Language Universals and Variation

cover next page > Cover title: author: publisher: isbn10 | asin: print isbn13: ebook isbn13: language: subject public...

Author: Mengistu Amberber | Peter Collins (Editors)

85 downloads 1612 Views 1MB 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

cover

next page >

Cover

title: author: publisher: isbn10 | asin: print isbn13: ebook isbn13: language: subject publication date: lcc: ddc: subject:

Language Universals and Variation Perspectives On Cognitive Science Amberber, Mengistu; Collins, Peter Greenwood Publishing Group 0275976823 9780275976828 9780313012488 English Language and languages--Variation, Typology (Linguistics) , Grammar, Comparative and general. 2002 P120.V37L348 2002eb 417/.2 Language and languages--Variation, Typology (Linguistics) , Grammar, Comparative and general.

cover

next page >

< previous page

page_i

next page >

page_i

next page >

Page i Language Universals and Variation

< previous page

< previous page

page_ii

next page >

Page ii Recent Titles in Perspectives on Cognitive Science Perspectives on Cognitive Science, Volume 1: Theories, Experiments, and Foundations Peter Slezak, Terry Caelli, and Richard Clark Perspectives on Cognitive Science, Volume 2: Theories, Experiments, and Foundations Janet Wiles and Terry Dartnall, editors Creativity, Cognition and Knowledge: An Interaction Terry Dartnall, editor

< previous page

page_ii

next page >

< previous page

page_iii

next page >

page_iii

next page >

Page iii Language Universals and Variation Edited by Mengistu Amberber and Peter Collins Perspectives on Cognitive Science Peter Slezak, Series Editor

< previous page

< previous page

page_iv

next page >

Page iv Library of Congress Cataloging-in-Publication Data Language universals and variation/edited by Mengistu Amberber and Peter Collins. p. cm—(Perspectives on cognitive science) Includes bibliographical references and index. ISBN 0-275-97682-3 (alk. paper)—ISBN 0-275-97683-1 (pbk.: alk. paper) 1. Language and languages—Variation. 2. Typology (Linguistics) 3. Grammar, Comparative and general. I. Amberber, Mengistu, 1961– II. Collins, Peter P120.V37 L348 2002 471′.2–dc21 2001058032 British Library Cataloguing in Publication Data is available. Copyright © 2002 by Mengistu Amberber and Peter Collins All rights reserved. No portion of this book may be reproduced, by any process or technique, without the express written consent of the publisher. Library of Congress Catalog Card Number: 2001058032 ISBN: 0-275-97682-3 0-275-97683-1 (pbk.) First published in 2002 Praeger Publishers, 88 Post Road West, Westport, CT 06881 An imprint of Greenwood Publishing Group, Inc. www.praeger.com Printed in the United States of America The paper used in this book complies with the Permanent Paper Standard issued by the National Information Standards Organization (Z39.48–1984). 10 9 8 7 6 5 4 3 2 1

< previous page

page_iv

next page >

< previous page

page_v

Page v Contents Preface 1 Quirky Alternations of Transitivity: The Case of Ingestive Predicates Mengistu Amberber Ingestives in a Cross-Linguistic Perspective Ingestives and Ambitransitivity Ingestives as Three-Place Predicates 2 Explaining Clitic Variation in Spanish José Camacho and Liliana Sánchez Overview of the Third Person Clitic Paradigm in Spanish Overview of the Etymological and the Referential Dialects Accounting for the Etymological Dialect and Referential A Dialects Contact Dialects 3 Slavic Passives, Bantu Passives, and Human Cognition Peter F.Kipka A Framework Slavic Bantu Small Clauses Prototypicality 4 The Split VP Hypothesis: Evidence from Language Acquisition Masatoshi Koizumi

