft: Typology and universals. Journal of Linguistics 27: 550–9. ‘‘Guwa.’’ In Gavan Breen (ed.), Salvage studies of western Queensland Aboriginal languages. Canberra: Paciﬁc Linguistics (B 105), 109–44. [With Gavan Breen.] Handbook of Australian languages, vol. 4. Melbourne: Oxford University Press. [Ed. with R.W.M. Dixon.] 1992 Review of H. Bowe: Categories, constituents and constituent order in Pitjantjatjara: an Aboriginal language of Australia. Linguistics 30: 428–31. Review of T. Givón. Syntax: a functional-typological introduction. Volume II. Journal of Linguistics 28: 495–500.

the published works of barry j. blake

xvii

‘‘Features of case.’’ La Trobe Working Papers in Linguistics 5: 7–11. ‘‘The case hierarchy.’’ La Trobe Working Papers in Linguistics 5: 1–6. 1993 ‘‘Ergativity in the Paciﬁc.’’ La Trobe Working Papers in Linguistics 6: 19–32. ‘‘Verb aﬃxes from case markers.’’ La Trobe Working Papers in Linguistics 6: 33–58. Review of J. Nichols: Linguistic diversity in space and time. Languages of the World 6: 50–3. 1994 ‘‘Language evolution.’’ In D. Horton (ed.), The Encyclopaedia of Aboriginal Australia. Canberra: Australian Institute of Aboriginal and Torres Strait Islander Studies, 598–600. ‘‘Case marking.’’ In D. Horton (ed.), The Encyclopaedia of Aboriginal Australia. Canberra: Australian Institute of Aboriginal and Torres Strait Islander Studies, 183–4. ‘‘Language.’’ In D. Horton (ed.), The Encyclopaedia of Aboriginal Australia. Canberra: Australian Institute of Aboriginal and Torres Strait Islander Studies, 592–5. Review of F.R. Palmer: Grammatical roles and relations. Journal of Linguistics 30: 297–8. ‘‘Relational Grammar.’’ In R. E. Asher (ed.). The Encyclopedia of Language and Linguistics. Oxford: Pergamon, 3514–20. ‘‘Language Classiﬁcation.’’ In R. E. Asher (ed), The Encyclopedia of Language and Linguistics. Oxford: Pergamon, 1952–57. ‘‘Australian Languages.’’ In R. E. Asher (ed.). The Encyclopedia of Language and Linguistics. Oxford: Pergamon, 266–73. ‘‘Australian Aboriginal Languages: introduction.’’ Australian Phrasebook. Melbourne: Lonely Planet, 89–95. ‘‘Languages of Victoria and New South Wales.’’ Australian Phrasebook. Melbourne: Lonely Planet, 154–60. (with Austin, P.K.) ‘‘Sound change in Kulin.’’ La Trobe University Working Papers in Linguistics 7: 1–14. [With J. Reid.] Case. Cambridge: University Press. 1995 ‘‘Classifying Victorian languages.’’ La Trobe Working Papers in Linguistics 8: 1–59. [With J. Reid.]

xviii

the published works of barry j. blake

‘‘Patient primacy in Balinese.’’ Studies in Language 21: 483–508. [With K. Artawa.] 1997 Review of F. Plank (ed.): Double Case. Journal of Linguistics 33: 204–11. 1998 ‘‘Climbing the Tower of Babel.’’ Review of R.M.W. Dixon, 1997, The Rise and Fall of Languages. In The Australian, 4 February 1998, p. 43. In press ‘‘Case.’’ In Morphology: A Handbook of Inﬂection and Word Formation. Berlin: Mouton de Gruyter (invited but refereed). ‘‘Nominal marking on verbs: some Australian cases.’’ Proceedings ICHL XII, Manchester, August 1995.

On the semantic frames of be and ‘possessive’ have KEITH ALLAN Monash University

1. Establishing common ground This chapter examines the semantic frames of the clause predicates be and ‘possessive’ have using a variant of the formalism of Role and Reference Grammar —hereafter RRG (Foley and Van Valin 1984, Van Valin 1990, 1993a, 1993b, Van Valin and Wilkins 1993, Van Valin and LaPolla 1996).1 Deriving from the role which an argument’s denotatum plays in the action, event, or state denoted by the predicate, thematic relations are a signiﬁcant part of the semantic frame of a predicate. Thematic relations are also referred to as ‘valencies’ in Tesnière (1959), ‘(deep) cases’ in Fillmore (1968, 1977), Anderson (1971), Blake (1977); ‘semantic roles’ in Katz (1972, 1977), Dillon (1977); ‘θ-/theta roles’ in Chomsky (1981), Marantz (1984), Wilkins (1988); ‘argument roles’ in Goldberg (1995); and ‘role archetypes’ by Langacker (1990). I shall adopt the RRG notion of macroroles (hereafter MR) in addition to thematic relations such as eﬀector (subsuming agent, instrument, force, and means), experiencer (perceivers, cognizers), theme, patient, and unnumbered locatives (cf. Landau and Jackendoﬀ (1993)). MR are described in Foley and Van Valin (1984), Van Valin (1993a); but the usage here is closest to that in Van Valin and LaPolla (1996). Dowty (1991) refers to something very similar as ‘proto-roles’. The number of MR a verb has is less than or equal to the number of arguments in its logical structure. The only two MR are ‘actor’ and ‘undergoer’; they roughly correspond to what is sometimes called ‘logical subject’ and ‘logical object’ respectively. If a verb has only one argument in its logical structure it will take one MR: If the verb has an activity predicate (DO or do′2) in its logical structure, the MR is actor; otherwise, the MR is undergoer. In an active clause, the actor is (to use Langacker’s (1987) terms) the trajector proﬁled in the clause and so be-

keith allan

2

comes clause subject and the undergoer or landmark is the direct object; in a passive clause the undergoer is the trajector and the actor is relegated to a peripheral PP. In the following examples, ‘actor’ is indicated by subscripting A, [NPA], ‘undergoer’ by [NPU]. The prototypical actor is an agent, and the prototypical undergoer is a patient.e (1) (a) [SueA-agent] slapped [HarryU-patient]. (b) [HarryU-patient] was slapped by [SueA-agent]. (2) (a) [Everyone thereA-perceivers] saw [the accidentU-theme]. (b) [The accidentU-theme] was seen by [everyone thereA-perceivers]. (3)

[

HarryA-possessor] kept [the CDU-theme].

(4)

[

(5)

[

WendyA-possessor] owns [a BMWU-theme]. The BMWU-theme] belongs to [Wendypossessor].

In (5) we see that belong to has only one MR, undergoer, and that is why it cannot passivize. Vendler (1967 Ch. 4) identiﬁed aspectual categories of verbs on the basis of the restrictions on their cooccurrence with adverbials and aspects, and certain of the entailments of the clauses in which they are predicates. Vendler’s grammatical tests were further reﬁned by Dowty (1979) in the context of Montague semantics, and later by Foley and Van Valin (1984), Van Valin (1990, 1993a), Van Valin and LaPolla (1996) in the context of RRG, giving rise to the ﬁve categories below. activities states

The water sparkled. The water is cool.

achievements

The balloon burst.

accomplishments The water cooled. causatives

Jo cooled the water.

The water does something The water is in a certain state of being There is a sudden change of condition There is a gradual change of condition Jo did something that caused something (else) to happen

Van Valin claims, citing studies from eight language families, that the classiﬁcation ‘has proved to be of great cross-linguistic validity’ (1993a:34). These relations are captured in the semantic structures of verbs, referred to as ‘logical structures’ in RRG, cf. Fig. 1 where (x or x,y) indicates either one or two arguments; pred is a variable to be replaced by the particular name of the predicate. All activity verbs contain do′; all achievements contain the operator INGR (cf. ingressive) indicating instantaneous change of state or activity, e.g. x shatters →

the semantic frames of be and have Verb class

Logical structure

activity

do′(x,[pred(x or x,y)])

state

pred(x or x,y)

achievement

INGR pred(x or x,y) or INGR do′(x,[pred(x or x,y)])

accomplishment

BECOME pred(x or x,y) or BECOME do′(x,[pred(x or x,y)])

causative

Φ CAUSE Ψ

3

Figure 1. The logical structures of verb classes in RRG INGR shattered′(x);3 all accomplishments contain become indicating change of state over some period of time, e.g. x melts → become melted′(x). There are problems distinguishing INGR from become (cf. Allan (ms)), but they are not relevant to the semantics of be and have. This completes the introduction to the metalanguage I shall use. In RRG, causatives occur when some proposition Φ, usually an activity, causes some proposition Ψ, usually an achievement or accomplishment, to come about. Causatives typically: (1) occur with progressive aspect; (2) occur with adverbs like vigorously, actively; (3) occur with adverbs like quickly, slowly; (4) occur in V for an hour, spend an hour Ving; (5) occur in V for an hour → V at all times in the hour; (6) allow of a causative paraphrase containing the same number of NPs as the sentence being paraphrased, e.g. Robin gives the book to Kim → ‘‘Robin causes Kim to come to have the book’’, but Sandy runs → / ‘‘Sandy causes Sandy to run’’.

2. The copula In RRG, locational states have the logical structure in (6): (6) be.on′(xlocative,yU-theme) ← The booky is on the tablex be.at′(xlocative,yU-theme) ← Harryy is at the ofﬁcex Note the unusual sequence in the logical structure that has the surface subject and undergoer coming last. This is acceptable because the semantic metalanguage supposedly applies to all languages and cannot therefore take surface word order into account. Argument positions in the semantic representation of the sentence are

keith allan

4

mapped to their positions in the syntactic representation via ‘linking theory’. Of more concern is that be.on′ and be.at′ would appear to be decomposable, cf. examples like [StateBE([ThingBARRY, [PlaceAT([ThingLATROBE])] in Jackendoﬀ (1983, 1990). Identiﬁcationals with a complement NP, such as (7), would usually be analyzed along the lines of (8) which, in eﬀect, locates Bob among the class of lawyers, cf. Gruber (1965, 1967), Anderson (1971), Jackendoﬀ (1983, 1990). (7) Bobx is a lawyery (8) be′(xtheme,ylocative) E.g. Jackendoﬀ (1983:194) has (9): (9) [StateBEIdent([Thing-tokenELISE], [Place ATIdent([Thing-typePIANIST])])] ← Elise is a pianist Since Schwartz (1993), the RRG analysis has the attribute of being a lawyer located at Bob. More generally, the attribute of being y is located at x, cf. (10). (10) be′(xU-locative,ytheme) The same analysis is given to attributive complements such as (11). (11) be′(xU-locative,ytheme) ← Paolox is happyy (11) is diﬀerent from Foley and Van Valin (1984:47), where tall′(x) and sick′(x) are described as one place predicates whose argument is a patient. In most formal semantic accounts, happy or tall would be treated as a predicate that denotes a set of entities of whom the predicate holds true (a predicate is function from entities in the domain to truth values—at least in declarative sentences): thus, in (11), Paolo would be a member of the set of tall entities. There is an obvious parallel with the usual (= other than RRG) analysis of (7). Why is the current RRG analysis of these copular sentences so very diﬀerent? Schwartz argues that copular clauses with locatives and themes contain two thematic roles, but only one role is linked to an argument position in the syntax of some languages[. . . . T]hese constructions [. . .] should be expected to behave consistently cross-linguistically with respect to the diagnostic of objectivity like concrete locative constructions and existential constructions such as a man is in the house and there is a man in the house respectively, which would have the same thematic structures. —However, in Italian, both concrete locatives and existential constructions have objective syntax, as shown in the examples in (23)–(25).

the semantic frames of be and have

5

(23) Molti turisti sono a Roma. many tourists are in Rome ‘Many tourists are in Rome.’’ (24) Ci sono multi turisti a Roma there are many tourists in Rome ‘There are many tourists in Rome.’’ (25) Ci ne sono stati molti alla festa there of-them were4 been many at the party ‘There were many of them at the party.’’ Example (23) shows a locative structure, which selects the auxiliary essere, like objectives. This can be inverted in an existential construction as in (24), and shows that the inverted ‘‘subject’’ can occur before the locative complement, a diagnostic of objectives in Burzio’s [1986] analysis. However, (25) shows that these constructions allow ne-insertion, unlike attributives and identiﬁcationals, demonstrating that attributives and identiﬁcationals do not consistently pattern cross-linguistically like concrete locatives. (Schwartz 1993:442 [Her numbering])

BE

H

Schwartz provides additional evidence in support of her position from Dakota, French, and Russian. The Schwartz analysis of (7) was given in (10), and there is a certain similarity with the Montagovian notion of ‘individual sublimation’ (cf. Dowty, Wall and Peters 1981:221ﬀ, Cann 1993:173f), i. e. an individual is deﬁned as the intersection of the set of all predications of it that give rise to true propositions, cf. Fig. 2 (note that this is consistent with the traditional analysis). At least with respect to proper names, there are insuperable problems with this Fregean notion, cf. Kripke (1972), Allan (1995). What Schwartz does is turn the

RIC

SP

EC

TA CL

ED

LAWYERS

ER

EN

AM

M

IC

AN S

BOB

Figure 2. Individual sublimation

keith allan

6 t 〈e,t〉 〈e,〈e,t〉〉

[be′+y](x) e

e

be′+y be′

xlocative ytheme

Figure 4. Attributive/identiﬁcational predicate creation in RRG

Montagovian deﬁnition of the individual on its head by presupposing the individual and locating the set of sets of properties at it. We can interpret Schwartz’s idea formally as (12), where b=Bob. (12) λ-converts to (13), and (13)≡(7). (12) λP[∨P(b)](∧lawyer′) (13) ∨∧lawyer′(b) This process corresponds to the RRG rule of ‘attributive/identiﬁcational predicate creation’ (Van Valin (1990:234), Schwartz (1993:447)) that incorporates the theme into the predicate. Although details of the procedure and of the formal properties of the resulting predicate are not speciﬁed in the RRG literature, I assume they are: Step 1. be′(xU-locative,ytheme) Step 2. Apply attributive/identiﬁcational predicate creation Step 3. [be′+ytheme](xU-locative), e.g. be. a. lawyer′(b) The eﬀect is to arrive at a structure reminiscent of a translation of (7) into the predicate calculus, where such propositions are treated as containing one-place predicates. The process is comparable with a categorial tree in Fig. 4 for an identity statement like Nimoyx is Spocky. Given the Montagovian characteristics of RRG’s metalanguage, this is not too surprising; however, outside of a formal theory of semantic types this proposal must be judged ad hoc—especially since the same compositional procedure is not adopted for other two place predicates. Even if we accept the attributive/identiﬁcational predicate creation rule, the question remains: Why should this be theme incorporation rather than locative incorporation? One can point to verbs like buy and sell that incorporate the theme ‘‘money’’ or ingestion verbs like eat and drink which optionally incorporate their patients ‘‘food’’ and ‘‘drink’’ respectively (cf. Jackendoﬀ 1990). Furthermore, verb+theme compounds exist, e.g. (14–15). (14) theme-head V -er ← bookgiver ‘‘someone who gives books’’ (15) patient-head V -er ← platebreaker

the semantic frames of be and have

7

But (16–17) don’t compute: (16) locative-head V -er (17) experiencer-head V -er A Fredgiver would not be ‘‘someone who gives to Freds’’ but ‘‘someone who gives Freds’’—in which ‘Freds’ is a theme. The incorporation of theme works for a verb like butter in (18), which—if it is synonymous with (19)—has the logical structure in (20), in which the ‘∅’ argument can be understood as ‘‘something’’, cf. Wilkins and Van Valin (1993), Van Valin and LaPolla (1996). (18) Harryx buttered the toasty. (19) Harryx put butterz on the toasty. (20) [do′(xA,∅)] CAUSE [BECOME be. on′(ylocative, ztheme)] This creates a problem for MR assignment, because in (18) ‘the toast’ is undoubtedly the undergoer and can be subject of a passive sentence: (21) The toastU was buttered. Yet as we see from (20), ‘the toast’ holds the thematic role of locative. This ability of locatives (including recipients) to function as undergoers even when there is an actor in the sentence may well account for the acceptability of sentences such as those in (22) in many dialects of English (including this author’s). (22) [The concertU] was listened to avidly by [thousands of peopleA]. [The foodU] was tasted by [all the kidsA] one after the other. [The bookU] had been written in. [This plantU] has been trodden on. [A paintingU] is meant to be looked at. [That bedU] was slept in by the late Princess Diana of Wales. Also, the verb remove can have an implicit source, and the verb travel has an implicit path, and so forth. No satisfactory account of these facts is yet oﬀered by proponents of RRG—though see Jolly 1993 and references cited there. Foley and Van Valin (1984) and Van Valin (1993a) identify (23) as a ‘condition’ (presumably a resultant condition). (23) broken′(xU-patient) ← The platex is broken It is by no means clear how conditions diﬀer from attributions which are analyzed as two place predicates. Why is the primitive verb in (23) merely broken′ and not be. broken′, like the states be. on′, be. at′? (23) is not analyzed as (10) —be′(xU-locative,ytheme)—like (11) (Paolox is happyy) is, because: (i) x is a prototypical patient; and (ii) broken′ derives from the verb break. Obviously, semantic be′ is

keith allan

8

only partly similar to English be. So, is the analysis of (11) correct? We can distinguish Paolo is happy as a habitual characteristic, for which (10) does seem the correct analysis, from Paolo is happy as a temporary state resulting from some event,5 in which case happiness is an achievement with the semantics in (24): (24) [happen′(∅)] CAUSE [BECOME be′(xU,y)] The diﬀerence between (11) and (23) cannot be explained away on the reasoning that NP complements of the copula are arguments, but adjectives and past participles are predicates. What is to be done with a copula like feel as in I feel a fool, I feel so stupid to which the attributive/identiﬁcational-predicate-creation rule in its currrent formulation does not apply as it would to I am a fool, I am so stupid? The RRG analysis of these stative constructions containing copula complements (Adj, Pp, NP, and even PP) leaves many unanswered questions.

3. Possession Possession in RRG is represented by (25) and (26) (Van Valin (1993a), Van Valin and LaPolla (1996)). (25) have′(xA-locative,yU-theme) ← Wendyx has a BMWy owns (26) have′(xlocative,yU-theme)

← The BMWy belongs to Wendyx

An objection to (25) is that own passivizes whereas possessive have does not; the latter has to be indicated in the lexical entry for have as a diﬀerence between semantic and syntactic valence. Furthermore, putting aside a diﬀerence in deﬁniteness in the theme, (27) and (28) have complementary semantic structures with diﬀerent arguments proﬁled; cf. Lyons (1967). (27) The dog is Jack’s. (28) Jack has a dog. Their semantic structures have identical thematic roles diﬀerently conﬁgured, resulting in the diﬀerent lexical verbs and concomitant proﬁling: (29) proﬁles the theme as trajector and requires semantic and lexical be; (30) proﬁles the possessor as trajector and requires have. (29) be′(xU-theme,ypossessor) (30) have′(yU-possessor,xtheme)

←The dogx is Jack’sy ←Jacky has a dogx

The ﬁrst argument of have′ is labelled ‘possessor’, which is no more unsatisfactory than the alternatives ‘holder’ and ‘haver’;6 it is more speciﬁc, and therefore prefera-

the semantic frames of be and have

9

ble to ‘locative’ (though it is a kind of locative). It seems that (27) asymmetrically entails (28) and therefore that (29) asymmetrically entails (30); but this would not always be the case. For instance, (31)–(33) either do not have an alternate with a proﬁled theme or else the alternate with be seems comparatively unnatural; in (34) the asymmetric entailment goes the opposite way from (27–28), and in (35) the entailments are symmetric. (In 34–35 the theme is a landmark, existential there being the trajector.) (31) (32) (33) (34) (35)

Harry has a cold ?→ The cold is Harry’s Sally has a pretty face ?→ The pretty face is Sally’s Sydney has a nice climate ?→ There is a nice climate in Sydney. The dog has ﬂeas ?→ There are ﬂeas on the dog. Sydney has the Opera House, Paris the Eiﬀel Tower ≡ There is the Opera House in Sydney, the Eiﬀel Tower in Paris.

Compare (29–30) with a revised semantics for two other predicates of possession, own and belong. Jacky owns a dogx (36) own′(yA-possessor,xU-theme) ← The dogx is owned by Jacky (37) belong′(xUtheme,[to′(ypossessor)])

← The dogx belongs to Jacky

Contrast (36–37) with (25–26). Belong to might look like a phrasal verb, but ‘to’ acts like a preposition because it can be preposed in a wh- question: To whom does this belong?, cf. To whom did he give it? *Up what did he wash? and With whom did you put up for ten years? ≠ You put up with who for ten years? Despite their formal diﬀerences, (36) and (37) are intuitively synonymous, and they asymmetrically entail (29) and (30). The asymmetry arises from the facts demonstrated in (38–40). (38) (a) I own an apartment in Manhattan → I have an apartment in Manhattan (b) I have an apartment in Manhattan, but I don’t own it, I rent it. *Harry owns a cold *The cold belongs to Harry

(39)

Harry has a cold → /

(40)

Sally has a pretty face → / ?*Sally owns a pretty face

(29) proﬁles the main clause theme as do the be sentences (41–45)—a diﬀerence is that in (29) the theme is trajector, but in (41–44) the landmark. ‘loc’=locative, ‘poss’=possessor, ‘ben’=beneﬁciary.

10

keith allan (41) be′(xU-theme,[in′(xtheme,yloc)]loc) ← There’s the Opera Housex in Sydneyy [the Eiﬀel Tower in Paris] (42) be′(xU-theme,[on′(xtheme,yloc)]loc) ← There’s a wreathx on the doory (43) be′(xU-theme,[in′(xtheme,yloc)]loc) ← There’s a spiderx in the bathy (44) be′(xU-theme,[on′(xtheme,[loc[have′(yposs,ztheme)]loc])]loc) ← There’s a ﬂyx on Sue’sy nosez

(45) be′([U[for′(xtheme,yben)]U-theme],[with′(xtheme,zposs)]loc) ← The bookx for Samy is with Fredz Notice the complexity of the locative argument of be′ in (41–45) and of its theme in (45). (46)–(50) correspond to (41)–(45) with the diﬀerence that they are have sentences that proﬁle as trajector the ‘possessor’ (which, remember, is a kind of locative). (46) have′(yU-poss,xtheme) ← Sydneyy has the Opera Housex, [Paris the Eiﬀel ← Tower] (47) have′(yU-poss,[on′(xtheme,yloc)]theme)

← The doory has a wreathx on ity

(48) have′(yU-poss,[in′(xtheme,yloc)]theme)

← The bathy has a spiderx in ity

(49) have′(yU-poss,[on′(xtheme,[loc[have′(yposs,ztheme)]loc])]theme) ← Suey has a ﬂyx on her nosez (50) have′(zU-poss,[for′(xtheme,yben)]theme) ← Fredz has the bookx for Samy The diﬀerence between (46) and (47–50) is perhaps explained by the alienability of the theme in (47–50) (cf. Chappell and McGregor (1996)). The grammar of alienability is culture speciﬁc, and in English it may well be deﬁned by the diﬀerence between inalienable have′(yU-poss,xtheme) and alienable have′(yU-poss,[locative(xtheme, yloc)]theme). A rough test is that inalienable (29–30) validate reference to [NP NP’spossessor Ntheme NP] i. e. Jack’s dog (29–30, 36–37), Harry’s cold (31, 39), Sally’s pretty face (32, 40) Sydney’s climate (33), the dog’s ﬂeas (34), Sydney’s Opera House (35, 41, 46), my apartment (38(a-b)). Alienable (47–50) do not validate respective reference to ??the door’s wreath, ??the bath’s spider, ??Sue’s ﬂy, or even ??Fred’s book—though the last is controversial. The inalienability of have′(yU-poss,xtheme) may explain the diﬀerence between (51)–(52). gave (51) Bill Fred the book. sent (52) *John put the table the book. In current RRG theory (51) presents a problem.

the semantic frames of be and have

11

4. Recipient and theme in ditransitives: which is the undergoer? In RRG, when there is an actor, the MR of undergoer is usually chosen on the basis of the rightmost member in the hierarchy in (53). (53) pred(x,. . . < pred(. . . ,y) < pred(x) When there is no actor, the sole MR will come from the leftmost member. This correlates well with the sequence of thematic roles in (54): (54) experiencer—locative—theme—patient Typical undergoers holding the speciﬁed argument in (53) are: pred(x, . . .) pred(. . . ,y)

pred(x)

location, perceiver, cognizer, attributant attribute, target of emotion/desire, stimulus, sensation, theme, object of knowledge/belief, possessed, performance, consumed, creation, implement, locus patient, existant

An exception to the choice of undergoer (cf. Van Valin (1993a:44)) is a three place predicate in English with the recipient immediately following the verb, as in (51). Nevertheless, in RRG the theme is the unmarked choice for undergoer in the logical structure of give (a hypothesis that I shall reject): (55) [do′(xA,∅)] CAUSE [BECOME have′(yposs,zU-theme)] ← [BillA] gave [FredU-recipient] [the booktheme]. Similarly with the accomplishment verb show in (56), where the experiencer ‘Sam’ is undergoer, despite the fact that the theme is the unmarked choice for undergoer in the logical structure in (57). (56) Maryx showed Samy her treasurez. (57) [do′(xA,∅)] CAUSE [BECOME see′(yexperiencer,zU-theme)] There is an implicit assumption that (51) (with give) and (56) are the marked alternatives for (58) and (59) respectively; but this is controversial, cf. Dryer (1986), Goldberg (1995). (58) BillA gave the bookU to Fred. (59) MaryA showed her treasureU to Sam. Van Valin gives three reasons for supposing that (58) and (59) are the base structures from which (51) and (56) derive. One is that English three place predicates like donate, announce and put have the theme as undergoer. Another is the fact that there is often a closer relationship between a predicate and the most typical

12

keith allan

undergoers, theme or patient, than there is between a predicate and other thematic roles. Recall the verb+theme compounds discussed in respect of (14–21) above; and, if we accept it, the rule of ‘attributive/identiﬁcational predicate creation’ that incorporates the theme into the predicate (cf. the discussion of (10)). A third arises from the diﬀerent consequences of questioning the arguments ‘Fred’ or ‘Sam’ and ‘the book’ or ‘the treasure’. Compare (60) with (61), bearing in mind that experiencers have a lot in common with animate locatives such as possessors and recipients: (60) (a) Whattheme did Bill give to Fred? Whattheme did Mary show to Sam? (b) Whattheme did Bill give Fred? Whattheme did Mary show Sam? (61) (a) Whorecipient did Bill give the book to? Whoexp did Mary show her treasure to? (b) ?*Whorecipient did Bill give the book? ?*Whoexp did Mary show her treasure? Similar conditions apply even to phrasal verbs like give NP a kick / polish, cf. (62): (62) (a) Max gave the table a polish. Max gave Sam a kick. (cf. 51, 56) (b) ??Max gave a polish to the table. ?*Max gave a kick to Sam. (cf. 58, 59) (c) What did Max give a polish to? Who did Max give a kick to? (cf. 61(a)) (d) ??What did Max give a polish? ??Who did Max give a kick? (cf. 61(b)) The preferred forms of question use the verbs kick and polish, e. g Who did Max kick, but the preference for marking the locative (including recipient) is nonetheless there. According to Van Valin (1993a:45), the most plausible solution is to postulate exceptions to the markedness constraint in the actor−undergoer hierarchy and let the recipient or experiencer outrank the theme in competition for the macrorole of undergoer with a few such verbs; presumably motivated by the personal and/or animacy hierarchy (Allan (1987), following many others). The preceding arguments are further supported from languages in which there is no controversy about a theme undergoer in the logical structure of three place predicates. So should we accept Van Valin’s solution for English ditransitives? Given the

the semantic frames of be and have

13

revised analysis to the semantic frames of be′ and have′ presented in this chapter, there is a better solution. For easy reference I copy (51)–(52) and add (63) and (64). gave (51) Billx Fredy the bookz. sent gave (63) Billx the bookz to Fredy. sent (52) *John put the table the book. (64) Johnx put the booky on the tablez. Suppose that: • (51) has the semantics in (51′) in which the recipient ‘Fred’ is the proﬁled object. • (63) has the semantics in (63′) in which the relocation of proﬁled theme ‘the book’ is indicated by go′; subscripts are introduced to clarify the difference between the locatives: loc identiﬁes the path locative, loc identiﬁes Π Σ the source location, loc identiﬁes the goal location. Γ • (64) has the semantics in (64′) in which the theme ‘the book’ is proﬁled. (51′) [do′(xA,∅)] CAUSE [BECOME have′(yU-poss,ztheme)] (63′) [do′(xA,∅)] CAUSE [go′(zU-theme,[loc from′(xloc ),[to′(yloc )]loc ])] Π Σ Γ Π (64′) [do′(xA,[take′(x,y)])] CAUSE [BECOME be′(yU-theme,[on′(y,zloc)]loc)] (64′) is surely the least controversial. (51′) is a revision of (55) to make it consistent with the analysis of have′ presented in (46–50) and with the facts concerning sentences containing ditransitives like those in (51). (63′) is consistent with both (51′) and (64′) in that the ﬁrst argument of Ψ in the structure [Φ CAUSE Ψ] is proﬁled as landmark in the main surface clause, and the dynamic relocation of the theme accounts for the occurrence of semantic to′ and lexical to. One justiﬁcation for the introduction of go′ is the answer in (65): (65) [In a discussion about who got what presents or bequests:] Q: Who got the book? A: The book went to Fred. Precedents for (63′) are examples like (66–67) in Jackendoﬀ (1990) (CS+ symbolizes ‘‘CAUSE with a successful outcome’’; GOPoss indicates change in possession; AFF indicates what Jackendoﬀ calls the ‘action tier’ which often corresponds to the MR in RRG). (66)

FROM ([α]) CS + ([HARRY]α, GOPoss ([BOOK], TO([SAM]β) AFF([α],[β]) Event ← Harry gave the book to Sam

keith allan

14 (67)

GOPoss([Thing BOOK], [TOPath ([Thing SAM]α)])] AFF([ ], [α]) ← Sam received a book (unspeciﬁed actor) Event

One objection to (63′) is the fact that it fails to distinguish between give and send; however, this can be rectiﬁed by replacing the ‘∅’ of [do′(xA,∅)] with a dissimilating speciﬁcation of the semantics of send. Another objection is that (51′) and (63′) seem too diﬀerent to be representations of the semantics of identical lexical verbs. This objection can be put aside as a confusion between forms in the object language and forms in the semantic metalanguage.7 Furthermore, the complex locative argument of go′—[loc from′(xloc ),[to′(yloc )]loc ]—is an abbreviation for (68), in which Π Σ Γ Π α<β symbolizes ‘‘α precedes β in time and space’’. (68) [locΠ[αbe′(ztheme,[at′(ztheme,xlocΣ)]loc)α] < [βBECOME [be′(ztheme,[at′(ztheme,ylocΓ)]loc)]β]locΠ] If we recognize that the possessor role is one kind of locative, and we expand (63′) to (69), taking into account the relationship already proposed between be and possessive have, the relationship between (51′) and (69) (and by implication, (51′) and (63′)) is much more obvious. (69) [Φdo′(xA,∅)Φ] CAUSE [Ψ[go′(zU-theme,[locΠ[αbe′(ztheme, [at′(ztheme,xlocΣ)]loc)α] < [βBECOME [be′(ztheme,[at′(ztheme,ylocΓ)]loc)]β]locΠ])]Ψ] It is also obvious that (63′) and (69) asymmetrically entail (51′), cf. (70). (70) [do′(xA,∅)] CAUSE [go′(zU-theme,[loc[from′(xloc1),[to′(yloc2)]loc])] → [do′(xA,∅)] CAUSE [BECOME have′(yU-poss,ztheme)] If these analyses are correct, (63), in which the proﬁled object is the theme, is grammatically prior to (51) in which the recipient/possessor is the proﬁled object. To what extent the analysis generalizes to other ditransitives awaits further research. With respect to the buttering of toast (18–21), (64′) would be the basis for (19) and also (71). (19) Harry put butter on the toast. (71) Jim put the book on the shelf. From both of these, the theme may be the subject in a passive sentence. As we would expect from the discussion above, (72–73) are perfectly good sentences:

the semantic frames of be and have

15

(72) The toast has butter on it. (73) The shelf has a(t least one) book on it. If (18) derives from (64′), then so should (74); but it doesn’t. Again, if (75) derives from (64′) so should (76); which it doesn’t. (18) (74) (75) (76)

Harryx buttered the toasty. Jimx booked the shelfy. Jimx shelved the booky. Harryx toasted the buttery.

I would suggest that (18) and (74–76), along with their passive counterparts, do not derive from (63′) at all, but rather: (18′) (74′) (75′) (76′)

[do′(xA,[butter′(x,yU)]) [do′(xA,[book′(x,yU)]) [do′(xA,[shelve′(x,yU)]) [do′(xA,[toast′(x,yU)])8

The lexical diﬀerence between the typically isomorphic noun and verb can be captured by comparing e.g. (77) with (18′), or (78) with (75′): (77) butter′(x) (78) shelf′(x)

5. Conclusion The analysis presented in this chapter was inspired by certain problems that arise in the logical structures proposed for English verbs be, have, own, and give by Robert D. Van Valin Jr and Linda Schwartz within RRG. I argued that possessive sentences with be and have contain the same thematic roles diﬀerently proﬁled: be proﬁles the theme, and have the possessor. A similar idea is to be found in Lyons (1967), and it was not new then. Contrary to RRG, I proposed that, in attributive and classiﬁcatory sentences too, be proﬁles the theme and not the locative. This is the traditional analysis. In (79–80) the logical structures match superﬁcial form by being ambiguous between characteristic happiness and a temporary state. That semantic distinction will be indicated by other means. (79) be′(xU,[happy′(x)]) ← Paolox is happy (80) be′(xU,[a x: λy[man′(y) & happy′(y)](x)])← Paolox is a happy man We do not have space to discuss the internal structure of complement NP in (78) (see Allan (forthcoming, ms)), but note that in this kind of copular sentence the

16

keith allan

second argument of be′ is partially identical with the ﬁrst. Own has the same thematic roles as be and have but two macroroles instead of just one. The semantic frames proposed for be and possessive have proved consistent with the hypothesis that there are two logical structures for ditransitives like give thus disposing of the argument over whether give NPtheme to NPrecipient derives from give NPrecipient NPtheme or vice versa. However, it was shown that the former semantically entails the latter.

Notes 1. I am deeply grateful to Robert D Van Valin Jr for discussing with me an earlier draft of this paper. However, I must take full responsibility for all remaining faults. 2. For the diﬀerence between DO and do′, see Wilkins and Van Valin (1993), Van Valin and LaPolla (1996). 3. ‘A → B’ and ‘B ← A’ symbolize ‘‘A (asymmetrically) semantically (logically) entails B’’; ‘A≡B’ means ‘‘A is semantically equivalent to B’’ (mutual semantic entailment). 4. Should be translated ‘‘are’’ not ‘‘were’’. 5. In Spanish the habitual characteristic would be expressed using the verb ser whereas the temporary state would be expressed with estar. 6. According to Meillet (1924), the various verbs for have′ in Indo-European languages all derive from verbs meaning ‘‘hold’’ or ‘‘take’’. 7. There is discussion of my views on the relationship between object language and metalanguage in Allan (ms). 8. The normal RRG analysis would probably be [do′(xA,∅)] CAUSE [BECOME toasted′(yU)], etc. I ﬁnd it counterintuitive that the past participle is taken to be the most primitive form of the verb.

References Allan, Keith. 1987. ‘‘Hierarchies and the Choice of Left Conjuncts (With Particular Attention to English).’’ Journal of Linguistics 23:51–77 ——. 1995. ‘‘What Names Tell About the Lexicon and the Enyclopedia.’’ Lexicology 1:280– 325. ——. Forthcoming. ‘‘The Semantics of English Quantiﬁers.’’ In Peter Collins and David Lee (eds.), The Clause in English: In Honour of Rodney Huddleston. Amsterdam and Philadelphia: John Benjamins. ——. ms. Natural Language Semantics: The study of meaning in human languages. Anderson, John M. 1971. A Grammar of Case: Towards a localistic theory. Cambridge: Cambridge University Press. Blake, Barry J. 1977. Case Marking in Australian Languages. Canberra: Australian Institute of Aboriginal Studies. Burzio, Luigi. 1986. Italian Syntax. A Government and Binding Approach. Dordrecht: Reidel.

the semantic frames of be and have

17

Cann, Ronnie. 1993. Formal Semantics. Cambridge: Cambridge University Press. Chappell, Hilary and William McGregor (eds). 1996. The Grammar of Inalienability: A Typological Perspective on Body Part Terms and the Part-Whole Relation. Berlin and New York: Mouton de Gruyter. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Dillon, George. 1977. Introduction to Contemporary Linguistic Semantics. Englewood Cliﬀs: Prentice-Hall. Dowty, David R. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel. —— 1991. ‘‘Thematic Proto-Roles and Argument Selection.’’ Language 67: 547–619 ——, Robert E. Wall and Stanley Peters. 1981. Introduction to Montague Semantics. Dordrecht: Reidel. Dryer, Matthew. 1986. ‘‘Primary Objects, Secondary Objects, and Antidative.’’ Language 62: 808–45. Fillmore, Charles J. 1968. ‘‘The Case for Case.’’ In Bach and Harms (eds), 1-88. ——. 1977. ‘‘The Case for Case Reopened.’’ Syntax and Semantics 8: Grammatical Relations. In Peter Cole and Jerry Sadock (eds), 59–82. Foley, William A. and Robert D. Van Valin Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Gruber, Jeﬀrey S. 1965. ‘‘Studies in Lexical Relations.’’ Ph.D. Thesis, Massachusetts Institute of Technology. Published by the Indiana University Linguistics Club (Bloomington) in 1970. Revised version in Lexical Structures in Syntax and Semantics, 1976: 1–210. Amsterdam: North Holland. Gruber, Jeﬀrey S. 1967. ‘‘ ‘Look’ and ‘See’ ’’. Language 43: 937–47. Jackendoﬀ, Ray. 1983. Semantics and Cognition. Cambridge MA: MIT Press. ——. 1990. Semantic Structures. Cambridge MA: MIT Press. Jolly, Julia A. 1993. ‘‘Preposition Assignment in English.’’ In Van Valin (ed.) 1993b: 275– 310. Katz, Jerrold J. 1972. Semantic Theory. New York: Harper and Row. ——. 1977. Propositional Structure and Illocutionary Force. New York: Thomas Crowell. Reprinted 1980. Cambridge MA: Harvard University Press. Landau, Barbara and Ray Jackendoﬀ. 1993. ‘‘ ‘What’ and ‘Where’ in Spatial Language and Spatial Cognition.’’ Behavioral and Brain Sciences 16: 217–65. Kripke, Saul. 1972. ‘‘Naming and necessity.’’ In Donald Davidson and Gilbert Harman (eds), Semantics of Natural Language. Dordrecht: Reidel, 253–355. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Vol. 1, Theoretical Prerequisites. Stanford: Stanford University Press. ——. 1990. Concept, Image and Symbol: The Cognitive Basis of Grammar. Cambridge: Cambridge University Press. Lyons, John. 1967. ‘‘A Note on Possessive, Existential and Locative Sentences.’’ Foundations of Language 3: 390–96.

18

keith allan

Marantz, Alec. 1984. On the Nature of Grammatical Relations. Cambridge MA: MIT Press. Meillet, Antoine. 1924. ‘‘Le développment du verbe ‘avoir’.’’ Antidoron: Festschrift Jacob Wackernagel. Göttingen, 9–13. Schwartz, Linda. 1993. ‘‘On the Syntactic and Semantic Alignment of Attributive and Identiﬁcational Constructions.’’ In Van Valin (ed.) 1993b: 433–63. Tesnière, Louis. 1959. Éléments de Syntaxe Structurale. Paris: Klincksieck. Van Valin, Robert D. Jr. 1990. ‘‘Semantic parameters of split intransitivity.’’ Language 66: 221–60. ——. 1993a. ‘‘A Synopsis of Role and Reference Grammar.’’ In Van Valin (ed.) 1993b: 1–164. ——. (ed.) 1993b. Advances in Role and Reference Grammar. Amsterdam: John Benjamins. ——and Randy LaPolla. 1996. Chapters 3 and 4 from Syntax: Structure, Meaning, and Function. A revised version published by Cambridge University Press (Cambridge) 1997. ——and David P. Wilkins. 1993. ‘‘Predicting Syntactic Structure from Semantic Representations: Remember in English and its equivalents in Mparntwe Arrernte.’’ In Van Valin (ed.) 1993b, 499–534. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca: Cornell University Press. Wilkins, Wendy (ed.) 1988. ‘‘Thematic Relation.’’ San Diego: Academic Press [Syntax and Semantics 21]. Wilkins, David P. and Robert D. Van Valin Jr. 1993. The Case for Case Reopened: Agents and agency revisited. Technical Report 93–2. Buﬀalo: Center for Cognitive Science. Abbreviated version ‘‘The case for ‘eﬀector’: Case roles, agents, and agency revisited’’. In Masayoshi Shibatani and Sandra A. Thompson (eds), Grammatical Constructions. Oxford: Oxford University Press. 1996, 289–322.

‘Crow is sitting chasing them’ Grammaticisation and the verb ‘to sit’ in the Mantharta languages, Western Australia PETER AUSTIN Department of Linguistics and Applied Linguistics University of Melbourne

1. Introduction1 In the Mantharta languages, traditionally spoken in the north-west of Western Australia, we ﬁnd the verb kumpa- ‘to sit’ used in an interesting type of construction. Consider the following sentence from a traditional mythological text in Jiwarli (Austin 1997):2 (1) Wakurra yana-nyja kumpa-yi pirru-wu-rru thika-rnu crow.nom go-pst sit-purp:ss meat-dat-now eat-impf:ss ngunha pirru wirntupinya-nyjaparnti pularntura-nyjarri-yi that.acc meat.acc kill-perf:ds type of lizard-pl-dat yanga-rnu kumpa-yi. chase-impf:ss sit-purp:ss ‘Crow went and ate the meat that had been killed, and chased pularntura lizards.’ [T43s51] In examples such as these, the verb kumpa- does not carry its regular lexical semantic meaning of ‘to sit’ but rather has a de-semanticised, almost auxiliary verb-like function. In this paper we will examine the syntax and semantics of this type of construction, and suggest a possible path of grammaticisation for the ‘‘sit’’ verb. We also look at similar constructions in other Australian languages, and elsewhere.

peter austin

20

2. The Mantharta languages The Mantharta languages are a group of Australian Aboriginal languages traditionally spoken in the north-west of Western Australia, along the Lyons and the upper Ashburton Rivers inland from the town of Carnarvon (see maps in Austin 1981b, 1988b, 1992). They comprise Jiwarli, Thiin, Warriyangka and Tharrkari (see Austin 1981b, 1988a). The languages show a high degree of lexical cognacy and appear syntactically to be identical in all major respects; in this paper examples will be presented from Jiwarli, the language for which we have the greatest amount of data (Austin 1997). The Mantharta subgroup is apparently fairly closely related to its western neighbours Payungu, Pinikura, Purduna and Thalanyji (members of the Kanyara subgroup). Among the more distant relatives of the Mantharta languages are the Ngayarta subgroup to its immediate north, including Martuthunira (Dench 1995), Panyjima (Dench 1991) and Yinyjiparnti (Wordick 1982). For further discussion of these genetic relationships see Austin (1996). Morphologically, the Mantharta languages show a rich system of nominal case marking of the split-ergative type (see Dixon 1979, 1994, Silverstein 1976, Austin 1995); the distribution of formal marking is determined by inherent lexical content (animacy) of the marked nominal. In Jiwarli: (i) the ﬁrst person singular pronoun ngatha (and optionally the second person pronoun nhurra) syncretises on a nominative/accusative pattern, i.e. there is one form for intransitive and transitive subject functions (abbreviated following Dixon 1979, 1994 as S and A respectively), and a diﬀerent form for transitive object (O) function; (ii) inanimate nominals and demonstratives syncretise on an ergative/absolutive pattern, i.e. there is one form for A function, and a diﬀerent form for S and O functions; (iii) all other nominals have three separate forms for A, S, and O functions. The following table shows examples illustrating this case form syncretism: Function ergative (A) nominative (S) accusative (O)

‘I’ ngatha ngatha ngathanha

‘we’ ngalilu ngali ngalinha

‘dog’ thuthungku thuthu thuthunha

‘sun’ jurungku juru juru

‘that’ ngulu ngunha ngunha

In any particular construction, by comparison across category types it is always possible to identify which function a given nominal is serving. Thus, in the morpheme-by-morpheme glosses a certain form may be glossed diﬀerently depending on the syntactic context, eg. ngatha is glossed as ‘I.erg’ or ‘I.nom’ depending on its function in the particular clause. Notice also that in Jiwarli all nominals bear case

grammaticisation and the verb ‘to sit’

21

regardless of whether they are adjacent or separated (forming discontinuous expressions). Case is formally marked locally depending on the animacy of the nominal referent. In addition to the core cases, there are cases with semantic functions: dative, locative, allative, ablative and causal (see Austin 1994 for details). Additionally, certain adnominal and adverbial modiﬁers, especially possessives, and ablatives, are marked twice for case, taking both their own case (such as dative, marking possession) and a further case marker in agreement with the modiﬁed head nominal (see Austin 1995 for discussion and examples). The verbal system of the Mantharta languages also shows morphological complexity. There are ﬁve verb conjugations, with a major distinction between inﬂectional aﬃxes attached to main clause verbs and those suﬃxed to dependent clause verbs. Main clause inﬂections mark tense-mood categories while dependent clause inﬂections code the type of dependency plus switch-reference (identity or non-identity of reference between the subjects in the two clauses—see Austin 1981c). For the topic of this paper, an important inﬂectional category is the imperfective (impf: ss—imperfective-same subject, and impf:ds—imperfective-diﬀerent subject) which codes relative present tense and imperfective aspect, ie. a situation occurring at the same time as the main clause situation and not yet a completed whole. These usually translate into English as relative clauses, or as gerundive or ‘while/when’ adverbial clauses. The same subject inﬂection has the allomorphs -rnu ~ -nhu ~ ngu depending on conjugation, and occurs on the dependent verb in construction with kumpa- ‘sit’ in the examples discussed here. Notice that there is also an interaction between dependent clause type and case marking: the transitive object of imperfective (and perfective) dependent clauses is placed in the dative case, regardless of animacy (ie. the usual main clause case coding outlined above is suspended and all O NPs take dative case). This explains why in example (1) above we ﬁnd pirru-wu ‘meat-dative’ and pularntura-nyjarri-yi ‘pularntura lizards-dative’ for the transitive objects of the two clauses. Syntactically, the Mantharta languages are of the non-conﬁgurational type (Austin and Bresnan 1996) with extremely free word order, lack of any evidence for a verb phrase constituent, and free omission of nominal arguments when reference is clear from the context. Interestingly, there are word order correlations peculiar to the kumpa- construction which are discussed in detail below.

3. The verb ‘to sit’ In Jiwarli and its close relatives the verb kumpa- ‘to sit’ is used as a main verb with a spatial locative meaning that can be translated into English as ‘to sit, camp, stay, live, be’, as in the following examples:3

22

peter austin A1. with an animate subject meaning ‘to live’: (2) Yawartamayi ngunha-pa nganthurra-mpa papu-thu Yawartamayi:nom that:nom-spec 1pl-dat:nom father:nom-top kumpa-inha yirrara-rru jirnti-ngka-rru. live-prs above:loc-now sky-loc-now ‘Yawartamayi our father lives up above in the sky.’ [T47s18] A2. with an animate subject meaning ‘to camp’, ‘to stay’: (3) Kala-pa kumpa-irarringu tharratharra thana. like this-spec camp-intent separated.nom 3pl.nom ‘They would camp separated oﬀ like this.’ [T48s5] (4) Warri kumpa-irarringu kayanu-ra ngurra-ngka. not stay-intent one-loc camp-loc ‘(They) wouldn’t stay in one camp.’ [T48s6] A3. with an animate subject meaning ‘to sit, be in a seated position’ (5) Kuwarti-thu yinha kumpa-inha purluu-rru nganyi-ngka now-top this.nom sit-prs face-to-face-now mother-in-law-loc karla-ngka kumpa-yi wangka-arni-rri-a. ﬁre-loc sit-purps:ss speak-non sing-coll-purps:ss ‘Now they sit facing the mother-in-law and sit at the ﬁre talking to one another.’ [T48s14] b. with an inanimate subject also meaning ‘to be, sit’ (6) Parlu-nyjarri ngunha kumpa-inha kartumayi-lu wantha-rninyjaparnti. stone-pl.nom that.nom sit-prs Kartumayi-erg put-perf:ds ‘The rocks are there that Kartumayi put down.’ [T47s34] (7) Warrara thurnti kumpa-artu. wild potato.nom vegetable food.nom be-usit ‘There used to be wild potatoes.’ [T52s10]

The verb kumpa- can also be used as a main clause verb with a dependent imperfective-same subject clause in several functions. These are discussed in the following sections. 2.1 Kumpa- with its main lexical meaning With its main lexical meaning of ‘be located’, as in (8) to (14) below. I have examined all the examples in the Jiwarli text collection (Austin 1997) and found that:

grammaticisation and the verb ‘to sit’

23

(a) there is just one example of the impf:ss clause preceding the ‘sit’ clause; (b) there are seven examples with the imperf:ss clause following the ‘sit’ clause; three involve intransitive dependent verbs and four transitive dependent verbs: (a)–intransitive dependent verb: sit sit S sit S

Vi Vi loc Vi

eg. (8) eg. (9) eg. (10)

(b) transitive dependent verb—dependent object, if present (and marked with dative case), always precedes the dependent verb. The two verbs may be separated by other words:4 sit sit sit S sit S

O-dat Vtr loc O-dat Vtr comit O-dat Vtr loc Vtr

eg. (11) eg. (12) eg. (13) eg. (14)

1(8) Ngali kumpa-ira kunyjil-arri-ngu. 1dl:nom sit-fut warm-inchoat-imprf:ss ‘Let’s sit down and get warm.’ [T20s11] 1(9) You know ngurru-martu-la kumpa-ja ngatha-thu you know old man-pauc-loc live-pst 1sg:nom-top juma-purra-thu parnka-rri-ngu. child.nom-time-top big-inchoat-impf:ss ‘You know I lived with the old men when I was a child growing up.’ [T70s33] (10) Kumpa-ira ngunha-pa karla-ngka-rru kunyjil-arri-ngu. sit-fut that:nom-spec ﬁre-loc-now warm-inchoat-impf:ss ‘They would sit by the ﬁre getting warm.’ [T61s41] (11) Kumpa-irarri pirru-wu kampa-rnu. sit-intent meat-dat cook-impf:ss ‘(We) would sit cooking meat.’ [T47s83] (12) Walangu ngunha kumpa-ira ngunhi-rru-pa bird:nom that:nom sit-fut there:loc-now-spec papa-wu paja-rnu. water-dat drink-impfss ‘The birds will sit there and drink the water.’ [T56s14]

24

peter austin (13) Kumpa-ja ngunha-pa mara-jaka-rru ngurnu-pa sit-pst that:nom-spec hand-com:nom-now that:dat-spec kala-ma-rnu murlimurli-rnu mara-jaka. like that-caus-impf:ss stir-impf:ss hand-com:nom ‘He sat there going like that with his hand, stirring it with his hand.’ [T56s19] (14) Ngunhi-pa kumpa-artu ngatha ngurru-nyjarri-la tharla-rnu there:loc-spec stay-usit 1sg:nom old man-pl-loc feed-impf:ss papa-jaka miramana-ngu jarntil-a papa-rla. water-com:nom lead-impf:ss walking stick-loc water-all ‘I used to stay with the old people feeding them water and leading them to the water with a walking stick.’ [T61s37]

2.2 Kumpa- bleached of its lexical semantics In constructions where the main lexical meaning is lacking and kumpa- seems to be bleached of all its lexical semantics, while the syntactically dependent verb expresses the primary predicate semantics. Here kumpa- serves mainly to carry the absolute tense and often has the function of expressing a non-punctual or continuous aspect.5 The following example illustrates this nicely: (15) Ngatha kumpa-artu tharla-rnu papa-jaka. 1sg:nom sit-usit feed-impf:ss water-com ‘I used to feed (him) with water.’ [T59s5] Contrast this with the punctual verb: (16) Karla-rla-laartu ngatha ngurnu-pa-thu ngurru-wu. ﬁre-fact-usit 1sg:erg that:dat-spec-top old man-dat ‘I used to make a ﬁre for that old man.’ [T59s6] All the verbs found in the Jiwarli text data that combine in imperfective-same subject form with kumpa- in its non-lexical function are listed in the Appendix below. The majority of these are atelic verbs, and a number are formed by the addition of the stem forming aﬃxes:6 (a) -rri- inchoative verbaliser, indicating entry into a state, ‘become . . . ’; (b) -marri- collective verbaliser, indicating entry by a non-singular group into a state, ‘become . . . ’; (c) -lparri- collective detransitiviser, indicating reciprocal action done ‘to each other’.

grammaticisation and the verb ‘to sit’

25

In the Jiwarli text data there are 63 examples of this construction type, subdivided as follows: 45 examples where kumpa- precedes the imperfss verb 20 imperfss verb intransitive verbs adjacent 6 imperfss verb intransitive verbs not adjacent 12 imperfss verb transitive verbs adjacent 7 imperfss verb transitive verbs not adjacent 18 10 2 6 0

examples imperfss verb intransitive imperfss verb intransitive imperfss verb transitive imperfss verb transitive

where kumpa- follows the imperfss verb verbs adjacent verbs not adjacent verbs adjacent verbs not adjacent

In the majority case, i.e., where kumpa- precedes the imperfss verb, the overwhelming tendency is for the dependent verb and the main verb to be adjacent. When they are not adjacent a locative, temporal or connective element and/or an intransitive subject (of kumpa-) may intervene, regardless of the dependent verb’s transitivity. If a transitive object is present, in eight out of fourteen instances it follows the dependent verb, in just six examples is the dependent verb object to the left of its mother—in four of these it is also to the left of kumpa-. In the minority case, i.e., where kumpa- follows the imperfss verb, in all but two instances the two verbs are adjacent, and in both of these the dependent verb is intransitive. The clear preference then is for verb adjacency. The following table sets out the array of possibilities: Where kumpa- precedes the impf:ss verb: 1. dependent verb intransitive verbs adjacent kumpaVi verbs non-adjacent kumpatemporal Vi kumpaS Vi kumpaS loc Vi kumpaconnective S Vi 2. dependent verb transitive verbs adjacent kumpakumpaO-dat kumpaO-dat S kumpa-

Vtr Vtr O-dat Vtr —Vtr

20 2 2 1 1

5 4 2 1

peter austin

26 verbs non-adjacent kumpakumpaO-dat kumpakumpakumpakumpa-

S S kumpaO adv adv loc

Vtr Vtr S Vtr Vtr Vtr Vtr

1 O-dat 1 Vtr 1 2 1 O-dat 1 O-dat 2

Where kumpa- follows the impf:ss verb: 1. dependent verb intransitive verbs adjacent Vi kumpaverbs non-adjacent Vi S kumpa2. dependent verb transitive verbs adjacent Vtr Vtr O-dat Vtr O-dat S Vtr

kumpakumpa- O-dat kumpakumpa-

10 2

1 3 2 1

The following examples illustrate separation of the two verbs, ﬁrstly when the dependent verb is intransitive: kumpa-—S—Vi (17) Ngatha kumpa-inha parna kulypa-nhu. 1sg:nom sit-prs head:nom be sore-impf:ss ‘I have a sore head.’ [T21s1] (18) Kumpa-artu ngurru-martu thika-rnu. sit-usit old man-pauc:nom eat-impf:ss ‘The old people used to eat (it).’ [T52s25] (19) Ngurnta-nhu thanyu-rru kumpa-inha. lie-impf:ss crippled:nom-now sit-prs ‘(They) lie crippled now.’ [T47s123] (20) Wuna-rru kumpa-inha kayanu warrkamu-rri-ngu. long time-now sit-prs alone:nom work-inchoat-impf:ss ‘(I) work here alone a long time.’ [T69s22]

grammaticisation and the verb ‘to sit’

27

kumpa-—temporal—Vi (21) Kumpa-ja muntu-parnti-purra warrkamu-rri-ngu kayanu. sit-pst morning-abl2-time work-inchoat-impf:ss one:nom ‘(I) have been working alone since morning.’ [T14s3] (22) Yinha-rru kumpa-inha kuwarti kurni-ngu-rru. this-now be-prs now look for-impf:ss-now ‘This is how (I) am looking for (them) now.’ [T61s45] kumpa-—connective—S—Vi (23) Kumpa-ja ngurnu-parnti-pa-thu ngurru-nyjarri sit-pst that.dat-abl2-spec-top old man-pl.nom julyu-nyjarri kurlkanyu-rri-ngu. grey hair-pl.nom thinking-inchoat-imperf:ss ‘After that the old grey-haired men thought.’ [T65s13] The following examples show separation when the dependent verb is transitive: kumpa-—S—Vtr (24) Kumpa-ja ngunha-rru jirril-jipa-rnu walangu-wu thurupu-la. sit-pst that.nom-now fear-caus-impf:ss bird-dat trough-loc ‘He frightened the birds away from the trough.’ [T56s16] kumpa-—O—Vtr (25) Malya-rru yana-nyja-rni jali ngunha nganaju fortunate-now go-pst-hence friend:nom that.nom 1sg:dat:nom kumpa-yi warrapa-wu-rru kurti-rnu mara-ngku-rru sit-purp:ss grass-dat-now pick up-impf:ss hand-erg-now mana-ngu walyparra-la juku-rnu. get-impf:ss wheelbarrow-loc throw-impf:ss ‘It’s a good job that friend of mine came to pick up the grass in his hands and get it to throw in the wheelbarrow.’ [T14s4] (26) Wakurra yana-nyja kumpa-yi pirru-wu-rru thika-rnu ngunha crow.nom go-pst sit-purp:ss meat-dat-now eat-impf:ss that.acc pirru wirntupinya-nyjaparnti pularntura-nyjarri-yi meat.acc kill-perf:ds type of lizard-pl-dat yanga-rnu kumpa-yi. chase-impf:ss sit-purp:ss ‘Crow went and ate meat that had been killed and chased pularntura lizards.’ [T43s51]

peter austin

28 kumpa-—locative—Vtr

(27) Warlartu ngarlpurri-nyja kumpa-yi martura yanga-rnu eaglehawk.nom run-pst sit-purp:ss middle.loc chase-impf:ss pirru-nyjarri-yi-rru wayurta-wu-rru. meat-pl-dat-now opossum-dat-now ‘Eaglehawk ran and chased possums in the middle.’ [T43s59] (28) Kumpa-ja juru-ngka-kunti-rru nhanya-ngu mathan-ku sit-pst sun-loc-sembl-now look-impf:ss hill kangaroo-dat warrkalarri-ya-wu parlu-ngka yirrara. crawl-impf:ds-dat hill-loc top ‘(You) could see the hill kangaroos crawling on top of the hill as if it were daytime.’ [T65s7] (29) Kumpa-ja ngunhi-pa juru-ngka-kunti-rru kartaju-la sit-pst there.loc-spec sun-loc-sembl-now night-loc nhanya-ngu kurrpirli-yi mathan-ku parlu-ngka see-impf:ss kangaroo-dat hill kangaroo-dat hill-loc warrkalarri-ya-wu papa-rla. crawl-impf:ds-dat water-all ‘(You) could see there in the night as if it were day the kangaroos crawling on the hill towards the water.’ [T65s12] In its most extreme form this kind of structure leads to interleaving of main and dependent clause arguments, as in the following instances: dependent O—main S—dependent V—main V: (30) Minga-nyjarri-yi-rru nhurra thika-rnu kumpa-ma. ant-pl-dat-now you.nom eat-impf:ss live-imper ‘You live eating ants.’ [T40s29] (31) Parlu-nyjarri-yi ngunha kumpa-irarri mana-ngu stone-pl-dat that.nom sit-intent get-impf:ss yirrangu-wu yajina-nyjarri-yi. stone knife-dat sweet food-pl-dat ‘He got stones, stone knives, berries.’ [T66s7] dependent O—main V—main S—dependent V: (32) Wartajunu-wu kumpa-artu ngunha-pa puka-rnu. poor fellow-dat be-usit that.nom-spec visit-impf:ss ‘They used to visit the poor things.’ [T61s49]

grammaticisation and the verb ‘to sit’

29

These examples are interesting for a number of reasons. Firstly, although the number of instances is not large, we can see a certain amount of syntactic rigidity in the examples compared to those where kumpa- has its full lexical meaning. Although word order is extremely free in Jiwarli, and these sentences seem to be no exception as far as the order of arguments and predicates is concerned, there appears to be preferential ordering for the co-occurrence of kumpa- and its dependent verb when kumpa- is bleached of its lexical semantics. Such word order preferences (and a tendency towards adjacency) are possibly indicative of incipient grammaticisation of kumpa- as an auxiliary verb coding continuous aspect (see further below). Secondly, the data are interesting from a further diachronic perspective because it is suggestive of a potential source for a shift from split-ergative case marking to a nominative-accusative system through auxiliarisation and the reanalysis of an old dative case marker as an accusative. Most of the Ngayarta languages immediately to the north of the Mantharta subgroup have developed nominative-accusative case marking where the modern accusative is the reﬂex of an old dative case (see Dench 1982 for discussion and reconstruction of the changes involved). For the Mantharta languages, the potential diachronic developments would be as follows. Firstly, note that imperfective-same subject transitive verbs require that their object be marked with dative case rather than the usual accusative case employed in main clauses (formally identical to the nominative case for all nominals except those with animate reference, giving rise to an ‘ergativity split’). Although kumpa- is the main verb syntactically, in these constructions the dependent verb carries the primary predicate semantics. A possible development would be for kumpa- to be grammaticised as an auxiliary carrying tense/mood and for the dependent verb to be reanalysed as the main verb (with a non-ﬁnite inﬂection). Since kumpa- is intransitive, its subject will always be in the nominative case. When the erstwhile ‘main’ verb is transitive, its object will be in the old dative case, which can now be reanalysed as an accusative case marking main clause transitive object function. The result would be a shift from split-ergative to nominative-accusative case marking, that is: [N-nom kumpa- [Vtr-impf:ss N-dat ]] subject main v dependent v object becomes: [N-nom kumpa- Vtr-nonﬁnite N-acc ] subject aux main v object This kind of reanalysis and auxiliarisation has been reported for other language groups. Whether or not the Mantharta languages might have developed in this way in the future is, of course, entirely speculative.

peter austin

30

4. Grammaticisation of ‘sit’ in other languages Bybee, Perkins and Pagliuca (1994: 127) present cross-linguistic evidence from a wide range of languages that: the majority of progressive forms in our database derive from expressions involving locative elements . . . The locative notion may be expressed either in the verbal auxiliary employed or in the use of postpositions or prepositions indicating location . . . The verbal auxiliary may derive from a speciﬁc postural verb, such as ‘sit’, ‘stand’, or ‘lie’, or it may express the notion of being in a location without reference to a speciﬁc posture but meaning only ‘be at’, ‘stay’, or, more speciﬁcally, ‘live’ or ‘reside’.

Our suggestion above that kumpa- in the Mantharta languages could be an incipient source for grammaticisation as a progressive auxiliary is consistent with this evidence. The development of ‘sit’ as a compounding or an auxiliary verb has actually occurred in a number of Australian languages, including Diyari and Ngamini, once spoken in the far north-east of South Australia (see Austin 1981a), and Yankuntjatjara from central Australia (Goddard 1985). We will now brieﬂy outline the situation in these languages. In Diyari and Ngamini verb phrases can consist of a main verb, a main verb compounded with a following verb in a serial construction, or a main verb or compound verb plus following auxiliary verb. The auxiliary verbs essentially mark tense and aspectual categories in Diyari; Ngamini has just three auxiliaries marking tense. In both languages the auxiliaries derive historically from grammaticised main verbs and must immediately follow the lexical main verb (which carries a non-ﬁnite inﬂection).7 Consider the following list of Diyari auxiliaries (Austin 1981a: 89):8 Auxiliary wanthiyi wapayi wapaya parrhaya wirrhiyi warrayi nganayi

Homophonous root ‘search’ ‘go’ ‘go’ ‘lie (inanimate)’ ‘enter’ ‘throw’ ‘be’

Function distant past tense habitual aspect intermediate past tense recent past tense yesterday past tense immediate past tense future tense

Examples of some of these are: (33) Ngathu ngara-rna wanthiyi karna-li warru warrapa-rnanhi I.erg hear-pple aux person-erg long ago relate-impf:ds ‘I heard people talking about (it) long ago.’ [Austin 1981a: 89, ex (87)]

grammaticisation and the verb ‘to sit’

31

(34) Maja ngathu nhinha nhayi-rna warrayi just I.erg 3sg:m:acc see-pple aux ‘I have just seen him.’ [Austin 1981a: 91, ex (95)] The auxiliary verbs are fully grammaticised and are phonologically linked to the lexical verb, being pronounced in the same intonation group, and being subject to contraction with the main verb in fast speech (Austin 1981a: 31), as in:9 ‘went long ago’ waparna wanthiyi pronounced as [wɔ´ pnunti] ‘went just now’ waparna warrayi pronounced as [wɔ´ pnuɾεi] Compound verb constructions in Diyari and Ngamini consist of a main verb (transitive or intransitive, taking a participial inﬂection) followed by a verb of stance or motion which provides adverbial speciﬁcation of the situation (Austin 1981a: 99). Like the auxiliaries, no phonological material can appear between the main verb and the following compounding adverbial verb. Examples are: Verb ngamathikaparlkatharangari-

Lexical meaning ‘to sit’ ‘to return’ ‘to travel’ ‘to go up’ ‘to go down’

Adverbial verb function action done while stationary action done while returning action done while in motion action directed upwards action directed downwards

as in: (35) Thanali thayi-thayi-tharrhi-rna ngama-yi 3pl.erg eat-eat-dur-pple sit-prs ‘They are sitting down eating.’ [Austin 1981a: 157, ex. (366)] (36) Nhawu tharrka-rna thara-rna warrayi 3sg:m:nom stand-pple go up-pple aux ‘He stood up.’ [Austin 1981a: 99, ex (110)] (37) Ngaldrra nhayi-rna ngari-yi paya pirna-ndrru 1dlincl:erg see-pple go down-pple bird big-ablat ‘We looked down from the aeroplane (lit. big bird).’ [Austin 1981a: 99, ex. (111)] In fast speech the compound verb phrases are pronounced as a single intonation unit with occasional phonetic reduction, including loss of the initial syllable of the adverbial verb (as we saw for the auxiliaries). Interestingly, in Breen’s 1976 description of auxiliaries in Ngamini he identiﬁes a ‘‘continuous tense form’’ -rnama- as in wapa-rnama-yi ‘is going’. This is clearly based on a reanalysis of the compound

peter austin

32

verb form waparna ngama-yi. Thus again we see ‘sit’ serving as a source for a continuous aspect verb form. A further instance of this phenomenon is to be found in Yankunytjatjara, spoken in central Australia. Here, according to Goddard (1985: 206): a ﬁnite verb acts as an auxiliary providing aspect-like modiﬁcation for a verb in serial form. The two must occur in a ﬁxed order (ﬁnite verb following) and be pronounced as a single intonation unit. The case of the subject is determined by the serial verb, consistent with the status of the ﬁnite verb as an auxiliary in this specialised use.

Just three verbs have this auxiliary function. One of them is nyina- ‘to sit’, serving to code a ‘‘customary’’ or generic situation, as in: (37) Wati-ngku karli at-ra nyina-nyi man-erg boomerang.acc chop-serial sit-prs ‘The man makes boomerangs.’ [Goddard 1985: 207, ex (6–66)] This does not appear to be semantically similar to the continuous aspectual coding of ‘sit’ in the other languages mentioned above, however Bybee et al (1994: 153–5) do note that ‘sit’ and ‘live’ can serve as grammaticisation sources for habituals. Finally, verb compounding with ‘sit’ occurs marginally in the Djapu dialect of Yolngu, north-east Arnhem Land.10 According to Morphy (1983: 89): ‘‘[a] small set of stative verbs, nhina-Ø‘sit’, yukarra-Ø ‘lie’, dhärra-Ø ‘stand’ and also marrtji-Ø ‘go/come’ may be used in conjunction with a main verb to denote durative aspect. In this function they always agree in inﬂection with the main verb . . . nhina-Ø ‘sit’ and dhärra-Ø ‘stand’ tend to be used only when the participants in an activity/event are sitting or standing respectively to perform the activity.’’ Morphy’s examples include: (36) Mukthu-rr nhini be quiet-pot sit.pot ‘Keep quiet!/Sit quietly!’ [Morphy 1983: 90, ex (118)] The Yolngu structure diﬀers from that found in all the other languages we have considered because here both verbs carry the same inﬂection and it is not clear which is the head of the construction. As Morphy (1983: 90) points out ‘‘[i]t is sometimes diﬃcult to decide which verb . . . should be considered the main verb.’’ It remains true however that a ‘sit’ verb is being put into service as a coverb-like element coding durative aspect.

5. Conclusion Text data from the Mantharta languages are suggestive of the potential emergence of an auxiliary verb construction based on the lexical root ‘sit’ and coding continu-

grammaticisation and the verb ‘to sit’

33

ous aspect. Although the data are not conclusive they seem to point to a preference for the two verbs in construction to be adjacent. In other Australian languages such as Diyari, Ngamini and Yankuntjatjara, verb compounding has developed where strict adjacency is required and ‘sit’ has become grammaticised as an auxiliary-like element. If the Mantharta languages had continued to be spoken into the future and grammaticisation taken place, it is possible that a shift in case marking towards a nominative-accusative pattern not unlike that of their immediate northern neighbours might have accompanied the reanalysis.

Notes 1. Earlier versions of this paper were presented at Stanford University, University of Hong Kong, and University of Melbourne. For helpful comments I am grateful to Alan Dench, Nick Evans, Ilia Pejros, and Elizabeth Traugott, none of whom is responsible for remaining errors. Data on Mantharta languages was collected 1978–85 during ﬁeldwork variously supported by University of Western Australia Department of Anthropology, La Trobe University School of Humanities, Australian Institute of Aboriginal and Torres Strait Islander Studies, and the Australian Research Council. My greatest debt is to Jack Butler who taught me all I know about Jiwarli. Finally, I take pleasure in oﬀering this paper in appreciation of ‘‘Bazza,’’ former colleague, co-researcher, and friend. 2. The Jiwarli transcription follows usual Australianist conventions: th, nh, lh are lamino-dentals, j, ny, ly are lamino-palatals, rt, rn, rl are apico-domals (retroﬂexes), ng is the velar nasal, rr is an alveolar ﬂap, rrh is a trill, and r is a post-alveolar continuant. Long vowels are written double 3. In most Australian languages the relevant verb root has the form nyina- or nyi- (see Dixon 1980: 407). A root nyina- occurs in the Kanyara languages, except for Pinikura, which has kumpa- like the Mantharta languages. A cognate and possible loan source is Panjima kumpa- ‘wait (for)’ (Dench 1991: 234, cf. panti- ‘sit’), perhaps via a semantic extension: ‘wait (for)’ > ‘sit and wait’ > ‘sit’. 4. In the following examples separated verbs are in bold face to highlight them. 5. I follow the deﬁnition in Bybee et al (1994: 127) that ‘‘continuous views the situation, whether it be dynamic or stative, as ongoing at the reference time.’’ 6. Note that -marri- derives an intransitive verb stem from a nominal root, while -lparri- derives an intransitive verb stem from a transitive verb root. 7. Two non-ﬁnite inﬂections occur, the choice of which attaches to the main lexical verb depends on the auxiliary. One of these, the participial inﬂection, is identical in shape with the imperfective-same subject dependent verb inﬂection (cf. Jiwarli). 8. In Ngamini we ﬁnd only wapini for distant past and habitual, warrayi for recent past, and nganayi for future. The related language Yarluyandi has just wapayarra for distant past and habitual and nganayarra for future (recent past is an inﬂection -ndu), suggesting a diachronic development toward the Diyari system via elaboration of categories of past tense discrimination.

34

peter austin

19. Early amateur recordings of Diyari by the policeman Gason (1874) clearly show the verb plus auxiliary written as a single word in his vocabulary list. 10. I am grateful to Nick Evans for bringing these data to my attention.

6. References Austin, Peter. 1981a. A Grammar of Diyari, South Australia. Cambridge: Cambridge University Press. ——. 1981b. ‘‘Proto-Kanyara and proto-Mantharta historical phonology.’’ Lingua 54: 41–77. ——. 1981c. ‘‘Switch-reference in Australia.’’ Language 57: 309–34. ——. 1988a. ‘‘Classiﬁcation of southern Pilbara languages’’ Papers in Australian linguistics No 17, 1–17. Canberra: Paciﬁc Linguistics. ——. 1988b. ‘‘Aboriginal languages of the Gascoyne-Ashburton region.’’ La Trobe University Working Papers in Linguistics 1: 43–63. ——. 1989. ‘‘Verb compounding in central Australian languages.’’ La Trobe University Working Papers in Linguistics 2: 43–71. ——. 1992. A Dictionary of Jiwarli, Western Australia. Melbourne: La Trobe University. ——. 1994. ‘‘A Grammar of the Mantharta Languages, Western Australia,’’ La Trobe University, MS. ——. 1995. ‘‘Multiple case marking in Kanyara and Mantharta languages, Western Australia.’’ In Frans Plank (ed.) Double case: Agreement by Sufﬁxaufnahme 363–79. Oxford: Oxford University Press. ——. 1996. ‘‘The Kanyara and Mantharta languages, Western Australia,’’ to appear in Patrick McConvell and Mary Laughren (eds) Where did the Western Desert language come from? ——. 1997. Texts in the Mantharta Languages, Western Australia. Tokyo: Institute for the Study of the Languages and Cultures of Asia and Africa. —— and Joan Bresnan. 1996. ‘‘Non-conﬁgurationality in Australian Aboriginal languages.’’ Natural Language and Linguistic Theory 14: 215–68. Breen, J. Gavan. 1976. ‘‘Ngamini and a note on Midhaga,’’ in R.M.W. Dixon (ed.) Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 745–50. Bybee, Joan, Revere Perkins and William Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: Chicago University Press. Dench, Alan. 1982. ‘‘The development of an accusative case marking pattern in the Ngayarda languages of Western Australia,’’ Australian Journal of Linguistics 2: 43–59. ——. 1991. ‘‘Panjima.’’ in R.M.W. Dixon and Barry J. Blake (eds), Handbook of Australian Languages, Vol 4. Melbourne: Oxford University Press, 125–243. ——. 1995. Martuthunira: A Language of the Pilbara Region of Western Australia. Canberra: Paciﬁc Linguistics, C-125.

grammaticisation and the verb ‘to sit’

35

Dixon, R.M.W. 1979. ‘‘Ergativity.’’ Language 55: 59–138. ——. 1980. The Languages of Australia. Cambridge: Cambridge University Press. ——. 1994. Ergativity. Cambridge: Cambridge University Press. Gason, Samuel. 1874. The Dieyerie tribe of Australian Aborigines: their manners and customs. Adelaide, pamphlet. Goddard, Cliﬀ. 1985. A Grammar of Yankunytjatjara. Alice Springs: Institute for Aboriginal Development. Morphy, Frances. 1983. ‘‘Djapu, a Yolngu dialect.’’ In R.M.W. Dixon and Barry J. Blake (eds), Handbook of Australian Languages, Volume 3. Canberra: Australian National University Press, 1–188. Silverstein, Michael. 1976. ‘‘Hierarchy of features and ergativity.’’ In R.M.W. Dixon (ed.) Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 112–71. Wordick, F.J.F. 1982. Yinjibarndi. Canberra: Paciﬁc Linguistics.

36

peter austin

Appendix Verbs occuring in Jiwarli text examples in construction with kumpa- where it is bleached of its main lexical semantics: jikalpajirril-jipajirril-marrikartakulypakunyakurlkanyu-rrikurlkayikurnikurtimanamarrkarrimuka-marringarlpurringurntangurrkuntha-rrinhanyapajawu-rriparlku-rripinya-rripirtipinyapukaputhi-lparritharlatharrpathikawalhi-rriwangkawangka-arni wanta-mawarntiwarrkamu-rriwiingkayangayanga-lparriyithi-lparri-

lift make afraid (fear-caus) become afraid (afraid-coll) write be sore lie think (thinking-inchoat) hear, listen to look for pick up get wait dance together (dance-coll) run lie be sorry (sorry-inchoat) look be angry (angry-inchoat) be heaped (heap-inchoat) ﬁght one another (ﬁght-coll) crack visit hit one another (strike-coll) feed enter eat become bad (bad-inchoat) speak speak to one another (speak-non sing) separate (separate-caus) get up work (work-inchoat) pull chase chase one another (chase-coll) scold one another (scold-coll)

7 instances 2 instances 2 instances 3 instances

8 instances 3 instances

2 instances

3 instances 3 instances 2 instances

2 instances 3 instances

Factors of typology in language acquisition: Some examples from Warlpiri EDITH L. BAVIN School of Psychological Science, La Trobe University

1. Introduction Berman and Slobin (1994: 110) provide evidence that young children are sensitive to the typological characteristics of their language in that they: (1) exploit the range of options oﬀered within a given domain; for example, if a language uses many tense/aspect distinctions, so do young children acquiring it; (2) treat the language as a system of interrelated formal options; for example, Turkish with both postnominal adjectives and postnominal relative clauses provides clear information to the child on the system of organizing information; and (3) treat the language as a system of interrelated expressive options; for example, English has an elaborate system of verb satellites to express ‘‘path,’’ but children acquiring a language which has a limited means for the grammatical encoding of ‘‘path’’ (e.g. Spanish) learn to ‘‘devote more narrative attention to locative descriptions.’’ In this paper, I ﬁrst outline some of the factors that must be considered in acquisition studies, particularly cross-linguistic studies, and then discuss typological properties of Warlpiri which are central in the acquisition process, with examples from children’s productions.1 It is only comparatively recently that diﬀerences across languages have been considered as an issue in the acquisition process. Discussions of Universal Grammar have inﬂuenced much of the research on ﬁrst language acquisition, with an emphasis on identifying similar patterns of acquisition across languages and testing theoretical assumptions of current linguistic theory. More cognitive based approaches have also been concerned with universal aspects of acquisition. For example, in the cross-linguistic studies reported in Slobin (1985, 1992), similarities in the beginning

38

edith l. bavin

stages of language acquisition are reported. Some of the data suggests that children across language groups tended to map a set of core concepts onto their early productions. For example, the distinction between ongoing and completed events, which is reﬂected in a basic linguistic distinction between perfective and imperfective aspect, is generally made by young children before they distinguish tense. Other evidence comes from the types of modiﬁcations made to the input data by children in the process of acquiring their language. For example, as summarized by Slobin (1985), there is underextension in the marking of transitivity; transitivity in Russian is ﬁrst marked on the objects of prototypical transitive verbs, not all transitive verbs, and young Kaluli children use ergative marking only on agents of prototypical transitive verbs at ﬁrst (Schieﬀelin 1985). Both groups of children will need to extend the speciﬁc marking to other transitive contexts. The example illustrates that young children do not necessarily attribute the same meanings to forms as do adults, but children across languages may pick out similar categories as salient, using language speciﬁc means to mark them.

2. Sources of variation While typological studies are concerned with the mature speaker’s grammar, the adult grammar is the end-product of acquisition; clearly a child does not have all knowledge of the grammar immediately. Since children develop grammatical knowledge only with exposure to a particular language, a major concern in a theory of acquisition, as opposed to a theory of language, is how a child determines the language form-function mappings for the particular language being acquired, that is how the child learns the linguistic forms and their range of uses. Karmiloﬀ-Smith views language as a formal problem-space for children (e.g. 1985, 1986). Berman (1985), amongst other researchers, has stressed that each language has particular areas that are central to the acquisition task. In Hebrew, for example, gender distinctions must be mastered because they are crucial in mastering the system of agreement. Bates and MacWhinney (e.g. 1989) argue that languages diﬀer in the cues they provide; for example, word order might be used to distinguish grammatical roles, or case marking, or a combination of semantic and morphological cues. Children become attuned to the cues provided by the particular language; based on distribution patterns they can make probability judgements about form-function mappings. Distribution patterns vary across languages; thus there are diﬀerences in language input, and this is reﬂected in patterns of acquisition across language groups. Berman and Slobin (1994: 425) argue that forms which are accessible (those that can be readily detected), and forms which are obligatory, draw the learner’s attention both

factors of typology in language acquisition

39

to the forms and the conceptual distinctions that must be made to use the forms appropriately. In all languages there are options. A task for the child is to learn in which contexts it is typical to use one form over another, or one structure as opposed to another. The acquisition literature based on production data shows children use the language speciﬁc patterns that are typical; they acquire language speciﬁc patterns of usage gradually alongside their developing grammatical knowledge and general cognitive development. There are many features of grammar that can be used to classify or type languages, features such as word order, word formation, and types of grammatical categories encoded (e.g see Mallinson and Blake 1981). In the Universal Grammar approach to language acquisition (following the Chomsky approach to language) parameters are proposed to capture variation across languages. The Null-subject (or Pro-drop) parameter is the one that is discussed most extensively in the acquisition literature (e.g. Bloom 1990, Hyams 1986, Valian 1990); a language will have the value (+) if it allows subject pronouns do be non-overt, as in Italian, and (–) if subject pronouns are required, as in English. While this binary division provides a neat typology at one level, it assumes a number of earlier stages in which the child must segment the language input, categorize the forms, and determine how a subject is represented in the language. It also ignores typology of use since even in non Prodrop languages there are contexts in which the subject will not be overt, as across adjacency pairs, and even in languages categorized as Pro-drop, there may be conditions, as in Hebrew in which ellipsis of subjects is only with past tense. In many languages there may also be the option of dropping noun subjects as well as other arguments, as in Warlpiri. In acquisition studies it is necessary to consider that the child works with the input data to construct a grammar, so typology of use is an important consideration in explaining acquisition For example, we need to consider whether subject ellipsis is conditioned by the same factors that inﬂuence ellipsis of other arguments, and whether ellipsis is conditioned by the same factors in the early stages of acquisition as in the later stages. Uziel-Karl and Berman (1996) have examined ellipsis of subjects and objects in simple clause structures in Hebrew. They distinguish permissibility (what must be deleted or overt), recoverability (information from context allows identiﬁcation of deleted element) and syntactic function (missing subjects are allowed more readily than missing objects). In addition to subjects ellipsis with ﬁrst and second person subjects of past tense verbs, subjects may be dropped in conversational adjacency pairs, where a missing subject is the topic established in prior utterance, and in extended discourse when the same subject is used across a series of utterances. Objects may also be missing, with some transitive verbs and in contexts where the situational context allows for recoverability. Uziel-Karl and Berman found that ellipsis in child Hebrew is guided ﬁrst by pragmatic factors, and

40

edith l. bavin

then by morpho-syntactic factors and by a combination at a third stage of development. This pattern follows that for the acquisition of other language knowledge. For example, in discussing development in the use of anaphoric expressions in English and French, Karmiloﬀ-Smith (1985, 1986) argues that children ﬁrst follow a datadriven strategy: children assume identity of referents, using anaphors and deﬁnite articles to introduce participants, whereas for the mature speaker participants are introduced with indeﬁnite expressions or through other language speciﬁc means. At a second stage, children rigidly follow a top-down strategy, following the language speciﬁc patterns for introducing referents and for subsequent mentions, but reserving anaphors for subject position for a default subject (e.g. ‘‘he’’ would refer to only one of two male participants). Finally there is more ﬂexibility in introducing participants and in maintaining reference across extended discourse, and speakers show variability within the choices allowed by their language. What is often overlooked in studies of subject ellipsis in early child language is whether transitive and intransitive subjects are omitted with equal frequency. The theory of preferred argument structure (e.g. DuBois 1987) proposes that when a new referent is introduced into the discourse it is most likely to be in object position or an intransitive subject. Since new referents are more likely to be mentioned than old, transitive subjects would be the best candidate for ellipsis. Thus if there is a greater frequency of transitive subject ellipsis in child language this might reﬂect the discourse structure of the input. Clancy (1993) has shown the relevance of preferred argument structure in accounting for argument ellipsis in the productions of young Korean children. Stromqvist and Ragnarsdottir (1996), investigated the use of subject arguments by young Swedish children, and report that when a subject NP is included in an utterance, it is more likely to be for a new referent, but if something is introduced in an utterance, there is a tendency for it to be the assumed subject in the following clause. Thus children are sensitive to discourse factors relating to given-new information at an early age. In addition, Allen and Schröder (1996) conclude that children acquiring Inuktitut, even 2 and 3 year-olds, are sensitive to relationships between syntactic position, pragmatic prominence and argument representation. That is, discourse factors must be considered in discussion of which arguments are overt.

3. Cross-linguistic comparisons One problem in comparing acquisition data across languages is that functions for particular linguistic structures may vary. For example, while Turkish, Spanish and English all mark progressive aspect, the aspect diﬀers in the range of meanings

factors of typology in language acquisition

41

conveyed because its usage is dependent on the system of tense/aspect contrasts within each language (Berman and Slobin 1994: 72). If a structure has a low functional load in a language, it is not a feature that a child acquiring the language will be speciﬁcally attuned to, as with the passive structure in English. In Sesotho, the passive has a greater functional load, and the diﬀerence across the two languages is reﬂected in the early acquisition of passives in Sesotho but not in English (Demuth 1989). Another factor which makes it somewhat diﬃcult to do cross-linguistic acquisition research is that semantic notions are grammaticized in diﬀerent ways across languages. Thus comparisons which focus on the acquisition of lexical items or on the acquisition of particular linguistic properties may miss important development patterns since a function word in one language may encode a semantic notion which is lexicalized in another. As discussed extensively by Talmy (e.g. 1983, 1991), manner and change of state are conﬂated into verbs of motion in Germanic languages (as with ﬂoat, jump), but direction is not; trajectories are generally encoded in a particle (as in fall down, stand up). In contrast to satellite-framed languages, verbs in Romance languages encode direction, while manner is encoded separately. Using narrative data based on a wordless picture book, Frog, Where are You? (Mayer 1969),2 Berman and Slobin (1994) show that by the age of 3 years children are inﬂuenced by this typological distinction. The data, collected from children (and adults) from 5 language groups, show that young children are inﬂuenced by the conceptual framework most readily grammaticized in their language. The English and German-speaking children elaborated trajectories much more than speakers of the other three languages, Turkish, Spanish and Hebrew. But there is evidence from naturalistic data from younger children. Choi and Bowerman (1991) compare English and Korean and show that even before the age of 2 years, children are inﬂuenced by how their language encodes spatial/locative information. Korean children must learn to distinguish spontaneous and caused motion events with respect to the category of path since diﬀerent verb roots are used; this they do. Children in both groups make overgeneralization errors, but these are consistent with their language typology. In order to construct the Frog Story, speakers must make inferences to link the pictures together, and how well they do this will depend on their world and linguistic knowledge. The discourse data have been used to study development within a language, since children can only use the linguistic means available to them, and cross-linguistically, since it allows comparison of the preferred structures for adults and the ages at which these are patterned in children’s productions. Speakers have a choice in how to express information; they can take diﬀerent perspectives, and there is ﬂexibility in the use of verb forms used to direct the listeners’ attention to the ﬂow of discourse. However, language speciﬁc preferences are evident. The

42

edith l. bavin

predominant pattern found in English is an assertion of actions, with result implied (as in He sat not He was seated), while in Spanish the result is asserted and the action implied. These diﬀerences, which relate to the availability of a number of aspectual forms in Spanish reﬂect distinctions in the typology of use. Even the youngest English-speaking children were found to be process-oriented in their stories, while Spanish-speaking children established the stative location of objects more. As Slobin (1989) argues, the child becomes sensitive to sets of interacting grammatical constructions and lexical forms that characterize the type of language being acquired. What should be clear, then, is that we should not use acquisition data to provide empirical support for a particular linguistic description without considering how the forms observed ﬁt into the system as a whole, how they are used in the adult system, and what the form means to the child at diﬀerent stages of development. By focussing on what is readily accessible in a language system to a child, we can consider how language typology and typology of use both contribute to the acquisition of a language. In this next section I outline some of the typological features of Warlpiri and discuss Warlpiri children’s productions, to illustrate some of the features which seem to be accessible to young Warlpiri children, and some of the developments noted. Many examples are drawn from Frog Story data, collected from Warlpiri children in the age range 2;6–16 years as well as adults; others are drawn from spontaneous productions from young children talking together.

4. Characteristics of Warlpiri Warlpiri is generally typed as a non-conﬁgurational language (e.g. Hale 1983, Laughren 1989). The main implication for acquisition is that grammatical relations can not be determined on the basis of word order heard by the child in the input. The language is agglutinating, and it has case marking and is morphologically ergative. Features of number and person for the subject and object are not marked as inﬂections on the verb, but in a clitic cluster which appears in second position in a ﬁnite clause. However, verbs are always inﬂected; there are 5 verb classes with inﬂections to mark past, non past, inﬁnitive, imperative and irrealis. One of the greatest diﬀerences between Warlpiri and other languages discussed in the acquisition literature is the use of the clitics which register subject and object. The clitics (bound pronominals) are distinct for these arguments The subject clitic precedes that for the object, and dative objects are marked over absolutives. However, both third singular subject and third singular absolutive object are zero in form, while the third singular dative object form is rla. Thus the past form of nyinami ‘see’ as an utterance, Nyangu, means ‘someone or something

factors of typology in language acquisition

43

saw someone or something’, Nyangu-pala means ‘two saw someone or something’, Nyangu-palangu means ‘someone or something saw two’, and Nyangu-rnapalangu means ‘I saw two.’ All these examples have the bound pronouns attached to the verb because there are no other items in the clause, but if another word appeared ﬁrst, they would be attached to that. Other elements may also appear in a clitic cluster, for example the aspect markers lpa and ka; the bound pronominals follow these. As indicated by the examples above, Warlpiri allows ellipsis of non-bound arguments, but when they are overt they are not in a ﬁxed word order in relation to the verb. Jelinek (1984) and Hale (1983), amongst others, have argued that the clitics are the core arguments and lexical items, if present, are adjuncts. If the Warlpiri bound pronominal forms are arguments, then Warlpiri is not a Pro-drop language; however, the third singular subject form, zero, must then be counted as an argument. If the lexical items are the arguments and the clitics agreement markers, then Warlpiri is a Pro-drop language but, unlike other languages for which acquisition data is available, the agreement markers are not inﬂections on the verb. The word orders used by mature speakers and the use or ellipsis of lexical arguments depend on pragmatic and discourse factors, and there is a great deal of individual variation. However, when a lexical subject is overt and its referent is assumed, the noun frequently comes at the end of the utterance in the data I have collected from adults, as in (1a). Frequently there is a pause before the nominal, as with right dislocation constructions in other languages. In (1b), lexical subject nouns and dative object appear late in the clause. New and contrastive information is more likely to come at or near the beginning of the sentence (e.g. the verb, and temporal and spatial information) although in a switch subject context the new subject is often sentence initial, as in the example in (2), in which maliki ‘dog’ is overt as subject in the third line. In (1c) the ‘two’ in line 1 are spelled out in the following two sentences, and there is then repetition of this information, something that is typical in the adult stories. There is also typically a lot of information given about spatial orientation, as in (a) and (d). These are all features of the language which the young Warlpiri child uses as a model. (1) a. Kankarla-rra warrka-rnu jurra-ngka wunu-ngka wirriya-ju up-dir climb-pst head-loc horn-loc boy-foc ‘(He) climbed on the horns on the head, the boy.’ b. Ngula-jangka-ju purla-mi ka-pala-rla wirriya that-from-foc call-npst imfv-3:du:sbj-3:dat:obj boy manu maliki jarlji-ki and dog frog-dat ‘Then they called for it, the boy and dog, for the frog.’

edith l. bavin

44

c. Ngula-jangka-ju wanti-ja-pala jirrama that-from-foc fall-pst-3:du:sbj two Wirriya wanti-ja. Maliki wanti-ja boy fall-pst dog fall-pst ‘After that they fell the two. The boy fell. The dog fell.’ d. Putu-lku ka puta yuka-mi puta-ngka. Maliki nyampu ka boot-now imfv boot enter-npst boot-loc dog this imfv yuka-mi kaninja-rra enter-npst down-dir ‘The boot now, (it) is entering the boot. The dog is entering down, away.’ (2)

Ya-nu-pala yurulku-kurra-lku. go-pst-3:du:sbj bush-all-then Purla-ja ‘‘Jarlji, jarlji, nyarrpara-npa ya-nu?’’ call-pst ‘‘Frog, frog where-2:sg:sub go-npst Maliki-rli panti-nya-ngu ngarlu-lku dog-erg smell-pst sugarbag-then Yanu ngula-jangka-ju jinta-kari-kirra-lku go-pst that-from-foc one-other-all-then ‘They went to the bush then. (He) called ‘‘Frog, frog, where did you go?’’ The dog smelled the sugarbag. Then (he) went to another one.’ (‘He’ is ‘ the boy’)

The proportion of lexical subjects and objects for 4 of the adult stories was calculated for comparison purposes. These are representative of the adult group. In total, 248 verbal utterances were used by the adult story tellers, and of these 37 per cent were transitive. Overall, 55 per cent sentences had an overt lexical subject; of the transitive verbs, 50 per cent had overt lexical subjects (ergative-marked), and 44 per cent had a lexical object. Only 22 per cent of the transitive verbs had both a subject and object. This does not include clitic arguments. While lexical subjects and objects are frequently non-overt in the adult stories, there are many utterances in which locative/spatial elements are overt. Case marked nominals, locative nouns and directional aﬃxes are all used. As an example, the total percentage of motion verbs with one or more locative elements is 75 per cent, with 83 per cent of the locative expressions case marked. The motion verb which typically lacks a separate locative element is juurly-pinyi ‘chase’; however, verbs for arise and emerge typically have a source argument speciﬁed and wantimi ‘fall’ a goal as well as a source argument for some speakers and/or a path (as in kanunju wantija ngapa-kurra = down fell water-all ‘(he) fell to the water’). In summary,

factors of typology in language acquisition

45

it is more usual to ﬁnd a locative argument with a motion verb than a subject with a verb or an object with a transitive verb. This pattern provides a model for the child, and illustrates that typology of use is a factor to consider when discussing the patterns children adopt. One of the main features central to the acquisition of Warlpiri is the case frames for verbs; one place predicates have absolutive subjects (zero marking), and two place predicates generally have ergative subjects and absolutive objects. However, some verbs require an absolutive subject and dative object, and a change of meaning from actual to attempted action is conveyed by using a dative object instead of an absolutive with an ergative/absolutive verb; for example, the use of the dative object clitic rla with the verb nyanyi ‘see’ would mean that the agent attempted to see someone or something, that is, they were not successful in seeing something. The referent could be speciﬁed with a dative-marked lexical item, as with jarntu ‘dog’ in (3): (3) Nya-ngu-rla jarntu-ku. see-pst-3:sg:dat:obj dog-dat ‘Someone looked for the dog.’ Inﬂections, not word order, provide the distributional evidence that could be used by children for distinguishing verbs from nouns, but these are not always available or reliable on nouns. For example, ergative case markers on transitive subjects are not always available in the input since lexical subjects are not always overt. An absolutive argument can be either a transitive object or intransitive subject, so knowing a verb’s case frame is essential for interpreting an utterance. For example, in (4), the absolutive jarntu is interpreted as the subject because nyinami ‘sit’ is an intransitive verb. There is no clitic attached to the base ka since the subject is third singular. (4) Nyina-mi ka jarntu sit-np impfv dog ‘The dog is sitting.’ Because any lexical item can appear clause initially, clitics might follow nouns (inﬂected or uninﬂected), verbs, or particles (which often have aspectual or modality meanings). However, verb inﬂections, always present, will be closest to the verb stem, so they are reliable as verb markers, as long as they can be detected. Because there may be bound pronominals, discourse markers, or directional aﬃxes or complementizers following the inﬂected verb stem, the inﬂections don’t always appear last on the word (e.g. ya-ni-rni = motion-np-dir ‘come’; ya-ni-rra =motionnp-dir ‘go’; paka-rnu-lku = run-p-then/now ). In the following sections, I give a brief overview of the young children’s

46

edith l. bavin

knowledge of case markings for arguments, verb inﬂections and clitics, and discuss some Frog Story data which illustrates the use or ellipsis of lexical arguments.