< previous page

page_v

next page > ix 1 4 10 12 21 22 22 25 28 41 42 45 49 54 56 61

next page >

< previous page

page_vi

Page vi The Split VP Hypothesis Preverbal Objects A Split VP Account Further Prediction Clausal Architecture 5 Syntactic Constraints in a “Free Word Order” Language Mary Laughren Composition of Warlpiri AUX Syntactic Constraints on the Position of AUX Negative AUX A Comparative Overview 6 On the Range and Variety of Cases Assigned by Adpositions Alan R.Libert Type I—Languages with One (Main) Adpositional Case Type II—Languages with More Than One Adpositional Case Toward a Minimalist Account of Adpositional Case 7 Optimality and Three Western Austronesian Case Systems Anna Maclachlan Background: Case System Typology, Optimality and Austronesian Three Western Austronesian Case Systems in OT 8 Affixes, Clitics, and Bantu Morphosyntax Sam Mchombo Morpholexical versus Morphosyntactic Processes Verbal Suffixation Verbal Prefixation Affixes versus Clitics Clitics and Inflectional Morphology On the Architecture of Universal Grammar Acquired Language Deficit Language Change Language Acquisition Parsing Strategies for Bantu 9 Two Types of Wh -In-Situ Masanori Nakamura Lexical Properties Syntactic Properties Toward a Nonunitary Account Null Operator Movement as Feature Movement 10 Vowel Place Contrasts Keren Rice

< previous page

page_vi

next page > 63 67 71 74 75 83 86 92 112 117 131 132 143 150 155 156 160 185 186 188 192 193 194 197 199 201 202 204 211 213 215 219 225 239

next page >

< previous page

page_vii

Page vii Evidence for Peripheral The Phonetic Realization of Peripheral Vowels Conclusions and Consequences Author Index Index of Languages and Language Families Subject Index About the Editors and Contributors

< previous page

page_vii

next page > 241 260 263 271 273 275 279

next page >

< previous page

page_viii

next page >

page_viii

next page >

Page viii This page intentionally left blank.

< previous page

< previous page

page_ix

next page >

Page ix Preface In mid-1999, Peter Slezak, the Series Editor of Perspectives on Cognitive Science, invited us to edit a linguistics volume addressing current issues and problems in the analysis of cross-linguistic data. We invited colleagues from Australia and overseas whose areas of expertise include phonology, lexical semantics, and morphosyntax to contribute a paper on the broad theme of formal approaches to language universals and variation. Most of the chapters in this volume investigate aspects of natural language variation from a formal theoretical perspective, including the Principles and Parameters/Minimalist Program, Lexical Functional Grammar, and Optimality Theory. It is assumed that the reader will have at least some basic familiarity with these frameworks. We thank all the authors for contributing to this volume. We would also like to thank David Adger, Heidi Harley, Larry Hyman, Géraldine Legendre, Andrew Radford, Peter Sells, Roumyana Slabakova, Margaret Speas, and Lindsay Whaley for their invaluable advice as reviewers. Thanks are also due to Peter Slezak for his encouragement, Debbie Carvalko, our acquisitions editor at Greenwood Publishing Group, for her professional help, and Marc Peake for his diligent assistance in preparing the volume for publication.

< previous page

page_ix

next page >

< previous page

page_x

next page >

page_x

next page >

Page x This page intentionally left blank.

< previous page

< previous page

page_xi

next page >

page_xi

next page >

Page xi Language Universals and Variation

< previous page

< previous page

page_xii

next page >

page_xii

next page >

Page xii This page intentionally left blank.

< previous page

< previous page

page_1

next page >

Page 1 1 Quirky Alternations of Transitivity: The Case of Ingestive Predicates Mengistu Amberber In a number of languages, verbs of ingestion—including verbs roughly equivalent to the English verbs eat, drink, swallow, taste, suckle —exhibit marked transitivity behavior. In languages where a causativizing morpheme is otherwise restricted to attach to intransitive verbs it can exceptionally appear with transitive ingestive verbs. It is unlikely that this phenomenon is due to an accidental property of individual languages given that it occurs in many genetically and typologically unrelated languages. This chapter will show that the marked transitivity pattern exhibited by ingestive predicates can be accounted for by appealing to a rich Lexical Conceptual Structure (LCS). I argue that contrary to appearance, ingestive predicates such as the English eat are ditransitive with an Agent, Theme/Patient, and Goal argument. The crucial property of ingestive predicates that is responsible for their marked transitivity pattern is that the Agent and Goal arguments are co-indexed at the level of LCS and thus normally only one argument (the Agent) is mapped onto the surface syntax. Due to the co-indexation of the Agent and Goal arguments, it is possible to suppress the Agent argument—as it is recoverable from the Goal— thus allowing the introduction of another Agent via morphological causativization. It is known that many languages employ a morphological strategy to encode transitivity alternations (cf. Comrie and Polinsky, 1993). A transitive verb may be derived by attaching an affix onto an intransitive stem. Likewise, transitive stems can be detransitivized via affixation, thus deriving intransitive constructions such as the anticausative, middle, passive, or antipassive among others. The class of verbs that may be affected by causativization or detransitivization may differ from language to