5. Warlpiri acquisition patterns Case marking There are four possible allomorphs that are used to mark an ergative nominal: rlu, rli, ngku and ngki. The appropriate form depends on the length of the nominal (two syllables or longer) and the vowel in the stem: wati-ngki ‘man-erg’, marlungku ‘kangaroo-erg’, wawirri-rli ‘kangaroo-erg’ and wirriya-rlu ‘boy-erg.’ As discussed elsewhere (Bavin 1992, 1995), it is not an easy task for children to work out which form is appropriate. Even though they do not overgeneralize ergative case forms to intransitive subjects, the children use a number of strategies before they are conﬁdent about which form to use. It is not the notion of morphological ergativity that is problematic. (See also Pye 1990). As with children faced with learning different forms to mark one function in other languages, overgeneralizations occur. However, vowel harmony alternations are not problematic. There are very few lexical subjects in the utterances of the children under 3;3, so ergative case forms do not appear. An example from a child of 3 years shows the form ngku, used on two syllable words, on ngaju ‘I’, which is irregular in that its ergative form is ngajulu-rlu. Some children pick the form ngku or ngki as the general ergative marker until they discover the pattern of use, while others use these for human nouns and the rlu/rli forms with other nouns. These and other strategies are used. Overgeneralizations are also made by Warlpiri children with the general locative case allomorphs ngka and rla which are distributed on the basis of word length, as with the ergative markers. However other case markings are not problematic. For example, the dative ku/ki appears appropriately in utterances from 2 year-olds, as does the allative kurra/kirra. One of the reasons for the diﬃculty in identifying the appropriate form to use for ergative subject may be that since lexical arguments may be omitted, they are not available for the children to work out the distribution of the forms. While the ellipsis of transitive subjects may reﬂect an avoidance strategy at one stage because of the uncertainty of case forms, avoidance in using them in early utterances could reﬂect lack of knowledge about argument structure, that is which verbs require ergative subjects. Acquisition of verbs involves not just acquiring their meanings and case frames, but also their morphological class. In this respect, the child is facilitated by the fact that verbs are always inﬂected.

factors of typology in language acquisition

47

Verb inﬂections Children acquiring English ﬁrst produce verb stems without inﬂections for past or progressive or third singular. They may also produce irregular past verbs such as ran and ate in their early utterances, and later regularize the endings using the forms runned and eated. This pattern seems to be related to the paucity of verbal inﬂection in English. Children acquiring languages in which verbs are generally inﬂected do not start with bare verb stems to which they add inﬂections at a later stage. For example, Clancy (1985) reports on the precocious control of verb inﬂections with Japanese children, as does Berman (1985) for Hebrew children. In both these languages, there is no citation form of a verb that children can rely on until inﬂections are acquired, and this seems to force the children to deal with the options and make choices amongst the inﬂections. That is, the verb morphology becomes a central issue for a child acquiring the language. Warlpiri children produce inﬂected verbs in their earliest verbal utterances. For example, a child of 24 months used the verbs wanti-ja ‘fall-p’ and paka-rni ‘ run-np’, and a number of imperatives. The forms do not reﬂect an understanding of the meaning of the inﬂections until a child is able to use several inﬂections with one stem and the same inﬂection with diﬀerent stems. There are ﬁve verb classes in Warlpiri, and even though most verbs fall into 2 of the ﬁve classes, the smaller classes include frequently used verbs, such as yani ‘go’, and ngarni ‘eat.’ Verbs are always inﬂected (although class 1 verbs in nonpast form may drop the mi inﬂection as in nyina or nyinami ‘sit-np’) and the Warlpiri child is attuned very quickly to this typological property. The shape of a verb becomes a good predictor as to its form in a diﬀerent context. As in Hebrew, there are clear formal cues in the shape of the word. For example a verb ending in mi will end in ja in past contexts. When children have the notion that the two forms are related in meaning but diﬀer in the contexts of use, they will selectively use the mi and ja forms in diﬀerent contexts and when a new verb is heard in one form, the other can be predicted. Another factor that facilitates children’s acquisition of verbs is the comparatively small number, about 150. Preverbs, which can be added to a clause to modify the meaning of the verb, are not frequent in young children’s productions. Their frequency increases with age. Bound pronominal forms Warlpiri children, like children acquiring other languages, focus on the here and now in their ﬁrst utterances. Utterances from 2 year-olds are related to the immediate rather than a displaced context, and so arguments that are not overt are usually recoverable from the non-linguistic context. When utterances relate to third person singular referents, it is not possible to determine if the child has acquired the knowledge that clitics for subject and object are required since the markers are null. However, dative objects are registered with rla, and when a child uses rla with some

edith l. bavin

48

verbs but not others, there is evidence for knowledge that some verbs take a dative object. In the naturalistic data I have collected, rla does not show up in utterances from 24 month-olds, but it appears in utterances from a child of 2;8. The ﬁrst bound pronominal form used by Warlpiri children is the ﬁrst person object ju, attached to an imperative (yungu-ju ‘give it to me’). This is in utterances of 24 month-olds. The form is rote-learned. Necessary evidence that ju has a 1st person object meaning rather than just functioning as part of yungu is that it appears in other contexts. By 2;6, there is evidence for a developing grammatical system in that the imperfective clitic base ka is used, although there is variability in use. With non-past verbs, ka is used to indicate progressive states or ongoing activities, as in the adult system; with the word kapu instead of ka, the non-past verb has a future meaning. The form ka appears in the appropriate position in examples 5 (c) and (d), but is missing from (a) and (b). These examples show all the verb forms in a frog story data from a child of 2;6. Note that the child uses ka with a non-past verb form if nothing else follows. In the ﬁrst two examples, ka is missing, which would only be appropriate with an immediate future interpretation. It may have a distinct function for the child, only appearing with a subset of dynamic verbs to mark on-going action, and not for perception verbs and verbs which represent on-going states, as in (a) and (b). Such undergeneralizations are a feature of developing grammatical knowledge. (5) a. Nya-nyi jinta-kayi-ji see-npst one other-foc ‘He sees another one.’ b. Jarntu nyi-na wiyapa dog sit-npst aff ‘The dog sits, poor thing.’ c. Yuka-mi ka enter-np impfv ‘It is entering’ (2 instances) d. Nyampu-ju, ya-ni-rra ka here/this-foc, go-npst-dir impf ‘This one, it’s going away.’ e. Wanti-ja fall-pst ‘It fell.’ (2 instances) f. Nya-ngka look-imp ‘Look!’

factors of typology in language acquisition

49

Children of 2;8 also show some variability in the use of ka, and there are individual diﬀerences in its use. However, never is it used with past form verbs. The examples in (6) are spontaneous productions from a child of 2;8. They show ka used in second position with a non-past verb but not with a past verb form (as in c) or imperative (as in d); in (b) ka appears initially. From this child and two others aged 3;0, there are examples of ka in initial position if there is a clitic attached, so making it longer than one syllable; presumably the children assume this gives it status as a word, since all words in Warlpiri are two syllables or longer. The restriction that bound pronominal forms appear in clause second position is illustrated in (c), in which rna ‘ﬁrst singular subject’ is attached to the ﬁrst word of the utterance, wiyi. The use of a bound object form is illustrated in (d), and both subject and object bound pronominal forms are used in example (e), in which they are ordered appropriately. (6) a. Yali ka karri-mi kuja there/that one impf stand-npst thus ‘It’s standing there like this/That one’s standing there like this.’ b. Ka-rna nga-rni imfv-1:sg:sbj eat-npst ‘I’m eating.’ c. Wiyi-rna wangka-ja ﬁrst-1:sg:sbj speak-pst ‘I spoke ﬁrst.’ d. Yu-nga-ju give-imp-1:sg:obj ‘Give it to me.’ e. Mani ka-rna-ngku hold-npst impf-1:sg:sbj-2:sg:obj ‘I’m holding it for you.’ Overt arguments and word order Overall, the data from narrative elicitation and spontaneous speech from children under 3 show that the central areas on which children focus in the early stages of acquisition are the verb inﬂections and case frames, the clitic cluster, and locative elements. Certainly not all clitic forms are acquired by age 3, but the notion that bound pronominal forms for subject and object appear in second position of the clause is well established before this age. Core lexical arguments are infrequent, but case markers appear on overt locative arguments. The number of ergative-marked subjects and absolutive arguments gradually increases from late in the fourth year. Pragmatic variability in word order is not problematic for children who are acquiring languages with this property. Berman (1985) reports that even 2 and 3–year

edith l. bavin

50

old children acquiring Hebrew use language speciﬁc patterns for topicalization that are also used by adults. Right dislocation, for example, is used by young Hebrew children to make it clear who is being talked about, and later left dislocation is used to establish the topic of their comment. Warlpiri children do not ﬁnd word order variability problematic since they use diﬀerent word orders in their ﬁrst multi-word utterances. They show growing awareness of the need to make it clear who is doing what to whom, but there is great ﬂexibility in what is assumed from the context, linguistic and non-linguistic. In order to give an overview, I will summarize a few of the stories from the younger children, stories which are representative of the group. The stories from the youngest children can not be said to be narratives since the events depicted in the pictures are not linked by appropriate linguistic means. It is only in the stories from the children of 5 or 6 and older that evidence begins to appear. However, by age 9, some of the speakers provide sophisticated narratives in content and structure. The stories collected from children aged from 2;6–3 have a low percentage of verbal utterances and these do not have a high percentage of lexical objects or subjects overt. Even though the listener does not see the pictures, children only gradually learn to take other people’s perspective and assume identity of the participants. Frog data from one child of 2;6 is illustrated in (7) and (5). The child used 26 utterances with a mean length of utterance of 2.5 morphemes. There were only 8 verb forms used, as shown in (5); that is, only about 30 per cent of the utterances had verbs. (‘‘Utterance’’ here means an expression bounded by silence.) Data from spontaneous utterances around this age show that children produce no more verbs than this. A count with two children aged 2;8 revealed only 25 per cent of utterances contained verbs. Other utterances are nominal, often deictic expressions such as the four examples in (7) and frequently the ﬁrst person pronoun with a dative marker (njaju-ku ‘for me.’) (7) a. Nyampu Here/this one. b. Nyampu jinta kari here one other ‘Here’s another one.’ c. Jarntu wita pardu dog small dim ‘Small dog.’ d. Nyampu wita pardu here little dim ‘Here’s a little one.’

factors of typology in language acquisition

51

Between 2;6 and 3, there is a great deal of development in language knowledge. It is in this age range that the children start using clitic clusters with bound pronominals. The development of language knowledge is observed in the elicited narrative data also. As an example, a 3 year-old used 29 utterances in his frog story, 21 with verbs, but only one transitive verb. No sentences had an overt lexical subject, and there was only one object noun (for ‘frog’). However, the child used the subject clitic for 1st plural subject (rnalu) when talking about what people called the frog. He also used one dative object clitic rla with the verb warrkarni ‘look for’ which has an ergative-dative case frame. Locative case markers were used (pirlingka ‘on the hill’ and walya-rla ‘on the ground’) and a directional suﬃx (yanu-rra ‘went away’). While these two examples show that children of 2;6–3 can interpret the pictures, and pick out salient features to comment upon, there is no evidence that they are organizing the utterances into a coherent narrative. In a narrative from a 4 year-old, there are 28 utterances with a mlu (Mean Length of Utterance) of 3.6 Of the 28, 27 utterances have verbs. Of these 14 (22 per cent) are transitive but only one has a subject (ergative marked). This is used for a deer who looks at the boy after it has thrown him into the water. The word order here is vs with no object. The boy is the default subject of most of the verbs used, but is never overtly mentioned in the whole narrative. In this respect, the zero clitic for third person singular functions like the free-standing pronouns for 3rd singular in English, which are used deictically rather than anaphorically at ﬁrst. Discussions of word order in Warlpiri are not as interesting as why lexical arguments are used in some contexts. The 4 year-old mentioned above used only one intransitive subject, for ‘dog’, used when there was a switch from a series of activities from the boy. The child’s strategy seems to be to use an overt subject if it is in a switch subject context. Only 4 lexical objects were overt for the 14 verbs (29 per cent). Three of these were used for ‘frog’, the object of the search, and one for ‘dog.’ In addition, the dative object clitic (rla) was used for ‘frog’ 3 times. That is, either a lexical absolutive object or a dative clitic were used for the focus of the search. Locative arguments were overt and marked, one with the locative case, pirli-ngka ‘hill-loc’, and two goals marked with the allative, ngapa-kurra ‘water-all.’ Another child of 5;7 produced a story with 25 utterances of which 24 contained verbs. Of these, 6 (25 per cent) were transitive and 3 of these had ergative-marked subjects. There were also 7 (38 per cent) intransitive subjects. Of the subjects 50 per cent came before the verb and 50 per cent at the end of the utterance, but all 3 ergative subjects were utterance ﬁnal. Three bound pronominal forms were used for dual subject when the boy and dog were involved in joint activity, and in these utterances no lexical subject was used. The locative elements used with motion verbs included case-marked nouns, as in (8), as well as directional aﬃxes.

edith l. bavin

52

(8) Pirli-wana ka ya-ni. Warrkarnu pirli-ngka watiya-ngka hill/rock-perl impf go-npst climb-pst hill/rock-loc tree-loc ‘(He) cwent by a rock. (He) climbed on a tree on the rock.’ Overall, the stories from the 5–6 year olds show that about 80 per cent of the utterances are verbal and of these about a ﬁfth have transitive verbs. Less than 40 per cent of all verbs have a lexical subject, less than a quarter of the transitive verbs have lexical subjects and less than 40 per cent of the transitive verbs have a lexical object overt. Only 5 per cent of the transitive verbs have both a lexical subject and object. Many of the young children used the verb kijirni ‘throw’ rather than wantimi ‘fall’ in the contexts where a sugarbag falls or is thrown from a tree, the boy falls or is thrown down, and the boy and dog fall or are thrown into water. These all provide contexts for agents, but typically the agent is not overt. Locative elements are used more consistently, especially with motion verbs. What conditions the use of a subject is not entirely predictable. Switch versus same subject is one factor: if the subject argument is the discourse topic, as with the boy in the frog story, it is often assumed; however, when there is a switch in subject, as when the dog’s action is mentioned, the subject is more likely to be mentioned. For example, in (9), a representative example, there is a switch to the dog as subject. (9) a. Yakarra-pardi-ja get up-pst ‘(He) got up.’ b. Patirli ka nya-nyi bottle impf see-npst ‘(He) is looking at the bottle.’ c. Jarntu ka yuka-mi patirli-rla dog impf enter-npst bottle-loc ‘The dog is going into the bottle.’ However, another factor is whether there is an overt subject clitic, that is if the subject is other than third singular. Following the sentences in (9), the child used the 3rd dual subject clitic pala, but with no lexical arguments.

6. Concluding remarks In conclusion, acquisition data from Warlpiri as from other languages clearly show that even very young children are sensitive to the typological characteristics of the language. When Warlpiri children start to use a signiﬁcant number of verbal utter-

factors of typology in language acquisition

53

ances, development in the use of the clitic cluster is rapid, and it is at this stage that we can say the grammar of the language is developing. Two-and-a-half is typically the age at which rapid developments in grammatical knowledge are noted across languages, and this is so for Warlpiri children. While children’s acquisition of their language is paced by general cognitive development, similarities are noted across languages in the type of information encoded in early language, but there are signiﬁcant diﬀerences in how the concepts are encoded. When a language provides clear distribution cues for a particular aspect of the language, acquisition is facilitated, as with morphological verb classes in Warlpiri. The type of language, in terms of structure, will inﬂuence acquisition with some aspects being more central, but typology of use is also a factor in the acquisition process since children make generalizations based on what they hear in the input. For example, locative elements are used frequently in the Warlpiri input, and locatives arguments are frequently used by young children; the locative case marker shows up in productions from children even before they use a signiﬁcant number of verbal utterances, and goal arguments of motion verbs show up in early verbal productions. In Warlpiri the bound pronominal forms for third person are zero and lexical arguments can be dropped. Yet the use of appropriate inﬂections for verbs and the use of dative clitics show that, even if lexical arguments are not overt in their productions, the children are mastering the case frames for individual verbs as well as their morphological class. Warlpiri allows ellipsis of lexical arguments and few of these are found in early verbal productions by Warlpiri children. There does not appear to be a preference for dropping subjects over objects or transitive over intransitive subjects, which might be expected. Individuals have options as to when to include them and in which position in the clause, and Warlpiri children exploit these options from a young age. In stories from the older children, overt lexical arguments are more probable in some contexts than others, for example, when there is a perceived need to disambiguate, and thus type of discourse will aﬀect their use.

Notes 1. The research was conducted in Yuendumu, N.T and was funded by the Australian Research Council. I am grateful to the community for allowing the research, and I am particularly grateful to Kay Napaljarri and Cecily Napanangka for their support and assistance over the years. 2. In the story book there are 24 pictures with no text. These represent a series of events involving a boy, his dog and a frog. The frog escapes and the boy and dog go on a journey, to ﬁnd the frog (we assume). On the way they encounter a series of adventures, ﬁnally ﬁnding a group of frogs from which

edith l. bavin

54

the boy takes one and walks away. Narrators often tell the story from the boy’s perspective but the story provides many contexts to switch subjects since the boy, the dog, the frog, an owl, bees, and deer are all doing things at various stages. Sometimes the boy and dog are involved in joint activity and sometimes separate.

References Allen, Shanley and Heike Schröder. (1996). ‘‘Preferred Argument Structure in Early Inuktitut Spontaneous Speech Data.’’ Paper presented at a workshop on ellipsis at the International Congress of Child Language, Istanbul, July. Bates, Elizabeth and Brian MacWhinney. 1989. ‘‘Functionalism and the Competition Model.’’ In Brian MacWhinney and Elizabeth Bates (eds), The Cross-Linguistic Study of Sentence Processing. Cambridge: Cambridge University Press, 3–73. Bavin, Edith L. 1992. ‘‘The Acquisition of Warlpiri.’’ In Dan I. Slobin (ed.), 309–71. ——. 1995. ‘‘Inﬂections and Lexical Organization: Some evidence from Warlpiri.’’ In H. Pishwa and K. Marold (eds.), The Development of Morphological Sysematicity. Tübingen: Gunter Narr Verlag, 39–53. Berman, Ruth 1985. ‘‘The Acquisition of Hebrew.’’ In Dan I. Slobin (ed.), 255–371. ——and Dan. I. Slobin. 1994. Relating Events in Narrative. Hillsdale, N.J.: Erlbaum. Bloom, P. 1990. ‘‘Subjectless Sentences in Child Language.’’ Linguistic Inquiry 21: 491–504. Choi, S. and Melissa Bowerman. 1991. ‘‘Learning to Express Motion Events in English and Korean: The inﬂuence of language speciﬁc lexicalization patterns.’’ Cognition 41: 83– 121. Clancy, Patricia. 1993. ‘‘Preferred Argument Structure in Korean.’’ In Eve Clark (ed.), 307–14. ——. 1995. ‘‘The Acquisition of Japanese.’’ In Dan I. Slobin (ed.), 373–524. Clark, Eve (ed.) 1993. Proceedings of the 25th Annual Child Language Forum. Stanford: CSLI. Du Bois, John. 1987. ‘‘The Discourse Basis of Ergativity.’’ Language 63: 805–55. Demuth, Katherine. 1989. ‘‘Maturation, Continuity and the Acquisition of Sesotho Passive.’’ Language 65: 56–80. Hale, Kenneth. 1983. ‘‘Warlpiri and the Grammar of Non-Conﬁgurational Languages.’’ Natural Language and Linguistic Theory 1: 5–47. Hyams, Nina. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: D. Reidel. Jelinek, Eloise. 1984. ‘‘Empty Categories, Case and Conﬁgurationality.’’ Natural Language and Linguistic Theory 2: 39–76. Karmiloﬀ-Smith, Anette. 1985. ‘‘Language and Cognitive Processes from a Developmental Perspective.’’ Language and Cognitive Processes 1: 61–85. ——. 1986. ‘‘From Meta-Process to Conscious Access: Evidence from children’s metalinguistic and repair data.’’ Cognition 23: 95–147.

factors of typology in language acquisition

55

Laughren, Mary. 1989. ‘‘The Conﬁgurationality Parameter and Warlpiri.’’ In László Marácz and Pieter Muysken (eds.), Conﬁgurationality. The Typology of Asymmetries. Dordrecht: Foris Publications. Mallinson, Graham and Barry Blake. 1991. Language Typology. Amsterdam: North Holland. Mayer, Mercer. 1969. Frog, Where are you? New York: Dial Press. Pye, Clifton. 1990. ‘‘The Acquisition of Ergative Languages.’’ Linguistics 28: 1291–330. Schieﬀelin, Bambi. 1995. ‘‘The Acquisition of Kaluli.’’ In Dan I. Slobin (ed.), 525–93. Slobin, Dan I. 1985. The Cross-Linguistic Study of Language Acquisition, vols 1 & 2. Hillsdale, N.J.: Erlbaum. ——. 1989. ‘‘Factors of Language Typology in the Cross-Linguistic Study of Acquisition.’’ Paper presented at the 10th Biennial Meeting of the International Society for the Study of Behavioural Development, Finland. ——. 1992. The Cross-Linguistic Study of Language Acquisition, vol. 3 Hillsdale, N.J.: Erlbaum. Stromqvist, S. and H. Ragnarsdottir. 1996. ‘‘On the Acquisition of Verb Argument Structure.’’ Paper presented at a workshop on ellipsis at the International Congress of Child Language, Istanbul, July. Talmy, Leonard. 1983. ‘‘How Languages Structure Space.’’ In H. Pick and L. Acredolo (eds) Spatial Orientation: Theory, research and application New York: Plenum, 225–81. ——. 1991. ‘‘Path to Realization: A typology of event conﬂation.’’ Proceedings of the Berkeley Linguistics Society 17: 480–519. Uziel-Karl, Sigal and Ruth Berman. 1996. ‘‘Where’s Ellipsis? Whether and Why there are Missing Arguments in hebrew Child Language.’’ Paper presented at a workshop on ellipsis at the International Congress of Child Language, Istanbul, July. Valian, Virginia. 1990. ‘‘Logical and Psychological Constraints on the Acquisition of Syntax.’’ In L. Frazier and J. de Villiers (eds.), Language Processing and Language Acquisition. Dordrecht: Kluwer Academic, 119–45.

Markedness and iconicity Some questions BYRON W. BENDER University of Hawai‘i at Ma¯noa

1. Introduction There has developed in linguistics a tradition for analyzing the members of grammatical categories and smaller, closed lexical categories in terms of features that are both privative and facultative, and that reveal in their design certain cross-linguistic tendencies that can be seen as iconic. It is part of what Greenberg (1966) characterizes as a ‘‘rich and complex set of notions . . . pertaining to marked and unmarked categories,’’ originating in Prague school phonology but shown, primarily in the writings of Roman Jakobson,1 as capable of extension ‘‘to the study of grammatical categories and to semantics.’’ Table 1 provides a summary of this set of notions, summarizing the characteristics set forth in Greenberg 1966, supplemented by the other sources indicated. This tradition has not been central to linguistics in America, either before or since 1957, and is sometimes not taken into account in discussions where it might aﬀord insight. It is but one of several approaches to the identiﬁcation of semantic features, some of which appear to have borrowed from it, while others seem oblivious to it.2 In this paper I take a brief look at several studies that have appeared since the midsixties and their relation to the tradition. I also note several questions that remain to be answered. There are several things to be noted about the tradition. One has already been mentioned—its origins in phonology. There were insights to be gained from analyzing the contrastive elements making up the secondary articulation of language into the substantive features of sound that some of them possessed and others lacked; similar insights could be gained from a parallel analysis of contrastive elements in the primary articulation into substantive features of meaning. It is important to note

byron w. bender

58

Table 1. Notions Pertaining to Marked and Unmarked Categories Notion

Marked manifestation

Unmarked manifestation

Facultative Expression

The marked member represents only the marked characteristic.

The unmarked member is ambiguous, and may represent what both it and the marked member have in common, or par excellence, only the opposite of the marked.

Zero Expression

‘‘A zero aﬃx cannot steadily be ‘‘. . . and a ‘non-zero’ (real) aﬃx [cannot assigned to a marked category . . .’’ steadily be assigned] to the unmarked category’’ (Jakobson 1966:270).

Syncretism

Distinctions existing in the unmarked member are often neutralized in the marked category.

Distinctions are maintained within the unmarked member.

Defectivation (a form of syncretism)

Example: That French lacks a future in the subjunctive (or is it a subjunctive in the future?) is evidence for the marked nature of both the future and the subjunctive.

The marked category lacks something that occurs in the unmarked. When a subjunctive occurs in the present, but not in the future, or a future occurs in the indicative but not in the subjunctive, this is evidence for the present and indicative being unmarked.

Deviance

Greater morphological regularity is to be found within the marked category.

Greater irregularity is to be found within the unmarked category.

Contextual Neutralization

In an environment in which an opposition is suppressed, it is the unmarked member that actually appears.

Dominance (a form of facultative expression)

The unmarked is chosen to represent both in a heterogeneous collection, e.g., Spanish los padres ‘the parents’ (literally, ‘the fathers’).

Frequency

Less frequent.

More frequent.

Iconicity

‘‘Increased morphological complexity is an icon of increased semantic complexity’’ (Haiman 1980:528). ‘‘The formal diﬀerence is quite striking: suﬃx versus no suﬃx, weightier versus lighter ending; periphrastic versus simple’’ (Matthews 1991:238).

‘‘The semantically unmarked term [in an opposition] . . . has no formal marker or . . . has shorter or less weighty exponents’’ (Matthews 1991:236).

that one began with formal, linguistic categories, and looked for substantive correlates, either phonetic or semantic.3 In both cases, the elements contrasted along a paradigmatic axis. In both cases they were members of relatively small, closed sets. On the one hand, the phonemes; on the other, the grammatical categories within a

markedness and iconicity

59

paradigm (and possibly, also, similar categories within lexical sets). Such analysis transformed the elements from a collection of monolithic entities into interrelated groupings, based on shared qualities. Generalizations could be made, based on the qualities. Networks could be identiﬁed, new characteristics noted, and new questions posed, based on ‘‘the rich and complex set of notions’’ thus revealed. The investigation seemed to lead deeper and deeper. Greenberg’s (1966) emphasis was on universals, on what was revealed about language generally. ‘‘What, if any, are the common features [that] would justify the equating of the concept of unmarked and marked categories in ﬁelds as diverse as phonology, grammar, and semantics? Is it possible to isolate one characteristic [that] might serve as deﬁnitional for this notion, which tends to take on Protean shapes? What is the connection between marked and unmarked categories and universals?’’ Greenberg pulled together the evidence that had been accumulating up until that time and posed questions such as these. He also added evidence concerning a closed lexical set of special interest to anthropologists, kin terms and their behavioral coordinates. But most of his questions, it would seem, remain unanswered at any deeper level.4 This is not to say that the ﬁeld has been vacant, or that his are the only questions that should be asked. Householder’s (1971) Linguistic Speculations contains several chapters that continue clearly in the direction set by Jakobson. Chapters 9 and 10 are commended by Matthews (1974: 153) as a source on marked/unmarked oppositions in phonology. Chapter 12 is especially helpful in fully understanding the tradition with respect to semantic oppositions. Here Householder shows how a feature analysis of the English prepositions might proceed. Jakobson, ‘‘unfortunately, never gave a full exposition of his procedures so that . . . it is diﬃcult for another linguist to duplicate his results. But we can easily see a number of them’’ (Householder 1971: 228). Householder shows how ‘‘the one reliable criterion for discovery, contrast (or opposition)’’ might be employed in analyzing this particular closed set as an example.5 Several authors in recent decades have approached the topic of iconicity in language quite generally, and have included markedness in their discussion (Haiman 1980, Matthews 1972, 1991, Croft 1990). In the ﬁrst edition of his textbook, Matthews (1972: 150–3) gives brief mention to markedness in an appendix to his chapter on ‘Properties and their exponents’, while also pointing the reader to Jakobson (1932, 1936), Lyons (1968), Householder (1971), and Benveniste (1966). He devotes an entire chapter in the second edition (1991) to iconicity in its syntagmatic as well as paradigmatic manifestations. He goes further in relating markedness to iconicity than other works I am aware of, but is slow to claim universal applicability. ‘‘To many scholars, it does not seem right to speak of explanation, even in what is plainly a branch of the humanities, unless there are laws that cover

byron w. bender

60

every instance. But we have stressed that there are no laws, only tendencies . . . A more realistic view [is that] the principle of iconicity will help to explain aspects of the history of languages’’ (1991: 243–4). Croft (1990), in a volume dedicated to Greenberg, continues clearly in the same tradition, exploring the role of markedness in implicational universals, and distilling the interrelated notions of Table 1 into four broad ‘‘criteria.’’ In response to the question he poses as to why some of the phenomena correlate so strongly, he cites both economic and iconic motivation (Croft 1990: 156, 165), with credits to Zipf (1935) in the ﬁrst instance, and to Haiman (1985) in both.

2. An instance of incongruity As an original contribution of this paper, I would like to call attention to a troublesome example from one language in which there is an incongruity in the picture associated with what seems to be a well-established tendency, that of plural to be marked and ‘‘singular’’ to be unmarked. This is found in the Marshallese demonstratives (Bender 1969: 76), where a human-nonhuman distinction is maintained in the plural but suspended in the nonplural, just the opposite of what is to be expected if plural is the marked category (see Table 2). Person categories serve as labels in Table 2 because of the close parallel with pronoun and subject marker distinctions: bok in ‘this book close to you and me’, bok e ‘this book close to me’, bok n¸e ‘that book close to you’; bok kan¸e ‘those books close to you’, armej ran¸e ‘those people close to you’, and so forth. This basic system is reinforced by a parallel set of emphatic demonstratives (bok n¸en¸e ‘that book right there close to you’, bok ka¯kan¸e ‘those books right there close to you’, armej ra¯ran¸e ‘those people right there by you’, etc.), and also by sets of person demonstratives, locative demonstratives, and copular demonstratives, all of which have emphatic counterparts. These extensive interlocking sets of demonstratives all reﬂect the basic set in Table 2, with its unexpected suspension in the singular of a

Table 2. The Marshallese Demonstratives

1 Incl 1 Excl 2 3 Remote

Plural Nonhuman

Human

Nonplural in e n¸e en¸ eo

kein ka¯ kan¸e kan¸ ko

rein ra¯ ran¸e ran¸ ro

/yin/ /yéy/ /n¸ey/ /yen¸/ /yew/

/kéyin/ /kay/ /kan¸ey/ /kan¸/ /kew/

/réyin/ /ray/ /ran¸ey/ /ran¸/ /rew/

markedness and iconicity

61

contrast made in the plural. Together they play a key role in noun phrase formation. It seems diﬃcult, or impossible, to view them as in any way peripheral, or transitional between states that better conform to the expected markedness complex. Even here, the plural has formal characteristics expected of marked members. Both the ‘1 incl’ and the ‘2’ forms have greater phonetic substance in the plural than in the singular, consisting as they do of kV- and rV- accretions to the nonplural. Also, the ‘1 excl’, ‘3’, and ‘remote’ forms have the full-consonant onsets in k and r, in contrast with the semi-consonant (glide) onset /y/ in the nonplural. Thus this example from Marshallese is not one that we should identify as a counterexample to the strong tendency elsewhere for plural to be the marked category. Instead, we must note an incongruity in one of the associated tendencies in the ‘‘complex set of notions.’’ In most of the Micronesian sister languages I have examined (Gilbertese, Ponapean, Mokilese, Trukese, Woleaian, Ulithian—all except Kusaiean) there is a partially cognate demonstrative system distinguished by person and number, with a k element throughout the plural, but no human distinction in either singular or plural. Distinguishing human in demonstratives thus appears to be a Marshallese innovation whose motivation is obscure. The only possibly parallel phenomenon I am able to relate it to at present is the distinctive treatment given human objects of transitive verbs in the plural. The 3pl object pronoun (er /yér/ ‘they’) must be used for human referents in anaphoric contexts, whereas singular human referents in parallel contexts may be represented simply by an object marker suﬃxed to the verb, just as both singular and plural nonhuman referents normally are (see Bender 1984:447–8). It may not be coincidental that this plural pronoun has an r element distinguishing it from the 3sg object pronoun e /yéy/. Both, like the 1sg and 2sg object pronouns, are used only for human referents. Use of the 3sg object pronoun in parallel anaphoric contexts is unusual, and thus emphatic.

3. Some possible aberrations and some further questions Less clearly central is Chafe’s (1970) work on semantics, which gives overt mention to features only when they are present. That they are intended as privative, however, is made clear by his formalisms, as for example, ‘‘V → state: The fact that the arrow has a broken shaft means that its application is optional. The fact that it has a double head means that it must be read is further speciﬁed as. . . . This rule says that a verb, abbreviated V, may be speciﬁed optionally as a state. By implication, if a verb is not so speciﬁed it is a nonstate or, informally, a ‘happening’ or event’’ (Chafe 1970: 99). Although he does not mention Jakobson, at one point he indicates familiarity with the work of Hjelmslev, and in discussing the features by which

62

byron w. bender

nouns may be further speciﬁed, says that ‘‘A human noun is given ‘gender’; that is, its sex is speciﬁed. There are certain complexities in this area, but I shall assume here that a human noun, if not otherwise speciﬁed, is understood to be what we would normally term masculine. In other words, I shall take masculine to be the unmarked state of a human noun, and feminine, then, to be the marked state. In general, not knowing the sex of a human noun, we treat it conceptually as male.’’ He thus intends not only the concepts ‘‘marked’’ and ‘‘unmarked,’’ but also that a facultative relation exists between them. His work thus can be counted within the tradition under discussion here. The inventory of features he proposes for the further speciﬁcation of nouns consists of count, potent, animate, human, feminine, and unique. Although developed in an English-centered discussion, this inventory seems intended for more universal application. None of these features needs to be replaced for Onondaga when he applies his model to that language, although additional features are added for the inﬂection of Onondaga nouns and the speciﬁcation of pronouns—features of person (ﬁrst, second, inclusive), number (plural, dual), and case (dative), for example. Another approach, one referred to as componential analysis or semantic decomposition and one that may at ﬁrst glance appear similar because of its use of privative features, is presented by O’Grady and Dobrovolsky in their textbook entitled Contemporary Linguistics.6 It is said to be ‘‘most useful for uncovering and representing similarities among semantically related words . . . A few simple features allow us to express the similarities and diﬀerences among subclasses of people—men, women, boys, and girls’’ (O’Grady 1997: 251–2). An accompanying ﬁgure for these four nouns shows all four as being [+human], two as being [+male], and two as being [+adult]. (Those not labeled [+male] and [+adult] are labeled [–male] and [–adult] respectively.) The most recent edition adds a parenthetical statement, ‘‘Nothing depends on the choice of feature names here; the analysis would work just as well with the feature [±female] as [±male]’’ (251). Presumably this extends to the choice of ‘‘adult’’ over ‘‘young.’’ An exercise at the end of the chapter asks the reader to attempt to provide the semantic features associated with dog, puppy, cat, and kitten, and also asks how the pairs dog–puppy and cat–kitten are diﬀerent from man–boy and woman–girl. The instructor’s manual to the 2nd edition (Aronoﬀ et al. 1993: 39) gives a solution that shows dog and cat as [±adult], and puppy and kitten as [–adult], and answers the second question by saying that ‘‘whereas man and woman are usually understood to be [+adult], dog and cat can be used for either [+adult] or [–adult] animals. Whether or not it is true that humans diﬀer from canines and felines in this regard, it is ironic that dog and cat are here characterized as what in markedness tradition would be the unmarked members, and puppy and kitten as marked, but with the speciﬁcation [–adult]. From a markedness point of view, this is evidence that the

markedness and iconicity

63

feature has been wrongly named, and that it should be instead [+young], with the signs switched from what they were when the feature was named [+adult]. Whether the human situation diﬀers basically—that man and woman are usually understood to be [+adult] and cannot be used for either [+adult] and [–adult] humans—is open to some question. ‘‘Is a girl a special kind of woman, or a woman a special kind of girl?’’ and ‘‘Is a boy a special kind of man, or a man a special kind of boy?’’ might be better ways to state the question. ‘‘Young woman’’ and ‘‘young man’’ are certainly common enough appellations, and fairly neutral in connotation, whereas ‘‘old boy’’ and ‘‘old girl’’ are narrow, dissonant, and pejorative by comparison. Although this question obviously needs further study, classic markedness theory will probably sustain the view that the terms boy and girl, as well as puppy and kitten, are marked in comparison with man, woman, dog, and cat, respectively, and that the feature involved is thus more properly [+young]. Clearly, componential analysis as here presented is separate from and seemingly uninﬂuenced by markedness tradition. ‘‘This approach,’’ it is said, ‘‘has long been used to analyze the meanings of certain types of nouns in terms of semantic features’’ and has as an obvious advantage that ‘‘it allows us to group entities into natural classes (much as we do in phonology)’’ (O’Grady 1997: 251). Thus the use of features such as ‘‘male’’ and ‘‘female’’ rather than ‘‘masculine’’ and ‘‘feminine’’ may well be intentional, as part of a quest for what is natural biologically rather than linguistically. On the other hand, where any distinction can be made, markedness tradition is concerned with how semantic distinctions are treated grammatically and lexically, rather than how they inhere in nature, paralleling the contrasts that often exist between folk taxonomies and scientiﬁc classiﬁcations. The distinctive feature analysis of grammatical and lexical categories was modeled originally on the distinctive feature analysis of phonological categories, and has been further inﬂuenced by the new deployment of given features by Chomsky and Halle (1968) in The Sound Pattern of English, and by the new contextual deﬁnitions given to ‘‘marked’’ and ‘‘unmarked’’ within their system (see Householder 1971: 153). No longer do features simply identify what distinguishes the phonemes of a language from each other. They are given the additional function of spelling out what is nondistinctive as well—all the contextual phonetic detail—the allophones. Although still referred to as phonological ‘‘distinctive features,’’ they have become in eﬀect ‘‘phonetic features’’ in this latter interpretive role. Although it is possible to say that [+nasal] is the unmarked value for vowels when they precede [+nasal] consonants, but that [+nasal] is the marked value for such consonants when they bear this value in a context in which [–nasal] consonants can also occur, it is important to recognize that this diverts markedness from its original mission of dealing only with what is distinctive.7 Some authors have attempted to make grammatical and lexical distinctive fea-

64

byron w. bender

tures contextual in parallel fashion, to have them serve as semantic features with nondistinctive interpretive functions, or to have their markedness dependent upon the semantic environment. Thus Bierwisch (1967: 244) says that ‘‘for several reasons one may consider neuter as the unmarked gender in German, as opposed to feminine, and that is certainly true for such nouns as Haus ‘house’, Dach ‘roof’, for all diminutives, such as Häuschen ‘small house’, Männchen ‘little man’ (perhaps even in Mädchen ‘girl’), but it cannot be considered unmarked in Weib ‘woman’.’’ He would have ‘‘nouns [such] as Haus and Dach . . . unmarked for the gender features in the lexicon, and . . . later be interpreted as characterizing neuters in the presence of the [semantic] feature [–Animate]. In the same way, nouns [such] as Frau ‘woman’, Tochter ‘daughter’, and Mutter ‘mother’ . . . remain unmarked for gender-features, these being interpreted as feminine in the presence of the semantic feature [+Female], while Weib on the other hand must be marked . . . for gender in such a way that the feature interpretation rules yield the irregular neuter in the presence of [+ Female].’’ Masculine is not mentioned in this discussion, except in a footnote, where he notes the even greater complexity created by ‘‘animate nouns [such] as Katze ‘cat’, Hund ‘dog’, where the feminine and masculine gender respectively behave in a sense as unmarked in representing either the female or male animal, respectively, or the whole species.’’ This approach might be summed up as saying that we will consider it normal or unmarked when grammatical gender coincides with biological gender—or what some authors refer to as natural gender—and nonnormal or marked when it does not. This is quite a diﬀerent matter from considering the three classes of nouns created by grammatical gender, and asking which of the three classes behaves as more marked grammatically, and which as less marked, according to the complex set of notions summarized in Table 1.8 Of course, one of the diﬃculties here is that we are not conﬁning our consideration to a closed grammatical set, or even to a closed lexical subset, those arenas in which the notions of Table 1 are most clearly manifest. Gender is unique among morphosyntactic categories in that it is primarily a reﬂection of the diﬀerential behavior of the members of a major lexical category, and its referential implications are often only partial. Classes of nouns are created by virtue of their having diﬀerent inﬂections and/or diﬀerent agreement patterns with the members of other lexical categories. The correlations between the classes thus created and real-world distinctions of some sort serve only as aids in gender assignment—determining the class to which a given noun belongs—and are invariable for any given noun.9 Aside from these correlations, genders are thus closer to declensions than they are to morphosyntactic categories such as number, person, and even case. Even when such real-world correlations exist, as with animacy or sex, gender is still

markedness and iconicity

65

Table 3. The so-called pronominal inﬂection of German (from Bierwisch 1967:245) Singular Masculine

Plural

Neuter

Feminine

Masculine/ Neuter/Feminine

Nominative

dieser

dieses

diese

diese

Accusative

diesen

dieses

diese

diese

Dative

diesem

diesem

dieser

diesen

Genitive

dieses

dieses

dieser

dieser

unique in its inherency. Whereas a given noun may be inﬂected for diﬀerent numbers, possessors, or cases, it is inherently a member of just one noun class. Bierwisch (1967: 248) himself seems to realize this when he later turns to ‘‘grammatical gender features,’’ and refers the reader to the discussion summarized above as having to do with ‘‘the markedness of lexical gender features.’’ Concerning ‘‘grammatical gender features [emphasis mine]’’ he invokes several of the notions found in Table 1. (1) He says that ‘‘there is certain evidence that the feminine gender must be marked as opposed to masculine and neuter: feminine always has the fewest case distinctions,’’ which I interpret to mean that the feminine (singular) shows the most syncretism among the cases, between nominative and accusative, and between dative and genitive. (Table 3 reproduces his display of a paradigm he refers to as ‘‘the so-called pronominal inﬂection of German.’’) (2) He says ‘‘this is somewhat in conﬂict with the fact that in the case of gender neutralization in the plural, the forms of the feminine appear, with the exception of only the dative,’’ referring to the notion of ‘Contextual Neutralization: In an environment in which an opposition is suppressed, it is the unmarked member that actually appears.’ If the plural forms were to coincide with the feminine forms completely, this would be evidence that the feminine is the least marked of the three genders. Because the match is less than complete, this evidence is questionable, and in any event equivocal. (3) He points out that ‘‘the masculine always shows the greatest case diﬀerentiation, which should be a characteristic of the unmarked gender . . . ,’’ referring to the notion of deviance. (4) Finally, as evidence of the neuter’s being unmarked, he says that ‘‘in almost all cases of syntactic and morphological neutralization of gender, the neutral [sic!] form shows up, for instance, in the neutral Es in Es is gut, daß du gehst ‘it is good that you are going’ ’’ (Bierwisch 1967: 248). At least two binary features are necessary to distinguish three entities, and he settles for a scheme that shows Neuter as least marked: Masculine

Feminine –

Masculine Neuter + –

byron w. bender

66 Feminine

+

–

–

Why he concludes that the evidence of (4) for the neuter’s being unmarked outweighs that of (3) for the masculine is not clear. One suspects that it might relate to the commonly held view of nature in which animacy and humanness are built upon a more basic inanimate order, such as that portrayed by Chafe (1970), in which the further speciﬁcations of animate, human, and feminine are built upon an inanimate, unmarked base. Although ‘‘it’’ may be count and—if an animate-like force of nature—potent, it is not speciﬁed as human, and certainly not as feminine. The following system is implied. it he she

Human Feminine – + – + +

Although this is a commonly held view of the natural order, is it what we ﬁnd in language? Peinovich (1979), in looking at the inﬂection of Old English nouns, ﬁnds two types of dialects, one in which nouns are subcategorized according to the natural gender distinctions of inanimate, animate, male, and female, and the other in which ‘‘O[ld] E[nglish] nouns are subcategorized according to grammatical gender as well. This requires a separate set of features [that] will indicate the grammatical gender, as opposed to the sex, of each noun in the lexicon. Since there are three grammatical genders, two binary features will be necessary. The two features used here, [±Neuter] and [±Feminine], are selected in accordance with Bierwisch’s universal markedness convention. The masculine nouns are the most numerous in the language, and since it is their inﬂectional system [that] survives into M[iddle] E[nglish] and N[ew] E[nglish], they will be designated the unmarked gender. The other two genders will be marked by one of the two features [Neuter] or [Feminine]. The system is represented in [the following ﬁgure]’’ (Peinovich 1979: 57–8). Masculine Feminine Neuter

Neut – – +

Fem – + –

The condition that is being invoked is ‘‘a general [one] imposed on rules of inﬂection’’ that according to Bierwisch (1967: 269) explains ‘‘certain widely discussed facts of asymmetry with respect to syncretization and diﬀerentiation in inﬂection [that] are assumed to be a universal characteristic of human language.’’ These facts turn out to be from among the complex of notions set forth in Table 1, attributed by Bierwisch to Jakobson and Hjelmslev, and secondarily to Greenberg. The condition

markedness and iconicity

67

is stated as follows, ‘‘Inﬂectional rules or subrules referring to unmarked values cannot precede rules referring to marked values’’ (Bierwisch (1967: 269). Bierwisch does not himself give it the name of ‘‘universal markedness convention,’’ but it serves his interpretive aim by identifying the less regular and usual before more general rules are applied. It is in this sense that Peinovich ﬁnds the masculine to be the most widespread pattern and the one that survives the transition from Old English to Middle English. We are left to wonder whether neuter as the marked category may have more widespread applicability beyond these possible instances in Germanic. We are reminded of the greater syncretism between the nominative and accusative shown by neuter nouns than by either masculine or feminine in other Indo-European languages such as Latin and Greek. This is a characteristic to be expected of a marked category, and if further investigation does show neuter to be more marked, especially in comparison with masculine, it will underscore the disparity between grammatical markedness and certain conceptions of the natural order.10

4. Conclusions It is possible to use binary features and privative terms, including the terms ‘‘marked’’ and ‘‘unmarked,’’ for a number of quite valid purposes in linguistics. One direction of investigation that continues to present a number of challenges, but promises much—and is the main focus of this paper—is the question as to what the complex set of notions in Table 1 can be made to reveal to us further about the iconicity and design of language. What is the origin of these tendencies that have been observed, and what is their function? Only a few tentative answers can be suggested at this point. For one thing, they bring to language a certain economy and eﬃciency.11 Inﬂection is by its nature obligatory. We are forced to choose from among small closed sets of markers in each use we make of a member of a lexical category that is inﬂected. But there is a way to sometimes avoid the choice, at least partially. The key to this avoidance lies in the asymmetrical nature of facultative expression, which makes it possible for those things that can go without saying to do so. That which is semantically unmarked need not be formally marked, thereby permitting expression with less phonetic substance, and sometimes even with zero expression. The result in the long run is shorter expression. If the unmarked can stand for both terms of an opposition when it is not important to diﬀerentiate between them, it will be opted for more frequently than the marked—both when it serves its purpose par excellence, and when it represents a suspended opposition. Although frequency thus may appear, at ﬁrst glance, to be a symptom of the asymmetry, it can more pro-

68

byron w. bender

foundly be seen as its raison d’etre. Shorter expression, and sometimes silence, are made possible more often. But why can we speak only of tendencies, and not of laws, as Matthews (1991: 243) reminds us? Why do the notions of Table 1 not always point in the same direction? What is the source of the inconsistencies, and of the incongruities such as the one detailed in Section 2? How and why do they appear, and how will they be worked out? We obviously have much to learn along these lines. We will do well to keep our investigation of markedness related to an interest in iconicity and language design in order to avoid being distracted by some of the other uses to which it has been put.

Notes 1. Greenberg also credits Hjelmslev (1953) with contributions to the tradition at a number of points in his exposition. 2. In fact, I am concerned that it is still not part of the education of linguists generally, for it would seem to have much to oﬀer to our understanding of semantics and of language design. I was happy to learn from conversations with Professor Blake when he visited Hawai‘i in the early nineties, and from reading Blake (1994), for example, that he is fully conversant with the tradition and has in fact made use of it in his own work. For this reason, I am especially pleased to have my brief discussion of the subject included in this volume. 3. Householder (1971: 169) stresses the importance of starting with the linguistic oppositions when working with phonological distinctive features. His advice can be extended to the subject of this paper by substituting ‘‘semantic’’ for ‘‘phonetic’’ at two places. After discussing in detail several questions that have been raised in connection with binary oppositions, he says, ‘‘What more do we need to consider? The importance of starting with oppositions, not with phonetic qualities, should not be forgotten, since it is reasonable to believe in a small number of kinds of opposition for all languages, but against all evidence to believe in a small number of phonetic qualities.’’ 4. Croft (1990) is something of an exception to this statement (see below: the ﬁnal paragraph of Section 1). 5. Householder knew Jakobson’s approach ﬁrsthand, having participated in seminars with him in New York City in the early forties. 6. This inﬂuential textbook appears to be attempting to live up to its name, having gone through three editions each in Canada and the USA. since the mid-eighties. 7. As Croft puts it, ‘‘Markedness has . . . been adapted by both the generative and the typological approaches to linguistic theory, not surprisingly in rather diﬀerent ways. As a consequence, markedness in generative grammar is considerably diﬀerent from markedness in typology . . .’’ (Croft 1990: 64).

markedness and iconicity

69

8. This would probably be easier to do for a language like Swahili, where the genders are more numerous and not so clearly related to nature. 9. See Corbett 1991 for a stimulating discussion of gender, and for details of gender assignment. 10. Greenberg (1966: 81) found, with evidence from Indo-European, Dravidian, and Algonkian languages, that ‘‘where a neuter exists alongside of a masculine and feminine, the neuter is the most marked category and can be opposed to the masculine/feminine.’’ The only counterevidence he mentions is that also cited by Bierwisch (1967), the German neuter es as representative of all three genders in Es is ein Tisch ‘It is a table’, and the Russian equivalent. 11. I am by no means the ﬁrst to make this observation. See the discussion of Croft (1990) at the end of Section 1.

References Aronoﬀ, Mark, Michael Dobrovolsky and Jane Rees-Miller. 1993. Instructor’s manual to accompany O’Grady et al. (eds.). Bender, Byron W. 1969. Spoken Marshallese. Honolulu: University of Hawai‘i Press. ——. 1984. ‘‘Object marking in Marshallese.’’ In Byron W. Bender (ed.), Studies in Micronesian Linguistics. Canberra: Australian National University, 443–65 [Paciﬁc Linguistics C–80]. Benveniste, E. 1966. ‘‘Structure des Relations de Personne dans le Verbe.’’ In Problèmes de Linguistique Générale, 225–36. Bierwisch, Manfred. 1967. ‘‘Syntactic Features in Morphology. General Problems of SoCalled Pronominal Inﬂection in German.’’ In To Honor Roman Jakobson: Essays on the Occasion of His Seventieth Birthday, vol. 1. The Hague: Mouton, 239–70. Blake, Barry. 1994. Case. Cambridge: Cambridge University Press. Chafe, Wallace L. 1970. Meaning and the Structure of Language. Chicago: University of Chicago Press. Chomsky, Noam and Morris Halle. 1968. The Sound Pattern of English. New York: Harper and Row. Corbett, Greville. 1991. Gender. Cambridge: Cambridge University Press. Croft, William. 1990. Typology and Universals. Cambridge: Cambridge University Press. Greenberg, Joseph H. 1966. ‘‘Language Universals.’’ In Thomas A. Sebeok (ed.), Current Trends in Linguistics, vol. 3, Theoretical Foundations. The Hague: Mouton, 61–112. Haiman, John. 1980. ‘‘The Iconicity of Grammar. Isomorphism and Motivation.’’ Language 56: 515–40. ——. 1985. Natural Syntax. Cambridge: Cambridge University Press. Hamp, Eric P., Fred W. Householder and Robert Austerlitz (eds.). 1966. Readings in Linguistics II. Chicago: University of Chicago Press.

70

byron w. bender

Hjelmslev, Louis. 1953. Prologomena to a Theory of Language. Baltimore: Waverly Press [International Journal of American Linguistics, Memoir 7, Indiana University Publications in Anthropology and Linguistics]. Householder, Fred W. 1971. Linguistic Speculations. Cambridge: Cambridge University Press. Jakobson, Roman. 1932. ‘‘Zur Struktur des Russischen Verbums.’’ Reprinted in Hamp et al., 22–30. ——. 1936. ‘‘Beitrag zur Allegemeinen Kasuslehre.’’ Reprinted in Hamp et al., 51–89. ——. 1966. ‘‘Implications of Language Universals for Linguistics.’’ In Joseph Greenberg (ed.), Universals of language, 2nd. ed. Cambridge, Mass.: MIT Press. Lyons, John. 1968. Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press, 263–78. Matthews, P. H. 1972. Inﬂectional Morphology. Cambridge: Cambridge University Press. ——. 1974. Morphology. An Introduction to the Theory of Word-Structure. Cambridge: Cambridge University Press. ——. 1991. Morphology. 2nd. ed. Cambridge: Cambridge University Press. O’Grady, William. 1997. ‘‘Semantics. The analysis of Meaning.’’ In William O’Grady and Michael Dobrovolsky (eds), Contemporary linguistics, 245–87. ——and Michael Dobrovolsky (eds). 1997. Contemporary Linguistics. An introduction. 3rd. ed. (US edition prepared by Mark Aronoﬀ.) New York: St. Martins Press. Peinovich, Michael P. 1979. Old English Noun Morphology: A diachronic study. New York: Elsevier North Holland [North Holland Linguistic Series 41]. Zipf, George. 1935. The Psychobiology of Language. An Introduction to Dynamic Philology. Cambridge, Mass.: MIT Press.

Throw the baby from the window a cookie English and Pennsylvania German in contact KATE BURRIDGE La Trobe University

1. Introductory remarks In 1945 in his now famous paper Amish Triple Talk, William Frey had this to say about the English of the Amish in Pennsylvania: The English used by the Amish . . . has all the earmarks of the type of speech found among any other Pennsylvania Dutch groups. It can be brieﬂy described as American English built on a framework of Pennsylvania Dutch phonemic patterns and interjected continually with whole or part loan-translations from the dialect.

Similar descriptions of the English of Pennsylvania German groups are still common place. Druckenbrod (1981: 10), for example, in the introduction to his teaching grammar of Pennsylvania German states: When the Pennsylvanian German [goes] to speak English (which [is] for him a foreign tongue), he [is] inﬂuenced by the German syntax or sentence structure. For some such English as ‘Run the steps once up’ is largely unintelligible unless you know that it’s really the Pennsylvania German ‘Schpring mol die Schdeeg nuﬀ’. This is the reason for many of the so-called ‘quaint’ or ‘cute’ sayings of the Pennsylvanian German trying[my emphasis] to speak English. ‘Throw the cow over the fence some hay’ is but another example of this process.

And more recently still, April McMahon, in her 1994 historical linguistics textbook, describes (citing Lehiste) how ‘‘speakers of Pennsylvania German . . . import the German order of modiﬁers into English, giving constructions like throw the baby from the window a cookie.’’ (p. 206). The only time you are likely to hear Pennsylvania German speakers produce

72

kate burridge

examples like these are when they themselves are imitating the sort of English which they imagine they speak—a type of English they describe as verhoodelt (or ‘mongrel’)! Otherwise such constructions are a ﬁctional stereotype—nonetheless a powerful stereotype and one which even the speakers themselves believe. In this paper I examine, as part of the whole question of interference between Pennsylvania German (PG) and English (E), where such a stereotype could have originated. In particular I look at the contact situation in Waterloo County, Canada (approximately 100 kms west of Toronto). To get the complete picture, I have considered both directions of interference: PG → E and E → PG. The paper divides itself into four main parts. The ﬁrst section provides some necessary socio-historical background details, which are important for understanding the full implications of the ﬁndings. The second section examines the question of PG-E interference and the third section the other side of the coin; namely, E-PG interference. Section four is a discussion of these ﬁndings. Here I need to stress two things. One, I have concentrated on the parameters of religious conservatism and to a lesser extent ethnicity, and have ignored other aspects of the social setting. Although I believe age and sex are also important social parameters in this culture, they are less revealing for what they can tell us about patterns of interference. Two, although the data has not been quantiﬁed, I believe that what it is able to indicate about the nature and direction of interference is both instructive and insightful.1 1.1 Setting the scene Like any speech community, the Pennsylvania Germans do not represent a totally homogenous group. What exists is a complex design of social, cultural and religious diversity which must be taken into account in any appraisal of the linguistic situation. This is particularly true for patterns of interference. The PG-speaking group examined here are the Mennonite Anabaptists of SwissGerman origin, who left Pennsylvania for Canada after the American War of Independence. The majority settled in Waterloo County, where they remain today (with the growth of the cities of Kitchener and Waterloo, however, some of the farming population are now moving out into more isolated areas). Since the 1870s, the Mennonites (more than any other Anabaptist group) have been experiencing continued factionalism and the result is now a complex pattern of diﬀerent splinter groups. These splits result from the question of conformity and the diﬀerent degrees to which members will interpret the scriptural quotation—And be not conformed to this world. It seems that biblical exactness is an important factor in maintaining their separate status and close sense of community, but, paradoxically, is also a major reason for this continuing splintering. To the outsider, it presents a confusing picture —diﬀerences may be as subtle as a diﬀerent dress or bonnet style, but of course can

throw the baby from the window a cookie

73

also involve more serious issues like the question of Sunday School, or adoption of new farming methods, for example. The Mennonite (and also Amish) community is normally divided into two major groups—the Plain Folkand the Non-Plain Folk.The Plain Folk are religiously the most conservative and are typically both rural and isolated. In Canada, they are known as the Old Order Mennonites (OOMs) or colloquially asthe horse and buggy people. They have a very distinctive style of dress which has changed little over the centuries, they drive only horse and buggies, and are opposed to modern conveniences like cars, radios, television, telephones and so on.2 Although clearly a purely religious denomination in origin, the OOMs are best now described as a distinct cultural-ethnic group with their own unique traditions, beliefs, customs, social practices and of course language. They are a religious sect, but also a distinct culturalethnic minority. Not surprisingly, their simple and conservative way of life attracts much attention from the wider community. The Non-Plain Folk comprise two main groups. Those in the ﬁrst least conservative group are the so-called Progressive Mennonites (PMs). Their members are generally indistinguishable from mainstream Canadians. The second group is really an intermediate group and its members are the Transitional Mennonites (TMs). They include, for example, the Markham Mennonites who follow many of the same beliefs and behaviour patterns as the OOMs but have accommodated more to modern ways. For example, black cars now replace the horse and buggy (earning them the nickname black-bumper Mennonites—until recently, the chrome on these cars was also painted black). Their dress, like their cars, is plain and without decoration, but not in the same distinctive tradition of the OOMs (for that reason, they are sometimes referred to as the Modern Plain). But this is painting a neat picture of what in reality is nowhere near as orderly. Clearly, the best way to view the situation is as a continuum—from the ultraconservative OOMs through the TMs to the least conservative PMs. It is also possible to plot the diﬀerent speakers along this continuum. Competence in PG accords generally with degree of religious conservatism. Of course it is not that religion directly bears on the linguistic abilities of these people, but it is on account of what their religion entails. The OOMs are bilingual E and PG. But what is crucial is that this bilingualism has the support of a stable diglossic situation. PG, as the L-variety, is usually only spoken and is the language of both home and community. E, as the H-variety, is read and written and only spoken when dealing with non-PG speaking outsiders.3 For the OOMs, language and religion are closely entwined. PG plays a crucial part in maintaining their separate and peculiar status and for this particular group, it is in no danger of dying. Among the Non-Plain Folk, however, language proﬁciency ranges from fully competent speakers to real semi speakers. But even the most competent speakers have not the support of diglossia—in other words, E

74

kate burridge

and PG do not have their own set of distinct social functions—and the situation is an unstable one. Ultimately, the shift to E is a certainty for this group (cf. Burridge 1997 and also papers by Huﬃnes 1980a, Louden 1987, Moelleken 1983 for a comprehensive account from the Pennsylvanian perspective). This then forms the backdrop to the following discussion of interference. The term interference is being used here in the more general sense of inﬂuence (it is of course not meant to be pejorative). An example of interference would therefore include any prosodic, phonological, grammatical or lexical element which appears in a language on account of contact with another. The speakers used for this particular study come from the above three Mennonite groups and are summarized below. The initial part of the paper, however, concentrates on the more proﬁcient PG speakers and relates speciﬁcally to taped conversational data collected from the ﬁrst two groups only (the ultraconservative OOMs and the less conservative TMs). The paper concludes with some discussion of data collected from the modern group. (a) Old Order Mennonites (OOMs)—10 speakers regular use of PG (family, community and work); infrequent use of E (outsiders only) (b) Transitional Mennonites (TMs)—7 speakers regular use of both PG and E (c) Progressive Mennonites (PMs)—7 speakers regular use of E (family, community and work); ability to use PG where necessary

2. Interference from PG in E As I mentioned in the introduction, in Canada and in Pennsylvania alike, there is a popular perception of the Pennsylvania Germans as speaking a form of mongrel or verhoodelt English. In the broader Canadian community, evidence for this stereotype exists in the form of jokes and anecdotes. They are full of examples of the sort of ‘‘quaint’’ and ‘‘cute’’ sayings that the Pennsylvania Germans (or ‘‘Pennsylvania Dutch’’ as they are popularly known) are supposed to use. The following are just some examples: Throw Father down the stairs his hat once. It wonders me if it don’t gif a storm. The faster I go, the behinder I get. Becky lives the hill just a little up.

throw the baby from the window a cookie

75

Yonnie stung his foot with a bee un it ouches him terrible. Ve get too soon oldt un too late schmart. Fictional clichés like these are also reproduced on tea towels, beer coasters, wall plaques and serviettes. They appear in linguistic descriptions too, where unfortunately they now get reproduced as fact (cf. examples in Section 1 above). Like most stereotypes, they have become part of the shared cultural knowledge of the area, despite the fact they don’t have any actual basis in reality. How this particular stereotype arose I will consider below. For the moment I want to show that there is surprisingly little in the way of interference in the English of these PG-speakers —certainly there is nothing remotely resembling verhoodelt Englisch.4 Without yet taking into account the question of socio-cultural diversity, consider ﬁrstly the basic patterns of interference found within the two groups of PG-speakers. These can be arranged into a simple hierarchy of decreasing strength. 2.1 Lexical The least amount of interference occurs at the lexical level. The few items which could be described as actual lexical transfers from PG to E are conﬁned to untranslatable concepts like Freindschaft (which inadequately translates into English as ‘‘family’’, ‘‘relations’’) and words to do with buggy technology, for example. 2.2 Grammatical At this stage, my ﬁndings suggest minimal grammatical interference. (Of course because of the nature of grammar, lack of occurrence is never conclusively negative evidence.) Nonetheless, for reasons which will be discussed later, the English of these speakers, if we put it against the standard, has remarkably ‘‘proper’’ grammar! The data shows little evidence of the syntactic modiﬁcations which have become the clichés of PG English. Word order is standard, and there are no examples of the so-called ‘‘redundant adverbs’’ like once. The only possible exceptions are the adverbs yet and already, which have become sentence-ﬁnal aspectual markers in PG E. However, these markers are also typical of the speech of Waterloo County people

Prosodic Phonological Grammatical Lexical

strong quite strong weak very weak

Figure 1. Interference PG → E.

kate burridge

76

generally (including those not of PG heritage). They undoubtedly have a German origin, but are by no means peculiarly PG. (Other German-speaking groups, who arrived in various waves during Ontario’s history—for example, the Roman Catholic and Lutheran communities—have assimilated into mainstream culture and have now surrendered their language.) My ﬁndings do suggest a greater preference for ﬁnal placement of verbal particles (I ate my dinner up versus I ate up my dinner). This follows the expected pattern of PG separable verbs. However, since both orders represent alternative patterns for E and are motivated primarily by discourse factors (considerations of topicality, for example; cf. Chen 1986), for the moment it is impossible to say, whether or not this particular word order is any more frequent in PG English than the standard variety of the area. Certainly, there is no evidence for the postposed prepositions of verhoodelt Englisch. Examples like Becky lives the hill just a little up never occur. 2.3 Phonological Most interference occurs at the prosodic level. In this case, it is a distinctive riserise-fall pattern which appears in many diﬀerent sentence types, including both questions and statements (cf. Huﬃnes 1986 for a detailed account of PG intonation). This pattern can spread across the whole utterance, but more often than not in the data here condenses into the last item, as the following examples show:

(1) What are you going to do for dinner?

(2) That is my niece. Figure 2 shows the segmental features of phonological interference. (These are based both on recorded material, as well as verhoodelt features). They are arranged in a tentative implicational hierarchy. For examples, speakers showing the ﬁrst feature, would show all other eight features and so on. The idea for this hierarchy comes from Enninger et al.’s (1984) table showing features of phonological interference found in Raith’s (1981) study5. These patterns of phonological interference distribute diﬀerently throughout the Mennonite community. Of the two groups being considered here, it is the less conservative TMs (as opposed to the OOMs) who have the greatest amount of interference. They show evidence of features 4–9. At ﬁrst consideration, this appears a surprising distribution—after all the TMs use more E and have the greatest contact

throw the baby from the window a cookie 1. 2. 3. 4. 5. 6. 7. 8. 9.

r realized as ð z, v b, d, g

i

Ø s z s, f p, t, k l υ, ɔ ε, ə

77

(word ﬁnally) (word ﬁnally)

(in unstressed positions)

Figure 2. Hierarchy of phonological features. with outsiders. In Section 4, I discuss the reasons for this apparent discrepancy. The OOMs, on the other hand, show much less interference. Only features 6–9 appear uniformly for all speakers. Neither group shows features 1–3. In studies coming out of Pennsylvania, there are two features which have additionally been identiﬁed as characteristic of PG English (cf. for example, Huﬃnes 1980b, 1984): v w

w v, ß

These are probably the two most clichéd features of PG English and the fact that they are totally absent from the data here shows once more how removed verhoodelt Englisch is from reality, at least in the Canadian context. In fact, the only stereotypical feature to show up consistently here is the devoiced version of [ ]. 2.1 Why verhoodelt Englisch? Many of these stereotypical features appear in the E found in recent German immigrants to Canada, as you might predict on the basis of a contrastive analysis of PG and E. The question is then, why have they come to be identiﬁed so closely with PG English? The answer lies in the early diglossic history of the Pennsylvania Germans. Up until well into the twentieth century, the diglossic situation which existed in Pennsylvania and Canada resembled that of modern-day Switzerland. PG was the L-variety, as it is still today, and High German was the H-variety—the language of education, worship and (with the exception of some folk literature) all writing. The Pennsylvania Germans were isolated and there was simply no need for them to learn English. One of the prefaces to Horne’s nineteenth century elocution manual, for example, describes ‘‘their [= the Pennsylvania German’s] entire ignorance of the English language [which for them] is as much a dead language as Latin and Greek’’ (p.7). In his manual, Horne suggests drills which exercise precisely those aspects

78

kate burridge

of phonology which are now ﬁxed in the modern-day characterizations of verhoodelt Englisch. The thing about stereotypes is that they never develop—they petrify. Over time they acquire an authenticity that seems immune to change. And of course, the horse and buggy community, being so highly visible and so eccentric to outsiders, are particularly vulnerable to this sort of stereotyping. In truth, their English is closer to the prescriptive ideal of ‘‘proper’’ English than the English of many of their stereotypers. (See discussion below on why this is so.) The question is why would the speakers themselves accept such a stereotype? As mentioned earlier, the OOMs willingly denigrate their verhoodelt Englisch. The answer to this probably lies in the importance which the Old Orders place on the need for what they call Demuut or ‘‘humility’’. Verhoodelt Englisch is much like their PG itself. PG suﬀers all the negative connotations so often associated with ‘‘dialect’’ or ‘‘vernacular’’. Under more usual circumstances this low esteem would spell almost certain decline and shift to E. But nothing could be further from the truth in this instance. In fact here, I have always believed, lies one of the main secrets of the language’s success at survival. For these people, PG has an important symbolic function within the community. The Old Orders view their dialect of German, much like their plain dress and buggy, as a sign of humility and therefore as a good and Godly thing. If High German is the word of God, the language of their Scriptures, then the low status of the dialect variety which they speak is seen as an appropriate symbol of this humility. It is not pride—pride is what they abhor the most—but it is, if you like, pride almost turned on its head. The plainer and the more basic the better. Verhoodelt Englisch would be much the same. Its low-status, like the low status of PG, could be seen as having a positive, almost sacred, value for the Old Orders.

3. Interference from E in PG What is remarkable when we consider interference from the other direction is that it is possible to take the earlier hierarchy and simply reverse it.6 This is shown in Figure 3. Lexical Grammatical Phonological Prosodic

strong quite strong weak very weak

Figure 3. Interference E → PG.

throw the baby from the window a cookie

79

3.1 Lexical The number of lexical borrowings into PG is enormous and this is commented on frequently by the Mennonites themselves. Typically they are needborrowings for new objects and concepts. This is hardly surprising—after all PG has been in North America for well over three centuries and cannot meet all current communicative needs. As continental German no longer represents a viable source of borrowing, there is no alternative but to borrow from E. Not all borrowings, however, could be said to arise out of need; PG has many E ﬁllers like anyway, in fact and well. There are also examples where an E word has replaced an original German word, as in Fens for original Zaun. Prestige is often given as the reason for unnecessary loanwords (cf. Weinreich 1953: 59–60), but clearly prestige is not a viable motivating force within the Amish-Mennonite context. The motivation behind these kinds of lexical borrowings into PG would make an interesting study. 3.2 Grammatical Grammatical interference also appears strong and I discuss some of the most interesting examples below. The ﬁrst feature involves word order. All Germanic languages have had at some stage in their development a feature of verb placement known as the verbal brace (Satzrahmen). In main clauses, this brace construction occurs only in complex tenses. It involves both the ﬁnite verb in second position and all non-ﬁnite verbal elements in ﬁnal position. These together form an imaginary kind of bracket or brace around all other sentence constituents. In subordinate clauses, it involves the initial subordinating conjunction and all clause ﬁnal verbal elements. These also form a kind of bracket around the other clausal constituents. What is happening increasingly in PG is violation of this brace construction; that is, elements are appearing more and more outside the verbal brace, bringing the discontinuous verbal constituents closer together. This is illustrated in Figure 4 (the verbal brace has been highlighted). This change is bringing PG syntax closer to the E, although it is impossible in this instance to tell whether E inﬂuence is directly responsible. Languages in the Germanic family are in the process of stabilizing SVO order and parallel word order changes have either already happened or are currently underway in all Germanic Main Clause Sub. Clause

S V1 X V2 Conj S X V

→

S V1 X V2 Y / S V1 V2 X) → Conj S X V Y / Conj S V X)

Figure 4. Changes in the verbal brace.

80

kate burridge

languages, including European Standard German. It is quite probable, therefore, that E is in this case simply accelerating a word order already immanent in PG. What is striking, however, is the nature of the material appearing after the ﬁnal verb. For one, all types of adverbials (including light elements) commonly appear outside the brace. In continental German dialects (and also Dutch), it is still unusual for anything other than ‘‘heavy’’ elements (or genuine afterthought material) to appear in this position. Examples (3) and (4) below suggest that contact with E is a possible accelerating inﬂuence here (although of course there will always be the chance of parallel but independent development). (3) Mir hen alsemols heemgemacht eiscream g’hat do we have before homemade icecream had here ‘We used to have homemade icecream here.’ (4) Er hot besser g’ﬁelt wann er sich gut hewe hot kenne aeryets he has better felt if he self good hold has can somewhere ‘He used to feel better, if he could get a good grip somewhere’ In addition, a number of subordinating conjunctions like weil ‘because’ and as ‘that’ occur with main clause word order. Again the exact role of English is diﬃcult to determine here, since the same phenomenon can also be found in continental dialects and is most certainly triggered by discourse factors like the assertiveness of the clause (cf. Burridge 1993 on main clause phenomena in medieval Dutch and German subordinate clauses). (5) No verblatz ich weil ich bin ganz arig vol then burst I because I am totally very full ‘Then I’ll burst, because I’m completely stuﬀed.’ The second feature involves new aspectual marking. PG has a progressive aspect involving some form of the verb sei ‘to be’ followed by am (a fused preposition and article; literally, an + em ‘on the’) and an inﬁnitival substantive. For example (the verb bischt is clause-ﬁnal in (6) because of the subordinator wann): (6) Wann du am singe bischt fer mich no daerfscht graad schtobbe if you on-the sing are for me then may-you at once stop ‘If you are singing for my beneﬁt, then you can stop right now!’ Although E inﬂuence is undoubtedly involved here, this construction is unlikely to be an actual E borrowing. For one, the construction also appears in other Germanic dialects, so the predisposition has always been present (compare modern German Er ist am Essen ‘he is eating’; cf. Lockwood 1968: 161–62). It also appears in Dutch (Hij is aan het eten). What does suggest E inﬂuence, however, is the fre-

throw the baby from the window a cookie

81

quency of the construction (cf. also Enninger 1980: 346–7 who argues this). In a very early study on PG aspect carried out in 1941, Reed states ‘‘the proportion of progressive forms used is extremely small’’ (Reed 1947: 9). This is no longer the case. The original construction has increased considerably in frequency and has expanded into a number of new contexts, including a new progressive passive construction (cf. Burridge 1992: 221–4). True, this increase in discourse frequency could be attributed to the on-going grammaticalization process—English may simply be helping the process along. But it is interesting to note in this regard, that the Canadian PG construction is following the E in showing signs of shifting from a progressive to a present tense. The third feature involves case. The original nominative and accusative cases have already merged in PG. This must have happened very early in its history, since it is a general feature of south eastern Rhine-Palatinate dialects (cf. Buﬃngton and Barba 1965; Keller 1961). There are now signs that the dative will also disappear, although ﬁndings here show that this change is not nearly as advanced in Canada as it is in Pennsylvania (cf. Huﬃnes 1989b). Some dialects of European German have also lost the dative distinction (cf. Keller 1961), but it cannot be an inherited pattern in PG, since the dative is still used freely among the Canadian OOMs. It is largely the transitional groups who are showing the signs of further case merging —or convergence to E. The same case syncretism appears in other varieties of German in contact with E (cf. Eikel 1949 and Gilbert 1965 for American German and Clyne 1972, 1989 for Australian German and also Dutch) and structural pressure from E is undoubtedly an issue. However, since this sort of morphological leveling or ‘‘grammatical stripping’’ (Markey 1987) is a general characteristic of languages in contact, once again the actual role played by E is diﬃcult to determine. The next feature is the passive. The following passive constructions are available for the diﬀerent tenses in PG. I should mention, however, that passive is extremely rare in my recorded material. (7) Present/Preterite: De Schtrump s/waar geschtoppt bei der Maem the stocking is/was darned by the mother (8) Present/Past Perfect: De Schtrump is/waar geschtoppt waerre bei der Maem the stocking has/had darned been by the mother (9) Progressive: Blaume sin am verkaaft waerre beim Kael am Eck plums are on-the sold being by-the chap at-the corner

82

kate burridge

There are a number of changes which have occurred in the PG passive and which suggest interference from E. First, the passive bei-phrase always appears outside the brace; that is, after the passive participle (cf. discussion above). The next two changes are really lexical borrowings. One, the preposition bei has now replaced original vun. Two, in present and preterite passives the auxiliary sei ‘to be’ has replaced original warre/waerre ‘to become’ (compare (7) with sentences (8) and (9)). Since the verb ‘‘to be’’ is one of the few verbs to maintain a preterite form (the preterite has otherwise disappeared from PG), this means that a preterite passive is now available. With the original auxiliary waerre, the only tenses possible were the present and perfect. The following examples show the passive before the changes. (10) De Schtrump wat vun der Maem geschtoppt (compare 7 above) (11) De Schtrump is/waar vun der Maem geschtoppt waerre (compare 8 above) With the verb ‘‘to be’’ as the new passive auxiliary, this produces the same ambiguity in PG as exists in E (the stocking is/was darned, for example, has both a passive reading and a stative reading, as does 7 above). The new progressive passive shown in (9) (undoubtedly inﬂuenced by E) overcomes this potential ambiguity. The ﬁnal feature involves a new inﬁnitival construction involving prepositions fer ‘for’. (12) Er hot sich immer g’faericht fer in de heh geh fer ebbes schaffe he has self always feared for in the high go for something do ‘He was always frightened to go up high to do something.’ (13) Es muss schee sei fer nix zu duh hawwe it must wonderful be for nothing to do have ‘It must be wonderful to have nothing to do.’ Although these clauses resemble E ‘for-to’ clauses, they are much more widespread. Fer ‘for’ has totally replaced original zu ‘to’ as the marker of an inﬁnitival complement. It may well be that it represents a direct borrowing from E (PG fer already showed semantic overlap with E for). But there are at least two other possibilities here. One, there is evidence of a similar construction in Continental Frankish (although it is conﬁned to purpose clauses; cf. Lockwood 1968: 154 who suggests that it is a likely calque from the French pour inﬁnitival construction). This particular dialect was one of the original input dialects to PG. The construction therefore may not have originated from English, although its spread does suggest at least inﬂuence in this regard. Another possibility is that this is a case of parallel but independent development. For example, given the modern Standard German construction Es wäre besser für

throw the baby from the window a cookie

83

mich, sogleich zu gehen (literally,‘‘It would be better for me, immediately to go), it is easy to imagine a change (as originally happened in E), whereby the close logical relationship existing between mich, the object of preposition, and gehen, the inﬁnitival verb, triggered the reanalysis of für as a complementizer heading a new subordinate clause. In short, the development of complementizer fer is a paradigm example of grammaticalization and the chance of parallel but independent development is very strong. (Note, the appearance of the Canadian inﬁnitival complementizer fer does not correlate with the subcategorization properties of the equivalent E verb, as Louden 1997 argues for Plain PG in Pennsylvania; this would appear to be an interesting diﬀerence between Canadian and Pennsylvanian PG.) 3.3 Phonological considerations Early borrowings did not have a great eﬀect on the phonology of PG. As discussed in the introduction, the speakers were not bilingual PG-E and borrowings simply assimilated to PG phonological patterns. But with the large numbers of borrowings entering PG today, this is now having some impact on the phonological structure of the language. For example, pie represents an early loanword. Although some older speakers still pronounce it [boi], younger speakers are now pronouncing it according to E structure [phei]. Borrowings like this, and others like teacher and car, are introducing initial aspirated stops. But one of the most interesting examples of interference involves the rhotic. This is because of its unusual distributional pattern. Whereas so far we have seen more intense contact interference between E and PG (in both directions) in the less conservative groups of Mennonites, the rhotic involves principally the most conservative group of speakers; namely, the OOMs. Early descriptions of Ontario PG describe it as essentially r-less (cf. for example, Kratz and Milnes 1953). The sound is pronounced (as a trill) when it appears before a vowel, and elsewhere as a schwa. But this is changing rapidly, particularly among the OOMs. Before consonants and word-ﬁnally after long vowels, it is becoming increasingly common to ﬁnd an r which closely resembles the North American retroﬂex; for example, PG Darm ‘intestine’ is now frequently pronounced [dærm]. And for many individuals there is evidence that the retroﬂex is replacing the trill in other positions. This is particularly true of the younger and middle age OOMs.7 If borrowings like car were solely responsible for introducing this pattern, this would not account for why the OOMs are ahead of the more progressive groups in this development. This pronunciation is certainly due to E inﬂuence, but in this case it is via the roundabout route of written High German (HG). HG is of course r-less, but when the OOMs read it or sing it, they pronounce r wherever it is written—and most

84

kate burridge

importantly, they pronounce it as a retroﬂex, not as a trill. In other words, since German is never a written language for them, the OOMs read HG as if it were E. In the Old Order parochial schools, this is the pronunciation now taught in instruction classes on HG. These special facts would account for why, contrary to all other aspects of interference, the OOMs show greater interference here than the NonPlain, who now only rarely encounter HG. The Non-Plain church service is predominantly E.

4. Why is this so—discussion of ﬁndings Why is it that we ﬁnd precisely these patterns of interference? Genetic or typological considerations are not much help here. Although they do predict that the similarity of the linguistic systems would presumably facilitate easy interference, they can not explain why interference seems to be sensitive to the diﬀerent levels of grammar and why, in particular, the reverse hierarchies obtain. To account for this we must look elsewhere. 4.1 The acquisition experience Consider ﬁrstly, the diﬀerent types of contact involved. PG is the language ﬁrst learned at home and E is generally acquired at school (from the ages of 5 or 6). The direction PG to E, therefore, involves a type of interference traditionally known as substratum interference—in other words, the transferral of language habits into a second language (so-called ‘‘improper learning’’). Such a situation involves potential or actual shift to E (as is currently happening among the transitionals and, to an even greater extent among the progressives). The other direction, however, E to PG, represents actual borrowing—the transfer of items to a group’s native language (this is a more restricted sense of the term borrowing than is usual; cf. Thomason 1986 and Thomason and Kaufman 1988). The patterns of interference we ﬁnd in the earlier hierarchies in Figures 1 and 3 are predictable when we take into account, ﬁrstly, these two diﬀerent types of contacts and, secondly, the order in which each diﬀerent linguistic level is acquired. Prosody (intonation and stress features) are the ﬁrst to be acquired and are therefore ‘‘the most deeply anchored’’ features (McLaughlin 1984). These and segmental features also are acquired during a child’s ﬁrst three years. The acquisition of grammatical features continues until much later, during the early school years, and vocabulary continues to be acquired throughout one’s life time. It is not surprising, therefore, that the most ‘‘deeply anchored’’ speech habits are the ones we ﬁnd most readily transferred from PG to E (from ﬁrst to second language) and—to consider

throw the baby from the window a cookie

85

the other side of the coin—are the ones which in PG show the greatest resistance to interference from E. Since vocabulary will keep on being acquired as the need arises, the enormous quantity of loans now entering PG from E is also no surprise. As discussed earlier, the majority of these are necessary borrowings. In short, social setting and the chronology of language learning dictate that in the ﬁrst instance we have substratum-induced interference, where the intensity of interference is greatest at the suprasegmental and segmental levels. And in the second instance, we have a case of borrowing, which typically begins with vocabulary. However, language learning does not predict one way or another for grammar. In fact, it suggests that grammatical interference could well be associated with either one of the types of contact. Research ﬁndings also support this. It appears that both borrowing and substratum-induced interference can result in grammatical changes (cf. Rayﬁeld 1970; Thomason 1986). Why is it then that grammatical interference is so much stronger from E to PG than in the other direction? The explanation for this lies in other aspects of the childhood acquisition experience. During the early school years, children would not yet have stabilized their PG grammatical structures. For that reason, structures like the passive and the inﬁnitival construction, for example, are particularly vulnerable to E interference at this time (cf. also Costello 1978, 1989 who describes syntactic innovations in the PG of Pennsylvanian children). Why the E structures in particular have such a strong inﬂuence, rather than the other way around, follows from the way in which it is acquired in the formal school setting. In Canada, school is an ‘‘E only environment’’ which admits no PG whatsoever (even during recess in the school yard!). There is also considerable prescriptivism attached to the learning of E—a strong emphasis on ‘‘proper English’’ (compare Enninger et al.’s 1984: 13 description of Amish parochial schools in Pennsylvania, where this is also the case). Similarly, outside of school, the children’s experience of E is within formal contexts only; i.e. in contact with outsiders and as a written language. In addition, they have no access to more colloquial varieties via things like television and radio, for example. As you might predict from this, there is strikingly little in the way of style-shifting in Pennsylvania German E. Speakers conﬁne themselves to a fairly formal register. All these factors work against the transference of PG features into their E. There would be very little in the way of ‘‘imperfect learning’’. This would not be true of E-PG interference, however. Socio-historical studies of change have shown that linguistic innovations have a much greater chance of taking hold in a language if they attract no attention and no resistance from speakers (cf. for example, Nadkarni 1975: 681 and Louden 1989, particularly pp. 34–5). While there would be considerable resistance to change in E because of the way in which it is acquired, because PG is not standardized and has no written form, people are quite used to variation and are tolerant of it. PG speakers constantly remark on the variation in their lan-

86

kate burridge

guage, often commenting: wie Englisch as mir sin ‘how English we are’. Despite their isolationist philosophy, these comments are never regretful, nor are they ever critical—even by the OOMs. The two types of language contact situations (substratum and borrowing) and the acquisition experiences that go hand in hand can account for how the structures are ﬁrst able to enter the language and take hold, but there is also the additional question of variation within the Mennonite community itself. Why do the less conservative TMs (particularly, the younger members) show more interference than the more conservative OOMs? This is surprising when you consider that the transitional groups use E more often and have much more contact with the E-speaking outside world. The answer lies in the diﬀerent linguistic inputs these groups receive as children. For one, the early acquisition experience of the TMs involves considerably more mixing of the two languages and for many E may well even come to dominate both linguistically and socially. Bilingual studies like McLaughlin (1984) show that both mixed input and imbalance more readily facilitate interference. The OOM children, however, keep the languages separated from start. They acquire the languages successively. PG is well established at home before E is later introduced in the school. No language appears to dominate over the other. Each have their own distinct functions. Also important here is the fact that the environment in which the OOM children learn E is a supportive one—something which may seem surprising in a community so intent on maintaining its separate and peculiar status.8 All of these factors—non-mixed input, absence of dominance and a supportive environment—have been shown to be important if linguistic structures are to remain diﬀerentiated and resist interference. 4.2 Diglossia—rigid compartmentalism Another important consideration has to do with the aspect of diglossia. I have said the OOM situation is one of stable bilingualism supported by diglossia. The rigid compartmentalism characteristic of diglossia, where languages are used in quite distinct and separate domains, appears to exclude this sort of code-mixing. This is reﬂected in the weak interference and also the little, if any, actual code-switching that takes place. The only examples of code-switching I have witnessed among the OOMs occur during translation tasks—an artiﬁcial situation for them, involving both languages. Of course, it is also highly unusual for them to have someone from the outside speaking PG. My presence necessarily upsets the usual strict compartmentalism. By comparison, the bilingualism of the Non-Plain Folk does not have the support of diglossia. For them, PG is showing ever shrinking domains, with English appearing more and more in contexts of usage which were traditionally PG (church ser-

throw the baby from the window a cookie

87

vices, for example). The breakdown of diglossia typically results in code-switching and general interference which is exactly what we are ﬁnding. Observations by Fishman and others support exactly this: As role compartmentalization and value complementarity decrease under the impact of foreign models and massive change, the linguistic repertoire also becomes less compartmentalized. Language varieties formerly kept apart come to inﬂuence each other phonetically, lexically, semantically, and even grammatically, much more than before. (Fishman 1972: 105)

4.3 Ethnic afﬁliation The nature of interference is also dependent on additional social and psychological factors. The TMs are in the process of shifting culturally and linguistically to mainstream Canada, but they are not yet part of the mainstream. These people keep close and continual contact with the OOMs and identify very strongly with them—the Modern Plain is really a more apt label for this group. However, while they share many of the same beliefs and behaviour patterns as the OOMs, they have lost the distinctiveness of this group. They no longer have the horse and buggy, nor the extraordinary dress which go to make the OOMs such a strikingly distinctive group. And now there is every sign that they are also losing PG. For these people then, PG features in their E have a value in signalling their PG ethnicity. As language attitude studies show, ethnic varieties of the dominant language can be ‘‘powerful markers of ethnic group belongingness’’ (cf. Giles 1977). But for the OOMs who still maintain ‘‘hard linguistic and non-linguistic boundaries’’ (to use Giles’ terminology), PG English has no role in ethnic diﬀerentiation and they place no value in it. (From Huﬃnes 1980b, 1984 description of the Pennsylvania Amish, it seems a similar situation holds there.) For those groups, where these boundaries are softening, however, PG features are becoming important ethnic markers. ‘‘The softer the perceived linguistic and non-linguistic boundaries existing between ethnic groups, the more likely speech markers will be adopted in order to accentuate ethnic categorization’’ (Giles 1977: 274). 4.4 The progressive Mennonites To conclude, let me say something about the modern group, the PMs, who have not yet featured in the discussion. As implied in the introduction, this group identiﬁes most strongly with the values of the dominant culture and are the most advanced in the shift to E. For them, social and economic advancement is important and can be via E only. They place no value in (and in fact go out of their way to avoid) any PG

88

kate burridge

markers in their E. With the exception of intonation—‘‘the most insidious form of interference’’ (Baetens-Beardsmore 1982: 62)—they show little in the way of interference. Given that PG is gradually dying within this group, the lack of interference is predictable. Studies have shown that in this sort of contact (involving inﬂuence from substratum in a dominant language), interference patterns only remain as long as the group itself remains bilingual. Rayﬁeld (1970: 106), for example, shows that the grand children, even the children, of Jewish immigrants in the USA speak E without any interference from Yiddish. Vogt (as discussed in Rayﬁeld p.106) predicts that interference in French from Breton will disappear when Breton itself disappears. Often it is only intonation features which survive as evidence that what is now a monolingual community has developed from precisely this type of bilingual contact situation. There are some signs, however, that this may change. It is my impression that in Waterloo County it is suddenly becoming fashionable to have PG heritage. If this does indicate the beginning of an ‘‘ethnic revival’’ (Fishman et al. 1985), we might actually see these features stabilize and ﬁnd their way into the E of surrounding areas. Up until now, this has deﬁnitely not been the case. For the progressives, PG suﬀers from all the negative associations of a dialect without a standard and without a written form (cf. Burridge 1997) and the fear of acquiring a ‘‘PG accent’’ has been a strong accelerating force behind the shift to E (cf. the problem of negative stereotyping mentioned earlier). The PG of the progressive group shows enormous variation, as one would expect of a language death situation (symptomatic of its lack of vitality; cf. Dorian 1981). Competence in PG ranges from the fully-ﬂedged bilingual to the real semi-speaker. As one would predict, the acquisition experiences are very varied for the members of this group and it is diﬃcult to generalize. Many acquire it passively. Parents, for example, report speaking PG with their own parents and among themselves but not with their own children (who also show no interest in learning the language). Of course, here too parameters involving socio-economic status, education, age, and sex are going to be crucial. Given these diﬀerent parameters, members within the progressive group will show to quite varying degrees the eﬀects of contact with E.

5. Conclusions There have been many linguistic constraints proposed on the diﬀerent ways in which languages can inﬂuence each other. Harris and Campbell (1995) for example, outline a number of the general claims and ‘‘universals’’ which have been put forward to explain linguistic interference, especially grammatical. But while these do contribute greatly to our understanding of the role of contact in language change,

throw the baby from the window a cookie

89

there is always a battery of counter-examples where the proposed universals simply fail. This is precisely because there are crucial social and psychological factors, which the constraints do not take account of. Without precise information about these factors, we cannot begin to predict when interference will take place and what the nature of it will be. Phonological, prosodic, or grammatical features can never be considered independently of the context in which they appear. For example, it is simply not possible to say that suprasegmental interference is the most pervasive, or that syntax is the most resistant to inﬂuence, or even that inﬂuence always begins with the lexicon. This is not to say that linguistic factors are irrelevant of course. For example, language internal considerations, like therapeutic borrowing, naturalness and markedness conditions all have a role to play. Closeness of linguistic systems and structural compatibility are also important—and yet, there are cases of interference between even the most divergent of systems. Why? Because there will always be social dynamics involved which play a crucial role in determining what kinds of linguistic features are involved, the direction and also extent of the interference. This is by no means an original observation of course. As Thomason and Kaufman (1988: 85) neatly put it, ‘‘it is the sociolinguistic history of the speakers, and not the structure of their language, that is the primary determinant of the linguistic outcome of language contact.’’ Social considerations, not linguistic ones, are what form the basis of their theory of contact-induced change. (See also Thomason 1986.) What this paper demonstrates is just how complex these social factors can be. Data here, even though preliminary, show quite clearly the importance of fully understanding the social setting, in order to be able to account for and predict changes occurring in a contact situation. To begin with it is the social setting which identiﬁes that, on the one hand, we have a case of substratum interference (PG-E) and, on the other, a case of borrowing (E-PG). Both situations, together with very diﬀerent acquisition experiences, have given rise to quite distinct interference results in the two languages. In addition, diﬀerent groups within the PG community show diverging degrees of interference. For example, one group’s bilingualism is supported by diglossia (the OOMs) and is a maintenance situation. The other group’s bilingualism is without diglossia (the TMs) and involves potential shift. This latter group shows the most intense interference. That is, in all matters except the Canadian rhotic. This is a surprise feature in the PG of the Old Orders, until their uniquely triglossic situation is factored in. E inﬂuence here is coming indirectly via HG. Finally, the various groups show quite diﬀerent attitudes towards their cultural and linguistic identity and this also has a profound eﬀect on the patterns of interference they show. In particular, the Transitionals are seeking to establish a more distinctive PG identity and they are doing this via their E. In sum, the

90

kate burridge

outcome of contact depends on all these diﬀerent factors, and not surprisingly, the various members within the PG-speaking community are showing the results of very diﬀerent types of changes. As has been emphasized throughout, this study is still in its initial stages. The data has not yet been quantiﬁed and the current analysis concentrates on the parameter of religious conservatism. Important aspects of the social setting are yet to be addressed. Within the close-knit and ultra-conservative OOM community, these are probably less crucial—patterns of interference seem to show up uniformly throughout and parameters like the speaker’s age and sex do not seem to be signiﬁcant for the broader picture (although obviously this is something which will be checked more thoroughly). But as the discussion has already hinted, this is not the case for the transitional and progressive groups. As one would expect of speakers in the midst of a shift (and in some cases a very rapid shift), factors like socio-economic status, age and education, are going to be crucial. In short, much remains to be done. Nonetheless, the results so far are encouraging and indicate worthwhile directions for future research.

Notes 1. The ﬁeld trip to Waterloo County, during which I collected the initial data for this paper, was made possible through the generous support of the Canadian Faculties Enrichment Award. It was additionally supported by an ARC grant jointly held by members of the Linguistics Department, La Trobe University. A much earlier draft of this paper appear in the La Trobe Working Papers 2. As always I am very grateful to the Mennonite community for their wonderful hospitality and their extraordinary patience in teaching me about their language. 2. Some accommodations are now being made; for example, in the matter of telephones and refrigeration (see discussion Burridge 1997 and also Hostetler 1980 for an account of the Amish situation in Pennsylvania). 3. If you include High German (HG) in this scenario, then the OOMs could well be described as triglossic. They are not trilingual, however. The OOMs cannot speak HG unless it is to quote from the Bible. In fact, it seems the younger members particularly can follow very little of a Bible reading in HG. In function, the language is clearly very restricted and in this model would best be described as the classical variety. 4. Studies on groups from Pennsylvania, like Enninger et al. (1984) and Huﬃnes (1980b, 1984) suggest the same—Pennsylvanian verhoodelt Englisch is also a modern-day ﬁction. 5. The results here show some similarity with both Enninger’s and Raith’s study (also Huﬃnes 1980b and 1984), although it is diﬃcult to compare ﬁndings, because of diﬀerent methodologies. Nonetheless, all show the same interesting trend; namely, it is the least conservative groups who show the most interference.

throw the baby from the window a cookie

91

6. This is basically the same pattern which Rayﬁeld (1970) records for a study of English-Yiddish contact. For Rayﬁeld’s ﬁndings also, similar hierarchies obtain. 7. There is enormous variety between speakers, as is typical of change in progress (e.g. the former OOM Buehler writes ﬂuctuating spellings like a’ahlich versus airlich ‘honest’; a’ahbs versus airbs ‘pea’ (1977). I have observed that the r-ful pattern is strongest among the Canadian Amish. In this case, observation is also backed by anecdotal evidence. The Mennonites often draw my attention to the rful nature of Amish PG and, when imitating Amish speech, they always pronounce postvocalic r’s, even where there are none! 8. It is a common misconception that the Amish and Mennonites are totally against any education. This is simply not true—the parents of Old Order children are very supportive of the training given at school. They realize the importance of learning English. After all, successful business with the outside world requires ﬂuency in English, and these people have a keen sense of business.

References Baetens-Beardsmore, Hugo. 1982. Bilingualism: Basic Principles. Avon: Tieto Ltd. Buehler, A. 1977. The Pennsylvania German Dialect and the Life of an Old Order Mennonite. Publication of the Pennsylvania Folklore Society of Ontario. Buﬃngton, A.F. and Barba, P.A. 1965. A Pennsylvania German Grammar. Allentown, Penn.: Schlechter. Burridge, Kate. 1992. ‘‘Creating Grammar: Examples form Pennsylvania German, Ontario.’’ In K. Burridge and W. Enninger (eds), Diachronic Studies on the Languages of the Anabaptists. Bochum: Universitätsverlag Dr. N. Brockmeyer, 199–241. ——. 1993. Syntactic Change in Germanic: With Particular Reference to Dutch. Amsterdam: John Benjamins. ——. 1997. ‘‘ ‘Separate and Peculiar’—the Survival of Pennsylvania ‘Dutch’ in Ontario, Canada’’. In K. Burridge, L. Foster and G. Turcotte (eds), Canada—Australia: Towards a Centenary of Partnership, Ottawa: Carlton Press, 247–66. Chen, P. 1986. ‘‘Discourse and Particle Movement in English’’, Studies in Language 10–1: 79–95. Clyne, Michael. 1972. Perspectives on Language Contact (based on a study of German in Australia). Melbourne: The Hawthorne Press. ——. 1989. German-English and Dutch-English Language Contact in Australia: Some Theoretical Implications (paper presented at La Trobe University). Costello, J.R. 1978. ‘‘Syntactic Change and Second Language Acquisition: The Case for Pennsylvania German.’’ Linguistics 213: 29–50. ——. 1989. ‘‘Innovations Increasing Syntactic Complexity in the Native Language of Bilingual Children from 5 to 10: The Case for Pennsylvania German.’’ In W. Enninger, J.A Hostetler, J. Raith, and K. Wandt (eds), Studies on the languages and the verbal behaviour of the Pennsylvania German II. Stuttgart: Steiner Verlag, 3–16.

92

kate burridge

Dorian, Nancy C. 1981. Language Death: The life cycle of a Scottish Gaelic dialect. Philadelphia: U. of Pennsylvania Press. Druckenbrod, R. 1981. Mir Lanne Deitsch. Allentown, PA. Eikel, F. 1949. ‘‘The Use of Cases in New Braunfels (Texas) German’’. American Speech 24: 278–81. Enninger, W. 1980. ‘‘Syntactic Convergence in a Stable Triglossia plus Trilingualism Situation in Kent County, Delaware, U.S.A’’. In H. Nelde (ed.), Sprachkontakt und Sprachkonﬂikt (Zeitschrift für Dialektologie und Linguistik, Beiheft 32. Wiesbaden: Steiner Verlag, 343–50. Enninger, W. et al. 1984. ‘‘The English of the Old Order Amish of Delaware: Phonological, Morpho-syntactical and Lexical variation of English in the Language Contact Situation of a Trilingual Speech Community’’. English World-Wide 5.1: 1–24. Frey, J.W. 1945. ‘‘Amish Triple Talk’’. American Speech 20: 85–98. Fishman, J.A. 1972. The Sociology of Language. Rowley, Mass.: Newbury House. ——. 1982. ‘‘Bilingualism and Biculturalism as Individual and as Societal Phenomena’’. In J.A. Fishman and G.D. Keller (eds), Bilingual Education for Hispanic Students in the United States. New York: Teachers College Press, 23–36. ——, M. Gertner, E. Lowy, and W.G Milan (eds). 1985. The Rise and Fall of the Ethnic Revival: Perspectives on Language and Ethnicity. Berlin: Mouton Publishers. Gilbert, G.G. 1965. ‘‘Dative Versus Accusative in the German dialects of Central Texas’’. Zeitschrift für Mundartforschung 32: 288–96. Giles, H. 1977. ‘‘Ethnicity Markers in Speech’’. In H. Giles (ed.), Language, Ethnicity and Ingroup Relations. London: Academic Press. Harris, A.C and L. Campbell 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Horne, A.R. 1910. Pennsylvania German Manual (3rd ed.). Allentown, Penn.: T.K. Horne Publishers. Hostetler, J.A. 1980. Amish Society (3rd ed.). Baltimore: The John Hopkins Uni Press. Huﬃnes, M.L. 1980a. ‘‘Pennsylvania German: Maintenance and Shift.’’ International Journal for the Sociology of Language 25: 43–57. ——. 1980b. ‘‘English in Contact with Pennsylvania German.’’ The German Quarterly 54: 352–66. ——. 1984. ‘‘The English of the Pennsylvania Germans: A Reﬂection of Ethnic aﬃliation.’’ The German Quarterly 57: 173–82 . ——. 1986. ‘‘Intonation in Language Contact: Pennsylvania German and English.’’ In W. Enninger (ed.), Studies on the Languages and the Verbal Behaviour of the Pennsylvania Germans 1. Stuttgart: Steiner Verlag, 25–36. ——. 1989a. ‘‘Convergence and Language Death: The Case of Pennsylvania German.’’ In Enninger et al. (eds), 17–28. ——. 1989b. ‘‘Case Usage among the Pennsylvania German Sectarians and Nonsectarians.’’ In N.C. Dorian (ed.), Investigating Obsolescence: Studies in Language Contraction and Death. Cambridge: Cambridge University Press, 211–66. Keller, R.E. 1961. German Dialects . London: Butler and Tannen.

throw the baby from the window a cookie

93

Kratz, H. and Milnes, H. 1953. ‘‘Kitchener German: A Pennsylvania German Dialect.’’ Modern Languages Quarterly 14: 184–98; 274–83. Lockwood, W.B. 1968. Historical German Syntax. Oxford: Clarendon Press. Louden, M.L. 1987. ‘‘Bilingualism and Diglossia: The Case of Pennsylvania German Dialect.’’ Leuvense Bijdragen 76: 77–36. ——. 1989. ‘‘Syntactic Variation and Change in Pennsylvania German.’’ In W. Enninger, et al. (eds), 29–40. ——. 1997. ‘‘Linguistic Structure and Sociolinguistic Identity in Pennsylvania German Society.’’ In J. R. Dow & M. Wolﬀ (eds), Languages and lives: Essays in honor of Werner Enninger. New York: Peter Lang, 79–92. Markey, T.L. 1987. Mistress of Many: How Dutch was Grammatically Stripped. Third Biennial Interdisciplinary Conference on Netherlandic Studies (1986). University of Michigan, Ann Arbor. McLaughlin, B. 1984. Second-Language Acquisition in Childhood: Preschool Children. New Jersey: Lawrence Erlbaum Associates. McMahon, A. 1994. Understanding Language Change. Cambridge: Cambridge University Press. Moelleken, W.W. 1983. ‘‘Language Maintenance and Language Shift in Pennsylvania German: A Comparative Investigation.’’ Monatshefte 75(2): 172–86. Nadkarni, M.V. 1975. ‘‘Bilingualism and Syntactic Change in Konkani.’’ Language 51: 172–86 Raith, J. 1977. ‘‘Pennsylvania German-American English Bilingualism: A Case Study.’’ In C. Malony et al. (eds), Deutsch in Kontakt mit anderen Sprachen, 104–28. Kronberg: Scriptor Verlag. Raith, J. 1981. ‘‘Phonologische Interferenzen im Amerikanischen Englisch der Anabaptistischen Gruppen deutscher Herkunft in Lancaster County.’’ Pennsylvania Zeitschrift für Dialektologie und Linguistik, 48–1: 35–52. Rayﬁeld, J.R. 1970. The Language of a Bilingual Community. The Hague: Mouton. Reed, Carroll E. 1947. ‘‘The Question of Aspect in Pennsylvania German.’’ Germanic Review. 20: 5–12 Thomason, S.G. 1986. ‘‘Contact-induced Language Change: Possibilities and Probabilities.’’ In W. Enninger and T. Stolz (eds), Akten des 2 Essener Kolloquiums zu Kreolsprachen und Sprachkontakte. Bochum: Studienverlag Dr. N. Brockmeyer, 261–84. Thomason, S.G. & T. Kaufman 1988. Language Contact, Creolization, and Genetic Linguistics. Berkeley: Uni of California Press. Weinreich, U. 1953. Languages in Contact. New York: Linguistic Circle of New York Publication No. 2.

The great Daghestanian case hoax BERNARD COMRIE and MARIA POLINSKY University of Southern California and University of California

1. Introduction1 Case marking systems have played a prominent role in Barry Blake’s linguistic work, from his early description of Kalkatungu (Blake 1968) and his overview of case systems in Australian languages (Blake 1977) to his comprehensive monograph on case as a linguistic category (Blake 1994). With that in mind, we are happy to expand his unmatched collection of data on case systems by the following material from an area particularly rich in case forms. The Daghestanian languages are well-known in the linguistic literature for their rich case systems. Indeed, this richness has passed beyond technical literature in linguistics. In The Guinness Book of Records for 1997 (Young 1997: 249), in a subsection of the general section on Language, we ﬁnd the following: Most complex language—Tabassaran, a language of Daghestan . . . uses the most noun cases, 48.

In the scientiﬁc literature, perhaps the most famous reference to the Tabasaran case system is Hjelmslev (1935: 138, 139), where Tabasaran is presented as having the ‘‘empirical maximum’’ number of cases. Hjelmslev analyzes Tabasaran as having 52 cases, but two of these occur only on adjectives, so the number of noun cases is 50. We do not dispute that Daghestanian languages have rich case systems. What we do ﬁnd questionable, however, is the high counts of the number of ‘‘cases’’ that are presented in both scientiﬁc and popular accounts. In order to demonstrate that the number of case morphemes is far lower than would be arrived at by the methodology implied by the above statement from The Guinness Book of Records, we have

96

bernard comrie and maria polinsky

chosen two Daghestanian languages as illustrations: Tabasaran, for the simple reason that is cited in the quotation given above, and Tsez (also known as Dido), the latter not only because it is the subject of our own ongoing research (see note 1) but also because, using the same methodology as is used to give about the 48 cases for Tabasaran, Tsez would end up having far more. As background information, we would note that the Daghestanian languages form one branch of the Nakh-Daghestanian or Northeast Caucasian language family, the other branch being the Nakh languages. The Daghestanian languages are spoken in Daghestan (Daghestan Republic), one of the republics in the Russian Federation, especially in the north and west of that republic, and in neighboring parts of Azerbaijan.2 Although the internal classiﬁcation of Nakh-Daghestanian is not without problems (especially in the light of new hypotheses advanced in Nikolayev and Starostin 1994), for the purposes of this paper it will suﬃce to note that Tabasaran and Tsez belong to diﬀerent subgroups of Daghestanian, namely Lezgic and Tsezic respectively. They are thus not particularly closely related to one another genetically within Daghestanian, nor are they in areal contact, so that their common property of a rich case system can be taken as symptomatic for Daghestanian as a whole. In presenting the data on the case systems of Tabasaran and Tsez below, we make a distinction between non-local (grammatical, abstract) and local cases, since the main source of morphological richness in Daghestanian case systems is provided by the local cases. As we shall see, there are, however, some cases that are on the border-line between local and non-local, so that the distinction should be considered, at least for present purposes, one of convenience rather than principle. Our main source for Tabasaran is Magometov (1965: 97–141), which marshals data from the various dialects, supplemented by Xanmagomedov (1967: 548–550), which deals primarily with the standard language. Our main source for Tsez is our own ongoing work on the language, of which a summary is provided in Comrie et al. (forthcoming); all claims relate to the Ceboru subdialect of the Asaq dialect of Tsez, although as far as we are aware these claims are of general applicability to the Tsez language (though precise morphological and phonological forms may vary from dialect to dialect). For earlier descriptions of Tsez (other dialects), see Bokarev (1959) and Imnajšvili (1963).

2. The case system of Tabasaran Given one way of counting cases, the total of 48 noun cases given in the quotation in Section 1 for Tabasaran seems at least close to correct. If one counts the total number of cases given by Magometov (1965), then this is 47 for the Southern dialects of Tabasaran, on which the standard language is based, and 53 for the Northern

the great daghestanian case hoax

97

dialects. It is thus conceivable that there is some variety of Tabasaran with exactly 48 cases, and certainly this ﬁgure is within the right range. But let us now turn to the structure of these ‘‘cases.’’ 2.1. Non-local cases in Tabasaran Tabasaran has 4 non-local or core (Blake 1994: 88–97, 119–44) cases: absolutive, ergative, genitive, and dative, although as we shall see in Section 2.2 the dative actually stands at the border-line between non-local and local cases. The absolutive case, which is unmarked, is the citation form of the noun. The ergative is formed by attaching a suﬃx to the absolutive, occasionally with minor morphophonemic change, but in the singular the form of this suﬃx is not always predictable. The regular formation of the ergative singular is by the suﬃx -i; the irregular or nonproductive formations include suﬃxes -di, -ri, -yi, -li, -ni, -u, -ru, -nu (Kibrik and Kodzasov 1990: 253–4, 285–7). For example, the ergative of ul ‘eye’ is ul-i, that of t’ub ‘ﬁnger’ is t’ub-ri, that of xwar ‘mare’ is xwar-u. In the plural, which has the suﬃx -ar (constant in all cases) attached to the absolutive singular, the ergative suﬃx is always -i, e.g. ul-ar-i, t’ub-ar-i. All other cases in Tabasaran are formed by attaching the appropriate ending to the ergative form, and this applies to both singular and plural. (There are occasional morphophonemic alternations of the vowel of the ergative suﬃx in the singular.) The genitive is formed by attaching -n to the ergative, e.g. ul-i-n, t’ub-ri-n, ul-ar-i-n. The dative is formed by attaching -z, e.g. ul-i-z, t’ub-ri-z, ul-ar-i-z. The dative is used, as in Daghestanian languages generally, for the semantic roles of recipient and experiencer; in Tabasaran it is also often used, however, to encode direction (motion towards), replacing more speciﬁc local cases, and in this sense we will need to return to the dative in Section 2.2. There are various ways of analyzing this part of the system, but all boil down to recognizing 4 argument cases (a similar analysis is given in Kibrik and Seleznev 1982: 21, 33), which are opposed to a number of local cases, typically used to encode adjuncts. The endingless absolutive is clearly distinct from the other case forms, and must therefore be considered a distinct case, whether one wants to assume that it has a zero (-Ø) case formative or that it has no case formative. In addition to analyzing the ergative as having a suﬃx, as done above, one might rather consider that the form of the ergative is a thematic stem variant (oblique stem per Kibrik and Kodzasov 1990), but either way it is distinct from all other cases. If one analyzes the ergative as having a suﬃx, then the genitive and dative each has composite case suﬃxes, – 〈ergative〉-n and – 〈ergative〉-z, respectively; if the ergative is analyzed as a stem variant, the genitive and dative each has a single suﬃx, -n and -z

98

bernard comrie and maria polinsky

respectively. Either way, each of the genitive and dative has a unique case formative, so we are dealing so far with a total of 4 cases. 2.2. Local cases in Tabasaran As in other Daghestanian languages, it is the local or adjunct cases in Tabasaran that provide motivation for assigning a large number of noun cases to the language. The basic system is as follows. Depending on dialect, either 7 or 8 series of local cases can be distinguished; the distinction is based on the concept of spatial orientation, alternatively described as the position of the reference point, as in Table 1. The distinction between the series ‘at’ and ‘near’ is neutralized in the southern dialects, and does not appear in the standard language. These suﬃxes are attached to the ergative form, e.g. cal ‘wall’, ergative cal-i, local cases: cal-i-ʔ ‘in the wall’, cal-i-q ‘behind the wall’, cal-i-kk ‘under the wall’, cal-i-h ‘by the wall’, cal-i-k ‘on (the vertical surface of) the wall’; ust’ul ‘table’, ergative ust’ul-i, local case: ust’ul-iʔin ‘on the table’; lik ‘leg’, ergative plural lik-ar-i, local case: lik-ar-i-γ y ‘between the legs’. (The ‘on (horizontal)’ marker is perhaps to be analyzed as a sequence of two markers, ‘in (hollow space)’ -ʔ followed by an ‘on (horizontal)’ formative -in, as suggested by Magometov (1965: 121). The number of case formatives that has to be recognized remains the same whether or not one adopts this analysis, the choice boiling down to whether the ‘on (horizontal)’ formative is -ʔin or -in. The argument is the same as used above for the genitive and dative, and of course this same argument is needed in any event for all the local cases to account for the appearance of the ergative form before the local suﬃx proper.) The local forms discussed so far have the meaning of essive, i.e. location at a place, alternatively described as absence of motion. In order to express direction (motion towards a place) or source (motion from a place), these local cases require a further suﬃx, respectively allative -na and ablative -an, to give triads like essive Table 1. Tabasaran morphemes encoding spatial orientation ‘in (hollow space)’ ‘on (horizontal)’ ‘behind’ ‘under’ ‘at’ ‘near, in front of’ ‘among’ ‘on (vertical)’

-ʔ -ʔin -q -kk -xy -h -γy -k

the great daghestanian case hoax

99

cal-i-q ‘(located) behind the wall’, allative cal-i-q-na ‘(to) behind the wall’, ablative cal-i-q-an ‘from behind the wall’. Each of these three possibilities—essive, allative, ablative—can be combined with each of the 7 or 8 series (according to dialect), to give a total of 21 or 24 ‘‘cases,’’ in the way in which cases have been conventionally counted in Daghestanian languages. It is clear, however, that this count is misleading. Tabasaran does not really have a primary 21- or 24-way distinction among local cases. Rather, it has two morphological positions in the nominal paradigm, one encoding spatial orientation (reference point) with a 7- or 8-way opposition, the other encoding type of motion with a 3-way opposition. There is thus, so far, a total of 10 or 11 local case morphemes. Some of these can be combined with one another to give the richness of the system, but it is crucial to recognize that this richness comes from the combination of case morphemes, and not from a particularly rich set of morphemes per se; compare Kibrik and Seleznev (1982), who also speak of local case forms rather than local cases in the Dubeq dialect of Tabasaran. In fact, the possibilities for combination are even richer, since each of the 21 or 24 local forms discussed so far, plus the dative discussed in Section 2.1, can take a further case suﬃx -di (translative). The general eﬀect of this suﬃx is to indicate more general rather than more speciﬁc location or motion; in somewhat more detail, attached to an essive it carries the meaning of ‘along, over, across’, to an allative (and the dative) the meaning of ‘in the direction of’, and to an ablative the meaning of ‘from the direction of’, e.g. (Northern dialect forms, cited from Magometov (1965: 129): nir ‘river, ergative nir-i: nir-i-q ‘at (on the bank of) the river’, nir-i-q-ri ‘along (the bank of) the river’, nir-i-q-na ‘to (the bank of) the river’, nir-iq-in-di ‘towards (the bank of) the river’, nir-q-an ‘from (the bank of) the river’, nirq-an-di ‘from the direction of (the bank of) the river’. (It will be noted that there is some morphophonemic alternation in these examples; see further Magometov (1965:129–130), who notes that at least in some dialects such alternations are optional.) To summarize the Tabasaran material, there are basically 14 or 15 case suﬃxes (depending on dialect), including the -Ø marker used for the absolutive (and also indicating location after an orientational suﬃx, and more speciﬁc location in contrast with -di): the non-local case suﬃxes absolutive -Ø, ergative -i (and other allomorphs), genitive -n, dative -z; the orientational local case suﬃxes ‘in’ -ʔ, ‘on (horizontal)’ -ʔin (or -in), ‘behind’ -q, ‘under’ -kk, ‘at’ -x y, ‘near, in front of’ -h (these last two distinct only in Northern dialects), ‘among’ -γy, ‘on (vertical)’ -k; two directional suﬃxes allative -na, ablative -an; and the general-locational suﬃx -di. The combinatorial richness is made up as shown in Table 2. However, these overwhelming numbers do not imply that there are anywhere between 47 and 53 cases in Tabasaran, but rather that there is a wide range of inﬂectional forms.

100

bernard comrie and maria polinsky Table 2. Tabasaran cases non-local cases not combining with others non-local case combining with -di local cases combining orientation with —direction with -di ——Southern dialects ——Northern dialects total ——Southern dialects ——Northern dialects

3×1 1×2

3 2

7×3×2 8×3×2

42 48

3 + 2 + 42 3 + 2 + 48

47 53

In Section 3, we will apply the principles of our discussion of Tabasaran to the Tsez language, which turns out to have an even richer system than that of Tabasaran.

3.–The case system of Tsez The case morphology of Tsez surpasses that of Tabasaran, both in terms of the range of oppositions and in terms of their formal exponency. The following preliminary remarks concern properties of the system that do not relate to the number of cases. While some Tsez nouns use the same stem in the absolutive and the oblique cases, many nouns use diﬀerent stems; with such nouns, the oblique stem usually involves the addition of a segment or segments to the absolutive form, but other alternations are also found, e.g. subtraction of a segment of the absolutive or internal vowel change. For several nouns, alternatives are allowed, and perhaps as a result of this some nouns use or may use the absolutive as the stem for some oblique cases, but not for others. In what follows, we simply take such stem variation for granted. The presence of the alternation is, however, sometimes useful as a criterion for identifying something as an oblique stem. In the plural, the situation is somewhat easier, in that, barring a handful of irregularities, the absolutive plural always has the suﬃx -bi, while the oblique plural always has the suﬃx -za (to which further suﬃxes expressing case are added). Which stem these plural suﬃxes attach to is not predictable, but at least the -bi/-za opposition provides a clear indication of absolutive versus oblique. In following the material below, it should be borne in mind that there is a certain amount of usually transparent morphophonemics at morpheme boundaries; general rules include one dropping vowels before another vowel, and a second inserting the vowel e to break up disallowed consonant clusters.

the great daghestanian case hoax

101

3.1 Core cases in Tsez The core non-local case system of Tsez contains 6 cases: absolutive, ergative, genitive-1, genitive-2, dative, and instrumental. The absolutive is identical to the citation form, i.e. has a -Ø suﬃx. The ergative of many nouns takes a distinct suﬃx -o, i.e. one that is diﬀerent from all other case suﬃxes and suﬃx combinations, as in ’it’ ‘scrap’, oblique stem ’it’-r-, ergative ’it’-r-o, and this suﬃx is productive, being found for instance with loans from Arabic ending in -at. Most nouns, however, use the essive of the ‘in’ local series (suﬃx -a¯) to express the ergative function (subject of a transitive verb), and this is also productive. Given that some nouns have a distinct ergative, and that this is a substantial and productive set of nouns, we assume that there is a distinct ergative case in Tsez. The dative and instrumental have straightforward forms, with suﬃxes -r and -d respectively, although we will need to return to the dative in the discussion of local cases, where it will be suggested that the dative is a special instance of the allative suﬃx. From besuro ‘ﬁsh’, we have ergative besur-a¯, dative besuro-r and instrumental besuro-d; from mec ‘tongue’, oblique stem mec-r-, we have ergative mec-r-a¯, dative mec-re-s, instrumental mec-re-d. In the plural of besuro we have absolutive besuro-bi, ergative besuro-z-a¯, dative besuro-za-r, instrumental besuroza-d. The genitive-1 and genitive-2 have the endings -s and -z respectively, e.g. besuro-s, besuro-z, besuro-za-s, besuro-za-z. The two genitives are distinguished functionally as follows: Genitive-1 is used when its head noun is in the absolutive case; genitive-2 is used when its head noun is in an oblique case—thus providing another criterion for distinguishing between absolutive and oblique cases. Functionally, one might consider genitive 1 to be, decompositionally, the absolutive of the genitive and genitive 2 to be the oblique of the genitive. However, this functional decomposition has no formal correlate, and so we treat genitive-1 and genitive-2 as two distinct cases. The distinction between genitive 1 and genitive 2 reﬂects the category of concord, which can be distinguished from agreement (Blake 1994: 197, 199.) For further discussion of this type of case agreement, against a more general typological background, see Kibrik (1995). There are some further suﬃxes that are at least good candidates for recognition as non-local cases. In particular, there are two equative forms, equative-1 in -ce and equative-2 in -q’a¯y. The equative-2 in -q’a¯y has all the properties of a case suﬃx, attaching to the oblique stem, and in particular requiring the -za allomorph in the plural; it also requires a dependent genitive to be genitive-2; it does not attach to anything other than a singular or plural nominal stem. The equative-1 in -ce also attaches to the oblique stem, requires the -za alternant of the plural marker, and

102

bernard comrie and maria polinsky

requires a dependent genitive to be genitive-2. It thus appears to be a case marker. However, unlike the equative-2, the equative-1 suﬃx can also attach to other case suﬃxes, indeed to all other case suﬃxes and suﬃx combinations, even to the equative-2 to give the combination -q’a¯y-ce. Thus if we assume that -ce is a case suﬃx, as seems required by the general considerations given above, then we must for consistency consider the combinations with -ce to be case combinations. If we assume that everything listed so far (absolutive, ergative, genitive-1, genitive-2, dative, instrumental, equative-1, equative-2) is a case, we have so far 8 non-local cases plus a further 6 case combinations (equative-1 combined with everything except absolutive—or rather, absolutive plus equative-1 gives simply -ce, what we have been considering uncombined equative-1—and itself). There are yet further possibilities, namely three suﬃxes that are on the borderline between case and derivational suﬃxes, namely -šay, -xu ‘characterized by’,3 and -tay ‘lacking’, treated in Comrie et al. (forthcoming) as derivational. One case property lacking with these suﬃxes is the following: Personal pronouns can form all other cases, to the extent that the result makes sense, e.g. di ‘me’, oblique stem da¯-, dative da¯-r ‘to me’, equative-1 da¯-ce ‘as me’, equative-2 da¯-q’a¯y ‘as me’. However, they do not allow the forms with these three suﬃxes: *da¯-šay, *da¯-xu, *da¯-tay. On the other hand, where semantically appropriate, these three suﬃxes can be used with a plural noun, in which case the noun takes the oblique plural suﬃx -za, e.g. (1) a. q’ot’ur-za-šay-ni ged button-pl-with-def shirt ‘shirt with buttons’ b. q’ot’ur-za-tay-ni ged button-pl-without-def shirt ‘shirt without buttons’ (2) a. cˇakar-yo-xu-ni cˇay sugar-th-with-def tea ‘tea with sugar’ b. cˇakar-yo-tay-ni cˇay sugar-th-without-def tea ‘tea without sugar’ In similar vein, one can ask whether these three suﬃxes can combine with a genitive-2—they certainly cannot combine with a genitive-1; unfortunately, the results are not unequivocal, examples with the positive suﬃxes -šay and -xu being judged just about possible, those with the negative suﬃx -tay impossible:

the great daghestanian case hoax (3) a.

103

gere-z q’ot’ur-za-šay-ni ged iron-gen2 button-pl-with-def shirt ‘shirt with iron buttons’

b. *gere-z q’ot’ur-za-tay-ni ged iron-gen2 button-pl-without-def shirt ‘shirt without iron buttons’ (4) a.

kuba-z cˇakar-yo-xu-ni cˇay Cuba-gen2 sugar-th-with-def tea ‘tea with Cuban sugar’

b. *kuba-z cˇakar-yo-tay-ni cˇay Cuba-gen2 sugar-th-without-def tea ‘tea without Cuban sugar’ (5) a.

andi-za-z ciyo-xu-ni re Andi-pl-gen2 salt-with-def meat ‘meat with Andi4 salt’

b. *andi-za-z ciyo-tay-ni re Andi-pl-gen2 salt-without-def meat ‘meat without Andi salt’. In what follows, we will exclude -šay, -xu, and -tay from the calculation of cases and case combinations, on the basis of the equivocal results of applying the relevant criteria. 3.2 Local cases in Tsez As in Tabasaran, the real richness of the Tsez case system, in particular with regard to case combinations, emerges primarily with consideration of local cases, since in Tsez too we ﬁnd a multiplicity of combinations resulting from the intersection of three parameters: orientation (with a 7-way distinction in Tsez), direction (with a 4-way distinction in Tsez), and distality (with a 2-way distinction). Distality, which involves a distinction between formally marked distal (‘over there’) and formally unmarked non-distal, is not found in Tabasaran, but on the other hand Tsez lacks the speciﬁc/general parameter of Tabasaran. Looking ahead, we can see that multiplying out these possibilities will give 56 case combinations, all of which are in fact found. And since each of these can in principle be followed by the equative-1 suﬃx -ce, this gives a grand total of 112 local case combinations, to which we can add the 14 case combinations from Section 3.1 for a grand total of 126.

bernard comrie and maria polinsky

104

Spatial orientation

Table 3. Tsez local case forms: non-distal.

‘in’ ‘among’ ‘on (horizontal)’ ‘under’ ‘at’ ‘near’ ‘on (vertical)’

essive

allative

Case ablative

-a¯ -λ -’(o) - -x(o) -de -q(o)

-a¯-r -λ-er -’o-r --er -xo-r -de-r -qo-r

-a¯y -λ-a¯y -’-a¯y --a¯y -x-a¯y -d-a¯y -q-a¯y

versative (‘towards’) -a¯γor -λ-xor -’-a¯γor, -’-a¯r --xor -x-a¯γor, -x-a¯r -d-a¯γor, -d-a¯r -q-a¯γor, -q-a¯r

Since the Tsez local case combinations are not quite as transparently segmentable as their Tabasaran counterparts, we present two tables giving the non-distal (Table 3) and distal (Table 4) case combinations; note that the alternative forms in the versative column of Table 3 are contracted forms in free variation with the fuller forms. The individual morphemes can now be analyzed as follows. The basic forms of the 7 orientation suﬃxes are: ‘in’ -a¯, ‘among’ -λ, ‘on (horizontal’ -’(o), ‘under’ -, ‘at’ -x(o), ‘near’ -de, ‘on (vertical)’ -q(o). For the forms given with o in parenthesis, the form without o is used word-ﬁnally after a vowel, e.g. besuro ‘ﬁsh’, besuro-x, but is ‘bull’, is-xo. When further suﬃxes are attached, the o is always present, e.g. besuro-xo-r, unless it drops regularly before another vowel, e.g. besuro-x-a¯y, is-x-a¯y. The directional suﬃxes are essive -Ø, allative -r, ablative -a¯y, versative -γor/-a. The essive suﬃx is thus identical with the absolutive suﬃx, and only one zero sufﬁx need be posited. The allative suﬃx is identical to the dative suﬃx, and therefore they must be subsumed as a single suﬃx, which we will arbitrarily call dative; thus, the so-called ‘‘dative’’ is the dative attached to the bare stem, while the so-called ‘‘allative’’ is the dative attached to a local stem bearing a suﬃx of local orientation. The ablative is basically -a¯y, but shortens to -ay after the vowel a¯ in a preceding inﬂectional suﬃx, as in the distal forms.5 The versative suﬃx is the only real problem, since it must be given two completely distinct representations, namely -γor in the non-distal (devoicing to -xor after a voiceless consonant), but -a in the distal. This is the only instance of suppletion, although it does not aﬀect segmentability. It should further be noted that these directional suﬃxes also occur, without a preceding orientation suﬃx, after certain nouns with inherently locational semantics, e.g. idu ‘home, at home’, idu-r ‘to home’, id-a¯y ‘from home’, idu-γor ‘towards home’. A remaining problem in the non-distal combination is the vowel a¯ that appears

the great daghestanian case hoax

105

Spatial orientation

Table 4. Tsez local case forms: distal.

‘in’ ‘among’ ‘on (horizontal)’ ‘under’ ‘at’ ‘near’ ‘on (vertical)’

essive

allative

Case ablative

-a¯z -λ-a¯z -’-a¯z --a¯z -x-a¯z -d-a¯z -q-a¯z

-a¯z-a-r -λ-a¯z-a-r -’-a¯z-a-r --a¯z-a-r -x-a¯z-a-r -d-a¯z-a-r -q-a¯z-a-r

-a¯z-ay -λ-a¯z-ay -’-a¯z-ay --a¯z-ay -x-a¯z-ay -d-a¯z-ay -q-a¯z-ay

versative (‘towards’) -a¯z-a -λ-a¯z-a -’-a¯z-a --a¯z-a -x-a¯z-a -d-a¯z-a -q-a¯z-a

in many forms at morpheme boundaries but has not been accounted for so far. In the versative forms, its occurrence can readily be predicted by a combination of phonological and morphological factors: If the orientation suﬃx ends in a vowel (which may be the ‘‘parenthetical o’’ of Table 3, plausibly so since this would normally show up before a further suﬃx), then a¯ is obligatory placed after the orientation suﬃxes; in other versative forms, it is disallowed. The distal suﬃx is -a¯z, located immediately after the orientation suﬃx. Before a following allative suﬃx, an inserted a (perhaps a¯, undergoing shortening after the preceding a¯) is obligatory. All other changes are regular: note in particular that the ‘in’ suﬃx -a¯ is regularly deleted before another vowel, as in its distal equivalent -a¯z (for -a¯-a¯z). We leave open whether the various instances of a¯ (‘in’; epenthetic/‘general’; perhaps versative, where a could be shortened from a¯) are to be identiﬁed with one another, since we have no evidence for or against such an identiﬁcation. As against the 126 case combinations, we have the following set of case morphemes: 8 non-local cases (including the absolutive in -Ø, and therefore not counting any further instances of -Ø as distinct); 7 orientation suﬃxes; 2 directional sufﬁxes (bearing in mind that allative -r is subsumed under dative); 1 distal suﬃx—for a total of 18 case suﬃxes, a respectable total, but hardly one meriting an entry in The Guinness Book of Records.

4. Implications for a theory of case We have tried to show that in both Tabasaran and Tsez, we have a moderately rich number of cases: 14 or 15 in Tabasaran, depending on dialect, and 18 in Tsez. The richness that gives rise to claims such as Tabasaran having 48, 47, or 53 cases, or Tsez having 126 cases, derives from the possibilities of combining these cases with

106

bernard comrie and maria polinsky

one another. It is not the case that children acquiring Tabasaran and Tsez have to learn 47, 53, or 126 cases; rather, they have to learn the 14, 15, or 18 suﬃxes and the general principles for combining them. There are two important considerations to bear in mind here, morphological and semantic transparency, both of which are brought up by Blake in his discussion of similar systems (Blake 1994: 153–4). 4.1 Morphological transparency First, the combinations are morphologically transparent, as one would expect from the generally agglutinative morphological structure of these languages. In other words, forms that we analyze as a combination of ‘‘behind’’ and ‘‘ablative’’ can be shown to have a formative that expresses ‘behind’ and a formative (possibly a zero formative, though always in contrast with one or more overt formatives) expressing ‘‘ablative.’’ There may be some morphophonemic alternation, but this is never suﬃcient to remove the possibility of identifying the two formatives. One might contrast this with two other languages that are analyzed as having rich case systems, namely Finnish and, especially, Hungarian. The comparison is interesting in that these two languages are also largely agglutinative, with number and case (and also possessor) in general readily identiﬁable in combinations. In Finnish, the relevant local case suﬃxes are as presented in Table 5, in a paradigm which one usually ﬁnds in Finnish grammars. It is possible to rewrite this paradigm into a combination of ‘‘orientation points’’ ‘in’ and ‘on’ and three cases. In terms of a componential analysis of the semantics, one can clearly identify a twoway opposition of orientation (‘in’ versus ‘on’) and a three-way opposition of direction (essive versus allative versus ablative), as in Table 6. All the ‘on’ forms have initial -l, so one might abstract this as the suﬃx for ‘on’ (though the remaining -le of -lle would not be identiﬁable with any independently occurring morpheme). This would suggest perhaps analyzing -lta as -l-ta, in which case -sta would plausibly be analyzed as -s-ta. The -s and -l recur in -ssa and -lla, which would then have to be analyzed as -s-Ca and -l-Ca, where C is a consonant slot assimilating to the preceding segment. This, incidentally, mirrors the historical Table 5. Finnish local case suﬃxes, Version 1 inessive illative elative adessive allative ablative

-ssa -hVn -sta -lla -lle -lta

the great daghestanian case hoax

107

Table 6. Finnish local case suﬃxes, Version 2 essive allative ablative

‘in’

‘on’

-ssa -hVn -sta

-lla -lle -lta

development, where Ca reﬂects *na (Hakulinen 1979: 103–105). The -hVn suﬃx is more problematic. It has allomorphs -hVn, -Vn (in both of which V takes on the quality of the immediately preceding vocalic segment), -seen (singular)/-siin (plural); etymologically, the s/h (for *s) is probably the same s as in the other ‘in’ forms, but the remainder is not to be identiﬁed with any part of the other local case suﬃxes (Hakulinen 1979: 103–104). Thus, with the Finnish material one can carry out a certain amount of internal analysis, but of the 6 forms one remains completely unanalyzed, one shows a suﬃx that occurs only in that form, and for the remaining 4 forms one needs 4 formatives; in other words, it is not clear that the decompositional analysis at a synchronic level has any real advantage over saying that there are just 6 unanalyzable case suﬃxes. There is a real diﬀerence between such a system and one as found in Tabasaran and Tsez, where the formal evidence for the decomposition is clear. Hungarian presents a richer set of local cases, since it distinguishes three orientations, with citation both of the case suﬃxes attached to nouns (Table 7) and the corresponding case forms of the third person singular pronoun (which also serve as bases for corresponding forms of the other person–numbers) (Table 8). But the picture is much the same as in Finnish. With the case suﬃxes and the pronouns, the three forms in the ‘at’ column are totally unrelated to one another. The case suﬃx Table 7. Hungarian local case suﬃxes essive allative ablative

‘in’

‘on’

‘at’

-ban -ba -ból

-n -ra -ról

-hoz -nál -tól

Table 8. Hungarian third person singular pronoun in local cases essive allative ablative

‘in’

‘on’

‘at’

benne belé belo˝le

rajta rá róla

hozzá nála to˝le

108

bernard comrie and maria polinsky

for essive ‘on’ bears no relation to the other two forms in that column. The ablative row items share a common element -ól (or its vowel-harmony variant), and bear a constant relation in the ‘in’ and ‘on’ series to the allative, but these are not in turn regularly related to the essive. Decomposition seems to give even less advantage than in the case of Finnish, and traditional Hungarian grammars are surely correct in listing these 9 forms as distinct cases. 4.2 Semantic transparency The second consideration is that the combinations of case morphemes are also semantically transparent, i.e. at least the basic, literal meaning of each combination is predictable from the combination of the meanings of the individual parts. Thus, a form like Tabasaran xul-xy-an-di ‘house-at-ablative-general’ has the meaning ‘from the direction of at the house’, predictable from the meaning of the individual components. This is not to deny that Daghestanian local cases often have non-local uses alongside their local uses, this in part compensating for the small inventories of non-local cases. In Tsez, for instance, the essive of the ‘on (vertical)’ in -q is also used to express the possessor in predicative constructions, for example: (6) žek’-qo-Ø ʕomoy yo. man-on-ess donkey:abs be:prs ‘The man has a donkey.’ In Tabasaran, the ablative of the ‘on (vertical)’ series in -k-an is used to translate English ‘about, concerning’. In Tsez, orientation ‘at’ in the essive case has the secondary meanings of goal or purpose, and orientation ‘near’ has the secondary interpretation of the comitative. We assume that these are secondary, at least etymologically metaphorical uses of the same case suﬃxes; many of them are paralleled in other languages, e.g. the use of the same form to express location and possession. It is of course conceivable that with time the non-local use could take over as the basic meaning of a case, as has happened with the so-called essive case of Finnish -na, which was originally a local case etymologically identical to the -Ca element hypothesized above, but in modern Finnish means ‘as, in the role of’, as in opettaja-na ‘as a teacher’; Matsumura (1994) argues that this has also already happened with the adessive in the closely related Estonian language. But this does not yet seem to have happened to the basic local cases in Tabasaran or Tsez. 4.3 Dative: core or non-core case? One problem that has arisen in our consideration of the case systems of Daghestanian languages has been the precise delimitation between grammatical and

the great daghestanian case hoax

109

semantic cases, in particular between grammatical and local cases—always assuming, of course, that there is a precise delimitation to be made. Indeed, part of our argument below will be that there is no such clear delimitation in Daghestanian languages, because of the tendency for cases and case forms from the rich set of local case forms to enter the domain that in other languages would be covered by grammatical cases. This can be illustrated particularly clearly by the dative. Blake (1994: 144–151, especially 145) sets out the range of functions that dative cases frequently cover cross-linguistically, claiming that the most typical functions are indirect object and, in a somewhat smaller set of languages, beneﬁciary, goal, and purpose.6 Other functions, in particular destination, are found even less typically, though still quite frequently. Blake does, however, also note that it is possible for datives to derive from local cases, as in the case of Romance a (Spanish, Italian a, French à), which both expresses indirect object and retains local functions of the kind associated with its Latin etymon ad ‘to’, expressing direction (Blake 1994:173). From the semantic viewpoint, the function of the protoypical dative can be decomposed into that of direction, as in the act of transfer, and that of the beneﬁciary (‘‘interested party,’’ as suggested in van Belle and van Langendonck 1996: xv– xviii). Arguably, both of these functions are consistent with the semantics of local cases. The directional function clearly implies a locative reading and is therefore consistent with the encoding of goal, destination, or purpose (as an abstract goal which has to be reached, attained, etc.). All these functions evoke the semantics of the allative, and we will refer to this function below as the allative function. The beneﬁciary bears resemblance to a possessor, and possession is often construed as location at or with the possessor, thus compatible with the semantics of the essive, hence the essive function below.7 In theory, one language can maintain both functions within the conﬁnes of one morphological form. However, since the functions, although semantically related, are still distinct, one of them may be suppressed and the other made prominent. Since the essive implies absence of motion, if the essive function is made prominent, that can easily allow for the reanalysis into a more abstract and therefore, more grammatical meaning. This is the situation in classical Latin or in Old Church Slavic, where multiple dative functions are all rather abstract and motion towards can be encoded by another argument case, the accusative, but not by the dative. Likewise, Hungarian has a clear dative (the case in -nak/-nek), with little or no locative semantics. If the essive function is suppressed and /or the allative function is made more prominent, one can expect that the directional semantics of the case would be maintained. This expectation is borne out by the above mentioned prepositional dative of modern Romance languages as well as by Finnish and Daghestanian, to which we now turn. The way of encoding indirect objects in Finnish is to use the so-called allative in -lle. However, this case clearly ﬁts into a set of local cases, and has the

110

bernard comrie and maria polinsky

literal meaning ‘onto’. Grammars of Finnish almost invariably give the local meaning as primary, which is surely correct historically, and given the tightly knit system of local cases may well also be correct synchronically, in which case the allative would be basically a local case which has nonetheless acquired the function of encoding indirect objects, i.e. a function typical of grammatical case. Let us now turn to Daghestanian languages, starting with Tabasaran. Magometov (1965) treats the Tabasaran dative along with the absolutive (in his terminology, nominative), ergative and genitive, i.e. as a grammatical case, and not with the local cases. However, Magometov also notes that the dative is frequently used in Tabasaran to express motion towards, in place of the more speciﬁc local cases with orientation suﬃxes, and, moreover, that the dative can be combined with the nonspeciﬁc suﬃx -di, which otherwise occurs only with local cases; the combination of dative with -di has clearly local meaning, ‘in the direction of, towards’, e.g. xal ‘house’ (oblique stem xul-a-), dative xul-a-z ‘(to) home’, xul-a-z-di ‘homewards’ (Magometov 1965: 128). The Tabasaran dative, then, seems to stand on the borderline between local and grammatical case. Another aspect in which it behaves more like a grammatical case is its use to encode the experiencer of certain verbs expressing psychological predicates (Kibrik 1985: 282–283), accidental events, etc., e.g. (7) izu-s b-iqun-is bay. I-dat anim-found-sg1 boy ‘I found the boy.’ In (7), the preﬁx b- on the verb agrees with the animate noun phrase bay in the absolutive case, while the suﬃx -is agrees with the dative pronoun izus, this agreement again suggesting grammatical rather than local status for the dative noun phrase. In Tabasaran, then, alongside the clearly grammatical absolutive, ergative, and genitive, the dative occupies an intermediate status between grammatical and local case, combining features that are otherwise typical of each set of cases. The nature of the dative in Tsez leads to similar conclusions, although the precise details are somewhat diﬀerent, in addition to which we have more information on the behavior of the dative than is available in the literature on Tabasaran. First, let us note the salient feature that makes it similar to a local case. The suﬃx -r, which we are somewhat arbitrarily labelling ‘dative’ can occur after orientation suﬃxes, and in this combination its meaning is that of motion towards, i.e. local. Only when it occurs without an orientation suﬃx does it acquire the function of encoding an indirect object (in addition to some other non-local functions noted below). Thus the single suﬃx with which we are concerned has both clearly local and clearly grammatical functions. Restricting ourselves now to the Tsez dative suﬃx without an orientational sufﬁx, we observe that, as in Tabasaran, it has a grammatical function in addition to

the great daghestanian case hoax

111

that of encoding indirect object, namely that of encoding the experiencer of psychological predicates, etc., e.g. (8) madina-r obiy Ø-eti-x. Madina-dat father II-like-prs ‘Madina likes father.’ (In (8), the null preﬁx on the verb shows agreement in class with the absolutive noun phrase obiy; only vowel-initial verbs show agreement in Tsez, and agreement is only with the absolutive noun phrase.) But in fact, in Tsez we ﬁnd even further evidence of the permeation between local and grammatical cases. Surely the most striking example is the fact that most nouns use the essive of the ‘in’ local series in ergative (transitive subject) function, as in: (9) obiy-a¯ magalu b-is-si father-in:ess bread III-buy-pstevid ‘Father bought the bread.’ (The word magalu ‘bread’ belongs to class III.) Note that Tabasaran, by contrast, has a distinct ergative case, as do most Daghestanian languages. A Tsez local case form from the ‘on (vertical)’ orientation series, namely the essive in -q(o), also plays a role as a grammatical case marker in ways similar to the dative in Tsez. In Tsez, indirect objects can take either -r or -q(o), the latter being the dative of the ‘on (vertical)’ series and expressing temporary possession by the recipient, while the simple dative without an orientation suﬃx expresses permanent possession, e.g. (10) už-a¯ kid-be-r eλu te-si. boy-erg girl-th-dat blueberry give-pstevid ‘The boy gave (the) blueberries to the girl (to keep).’ (11) už-a¯ kid-be-q eλu te-si. boy-erg girl-th-on:vert:ess blueberry give-pstevid ‘The boy gave (the) blueberries to the girl (for a while).’ Just as the dative is used to express the subject of certain verbs, such as psychological predicates, so too the essive of the ‘on (vertical)’ series is used to express the subject in a construction expressing accidental action, as in (12): (12) da¯-q cˇ’ikay y-exu-s. I-on:vert:ess glass II-break-pstevid ‘I accidentally broke the glass.’ (The word cˇ’ikay ‘glass’ belongs to class II.) What these data suggest is an increased tendency in languages with rich case

112

bernard comrie and maria polinsky

systems for local cases to permeate the domain of grammatical cases. Only further investigation will tell whether this typological correlation does indeed hold up, and also what other typological correlates there may be in this area.

5. Conclusion As a ﬁnal note, we observe that The Guinness Book of Records, in discussing complexity in verbal morphology, says (Young 1997: 249), in the same paragraph cited in Section 1: The Ample language of Papua New Guinea has over 69,000 ﬁnite forms and 860 inﬁnitive forms of the verb.

Crucially, this statement does not refer to the number of tenses, moods, or whatever, but simply to the number of forms, i.e. the number of possible combinations. Had its reference been to noun forms, rather than to cases, then the number for Tabasaran might have seemed both more plausible, and indeed rather low: given that Tabasaran has distinct singular and plural forms, its dialects have 94 or 106 forms, while by the same token Tsez has 252. But crucially, we are here dependent on the power of multiplication, so that even a small number of morphological oppositions that can combine with one another can soon give a large number of forms. Imagine a somewhat idealized Turkic language, with 2 numbers, 6 cases, and 6 sets of possessive suﬃxes (for each of three persons and two numbers). None of the individual morphological oppositions is particularly rich, but the total number of combinations is already 72. Add in one more number, i.e. a dual, and the total goes up to 162 (surpassing Tabasaran). The moral of this is that the enterprise of counting combinations of morphological categories, while no doubt in some sense ‘‘fun,’’ may detract from the more serious task of identifying the number of morphological categories and the principles that permit their combination.

Notes 1. This paper is based in part on work supported by the National Science Foundation under grant SBR9220219. It incorporates material provided by Ramazan Rajabov, Research Assistant to this project. Some of the material on the case system of Tsez was presented by Comrie at a meeting of the research group Frontiers in Morphology at the University of Surrey, England, in June of 1996; we are grateful to other participants in the meeting, and also to Margarte Langdon, for their comments. 2. The full last clause of the quotation given in the text is ‘‘Tabassaran, a language of Daghestan, Azerbaijan, uses the most noun cases.’’ As indicated in the text, Daghestan is not part of Azerbaijan,

the great daghestanian case hoax

113

and indeed Tabasaran (also spelled: Tabassaran) is spoken in Daghestan, not in Azerbaijan, though perhaps confusingly Azeri (Azerbaijani) is used as a lingua franca by speakers of northern varieties of Tabasaran. The misleading geo-political information is not included in the British edition of The Guinness Book of Records. 3. The suﬃx -šay indicates inseparability, -xu separability, but the two tend not to be distinguished in modern Tsez, and the distinction is not made in the examples cited in the text. 4. The Andi are another Daghestanian people. ʕ 5. It does not, however, shorten after an a¯ that is part of the stem, e.g. ’ a¯-’-a¯y ‘oﬀ the roof’, where ʕ ʕ ’ u ‘roof’ has the oblique stem ’ a¯-. The alternation is thus both phonologically and morphologically conditioned. 6. One of the problems with such an account seems to be confusion between a grammatical function (indirect object) and several semantic functions commonly but not necessarily associated with that grammatical function. This underscores the problems inherent in a cross-linguistic categorization of the dative (see also van Belle and van Langendonck 1996: xvi–xvii). 7. The literature on this particular issue is unusually extensive and we will limit ourselves to the classical work by Benveniste (1960), which has subsequently inspired a large number of studies, including purely structural analyses (Hoekstra 1995).

References Benveniste, Émile. 1960. ‘‘ ‘Être’ et ‘avoir’ ’’ dans leurs Fonctions Linguistiques. Bulletin de la Société de linguistique de Paris 55/1: 113–34. Blake, Barry J. 1968. A Brief Description of the Kalkatungu Language. Canberra: Australian Institute of Aboriginal Studies. ——. 1977. Case Marking in Australian Languages. Canberra: Australian Institute of Aboriginal Studies. ——. 1994. Case. Cambridge: Cambridge University Press. Bokarev, Evgenij A. 1959. Cezskie (Didojskie) Jazyki Dagestana [The Tsezic (Didoic) languages of Daghestan]. Moscow: Izd. AN SSSR. Comrie, Bernard, Maria Polinsky and Ramazan Rajabov. Forthcoming. Tsezian Languages. In Alice Harris & Rieks Smeets (eds): The Caucasian Languages. Edinburgh: Edinburgh University Press. Hakulinen, Lauri. 1979. Suomen Kielen Rakenne ja Kehitys [Structure and Development of the Finnish language], 4th ed. Helsinki: Otava. Hjelmslev, Louis. 1935. La Catégorie des Cas. Étude de Grammaire Générale. Première partie. (Acta Jutlandica VIII.1). Aarhus: Universitetsforlaget i Aarhus. Hoekstra, Teun. 1995. ‘‘To Have to Be Dative.’’ In Hubert Haider, Susan Olsen and Sven Vikner (eds.): Studies in Comparative Germanic Syntax. Dordrecht: Kluwer, 119–38.

114

bernard comrie and maria polinsky

Imnajšvili, David S. 1963. Didojskij Jazyk v Sravnenii s Ginuxskim i Xvaršijskim Jazykami [The Dido Language in Comparison with Ginux and Xwarši]. Tbilisi: Izd. AN Gruzinskoj SSR. Kibrik, Aleksandr E. 1985. ‘‘Toward a Typology of Ergativity.’’ In Johanna Nichols and Anthony C. Woodbury (eds.), Grammar Inside and Outside the Clause. Cambridge: Cambridge University Press, 268–323. ——. 1995. Direct–Oblique Agreement of Attributes in Daghestanian. In Frans Plank (ed.), Double Case: Agreement by Sufﬁxaufnahme. New York: Oxford University Press, 216–229. —— and Sandro V. Kodzasov. 1990. Sopostavitel’noe izucˇenie dagestanskix jazykov. Moscow: Izd-vo MGU. —— and Mixail G. Seleznev. 1982. ‘‘Sintaksis i Morfologija Glagol´nogo Soglasovanija v Tabasarnskom Jazyke’’ [The morphosyntax of verbal agreement in Tabasaran]. In Vladimir A. Zvegincev (ed.): Tabasaranskie Ètjudy. [Essays on Tabasaran]. Moscow: Izdvo MGU, 17–33. Magometov, A. A. (1965). Tabasaranskij Jazyk (Issledovanie i Teksty) [The Tabasaran Language (Investigation and Texts)]. Tbilisi: Izd-vo ‘‘Mecniereba.’’ Matsumura, Kazuto (1994). Is the Estonian Adessive Really a Local Case? Journal of Asian and African Studies 46/47: 223–235. Nikolayev, Sergey L. and Sergey A. Starostin 1994. A North Caucasian Etymological Dictionary. Moscow: Asterisk. van Belle, William, and Willy van Langendonck (eds.). 1996. The Dative. Vol. 1: Descriptive Studies. [Case and Grammatical Relations Across Languages 2]. Amsterdam: John Benjamins. Xanmagomedov, B. G.-K. (1967). ‘‘Tabasaranskij Jazyk’’ [The Tabasaran Language]. In E. A. Bokarev, K. V. Lomtatidze, Ju. D. Dešeriev, G. B. Murkelinskij, M. A. Kumaxov, S. M. Xajdakov and A. K. Šagirov (eds.), Iberijsko-Kavkazskie Jazyki [Ibero-Caucasian Languages], 545–561. (Jazyki narodov SSSR [Languages of the Peoples of the USSR], 4.) Moscow: Izd-vo ‘‘Nauka.’’ Young, Mark C. (ed.) (1997). The Guinness book of records 1997. (US edition.) New York: Bantam Books.

Iwaidja mutation and its origins NICHOLAS EVANS 1 Linguistics & Applied Linguistics University of Melbourne

1. Introduction2 In this paper I discuss the sole known case of an Australian language that has developed initial mutation as a grammatical process3. In addition to regular mutations still operating in the language, many other lexemes display evidence of frozen mutations that, unless the historical morphology is understood, wreak havoc with any attempt to reconstruct the regular sound correspondences that are a prerequisite to understanding the historical phonology of the family. I shall argue that the predominant cause of mutation is a relic of an old gender preﬁx that has, by and large, been dropped as Iwaidja has lost the ancestral gender system; mutated forms continue morphophonemic alternations to the initial of the root which were conditioned by the erstwhile preﬁx. 1.1 The Iwaidjan family The Iwaidjan family is a group of non-Pama-Nyungan languages spoken in Western Arnhem Land in the Cobourg Peninsula region (see Map). Only Iwaidja and Maung are still being passed on to children; each of these has around 150 speakers concentrated in the communities of Minjilang (Croker Island) and Warruwi (Goulburn Island) respectively. The other languages are all near extinction: Marrgu has one speaker only, Garig and Ilgar (two almost identical dialects4) about three full speakers between them, Amurdak three full speakers, and Wurrugu, probably the ‘Popham Bay language’ recorded by Earl (1846), has one remaining rememberer (Evans 1996). I have no direct information about Manangkari, which Capell and

nicholas evans

116

N ARAFURA

SEA MARRGU

WURRUGU

Croker I.

ILGAR GARIG C o b o u rg Pe n i nsu

l

a

Melville Island

IWAIDJA MAUNG

Va n D i e m e n Gulf

AMURDAK

TIMOR SEA

Darwin 0

50 km

Map 1. Location of the Iwaidjan languages. Hinch (1970) call a dialect of Maung; Charlie Wardaga, a speaker of Ilgar, Iwaidja and Maung, says it is ‘‘like Maung but a bit diﬀerent,’’ but I have yet to get details. Finally, two sources mention a further sub-dialect of Iwaidja, using the terms Mangawulu (Berndt and Berndt 1970: 9) and Manganawal (Pym n.d.), which may or may not be identical; nothing further is known about this variety. Sources for the languages are: Amurdak (Am): Garig (Ga): Ilgar (Ilg): Iwaidja (Iw): Marrgu (Mrg): Maung (M): Wurrugu (W):

Handelsmann (1991) and subsequent unpublished work. Whitehead (1991), Pym (1977), Evans (ﬁeld notes). Evans (ﬁeld notes). Pym and Larrimore (1979); Evans (ﬁeld notes). Capell (1963), Hinch (1966), Evans (ﬁeld notes). Capell and Hinch (1970). Evans (1996) and Earl (1846).

The internal relationships of the Iwaidjan languages are shown in Figure 1. While there are various shared morphological patterns with other Arnhem Land families such as Gunwinyguan, the Maningrida group, the Wagaity group, Tiwi and Umbugarla, it is still unclear which is the closest coordinate family. M and Iw vocabularies have borrowed heavily from Kunwinjku, and all Iwaidjan languages have also borrowed from Macassarese and Malay (see Evans 1992, 1997). At present we

iwaidja mutation and its origins

117

proto-Iwaidjan

Iwaidjic

Warrkbi

Wurrugu

Marrgu Iwaidja Garig Ilgar Manangkari

Maung Amurdak

Figure 1. Genetic relationships of the Iwaidjan languages. lack reliable vocabularies, or extensive grammatical information, for Marrgu, Amurdak or Wurrugu; as a result most comparative statements made in this paper will be conﬁned to the Iwaidjic subgroup. However, the existence of cognates in other north Australian families for most of the gender morphology we will be discussing suggests that Iwaidjic is relatively conservative, at least in this regard. Typologically, the Iwaidjan languages have subject and object preﬁxes to the verb but no noun incorporation; a ﬁve-class gender system is reconstructable for free nouns and for subject and object preﬁxes, but only M retains all ﬁve genders. Ilg and Ga retain the masculine/feminine contrast; along with Iw they retain some traces of the old neuter and vegetable genders in conjugational irregularities of the verbal preﬁx system. 1.2. Iwaidjan phonologies Figure 2 summarizes the consonant systems of the Iwaidjan languages, giving IPA symbols for the phonemes followed by the practical orthography used in the current sources on these languages5, and in the rest of this paper. Like most Australian languages, the Iwaidjan languages have paired stops and nasals, several points of articulation for laterals. The presence of the velar approximant phoneme /γ/ is unusual in Australia (Evans 1995) but an areal feature of the Western Arafura coast, shared with Tiwi to the west and Gunbarlang to the east. All the Iwaidjan languages except M are notable for the proliferation of liquid phonemes, superimposing a phonemic ‘‘ﬂapped vs non-ﬂapped’’ contrast on

nicholas evans

118

Stop Nasal Approximant Tap Trill Lateral a Flapped Lateral a b

Bilabial

ApicoAlveolar

ApicoLaminoRetroﬂex Dentalb

LaminoAlveolar

Velar

b (b) m (m) w (w)

d (d) n (n)

(rd) (rn) ɹ (r) (rt)

t (j) (ny) y (y)

k (k) (ng) γ (h)

ˇl (ld)

r (rr) l (l) ˇ (rld)

d (dh) n (nh)

(rl)

λ (ly)

absent from M Marrgu only

Figure 2. Iwaidjan consonant systems. the alveolar and retroﬂex laterals6. A minimal pair from Iw is /kui/ ‘skin’’ vs /kuˇi/ ‘cheeky yam,’ (the corresponding Ilg forms are yiui and kuˇi); further examples of sub-minimal pairs from Iw are /lama/ ‘shovel-nosed spear’ vs /lˇla/ ‘and,’ and /gau/ ‘no’ vs /gaˇuri/ ‘meat’ (Pym and Larrimore 1979: 5), and from Amurdak are /lalmana/ ‘ancestral creator being’ vs /landa/ ‘dingo,’ and /auma/ ‘mother’ vs /guˇi/ ‘cheeky yam type’ (Handelsmann 1991: 7–11). Further complications to the set of phonemic liquids come from the presence of three rhotics: cf Iw /maɹu/ ‘who,’ /maru/ ‘cabbage palm’ and /mau/ ‘bandicoot.’ All have but one stop series, lacking the long/short distinction of their neighbours to the south and east, and no glottal stop. All but M have just three vowels: /a/, /i/ and /u/; the extra M vowels /o/ and /e/ appear mostly in loans from Kunwinjku and Kunparlang. Mrg has an interdental series lacking in the other Iwaidjan languages; it also has phonetic dental and labial fricatives whose phonemic status is still unresolved. 1.3. Sound correspondences within Iwaidjan There are many regular correspondences between Iw and the other Iwaidjic languages; in most, the initials are identical.7 Iwaidja form (1) m:m marrk ‘desire’ martan ‘small’

Other Iwaidjan form M marrk ‘want’ M martan ‘small’

iwaidja mutation and its origins (2) ng:ng ngabi ‘I’ ngalhij ‘liver’

M, Ilg ngabi ‘I’ M ngalhij ‘liver’

(3) w:w wirrhala ‘throwing stick’ waharti ‘elbow’

M wirrhala ‘throwing stick, boomerang’ M wahari ‘elbow’

(4) y:y yanyjuk ‘milk, breast’ yidbilk ‘ﬂying ﬁsh’

M yanyjuk ‘milk, breast’ M yidbirl ‘ﬂying ﬁsh’

(5) a:a ajbud ‘beach, sand’ alij ‘spear’

M, Ilg ajbud ‘beach, sand’ M alijalij ‘harpoon spear’

119

Initial /ld/ in Iw corresponds to the identical phoneme in Ilg, against /l/ in M: (6) ld:l lda ‘and’ ldirri ‘anger, ﬁght’

M la , Ilg lda ‘and’ M lirri ‘hot, anger, ﬁght,’ Ilg -ldirrindirri ‘cheeky’

However, we also have correspondences in which the Iw form undergoes initial mutation, of nasal to stop, approximant to stop, or lateral to retroﬂex glide; where the M word begins with an initial /a/ Iw has an epenthetic initial /w/. Note that in some cases the relevant comparison is with a root (which takes a gender preﬁx in M and Ilg/Ga—see Section 1.4) rather than with an independent word. Table 1 summarizes the attested correspondences. Iwaidja form

Other Iwaidjan form

8

(7) b:m barryun ‘young man’ bawurr ‘arm’ (8) k:ng kartalk ‘tongue’ kijalk ‘fruit’ kurljak ‘rich person, important person’

M marryun ‘boy’ M -mawurr ‘arm,’ Ilg yi-mawurr ‘his arm muscle’ M ngartalk ‘tongue,’ Ilg yi-ngartalk ‘his tongue’ M -ngijalk ‘body,’ Ilg yi-ngijalk ‘his body’ M -ngurlyak ‘important, famous’

nicholas evans

120

(9) b:w burrwurr ‘young person’ M wurrwurr ‘new’ birrkalk ‘hand, ﬁnger’ Ilg yi-wirrkalk ‘his hand, ﬁnger’ (10) k:h (/γ/) kamung ‘sandhill’ karlu ‘not’

Ga hamung ‘sandhill’ (Pym 1977) Ilg yi-harlu ‘no, not’

(11) k:w kuwarr ‘Ubarr cer.’ kujurn ‘white clay, beautiful’

M wuwarr ‘Ubarr ceremony’ A ujurn ‘white clay,’ M wujurn ‘beautiful,’ kujun ‘white clay’

(12) j:y jihi ‘tooth’ jamin ‘his/her turn’ (13) r:l rakbi ‘heavy’ rakbirrij ‘lips’

M yihi ‘tooth, point of spear,’ Ilg yi-yihi ‘his tooth’ M yamin ‘3 sg recip/reﬂ’ M lakbi ‘heavy’ M -lakbirrij ‘mouth, opening,’ Il yi-lakbirrij ‘mouth’

(14) wa:a wartad ‘one’ wartak ‘far’

Ilg artad ‘one,’ M -arta ‘other’ M -artak ‘across’

Except for words beginning with k, there are few if any non-loan cognates between Iw and M with initial stops (Table 2). This is because the sound change of initial lenition, which has applied in all the Iwaidjic languages, aﬀected all stops except initial k and even there it lenited initial stops forming part of preﬁxes of form Table 1. Attested initial correspondences between Iwaidja, M and Ilg/Ga. Regular M, Ilg m ng w y l (M), ld (Ilg)

Mutated Iw m ng w y ld

M, Ilg m ng w h y ø (_a) l (M)

Iw b k k k j w r

iwaidja mutation and its origins

121

Table 2. Percentage of known loan sources for initial stop: stop correspondences. Initial correspondence

Total

< other Austra- < Macassarese lian languages or Malay

< English

No known loan source

b:b d:d or d:rd j:j k:k

16 1 10 51

0 0 0 8

2 0 0 0

5 = 31% 0 = 0% 0 = 0% 35 = 71%

9 1 10 8

ku- or ki- (see Evans 1997 for details and examples). This removed indigenous words with initial stops from both M and Iw; this phonotactic type has gradually been replenished in both languages with loans from other Australian languages (especially Mayali/Kunwinjku),9 Macassarese and Malay, and English. There are also morphophonemic alternations in Iw between root-initials in the historically expected form, and root-initials displaying the same pattern of initial mutation; the only gap with respect to the mutated correspondence sets is the rare k:h alternation, and there is also only one example of a k:w alternation. Generally, the expected initials appear after plural preﬁxes, while the mutated forms appear on singular nouns, and on the third person singular form of verbs and adjectives, although there are further conditions on mutation, to be discussed below. There is virtually perfect correspondence between the initials of the underlying roots in Iw, established on the basis of the plural form, and the initials of the cognate roots in M and Iw, so that some words in the following lists also occurred in the lists of cognates given above. Except for the k:w alternation, dozens of instances of each type exist in Iw; but for reasons of brevity I conﬁne myself here to a couple of examples of each type: (15) b:m baju ‘(s)he is tired, sick; dies’, pl. a-maju banga ‘throat,’ a-manga ‘throats’ (16) k:ng kartalk ‘tongue,’ a-ngartalk ‘tongues’ kirlirrk ‘bone, skeleton,’ a-ngirlirrk ‘bones, skeletons’ (17) b:w baharl ‘head,’ a-waharl ‘heads’ bani ‘(s)he sits, stays, lives, exists’, pl. a-wani (18) k:w kurtwiny ‘shy, embarrassed (sg),’ pl. a-wurtwiny (19) j:y

jigi ‘tooth, ﬁshhook,’ a-yigi ‘teeth, ﬁshhooks’ jinmul ‘nose,’ a-yinmul ‘noses’

(20) r:ld rakbi ‘heavy (sg.),’ a-ldakbi ‘heavy (pl.) ralal ‘stomach, bottom of boat (sg.),’ pl. a-ldalal

122

nicholas evans

(21) wa:a warnangkad ‘strong (sg.),’ arnangkad ‘strong (pl.)’ warndi ‘(s)he is high,’ arndi ‘they are high’

2. Gender and mutation in Iwaidja grammar Since the synchronic alternations described above give a clue to the historical origins of initial mutation, it will be necessary to examine the grammatical functions of mutation in Iw more thoroughly. Before doing so, however, it is useful to look at the behaviour of gender preﬁxes in those languages (M and Ilg/Ga) which retain the ancestral system more fully. 2.1 Gender systems in the Iwaidjan family Proto-Iwaidjan almost certainly had ﬁve genders (masculine, feminine, vegetable, neuter and miscellaneous). (22) ja ilijab ja arrarrkbi k-i-wani M m:dem m:small m:dem man prs-3.S-sit ‘That small man sits.’ (23) jida ninyalijab jida warramungbik k-iny-bani M f:dem f:small f:dem woman prs-3f.S-sit ‘That small woman sits.’ (24) mada ma-lijab mada warlk ka-ma-wani M ve:dem ve-small ve-dem tree prs-3ve.S-sit ‘That small tree sits.’ (25) da wu-lijab da kunak k-ang-bani M nt:dem nt:small nt:dem camp prs-3nt.S-sit ‘That small camp sits.’ (26) da awa-lijab da walij k-a-bani M mis:dem mis:small mis:dem food prs-3mis.S-sit ‘That small (amount of) food sits.’ (27)–(31) illustrate gender agreement with objects, with masculine subjects in the left hand column and feminine (and other) subjects in the right hand column. (27) yi-ni-wung M 3m.O-3m.A-hit ‘he hit him’

yi-nga-wung 3m.O-3f.A-hit ‘she hit him’ (also used if subject is in ve, nt or mis class)

iwaidja mutation and its origins (28) yiny-i-wung M 3f.O-3m.A-hit ‘he hit her’

yiny-nga-wung 3f.O-3f.A-hi ‘she hit her’ (also used if subject is in ve, nt or mis class)

(29) ma-ni-wung M 3ve.O-3m.A-hit ‘he hit it (ve)’

ma-nga-wung 3ve.O-3f.A-hit ‘she (etc.) hit it (ve)’

(30) a-ni-wung M 3.nt.O-3m.A-hit ‘he hit it (nt)’

a-nga-wung 3nt.O-3f.A-hit ‘she (etc.) hit it (nt)’

(31) [form not attested] M

aga10-wung 3misS.O:3femM.A-hit ‘she (etc.) hit it (mis)’

123

Consistently in the verbal preﬁxes, and less consistently within the np, the ﬁve gender preﬁxes form the third person singular subset of a paradigm distinguishing person (four-way, with an inclusive vs exclusive contrast in the ﬁrst person) and number (singular vs plural). Our focus will not be on these forms. The other languages have all simpliﬁed this system in diﬀerent ways. Ilg/ Ga has retained the most: for adjectives and most body part nouns within the np, they retain the masculine vs feminine distinction; nouns which would be in other genders in M (e.g. vegetable or neuter) have fallen into an expanded masculine class. To exemplify from Ilg: (32) a. raka arrarrkbi yi-ldujumartal Ilg that man 3m-small ‘‘that man is small’ b. raka wurruwajba yiny-jujumartal that woman 3f-small ‘‘that woman is small’ c. raka arlirr / inyali yi-ldujumartal that tree / yam 3m-small ‘‘that tree / yam is small’ (33) yi-ngartalk yiny-ngartalk Ilg 3m-tongue 3f-tongue ‘his tongue; tongue (of a man)’ ‘her tongue; tongue (of a woman)’ Ilg/Ga have also retained distinct masculine and feminine forms for subject and

nicholas evans

124

object preﬁxes to the verb; as in M, object preﬁxes precede subject preﬁxes when both are third person,11 and there is an absolutive system with most object (O) preﬁxes formally identical to intransitive subject (S) preﬁxes, and distinct forms for transitive subject (A) (cf 34, 35). A few intransitive verbs, whose subjects are trees, grass or wood, take a reﬂex of the M vegetable-class preﬁx (36). There are also a few vestiges of the vegetable and neuter classes as ﬁxed object preﬁxes with certain verbs, to be discussed in Section 3.2.2 below. (34) Ilg

yi-wani 3msgS-sit ‘he sits’

yiny-bani 3fsgS-sit ‘she sits’

(35) a. yi-ni-yalmang Ilg 3msgO-3msgA-looks.for ‘he looks for him’

yi-nga-yalmang 3msgO-3fsgA-looks.for ‘she looks for him’

b. yiny-i-yalmang 3fsgO-3msgA-looks.for ‘he looks for her’

yiny-nga-yalmang 3fsgO-3fsgA-looks.for ‘she looks for her’

(36) a. ma-ldu mirlak Ilg 3vesgS-smokes grass ‘The grass is smouldering.’ b. ma-ldi raka arlirr 3vesgS-stands that tree ‘There is a tree.’ c. ma-rldarramang mirlak 3vesgS-burns grass ‘The grass is burning.’ In all other Iwaidjan languages the noun class system has been lost, though traces survive in a variety of ways. In Iw, the main subject of this article, traces of some noun classes survive in frozen forms in some nouns derived by change of noun class, but more importantly as mutation triggers, and as alternative (conjugationally determined) preﬁx sets on certain verbs. 2.2 Grammatical functions of mutation in Iwaidja My discussion in this section draws on the exposition of Iwaidja morphophonemics in Pym and Larrimore (1979), though I have organized it somewhat diﬀerently. The set of formal changes labelled above as mutation serve the following grammatical functions in Iw.

iwaidja mutation and its origins

125

2.2.1 Formation of singular intransitive verbs The mutated form is the 3sg value of a paradigm containing all seven person/number combinations, exempliﬁed here for the verbs -mirtan ‘sing,’’ -ngardbuni ‘fall,’ -wani ‘sit’ and -yamang ‘work.’ Note that the same changes to the root-initial consonant are triggered by the preﬁx-ﬁnal stop d in the 1pl. and 2pl. preﬁxes. (37) 1 sg 1 inc. pl. 1 exc. pl. 2 sg. 2 pl. 3 sg. 3 pl.

‘sing’ nga-mirtan ad-birtan ngad-birtan ang-mirtan gud-birtan birtan a-mirtan

‘fall’ nga-ngardbuni ad-kardbuni ngad-kardbuni ang-ngardbuni gud-gardbuni kardbuni a-ngardbuni

‘sit’ nga-wani ad-bani ngad-bani ang-bani gud-bani bani a-wani

‘work’ nga-yamang ad-jamang ngad-jamang an-jamang gud-jamang jamang a-yamang

In the other Iwaidjan languages, the reﬂexive is formed by using the intransitive instead of the transitive preﬁx set. The presence of this principle in Iw means that for verbs whose stem begins with one of the relevant segments, the 3sg reﬂexive form is mutated: (38) ri-winybun Iw 3m/3-washnp ‘he washes it’

ri-wujirran 3m/3smellnp ‘he smells it’

ri-ldalkun 3m/3-cutnp ‘he cuts it’

(39) binybun Iw 3:washnp ‘(s)he washes (him/herself)’

bujirran 3:smellnp ‘(s)he/it smells, is smelly’

ralkun 3cutnp ‘(s)he cuts himself’

2.2.2 Third person singular object marker on transitive verbs with non-third person subjects Mutation in the morpheme following the preﬁx is found with the forms having a third person object, as long as the subject is not also third person. Examples with ‘take,’ whose root is m-initial and mutates to b, and ‘‘ﬁnd,’’ whose stem is vowelinitial and preﬁxes a w, are given below; verbs with other initials behave according to the same pattern of mutation we have already seen (see Pym and Larrimore 134–143 for further examples). This is a further example of the absolutive distribution of pronominal preﬁxes in Iwaidjan—the absolutive pattern is found with actual gender preﬁx forms in M and Ilg/Ga, but with the distribution of mutation in Iw.

nicholas evans

126 (40) Subject 1sg 1 pl inc 1 pl exc 2 sg 2 pl 3 sg.M 3 sg.fem 3 pl

‘take, carry (3sgO)’

‘take, carry (3plO)’

‘ﬁnd (3 sg O)’

‘ﬁnd (3 pl O)’

aban arruban ngarruban kaban kurruban riman kaman buman

ayunman adbunman ngadbunman angmunman kudbunman animan anngaman anduman

awadbi arruwadbi ngarruwadbi kawadbi kurruwadbi radbi kadbi badbi

ayunadbi adbunadbi ngadbunadbi angmunadbi kudbunadbi anadbi anngadbi anbadbi

2.2.3 Formation of the (third person) singular of body parts, adjectives, a few other nouns, and pronouns Note that for all of these categories except the inalienably possessed category this change does not apply to all eligible words; rather, a subset need to be memorized as undergoing it. In addition to the many examples given in Section 2.2, here are one or two examples of each type: Singular of body part—This forms part of an inalienably possessed series, all taking pronominal preﬁxation for the possessor, and are often better translated as ‘‘his/her X, their Xes’’ rather than simply by a singular/plural opposition. (41) a-mawurr ‘(their) arms’ bawurr ‘(his/her) arm’ Iw [cf nga-mawurr ‘my arm,’ ang-mawurr ‘your arm’ etc.] Singular of adjective—Here again, there are two alternative glosses: preﬁxed forms may sometimes be glossed as predicates with a relevant subject pronoun; in other cases simply as an attributive pronoun. (42) a-wurrul-ad12 burruli a-ldirrindirri rirrindirri

‘they are good; good, well (pl.)’ ‘(s)he is good, well; good, well (sg.)’ ‘they’re angry; angry (pl.)’ ‘(s)he’s angry; angry (sg.)’

Singular of miscellaneous noun—There are only two examples of this type given by Pym and Larrimore, though there may be a few more in the language.13 (43) marryun ‘young men’ barryun ‘young man’ aldujudbularr ‘young, plu’ rujud ‘young, sg’ [cf ngunbuj ‘younger sibling,’ plural ngunbujun; kamu ‘mother,’ plu. kamularr; etc.]

iwaidja mutation and its origins

127

Singular of preposition—An unusual feature of Iw and the other Iwaidjic languages is a small set of inﬂected prepositions (probably originating as serial verbs). Two of these are predominantly adnominal—‘‘having; with’’ and ‘‘lacking, without’’ while two more mainly code relations at clause level—‘‘from’’ and ‘‘on, on top of.’’ Unlike in Welsh or Hungarian, where prepositions inﬂect for features of their dependent, in Iw they inﬂect for person and number of their head. The four relevant prepositions are: a-wunyak-ud bunyak a-wurnari burnari a-wurran burran arndi w-arndi ang-arndi

‘(3 pl.) possessing, with,’ ‘(3sg) possessing, with’ ‘(3 pl.) not having, lacking, without’ ‘(3sg) not having, without’ ‘(3 pl.) from’ ‘(3sg) from’; cf Ilg yi-wurran ‘3sg.M-from’ ‘(3 pl.) on, on top of’ ‘(3sg) on, on top of’ ‘(2sg) on, on top of’

Third singular pronoun—Note that this doesn’t precisely follow the morphophonemic pattern, since the plural form is wanad rather than the form a-yanad that would yield janad by regular mutation. (44) wanad ‘they’ wamin ‘they (contrastive)’

janad ‘(s)he’ jamin ‘(s)he (contrastive)’

2.2.4 Formation of a few derivational nouns Certain types of metaphorical extension, signalled in other languages of the region by switching gender (see 55–57 below for some M examples), are indicated by mutation in Iw. An example is bangartalk ‘foliage,’ apparently a mutated form of mangartalk ‘ﬂame,’ itself formed from -ngartalk ‘tongue’ by preﬁxing the now non-productive vegetable-class preﬁx ma-. 2.3 The synchronic morphophonemics of Iwaidja mutation The treatment of Iw mutation adopted by Pym and Larrimore (1979) is to postulate a morphophoneme, which they label K, that is part of the underlying representation of certain verbal preﬁxes and triggers the appropriate morphophonemic changes. This morphophoneme may occur alone, as in the 3sg preﬁx K-, or as the ﬁnal element of a preﬁx following ‘‘real’’ phonemic material, as in the 1A/3sgO preﬁx aKor the 2pl/3sg preﬁx kurruK-; it may also precede other pronominal preﬁxes, as in the combination K-nga- [3sgobj-3sgfem.subj].

128

nicholas evans

Examples of each change, as instantiated by derivations in their grammar, are: (45) Km → b,

cf. ri-man [he/3sg-carry] ‘he carries it’ but %kurruK-man% [2pl/3sg-take] → —kurruban ‘you pl. are taking it’ [ p. 124]

(46) Kw → b,

cf. ri-wun [he/3sg-hit] ‘he hits it’ but %aK-wun% [1/3sg-hit] → abun ‘I hit it’

(47) Kng → k,

cf. ri-ngulda [he/3sg-make] ‘he makes it’ %aK-ngulda% → a-kulda ‘I make it’

(48) Ky → j,

cf. ri-yalman ‘he hunts for it’ %aK-yalman% → a-jalman ‘I hunt for it.’

(49) Kld → r,

cf. nga-ldimbarni ‘I sit down,’ %K-ldimbarni% → rimbarni ‘he sits down’

(50) K→ w / _V, cf. %aK-unma% [1/3sg-count] → awunma ‘I count’14 %K-ara % [3sg-go] → wara ‘(s)he goes’ Pym and Larrimore do not attempt to justify why they label this K rather than say D, although the alternation with w before vowels at least demonstrates it is a noncoronal. However, there is one set of data which do not ﬁt the otherwise consistent picture. On p. 81 they point out that when the root-initial vowel is a, the resultant consonant may be either w or rd, and that cannot be predicted other than by specifying the verb; some verbs even allow both. Thus in addition to the w- in wara ‘(s)he goes,’ some verbs take rd (medially) or d (initially), e.g. (51) aK-ayan → ardayan ‘I see it’ K-arryu → darryu ‘it is open’ aK-ajukun → ardujukun or awujukun ‘I wait for him’ (Pym and Larrimore 1979: 81–2)15 Apart from this problematic set, I believe Pym and Larrimore’s to be the correct treatment of the verb paradigms, and a vast advance on the unsatisfactory analysis in Capell 1962: 133–4, which vacillates between taking the mutated and non-mutated forms as underlying.16 However, Pym and Larrimore do not make full use of the morphophoneme they postulate. For example, they do not extend it to account for mutation in singular/plural alternations like bawurr /a-mawurr. A treatment that would account for this naturally would be to take the root as mawurr, the plural preﬁx as a-, and the singular preﬁx as K-:

iwaidja mutation and its origins (52) K-mawurr → bawurr sg-arm ‘arm’

129

a-mawurr pl-arm ‘arms’

It would also allow a more natural account of sets such as (53) kartalk angartalk mangartalk bangartalk

‘tongue’ ‘tongues’ ‘ﬂame’ ‘foliage’

In their account no relation to the latter two lexemes is mentioned, but this is clearly a case of residual use of old noun-class morphology (in this case, the vegetable class, then feeding the K-preﬁx) to derive new lexemes by metaphorical extension; on my treatment these would be represented as three single preﬁxes, and a combination, (underlyingly) to the same root: (54) K-ngartalk a-ngartalk ma-ngartalk K-ma-ngartalk Supplemented in this way, their account of Iw morphophonemics is largely satisfactory, although there are a couple of remaining irregularities, such as the singular form of some a-initial substantives, which preﬁx aw-: examples were given above. Pym and Larrimore do not make any diachronic claims about the source of K, and it is to this question that we now turn, by looking at its source as the ‘‘miscellaneous’’ noun class preﬁx in M and presumably in proto-Iwaidjic.

3. Tracking down big K in proto-Iwaidjic First note that initial mutation in Iw generally only occurs where noun class preﬁxes can occur in Ilg/Ga and M: on adjectives and a few class-preﬁxed nouns, 3rd singular pronouns, and pronominal preﬁxes to verbs. The only place where Iw mutation doesn’t occur in a noun-class type environment is with 1st plural preﬁxes; another explanation must be found for these. 3.1 Parallels to aK- in the Maung noun class system If we now look for formal parallels to aK- in M, we ﬁnd that what Capell and Hinch call the ‘‘VI class marker’’,17 which marks the ‘‘miscellaneous’’ class has a form

130

nicholas evans

that we will argue below should be represented as aK-. The miscellaneous class preﬁx goes on names of food plants and vegetable foods, introduced foodstuﬀs, winds and clouds; human body and certain parts of it; some abstract nouns; ceremonial terms. Note the following M cross-classifying series, for example, where miscellaneous-class markers produce initial mutation after their initial a-: (55) yi-nimi a-dimi

‘backbone of male’ ‘main radicale of potato plant or yam vine’

(56) i-mawurr a-bawurr

‘arm of male’ ‘arm or tendril of vine’

(57) i-ngijalk ma-ngijalk a-gijalk wu-ngijalk

‘body’ (masculine) ‘fruit’ ‘body of food trees, i.e. fruit’ ‘sea; north’

Capell and Hinch give the form of the miscellaneous noun class preﬁx as aw-, and formulate changes hardening w and ng to k before consonants. An alternative would be to treat it as ak- before consonants (with subsequent cluster simpliﬁcations) and aw- before vowels. For now I will represent it as aK-, which has the realization aw before vowels, and before consonants takes the form a plus a range of morphophonemic changes to be outlined below. In fact, Capell and Hinch’s treatment of morphophonemic changes involving w as ﬁrst element (see their chart on p. 36, which sometimes represents the VI object form as if it ended in b) is inconsistent with data they give elsewhere, but the following changes have been excerpted from various parts of their grammar. The morphophonemic changes triggered by the miscellaneous preﬁx aK- are: (58) %Km% → b, e.g. %naK-malal% → nabalal [VI-good] (p. 56)18 %Kw% → b, e.g. %aK-wijarr% → abijarr [VI-dry] (p. 57) % Kn% → d, e.g. %aK-nimi% → adimi [VI-backbone] %Kl% → d, e.g. %aK-ludbuj% → adudbuj [VI-short] (p. 57) %Kng% → k, e.g. %kaK-nga-ma% → kakama [VIobj-IIsubj-take] (p. 37) %Kg% → k, e.g. %aK-garra% → akarra ‘plenty’ (p. 58) %Ky% → j, e.g. %aK-yanad% → ajanad ‘theyVI’ (p. 59) Some adjectives fail to undergo the expected initial mutation; with these the resultant preﬁx is awa- (cf the preﬁxation of a- to some adjectives in Iw): (59) aK- + -lijab ‘little’ → awa-lijab [p. 57] At a deeper level, we can relate the alternations it triggers to a sound change of intervocalic lenition that applied at an early stage in the history of Iwaidjic, leniting

iwaidja mutation and its origins

131

intervocalic stops according to the schema set out in (60); see Evans (1997) for fuller discussion, and illustrative cognate sets. (60) Intervocalic lenition: in the environment V(L)_V, Medial p t k becomes w ɹ y γ aK-, as a consonant ﬁnal preﬁx, would have protected the following segment from the application of intervocalic lenition, but it would have also induced special processes of cluster simpliﬁcation and assimilation. Of the set given in (58), we can identify the %Kw% → b and %Kg% → k rules as preserving the etymologically original consonant in a cluster enviroment, while the %Km% → b and %Kng% → k rules induce an alternation away from the etymologically original nasal roots; words conforming to the %Ky% → j rule are a mixture of original j-roots and original y-roots. We do not know enough yet to decide on the etymological status of the roots participating in the %Kl% → d rule. The matching of M aK- with Iw K- , then, is shown by the parallelism of the rules in (58) with the Iw alternations discussed in Section 2. The only additional step required is to propose that a destressing of the initial preﬁxal syllable led to loss of the a, and we get all the other phonological changes by regular morphophonemic processes known to have occurred in Iw.19 The formal equation of Iw K- ~ aK- with the M miscellaneous gender preﬁx aKthus seems safe. However, a semantic problem remains. Body part nouns in M (which are the class of nouns with the biggest set of mutated cognates in Iw) do not take the miscellaneous class, but rather the class of their possessor, though of course there are some cases, such as when talking about the ‘‘arms’’ (i.e. tendrils) of miscellaneous class nouns suchs as vines, where the class of the possessor is the miscellaneous class. But generally, for nouns designating body parts of humans, the class of the possessor will be masculine or feminine, so that the M body part nouns will have forms like (61) i-mawurr M-arm iny-mawurr F-arm

‘his arm’ ‘her arm’

as do their Ilg/Ga cognates, which are identical to those just given. Iw ‘arm,’ by contrast, has the form bawurr, which is closest formally to M aK-mawurr > abawurr ‘arm or tendril of vine.’ To account for this discrepancy we need to assume that, after the breaking oﬀ of Iw from the other Iwaidjic languages Ga and Ilg, pre-Iw shifted away from marking body parts in terms of the gender of their possessor, towards marking them in terms of their intrinsic gender. Now such an ambiguity in the gender of body parts is widely present in Arnhem

132

nicholas evans

Land languages. In Umbugarla, for example, body parts agree in gender with their possessor,20 but have their own intrinsic gender when it comes to governing their own modiﬁers; two examples are (62) and (63), in which the body part nouns receive the feminine gender appropriate to their possessor, but govern neuter agreement on modifying adjectives. (62) marmu kw-arik ki-mira wife nt-bad fem-back ‘My wife’s got a sore back.’ (63) ki-jamark kw-arik fem-teeth nt-bad ‘(a woman’s) bad teeth’ In Eastern Kunwinjku, body part nouns can be formed on two patterns, with both in free variation: they can be preﬁxed with the neuter noun class preﬁx, or they can be suﬃxed with the ‘‘possessed noun’’ suﬃx. Thus ‘nose’ can be either kun-keb [nt-nose] or keb-no [nose-his/hers/its], ‘‘eye’’ can be kun-mim [nt-eye] or mim-no [eye-his/her/its] and so on. Preﬁxation with the neuter is the only pattern available in more westerly dialects, such as Gun-djeihmi, while the possessor-suﬃxed pattern is the only pattern available in the Dalabon and Rembarrnga, two neighbouring languages that are closely related to Kunwinjku and have exerted a number of structural inﬂuences on its eastern dialects. The pattern of alternation in Eastern Kunwinjku thus represents the intersection of two construction types; its emergence at the dialect level shows how rapidly a shift can take place from one type to the other. These examples, which could be multiplied (see Evans 1994 for fuller discussion than is possible here), indicate that a shift from possessor class to intrinsic class can rapidly be made within a language system, and the non-correspondence of class between M, Ilg and Ga body parts on the one hand, and Iw on the other, can be seen to result from such a shift. With other types of noun the shift is more of a puzzle. Why, for example, should the Iw word for ‘boy,’ or the third person singular pronoun, have shifted into what is formally the miscellaneous class? It is not enough to say simply that the miscellaneous class has been generalized (as a preﬁx to head nouns), since there is a large class of nouns which have not received it. We are faced with claiming that a formal shift of class has occurred in a large set of nouns, which has a semantic basis in body part nouns but none in other nouns. I return in Section 4 to what I regard as the most likely explanation: that the correspondences in body part nouns between Iwaidja and Garig/Ilgar served as a sort of template for subsequent analogical changes motivated by conscious social diﬀerentiation, with a rule roughly like ‘where they say yi-X, we say K-X’.

iwaidja mutation and its origins

133

Although our main focus has been the relic forms of the miscellaneous class, which produces initial mutation, there are relic forms of other classes as well. We already saw the example of mangartalk (ﬂame) as an example of a noun retaining the old vegetable class preﬁx ma-; a number of other plant names begin with ma- but we cannot prove the segmentation because there is no alternation with an unpreﬁxed version, e.g. mamungarr ‘ eucalyptus sp.’, marlinkarrk ‘ macrozamia palm. ’ (There are also many loans from Kunwinjku with man- vegetable preﬁx, e.g. manburrwa ‘material, cloth,’ manjad ‘straight’ ). Nouns with frozen masculine preﬁx i- include imajak ‘ bird’s wing’ (cf Kunwinjku medjek ‘goosewing fan’), while a frozen feminine preﬁx iny- is found in inybarr ‘stones uncovered at low tide’ ; cf Dalabon bad-no ‘rock.’ A possible example of a frozen wu- preﬁx is the demonstrative set: (64) wubaka wubaki wuka wuki

‘somewhere, over there’ ‘somewhere, over there’ ‘over there, that way’ ‘this way, over here’

3.2 Mutation in Iwaidja verbs Let us now turn to another grammatical trigger to mutation in Iw, verbal preﬁxes. We begin with intransitive and reﬂexive verbs with third person subjects, then pass to the object preﬁxes. 3.2.1 Intransitive and reﬂexive verbs Intransitive verbs in Iw, as we saw in Section 2.2, preﬁx K- in the third person singular, as with nga-wani ‘I sit,’ bani ‘(s)he sits.’ Within the M paradigm, the rootinitial alternations with the miscellaneous gender verbs (exempliﬁed here in third person singular present forms) correspond closely to the Iw pattern: (65) M F nt ve mis

talk prs kinginka kinynginka kanynginka kamanginka kakinka

sit down, stay prs kiwani kinybani kangbani kamawani kabani

lie down, sleep prs kiyu kin(y)u kanyu kamau kaju

With reﬂexive verbs, which are formed in Iw simply by substituting intransitive pronouns for transitive pronouns, the pattern does not correspond so well to M, which has a special set of preﬁxes to the verb (Capell and Hinch 1970: 80–81) that don’t correspond to the Iw construction. Note, though, that it is possible to substitute

nicholas evans

134

intransitive preﬁxes onto transitive verbs in M with certain eﬀects, such as the formation of mediopassives (ibid: 85): (66) ngungmadjbud mada malakbirrij 1–nt-close ve door ‘I closed the door’21 (67) kangmabud22 mada malakbirrij 3sgnt-close ve door ‘the door closed’ Although their description contains only this example of the construction, with vegetable/neuter class subject, presumably it is also available with miscellaneous class subjects, and if this construction went back to proto-Iwaidjic this would have been the source of the Iw reﬂexive, though it would have required two changes in Iw: a generalization of this mediopassive construction to all reﬂexives, and a generalization of the miscellaneous noun class as the third person preﬁx in intransitives. Compared to the situation with nouns (where the tension between inherited and intrinsic gender in body parts provided the necessary transition), we have more diﬃculty in accounting for why the rare miscellaneous class should be the form generalized on verbs. Was it simply analogy from the nominal morphology, or did it spread from constructions with miscellaneous-class subjects? Against the second route, we may cite the extreme rarity of miscellaneous-class subjects in M texts. An examination of the ﬁrst three texts in Capell and Hinch (1970) revealed the following distribution of noun class preﬁxes to the verb in A (transitive subject), S (intransitive subject), and O (object) functions, from a total of 251 occurrences. The miscellaneous class preﬁx does not occur even once in the data; there are also no examples of reﬂexive constructions. The rarity of aK- class verbal preﬁxes in the corpus makes it highly unlikely that the adoption of the miscellaneous-class preﬁx as the sole verbal preﬁx occurred through generalization across the verbal preﬁx set. At present, then, we are forced to postulate a certain development on formal Grammatical Function

M

F

A S O

46 77 97

1 11 2

ve

nt

mis

0 1 1

0 4 11

0 0 0

Figure 3. Frequency of occurrence of gender preﬁxes to the verb in A, S and O functions for the ﬁrst three texts in Capell and Hinch (1970: 107–117)

iwaidja mutation and its origins

135

grounds, but the actual steps involved seem counter-intuitive unless we assume analogic spread from the nominal paradigm. 3.2.2 Transitive verbs and object preﬁxes It is in the system of transitive verb preﬁxes that Iw has retained the most numerous traces of the proto-Iwaidjic (pI) gender system. Although some preﬁxes with third singular objects trigger mutation, and again we shall trace these preﬁxes to the miscellaneous-gender object markers in pI, there are also survivals of the other gender forms from pI, though no longer functioning as a gender system. To simplify the exposition I will concentrate on the non-future forms, but similar points can be made for the future set, which are derived from the non-future set through the insertion of a further preﬁx ba(na)- ~ mana- following the subject/object preﬁxes. Firstly, the masculine/feminine subject distinction survives when the object is third person, whether singular or plural. As with M and Ilg (already illustrated in Section 2.1), the preﬁx ordering is obj-subj if both are third person. (data from Pym and Larrimore p. 138). (68) a-ni-man 3plO-3sgM.A-take/carryprs ‘he takes/carries them’ (69) an-nga-man 3plO-3sgF.A-take/carryprs ‘she takes/carries them’ (70) ka-man %K-nga-man% 3O-3sg.F.A-take/carryprs ‘she takes/carries him/her/it’ As these illustrate, the masculine/feminine distinction is lost for objects, with all being represented by K. The postulation of initial K as the object marker is made in Pym and Larrimore (1979: 126–7), where they analyse ri- < K-i- and ka- < K-nga. The second derivation is morphophonemically regular, but the ﬁrst is otherwise unattested.23 In more complex preﬁx sequences, which we do not consider here, the 3sg object preﬁx does not immediately precede the 3sg subject preﬁx; an example is: k-ana-nga-wunya [3obj/3fem.subj-imp-she-cook] ‘let her cook it’. Note that K is always positioned before the subject preﬁx, reﬂecting the Obj-Subj-V order universal in the Iwaidjic languages when both subject and object are third person. When the subject is non-third person, the Iw morpheme ordering is Subj-Obj-V, again as for M and Ilg. This is clearly illustrated by (71)–(72), with third person plural objects and non third person subjects, and (73), with third singular object and

nicholas evans

136 Ilg/Ga

3sgS/O

pI and M

F

iny-

F

iny-

M

yi-

ve

ma-

nt

ang-

mis

aK-

M

yi-

Iwaidja

3sgS/O

-(u)K

Figure 4. Collapse of proto Iwaidjan gender systems in Iw and Ilg/Ga (simpliﬁed). non-third person subject. Note that the third person object marker is again (u)K-, which takes the form /w/ or /uw/) before vowel-initial stems, and u plus hardening of eligible consonants before C-initial stems (data from Pym and Larrimore 1979: 135 and 138): (71) a-yun-adbi 1sgA-3plO-ﬁndprs ‘I ﬁnd them’

a-yun-man 1sgA-3plO-take/carryprs ‘I take/carry them’

(72) awadbi %a-K-adbi % 1sgA-3sgO-ﬁndprs ‘I ﬁnd him/her’

aban % a-K-man % 1sgA-3sgO-take/carryprs ‘I take/carry him/her’

(73) kurr-uw-adbi %kurr-uK-adbi% 2plA-3sgO-ﬁndprs ‘you pl. ﬁnd him/her’

kurr-u-ban %kurr-uK-man% 2plA-3sgO-take/carryprs ‘you pl. take/carry him/her’

With most verbs, then, 3sg Object is represented by K, the Iw reﬂex of the pI miscellaneous object marker. In other words, the ﬁve distinct genders distinguishable for objects in proto-Iwaidjic have been neutralized in Iw in favour of a form that continues the miscellaneous gender. Ilg/Ga, despite its close similarities to Iw, has collapsed the system in a diﬀerent way, to a two-way choice between masculine and feminine for both subjects and objects, as discussed in Section 2.1. The diﬀerential eﬀects of the collapse in Iw and in Ilg/Ga may be diagrammed as in Figure 4. The discussion above holds for most verbs in Iw and Ilg/Ga. However there are some verbs, both in Iw and in Ilg/Ga, that follow a diﬀerent pattern. Eﬀectively, the pI vegetable and neuter object preﬁxes have been reanalysed as a sort of preﬁxal conjugation marker, occurring with a small class of verbs regardless of the gender of the object. Thus there are six verbs in Iw belonging to the so-called ‘mam-verb

iwaidja mutation and its origins

137

2sg

2 pl

3sg. M

3sg fem25

3 pl

ngi-

ku-

kurri-

(k)26ini-

(k)inga-

(k)awuni-

a–27

ku–

kurri–

inii- ~ ri-

inganga- ~ ka-28

iwu-

kuny-

kurriny- (k)inyi-

(k)inynga-

kuny-

kurriny- (y)inyi-

(k)inydja-/ (k)inybinybu-

Sub 1sg Obj 3 M: —Maung 3 M: Ilg 3 sg: Iw

3 fem: nginy—Maung 3 fem: Ilg any3 veg: —Maung 3 mam: Ilg 3 mam: Iw 3 neut: M

anyanyngung-

kunykunykung-

kurrukurrunykurrung-

3 ang: Ilg

ang-

kung-

kurrung-

3 ang: Iw

ang-

kung-

3 misc: —Maung 3 sg: Iw

ngaw-

kaw-

aK-

kaK-

kurruK-

3 pl: —Maung 3 pl: Ilg 3 pl: Iw

nganwa- kannga- kudb-

inynga- ~ inga

(ka)mani-

(ku)manga-

kurrung-

kambu-/ pst mambuman(i)mangmambumambumambumambukung(k)a(ng)nga(k)angyu-/ (f. angbani-) (k)angbanianga- (G annga-) angbu(<%angni%) (<%anng-nga) (Ga anbu-) angbuangbuangbu-

kurraw-

(k)aw-

(k)aw-

(k)aw-

ngawun- kanbun- kudbun- (k)awuni-

(k)awunnga-

ayunayun-

anngaannga-

(k)awundu-/ (k)awunbanbuanbu-

angmun- kudbun- aniangmun- kudbun- ani-

Figure 5. Fuller set of Iwaidjic object preﬁxes, showing survival of vegetable and neuter preﬁxes in Iw and Ilg.

class’ (P and L 90), and around a dozen verbs belonging to the so-called ‘ang-verb’ class (P and L 88–9). This results in the preservation of many of the vegetable and neuter object preﬁx forms in Ga/Ilg and Iw, as summarized in Figure 5. Note that future forms are not shown here, except where irregularities are relevant for comparative purposes. In all three languages futures are formed semi-regularly by adding, to the past/present preﬁx, a reﬂex of *bana-; this mostly has the form bana- in Ilg, ba- in M and mana- in Iw. To make the workings of these verb classes clearer, we will illustrate with the verbs marrun ‘eat’ (Iw) and -urrun ‘burn oﬀ’ (Ilg) for the mam-class, and -ldijbun ‘collect water’ (Iw) and wurrwun ‘know’ (Ilg) for the ang-class. To underline the

138

nicholas evans

fact that these preﬁxes no longer function as gender (i.e. they are not produced by agreement) I shall gloss them as mam- (Iw) / ma- (Ilg) and ang- (Iw/Ilg), rather than by their etymological genders, which would be vegetable and neuter respectively. Consider ﬁrst the ‘mam-verb’ class. With ﬁrst or second person subjects these take a third person object preﬁx nya- in place of the expected uK-; the nya- is a reﬂex of the pI third person feminine object preﬁx, otherwise unattested in Iw (see Figure 5). With third person subjects, the form is mambu- regardless of the number and gender of the subject; this continues the old 3plsubject > 3 vegetable object form found in the past in M, with an otherwise attested extension from plural to singular in Iw. (74) Iwaidja-Mam verb a. a-nya-marrun 1sgA-3sgmam.O-eat.meat ‘I eat meat.’ c. gurru-nya-marrun 2plA-3sgmam.O-eat.meat ‘You (pl.) eat meat.’

b. arru-nya-marrun 1plA-3sgmam.O-eat.meat ‘We (incl.) eat meat.’ d. mam-bu-marrun 3sgmam.O-3sg/plA-eat.meat ‘He/she/they eat meat.’

The situation in Ilg is similar, but with two important diﬀerences. Firstly, the nypreﬁx, with non-third person subjects is only found in second singular form (75a);29 in the present it is replaced by n- (75b) or u- (75c). Secondly, distinct forms of the object preﬁx, based on ma-, appear with all third person subjects (75 d, e, f); they can be derived from an underlying sequence mang-ni/nga/bu [ang-3sgM/3sgF/3pl] by regular morphophonemic rules. Apart from the present non-third person forms, all the Ilg forms continue the pI forms attested in M. (75) Ilgar-Ma verb a. ku-ny-urrun 2sgA-3m.O-make.ﬁrenp ‘You (sg) are burning oﬀ.’

b. a-n30-urrun 1sgA-3sgM.O-make.ﬁrenp ‘I am burning oﬀ.’

c. arr-urrun (< arru-urrun) 1pl.incA/3sgM.O-make.ﬁrenp ‘We are burning oﬀ.’

d. ma-n-urrun %maN-n-urrun% 3sgM.O-3sgM.A-make.ﬁrenp ‘He’s burning oﬀ.’

e. ma-ng-urrun 3sgM.O-3sgF.A-make.ﬁrenp ‘She’s burning oﬀ.’

f. mam-b-urrun 3sgmapl.A-make.ﬁrenp ‘They’re burning oﬀ.’

With so-called ang-verbs, the preﬁx forms in Iw can all be derived by postulating a third person object marker with form ang- ~ ung- (the second appearing after

iwaidja mutation and its origins

139

consonants, and the ﬁrst initially or after vowels) in place of the regular third person singular object marker uK-, and again extending the third person plural subject forms to replace the third person singular forms. The ﬁnal ng- then assimilates in place of articulation to following apicals, as shown in the examples in (76). (76) a. an-dijbun %a-ng-ldijbun% 1sgA-3sg.ang.O-collect.water ‘I collect water.’

b. arrun-dijbun %arr-ung-ldijbun% 1plA-3sg.ang.O-collect.water ‘We exc. collect water.’

c. kurrun-dijbun %kurr-ung-ldijbun% 2plA-3sg.ang.O-collect.water ‘You pl. collect water.’

d. ang-bu-ldijbun 3sgANG.O-3plA-collect.water ‘He/she/they collect water.’

With ang- verbs in Ilg and Ga,31 ang- ~ ung- occupies the regular third person singular object slot (i.e. after ﬁrst and second person subject preﬁxes, and before third person subject preﬁxes), then undergoes regular morphophonemic reduction to to a before nga- and ni, resulting in the third person forms ani- (3 sg. M. subject), anga- (3 sg fem subject) and angbu- (3 pl subject). (77) Ilgar-Ang class a. ang-burrwun %a-ang-burrwun% 1sgA-3ang.O-know ‘I know.’

b. arr-ung-burrwun 1pl.inc.A-ang.O-know ‘We know.’

c. kurr-ung-burrwun 2.pl.A-ang.O-know ‘You (pl.) know.’

d. a-ni-wurrwun %ang-ni-wurrwun% 3sgANG.O-3sgM.A-know ‘He knows .’

e. a-nga-wurrwun %ang-nga-wurrwun% 3sgang.O-3sgF.A-know ‘She knows.’

f. anggu-wurrwun %ang-bu-wurrwun 3sgang.O-3plA-know ‘They know.’