< previous page

page_1

next page >

< previous page

page_2

next page >

Page 2 language. For instance, in many languages the passive derivation typically applies to dynamic transitive verbs (e.g., cut -> be cut ) but not to intransitive verbs (walk -> * be walked) . However, in a number of languages intransitive verbs can be passivized, giving a construction known as impersonal passive (or “pseudo-passive”), as in Dutch (cf. Kirsner, 1976). Analogous variation can be found in the case of causativization. In some languages, the causative morpheme is attached only to an intransitive stem to derive a transitive verb, whereas in other languages the causative morpheme may attach to both intransitive and transitive stems, in the latter case deriving a ditransitive verb (cf. Haspelmath, 1993). From a broad typological perspective, it appears that there are two types of languages with respect to the distribution of causative morphemes: languages with a single causativizing morpheme (Type A) and languages with two (perhaps more) causativizing morphemes (Type B). Type A languages can be further subclassified in terms of whether the causative morpheme can attach to only intransitives (Type A1) or to both intransitive and transitives (Type A2). In Type B languages, we find one causative morpheme exclusively for causativizing intransitive verbs and another causative morpheme that can causativize both intransitive and transitive verbs (deriving a “double” causative in the latter case). The classification is summarized in (1) below with some examples of representative languages. (1) Causative Type Language Type A: a single CAUS Type A1: CAUS + √Intr Berber Type A2: CAUS + √Intr/Tr Malayalam Type B: two CAUS (CAUS1 + √Intr; CAUS2 + √Intr/Tr) Amharic When languages have two or more causativizing morphemes, their distribution is often circumscribed: One causative morpheme attaches only to intransitive verbs whereas the other causative morpheme attaches to both intransitive and transitive verbs.1 For example, Amharic has two productive morphological causative prefixes: a- and as- . The causative a- attaches to intransitive verbs as in (2), whereas the causative as- can attach to both intransitive and transitive verbs as in (3):2 (2) a. k’om ‘stand (intr)’ a-k’om ‘stand (tr)’ b. k’ ll t’ ‘melt (intr)’ a-k’ ll t’ ‘melt (tr)’ c. k’W rr t’ ‘cut’ *a-k’W rr t’ (3) a. m t’t’a ‘come’ as-m t’t’a ‘make x come’ b. k’W rr t’ ‘cut’ as-k’w rr t’ ‘make x cut y’

< previous page

page_2

next page >

< previous page

page_3

next page >

Page 3 The fact that the causative a- cannot attach to transitive verbs can be seen in the ill-formed derivation in (2c). Therefore, in Amharic the distribution of the two causative affixes is predictable. While the causative affix as- can attach to either a transitive or an intransitive stem, the causative affix a- can attach to intransitive stems only. However, further investigation reveals that there is one exception to this generalization. Some verbs exhibit an unexpected causativization pattern: They can take the causative affix a- despite the fact that they are already transitive. For example, consider the verb b lla “eat.” This verb is transitive as it occurs with two arguments—Agent and Patient. As would be expected for any other transitive verb, this verb can be causativized by the causative prefix as-, deriving the factive meaning “cause someone to eat” as in (4): (4) aster l mma-n dabbo as-b lla-čč-iw A. L.-ACC bread CAUS-eat.PF.-3F-3MO ‘Aster made Lemma eat some bread.’ Now given that the verb which means “to eat” in (4) is transitive and that transitives do not take the causative prefix a-, we would not expect the verb b lla “eat” to occur with a- and yet this is precisely what we find in (5): (5) aster l mma-n dabbo a-b lla-čč-iw A. L.-ACC bread CAUS-eat.PF.-3F-3MO ‘Aster fed Lemma some bread.’ The verbs that behave like b lla “eat” are very few. There are about 10 or so verbs, and they are listed in (6) below (see also Demoz, 1964; Leslau, 1995): (6) a. b lla ‘eat’ a-b lla ‘feed’ b. t’ t’t’a ‘drink’ a- t’ t’t’a ‘give to drink’ c. las ‘lick’ a-las ‘give to lick’ d. t’ bba ‘suck’ a-t’ bba ‘to suckle’ e. k’ mm s ‘taste’ a-k’ mm s ‘give to taste’ f. l k’k’ m ‘pick up’ a-l k’k’ m ‘graze’ g. gw rr s ‘take a mouthful’ a-gw rr s ‘give a mouthful’ h. wat’ ‘swallow’ a-wat’ ‘give to swallow’ i. k’am ‘eat large mouthfuls of’ a-k’am ‘give large mouthfuls of grain’ grain j. gat’ ‘graze’ a-gat’ ‘let graze’ A closer look at the meaning of the class of verbs that exhibit the unexpected causativization pattern reveals that they share a common