Though the Ilg forms show the etymological relation more clearly than those in Iw, owing to the Iw extension of third person plural forms into the singular, it is clear in both languages that the ang- preﬁx sets continue the neuter object forms represented by the forms ang- ~ ung- in M.32 The mam- and ang- preﬁxes thus represent, as signiﬁants, the direct continuation of the proto-Iwaidjan vegetable and neuter preﬁxes. This was pointed out by Pym and Larrimore (1979: 88–9):

140

nicholas evans

Here, in the Iw ang- and mam- verbs, are to be found the only resemblances to the noun classifying system of the related language, M. There is a resemblance in form between Iw ang- verbs and M verbs with a class 4 object. In M, class 4 nouns are ‘‘ground objects’’ connected with ‘‘ground’’ and abstract ideas. There is a marked correlation between the general meaning of this class and Iw ang- verbs.

There is also a resemblance in form between Iw mam- verbs and M verbs with a class 5 object. In terms of their combinatorics, as well, they form part of the same paradigmatic series as the third person object preﬁxes, and in Ilg more precisely of the masculine and feminine object preﬁxes. In both Ilg and Iw, mam- and ang- are positioned wherever third person singular object preﬁxes would be positioned in normal verbs. In terms of their signiﬁés, however, they no longer function as gender markers in either Ilg or Iw: they do not agree with the grammatical gender, or for that matter any aspect of the semantics, of the object noun phrase they cross-reference.33 At best, they occur with verbs whose typical objects belong to the vegetable and neuter genders in M and presumably did so in proto-Iwaidjic. Pym and Larrimore suggest that this semantic connection holds better for mam- verbs than for ang-verbs: In M, class 434 [i.e. neuter—N.E.] nouns are ‘‘ground objects’’ connected with ‘‘ground’’ and abstract ideas. There is a marked correlation between the general meaning of this class and Iw ang- verbs . . . . Class 5 nouns in M are trees and objects connected with wood, but there does not seem to be any semantic correlation with Iw mam- verbs . . . .35 Some of the verbs which take ang- have meanings directly associated with the ground: ‘take a short cut,’ ‘collect water,’ ‘sweep,’ ‘live on land,’ ‘dig,’ ‘make cement,’ ‘walk,’ ‘go hunting’; while others have no such association: ‘make a noise,’ ‘think,’ ‘tell a story,’ ‘stir,’ ‘want,’ ‘eat,’ ‘drink,’ ‘breathe.’

The most likely sequence of events is the following three-step process. Initially, the ang- and ma- preﬁxes formed part of the regular gender system, as in contemporary M. In the second phase, the gender preﬁxes became lexicalized with some verbs, so that given verbs, or senses of verbs, always selected ang- or mam- preﬁxes, regardless of their actual object. In some but not all cases this was based on the semantics of their prototypical objects, i.e. it represented an extension of gender agreement with their prototypical object to what we might call ‘cognate object gender’, where cognate objects are arguments which extend, or make explicit the activity denoted by the verb, and typically belong to a small or one-member set owing to the predictability of the object from the activity, e.g. sing a song (Austin 1982). While the use of cognate object genders (as opposed to their subsequent

iwaidja mutation and its origins

141

lexicalization) is not directly attested for any Iwaidjan language, it has been clearly described for Gaagudju (Harvey 1992), not closely related but spoken just south of the Iwaidjan family. Gaagudju has the standard four genders (masculine, feminine, vegetable and neuter) found in Australian languages, and cross-references them on the verb through subject and object agreement; an example of standard agreement with a vegetable class object is (78): (78) djaamu ma-’rree-yaba=yu tucker 3ve.abs-1.erg-sendpp=3io ‘I sent the tucker to her.’ (Harvey 1992: 336) In addition to such standard uses, Harvey describes a number of cases where the use of object pronominal preﬁxes involve lexicalized choices of the vegetable and neuter preﬁxes, with varying degrees of possible linkage to actual nouns with which agreement can be postulated: in a case like (79) the vegetable class preﬁx could be agreeing with the ellipsed vegetable-class noun maba’laabala ‘corroboree, song’, and in (80) with the ellipsed neuter-class noun dja’gaardu ‘language, noise, word’. (79) ma-nga-n-ba’rlaa-bu 3ve.abs-3merg-fut-sing-aux ‘He will sing.’ (Harvey 1992: 338) (80) gu-nga-n-ba’laa-bu 3nt.abs-3merg-fut-talk-aux ‘He will talk’. (Harvey 1992: 338) But in other cases there is no clear-cut noun responsible for triggering agreement; an example is (81), in which the choice of neuter object preﬁx with the verb ga’lamarrwa ‘to be jealous’ cannot be linked to any clear referent, and ‘the Class IV [= neuter - N.E.] preﬁxing found with ga’lamarr-wa must simply be analysed as lexicalised’ (Harvey 1992:340). Harvey also discusses a few cases in which subject preﬁxes are lexicalised in similar ways. (81) gu-nga-n-ga’leemarr-wa=nu 3nt.abs-3m.erg-fut-jealous-aux=3m.io ‘He will be jealous of him.’ The signiﬁcance of the Gaagudju data is its clear demonstration of a parallel to noun incorporation whereby verbs can incorporate, instead of nouns, gender preﬁxes, with their much higher level of semantic abstraction. It is likely that pre-Arrkbi had a similar system, and that during this phase the mam- and ang- preﬁxes would have

142

nicholas evans

been structurally ambiguous, as they are in Gaagudju: they could function either as regular gender markers, agreeing in gender with the verb’s object, or they could be subcategorized by the verb and fail to agree with the verb’s object. In the ﬁnal phase, the mam- and ang- preﬁxes become delinked from the system of grammatical gender, even though in Ilg they continue to occupy the same morphological slot as the surviving masculine and feminine gender markers. At this point they depend entirely on the verb, and are best treated, from the point of view of what selects their form, as conjugationally-determined preﬁxal variants limited to particular verbs. In both Iw and Ilg/Ga the loss of vegetable and neuter genders outside the object preﬁx system would have left the object preﬁx forms isolated and unmotivated. As a result, a process of morphological reanalysis has set in, leading to the replacement of ang- forms by -ny forms originating in the feminine gender in both languages; presumably this occurred in Iw before the loss of the feminine gender from the rest of the paradigm.

4. Conclusions We have seen that the development of initial mutation in Iw was closely tied up with the loss of the ancestral gender system, as a result of the fact that the ‘triumphant’ miscellaneous gender was the only one of the ancestral gender preﬁx set that triggered certain changes in root-initials - basically the hardening of root-initial nasals and approximants, and the conversion of initial ﬂapped laterals to the retroﬂex approximant. Traces of the ancestral gender system also survive as ‘preﬁxal conjugations’ with a small set of verbs, in third person singular transitive subjects, and as non-productive preﬁxes in nouns. Though the formal matching of Iw mutated forms to their proto-Iwaidjic miscellaneous-gender precursors is clear, accounting for what steps and motivations led to this outcome is more problematic. To recapitulate the steps that occurred between proto-Iwaidjic and Iw, the ﬁrst was almost certainly the extension of the miscellaneous gender preﬁx to the exclusion of the other genders on body part nouns. This is the only part of the grammar in which a plausible reason can be found for extending such a highly-marked gender in this way - namely, to resolve the conﬂict between two principles for assigning gender to body-parts: through marking the person/number/gender features of their possessor (inherited gender), or through marking their intrinsic gender. Presumably both options co-existed in proto-Iwaidjic (as in eastern Kunwinjku), with the ‘inherited gender’ pattern winning out in M and Ilg/Ga, and the ‘intrinsic gender’ pattern winning out in pre-Iwaidja, leading to the use of the miscellaneous gender preﬁx aK- on all body part nouns. To account for the fact that all the Arrkbi languages severely curtailed their gender systems, though in diﬀerent ways, we need to postu-

iwaidja mutation and its origins

143

late a substantial restriction on the gender systems, including the loss of the vegetable and neuter systems, by the time of proto-Arrkbi. The next problem is to account for the extension of the miscellaneous preﬁx to other environments, such as verbal preﬁxes, adjectives, inﬂected prepositions and pronouns. The diﬃculty here is the highly marked nature of the miscellaneous gender in the one Iwaidjic language that attests it, M. As mentioned in Section 3.2.1, a text count of 251 tokens of gender preﬁxes in M texts failed to include a single example of the miscellaneous preﬁx, making it impossible to invoke the extension of an unmarked form as the reason for the extension of the miscellaneous preﬁx. It is also highly unusual cross-linguistically for neuter forms of pronouns to replace human-gender forms, but this appears to have happened in Iw: the third singular pronoun janad ‘he, she, it’, which has replaced the rival forms y-anad ‘he’ and inyanad ‘she’ found in M and Ilg, derives from the miscellaneous form ad-janad ‘it (miscellaneous)’. Appeals to the extension of an unmarked gender being inappropriate, a more likely explanation is an extreme case of what Alpher and Nash (1984) have called ‘correspondence mimicry’.36 In speech communities such as those of the Coburg area37 and Australia more generally, where multilingualism is all-pervasive (see Brandl and Walsh 1982), it is common for speakers to be aware of correspondence patterns between their own language and its neighbours, and to use this awareness to extend such patterns analogically through the vocabulary. Sporadic examples of this process have been reported by Alpher and Nash (1984), Nash (1997) and Koch (1997) for languages elsewhere on the Australian continent, but it appears to have taken a much more extreme form in Iw. I am proposing, in other words, that at some point in the history of Iw, its speakers took the correspondence that existed between the miscellaneous-preﬁxed body part nouns in their language and the possessor-gender forms in neighbouring Ilg and M, and generalized this to other environments, notably verbs, pronouns and prepositions. A ﬁnal phonological change in Iw would have been the loss of the initial a-vowel, except before monosyllabic roots, and the blending of K- with the following segment led to the pattern of mutation. Note that since the two monosyllabic roots retaining initial a-, namely aju ‘(s)he lies’ and ari ‘(s)he stands’ (vs a-ldi ‘they stand’) are both verbs, we must assume that the conditioned loss of initial a- followed the generalization of the miscellaneous as the sole gender from nouns to verbs, otherwise there would be no basis to explain the survival of the a-preﬁx in these verbs. While the above explanation accounts for all the formal morphological and morphophonemic correspondences between Iw and its relatives, it has the disadvantage of placing at centre stage a process—correspondence mimicry—that has hitherto only been reported as operating marginally, on small numbers of lexemes. The

144

nicholas evans

series of changes reported in this paper can only be regarded as clearly established if we can go on to widen our attestation of wholesale correspondence mimicry as a major process in historical linguistics.

Notes 1. It is a great pleasure to pay tribute, with this paper, to Barry Blake’s many contributions to Australian linguistics, which have done so much to form a coherent picture out of the vast range of complex morphological data Australian languages oﬀer. In particular, his beautiful work on the use of intragenetic typology in Australia, and on large-sample comparisons of morphological forms, have enormously advanced our understanding of the typological and genetic classiﬁcations of Australian languages. 2. Earlier versions of this paper were delivered at the Melbourne University Australian languages working group, and the Second compaust workshop at Stanwell Tops. I am grateful to the participants at those seminars, in particular Ilia Pejros, Bill McGregor, Mark Harvey and Nick Reid, as well as to Jae Jung Song and Anna Siewierska, for their comments. My understanding of Amurdak and Garig has beneﬁtted greatly from discussions with Robert Handelsmann and Oscar Whitehead respectively. Research on the Iwaidjan languages was supported by three Grants from the Australian Research Council: ‘Non-Pama-Nyungan Languages of Northern Australia’, ‘Polysemy and Semantic Change in Australian Languages’ and ’Describing Australian Aboriginal Languages’, and the ﬁnal version of this paper was revised while on a Humboldt Fellowship; I gratefully acknowledge the support of these bodies. My most important debt is to Charlie Wardaga and the late Mick Yarmirr for teaching me something about Ilgar and Marrgu. 3. Initial data suggests that some Marrgu verbs also have mutation – cf girri-lawudhi ‘I talk’, nyi-lawudhi ‘you talk’, lawudhi ‘he talks’ but dhawudhi ‘you and I talk’. But our documentation and analysis of Marrgu is still at too basic a level to comment further at this stage. 4. Apart from a diﬀerent intonation and voice quality, which Charlie Wardaga gives as the most salient diﬀerence between these two dialects, the major diﬀerences between Ilgar and Garig are: (a) the verb ‘eat’ and its subcategorized object preﬁx (see 3.2.2) (b) diﬀerent forms for the third person singular masculine and feminine ‘towards’ preﬁxes to the verb: cf Ilgar ani- ‘he.non.future.towards’, Garig a- ‘id.’; Ilgar anga- ‘she.non.future.towards’, Garig any- ‘id.’; formally both include a ‘towards’ preﬁx a-, but combined with the intransitive subject preﬁx in Garig (masculine i- and feminine iny-, but with the vowel lost through vowel elision) and with what is normally the transitive subject preﬁx in Ilgar (masculine ni- and feminine nga-) (c) distinct free pronoun forms for ‘he’: Ilgar anad but Garig yanad (d) a handful of diﬀerent lexical forms. 5. Based on the Iwaidja practical orthography developed by Pym and Larrimore (1979). 6. The exact phonetic and phonological status of this contrast is still unclear, and Handelsmann (1991) has suggested that they be analysed as prestopped laterals in Amurdak. 7. Note though that some of these initials are not original but represent lenited forms of stops attested in cognates in neighbouring languages. Examples are wula (< older bula, the form found in Mayali and Jawoyn), wirrhala (< older birrkala, as in Mayali birrkala ‘throwing stick’), and yanyjuk (< older

iwaidja mutation and its origins

145

janyjuk or jenyjok, as in Mayali djenjdjok ‘milkwood tree’) – see Evans 1997 for the relevant cognate sets. For reasons of space I conﬁne myself to giving at most 2 cognate sets per correspondence. 8.

Note also Iw baldak ‘small string bag, biting bag’, Mayali malakka ‘small ‘ﬁghting dillybag’, bitten by men during moments of anger’. This is most likely a loan from Mayali that predates the operation of mutation in Iwaidja.

19. Bear in mind that little is known about the other non-Gunwinyguan Australian languages originally spoken to the south of the Iwaidjan family, such as Mengerr, Erre, Ngarduk and Umbugarla, so the percentage of identiﬁed Australian loans is likely to be a serious underestimate. 10. Analysable into object preﬁx aK- (see below on the morphophoneme K) and feminine subject preﬁx nga-. 11. First and second person preﬁxes precede third person preﬁxes, regardless of their grammatical role. 12. -ad is a plural suﬃx found on some nouns and adjectives. 13. The presence of deverbal nouns like buhaldak ‘greedy person’ based on the the 3sg verbal form (cf buhaldak ‘(s)he is greedy’, a-wugaldak ‘they are greedy’) suggests their plurals would be based on the 3 pl form, and hence be subject to mutational alternations. More work is needed on Iwaidja to conﬁrm this. 14. This is Pym and Larrimore’s gloss; the exact semantic status of the object suggested by the 1/3sg preﬁx form (should it be ‘I count it’?) is unclear at this stage. 15. Their discussion of this data is not very satisfactory. It is possible, for example, that these have an initial n- (otherwise lacking from our data), so that ardayan is really aK-rnayan. This is plausible given the widespread cognates of see in na- or rna- (e.g. Mayali and Dalabon na- ‘see’). Should this not be the case, it seems wrong to derive the form from aK- since the morphophoneme itself lacks the information to predict between apical and velar forms; a better analysis would be to have a variant set of preﬁxes with a diﬀerent morphophoneme, call it RD-, thus 1/3sg aRD- and 3sg RD-. We shall see below that some verbs select other anomalous preﬁxes, namely ang- and mam-, so this solution would be consistent with the general existence of conjugationally selected preﬁx variants. Without further data no more can be said about this problem. 16. E.g. the following passage, which takes the mutated form as underlying for the ﬁrst verb, and the unmutated form as underlying for the second: ‘In verbal forms used as subject preﬁxes: root /-bi-/, ‘say’ > /a-mi-n/, ‘I say’, /a-mi-n/, ‘you say’; /-wani-/, ‘sit’ > /a-wani /, ‘I sit’, but /a-bani /, ‘you sit’, are also irregular.’ (Capell 1962:134). Capell’s formulation overlooks the fact that the diﬀerences in form stem from the ﬁrst verb having underlying root-initial m, and the second underlying root-initial w. 17. Its treatment as a ‘sixth noun class’ comes from Capell and Hinch’s inclusion of ‘plural’ with the noun classes (as their class III). Their terminology matches up with the contemporary numbering system for Arnhem Land languages in the following way: I = I (masculine), II = II (feminine), III = plural, IV = IV (neuter), V = III (vegetable), and VI = V (miscellaneous). 18. The initial n- here is because this adjective belongs to a small class taking n(u)- as 3sg preﬁx

146

nicholas evans

throughout; the implied breakdown is n(u)+ NCl- (Capell and Hinch 1970:56). 19. In fact, there is one environment in which the initial a of aK- is preserved in Iwaidja: in the two known intransitive verb roots with monosyllabic present tense forms, the third person preﬁx is aKrather than the K- found elsewhere: aK-yu → aju ‘(s)he lies’ aK-ldi → ari ‘(s)he stands’ —It seems likely that destressing and loss of the initial vowel was blocked by a minimum disyllabic template (with stress on the ﬁrst syllable) for major word classes, leading to the survival of the entire aK- preﬁx in these two cases. Owing to the lack of monosyllabic body-part roots, there are no examples of the aK- preﬁxing surving with nouns. —Pym and Larrimore do not discuss this exception in their 1979 book, but include the relevant data in their unpublished dictionary. Both commonly occur in place names as part of the formula ‘X lies/ stands there’, where X designates some mythical ﬁgure preserved as a ‘dreaming site’. 20. These examples, taken from ﬁeldwork by Gavan Breen with Butcher Knight, are drawn from Davies (1989). 21. Capell and Hinch do not explain why this governs vegetable agreement in the demonstrative, but neuter in the object preﬁx on the verb. 22. The discrepancy in the root form of the verb, with respect to (66), is in Capell and Hinch. 23. An alternative they do not consider is to derive ri- from K-ni; ni- is the full allomorph of the 3sg masculine subject forms. This at least accounts for the alveolar articulation in ri-. Such a proposal would need to be attested against other instances of K before n, but so far I have yet to ﬁnd any. 24. In Maung the 3sg fem subject form is also used for all other genders except the masculine. 25. Here and elsewhere for the Maung forms, the presence of the k- marks present, and its absence the past, e.g. present kini-, past ini-. Where the choice is irregular both forms are shown. 26. The masculine/feminine object distinction in Iw is only made with 3sg subjects, so the cells with other subjects are not shown here; the Iwaidja 3sg object forms for other subject values are given in the second last row, together with the three miscellaneous forms from Maung. 27. The choice between ri- and i-, and ka- and nga-, depends on the whether the K- object preﬁx surfaces immediately before the preﬁx or earlier; Pym and Larrimore analyse ri- < K-i- and ka- < K-nga. The second derivation is morphophonemically regular, but the ﬁrst is otherwise unattested. Examples of situations where the 3sg object preﬁx does not immediately precede the 3sg subject preﬁx are k-ana-nga-wunya [3obj/3fem.subj-imp-she-cook] ‘let her cook it’, and n-an-i-wu [3ss-imp-he-hit] ‘let him hit it!’. See Pym and Larrimore (1979:126-7) for further examples and discussion. 28. It also appears in some of the future forms, e.g. a-ny-ban-irimanbun [1sgA-3sgmam.obj-fut-smoke] I will smoke it’. 29. Here and elsewhere, the vowel of the preﬁx is lost by vowel elision before a following root-initial vowel.

iwaidja mutation and its origins

147

30. One of the clearest of the few diﬀerences between Ilgar and Garig is with the verb ‘drink’: in Garig this is the ang verb ang-. . .-lda, while in Ilgar this meaning is expressed by the root -arrun plus a subcategorized feminine pronoun form: cf Garig anilda [< ang-ni-lda ] ‘he drinks’, Ilgar inyarrun [< iny-i-arrun] ‘he drinks’. 31. The forms ang- and ung- form part of a wider set of neuter allomorphs in Maung: ang- on intransitive verbs, and variously u-, wu- and nung- on adjectives (Capell and Hinch 1970:56). The adjectival preﬁx forms link the Maung neuter preﬁxes to widespread non-Pama-Nyungan forms in ku(Heath 1987) and somewhat less widespread forms in nung- (Evans in prep.). 32. The sole exception to this, for Iwaidja, is reported by Pym and Larrimore (1979:89): One instance has been noted of ang- being attached to the verb -ayan ‘to see’ (a regular transitive) when it was used of looking at the ground. Men in a circling plane were described as ang-b-ayan ground-3pl-see ‘they’re looking at the ground’ Synchronically, this is best accommodated by having a derived verb ang- -ayan ‘to look at the ground’, derived from -ayan ‘to see, look at’ by switching preﬁxal conjugation. However, etymologically it represents a frozen use of object gender agreement. 33. Pym and Larrimore follow the Capell system for numbering noun classes, which includes plurals in the series:1: masculine, 2: feminine, 3: plural, 4: neuter, 5: vegetable, 6: miscellaneous. 34. The six verbs taking the mam- preﬁx set in Iwaidja are: amarrung ‘to eat meat,’ awungmiri ‘to keep away,’ awurrurdban ‘to lead on a rope,’ madbalang ‘to open,’ madbunggu ‘to open,’ and mirrhuran ‘to close.’ In Ilgar the verbs attested with ma- preﬁxes are -arrun ‘eat’, -irimanbun ‘smoke’ and -urrun ‘burn’, all involving consumption of the object, and in the second and third cases involving objects that would belong to the vegetable class in Maung. Research on Ilgar is at an early stage, however, and future work is likely to uncover a larger set of verbs in this class. 35. Recalling Baudouin de Courtenay’s (1885) earlier term ‘correspondence simultaneity’. 36. My two main language teachers are each highly multilingual: Mick Yarmirr spoke Maung, Iwaidja, English , Indonesian and some Kunwinjku and Kunbarlang as well as his ‘own’ language Marrgu, which he spoke less ﬂuently than the others as a result of being removed from his home to a mission on Goulburn Island at the age of seven, while Charlie Wardaga speaks Ilgar/Garig, Maung, Manangkari, Iwaidja, Kunwinjku and English. Another senior man from the region, Nelson Mulurinj, speaks Amurdak, Garig, Maung, Iwaidja, Kunwinjku, English, Indonesian and (reportedly) Umbugarla.

References Abbreviations: aseda: Aboriginal Studies Electronic Data Archive. sil-aab: Summer Institute of Linguistics: Australian Aborigines Branch.

148

nicholas evans

Alpher, Barry and David Nash. 1984. ‘‘Lexical Replacement and Cognate Equilibrium in Australia.’’ In precirculated papers for Australian Linguistics Society Conference, Alice Springs, August 1984, 17–44. Austin, Peter. 1982. ‘‘Transitivity and Cognate Objects in Australian Languages.’’ In Paul J. Hopper and Sandra A. Thompson (eds), Studies in Transitivity. New York: Academic Press, 37–48. [Syntax and Semantics, vol. 15] Berndt, R.M. and Berndt, C.H. 1970. Man, Land and Myth in North Australia. Sydney: Ure Smith. Brandl, M.M. and M. Walsh. 1982. ‘‘Speakers of Many Tongues: Toward understanding multilingualism among Aboriginal Australians.’’ International Journal of the Sociology of Language 36: 71–81. Capell, Arthur. 1962. Some Linguistic Types in Australia. Sydney: Oceania Linguistic Monographs, No. 7. ——. 1963. Marrgu ﬁeld notes and word list. —— and Heather E. Hinch. 1970. Maung grammar, Texts and Vocabulary. The Hague: Mouton. de Courtenay, Baudouin. 1885. [1972]. ‘‘An Attempt at a Theory of Phonetic Alternations: A chapter from psychophonetics.’’ Chapter VI, 187–193: ‘‘Foreign alternations, i.e. alternations which are due to the inﬂuence of another language.’’ Part VII, 144–212. In A Baudouin de Courtenay Anthology. The beginnings of Structural Linguistics, ed. and transl. by Edward Stankiewicz. Bloomington, London: Indiana UP. Davies, Jennifer. 1989. ‘‘Umbugarla: A Sketch Grammar.’’ Unpublished Honours Thesis, University of Melbourne. Earl, G.W. 1846. ‘‘On the Aboriginal Tribes of the Northern Coast of Australia.’’ Journal of the Royal Geographical Society of London 16: 239–51. ——. 1853. The Native Races of the Indian Archipelago. London. Evans, Nicholas. 1992. ‘‘Macassan Loanwords in Top End Languages.’’ Australian Journal of Linguistics 12: 45–91. ——. 1991–4. Marrgu Field Notes, Recorded from Mick Yarmirr. ——. 1994–6. Ilgar Field Notes, Recorded from Charlie Wardaga. ——. 1994. ‘‘The Problem of Body Parts and Noun Class Membership in Australian Languages.’’ University of Melbourne Working Papers in Linguistics 14: 1–8. ——. 1995. Current Issues in Australian Phonology. In John Goldsmith, ed., Handbook of Phonological Theory. Oxford: Blackwells, 723–61. ——. 1996. ‘‘First—and last—Notes on Wurrugu.’’ University of Melbourne Working Papers in Linguistics 16: 91–8. ——. 1997. ‘‘Macassan Loans and Linguistic Stratigraphy in Western Arnhem Land.’’ In McConvell and Evans, eds. ——. (in prep.). ‘‘Paradigm Regained: Towards the reconstruction of case: gender portmanteaux in northern Australian languages.’’ Handelsmann, Robert. 1991. ‘‘Towards a Description of Amurdak: A language of Northern Australia.’’ Unpublished Honours Thesis, University of Melbourne.

iwaidja mutation and its origins

149

Harvey, Mark. 1992. ‘‘The Gaagudju People and Their Language.’’ Unpublished Ph.D. Thesis, University of Sydney. Heath, Jeﬀrey. 1987. ‘‘Story of *-n-: *CV- vs *CV-n- Noun-Class Preﬁxes in Australian languages.’’ In Donald C. Laycock and Werner Winter (eds), A world of language: papers presented to Professor S.A. Wurm on his 65th birthday. Paciﬁc Linguistics C-100, 233– 243. Hinch, H. 1966. ‘‘Marrgu ﬁeld notes.’’ Unpublished manuscript. Koch, Harold. 1997. ‘‘Comparative Linguistics and Australian Prehistory.’’ In McConvell and Evans (eds). McConvell, Patrick and Nicholas Evans (eds). 1997. Archaeology and Linguistics: Global perspectives on Ancient Australia. Melbourne: Oxford University Press. Nash, David. 1997. Comparative Flora Terminology of the Northern Territory. In McConvell and Evans (eds), 187–206. Pym, Noreen. 1977. ‘‘Garig Word List for two Month Survey.’’ SIL-AAB: Unpublished MS. ——. (with Bonnie Larrimore). 1979. Papers on Iwaidja phonology and grammar. Darwin: SIL-AAB. Series A, Volume 2. Pym, Noreen. (n.d.). Iwaidja word-list. Data ﬁle deposited in ASEDA archive. Whitehead, Oscar. 1991. Garig ﬁeld notes, recorded from Nelson Mulurinj.

Functional control with and without structure-sharing1 RICHARD HUDSON University College London How should the syntactic structure around a transitive verb such as persuade or force be analysed? Take an example like (1). (1) We persuaded Pat to come. How should the second NP Pat be related syntactically to the inﬁnitive to come? Current opinion is split between two answers: that Pat is the subject of to come, and that it is not. These views are associated with diﬀerent general theories of syntax, and important matters of general theory are certainly at stake, but I shall argue that it is ultimately an empirical matter. We can distinguish two diﬀerent families of analysis, which we can call respectively the ‘non-sharing’ analyses and the ‘structure-sharing’ analyses; and I shall argue, on the basis of data from Icelandic and Ancient Greek, that both these approaches may be correct for English. According to non-sharing analyses, Pat is not itself the subject of to come, whose syntactic subject is generally assumed to be some kind of null pronoun. In Government and Binding theory (GB) this is PRO, while in Head-Driven Phrase Structure Grammar (HPSG) it is unnamed (Pollard and Sag 1994: 135). This null pronoun mediates the link between Pat and to come by referring anaphorically to Pat, which is a semantic relationship; in terms of syntax they are simply two separate NPs. This is like the ‘anaphoric control’ of Bresnan (1982: 326–43) illustrated with examples like (2): (2) Tom felt sheepish. Pinching those elephants was foolish. He shouldn’t have done it. In the analysis of sentences like the second of these, though not of our main example (1), Bresnan recognises no subject at all in the subordinate clause’s c-structure, though the more semantically-oriented f-structure gives it a pronoun-like subject ‘PRO’ which can be linked by a combination of pragmatic and grammatical rules to its antecedent.

152

richard hudson

In contrast, the structure-sharing analyses of (1) allow the noun-phrase Pat to be ‘shared’ by both the verbs—i.e. to act not only as the object of persuaded, but also as the syntactic subject of to come. In Relational Grammar (RG) it is shown by ‘igloo’ diagrams in which one NP node is linked by arcs to two diﬀerent clause nodes (Blake 1990: 90). In Lexical-Functional Grammar (LFG) the f-structure slots for the object of . . . persuade . . . and the subject of to come are linked by an arc to show that they share the same ﬁller (Bresnan 1982: 322). We shall follow Bresnan (1982: 321) in calling verbs like persuade ‘functional control’ verbs, in contrast with what she calls (ibid: 374) ‘raising’ verbs such as expect, as in (3): (3) We expected Pat to come. What the structure-sharing analyses have in common is that they give the same syntactic structure to functional-control verbs as they do to raising verbs. The term ‘structure-sharing’ is borrowed from the HPSG analysis of such ‘raised’ structures as (3) (Pollard and Sag 1994;136), where the attribute-slots for the object of expected and the subject of to come both have the NP Pat as their values. An alternative way to show the same relationships is to introduce a special kind of syntactic relationship between Pat and expected, ‘exceptional case-marking’, as in GB (Chomsky 1986: 190). For present purposes I shall count this as a kind of ‘structuresharing’; indeed, the exceptional case-marking machinery is quite similar to the explicit structure-sharing of Relational Grammar, LFG and HPSG, to the extent that case-marking can be considered a syntactic relationship. At one level, then, the question is simply a matter of how to divide the territory between two structures, both of which are independently needed and uncontroversial: between structure-sharing and anaphoric control. However the choice also has important theoretical implications because the shared NP has semantic relationships to both verbs: Pat is the ‘persuade-ee’ as well as the ‘come-er’. The non-sharing analysis provides a separate NP for each of these semantic roles—Pat as persuade-ee, PRO as come-er—which invites a broad generalisation: no single NP can have more than one semantic role (the ‘theta-criterion’2 of Chomsky 1981: 36). This generalisation is compatible with structure-sharing in the raising constructions after expect, because the raised NP has no semantic link to expect: Pat is the come-er, but not the ‘expect-ee’. Functional control, however, is a real challenge for the theta-criterion if it really does involve structure-sharing, because the NP which belongs to two syntactic constructions also has two diﬀerent semantic roles. Which of these two analyses is correct? The answer is not at all obvious, and I myself have oscillated between them over the years. In Hudson (1971: 180) I said that expect and persuade had diﬀerent syntactic structures (the non-sharing analysis), and I repeated the same suggestion in Hudson (1976: 124, 130). By Hudson (1984: 113), however, I had changed my mind in favour of the structure-sharing

functional control

153

analysis, in which expect and persuade have the same syntactic structure and diﬀer only in the semantics; and I still thought the same in Hudson (1990: 237). However I have to admit that I had private doubts about the evidence for the later analysis. The trouble was that both analyses were permitted by the theoretical framework I assumed (Word Grammar, WG), and there was very little to choose between them. The structure-sharing analysis was easy, because the syntactic machinery was already in place for expect, and persuade was even easier to map onto semantic structure because each dependency in the syntax corresponded to one in the semantics. On the other hand, the non-sharing analysis was easy too because it exploited the machinery for anaphoric control. The grammar would have to allow non-ﬁnite verbs to be used without any subject at all (e.g. To err is human), and would have to provide fairly powerful rules or principles for reconstructing these missing subjects which would surely be powerful enough to link Pat to to come after persuaded. Either analysis would do equally well. The evidence in this paper suggests that this conclusion may have been right; there really is no evidence either way in English, and both analyses must be allowed by general syntactic theory. This leaves us in the happy position where both analyses are right, though the theta-criterion (and its successors) must be wrong, since it is incompatible with the structure-sharing analysis. There are some constructions in other languages which unequivocally demand the structure-sharing analysis, and there are others which equally clearly demand the non-sharing one. The evidence turns crucially on the existence of inﬂectional case and case-agreement rules in these languages; there is certainly no case-agreement in English, and I have argued elsewhere that there is no inﬂectional case (Hudson 1995b), so we cannot run the same arguments for English. The conclusion is that general syntactic theory must allow both kinds of analysis, which means that both are available for English sentences like (1) so such sentences are syntactically ambiguous—they allow two distinct structures which both have the same meaning. To summarise, the question is whether persuade is more like expect or more like, say, meet in examples like the following: (4) I expected Pat to buy a book. (5) I persuaded Pat to buy a book. (6) I met Pat buying a book. In each case the non-ﬁnite form of buy has Pat as its ‘understood subject’ (i.e. the buy-er is Pat); but is the word (or noun-phrase) Pat structure-shared with persuaded, as after expected, or simply missing, as after met? Figure 1 shows the model syntactic structures with expected and met in WG notation, in which arrows point towards syntactic dependents and labels indicate grammatical relations (the WG

154

richard hudson

‘sharer’ relation is the same as the LFG ‘xcomp’). The arrows are arranged either above or below the words so as to provide a coherent tangle-free structure above the words; this is explained in Hudson 1994. The vertical arrow points at the one word which does not depend on any other, the ‘root’ of the whole sentence.

Structure=sharing

Anaphoric control

s

r r o c o I expected Pat to buy a book. s s

s

a o c o I met Pat buying a book. Key: s subject; o object; c complement a adjunct; r sharer

Figure 1. Figure 2 shows the two alternative analyses with persuaded. The only diﬀerence is the presence or absence of the two extra subject-arrows below the words. When they are present they represent a structure-sharing analysis in which Pat is subject of to buy as well as object of persuaded. When absent, to buy has no subject, which is the WG equivalent of the anaphoric-control analyses which invoke PRO.

Functional Control =Structure-sharing s

r r o c o I persuaded Pat to buy a book. s s

=Anaphoric control s

? r o c o I persuaded Pat to buy a book.

Figure 2.

2. Case as evidence for subject-sharing Although it is hard to ﬁnd evidence in favour of either of these analyses in English, other languages are more helpful. What is particularly helpful is a rich system of inﬂectional case, and the evidence that follows will all be taken from languages of this type—Icelandic and Ancient Greek. Case has often been used in this way as

functional control

155

evidence for ‘sharing’ analyses; all that this paper will do is to extend the discussion to functional control structures like the ones found with persuade, and to draw some general theoretical conclusions. For instance, take the evidence for ‘raising’ analyses of some verbs in Icelandic surveyed in Andrews (1982: 445), where subscripted N and A stand for nominative and accusative case. (7) a. HúnN er vinsælN. ‘She is popular.’ b. Þeir segja hanaA (vera) vinsælaA. they say her to-be popular ‘They say she is popular.’ Clearly the case of vinsæl varies with the case of the pronoun. It is hard to dispute the conclusion that Andrews draws, namely that the pronoun is the adjective’s subject (as well as subject of the copula) which is separated from it by the application of raising (or its theoretical equivalent). Given this analysis the rule for the case (and other features) of a predicative adjective in Icelandic is very simple: it must agree with its subject. However, the consequence of this conclusion is that the pronoun must be the subject of more than one word. In addition to being the subject of vinsæl, we can be sure that it is the subject of er in (7a) because it is in the subject position, it is in the expected nominative case, etc. Similarly in (7b): hana must be the subject of vera as well as of vinsæla. But hana has accusative case either because it is the object of segja, or (in the GB analysis) because it is exceptionally case-marked by segja—a kind of exceptional object relationship. Either way, hana is simultaneously (exceptional) object of segja and subject of vera—pure structuresharing. The WG analyses for these two examples are shown in Figure 3 where it

s

r

Hún er vinsæl/ N. s

s

r r o Þeir segja hana /A vera vinsæla /A. s s Figure 3.

156

richard hudson

can be seen that the shared pronoun is the subject of the adjective as well as of the inﬁnitive verb. The main point of this argument is that it rests crucially on the assumption that case, unlike categories like person, number and gender, is purely syntactic; that is, it is a characteristic of words (or phrases), but not of meanings, so it can only be carried by words. The word hana itself can easily receive case (by government) from segja, and then transmit it (by agreement) to vera and vinsæla. But suppose the subject of vera had been PRO, not hana. In the syntactic structure PRO and hana are two distinct ‘words’, and although they are related semantically by anaphoric control, there is no reason to expect them to have the same case. In typical anaphoric control, the case of the antecedent is quite irrelevant to the PRO, so even if PRO does have a case, it is not tied to the case of its antecedent so PRO cannot transmit its antecedent’s case to any other part of the sentence. Reversing the argument, then, any NP that does transmit case from a higher verb to a lower one must be linked to them both by structure-sharing, and not by anaphoric control. Therefore hana must be shared by segja and vera in the syntactic structure, but not in the semantic structure. Icelandic case favours sharing analyses in other ways as well. In particular, the facts about ‘quirky’ cases have been quoted as evidence for sharing. The data are summarised conveniently by Pollard and Sag (1994: 138). For example, vanta, ‘lack’, requires its subject to be accusative, but if this verb is the complement of virðist, ‘seems’, the latter’s subject also has to be accusative (Andrews 1982: 462, Pollard and Sag 1994: 138): (8) a. DrenginaA vantar matA. the-boys lacks food ‘The boys lack food.’ b. HanaA virðist vanta peningaA. her seems to-lack money ‘She seems to lack money.’ In the ﬁrst example vantar governs the case of the ﬁrst noun, which there are excellent reasons for taking as its subject (Andrews 1982). The same government relationship exists across ‘seems’, and there are equally good reasons for taking hana as the latter’s subject too, so we have another clear example of sharing, with hana shared as subject by both virðist and vanta. Figure 4 presents the WG analyses. In both these constructions the evidence for sharing came from a purely syntactic rule for inﬂectional case, but the ﬁrst one involved case-agreement while the second involved case-government. An important diﬀerence between the two is that the latter can lead to case-conﬂicts. The normal case for an Icelandic verb’s subject is

functional control

s

157

o

Drengina /A vantar mat /A. s

r

o

Hana /A virðist vanta peninga /A. s Figure 4. nominative, so one would expect the subject of virðist to be nominative; but vanta requires its subject to be accusative. The fact that this conﬂict is resolved in favour of vanta suggests that the mechanism is default inheritance—the subject of a verb inherits the default case3 (nominative, required by the typical verb) unless this is overridden by a more speciﬁc case-requirement (accusative, required by the verb vanta). Diﬀerent theories provide diﬀerent mechanisms for this kind of conﬂictresolution, but the basic insight is probably uncontentious. In WG the mechanism is default inheritance itself (Hudson 1990, chapter 3; Fraser and Hudson 1992). The main conclusion is that case provides strong evidence for sharing in Icelandic, whether we consider the facts of case-agreement between a predicate nominal (noun or adjective) and its subject or the facts of case-government by non-ﬁnite verbs. So far as I know there is no serious disagreement about this conclusion so long as we concentrate on examples where the sharing is purely syntactic, without any sharing in the semantic structure. The verbs responsible for the sharing were all ‘raising’ or ‘ECM’ verbs, in whose semantic structure the syntactically shared element played no part. It is generally agreed, then, that one word (or phrase) can have two diﬀerent syntactic roles (e.g. as subject of two diﬀerent words). The outstanding question is whether such a word may also have two diﬀerent semantic roles.

3. Functional control verbs: Icelandic Consider the following examples (Anderson 1992: 116). (‘D’ stands for dative.) (9) Ég bað hannA að vera góðanA /góðurN /*góðumD. ‘I asked him to be good.’ (10) Hann skipaði honumD að vera góðumD /góðurN /*góðanA. ‘He ordered him to be good.’

158

richard hudson

In both examples the adjective may be in either of two cases: nominative, or the same as the preceding pronoun. This choice is easy to explain if we assume a structural ambiguity between a sharing analysis and a non-sharing analysis. In the sharing analysis, hann in (9) is not only the object of bað but also the subject of the inﬁnitive (and therefore also of góðan). In the non-sharing analysis hann is merely coreferential with the inﬁnitive’s understood subject, so the adjective has no overt subject and takes the default case for subjects, nominative. This is the pattern found in examples where sharing is out of the question such as the following (Maling and Sprouse 1995): (11) Að vera kennariN /*kennaraA er mikilvægt. to be teacher is important ‘It is important to be a teacher.’ In short, the predicative adjective always agrees with its subject, and the case of a ‘covert subject’ is always nominative. (We shall discuss the problem of covert subjects in the next section.) Figure 5 shows the two structures that we can assume for example (9).

Ég bað hann /A að vera goðan /A

Ég bað hann að vera goður/N. Figure 5. What is important in these analyses is that hann is semantically (as well as syntactically) related to bað, ‘asked’. In the sharing analysis the sharing is also syntactic, so we have evidence here that one word may be shared by two other words in the semantic structure as well as in the syntactic. Unfortunately this conclusion is contested by Andrews (1982: 432), who argues that verbs in Icelandic, unlike their English counterparts, never share their objects with their inﬁnitival complements by functional control. They only take inﬁnitives with PRO subjects, which are linked to the intervening NP by anaphoric control. Anaphoric control involves coreference (in semantics), but no syntactic identity, so it predicts that predicate nouns and adjectives should all be in the default nominative case. Why, then, do we ﬁnd any examples of other cases in apparent agreement with

functional control

159

the case of the controlling NP (henceforth ‘NPc’)? Andrews’ explanation is that this is a performance phenomenon, ‘case attraction’ (Andrews 1982: 452), so the sentences containing it are ungrammatical. My reasons for rejecting this conclusion are as follows: • Andrews does not consider the possibility of syntactic ambiguity, so his evidence against a structure-sharing analysis can be reinterpreted as showing that there are some structures that do not allow structure-sharing, rather than that there are no structures that do. For example, if there is an intervening NP between NPc and the inﬁnitive this blocks agreement: (12) Þeir telya hanaA hafa lofað honum að vera góðN/*góðaA. they believe her to-have promised him to be good ‘They believe she promised him to be good.’ According to the syntactic ambiguity account, these data show that the verb lofa ‘promise’ allows only subject-less inﬁnitives—i.e. it does not allow structuresharing. (A similar analysis for the English verb promise would explain why the usual sharing of lower subject and higher object does not apply.) Such examples are irrelevant to the presence of structure-sharing with other verbs. • If the apparent case-agreement were a performance phenomenon, involving shortterm memory limitations, one would expect it to be insensitive to abstract structure; why not, for example, a dative adjective in (12) under the inﬂuence of the pronoun honum? According to Andrews this is not possible, although a dative adjective is possible with lofa when it means ‘allow’ (ibid): (13) Hún lofaði honumD að vera góðumD. ‘She allowed him to be good.’ A more plausible explanation for the diﬀerence between (12) and (13) is that when lofa means ‘promise’ it takes a subject-less inﬁnitival complement, in contrast with the sharing complement that it takes when it means ‘allow’. If this is right, it is the grammar rather than the production system that makes ‘good’ agree with ‘him’ in one case but not in the other. • The predicative adjective is more likely to agree with NPc if it is accusative than if it is dative (Andrews 1982: 452). Why should this be if the agreement is a performance matter? Case-attraction should apply to datives as strongly as to accusatives. An alternative explanation for the facts that Andrews reports—patterns of responses from informants and text-frequencies—is that the grammar favours the sharing of accusative objects. This would not be surprising given that the raising/ECM constructions always have accusative objects. In contrast dative objects

160

richard hudson

are more oblique and more likely to exert anaphoric rather than functional control. • In some examples the predicate nominal does not even allow nominative case. With some verbs that allow a bare inﬁnitive (without að), the accusative of the object is highly favoured, or even obligatory (Andrews 1982: 453). An example is BIðJA, ‘ask’: (14) Ég bað MaríuA vera ?góðN /góðaA. ‘I asked Mary to be good.’ If the accusative case was a performance error it would have to be an obligatory performance error, which would be hard to distinguish from a grammatical rule. A much easier solution would be to say that the verb BAð ‘ask’ takes an accusative object which doubles up as subject of the inﬁnitival complement. • The supposed case-attraction seems to apply much less readily, if at all, when the functional-control verb is intransitive (in a pattern NP + V1 + V2). Most of these verbs (as V1) take ordinary nominative subjects, but some take subjects with quirky case (e.g. langar, ‘longs’ takes an accusative subject). Whatever the case of V1’s subject, the preferred case for the predicative nominal after V2 is nominative, and according to Andrews it is doubtful whether it would ever be accusative (Andrews 1982: 454). The following are Andrews’ examples and judgements: (15) a. ÉgN vonast til að vera vinsællN. I hope for to be popular ‘I hope to be popular.’ b. HanaA langar að vera ríkN /?ríkaA. ‘She longs to be rich.’ Why should case-attraction not apply in examples like this? Admittedly the presence of the intervening V1 might weaken the contaminating eﬀect of NP, but not to this extent. Furthermore, if mere adjacency was relevant, subject-inversion should produce a subject whose inﬂuence should be as strong as that of an object; so an accusative adjective should be as likely in (16a) as in (16b) (=13): (16) a. Langar hanaA að vera ríkN /ríkaA? longs she to be rich ‘Does she long to be rich?’ b. Hún lofaði honumD að vera góðumD. ‘She allowed him to be good.’

functional control

161

It would be surprising if this turned out to be true. • Furthermore, Sigurðsson (1991) shows that the case of V1’s subject is irrelevant to V2’s dependents. For example, a ﬂoated quantiﬁer always agrees with the expected subject of V2, and is never ‘contaminated’ by the nominative of V1’s subject instead. The crucial point of the following examples is the accusative case of alla in the last one, which can only be explained if we assume that the understood subject of the inﬁnitive vanta is diﬀerent from the nominative subject of vonast, ‘hope’. (17) a. StrákarnirN komust allirN í skóla. the boys get all to school ‘The boys all managed to get to school.’ b. StrákanaA vantaði allaA í skólann. the boys lacked all in the school ‘The boys were all absent from school.’ c. StrákarnirN vonast til að komast allirN í skóla. the boys hope for to get all to school ‘The boys hope to all manage to get to school.’ d. StrákarnirN vonast til að vanta ekki allaA í skólann. the boys hope for to lack not all in the school ‘The boys hope not to all be absent from school.’ Sigurðsson’s evidence shows very clearly that at least some intransitive verbs are absolute barriers to ‘case-attraction’. This is not what we should expect of a performance inﬂuence, but unsurprising if case is determined by grammatical structure. The natural conclusion (which Sigurðsson takes as given) is that verbs like ‘hope’ in Icelandic take an inﬁnitive whose covert subject has the normally expected case—nominative by default, but quirkily accusative with vanta. My conclusion, then, is that Icelandic allows either sharing or subject-less patterns after most transitive functional-control verbs, though it only allows subject-less ones after intransitives. The main point is the existence of some transitive functionalcontrol verbs which demonstrably do allow sharing—i.e. which allow their object to act as the syntactic subject of their inﬁnitival complement. If this conclusion is correct, then it establishes the general principle that syntactic sharing can be combined with semantic sharing, because in all these examples NPc is very clearly a semantic argument of the ﬁrst verb as well as of the second. This principle is controversial as it conﬂicts with the theta-criterion, but we shall ﬁnd further support for it in Ancient Greek.

richard hudson

162

4. Functional control verbs: Ancient Greek Another language in which transitive verbs seem to allow sharing is Ancient Greek, pace Andrews (1982: 452). For example, take the following example from Andrews (1971: 130), with ‘G’ standing for genitive: (18) Ku:rouG edeonto ho:s prothumotatouG genesthai of-Cyrus they-begged maximally devoted to-be ‘They begged Cyrus to be as devoted (to them) as possible.’ The obvious explanation for the genitive case on prothumotatou, ‘devoted’, is that it is agreeing with Ku:rou; in general predicate nominals agree with their subjects (which are also the subject of the associated copula verb, in this example genesthai, ‘to be’), so Ku:rou must be the subject of prothumotatou. But Ku:rou gets its case by government from edeonto, ‘they begged’, so it must be the latter’s object. Therefore it must be shared by the two verbs. The assumed structure is shown in Figure 6.

r o

a

r

Ku:rou/G edeonto ho:s prothumotatou /G genesthai. s s Figure 6. This example illustrates a pattern which is widely used in Ancient Greek, as witness the following examples. The ﬁrst (provided by Anna Morpurgo Davies) is from Herodotus VII 160, the second (via James Tauber) from the New Testament (Titus 2.4): (19) sù . . . ou meA épeisas askhe:monaA . . . genésthai you . . . not me persuaded indecorous . . . to-be ‘You have not persuaded me to be indecorous.’ (20) so:fronizo:sin tas neasA philandrousA einai they-may-train the young-women husband-loving to-be ‘They may train the young women to be devoted to their husbands.’ The explanation for these sentences should be seen in the context of other constructions in Ancient Greek, where similar agreement patterns can be found. As we might expect, structure-sharing is found where the lower verb’s subject has no semantic link to the higher verb (as in English raising/ECM structures). The following

functional control

163

is an example (Andrews 1971: 141) in which the higher verb imposes dative case on the lower verb’s subject: andriD genesthai. (21) nu:n de soiD eksestin now ? to-you it-is-possible man to-be ‘Now it is possible for you to be a man’ The dative case of andri, ‘man’, reveals its syntactic link to soi, ‘to you’, similar to the link in example (18). However, there is a range of other possibilities which show that the higher verb’s object does not always determine the case of the lower verb’s subject. To start with, the same verb may allow two diﬀerent structure-sharing patterns, one with its (indirect) object and the other with its subject. This explains the examples which Andrews considers really problematic (Andrews 1971: 149): (22) a. sunoida emauto:iD e:dike:meno:iD b. sunoida emauto:iD e:dike:menosN I-am-aware-of to-myself having-been-wronged ‘I am aware that I have been wronged’ Two verbs, both meaning ‘to be aware of’, allow both these possibilities. In example (a) the participle’s subject is presumably the higher indirect object emauto:i ‘to myself’, in contrast with example (b) where it structure-shares with the higher subject. The latter pattern is also found with predicative adjectives as in the next example (Andrews 1971: 144): (23) doko: moiD adunatosN einai I-seem to-myself unable to-be ‘I seem to myself to be unable’ And it is found, without the dative pronoun, with intransitive verbs (ibid: 132): (24) Perse:sN einai ephe: Persian to-be he-said ‘He said that he was a Persian’ In short, the subordinate verb or predicative may agree with the main verb’s subject even when the main verb also has some kind of object which could have controlled its case as in the pair of examples in (22). The examples just discussed raise an important theoretical side-issue: the status of ‘covert subjects’. In all the examples where the subordinate verb or predicative is nominative it must be agreeing with the main verb’s nominative subject, even though the latter is only covert. As with Icelandic, the most natural way to explain

richard hudson

164

r

r

Perse:s/N einai ephe:. s s s w/N Figure 7. predicative case is to say that the case of a predicative (or of a case-marked verb such as a participle) is always the same as the case of its subject. Agreement with a covert subject is easy to express in theories which allow empty categories, but problematic in theories such as WG which eschew them. It is hard to think of a better explanation for the facts, so the theory of WG needs to change. As a start I hereby introduce a notation for covert dependents: a vertical arrow that points down below the words at a ‘virtual’ word ‘w’ whose case is determined by the usual rules. Structure-sharing involves another arrow pointing at the same spot, as shown in the structure for example (24) in Figure 7. Unlike Icelandic, we cannot take nominative as a default case for all subjects. This is because the case of a predicative varies according to the ﬁniteness of the verb whose subject it shares. As the examples quoted above show, ﬁnite verbs have nominative subjects and nominative predicatives. But non-ﬁnite verbs have accusative subjects, and accusative predicatives. In other words, a non-ﬁnite verb may have a subject of its own which is not structure-shared with a higher verb, so it determines its own case; and when that is true, the case is accusative. This can be seen clearly with verbs which allow an accusative NP + inﬁnitive: when the higher verb is passivized, the NP may turn into its nominative subject, but it may also stay accusative (Andrews 1971: 133): (25) a. legousin tousA andrasA elthein they-say the men to-have-come ‘They say the men came.’ b. legontai hoiN andresN elthein they-are-said the men to-have-come ‘The men are said to have come.’ c. legetai tousA andrasA elthein it-is-said the men to-have-come ‘It is said that the men came.’ Examples like these are easy to explain if we allow such verbs to subcategorize either for an object and an inﬁnitive, or only for an inﬁnitive, and then allow an

functional control

165

r c o legousin tous /A andres/A elthein. s

s

?

c legousin tous /A andres/A elthein. Figure 8.

inﬁnitive to have a subject of its own. This means that examples like (a) are syntactically ambiguous. The sequence V1 + N + V2 allows two diﬀerent analyses, as it did in Icelandic; and as in Icelandic, one of the possibilities is that N is structureshared by the two verbs (as object of V1 and subject of V2). But unlike Icelandic, in Greek the other possibility is that N belongs exclusively to V2. Figure 8 shows the two competing structures for example (25a). If this explanation is right, it is possible for a non-ﬁnite verb to have a subject ‘of its own’, for whose case it alone is responsible. This conﬂicts with the view (Chomsky 1986: 74) that non-ﬁnite verbs cannot assign Case to their subjects. In our terms, Chomsky’s claim was that a non-ﬁnite verb’s subject must always be structure-shared; but examples like (25) show that this is not true. A non-ﬁnite verb’s subject need not belong syntactically to any other verb. The same conclusion is supported by other evidence. Andrews quotes a particularly fascinating example (ibid: 133) in which two coordinated participles have diﬀerent cases because the second has a subject with which it must agree: (26) heo:ro:n ou katorthountesN kai tous stratio:ta:sA akhthomenousA they-saw not succeeding and the soldiers aggrieved ‘They saw that they were not succeeding and that the soldiers were aggrieved.’ Presumably the ﬁrst participle katorthountes, ‘succeeding’, is nominative because its subject is the same as that of the ﬁnite main verb, heo:ro:n, ‘they saw’, which of course must also be nominative. The accusative NP tous stratio:ta:s, ‘the soldiers’, cannot be the object of heo:ro:n because the ﬁrst conjunct shows that heo:ro:n has no object; therefore it must be the unshared subject of the second participle. Figure 9 shows the structure for this example. The brackets {. . .} and [. . .] demarcate the coordinate structure and its conjuncts. (In WG, coordination is a

richard hudson

166

?

s c a heo:ro:n {[ou katorthountes/N] [kai tous stratio:ta:s/N akhthomenous/A]} Figure 9. separate kind of structure from dependency structures; see Hudson 1990: 404–20.) A non-ﬁnite verb in Ancient Greek therefore may have at least two possible kinds of subject. It can have a structure-shared subject which is either subject or object of the higher verb, and whose case is ﬁxed by the higher verb; or it can have its own separate overt subject, whose case is accusative. However there is a third possibility, which is that its own separate subject may be covert. Even in this case, though, the case of a predicative shows that the covert subject is accusative. This can be seen in the next example (ibid: 148), whose structure is shown in Figure 10. (27) sumpherei autoisD philousA einai it-is-advantageous to-them friends to-be ‘It is advantageous to them to be friends.’ The possibility of covert subjects with non-ﬁnite verbs is hardly surprising, especially in a language like Greek where ﬁnite verbs allow them, but it has an important consequence for our present topic. We are surveying the possible syntactic analyses of a surface string containing a functional-control verb (V1) such as ‘ask’ or ‘persuade’, plus the latter’s object (N) and an inﬁnitive (V2). In Icelandic we found clear evidence for two possible structures: a. V1 and V2 share N (as object of V1 and subject of V2), b. N is the object of V1 but not the subject of V2, whose subject is covert. In our discussion of Ancient Greek we have so far found evidence for the ﬁrst structure. The main example was (18), where V2 had a predicative which agreed in case with the genitive N, the object of V1. We now turn to an example (ibid: 148) which illustrates the second structure, in which V2 has its own covert subject. As we

?

? r

sumpherei autois/D philous/A einai s s w/A Figure 10.

functional control

167

should expect from the preceding discussion, the case of the predicative shows that this subject is accusative, in contrast with the nominative that we found in Icelandic: (28) Athe:naio:nG edee:the:san sphisi boe:thousA genesthai. the-Athenians they-asked to-them helpers to-become ‘They asked the Athenians to become their helpers.’ In short, Ancient Greek allowed just the same range of structures for functionalcontrol verbs as we found in Icelandic: with or without structure-sharing.

5. Conclusion This is an important conclusion for general linguistic theory because it oﬀers two distinct precedents for English. As we have seen, some theories treat English functional control without syntactic structure-sharing (e.g. GB, HPSG), and these theories can be justiﬁed by pointing to Greek examples like (28) as evidence that this kind of analysis is certainly found in some languages. But the theories that do invoke syntactic structure-sharing (RG, LFG, WG) ﬁnd equally strong support in examples like (18). In short, the general theory of syntax must allow both possibilities (and must even allow them to coexist within a single language, which means that this is not yet another parameter of language variation). The two alternatives are presented in WG notation in Figure 11 which is the same as Figure 2 except for the arrows pointing down at the covert subject.

Functional Control =Structure-sharing

=Anaphoric control

r r o c o I persuaded Pat to buy a book. s s

? r o c o I persuaded Pat to buy a book. s s w

s

s

Figure 11. The data surveyed also support some other general conclusions, which can be summarised brieﬂy here: • Syntactic theory should recognise something like ‘empty categories’ (in particular, the notion ‘covert subject’). Without ‘covert subjects’ that have either nomi-

168

richard hudson

native or accusative case it would be very hard to explain the cases on predicatives in Ancient Greek. This is a challenge for WG and other theories which only recognise overt syntactic units. • Syntactic theory should allow non-ﬁnite verbs (and other predicatives) to assign a case to their subject, whether overt or covert. This challenges the GB claim that only a ﬁnite verb can assign Case to its subject. • Syntactic theory should allow syntactic and semantic structure-sharing to cooccur, rather than to be in complementary distribution (as in GB and HPSG). This challenges the GB Theta Criterion and its descendants. • Syntactic theory should allow typological variation of the kind illustrated by Icelandic and Ancient Greek. The subject of a non-ﬁnite verb is nominative in Icelandic, but accusative in Greek; and if it is not structure-sharing, it must be covert in Icelandic, but may be overt in Greek.

Notes 1. This paper derives from Hudson 1995a, but discusses the Ancient Greek data more fully and draws diﬀerent conclusions. For data and/or comments on the earlier paper I should like to thank the following: Richard Alderson, Stefanie Anyadi, Leslie Barrett, Loren Allen Billings, Alan Cienki, Bernard Comrie, Leo Connolly, Annabel Cormack, Anna Morpurgo Davies, William Diver, Alex Eulenberg, Claire Grover, Robert Hoberman, Elisa Konstantinou, Jacqueline Lecarme, Knud Lambrecht, Nigel Love, Alex Manaster-Ramer, Philip Miller, Regina Moorcroft, Bert Peeters, David Pesetsky, Carsten Peust, Karen Robblee, Rex Sprouse, James Tauber, Dimitra Tzanidaki, Cynthia Vakareliyska, Max Wheeler. 2. Chomsky’s 1981 presentation of the theta-criterion allows only one theta-role (not theta-position) per argument NP, and this version has been perpetuated in various introductory works (Radford 1988: 391, McCloskey 1988: 51, Manzini 1994: 502). However, in Chomsky (1986) he is careful to distinguish ‘theta-position’ from ‘theta-role’, on the grounds that a single D-structure position may be assigned more than one theta-role as in John left the room angry. 3. It is interesting to notice that an Icelandic verb’s subject has the same default case regardless of the verb’s ﬁniteness, but is aﬀected by lexical variation between verbs. We shall see in the discussion of Ancient Greek that the subject’s case does vary with ﬁniteness, but so far as I know it does not vary lexically. One wonders whether any languages allow both kinds of inﬂuence on the subject’s case.

References Anderson, Stephen. 1992. A-Morphous Morphology. Cambridge: Cambridge University Press.

functional control

169

Andrews, Avery. 1971. ‘‘Case Agreement of Predicate Modiﬁers in Ancient Greek.’’ Linguistic Inquiry 2: 127–51. ——. 1982. ‘‘The Representation of Case in Modern Icelandic.’’ In Bresnan, Joan (ed.), The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press, 427–503. Blake, Barry. 1990. Relational Grammar. London: Routledge. Bresnan, Joan (ed.) 1982. The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. ——. 1986. Knowledge of Language. Its nature, origins and use. New York: Praeger. Fraser, Norman and Richard Hudson. 1992. ‘‘Inheritance in Word Grammar.’’ Computational Linguistics 18: 133–58. Hudson, Richard. 1971. English Complex Sentences. An introduction to Systemic Grammar. Amsterdam: North Holland. ——. 1976. Arguments for a Non-Transformational Grammar. Chicago: University of Chicago Press. ——. 1984. Word Grammar. Oxford: Blackwell. ——. 1990. English Word Grammar. Oxford: Blackwell. ——. 1994. ‘‘Discontinuous Phrases in Dependency Grammar.’’ UCL Working Papers in Linguistics 6: 89–124. ——. 1995a. ‘‘Con PRO, or the Virtues of Sharing.’’ UCL Working Papers in Linguistics 7: 277–96. ——. 1995b. ‘‘Does English Really have Case?’’ Journal of Linguistics 31: 375–92. McCloskey, James. 1988. ‘‘Syntactic Theory.’’ In Frederick J. Newmeyer (ed.), Linguistics: The Cambridge survey, vol. 1. Cambridge: Cambridge University Press. Maling, Joan and Sprouse, Rex. 1995. ‘‘Structural Case, Speciﬁer-head relations and the Case of Predicate NPs.’’ In Hubert Haider, Susan Olsen and Sten Vikner (eds.) Studies in Comparative Germananic Syntax. Dordrecht: Kluwer, 167–86. Manzini, Rita. 1994. ‘‘Chains.’’ In Ronald Asher (ed.), Encyclopedia of Language and Linguistics. Oxford: Pergamon Press, 501–3. Pollard, Carl and Sag, Ivan. 1994. Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press. Sigurðsson, H. 1991. ‘‘Icelandic Case-Marked PRO and the Licensing of Lexical Arguments.’’ Natural Language and Linguistic Theory 9: 327–63.

Applicative constructions in Warrwa WILLIAM McGREGOR Leuven University

1. Introduction1 Traditionally spoken to the north and east of the township of Derby in the Kimberley region of Western Australia, Warrwa is an Eastern Nyulnyulan language (Stokes & McGregor forthcoming). It shows (like the other languages in the family) two grammatically distinct applicative constructions, a comitative applicative and an instrumental applicative. Both of these are marked by the suﬃx -ngany—from protoNyulnyulan *-ngany—which attaches to both nominal and verbal roots. Nominal -ngany is a case-marking postposition which serves principally as an instrumental marker; verbal -ngany marks, in addition to the applicative constructions, an implicated construction, which basically signals that the referent situation is projected or intended for the future.ə The aim of this paper is to provide a detailed description of the meanings and uses of the two applicative constructions in Warrwa. In addition proposals are made concerning the grammatical structures involved in them, employing the framework of Semiotic Grammar (SG) (McGregor 1997). Nominal uses of -ngany are not discussed, and nor is its use in the implicated construction. (McGregor 1995 provides discussion of these uses.) Warrwa is in a critical state, having only two full speakers, along with a handful of part speakers. Almost all of the data is my own, from ﬁeldwork conducted in Derby in 1992, 1994, 1995 and 1996 with both speakers. The only other data I have used are a few texts narrated by one of the full speakers to Bronwyn Stokes in 1979; I am grateful to her for making the recordings available to me. The corpus thus shows many limitations, only some of which are likely to be resolved by further ﬁeldwork.

172

william mcgregor

2. Some aspects of Warrwa grammar Before we embark on our account of the applicative constructions, it is necessary to lay some groundwork, and provide information about some aspects of Warrwa grammar. (See McGregor 1994 for further information.) First, like other Nyulnyulan languages Warrwa shows a set of around a dozen postpositions, bound phrase-level enclitics, which are typically attached to the ﬁrst word of the NP (McGregor 1994: 14, 26). Most of these mark case-relationships; there are distinct ergative, instrumental, dative, locative, allative, ablative, perlative, and comitative postpositions in the language. Second, Warrwa shows two types of verbal construction: simple and compound. Simple verbs consist of a single inﬂecting verb (IV), a morphologically complex form involving various preﬁxes and suﬃxes marking tense, aspect, mood, person and number of the subject and object, etc.. The applicative marker -ngany is one such suﬃx; it occurs in the antepenultimate order-class. Compound verbs, by contrast, consist of an IV together with an almost invariant preverb (PV), which usually conveys the main bulk of the lexical meaning. The IV serves as a classiﬁer of the PV in the compound verb construction. Third, Warrwa distinguishes three primary clause types: intransitive (e.g. (1)), which includes reﬂexive/reciprocal as a subtype (e.g. (2)); transitive (e.g. (3)); and middle (e.g. (4)).2 (1) juwa jawu minjan3 you swim you:do ‘You are swimming.’ (2) ngayu waarru ngambanyjiny I scratch I:did:ref ‘I scratched myself.’ (3) yila-na kujuk nankany warli dog-erg swallow it:carried meat ‘The dog swallowed the meat.’ (4) ngayu-na ngamurunguny-jina kinya wamba I-erg I:looked-3minobl this man ‘I looked for this man.’ Following McGregor (1994: 55–57) and (1997: 93–106), it is proposed that these three clause types may be characterised in terms of the number and types of inherent experiential roles which they exhibit, as follows, where the order of roles is immaterial:

applicative constructions in warrwa Intransitive Transitive Middle

173

Actor/Medium and SoA; Actor/Agent, Undergoer/Medium and SoA; Actor/Agent, Aﬀected/Medium and SoA.

The roles may be described roughly as follows (see McGregor 1990: 324 for further explanation). The Actor is cross-referenced by a nom preﬁx to the IV, and designates someone or something engaged in a situation; the Undergoer is crossreferenced by an acc pronominal enclitic, and designates someone or something that suﬀers from the action, typically as a patient; the Aﬀected is cross-referenced by an obl enclitic, and indicates someone aﬀected by the action. These three roles will be referred to as participant roles; they are subject to a role-uniqueness stipulation. Each conﬂates with (indicated by the slash) another role, either Medium or Agent. The Medium, realised by an unmarked NP, designates something through which the situation comes into being; the Agent role, typically discharged by an erg PP (sometimes the erg postposition is omitted), indicates someone or something that engages in goal-directed activity. Non-participant roles are not subject to a uniqueness requirement. Finally, the SoA is the role served by the verbal construction; it indicates the action, event, state, or whatever that is being referred to.

3. The comitative applicative construction Two distinct types of applicative construction are discernible in Warrwa, a comitative applicative and an instrumental applicative. In comitative applicatives -ngany can be attached to the IV of some intransitive and transitive clauses, but not, apparently, of middle clauses. Generally a comitative applicative clause has one more role than the agnate clause without -ngany. This is almost always a Medium, which, in many instances, conﬂates with the Undergoer participant role. Thus in most cases the app suﬃx may be considered to mark an increase in valence or transitivity of a clause, as per generally accepted wisdom (Shibatani 1996: 159). However, two points must be stressed. First, -ngany app marks a clausal rather than verbal category; verbal transitivity is never aﬀected (cf. Tsunoda this volume)). Thus when attached to an intransitive clause, the transitive preﬁx in the verb never occurs; nor does it condition the omission of reﬂexive/reciprocal aﬃxes. Second, there is a small, but not insigniﬁcant, class of instances in which the comitative applicative construction has the same number of roles as its non-comitative agnate. The discussion of the comitative applicative construction is divided into subsections according to the transitivity of the agnate clause without the app suﬃx. Discussion of the instrumental applicative is left for Section 4.

174

william mcgregor

3.1 Intransitive clauses The applicative suﬃx -ngany may be attached to the IV in intransitive clauses of three main semantic types, and one grammatical type: motion, stance, speech, and reﬂexive/reciprocal. We discuss these in order in the following Subsections. 3.1.1 Clauses of motion In clauses of motion -ngany app typically indicates that the movement occurred with, or in the company of somebody or something; the latter thing is generally taken along by the Actor, who serves as the primary mover and instigator of the action. Perhaps the most common clause of motion which the app occurs in involves the compound verb nguy . . . -JI ‘return’. Without -ngany app the clause is intransitive, as (5) shows; with -ngany app it is transitive, as demonstrated by (6), which involves an ergatively marked NP and an unmarked NP. (Here -ngany clearly serves as a prototypical applicative marker, indexing the presence of a second participant role in the clause, an Undergoer/Medium.) (5) linyju nguy jina policeman return he:did ‘The policeman returned.’ (6) linyju-na nguy jina-ngany-jirr, kuya jirra, policeman-erg return he:did-app-them mother their ‘The policeman brought back their mother.’ Other lexical verbs of motion that are attested in the applicative construction include: -BULA ‘emerge, arrive’, bud . . . -BULA ‘come to surface’, kudii . . . -BULA ‘run away’, -JALU ‘fall’, (inyja) . . . -NDA ‘walk, go walking’, kurrak . . . -JI ‘go away/across’, and jawu . . . -JI ‘swim’. In each case the app construction conveys the meaning that the Actor engaged in the movement is in the company of another entity; because it is transitive, the Actor is also an Agent, the other entity an Undergoer and Medium. The former is always animate; the latter may be animate or inanimate. As far as I can ascertain, both participants in app clauses of motion necessarily change position. Thus it is not obvious from a logical point of view whether it is the Actor/Agent or the Undergoer/Medium that corresponds to the Actor/Medium of the agnate intransitive clause. However, two facts suggest that it is the Actor/Agent (see also Austin forthcoming, Section 2.2). First, examination of the corpus reveals no examples of nguy . . . -JI ‘return’ with an inanimate Actor incapable of independent movement. (The same may well hold for the other verbs, but I have not exhaustively checked the corpus.) Second, while it is true that both participants in the transitive

applicative constructions in warrwa

175

clause move, it does not appear to be true that their movement would necessarily be describable in terms of the agnate intransitive clause. For instance, the dog referred to in example (9) was dead, and jawu jan ‘it is swimming’ could not (I believe) be truthfully said of it, although it could have been said truthfully of the crocodile. Thus, in the applicative clause the animate being who is the primary moving entity, also brings about the motion of the other object. These are thus comitative, not causative, applicatives. The intransitive clauses which agnate with applicative clauses of motion such as (6) usually contain a -barri comit 1 PP4 as in (7)—rarely a -nyarri comit 2 PP, but never an inst PP. This corresponds to the Undergoer/Medium of the applicative clause. (7) nguy jiny jina-warri waangu kinya wamba this man return he:did his-com 1 wife ‘He came back with his wife.’ In examples such as (7) the Actor and the accompanying thing are represented as jointly engaged in the action, although the latter typically refers to a lower status entity, often a lower-order animate. What is important is that the Actor is not represented as directing their activity towards the accompanying thing, which (presumably) moves under its own volition; it does not undergo action instigated by the Actor. Matters are, however, somewhat more complicated than the above account suggests. As (8) shows, the Actor in what looks like an intransitive clause with a comit 1 marked NP designating an accompanying entity, is sometimes marked by the erg postposition. On the basis of form alone it might be hypothesised that such clauses are intermediate in transitivity, lying somewhere between transitive and intransitive. Apparently examples such as (8) represent the accompanying thing as being brought along with the speaker, who remains in overall control of the action. However, in contrast to (6), the action was not necessarily as strongly or forcefully directed towards the accompanying thing brought—it came along relatively easily, inadvertently, or willingly, and not by coercion. The dog in (8) might be expected to move under its own steam, and not under duress; a similar example exists with the accompanying thing as the speaker’s wife, and other examples involve horses. By contrast, as the surrounding narrative made clear, the woman referred to in (6) did not willingly accompany the police, who needed to direct action more forcefully to her. This lends some support to the suggestion that (8) is lower in transitivity than (6)—but higher than an intransitive such as (5). (8) ngayi-na inyja ngarndany yila-warri I-erg walkabout I:went dog-com 1 ‘I went with my dog.’

176

william mcgregor

It seems that choices among the various modes of expression hinge on three primary considerations: (i) whether the accompanying entity is involved in the ‘‘direction’’ of the action; (ii) whether the accompanying entity is simultaneously engaged in motion under its own volition; and (iii) whether or not there is a diﬀerence in relative status of the Actor and the accompanying entity. The applicative construction is employed when (i) obtains: that is, when the accompanying thing is involved in the direction of the action. This condition is met in two main circumstances: either the accompanying item is incapable of independent motion—e.g. it is dead, as in (9), or inanimate, as in (10)—and thus needs to be moved, or it is capable of independent motion, but resists movement, or moves unwillingly, as in (6). Intransitive clauses with a comit PP are employed when (ii) obtains; the Actor of this clause is sometimes ergatively marked to reinforce or underline its agentivity. Finally, as regards (iii), when there is a diﬀerence in the relative status of the entities, comit 1 is used; when they share the same status, comit 2 is used. (9) linykurra-ni jawu jan-ngany yila nilirr-kany jina crocodile-erg swim it:does-app dog mouth-loc his ‘Crocodile is swimming along with a dog in its mouth.’ (10) kinya-na wajbal bulany-ngany-janu this-erg white:person he:came-app-1minobl ‘That white man brought money to me.’ Diagrammatic schemata similar to those employed in cognitive linguistics (e.g. Langacker 1987) provide a useful way of conceptualising the contrast between these various modes of expression. Figure 1 provides tentative representations of the main construction types we have discussed so far. In these diagrams the small circles represent two entities, A (Actor) and B (accompanying entity), involved in some situation; the surrounding ellipse indicates an associative relationship between the two entities. The headless horizontal lines represent motion, while the headed vertical lines represent directed (transitive) activity. Unbroken lines signify actions which appear to be explicitly marked, as distinct from those which are implied, which are represented by the broken lines. Grey indicates secondary status. If these representations are plausible, the contrasts among the four constructions can be reduced from three to two, by combining (i) and (ii): the status of A relative to B; and whether A directs action to B. In (a) and (c) B is of lower status, judged as less important than A, whereas in (b) and (d) they are accorded equal status. On the other hand, (a) and (b) contrast with (c) and (d) in the respect that the ﬁrst two show no directed action, no transfer of action, between A and B. What we have seen so far suggests that -ngany app registers the presence of an

applicative constructions in warrwa

177

A

A

A

A

B

B

B

B

(a) Intransitive clause with comit1 PP.

(a) Intransitive clause with comit2 PP.

A

A

A

A

B

B

B

B

(c) Intransitive clause with comit1 PP and ergatively marked actor.

(d) Applicative clause (transitive).

Figure 1. Diagrammatic representation of various types of clauses of motion.

Undergoer/Medium in clauses of motion, and signals that the clause is transitive. The Actor/Agent directs activity to the Undergoer/Medium, bringing about its change of position; simultaneously, the latter accompanies the former, who is represented as being in ultimate control of the action. As in transitive clauses generally, the ergative postposition does not always mark the Actor/Agent, as illustrated by (11). (11) nguy jina-ngany jina waangu kinya wamba return he:did-app his wife this man ‘He came back with his wife.’ How this contrasts with the agnate clause in which the erg postposition is attached to kinya ‘this’ (also recorded) is not certain. Possibly it has something to do with the expectedness and/or agentivity of the Agent (e.g. McGregor 1992), in which case the clause remains transitive. Unfortunately, this issue cannot be pursued due to paucity of data. To conclude the discussion, it is acknowledged that there are a few problematic examples, including: (12) angki mirndany-ngany what you:went-app ‘Who did you go with?’ (13) jalun-ngany baalu wamba he:falls-app branch man ‘He falls to the ground with the branch.’

178

william mcgregor

(14) wamba lagarr jiny-ngany baalu man climb he:did-app tree ‘The man climbed the tree (by himself).’ (12) was elicited in response to the free translation shown. It is conceivably an inappropriate translation, possibly indicating misunderstanding of the prompt; however, it is not the only such example, and cannot be ruled out. (13), along with two other examples involving the same VP, was elicited from the younger speaker, but rejected by the older speaker, who consistently rejected all similar examples which I tested with him. Regarding (14), however, there is no doubt as to its acceptability. And it may be that the semantic description given above can be modiﬁed slightly to account for it. Basically, all that is required is the removal of the bottom horizontal line in Figure 1 (d) indicating the movement of the Undergoer—the Actor is moving, in the ‘‘company’’ of the tree, and directing action towards it, although it remains stationary.5 Moreover, the absence of the erg postposition may signal reduced agentivity of the Actor/Agent vis-a-vis Actor/Agents which induce and/or control the movement of the accompanying entity. Some such reduction in agency might also provide an explanation for (12) and (13): these examples might, for instance, be appropriate to contexts in which the person was, say, holding onto the accompanying thing, and thus directing action to it, although not causing it to move. If this suggestion is sustained by further ﬁeld investigations, clauses such as (12)– (14) may well be transitive. The app suﬃx could signal that the clause is transitive, and that the Undergoer/Medium accompanies the Actor/Agent. This is entirely consistent with the remainder of the grammar of Warrwa, and for now I presume it to be the correct analysis. 3.1.2 Clauses of stance Just two verbs of stance are attested with -ngany app, yuk . . . -JI ‘lie, camp’, and (mijala) . . . -NI ‘sit’. The function of the suﬃx in this environment is similar to when it is attached to a verb of motion: it signals the presence of a second participant role, additional to the Actor of the agnate intransitive clause. Action is directed by the Actor towards this additional entity, which simultaneously accompanies it. This emerges particularly clearly in example (15), where the normal implication is that the Actor would be having sex with the Undergoer, just as in the English translation sleep with. (15) marlu yuk wilyina-ngany, not camp he:might:have:done-app ‘The old woman wouldn’t let him sleep with the young one.’ (Literally: ‘He couldn’t camp with her’.)

applicative constructions in warrwa

179

A

A

A

A

A

A

B

B

B

B

B

B

(a) Intransitive clause with locative PP.

(b) Applicative clause.

Figure 2. Diagrammatic representation of two types of clause of stance. Unfortunately, only a few examples involving yuk . . . -JI ‘sleep, lie’ are available, and none has an overt Actor NP. Hence it is not known for certain that this NP can be marked by the erg postposition, although there is no reason to believe that it could not be, particularly since the Actor in a corresponding applicative clause involving mijala . . . -NI ‘sit’ can be (see below). Thus (15) may reasonably be regarded as transitive, with an Actor/Agent and an Undergoer/Medium. I have not been able to ﬁnd an example of an agnate intransitive clause in which the accompanying thing is marked by a comit postposition. However, (16) is an intransitive agnate with a loc PP designating the accompanying entity. The contrast with (15) is manifest: whereas the latter suggests sexual liaison between the participants, the former does not; it merely concerns the physical location of the addressee. (16) nyin-kardiny yuk walyi, ngaya-n, here-side camp you:turn me-loc ‘‘‘Sleep this side, with me!’’ ’ Figure 2 represents the contrast between these two modes of expression in schematic terms. (The dotted ellipse in (a) represents the relationship of location.) For (mijala) . . . -NI ‘sit’ a set of ﬁve agnate clauses is available. (17) and (18) are intransitives with comitative PPs indicating entities accompanying the Actor, sitting with him; no action is directed from the former to the latter. (19)–(21) are applicatives. It is suggested that (20) and (21) can be characterised as per Figure 1 (d), the horizontal lines representing motion being, of course, absent. (19), however, is problematic: the Actor cannot, it seems, be ergatively marked, and the other NP is marked by the comit 1 postposition. Apparently it constitutes a clause type of intermediate transitivity; how it should be appropriately characterised diagrammatically remains unclear. (17) wamba mijala ingan, waangu-warri; jina, man sit he:is wife-com 1 his ‘The man is sitting with his wife.’

180

william mcgregor

(18) mijala ingan waangu-nyarri jina wamba his man sit he:is wife-com 2 ‘Man is sitting with his wife.’ (19) mijala ingan-ngany waangu-warri jina, wamba *-na, sit he:is-app wife-com 1 his man *-erg ‘The man is sitting with his wife.’ (20) ingana-ngany kinya-na wamba mijala he:was-app this-erg man sit ‘He was sitting with him.’ (21) kinya wamba mijala ingana-ngany-ngayu this man sit he:was-app-1minacc ‘This man was sitting with me.’ In contrast with applicative clauses of motion, the Actor in an applicative clause of stasis is infrequently ergatively marked; usually, as in (19) and (21), it is represented by an unmarked NP. This is consistent with the above suggestion that the Actor/Agent is ergatively marked to foreground its agency, especially when action is directed from that entity to the accompanying one—which circumstance does not normally emerge with the action of sitting. In fact, (20) is the only example in which the Actor is ergatively marked; perhaps, as discussed in relation to (14) above, it suggests the Actor was holding onto the accompanying person—or perhaps, less materially, holding them in conversation. (On other occasions, both speakers rejected examples with the ergative postposition on the Actor.) 3.1.3 Clauses of speech The general verb for speaking in Warrwa is the inﬂecting verb -NGANKA ‘talk, speak’, which is homophonous with the preverb nganka ‘talk, speak’, and nominal nganka ‘word, speech, language, etc.’. It is a monovalent inﬂecting verb, and thus normally occurs in an intransitive clause, the Actor of which is usually realised by an unmarked NP, as in (22). However, the Actor NP is occasionally ergatively marked, as in (23), presumably to stress the agentivity of the Actor, highlighting the fact that they are engaged in directed, reciprocating, action. (22) ngayu layi-mirri ngangankany I alone-emp I:spoke ‘I was talking to myself.’ (23) kujarra-nma yangankan two-erg they:talk ‘The two are talking together.’

applicative constructions in warrwa

181

When -ngany app is attached to the inﬂecting verb -NGANKA ‘speak’ the valence of the clause is increased by one, and it becomes transitive. The Undergoer is the addressee, and is referred to by an unmarked NP (as in (24)) which is crossreferenced by an acc pronominal enclitic (as in (25))—rather than an OBL enclitic, as in middle clauses of speech. (24) wardal ngirrngankany-ngany boss they:talk-app ‘They spoke with the boss.’ (25) wamba ngirrngankan ngirrngankan, ngirrngankan-ngany-jirr, man they:speak they:speak they:speak-app-3augacc nganka ngirrarliny word they:ate ‘They speak together with them, discussing things.’ Unfortunately, there are no examples in the corpus in which the Actor is not ellipsed, but presumably it may be either unmarked or ergatively marked, depending on the agentivity of the Actor/Agent. Applicative transitive clauses of speech apparently translate as ‘speak with, talk with’. The action does not strongly aﬀect the Undergoer, certainly to a lesser degree than in non-applicative transitive clauses of speech which translate as ‘speak to’, ‘tell’, or ‘preach’, as in example (26). By comparison, in (24) the Actors are represented as speaking with the boss, rather than telling him something; and in (25) they are clearly discussing things together, some with others. Regular transitive clauses of speech are thus higher in transitivity—are more undergoer-eﬀective—than applicative clauses of speech. (26) ngayi-na ngangka-wan nganamany kinya iri I-erg I:spoke cont I:put this woman ‘I was speaking to that woman.’ Finally, the normal way of asking what someone is doing is by an intransitive clause involving the compound verb jana-ngkay . . . -NI (where-cont . . . be) ‘do what’, as in (27). To ask what someone is doing to someone or something else, the comitative applicative construction is employed, as illustrated by (28). This is entirely consistent with the above discussion. (27) jana-ngkay mingan where-cont you:are ‘What are you doing?’ (28) jana-ngkay mingan-ngany kinya baawa juwa-na where-cont you:are-app this child you-erg ‘What did you do with the child.’

182

william mcgregor

3.1.4 Reﬂexive/reciprocal intransitive clauses Just a few reﬂexive/reciprocal verbs are attested with the comitative applicative ngany: -KA ‘ﬁght’, -YAMA ‘argue’, and jirrmbil . . . -BANYJI ‘meet together’. Here also this suﬃx apparently registers an increase in transitivity. As (29) shows, the Actor/Agent can be referred to by an erg PP, the other participant by an unmarked NP, which is cross-referenced in the verb (see (30)). (29) yirra-na wardal jirrmbil ngirrwanyjiny-ngany they-erg boss meet they:did:ref-app ‘‘They met up together longa him.’’ (speaker’s gloss) (30) yirra-na jirrmbil ngirrwanyjiny-ngany-ngayu they-erg meet they:did:ref-app-1minacc ‘They all met up with me.’ What (29) and (30) indicate is that not only is action being exchanged among the Actors themselves as a group, but it is also being directed simultaneously to another individual who is involved in the situation as an Undergoer/Medium. The person who is being met up with is not (represented as) doing anything to bring about the meeting; their involvement is a consequence of the involvement of the others. This point is reinforced by the speaker’s gloss for another utterance almost identical with (30), ‘‘They all came to me mixed.’’ Similarly for examples (31) and (32), which also appear to be transitive. In (31), rather than the action being reciprocated among the Actors, and simultaneously between them as a group and an Undergoer, it seems—given the narrative context—that the action is reciprocated between each of the workers and their boss, but not amongst the workers themselves. And in (32) the ﬁghting is presumably reciprocal between the Actor and various members of the set of Undergoers, separately; but there is no suggestion that there was any ﬁghting amongst the latter group (or that reﬂexive action was involved). (31) ngirrmayamanyjina-ngany kinya wardal, jirra, they:argued:ref-app this boss they ‘They argued together with him.’ (32) wajbal mimankanyjin-ngany-jirr, white:person you:fought:ref-app-them ‘‘‘You used to ﬁght with the white people.’’ ’ The three diﬀerent interpretations we have just outlined for these examples can be represented schematically as in Figure 3 (a)–(c), where boxes surround entities grouped together by the relationship ‘and’. (a) indicates that A, B, C, D, E and F both interact with one another, and collectively on X, who serves as an Undergoer;

applicative constructions in warrwa

183

(b) that they collectively act on X, and at the same time, X reciprocally interacts with each of them; and (c) that X acts on them collectively, and simultaneously reciprocally interacts with each of them. (d) represents an expected (but not attested) scheme, a variant of (a) in which the action is directed from the single individual to the interacting group; doubtless there are other schemes which are also describable by comitative applicatives of reﬂexive/reciprocal clauses (e.g. where both participants constitute interacting groups). In each case, two perspectives are provided, a reciprocal and a transitive one. There is nothing unusual about this in the case of (a); however, for (b) and (c), there is an apparent inconsistency: A has both double headed and single headed arrows attached to it. In fact, the inconsistency is more apparent than real. It is only the black arrow that indicates the Agent role; the double-headed arrows indicate part of the semantics of the reﬂexive/ reciprocal verb—they have nothing to do with the meaning expressed at clause level. What we have in both cases is mutual interaction between (a subset) of the actants in the situation, but a subset is singled out as bearing responsibility for the situation. In (29) and (30) ‘‘they’’ are responsible for the meeting; in (31) ‘‘they’’ instigate the argument; and in (32) the addressee is indicated as causing the ﬁghts. Any referent situation satisfying the schemata of Figure 3 also satisﬁes the schema for a reﬂexive/reciprocal. But in the applicative, one of the entities is singled out from the others, and represented as ultimately responsible. As far as I am aware, there are no agnate intransitive clauses in which the non-Actor is referred to by a comit PP, and thus could be said to be ‘with’ the Actor, as in the case of applicative-marked intransitives of motion and stasis (sections 3.1.1 and 3.1.2). Nevertheless, the Actor of the agnate intransitive clause involves the union of the two sets of actants in the applicativised reﬂexive/reciprocal, and thus the Actor and Undergoer may be said to be together. (See also McGregor 1997, which suggests that ‘and’ and ‘with’ are subtypes of the general logical relationship of extension.) 3.2 Transitive clauses In a small number of examples comitative applicatives clause agnate with a transitive clauses. Consider examples (33)–(35): (33) yab ngindan jina wanangarri naman-ngany-ngayu away he:goes his money he:put-app-1minacc ‘He went away, leaving (putting/having put) his money with me.’ (34) yab ngirrangulany-ngany jina babala away they:sent-app his brother ‘They sent him away with his brother.’

william mcgregor

184

A

B D

E

A C

F

B D

E

C F

X

X

(a) Example (30): ‘meet together with’

(b) Example (31): ‘argue together with’

A

B D

E

A C

F

B D

E

C F

X

X

(c) Example (32): ‘ght (together) with’

(d) Hypothesized alternative to (a)

Figure 3. Schematic diagrams for three clauses with reciprocal APP verbs. (35) ngirrawulany-ngany baalu yirra-na kinya wamba they:tied-app tree they-erg this man ‘They tied this man up to the tree.’ In each of these examples there appear to be two Mediums, which may be designated by unmarked NPs: in (33), both the money and the speaker; in (34), both the

applicative constructions in warrwa

185

man and his brother; and in (35) both the man and the tree. One of these Mediums only may be cross-referenced by an acc enclitic to the IV, and thus only one may be conﬂated with the Undergoer role. Typically, as (33) shows, this is the one which is highest in animacy, even though it may not be the entity most directly acted upon. In each of the above examples there is clearly an associative relationship between the two Mediums. Presumably there are agnate transitive clauses with a single Medium/Undergoer, and a PP corresponding to the second Medium of the applicative clause. Corresponding to (34) is a plain transitive clause in which the accompanying person is referred to in a comit 1 PP; and corresponding to (33) and (35) are transitive clauses in which the accompanying thing is designated by a loc PP. There are just a few examples that do not ﬁt into the above pattern. First, there are a small number of cases in which the applicative does not have a second Medium role, but instead shows a comit 1 PP designating the accompanying item, as in (36)—which is not too surprising given the discussion of section 3.1.1 above. (36) ngayi-na nganjalany-ngany kinya wamba wariny-warri wamba I-erg I:saw-app this man one-com 1 man ‘I saw him with his mate.’ In fact, all instances bar one of -JALA ‘see’ in the comitative applicative construction are like this. These are obviously transitive clauses; but the associative relationship is between the Undergoer/Medium and the comit 1 PP, rather than a second Medium. Precisely how these should be incorporated into the present account remains unclear. Second, there are a few examples in which the applicative clause has the same number and types of role as the agnate transitive clause. This seems to be the case in (37), which has the same role-conﬁguration as the agnate plain transitive (38).6 (37) stockman-ni wajbal, kirlay-kirlay ngirraana-ngany waangu jina, stockman-erg white:person chase-chase they:gave-app wife his ‘The stockmen chased his wife all the time.’ (38) yawarda-ngany kilay nangkaana-yirra baabala-na jirra, horse-inst chase he:gave-them brother-erg their ‘He chased them on a horse, their brother.’ How can examples like (37) be accounted for? One possibility (which accords neatly with the framework outlined above) is that the Undergoer/Medium is actually with the Actor/Agent: that the stockmen and the Aboriginal woman were in close physical proximity throughout the performance of the action—and indeed, it is clear from the text that the stockmen chased the woman for sexual purposes, and there was probably no great distance between the interactants. By contrast a much greater

william mcgregor

186

X

X

X

A A A B

B

B

(a) Example (33): ‘put (leave) with’

X

X A

X A

A

(b) Example (37): ‘chase after’ Figure 4. Diagrammatic representation of two types of transitive applicatives. physical distance would presumably have separated the Actor/Agent and the Undergoer/Medium in (38). Similarly (39) shares the same conﬁguration of roles as the agnate plain transitive clause involving -WA ~ ø ‘give’—namely, two participant roles, an Actor/Agent and an Undergoer/Medium, as well as a non-participant Medium (McGregor 1994: 57; see also Rumsey 1982: 144 and McGregor 1990: 335). Although the applicative is clearly of the same valence and transitivity as the agnate non-applicative, it may be that the former construction establishes an association between the Undergoer/Medium and the non-participant Medium. This is perhaps what the speaker was trying to convey with the turn of phrase passed over in his translation: possibly that the recipient had strong rights to the object. (39) ngirrana-ngany they:gave:it-app ‘They passed something over to him.’ Figure 4 is an attempt to represent diagrammatically the two main patterns identiﬁed in this section for applicativised transitive clauses. (a) shows a schema for those

applicative constructions in warrwa

187

examples which involve the association of Mediums; the single arrow indicates that there is a single Undergoer, B, and that the action is directed and actualised through A, which serves as a type of intermediary. (b) is a schema for the rare examples showing an association between the Medium and Agent. 3.3 Concluding remarks The following conclusions may be drawn from the discussion of this section. First, despite being a verbal suﬃx, -ngany app marks a clausal, rather than a verbal category. Its meaning has to do with clausal transitivity in the widest sense of the word, embracing role structure. However, it is not simply a morphological ‘‘device’’ registering the addition of an extra participant role to the clause. Although it does do this in many cases, as we have seen, there are certain circumstances in which no new participant role is added and there is no apparent diﬀerence in transitivity between clauses with and clauses without verbs marked by the app suﬃx. This is especially the case for the applicative variants of transitive clauses. What -ngany app appears to signify is that the clause has a Medium which is closely associated with the Actor/Agent in applicativised intransitive clauses (which corresponds to the Actor/Medium of the agnate intransitive), and with another Medium in most applicativised transitive clauses—although on rare occasions the association is with the Actor. In this respect the comitative applicative construction appears to be ergatively oriented (see, however, below). The associated Medium is

Clause [comitative applicative]

Actor/Agent SoA Undergoer/Medium

linyju-na nguy jina-ngany-jirr kuya jirra policeman-erg return he:did-app-them mother their

&with Figure 5. Grammatical structure of example (6).

william mcgregor

188

Clause [comitative applivative]

Actor/ Undergoer/ SoA Agent Medium yirra-na wardal jirrmbil ngirrwanyjiny-ngany they-erg boss meet they:did:ref-app &with Figure 6. Grammatical structure of example (29). in close physical proximity to the Actor/Agent or Undergoer/Medium during the performance of the action, which association could be represented by means of either a comit or loc postposition attached to the relevant NP in an agnate clause —except in the case of reﬂexive/reciprocal clauses, where the expression would be by means of conjoined NPs. A more semantically revealing designation for the comitative applicative construction would thus be the associated medium construction.

Clause [comitative applicative]

SoA

Medium

Undergoer/ Medium Actor/ Agent

ngirrawulany-ngany baalu yirra-na kinya wamba they:tied-app tree they-erg this man

× at Figure 7. Grammatical structure of example (35).

applicative constructions in warrwa

189

McGregor 1997 proposes that the roles Actor, Undergoer, Agent, and Medium are experiential roles, and that the NPs and PPs that serve in them enter into constituency relationships within their clause. By contrast, the associative relationship is a logical relationship—a type of extension—realised grammatically by a dependency relationship. SG analyses of representative examples—(6), (29) and (35) above—are shown in Figures 5, 6 and 7 respectively. In these diagrams constituency relationships are represented in the usual way, and dependency relationships by arcs; &with indicates that the relationship is an associative subtype of extension, and ×at that it is a locative subtype of enhancement (see further McGregor 1997: 137ﬀ). The hand-held pencil indicates a grammatical marking relationship, signifying that -ngany marks the clause as being a comitative applicative. Most examples can, I submit, be accommodated in this characterisation. The main exceptions are the small number of examples, including (19) and (36), in which a comit 1 PP is found instead of an NP. It is not certain that the comit 1 PP in these examples discharges the grammatical role of Undergoer, although the clause does clearly have an Agent/Actor.

4. The instrumental applicative construction Instrumental applicatives agnate with reﬂexive/reciprocal intransitive and transitive clauses only (cf. the situation in Warrungu, where they are restricted to transitive clauses—Tsunoda (this volume). These two subtypes are discussed in order in the following subsections. 4.1 Reﬂexive/reciprocal clauses Again, just a few reﬂexive/reciprocal-marked verbs are attested with the applicative suﬃx: the inﬂecting verbs -KA ‘hit’, -BULA ‘tie up, get dressed’, and -WANDIWANDI ‘cover’; and the compound verb ngul . . . -BANYJI ‘spear one self/one another’. (40) and (41) are examples. (40) ngirrmangkanyjina-ngany kinya-ngany they:fought:ref-app this-inst ‘They bin ﬁghting with it now.’ (41) ngamawandiwandi-ngany blanket-ngany I:covered:ref-app blanket-inst ‘I covered myself with a blanket.’ There are, of course, agnate plain reﬂexive/reciprocal intransitive clauses. These

190

william mcgregor

may have an inst PP designating the instrument employed to eﬀect the action. But there is never an agnate reﬂexive/reciprocal clause whose Actor/Agent consists of the union of the two actant sets of the applicative clause, as is the case for comitative applicatives. Furthermore, the applicative clause remains intransitive, and admits both reﬂexive and reciprocal interpretations, as illustrated by (40) and (41) respectively. By contrast, comitative applicatives of reﬂexive/reciprocals are transitive, and (as far as I can tell) admit only reciprocal interpretations for the interaction between one of the participant groups. It seems that instrumental applicative clauses with reﬂexive/reciprocal verbs have three inherent experiential roles, an Actor/Medium, a SoA, and an Instrument, in contrast with reﬂexive/reciprocal intransitive clauses in which only the ﬁrst two are inherent. Here also -ngany apparently signals the presence of an additional inherent role in the clause—except that it is an Instrument rather than a Medium, and there is no eﬀect on clausal transitivity. It is suggested that -ngany app serves to indicate an associative relationship between the Actor/Agent and Instrument. There are insuﬃcient examples to make an entirely convincing case for this suggestion, although examples such as (41) are clearly consistent with it. (42) and (43) lend further credence to this suggestion: in the former the Actors appear to be closely related to the spears, as owners and manufacturers (in fact, the same holds true of (40)), while in the latter the clothing will obviously be closely associated with the wearer. (The inst PP designating the clothing in the second clause has presumably been ellipsed, as it conveys given information.) (42) kanka ngul ngirrwanjina-ngany kinya; kinya-ngany yina, that spear they:did:ref-app this this-inst his ‘They speared one another with those ones now.’ (i.e. the good spears which they had made).’ (43) balya mamawulanyjiny, mamawulanyjiny-ngany, clothes tying:to:ref tying:to:ref-app ‘He got dressed; he is wearing clothes now.’ 4.2 Transitive clauses Examples (44)–(47) illustrate the instrumental applicative construction for transitive clauses. (44) ngirrawuluna kinya jinal, kinya-ngany, they:tied this spear this-inst ‘They tied up the spear with this one [i.e. sinew].’

applicative constructions in warrwa

191

(45) wurr ngirrawarrany-ngany warli birrinybirri-ngany rub they:rubbed-app meat salt-com ‘They rubbed salt on the meat.’ (46) judu-ngany warli nganamarrany-ngany fryingpan frying:pan-inst meat I:cooked-app fryingpan ‘I cooked it in a frying pan.’ (47) biina jirrwadaj jan-ngany, bakal-ngany, ant:eggs yandy she:does-app coolamon-inst ‘She yandies (separates by shaking) the ant eggs in a coolamon. Unfortunately, not a lot can be said with certainty about this construction: examples are scarce, and almost all are elicited; moreover, opinions concerning the acceptability of apparently similar—sometimes, identical—examples vary considerably between the speakers, and even on diﬀerent occasions for a single speaker. The following remarks are therefore tentative, and demand further checking in the ﬁeld. Only a small subset of the range of ‘‘instrument’’ types designated by inst PPs are found in instrumental applicatives: some means (as in (45)), intermediaries (as in (44)), and vehicles ((46) and (47)). Other types of instrument, such as tools, body part instruments, animate instruments, and so forth never occur—and thus the unacceptability of (48) (rejected both with and without -ngany inst) and (49). There is a single exception in the corpus, (50), which I constructed from a corresponding simple transitive clause proﬀered by one of the speakers; it was accepted without hesitation. (48) *ngul nganamany-ngany waya (-ngany) spear I:put-app wire:spear (-inst) [‘I speared it with a spear.’] (49) *jurrmbul ngandiny-ngany mayi wila-ngany mix I:caught-app food water-com [‘I mixed ﬂour and water.’] (50) jurrmuk ngandiny-ngany buru punch I:caught-app ground ‘I hit the ground with the axe.’ It seems that it is only when there is a particularly close connection between the ‘‘instrument’’ and the Undergoer/Medium—normally close physical contiguity throughout the situation—that the instrumental applicative construction is acceptable. Such an association obtains for many mediums, intermediaries, and vehicles, but rarely for other types of instrument, which are more likely to be associated with

192

william mcgregor

the Actor/Agent. The acceptability of (50) may well relate to this observation: an axe is more likely to remain in physical contiguity with the ground as a result of hitting it, than with a branch or tree. Additionally, the associated instruments must usually be distinct, potentially individuated entities. Mass-entities such as water, sand, and the like do not maintain suﬃcient separateness to be said to be in associative relationships with the Undergoer/Medium—and thus the unacceptability of (49). It appears that the Instrument role is inherent in the transitive instrumental applicative construction, as it is in the corresponding reﬂexive/reciprocal intransitive applicative construction—and hence the appropriateness of the designation ‘‘instrumental applicative.’’ In fact, an inst PP is almost always present in instrumental applicative transitive clauses. (50) above is one of the few exceptions. In almost all cases speakers rejected my constructed examples which did not involve an overt inst PP, even when a suitable contender for the role of Instrument was available in the immediately prior discourse. Finally, there is one example in which what appears to be an expression designating an instrument in an applicative transitive clause is an unmarked NP rather than an inst PP: (51) ngirrawulany-ngany kinya baalu nalma-nyarri, nimala-nyarri jina, they:tied-app this thing head-comit 2 hand-comit 2 his ‘They tied him up with this thing, by the head, hands, and so on.’ Whether kinya baalu serves as a Medium or Instrument in this example is not obvious. Both possibilities make sense semantically: something used for tying may be conceived of as either being involved as a target of the action (the rope is most directly acted on by the tie-er) or as being involved instrumentally in bringing the situation into eﬀect. (51) might be either a comitative or instrumental applicative. 4.3 Concluding remarks The instrumental applicative suﬃx -ngany has no eﬀect on the transitivity of a clause. However, instrumental applicative clauses do show one more inherent role than their plain (non-comitative) agnates, an Instrument—i.e. a non-participant role—which is, of course, permissible in the plain agnate clause. It is important to observe that the instrumental applicative of Warrwa does not involve anything akin to ‘‘advancement’’ of the Instrument to ‘‘object,’’ as in some Pama-Nyungan languages of Australia which show such a construction—e.g. Warrungu (Tsunoda [this volume]), Kalkatungu (Blake 1979: 88), Dyirbal (Dixon 1972: 95), and Yidiny (Dixon 1977: 310). Additionally, the instrumental applicative clause contains an inherent dependency

applicative constructions in warrwa

193

Clause [instrumental applicative]

Actor/ Agent

SoA

Instrument

O ngirrmangkanyjina-ngany kinya-ngany they they:fought:ref-app this-inst

&with Figure 8. Grammatical structure of example (40). relationship, linking the Instrument with the Actor/Agent of an intransitive reﬂexive/reciprocal clause, and the Undergoer/Medium of a transitive clause—ergative patterning again. This dependency relationship may be an associative one, as appears to be normally the case in reﬂexive/reciprocal and in a few transitive clauses;

Clause [instrumental applicative]

Actor/ Agent

SoA Instrument

Undergoer/ Medium O biina jirrwadaj jan-ngany bakal-ngany she ant:eggs yandy she:does-app coolamon-inst

× at Figure 9. Grammatical structure of example (47).

194

william mcgregor

or it may be a locative one, as appears to be the case in most transitive clauses (e.g. (46), (47), and (50) above). Figures 8 and 9 provide diagrammatic representation of the grammatical structures of reﬂexive/reciprocal intransitive and transitive instrumental applicative constructions.

5. Further remarks on the grammar of Warrwa applicative constructions Applicatives are generally seen as argument changing and/or valence increasing processes which increase the transitivity of a verb or clause (e.g. Comrie 1985, Baker 1988, Austin forthcoming). It is frequently considered that applicatives involve the promotion of some ‘‘underlying’’ non-argument role to ‘‘object’’ (Undergoer), which will be accompanied by a demotion of the ‘‘underlying object’’ (if there is one) to a non-argument position. I reject the derivational account of applicatives, for both theoretical and descriptive reasons. First, it makes no sense in SG, where (as in other functionally-oriented grammars) structure-changing rule are not permitted (see also Shibatani 1996: 162). The applicative constructions can and should be treated as constructions in their own right, and in no way derivative.7 Although applicative constructions are more marked semantically/functionally and in terms of frequency than their plain agnates, this does not argue for the derivational approach; it can be seen as a correlate of the formal markedness typical of the applicative. Second, the derivational account does not provide an adequate description of the facts of Warrwa. As we have seen, comitative applicatives are frequently higher in transitivity than their non-comitative agnates (e.g. when the latter are intransitive). However, it is not the case for comitative-applicativised transitive clauses, many of which show the same role structure as the corresponding plain transitive clauses; it is diﬃcult to see how a derivational account could account for examples such as (37) and (39) consistently with (33)–(35) and applicativised intransitives and reﬂexive/reciprocals. Nor does the derivational account throw any light on the semantics of the applicative constructions. For an SG account of applicative constructions it is necessary to provide a characterisation in terms of the semiotically signiﬁcant grammatical relationships which are involved. This is possible along the following lines. Comitative applicatives, show two crucial characteristics: they are always transitive (and thus satisfy the characterisation of Section 2), and they always involve an inherent de-

applicative constructions in warrwa

195

Experiential structure Clause [comitative applicative]

Actor/ Undergoer/ SoA Agent Medium

Logical structure

&with / × at Schema for comitative applicative construction. pendency relationship (as per McGregor 1997) between two entities in the clause. The dependency relationship is typically an extending one, a close associative relationship; occasionally it is an enhancing relationship of spatial location. Both characteristics are crucial. Contra Shibatani (1996: 165), they must satisfy more than the ‘‘prototype of transitive clauses.’’ It is the inherent presence of a particular sort of dependency relationship which distinguishes comitative applicatives from ordinary transitive clauses. Comitative applicative clauses thus must satisfy the characterisation shown in Figure 10. Of course, there may be other grammatical relationships as well; what are shown are the necessary ones.) Clearly this is too unconstrained: no restrictions are imposed on the units that may be linked by the dependency relationship, although we have seen that there are such restrictions. These may be characterised as follows: (52) The dependency relation always obtains between a Medium and another Medium if there is one, or otherwise, with an Agent. This permits us to account for almost all examples cited in sections 3 and 4 above. In particular, problematic examples such as (37) and (39), which show no diﬀerence

196

william mcgregor

Experiential structure Clause [instrumental applicative]

Actor/ Agent

Instrument SoA

Logical structure

&with / × at Figure 11. Scema for instrumental applicative construction.

from ordinary transitive clauses in terms of their conﬁguration of participant roles, are rendered explicable: for the former, there is a single Medium in the clause, and so the dependency relationship must be between it and the Agent; for the latter, there are two Mediums, the recipient and the gift, and so these must be linked by the dependency relationship. Furthermore, this shows that the apparent ergative orientation of the comitative applicative is an epiphenomenon. The only examples that are not accounted for now are those rare ones which, like (19) and (36), have a comit 1 PP instead of an unmarked NP. (52) can be modiﬁed (admittedly in an ad hoc way) by adding in the qualiﬁcation that if there is a comit 1 PP in the clause the Medium must be related to it by the dependency relationship. The instrumental applicative construction can be characterised in almost identical terms, as shown in Figure 11, and the stipulation of (53). The dependency relation always obtains between the Instrument and the Medium if there is one, or otherwise, with the Agent. The discussion of this section supports Shibatani’s suggestion (1996: 173) that applicative constructions are closely related to benefactives. The critical diﬀerence

applicative constructions in warrwa

197

lies in the nature of the dependency relationship: for benefactives it is an extending relationship of possession, rather than of association.

6. Conclusions In this paper we have discussed the semantics and grammatical structures involved in the two Warrwa applicative constructions, the comitative and instrumental, and seen that they are closely related in both respects. Semantically, they share the property that a close association is asserted to hold between one role in the clause—usually a Medium, in the case of comitative applicatives, an Instrument in the case of instrumental applicatives—and some other role, normally a participant role. Cognitive Linguistics style diagrams provide convenient representations of the meaning of the applicative constructions. Grammatically, both have, in addition to the normal inherent experiential roles, an inherent logical relationship, which obtains between a pair of these experiential roles, one of which is normally a Medium. We have further argued that applicatives in Warrwa are not derived constructions, and nor do they always involve valence increase through the addition of a new role to the clause. As mentioned at the beginning of the paper, the verbal morpheme -ngany which marks the applicative construction is phonologically identical with a nominal postposition which marks instrumental case, and a verbal suﬃx which marks the category implicated. This raises the question as to whether -ngany represents a single morpheme with a number of perhaps related uses, or two or more homophonous morphemes. Considerations of length have precluded investigation of this interesting question. I think it most likely that two morphemes should be identiﬁed in the modern language, but they almost certainly derived historically from a single comitative morpheme. This story will be told in a future publication.

Notes 1. This is a revised version of a paper presented to the Linguistics Department, Monash University, 26th May 1995 (McGregor 1995) I am grateful to the audience, particularly Keith Allan, for useful comments, and to Tasaku Tsunoda for many detailed and helpful comments on an earlier draft. The ﬁeldwork on which this investigation is based was supported by grants from the Australian Research Council (Grant A58930745 and A59332055), and the paper was written under an ARC Research Fellowship (A9324000). My greatest debts are of course to my Warrwa teachers, Maudie Lennard and Freddy Marker—to whom I owe many apologies for asking far too many very stupid questions about -ngany!

198

william mcgregor

2. Thus the binary distinction between transitive and intransitive clauses traditionally made in Australian linguistics is inadequate (see also McGregor 1990: 317ﬀ, and Austin forthcoming, Section 21). Furthermore, in Warrwa (as in other languages in the vicinity) transitivity is a property of clauses, not verbs. 3. Inﬂecting verbs are cited in all capitals; grammatical roles are indicated by initial capitals. A comma is used to mark the end of an intonation unit (this usually corresponds to a pause) in textual examples; and a slash is used to indicate conﬂation of grammatical roles. The numbers 1, 2 and 3 represent the three person categories. 4. There are, it should be noted, no examples to hand of agnate clauses involving nguy ‘return’ with a -nyarri comit 2 PP, although it is expected that they should be permissible, and that they should convey the meaning that the Actor returned in someone’s company, where this is a person of similar status to the Actor, and/or both are equally in control of their actions Thus, it is expected that the following is acceptable, and conveys no suggestion of coercion by the policeman. Note that there exist other clauses of motion involving comit 2 PPs. (i) ?linyju nguy jina kuya-nyarri jirra, policeman return he:did mother-com 2 their ‘The policeman came back with (in the company of) their mother.’ 5. This of course adds support to my contention that applicative clauses of motion are not causatives, and it is the Actor/Agent that is the one who is primarily asserted as moving 6. It is possible that this is an instrumental applicative (see Section 4), rather than a comitative applicative, and that the instrumental PP—‘with/by horse’—has been ellipsed This, however, seems highly unlikely given the textual environment. 7. This is why I have spoken throughout this section of agnate applicative and non-applicative clauses, even when this results in somewhat cumbersome expression The point is too fundamental to fudge the issue by employing misleading, albeit convenient and evocative, wording.

References Austin, Peter. Forthcoming. ‘‘Causatives and Applicatives in Australian Aboriginal Languages.’’ To appear in Toru Hayasi & Kazuto Matsumura (eds), Dative and Related Phenomena. Tokyo: Hitsuji Shobo, 1–61. Baker, Mark. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Blake, Barry. 1979. A Kalkatungu Grammar. Canberra: Paciﬁc Linguistics. Comrie, Bernard. 1985. ‘‘Causative Verb Formation and Other Verb-Deriving Morphology.’’ In Timothy Shopen (ed.), Language Typology and Syntactic Description, vol. 3. Cambridge: Cambridge University Press, 309–348. Dixon, Robert M. W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press.

applicative constructions in warrwa

199

——. 1977. A Grammar of Yidi. Cambridge: Cambridge University Press. Langacker, Ronald. 1987. Foundations of Cognitive Grammar, vol. 1. Stanford: Stanford University Press. McGregor, William. 1990. A Functional Grammar of Gooniyandi. Amsterdam: John Benjamins. ——. 1992. ‘‘The Semantics of Ergative Marking in Gooniyandi.’’ Linguistics 30: 275–318. ——. 1994. Warrwa. Munich: Lincom Europa. ——. 1995. ‘‘Nominal and Other Uses of the Instrumental Postposition in Warrwa and Other Nyulnyulan Languages.’’ Paper presented to Linguistics Department, Monash University, May 1995. ——. 1997. Semiotic Grammar. Oxford: Oxford University Press. Rumsey, Alan. 1982. An Intra-Sentence Grammar of Ungarinjin, North-Western Australia. Canberra: Paciﬁc Linguistics. Shibatani, Masayoshi. 1996. ‘‘Applicatives and Benefactives: A cognitive account.’’ In Masayoshi Shibatani and Sandra A. Thompson (eds.), Grammatical Constructions: Their form and meaning. Oxford: Clarendon Press, 157–194. Stokes, Bronwyn and William B. McGregor. Forthcoming. ‘‘Classifying the Nyulnyulan languages.’’ In Nicholas Evans (ed.), Comparative Studies in Non-Pama-Nyungan. Canberra: Paciﬁc Linguistics. Tsunoda, Tasaku. [this volume] ‘‘Applicative Constructions in Warrungu of Australia.’’

From event sequence to grammar: Serial verb constructions in Kalam ANDREW PAWLEY and JONATHAN LANE Australian National University

1. Introduction The name ‘‘Kalam’’ is used of two closely related languages spoken by about 20,000 people living around the junction of the Bismarck and Schrader Ranges on the northern margins of the Central Highlands of Papua New Guinea.1 These languages belong to the Madang subgroup of the large Trans New Guinea phylum, which includes perhaps two thirds of the 750 or so Papuan languages of New Guinea. The two Kalam languages are known to their speakers as Etp Mnm and Ti Mnm, respectively.2 This paper will use examples only from Ti Mnm, which we will refer to simply as Kalam.3 Kalam speakers make heavy use of serial verb constructions (SVCs), in which one or more bare verb stems precede a verb inﬂected for tense/aspect/mood and subject reference. There is in principle no grammatical limit to the number of verb stems that may be combined in an SVC, though in practice the upper limit is around 9 or 10. In (1) and later examples the verb stems appear in bold face both in the Kalam text and in the English morpheme-by-morpheme glosses.4 Hyphens denote morpheme breaks. (1) Bin ak ñapana anup sop ak wik d ap tan woman the child him soap the rub touch come descend d ap yap g-e-b. touch come ascend do-prs:prog-3sg ‘The woman is soaping her child.’ After surveying the main types of SVCs in Kalam we will focus on one much-used type, the ‘‘multi-scene’’ SVC, which has no counterpart in the literature on serialisation. A multi-scene SVC speciﬁes a series of events beginning with the actor

202

andrew pawley and jonathan lane

moving to the scene of a pivotal action in the series. The ﬂavour of such constructions is indicated by (2). (2) Ognap am mgan kti kapkap su d am l-p-al. Sometimes go inside they quietly bite get gop ut-perf-3pl ‘Sometimes they go and sneakily burrow out through an escape tunnel.’ Several questions will be asked about multi-scene SVCs. What are their functions? How have they arisen? How do they ﬁt into the typology of construction types developed by Foley and Olson (1985) and Foley and van Valin (1984)? We argue that SVCs illustrate the general principle that speakers will ﬁnd short and standardised ways of reporting highly recurrent complex events and situations. However, the kinds of information which must be explicitly mentioned in well-formed reports of events diﬀer considerably across languages. A central argument will be that multiscene SVCs preserve the essentials of Kalam discourse structure rules while achieving a high degree of economy and conventionalisation. In many respects SVCs, and especially multi-scene SVCs, look like sequences of clauses that have been compressed into a single clause. However, the constraints on the kinds of peripheral arguments and other material that can intervene between verb stems in multi-scene SVCs are not so strict as in canonical clauses. Yet, paradoxically, particular SVC strings also exhibit some of the properties of lexemes. The conventional breakdown of particular utterances or bits of text into mutually exclusive types of unit—‘‘lexeme,’’ ‘‘clause structure’’ or ‘‘discourse structure’’—is seen to be artiﬁcial in this case. Kalam SVCs represent a powerful argument for seeking a new model of linguistic analysis, one which links constraints on discourse structure, syntax, idiomaticity and the formation of lexemes.

2. Notes on Kalam grammar and types of serial verb construction A few summary notes on Kalam grammar are in order. Case-marking is nominativeaccusative. Preferred word order is SOV but is not rigid. OSV and SVO constructions occur. Locatives follow the verb as often as they precede it. Verb stems are clearly diﬀerentiated from other word classes on morphological grounds. Verbs can appear either uninﬂected (that is, as bare stems), e.g. am- ‘go’ or inﬂected, e.g. am-l ‘having gone’, am-eb-yn ‘I am going’. Inﬂected verbs can be further subdivided into dependent and independent verbs. Independent verbs are fully speciﬁed for tense/mood/aspect (TMA) and for person and number of the subject, e.g. am-eb-in (go-pres:prog-1sg) ‘I am going’ am-elgp-al (go-past: habitual-3pl) ‘they used to go’. A clause whose ﬁnal verb is independent is an independent clause. The suﬃxes on dependent verbs specify two things: one, whether the subject of

from event sequence to grammar

203

the dependent verb is the same as (SS) or diﬀerent from (ds) the subject of the next inﬂected verb and two, relative tense: whether the action of the dependent verb occurs prior to (prior), simultaneous with (sim), or future to (fut) the action of the next inﬂected verb. A clause whose inﬂected verb is dependent is a dependent clause. Since Kalam normally requires the identity or non-identity of subject referents to be marked on non-ﬁnal clauses (other than embedded clauses), we can say it uses a switch reference system when building complex sentences, of the same general kind as is found in many other Papuan languages (Roberts 1997). (3) Same subject, Prior Action (ss:prior) Np n-l, a-b-al. 2sg:obj perceive-ss:prior go-perf-3pl ‘Having seen you they went (earlier today)’/‘They went after seeing you.’ (4) Different Subject, Prior Action (ds:prior) Kun g-e-y, sl ag-e-b thus do-ds:prior-2sg weeping say-pres:prog-3sg ‘She is weeping because of what you did. (You having done thus, she is weeping.)’ As in many Papuan languages of the Trans New Guinea family, Kalam discourse often contains long chains of dependent clauses—i.e. clauses containing dependent verbs, with marking of relative tense relative identity of subject— followed by an independent clause, with absolute marking of tense and subject person/number. Kalam is somewhat unusual in having a closed verb class numbering about 125 verb stems.5 A small subset of these verbs, known as generic verbs, are characterised by their very high frequency—some 15 generic verbs account for nearly 90 per cent of all verb tokens in texts—and their broad meanings. The most common verb stems (with very approximate glosses) include ag- ‘make a sound, say’, am- ‘go’, ap- ‘come, appear’, d- ‘control, constrain, hold, get’, g- ‘do, act, make’, jak- ‘attain an elevated or targeted point: stand, rise, attain, arrive, reach’, l- (ay- in Etp Mnm) ‘stabilise, become, put’, md- ‘exist, stay, live’, n-‘perceive, know, be aware, see, hear, etc.’, ñ- ‘transfer, give, connect, apply’, ñb- or ñ- ‘consume, eat, drink, etc.’, pak- (pk- in Etp Mnm) ‘strike, kill’, tan- ‘rise’, yap- ‘descend’, tk- ‘cause a hiatus, separate, sever, cross, change suddenly’. All verb stems can stand alone as the head of a clause. However, it is common for a verb stem to be paired with a verb adjunct. An adjunct is a noun-like or adjective-like element that always occurs followed by a generic verb, whose meaning it modiﬁes, e.g. the verb ag- ‘make a sound, say’ occurs with many adjuncts denoting kinds of sounds, such as bu ‘explosion’, gu ‘thudding’, si ‘crying (weeping)’, sabok ‘whistling’, wal ‘scream’, as bu ag- ‘to explode’, si ag- ‘to cry’, etc.

andrew pawley and jonathan lane

204

In SVCs an inﬂected verb stem follows one or more bare verb stems. The lexical core of an SVC is a sequence of verb stems (SVS), one or more of which may also be paired with a verb adjunct. Under certain circumstances non-verb material (modiﬁers or nominal arguments other than verb adjuncts) may intervene within the string of verb stems. For example, in the SVC in (5) the object NP mon ‘wood’ follows am ‘go’, the ﬁrst verb stem in the SVC. (5) B ak am mon p-uk d ap la-k. man that go wood hit-smash get come put-3sg-pst ‘The man fetched some ﬁrewood.’ It is useful to divide SVCs into single scene and multi-scene SVCs. Single scene SVCs have the semantic property that they refer to a series of acts which take place at the same scene (or site). Multi-scene SVCs refer to a series of acts which take place at diﬀerent scenes (or sites), i.e. the subject moves from one place to another. The two types also diﬀer to some extent in their syntactic and prosodic restrictions. Some examples of simple SVSs follow: (6) aglagñag tkag nñb nd npak npak wkap yap pakpui ju yokpui md aypui pag ykyk ask ask aywog g ym ñ-

(say stabilise)

‘ask to remain, make an appointment’ (say transfer) ‘tell, inform’ (say sever) ‘interrupt’ (say perceive)’ ‘request, ask, inquire’ (consume perceive) ‘taste’ (hold perceive) ‘feel (by touching)’ (strike perceive) ‘nudge’ (strike shatter) ‘smash’ (come fall strike) ‘fall down’ poke withdraw displace) ‘prise out’ (poke stay stabilise) ‘insert, ﬁx in’ (poke disturb open) ‘prise open’ (open free free stabilise) ‘open up, reveal’ (garden do plant eat) ‘make gardens, be a farmer’

Some examples of multi-scene SVSs: (7) d apd amam d apd am yokap jak amd ap tan d ap yap g-

(hold come) (hold go) (go hold come) (hold go displace) (come arise go) (get come ascend get come descend do)

‘bring’ ‘take’ ‘fetch’ ‘get rid of’ ‘come up, emerge’ ‘go back and forth, go up and down’

from event sequence to grammar

205

Single scene SVCs are subject to all of the syntactic and prosodic restrictions listed in (8) below. Most of these have been observed in a number of West African, Southeast Asian, Oceanic and other Papuan languages and can be assumed to be characteristic of serial verb constructions worldwide. (8) Deﬁning features of single scene SVCs in Kalam (a) There is only one subject, coded on the inﬂected verb, and sometimes also by one overt NP. (b) There is only one inﬂection per SVC. (c) No subordinating or coordinating morphemes occur within the construction. (d) There is only one negative morpheme per construction. (e) The order of the verbs stems must match the temporal order of the actions which they denote (f) The SVC is spoken under a single intonational contour, without perceptible internal pause between stems. (g) Transitive verb stems share an object. (h) Stems are not normally separated by NPs or adverbs such as kasek ‘quickly’. In general, the only type of non-verb material that can occur within the string of verbs in simple SVCs consists of modiﬁers known as ‘verb adjuncts’. These are adverbial or nominal complements which are added to generic verbs to make more speciﬁc, or at least diﬀerent meanings, e.g. wsn ‘sleeping’ in the collocation wsn n- (sleeping perceive) ‘dream’. These adjuncts (an ill-deﬁned class which can be broadened to include tep ‘good, well’ in a recurrent construction type tepg- ‘do well’) form constituents with the following verb stems. Adjunct + verb stem combinations are in some circumstances distributionally equivalent to single verb stems. (i) None of the non-verbal morphemes which can be postposed to inﬂected verb stems (such as tek ‘like, positive question marker’, aka- ‘or, negative question marker’, or the conjunctions yp ‘with’ or pen ‘but’) can follow a bare verb stem. All serial verb sequences are probably lexicalised to some degree. One way in which the language expands its small repertoire of fully lexicalised verb meanings is by combining two or more purely lexical verb stems without intervening material. to form a conventional expression. Thus tb ‘cut’ combines with tk’separate, sever’ to create tb tk- ‘cut oﬀ’: (9) Mab alk tb tk-eb-in. tree branch cuts ever-prs:prog-1sg ‘I’m cutting oﬀ the branches.’ The features of lexicalisation which they exhibit are, ﬁrstly, those listed in (8a–i).

206

andrew pawley and jonathan lane

In addition, most if not all are stereotyped expressions, each being a standard way of expressing a familiar complex idea. Sometimes the participant stems of a compound SVC can or have coalesced phonologically into a single word, as for example, with p-wuk ‘smash, break up’, which derives from pak’hit’ (pk in Etp Mnm) plus wuk shatter.’ A special type of single scene SVC is that termed the g-support construction. In this the initial elements are a verb stem plus the modiﬁer tep ‘well, good’ and the ﬁnal element is the verb g- ‘do, make, create, work, etc.’, as e.g. md tep g- (exist well do) ‘prosper, do well’, n tep g- (perceive good do) ‘consider, think over, know (sth) well’ and pk tep g- (hit well go) ‘hit (sth.) well’. We may also mention in passing grammaticalised SVCs. In these, the ﬁnal verb stem can be seen as carrying out a grammatical role. For instance, d- ‘constrain, control, get, hold, etc’ can act as a completive aspect marker when it occurs as the last verb of an SVC: (10) mnm ag d-p-al. word say get/complete-perf-3pl ‘They have ﬁnished talking.’ While it seems clear, on intuitive semantic grounds at least, that d- in the previous example has become grammaticised in this function, in other cases the distinction between literal meanings and grammatical functions of serialised verbs, e.g. between md- meaning ‘exist, stay, persist’ and md- marking persistence of an action (but not simple progressive, which is marked by a verbal suﬃx) is less clearcut. For instance, compare mnm ag md-p-ay ‘they are still talking, they remain talking’ with (10). Features (a–i) in (8) all demonstrate ways in which certain types of Kalam SVCs resemble the prototypical clause, at least prosodically and syntactically. But ‘‘multiscene SVCs’’ are not bound by all the restrictions in (8), speciﬁcally not by (f)–(i). Multi-scene SVCs are prime examples of the interaction between the syntactic rules of Kalam and its discourse structure. We will return to this class of constructions after a brief survey of some ideas about SVCs in the general literature.

3. Some notes on the literature on SVCs In our use of the term ‘‘serial verb construction’’ do we mean the same thing as other writers on the same topic? Should the diverse range of entities for which this term is used really be grouped together? Let us take a brief look at other studies of verb serialisation.

from event sequence to grammar

207

First of all, it should be noted that the literature on serialisation is rather sprawling, covering a wide range of languages, data and theoretical approaches. Serialising languages have been identiﬁed in locations and language groups as diverse as Benue-Kwa (West Africa) and Atlantic and Indian Ocean Creoles, through various Dravidian and South-east Asian languages (including Thai and Mandarin Chinese), to Papuan languages, varieties of Melanesian Pidgin, and some Oceanic languages. The criteria used to deﬁne SVCs have varied from being quite restrictive to quite broad in scope. Yet it is possible to identify common features in the constructions that have been identiﬁed as SVCs. Further, there are common threads in the recent literature. Much recent work on serialisation can be seen as stressing the interaction between serialisation and the deﬁnition of the clause. Early descriptive work on serialisation concentrated on the Benue-Kwa languages of West Africa. The earliest work in generative grammar frameworks, in the late 1960s and early 1970s, tended to view at least some types of SVCs as transformationally derived from underlying multi-clause constructions. Later treatments argued for the basegeneration of all SVCs, typically from conjoined or embedded VPs. More functionally orientated work by Foley and others (e.g. Foley and Van Valin 1984, Foley and Olson 1985, Foley 1986, 1991, Crowley 1987) considers serialisation as one of the possible ways of joining ‘‘layers’’ of the clause together. This can be seen as part of an increasingly widespread view of SVCs as part of a continuum between singlepredicate constructions and constructs involving multiple clauses. Similar views have been advanced by Bruce (1984) and Pawley (1987). Givón (1990) reaches similar conclusions on the basis of a psycholinguistic experiment. The distribution of pauses within SVCs and other methods of combining verbs in three languages from Papua New Guinea, including Kalam, Tairora and Tok Pisin, were analysed. One of Givón’s ﬁnding was that there was a continuum of pause probability from SVCs through SS-marked clauses to clauses marked with more ﬁnite morphology. The notion of SVCs as somewhere between a single-verb clause and a set of conjoined clauses is also implicit in much recent generative grammar work, since it tends to claim that the multiple verbs in an SVC belong ultimately to a single verb phrase. In the previous section, we identiﬁed a number of diﬀerent types of SVC in Kalam. We initially made a division between simple and multi-scene SVCs. Simple SVCs are, we claimed, subject to a wide range of constraints (a–i). These restrictions are shared by other serialising languages. (6a–i) can be recast in the following manner (which closely follows the wording of Bradshaw (1982: 28) and Crowley (1987: 38, 1990: 60–1): (11) SVCs are constructions in which there is more than one verb and, in addition:

208

andrew pawley and jonathan lane

(i) There is no contrast in verb inﬂections (e.g. for categories such as tense, mood, aspect, person/number of subject or actor (and sometimes other grammatical or semantic roles), directionals, transitivity and negation). (ii) No overt morphemes characteristic of clause boundaries are present (i.e. no subordinating/coordinating/switch reference morphology). (iii) There is no intonational evidence of clause boundary. (iv) There are restrictions on core arguments (actor and undergoer), according to two main patterns (either both actors/subjects are identical, or the undergoer of the ﬁrst verb becomes the actor of the next). Further, for an identiﬁed subclass of SVC, (v) Negation and adverbs whose scope is the (prototypical) clause have scope over all verbs in the construction. These criteria cover at least an important subset of the entities which have been identiﬁed as SVCs. The way in which they are implemented in the wide range of serialising languages, which have diverse morpho-syntactic properties, will necessarily be far from uniform. For instance, in Kalam, according to the deﬁnition given earlier, there can only be one inﬂected verb per SVC. This exempliﬁes one common method of implementing restriction (i) above (no contrast in inﬂectional categories). The other major pattern is for all verbs to share the same inﬂection, as in the following example from Akan (A Kwa language of West Africa): (12) Mekooe mebaae. I:went(pst) I:came(pst) ‘I went and came back.’ (Schachter 1974: 259) Other diﬀerences between serialising language are created by the possibility of coding the same types of grammatical category using either bound or free morphemes. More signiﬁcant than such details of implementation is the issue raised by point (v) above, which suggests that subtypes of serialisation can be identiﬁed with respect to the scope of negatives and adverbs. An example is given by Barai, a Papuan language, which has been identiﬁed as possessing two types of serialisation, termed nuclear serialisation and core serialisation. In nuclear serialisation, the negative morpheme must precede the ﬁrst verb in the SVC, and the semantic scope of negation is necessarily over both verbs: Nuclear serialisation: (13) a. Fu fase naebe ﬁ isoe. he letter neg sit write ‘He did not sit and write a letter’

from event sequence to grammar

209

b. *fu fase ﬁ naebe isoe he letter sit neg write In core serialisation, on the other hand, either verb may be negated separately: Core serialisation: (14) a. Fu naebe ﬁ fase isoe. he neg sit letter write ‘He did not sit down, but did write a letter.’ b. Fu ﬁ fase naebe isoe. he sit letter neg write ‘He sat down, but did not write a letter.’ (Foley and Olson 1985: 40) The distinction between nuclear and core serialisation, used by Foley and other writers, including Crowley (1987) and Durie (1997), can be interpreted in the following way. nuclear serialisation joins verbs together to create a complex predicate with a single set of arguments. core serialisation joins groups of verbal inner arguments into a single construction, subject to the constraint that all the verbs in the construction share at least one inner argument. The problem with classing core SVCs in Barai as a type of serialisation is that they violate more than just condition (v) above. For instance, the condition that there is only a single intonational contour in SVCS (condition iii) applies to nuclear SVCs in Barai, but core SVCs have two partially overlapping intonation contours. Other exceptions to conditions (i)–(v) can be found in serialising languages. For instance, individual verbs in SVCs may take their own aﬃxes, as in the following example from the Papuan language Yimas. (15) Wa-yarim-ak-ni-n-m-ko imp-dir-get-go-2sgA-3plU-dir ‘Get them toward me and go [up there]’ (Foley and Van Valin 1984: 248) In this nuclear SVC, ak ‘get’ is modiﬁed by the directional preﬁx yarim, while ni ‘go’ is modiﬁed by a separate directional, the suﬃx ko. At the same time, this group of verbs seems to obey the other conditions (ii)–(iv)—they are not separated by clause-joining morphemes, they are part of the same phonological word, and they share the same actor. Even condition (ii), which seems on the face of it to be a necessary condition for serialisation, is not met by some constructions which have been classed as SVCs. For instance, Yimas allows constructions which behave rather like the combinations of roots shown in the previous example, but in which constituent verbs are not juxtaposed but rather are separated by the suﬃx -mpi, which also doubles as a sequential action suﬃx in the clause-chaining system. Foley (1991) classes such constructions as SVCs provided the verb roots belong to the

210

andrew pawley and jonathan lane

same phonological word – in our terms, as long as they conform to condition (iii). Constructions such as the previous two examples, which consist of juxtaposed verbs but which constitute apparent exceptions to some of the criteria given above, make a hard-and-fast universal deﬁnition of serialisation diﬃcult to obtain. One possible solution to this dilemma is to consider conditions (i)–(v) as a pool of features, possibly with diﬀerent weights, from which SVCs draw. A necessary consequence of such an approach is that the boundaries between SVCs and other types of constructions are not clearly set. Unfortunately, while such fuzziness may well be part of the reality of human language, it is not necessarily very appealing to most linguists, who tend to prefer things to be more precise. How do Kalam SVCs ﬁt into this vaguely delineated territory? We said earlier that simple Kalam SVCs were subject to a range of restrictions, essentially those recast in more general terms as (i)–(v) above. Since they obey all these restrictions, they can be viewed as essentially prototypical SVCs. More complex Kalam SVCs are more problematic. To start oﬀ with, verb stems can be repeated, along the lines of: . . . am am am am am am ng-a-k . . . go go go go go go see-3sg-past ‘. . . (he) went a long, long, long way and saw . . .’ (KHT I #114) Although such repeated verbs have the combinatorial possibilities of single verb stems, doubts have been raised about whether they should be treated as SVCs. And indeed, the same question applies to multi-scene SVCs, as we will see later. Since the cut-oﬀ point between simple and multi-scene SVCs is not a sharp one, and since all the constructions in which bare verb stems precede an inﬂected verb in Kalam share features (6a–e), we treat both simple and multi-scene constructions as SVCs in this paper.

4. Multi-scene serial verb constructions: from discourse to grammar 4.1 Multi-scene SVCs Multi-scene SVCs diﬀer from the simpler types of SVC introduced above in that they more closely resemble various types of syntactically productive clause sequences. At the same time, while we contrast the complex multi-scene SVC with the simple compound and grammaticalised types discussed above, it should be noted that there is no clear cut-oﬀ point between the types, or between the notion of lexeme and that of syntactically productive construction in Kalam.

from event sequence to grammar

211

A multi-scene SVC of relatively simple structure was given in (2) above—this contained a string of ﬁve verbs interrupted only by a single locative and adverbial modiﬁer. Five more examples follow. The ﬁrst, (16), is only slightly more complex in structure than (2). It contains a sequence of ﬁve verb stems with a single nominal (kab ‘stone’, the object of agl ‘heat’) interrupting the sequence. The ﬁve verbs share the same subject (cn ‘we’). Three of the transitive verbs share the same object (wgi ogok ‘the bandicoots’). (16) . . . ksen cn wgi ogok d am kab agl ad ñb-lg, . . . . . . later we bandicoot the:pl get go ovenstones heat bake eat-ss:sim ‘. . . later (when) we take the bandicoots and go and heat ovenstones and cook and eat (the bandicoots), . . . ’ The next four show rather more material intervening between some verbs—a locative phrase in (17), a locative object in (18), a complex locative phrase in (19) and a complex locative phrase and a patient object in (20). (17) . . . mj bep tk d ap nb okya yok-l, . . . . . . leaf plant:sp break get come place below displace-ss:prior ‘ . . . having broken oﬀ bep (and) brought (and) tipped (it) in below, . . . ’ (KHT: : I #72) (18) . . . su d am d am d am, kje day alud yk-l, . . . . . . bite get go get go get go, path section above open-ss:prior ‘. . . it burrows on and on and opens up a tunnel [kje day] above its nest, . . . ’ (19) . . . d am ñg klam klam okok d ap ym-e-l, . . . . . . get go water spring spring around get come plant-ds:prior-3pl ‘. . . they took (it) to springs thereabouts and brought it and planted (it), . . . ’ (20) kmn tma ak am Sapkoy-Sagaym kab alim game:mammal cuscus this go S-S [place name] rock down:valley kañm ak ñb-l, . . . banana this eat-ss:prior ‘This cuscus, having gone to the rocks down below at Sapkoy-Sagaym and eaten the bananas, . . . ’ We will take a closer look at the syntactic structure of multi-scene SVCs in Sections 4.3–4.5. But ﬁrst a brief excursion into the domain of Kalam discourse structure conventions.

212

andrew pawley and jonathan lane

4.2 Some Kalam discourse rules for reporting events To understand the nature of Kalam SVCs it is necessary to understand how the Kalam report events.6 We will deal here only with the kinds of event reports immediately relevant to our concerns, namely those which speakers treat as wilful actions, or as a sequence of wilful actions,7 such as those actions we might gloss, in English, as ‘make a garden’, ‘collect ﬁrewood’, ‘gather pandanus fruit’, or ‘hunt possums’. A minimal well-formed report of a single deliberate action in Kalam should make reference to a sequence of associated actions. Speciﬁcally it should answer the questions: 1. Did the actor move to the scene of the action in question (call it the ‘‘pivotal action’’) or was he already there? 2. What did he do there? 3. Did he move on to another place and if so, did he take the aﬀected object with him? 4. What was the outcome, e.g. how was the aﬀected object disposed of? The conventional realization of conceptual elements 1–4 in Kalam discourse is summarized in the following formula. (21) Core elements of deliberate action event reports 1 2 3 4 move/stay pivotal act move (with outcome affected object) Such an analytic treatment of familiar events is very diﬀerent from English, which favours a metonymic reporting. That is to say, in English (and most languages) it is usual to let a single component action stand for a familiar complex event in which a single actor performs an integrated series of actions. For instance, in answer to the question ‘‘What did you do today?’’ one can simply say ‘I went to the supermarket/ movies/doctor’’ without specifying any of the essential consequences, or ‘‘John and I baked bread/cut ﬁrewood/wrote letters’’ without mentioning any of the prerequisite or consequent actions.8 In Kalam, however, you should say what you did at the supermarket or what John and you did with the bread, ﬁrewood or letters.9 In addition to this very general formula we can deﬁne scores of derivative formulas applying to particular kinds of wilful action sequences. A number that have to do with hunting are well represented in the Kalam text of Majnep and Bulmer’s (1990) Kalam Hunting Traditions. The Kalam do not have a simple verb corresponding to ‘hunt’ or ‘go hunting’. When referring to hunting game mammals in general (as opposed to hunting a particular kind) they use a discourse formula with the following conceptual elements:

from event sequence to grammar

213

(22) Discourse formula for generic reports of hunting game 1 2 3 4 [move/stay] [game: mammal [(carry come)] [(cook) eat] (search) kill] Constituent 3 refers to bringing the game back to a ritual cooking and camping place in the forest, where the game is baked in an earth oven with appropriate protocols. Although the notation in (21) distinguishes between optional and obligatory constituents, a more complex notation would be needed to specify contextual conditions on the form of event reports (what kind of game is hunted, how it is cooked, etc.) and co-occurrence restrictions on constituents (for example, if the speaker mentions constituent 3 (bringing the game back to the cooking site) he will also mention constituent 4—baking it (in an earth oven) and/or eating it). When the subject is hunting a particular kindof game mammal, other formulas are used which refer to the particular methods of search and capture. For game taken in trees, a frequent discourse formula is: (23) 1 2 3 4 5 [go] [tree climb] [animal:name [(carry come)] [(cook) eat] (find) kill] For game hunted by ground-search a frequent formula is: (24) 1 2 3 4 5 [go] [animal: (search walk:about)] [(carry [(cook) eat] name [find kill] come)] For game taken by digging, a frequent formula (see Section 4.3 for more extended discussion) is: (25) 1 2 3 4 [go] [animal: name dig kill] [(carry come)] [(cook) eat] For game taken in snares, one of the formulas is: (26) 1 2 3 4 [go] [animal:name snare [(carry come)] [(cook) eat] set catch kill] For frogs, which are unimportant animals, not needing to be brought back to a ritual cooking-place before eating, there is a simple formula: (27) 1 2 3 [go] [frog search find] [eat]

214

andrew pawley and jonathan lane

4.3 Syntactic options for reporting multi-scene events From this account the reader might think that Kalam reports of multi-scene events are necessarily long and cumbersome. No so. When mentioning a familiar kind of multi-scene event Kalam speakers have the option of giving (i) an elaborate description which runs across many clauses or sentences, (ii) a relatively streamlined multiclause description or (iii) a highly conventional description using a serial verb construction. A look at some extracts from Ian Saem Majnep’s texts in Majnep and Bulmer (1990) will illustrate. The examples all have to do with the hunting and cooking of two broad classes of wild animals—kmn (large game mammals, i.e. the larger marsupials and rodents) and as (small game, principally very small marsupials and rodents and frogs)—we will refer to this as ‘‘the hunting sequence.’’ In these multiclause extracts each clause or clause-like constituent is given a line to itself and the clauses are labelled (a), (b), etc. Bare verb stems are assigned to the same clause as the following tensed verb. Inﬂected verbs are assigned to separate clauses. In the ﬁrst example the hunting sequence is spread over the four clauses (c)–(f). (28) a. mñab ak l g-l land that establish occur-ss.prior ‘After that land came into existence b. md-e-k, exist-ds:prior-3sg c. kmn ak pak dad ap-l, game:mammal that kill carrying come-ss:prior (the ﬁrst hunter) killed and brought game mammals, d. ti ti g-l what what do-ss:prior then after performing the customary rituals, e. ad-l bake-ss:prior he cooked f. ñb-e-k, . . . eat-ds:prior-3sg and ate (the game) . . . ’ In (29)–(34) the hunting sequence is in each case spread over the three clauses (a)–(c).

from event sequence to grammar

215

(29) a. basd yes ogok md-l, ancestor distant those stay-ss:prior ‘. . . the ancestors living there, b. kti am kmn pak dad ap-l-, they go game:mammal kill carrying come-ss:prior used to go and kill game mammals and having brought them c. nb okok ad ñb-elgpal. so thereabouts bake eat-past:hab-3pl to places nearby they would bake (them in earth ovens) and eat (them)’ (30) a. bin-b ak ned md-l sblam mey ognl man-woman that ﬁrst stay-ss:prior cordyline aforem’d those ‘. . . the cordyline enclosures where the ﬁrst people had their house sites, b. kmn pak dad ap-l game:mammal kill carrying come-ss:prior where, having killed and brought game mammals c. ad ñbl katp se ognl, . . . bake eat-ss:prior house old:site those they cooked and ate them, . . . ’ (31) a. . . . numd-ney kmn sblam yb ak mey, . . . crosscousin-his game:mammal cordyline true that aforem’d ‘. . . the close relative of the real game mammal cordyline, the one (used in ovens)’ b. kmn ak pak dad ap-l, game-mammal that kill carrying come-ss:prior game mammals, after being killed and brought there, c. adñ-b-al. bake eat-prs:hab-3pl are cooked and eaten’. In (32) the hunting sequence extends over four clauses but the major part of it is compressed into a single clause-like constituent, (c), which contains a string of ﬁve verbs. (32) a. Mey basd skop yes ogok amt ag-l, aforem’d ancestor group distant the:pl go travel:about-ss:prior ‘When our forefathers went to (the forest) and walked about (catching game mammals) b. sblam mgan nb ak kn-l cordyline enclosure thus that sleep-ss:prior they used to sleep in these same cordyline enclosures

216

andrew pawley and jonathan lane c. kmn pak d ap ad ñb-l game: hit get come bake eat-ss:prior and having killed game and brought (it there) and cooked and eaten (it) d. ap-elgp-al. come-pst:hab-3pl they used to return (home)’.

What is striking about this (very typical) group of texts is that whether Saem is speaking generally about the hunting sequence, as in (28), focusing on one particular activity in the hunting sequence, e.g. cooking and eating the animals, as in (29)– (33), using the proper ‘‘game:mammal’’ variety of cordyline when cooking, as in (31) or bringing the game back to the cordyline enclosures, as in (32) he mentions the same four or ﬁve core actions, in the same order. Focus on certain parts of the hunting sequence is not accomplished by changing the order of the constituents. It is rarely done by leaving out actions that are predictable and not in focus. It is mainly done by elaborating the important constitituents, by saying more than the minimum required by the formula. These texts illustrate conventions of discourse structure at several diﬀerent levels of abstraction. At the highest level, the texts conform to the general conceptual structure for talking about deliberate action events, summarised in (21). At the next level down the narrator follows the standard conceptual schema for reporting a speciﬁc kind of deliberate action, namely hunting game mammals, summarised in (22). Finally, we come to the choice of syntactic structures and words. At this level there is considerable variation between examples but, as we shall see, there are also strong family resemblances. The two principal syntactic mechanisms for handling multi-clause events are serialisation and the switch-reference/clause-chaining system. Both were seen at work in examples (28)–(32). To a large extent these two systems do similar work, yet each has aspects of event reporting that it is more suited to. To illustrate the ways in which the switch reference system and serialisation overlap, here are two examples from Majnep’s description of hunting Kalam rodents. (33) . . . n-l, yg-l, pak-l, dad o-p-al . . . . . . see-ss:prior dig-ss:prior, hit-ss:prior carrying come-perf-3pl ‘. . . they see (it) and dig, and hit (it), and carry it back . . . ’ (34) Bin pataj ogok am yg pak dad ap-elgp-al . . . woman young the:pl go dig hit carrying come-pst:hab-3pl ‘Young women used to go and dig and hit and bring back (these animals)’ (KHT XIII #29)

from event sequence to grammar

217

Both examples mention digging (yg-), hitting (i.e. killing; pak-) and carrying and coming (dad ap-; note that dad is a verb adjunct, i.e. a kind of verb modiﬁer, rather than a verb). In example (33), each verb is in a separate clause, and all verbs except the last are inﬂected for same subject:prior. Example (34), in contrast, places the same verbs into a serial verb construction. Most logically possible combinations of serialisation and clause chaining have been encountered in texts describing such kinds of hunting. In a text containing reports of 70 digging and hitting events, the two core verbs of the formula, yg- and pak-, occurred serialised together in 39/70 reports (56 per cent), and in diﬀerent clauses 31/70 (47 per cent) of the time. Despite the overlap between SVCs and switch reference clause chains, however, the nature of serialisation makes it particularly eﬀective for certain types of description and unusable for others. A chief advantage of serialisation is its eﬃciency. In at least two respects (34) encodes the same essential information more economically than (33). First, (34) dispenses with overt verbal morphology. Second, it has also been found (Givón 1990) that speakers are much less likely to pause between verbs within SVCs in Kalam than between clauses. 4.4 Some constraints on multi-clause SVCs We turn now to constraints on the internal structure and use of multi-clause SVCs. Mention was made earlier of certain restrictions which apply to simple SVCs. Several of these, namely (8a–e) apply also to multi-scene SVCs. For example, the negative verbal preﬁx ma- has scope over the entire SVC rather than over individual predicates. So to describe an event in which one hits an animal but doesn’t eat it, a sequence of clauses would be required; an SVC would not be appropriate. Further, the subject and tense (relative or absolute) marked on the inﬂected verb are shared by all verb stems in the SVC. Similarly there can (by deﬁnition) be only one subject NP per SVC. An event description in which, say, the actor kills the animal and then another actor eats it would necessarily involve at least two clauses. It’s worth noting that many serialising languages are freer than Kalam in allowing ‘‘switch-subject’’ SVCs, in which the object of one verb becomes the subject of the next. But even these types of SVCs have tighter restrictions on changes in subject reference than do various devices for joining clauses. Another characteristic of Kalam is that modiﬁers such as kasek ‘quickly’ or kapkap ‘carefully’ are highly restricted in their ability to intervene between verb stems in an SVC. For instance, they would not be able to appear between yg- ‘dig’ and pak- ‘hit’ in an SVC such as (28). In semantic/pragmatic terms, individual predicates within SVCS cannot be highlighted in these cases. To summarize, in order for SVCs to be applicable, the actions coded by the serialised verbs must be:

218

andrew pawley and jonathan lane

(a) either all asserted or all denied (b) all performed by the same subject (c) of roughly equal importance. If any of these conditions do not hold—that is, if the sequence of events diﬀers from the norm in any way—then interclausal syntax must be used. Thus the implementation of the requirements of Kalam discourse involves a trade-oﬀ between the eﬃciency of serialisation and the ﬂexibility of interclausal syntax.

4.5 The role of orientation verbs in multi-scene SVCs The key grammatical diﬀerences between simple and multi-scene SVCs have to do with the role played by orientation verb stems in the latter. Orientation verbs occur in positions 1 and/or 3 of the formula in (21), indicating whether the actor moved to the scene of the next action or was there already. The most common orientation verbs are the simple verbs of motion am ‘go’ and ap ‘come’. Others include tag ‘walk about, travel’, tan ‘ascend’, yap ‘descend’, the compound verbs d am and dad am ‘take s.th. (go carrying)’, d ap and dad ap’‘bring s.th. (come carrying)’, ap tan‘come upwards’ and ap yap- ‘come downwards’, am tag ‘go walking about’ and piow tag ‘go searching’ and the verbs of staying md ‘stay’, l ‘stabilise, put’ and kn ‘sleep’. This group of verbs possesses certain combinatorial privileges not shared with non-ﬁnal verbs in simple SVCs, as follows: 1. The orientation verbs when they appear non-ﬁnally in SVCs, can be followed by object and locative arguments and by certain adverbial modiﬁers. Non-orientation verbs appearing non-ﬁnally in an SVC have much stronger restrictions on the material which can follow this. Recall that in Section 2 it was noted that nonverb material following other verbs will, in general, consist of one of a small group of modiﬁers which creates a constituent with the ﬁnal (inﬂected) verb of the SVC. —Examples of nominal phrases following a motion verb in position 1 in the SVC structure can be found in examples (2), (5), (16), (18)–(20) and (35). Sentence (17) exempliﬁes such material following a motion verb in position 3. In clauses (b), (c) and (d) of the following extract we see three separate examples of extensive material (marked oﬀ here by square brackets) intervening between am and the next verb. (35) a. . . . as nb ak yg pak ñb-l, . . . . . . small:game thus the dig kill eat-ss:prior ‘After digging up and killing and eating game of this kind

from event sequence to grammar

219

b. am[maj wog ksen y-b-al ak] ma-a-b-al, go [sweet: potato garden new plant-perf-3pl] neg-go-perf-3pl (people) don’t go into newly planted sweet potato gardens c. ji am[okok m wog y-b-al] therefore go [thereabouts taro garden plant-perf-3pl] nor do they go into newly planted taro gardens d. ognap am[nb okok kpl ak] ma-g-p-al. sometimes go [place there weeding the] neg-do-perf-3pl and sometimes they don’t go and do any weeding..either.’. (KHT XIII, #32) Examples follow of SVCs with initial verb of staying followed by other material: (36) a. ñluk l-l lying:in:wait put-ss:prior ‘. . . (the two men) lay in wait b. md md [mey bin ak awl] yap pak-e-k stay staythere woman the here descend hit-ds:prior-3sg:pst time passed and then the woman fell down here (in front of them)’ (37) a. . . . abn ak tk g md md . . . undercroft the cut do stay stay ‘. . . we kept on cutting into the undercroft (galleries under the litterlayer on the forest ﬂoor) b. [suatg ak pat nb ak] pak-tu-k [marsupial:cat this long such this] kill-1dl-pst and we killed this big quoll(marsupial cat)’ (XI #15) 2. Am- and ap-, plus compound forms such as ap tan and ap yap can combine with either the verb d- ‘get’ or the adjunct dad ‘carrying’ to allow mention of the transportation of syntactic objects associated with a previous action. In (38) d ap‘get come’ indicates the return of the actor with aﬀected object (kmn ‘game mammal’) from scene of the pivotal action to the forest camp where the game will be cooked. (38) . . . kmn pak d ap ad ñb-l . . . . . . game:mammal hit get come bake eat-ss:prior ‘. . . having hit (i.e. killed) and brought (got and come with) and cooked and eaten the game mammals . . . ’ (KHT Intro #22) Plain motion stems or d- + verb of motion, or dad + verb of motion can only appear in set places within SVCs, in accordance with the discourse requirements

220

andrew pawley and jonathan lane

of Kalam. For example, transitive stems (e.g. ag- ‘shoot’) will be followed by dad am- or d ap- rather than by bare motion stems, except when the object of the transitive verb could not accompany the actor to the next scene (as is the case with the transitive verb ktg- ‘abandon’). 3. The orientation verbs, and especially am- ‘go’ and ap- ‘come’, can be followed by an extremely wide range of verb stems, considerably wider than other verbs, which tend to enter into lexical combinations with a restricted subset of verbs. 4. Orientation verbs are unique in that speakers sometimes pause after non-ﬁnal instances of such stems in SVCs; this often happens if a nominal phrase follows the verb stem. That is, such a stem can be part of a diﬀerent intonation contour from that of the following verb or verbs. In contrast, sequences of non-orientation verbs are normally spoken under a single intonational contour, even followed by non-verb material. In SVCs where motion stems (or the verb l- or ay- ‘stabilise, put’ etc.) aren’t involved, the length and probability of pauses is comparable to that between non-verbal words in a clause, and contrasts markedly with the pauses between clauses. These features add up to allow SVCs containing orientation verbs the potential for considerable internal syntactic complexity. Mention has already been made of the resemblance between multi-scene SVCs and clause chains where the non-ﬁnal clauses are marked for prior action by the same subject. Compare the ﬁrst parts of (39) and (40), which say almost the same thing using an SVC in (39) and a sequence of clauses in (40): (39) . . . kti am kmn pak dad ap-l nb okok, . . . . . . they go game: mammal hit carrying come-ss:prior placethese ‘. . . they having gone (and) hit and carried back game mammal to these places’ (Intro #17) (40) . . . b ak am-l, kmn pak-l, . . . . . . man this go-ss:prior game:mammal hit-ss:prior ‘. . . the men having gone, having hit game mammals . . . ’ (KHT III #97) In both examples, kmn ‘game mammal’, the object of pak- ‘hit’, follows the motion verb and precedes pak-. Orientation verbs are the only bare stems which consistently precede the objects of a following verb or verb series. In our data there are multi-scene SVCs with two locatives in two syntactically distinct positions, each locative belonging to a diﬀerent verb or group of verbs. This contravenes one of Foley and Olson’s (1985: 57) constraints on SVCs, namely that all the verbs in an SVC must share a single set of peripheral arguments. That is,

from event sequence to grammar

221

only core layer constructions can be serialised and keep their separate arguments. In (41) the SVC is in clause (c), with the locatives enclosed in square brackets: (41) a. Kmn nb ak ney nb g-u-p, game:mammal such this it thus do-3sg-perf ‘This type of game mammal behaves like this b. k-ng g-u-p. sleep-ss:fut do-3sg-perf so that it will sleep. c. [mab-yb al-ya] su tk d am [ok-ya] l-l, . . . tree-epiphyte below bite sever get go distant-below put-ss:prior Having burrowed down into a clump of ephiphytes, . . . ’ Here the locative mab-yb alyaa ‘the clump of epiphyte below’ serves as the direct object of the serial verb string su tk (bite sever) ‘burrow’, while okya ‘below (more distant than bya9), can be interpreted as the place where the burrowing stops. (42) a. . . . tap maj wad ogok, wad yg-l, . . . food sweet:potato string:bag these string:bag ﬁll-ss:prior ‘. . . (when they have ﬁlled up bags with sweet potatoes b. d ap [kapk mgan okok] yp, d ap [okok] l-l, . . . get come oven hole around with get come around put-ss:prior (and) brought them to oven pits around about, or other places around (the houses) . . .’ In (42) the SVC is in clause (b). The ﬁrst and second instances of d ap ‘bring’ are each followed by a locative argument. Serialisation across peripheral junctures has not previously been recorded and Foley and Olson, if we understand them correctly, regard any sequence of peripheral layer constructions as a sequence of clauses rather than a single clause. It has been suggested to the authors that one way around the problem of serialisation across peripheral junctures is to say that locatives are direct, rather than peripheral, arguments of motion verbs. This hypothesis seems plausible, given that motion verbs in Kalam clearly behave diﬀerently from other verbs with respect to locative NPs. However, more work needs to be done to establish whether this diﬀerential treatment applies outside of SVCs. 4.6 Lexicalised multi-scene SVC strings It is surely a general principle that speakers will seek short and standardised ways

222

andrew pawley and jonathan lane

of reporting highly recurrent complex events and situations. Multi-scene SVCs preserve the essentials of Kalam’s distinctive discourse structure rules while achieving a high degree of economy and conventionalisation. The most economical coding of the several discourse formulas for kinds of hunting is as a serial verb construction with highly conventional wording. There are conventional expressions which stand for a part of the one, two, three or all four scenes in the hunting sequence summarised in (22). For instance, (kmn) pak d ap and (kmn) pak dad ap ‘kill and bring (game mammals)’ are standard representations of scenes 2 and 3 while ad ñb ‘bake and eat’ and kab agl ñb ‘make earth oven and eat’ are standard representations of scene 4. The formulas for scene 2, 3 and 4 may be combined into a single long SVC, as in (38), where we ﬁnd kmn pk dad ap ad ñb ‘kill and bring and bake and eat game mammals’. The minimal (generic) description of all four scenes in the hunting sequence is achieved by adding am ‘go’ to the previous formula,as in example (38). The corresponding description for taking game by digging takes the form am (animal name) yg pak dad ap, as in (34). Only the name of the animal interrupts the string of verbs. These individual SVCs can be regarded as showing a high degree of lexicalisation in the same sense as the expressions listed in (7).10

5. Conclusions We have identiﬁed several contrasting levels of bonding between verb stems in Kalam SVCs. On the one hand there are tight-knit SVCs in which the serial verb string may be regarded as a single complex verb nucleus, a compound verb. In these constructions each verb stem is contiguous with another verb stem or with a lexemelike combination of nominal and adverbial + verb stem and the verb group is subject to a raft of constraints (8a–i). Contrasting with such constructions are multi-scene SVCs which allow considerable freedom for various types of NP to occur within the SVC and which are often spoken with a pause after the initial orientation verb constituent. In this respect multi-clause SVCs resemble sequences of clauses, especially sequences in which the non-ﬁnal clauses are marked with ss:prior morphemes. Should multi-scene SVCs be considered a single clause?11 The limited range of grammatical and semantic distinctions between verb stems allowed in SVCs—for example, the requirements of iconicity between the ordering of verbs and the temporal ordering of actions, and agreement of tense/aspect and subject identity and in negation or assertion between all verbs—means that they cannot do all the tasks performed by interclausal syntax. On the other hand, they can do much more than a canonical, one-verb clause.

from event sequence to grammar

223

They can report a series of actions taking place in diﬀerent places at diﬀerent times. The actor needs to be the same for all the actions but under some circumstances the patients may diﬀer. The multi-scene SVCs discussed above, in which NPs which followed motion stems were frequently associated with a pause, often have close paraphrases where the only diﬀerence is that NPs precede the motion stems. Note the diﬀerence in delivery speed. The only possible answer seems to be that multiscene SVCs are neither one canonical clause nor a canonical clause sequence. They are an intermediate type, showing family resemblances to both. Foley and Olson (1985) and Foley and Van Valin (1984) have developed a typology of clause types which accomodates certain kinds of multi-predicate constructions as well as single predicate clauses. In this typology there is a place for a clause type which shows fusion of two or more core layers, each core layer being a verb phrase, i.e. a verb with its direct arguments (subjects and direct objects). Such a combination of verb phrases by deﬁnition, can incorporate a wider range of arguments than a clause with a single predicate or a compound predicate. Most of the multi-scene SVCs in our data ﬁt this description. However, those SVCs with two peripheral locatives do not. Should we therefore regard compound SVCs as single clauses while treating (at least some) multi-scene SVCs as clause sequences? The problem here is that ‘‘clause’’ is a moveable feast. Foley and his co-authors have redeﬁned the term to suit their purposes. They have extended the traditional rather narrow notion of clause to accomodate a wider range of constructions (specifically SVCs) while stopping short of admitting a type which contains two or more verbs with separate peripheral layer arguments. The value of their typology is to make explicit points of similarity and diﬀerence between simple clauses and SVCs and between both and clear cases of clause sequences. However, this terminological shift does not alter the fact that the various types of multi-predicate constructions are diﬀerent from single-predicate constructions and from each other. We suggest that in some languages there is no natural break between multi-clause and single clause constructions, only a continuum of types. The trend of historical change in individual languages may indeed be for SVCs to become more and more like canonical clauses—i.e. the clause structure may serve as a target for speakers of the language.12 But in the meantime, the more complex SVCs, such as Kalam multi-scene SVCs, are a long way from this target. How have multi-scene SVCs arisen? We suggest that they strike a nice balance between two opposing requirements in language use: (1) the need to economise or streamline, (2) the need to give details or elaborate. In this case the competition concerns the reporting of complex events. The lack of morphological baggage within SVCs allows them to code certain kinds of event sequences more economically than clause sequences. At the same time the structure has much more expres-

224

andrew pawley and jonathan lane

sive power than the canonical clause, at least in regard to the description of complex events. But most other languages do not handle the competition between economy and detail in the way that Kalam does.13 We noted that English speakers prefer the metynomic strategy when reporting complex events—one component event conventionally stands for the whole sequence. Why have the two languages arrived at such diﬀerent solutions? Part of the answer has to do with discourse structure rules. If you are to talk idiomatically about a subject matter in a given language you need, in the ﬁrst place, to follow the general rules in that language for constructing discourse about that subject matter. Kalam has well-deﬁned discourse structure rules for reporting events in an elaborate way, rules which diﬀer markedly from those of English. Multi-scene SVCs preserve the essential details of Kalam discourse rules. But more than that, you need to know how to say speciﬁc things—in this case, to report particular events or kinds of events in ways that are nativelike. The phrasing and wording must be right. And it turns out that the Kalam discourse structure rules are even preserved in the lexicalised expressions for denoting familiar, much talked about events. The struggle with multi-scene SVCs has brought home to us a few points of wider signiﬁcance. To understand how these constructions work it is clear that we need to study the links between those phenomena we have glibly put into separate boxes labelled ‘‘discourse structure rules,’’ ‘‘syntax,’’ ‘‘idiomaticity’’ and the ‘‘lexicalization process.’’ There is, as far as we know, no formal model of language which allows these facets of language to be uniﬁed. Instead, linguists by and large persist with models which encourage us to treat these interrelated phenomena as mutually exclusive and to look at just one box at a time (mainly it is syntax). This modular approach goes together with linguists’ age-old penchant for viewing utterances as bits of structure subject only to abstract architectural constraints, rather than as bits of behaviour subject to processing and cultural pressures.

Notes 1. We are delighted to oﬀer this paper to Barry Blake. An earlier version was presented at the 3rd International Conference on Papuan Linguistics, Madang, September 1992. Pawley’s ﬁeld research on Kalam was supported by grants from the Wenner Gren Foundation for Anthropological Research and the Universities of Auckland and Papua New Guinea. 2. Etp mnm is spoken chieﬂy in the Upper and Middle Simbai Valley and Kaiment Valleys and much of the Upper Kaironk. Ti Mnm is spoken in the Asai Valley and parts of the Upper Kaironk. The two diﬀer considerably in morphological forms and lexicon but very little in basic structure. The Kalam

from event sequence to grammar

225

give separate names to several regional varieties in addition to those called Etp Mnm and Ti Mnm. However, they also recognise that the major division is between Etp Mnm and Ti Mnm and that all other named varieties can be considered to fall within one or the other major division. Linguistic analyses of Kalam include a detailed study of serial verb constructions by Lane (1991) and a general grammar by Pawley (1966) and treatments of particular topics by Givón (1990), Kias and Scholz (1991), Pawley (1987, 1991, 1993, 1994), Scholz (1979), Sugimoto (1975). There is an unpublished dictionary (Bulmer and Pawley 1993). 3. Examples used in this paper are largely taken from the Kalam text of Kalam Hunting Traditions (Majnep and Bulmer 1990), which contains some 200 pages about diﬀerent wild animals hunted by the Kalam. Kalam Hunting Traditions consists of 13 chapters, with numbered paragraphs, compressed into six Working Papers. Following each example from this source the number and the chapter of the relevant Working Paper are cited together with the paragraph number. 4. It will be seen that the free translation departs radically from the literal one indicated by the morpheme glosses—the free translation aims at a pragmatic or idiomatic equivalence in English. 5. The list of about 125 verb stems in Kalam is likely to be fairly exhaustive. Research on Kalam lexicon has been carried out intermittently over some 30 years by a number of scholars. 6. We will use ‘‘event’’ in the ‘‘subjective’’ sense of a person’s conceptual construction or perception of a bounded happening. Virtually all such events are potentially analysable as a sequence of component events or sub-events. ‘‘Report’’ (or ‘‘event-report’’) is used here to refer to any verbal act specifying the details of an event, e.g. ‘‘mention,’’ ‘‘describe,’’ ‘‘predict,’’ ‘‘ask about,’’ etc. Givón (1990) calculated the relative frequency of serial verb constructions used by speakers describing the same visual events (in a short ﬁlm) in three New Guinea languages: Kalam, Tairora and Tok Pisin. 7. For some observations on how Kalam talk about other kinds of events see Pawley (1987, 1991 and 1993). 8. Unless something unusual occurred, Bill does not need to say what he did after going to the supermarket (bought supplies, brought them home, put them away). Jill does not need to say that she ﬁrst went to the golf course or that she went home afterwards. Gricean principles apply. Speakers say enough to get the meaning across but not too much. 9. What counts as enough or too much in an event-report? Given a universe of discourse in which certain things are signiﬁcant, or potentially signiﬁcant, the following principle seems to apply in English discourse (excluding special types) to the reporting of stages of an event (sub-events): only unpredictable outcomes or stages need be mentioned. If someone reports an act of searching for a child, the audience wants to know whether or not the child was found. If a baseball commentator remarks that a pitch has been thrown, the audience craves knowledge of the outcome: did the batter swing? Did he make contact? If not, was the ball a strike? etc. What counts as a well-formed or complete report is partly a matter of shared knowledge about the structure of that world (e.g. searching events, baseball games) including the predictability of outcomes. —Why do the Kalam consistently mention certain sub-events that are routine components of an event, such as outcomes that are predictable in the normal course of aﬀairs? We hesitate to say that Kalam event-reports are not constructed on Gricean principles. One might argue that the Kalam crave knowledge of actors’ comings and goings because these things are more signiﬁcant in the Kalam world,

226

andrew pawley and jonathan lane

for whatever reason. But we are doubtful of that explanation. Arguably, some requirements of discourse structure, like some syntactic conventions, are arbitrary conventions of a language. Compare the obligatory marking of tense-aspect and deﬁniteness in English discourse. The Kalam emphasis on explicitly stating movements to and from, etc. may well be related to the reliance of a small number of verbs. Reporting complex events in terms of component events is often necessary, just as it is sometimes necessary to specify the time of an action. A response to a periodic need may grow into a preference and then into a ﬁxed habit or rule. 10. Kobon, a language quite closely related to Kalam, appears to have SVC constructions very similar to Kalam (Davies 1981), including both the single and multi-clause variations. 11. A logical question to ask at this point is whether the more loosely bound multi-scene SVCs should even be treated as SVCs. Some restrictive deﬁnitions of SVC would in fact exclude these constructions. But in fact there is no sharp cutoﬀ point between simple and complex SVCs in Kalam. The family resemblances between them are very strong—in particular, they share restrictions (8a–e) . And individual serial verb strings show a tendency to lexicalise. 12. Foley and Olson (1985) point out some ways in which complex clauses are restructured to become more like simple clause as target. Givón (1980) oﬀers some nice observations on the continuum between multi-clause and simple clause constructions. 13. See Bruce (1984) for comments on similar lines.

References Bradshaw, Joel. 1982. ‘‘Word Order Change in Papua New Guinea Austronesian languages.’’ Ph.D. thesis. Department of Linguistics, University of Hawaii. Bruce, Les. 1984. ‘‘Serialisation: The interface of syntax and lexicon.’’ Papers in New Guinea Linguistics 24: 21–37. Canberra, Australian National University [Paciﬁc Linguistics A70]. Bulmer, Ralph & Andrew Pawley. 1993. ‘‘A Dictionary of Kalam.’’ Printout. Dept. of Linguistics, Research School of Paciﬁc Studies, Australian National University. Crowley, Terry. 1987. ‘‘Serial Verbs in Paamese.’’ Studies in Language 11: 35–84. ——. 1990. ‘‘Serial Verbs and Prepositions in Bislama.’’ In Verhaar (ed.), 57–89. Davies, John. 1981. Kobon. Amsterdam: North-Holland [Lingua Descriptive Series 3]. Durie, M. 1997. ‘‘Grammatical Structures in Verb Serialization’’ In A. Alsina, J. Bresnan, and P. Sella (eds), Complex Predicates. Stanford: SCLI Publications, Stanford University, 289–354. Foley, W. 1986. The Papuan Languages of New Guinea. Cambridge: Cambridge University Press. ——. 1991. The Yimas Language of Papua New Guinea. Stanford: Stanford University Press. —— (ed.) 1993. The Role of Theory in Linguistic Description. Berlin: Mouton de Gruyter.

from event sequence to grammar

227

Foley, W. and Mike Olson. 1985 ‘‘Clausehood and Verb Serialization.’’ In Johanna Nichols and Anthony C. Woodbury (eds), Grammar Inside and Outside the Clause. Cambridge: Cambridge University Press, 17–60. —— 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 5: 163–93 ——. 1990. ‘‘Verb Serialisation in Tok Pisin and Kalam: A comparative study of temporal packaging.’’ In Verhaar (ed.). Kias, John and Lyle Scholz, 1991. ‘‘We Threw Away and Bows and Axes.’’ In Pawley (ed.), 412–20. Lane, Jonathan. 1991. ‘‘Kalam Serial Verb Constructions.’’ MA thesis, Dept. of Anthropology, University of Auckland. Majnep, Ian Saem and Ralph Bulmer. 1990. Aps Basd Skop Kmn Ak Pak Nbelgpal. Kalam HuntingTraditions. (ed. Andrew Pawley.) University of Auckland. Dept. of Anthropology Working Papers 85–90. Pawley, Andrew. 1987. ‘‘Encoding Events in Kalam and English: Diﬀerent logics for reportingexperience.’’ In R. Tomlin (ed.) Coherence and Grounding in Discourse. Amsterdam: Benjamins [Typological Studies in Language 11], 329–60. ——. 1991. ‘‘Saying Things in Kalam: Reﬂections of language and translation.’’ In Pawley (ed.), 432–44. ——. (ed.) 1991. Man and a Half: Essays in Paciﬁc Anthropology and Ethnobiology in Honour of Ralph Bulmer. Auckland: Polynesian Society. ——. 1993. ‘‘A Language Which Deﬁes Description by Ordinary Means.’’ In W. Foley (ed.), 87–129. ——. 1994. ‘‘Kalam Exponents of Semantic and Lexical Primitives.’’ In Cliﬀ Goddard and Anna Wierzbicka (eds), Semantic and Lexical Universals. Amsterdam: Benjamins, 377–421. Roberts, John R. 1997. ‘‘Switch-Reference in Papua New Guinea: A preliminary survey.’’ In A. Pawley (ed.) Papers in Papuan Linguistics No. 3, 101–241. Canberra: Paciﬁc Linguistics A–87. Schachter, Paul. 1974. ‘‘A Non-Transformational Account of Serial Verbs.’’ Studies in African Linguistics, supplement 5: 253–70. Scholz, Lyle, 1979. ‘‘Kalam Lower Level Grammar Sketch.’’ Ms. Summer Institute of Linguistics, Ukarumpa,, Papua New Guinea. Sugimoto, Takashi. 1975. ‘‘Notes on a Kalam Relative Clause Construction and Some Related Problems. University of Hawaii Working Papers in Linguistics. Verhaar, John (ed.) 1990. Melanesian Pidgin and Tok Pisin. Amsterdam: Benjamins [Studies in Language Companion Series 20].

Passive-to-ergative versus inverse-to-ergative ANNA SIEWIERSKA Lancaster University

1. Introduction Among the languages that exhibit overt case marking of verbal arguments, two case marking patterns are most common, namely the accusative and the ergative. These are shown graphically in Figure 1 where, following Dixon (1972) and Comrie (1978), S denotes the sole argument of an intransitive clause, A the agentive argument of a transitive clause and P the patient argument of a transitive clause. In an accusative case marking system the S and A appear in the nominative case and the P in the accusative. In an ergative system of case marking the S and P take the absolutive case and the A the ergative. Though there are languages in which both the nominative and accusative are overtly marked and also languages in which the nominative is overtly marked while the accusative is not, typically the nominative takes zero marking and the accusative overt marking as in the examples in (1) from Kannada.

Nominative-Accusative

A

O S

Ergative-Absolutive

A

O S

Figure 1.

anna siewierska

230

(1) Kannada (Sridhar 1990: 159, 160) a. HuDuga-ø o:Diho:da boy-nom 1 run:pp:go:pst: 3sg:m ‘The boy ran away.’ b. HuDuga-ø vis’ala:kSiy-annu maduveya:danu boy:nom Vishalakshi-acc Mary:pst: 3sg:m ‘The boy married Vishalakshi.’ In ergative case marking, on the other hand, it is the ergative which is always overtly marked, while the absolutive is typically zero as in (2) from Watjarri. (2) Watjarri (Douglas 1981: 217, 214) a. Mayu-ø yanatjimanja kurl-tjanu child-abs come:prs school-abl ‘The child is coming from school.’ b. Mayu-ng(k)u tjutju-ø pinja child-erg dog-abs hit:pst ‘The child hit the dog.’ While most linguists recognize that both accusative and ergative case marking may be the result of several diﬀerent diachronic developments, ergative case marking, as the less common of the two, has excited special interest.2 Of the sources of ergative case marking that have been proposed in the literature, the one most frequently cited is the passive.3 And indeed when we compare the case marking of an active ergative clause such as (2b) with the case marking found in a typical passive clause with an overt agent such as (3), the similarity in case marking is quite clear. (3) Ngarluma (Blake 1977: 7, 27) Makula-ø pilya-nali-na yukuru-la child-nom bit-pass-pst dog-loc/inst ‘A child was bitten by a dog.’ As shown schematically in (4), in both the active/ergative (2b) and the passive (3) the patient occurs with zero marking, the agent with overt marking. (4) a. pass b. erg

Patientø Agentobl Agenterg Patientø The overt marking of the agent and zero marking of the patient characteristic of ergative case marking is not only remeniscent of the case marking found in passive

passive-to-ergative versus inverse-to-ergative

231

clauses but also of that found in certain types of inverse constructions. Inverse constructions are best known from the Algonquian languages in which the direct voice is used if the agent is more topical or ontologically salient than the patient, and the inverse if the patient is more topical or ontologically salient than the agent. Traditionally the more salient or topical participant is called the proximate and the less salient or topical one the obviative. This direct/inverse voice opposition is illustrated in (5) on the basis of Plains Cree. (5) Plains Cree (Wolfart 1973: 25) a. sekih-ew napew antim-wa scare-dir man:prox dog-obv ‘The man scares the dog.’ b. sekih-ik napew-a antim scare-inv man-obv dog:prox ‘The man scares the dog.’ In Plains Cree and in other Algonquian languages, in clauses with two nominal participants the proximate participant occurs with no morphological marking while the obviative takes a special marker, -(w)a in (5). Thus the nominal marking in the inverse, i.e. zero marking of the patient and obviative marking of the agent is analogous to what we ﬁnd in ergative constructions as shown in (6). (6) a. inverse Patientø Agentobv b. ergative Agenterg Patientø Givón (1994a) has recently suggested that the inverse constitutes a more promising source of ergative case marking than the passive, since inverse clauses are functionally more similar to active ergative clauses than are passive clauses. And indeed in terms of Givon’s functional pragmatic deﬁnitions of the active, inverse and passive, cited in (7), this is so. (7) Active: The agent is more topical than the patient but the patient retains considerable topicality. Inverse: The patient is more topical than the agent but the agent retains considerable topicality. Passive: The patient is more topical than the agent and the agent is extremely non-topical (suppressed, demoted). The passive diﬀers from the active more than the inverse does, due to the fact that the agent in the passive, if at all present, is nontopical, while in the inverse it retains considerable topicality.

232

anna siewierska

Though Givón does not deny that some instances of ergative case marking may have evolved from the reanalysis of passive clauses, he contends that the inverse constitutes a necessary stage of any such reanalysis, i.e. that the passive-to-ergative reanalysis is actually a passive-to-ergative via inverse reanalysis. Needless to say, this follows from the deﬁnitions in (7). In order for the passive to become the unmarked active/ergative construction, the passive must begin to be used ﬁrst with relatively topical agents, and then also with even more topical ones. And once it is thus used it will no longer be a passive but an inverse. If the use of the inverse is then extended from clauses with relatively topical agents to those with topical ones, the reanalysis of the former passive-turned-inverse as an ergative will be complete. While Givón’s deﬁnition of the inverse renders the inverse a necessary stage of the passive-to-ergative reanalysis, the inverse is also viewed by Givón as a source of ergative nominal marking independent of the passive. Thus in addition to the historical scenario in (8a), Givón also proposes the historical scenario in (8b). (8) a. passive → inverse → ergative b. inverse → ergative Given that inverse constructions have not as yet been extensively studied, it is by no means clear whether the passive and inverse can be distinguished from each other systematically on structural or even, as Givón contends, on functional grounds.4 However, assuming that they can, and that both of the scenarios in (8) are possible, is there any way of determining whether the ergative marking in a language originates from a passive turned inverse as opposed to an inverse? In this paper I would like to consider this issue in relation to two types of ergative languages, namely: those exhibiting split ergativity conditioned by the semantics of nominals and those exhibiting ergative verbal agreement. Split ergativity conditioned by the semantics of nominals is viewed by Givón as the clinching argument for the universality of the inverse-to-ergative diachronic pathway. Ergative agreement, on the other hand, is typically considered as suggestive of a passive origin of ergative marking. In what follows I will attempt to establish to what extent the two types of ergative marking may indeed be seen as diagnostic of the two sources of ergative marking, the inverse and the passive respectively. In Section 3 I will consider the inverse and passive reanalyses in the context of languages manifesting split ergativity conditioned by the semantics of nominals. In Section 4 I will seek to determine whether ergative agreement can be derived from an inverse. But ﬁrst let me brieﬂy clarify what sort of passive and inverse constructions will be assumed to be involved in the two diachronic sources of ergative marking.

passive-to-ergative versus inverse-to-ergative

233

2. Promotional and non-promotional passives and inverse constructions If there were to be no structural diﬀerences between the passive and the inverse, Givón’s contention that the inverse constitutes a source of ergative nominal marking independent of the passive would be rendered vacuous, as there would be no means of ever determining whether the ergative marking originates in a passive turned inverse or an inverse. Therefore in order to proceed with the investigation we must ﬁnd some means of diﬀerentiation between the two constructions. Passive and inverse constructions may be classiﬁed along several dimensions but the classiﬁcation pertinent to our discussion is the distinction between promotional and non-promotional ones. In the former the patient of the corresponding active/ direct voice is promoted to subject in the passive/inverse. In the latter there is no such promotion, the patient is not the subject in the passive/inverse. In terms of the subjecthood of the patient the promotional passive is thus indistinguishable from the promotional inverse, and the non-promotional passive from the non-promotional inverse. The two promotional and the two non-promotional constructions do, however, diﬀer from each other in regard to the status of the agent. Whereas the agent in the passive is a syntactic adjunct, the agent in the inverse is a syntactic argument. This is evinced by the obligatoriness of the agent in the inverse as opposed to the passive and by the nature of the verbal agreement marking that the two constructions display. There is no agreement between the verb and the agent in the passive, but agent agreement may occur in the inverse. For reasons which will be speciﬁed later such agreement is not always obvious, but we see it clearly in (9), where the verb is marked for the third person plural obviative agent ‘they’. (9) Cree Ki-wapam-ikw-ak 2-see-inv-3pl ‘They see you (sg).’ The passive is thus a monovalent, intransitive construction, while the inverse is bivalent and transitive or potentially de-transitivized, but not intransitive. The above distinction between the passive and the inverse constitutes the only structural diﬀerence between the two constructions if we take both the promotional and non-promotional variants of the construction into account. However, if we were to consider only the promotional passive and the non-promotional inverse, the two constructions would also be further diﬀerentiated by the subject vs non-subject status of the patient. In the context of a discussion of the passive as opposed to

234

anna siewierska

inverse source of ergative nominal marking, restricting the passive and the inverse in such a way is not unjustiﬁed. The type of passive assumed to be involved in the passive-to-ergative scenario is the promotional passive and not the non-promotional one. Though the passive-toergative reanalysis could produce ergative morphological marking in a language previously lacking case marking, such as Puget Salish, for example, the passive-toergative scenario is primarily understood as involving a change from accusative to ergative marking as shown in (1O). (10) a. active b. passive c. ergative

Agentnom Patientacc Agentobl Patientnom Agenterg Patientabs

Needless to say the change of an overtly marked accusative patient to a zero absolutively marked subject could not be achieved via the non-promotional passive. While Givón does not actually state that it is the non-promotional rather than the promotional inverse that constitutes a source of ergative nominal marking independent of the passive, the fact that he sees split ergativity as the best evidence bearing out the inverse-to-ergative diachronic pathway strongly suggests that it is the nonpromotional inverse that he has in mind. Moreover, the typical Amerindian inverse as found in the Algonquian languages and also in the Athabaskan and Tanoan as well as in Kutenai, Nootka and Shapatin, is a non-promotional inverse. And actually, so far, the only language with a promotional inverse clearly distinct from the passive is Chamorro. In view of the above, in the ensuing discussion the passive-to-ergative scenario will be taken to involve the promotional passive and the inverse-to-ergative scenario the non-promotional inverse. This already allows us to identify one set of circumstances amenable to a passive-to-ergative reanalysis but not to an inverse-to-ergative one, namely the complete accusative-to-ergative change which provides the basic motivation for the passive-to-ergative scenario. Since such a direct accusativeergative change is impossible for a non-promotional inverse, I will concentrate on more complex inputs and/or outputs.

3. Split-ergativity conditioned by the semantics of nominals In languages in which split ergativity is conditioned by the semantic features of nominal constituents, as opposed to tense or aspect or main vs subordinate clause or word order, the more ontologically salient constituents manifest accusative marking, the less ontologically salient ones ergative marking. This may be captured in the hierarchy in (11) based on Silverstein (1976).

passive-to-ergative versus inverse-to-ergative (11) acc erg

235

1 2 3 kin & personal nouns human animate inanimate > > > > > > > < < < <

The arrows in (11) indicate the direction of spread of accusative and ergative marking, from left to right in the case of accusative marking and from right to left in the case of ergative. The arrow heads indicate possible cut oﬀ points, the most common of which are accusative marking of pronouns and ergative marking of nouns or accusative marking of ﬁrst and second person pronouns and ergative marking of all other constituents. Givón (1994a: 33) argues that the use of accusative nominal morphology for highly salient agents and ergative morphology for agents low in saliency makes little synchronic sense for an active transitive clause, but perfect sense for an inverse voice clause. While he is indeed correct, this does not entail that split accusative/ ergative marking conditioned by the semantics of nominals necessarily originates from the reanalysis of an inverse as opposed to a passive. The inverse constitutes a more viable source of split acc/erg marking conditioned by the semantics of nominals than the passive only under a very speciﬁc set of conditions, namely: if the accusative marking of the constituents on the left of the hierarchy in (11) is already in existence at the time of the emergence of ergative nominal marking and, furthermore, if the passive or inverse is used not only with nominal agents and patients but also with pronominal ones or at least a pronominal patient and a nominal agent. Both of the above assumptions are reasonable ones. Accusative marking, particularly of pronouns, is highly common and languages with case marking of pronouns but not of nouns clearly outnumber those in which the former but not the latter display case marking. As for the use of pronouns in the passive or inverse, note that in most types of texts clauses with two nominal participants are much rarer than those involving a pronoun and a noun or two pronouns. Therefore if the passive or inverse were to be restricted to clauses with two nominal participants, the use of neither the passive nor the inverse would be frequent enough to warrant it being reanalyzed as the basic transitive construction. Assuming that the pronouns are accusatively marked and that the passive can be used with pronominal subjects, the reanalysis of passive clauses as ergative ones will necessarily destroy the accusative marking, by transforming a O into a S. Accordingly, under the passive-to-ergative scenario, accusative marking in a split accusative/ergative language must be assumed to be a development subsequent to the emergence of ergative marking. By contrast, in the inverse the O remains a O and therefore an accusatively marked pronominal O in the direct voice will maintain its accusative marking in the

anna siewierska

236

inverse. The inverse is thus fully compatible with the prior existence of accusative marking. In fact if there is no pronominal case marking the reanalysis of the inverse as an active ergative will result in ergative marking of nominals and no marking of pronominals, or ergative marking of both nouns and pronouns and not split accusative/ergative marking. The inverse-to-ergative scenario most directly leading to split accusative/ergative marking would involve a language which already has accusative marking of pronouns, no overt marking of nouns in the direct voice, obviative marking of the agent in the inverse and a direct/inverse voice opposition restricted to clauses with nominal or mixed participants but not to pronominal participants. The reanalysis of inverse clauses as active ergative clauses in such a language is shown schematically in (12) and (13). (12) a. direct b. inverse c. ergative

N Agentø N Agentobv N Agenterg

N Patientø N Patientø N Patientabs

(13) a. direct b. inverse c. split

Pro Agentnom N Agentobv N Agenterg

N Patientø Pro Patientacc Pro Patientacc

If in the above type of language the direct/inverse opposition were to also include clauses with two pronominal participants, the reanalysis of the inverse as the basic transitive construction would produce ergative marking with nouns and tripartite marking of pronouns, i.e. separate marking for S, A and P as shown schematically in (14). (14) a. b. c. d.

direct inverse ergative intransitive

Pro Agentnom Pro Agentobv Pro Agenterg Pronom

Pro Patientacc Pro Patientacc Pro Patientacc

The tripartite marking of pronouns resulting from the reanalysis of an inverse with two pronominal participants is of particular interest because it provides a strong argument for the inverse source of ergative nominal marking in the languages of Australia. Though Givón cites the Australian languages as prime candidates for the inverse source of ergative nominal marking, he bases his claim on the current accusative/ergative split marking found in many Australian languages. However, according to Dixon (1980) and also Blake (1987), Proto-Australian was not split accusative/ergative but split tripartite/ergative. The current accusative marking of pronouns therefore postdates rather than predates the emergence of ergative marking. The forms of ﬁrst and second person A and S pronouns bear clear traces of the same ergative marker as has been reconstructed for nouns, namely *lu, which suggests

passive-to-ergative versus inverse-to-ergative

237

that the accusative marking of pronouns is the result of the ergative marking of A pronouns being extended to the S pronouns.5 Since the A pronouns, just like the A nouns, once bore ergative marking, and ergative marking of both nouns and pronouns followed by the development of accusative pronominal marking is compatible with both a inverse and a passive source of ergative morphology, the current accusative/ergative split in the languages of Australia does not in itself constitute an argument for the inverse source of the ergative marking. By contrast, the tripartite/ergative split of the proto-language does. Given that the passive converts a O into a S, it destroys the accusative marking of the O. Therefore, the emergence of ergative nominal marking in the proto-language via the reanalysis of the passive would have resulted in ergative and not tripartite pronouns. An inverse which does not promote a O to an S, on the other hand, allows for the accusative marking of a pronominal O. Consequently, if the inverse is reanalyzed as the basic transitive construction, both the A and the O will emerge with overt marking, precisely as appears to have been the case in Proto-Australian. To the best of my knowledge the possibility that the tripartite/ergative split in Proto-Australian may have arisen via the reanalysis of an inverse has not been previously entertained. But clearly the reanalysis of an inverse is a more promising source of the split marking than the reanalysis of a passive. As suggested by the above discussion, split acc/erg marking conditioned by the semantics of nominals is in principle compatible with both a passive and an inverse source of ergative nominal marking, provided that the accusative marking is subsequent to the emergence of the ergative. However, if the accusative marking predates that of the ergative, the inverse constitutes a much more viable source of ergative nominal marking than the passive. While the above suggests that the inverse source of ergative nominal marking provides a better account of split ergativity conditioned by the semantics of nominals than the passive source, the restricted set of conditions under which this holds simmultaneously counter Givón’s contention that the mere existence of such split ergativity constitutes a clinching argument for the universality of the inverseto-ergative diachronic pathway. In the absence of evidence for the greater antiquity of accusative marking and especially for the form of the accusative case than the ergative, the two sources of ergative marking are essentially indistinguishable from each other.

4. Ergative agreement Ergative agreement in person, either just with the S and O or with O and P and also A, is considerably less common cross-linguistically than ergative nominal marking.6 This may to a large extent be attributed to what is considered to be the normal route

anna siewierska

238

for the development of person agreement, i.e. the reanalysis of unstressed pronouns occurring in topicalized constructions such as those in (15) or (16). (15) a. John, he left ages ago. b. The boy, he wrecked the car. c. The car, the boy really wrecked it. (16) a. As for you, you should go. b. As for me, I like the man. c. As for me, that won’t stop me. Since, as we have seen, in languages with split ergativity conditioned by the semantics of nominals, the pronouns are accusatively marked, it follows that when they turn into agreement markers bound to the verb, the resulting agreement system will also be accusative. There is an additional factor which strongly favours accusative agreement over ergative. In the vast majority of languages, if not in all, given information is associated with the A and to a lesser extent with the S, while new information is associated with the O and oblique constituents. If this is so, the development of agreement from unstressed pronouns in topicalizations is most likely to produce agreement with the most probable topics, i.e. with the A and S rather than P and S. In other words the agreement which is likely to emerge is accusative not ergative.7 If the forms of the A and S pronouns which constitute the source of the agreement markers are distinct, i.e. if they pattern ergatively, the resulting agreement system would be neither accusative nor ergative.8 But accusative agreement would emerge if the agreement marker of the A is extended to the S. According to Harris and Campbell (1995: 249), this is precisely what has happened in the northern dialects of the Daghestanian language Tabasaran. In the southern dialects the former distinction between A and S pronouns is partially preserved in the agreement system, as shown in (17).9 (17) Southern Tabasaran a. uzu gak’wler urgura-za I ﬁrewood:abs burn-1sg(erg) ‘I burn ﬁrewood.’ b. uzu urgura-zu I burn-1sg(abs) ‘I am on ﬁre.’ In the northern dialects, on the other hand, the agreement system is entirely accusative, the agreement marking of the A having been extended to the S, as we see in (18).

passive-to-ergative versus inverse-to-ergative

239

(18) Northern Tabasaran a. izu bisnu-za zaq’a (change) I catch-1sg bird ‘I caught a bird.’ b. izu t’irxnu-za I ﬂy-1sg ‘I ﬂew.’ In all, ergative SO agreement is highly unlikely to arise from unstressed pronouns in topicalized constructions such as those in (15) and (16). Ergative SO agreement is, however, a natural consequence of the passive-toergative scenario applied to languages with pre-existing accusative agreement. Since passive clauses are intransitive, the agreement marking in the passive is the same as in active intransitives, i.e. with the S. As a result of the reanalysis of the passive as active ergative, the S is reinterpreted as a P. Consequently, the agreement marker of the former S and now P is the same as that of the intransitive S. While the passive may not be the only source of ergative agreement, ergative agreement clearly cannot arise from a pre-existing accusative agreement system under the inverse-to-ergative scenario. Given that the patient in the inverse is a O not an S, i.e., that the inverse is not intransitive, once the inverse is reanalysed as an active ergative, accusative agreement marking of the O will not be the same as that of the S. And if the agent in the inverse also manifests agreement, the resulting agreement will be accusative, i.e. the same marker will be used for the A and S. In fact the preservation of accusative agreement as well as accusative pronominal marking, extends the scope of the inverse source of ergative nominal marking to languages with split ergativity involving ergative case and old as opposed to recent accusative agreement marking. If ergative agreement cannot evolve from the reinterpretation of a direct/inverse voice opposition in a language with a pre-existing accusative agreement system, but can evolve from the reanalysis of an active/passive voice opposition, ergative agreement provides a potentially strong argument for distinguishing the passive from the inverse source of ergative nominal marking. However, before we conclude that this is indeed so, we must yet consider whether the inverse-to-ergative reanalysis could not produce ergative agreement from the type of agreement system that languages with direct/inverse voice oppositions appear to favour. Though accusative agreement is cross-linguistically the most common type of agreement system, the languages which currently display direct/inverse voice oppositions tend to have either no agreement or hierarchical rather than accusative agreement. Hierarchical agreement is a type of agreement where the participant displaying agreement is determined by the ranking of the participant on the personal hierarchy, not by its grammatical relations or semantic role. For instance, in the

anna siewierska

240

Tanoan languages the hierarchy is 1 > 2 > 3. In the Algonquian languages it is 2 > 1 > 3 as shown by the examples in (19) and (20) from the previously mentioned Plains Cree. (19) Plains Cree a. ki-tasam-in 2-feed-dir ‘You feed me.’ b. ki-tasam-itin 2-feed-inv ‘I feed you.’ (20) a. ni-tasam-aw 1-feed-dir ‘I feed him.’ b. ni-tasam-ik 1-feed-inv ‘He feeds me.’ We see that in (19) the preﬁxal agreement marker ki- is the same in both the (a) and the (b) clause, though in the (a) clause the second person is the agent, while in the (b) clause it is the patient. The same holds for the clauses in (20). The agent vs patient status of the agreement preﬁx is indicated by the suﬃxal direct vs inverse markers. In clauses with two third person participants such as those given earlier in (5) and repeated for convenience in (21) there is no agreement preﬁx or alternatively the preﬁx is zero. (21) a. sekih-ew napew antim-wa scare-dir man:prox dog-obv ‘The man scares the dog.’ b. sekih-ik napew-a antim scare-inv man-obv dog:prox ‘The man scares the dog.’ Whether the clause is direct or inverse is indicated solely by the direct and inverse markers respectively. (Note that the direct and inverse markers are partially sensitive to person, i.e. for clauses involving only ﬁrst and second person participants, i.e. speech act participants (SAP), the markers are -in and -itin, while for all other clauses the markers are -aw/ew and -ik.) In intransitive clauses agreement is indicated by the same set of preﬁxes as in the direct and inverse: ni- for ﬁrst person, ki for second and zero for third. E.g.

passive-to-ergative versus inverse-to-ergative

241

(22) a. ni-pimipahta-n 1-run-dir ‘I run.’ b. pimoht-ew napew walk-dir man ‘The man is walking along.’ If the inverse clauses with nominal participants were to be reinterpreted as ergative with the obviative agent functioning as the transitive subject, the inverse marker -ik could (a) disappear; (b) be reanalyzed as a transitivity marker; (c) be reanalyzed as a portmanteau 3rd/3rd A/O (subject and object) agreement marker; or (d) be reanalyzed as a A agreement marker. If it were to be indeed reanalyzed as an agreement marker, irrespective of the actual analysis, the resulting agreement system would be neither accusative nor ergative, since in intransitive clauses a diﬀerent marker is used, i.e., the direct marker. Signiﬁcantly, the only way that ergative agreement could emerge is if the inverse marker were to be reanalyzed as a O agreement marker and then this marker were to be extended to intransitive clauses. However, given that in clauses involving mixed participants, i.e. ﬁrst or second person and third person such as (20b), the inverse marker could only be interpreted as an agent or A marker, it could hardly be interpreted as a O marker in clauses with third person participants such as (21b). In all, ergative agreement marking is highly unlikely to emerge from the type of agreement marking found in Algonquian or Tanoan inverse clauses with two nominal participants or two third person pronominal participants, for the matter. At ﬁrst sight the situation looks more promising with respect to ergative agreement if we take clauses with mixed participants (ﬁrst or second person plus third person) into account as in (20b) or (23b). (23) a. ni-sekih-a atim10 1-scare-dir dog ‘I scare the dog.’ b. ni-sekih-iko atim 1-scare-inv dog The dog scares me. Recall that in such inverse clauses the ﬁrst or second person is a patient and the third person an agent. Furthermore, it is always the ﬁrst or second person which is marked by the verbal preﬁx. Recall also that the same preﬁx occurs in intransitive clauses. Therefore if inverse clauses with mixed participants were to be reinterpreted as ergative, by analogy with clauses involving third person participants, the agree-

242

anna siewierska

ment preﬁx would be the same for the O in transitive clauses and the S of intransitives. We would thus have ergative SO agreement for the ﬁrst and second person. While a reanalysis such as the above could indeed produce ergative agreement, the consequences of the reanalysis are too drastic for it to ever take place. Note that if inverse clauses with mixed participants are reinterpreted as ergative, the ﬁrst and second person preﬁxes would be open solely to a patient reading. Therefore, in the absence of free pronouns, the language would have no means of expressing a situation where a ﬁrst or second person agent acts on a third person patient. In other words, it would be impossible to say (24) a. I hit him. or I hit the dog. b. You hit him. or You hit the dog. since a preﬁx occurring with a verb would always be interpreted as the O or the S but never as the A, The same, of course, applies to clauses involving only ﬁrst and second person participants. If the direct voice were to be lost, so to speak, and the second person preﬁx occurring in inverse clauses such as (19b) were to be reanalyzed as a O preﬁx, a clause with such a preﬁx could only mean I hit you but not You hit me. Needless to say, no language would tolerate a situation in which it would be impossible to express a ﬁrst or second person acting on a third or a second person acting on a ﬁrst. We can thus reaﬃrm our previous conclusion that ergative agreement is unlikely to arise from hierarchical agreement system as manifested currently in the Algonquian and also Tanoan languages and arguably any other language displaying hierarchical agreement in which the agreement markers are not sensitive to grammatical relations or semantic role. If the nominal marking in inverse clauses in such languages is reinterpreted as ergative, the agreement system will either remain hierarchical, change into an accusative one or end up as neither accusative nor ergative by virtue of portmanteau transitive A and O forms in transitive clauses. Since under the inverse-to-ergative scenario ergative agreement cannot arise from a pre-existing accusative or hierarchical agreement system, but is a natural consequence of a passive-to-ergative reanalysis, ergative agreement emerges as a pretty good diagnostic of the passive source of ergative nominal marking.