< previous page

page_3

next page >

< previous page

page_4

next page >

Page 4 semantic core. Indeed, Demoz (1964) classifies them under a single heading “ingestive” as the event expressed by the verbs has something to do with taking food or edible substance. A number of interesting questions arise regarding ingestive verbs including the following: (a) Is there any cross-linguistic evidence to support the assumption that the verbs in question form a distinct lexical semantic class? (b) What are the lexical semantic properties of the ingestive verbs that warrant their quirky causativization behavior? These and related questions are addressed in this chapter. The rest of the chapter is organized as follows. In the next section some cross-linguistic evidence regarding the class of ingestive verbs and other similar verbs that exhibit a marked causativization pattern is presented. In the section Ingestives and Ambitransitivity, the ambitransitive nature of ingestives is discussed. In the section Ingestives as Three-Phase Predicates, a lexical-semantic analysis of the ingestive predicates is proposed mainly on the basis of the Amharic data. INGESTIVES IN A CROSS-LINGUISTIC PERSPECTIVE Ingestive predicates seem to exhibit marked behavior in a number of genetically and typologically diverse languages including Malayalam (Mohanan, 1983:105–106), Berber (Guerssel, 1986:36ff), Tariana (Aikhenvald, 2000), Jarawara (Dixon, 2000), and Chichewa (Baker, 1988:461). In the Dravidian language Malayalam (Mohanan, 1983), there is a productive causativization process that derives causative predicates both from intransitives and transitives. However, intransitives and transitives differ in the syntactic realization of the causee. In the causativization of intransitive verbs, the causee (the original subject) becomes a primary object marked by accusative case, whereas in the causativization of transitive verbs, the causee occurs in an instrumental phrase. Thus, consider the following contrast (from Mohanan, 1983:58–59): (7) a. ku i ka aññu child-N cried ‘The child cried.’ b. acchan ku iye ka ay-icc-u Father-N child-A cry-CAUSE-PAST ‘The father made the child cry.’ (8) a. ku aanaye nu i child-N elephant-A pinched ‘The child pinched the elephant.’

< previous page

page_4

next page >

< previous page

page_5

next page >

Page 5 b. amma ku iye-kkonte aanaye u -icc-u mother-N child-with elephant-A pinch-CAUSE-PAST ‘The mother made the child pinch the elephant.’ In the causative of a transitive verb, as in (8b), the causee must appear as an instrumental with the postposition konte “with.” The only exception to the above generalization comes from a small class of transitive verbs that Mohanan also refers to as ingestive. Consider the following examples (Mohanan, 1983:105): (9) a. ku i coor i u child-N rice-N ate ‘The child ate the rice.’ b. amma ku iye coor iiti mother-N child-A rice-N eat-CAUSE-PAST ‘The mother fed the child rice.’ As (9b) shows, the causee of the verb i “eat” behaves as the causee of an intransitive verb: It occurs with the accusative case instead of the instrumental adposition. Thus, even though the verb is transitive, its causativization pattern is that of an intransitive verb. Mohanan does not offer any explanation for what he calls the “mystery of ingestive verbs” (Mohanan, 1983:106). It is interesting that the Malayalam ingestive class includes not only verbs of eating, such as i “eat,” ku ikk “drink” but also verbs such as kaa “see,” and pa hikk “learn.” Mohanan (1983:106) notes that in the Dravidian literature the term “ingestive” is used to encode the meaning of “taking something either literally or metaphorically.” According to Mohanan, this class of verbs exhibits similar behavior in other Indian languages as well. Apparently, the existence of the ingestive class of verbs had been noted as far back as Panini in the study of classical Sanskrit. The ingestive verbs also exhibit unexpected patterns of transitivity alternation in Berber, an Afroasiatic language (Guerssel, 1986). Berber has a productive morphological process that derives causative verbs from intransitive verbs. Thus, according to Guerssel (1986), “active” (unergative) monadic verbs such as bedd “stand” and “stative” (unaccusative) monadic verbs such as zyert ‘be long’ can be causativized by the causative prefix ss- (Guerssel, 1986:14–15): (10) a. y-bedd wrba 3ms-stand boy:cst ‘The boy stood up.’