5. Conclusion In this paper I have sought to determine whether the inverse and passive sources of ergative marking may be distinguished from each other. I have attempted to do so by considering to what extent the two types of ergative marking claimed to be associated with each diachronic scenario, i.e. split ergativity conditioned by the seman-

passive-to-ergative versus inverse-to-ergative

243

tics of nominals in the case of the inverse, and ergativie agreement in the case of the passive, may be seen as indeed favouring the inverse and the passive respectively. My considerations reveal that while split ergativity conditioned by the semantics of nominals is in principle compatible with both an inverse and a passive source of the ergative marking, ergative agreement is strongly suggestive of the passive source of ergative marking. I have argued that the reanalysis of the inverse constitutes a more viable source of split acc/erg marking conditioned by the semantics of nominals than the reanalysis of the passive provided the accusative marking and especially the actual form of the accusative case predates that of the emergence of the ergative. But if there is no evidence of the greater antiquity of the accusative case form than that of the ergative, the two potential sources of split acc/erg marking conditioned by the semantics of nominals are essentially indistinguishable from each other. I have also argued that whereas ergative agreement is a natural consequence of the passive-to-ergative reanalysis applied to a language with a pre-existing accusative agreement system, it is highly unlikely to emerge from the reanalysis of an inverse. This should not be interpreted as implying that languages currently displaying ergative case marking and accusative agreement could have only evolved from the reanalysis of an inverse as opposed to a passive, since the accusative agreement may be subsequent to the emergence of the ergative nominal marking. But it does imply that the reanalysis of an inverse is not a promising source of current ergative agreement. In the preceding discussion I did not take into account the origins of split ergativity conditioned by tense and aspect. As far as I can see, the inverse is an unlikely source of ergative nominal marking in languages with such split ergativity. In languages with split ergativity determined by tense and aspect, the ergative marking occurs in the perfective or past, the accusative or other marking in the nonperfective or nonpast. In the case of the Indic and Iranian languages this split in case marking has been traced to the reanalysis of the periphrastic passive in the perfective (Anderson 1977: 336; Dixon 1994: 190). Though we have no historical records for other languages manifesting split ergativity conditioned by the semantics of nominals, the passive constitutes a more viable source of the ergative marking than the inverse, since there is a semantic similarity between the passive and the perfect but none between the inverse and the perfect.11 If both split ergativity conditoned by tense/aspect and ergative agreement clearly favour the passive source of ergative marking over the inverse, as I have argued, and furthermore split ergativity, be it of nouns vs pronouns or of case vs agreement marking, is in principle compatible with either diachronic scenario, the status of the inverse as a source of ergative marking independent of the passive emerges as somewhat questionable. Nonetheless, it would be premature to disregard such a

244

anna siewierska

possibility altogether. As discussed in section 3, a promising instance of ergative nominal marking attributable to the inverse as opposed to the passive-turned-inverse is that of the tripartite/ergative split reconstructed for Proto-Australian. Also promising are languages with ergative nominal marking and hierachical agreement marking. An ergative/hierarchical split is found among the Sino-Tibetan languages such as Nocte (Das Gupta 1971), Tangut (Ebert 1987), Limbu (van Driem 1987) and Chepang (Caughley 1982) and some traces of nominal ergativity can be discerned among several of the hierarchical agreement marking Carib languages, for instance, Kuikúro, Waiwai and Apalaí (Franchetto 1990). We have seen that the reanalysis of an inverse as an active ergative is not only likely to leave a pre-existing accusative but also a pre-existing hierarchical agreement system intact, so to speak. This suggests that the presence of hierarchical agreement may be a potential indicator of an inverse origin of ergative nominal marking. In any case, such a possibility is worth investigating. In sum, while the inverse undoubtedly needs to be taken into account as a potential source of ergative marking, the instances of ergative marking attributable to the reanalysis of an inverse as opposed to a passive-turned-inverse appear to be rather limited.

Notes 1. The samples of Nichols (1992) and Siewierska (1996) suggest that accusative case marking is about twice as common as ergative, the relevant ﬁgures being 61 per cent vs. 39 per cent (Nichols) and 53 per cent vs. 34 per cent (Siewierska). 2. The other major source of ergative nominal marking typically considered in the literature is the reanalysis of nominalizations in which the agent is expressed by means of a possessive phrase such as the enemy’s destruction of the city. Comrie (1978), however, questions whether this is a source of ergative marking independent of the passive since such nominalizations may well have been used as a device for forming passive constructions. Another source of ergative nominal marking suggested more recently by Garrett (1990) is that of oblique instrumental NPs in transitive clauses with covert As. This is a highly likely source of split ergativity conditioned by the semantics of nominals. In view of the fact that following discussion will be conﬁned to a consideration of the passive and inverse sources of ergative nominal marking, these other potential sources of ergative marking will not be considered. 3. Givón’s functional pragmatic deﬁnition of the inverse is somewhat controversial. Note that his deﬁnition encompasses OVS clauses in the Slavic languages, for example, in which the patient is typically more topical than the agent, which in turn may retain considerable topicality. 4. Sands (1996) argues that the basic allomorph of the ergative is actually *-Dhu and that *lu is an morphologically conditioned allomorph following nominals which are not common nouns. However,

passive-to-ergative versus inverse-to-ergative

245

as far as I can see, this does not aﬀect the current argument about the previous ergative marking of A pronouns. 5. Of the languages with agreement in Siewierska’s (1996) sample only 16% display ergative or split ergative agreement. The corresponding ﬁgure in the sample of Nichols (1992) is 11%. 6. Note that agreement with only the A but not the S, though possible, is less likely since intransitive clauses are more common in discourse than transitive. Agreement with only the A would be an instance of ergative agreement, but not of the type generally manifested in languages; languages with ergative agreement tend to display agreement with the S and O and also the A and more rarely agreement only with the S and P. In any case we would expect agreement solely with the A to be extend to the S, as outlined below. 7. Actually an agreement system in which the form of the A marker diﬀers from that of the S marker but in which there is no agreement with the O may be viewed as accusative since the S and A are grouped together in opposition to the O by virtue of displaying agreement. Note that such a system would not qualify as tripartite since no agreement marking of the O, unlike no O case marking, must be interpreted as absence of agreement rather than as agreement by means of a zero morpheme. 8. According to Kibrik (1979: 75) the southern dialects of Tabasaran actually display active agreement marking, i.e. the S has two types of agreement markers corresponding to the marking of the A and to the O respectively, the latter marker being the same as the original S pronoun. Thus whereas with some verbs the form of the ﬁrst person S agreement marker is -zu, as in (17b) with a verb like ‘ﬂy’ it is -za just as in (17a). Thus the extension of A to S marking found in the northern dialects is also partially evinced in the southern dialects. 9. In Algonquian the third person participant does not take obviative marking in clauses with mixed participants. Therefore such clauses are not in fact good candidates for reanalysis as ergative. However, in Tanoan languages the obviative marking on overt nominals occurs irrespective of whether only third person participants or both third and non-third are involved. 10. Passives, particularly periphrastic passives built on the auxiliary verb ‘be’ tend to focus on the state in which the patient is in, while perfects express the state resulting from a previous action. This ‘‘stative’’ nature of the passive is also partially due to the supression or the demotion of the agent. In inverse caluse, on the other hand, the agent is not supressed. Moreover, the traditional inverse, as found in the Amerindian languages, is never built on a participle.

References Anderson, Stephen R. 1977. ‘‘On Mechanisms by Which Languages Become Ergative.’’ In C. Li (ed.), Mechanisims of Syntactic Change. Austin: University of Texas Press, 317–63. Blake, Barry. 1977. Case Marking in Australian Languages. Canberra: Australian Institute of Aboriginal Studies. ——. 1987. Australian Aboriginal Grammar. London: Croom Helm.

246

anna siewierska

Caughley, Ross C. 1982. The Syntax and Morphology of the Verb in Chepang. Canberra: The Australian National University. Comrie, Bernard. 1978. ‘‘Ergativity.’’ In W.P. Lehmann (ed.) Syntactic typology: Studies in the phenomenology of language. Austin: University of Texas Press, 329–394. Das Gupta, K. 1971. An Introduction to the Nocte Language. Shillong: North-East Frontier Agency. Dixon, R.M.W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. ——. 1980. The Languages of Australia. Cambridge: Cambridge University Press. ——. 1994. Ergativity. Cambridge: Cambridge University Press. Douglas, W.H. 1981. ‘‘Watjarri.’’ In R.M.W Dixon and B. Blake (eds.) Handbook of Australian Languages, vol.2. Amsterdam: John Benjamins, 196–272. Ebert Karen H. 1987. ‘‘Grammatical Marking of Speech Act Participants in TibetoBurman.’’ Journal of Pragmatics 11: 473–482. Franchetto, Bruna. 1990. ‘‘Ergativity and Nominativity in Kuikúro and Other Carib languages.’’ In D. Payne (ed.) Amazonian Linguistics. Studies in lowland South American languages. Austin: University of Texas Press, 407–428. Garrett, Andrew. 1990. ‘‘The Origin of NP Split Ergativity.’’ Language 66, 261–96. Givón, Talmy. 1994a. ‘‘The Pragmatics of De-Transitive Voice: Functional and typological aspects of inversion.’’ In T. Givón (ed.), 3–44. ——. (ed.), 1994b. Voice and Inversion. Amsterdam: John Benjamins. Harris, Alice C. & Lyle Campbell 1985. Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Kibrik, Aleksandr. 1979. ‘‘Canonical Ergativity and Daghestan Languages.’’ In F. Plank (ed.), 61–77. Nichols, Johanna. 1992. Linguistic Diversity in Space and Time. Chicago: The University of Chicago Press. Plank, Frans (ed.) 1979. Ergativity. New York: Academic Press. Sands, K. 1996. The Ergative in Proto-Australian. Munich: Lincolm Europa. Siewierska, Anna. 1996. ‘‘Word Order Type and Alignment Type.’’ Sprachtypologie und Universalienforschung 49.2: 149–176. Silverstein, Michael. 1976. ‘‘Hierachy of Features and Ergativity.’’ In. R.M.W. Dixon (ed.), Grammatical categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 112–171. Sridhar, S.N. 1990. Kannada. London: Routledge. van Driem, George. 1987. A Grammar of Limbu. Berlin: Mouton de Gruyter Lobster Publications. Wolfart, H. C. 1973. ‘‘Plains Cree: A grammatical study.’’ Transactions of the American Philological Society n.s. 63, pt. 5.

Benefactive marking in Oceanic languages: From possessive classiﬁers to benefactive markers JAE JUNG SONG University of Otago

1. Introduction1 Pawley (1973: 120–6, 128, 140, 146) reconstructs the verbal suﬃx *-aki, and the prepositional verb *pani as two benefactive markers for Proto-Oceanic (or POc) (also see Harrison 1982, and Lichtenberk 1986).2 The Oceanic languages on which his reconstruction is based indeed express benefactive role by means of verbal sufﬁxes, prepositional verbs or other verbal devices (which can hereafter be referred to collectively as the V[erbal] type of benefactive marking). Thus, one may easily be led to think that the benefactive marking in contemporary Oceanic languages is largely verbal. This view, however, is premature and should not be taken to be conclusive in the absence of suﬃcient evidence. In this chapter, therefore, I will present a more accurate or balanced picture of the Oceanic benefactive marking by bringing together data from forty Oceanic languages. The investigation will reveal that the Oceanic group enjoys a much richer variety of benefactive marking than has hitherto been realized. In addition to the V type, the Oceanic group makes use of at least four nonverbal types of benefactive marking, namely (i) possessive classiﬁer type, (ii) preposition type, (iii) possessive suﬃx type, and (iv) case suﬃx type. The benefactive marking in Oceanic languages will also turn out to be more nonverbal than verbal, because more subgroups take advantage of the nonverbal types than of the V type. In particular, the possessive classiﬁer system, which generally encodes what is known as alienable possession in the literature (Lichtenberk 1983a: 148, Harrison 1988: 66, and Lynch 1973, in press), is exploited for benefactive marking in more subgroups than any of the other types (this type of benefactive marking will hereafter be referred to as the P[ossessive Classiﬁer] type).3 From a typological perspective, it comes as no surprise to ﬁnd that verbal ele-

248

jae jung song

ments (i.e. verbal aﬃxes, prepositional verbs, etc.), prepositions, and case suﬃxes are used for benefactive marking in the Oceanic group. But little has been written or understood about the other two types: the P type and the possessive suﬃx type. To say the least, they are typologically unusual or rare types of benefactive marking.4 I will, therefore, devote part of this chapter to discussion of the P type, and demonstrate that the possessive classiﬁer system has been grammaticalized as a very productive device for benefactive marking in widely distributed Oceanic languages. Unfortunately, I do not have anything to say about the possessive suﬃx type at the moment. The rest of the chapter is organized as follows. In Section 2, a survey of the benefactive marking in Oceanic languages will be presented. In Section 3, selected Oceanic languages will be closely examined in an attempt to characterize the development of the possessive classiﬁer into the benefactive marker. The chapter concludes with a brief summary in Section 4.

2. The survey For purposes of the present survey, I assumed Pawley and Ross’s (1995: 51–7) classiﬁcation of Oceanic, in which at least nine high-order or primary subgroups are recognized, as reproduced in Figure 1.5 I looked upon the Oceanic languages referred to or discussed in Lichtenberk (1985, 1986), and in Ross (1988) as a kind of ‘convenience’ database, from which

PROTO-AUSTRONESIAN PROTO-OCEANIC 1. Admiralty Islands 2. St. Matthias Islands 3. Western Oceanic 4. Southeast Solomonic 5. North/Central Vanuatu 6. South Vanuatu 7. Southern Oceanic 8. Central Pacic 9. [Nuclear] Micronesian Figure 1.

benefactive marking in oceanic languages

249

languages could freely be selected for purposes of the survey.6 I managed to collect relevant data on twenty nine of the languages cited there plus nine additional ones. Also chosen for inclusion in the survey were two uncontroversially Austronesian languages of the Eastern Outer Islands (the Solomon Islands), viz. Asumboa, and Tanimbili (Tryon 1994; cf. Lynch and Tryon 1985); the Austronesian languages of the area may possibly form a high-order subgroup within Oceanic in their own right (Malcolm Ross, pers. comm.). The resulting sample of forty languages covers all the high-order subgroups in Figure 1, except the Admiralty Islands, and St Matthias Islands subgroups, for which no relevant data are available at present. The names of the subgroups represented in the sample, followed by the actual number of surveyed languages in parentheses, are: Western Oceanic (11), Southeast Solomonic (7), North/Central Vanuatu (5), South Vanuatu (2), Southern Oceanic (4), Central Paciﬁc (5), Nuclear Micronesian (4), plus the Eastern Outer Islands (2). 2.1 Western Oceanic The languages in the sample belonging to this subgroup are: Kairiru (Wivel 1981), Manam (Lichtenberk 1983b), Mangap-Mbula (Bugenhagen 1986), Labu (Siegel 1984), Nakanai (Johnston 1980), Tigak (Beaumont 1979), Tolai (Mosel 1984), Balawaia (Kolia 1975), Nissan (Todd 1978), Banoni (Lincoln 1976), and Mono-Alu (Fagan 1986). Insofar as benefactive marking is concerned, the Western Oceanic subgroup seems to be the most diverse of all the high-order Oceanic subgroups to be surveyed in this chapter (cf. Ross 1988: 28, 382). The V type is found to play a signiﬁcant role in benefactive marking (Kairiru, Manam, Labu, Tigak, and Nissan). But the P type is almost equally common (Mangap-Mbula, Tolai, Balawaia, and Mono-Alu). Prepositions are also found to be in use in Tolai, Banoni, and possibly Labu. In addition, benefactive nominals are marked by adding inalienable possessive pronominal suﬃxes directly to verbs in Nakanai, whereas in Manam, a case suﬃx is used under certain circumstances to encode benefactive role. In Kairiru, the lexical verb ny- ‘to give’, which reﬂects POc *pani (Lichtenberk 1986: 15), is used to mark benefactive nominals, as illustrated in (1) and (2) (Wivel 1981: 43, 141).7 (1)

Kairiru Maki muli srri tai a-muom Maki orange small.group one 3sg:sbj-pick.oﬀ.a.tree ny-i Smouwai ben-3sg:obj Smouwai ‘Maki picked a small cluster of oranges for Smouwai.’

250 (2)

jae jung song Kairiru (ei) lil spai a-ny-am (kyau) (he) rice some 3sg:sbj-give-1sg:obj (I) ‘He gave me some rice.’

As can be seen in (1), the benefactive marker ny- takes object person/number marking, as does the lexical verb ‘to give’ in (2). For this reason, Wivel (1981: 143) calls ny- a ‘‘prepositional verb’’. Kairiru is classiﬁed as using the V type of benefactive marking. In Manam, benefactive role is expressed by adding the suﬃx -n (-m, -N) to the verb (Lichtenberk 1983b: 162, 165). This suﬃx is also taken to be a reﬂex of POc *pani (Lichtenberk 1986: 15). (3)

Manam tanépwa ʔáti da-sabár-Ø-a-n-i chief canoe 3pl:sbj:irls-hew-3sg:obj-bf-ben-3sg:obj ‘They will make (hew) a canoe for the chief.’

In addition, Manam uses an oblique case suﬃx -lo to mark benefactive nominals (Lichtenberk 1983b: 356–8). (4)

Manam ʔeʔá-i-lo da-té-ʔo 1pl:exc-bf-ben 3pl:sbj:irls-ﬁnd-2sg:obj ‘They will ﬁnd you for us.’

The exact meaning of the suﬃx -lo is determined on the basis of the linguistic or extralinguistic context. For this reason, Lichtenberk (1983b: 356) refers to it as a ‘‘general’’ suﬃx. Among its meanings are location, destination, point of origin in space, location in time, source, and instrument. Mangap-Mbula seems to have two possessive classiﬁers le and ko/ka. However, these possessive classiﬁers are hardly used to express possession ‘‘with but one very small set of exceptions’’ (for an actual example, see Bugenhagen [1986: 153]). In fact, they are rather used to express benefactive nominals. (5)

Mangap-Mbula Silas i-taara le-ng ke Silas 3sg:sbj-cut clf-1sg:poss tree ‘Silas cut a tree for me (to have/control).’

Siegel (1984: 105–6, 114) reports that Labu (or Hapa) has a benefactive preposition kô, which he believes to be diachronically derived from a lexical verb kô ‘to go to or to move to or towards’. Unlike verbs, however, this preposition does not host

benefactive marking in oceanic languages

251

verbal preﬁxes. So, it is possible that kô may be a full-ﬂedged preposition. (6)

Labu ini ô-sôhô hanô kô ai he 3sg:nrls-build house ben me ‘He’s building a house for me.’

Interestingly enough, Labu also seems to use a lexical verb -ta ‘to sit, to stay, or to reach’ for benefactive marking (Siegel 1984: 114). The sentence in (7) is the only example available from Siegel’s description. (7)

Labu ini gwa mala nê-ta ainialô he fut guardian 3sg:irls-reach boy ‘He’ll be guardian for the boy.’

Note that in (7), the benefactive marker -ta bears the subject/tense/modality preﬁx nê, as do ordinary lexical verbs. In Nakanai, the inalienable possessive pronominal suﬃx appearing on the verb seems to mark benefactive nominals. This possessive suﬃx must agree with the benefactive nominal in terms of number, person, and inclusiveness (Johnston 1980: 29), as in (8) (Johnston 1980: 40). (8)

Nakanai8 amiteu vulalai-a-la-e Luluai we (pl:exc) cut.garden-3sg-3sg:poss-nm village.leader ‘We cut a garden area for the village leader.’

The third person singular suﬃx -a, which is treated as the accusative marker, cooccurs obligatorily with the benefactive nominal (Johnston 1980: 28–9). Note the absence of both the accusative marker and the benefactive-cum-possessive suﬃx in (9), which lacks a benefactive nominal (Johnston 1980: 40). (9)

Nakanai amiteu vulalai we (pl:exc) cut.garden ‘We cleared a garden area.’

It is not clear from Johnston (1980) as to how the inalienable possessive suﬃx has actually come to be used for benefactive marking in Nakanai. Tigak has what Beaumont (1979: 43) calls a ‘‘referential relator’’, or ani (perhaps reﬂecting the POc *pani-). It is a kind of prepositional verb, marking benefactive nominals, as in (10) (Beaumont 1979: 43).

252 (10)

jae jung song Tigak ga aigot-i pok an-iri she:pst prepare-it food ben-3pl:obj ‘She prepared food for them.’

In Tolai, there are two possessive classiﬁers, ka- and a-, the former marking dominant possession, and the latter edible or subordinate possession (Mosel 1984: 34–9; and also Pawley 1973: 158–63). The possessive classiﬁer system is exploited to encode benefactive nominals, as in (11) ‘‘if it is understood that the beneﬁciary owns or will own what is aﬀected or produced by the action, or that it is determined for the beneﬁciary’’ (Mosel 1984: 172–3). (11)

Tolai i ga igir a-dir he ta cook.vegetables clf-3du:poss ‘He cooked vegetables for them.’

Tolai also seems to have a benefactive preposition, ta (Mosel 1984: 171–2; perhaps < POc preposition *ta [Pawley 1973: 148–9]). Balawaia is another ﬁne example of the P type, since benefactive role is ‘‘expressed in Balawaia by possessive phrases’’, as in (12), where the possessive classiﬁer suﬃxed by an appropriate possessive pronoun is exploited for the expression of benefactive role. In fact, ‘‘there are no [other] forms corresponding to ‘to’ [or ‘for’] in English’’ (Kolia 1975: 124–6, 142–3, 160). (12)

Balawaia tama-gu tari-gu ge-na gio kalato father-my brother-my clf-3sg:poss spear he:made:cmp ‘My father made a spear for my younger brother.’

Nissan makes use of the V type of benefactive marking (Todd 1978: 1200, 1205). (13)

Nissan ingo u pusaka weleher tar iana tomua I aux clean ben art ﬁsh you ‘I will clean a ﬁsh for you.’

Todd (1978: 1205) seems to regard the sequence of u pusaka weleher in (13) as a complex predicate group, thereby suggesting that weleher itself may be a verb. In fact, weleher is listed as a kind of intransitive verb in the lexicon (Todd 1978:1237), although its exact source is not identiﬁed at all. Banoni has a benefactive preposition ma, as in (14) (Lincoln 1976: 91–2, 101).

benefactive marking in oceanic languages (14)

253

Banoni na ko ta ghano sapaia ma ne borogho I cmp def.fut dig potatoes ben apl pig ‘I am digging potatoes for the pigs.’

Mono-Alu exploits the possessive classiﬁer system for benefactive marking, as exempliﬁed in (15). Like Tolai, this language has two possessive classiﬁers, sa(dominant) and e- (subordinate) (Fagan 1986: 19–40). (15)

Mono-alu boi tapoina sa-gu tala fana-lapu loai day many clf-1sg:poss men 1sg:fut-kill wild.thing ‘whenever I kill wild (pigs) for my men.’

2.2 Southeast Solomonic This subgroup is represented by Arosi (Capell 1971), Bugotu (Ivens 1933–5), Kwaio (Keesing 1985), Kwara’ae (Deck 1933–4), Inakona (Capell 1930), To’aba’ita (Lichtenberk 1984), and Ulawa (Ivens 1913–4). These languages are all distinctly of the V type, although To’aba’ita is, strictly speaking, a hybrid of the V and preposition types. Arosi has a benefactive marker tana-, which Capell (1971: 77) lists as one of what he calls ‘‘verbal relators’’, some of which in fact sometimes function as full verbs. No actual examples are available. In Bugotu, vani-, which Lichtenberk (1986: 24) relates historically to POc *pani, seems to be used as the benefactive marker (cf. Ivens 1933–5: 153, 169). In Kwaio, fa- or fe’a-, which is also a reﬂex of POc *pani (Keesing 1985: 165), is used for benefactive marking (Keesing 1985: 131, 149). (16)

Kwaio ngai e beri-a la’u mola 3sg:focal 3sg:sbj:ref steal-3sg:obj prt prt boo fa-na gila ifa-na pig ben-3sg:poss pl brother.in.law-3sg:poss ‘He stole pigs again for his brothers-in-law.’

(17)

Kwaio taunga’i fa-gu work ben-1sg:poss ‘work for me.’

Interestingly enough, the verbal-based benefactive marker in Kwaio bears posses-

254

jae jung song

sive suﬃxes, not object suﬃxes (for a plausible explanation as to how this came about by way of grammatical reanalysis, see Lichtenberk 1986: 63–8; cf. Nakanai in (8), wherein possessive suﬃxes alone express benefactive role). Deck (1933–4: 98–9, 137–9) reports that in Kwara’ae the benefactive marker is fua-, which is also a reﬂex of POc *pani- (cf. Lichtenberk 1986: 35), as illustrated in (18). (18)

Kwara’ae nia mae fua-ku he die ben-1sg:poss ‘He died for me.’

Kwara’ae is similar to Kwaio in that its benefactive marker also carries possessive suﬃxes, not object suﬃxes. Capell (1930: 123, 127) shows that the benefactive marker in Inakona is vani-, again a reﬂex of POc *pani- (Lichtenberk 1986: 11), as shown in (19). Unlike Bugotu or Kwara’ae, however, this language also uses vani- as a full lexical verb meaning ‘to give’, as in (20). (19)

Inakona ni nau vani-go na Lord he do ben-thee art Lord ‘The Lord has done (it) for you.’

(20)

Inakona igia ga vani-go na uvi he he:fut give-thee art yam ‘He will give you a yam.’

To’aba’ita exhibits a slight variation on the preceding Southeast Solomonic languages in that it combines fa/fe, a reﬂex of POc *pani, and a ‘‘semantically empty oblique’’ preposition i (Lichtenberk 1984: 58, 61–2). Also in common with Kwaio, and Kwara’ae, the benefactive sequence i fa/fe in To’aba’ita bears possessive, not object, suﬃxes (Lichtenberk 1986: 33), as in: (21)

To’aba’ita ku usi-a te’e ’abola faalu ’i fa-na kuke’e 1sg buy-3sg:obj one loincloth hew prep ben-3sg:poss old.woman nau 1sg ‘I bought a new loincloth for my wife.’

In Ulawa, the benefactive marker muni is also verbal in origin. For instance, it

benefactive marking in oceanic languages

255

bears object pronominal suﬃxes, just as transitive verbs do. This benefactive marker seems to be related to *muni, one of the prepositional verbs which Pawley (1973: 147) is rather reluctant to attribute to POc (see Ross 1988: 108, who does not reconstruct *muni for POc at all). 2.3 North/Central Vanuatu The North/Central Vanuatu languages in the sample are Atchin (Capell and Layard 1980), Nguna (Schütz 1969), Paamese (Crowley 1982), Big Nambas (Fox 1979), and Sakao (Guy 1974). This subgroup exhibits a relatively high degree of diversity: the P type (Atchin, Paamese, and Sakao), the V type (Nguna, and Paamese), the use of possessive pronominal suﬃxes (Big Nambas), and possibly a preposition (Atchin). Atchin may have a benefactive preposition i (perhaps < POc *(q)i, < Proto-Austronesian [or PAn] *(q)i), the exact distribution and function of which are yet to be determined (Capell and Layard 1980: 63). But it certainly puts its possessive classiﬁer system to use for benefactive marking (Capell and Layard 1980: 54–8). There are four possessive classiﬁers, sa- for general, na- for cooked food, ra- for raw food, and ma- for drink (Capell and Layard 1980: 56). To these possessive classiﬁers the possessive pronominal suﬃx is attached, regardless of whether or not the possessor is expressed by means of a full nominal. The use of the P type of benefactive marking is illustrated in (22) (Capell and Layard 1980: 57). (22)

Atchin wiʃewin ti-e-Ra mwi lai na-rän pwer-pwer’a woman of-Raga 3sg:pst take clf-3sg:poss sow ‘The Raga woman took food for that sow.’

Schütz (1969: 38, 59–60) identiﬁes magi- as the benefactive marker in Nguna. He (1969: 38) thinks that it is basically a verb, since it takes object pronominal suﬃxes, as in (23) (Schütz 1969: 60). (23)

Nguna p˜ a magi-nami soso na-atamoli ˜ taare sikai imp ben-us call person white one ‘Call a white man for us.’

Like Atchin, Paamese also has four possessive classiﬁers, aa- for food, mo- for drink or domestic purposes, so- for location or domesticated animals, and ono- for general (Crowley 1982: 211–5). These possessive classiﬁers are all exploited for benefactive marking, as exempliﬁed in (24) (Crowley 1982: 212, 215).

256 (24)

jae jung song Paamese tomakii vaa muumo one-nV asuvo-neke Tomaki 3sg:rls:go 3sg:rls:work clf-const chief-dis ‘Tomaki went and worked for that boss.’

In Paamese, there are also benefactive prepositional verbs, mini and veni (< POc prepositional verbs *muni, and *pani [Crowley pers. comm.]), which Crowley (1982: 195) calls ‘‘punctual dative case’’ and ‘‘areal dative case’’, respectively. One of the functions of these dative cases is to mark benefactive nominals. The ﬁrst is used to encode ‘‘the animate beneﬁciary of an action not involving the transfer of something’’, as in (25a) (Crowley 1982: 199), and the second to express the beneﬁciary ‘‘who is the one who receives the beneﬁt of some action that is not being done in one’s place but purely for one’s beneﬁt’’ (Crowley 1982: 199), as in (25b). (25)

Paamese a. kaiko ki-tai-e mini-e 2sg 2sg-dis-chop ben-3sg ‘You cut it for him (as he can’t do it).’ b. au-volu tahii veni pitiee 3pl:rls-dance Tahi ben bda ‘They danced at Tahi for bda [British District Agent].’

Big Nambas is similar to Nakanai from the Western Oceanic subgroup discussed in Section 2.1 in that the possessive pronominal suﬃxes are used as benefactive markers and attached to verbs, as exempliﬁed in (26) (Fox 1979: 82). (26)

Big Nambas nə-v-təhtah-ar 1:rls-pl-burn.oﬀ.garden-3pl:poss ‘We burnt oﬀ their garden for them.’

Finally, Sakao is not so diﬀerent from Atchin, and Paamese in that it also exploits its possessive classiﬁer system for benefactive marking (Guy 1974: 42–4). This language has seven possessive classiﬁers, na- for food, nne- for drink, hÅ- for general, ia- for shadow or vomit, βa- for smell, na- for shelter or location, and $nafor other functions.9 But only the ﬁrst three na-, nne-, and hɒ- are used for benefactive marking. (27)

Sakao a-kep tara ité hœ-m 1sg:irls-take pig art clf-2sg:poss ‘I will take a pig for you.’

benefactive marking in oceanic languages

257

2.4 South Vanuatu This subgroup is represented by Lenakel (Lynch 1978) and Anejom ˜ (John Lynch, pers. comm.). Both languages make use of the P type of benefactive marking. In Lenakel, the dative preposition kam is occasionally used to express benefactive nominals, as in (28) (Lynch 1978: 72). (28)

Lenakel r- m-hat k-pn n-aug n-aan kam kova ka 3sg-pst-lay.down-there nr-eat-nr ben child that ‘He laid down food for that child.’

However, the expression of benefactive nominals is carried out mainly by taha-, which also happens to be the general possessive classiﬁer in Lenakel, as exempliﬁed in (29) (Lynch 1978: 72, 93). There are four other possessive classiﬁers as listed in (30), but they are not pressed into service to express benefactive role (see Section 3.4 for further discussion). (29)

Lenakel a. uus-suaas ka r- m-am-asumw taha man-small that 3sg-pst-cont-garden ben r m-n iuok t to nimwa taha-k father-3sg:poss near loc house clf-1sg:poss ‘That boy was gardening for his father near my house.’ b. i- m- lh n m ker le n ki-nhamra taha-m 1:exc-pst-pick breadfruit one loc loc-bush ben-2sg:poss ‘I picked a breadfruit for you in the bush.’

(30)

Lenakel 1. Edible 2. Drinkable 3. Plant 4. Location 5. General

n kn mwneiimwa- (not obligatory; taha- may instead be used) taha-

Anejom ˜ is similar to Lenakel in that its possessive classiﬁer system is used to mark benefactive nominals. But all its ﬁve possessive classiﬁers—u-/uwu(general), lida- (suck-juice-from), uma˜ (place), luma˜ (drink), and inca- (food) —participate in benefactive marking, as illustrated in (31) (John Lynch, pers. comm.).

jae jung song

258 (31)

Anejom ˜ ek aworitai uwu-n 1sg:aor garden clf-3sg:poss ‘I am gardening for him.’ (i.e., I am helping him out with gardening)

2.5 Southern Oceanic This subgroup is represented by Dehu (or Drehu) (Tryon 1968a), Houailou (or Ajië) (Lichtenberk 1978), Iai (or Iaai) (Tryon 1968b), and Nengone (Tryon 1967). These languages all use prepositions to express benefactive role. The relevant prepositions are a t awan

or a t awai

in Dehu (32) (Tryon 1968a: 85), ki in Houailou (33) (Lichtenberk 1978: 90), hobi kö (for a pronoun or proper noun) or hobi kwɔu (for a common noun) in Iai (34) (Tryon 1968b: 94–5), and, ﬁnally, so (for pronouns) or sɔn (for nouns) in Nengone (35) (Tryon 1967: 74–5). (32)

Dehu eni a xom la tusi a tawan

la fø I take the book ben the woman ‘I take the book for the woman.’

(33)

Houailou na bã e˜ r˜e ki pãgara 3sg hab say ben European ‘He always gloriﬁes Europeans’

(34)

Iai a. oge l îɔ ke tεp hobi kwɔu kuli I killed a rat ben dog ‘I killed a rat for the dog.’ b. a l îɔ ke tεp hobi kö ña he killed a rat ben me ‘He killed a rat for me.’ c. a l îɔ ke uñ hobi kö david he killed a turtle ben David ‘He killed a turtle for David.’

(35)

Nengone a. inu n a udul ɔre wεge so bɔn I transported the raft ben him ‘I transported the raft for him.’

benefactive marking in oceanic languages

259

b. inu cˇi cˇεo kaka sɔn ɔre pailai I seek food ben the dog ‘I seek food for the dog.’ The benefactive prepositions in these Southern Oceanic languages are not at all like the prepositional verb, for instance, in Kwaio (see 2.2) in that they do not bear object pronominal suﬃxes when the beneﬁciary is expressed by means of a full nominal (as opposed to a pronominal). 2.6 Central Paciﬁc This subgroup is exempliﬁed by the following languages: Rotuman (Churchward 1940), Ma¯ori (Bauer 1993), Niuean (Seiter 1980), Rapanui (Du Feu 1996), and Samoan (Mosel and Hovdhaugen 1992). The subgroup seems to have one exponent of the P type of benefactive marking, namely Rotuman. The other four languages use benefactive prepositions. Rotuman has three ‘‘possessive prepositions’’, ne, ‘o/‘on, and ‘e/‘en (Churchward 1940: 31). The last of these possessive prepositions, ‘e/‘en, being reserved for food, drink, and a turn or an occasion to do X (Churchward 1940: 31), is reminiscent of the possessive classiﬁers of the other Oceanic languages which have so far been discussed. It is indeed understood to be derived historically from the POc possessive classiﬁer *ka- ‘edible/subordinate’ (Pawley 1972: 86; 1979: 21–2). Churchward (1940: 142) indicates that these possessive prepositions are sometimes used for benefactive marking, especially after a verb. Rotuman can thus be regarded as a language of the P type. (36)

Rotuman a. gou re¯ te¯ la ‘a¯ ma hoa‘ ‘e le Tausia I get thing purp eat and take clf prt Tausia se h aspet ta to hospital prt:sg ‘I got some food ready and took it to the hospital for Tausia.’ b. gou ho‘a-m ‘eu kalöf I take-dir you:poss egg:indef ‘I have brought some eggs for you.’

In Ma¯ori, benefactive nominals are marked by the preposition moo or maa, the choice between which depends largely on the same distinction as that between socalled O-possessive (or non-dominant/non-controlled) and A-possessive (or dominant/controlled) marking (Bauer 1993: 209–15, 271–2, 282–3 for further discussion of this distinction; also see Clark 1976, and Wilson 1976).

260 (37)

jae jung song Ma¯ori ka haere a Rona ki te tiki wai moo ana tamariki ta move pers Rona to the fetch water ben pl:gen:3sg children ‘Rona went to fetch water for her children.’

Niuean also makes use of a preposition, ma, to mark benefactive nominals (Seiter 1980: 36–7).10 (38)

Niuean tunu e au e ika ma Sione cook erg I abs ﬁsh ben Sione ‘I’m cooking the ﬁsh for Sione.’

When the beneﬁciary is expressed in pronominal form, possessive suﬃxes must be chosen (but cf. Ma¯ori), and ha- or what Seiter calls the ‘‘possessive marker’’ is used optionally. Rapanui can be said to be in between Ma¯ori and Niuean in that the choice between its benefactive prepositions, mo and ma, is symptomatic of the distinction between O-possessive and A-possessive marking (Du Feu 1996: 102, 119). And with pronominal benefactives these prepositions co-occur with possessive suﬃxes rather than free pronouns (Du Feu 1996: 66). (39)

Rapanui a. he ma’u takoa mai mo’oku acn bring also towards ben-1sg:poss ‘He brought one in for me too.’ b. ma’aku mau’a te haraoa nei i tunu ai ben-1sg:poss emph spe ﬂour ppd pst cook pho ‘It’s for me that she has baked the roll.’

In Samoan, the preposition mo or ma is used to express benefactive role. According to Mosel and Hovdhaugen (1992: 146), the ﬁrst preposition ‘‘introduces arguments denoting persons who are given something or for whom something is done’’ whereas the second ‘‘introduces arguments which refer to a person who beneﬁts from a state of aﬀairs, to the purpose and aim of an action, or to the duration of a state of aﬀairs.’’11 (40)

Samoan saka fa’a-malu¯lu¯ le talo ma¯ le ma’i boil caus-tender art taro ben art ill ‘Boil and mash the taro for the patient.’

benefactive marking in oceanic languages

261

2.7 Nuclear Micronesian The Nuclear Micronesian languages included in the survey are Kusaiean (Lee 1975), Mokilese (Harrison 1976), Marshallese (Bender 1969, Zewen 1977, and Pagotto 1987), and Gilbertese (Groves, Groves, and Jacobs 1985, Harrison in prep.). It is well known among Oceanists that all Micronesian languages except Gilbertese possess a very large number of possessive classiﬁers, ranging between ﬁfteen and twenty (Harrison 1988: 63). The use of the possessive classiﬁer system for benefactive marking is observed in Kusaiean and Mokilese. Marshallese is a language of the V type, and Gilbertese makes use of a nominal-based benefactive preposition. Kusaiean is a prime example of the P type of benefactive marking, since it employs no other ways of expressing benefactive role. This language has a very rich variety of possessive classiﬁers, in fact, 19 in all ranging from transportation to decoration (for detailed discussion, see Lee 1975: 110–8). This possessive classiﬁer system has ‘‘another important function, to indicate benefactors, in a verb phrase’’, as in (41) (Lee 1975: 261–3). (41)

Kusaiean nga mole-lah rais ah la-l Sohn 1sg:sbj buy-asp rice det clf-3sg:poss John ‘I have bought the rice for John.’

There are at least fourteen commonly used possessive classiﬁers in Mokilese (Harrison 1976: 130). The fact that the possessive classiﬁer system is exploited for benefactive marking is illustrated in (42) (Harrison 1976: 133). (42)

Mokilese ngoah insingeh-di kijinlikkoauoaw nih-mw 1sg:sbj write-asp letter clf-2sg:poss ‘I wrote a letter for you.’

The sequence of the direct object nominal and the benefactive nominal in (42) can be switched around without causing any change in meaning, as in (43), although such a change of word order may occasionally bring about a change in meaning (Harrison 1976: 133).12 (43)

Mokilese ngoah insingeh-di nih-mw kijinlikkoauoaw 1sg:sbj write-asp clf-2sg:poss letter ‘I wrote a letter for you.’

Unlike the foregoing two Micronesian languages, Marshallese does not exploit its possessive classiﬁer system for benefactive marking. It instead makes full use

jae jung song

262

of a benefactive preposition ñan (Pagotto 1987: 266–76), as in (44) (Pagotto 1987: 262). (44)

Marshallese i-ar wia-ik pinjel eo ñan Mary 1sg:sbj-cmp buy-tr pencil the ben Mary ‘I bought the pencil for Mary.’

Although it is categorized as a preposition, Bender (1984: 454) points out that ñan has a verbal origin (cf. Pagotto 1987: 272–3, 756–60; perhaps reﬂecting Proto-Micronesian [or PMc] *anga-ni [Harrison 1977: 182–4]).13 Indeed, just as bona ﬁde transitive verbs, it may bear the object agreement marker -e (for third person singular human or nonhuman referents) or -i (for third person plural nonhuman referents) (Bender 1984: 443), as in (45) (Pagotto 1987: 754). (45)

Marshallese lökake ok l ñan-e take(tr) dir to-3sg:obj ‘Take it to him.’

I will thus take ñan to be a prepositional verb, thereby classifying Marshallese to be an example of the V type. Gilbertese uses a body part term as the basis of its benefactive marker, that is buki- ‘buttock’ or ‘end’ (Groves, Groves, and Jacobs 1985: 23, Harrison 1989: 218, 220). This body part term combines with a general oblique preposition i (Harrison 1989: 230–3; < POc *(q)i, Pawley 1972, 1973, Hooper 1985; < PAn *(q)i, Dahl 1975, Pawley and Reid 1979, Blust 1974, 1977, Reid 1978, 1981) to form the benefactive marker. (46)

Gilbertese te mm’aane e a tia ni katea te auti i art man 3sg:sbj pfv build art house prep buki-n te aine prep-3sg:poss art woman ‘The man built a house for the woman.’

For purposes of the present survey, i buki- is taken to be a preposition, albeit nominal-based. 2.8 Eastern Outer Islands At present there is little reliable data on the languages of this area, viz. Nembao, Asumboa, Tanimbili, Buma, Vano, and Tanema (Tryon 1994: 613). The data in

benefactive marking in oceanic languages

263

Tryon (1994; also pers. comm.) suggest that these languages all employ possessive classiﬁers for benefactive marking. Two cases in point from Asumboa and Tanimbili are illustrated in (47), and (48) (Tryon 1994: 623, 629). (47)

Asumboa nyigo-ge na-u numuo e-ge sika want-I I-build house clf-my one ‘I want to build a house for myself.’

(48)

Tanimbili name nyi-u nomwa i-gu fut I-build house clf-my ‘I shall build a house for myself.’

2.9 Summary The preceding survey has revealed that there are at least ﬁve diﬀerent types of benefactive marking in use in the Oceanic group: (i) V type, (ii) P type, (iii) preposition type, (iv) possessive suﬃx type, and (v) case suﬃx type. The ﬁrst two are the most common types of benefactive marking in the Oceanic group, since more Oceanic languages employ the V type or the P type than any other type. The use of benefactive prepositions is almost as common as the V or P type. Possessive and case sufﬁxes are the least common devices for benefactive marking in Oceanic, the former found in Nakanai and Big Nambas, and the latter in Manam. The distribution of the types of benefactive marking in the Oceanic group is summarized in Table 1. It is clear from Table 1 that benefactive marking in Oceanic Table 1. Western Oceanic Southeast Solomonic North/Central Vanuatu South Vanuatu Southern Oceanic Central Paciﬁc Nuclear Micronesian Eastern Outer Islands

V

P

prep

poss

case

+ + +

+

+

+

+

+ +

? + + + +

+

+

+ + +

Note: V = verbal type, P = possessive classiﬁer type, prep = preposition type, poss = possessive suﬃx type, case = case suﬃx type, + = attested, ? = possibly present

264

jae jung song

languages is not as verbal as one is led to expect on the basis of the previous studies alluded to at the beginning of this chapter. In particular, the P type is more widespread in terms of subgroups than the V type, i.e. six subgroups versus four (see Song [1997] for the proposal that the benefactive marking function of the possessive classiﬁer system be reconstructed for POc, and for its implications for the history of Nuclear Micronesian possessive classiﬁers). It may be said, therefore, that the P type is more characteristic of the Oceanic group than the V type.

3. The P-Type of benefactive marking I have so far indicated simply that in languages such as Tolai, Lenakel, Paamese, Kusaiean, etc., the possessive classiﬁer system is enlisted in the service of expressing benefactive role. In this section, I will examine additional data from some of the P-type languages with a view to gaining an understanding of the possessive classiﬁer-cum-benefactive marker from the perspective of grammaticalization. I will make no attempt here to adumbrate the semantic and/or cognitive relationship between possessive and benefactive (but see n. 4), but conﬁne discussion to grammatical processes associated with the grammaticalization in question. 3.1 From possessive classiﬁers to benefactive markers The possessive classiﬁer normally appears with the possessor and the possessed, linking the latter two in terms of semantic or culturally signiﬁcant relationships (see n. 3). But as the possessive classiﬁer is exploited for the expressing of benefactive role, the possessor is interpreted as the beneﬁciary, and the possessed as the entity that is intended for the beneﬁt of the beneﬁciary, as is illustrated in Figure 2, which involves a typical transitive clause with a direct object possessive nominal phrase, e.g. A repaired B’s canoe. Stage I represents the situation wherein the possessive classiﬁer carries out its original function or expresses alienable possession, whereas Stage II depicts the situation wherein the possessive classiﬁer functions as the benefactive marker. The original possessive nominal phrase (enclosed within Ys in Stage I) is seen here to break up into three diﬀerent parts in the process of grammaticalization (Stage II), whereby the possessed and possessor nominals become separate nominals with the possessive classiﬁer being reanalysed as the benefactive marker. As far as their categorial status is concerned, there is not much change involved with the possessed and possessor nominals, both of which remain as nominals in Stage II. But the grammatico-functional change—or the semantic role change as the case may be—of the possessor nominal is nothing but dramatic. The possessor assumes a separate role of benefactive in Stage II although the possessed nominal inherits the grammat-

benefactive marking in oceanic languages

265

Stage I: np[X]np

VERB np[Y[POSSESSED, POSS-CL, POSSESSOR]Y]np

Stage II: np[X]np

VERB

np[Y]np BEN-MARKER

np[Z]np

Note: x=subject, y=direct object, z=beneﬁciary, poss-cl=possessive classiﬁer, ben-marker=benefactive marker, →=direction of change. Figure 2. ical function of the original possessive nominal phrase (that is, direct object). The development of the possessive classiﬁer into the benefactive marker can further be studied in terms of: (i) word order change; (ii) generalization (Bybee and Pagliuca 1985: 62–3, 67, 71–5); and (iii) specialization (Bybee and Pagliuca 1985: 62–3, 67, 71–5, Hopper 1991: 25–8, Heine, Claudi, and Hünnemeyer 1991: 109, 157, and Hopper and Traugott 1993: 100–3, 113–6). 3.2 Word order change In Stage I, the possessor is part of the whole direct object possessive nominal phrase, or is associated directly with the possessed (i.e. the possessed and the possessor being the head and dependent, respectively, of the possessive nominal phrase). But in Stage II, it is related structurally to the verb as the benefactive nominal. This change may also be reﬂected in the diﬀerent positions which the sequence of the possessive classiﬁer and the possessor nominal occupies within the clause, depending on whether the possessive classiﬁer encodes alienable possession or benefactive role. This clearly is the case in Kusaiean. In (49), the sequence, being used to express alienable possession, is part of the whole possessive nominal phrase, rais la-l Sohn ah ‘John’s rice’. This is why the sequence in question appears to the left of or before the determiner ah, which must occupy the ﬁnal position of the noun phrase in Kusaiean (Lee 1975: 237). (49) Kusaiean nga mole-lah rais la-l Sohn ah 1sg:sbj buy-asp rice clf-3sg:poss John det ‘I have bought John’s rice.’ In (41) repeated below as (50), on the other hand, the same sequence encodes a benefactive nominal. In other words, it is no longer part of the possessive nominal phrase, thereby appearing to the right of or after the determiner ah. Its position indicates unequivocally that it does not belong to the direct object nominal phrase.

266

jae jung song

(50) Kusaiean nga mole-lah rais ah la-l Sohn 1sg:sbj buy-asp rice det clf-3sg:poss John ‘I have bought the rice for John.’ The development of the possessive classiﬁer into the benefactive marker has thus resulted in the possessor nominal physically moving out of the original possessive nominal phrase (along with the ‘new’ benefactive marker), as it were, and becoming a separate benefactive nominal, now associated directly with the verb. The benefactive marking function of the possessive classiﬁer system in Kusaiean is further supported by (51) (Lee 1975: 262). (51) Kusaiean Sohn el mole-lah ik la-l Sepe ah la-l Sru John 3sg:sbj buy-asp ﬁsh clf-3sg:poss Sepe det clf-3sg:poss Sru ‘John has bought Sepe’s ﬁsh for Sru.’ In (51), the same possessive classiﬁer la- is used twice in the same clause, once for the encoding of possession, and once again for the expressing of the benefactive nominal. A similar change in word order also seems to have taken place in Lenakel. Lynch (1978: 93) points out that in (29.b) renumbered here as (52), if the taha-marked nominal is placed immediately after the direct object nominal nIm ker ‘a breadfruit’, it is more likely to be interpreted as the possessive classiﬁer proper plus the pronominal suﬃx (i.e. alienable possession) than as a benefactive nominal. Note that in (52) the sequence of the ‘‘possessive classiﬁer’’ and the ‘‘possessor’’ is separated from the ‘‘possessed’’ by a locative phrase, le nIki-nhamra (note that the erstwhile functions of the possessive classiﬁer, possessor, and possessed are hereafter indicated within double quotation marks). (52)

Lenakel i- m- lh n m ker le n ki-nhamra taha-m 1:exc-pst-pick breadfruit one loc loc-bush ben-2sg:poss ‘I picked a breadfruit for you in the bush.’

Compare (52) with (53), wherein the possessive classiﬁer only encodes alienable possession. (53)

Lenakel kuri miin taha uus mil aan k-n-ai-ak mw ita dog pl clf man du that 3nsg-pfv-pl-run.away already ‘Those two men’s dogs have run away.’

In (53), the possessed is followed immediately by the sequence of the possessive classiﬁer and the possessor.

benefactive marking in oceanic languages

267

3.3 Generalization Once it has been realigned structurally with the verb as the full-ﬂedged benefactive nominal, the ‘‘possessor’’ nominal may also begin to be used on its own, without the ‘‘possessed’’ nominal being present in the same clause, as long as the ‘‘possessive classiﬁer’’ is there to mark its beneﬁciary role. That is to say that the use of the possessive classiﬁer for benefactive marking may also be extended or generalized to intransitive verbs, which lack direct object nominals. An early sign of this generalization is also observed in Kusaiean, wherein the possessive classiﬁer can be used for benefactive marking in conjunction with intransitive verbs as well, although the latter must co-occur with their ‘‘included’’ or incorporated direct object nominals (Lee 1975: 270–7; see Sugita 1973 for object incorporation in Micronesian). Being amalgamated with the verb, the included direct object nominal (or the ‘‘possessed’’) is no longer associated with the ‘‘possessor’’. In this respect, Mokilese is more advanced than Kusaiean in that the exploitation of the possessive classiﬁer for benefactive marking is not only found in transitive clauses, as in (42) repeated below as (54), but also in clauses without overt direct object (or ‘‘possessed’’) nominals or in ordinary intransitive clauses, as in (55) (Harrison 1976: 133).14 (54)

Mokilese ngoah insingeh-di kijinlikkoauoaw nih-mw 1sg:sbj write-asp letter clf-2sg:poss ‘I wrote a letter for you.’

(55)

Mokilese lih-o doadoa a-h woman-that sew clf-3sg:poss ‘That woman sews for him.’

In other words, being subject to fewer restrictions on its occurrence the use of the possessive classiﬁer for the expression of benefactive role is permitted in a wider range of grammatical contexts in Mokilese than in Kusaiean, thereby suggesting a higher degree of grammaticalization in the former (cf. Heine, Claudi, and Hünnemeyer 1991: 109, 157; and Hopper and Traugott 1993: 100–3). 3.4 Specialization Further evidence for the development of the possessive classiﬁer as the benefactive marker comes from the fact that only one of the multiple possessive classiﬁers, which are all used to express alienable possession in a given language, is chosen and

jae jung song

268

enlisted in the service of benefactive marking. In other words, one of the possessive classiﬁers is specialized for purposes of benefactive marking. Recall that such is the case in Lenakel, wherein of the ﬁve possessive classiﬁers (cf. 30), only the general one is pressed into service to express benefactive role. (56)

Lenakel a. uus-suaas ka r- m-am-asumw taha r m-n man-small that 3sg-pst-cont-garden ben father-3sg:poss iuok t to nimwa taha-k near loc house clf-1sg:poss ‘That boy was gardening for his father near my house.’ b. i- m- lh n m ker le n ki-nhamra taha-m 1:exc-pst-pick breadfruit one loc loc-bush ben-2sg:poss ‘I picked a breadfruit for you in the bush.’

In (29a) repeated here as (56a) (Lynch 1978: 72, 93), there are no ‘‘possessed’’ nominals which the taha-marked nominals could be associated with, even if tahawere to be interpreted as the general possessive classiﬁer (i.e. possession). In fact, taha- is used here twice, once as the benefactive marker and once again as the possessive classiﬁer proper (cf. 51). It is also important to note that the semantics of the general possessive classiﬁer taha- must be abstract enough—‘‘it lacks certain speciﬁc features of meaning’’ (Bybee and Pagliuca 1985: 63)—to be compatible with that of the nominal which would otherwise be associated with it as the possessed, e.g. the direct object nominal nIm ker ‘a breadfruit’ in (56b). Lenakel thus contrasts with Kusaiean, and Mokilese, wherein diﬀerent possessive classiﬁers must be selected, depending on the semantic or cultural relationship between the benefactive and direct object nominals. In these languages, the possessive classiﬁers can only be used to express benefactive role in the same limited set of environments in which they express possession. This suggests strongly that the possessive classiﬁer taha- in Lenakel is at a far more advanced stage of grammaticalization as a benefactive marker. It is found in a far wider range of semantic contexts.

4. Conclusion In this chapter, a survey of benefactive marking in Oceanic languages has been presented. The survey has exposed two points of empirical importance. First, there is a far richer variety of benefactive marking in Oceanic than has hitherto been assumed. Second, the P type of benefactive marking is more widespread in Oceanic languages than the V type of benefactive marking, which was reconstructed for POc

benefactive marking in oceanic languages

269

or taken to be characteristic of Oceanic languages in the previous studies. In the second part of the chapter the exploitation of possessive classiﬁers for benefactive marking has been discussed from a perspective of grammaticalization under the headings of (i) word order change; (ii) generalization; and (iii) specialization. The possessor nominal may be separated physically from the possessive nominal phrase as the possessive classiﬁer develops into a benefactive marker. Moreover, the use of the possessive classiﬁer for benefactive marking may be extended to intransitive clauses, which lack direct object (possessive) nominal phrases; and one of the possessive classiﬁers may also be selected and specialized for the service of benefactive marking.

Notes 1. I am grateful to Byron Bender, Terry Crowley, Terab’ata Groves, Frank Lichtenberk, John Lynch, Louise Pagotto, Malcolm Ross, and Darrell Tryon for answering a number of questions that I had about the languages that they specialize in. I am also indebted to Anna Siewierska for her comments and suggestions on an earlier draft. I alone bear full responsibilities for claims and conclusions made here as well as for errors of fact or judgment that may still remain. 2. Pawley’s work (1973) is based on a convenience sample of twenty odd languages. Furthermore, both *aki- (Harrison 1982: 183–93) and *pani (Pawley 1973: 143; Lichtenberk 1986: 52–6) are, in fact, claimed to have been independent lexical verbs in POc. John Lynch (pers. comm.) is of the opinion, however, that there is no general agreement as to whether *-aki was an independent lexical verb in POc. 3. The possessive classiﬁer construction speciﬁes a semantic or culturally signiﬁcant relation between the possessor and the referent the nominal associated with the possessive classiﬁer refers to (that is, the possessed), as exempliﬁed in (i) below (na- ‘edible’, suhno- ‘plant’, and osrwac- ‘raw, uncooked’). (i) Kusaiean a. nu na-k coconut clf-1sg:poss ‘my eating coconut’ b. mos suhno-m breadfruit.tree clf-2sg:poss ‘your breadfruit tree’ c. ik osrwac-l ﬁsh clf-3sg:poss ‘his (or her) raw ﬁsh’ For example, given possessor X and possessed Y, the ‘edible’ possessive classiﬁer expresses that Y is food to X. It is also possible that possessed Y, e.g. coconut, can enter into more than one relationship, e.g. food, drink, and plant, with possessor X (Lynch 1973: 76 refers to this phenomenon as

270

jae jung song

‘‘overlap’’). This is why Lichtenberk (1983a: 148) chooses to refer to possessive classiﬁers as relational classiﬁers as opposed to ‘‘sortal’’ (e.g. Allan 1977) or ‘‘mensural’’ classiﬁers. 4. Terry Crowley points out that the use of possessive for benefactive marking may be fairly widespread in unrelated languages. However, I am not aware of any typological studies that conﬁrm this. Although there is a fair amount of evidence for the syncretism of possessive and dative being crosslinguistically common (e.g. Blake 1994: 45–6 on this point), it remains to be seen whether the same can be said of possessive and benefactive. For example, many Australian languages do not have possessive separate from dative, and dative-possessive also covers benefactive. But this is not surprising since dative tends to be metaphorically derived directly from benefactive, whereas possessive in turn tends to be metaphorically derived directly from dative (Heine, Claudi, and Hünnemeyer 1991:150–61). In Oceanic languages, however, dative tends to be marked diﬀerently if the possessive classiﬁer is used to express both possessive and benefactive. Furthermore, the direction of the functional extension associated with the possessive classiﬁer system in Oceanic languages seems to be from possessive to benefactive, because not all languages with the possessive classiﬁer system use that system for benefactive marking. 5. In Pawley and Ross’s subgrouping of Oceanic in Figure 1, the Eastern Outer Islands languages (especially those of Utupua and Vanikoro), the Oceanic languages of Irian Jaya, and Yapese are not represented due to lack of evidence (Pawley and Ross 1995: 65). Malcolm Ross (pers. comm.) also informs me that the languages of Utupua and Vanikoro were omitted from the subgrouping in Figure 1 because they did not ﬁt anywhere except as a potential high-order subgroup. 6. I have decided to ignore the Santa Cruz/Reefs languages such as Malo, Nemboi, Reefs, and Nooli, which Lichtenberk (1985: 132–3) includes in his sample as Oceanic languages (also see Lincoln 1978), because their genetic position seems to be very much in dispute (Wurm 1969, 1978; but cf. Tryon 1994: 638). 7. Lichtenberk (1986: 36–7) argues that a doublet of *pani and *pañi must be reconstructed for POc in lieu of Pawley’s (1973) single protoform *pani. 8. The third person singular suﬃx -a is a personal pronoun root. The subject pronouns are produced by adding the personal noun marker e- to personal pronoun roots such as -a. In singular object marking, only the pronoun root is suﬃxed directly to the verb, whereas in plural object marking, the compound form (consisting of the pronoun root and the personal noun marker) appears on its own (Johnston 1980: 180–1). 9. The symbol ‘$’ represents what Guy (1974: 17) refers to as ‘‘a mater vocalium’’, a vowel which is ‘‘unspeciﬁed for height, fronting or backing, rounding or unrounding’’ and distinctive features of which ‘‘are wholly determined by its phonological environment.’’ 10. The distinction between O-possessive and A-possessive marking is found in all the Polynesian languages except Niuean (Biggs 1971: 470). 11. Duranti (1981: 170–4) observes that in Samoan the possessive noun phrase can actually be interpreted as expressing benefactive role. So, (i) and (ii) both contain a possessive noun phrase each, but the ‘possessor’ is understood to be a beneﬁciary, not a possessor.

benefactive marking in oceanic languages

271

(i) Samoan sou lu ai e ea’ai amaiki ’oga I o omp ake rt ood ﬁds chool ‘I am going to make some food for the kids at school.’ (ii) Samoan . . . ai e akou e’epe’e . . . ake ome ur ream ‘. . . make some cream for us all.’ Other Central Paciﬁc languages do not seem to be like Samoan, although this has to be empirically tested. 12. Harrison (1976: 133) says that he does not have any explanation for the conditions under which such a change in word order causes a change in meaning, as shown in (i) and (ii) below. (i) Mokilese Ngoah dupuk-la woaroa-mw pohspas 1sg:sbj buy-asp clf-2sg:poss boat ‘I bought a boat for you.’ (ii) Mokilese Ngoah dupuk-la pohspas woaroa-mw 1sg:sbj buy-asp boat clf-2sg:poss ‘I bought a boat from you.’ 13. Marck (1991: 226) only proposes *fanga for the PMc verb meaning ‘to give’, whereas Harrison (1977: 184) reconstructs a doublet PMc *fanga/*anga-ni, the former functioning as a main verb ‘to give’, and the latter more frequently as a prepositional verb meaning ‘towards’. 14. It is not clear at all from Harrison (1976) whether or not all of the possessive classiﬁers are used for benefactive marking in conjunction with intransitive clauses.

References Allan, Keith. 1977. ‘‘Classiﬁers.’’ Language 53: 285–311. Bauer, Winifred. 1993. Ma¯ori. London: Routledge. Beaumont, Clive H. 1979. The Tigak Language of New Ireland. Canberra: Australian National University [Paciﬁc Linguistics B–58]. Bender, Byron W. 1969. Spoken Marshallese. Honolulu: University of Hawaii Press. ——. 1984. ‘‘Object Marking in Marshallese.’’ In Byron W. Bender (ed.), Studies in Micronesian Linguistics. Canberra: Australian National University [Paciﬁc Linguistics C–80], 443–65. Biggs, Bruce. 1971. ‘‘The languages of Polynesia.’’ In Thomas A. Sebeok (ed.), Current Trends in Linguistics, vol. 8: Linguistics in Oceania. The Hague: Mouton, 466–505. Blake, Barry J. 1994. Case. Cambridge: Cambridge University Press.

272

jae jung song

Blust, Robert. 1974. ‘‘Proto-Austronesian Syntax: The ﬁrst step.’’ Oceanic Linguistics 13: 1–15. ——. 1977. ‘‘The Proto-Austronesian Pronouns and Proto-Austronesian: A preliminary report.’’ University of Hawai’i Working Papers in Linguistics 9.2: 1–15. Bugenhagen, Robert D. 1986. ‘‘Possession in Mangap-Mbula: Its syntax and semantics.’’ Oceanic Linguistics 25: 124–66. Bybee, Joan L. and William Pagliuca. 1985. ‘‘Cross-Linguistic Comparison and the Development of Grammatical Meaning.’’ In Jacek Fisiak (ed.), Historical semantics: Historical word formation. Berlin: Mouton, 59–83. Capell, Arthur. 1930. ‘‘The Language of Inakona, Guadalcanar, Solomon Islands.’’ Journal of the Polynesian Society 39: 113–36. ——. 1971. Arosi Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–20]. —— and John Layard. 1980. Materials in Atchin, Malekula: Grammar, vocabulary and texts. Canberra: Australian National University [Paciﬁc Linguistics D–20]. Churchward, C. Maxwell. 1940. Rotuman Grammar and Dictionary. Sydney: Methodist Church of Australasia. Clark, Ross. 1976. Aspects of Proto-Polynesian Syntax. Auckland: Linguistic Society of New Zealand. Crowley, Terry. 1982. The Paamese Language of Vanuatu. Canberra: Australian National University [Paciﬁc Linguistics B–87]. Dahl, Otto C. 1973. Proto-Austronesian. Stockholm: Studentlitteratur. Deck, Norman C. 1933–4. ‘‘A Grammar of the Language Spoken by the Kwara’ae People of Mala, British Solomon Islands.’’ Journal of the Polynesian Society 42: 33–48, 133–44, 241–56; 43: 1–16, 85–100, 163–70, 246–57. Du Feu, Veronica. 1996. Rapanui. London: Routledge. Duranti, Alessandro. 1981. The Samoan fono: A sociolinguistic study. Canberra: Australian National University [Paciﬁc Linguistics B–80]. Fagan, Joel L. 1986. A Grammatical Analysis of Mono-Alu (Bougainville Straits, Solomon Islands). Canberra: Australian National University [Paciﬁc Linguistics B–96]. Fox, G. J. 1979. Big Nambas Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–60]. Groves, Terab’ata R., Gordon W. Groves and Roderick Jacobs. 1985. Kiribatese: An outline description. Canberra: Australian National University [Paciﬁc Linguistics D–64]. Guy, Jaques B.M. 1974. A Grammar of the Northern Dialect of Sakao. Canberra: Australian National University [Paciﬁc Linguistics B–33]. Halim, Amran, Lois Carrington and Stephen A. Wurm (eds). 1982. Third International Conference on Austronesian Linguistics, vol. 1: Currents in Oceanic. Canberra: Australian National University [Paciﬁc Linguistics C–74]. Harrison, Sheldon P. 1976. Mokilese Reference Grammar. Hawaii: University Press of Hawaii. ——. 1977. Some Problems in the History of Mokilese Morpho-Syntax. Ph.D. dissertation, University of Hawai’i.

benefactive marking in oceanic languages

273

——. 1982. ‘‘Proto-Oceanic *aki(ni) and the Proto-Oceanic Periphrastic Causative.’’ In Halim et al. (eds), 179–230. ——. 1988. ‘‘A Plausible History for Micronesian Possessive Classiﬁers.’’ Oceanic Linguistics 27: 63–78. ——. 1989. ‘‘Lexically-Governed Preposition Choice in Gilbertese.’’ In Ray Harlow and Robin Hooper (eds), VICAL 1, Oceanic Languages: Papers from the ﬁfth international conference on Austronesian linguistics. Auckland: Linguistic Society of New Zealand, 211–36. ——. In prep. Gilbertese Reference Grammar. Heine, Bernd, Ulrike Claudi and Friederike Hünnemeyer. 1991. Grammaticalization: A conceptual framework. Chicago: University of Chicago Press. Hooper, Robin. 1985. ‘‘Proto-Oceanic *qi.’’ In Andrew Pawley and Lois Carrington (eds), Austronesian Linguistics at the 15th Paciﬁc Science Congress. Canberra: Australian National University [Paciﬁc Linguistics C–88], 141–67. Hopper, Paul. 1991. ‘‘On Some Principles of Grammaticization.’’ In Elizabeth Closs Traugott and Bernd Heine (eds), Approaches to Grammaticalization, vol. 1: Focus on theoretical and methodological issues. Amsterdam: John Benjamins, 17–35. Hopper, Paul and Elizabeth Closs Traugott. 1993. Grammaticalization. Cambridge: Cambridge University Press. Ivens, W. G. 1913–4. ‘‘Grammar of the Language of Ulawa, Solomon Islands.’’ Journal of the Polynesian Society 22: 28–35, 96–103, 134–40, 219–24; 23: 21–7. ——. 1933. ‘‘A Grammar of the Language of Bugotu, Ysabel Island, Solomon Islands.’’ Bulletin of the School of Oriental and African Studies 7: 141- 177. Johnston, Raymond L. 1980. Nakanai of New Britain: The grammar of an Oceanic language. Canberra: Australian National University [Paciﬁc Linguistics B–70]. Keesing, Roger M. 1985. Kwaio Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–88]. Kolia, J. A. 1975. ‘‘A Balawaia Grammar Sketch and Vocabulary.’’ In Tom E. Dutton (ed.), Studies in Languages of Central and South-East Papua. Canberra: Australian National University [Paciﬁc Linguistics C–29], 107–226. Lee, Kee-dong. 1975. Kusaiean Reference Grammar. Honolulu: University Press of Hawaii. Lichtenberk, Frantisek. 1978. ‘‘A sketch of Houailou Grammar.’’ University of Hawai’i Working Papers in Linguistics 10.2: 74–116. ——. 1983a. ‘‘Relational Classiﬁers.’’ Lingua 60: 147–76. ——. 1983b. A Grammar of Manam. Honolulu: University of Hawaii Press. ——. 1984. ‘‘To’aba’ita Language of Malaita, Solomon Islands.’’ Working Papers in Anthropology, Archaeology, Linguistics, and Maori Studies, No. 65. Department of Anthropology, University of Auckland. ——. 1985. ‘‘Possessive Constructions in Oceanic Languages and in Proto-Oceanic.’’ In Andrew Pawley and Lois Carrington (eds), Austronesian Linguistics at the 15th Paciﬁc Science Congress. Canberra: Australian National University [Paciﬁc Linguistics C–88], 93–140.

274

jae jung song

Lichtenberk, Frantisek. 1986. ‘‘Syntactic-Category Change in Oceanic Languages.’’ Oceanic Linguistics 24: 1–84. Lincoln, Peter C. 1976. Describing Banoni, an Austronesian Language of Southwest Bougainville. Ph.D. dissertation, University of Hawai’i. ——. 1978. ‘‘Reefs-Santa Cruz as Austronesian.’’ In Stephen A. Wurm and Lois Carrington (eds), Second International Conference on Austronesian Linguistics: Proceedings. Canberra: Australian National University [Paciﬁc Linguistics C–61], 929–67. Lynch, John. 1973. ‘‘Verbal Aspects of Possession in Melanesian Languages.’’ Oceanic Linguistics 12: 69–102. ——. 1978. A grammar of Lenakel. Canberra: Australian National University [Paciﬁc Linguistics B–55]. ——. 1982. ‘‘Towards a Theory of the Origin of the Oceanic Possessive Constructions.’’ In Halim et al. (eds), 243–68. ——. In press. ‘‘Proto Oceanic Possessive Marking.’’ In John Lynch and Fa’afo Pat (eds), Oceanic Studies: Proceedings of the First International Conference on Oceanic Linguistics. Canberra: Australian National University. ——and Darrell T. Tryon. 1985. ‘‘Central-Eastern Oceanic: A subgrouping hypothesis.’’ In Andrew Pawley and Lois Carrington (eds), Austronesian Linguistics at the 15th Paciﬁc Science Congress. Canberra: Australian National University [Paciﬁc Linguistics C–88], 31–52. Marck, Jeﬀrey C. 1991. ‘‘Sixteen Nuclear Micronesian Verbs.’’ In Robert Blust (ed.), Currents in Paciﬁc Linguistics: Papers on Austronesian languages and ethnolinguistics in honour of George W. Grace. Canberra: Australian National University [Paciﬁc Linguistics C–117], 223–39. Mosel, Ulrike. 1984. Tolai Syntax and its Historical Development. Canberra: Australian National University [Paciﬁc Linguistics B–92]. ——and Even Hovdhaugen. 1992. Samoan Reference Grammar. Oslo: Scandinavian University Press. Pagotto, Louise. 1987. ‘‘Verb Subcategorization and Verb Derivation in Marshallese: A localistic lexicase analysis.’’ Ph.D. dissertation, University of Hawai’i. Pawley, Andrew. 1972. ‘‘On the Internal Relationships of Eastern Oceanic Languages.’’ In R. C. Green and M. Kelly (eds), Studies in Oceanic culture history. Honolulu: Bernice P. Bishop Museum, 1–142. ——. 1973. ‘‘Some problems in Proto-Oceanic grammar.’’ Oceanic Linguistics 12: 103–88. ——. 1979. ‘‘New Evidence on the Position of Rotuman.’’ Working Papers in Anthropology, Archaeology, Linguistics, and Maori Studies, No. 56. Department of Anthropology, University of Auckland. —— and Lois Carrington (eds). 1985. Austronesian Linguistics at the 15th Paciﬁc Science Congress. Canberra: Australian National University [Paciﬁc Linguistics C–88]. —— and Lawrence A. Reid. 1979. ‘‘The Evolution of Transitive Constructions in Austronesian.’’ In Paz B. Naylor (ed.), Austronesian studies: Papers from the Second Eastern Conference on Austronesian Languages. Ann Arbor: Center for South and Southeast Asian Studies, 103–30.

benefactive marking in oceanic languages

275

—— and Malcolm Ross. 1995. ‘‘The Prehistory of the Oceanic languages: A current view.’’ In Peter Bellwood, James J. Fox and Darrell Tryon (eds), The Austronesians. Canberra: Department of Anthropology, Australian National University, 39–74. Reid, Lawrence A. 1978. ‘‘Problems in the Reconstruction of Proto-Philippine Construction Markers.’’ In S. A. Wurm and Lois Carrington (eds), Second International Conference on Austronesian Linguistics: Proceedings. Canberra: Australian National University [Paciﬁc Linguistics C–61], 33–66. ——. 1981. ‘‘Proto-Austronesian Genitive Determiners.’’ In Andrew Gonzalez and David Thomas (eds), Linguistics across continents: Studies in honor of Richard Pittman. Manila: Linguistic Society of the Philippines, 97–105. Ross, M. D. 1988. Proto Oceanic and the Austronesian Languages of Western Melanesia. Canberra: Australian National University [Paciﬁc Linguistics C–98]. Schütz, Albert J. 1969. Nguna Grammar. Honolulu: University of Hawaii Press. Seiter, William. 1980. Studies in Niuean Syntax. New York: Garland. Siegel, Jeﬀ. 1984. ‘‘Introduction to the Labu Language.’’ Papers in New Guinea Linguistics 23: 83–157. [Paciﬁc Linguistics A-69]. Song, Jae Jung. 1997. ‘‘The History of Micronesian Possessive Classiﬁers and Benefactive Marking in Oceanic Languages.’’ Oceanic Linguistics 36: 29- 64. Sugita, Hiroshi. 1973. ‘‘Semitransitive Verbs and Object Incorporation in Micronesian Languages.’’ Oceanic Linguistics 12: 393–406. Todd, Evelyn M. 1978. ‘‘A Sketch of Nissan (Nehan Grammar).’’ In Wurm and Carrington (eds), 1181–239. Tryon, D. T. 1967. Nengone Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–6]. ——. 1968a. Dehu Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–7]. ——. 1968b. Iai Grammar. Canberra: Australian National University [Paciﬁc Linguistics B–8]. ——. 1994. ‘‘Language Contact and Contact-Induced Language Change in the Eastern Outer Islands, Solomon Islands.’’ In Tom Dutton and Darrell T. Tryon (eds), Language contact and change in the Austronesian world. Berlin: Mouton de Gruyter, 611–48. Wilson, William H. 1976. ‘‘The O/A Distinction in Hawaiian Possessives.’’ Oceanic Linguistics 15: 39–50. Wivel, Richard. 1981. ‘‘Kairiru Grammar.’’ M.A. thesis, University of Auckland. Wurm, Stephen A. 1969. ‘‘The Linguistic Situation in the Reef and Santa Cruz Islands.’’ Paciﬁc Linguistics (A) 21: 47–105. ——. 1978. ‘‘Reefs-Santa Cruz: Austronesian, but . . . !’’ In Wurm and Carrington (eds), 969–1010. ——and Lois Carrington (eds). 1978. Second International Conference on Austronesian Linguistics:Proceedings.Canberra:AustralianNationalUniversity[PaciﬁcLinguisticsC–61]. Zewen, François X.N. 1977. The Marshallese Language: A study of its phonology, morphology and syntax. Berlin: Dietrich Reimer.

Ergativity, transitivity, and clitic coreference in four Western Austronesian languages1 STANLEY STAROSTA University of Hawaii

1. Ergativity and theory in Western Austronesian In her University of Hawaii doctoral dissertation, Syntactic derivation of Tagalog verbs (DeGuzman 1976), Videa DeGuzman proposed that a subset of Tagalog verbs should be analyzed as ergative in their case-marking properties. Subsequent work on Philippine languages by linguists such as Gary Byma (Byma 1986), Stanley Starosta (Starosta 1986), William O’Grady (O’Grady 1987), Donna Gerdts (Gerdts 1988), and DeGuzman (DeGuzman 1988),2 on Proto-Austronesian by Stanley Starosta, Andrew Pawley and Lawrence Reid (Starosta, Pawley and Reid MS (1981), and on Formosan languages by Starosta (Starosta 1988), Arlene Ho (1990, 1993), and Lillian Huang (Huang 1994) have contributed to a rather remarkable increasing cross-theoretical consensus that Philippine-type ‘‘focus’’ languages, formerly regarded under the old ‘‘focus’’ analysis as something quite special, are actually just morphologically elaborate ergative languages.3 One of the contributions of the honoree of this volume (Blake 1988) was to show that they are in fact rather close syntactic and morphological analogues of the same kind of ergative grammar that can be found in Australian languages such as Kalkatungu. An ergative analysis of Philippine and Formosan languages and Proto-Austronesian makes possible some neat and insightful accounts of facts that can only be stipulated under the old focus analysis. It is the purpose of this paper to look at one set of such facts, the patterns of clause transitivity and clitic pronoun agreement in three Formosan languages, Atayal (ATA), Tsou (TSO), and Yami (YAM)4, and then extend the same analysis to Acehnese (ACH). It is hoped that the results will be interesting for two reasons: 1. the analyses are formulated within the Lexicase framework, a version of depen-

278

stanley starosta

dency grammar whose case relations are deﬁned in terms of grammatical patterns rather than situations. Some of the assumptions made in this approach to grammatical relations are diﬀerent from assumptions made in other frameworks, and to the extent that these assumptions make it possible to capture more and better generalizations than other approaches, they constitue a conﬁrmation of the superiority of a Lexicase analysis; and 2. Lexicase principles allow for only two kinds of case marking systems:5 ergative and accusative. To the extent that the analysis succeeds in capturing generalizations about transitivity and clitic agreement in Acehnese, it constitutes a refutation of Mark Durie’s contention (see especially Durie 1987), stated within a Role and Reference grammar framework, that Acehnese is neither ergative nor accusative, but rather belongs to a third type of case marking which is characteristic of what he refers to as ‘‘Active languages.’’ If this ‘‘active language’’ can better be analyzed as ergative, then maybe other ‘‘active languages’’ should be analyzed this way too.

2. Characterization of ergativity 2.1 Conventional characterization of ergativity A familiar characterization of the distinction between ergative and accusative languages goes back to R.M.W. Dixon’s SAO analysis (Dixon 1972; cf. Dixon 1994: 8–9). Dixon sets up three pre-theoretical situational roles, S (the sole obligatory participant (‘‘subject’’) in an intransitive clause), A (the instigator (loosely deﬁned) of the action in a transitive clause), and O (roughly, the patient or undergoer of the action described in a transitive clause). Ergative languages can then be distinguished from accusative ones in terms of whether the S role groups grammatically with O, as in ergative case-marking systems such as Dyirbal (indicated by the loop around S and O; see Figure 1), or with A, as is familiar from accusative languages such as English (indicated by the loop around S and A): 2.2 Ergativity assuming Patient centrality Three independently motivated assumptions of lexicase syntax make this dichotomy somewhat more transparent: 1. Patient centrality: every verb has one common case relation in its case frame, or ‘‘theta grid’’ in Chomskyan terminology. This case relation is referred to in lexicase as Patient (pat);

ergativity, transitivity, and clitic coreference

Ergative languages

Intransitive clauses

Accusative languages

S

S

absolutive

Transitive clauses

O

279

nominative

A

O

ergative

accusative

A

Figure 1. 2. one of the actants in a verb’s case frame is assigned the ‘‘actor’’ (actr) macrorole, a term and concept borrowed partially from Role and Reference grammar; and 3. every transitive verb has an Agent (agt) in its case frame, and conversely, every verb with an Agent in its case frame is transitive. The actor is marked on the same actant as the Patient of an intransitive clause or the Agent of a transitive clause. Terms like agent, patient, and actor are reminiscent of ‘‘case relations’’ or ‘‘-roles’’ in Fillmorean and Chomskyan theories and their derivatives, but there is an important diﬀerence: other theories deﬁne these relations situationally, but the lexicase case relations are deﬁned grammatically: they are drawn from a very limited inventory (pat, agt, loc, mns, cor and actr), and are assigned in a way that facilitates the capture of language-internal and cross-linguistic generalizations. That is, the correct assignment of case relations onto grammar is an empirical question rather than a subjective and arbitrary one. The Lexicase analysis of ergative versus accusative case marking is an example of an area where a suitable case analysis proves its worth. Taking the Nominative case form to be a kind of display window in which a language places its favorite case relation, an ergative language is one which displays the pat of each verb in its Nominative showcase, and an accusative language is one which displays the actr; see Figure 2. By Lexicase assumptions, then, every verb has a pat and an actr. With intransi-

stanley starosta

280

Ergative languages

Intransitive clauses

Accusative languages

Nom PAT actr

Nom PAT actr

(‘absolutive’)

Transitive clauses

nominative

Nom PAT

Erg AGT

Acc PAT actr

ergative

accusative

Nom ↔ PAT

Nom AGT actr

Nom ↔ actr

Figure 2. tive verbs, the sole obligatory argument is both pat and actr, but with transitive clauses, these two relations diverge: actr goes with the transitive Agent, and pat is the other pole of the transitive axis. As can be seen from the diagram, the formal correlate of both the ‘‘absolutive’’ and the ‘‘nominative’’ loops in Figure 2 is the Nominative (Nom) case form, and the diﬀerence between ergative and accusative languages can accordingly be stated very concisely: in an ergative language, the Nominative-marked actant (the ‘‘grammatical subject’’) is always the pat , while in an accusative language, the Nominative-marked actant is always the actr. The elimination of the absolutive-nominative distinction also makes possible the capture of a number of other cross-linguistic generalizations: Nominative is morphologically and syntactically the least marked case form, the one that is always relativizable, the one that takes precedence in word order, and one of the two categories that agrees with the verb if only one actant does. (The other is the actor.) The ‘‘ergative’’ case form in this schema is commonly, though not always, identical to a case form used in other functions in the same language.6 Thus the ‘‘ergative’’ in Avar, Dyirbal and Tibetan is completely identical in form to the instrumental case form, and the ‘‘ergative’’ in Philippine and Formosan languages is completely identical to the Genitive case form. Linguists working on ergative languages seem to like to give distinctive labels to these forms when they are

ergativity, transitivity, and clitic coreference

281

used in diﬀerent functions (see Blake 1994 for Avar7, Maclachlan 1996 for Tagalog8 and Dixon 1994 for Dyirbal9), thereby losing several important languageinternal and cross-linguistic grammatical generalizations. By contrast, the convention in the description of European case-marking languages is to give the same name to the same form, and this is the practice I will follow here. Thus I will refer to the ‘‘ergative’’ case in Austronesian languages considered in this paper as ‘‘Genitive.’’ In this approach, the nominative-marked agt or pat may be referred to for convenience as the ‘‘(grammatical) subject’’, but it should be remembered that in the Lexicase analysis, ‘‘subject’’ is not a primitive, and plays no special role in grammatical generalizations. It should also be noted that this is quite diﬀerent from the notion of ‘‘subject’’ in Chomskyan discussions of ergativity, where ‘‘subject’’ is identiﬁed in practice as whatever constituent translates the English grammatical subject. The language-speciﬁc and universal generalizations lost by this alternative deﬁnition will be obvious in any objective attempt to seriously apply the two analyses to the same set of data.

3. Case marking assuming Patient centrality 3.1 General pattern This pattern of case marking depicted in Figure 2 can alternatively be represented in terms of schematic clause structures as shown in Figure 3. Vt indicates a grammatically transitive verb, and Vi a grammatically intransitive verb. Figure 3 Ergative pattern: (Tagalog; Dyirbal (nouns10)) V S (Oblique) i, Nom (CFi) Pat (CRj) actr

Accusative pattern: (Chinese, English; Dyirbal (pronouns)) Vi, S (Oblique) Nom (CFi) Pat (CRj) actr

Vt,

Vt, S Nom Agt actr

O Nom Pat

A Gen Agt actr

(Oblique) (CFi) (CRj)

O Acc Pat

(Oblique) (CFi) (CRj)

Notice that a consequence of the Lexicase ergative analysis is that S, the sole argu-

stanley starosta

282

ment of an intransitive clause, is the ‘‘grammatical subject’’ in both systems, but for diﬀerent reasons: it is Nominative in ergative languages because it is the Patient, and it is Nominative in accusative languages because it is the actor. In transitive clauses, this distinction results in the divergent case marking that characterizes the ergative-accusative dichotomy. An important point to note here is that an intransitive clause is not necessarily a one-argument clause. It can have as many complements and adjuncts as it likes and still count as intransitive, as long as none of them is an Agent. Thus in Philippine-type languages, constructions which have been called ‘‘Actor Focus’’ (AF) in the old ‘‘focus’’ analysis are grammatically intransitive in an ergative analysis. Herewith some examples from English, a Western Indo-European creole of the Britannic Archipelago, and Atayal, an aboriginal Austronesian language of northern Taiwan.11 3.2 Examples Figure 4. One-component intransitive sentences

English A1. I ﬁred1. Nom-trns pat actr

Atayal E1. pima sakuʔ wash I -trns Nom pat actr ‘I am going to wash’

One-component intransitive sentences A2. E2.12 I ﬁred2 at them Nom-trns Lcv pat loc

ppima sakuʔ sunan wash I to you -trns Nom Lcv pat loc actr ‘I am going to wash you (partially).’

ergativity, transitivity, and clitic coreference Transitive sentenes A3. I ﬁred3 her Nom+trns Acc agt pat actr

283

E3. pman sakuʔ nyaʔ wash I by him +trns Nom Gen pat agt actr ‘He is going to bathe me.’

In older terminology, E2 is ‘‘AF’’ and E3 is ‘‘OF’’ or ‘‘GF’’ (‘‘Object Focus’’ or ‘‘Goal Focus’’). In the ergative analysis, note that this means that E2, despite the fact that is has two arguments and translates into an English transitive, is in fact grammatically intransitive. Initial indications of the validity of this analysis appear already in just these three examples: E2 groups with E1 morphologically (both lack the -an suﬃx), syntactically (both mark their actors as Nominative), and semantically (E3 is semantically more transitive than E1 or E2 by the Hopper-Thompson semantic transitivity metric). The claim that E2-type constructions, which I will refer to henceforth as ‘‘pseudotransitives’’, are grammatically intransitive, is consistent with the analysis I will propose of clitic coreference: the two analyses are mutually supportive, and together constitute evidence for the ergative-transitive analysis of Formosan languages as well as Acehnese and for the validity of the Lexicase analyses of these phenomena. In the remainder of this paper, I will apply this analysis to clitic pronoun constructions in all four languages, contrast it with the ‘‘active’’ analysis of Acehnese, and draw conclusions for linguistic theory.

4. Clitic pronoun constructions 4.1 General pattern Most or all of the Formosan aboriginal languages, including Atayal, have clitic pronouns (cf. Huang et al. 1996), and these forms are quite similar in their syntactic properties to for example clitic pronouns in Romance languages. The term ‘‘clitic pronoun’’ in this paper refers to pronouns which are phonologically bound but syntactically free. They can only be dependents of the root predicate in their clauses, but may be coreferential with another overt or covert complement of a root or nonroot predicate in that same clause. If the coreferring actant is a complement of a predicate which is a dependent of the host root predicate, the phenomenon is sometimes referred to in a transformational analysis as ‘‘clitic climbing.’’

stanley starosta

284

(i) ‘actr-to-actr’

(ii) ‘Nom-to-Nom’

verb +root

verb +root prnni actr +cltc

nounj actr –cltc

prnni Nom +cltc

nounj Nom –cltc

actri coref actrj Nomi coref Nomj Figure 5. Coreference patterns, single-verb clause The ﬁrst part of this paper will be concerned with presenting and illustrating a two-part well-formedness constraint on anaphoric relations in clitic constructions in the four languages we are concerned with. In brief, given the Lexicase deﬁnitions of grammatical relations, we can state that 1) if a sentence contains a clitic pronoun, and if that pronoun bears the actor macrorole, it must corefer to an (overt or covert) non-clitic argument which also bears the actor macrorole (‘‘actor-to-actor’’), while if it bears the Nominative case form, it must corefer with a non-clitic Nominative argument (‘‘Nom-to-Nom’’).13 This is represented schematically in Figure 5. Formosan languages are wonderful examplars of Háj Ross’s ‘‘Auxiliaries as main verbs’’ analysis (Ross 1969): if an auxiliary verb appears in a clause, it takes precedence over non-auxiliaries as the root of the clause and the host of clitic verbs. In this case the exact same coreference conditions apply, but the non-clitic arguments to which the clitics corefer appear (overtly or covertly) in the non-root clause (cf. Figure 6). As we will see below, this analysis requires auxiliary verbs to be marked for transitivity, and auxiliary verbs to agree in transitivity with the non-root nonauxiliary verbs they cap-command. I will explain below how and why this is the case, and give some additional evidence for it from Tsou. Of course, not all languages are the same, but in the case of the four languages under consideration here, at least, some of the diﬀerences can be stated quite simply in terms of independently motivated well-formedness constraints, as shown in Figure 7. To paraphrase: 1. The Non-root clitic ﬁlter states that in the three Formosan languages, clitic pronouns can only occur with root verbs. This condition is the motivation for what is sometimes referred to as ‘‘pronoun attraction’’ or ‘‘clitic climbing’’ in transformational analyses. 2. The Tsou Aux clitic ﬁlter states that clitic pronouns can only appear with auxiliary pronouns. That is, they never appear in the pattern illustrated in Figure 5.

ergativity, transitivity, and clitic coreference

(i) ‘actr-to-actr’

(ii) ‘Nom-to-Nom’

verb +root +xlry

verb +root +xlry

prnni actr +cltc

verb actr ±xlry nounj actr –cltc

prnni Nom +cltc

285

verb –root ±xlry nounj Nom –cltc

actri coref actrj Nomi coref Nomj Figure 6. Coreference patterns, clauses with an auxiliary verb

Non-root clitic lter (ATA, TSO, YAM) * –root

Aux clitic lter (TSO only)

Aux clitic lter (ACH only)

* –xlry

AGT-PAT lter (universal) *[AGTi=PATi]

+xlry

+cltc +cltc +cltc Figure 7. Well-formedness constraints (‘ﬁlters’)

3. The Acehnese Aux clitic ﬁlter is in a sense the converse of the Tsou Aux clitic ﬁlter. I have so far found no evidence that Acehnese auxiliary pronouns ‘‘attract pronouns’’, and this ﬁlter makes that an empirical hypothesis: Acehnese verbs do not occur in the pattern illustrated in Figure 6. 4. The agt-pat ﬁlter is a kind of consistency condition that that states that entities encoded as agt and pat are not perceived as coreferential.14 It functions as a boundary condition on clitic - non-clitic coreference that guarantees that, say, a pat in the auxiliary verb clause will not be chained to an agt in the non-auxiliary clause in the pattern illustrated in Figure 6. A consequence of this condition is that a verb and its regent auxiliary verb must agree in transitivity, since otherwise they would run afoul of this ﬁlter. This is not a theory-internal artifact, as shown by the morphological marking of auxiliary verbs in Tsou, for example. All of the well-formedness constraints stated above apply without special stipulations to transitive and intransitive clauses.

stanley starosta

286

Figure 8a. Intransitive constructions, no auxiliary verb

Figure 8b. Intransitive constructions with an auxiliary verb

verb +root –xlry –trns

verb +root +xlry –trns

prnni (noun) +cltc Nom PAT actr

nouni –cltc Nom PAT actr

prnni +cltc Nom PAT actr

verb –root ±xlry (noun) –trns

nouni –cltc Nom PAT actr

The schematic ﬁgures below illustrate these points. (Linear order in these schematic diagrams is not intended to be signiﬁcant, though all four languages are strongly verb-initial in basic clause structure.) Lexicase is a monostratal theory, so clitic pronouns are generated in situ as direct dependents of the root verb, not moved into their positions by transformation, and they must obey the same case-marking rules as other NPs. Thus we can tell for example that the clitic actr must be Nominative and Patient if the regent root verb is intransitive (cf. Figure 8a, b), but Genitive and Agent if the regent root verb is transitive (cf. Figures 9a,b, 10a, b). In either event, the coreference rules apply in exactly the same way. The four languages considered here divide into two pairs in terms of how many clitic pronouns they allow in grammatically transitive clauses: Atayal and Acehnese allow two (Figures 9b, 10b), but Tsou and Yami allow only one (Figures 9a, 10a). For all the languages, regardless of transitivity, one of the clitics may be an actr, and if there is only one clitic, it can only be an actr.15 In an intransitive clause, actr universally matches pat, and in a transitive clause, actr matches agt. In an ergative language, then, the obligatory actr clitic will be Nom in an intransitive sentence (Figures 9a, 10b) and Gen in a transitive one (Figures 9a,b, 10a,b). The additional clitic in a transitive clause in Atayal and Acehnese will then be a Nominative Patient (Figures 9b, 10b)16 The schematic examples in Figures 8–10 are followed by examples of each of these patterns in the four languages. In a formal Lexicase analysis, the rules which coindex clitic and non-clitic actants apply completely within the lexical matrix of the main verb, copying an index associated with one contextual feature onto another contextual feature (cf. Lee 1989, Savetamalya 1989). To make the notation a bit easier to grasp in the examples here, I have externalized unlinked contextual fea-

ergativity, transitivity, and clitic coreference

Figure 9a. Transitive constructions, no auxiliary verb, one clitic (Yami)

verb +root –xlry +trns

prnni (noun) +cltc Gen AGT actr

noun –cltc Nom PAT

nouni –cltc Gen AGT actr

Figure 9b. Transitive constructions, no auxiliary verb, two clitics (Yami, Atayal)

verb +root –xlry +trns

prnni +cltc Nom PAT

prnn (noun) +cltc Gen AGT actr

nouni –cltc Gen AGT actr

noun –cltc Nom PAT

Figure 10a. Transitive constructions with an auxiliary verb, one clitic (Tsou, Yami)

verb +root +xlry +trns

prnni +cltc Gen AGT actr

verb –root ±xlry (noun) +trns

nouni –cltc Gen AGT actr

noun –cltc Nom PAT

287

stanley starosta

288

Figure 10b. Transitive constructions with an auxiliary verb, two clitics (Atayal)

verb +root +xlry +trns

prnni +cltc Nom PAT

prnnj +cltc Gen AGT actr

verb –root ±xlry (noun) +trns

nounj –cltc Gen AGT actr

nouni –cltc Nom PAT

tures as ∆s. This is just a courtesy to the theoretically disinclined reader, and should not be taken to mean that Lexicase allows empty constituents, which it doesn’t.17 4.2 Atayal According to Søren Egerod (Egerod 1966: 251), Atayal clitic pronouns18 fall into two sets, Nominative and Genitive, but in his next paper he adds tertiary (locative). Free pronouns are called nominalized (1966: 347–8). Huang’s earlier analysis is similar in setting up these four classes, though in Huang et al. (1996) they are instead considered free accusative-locative pronouns. I believe that Egerod (1966) and Huang (1989) are correct about the bound status of locative pronouns. The labels Nom, Gen, and Lcv will be used for the three sets in this paper. 4.2.1 Intransitive (1) Intransitive, without an auxiliary verb (8a) pima sakuʔ ∆ wash I –trns +cltc –cltc +Nom Nom actr actr pat pat ‘I am going to wash.’

ergativity, transitivity, and clitic coreference

289

(2) Intransitive, with an auxiliary verb (8b) musa going +root +xlry –trns

sakuʔ [mima I wash +cltc –root +Nom –trns actr pat ‘I am going to wash.’

∆

]

–cltc Nom actr pat

4.2.2 Pseudotransitive The important thing to notice about these ‘‘pseudotransitive’’ examples is that although they are translated as transitive English sentences, their clitic patterns are intransitive in that they do not distinguish actor from Patient, and in that the notional ‘‘object’’ is marked with an oblique case form. Since pseudotransitives never take an Agent argument, no genitive pronouns may ever appear in these constructions. It is not obvious how a theory which considers these constructions to be transitive would account for this fact. (3) Pseudotransitives without an auxiliary verb (8a) ppima sakuʔ sunan ∆ wash I to you –trns +cltc +cltc –cltc +Nom Lcv Nom pat loc pat actr actr ‘I am going to wash (to) you. (4) Pseudotransitives with an auxiliary verb (8b) sakʔ [mima tε sunan ∆] I wash to you +cltc –root Lcv –cltc +Nom –trns Loc Nom pat pat actr actr ‘I am going to wash (to) you.’

musa going +root +xlry –trns

4.2.3 Transitive Like pseudotransitive constructions, true transitive clauses have two (or more) NP complements. Moreover, as transitives, they distinguish actor from Patient, and in

stanley starosta

290

Atayal and Acehnese, each can be separately encoded as a Genitive or Nominative clitic pronoun respectively. (5) Transitive clauses without an auxiliary verb (9b) pman sakuʔ nyaʔ ∆ wash I by him +trns +cltc +cltc -cltc +Nom Gen Gen pat agt agt actr actr ‘He is going to wash me.’

∆ -cltc Nom pat

(6) Transitive clauses with an auxiliary verb (10b) nyuxu is +root +xlry +trns

sakuʔ nyaʔ I by him +cltc +cltc +Nom Gen pat agt actr ‘He is washing me.’

[pman ∆ wash -root -cltc +trns Nom pat

∆

]

-cltc Gen agt actr

It is in the nature of pronouns that they are suﬃcient unto themselves in a given discourse, so it should not be surprising that so far there have been no examples of an overt non-clitic matching one of the clitics. However, we will have lots of examples of third person clitics matching overt non-clitic arguments, and occasionally there are examples of matching non-third clitic - non-clitic pairs, such as the following: (7) [Egerod 1966: 358] kialun taʔ niaʔ speak we by him +root +cltc +cltc +trns +Nom Gen pat agt actr ‘He speaks to us.’

∆

itaʔ we –cltc –cltc Gen Nom agt pat actr

Here ita/ is the non-clitic counterpart of ta/, appearing redundantly presumably for emphasis. As expected, Nominative marks the grammatical subject in transitive as well as intransitive clauses.

ergativity, transitivity, and clitic coreference

291

4.3 Tsou 4.3.1 Intransitives (8)19 Intransitives without auxiliary verbs (a): impossible in Tsou Intransitives with auxiliary verbs (b): ʔo [sʔno I get angry +cltc –root +Nom –trns pat actr ‘I got angry.’ mi did +root +xlry –trns

aʔo ] I –cltc Nom pat actr

This example illustrates two points in addition to the general exempliﬁcation of actor-to-actor and Nom-to-Nom: (1) again we have an overt non-clitic pronoun aʔo coreferencing the actor clitic ʔo, and (2) the initial m- on the auxiliary mi is signiﬁcant. I have claimed above that auxiliary verbs, like other verbs, participate in the +transitive classiﬁcation, in order to explain the distribution of clitic pronouns. In Tsou, there is direct morphological evidence to support this analysis: there are two sets of Tsou non-future auxiliary verbs, a set with initial m- which co-occurs only with intransitive verbs (as deﬁned in the Lexicase analysis, of course) and a parallel set without the initial m- which occur only with transitives. Gratifyingly, the m- set also co-occur with ‘‘pseudotransitive’’ verbs, conﬁrming the claim that the latter are grammatically intransitive. 4.3.2 Pseudotransitives (9) Pseudotransitives without auxiliary verbs (8a): impossible in Tsou Pseudotransitives with an auxiliary verb (8b) mi did +root +xlry –trns

ta [sʔno aʔo he get angry me +cltc –root –cltc +Nom –trns Gen pat cor actr ‘The old man got angry at me.’

e

mameoi ] old man –cltc Nom pat actr

In this example, the clitic ta is coreferential with the grammatical subject, e mameoi.

stanley starosta

292

The Tsou nominative and genitive clitic pronouns are homophonous except in the third person, where Genitive si has no nominative counterpart. We know that ta is Nominative here because it cannot be replaced by Genitive si. The grammatical evidence thus supports the analysis of this construction as intransitive. 4.3.3 Transitives Transitives without an auxiliary verb (a): impossible in Tsou Transitives with an auxiliary verb (a) Note (1) that the auxiliary verbs in the transitive examples are drawn from the transitive set, without the antipassive m- preﬁx, and (2) that the clitic pronouns coreference the Genitive actant if any, not the Nominative one. This is exactly what we expect under the Lexicase ergative analysis. Note also the -a suﬃx on suʔnova and cobaka in examples (10) and (11), and contrast it with the absence of -a on sʔno in example (8) and (9). The -a is the Tsou simple transitive suﬃx, and has corresponding forms especially in subordinate clauses in related Philippine and Formosan languages. It is the functional counterpart of -Vn in main clauses in many of these same languages (cf. Starosta, Pawley, and Reid 1982: 154, 164). ʔo [sʔnova by me get angry at +cltc –root +Gen +trns agt actr ‘I got angry at this woman.’

(10) os did +root +xlry +trns

ʔo [eobaka by me hit +cltc -root +Gen +trns agt actr ‘I hit you.’

(11) os were +root +xlry +trns

∆

e this

–cltc Gen agt actr ∆ –cltc Gen agt actr

mamespii] woman –cltc Nom pat

suu ] you –cltc Nom pat

4.4 Yami Yami is spoken on Botel Tobago (Orchid Island) oﬀ the southeast coast of Taiwan. It is genetically a Philippine rather than a Formosan language, closely related to Ivatan, but the analysis proposed here and supported by evidence from Atayal and Tsou works equally well here.

ergativity, transitivity, and clitic coreference 4.4.1 Intransitives (12) Intransitive clause without an auxiliary verb (8a) du izara ∆ maey ku go I to Orchid Island –trns +cltc –cltc –cltc +root +Nom Lcv Nom pat loc pat actr actr ‘I am going to Orchid Island.’ (Ho 1990: 115) Note the intransitive maN- preﬁx on the verb. (13) Intransitive clause with an auxiliary verb (8b) ku [matava ∆ ] nu ku madey kapalalayu am ta if I every day run LIG not I fat –trns +cltc –trns –cltc +root +Nom –root Nom +xlry pat –xlry pat actr actr ‘If I jogged every day, I would not be fat.’ (I do not jog every day, and I am fat now.) (Ho 1990: 100) 4.4.2 Pseudotransitive (14) Pseudotransitive without an auxiliary verb (8a) manii namen su kadai ∆ sift we millet –trns +cltc –cltc –cltc +Nom Abl Nom pat mns pat actr actr ‘We will sift millet.’ (Ho 1990: 96) (15) Pseudotransitive with an auxiliary verb (8b) ya namen [manii su kadai are we sift millet +root +cltc –root –cltc +xlry +Nom –trns Abl pat mns actr ‘We are sifting millet.’ (Ho 1990: 91)

∆] –cltc Nom pat actr

293

stanley starosta

294

Again, note the intransitive m- preﬁx on both verbs. The Yami examples here are closely analogous to the Tsou examples above. Namen is drawn from the Nominative pronoun set, as expected. Arlene Ho (Ho 1990) analyzes examples such as (16) if they had an invisible initial auxiliary verb, and in fact such an analysis ﬁts perfectly into the analysis I propose here except for the fact that a Lexicase grammar has no empty categories. The only reasonably promising way to ﬁt such examples into a strict Lexicase analysis is to treat namen in (16) as a contraction of ya plus the nominative clitic pronoun namen. (16) Pseudotransitive with a phantom auxiliary verb (8b) ∆ are +root +xlry

namen [manigi su kadai ∆ ] we sift millet +cltc -root -cltc -cltc +Nom -trns Abl Nom pat mns pat actr actr ‘We are sifting millet.’ (Ho 1990: 92)

4.4.3 Transitive (17) Transitive without an auxiliary verb (9a) ∆ u anakna bakbakan na hit by him his child +trns +cltc -cltc Nom +Gen Gen pat agt agt actr actr ‘He will hit his child.’ (Ho 1990: 96) (18) nibakbakan na ni mapapu hit (perf) by her of Mapapu +trns +cltc -cltc +Gen Gen agt agt actr actr ‘Mapapu hit me.’ (Ho 1990: 117)

yaken I -cltc Nom pat

ergativity, transitivity, and clitic coreference

295

ku imu by me you +cltc -cltc Gen Nom agt pat actr ‘I will hit you.’ (Ho 1990: 117)

(19) bakbakan hit +trns

The third person singular Genitive clitic pronoun na and ﬁrst person singular Genitive clitic pronoun ku mark these sentence as transitive, as does the -an suﬃx on the verb. In contrast to the pseudotransitives above, the clitic pronoun is coreferential not with the Nominative but with the Genitive actant if any. As Yami is a one-clitic language, like Tsou, non-actor pronouns such as yaken ‘I’ and imu ‘you’ must appear as non-clitics. They have the same syntactic distribution as free nominative NPs such as u anakna in ), and do not participate in clitic coreference patterns. (20) Transitive with an auxiliary verb (10a) na [nibakbakan ni mapapu by her hit Mapapu +cltc –root –cltc +Gen +trns Gen agt agt actr actr ‘Mapapu has hit me.’ (Ho 1990: 72) ya is +root +xlry +trns

yaken ] I –cltc +Nom pat

4.5 Acehnese ‘‘Acehnese is an Austronesian language spoken by some two and a half million people in the Province of Aceh in northern Sumatra.’’ (Asyik 1982: 1) It has been regarded by Mark Durie (Durie 1985: 186–7) as belonging to the Active language type, neither ergative nor accusative. According to Durie, working within the Role and Reference framework, such languages lack any category corresponding to the usual S of intransitive clauses, and instead divide intransitive verbs into two groups, those with A(ctor) subjects and those with U(ndergoer) subjects, rather like the unaccusative-unergative split in Relational Grammar (cf. Figure 11). If this analysis is correct, it constitutes an embarrassment to the Lexicase framework, which claims that all intransitive subjects must be analyzed in the same way in all language types. What I propose to do in the remainder of this paper is to assume that in fact Acehnese is an ergative language rather than an ‘‘active’’ one. I will apply the same tests and analysis to Acehnese that I applied to Atayal, Tsou, and Yami. What I

stanley starosta

296

Ergative languages

Intransitive clauses

Accusative languages

S

S

absolutive

Transitive clauses

O

Active languages

A

U

A

U

nominative

A

O

A

ergative accusative Figure 11. Active case marking (adapted from Durie 1987: 365) hope to demonstrate is that in fact Acehnese ﬁts perfectly into an ergative analysis as formulated within Lexicase theory, with one or two surprises but with no undue forcing. The following data, taken from Asyik (Asyik 1982) and Durie (Durie 1987), parallel the examples from the other three languages. Acehnese, like Atayal, allows clitic pronouns to attach before or after non-auxiliary verbs. I will contend here that Acehnese clitic pronouns always follow intransitive verbs and precede transitive ones, and that in fact this is a completely reliable test for transitivity in Acehnese.20 To the extent that I can capture the same generalizations with this analysis which I captured with the other three languages, I will consider my hypothesis to have been conﬁrmed at both a language-speciﬁc and universal level. (21) Intransitive clauses (8a) opnyan rhët he/she fall -cltc -trns +them

euh ∆ he/she +cltc -cltc +Nom Nom pat pat actr actr ‘(S)he falls.’ (Durie 1987: 369)

ergativity, transitivity, and clitic coreference

297

(22) Pseudotransitive clauses (8a) h’an [alak euh ∆ keu u muda ] not like he/she to coconut young –trns +cltc –cltc –cltc +Nom Nom Gen pat pat mns actr actr’ ‘(S)he doesn’t like young coconut.’ (Durie 1987: 372) (21) is an example of one of Durie’s U-subject intransitive verbs. If the initial free pronoun gopnyan is analyzed as a ‘‘displaced’’ topic (+them ‘‘theme’’), this example ﬁts the pattern of Atayal intransitive verbs exactly. As with Atayal pseudotransitives, Acehnese pseudotransitives may be translated by English transitives but share grammatical properties with typical one-argument intransitives. As in Atayal, the notional ‘‘object’’ of such verbs appears in an oblique case form, either marked by an overt preposition, as in (22), or ‘‘bare’’, as in (23), where only the lack of agreement tells us that u muda is not the grammatical subject. euh ∆ u muda ] he/she coconut young +cltc -cltc -cltc +Nom Nom Gen pat pat mns actr actr actr’ ‘(S)he doesn’t like young coconut.’ (Durie 1987: 372)

(23) h’an [alak not like -trns

4.5.1 Transitive (9a) As in Atayal, Acehnese transitive verbs take both a Nominative Patient clitic pronoun, as in intransitives, and a Genitive Agent actor clitic pronoun. The Nom-pat clitic pronoun follows the verb, as in Atayal, and the Gen-agt-actr clitic pronoun precedes the verb, exactly as it does in the Yami ‘‘phantom auxiliary’’ construction, the kind of pattern from which it probably evolved. This is illustrated by examples (24) and (25). Examples such as (25) are the point where Durie’s analysis diﬀers most radically from mine. For him, (25) would be an intransitive sentence with an A(ctor) subject, with the pre-verbal position of the clitic pronoun traceable to the role of the subject. In the Lexicase analysis being proposed here, on the other hand, the verb jak ‘‘go’’ in (25), and similar verbs in similar constructions, are grammatically transitive but impersonal.21 A grammatically comparable English pattern would be ‘‘Father made it to the oﬃce’’, where it is a non-referential pat which would be the grammatical subject if English were an ergative language.

stanley starosta

298 (24) Personal (9)

kèe h’an [eu– pateéh kuh ∆ opnyan ] she not he/she believe me he/she +cltc +trns +cltc –cltc –cltc +Gen –mprs +Nom Nom Gen agt pat pat agt actr actr’ ‘(S)he doesn’t believe me.’ (Durie 1987: 370) (25) Impersonal (9a) ha [u jak ∆ ∆ u kanto (perf) he/she go to oﬃce +cltc +trns cltc –cltc Lcv +Gen +mprs Nom Nom loc agt pat pat actr +mbnt +mbnt ‘Father’s gone to the oﬃce.’ (Asyik 1982: 8)

Ayah ] father –cltc Gen agt actr

The innovation made here of course is in daring to think that a one-argument verb could be transitive. This is on ﬁrst blush implausible, but hopefully we have learned from the analysis of pseudotransitives that you can’t judge a verb by its English translation or by the number of overt arguments it carries around with it. Instead, such a proposal must be evaluated according to how many language-speciﬁc and cross-linguistic generalizations it allows us to capture. Here are some. The reader is invited to keep score.

5. Active versus ergative analysis 5.1 Contrasting the claims In the following section, I will try to summarize some of the ways that the Lexicase ergative analysis proposed in this paper is able to integrate and interrelate facts about semantic transitivity, morphology, and clitic pronoun occurrence and coreference in the four languages under consideration. In particular, I will show that if we assume, contrary to Durie’s claim, that Acehnese is an ergative language like the other three surveyed here, then the same general and language-speciﬁc generalizations apply without alteration. To the extent that this attempt is successful, it will constitute a disconﬁrmation of Durie’s claim, a conﬁrmation of the Lexicase

ergativity, transitivity, and clitic coreference

299

ergative-transitive language analysis, and a threat to ‘‘active’’ case marking analyses in general. 5.2 Transitives versus pseudotransitives Exhibit A draws on Hopper and Thompson’s discussion of transitivity (Hopper and Thompson 1980). They cite a number of properties of clauses which they label as semantically more or less transitive, and establish a scale of semantic transitivity something like the one in Figure 12.

transitive ‘objects’ tend to be de nite, totally affected; single; clauses tend to be perfective

‘objects’ tend to be inde nite, partially affected; multiple; clauses tend to be imperfective intransitive Figure 12. 5.2.1 Semantic transitivity Their paper was in my opinion one of the most insightful and useful linguistic papers written in recent times. It clearly demonstrates that while semantic transitivity and ‘‘morphosyntactic’’ (grammatical) transitivity are not the same thing, there is a signiﬁcant correlation between the two, and that if two closely similar constructions diﬀer in semantic transitivity, the one which is more transitive semantically is likely to be more transitive grammatically as well. 5.2.2 Atayal In the Lexicase framework, the issue is clear-cut: grammatical transitivity is polar, not scalar, and if two constructions diﬀer in semantic transitivity, the null hypothesis is that they also diﬀer in syntactic transitivity. Here are Atayal examples taken from a paper by Lillian Huang to illustrate this point:

stanley starosta

300 (28) Pseudotransitive (8a)

hira mnulu sakuʔ kupuʔ qani ∆ found I cup this yesterday –trns +cltc –cltc –cltc +Nom Gen Nom pat mns pat actr actr ‘I found this cup yesterday.’ ‘Relatively unintentional Agent (not on purpose; happened to ﬁnd it)’ (Huang 1989: 127) (27) is more transitive semantically than (26), in that it describes an action carried out intentionally and to a successful conclusion. (27) is also the construction which is independently selected as transitive by the Lexicase ergative analysis. (27) Transitive (9b) kupuʔ qani hira lwan makuʔ ∆ found by me cup this yesterday –trns +cltc –cltc –cltc +Gen Gen Nom agt agt pat actr actr ‘I found this cup yesterday’ ‘more involved AGENT. . . . I was looking for it and ﬁnally found it’ (Huang 1989: 127) 5.2.3 Tsou Similar evidence can be found in the other languages. For example, the following two Tsou sentences diﬀer in the deﬁniteness of their objects, and the semantic difference parallels the grammatical diﬀerence as represented in the Lexicase analysis of these examples: (28) Pseudotransitive (8b) mio did +root +xlry –trns

∆

[eoʔvako to fkoi o mameoi ] hit of snake old man +cltc –root –cltc –cltc +Nom –trns Gen Nom pat mns pat actr actr ‘The old man hit a snake.’

ergativity, transitivity, and clitic coreference

301

(29) Transitive te will +root +trns +xlry

to nʔa [eobaka we beat +cltc –root +Gen +trns agt actr Let’s go beat the snake.’

∆

o fkoi ] snake –cltc –cltc +Gen +Nom agt pat actr’

5.2.4 Yami Arlene Ho’s MA thesis provides a number of examples from Yami in which semantic transitivity diﬀerences parallel grammatical transitivity diﬀerences; e.g., example (30). In this example, the absence of a clitic pronoun after an auxiliary verb is treated as grammatically equivalent to a third person nominative clitic pronoun. This particular paradigm gap (third person Nominative) is ubiquitous though not universal in Formosan languages. (30) Pseudotransitive (8b) ya is +root +xlry –trns

∆

[kuman si mapapu su eat Mapapu +cltc –root –cltc +Nom +Nom –trns pat pat actr actr ‘Mapapu is eating taros.’ (Ho 1990: 74)

suli ] taro –cltc +Gen mns

(31) Transitive (10a) ya is +root +xlry +trns

na [nikan ni mapapu u suli by her eaten Mapapu taro +cltc –root –cltc –cltc +Gen +trns Gen Nom agt agt pat actr actr ‘Mapapu has eaten up the taros.’ (Ho 1990: 74)

5.2.5 Acehnese Interestingly, closely parallel data can be found in Acehnese, and they support the thesis of this part of the paper: Durie’s A-argument intransitive verbs are grammatically transitive, and ﬁt comfortably into a Lexicase ergative analysis:

stanley starosta

302 (32) Pseudotransitive (8a)

alak keuh nyan ∆ like you that –trns +cltc –cltc –cltc +Nom +Gen Nom pat mns pat actr actr ‘Please don’t like that.’ [weak] (Asyik 1982: 16)

bèk don’t

(33) Transitive (9b) bèk kaalak ∆ ∆ nyan don’t you like that +cltc +trns cltc –cltc –cltc Nom Gen +Nom +Gen agt pat agt pat actr actr ‘Don’t (make an effort to) like that.’ [stronger] (Asyik 1982: 16) (34) Pseudotransitive (8a) [banci lən keu mie nyan ∆ ] hate I at cat that -trns +cltc -cltc -cltc +Nom +Lcv Nom pat loc pat actr actr ‘I don’t hate the cat.’ [neutral statement] (Asyik 1982: 16–17)

hana not

(35) Transitive impersonal (9b) hana [lən banci ∆ ∆ keu mie nyan not I hate at cat that +cltc +trns cltc -cltc +Gen +mprs Nom Gen +Lcv agt pat agt loc actr -mbnt actr ‘I don’t (make an eﬀort to) hate the cat.’ [denying accusation] (Asyik 1982: 16) According to Asyik (1982), there is a diﬀerence in deliberateness and intentionality between the pairs of examples. (33) is a stronger command, and (33) is the one which, by the ergative analysis, is transitive. Similarly, (34) contrasts with (35) in

ergativity, transitivity, and clitic coreference

303

the degree of intentionality. We can say that (35) is semantically more transitive, and (35) is the one which would be analyzed as grammatically transitive by the ergative analysis.

6. Conclusion This is of course not the ﬁnal word on ergativity in Acehnese, but it is a rather suggestive word. The ergative analysis of transitivity which works nicely in accounting for some properties of three ergative Formosan languages seems to work just as well for supposedly ‘‘active’’ Acehnese. The examples of inﬁnitival complement constructions which Durie presented in support of his ‘‘Active’’ analysis ﬁt at least as nicely into the ergative analysis, though I will not be able to present that part of the analysis in the space available here. If it is true that all of Durie’s data turn out to be consistent with an ergative analysis, then Acehnese is not a counterexample to the Lexicase axiom that only two kinds of case-marking systems are possible. On the contrary, the theory has made a prediction, and the data when looked at in accordance with the predictions turn out to be consistent with it and in fact to provide further evidence for the theory. The prosecution rests.

Notes 1. An earlier version of this paper was presented at the Symposium on Grammatical Relations in Austronesian, Sixth International Conference on Austronesian Linguistics, Honolulu, May 1991. 2. Anna Maclachlan (Maclachlan 1996), working in a Chomskyan framework, comes to the conclusion that Tagalog is a hybrid of ergative and accusative case-marking types. 3. This is not to say that everyone has ﬁnally got the word. Thus the old focus analysis can still be found in work by Chomskyan linguists such as for example Guilfoyle, Hung, and Travis (1992), Mei (1994), Holmer (1996), and Chang (1996). 4. Yami is actually genetically a Philippine language, although administratively Orchid Island/Botel Tobago, where the Yami language is spoken, is part of Taiwan. 5. It also allows for both systems to be present in a single language, such as the two distinct systems of case-marking for nouns and pronouns in Dyirbal (Dixon 1972). However, I contend that most reported cases of ‘‘hybrid ergativity’’ (Maclachlan 1996) and ‘‘split ergativity’’ (cf. Harris 1981: 46, cited in Blake 1994: 125–26) are just mis-analyses resulting from failure to properly identify grammatical transitivity in the languages in question. 6. ‘‘It is common to ﬁnd that in ergative languages the so-called ergative marks a peripheral function such as genitive, locative or instrumental as well as A.’’ (Blake 1994: 122)

304

stanley starosta

17. ‘‘The case marker glosses as erg(ative) is the one that also indicates instruments.’’ (Blake 1994: 122) 18. ‘‘The two ngs, ngA and ngP, are considered here to be diﬀerent Cases structurally, erg and acc respectively, despite the fact that they are homophonous.’’ (Maclachlan 1996: 82) ‘‘It is not inconsistent to make a structural distinction between the abstract Case associated with ngA and ngP and yet recognize that one and the same case marker ng is used for these two strcuturally distinct cases.’’ (Maclachlan 1996: 83) 19. ‘‘Ergative and instrumental cases have identical realisation but rather diﬀerent syntactic behaviour: . . .’’ (Dixon 1994: 170). 10. The term ‘‘noun’’ when used to contrast with the term ‘‘pronoun’’ should be taken to refer speciﬁcally to non-pronominal nouns. 11. All Atayal examples in this paper are taken from Lillian Huang, p.c., unless otherwise indicated. The morphological and syntactic analyses of all the examples in this paper are my own. 12. Glosses for E2 and E3 adapted to ﬁt the analysis and observations in Huang (1994). 13. This same pattern has also been identiﬁed in Gujarati, an ergative Indoaryan language (Starosta, to appear), suggesting that it is associated with non-root clauses in ergative languages. Students in my ‘‘Formosan grammar’’ class in the Graduate Institute of Linguistics, National Taiwan University (Fall 1996) have also found some apparent exceptions, which may turn out to be examples of distinct inﬁnitival rather than non-root constructions. 14. This ﬁlter implies that transitive reﬂexive constructions are constructions in which a single entity in objective external reality is encoded as two perceptually distinct entities. 15. ‘‘The clitic system is unusual for Formosan languages in that the pronominal clitics invariably refer to the highest case in the accusative subject hierarchy, regardless of whether the sentence is active or passive.’’ (Starosta 1974: 347) 16. This last statement needs to be modiﬁed if Wulai Atayal has locative clitic pronouns. According to Huang (1989: 117, 128, 130), it does, but according to Huang et al. 1996, it doesn’t. I think Huang (1989) makes a more compelling case that it does. 17. This statement applies to well-formed sentences. However, in the treatment of fragments, a discourse-level phenomenon, Lexicase does assume the equivalent of empty categories; cf. Starosta (1988: 253–55), Sak-Humphry (1992), Springer and Starosta (in progress). 18. He doesn’t use the term ‘‘clitic’’, but it can be inferred from his examples and from the fact that he refers to his ‘‘nominalized’’ pronouns as occurring as ‘‘complements’’, implying that the others don’t. 19. Tsou examples are taken from Starosta’s ﬁeld notes. 20. A number of Formosan and Western Austronesian languages have a similar pattern, a consequence of ‘‘Aux axing’’ (Starosta, Pawley, and Reid MS (1981)), the loss of an initial auxiliary verb in

ergativity, transitivity, and clitic coreference

305

transitive clauses in one-clitic languages. The ‘‘phantom auxiliary’’ constructions in Yami can be seen as a transitional stage in this process. 21. Similar constructions occur for example in Amis (Taiwan; Starosta’s ﬁeld notes): nira namaka patiamay tara i natawran of him from Hualien to at Natawran +Gen agt actr ‘He walked from Hualien (‘busy place’) to Nataoran.’ rakatən walk +trns +mprs

Here the -ən suﬃx and the Genitive actor mark this sentence as grammatically transitive, even though the English gloss is intransitive.

References Asyik, Abdul Gani. 1982. ‘‘The Agreement System in Acehnese.’’ Mon-Khmer studies XI: 1–33. Blake, Barry J. 1988. ‘‘Tagalog and the Manila–Mt. Isa axis.’’ La Trobe Working papers in Linguistics 1: 77–90. Melbourne: Linguistics Department, LaTrobe University, 77–90. ——. 1994. Case. Cambridge: Cambridge University Press. Byma, Gary F. 1986. ‘‘Government and Binding in Tagalog: An ergative analysis.’’ M.A. thesis. Calgary: Department of linguistics, University of Calgary. Chang, Yung-li. 1996. ‘‘Agent/non-agent focus asymmetries in Seediq.’’ In Yuen-mei Yin, I-li Yang, Chien-ching Mo, and Hui-chen Chan (eds), Proceedings of the ﬁfth international symposium on Chinese languages and linguistics. Mucha: Graduate program in linguistics, National Chenchi University, 81-100. DeGuzman, Videa P. 1976. ‘‘Syntactic Derivation of Tagalog Verbs.’’ Ph. D. dissertation. Honolulu: University of Hawai’i. ——. 1988. ‘‘Ergative Analysis for Philippine Languages: An analysis.’’ In McGinn (ed.), 323–345. Dixon, R.M.W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. ——. 1994. Ergativity. Cambridge: Cambridge University Press. [Cambridge Studies in Linguistics 69.] Durie, Mark. 1985. ‘‘A Grammar of Acehnese on the basis of a Dialect of North Aceh.’’ Verhandelingen van het Koninklijk Instituut voor Taal-, Land- en Volkenkunde, 112. Dordrecht, Holland: Foris Publications. ——. 1987. ‘‘Grammatical Relations in Achenese.’’ Studies in Language 11:1–2, 365–399. ——. 1988. ‘‘Preferred Argument Structure in an Active Language.’’ Lingua 74: 1–25. Egerod, Søren. 1966. ‘‘Verb Inﬂexion in Atayal.’’ Lingua 15: 251–82. Gerdts, Donna B. 1988. ‘‘Antipassives and Causatives in Ilocano: Evidence for an ergative analysis.’’ In McGinn (ed), 295–321.

306

stanley starosta

Guilfoyle, Eithne, Herietta Hung and Lisa Travis. 1992. ‘‘Spec of IP and Spec of VP: two subjects in Austronesian languages.’’ Natural language and linguistic theory 10.3: 374– 414. Harris, Alice C. 1981. Georgian Syntax: A study in relational grammar. Cambridge: Cambridge University Press. Ho, Arlene Y.-L. 1990. ‘‘Yami Structure: A descriptive study of the Yami language (English version).’’ M.A. thesis, Hsinchu, Taiwan: Institute of Linguistics, National Tsing Hua University. ——. 1993. ‘‘Transitivity, Focus, Case and the Auxiliary Verb Systems in Yami.’’ Bulletin of the Institute of History and Philology 62.1, 83–147. Taipei: Academia Sinica. Holmer, Arthur J. 1996. A parametric grammar of Seediq. Lund: Lund University Press [Travaux de l’Institut de Linguistique de Lund 30]. Hopper, Paul J., and Sandra A. Thompson. 1980. ‘‘Transitivity in Grammar and Discourse.’’ Language 56: 251–99. Huang, Lillian M. 1989. The Pronominal System in Atayal. Studies in English Literature and Linguistics 15:115–33. Taipei: National Taiwan Normal University. ——. 1994. Ergativity in Atayal. Oceanic Linguistics 33:1, 129–143. ——. 1995. A study of Mayrinax syntax. Taipei: The Crane Publishing Company. ——. Elizabeth Zeitoun, Marie M. Yeh, Anna H. Chang, and Joy J. Wu. 1996. ‘‘A Typological Study of Pronominal Systems in Some Formosan Languages.’’ To appear in H. Samuel Wang (ed.), Proceedings of the Fifth International Conference on Chinese Linguistics (ICCL-5). Hsinchu, Taiwan: National Tsing Hua University. Lee, Nagiko Iwata. 1989. ‘‘Complementation in Japanese: A Lexicase analysis.’’ Ph.D. dissertation. Honolulu: University of Hawai’i. McGinn, Richard (ed.). 1988. Studies in Austronesian Linguistics. Athens, Ohio: Ohio University Press. Maclachlan, Anna E. 1996. ‘‘Aspects of Ergativity in Tagalog.’’ Ph.D. dissertation. Montreal, Canada: Department of Linguistics, McGill University. Mei, Kuang. 1994. ‘‘Word order, case markers and theta agreement in Mayrinax (in Chinese).’’ Paper presented to the Conference on Austronesian Languages in Taiwan, May 20–22, 1994, Academia Sinica, Taiwan. O’Grady, William. 1987. ‘‘The Nature of Philippine Ergativity.’’ M.A. thesis(?). Cambridge, Mass.: Harvard University. Ross, John Robert. 1969. ‘‘Auxiliaries as Main Verbs.’’ In W. Todd (ed.), Studies in Philosophical Linguistics (Series 1), 77–102. Carbondale, Ill: Great Expectations Press. Sak-Humphry, Chhany. 1992. ‘‘The syntax of nouns and noun phrases in dated Pre-Ankorian inscriptions.’’ MA thesis. Honolulu: Department of Linguistics, University of Hawai’i. Savetamalya, Saranya. 1989. ‘‘Thai Nouns and Noun Phrases: A Lexicase analysis.’’ Ph.D. dissertation. Honolulu: University of Hawai’i. Springer, Hisami, and Stanley Starosta. To appear. ‘‘A dependency analysis of coordination in English and Japanese.’’ In B. Caron (ed), Proceedings of the 16th International Congress of Linguists. Paris.

ergativity, transitivity, and clitic coreference

307

Starosta, Stanley. 1986. ‘‘Focus as Recentralisation.’’ In Paul Geraghty, Lois Carrington, and S.A. Wurm (eds), FOCAL I: Papers from the Fourth International Conference on Austronesian Linguistics, 73–95. [Paciﬁc Linguistics C-93]. ——. 1988. ‘‘The Case for Lexicase: An outline of Lexicase grammatical theory.’’ London: Pinter Publishers. [Open Linguistics Series.] ——, Andrew K. Pawley, and Lawrence A. Reid. MS (1981). ‘‘The Evolution of Focus in Austronesian.’’ Honolulu: Department of linguistics, University of Hawaii.

A discourse explanation for the cross-linguistic diﬀerences in the grammar of interrogation and negation* SANDRA A. THOMPSON Dept. of Linguistics, UC Santa Barbara

1. Introduction One of the major questions to which functional research has devoted itself in the last ﬁfteen years or so is the question of what makes grammars the way they are. What sorts of principles account for the massively recurrent grammatical patterns that we can notice across languages? A primary answer to these questions can be found in the interactional patterns that emerge from studies of large corpora, on the assumption that the organization of grammar is motivated by the way language is used. In fact, I would suggest that it is in the study of actual interactions that we can most readily ﬁnd the ‘‘functions’’ that are appealed to in ‘‘functional linguistics.’’ In this paper I will investigate the grammar of interrogation and negation across languages with the aim of showing how recurrent diﬀerences in the grammar of these two constructions can be explained by consistent and large-scale diﬀerences in the interactional functions associated with these two types of constructions. My decision to pursue the relationship between form and function in interrogatives and negatives was motivated by the striking diﬀerences in the grammar of these two construction types that emerge in the typological literature. In particular, the grammar of interrogation typically involves a number of strategies that are not found in the grammar of negation, and vice versa. Yet they have sometimes been discussed together in typological studies, as they may share certain grammatical features, and, to many linguists, especially those trained in the 60’s and 70’s, they appear to involve similar sorts of alterations, or ‘‘mutations,’’ from a ‘‘kernel’’ structure (Chomsky 1957).

310

sandra a. thompson

Before beginning the actual discussion, I wish to make one important assumption, and to restrict the investigation in three ways. Central to the discussion will be the assumption that a basic unit of spoken language is a prosodic unit (Altenberg 1987, Maynard 1989, Oreström 1983); i.e., people talk in ‘‘spurts’’ (Chafe 1987). Various proposals for prosodic units have been recognized in the literature on natural speech; for my purposes the ‘‘intonation unit’’ (Chafe 1987, 1988, 1993, 1994, Du Bois 1991, Du Bois et al. 1993, SchuetzeCoburn 1992, to appear, Schuetze-Coburn et al. 1991, Tao 1996) will serve as a prototype, roughly characterizable as ‘‘a stretch of speech uttered under a single coherent intonation contour’’ (Du Bois et al. 1993). There are issues regarding the size of prosodic units and their identiﬁcation in various languages. I readily acknowledge the need for much further research into prosodic units in spoken discourse from as many languages as possible; for this paper, I will focus on ‘‘intonation units’’ as deﬁned for English in the studies cited as a prototype of the type of prosodic unit that seems to be signiﬁcant for interaction, and use the term ‘‘prosodic unit’’ as a cover term. Recent studies of everyday interaction have shown that prosodic units have profound interactional consequences (Oreström 1983, Ford 1993, Ford and Thompson 1996, Maynard 1989, Schegloﬀ 1988, Schuetze-Coburn to appear, Tao 1996). Speciﬁcally, recent research (Ford and Thompson 1996) has shown that prosodic units play a critical role in the smooth exchange of conversational turns; points at which prosodic unit completions occur strongly tend to be points at which turn transition occurs or could occur. Ford and Thompson found that the majority of speaker changes (71 per cent) occur at points of prosodic completion, strongly suggesting that prosodic units play a major role in conversational exchanges. This ﬁnding will play a major role in my argument. The second assumption I will make is a terminological one: I will use the terms ‘‘interrogative’’ and ‘‘interrogation’’ as formal terms and ‘‘question’’ and ‘‘questioning’’ as functional terms.1 The restrictions I have chosen to place on the scope of my investigation are these: 1. I will restrict this study to clause-level interrogation and negation, as opposed to constituent interrogation (including interrogative-word questions) and constituent negation.2 2. My hypothesis will involve, as much as possible, ‘‘dedicated’’ interrogative and negative morphosyntax, but not necessarily morphology expressing interrogation or negation among other meanings.3 3. I will leave for future research to explore the implications of my claims for negative interrogatives and negative imperatives, as fascinating as such an exploration might be.

differences in interrogation and negation

311

What I will try to show, then, is that it is the diﬀerences in interactional functions between interrogatives and negatives that underlie the pervasive cross-linguistic diﬀerences in their forms.

2. The typological facts Let’s begin with a summary of these cross-linguistic diﬀerences. 2.1 Interrogative grammar The studies of the cross-linguistic patterns in interrogation carried out by Moravcsik (1971) and Ultan (1978) have identiﬁed the following ﬁve strategies for interrogating clauses:4 A.–Intonation B.–Interrogative morphemes C.–Tags D.–Non-intonational phonological marking E.–Inversion A sixth strategy, namely the ‘‘V-not-V’’ type, will be brieﬂy considered in Section 5.1. A. Intonation Most languages seem to be able to indicate interrogation by means of conventionalized intonation patterns; according to Ultan (1978: 218), ‘‘among clause-level Q-features, intonation holds the ﬁrst rank.’’ B. Interrogative morphemes: particle, afﬁx, or clitic According to Ultan, the ‘‘interrogative morpheme’’ strategy is the most commonly found after intonation for forming interrogatives. Intriguingly and importantly, the interrogative morpheme strongly tends to occur in one of two places: either at or very near (typically ‘‘second position’’) the beginning of the utterance, or at the very end. An example of a particle occurring in utterance-initial position is: Zapotec (1) a. nuu bisoze-lu cop father-you ‘Your father is there.’

312

sandra a. thompson b. nee nuu bisoze-lu q cop father-you ‘Is your father there?’

An example of the ﬁnal-particle strategy can be seen in Mandarin: Mandarin (2) a. Yanmei shi wo-de pengyou Yanmei cop 1sg-gen friend ‘Yanmei is my friend.’ b. Yanmei shi wo-de pengyou ma? Yanmei cop 1sg-gen friend q ‘Is Yanmei my friend?’ In some languages, the interrogative morpheme is an inﬂectional aﬃx in the verb; here is an example: W. Greenlandic (3) a. neri-vu-q eat-ind-3sg ‘He ate.’ b. neri-va-ø eat-interr-3sg ‘Did he eat?’ (Sadock 1984) Such aﬃxes have been found, subsequently to the publication of Ultan (1978), to tend to occur in ﬁnal position (Bybee 1985). C. Tag questions Tag questions are cited as ‘‘common’’ by Ultan (1978: 223), though they are often not described in reference grammars. They are appended (typically as a separate prosodic unit) at the end of an utterance: English (4) a. She’s leaving, isn’t she? b. She’s leaving, right? Lamani (Indo-Aryan, India) (5) U jaa-wa cha, koni ka? he goes-he prs neg q ‘He’s going, isn’t he?’ D. Non-intonational phonological markers Ultan (p. 240) mentions non-intonational phonological markers (e.g. glottal stop,

differences in interrogation and negation

313

voicing) associated with ﬁnal phonological segment(s), though he does not name any languages which actually use such a strategy. I suggest that Nchuﬁe (Byrd 1992), a Grassﬁeld Bantu language of Cameroon, in which ‘‘the ﬁnal rhyme in the question domain is lengthened almost 70 per cent compared to the declarative statement’’ (p. 19), is a good example. E. Inversion This strategy is familiar from Indo-European languages, though it does not appear to be common outside Indo-European. German (6) liebt er diese Frau? love:3sg he this woman ‘Does he love this woman?’ This process is often described as ‘‘subject-verb’’ inversion, but is probably better described as ‘‘verb preposing’’ since the ‘‘subject’’ isn’t really relevant; what’s relevant is that the verb (or some verbal element) is placed at or near the beginning of the utterance. Before comparing the strategies used for interrogation with those for negation, I would like to problematize the issue of the locus of the set of interrogative strategies. What is striking about this group of strategies is that they all involve special marking either at the beginning or end of the clause/sentence/utterance, or involve prosodic patterns that characterize the entire clause/sentence/utterance. That is, in order to describe the grammar of interrogation, appeal must be made to some unit other than the clause. The problem in stating the exact nature of the ‘‘unit’’ which is the actual locus of interrogative marking is that grammars are often unclear about this issue, and there is little consistency from one grammar to another. Often the distribution of segmental non-aﬃxal markers for interrogation, such as particles and clitics, is described in terms of ‘‘utterances’’ rather than in terms of grammatical ‘‘predicates,’’ ‘‘clauses,’’ or sentences. In other words, the ‘‘unit’’ in terms of which interrogation must be described seems to be not a grammatical one like ‘‘clause,’’ but a prosodic one like ‘‘intonation unit.’’ This leads me to think that if actual spoken discourse data were considered for many of these languages, the locus of interrogation would in fact turn out to be a prosodic unit such as the intonation unit, as it has turned out to be for those languages for which such data are available. For example, studies of Japanese and Mandarin spoken discourse reveal that question particles (like other ‘‘sentence-ﬁnal’’ particles) regularly occur at the ends of prosodic units, not at the ends of ‘‘predicates,’’ ‘‘clauses,’’ ‘‘sentences,’’ or any other grammatical unit (Clancy 1982, Cook 1992, Maynard 1989, Squires 1994, Tao 1996).

314

sandra a. thompson

Thus, Tao (1996: 52) reports that in one conversational Mandarin database of 1,284 intonation units, 232 out of 233 ‘‘ﬁnal’’ particles occur at the ends of intonation units, where these intonation units include predominantly predicates, clauses, and noun phrases. Example (7) shows an NP intonation unit in the ﬁrst line, and a predicate intonation unit in the second line (see the Appendix for transcription symbols): Mandarin (7) gaozhong luqu fenshuxian ne=, high:school admission score prt ‘The minimum score for high school admission jiu biancheng duoshao ne=, then become how:many prt came to be how many,’ Example (8) shows a particle at the end of a noun phrase intonation unit: (8)

zhenghao bus siji ne, jiu zuozheng shuo, promptly bus driver prt then testify say ‘At that moment, the bus driver testiﬁed, saying . . .’

(9) shows the Mandarin interrogative particle ma at the end of a predicate intonation unit: (9)

bu gen ni shuo le ma? neg to 2sg say prt q:prt ‘Didn’t [anyone] tell you about that already?

I thus suggest that, while most of the available grammatical descriptions, as well as the typological generalizations based on them, are unclear about the locus of the interrogative grammar, recent research clearly suggests that the actual locus of interrogation is a prosodic unit such as the intonation unit. Again, by this I mean that in order to describe the grammar of interrogation for a given language, appeal must be made to a prosodic unit rather than to a grammatical unit. There are several additional pieces of evidence in favor of this claim. First and foremost, all of the ﬁve interrogative strategies discussed by Ultan must appeal to prosodic, rather than grammatical, units for their description. Strategies A, C, and D, ‘‘Intonation,’’ ‘‘Tags,’’ and ‘‘Non-intonational phonological marking’’ must obviously be described in terms of prosodic units rather than ‘‘clauses’’ or ‘‘sentences.’’ As noted above, interrogation is commonly, perhaps universally, signalled by a special intonation contour. The domain of an intonation contour is precisely a prosodic unit such as the intonation unit, and the ‘‘tag’’ question, can only be de-

differences in interrogation and negation

315

scribed in terms of its position with respect to a preceding prosodic unit rather than, say, ‘‘clause,’’ since its primary criterial attribute is that it constitutes a separate prosodic unit. Strategy D, ‘‘Non-intonational phonological marking’’ is also a strategy that depends in no way on any grammatical unit, but operates on the ﬁnal phonological segment of the utterance. Strategies B and E, ‘‘Interrogative morphemes’’ and ‘‘Inversion,’’ likewise have prosodic unit implications, though these are less obvious than is the case with the other three strategies. As noted above, interrogative morphemes are typically at the beginning, in second position, or at the end of an utterance. This disjunction of properties does not deﬁne a natural grammatical class, but it does deﬁne a natural prosodic unit class. And while ‘‘Inversion’’ must be described in terms of the grammatical ‘‘predicate,’’ its eﬀect is also to create an utterance with an interrogative ‘‘signal’’ either in ﬁrst or second position. What these arguments suggest, then, is that interrogation is strongly prosodicunit-oriented. In the next section we consider the grammar of negation, which presents a very diﬀerent picture. 2.2 Negation grammar The cross-linguistic grammar of negation is a topic with a rich literature (cf., e.g., Dahl 1979, Givón 1979, 1982, 1984, 1989, and 1995, Payne 1985, Dryer 1989, Kahrel and van den Berg 1993, and Bernini and Ramat 1996). From this literature, as well as a number of language-speciﬁc studies, it is clear that the grammar of negationoperatesaccordingtodiﬀerentprinciplesthandoesthegrammarofinterrogation. Remarkably, when we compare the grammar of standard negation with the grammar of interrogation, we ﬁnd that the grammar for standard negation does not involve initial or ﬁnal particles, tags, intonation, or phonological markers. If we restrict our attention to what Payne (1985: 198) terms ‘‘standard negation,’’ we ﬁnd the following cross-linguistic strategies for marking negation, as discussed by Dahl (1979) and Payne (1985: 198): A. Negative particles B. Negative verbs C. Negative aﬃxes A. Negative particles According to Dahl and Payne, negative particles are associated with the verb (which I am more broadly referring to as the ‘‘predicate’’), in the sense that their distribution must be described in terms of the predicate. These particles are distinguished from aﬃxes by the usual criteria: they are free morphemes rather than bound to a lexical stem (see Dahl 1979 for an important discussion of these criteria).

sandra a. thompson

316 Russian (10) on ne igraet he neg play ‘He doesn’t play.’

Mandarin (11) ta bu he jiu 3sg neg drink wine ‘S/he doesn’t drink.’ B. Negative verbs The ‘‘negative verbs’’ discussed by Payne (Dahl’s 1979 ‘‘ﬁnite element’’) are either diachronically or synchronically ‘‘higher’’ verbs taking a ‘‘complement’’ whose predicate is the ‘‘lexical’’ (what I will call the ‘‘aﬃrmative’’) verb. The negative verb typically has properties characteristic of verbs in the language: it inﬂects like a verb, it takes the same position as verbs do if the language has a relatively rigid word order, and the aﬃrmative verb often becomes a less ﬁnite ‘‘reduced complement’’ (Noonan 1985). Languages with higher negative verbs include Tongan, Evenki, Kawaiisu, Minnan Chinese, and Finnish5. Here is an example from Evenki: Evenki (Tungus, Siberia) (see also Nedyalkov 1993): (12) a. Bi dukuwu˜n-ma duku-ca¯-w I letter-acc write-pst-1sg ‘I wrote a letter.’ b. Bi dukuwu˜n-ma ə-cə-w duku-ra I letter-acc neg-pst-1sg write-pple ‘I didn’t write a letter.’ C. Negative afﬁxes The ﬁnal standard negation strategy considered by Payne is negative aﬃxes (Dahl’s 1979 ‘‘Morphological Neg’’). A negative morpheme is considered an aﬃx if its position is bound to the verb word. Turkish (see also van Schaaik 1993) (13) a. git-ti-m go-pst-1sg ‘I went.’ b. git-me-di-m go-neg-pst-1sg ‘I didn’t go.’

differences in interrogation and negation

317

2.3 Summary of typological facts What we have seen in this survey of the ways in which grammars mark interrogation and standard negation is two very diﬀerent proﬁles. The grammar of interrogation involves one or more of these ﬁve strategies for forming clausal interrogatives: (1) conventionalized intonation; (2) interrogative morphemes at the beginning or end of the utterance; (3) tags; (4) phonological markers on ﬁnal segments; (5) verb inversion. As I suggested at the end of the discussion of the typology of interrogation, what is striking about this group of strategies is that they all involve special marking either at the beginning or end of the clause/sentence/utterance, or involve prosodic patterns that characterize the entire clause/sentence/utterance. In contrast, the grammar for standard negation tends not to involve initial or ﬁnal particles, and does not involve tags, intonation, or phonological markers (except as ‘‘secondary modiﬁcations’’). Instead, standard negation for clauses is generally indicated by the use of a special word or aﬃx which is positioned with respect to, and must be described in terms of, the predicate, not with respect to the beginning or end of something, or to anything prosodic. I suggest that these large-scale diﬀerences can be interpreted in terms of the domain, or locus, of the two grammatical phenomena we are considering. I have suggested that the actual domain of interrogation is the prosodic unit. The typological facts for negation suggest that the domain for standard negation is the clause, or more speciﬁcally the predicate, which I take to be the deﬁning feature of ‘‘clause.’’ In the next section, I will show that these diﬀerences in domain ﬁnd an explanation in the diﬀerences between what interrogation and negation do in conversational interactions.

3. Explanation The facts outlined in Section 2, highlighting a signiﬁcant diﬀerence in the locus of the grammaticization of interrogation vs. negation, do not ﬁnd a ready structural explanation: there is no structural reason why interrogation makes use of prosodic-unit-oriented strategies but negation makes use of predicate-centered strategies. Horn has proposed a semantic solution to this very fact. In his view, this striking formal diﬀerence between interrogation and negation reﬂects a semantic asymmetry between the two categories. Crucially, the yes-no question operator applies to a fully formed proposition . . . , arguably in the manner of the illocutionary-force

318

sandra a. thompson

indicating devices of speech act theory . . . . But descriptive negation does not constitute a speech act . . . , or indeed a propositional operator. (Horn 1989:473)

Some such view seems to be widely shared,6 but it has several problems. First, it is not clear to me what it would take to empirically deﬁne ‘‘operators’’ or ‘‘propositions,’’ or what it means to ‘‘apply’’ an ‘‘operator’’ to a ‘‘proposition.’’ These notions have indeed been discussed in the rich logical semantic literature, but, in contrast to ‘‘prosodic unit,’’ they are theoretical constructs which have not arisen from studying actual conversational interactions. When, in fact, one turns to the spoken language data, it is not clear what would count as a ‘‘proposition,’’ not to mention a ‘‘fully formed proposition.’’ Second, similar problems emerge in trying to apply terms from speech act theory to the analysis of actual conversational data (see especially Schegloﬀ 1992 on the diﬀerence between ‘‘empirical/sociological’’ and ‘‘conceptual/philosophical’’). A similar proposal has been suggested to me, that the interactional explanation I am proposing can be replaced by a simpler one grounded in ‘‘scope’’ properties of interrogation. This argument runs something like this: since interrogatives have an entire ‘‘utterance’’ in their scope, while (standard) negation has only the predicate in its scope, we can predict that interrogatives would have the utterance as their locus of grammaticization, while negation would have only the grammatical predicate or clause as its locus of grammaticization. To me, the primary problem with both these proposals is that they push the explanation back one major step: they have nothing to say about why interrogation should have diﬀerent ‘‘scope’’ properties from negation, or why interrogation would be a ‘‘speech act,’’ while negation would not. But there are two further speciﬁc problems with this second proposal. First, empirical evidence from actual interactions supporting such a claim is diﬃcult to ﬁnd, and has, to my knowledge, never been oﬀered, particularly with respect to negation. Although most linguists appear to believe strongly that ‘‘intuitively’’ an utterance such as ‘‘Sally isn’t here today’’ has only the predicate but not Sally in its ‘‘scope,’’ this is not apparent to me, and I have neither seen any arguments that this is so, nor have I had success in ﬁnding any empirical grounds for it in my database.7 In example (14), for instance, it isn’t clear what it would mean to argue that the ‘‘scope’’ of negation includes the predicate gonna have money, but not the subject I, or how we would test such a claim.

differences in interrogation and negation

319

(14) (The family members have been talking about the new vacuum cleaner Kendi has just received for a birthday present, and Marci oﬀers Kendi the following information) (transcription follows Du Bois et al. 1993; see Appendix for a summary): Marci: (H) you can get the ba=gs at, . . at . . United Vacuum? Kendi: I have to buy ba=gs? Marci: . . . nyeah, → Kendi: I’m not gonna have money. → not budgeted. But there is a related and even more disturbing problem: what has been described as ‘‘scope’’ in the linguistic literature appears to be nothing more or less than the socially situated mutual construction of the meanings of utterances, which is clearly heavily dependent on context. In fact, in actual discourse, exactly what is being denied by a negative utterance is not directly recoverable from the structure of the utterance. In considering actual data, therefore, ‘‘scope’’ issues take a much less prominent position than is the case in some of the literature considering only constructed examples. Appealing to ‘‘scope’’ as if it were a ‘‘ﬁxed’’ aspect of linguistic structure, without considering the actual interactional context of an utterance, has not proven to be helpful in describing or explaining the structure of real conversational interactions.8 An excellent further study might pursue the ways in which scope issues do arise in ordinary conversations; based on my study of conversational data, I suggest that ‘‘scope’’ is highly indeterminate in interaction. In order to come closer to an explanation of the typological features of interrogation and negation, I suggest that it is at least worth considering an interactional perspective, which looks at what interrogatives and negatives actually do interactionally in conversational encounters. As Schegloﬀ has suggested (1989: 143): If the conduct of language as a domain of behavior is biological in character, then we should expect it (like other biological entities) to be adapted to its natural environment . . . Transparently, the natural environment of language use is talk-in-interaction, and originally ordinary conversation. The natural home environment of clauses and sentences is turns-at-talk. Must we not understand the structures of grammar to be in important respects adaptations to the turn-at-talk in a conversational turn-taking system with its interactional contingencies? [Emphasis added—SAT]

To carry out an interactional investigation, of course, corpora are essential; introspection can be a valuable hypothesis-generating mechanism, but we simply cannot know how interrogation and negation are used in everyday talk without examining conversational data. Here we are severely limited in the availability of extensive

320

sandra a. thompson

conversational corpora from a variety of languages. Given this limitation, I am constrained to use corpora from English as my data, and to refer to studies based on English corpora. My ﬁndings are thus suggestive, subject to conﬁrmation for a wide range of languages, but in my opinion, the implications from the English corpora are clear. The ‘‘Q-N Hypothesis,’’ which I will elaborate and defend in section 3.3 is this: The interactional diﬀerences between interrogatives and negatives motivate the diﬀerential way in which these two types of constructions have grammaticized and the diﬀerential loci for these two types of constructions.9 But ﬁrst, we must consider the conversational data to determine, as best we can, what the interactional functions of interrogation and negation actually are. 3.1 Interrogatives seek information A recent corpus study of interrogation in English conducted by Freed (1994) suggests that interrogatives can be categorized into 16 groups, according to type of information sought. These groups fall along a continuum, ranging from the most narrowly factual questions to expressions of speaker stance, and can be conﬂated into four broader classes.10 Freed’s study includes constituent (question-word) questions, but I will limit my discussion of her ﬁndings to non-WH-form interrogation. 1. Seek information external to conversation (15) (16)

They only make that . . . with Nutrasweet though, do—don’t they. Are you going home tonight?

2. Seek information about the talk itself11 (17) A: No. He’s got a grandkid but it’s not from his kid. It’s from his second wife. From his, yeah, not his own grandchild. →B: Oh, he’s remarried? 3. Seek open-ended information about speaker/hearer (18)

And I went, did I tell you I went to, uh, away for a weekend?

4. Seek no information, but can be said to convey information (19) A: . . . Normally I leave 45 minutes before a class. Just to make sure. B: That long? →C: Yeah. It’s too much, isn’t it? Then I get here early. Of these four broad types of questioning functions, based on type of information

differences in interrogation and negation

321

sought, we can say that the ﬁrst three seek information. Counting only the ‘‘nonWH interrogation’’ types (including Freed’s yes-no, declarative questions with question intonation, and tags), we ﬁnd that these ﬁrst three categories account for 80 per cent of the non-WH interrogatives (n = 816) in Freed’s data. So a non-WH interrogative form, 80 per cent of the time, is involved in interactions which request a response on the part of the interlocutor. In Conversation Analysis terms, the information-seeking question is described as the ﬁrst member of an ‘‘adjacency pair,’’ which establishes ‘‘conditional relevance’’ on whatever follows it (Schegloﬀ 1968, Schegloﬀ and Sacks 1973).12 That is, crucially the ﬁrst ‘‘pair part’’ establishes a ‘‘conditional relevance’’ on anything that occurs in the slot that follows; whatever comes to be said there will be inspected to see how it might serve as an answer, and if nothing is said, then the resulting silence will be taken as notable. (Goﬀman 1976 (1981: 7))

Or, as Schegloﬀ (1984: 35) puts it, One thing one might mean by an utterance being interactionally or conversationally a question is that it lays constraints on the next slot in the conversation of a sort special to the Q-A pair type of adjacency pairs.

Seen in these terms, and extrapolating from the English data, it is not surprising that the prosodic unit is the locus for the grammaticization of interrogation. The reasoning behind this claim goes as follows: as noted above, recent research has shown that prosodic units play a critical role in the smooth exchange of conversational turns; points at which prosodic unit completions occur strongly tend to be points at which turn transition occurs. Since precisely what interrogation does is to call for a turn transition in that a response from another speaker is implicated, then it makes sense that the prosodic unit is the natural locus for the grammaticization of such interactional moves, since these moves help deﬁne points of possible turn transition, and, as noted above, recent research strongly suggests the prosodic unit as playing a major role in deﬁning these points of turn transition. We ﬁnd, then, that the types of grammar found cross-linguistically to be associated with seeking information are precisely those whose domain is what grammars often label ‘‘the entire utterance,’’ or what I am suggesting will turn out to be, with examination of more data from conversational language use, a prosodic unit. That is, interrogative grammar must be described in terms of dedicated interrogative intonation covering the entire prosodic unit, prosodic unit tags, and special grammar at the beginning or at the end of the prosodic unit. In sum, the great majority of non-WH interrogative forms in English conversation anticipate responses, are ﬁrst ‘‘pair parts’’ of adjacency pairs, and have prosodic

322

sandra a. thompson

and segmental signals whose distribution can best be described in terms of prosodic units, rather than grammatical units. Now, let’s compare the interactional functions of negation. 3.2 Negatives do denying and rejecting Tottie (1991) is a corpus study of negation in spoken and written English. In addition to diﬀerences between these two modalities, Tottie is also interested in factors motivating speakers’ choice of aﬃxal (e.g., unhappy) vs. clausal (she’s not happy) negation. Consistent with my purposes in this paper, I will consider only the results from the conversational portion of her data (taken from the London-Lund corpus (Svartvik and Quirk 1980), and only the clausal (non-constituent) negation ﬁndings. One of the key points in Tottie’s analysis is her functional classiﬁcation, justiﬁed at length, of negative sentences as follows (p. 22): (i) (ii)

REJECTIONS (including REFUSALS) DENIALS: (a) EXPLICIT (b) IMPLICIT

I will not discuss ‘‘rejections’’ here since they constitute a small fraction of the negative utterances in Tottie’s data (7 per cent), but will rather focus on the category of ‘‘denial.’’ Denials, according to Tottie, are of two types: the ﬁrst is a denial of something which has just been uttered (‘‘explicit’’ denial), the second is a denial of ‘‘something which has not been explicitly asserted’’ (‘‘implicit denial’’) (p. 21). The idea of ‘‘explicit’’ and ‘‘implicit’’ denial has been discussed extensively, though not with empirical support, by Givón (1979 (3.2.2), 1982, 1984 (chap. 9), 1989 (4.5), and 1995 (4.2)), who argues that ‘‘the speaker uttering a negative sentence assumes that the hearer knows that the corresponding aﬃrmative was likely or has been previously mentioned’’ (1979, p. 104). In fact, as we shall see, Givón’s characterization does not quite ﬁt the data, but the idea that what is being denied is typically not explicitly mentioned is signiﬁcant. Table 1 summarizes Tottie’s ﬁndings for the conversational English portion of the London-Lund corpus (adapted from her Table 3.3, p. 35). Two highly relevant ﬁndings emerge from Table 1. First, 81 per cent of the 427 negative clauses in her conversational database are either explicit or implicit denials. Second, it is clear that the largest category of negative clauses in the conversational data is that of ‘‘implicit denial,’’ which I will discuss further just below.13 In Tottie’s analysis, the category of ‘‘implicit denial’’ characterizes a substantial majority of the negative clauses in the data. Remember that for Tottie, all that is

differences in interrogation and negation

323

Table 1. Types of negatives in Tottie’s conversational database. Denials Rejection Other* Total

Impl. 286 Expl. 63 7 71 427

67 14 2 17 100

81

83

*Includes negative imperatives and interrogatives.

meant by ‘‘implicit denial’’ is a denial of ‘‘something which has not been explicitly asserted.’’ She does not say that ‘‘implicit denial’’ ‘‘denies something which is implicit.’’ This distinction is important: though Tottie gives relatively few corpus examples, conversational data from the Corpus of Spoken American English (= CSAE; see Note *) show clearly that there is typically no sense in which a negative clause denies anything, either explicit or implicit, in the conversation; these data thus do not support Givón’s claim that the aﬃrmative is ‘‘assumed to be likely.’’ To the extent that a negative clause denies something that is ‘‘‘likely,’’ this ‘‘something’’ is generally neither explicitly mentioned nor ‘‘‘‘assumed’’ or ‘‘‘implicit’’ in any sense. Let us re-consider example (14) again from the CSAE, in which a good case can be made for ‘‘implicit denial’’ of a ‘‘likely’’ corresponding aﬃrmative. (14) (The family members have been talking about the new vacuum cleaner Kendi has just received for a birthday present, and Marci oﬀers Kendi the following information) Marci: (H) you can get the ba=gs at, . . at . . United Vacuum? Kendi: I have to buy ba=gs? Marci: . . . nyeah, → Kendi: I’m not gonna have money. → not budgeted. Here one could argue that Marci’s information about where to get the bags strongly implies that Kendi will have the money to buy bags, and it is this implicit proposition that her two negative clauses deny. In this case, Givón’s claim that the aﬃrmative is ‘‘assumed to be likely’’ seems to be valid. However, more typical for the CSAE database are cases such as the following, in which the ‘‘implicit’’ proposition being denied would be diﬃcult to identify, and could hardly be said to be ‘‘assumed to be likely’’:

324

sandra a. thompson

(20) (Marci, Kendi, and Wendy are talking about a can of cream soda that Marci is about to open) 1 Marci: . . why do these ca=ns . . get so warped. only the – – . . only the Sam’s Club cans get so warped. 2 Wendy: (H) % I think you should [write– – 3 Kendi: [1cause they’re1] [2cheap cans, 4 Wendy: [2%just %= write a letter to em2], 5 Kendi: that’s why2]– – 6 Wendy: and complain Marci. → 7 Marci: . . (H) = what I’m gonna complain about, → 8 is that they don’t make white grape. I have given a bit of previous context for the negative sentence in lines 7–8 speciﬁcally to show the typical way in which ‘‘implicit denial’’ negatives occur in the CSAE database. In this example, it would be diﬃcult to claim that the idea that ‘‘they do make white grape’’ is ‘‘assumed to be likely,’’ let alone mentioned, in the preceding context. No one had mentioned anything having to do with any aspect of white grape cream soda. Let us consider another example: (21) (Kendi is opening her birthday present, a set of blue plastic baking implements) Marci: oh that’s=. in blue. Kendi: in blue=. → [that’s not my color]. → Wendy: [it’s not green], I’m sorry. No mention of any colors had been made in the conversation up to this point. The family members present at this birthday party know what Kendi’s favorite colors are, and can, of course, readily interpret the two negative utterances at the arrows. The interpretation of negative utterances clearly has much to do with expectations. The point is that these expectations are typically not in the linguistic context in such a way that they can be said to participate in any turn-taking or interactional sequential organization, as interrogation can. Here is a ﬁnal example from the same transcript.

differences in interrogation and negation

325

(22) The family members have been talking about the new vacuum cleaner Kendi has just received for a birthday present, and where to buy the bags for it, when Kevin says: You guys won’t believe what happened to us in the parking lot of the mall the other day. It is easy to see that Kevin is oﬀering a ‘‘story preface’’ (Sacks 1974, 1992: vol. II, part I, lectures 1, 2, and part IV, lecture 2), which happens to be framed with a formulaic negative epistemic phrase. Again, what this clause ‘‘denies’’ is not only not ‘‘implicit,’’ but is very nearly nonsensical. It might be argued that we should ‘‘exclude’’ or ‘‘count separately’’ such near-formulas as ‘‘you won’’t believe’. But even if we were to do this, their numbers are relatively small, and examples of negative clauses whose aﬃrmative counterparts are not ‘‘assumed,’’ as illustrated by (20) and (21), abound in the data. I have taken some space to illustrate the types of contexts in which negation occurs in the data, because many linguists, hearing and reading earlier versions of this paper, have expressed their convictions that negative clauses constitute a ‘‘response’’ to something. As I have emphasized, the data do not bear out this claim.14 In stark contrast to interrogation, then, where the data show an 80 per cent tendency for an interrogative clause to seek information, and thus to be the ﬁrst part of an adjacency pair, there does not seem to be such a constraint on the use of ‘‘standard’’ negative clauses. The data do not provide any evidence that standard negative clauses participate in any adjacency pair structure, as interrogative clauses systematically do. Notice that even a case such as (14), where the implicit ‘‘proposition’’ being denied could be readily identiﬁed, cannot be considered an example of an adjacency pair, or any other systematic interactional pattern, since the set of utterances leading up to the denial cannot in any way be said to create an expectation of a certain type of next interactional move. That is, the critical criterion of ‘‘conditional relevance’’ for adjacency pair structure is missing with negative utterances; referring again to Goﬀman’s characterization of questions as ﬁrst pair parts of adjacency pairs, we recall that ‘‘the ﬁrst pair part establishes a ‘conditional relevance’ upon anything that occurs in the slot that follows,’’ such that ‘‘whatever comes to be said there will be inspected to see how it might serve as an answer.’’ This does not appear to be the case with ‘‘standard’’ negative utterances; negative clauses function largely to deny, but what they deny is typically not explicitly present in the conversation. Nor do the denials show any signs of themselves establishing a ‘‘conditional relevance’’ on what follows them.

326

sandra a. thompson

I suggest that this is the interactional diﬀerence between interrogation and negation: Interrogatives are interactive: they can be empirically shown to function to do questioning, anticipating answers and serving as the ﬁrst pair part of a systematic interactive format known as an adjacency pair. The locus of interrogative morphosyntax is the prosodic unit, which previous research has shown to be a basic unit of interaction, playing as it does a major role in the way in which speakers exchange turns in conversation. (Standard) negation has not been found to participate in any systematic interactional formats; negation can be empirically shown to serve to deny some state or event which is usually neither explicit nor ‘‘implicit’’ in the context. The locus of negation is the predicate, which is the center of the basic unit of reporting, or denying, the occurrence of events or state, the clause.

4. Putative problems Having outlined the broad-scale typological properties of interrogatives and negatives, and proposed an explanation for these properties grounded in the interactional contexts in which they are used, I would like to brieﬂy address two apparent problems with the proposed explanation. The ﬁrst concerns the possible existence of languages without Q-A adjacency pair organization. Walsh (1994) suggests that northern Australian aboriginal groups tend to ‘‘broadcast’’ information to groups of listeners rather than directly address questions at speciﬁc addressees, and that the dissemination of information is tightly controlled. Walsh’s point is very well taken and highly suggestive. One might ask, then, whether my account, which ﬁnds a motivation for the grammar of interrogation in the systematic organization of adjacency pairs, would be less viable in view of cultures such as those described by Walsh. While it is uncontroversial that cultural diﬀerences exist in the use of various types of adjacency pair organization (Ochs 1976, Wierzbicka 1985, inter alia), this question is unfortunately one for which there simply is not, to the best of my knowledge, suﬃcient evidence to begin to answer at this time. It would be important to examine data comparable to Freed’s for a variety of speech communities to see how interrogatives are used in everyday interactions and whether similar results obtain, but so far no such studies have been conducted. The second potential problem concerns interrogative markers in intonation-unit-

differences in interrogation and negation

327

ﬁnal position. The question goes like this: if prosodic units are the locus of interrogative grammar, then marking in intonation-unit-initial position would be predicted, as would whole routinized intonation contours, but marking in intonation-unit-ﬁnal position (such as ﬁnal interrogative particles) would seem to be problematic for this claim, since interlocutors would not be able to ‘‘wait’’ until the end of the prosodic unit to ‘‘ﬁnd out’’ whether the utterance is an interrogative or not. My response to this question is as follows: there is a diﬀerence between claiming that the prosodic unit is the locus for interrogative grammar, and claiming that interlocutors rely exclusively on such grammar for interpreting utterances as interrogatives. Much research on conversational organization has shown clearly that there is a wide range of factors shaping how utterances are in fact interpreted, and how their trajectories are projected, including prosodic, rhythmic, sequential, gestural, pragmatic, morphological, and syntactic factors (see, e.g., Auer and Di Luzio 1992, Couper-Kuhlen 1993 (especially chaps. 7, 8), Goodwin 1981, Ochs et al. 1996, Stenström 1984, Weber 1993, and references cited in these works). Thus, it is almost certainly not the case that interlocutors ever have to actually ‘‘wait’’ until they hear an intonation-unit-ﬁnal particle or aﬃx to know that the utterance being produced is, or is likely to be, an interrogative. Thus, what is crucial is seeing the prosodic unit as the locus for the grammaticization of interrogative marking, rather than assuming that listeners actually have to ‘‘wait’’ until the end of such a unit to determine the signiﬁcant interactional properties of this unit. Having considered and attempted to put to rest the putative problems that I am aware of with my hypothesis, I will now turn to two interesting counterexamples, and then to my conclusions.

5. Counter-examples As noted above, the typological picture of interrogation presented by Ultan and the typological picture of negation presented by Payne each has one clear counterexample. These must be taken seriously, and yet, the facts which support my hypothesis seem compellling enough to also be taken seriously. It is my hope that a deeper understanding of cross-linguistic regularities might emerge from a consideration of all of these facts, as well as those which will undoubtedly emerge from further studies of everyday language use. 5.1 ‘‘V-not-V’’ Interrogatives A prominent interrogative type not mentioned by Ultan is the ‘‘V-not-V’’ interrogative, found mostly in Chinese and continental southeast Asian languages. This type,

328

sandra a. thompson

as is well-known, involves the grammaticized juxtaposition of an aﬃrmative and a negative verb or verbal element: Mandarin (23) ni chou-yan bu chou-yan? 2sg smoke neg smoke ‘Do you smoke?’ (24)

ta hui bu hui youyong? 3sg able neg able swim ‘Can s/he swim?’

In this type of interrogative structure, the locus for the grammar marking the utterance as an interrogative is not the prosodic unit, but clearly the grammatical predicate. There are obvious historical motivations for this, since the interrogative morphosyntax is actually negative morphosyntax, which would be predicted to be predicate-centered. The number of languages employing this strategy is small and their geographical range is limited.15, 16 5.2 SVO Languages with ‘‘clause-peripheral’’ negative markers Dryer (1989), p.c. has noted a group of languages which are genuine counterexamples to the typological generalizations for negation described in Section 2. These are SVO languages for which a negative particle is described as occurring in ‘‘sentence-ﬁnal’’ position rather than associated with the verb, which is the predicted situation. In Dryer’s database, as of July, 1995, there were 37 such languages; here is an example from Ngizim (Schuh 1972): Ngizim (25) ii naa duuka bai 1sg have horse neg ‘I don’t have a horse.’ Such cases are indeed counterexamples to the generalizations which this paper seeks to explain. However, in my favor, the number of languages (37) in which this occurs is a small fraction (5 per cent) of the total number of languages in Dryer’s database (775). Furthermore, the geographical distribution of these languages in his database is restricted essentially to two areas in the world: central Africa and Southeast Asia and Oceania.17 SVO languages with initial negative particles are quite rare; Dryer found 5 in his 775-language database; for these, it would be important to consider the

differences in interrogation and negation

329

diachronic development of such particles. As discussed in connection with the Finnish negative in Section 2.2.2, such synchronic negative particles may well have grammaticized from older negative verbs, which accounts for their position.

6. Unexpected support for the Q-N Hypothesis 6.1 Afﬁx order My hypothesis makes a speciﬁc claim regarding the locus of grammaticization for interrogation and negation, namely that the locus for the grammaticization of interrogation is the prosodic unit, while the locus for the grammaticization of negation is the predicate. We have seen that, by and large, the typological facts support this hypothesis. However, one could take the hypothesis one step farther, consider only aﬃxing languages, and make certain predictions regarding the expression of negative and interrogative aﬃxes. Acknowledging the very important factors inﬂuencing the grammaticization of aﬃxes and their order (see especially Bybee 1985 and Bybee et al. 1994), the type of aﬃxation found nicely supports the generalization that interrogative marking favors the periphery of the utterance in a way that negative aﬃxes do not. I will very brieﬂy outline the three major arguments in which this hypothesis is conﬁrmed. A. Interrogative aﬃxes should show a stronger tendency to appear as suﬃxes than preﬁxes in verb-ﬁnal languages, since a preﬁx in a verb-ﬁnal language would not be in prosodic-unit-ﬁnal position. This hypothesis is strongly conﬁrmed in Bybee (1985) and in Dryer’s database, where 46 verb-ﬁnal genera show interrogative suﬃxes while only three show interrogative preﬁxes. In Dryer’s sample of 775 languages, V-ﬁnal languages show an overwhelming preference for interrogative suﬃxes, as expected: 46 genera to 3.18 B. Bybee (1985) notes that when interrogative markers are suﬃxes, they tend to occur as the ﬁnal suﬃx on the verb. This is strong support for my hypothesis, which predicts interrogative markers in intonation-unit-ﬁnal position.19 C. For verb-ﬁnal languages, interrogative preﬁxes should be much less common than negative preﬁxes, since negative preﬁxes conform to the generalization that negative markers tend to be associated with the verb, but interrogative preﬁxes in verb-ﬁnal languages would not be in intonation-unit-ﬁnal position. This hypothesis is also strongly conﬁrmed, as imagined, since so few verb-ﬁnal languages have interrogative preﬁxes to begin with: the percentage of verb-ﬁnal

sandra a. thompson

330

genera with negative preﬁxes is 25 per cent, while the percentage of verbﬁnal genera with interrogative preﬁxes is only 6 per cent. 6.2 ‘‘Metalinguistic’’ negation There has been some discussion in the literature of a ‘‘special’’ type of negation that typically occurs ‘‘sentence-initially’’ (see, e.g., Bossuyt 1983 and Yeh 1995): English (26) A: You trapped two mongeese B: Like hell I trapped two mongeese, I trapped two mongooses (Drozd 1995) Mandarin (27) bushi Lao Zhang lai le, shi Lao Li neg:cop Old Z. come prt cop Old L. ‘It wasn’t Old Zhang who came, it was Old Li.’ As can be seen from the examples, this ‘‘special’’ type of negation appears to express something like what Horn (1989) calls ‘‘metalinguistic negation,’’ ‘‘a device for objecting to a previous utterance on any grounds whatever’’ (p. 363).20 Although I know of no corpus-based studies of this phenomenon for adult interactions, Drozd’s (1995) study of child English exempliﬁes this phenomenon nicely. It also might appear to be a counterexample to the Q-N hypothesis, since its distribution seems to be best described as ‘‘utterance-initial,’’ rather than in terms of the predicate, as is predicted for ‘‘standard’’ negation. But in fact it also strongly supports the hypothesis, since the utterance-initial position of ‘‘metalinguistic negation’’ can be explained in precisely the same way as the utterance-based grammar of interrogation: although further corpus studies are needed, Drozd’s data clearly show that metalinguistic negation is interactional; it responds to a previous utterance. While some examples of ‘‘standard’’ negation do this as well, one could argue that ‘‘metalinguistic negation’’ is a conventionalized form of what Tottie calls ‘‘explicit denial.’’ Related to ‘‘metalinguistic negation’’ is the ‘Wayne’s World NOT!’. I have presented versions of this paper in the US, Europe, Asia, and Australia, and have yet to make a presentation in which someone does not ask about this English clauseﬁnal not!, which seems to have entered the language in expressions such as the following via the movie ‘‘Wayne’s World.’’ (28)

Nice tie, Dad—NOT!

While this construction was a part of popular slang in the early nineties in the US, its popularity appears to already be on the wane. In any case, similar arguments for

differences in interrogation and negation

331

its ﬁnal position in the prosodic unit can be made: this not! is also not ‘‘standard’’ negation; it can be analyzed as a mini-dialogue conversational routine whereby the not! is an explicit denial of the utterance which immediately precedes it.

7. Conclusions The grammar for interrogative and negative utterances diﬀers cross-linguistically in consistent and characteristic ways. With certain areally-restricted exceptions, interrogative grammar tends to be prosodic-unit-based, while negative grammar tends to be clause-based. This diﬀerence can be directly related to the diﬀerence in their interactional functions, as indicated by empirical studies of large corpora for conversational English. Interrogative grammar arises from the interactional function of questioning, which is an interactional move that ‘‘lays constraints on the next slot in the conversation’’ (Schegloﬀ 1984). Thus the locus of interrogative grammar is the prosodic unit, a primary unit of interactional turn-taking. In contrast, negative grammar arises from the function of denying a state or event, which shows no such interactional structuring. Its locus is thus the clause, the locus of reports of states and events. Grammar has been persuasively argued to be the sedimentation of patterns of language use (Du Bois 1985, 1987, Hopper 1987, 1988, Hopper and Thompson 1993). If this is so, grammarians wishing to understand grammar have no choice but to investigate the conventionalization and automatization of linguistic collocations that people use very frequently by studying what it is that they actually do in everyday talk. Cross-linguistic generalizations must also ultimately be explainable in terms of language use (e.g., Bybee et al. 1994). What I hope to have shown in this study is one way in which one set of cross-linguistic generalizations can be explained in terms of patterns of language use. While we are limited at present to investigating such patterns for languages for which we have extensive interactional corpora, this study may at least point towards directions which might fruitfully be taken to come to an understanding of why grammars are the way they are.

Appendix Symbols for Discourse Transcription from Du Bois (1991) and Du Bois et al. (1993) UNITS Intonation unit Truncated intonation unit Word Truncated word

{carriage return} –– {space} –

SPEAKERS Speaker identity/turn start Speech overlap

: [ ]*

TRANSITIONAL CONTINUITY Final Continuing Appeal

. , ?

LENGTHENING

=

PAUSE Long Short Latching

... .. (0)

VOCAL NOISES Vocal noises Inhalation Exhalation Glottal stop Laughter

() (H) (Hx) % @

*Certain brackets are indexed with numbers to clarify which speech overlaps with which.

differences in interrogation and negation

333

Notes * I am grateful to the following people who have discussed the ideas in this paper with me: Mira Ariel, Mark Aronoﬀ, Melissa Bowerman, Joan Bybee, Hilary Chappell, William Croft, Angélica Cunha, Eve Danziger, Kenneth Drozd, Mark Durie, Leonard Faltz, Alice Freed, Nikolaus Himmelmann, ChuRen Huang, Randy LaPolla, Stephen Levinson, Edith Moravcsik, Thomas Payne, Russell Schuh, Susan Steele, Hongyin Tao, R. McMillan Thompson, Elizabeth Weber, Sören Wichmann, and David Wilkins. None of these people is responsible for the use I have made of their suggestions. —I am especially grateful to the Santa Barbara Corpus of Spoken American English (CSAE) team, especially its director, John Du Bois, and Robert Englebretson, for their work in gathering and transcribing the data referred to in this paper, and to Matthew Dryer, who gave me invaluable information from his stratiﬁed database. 1. I am following Jespersen (1924) for this distinction for ‘‘question’’ and ‘‘interrogative.’’ For further discussion, see also Hiz˙ (1978) and McHoul (1987). 2. In fact, there are interesting parallels between constituent interrogation and constituent negation, which would be important to pursue. But I am constrained to keep this investigation to manageable proportions. This restriction is also justiﬁed by the lack of empirical studies of constituent negation in actual interactions, and by the relative infrequency of constituent interrogation and negation in ordinary conversation. In particular, based on corpus studies in English, constituent negation appears to be extremely rare: in one English conversational database containing 80 negatives, I found no occurrences of constituent negation. The proportion of constituent interrogation has also been shown in at least one databaase to be lower than that of non-constituent interrogation, but by a much smaller margin: Freed (1994: 635) reports that 41 per cent of 1275 interrogative forms in her conversational database were ‘‘yes-no’’ interrogatives, while 31per cent were constituent interrogatives. 3. This means that I will not consider languages in which interrogation might be expressed by a general mood marker expressing something like ‘‘irrealis,’’ as discussed, for example, for Caddo by Chafe (1995). 4. For a survey of interrogatives in several languages, see Chisholm (1984). 5. As noted by Dahl (1979), Payne (1985: 209–11), and M-L Helasvuo, p.c., the Finnish ﬁnite ‘‘higher’’ verb e- does not have all the characteristics of an ordinary lexical verb in Finnish, and can be described as grammaticizing as a negative particle. 6. For example, a number of models of linguistic structure have been designed to capture this diﬀerence; perhaps the most explicit is the ‘‘layered’’ model represented in Dik (1989), where negation is an operator with only the ‘‘predication’’ in its scope, while interrogation is an ‘‘illocutionary’’ operator, which takes the ‘‘whole clause’’ in its scope. Similar arguments can be found in Foley and Van Valin (1984). 7. In fact, one anonymous reader of this paper has suggested that it could as easily be argued that in ‘‘Is Sally here today?,’’ Sally also isn’t in the scope of the interrogative. This reader goes on to suggest, citing the constructed examples (i) and (ii) below, that Horn’s account actually suﬀers from the fact that whatever the ‘‘scope’’ issues are with negation, they can be argued to be the same for interroga-

sandra a. thompson

334

tion; I hope that an empirical study can be carried out soon to investigate the ways in which such interpretations are actually negotiated in ordinary conversations. (i) (ii)

Did Bill buy Daphne a present? Bill didn’t buy Daphne a present.

8. For a similar point with regard to claims about the ‘‘scope’’ of the Japanese ﬁnal interrogative particle ka, see Hinds (1984: 180–2). 9. A description of the range of grammaticization pathways by which these grammatical and prosodic regularities consistently arise remains a topic for future study. 10. For additional corpus studies of various aspects of interrogation, see Athanasiadou (1991), Heritage and Roth (1995), Selting (1992), Stenström (1984, 1988), Tsui (1992), and Weber (1993). 11. This function is often referred to in the literature as ‘‘next-turn repair initiation.’’ Extensive discussion can be found in Couper-Kuhlen (1992), Geluykens (1987, 1994), Jeﬀerson (1974), Schegloﬀ (1979), and Schegloﬀ et al. (1977). 12. Other sequences which can be understood in adjacency pair terms are greetings, or oﬀers and invitations and their acceptances or rejections. For further discussion, see also Clark (1992), chap. 5, Fox (1987), Goﬀman (1976), Pomerantz (1984), Sacks (1987, 1992) (especially Vol. II, Part VIII), Sacks et al. (1974), Schegloﬀ (1984, 1988), Schegloﬀ et al. (1977), and Tsui (1991). For additional support for this view of ‘‘conditional relevance,’’ though not phrased in terms of ‘‘adjacency pairs,’’ see Labov and Fanshel (1977), Moravcsik (1971), Sinclair and Coulthard (1975), Stenström (1984, 1988), and Stubbs (1983). 13. An independent analysis of a small corpus taken from the Corpus of Spoken American English yields nearly identical percentages, as shown in Table 2. Table 2. Types of negatives in Thompson’s conversational database. Denials Rejection Other* Total

Impl. Expl.

98 7 0 24 129

67 5 0 19 100

81

81

*Includes negative imperatives and interrogatives. 14. The idea that perhaps negative clauses might be first pair parts of some adjacency pair structure has also been speciﬁcally entertained by Givón (1995: 114), who claims that ‘‘challenge from the hearer is anticipated.’’ I have not found evidence to support this view either, and, unfortunately, Givón does not provide any. 15. For further discussion of the diachrony and areal nature of this interrogative type, see Clark (1985, 1989), Li (1992), Sanders (1991), Yue-Hashimoto (1993), and Zhu (1985, 1990), and references cited in these works.

differences in interrogation and negation

335

16. The V-not-V interrogative clearly also involves clause-combining syntax, which has been ritualized (Haiman 1991, 1994), grammaticized, and reduced, and in fact, there is some interesting discussion in the literature as to whether explicit disjunctions of aﬃrmative and negative clauses are interactionally distinct from non-disjunctive interrogatives within a given language. For at least two such languages, the answer, not surprisingly, seems to be ‘‘yes’’: Mandarin, where the disjunction has been grammaticized (Li and Thompson 1984), and English, where the disjunction has not been grammaticized to the same extent (Bolinger 1978). For extensive cross-linguistic discussion of ‘‘remnants’’ of disjunctive morphosyntax in interrogative grammar, see Moravcsik (1971). 17. Dryer gives the following list: Africa: ADAMAWA-UBANGIAN (Mumuye, Mbum, Gbaya Kaka, Gbeya Bossangoa, Sango, Nzakara), NUPOID (Nupe), PLATOID (Jukun, Birom), KAINJI (Duka), EASTERN JEBEL (Gaam), DAJU (Shatt), BONGO-BAGIRMI (Kresh*, Aja, Bongo, Jur), MÖDÖ (Ngambay, Mbaye, Bagirmi, Kenga, Kara), EAST CHADIC (Kera), BIU-MANDARA (Margi, Musgu, Mbara, Buduma, Logone), WEST CHADIC (Angas, Ngizim). SEAsiaandOceania:KAREN(EasternKayah),CENTRAL-EASTERNMALAYO-POLYNESIAN (Alune, Solor, Mor, Kaliai-Kove, Yabem, Patep, Loniu). * For a discussion of negation in Kresh, see Brown (1993); interestingly, and in my favor, he notes that although Kresh has a clause-ﬁnal negative particle, it also has ‘‘a particle bãá which may optionally be used clause-initially or pre-verbally to presage the negation that will follow’’ (p. 165). (i) Bãá Kôkó ãmbá Gõkó ’dı˜ neg Koko he-hit Goko not ‘Koko didn’t hit Goko’ Signiﬁcantly, however, there are at least three languages outside of these areas for which ‘‘clauseﬁnal’’ negative markers have been described: French (Ashby 1981), Brazilian Portuguese (Cunha to appear), and the (possibly) creolized language of northern Colombia, Palenquero, with the same pattern (thanks to Loretta O’Connor; cf. de Friedemann and Posselli 1983). 18. An analogous prediction for verb-initial languages is not as likely to be conﬁrmed, since there is a tendency, as noted in Section 2, for interrogative markers to occur in utterance-second position in any case. Thus an interrogative suff ix in a verb-initial language is not the anomaly that an interrogative pref ix would be in a verb-ﬁnal language. Indeed, in Dryer’s database, 4 verb-initial genera show interrogative suﬃxes and 3 show interrogative preﬁxes. In any case, as Dryer suggests (p.c.), in line with my prediction, we can safely say that the ratio of interrogative suﬃx to preﬁx in verbﬁnal languages is much higher than it is in verb-initial languages. 19. Matthew Dryer (p.c.), agreeing with the Q-N hypothesis that negative markers tend to be more ‘‘clause-bound’’ than interrogative markers, points out that his data suggest that they are, however, less ‘‘clause-bound’’ than tense-aspect markers. This is also inferrable from Bybee (1985) and Bybee et al. (1994). 20. But see McCawley (1991) for a critique of this view.

336

sandra a. thompson

References Altenberg, Bengt. 1987. Prosodic Patterns in Spoken English: Studies in the correlation between prosody and grammar for text-to-speech conversion. Lund Studies in English 76. Lund: Lund University Press. Ashby, William. 1981. ‘‘The Loss of the Negative Particle ‘ne’ in French: A syntactic change in progress.’’ Language 57: 674–87. Athanasiadou, Angeliki. 1991. ‘‘The Discourse Function of Questions.’’ Pragmatics 1.1: 107–22. Auer, Peter and Aldo di Luzio (eds). 1992. The Contextualization of Language. Amsterdam: John Benjamins. Bernini, Giuliano and Paolo Ramat (eds). 1996. Negative Sentences in the Languages of Europe: A typological approach. Berlin: Mouton de Gruyter. Bolinger, Dwight. 1978. ‘‘Yes-No Questions are not Alternative Questions.’’ In Henry Hiz˙ (ed.), Questions, 87–105. Dordrecht: D. Reidel. Bossuyt, Alain. 1983. ‘‘Historical Functional Grammar: An outline of an integrated theory of language change.’’ In Simon Dik (ed.), Advances in Functional Grammar, 301–25. Dordrecht: Foris. Brown, D. Richard. 1993. ‘‘Kresh.’’ In Peter Kahrel and René van den Berg (eds), 163–89. Bybee, Joan. 1985. Morphology. Amsterdam: John Benjamins. ——, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar. Chicago: University of Chicago Press. Byrd, Dani. 1992. ‘‘Pitch and Duration of Yes-No Questions in Nchuﬁe.’’ Journal of the International Phonetic Association 22: 12–26. Chafe, Wallace. 1987. ‘‘Cognitive Constraints on Information Flow.’’ In Russell Tomlin (ed.), Coherence and Grounding in Discourse. Amsterdam: John Benjamins, 21–51. ——. 1988. ‘‘Linking Intonation Units in Spoken English.’’ In John Haiman and Sandra A. Thompson (ed.), Clause combining in grammar and discourse. Amsterdam: John Benjamins, 1–27. ——. 1993. ‘‘Prosodic and Functional Units of Language.’’ In Edwards and Lampert (eds), 33–43. ——. 1994. Discourse, Consciousness, and Time: The ﬂow and displacement of conscious experience in speaking and writing. Chicago: University of Chicago Press. ——. 1995. ‘‘The Realis–Irrealis Distinction in Caddo, the Northern Iroquoian Languages, and English.’’ In Joan Bybee and Suzanne Fleischmann (eds), Modality and Grammar in Discourse. Amsterdam: John Benjamins, 349–65. Chisholm, Willaim S. 1984. Interrogativity. Amsterdam: John Benjamins. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Clancy, Patricia M. 1982. ‘‘Written and Spoken Style in Japanese Narratives.’’ In Deborah Tannen (ed.), Spoken and written language: exploring orality and literacy. Norwood, NJ: Ablex, 55–76. Clark, Herbert H. 1992. Arenas of Language Use. Chicago: University of Chicago Press.

differences in interrogation and negation

337

Clark, Marybeth. 1985. ‘‘Asking Questions in Hmong and other Southeast Asian languages.’’ Linguistics of the Tibeto-Burman Area 8.2: 60–7. ——. 1989. ‘‘Hmong and Areal South-East Asia.’’ In David Bradley (ed.), Papers in Southeast Asian linguistics 11: 175–214. Cook, Haruko Minegishi. 1992. ‘‘Meanings of Non-Referential Indexes: A case study of the Japanese sentence-ﬁnal particle ne.’’ Text 12.4: 507–39. Couper-Kuhlen, Elizabeth. 1992. ‘‘Contextualizing Discourse: The prosody of interactive repair.’’ In Peter Auer and Aldo di Luzio (eds), The Contextualization of Language. Amsterdam: John Benjamins, 337–364. ——. 1993. English Speech Rhythm. Amsterdam: John Benjamins. Cunha, Maria Angélica Furtado da Cunha. To appear. Strategies of Negation in Written and Spoken Brazilian Portuguese. Dahl, Östen. 1979. ‘‘Typology of Sentence Negation.’’ Linguistics 17: 79–106. Dik, Simon. 1989. The Theory of Functional Grammar. Dordrecht: Foris. Drozd, Kenneth F. 1995. ‘‘Child English Presentational Negation as Metalinguistic Exclamatory Negation.’’ Journal of Child Language 22.3: 583–610. Dryer, Matthew. 1989. ‘‘Universals of Negative Position.’’ In E. Moravcsik, J. Wirth, and M. Hammond (eds), Studies in Syntactic Typology. Amsterdam: John Benjamins, 93–124. Du Bois, John W. 1985. ‘‘Competing Motivations.’’ In John Haiman (ed.), Iconicity in Syntax. Amsterdam: John Benjamins, 343–65. ——. 1987. ‘‘The Discourse Basis of Ergativity.’’ Language 63.4: 805-855. ——. 1991. ‘‘Transcription Design Principles for Spoken Discourse Research.’’ Pragmatics 1: 71–106. ——, Stephan Schuetze-Coburn, Danae Paolino, and Susanna Cumming. 1993. ‘‘Outline of Discourse Transcription.’’ In Edwards and Lampert (eds), 45–89. Edwards, Jane A. and Martin D. Lampert (eds). 1993. Talking Data: Transcription and coding methods for language research. Hillsdale, NJ: Lawrence Erlbaum. Foley, William and Robert Van Valin. 1984. Functional syntax and Universal Grammar. Cambridge: Cambridge University Press. Ford, Cecilia. 1993. Grammar in Interaction: Adverbial clauses in American English conversations. Cambridge: Cambridge University Press. —— and Sandra A. Thompson. 1996. ‘‘Interactional Units in Conversation: Syntactic, intonational, and pragmatic resources for the projection of turn completion.’’ In Elinor Ochs, Emanuel Schegloﬀ and Sandra A. Thompson (eds). Grammar and Interaction. Cambridge: Cambridge University Press, 134–84. Fox, Barbara A. 1987. Discourse Structure and Anaphora. Cambridge: Cambridge University Press. Freed, Alice F. 1994. ‘‘The Form and Function of Questions in Informal Dyadic Conversation.’’ Journal of Pragmatics 21: 621–44. de Friedemann, Nina S. and Carlos Patino Rosselli. 1983. Lengua y Sociedad en el Panenque de San Basilio. Bogotá: Publicaciones del Instituto Caro y Cuervo.

338

sandra a. thompson

Geluykens, Ronald. 1987. ‘‘Tails (Right-Dislocations) as a Repair Mechanism in English Conversation.’’ In J. Nuyts and G. De Schutter (eds), Getting one’s words into line: On word order and functional grammar. Dordrecht: Foris, 119-129. ——. 1994. The Pragmatics of Discourse Anaphora in English: Evidence from conversational repair. Berlin: Mouton de Gruyter. Givón, Talmy. 1979. On Understanding Grammar. New York: Academic Press. ——. ‘‘Logic vs. Pragmatics, with Human Language as the Referee.’’ Journal of Pragmatics 6.1: 81–113. ——. 1984. Syntax. Amsterdam: John Benjamins. ——. 1989. Mind, Code and Context: Essays in pragmatics. Hillsdale, NJ: Lawrence Erlbaum. ——. 1995. Functionalism and Grammar. Amsterdam: John Benjamins. Goﬀman, Erving. 1976. ‘‘Replies and Responses.’’ Language in Society 5: 257–313. Reprinted in Erving Goﬀman, Forms of talk, 1981, 5–77, University of Pennsylvania Press. Goodwin, Charles. 1981. Conversational Organization. New York: Academic Press. Haiman, John. 1991. ‘‘Motivation, Repetition, and Emancipation: The bureaucratisation of language.’’ In H.C. Wolfart (ed.), Linguistic studies presented to John L. Finlay. Algonquian and Iroquoian Linguistics, Memoir 8, 45–69. ——. 1994. ‘‘Ritualization and the Development Of language.’’ In William Pagliuca (ed.), Perspectives on grammaticalization. Amsterdam: John Benjamins, 3–28. Heritage, John C. and Andrew L. Roth. 1995. ‘‘Grammar and Institution: questions and questioning in the broadcast news interview.’’ Research on language and social interaction 28.1: 1–60. Hinds, John. 1984. ‘‘Japanese.’’ In Chisholm (ed.), 145–88. Hiz˙, Henry. 1978. ‘‘Diﬃcult Questions.’’ In Henry Hiz˙ (ed.), Questions, 211–26. Dordrecht: D. Reidel. Hopper, Paul J. 1987. ‘‘Emergent Grammar.’’ Proceedings of the thirteenth annual meeting, Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 139-157. ——. 1988. ‘‘Emergent grammar and the a priori grammar postulate.’’ In Deborah Tannen (ed.), Linguistics in Context. Norwood, NJ: Ablex, 117–34. —— and Sandra A. Thompson. 1993. ‘‘Language Universals, Discourse Pragmatics, and Semantics.’’ Language Sciences 15: 357–76. Horn, Laurence R. 1989. A Natural History of Negation. Chicago: University of Chicago Press. Jeﬀerson, Gail. 1974. ‘‘Error Correction as an Interactional Resource.’’ Language in Society 2: 188–99. Jespersen, Otto. 1924. The Philosophy of Grammar. New York: Holt and Co. Kahrel, Peter and René van den Berg (eds). 1993. Typological Studies in Negation. Amsterdam: John Benjamins. Labov, William and David Fanshel. 1977. Therapeutic Discourse. New York: Academic Press. Li, Charles N. and Sandra A. Thompson. 1984. ‘‘Mandarin.’’ In Chisholm (ed.), 47–61.

differences in interrogation and negation

339

Li, Ying-che. 1992. ‘‘Aspects of Comparative Syntax Between Mandarin and Taiwanese: Use of negatives in questions.’’ Chinese languages and linguistics I: Chinese dialects. Symposium series of the institute of history and philology, Academia Sinica, Number 2. Taipei: Academia Sinica, 437–48. Maynard, Senko. 1989. Japanese Conversation. Norwood, NJ: Ablex. McCawley, James. 1991. ‘‘Contrastive Negation and Metalinguistic Negation.’’ CLS 27 Parasession on negation. Chicago: Chicago Linguistic Society, 189–206. McHoul, A. W. 1987. ‘‘Why There are no Guarantees for Interrogators,’’ Journal of Pragmatics 11: 455–71. Moravcsik, Edith. ‘‘Some Cross-Linguistic Generalizations about Yes-No Questions and Their Answers.’’ Working Papers on Language Universals 7. Stanford: Stanford University, 45–193. Nedyalkov, Igor. 1993. ‘‘Evenki.’’ In Peter Kahrel and René van den Berg (eds.), 1–34. Noonan, Michael. 1985. ‘‘Complementation.’’ In Timothy Shopen (ed.), Language typology and syntactic description, vol. 2. Cambridge: Cambridge University Press, 42–140. Ochs, Elinor. 1976. ‘‘The Universality of Conversational Postulates.’’ Language in Society 5: 67–80. ——, Emanuel Schegloﬀ, and Sandra A. Thompson (eds). 1996. Interaction and Grammar. Cambridge: Cambridge University Press. Oreström, Bengt. 1983. Turn-taking in English Conversation. Lund: CWK Gleerup [Lund Studies in English 66]. Payne, John R. 1985. ‘‘Negation.’’ In Timothy Shopen (ed.), Language Typology and Syntactic Description, Vol. 2, 197–242. Cambridge: Cambridge University Press. Pomerantz, Anita. 1984. ‘‘Agreeing and Disagreeing with Assessments: Some features of preferred-dispreferred turn shapes.’’ In J. Maxwell Atkinson and John Heritage (eds), Structures of Social Action. Cambridge: Cambridge University Press, 57–101. Sacks, Harvey. 1974. ‘‘An Analysis of the Course of a Joke’s Telling in Conversation.’’ In R. Bauman and J. Sherzer (eds), Explorations in the Ethnography of Speaking. Cambridge: Cambridge University Press, 337–53. ——. 1987. ‘‘On the Preference for Agreement and Contiguity in Sequences in Conversation.’’ In Graham Button and John R.E. Lee (eds), Talk and Social Organization, 54–69. Clevedon, Avon: Multilingual Matters. ——. 1992. Lectures on Conversation. Edited by Gail Jeﬀerson. Oxford: Blackwell. ——, Emanuel Schegloﬀ, and Gail Jeﬀerson. 1974. ‘‘A Simplest Systematics for the Organization of Turn-Taking for Conversation.’’ Language 50.4: 696-735. Sadock, Jerrold M. 1984. ‘‘West Greenlandic.’’ In Chisholm, 189–214. Sanders, Robert. 1991. ‘‘Yes-No Questions in Taipei and Peking Mandarin.’’ Sino-Platonic Papers 27 [Festschrift for John DeFrancis]: 147–56. Schegloﬀ, Emanuel. 1968. ‘‘Sequencing in Conversational Openings.’’ American Anthropologist 70.6: 1075–95. ——. 1979. ‘‘The Relevance of Repair to Syntax-for-Conversation.’’ In Talmy Givón (ed.), Discourse and syntax, 261-286. New York: Academic Press.