< previous page

page_5

next page >

< previous page Page 6

page_6

next page >

y-ss-bedd wryaz arba 3ms-TRANS-stand man:cst boy ‘The man made the boy stand up.’ (11) a. y-zyert wfuli 3ms-be long string-cst ‘The string is long.’ b. y-ss-zyert wrba fuli 3ms-TRANS-belong boy-cst string ‘The boy lengthened the string.’ On the other hand, causativization cannot apply to typical transitive verbs such as wt “hit” (Guerssel 1986:18): (12) *y-ss-wt wmddakkwl-inw mucc aryaz 3ms-CAUSE-hit friend:cst-my cat man ‘My friend made the man hit the cat.’ The only exception to the generalization that Berber transitive verbs cannot be causativized comes from a class of verbs that Guerssel (1986:36) refers to as the eat class, which includes verbs such as ttc “eat” sw “drink” jjawn “be satiated with food” and tted “suckle.” Consider the examples in (13)–(14): (13) a. Y-ttcu wqqzin 3ms-eat dog:cst ‘The dog ate.’ b. Y-ss-ttc wryaz aqqzin 3ms-TRANS-eat man:cst dog ‘The man fed the dog.’ (14) a. Y-ttcu wqqzin aysum 3ms-eat dog:cst meat ‘The dog ate the meat.’ b. Y-ss-ttc wryaz aysum i-wqqzin 3ms-TRANS-eat:per man:cst meat dat-dog:cst ‘The man fed meat to the dog.’ Notice that, like most other languages, the Berber verb ttc “eat” can be used intransitively, as in (13a). However, the interesting example is (14b), where the transitive variant of the verb ttc “eat” is causativized, in a marked departure from the causativization pattern of Berber. b.

< previous page

page_6

next page >

< previous page

page_7

next page >

Page 7 In Tariana (North-Arawak), morphological causatives typically attach to intransitive verbs (Aikhenvald, 2000:154–155). In the following examples the affix -i(ta) is employed to derive transitive verbs from intransitives: (15) a. -eku ‘run’ -eku-ita ‘make run’ b. -yena ‘pass’ -yeneta (
< previous page

page_7

next page >

< previous page

page_8

next page >

Page 8 (20) okaki owa na-jana IsgPOSS+grandmother(f) IsgO CAUS-grow up + f ‘My grandmother brought me up.’ According to Dixon (2000), the causative prefix is normally attached to intransitive verbs. However, it is “occasionally used with a transitive verb.” The only example given by Dixon refers to the ingestive meaning “to drink” (Dixon, 2000:28): (21) a. inamatewe remejo fawa-ke child (f) medicine (f) drink-DECf ‘The child (feminine) drinks the medicine.’ b. inamatewe mati na-fawa-ke child (f) 3sgPOSS+mother(f) CAUS-drink-DECf remejo jaa medicine PeRI ‘Her mother made the child drink the medicine.’ It is interesting to note that in another Amazonian language, Urubu-Kaapor (Tupí-Guaraní, cf. Kakumasu, 1986:342), causatives are formed by attaching the prefix mu- to an intransitive verb (and also to some nouns). It appears that in Urubu-Kaapor, the causative does not attach to a transitive verb. According to Kakumasu (1986:342), the equivalent of “the mother caused the child to eat” is expressed either by “paraphrasing in terms of commanding (…) or using a different verb, e.g. jopói “give food to” (used of babies, pets, etc.)”: (22) a. e-’u ma’e aja tipe i-mai 2SQ.IMP-eat something thus FRUST 3-mother ‘His mother said in vain, “Eat something”.’ b. ihe rendyr riki ma’e so’o te’e jopói reko my sister EMPH some game freely 3+feed AUX ‘My sister is feeding some kind of game (animal).’ In Apalai (Carib family, cf. Koehn and Koehn, 1986:50–51), some intransitive verbs can take two causative suffixes. The only example given by Koehn and Koehn (1986) involves the ingestive verb otuh “eat.” This verb can be transitivized by the causative suffix -ma: (23) a. poeto otuh-noko mana child eat-CONT 3+be+PRES ‘The child is eating.’