340

sandra a. thompson

Schegloﬀ, Emanuel. 1984. ‘‘On Some Questions and Ambiguities in Conversation.’’ In J.Maxwell Atkinson and John Heritage (eds.), Structures of Social Action, 28–52. Cambridge: Cambridge University Press. ——. 1988. ‘‘Goﬀman and the Analysis of Conversation.’’ In Paul Drew and Anthony Wootton (eds.), Erving Goffman: exploring the interaction order. Cambridge: Polity Press, 91–135. ——. 1989. ‘‘Reﬂections on Language, Development, and the Interactional Character of Talk-in-Interaction.’’ In Marc H. Bornstein and Jerome S. Bruner (eds.), Interaction in Human Development. Hillsdale, NJ: Lawrence Erlbaum, 139–53. ——. 1992. ‘‘To Searle on Conversation.’’ In Searle, John R. et al., (On) Searle on Conversation. Amsterdam: John Benjamins, 113–28. —— and Harvey Sacks. 1973. ‘‘Opening Up Closings.’’ Semiotica 7: 289–327. Reprinted in John Baugh and Joel Sherzer (eds), 1984, Language in Use: Readings in sociolinguistics. Englewood Cliﬀs, NJ: Prentice-Hall, 69–98. ——, Gail Jeﬀerson, and Harvey Sacks. 1977. ‘‘The Preference for Self-Correction in the Organization of Repair in Conversation.’’ Language 53: 361-382. Schuetze-Coburn, Stephan. 1992. ‘‘Prosodic Phrase as a Prototype.’’ Proceedings of the IRCS Workshop on Prosody in Natural Speech, University of Pennsylvania, August, 1992. (Institute for Cognitive Research Report, 92–37.) ——. To appear. Prosody, Grammar, and Discourse Pragmatics: Organizational principles of information ﬂow in German conversational narratives. Amsterdam: John Benjamins. ——, Marian Shapley, and Elizabeth G. Weber. 1991. ‘‘Units of Intonation in Discourse: A comparison of acoustic and auditory analyses.’’ Language and Speech 34.3: 207–34. Schuh, Russell. 1972. ‘‘Aspects of Ngizim Syntax.’’ Unpublished PhD dissertation, UCLA. Selting, Margret. 1992. ‘‘Prosody in Conversational Questions.’’ Journal of Pragmatics 17: 315–45. Sinclair, John M. and Malcolm Coulthard. 1975. Towards an Analysis of Discourse. London: Oxford University Press. Squires, Todd. 1994. ‘‘A Discourse Analysis of the Japanese Particle sa.’’ Pragmatics 4.1: 1–29. Stenström, Anna-Brita. 1984. Questions and Responses in English Conversation. Lund: Lund University Press. ——. 1988. ‘‘Questioning in Conversation.’’ In M. Meyer (ed.), Questions and Questioning. Berlin: Walter de Gruyter, 304–25. Stubbs, Michael. 1983. Discourse Analysis: The sociolinguistic analysis of natural language. Chicago: University of Chicago Press. Svartvik, Jan and Randolph Quirk (eds). 1980. A Corpus of English Conversation. Lund: CWK Gleerup [Lund Studies in English 56]. Tao, Hongyin. 1996. Units in Mandarin Conversation: Prosody, discourse, and grammar. Amsterdam: John Benjamins. Tottie, Gunnel. 1991. Negation in English Speech and Writing. New York: Academic Press. Tsui, Amy B.M. 1991. ‘‘Sequencing Rules and Coherence in Discourse.’’ Journal of Pragmatics 15: 111–29.

differences in interrogation and negation

341

Tsui, Amy B.M. 1992. ‘‘A Functional Description of Questions.’’ In M. Coulthard (ed.), Advances in Spoken Discourse Analysis. London: Routledge, 89–110. Ultan, Russell. 1978. ‘‘Some General Characteristics of Interrogative Systems.’’ In Joseph Greenberg (ed.), Universals of Human Language. Stanford: Stanford University Press, 211–48. van Schaaik, Gerjan. 1993. ‘‘Turkish.’’ In Peter Kahrel and René van den Berg (eds), 35–50. Walsh, Michael. 1994. ‘‘Interactional styles in the courtroom: an example from northern Australia.’’ In John Gibbons (ed.), Language and the Law. London: Longman, 217–33. Weber, Elizabeth G. 1993. Varieties of Questions in English Conversation. Amsterdam: John Benjamins. Wierzbicka, Anna. 1985. ‘‘Diﬀerent Cultures, Diﬀerent Languages, Diﬀerent Speech Acts.’’ Journal of Pragmatics 9: 145–78. Yeh, Ling-hsia. 1995. ‘‘Focus, Metalinguistic Negation, and Constrastive Negation.’’ Journal of Chinese Linguistics 23.2: 42–75. Yue-Hashimoto, Anne. 1993. ‘‘The Lexicon in Syntactic Change: Lexical diﬀusion in Chinese syntax.’’ Journal of Chinese Linguistics 21.2: 213–54. Zhu, Dexi. 1985. ‘‘Hanyu Fangyanli de Liangzhong Fanfu Wenju [The Dialectal Distribution of Two Types of V-Not-V Interrogative Sentence Patterns in Chinese].’’ Zhongguo Yuwen [Chinese linguistics] 184: 10–20. ——. 1990. ‘‘A Preliminary Survey of the Dialectal Distribution of the Interrogative Sentence Patterns V-Bu-VO and VO-Bu-V in Chinese.’’ Journal of Chinese Linguistics 18.2: 209–30.

Applicative constructions in Warrungu of Australia TASAKU TSUNODA University of Tokyo

1. Introduction1 1.1 General notes Warrungu (also spelt Warungu) is an Australian Aboriginal language once spoken in the upper Herbert River area of north Queensland. It became virtually extinct in the early 80’s. Most of the data on it was collected by me on Palm Island, from 1971 to 1974, from the late Alf Palmer (Warrungu name: Jinpilngkay). Furthermore, a small amount of data was obtained from a few other speakers by R.M.W. Dixon, by Peter Sutton and by me. The present work is based on a ﬁnite corpus (which includes six hours of running texts by Alf Palmer), and it contains many points that can no longer be answered. My works on Warrungu include Tsunoda (1974, 1976a, 1976b, 1988, 1990, 1992, forthcoming-a, -b). 1.2 Typological sketch The Warrungu phoneme inventory, written in a practical orthography, is as follows: /p, t, j, k, m, n, ny, ng, rr, r, l, w, y, a, i, u, aa/. Warrungu has the following cases: ergative, nominative, accusative, dative, genitive, locative, ablative, nominal comitative, and nominal instrumental. The case system exhibits a kind of split ergativity. With pronouns, the ergative and the nominative are identical, in distinction from the accusative. With nouns, the nominative and the accusative share the same form, as opposed to the ergative. This is, however, an oversimpliﬁcation. For instance, the ‘3du’ and ‘3pl’ can optionally take the ergative suﬃx, cf. jana-ngku ‘3pl-erg’ in (25). ‘‘Adjectives’’ are diﬃcult to distinguish from nouns; they are really a type of noun.

tasaku tsunoda

344

Table 1. Case-marking patterns

ergative nominative accusative

pronouns

nouns

‘1sg’

‘3pl’

‘man’

ngaya ngaya nganya

jana-Ø, jana-ngku jana-Ø jana-nya

pama-ngku pama-Ø pama-Ø

There are three conjugational classes: L-class, Y-class and Ø-class. The diﬀerences among them are clearly seen in certain non-future forms: palka-l ‘hitnonfut’, nyina-y ‘sit-nonfut’ and watali-Ø ‘run-nonfut’. I shall indicate the class membership of verb roots and derivational suﬃxes wherever information is available. An uncertain class membership is indicated by a question mark. Word order in Warrungu is not rigid (Tsunoda 1990). Warrungu discourse is highly elliptical. 1.3 Outline of the paper This paper describes the morphology, syntax, semantics and pragmatics of applicative constructions in Warrungu, and also provides brief comparative notes on them. Applicative verbs based on intransitive roots become transitive. They have a locative-like or a comitative-like meaning, hence ‘‘comitative verbs.’’ Those derived from transitive roots remain transitive. They generally have an instrumental-like meaning, hence ‘‘instrumental verbs.’’ Instrumental verbs exhibit unusual syntactic possibilities. Both comitative and instrumental verbs appear to lack a benefactivelike meaning, like applicatives in some other Australian languages but unlike certain applicatives reported from outside Australia.

2. Applicative verbs Warrungu has the verbal derivational suﬃx -ri-L ‘applicative’ (app), which isgenerally added to (intransitive or transitive) verb roots. (There are just a couple exceptions, in which a verb stem or a noun is involved; cf. 2. and 3. in Section 3.1.) The resultant stems are transitive. They belong to the L-class. (Virtually all the transitive roots and stems in Warrungu belong to the L-class.) Virtually all the verb roots are disyllabic, and all the verb roots are vowel-ﬁnal. Likewise, in all the applicative verbs, the elements to which -ri-L is added are disyllabic and vowel-ﬁnal; there are certain complications, however; cf. 2. and 3. in Section 3.1.

applicative constructions in warrungu

345

Suﬃxation of -ri-L ‘app’ to intransitive roots produces comitative verbs. Their valence increases by one. Suﬃxation of -ri-L ‘app’ to transitive roots creates instrumental verbs. Here, it is not clear whether or not their valence increases. It seems that -ri-L ‘app’ cannot be added to a derivational suﬃx. In this respect, it resembles derivational suﬃxes such as -wa-Y ‘reciprocal’ and diﬀers from those such as -karra-Y ‘iterative’ and -kali-Ø ‘reﬂexive, antipassive’. It also seems that -ri-L ‘app’ cannot be followed by a derivational suﬃx. This paper ﬁrst looks at the comitative use (‘‘V.com’’), and then the instrumental use (‘‘V.inst’’), followed by comparative notes. It will be shown that applicative constructions are highly marked as against corresponding clauses in terms of verb morphology, semantics, syntax, pragmatics and also frequency. For applicatives in Australian languages, see Blake (1987: 69–76) and Austin (1997).

3. Comitative verbs: Vi-ri-L 3.1 Formation The suﬃx -ri-L ‘V.com’ is generally added to intransitive roots. It also appears to be added to one intransitive stem and one noun. A full list of the forms taking the V-form suﬃx is given below, together with the number of instances of each found in the textual corpus. -ri-L ‘V.com’ is not productive. It is attested with only 15 verb roots/stems and one noun. 1. Intransitive roots. (a) Y-class: nyina-Y ‘sit (down)’ (19 examples), nyampa-Y ‘dance’ (10 examples), wuna-Y ‘lie (down), sleep’ (7 examples), jana-Y ‘stand (up)’ (2 examples), janta-Y(?) ‘wade’. (b) L-class: panta-L ‘come out’, jarka-L(?)‘go in’, wata-L(?) ‘play’, waka-L ‘get up’, wanti-L(?) ‘rest’, wara-L ‘jump’. (c) Ø-class: yuti-Ø(?),-L (?) ‘swim’. (d) other: yani-L/Y ‘go, come, walk’ (2 examples). 2. Intransitive stem: watali-Ø ‘run’. The comitative verb wata-ri-L ‘run with (shoes on)’ seems to be derived from the stem watali-Ø ‘run’, with the concomitant deletion of -li-Ø. (Etymologically, this stem probably contains the root wata-L(?) ‘play’.) 3. Noun: makul ‘work’. The comitative verb maku-ri-L ‘work together with (someone)’, appears to be formed by the addition of -ri-L ‘V.com’ to the noun makul ‘work’, with the concomitant deletion of the root-ﬁnal l. (It seems that -ri-L can not be added to a consonant.) There is no such verb as maku- ‘work’. However, there is a derived intransitive verb makul-i-Ø ‘work’; -i-Ø is an in-

tasaku tsunoda

346

transitive-stem-forming suﬃx. Therefore, maku-ri-L ‘work with’ may also be said to be derived from the stem makul-i-Ø ‘work’, accompanied by the deletion of -l-i-Ø. (Apparently -ri-L cannot be added to a derivational suﬃx (cf. Section 2), and consequently it cannot be added to makul-i-Ø ‘work’, which contains a derivational suﬃx, i.e. -i-Ø. See also the comment on the comitative verb wata-ri-L ‘run with (shoes on)’ in 2. above.) Comitative verbs are attested with the following inﬂectional suﬃxes: -n ‘nonfuture’, -l ‘non-future’, -lku/-yal ‘purposive’, -ya ‘imperative’, -nyu ‘non-ﬁnite’. They are not attested with other inﬂectional suﬃxes, such as -lka ‘apprehensional’ or -nji ‘non-ﬁnite’. Thus, comitative verbs seem to have a somewhat defective paradigm. There is one example with the zero suﬃx, possibly a third non-future form, viz. wuna-ri-Ø ‘lie-V.com-nonfut(?)’. There is also one instance of a comitative verb followed by the ablative suﬃx -ngumay, viz. wuna-ri-ngumay ‘lie-V.comabl’. Unfortunately, there is no sentential example of either of them. 3.2 Meaning Comitative verbs can often be translated involving together with (hence, the name ‘‘comitative verbs,’’ cf. Dixon 1972: 96), but they may have some other meanings, such as ‘on (top of)’, ‘in’ and ‘into’. The attested comitative verbs have the following meanings. (In the following list, the gloss ‘V.com’ is omitted.) (i) nyina-ri-L ‘sit/stay-’: ‘sit/stay with (someone)’, cf. (1c), ‘sleep with (a man, a woman)’, ‘mind, look after (someone)’, ‘be married to (a man, a woman)’, cf. (5), ‘sit on (a saddle, a swag, a groundsheet)’, ‘sit/stay in (a camp, a house, a shade)’, cf. (2b), (10), (21), (22), ‘sit/be with (clothes on)’, cf. (3b). (ii) nyampa-ri-L ‘dance-’: ‘dance with (someone)’, ‘dance with (ornaments)’, ‘dance to (a song, a beating rhythm)’, cf. (15). (iii) wuna-ri-L ‘lie (down), sleep-’: ‘sleep with (a man, a woman)’, cf. (9), (13), (17), (56), ‘sleep on (a blanket)’, ‘sleep in (a camp, a house)’. (iv) jana-ri-L ‘stand (up)-’: ‘stand by (a ﬁre)’, ‘stand on (a road)’, ‘stand with (one’s own feet)’, ‘stand near (someone)’, ‘look after, mind, keep an eye on (someone)’, cf. (6), ‘look after, keep (cattle, goats)’, ‘hold in arms and soothe (a crying baby)’. (v) panta-ri-L ‘come out-’: ‘come out with (someone ?)’. (vi) janta-ri-L ‘wade-’: ‘carry (a child on the shoulder) and wade’. (vii) jarka-ri-L ‘go in-’: ‘go into (a house)’. (viii) waka-ri-L ‘get up-’: ‘get up with (something in a hand)’. (ix) wata-ri-L ‘play-’: ‘play in (a playground)’. (x) wata-ri-L ‘run-’: ‘run with (shoes on)’.

applicative constructions in warrungu

347

(xi) wanti-ri-L ‘rest (on something)-’: ‘lie on top of (a woman)’. (xii) wara-ri-L ‘jump-’: ‘jump with (something in a hand)’. (xiii) yuti-ri-L ‘swim-’: ‘swim with (someone)’, cf. (12), ‘swim with (clothes on)’, ‘carry (something) and swim’. (xiv) yani-ri-L ‘go, come, walk-’: ‘go, walk with (someone)’, ‘carry (a swag)’, cf. (4c), ‘take away, steal (a swag)’, cf. (4b), ‘go into (a scrub)’, ‘walk on (a road)’. (xv) maku-ri-L ‘work-’: ‘work together with (a co-worker)’, cf. (14). In all the examples of comitative verbs, the A NP is human.2 (In non-derived intransitive clauses a non-human S NP (e.g. ‘‘a dog/log is lying there’’) is possible.) In virtually all the attested comitative verbs, the roots and the derived comitative stems seem to denote volitional activities. Perhaps the only exceptions are the root wuna-Y in the sense of ‘sleep’ rather than ‘lie (down)’ and the comitative stem wuna-ri-L ‘sleep-V.com-’, ‘sleep with (a man, woman), sleep on (a blanket), sleep in (a house, camp)’. (Note, however, the act of sleeping with a man/woman may be a volitional one, M. Shibatani p.c.)3 Comitative verbs such as the following are not attested: wula-ri-L ‘die with (someone), die in (a place)’ and kalngka-ri-L ‘fall down with (someone, something)’. It may be signiﬁcant that wula-Y ‘die’ and kalngka-Y ‘fall’ appear to describe non-volitional acts. In certain instances the addition of -ri-L produces causative verbs, rather than applicative verbs. Speciﬁcally: (a) intransitive roots: (i) jana-Y ‘(someone, something) stand’ → jana-ri-L ‘stand (something) up’, (ii) waka-L ‘get up’ → waka-ri-L ‘make (a penis) big(?)’, (iii) wanpa-L ‘be afraid’ → wanpa-ri-L ‘frighten’, (iv) panta-L ‘crack (of noise)’ → panta-ri-L ‘make (something) crack (of noise)’, and possibly (v) janta-ri-L ‘make (someone) wade’. (b) an ‘‘adjective’’: kakal ‘big’ → kaka-ri-L ‘grow (something) big’. The formation of kaka-ri-L ‘grow (something) big’ involves deletion of the ﬁnal l of kakal; cf. the formation of maku-ri-L ‘work with’ (Section 3.1). The following verbs may be either causative or applicative: (i) jana-ri-L ‘stand (something) up’ and ‘stand by/near (someone, something)’, (ii) waka-ri-L ‘make (a penis) big(?)’ and ‘get up with (something)’, and possibly (iii) janta-ri-L ‘make (someone) wade (?)’ and ‘wade with (someone)’. That is, with these verbs, suﬃxation of -ri-L may produce either a causative verb or an applicative verb. Austin (1997) seems to propose a strict dichotomy between causativization and applicativization for Australian languages. However, this does not apply to the Warrungu data. (This observation is largely due to William McGregor, p.c.)

tasaku tsunoda

348 3.3 Case frames

V.com constructions are transitive; they take the erg-acc case frame (cf. Section 3.4). They can be classiﬁed as in Table 2, in terms of the case frame of corresponding clauses (hereafter, ‘‘basic clauses’’). (The nominal comitative case can mean ‘with, having’, with reference to humans, animates and inanimates (see Tsunoda 1976a). The locative case can have meanings such as ‘(sit, etc.) with (someone)’, ‘(jump) onto (someone)’, ‘(sit) in (a place)’, ‘(go) into (a place)’, ‘(sit) on (top of something)’.) Table 2. Verbal comitative constructions and basic intransitive clauses V.com constructions erg acc Vi-ri-L basic intransitive clauses (a) nom loc or N.com Vi (b) nom loc Vi (c) nom N.com Vi (d) nom ? Vi To use the terminology of Relational Grammar, in type (a), either an N.com NP or a loc NP is ‘‘promoted’’ to the acc position. Similarly for (b) and (c). For (d), it is diﬃcult to set up a basic clause, and I cannot suggest any NP which would be promoted to the acc position4 Examples of each type follow. (a) ‘‘nom-loc or nom-N.com.’’ There are very few examples of this type. As a set of examples, compare: (1) a. yarru-n-ta yinta kaya-ngka nyina-ya this-link-loc 2sg.nom father-loc sit-imp ‘Sit/stay with [your] father in this [place, i.e. here]. b. yinta nyina-n warrngu-yi-Ø 2sg.nom sit-nonfut woman-N.com-nom ‘You are sitting with a woman.5 c. rayi-Ø nyula nyina-ri-n girl-acc 3sg.erg sit-V.com-nonfut ‘He is sitting with a girl.’ Another example is nyampa-ri-L when it means ‘dance to (a song)’. See (15) and (16). (b) ‘‘nom-loc.’’ Many instances of V.com constructions correspond to nom-loc

applicative constructions in warrungu

349

clauses only, and not nom-N.com clauses. Thus, consider the following two successive sentences from the running texts. (2) a. ngana-Ø nyina-n nguna-n-ta=wa wuma-ngka 1pl-nom sit-nonfut that-link-loc=clc shade-loc ‘We sat in that shade.’ b. [?ngana-Ø TT] jarripara-Ø wuma-Ø nyina-ri-n [1pl-erg] good-acc shade-acc sit-V.com-nonfut ‘We sat in a good shade.’ (For the question mark in (2b), see the third paragraph below (14) in Section 3.5.4.) If (2b) were taken to correspond to a nom-N.com clause, in (2a) we would have wuma-yi-Ø ‘shade-N.com-nom’ in place of wuma-ngka ‘shade-loc’. The resultant sentence would be translated ‘We sat with a shade’, but I am not certain if this sentence is acceptable. (c) ‘‘nom-N.com.’’ Some instances of V.com constructions seem to correspond to nom-N.com clauses only, and not nom-loc clauses. There are very few examples of this type. (This is despite the widely accepted use of the label ‘‘comitative’’ for such verbs/constructions in Australian languages.) As examples, compare: (3) a. TT yarru-yi-Ø kampi-yi-Ø ngaya nyina-n this-N.com-nom clothes-N.com-nom 1sg.nom sit,stay-nonfut ‘I am sitting with these clothes on.’ b. (‘‘I am cold and shivering.’’) yarru-Ø=kul kampi-Ø ngaya nyina-ri-n this-acc=only clothes-acc 1sg.erg sit-V.com-nonfut ‘I am sitting with only these clothes on’, i.e. ‘I have only these clothes on.’ As additional examples, compare: (4) a. nyula yani-Ø kungkarri-ngal kulkurra-yi-Ø 3sg.nom go-nonfut north-to swag-N.com-nom ‘He went north with [his] swag.’ b. pama-ngku yani-ri-n kulkurra-Ø man-erg go-V.com-nonfut swag-acc ‘The man walked away with/took away/stole a swag.’ c. pama-ngku yani-ri-n nyungu kulkurra-Ø man-erg go-V.com-nonfut 3sg.gen swag-acc ‘The man went with/carried his swag.’ Literally, (4b) would be expected to mean ‘The man went with/carried a swag’. But

350

tasaku tsunoda

Alf Palmer’s glosses for it are ‘walk away with it’, ‘take away’ and ‘steal’. He stated that in order to say that the man carried his swag, and not stole it, the genitive nyungu has to be included, as in (4c). That is, there can be a semantic discrepancy between the basic clause ‘‘nom-N.com yani-L’’ and the V.com construction ‘‘erg-acc yani-ri-L.’’ (d) ‘‘nom-?.’’ For certain instances, it is diﬃcult to set up an NP corresponding to the acc NP of a V.com construction. The meaning of the comitative verb concerned is not ‘‘regular’’ and is diﬀerent from the expected ‘‘literal’’ meaning. Such comitative verbs include yani-ri-L ‘go-’ for ‘steal’, e.g. (4b); nyina-ri-L ‘sit-’ expressing ‘look after’, ‘be married to’, e.g. (5), and; jana-ri-L ‘stand-’ meaning ‘look after, mind, keep an eye on’, e.g. (6). (5) nguna-Ø nyula kalpin-Ø nyina-ri-n ngayku that-acc 3sg.erg niece-acc sit-V.com-nonfut 1sg.gen ‘He is married to that niece of mine.’ (6) pirku-Ø ngayku yarru-n-ta jana-ri-ya . . . . wife-acc 1sg.gen this-link-loc stand-V.com-imp ‘Please watch/mind/keep an eye on my wife in this [place, i.e. here].’ Needless to say, all of those comitative verbs which have an ‘‘irregular’’ meaning can also have a ‘‘regular’’ meaning. See Section 3.2. 3.4 Transitivity As noted above, comitative verbs are transitive. This can be seen in terms of the two criteria discussed below. 1. Case marking of the A NP. Nouns have the ergative suﬃx (-ngku) distinct from their nominative suﬃx (zero), and also the ‘3du’ and ‘3pl’ can optionally take the ergative suﬃx (cf. Section 1.2). The transitive status of V.com constructions is clearly seen when the A NP has the ergative suﬃx, e.g. pama-ngku ‘man-erg’ in (4b, c). 2. Modifying verbs. Certain verbs may modify another verb. Those verbs with the transitive-stem-forming -nga-L may modify a transitive verb, e.g. (7), while those formed with the intransitive-stem-forming -pi-L may modify intransitive verbs, e.g. (8). (This method of ascertaining the transitivity of verbs is due to Dixon 1972: 150, 1977: 252.) (7) ngaya nyunya makan-nga-n mayka-n 1sg.erg 3sg.acc false-tr-nonfut tell-nonfut Lit. ‘I did falsely, told her’, i.e. ‘I told her a lie.’

applicative constructions in warrungu

351

(8) nyula makan-pi-n wula-n 3sg.nom false-intr-nonfut die-nonfut Lit. ‘He did falsely, died’, i.e. ‘He pretended to be dead.’ Now, comitative verbs may be modiﬁed by a verb with -nga-L ‘tr’, but not by a verb with -pi-L ‘intr’. In this respect as well, comitative verbs are transitive. Thus: (9) nyunku-tu-nga-n ngaya [nyunya TT] wuna-ri-n one-N.inst-tr-nonfut 1sg.erg [3sg.acc] sleep-V.com-nonfut ‘I slept with [her] once.’ (nyunkul ‘one’.) These two criteria are logically independent of each other, but they yield the same result and they never contradict each other. (This observation was triggered by Motoyasu Nojima’s comment.) In view of what has been stated in Section 3.3 and Section 3.4, it is safe to say that the valence of comitative verbs increases by one. 3.5 Use of V.com constructions 3.5.1 Frequency As noted in Section 3.1, V.com constructions are not productive. They are far less frequent than basic clauses (if there are any corresponding basic clauses at all). Thus, consider (2a) ‘‘nom loc nyina-Y ‘sit’ ’’ and (2b) ‘‘erg acc nyina-ri-L ‘sitV.com’ .’’ They have much the same meaning: ‘sit in (a place)’. In the six hours of running texts, there are only four instances of nyina-ri-L ‘sit-V.com’ with the meaning of ‘sit in (a place)’, as in (2b), while on the other hand there are as many as 167 instances—more than 40 times as many—of nyina-Y ‘sit’plus a loc NP with the meaning of ‘sit in (a place)’, as in (2a). Due to their very low frequency, it has proved very diﬃcult to pinpoint factors that may condition the use of V.com constructions, as against basic clauses. Nonetheless, a few conditioning factors have emerged which may be semantic, pragmatic and/or syntactic in nature. They are dealt with below. 3.5.2 Lexicalized comitative verbs As noted and exempliﬁed in (d) of Section 3.3, certain V.com constructions have a meaning that is diﬀerent from the expected ‘‘literal’’ meaning. They may be considered as lexicalized. Such V.com constructions seem to lack a basic clause. They would have to be used in place of a non-existent basic clause. Sometimes, though this is decidedly uncommon, there is a transitive verb that has a very similar meaning to that of a comitative verb. Thus:

352

tasaku tsunoda

(a) yani-ri-L ‘steal’ (lit. ‘go-V.com’), e.g. (4b), and yupaynga-L ‘steal (e.g. a woman)’. (b) yani-ri-L ‘carry’ (lit. ‘go-V.com’), e.g. (4c), and kanyji-L ‘carry’. (c) jana-ri-L ‘look after, mind, keep an eye on’ (lit. ‘stand-V.com’), e.g. (6), and nyaka-nyaka-L ‘watch, mind, keep an eye on’. 3.5.3 Word order Whereas among Warrungu transitive clauses in general two-thirds have A-initial orders (see Tsunoda 1990), among V.com constructions two-thirds have O-initial orders. Examples of such O-initial order include (1c), (5) and (10). (10) (‘‘My uncle and my father brought me some food, I am sitting here and eating it.’’) ngayku yarru-Ø (pause) warayi-Ø yampa-Ø nyina-ri-n 1sg.gen this-acc one’s own-acc camp-acc sit-V.com-nonfut ngaya 1sg.erg ‘I am sitting in my own camp.’ The tendency to place the accusative NP in sentence-initial position in V.com constructions, also contrasts with the typical location of locative or comitative NPs in basic clauses. Though such NPs may occur sentence-initially, as in (11), in twothirds of the cases they do not (see, e.g., 2a). (11) yampa-ngka nyina-n ngaya camp-loc sit-nonfut 1sg.nom ‘I sat in the camp.’ While it is tempting to view the frequent sentence-initial placement of the O in V.com constructions as a reﬂection of its topic or focus status, unfortunately not all instances of such placement (e.g. 10) are amenable to such an analysis. 3.5.4 Use in subordinate clauses All of the inﬂectional categories attested with comitative verbs (Section 3.1)— except for -nyu ‘non-ﬁnite’—can be used in independent clauses or main clauses, e.g. non-future in (10); imperative, e.g. (6); and purposive, e.g. (12). (12) ngaya nyunya yuti-ri-lku 1sg.erg 3sg.acc swim-V.com-purp ‘I will swim with her.’ In addition, the purposive can be used in subordinate clauses as well.

applicative constructions in warrungu

353

The Warrungu purposive in general indicates: (a) intention, future, etc. when used in independent clauses or main clauses, e.g., kanyji-lku in (13) and paya-lku in (15), and (b) a purpose, result, successive action or the like when used in subordinate clauses. This applies to comitative verbs as well. Examples of (a) include (12), and those of (b) include the subordinate clause in each of (13) and (14). (13) (Two men found a group of women, and one of them said: ) ngaya pulari-Ø kanyji-lku wuna-ri-lku ngaya pulari-Ø 1sg.erg two-acc carry-purp sleep-V.com-purp 1sg.erg two-acc ‘I will take [the ?] two [women] so that I can sleep with [the ?] two [women].’ (14) ngaya yarru-Ø pama-Ø kanyji-n [?ngaya TT] 1sg.erg this-acc man-acc carry-nonfut [1sg.erg] maku-ri-lku [?yarru-Ø pama-Ø TT] work-V.com-purp [this-acc man-acc] ‘I am taking this man so that [I] can work [together] with [this man].’ Before discussing the use of comitative verbs in subordinate clauses, it is ﬁrst necessary to look at the patterns of coreferential deletion in Warrungu. In Warrungu in general, when there are two coreferential NPs in two successive clauses—typically main and subordinate clauses, in this order—the second occurrence is as a rule deleted. In my view this deletion is a matter of stylistic preference, rather than a rigid grammatical rule. (I should add, however, that the syntactic phenomenon of coreferential deletion and the discourse phenomenon of ellipsis may be conceptually distinct, but that in practice they are not always easy to distinguish as far as Warrungu is concerned. See Tsunoda 1988: 616–19.) First, although this is not frequent, the second occurrence of the coreferential NP does occur un-elided. For example, see ngaya ‘1sg.erg’ pulari-Ø ‘two-acc’in the second clause of (13) (an example taken from the texts) and kiya-Ø ‘hook-acc’and yinta ‘2sg.erg’in the second clause of (24) (an example obtained during an elicitation session). Second, those sentences—not necessarily V.com constructions—which I made up and in which the second occurrence of coreferential NP occurs, were generally approved by Alf Palmer. For the sake of clarity, I supply the words which are absent due to ‘‘coreferential deletion.’’ I suspect that the resultant sentences are not stylistically perfect, although they are not ungrammatical either. Such words are preceded by a question mark, cf. (14) to (17). The preferred patterns in coreferential deletion in two successive clauses are the following (Tsunoda 1988: 610):

354

tasaku tsunoda

(a) neutral patterns: A=A, O=O, S=S (b) ergative patterns: S=O, O=S The following patterns are not favoured: (c) accusative patterns: S=A, A=S (d) aberrant patterns: A=O, O=A (e) any other patterns: O(acc)=loc, O(acc)=N.com, etc. That is, the ergative patterns are favoured over the accusative patterns. (The accusative and aberrant patterns are generally turned into one of the preferred patterns by means of antipassivization. See Tsunoda 1988.) The use of purposive comitative verbs in subordinate clauses seems to facilitate coreferential deletion. That is, it seems to have a syntactic, rather than semantic or pragmatic, motivation. As examples, compare (15) with (16). (15) ngaya yinu ngaya paya-lku kama-Ø yinta 1sg.erg 2sg.gen 1sg.erg sing-purp Kama-acc 2sg.erg nyampa-ri-lku [?kama-Ø TT] nguni-n-ta dance-V.com-purp [Kama-acc] there-link-loc ‘I’ll sing the Kama song for you so you can dance to [the Kama song] there.’ (16) TT ngaya yinu paya-lku kama-Ø yinta 1sg.erg 2sg.gen sing-purp Kama-acc(O) 2sg.nom nyampa-yal kama-yi-Ø (or kama-ngka) nguni-n-ta dance-purp Kama-N.com-nom ( Kama-loc) there-link-loc Intended meaning: As above. Probably, (16) is stylistically not perfect, since there are two coreferential NPs. In order for it to be stylistically well-formed, one of the coreferential NPs (typically the second occurrence) has to be deleted. However, the coreference pattern is acc(O) = N.com-nom (or acc(O) = loc), and this does not ﬁt in any of the preferred patterns. Consequently, if kama-yi-Ø ‘Kama-N.com-nom’ (or kama-ngka ‘Kama-loc’) is deleted, the resultant sentence will mean ‘. . . so you can dance’, rather than ‘. . . so you can dance to [the Kama song]’. That is, the resultant sentence will not be considered as one in which that N.com NP or that loc N has been deleted. Promotion of the N.com NP (or the loc NP) to the acc position creates the acc(O)=acc(O) coreference pattern, and enables the deletion of its second occurrence, as in (15). In each of (13) to (15), the main clause is transitive, and the coreference pattern is acc(O)=acc(O). Now, there is just one example in which the main clause is intransitive (a ‘‘verbless clause’’). The coreference pattern is S=O, which is a preferred deletion pattern. Here, coreferential deletion is straightforward:

applicative constructions in warrungu

355

(17) yinu-n-ku nyula walngkayi-Ø yinta 2sg-link-dat 3sg.nom(S) fond,keen-nom 2sg.erg wuna-ri-lku [?nyunya TT] sleep-V.com-purp [3sg.acc(O)] ‘She is fond of you and keen for you to sleep with [her].’ (Lit. perhaps, ‘She is fond of you so that you will sleep with [her].’) 3.5.5 ‘‘Nice to verb ’’ constructions Consider: (18) mulkungurru-Ø jarripara-Ø muja-lku squatter pigeon-acc good-acc eat-purp ‘Squatter pigeons are nice to eat.’ It may look as if (18) is an elliptical version of a sentence such as: (19) TT ngana-Ø mulkungurru-Ø jarripara-Ø muja-lku 1pl-erg squatter pigeon-acc good-acc eat-purp ‘We will eat nice squatter pigeons.’ (It is not signiﬁcant that the A NP is the ‘‘1pl.’’ Its choice is arbitrary.) However, the meaning of (18) is rather diﬀerent from that of (19), and probably (18) should not be considered as an elliptical version of (19). There are several examples like (18), which are all reminiscent of the English tough-constructions. We should probably recognize for Warrungu the following construction type: (20) ‘‘Nice to verb’’ constructions: Noun-acc ‘‘Adjective’’-acc Transitive verb-purp (As noted in Section 1.2, ‘‘adjectives’’ are diﬃcult to distinguish from nouns.) This pattern occurs in V.com constructions as well. (21) (‘‘We will sit by the ﬁre.’’) yampa-Ø yarru-Ø kitu-Ø nyina-ri-yal camp-acc this-acc cold-acc sit,stay-V.com-purp ‘This camp is cold to stay in.’ (22) (This sentence follows (2a, b): ‘‘We sat in that shade. We sat in a good shade.’’) yampa-Ø ngayku yarru-Ø jarripara-Ø nyina-ri-lku camp-acc 1sg.gen this-acc good-acc sit,stay-V.com-purp ‘This camp of mine is good to stay in.’

356

tasaku tsunoda

In these examples, too, the O NP (acc) occurs initially. In each example, the context suggests that the O NP indicates topic. That is, this use of V.com constructions seems to promote a N.com NP or a loc NP, place it sentence-initially and make it topical. If the word for a place is in the locative, the resultant sentence seems unlikely to have the meaning ‘‘nice to verb.’’ Consider (23), which I have made up: (23) TT yarru-n-ta yampa-ngka jarripara-ngka nyina-yal this-link-loc camp-loc good-loc sit-purp ‘[Someone] will sit in this good camp.’ This sentence is certainly unlikely to mean ‘This camp is nice to stay in’. That is, in order to express ‘nice to stay in’, apparently the V.com version, rather than the basic clause, has to be used. (This pattern is probably acceptable with other comitative verbs as well, although there is no suitable example.) 3.5.6 Use of V.com constructions: summary It seems that V.com constructions rather than basic clauses have to be used under any of the following circumstances: (a) when comitative verbs are lexicalized (Section 3.5.2); (b) in subordinate clauses: to facilitate coreferential deletion (Section 3.5.4), and (c) when ‘Nice to stay in’ is expressed (Section 3.5.5).

4. Instrumental verbs: Vt-ri-L 4.1 Formation The suﬃx -ri-L ‘V.inst’ is added to transitive roots. It is attested with 27 roots. In this respect, it is more productive than -ri-L ‘V.com’, which is attested with 14 verb roots/stem and one noun. Instrumental verbs are generally attested with only one inﬂectional category: purposive -lku (used in the L-class only). That is, they have a highly defective paradigm. Instrumental verbs do not seem to occur with any other inﬂectional category—possibly except for -lka ‘apprehensional’ (cf. Section 4.5.1) and -l, one of the non-future suﬃxes for the L-class (cf. Section 1.2). In this respect, instrumental verbs are less productive than comitative verbs. My ﬁeldnotes contain the following forms, which involve the ablative case suﬃx -ngumay: papa-ri-l-ngumay ‘bite-V.com-l-abl’ papa-ri-ngumay ‘bite-V.comabl’. The element -l- may be one of the non-future suﬃxes for the L-class of verbs. Unfortunately, there is no sentential example.

applicative constructions in warrungu

357

4.2 Meaning Instrumental verbs typically denote an instrument (hence, the term ‘‘instrumental verbs’’), in which case their English translations typically involve with or by means of. They can also describe a destination, a place, etc. and they may be translated involving in, into or on. They can even involve from or in return for. All the attested instrumental verbs, together with their meanings, are listed below. The number of examples of each verb in the running texts is shown in the parentheses. (The gloss ‘V.inst’ is omitted.) (i) papa-ri-L ‘stab, dig, sew, wash-’ (8 examples) : ‘stab with (a spear, a penis)’, cf. (28), (52), (55), ‘dig with (a stick)’, ‘sew with (a needle)’, cf. (26), ‘wash in (a laundry)’, cf. (33). (ii) kanyji-ri-L ‘carry-’ (8 examples): ‘carry in (a vessel)’, cf. (31), ‘carry on (wheels)’, ‘drove (cattle) with (a whip)’. (iii) palka-ri-L ‘hit, kill-’ (5 examples) : ‘hit/kill with (a stick, a club)’, ‘hit in (a ﬁghting ground)’, ‘kill (cattle) in (a yard)’. (iv) kampa-ri-L ‘cover-’ (3 examples): ‘cover with (a handkerchief, clothes, a bark blanket, sand)’. (v) jarka-ri-L ‘put in-’ (3 examples): ‘put (meat) in (a store room)’, ‘put (cattle) in (a yard)’, ‘put (water) in (a bark container)’, cf. (41), ‘put (a penis) in (a vagina)’. (vi) waju-ri-L ‘cook-’ (3 examples): ‘cook on (a ﬁre, heated stones)’, cf. (25), ‘cook in (a bark container)’. (vii) karpa-ri-L ‘cover-’ (1 example): ‘cover with (a bark blanket)’. (viii) kunpa-ri-L ‘cut-’ (1 example): ‘cut with (a knife, a stone axe)’, cf. (27), (32), (39). (ix) kuypa-ri-L ‘throw, give-’ (1 example): ‘throw (meat) onto (a ﬁre)’, cf. (43). (x) ngapa-ri-L ‘soak-’ (1 example): lit. ‘soak (honey) in (water)’, i.e. ‘mix honey with water’. (xi) paja-ri-L ‘bite-’: ‘bite with (teeth)’, cf. (29), (49). (xii) palmpi-ri-L ‘smell-’: ‘smell with (the nose ?)’. (xiii) paya-ri-L ‘sing-’: ‘sing with (a music stick ?)’, ‘sing to (the rhythm of a stick)’, cf. (46). (xiv) pija-ri-L ‘drink-’: ‘drink with (a cup)’, ‘drink from ( a leaf)’. (xv) pinta-ri-L ‘put down-’: ‘put (cattle) in (a yard)’. (xvi) pirri-ri-L ‘send-’: ‘send (someone) away with (a weapon)(?)’. (xvii) jaynyja-ri-L‘copulatewith-’:‘copulatewith(someone)in(aﬁghtingground)’, cf. (30), (35), ‘copulate with (someone) in return for (a gift)’, cf. (45).

358 (xviii) (xix) (xx) (xxi) (xvii) (xxiii) (xxiv) (xxv) (xxvi) (xxvii)

tasaku tsunoda jingka-ri-L ‘punch-’: ‘punch with (something)’. jupi-ri-L ‘rub-’: ‘rub with (a hand)’. junta-ri-L ‘kiss-’: ‘kiss with (the mouth)’. kalka-ri-L ‘pour-’: lit. ‘pour (urine or faeces, i.e. urinate or defecate) into (something)’. kipa-ri-L ‘scrape-’: ‘scrape with (something)’, cf. (53). mayka-ri-L ‘tell-’: ‘talk with (a microphone)’, cf. (34). muja-ri-L ‘eat-’: ‘eat (food) with/from (a plate)’, cf. (36). muka-ri-L ‘catch-’: ‘catch/grab (a woman) with (a hand)’, cf. (38), ‘catch (ﬁsh) with (a ﬁsh net)’. nyaka-ri-L ‘see, look at-’: ‘look at (something) with (eyes ?)’. yilmpu-ri-L ‘pull-’: ‘pull (ﬁsh) with (a hook)’, cf. (24), ‘pull (ﬁsh) in (a ﬁsh net)’.

4.3 Case frames and transitivity In terms of the case frames employed, V.inst constructions correspond to basic transitive clauses as shown in Table 3. That is, the N.inst, loc or dat NP in the basic clause is promoted to the acc (O) position in the V.inst construction. For certain instances, however, it is diﬃcult to set up a corresponding case frame. We shall brieﬂy look at each of these patterns. (a) In the vast majority of the examples, instrumental verbs indicate an instrument or the like, the N.inst NP being promoted to the acc (O) position, cf. (39).6 (b) In a fair number of instances, the loc NP is promoted to the acc (O) position. They have meanings such as ‘carry in or on (something)’, ‘cook on or in (a ﬁre)’, ‘hit or copulate in (a place)’, ‘put in or into (something)’, cf. (41). (c) With kuypa-ri-L ‘throw-V.inst’, ‘throw onto (something)’, the dat NP is promoted to the acc (O) position, cf. (43). The dative case exhibits a wide range of meanings (Tsunoda 1976b). However, this is the only example of the datto-acc(O) promotion with any instrumental verb. (d) For ‘in return for’ of jaynyja-ri-L ‘copulate with (someone) in return for (money)’, e.g. (45), and paya-ri-L ‘sing to (a stick’s rhythm), e.g. (46), it is not known what corresponding case frame to set up. As Table 3 shows, while the N.inst, loc or dat NP is promoted to the acc (O) position, the basic acc NP is demoted to the dat position. Such demotion in applicative constructions occurs, for instance, in Dyirbal (Dixon 1972: 95), in a slightly diﬀerent way in Yidiny (Dixon 1977: 310), in Kalkatungu (Blake 1979b: 88, 1987: 71), and, from outside Australia, in Chamorro (Baker 1988: 248, cited from Gibson

applicative constructions in warrungu

359

1980). It may not be irrelevant to mention here that Warrungu lacks a case frame of the type ‘‘erg acc acc,’’ with ‘‘double objects,’’ such as John gave Mary the book. With comitative verbs, it is safe to say that their valence increases by one (cf. Section 3.4). However, with instrumental verbs, it is not certain that their valance increases at all. This is because the acc NP of a basic transitive clause is ‘‘demoted,’’ turning into a dat NP, and it is not clear if this dat NP is still an argument of the verb. 4.4 Use of V.inst constructions 4.4.1 Syntactic possibilities As noted in Section 4.1, instrumental verbs generally occur with the purposive -lku only. Note in particular that they do not occur in ﬁnite forms, such as non-future. They occur in various types of sentences, as shown in Table 4. The pattern (a) is probably not very good stylistically, since there is a repetition of a coreferential word, viz. the acc NP. Recall that repetition of coreferential NPs in successive clauses is stylistically not favoured, generally the ﬁrst occurrence being retained and other occurrence(s) deleted. See Section 3.5.4. Nonetheless, in a sense, (a) exhibits the full, non-elliptical pattern, and, for the purpose of exposition, we shall start with (a) and compare other patterns with it. In V.inst constructions, the two occurrences of the acc NP have to be coreferential and the second occurrence (which is the ‘‘promoted’’ O NP) is almost always deleted7 (The only exception is (a).) The two occurrences of the erg NP do Table 4. Verbal instrumental constructions main clause

subordinate clause

(a)?

erg

acci

Vt,

erg

acci

dat

Vt-ri-lku

(b) (c) (d) (e)

erg erg erg erg

acci acci acci acci

Vt, Vt, Vt, Vt,

erg erg [erg] [erg]

[acci] [acci] [acci] [acci]

dat [dat] dat [dat]

Vt-ri-lku Vt-ri-lku Vt-ri-lku Vt-ri-lku

(f) (g) (h) (i) (j) (k)

[erg] [erg] [erg] [erg] [erg] [erg]

acci acci acci acci [acci] [acci]

[Vt], [Vt], [Vt], [Vt], [Vt], [Vt],

erg erg [erg] [erg] [erg] [erg]

[acci] [acci] [acci] [acci] [acci] [acci]

dat [dat] dat [dat] dat [dat]

Vt-ri-lku Vt-ri-lku Vt-ri-lku Vt-ri-lku Vt-ri-lku Vt-ri-lku

(l) [erg] (m)? [erg]

[acci] [acci]

[Vt], [Vt],

erg erg

[acci] acci

[dat] [dat]

Vt-ri-lku Vt-ri-lku

360

tasaku tsunoda Table 3. Verbal instrumental constructions and basic transitive clauses V.inst constructions

erg

acc

dat

Vt-ri-L

basic transitive clauses (a) (b) (c) (d)

erg erg erg erg

N.inst loc dat ?

acc acc acc Acc

Vt Vt Vt Vt

not have to be coreferential. When they are, the second occurrence is generally deleted (again the only exception being (a)). Examples follow. The only example of (a) is: (24) kiya-Ø yinta muka-Ø kiya-Ø nguna-wu hook-acc 2sg.erg hold-imp hook-acc that-dat yinta yilmpu-ri-lku 2sg.erg pull,drag-V.inst-purp ‘Get a/the hook so that you can pull that with the hook.’ As mentioned above, probably (a) is stylistically not very well-formed8 (This is indicated by the question mark in the table.) For it to be well-formed, the second occurrence of the acc NP will have to be deleted. (In (24), the second erg NP happens to be coreferential with the ﬁrst erg NP, and this, too, will have to be deleted for the sentence to be well-formed. The resultant sentence will be an instance of (d).) The only example of (b) is: (25) puri-Ø ngaya muka-n ﬁrewood,ﬁre-acc 1sg.erg get-nonfut [?puri-Ø TT] manyja-wu waju-ri-lku jana-ngku [ﬁrewood-acc] food-dat cook-V.inst-purp 3pl-erg ‘I obtained ﬁrewood so that they could cook food on [the ﬁre].’ In the subordinate clause—in addition to the acc NP—the dat NP, too, may be elided, resulting in (c), e.g. (26); the erg, too, may be ellipsed, resulting in (d), e.g. (27); or both of them, too, may be understood, resulting in (e), e.g. (28). (26) yinta nguna ngayku kuypa-Ø ngaya papa-ri-lku 2sg.erg that-acc 1sg.gen give-imp 1sg.erg sew-V.inst-purp ‘Give me that [needle] so that I can sew [something] with [it].’

applicative constructions in warrungu

361

(27) pama-ngku pilkuru-Ø kanyji-n manyja-wu kunpa-ri-lku man-erg knife-acc carry-nonfut food-dat cut-V.inst-purp a. ‘The man carried a knife so [he] could cut food with [it].’ b. ‘The man carried a knife so [someone else] could cut food with [it].’ c. ‘The man carried a knife to cut food with [it].’ The above three interpretations are possible, since the elided erg NP of the subordinate clause does not have to be coreferential with that of the main clause. (28) ngaya pangkay-Ø muka-n papa-ri-lku 1sg.erg spear-acc get-nonfut stab-V.inst-purp ‘I got a spear to stab [someone] with.’ It seems reasonable to say that (c), (d), and (e) are elliptical versions of (a). The other patterns may not be adequately considered as elliptical versions of (a). If for the sake of exposition they are considered as elliptical versions at all, they can be shown as (f) to (l) in Table 4. Examples are (29) for (f); (30) for (g); (31) for (h); (32) for (i); (33) for (j), and; (34) for (k). (29) rirra-Ø ngaya paja-ri-lku manyja-wu tooth-acc 1sg.erg bite-V.inst-purp food-dat Informant’s translation: ‘I [have—TT] got my tooth to bite tucker.’ (30) (Alf Palmer described the ﬁghting ground rather jocularly as follows.) purun-Ø jaynyja-ri-lku kalpiri-ngku ﬁghting ground-acc copulate with-V.inst-purp children-erg ‘The ﬁghting ground is for children to copulate with [someone] in.’ (31) nupa-Ø kamu-wu kanyji-ri-lku bark container-acc water-dat carry-V.inst-purp ‘A bark container is to carry water in’ or ‘a bark container is for carrying water.’ (32) pilkuru-Ø kunpa-ri-lku knife-acc cut-V.inst-purp ‘A knife is to cut with’, or ‘a knife is for cutting.’ (33) (Alf Palmer described a laundry as follows.) kampi-wu papa-ri-lku clothes-dat wash-V.inst-purp ‘For washing clothes in.’ (34) (Alf Palmer described the use of a microphone as follows.) mayka-ri-lku tell-V.inst-purp ‘[A microphone is] for talking with.’

362

tasaku tsunoda

The patterns (f) to (k) are common. They are generally used to describe the use/ function of a tool or the like. In my view they should not be considered as elliptical versions of (a), but as separate constructions: ‘‘Noun is for verbing with.’’ See Section 4.4.4. Examples of the remaining patterns follow: (35) for (l), and (36) for (m). (35) (Alf Palmer described the ﬁghting ground jocularly as follows.) yinta yinta jaynyja-ri-lku 2sg.erg 2sg.erg copulate with-V.inst-purp ‘[The ﬁghting ground] is for you to copulate with [someone] in.’ (36) ngaya yarru-Ø muja-ri-lku 1sg.erg this-acc eat-V.inst-purp ‘I eat with/from this [plate]’, or ‘This [plate] is for me to eat with/from.’ Example (36) was uttered by Alf Palmer. However, this pattern does not seem very well-formed (hence the question mark for (m) in Table 4). The reason is as follows. On the analogy of examples such as (36), a sentence such as (37) would be expected: (37) TT *pama-ngku yarru-Ø paja-ri-lku man-erg this-acc bite-V.inst-purp Intended meaning: ‘The man will bite [something] with these [teeth of his]’. However, it was rejected by Alf Palmer. Also, I suspect that (l), too, is not very well-formed. These two patterns (l, m) diﬀer from the (common and acceptable) patterns (f) to (k) in that the erg NP of the instrumental verb occurs sentence-initially. See also the last paragraph of Section 4.4.4. As shown by examples (24) to (28), the main clause is virtually always transitive. There is no unequivocal example of an intransitive clause, such as ‘‘The axe arrived for me to chop the tree with’’ (the English sentence by William McGregor, p.c.). In virtually all of the relevant examples, the action described by the main clause is done in order that the action described by the subordinate clause is possible, i.e. ‘in order that’; see (24) to (28). The only exception is: (38) (When Alf Palmer and I were sitting at the ﬁre, he advised me not to burn my hands.) yinta mara-Ø wuyji-ngka . . . rayi-wu muka-ri-lku 2sg.erg hand-acc burn-appr girl-dat catch-V.inst-purp ‘You might burn [your] hands that you catch [girls] with.’ (This is an example of (d).) This sentence does not mean ‘You might burn your hands in order that you can catch girls’. Here, the instrumental verb is probably used

applicative constructions in warrungu

363

to characterize the hands in a way similar to examples like (29) to (34), i.e. ‘hands for catching girls’. (I owe this observation to William McGregor, p.c.) V.inst constructions with the meaning ‘in order that’ are reported, for instance, from Dyirbal (Dixon 1972: 96), Yidiny (Dixon 1977: 310), and Kalkatungu (Blake 1979b: 88). But examples such as (38), which do not contain the meaning ‘in order that’, seem rare in those languages; the only reported example is the example (583) in Dixon (1977: 311). As Table 4 demonstrates, sentences involving an instrumental verb exhibit a continuum from full two-clause sentences, e.g. (24), to very fragmentary—oneclause and one-word—sentences. Since they constitute a continuum, it is diﬃcult to draw a line between monoclausal and biclausal sentences. This in turn makes it diﬃcult to adequately characterize subordination in Warrungu. 4.4.2 Use in subordinate clauses Like V.com constructions (Section 3.5.4), V.inst constructions may be used in subordinate clauses to facilitate coreferential deletion, turning a nonfavoured pattern into a preferred pattern. Thus, (39) may be considered to be derived from (40), which is probably not perfect stylistically. (39) pama-ngku pilkuru-Ø kanyji-n manyja-wu kunpa-ri-lku man-erg knife-acc carry-nonfut food-dat cut-V.inst-purp ‘The man carried a knife so that [he] could cut food with [it].’ (40) pama-ngku pilkuru-Ø kanyji-n pama-ngku manyja-Ø kunpa-lku man-erg knife-acc carry-nonfut man-erg food-acc cut-purp pilkuru-ngku knife-N.inst Intended meaning: as above. This pair demonstrates N.inst-to-acc(O) promotion. The coreference pattern is ‘‘acc(O)=N.inst’’ in (40), and ‘‘acc(O)=acc(O)’’ in (39). That is, this promotion facilitates coreferential deletion, turning a nonfavoured pattern ‘‘acc(O)=N.inst’’ into a preferred pattern ‘‘acc(O)=acc(O).’’ (In (39), the second occurrence of pama-ngku ‘man-erg’, too, is deleted.) Similarly for loc-to-acc and dat-to-acc(O) promotions. An example demonstrating the promotion of a loc NP is: (41) nyula kanyji-n parrawu-Ø kamu-wu 3sg.erg carry-nonfut bark container-acc water-dat jarka-ri-lku put in-V.inst-purp ‘She carried a bark container to put (i.e. pour) water into.’

364

tasaku tsunoda

The second clause corresponds to: (42) TT nyula kamu-Ø parrawu-ngka jarka-lku 3sg.erg water-acc bark container-loc put in-purp ‘so that she could put water in the bark container’ The only example of the promotion of a dat NP is: (43) (‘‘I caught a kangaroo, and made a ﬁre to cook it.’’) puri-Ø nguna-Ø kakal-Ø waju-n ﬁrewood,ﬁre-acc that-acc big-acc burn-nonfut kuypa-ri-lku . . . yuri-wu . . . . throw-V.inst-purp kangaroo-dat ‘[I] made that big ﬁre to throw the kangaroo onto.’ The second clause corresponds to: (44) TT ngaya yuri-Ø kuypa-lku puri-wu 1sg.erg kangaroo-acc throw-purp ﬁre-dat ‘so that I could throw the kangaroo onto the ﬁre’ 4.4.3 Lexicalized instrumental verbs There are at least two instrumental verbs for which I am unable to set up a corresponding clause (cf. (d) in Section 4.3). They will have to be used in place of a nonexistent basic clause. They may be considered lexicalized. One of them is jaynyjari-L ‘copulate with (someone) in return for (a gift, payment)’, e.g.: (45) yinta kajarra-Ø kuypa-n jaynyja-ri-lku 2sg.erg possum-acc give-nonfut copulate with-V.inst-purp ‘You gave [her] a possum to copulate with [her] in return for [it].’ jaynyja-ri-L ‘copulate with-V.inst’ is the only instrumental verb that is attested with the meaning ‘in return for’. (There are 6 examples of jaynyja-ri-L ‘copulateV.inst’ with this meaning.) The other lexicalized instrumental verb is paya-ri-L ‘sing to (a stick’s rhythm)’, e.g.: (46) jalimpirri-Ø muka-Ø paya-ri-lku music stick-acc get-imp sing-V.inst-purp ‘Get a music stick to sing [a song] to [its rhythm].’ In intransitives clauses, the N.com and the loc can express ‘dance to (music)’, cf. (16). However, it is not certain if this is possible in transitive clauses. I suspect that this is not possible.

applicative constructions in warrungu

365

4.4.4 ‘‘Noun is for verb ing with’’ constructions In Section 3.5.5, I proposed to recognize ‘‘Nice to verb’’ constructions. They may involve either a basic transitive verb or a comitative verb. Likewise, I propose to set up ‘‘Noun is for VERBing with’’ constructions. They always involve an instrumental verb, and not a basic transitive verb. They are exempliﬁed by the patterns (f) to (k) of Table 4, cf. (29) to (34). These patterns are used to describe the use or function of a tool or the like. The acc NP, which denotes a tool or the like, always occurs sentence-initially, and they seem to indicate a topic. Compare (31), one such example, with: (47) TT nupa-ngku kamu-Ø kanyji-lku bark container-N.inst water-acc carry-purp This sentence will simply mean ‘[Someone] will carry water in a bark container’, and it is decidedly unlikely to mean ‘A bark container is for carrying water in/with’. That is, a basic clause will not be used to describe the use or function of a tool, and the corresponding ‘‘Noun is for verbing with’’ version will have to be used instead. ‘‘Nice to verb’’ and ‘‘Noun is for verbing with’’ constructions are similar in that: (a) the sentence characterizes (the referent of) the acc NP; (b) the acc NP occurs sentence-initially, and; (c) the acc NP seems to indicate topic. It may be relevant that in (37), which was rejected by Alf Palmer, the acc NP does not occur sentence-initially. 4.4.5 Use of V.inst constructions: summary It seems that V.inst constructions rather than basic clauses have to be used under any of the following circumstances: (a) in subordinate clauses: to facilitate coreferential deletion (Section 4.4.2); (b) when instrumental verbs are lexicalized (Section 4.4.3), and; (c) when ‘‘Noun is verbing with’’ is expressed (Section 4.4.4). 4.5 Restrictions (?) on the use of instrumental verbs 4.5.1 Inﬂectional sufﬁxes As noted in Section 4.1, instrumental verbs almost always occur with the purposive suﬃx (-lku for the L-class) and generally do not occur with any other inﬂectional category, such as non-future or apprehensional. Alf Palmer rejected made-up examples with the non-future suﬃx, but was more liberal in regard to those with

366

tasaku tsunoda

the apprehensional suﬃx, though he himself produced only one such example, namely (52). (48) papa-ri-lka stab-V.inst-appr ‘By and by [he] might spear [someone]. [He is] no good.’ V.inst constructions with the meaning ‘in order to/that’ are reported, for instance, from Dyirbal (Dixon 1972: 96), Yidiny (Dixon 1977: 310), and Kalkatungu (Blake 1979b: 88). However, examples with the opposite meaning ‘lest, in case’, do not seem to be reported from those languages. 4.5.2 Pronouns for the ‘‘demoted’’ dat NP In virtually all the examples, the ‘‘demoted’’ dat (which is derived from the ‘‘basic’’ acc NP) is a noun or a demonstrative (a type of noun), and not a (personal) pronoun. There is only one example of a pronoun: (49) yinu-n-ku kipa-ri-lku 2sg-link-dat scratch-V.inst-purp ‘To scratch you with.’ However, it appears that a pronoun is not favoured in this position. Thus, when I suggested: (50) TT papa-ri-lku nyungu-n-ku stab-V.inst-purp 3sg-link-dat Intended meaning: ‘To stab him/her with.’ Alf Palmer did not approve it and instead he gave the following, which involves the demonstrative (a type of noun) yarru ‘this’: (50) papa-ri-lku yarru-wu stab-V.inst-purp this-dat ‘To stab this [man] with.’ In V.com constructions, the ‘‘promoted’’ acc is almost always expressed by a noun, but there are a very small number of examples involving a pronoun, cf. nyunya ‘3sg.acc’ in (12). It is not known whether a pronoun is disfavoured in this position.

5. Restrictions on V.com and V.inst constructions There are a few restrictions or tendencies that (appear to) apply to both V.com and V.inst constructions.

applicative constructions in warrungu

367

5.1 Derivational sufﬁxes As noted in Section 2, apparently -ri-L ‘V.com’/’V.inst’ cannot be added to a derivational suﬃx9 Thus, we cannot have sequences such as: rec-V.com, rfl-V.com, anti-V.com rec-V.inst, rfl-V.inst, anti-V.inst This is despite the fact that -ri-L ‘V.com’ are generally added to intransitive roots and that reciprocal, reﬂexive and antipassive verbs are intransitive; they take S NP (in the nominative), and not A NP (in the ergative). Also, apparently -ri-L ‘V.com’/’V.inst’ cannot be followed by a derivational suﬃx. Thus, we cannot have sequences such as: V.com-rec, V.com-rfl, V.com-anti V.inst-rec, V.inst-rfl, V.inst-anti This is despite the fact that both comitative and instrumental verbs are transitive (they take A NP) and that the reciprocal (-wa-Y), reﬂexive (-kali-Ø, -li-Ø) and antipassive (-kali-Ø) are added to transitive roots/stems (but not to intransitives)10 Thus, consider: (51) ngali-Ø wuna-ri-n pirri-pirri 1du-erg lie-V.com-nonfut close-close ‘We slept very close to each other.’ As the translation indicates, (51) describes a reciprocal situation. However, the simple comitative verb wuna-ri-n is used in place of the expected but unacceptable reciprocal version *wuna-ri-wa-n ‘sleep-V.com-rec-nonfut’. In my view (51) is transitive, like other V.com constructions and unlike reciprocal constructions, and that ngali-Ø is ergative rather than nominative. However, it is no longer possible to test this. 5.2 Animacy of A NP In all the examples of V.com and V.inst constructions, the A NP is human; there is no example of non-human animate (e.g. ‘‘dog’’) or inanimate (e.g. ‘‘log’’) A NP. Applicative constructions, whether ‘‘V.com’’ or ‘‘V.inst,’’ almost always describe a volitional act. The only possible exception is wuna-ri-L ‘sleep-V.com-’, ‘sleep with (a man, woman)’. See Section 3.2.

368

tasaku tsunoda

5.3 Meaning of the ‘‘basic’’ NPs Abstract nouns. In a fair number of instances, the N.com suﬃx (C-ji/V-yi) is added to an abstract noun, resulting in an ‘‘adjective.’’ They generally describe the state of body or emotion. Thus, murran ‘illness’ and murran-ji ‘ill’, e.g. (57). (52) murran-ji-Ø nyula wuna-n illness-N.com-nom 3sg.nom lie-nonfut ‘She is lying with illness’, i.e. ‘She is lying ill’. It seems that an abstract noun cannot be promoted. (At least, there is no example.) This is despite the existence of a corresponding comitative verb, e.g. wuna-ri-L ‘lieV.com’ for (52). Similarly, in V.inst constructions, the ‘‘promoted’’ acc(O) NP, which indicates an instrument or the like, has a concrete referent, and is never an abstract noun. Probably abstract nouns cannot be promoted. Body part nouns. The N.com suﬃx may be added to body part nouns. Thus: (58) jana-Ø . . . pulu-yi-Ø wuna-n 3pl-nom belly-N.com-nom lie-nonfut ‘They are lying satisﬁed with food’. (Lit. ‘They are lying with the belly.’11) However, it seems that a ‘‘body part’’ N.com NP cannot be promoted. (At least, there is no example.) This is despite the existence of a corresponding comitative verb, e.g. wuna-ri-L ‘lie-V.com’ for (53). In contrast, V.inst constructions have yielded several examples of the promotion of ‘‘body part-N.inst’’ to the acc(O) position, e.g. rirra-Ø ‘tooth-acc’ in (29). Beneﬁciaries and possessors. Cross-linguistically, applicative verbs appear often to have a ‘‘benefactive’’ meaning (cf. Comrie 1985: 312–19, Baker 1988: 236, Spencer 1991: 253–54, Palmer 1994: 161–63). (I owe Masayoshi Shibatani the reference to Baker’s work.) They may also refer to possessors (Baker 1988: 269–73, Spencer 1991: 254). However, Warrungu applicatives—whether comitative or instrumental—do not seem to indicate a beneﬁciary or a possessor. In Warrungu, both a beneﬁciary and a possessor may be marked by the dative, e.g. jumupuru-wu ‘cattle-dat’ in (59) (a transitive clause), or by the genitive, e.g. yinu ‘2sg.gen’ in (15) (again, a transitive clause). (54) (‘‘On a cold night:’’) waypala-ngku jumupuru-wu yaku-Ø waju-n whiteman-erg cattle-dat grass-acc burn-nonfut ‘The whitemen burned grass for the cattle (in order to keep them warm).’

applicative constructions in warrungu

369

Table 5. Applicatives in Australian languages

Yir-Yoront Djabugay Yidiny Dyirbal Wargamay Warrungu Biri Kalkatungu Pitta-Pitta Ngiyambaa Diyari

V.com (Vi-app) indep. sub-purp

V.inst (Vt-app) indep.

VERB-ing with

sub-purp

+ + + + + + + rare + + +

? ? + + ? – ? rare ? ? ?

? ? ? ? ? + ? + ? ? ?

? ? + + + + ? + ? ? ?

? ? ? ? + + ? + ? ? ?

There is no example of promotion of a beneﬁciary or a possessor NP. This is despite the existence of a corresponding instrumental verb, e.g. waju-ri-L ‘cook, burnV.inst’ for (59), and paya-ri-L ‘sing-V.inst’ for (15). This is also despite the fact that a dat NP can be promoted when it means ‘(throw) onto’; see (43). I feel conﬁdent to say that, in Warrungu, beneﬁciary and possessor NPs cannot be promoted. Other meanings of the nominal comitative and the locative. The N.com case (see Tsunoda 1976a) and the loc case can have various meanings, but apparently such NPs cannot be promoted when they have any one of the following meanings. At least, there is no example. This is often despite the existence of a corresponding applicative–comitative or instrumental–verb. (a) N.com case: (i) cause, reason (example (53) may be an instance of this restriction); (b) loc case: (i) time, (ii) place name, (iii) cause, reason12

6. Comparative notes Table 5 looks at a selected sample of applicative constructions in Australian languages. The sources are as follows (the applicative suﬃxes employed are shown in the parentheses): Yir-Yoront (-on), Alpher 1991: 48; Djabugay (-rri), Patz 1991: 283–84; Yidiny (-nga-l), Dixon 1977: 293–322; Dyirbal (-ma-l, -mpa-l), Dixon 1972: 95–99; Wargamay (-ma), Dixon 1981: 77–80; Biri (-ri), Beale 1974: 25; Kalkatungu (-nti), Blake 1979b: 87–89; Pitta-Pitta (-la, -ri), Blake 1979a: 204–06; Ngiyambaa (-pa-l), Donaldson 1980: 163; and, Diyari (-lka), Austin 1981: 72–73,

370

tasaku tsunoda

157–59. I have excluded those applicatives which have a benefactive meaning; they occur in Pitta-Pitta. Biri and Pitta-Pitta have a V.com suﬃx -ri, which is possibly cognate with the Warrungu suﬃx -ri-L. Also, the following generalizations seem possible. (a) If a given suﬃx has a V.inst function, then it has a V.com function as well. (b) If given comitative verbs can be used in purposive subordinate clauses (Section 3.5.4), then they can be used in independent clauses as well. (c) In contrast, if given instrumental verbs can be used in independent clauses, then they can be used in purposive subordinate clauses (cf. Section 4.4.2) as well. (Here, we deal with what may be considered as prototypical independent clauses, cf. [1] in Section 4.5.1, and ignore the use of ‘‘Noun is for verbing with.’’) (d) Instrumental verbs: their use for ‘‘Noun is for verbing with’’ (Section 4.4.4) is uncommon. It is reported from two languages only, viz. Kalkatungu and Warrungu. The remarks in (a), (b) and (c) suggest possible implications for a comparativehistorical study of applicative constructions. That is, it will be: (a) comitative verbs rather than instrumental verbs; (b) with comitative verbs: use in independent clauses rather than in purposive subordinate clauses, and (c) with instrumental verbs: use in purposive subordinate clauses rather than in independent clauses that develop ﬁrst, that are diﬀused ﬁrst, and/or that are lost last.

Notes 1. I dedicate this paper to Barry J. Blake, who supervised my postgraduate studies at Monash University, in my deepest appreciation of his invaluable guidance and support. My work on Warrungu was ﬁnanced by the then Australian Institute of Aboriginal Studies and by Monash University. I wish to thank Alf Palmer for teaching me his language. I am also grateful to (i) Peter Sutton for making his Warrungu data available to me, (ii) Bernard Comrie, Motoyasu Nojima, Asako Shiohara, Masayoshi Shibatani and Kevin Tuite for commenting on an early version of this paper, and (iii) Gavan Breen and William McGregor for providing very detailed comments. 2. William McGregor informs me that in Warrwa of Western Australia (see McGregor, this volume) the O NP in applicative constructions is in some way aﬀected by the action, in contrast with the corresponding non-applicative examples. This is an attractive suggestion, but unfortunately it is no longer possible to this for Warrungu.

applicative constructions in warrungu

371

13. In this paper, I use the terminology of transformational grammar or relational grammar, such as ‘‘promotion,’’ ‘‘demotion’’ and ‘‘deletion.’’ This is merely because these theories are convenient for handling phenomena such as voice, e.g. applicatives (cf. Blake 1987: 69–76, Baker 1988: 229–304) and intra-clausal anaphora. I do not consider myself a transformational or relational grammarian. 14. The following additional abbreviation are employed:TT marks those words which I have added to the examples provided by Alf Palmer, or, those sentences which I have made up. See also the third paragraph below (14) in Section 3.5.4. Enclitics are preceded by an equation sign, while other morpheme boundaries are shown by a hyphen. In the case of fusion, morpheme boundaries are not indicated and glosses are given as in ngaya ‘1sg.nom’. 15. The N.com case suﬃx (C-ji/V-yi) can be followed by any one of other case suﬃxes except for the ablative: nominative (zero), accusative (zero), ergative, nominal instrumental, dative and locative. NPs in the N.com case often modify another NP, showing agreement in terms of case. Thus, in (1b), warrngu-yi-Ø ‘woman-N.com-nom’modiﬁes yinta ‘2sg.nom’ (both are in the nominative) and the sentence may also be translated as follows: ‘You, who are with a woman, are sitting’. —It is in fact controversial whether to regard such a suﬃx as a case suﬃx or as a derivational suﬃx (Blake 1987: 73–74). In this paper, I treat it as a case suﬃx. This treatment enables us to state the promotion in terms of cases only (i.e. promotion from the N.com case or from the loc case), rather than having to refer to a derivational suﬃx (cf. Blake 1987: 74). 16. With muja-ri-L ‘eat-V.inst’, ‘eat with/from (a plate)’, e.g. (36), and also pija-ri-L ‘drink with/ from (a cup, a leaf)’, it may also be possible to say that an ABL NP is promoted to the acc position. 17. In the vast majority of the examples, the subordinate clause follows the main clause, but there are a few examples in which the subordinate clause may be considered to precede the main clause. An example is: (i) pangkurru-wu papa-ri-lku nguna-Ø ngaya miranga-lku turtle-dat stab-V.inst-purp that-acc 1sg.erg make-purp ‘I will make that to stab turtles with’. In terms of the classiﬁcation in Table 4, I tentatively assign (i), for instance, to the pattern (d), although the main clause precedes in (d). 18. (24) was obtained during an elicitation session. Probably, Alf Palmer dictated this sentence, slowly and word by word, and this explains the repetition of the erg and acc NP’s. 19. My ﬁeldnotes contain two instances in which -ri-L ‘V.inst’ is added to a derivational suﬃx, viz. the transitive-stem-forming suﬃx -nga-L in each instance: (a) mira-nga-L Vt‘make’ → mira-nga-riL ‘make (something) with (something)’, and (b) yakay ‘Ouch !’ → yakay-nga-ri-L ‘make (someone) scream with (something): (i) yinta jumpi-Ø kanyji-n warrngu-wu yakay-nga-ri-lku 2sg.erg penis-acc carry-nonfut woman-dat Ouch!-tr-V.inst-purp ‘You have a penis with which to make a woman scream.’

372

tasaku tsunoda

However, I am not certain if these two instrumental verbsare genuine Warrungu words. 10. In contrast with Warrungu, sequences such as the following are attested in other languages: (a) anti-V.com in Dyirbal (Dixon 1972: 97), (b) rfl-V.com in Dyirbal (Dixon 1972: 97), (c) antiV.com in Yidiny (Dixon 1977: 304, 320), (d) anti-V.inst in Yidiny (Dixon 1977: 294, 310). 11. If the noun refers to a body part that people generally have, e.g. ‘‘belly’’ and ‘‘foot,’’ rather than a body part that not everyone has, e.g. ‘‘grey hair,’’ the resultant word means ‘‘something is abnormal/not usual with the body part,’’ e.g. ‘satisﬁed with food’ in (55); see Tsunoda (1976a: 220). 12. Baker (1988: 244–45, 261) notes that cross-linguistically adverbial phrases such as the following cannot be promoted in applicative constructions: temporal phrases, manner phrases, and reason phrases.

References Alpher, Barry. 1991. Yir-Yoront Lexicon: Sketch and dictionary of an Australian language. Berlin: Mouton de Gruyter. Austin, Peter. 1981. A Grammar of Diyari, South Australia. Cambridge: Cambridge University Press. ——. 1997. ‘‘Causatives and Applicatives in Australian Aboriginal Languages.’’ In Kazuto Matsumura and Tooru Hayasi (eds.), The Dative and Related Phenomena. Tokyo: Hitsuji Shobo, 165–225. Baker, Mark C. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Beale, Tony. 1974. A Grammar of the Biri Language of North Queensland. MS. Canberra: Australian National University. Blake, Barry J. 1979a. ‘‘Pitta-Pitta.’’ In R.M.W. Dixon and Barry J. Blake (eds), Handbook of Australian Languages, vol.1. Canberra: Australian National University Press, 183–242. ——. 1979b. A Kalkatungu Grammar. Canberra: Paciﬁc Linguistics, The Australian National University. ——. 1987. Australian Aboriginal Grammar. London: Croom Helm. Comrie, Bernard. 1985. ‘‘Causative Verb Formation and Other Verb-Deriving Morphology.’’ In Timothy Shopen (ed.), Language typology and syntactic description, III, Grammatical Categories and the Lexicon. Cambridge: Cambridge University Press, 309–48. Dixon, R.M.W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. ——. (ed.). Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, and New Jersey: Humanities Press, ——. 1977. A Grammar of Yidin. Cambridge: Cambridge University Press. ——. 1981. ‘‘Wargamay.’’ In R.M.W. Dixon and Barry J. Blake (eds), Handbook of Australian Languages, vol. 2. Canberra: Australian National University Press, 1–44. Donaldson, Tamsin. 1980. Ngiyambaa: The language of the Wangaaybuwan. Cambridge: Cambridge University Press.

applicative constructions in warrungu

373

Gibson, J. 1980. Clause Union in Chamorro and in Universal Grammar. Ph.D. thesis. University of California, San Diego. Palmer, F.R. 1994. Grammatical Roles and Relations. Cambridge: Cambridge University Press. Patz, Elizabeth. 1991. ‘‘Djabugay.’’ In R.M.W. Dixon and Barry J. Blake (eds), Handbook of Australian Languages, vol. 4, Melbourne: Oxford University Press, 245–347. Spencer, Andrew. 1991. Morphological Theory. Oxford: Basil Blackwell. Tsunoda, Tasaku. 1974. A Grammar of the Warungu Language, North Queensland. MA thesis. Melbourne: Monash University. ——. 1976a. ‘‘The Derivational Aﬃx ‘Having’ in Warungu.’’ In Dixon (ed.), 214–25. ——. 1976b. ‘‘The Bivalent Suﬃx -ku in Warungu.’’ In Dixon (ed.), 456–64. ——. 1988. ‘‘Antipassives in Warrungu and Other Australian Languages.’’ In Masayoshi Shibatani (ed.), Passive and Voice. Amsterdam: John Benjamins, 595–649. ——. 1990. ‘‘Typological Study of Word Order in Languages of the Paciﬁc Region (5): Warrungu (Australia).’’ The Journal of the Faculty of Letters Nagoya University, No.106: 13–47. ——. 1992. ‘‘Positions of Warrungu Interrogative Words.’’ In Tom Dutton et al. (eds), The Language Game: papers in memory of Donald C. Laycock, Canberra: Paciﬁc Linguistics, The Australian National University, 483–90. ——. Forthcoming-a. ‘‘Reciprocal Constructions in Warrungu.’’ In Vladimir P. Nedjalkov (ed.), Typology of reciprocal constructions Munich: Lincom Europa. ——. Forthcoming-b. ‘‘Aspect and Transitivity of Iterative Constructions in Warrungu.’’ In Werner Abraham and Leonid Kulikov (eds), Aspects of Aspect: Tense and event categories in typological perspectives. Papers presented to Vladimir Nedjalkov on the occasion of his 70th birthday. Amsterdam: John Benjamins.

Language Index Acehnese–277–8, 283, 285–6, 290, 295–8, 303, 305 Akan–208 Algonquian–231, 234, 240–2, 245 Amurdak–115–8, 144, 148 Ancient Greek–151, 154, 161–2, 166–8 Apalai–244 Arosi–253 Asumboa–249, 262–3 Atayal–277, 282–3, 286, 288, 290, 292, 295–7, 299, 306 Atchin–255–6 Athabaskan–234 Avar–280–1 Balawaia–249, 252 Banoni–249, 252–3 Barai–208–9 Big Nambas–255–6, 263 Biri–367, 370 Bugotu–253–4 Buma–262 Carib–244 Chamorro–234, 358 Chepang–244 Daghestanian–95–9, 108–11, 113, 238 Dehu–258 Diyari–30–4, 369 Djabugay–367 Djapu–32 Dravidian–68, 207 Dyirbal–192, 278, 280–1, 305, 358, 363, 366, 369 English–7, 8, 10–2, 15, 21, 37, 39–42, 47, 51, 59, 62, 63, 66–7, 71–2, 74–8, 80–2, 86–7, 91, 108, 121, 148, 151, 153–4, 158–9, 162, 167, 178, 201, 212, 224–6, 278, 281–3, 289, 297–8, 307, 310, 312, 320–23, 330–5, 355, 357, 362 Evenki–314 Finnish–106–10, 316, 328, 333

Formosan–277, 280, 283–4, 292, 303, 305, 306 Garig–115–6, 132, 144, 147–8 German–41, 64–5, 68, 71–2, 76–85, 90, 313 Gilbertese–61, 261–2 Hebrew–38–9, 41, 47, 50 Houailou–258 Hungarian–106–9, 127 Iai–258 Icelandic–151, 154–8, 161, 163–8 Ilgar–115–6, 132, 139–0, 145, 147 Inakona–253–4 Iwaidja–115–6, 118–20, 122, 124, 127, 132–3, 138, 143, 145–8 Japanese–47, 313, 334 Jiwarli–19–22, 24, 29, 33 Kairiru–249–50 Kalam–201–3, 205–8, 210–2, 214, 216–8, 220–6 Kalkatungu–95, 192, 277, 358, 363, 366, 369, 370 Kaluli–38 Kawaiisu–314 Kuikuro–244 Kunparlang–118 Kunwinjku–116, 118, 121, 132–3, 143, 148, 148 Kusaiean–61, 261, 264–9 Kutenai–234 Kwaio–253–4, 259 Labu–249–51 Lamani–310 Latin–67, 77, 109 Lenakel–257, 264, 266, 268 Limbu–244 Macassarese–116, 121 Malay–116, 121 Manam–249–50, 263 Mandarin–207, 312–4, 316, 328, 330, 335

376

language index

Mantharta–19–21, 29, 30, 32–3 Maori–259–60 Marshallese–60–1, 261–2 Martuthunira–20 Maung–115–6, 129, 147–8 Micronesian–61, 249, 261–2, 264, 267 Mokilese–61, 261, 267–8, 271 Mono-alu–249, 253 Nakanai–249, 251, 254, 256, 263 Nchuﬁe–311 Nembao–262 Nengone–258 Ngamini–30–1, 33 Ngiyambaa–369 Ngizim–326, 335 Nguna–255, 350 Nissan–249, 252 Niuean–259–60, 270 Nocte–244 Nootka–234 Oceanic–205, 207, 247–9, 256, 258–9, 263–4, 268–70 Onondaga–62 Paamese–255–6, 264 Panyjima–20 Payungu–20 Pennsylvania German–71–2, 85 Philippine–277, 280, 282, 292, 305 Pinikura–20, 33 Plains Cree–231, 240 Ponapean–61 Purduna–20 Rapanui–259–60 Romance–41, 109, 283 Rotuman–259 Russian–5, 38, 68, 96, 316 Sakao–255–56 Samoan–259–60, 270–71 Shapatin–234 Swahili–68

Tabasaran–95–100, 103–8, 110–3, 238–39, 245 Tairora–207, 225 Tanema–262 Tangut–244 Tanimbili–249, 262–3 Tanoan–234, 240–2, 245 Thalanyji–20 Tharrkari–20 Thiin–20 Tibetan–244, 280 Tigak–249, 251–52 Tiwi–116–7 To’aba’ita–253–4 Tok Pisin–207, 225 Tolai–249, 252–3, 264 Tongan–314 Trukese–61 Tsez–96, 100–01, 103–8, 110–3 Tsou–277, 284–6, 291–2, 294–5, 302, 306 Turkish–37, 40–1, 316 Ulawa–253–4 Ulithian–61 Umbugarla–116, 132, 145, 148 Vano–262 Waiwai–244 Wargamay–369 Warlpiri–37, 39, 42–3, 45–53 Warriyangka–20 Warrungu–189, 192, 343–4, 347, 352–3, 355, 358, 363, 368–73 Warrwa–171–3, 178, 180, 192, 194, 197, 370 Welsh–127 Woleaian–61 Wurrugu–115–7 Yami–277, 286, 292, 294–5, 297, 303, 305, 307 Yankuntjatjara–30, 33 Yidi–192, 358, 363, 366, 369, 372 Yimas–209 Zapotec–309

Subject Index accomplishments–3 achievement–2, 3 acquisition–84, 85 active alignment–230–6, 239, 244–5, 295, 298–9, 303 voice–1, 230–1, 239 activity–1–3, 141, 173, 176 activity verbs–2 actor–1, 174 ﬀ, 279 ﬀ actor–220–5 Actors–181, 283 actor−undergoer hierarchy–12 adjacency–26, 29, 160 strict–33 verb–25 adjacency pairs–39, 321, 325, 326, 334 aﬀected–173, 212, 219 agent–2, 52, 173 ﬀ, 189, 230, 231, 279, 282, 286, 289, 300 agents–38, 231ﬀ agreement–64, 110–1, 123, 132, 138, 147–8, 156, 159, 222, 297 acquisition–38, 43 agent agreement in inverse–233 case–101, 153, 157, 159, 369 class–111 clitic–278 ergative–232, 245 ergative–237ﬀ–237 gender–122, 138, 141, 148 hierarchical marking–244 neuter–132 object–262 pronoun–277 subject–164 subject and object–141 vegetable–147 verbal–233 vs concord–101 allative–21, 46, 51, 98, 99, 101, 104 ﬀ animacy–365 and case–21 hierarchy–12, 185 antipassive–364 assumption–11 auxiliarisation–29, 33

auxiliary–29, 31–3, 82 auxiliary verb–19, 29, 31–33, 82, 284, 288–92 and passive–245 phantom auxiliary–305 benefactive–196, 342, 366–7 bound pronoun–42–3, 45, 47–9, 51, 53, 117, 124, 127, 133–5, 250, 254, 255, 277 ﬀ (see also agreement) case–21, 33, 42, 45, 46 attraction–153, 159 causative–2–3, 175, 345 classiﬁer–172, 247 ﬀ clitics–42 comitative–172, 173, 175, 181ﬀ, 342, 345–8, 350, 364, 366 comitative verbs–349–52, 354, 356, 362, 365, 367 copular demonstratives–60 copular sentence–4, 15 core serialisation–208, 209, 221 coreference pattern–295, 352, 361 coreferential deletion–351–2, 354, 360, 363 cross-reference (see agreement) dative–21, 29, 45, 48, 50, 81, 97, 101, 104, 108–11, 113, 159, 163, 256, 270, 356, 366 depency grammar–278 dependency–153, 166, 189, 192, 193, 195, 196 diglossia–73, 77, 86–7, 89 direct voice–231, 233, 235, 236, 239–42 marker–241 discourse factors–76 discourse structure–40, 202, 206, 211, 216, 222, 224, 226 distality–103, 104 eﬀector–1 ellipsis–39, 40, 43, 46, 53, 351 equative–101–3 ergative–229 ﬀ, 280, 283, 286, 292, 341, 348, 351 acquisition–38, 44 alignment–20, 29, 187, 193, 196, 278, 295 alignment–229 ﬀ case–45, 97, 98, 101, 111, 174, 176, 179–81, 277, 279, 281

378

subject index

ergative (cont.) source–244 vs active–298 vs. accusative–280, 282 ergative/hierarchical split–244 essive–98, 99, 104, 108, 109 event reports–212, 213, 216 experiencer–11, 12, 110 functional load–41 gender–64, 68, 115, 117, 119, 125, 127, 131, 132, 134–8, 138, 140, 141, 143 acquisition–38 agreement–148 grammatical–65, 66, 142 natural–66 generic verbs–203, 205 genitive–97, 98, 101, 102, 162, 280, 281, 295, 297, 305, 366 Government and Binding–151, 152, 155, 167, 168 GPSG–156 grammatical vs natural–64 grammaticalization–19, 29–33, 41, 81, 83, 206, 210, 248, 264, 267–9 Head-Driven Phrase Structure Grammar– 151, 152, 167, 168 humanness–46, 60–3, 66, 131, 143, 235 iconicity–59 ﬀ, 222 incorporation locative–6 noun–117, 142 object–267 theme–6, 7 instrumental–101, 171, 197, 244, 280, 342, 369 applicative–171, 173, 189 ﬀ, 198 verb–343, 354 ﬀ interaction–183, 190, 206–7, 309–10, 317–8, 320 interference–72, 74–6, 78–9, 82–90 interleaving–28 intonation unit see prosodic unit intransitive–23 ﬀ, 40, 45–6, 51, 124 ﬀ, 133–4, 145–7, 160–1, 172–3, 187, 190, 198, 229, 233, 236, 240–1, 245, 267, 269, 279, 281–3, 286, 288–9, 291, 293–6, 301, 342, 345–6, 348, 362, 364 inverse–229 ﬀ, 231, 234 marker–240, 241

inversion–160, 311 ﬀ Lexical Functional Grammar–152, 154, 156, 167 lexicalization–41, 141, 205, 221 ﬀ, 349, 354, 362–3 Lexicase–277ﬀ locative–5, 9, 10, 21, 41, 44, 45, 51–3, 109, 211, 218, 220, 266, 346, 350, 353 locative structure–5 locatives–13, 202, 221 markedness–12, 59 ﬀ, 60–2, 89, 194 mediopassive–134 medium–173–5, 177–9, 182, 184–93, 195–7 middle clause–172, 173, 181 motion–31, 41, 44, 51, 52, 97–9, 109, 110, 174–6, 198, 219–21, 223 motion verbs–53, 174 mutation–119, 121, 135 negation–102, 205, 208, 217, 222, 315 ﬀ nominative–20, 29, 65, 67, 81, 157–8, 161, 163–5, 167, 229, 279–82, 284, 288, 290, 292, 295, 297, 341 non-conﬁgurational–21, 42 noun class–124, 129, 134, 140, 146, 148 nuclear serialisation–208, 209 oblique–100, 101, 238, 244, 281 participant vs non-participant roles–173 passive–2, 7, 14, 41, 81, 82, 229 ﬀ patient–4, 7, 211, 223, 229 ﬀ, 278, 279, 281, 282, 286, 289, 297 possession–8, 9, 13, 108, 111, 197, 247, 250, 252, 264–8, 270 possessive classiﬁer–247–8, 250, 252, 253, 255–7, 259, 261, 263–9, 271 possessor–10 preferred argument structure–40 pro-drop–39, 43 prosodic unit–310, 313–5, 318, 321, 322, 326–30 pseudotransitive–283, 289, 291, 293–4, 297–302 purposive–350, 352, 363, 367 recipient–11–4, 111, 196 reciprocal–24, 172–4, 182, 183, 188–90, 194, 364 reﬂexive–125, 134, 172–4, 182–3, 188–90, 192–4, 304, 364 verbs–133

subject index rejections–322, 334 Relational Grammar–152, 167 Role and Reference Grammar–1–4, 6–8, 10, 11, 15, 16, 202, 223, 279 scope–208, 217, 239, 318, 319, 333 semantic role–152, 157, 240, 242, 264 (see also thematic role) Semiotic Grammar–171, 189, 194 serial verb construction–30, 32, 127, 201 ﬀ socio-cultural diversity–87, 89 split-ergative–20, 29 source–244 structure-sharing–151, 162 thematic relations–1 thematic roles–4, 8 theme–6, 10, 13, 297 theta-criterion–152, 153, 161, 168 topic–52, 59, 76, 231, 232, 238, 244, 297, 350, 362 topicalization–50, 238, 239

379

transference–85 transitivity–38, 172, 173, 175, 179, 181, 182, 186, 187, 194, 198, 277 ﬀ, 348, 356 tripartite alignment–236, 237, 245 tripartite/ergative split–237, 244 undergoer–1–3, 7, 11, 12, 173–5, 177, 181 ﬀ, 194, 208, 278 Universal Grammar–37, 39 universals–37, 59, 60, 66, 88, 89, 135, 210, 281, 314 verb control–158 versative–104, 105 volitional activities–345, 365 Word Grammar–153–7, 164 ﬀ word order–3, 21, 29, 38, 43, 45, 49–51, 75, 76, 79, 80, 202, 234, 261, 265–6, 269, 271, 342, 350 zero marking–42–3, 45, 51, 53, 67, 97, 104, 106, 229–31, 234, 240, 245, 344, 348

In the series TYPOLOGICAL STUDIES IN LANGUAGE (TSL) the following titles have been published thus far: 1. HOPPER, Paul J. (ed.): Tense-Aspect: Between semantics & pragmatics. 1982. 2. HAIMAN, John & Pamela MUNRO (eds): Switch Reference and Universal Grammar. Proceedings of a symposium on switch reference and universal grammar, Winnipeg, May 1981. 1983. 3. GIVÓN, T.: Topic Continuity in Discourse. A quantitative cross-language study. 1983. 4. CHISHOLM, William, Louis T. MILIC & John A.C. GREPPIN (eds): Interrogativity: A colloquium on the grammar, typology and pragmatics of questions in seven diverse languages, Cleveland, Ohio, October 5th 1981-May 3rd 1982. 1984. 5. RUTHERFORD, William E. (ed.): Language Universals and Second Language Acquisition. 1984 (2nd ed. 1987). 6. HAIMAN, John (Ed.): Iconicity in Syntax. Proceedings of a symposium on iconicity in syntax, Stanford, June 24-26, 1983. 1985. 7. CRAIG, Colette (ed.): Noun Classes and Categorization. Proceedings of a symposium on categorization and noun classification, Eugene, Oregon, October 1983. 1986. 8. SLOBIN, Dan I. & Karl ZIMMER (eds): Studies in Turkish Linguistics. 1986. 9. BYBEE, Joan L.: Morphology. A Study of the Relation between Meaning and Form. 1985. 10. RANSOM, Evelyn: Complementation: its Meaning and Forms. 1986. 11. TOMLIN, Russel S.: Coherence and Grounding in Discourse. Outcome of a Symposium, Eugene, Oregon, June 1984. 1987. 12. NEDJALKOV, Vladimir (ed.): Typology of Resultative Constructions. Translated from the original Russian edition (1983). English translation edited by Bernard Comrie. 1988. 14. HINDS, John, Shoichi IWASAKI & Senko K. MAYNARD (eds): Perspectives on Topicalization. The case of Japanese WA. 1987. 15. AUSTIN, Peter (ed.): Complex Sentence Constructions in Australian Languages. 1988. 16. SHIBATANI, Masayoshi (ed.): Passive and Voice. 1988. 17. HAMMOND, Michael, Edith A. MORAVCSIK and Jessica WIRTH (eds): Studies in Syntactic Typology. 1988. 18. HAIMAN, John & Sandra A. THOMPSON (eds): Clause Combining in Grammar and Discourse. 1988. 19. TRAUGOTT, Elizabeth C. and Bernd HEINE (eds): Approaches to Grammaticalization, 2 volumes (set) 1991 20. CROFT, William, Suzanne KEMMER and Keith DENNING (eds): Studies in Typology and Diachrony. Papers presented to Joseph H. Greenberg on his 75th birthday. 1990. 21. DOWNING, Pamela, Susan D. LIMA and Michael NOONAN (eds): The Linguistics of Literacy. 1992. 22. PAYNE, Doris (ed.): Pragmatics of Word Order Flexibility. 1992. 23. KEMMER, Suzanne: The Middle Voice. 1993. 24. PERKINS, Revere D.: Deixis, Grammar, and Culture. 1992. 25. SVOROU, Soteria: The Grammar of Space. 1994. 26. LORD, Carol: Historical Change in Serial Verb Constructions. 1993. 27. FOX, Barbara and Paul J. Hopper (eds): Voice: Form and Function. 1994. 28. GIVÓN, T. (ed.) : Voice and Inversion. 1994. 29. KAHREL, Peter and René van den BERG (eds): Typological Studies in Negation. 1994.

30. DOWNING, Pamela and Michael NOONAN: Word Order in Discourse. 1995. 31. GERNSBACHER, M. A. and T. GIVÓN (eds): Coherence in Spontaneous Text. 1995. 32. BYBEE, Joan and Suzanne FLEISCHMAN (eds): Modality in Grammar and Discourse. 1995. 33. FOX, Barbara (ed.): Studies in Anaphora. 1996. 34. GIVÓN, T. (ed.): Conversation. Cognitive, communicative and social perspectives. 1997. 35. GIVÓN, T. (ed.): Grammatical Relations. A functionalist perspective. 1997. 36. NEWMAN, John (ed.): The Linguistics of Giving. 1998. 37. RAMAT, Anna Giacalone and Paul J. HOPPER (eds): The Limits of Grammaticalization. 1998. 38. SIEWIERSKA, Anna and Jae Jung SONG (eds): Case, Typology and Grammar. In honor of Barry J. Blake. 1998. 39. PAYNE, Doris L. and Immanuel BARSHI (eds.): External Possession. 1999. 40. FRAJZYNGIER, Zygmunt and Traci S. CURL (eds.): Reflexives. Forms and functions. 2000. 41. FRAJZYNGIER, Zygmunt and Traci S. CURL (eds): Reciprocals. Forms and functions. 2000. 42. DIESSEL, Holger: Demonstratives. Form, function and grammaticalization. 1999. 43. GILDEA, Spike (ed.): Reconstructing Grammar. Comparative Linguistics and Grammaticalization. 2000. 44. VOELTZ, F.K. Erhard and Christa KILLIAN-HATZ (eds.): Ideophones. n.y.p. 45. BYBEE, Joan and Paul HOPPER (eds.): Frequency and the Emergence of Linguistic Structure. 2001. 46. AIKHENVALD, Alexandra Y., R.M.W. DIXON and Masayuki ONISHI (eds.): Noncanonical Marking of Subjects and Objects. 2001. 47. BARON, Irene, Michael HERSLUND and Finn SORENSEN (eds.): Dimensions of Possession. n.y.p. 48. SHIBATANI, Masayoshi (ed.): The Grammar of Causation and Interpersonal Manipulation. n.y.p. 49. WISCHER, Ilse and Gabriele DIEWALD (eds.): New Reflections on Grammaticalization. n.y.p.

Case, Typology and Grammar (Typological Studies in Language)

Creoles, their Substrates, and Language Typology (Typological Studies in Language)

Studies in African Linguistic Typology (Typological Studies in Language)

Passivization And Typology: Form And Function (Typological Studies in Language)

Clusivity: Typology and Case Studies of Inclusive-Exclusive Distinction (Typological Studies in Language)

A Typology of Purpose Clauses (Typological Studies in Language)

Benefactives and Malefactives: Typological Perspectives and Case Studies (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Prototypical Transitivity (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Coordinating Constructions (Typological Studies in Language)

Dimensions of Possession (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Deconstructing Creole (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Interaction of Morphology and Syntax: Case studies in Afroasiatic (Typological Studies in Language)

Studies in Evidentiality: v. 54 (Typological Studies in Language)

Theoretical and Empirical Issues in Grammaticalization (Typological Studies in Language)

Reflexives: Forms and Functions (Typological Studies in Language)

Language Complexity: Typology, contact, change (Studies in Language Companion Series)

Adpositions: Pragmatic, Semantic and Syntactic Perspectives (Typological Studies in Language)

The Grammar of Causation and Interpersonal Manipulation (Typological Studies in Language)

The Grammar of Causation and Interpersonal Manipulation (Typological Studies in Language)

New Reflections on Grammaticalization Vol 49 (Typological Studies in Language)

The Linguistics of Giving (Typological Studies in Language)

Modality-Aspect Interfaces: Implications and typological solutions (Typological Studies in Language)

Rethinking Grammaticalization: New perspectives (Typological Studies in Language)

Formulaic Language, Vol. 1: Distribution and historical change (Typological Studies in Language)

Recursion and Human Language (Studies in Generative Grammar)

Formulaic Language, Vol. 2: Acquisition, loss, psychological reality, and functional explanations (Typological Studies in Language)

Language Typology: A Functional Perspective

Case, Typology and Grammar (Typological Studies in Language)

Creoles, their Substrates, and Language Typology (Typological Studies in Language)

Studies in African Linguistic Typology (Typological Studies in Language)

Passivization And Typology: Form And Function (Typological Studies in Language)

Clusivity: Typology and Case Studies of Inclusive-Exclusive Distinction (Typological Studies in Language)

A Typology of Purpose Clauses (Typological Studies in Language)

Benefactives and Malefactives: Typological Perspectives and Case Studies (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Prototypical Transitivity (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Coordinating Constructions (Typological Studies in Language)

Dimensions of Possession (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Deconstructing Creole (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Interaction of Morphology and Syntax: Case studies in Afroasiatic (Typological Studies in Language)

Studies in Evidentiality: v. 54 (Typological Studies in Language)

Theoretical and Empirical Issues in Grammaticalization (Typological Studies in Language)

Reflexives: Forms and Functions (Typological Studies in Language)

Language Complexity: Typology, contact, change (Studies in Language Companion Series)

Adpositions: Pragmatic, Semantic and Syntactic Perspectives (Typological Studies in Language)

The Grammar of Causation and Interpersonal Manipulation (Typological Studies in Language)

The Grammar of Causation and Interpersonal Manipulation (Typological Studies in Language)

New Reflections on Grammaticalization Vol 49 (Typological Studies in Language)

The Linguistics of Giving (Typological Studies in Language)

Modality-Aspect Interfaces: Implications and typological solutions (Typological Studies in Language)

Rethinking Grammaticalization: New perspectives (Typological Studies in Language)

Formulaic Language, Vol. 1: Distribution and historical change (Typological Studies in Language)

Recursion and Human Language (Studies in Generative Grammar)

Formulaic Language, Vol. 2: Acquisition, loss, psychological reality, and functional explanations (Typological Studies in Language)

Language Typology: A Functional Perspective

Recommend Documents