< previous page

page_8

next page >

< previous page

page_9

next page >

Page 9 b. poeto otuh-ma-Vko mana child eat-CAUS-CONT 3+be+PRES ‘He is feeding the child.’ The verb so transitivized can further take a second causative suffix: (24) aimo otuh-ma-po-Vko mana boy eat-CAUS-CAUS-CONT 3+be+PRES ‘He is getting (someone) to feed the boy.’ Although the meaning that corresponds to the English feed seems to be a causative of the transitive eat, languages can lexicalize this meaning so that the verb can be predicated of only certain arguments. Thus, in Sre (Mon Kmer) the verb po means “to feed at the breast” and is predicated of infants (cf. Manly, 1972:44). Interestingly, this verb is intransitive—along with verbs such as dùn “to fall,” and lik “to come out”—and can be causativized by the prefix t n- and the derived meaning is “to suckle”: (25) a. po ‘to feed at the breast (of infants)’ b. t n-po ‘to suckle’ It is clear therefore that in several languages some verbs appear to behave in a marked manner with respect to causativization processes or some morphosyntactic operations. In Amharic, the causative affix a- normally takes intransitive verbs. However, this affix can exceptionally take transitive ingestive verbs. In Malayalam, the causee of a transitive verb is always realized as an instrumental. With the exception of ingestive verbs, there is no other transitive verb whose causee can appear in an accusative case. In Berber, transitive verbs cannot be causativized. The only exception to this comes from ingestive verbs. In Tariana, the morphological causative is normally based on intransitive verbs. In a few instances, however, the morphological causative can be based on transitive verbs. Most of the verbs that belong to this exceptional class of verbs refer to ingestion. Likewise, in Jarawara, while the unmarked use of the causative morpheme is with intransitive verbs, it can occasionally occur with some transitives such as the verb meaning “to drink”—a typical ingestive predicate. The cross-linguistic facts suggest that there are three ways of deriving the causative of ingestive predicates (not surprisingly parallelling the three types of causativization strategies (cf. Comrie, 1989). (26)a.Lexical (suppletive forms are employed as in English) b.Morphological (a causative morpheme is employed even if it means relaxing the selectional restriction on the causative affix as in Amharic)

< previous page

page_9

next page >

< previous page

page_10

next page >

Page 10 c. Periphrastic (an independent verb, e.g., give, is employed as in Urubu-Kaapor) The cross-linguistic behavior of ingestive verbs militates against analyzing the verbs in question as quirks of individual languages. Thus, a more satisfactory reason should be provided by investigating the semantic property of the verbs. INGESTIVES AND AMBITRANSITIVITY The Implicit Object One characteristic property of ingestive verbs in general is the fact that they can be both transitive, as in (27a), and intransitive, as in (27b): (27) a. John ate the sandwich b. John ate The verb eat in (27a) occurs with its patient argument whereas in (27b) it occurs as a monadic predicate with only one argument. Such ambitransitivity is also found in Amharic: (28) a. l mma dabbo b lla L. bread eat.PF.3M ‘Lemma ate some bread.’ b. l mma b lla L. eat.PF.3M ‘Lemma ate.’ Even though the (b) sentences in (27) and (28) appear to be intransitive, it is the intuition of speakers that there is an implicit object argument that is prototypically understood as something edible or more specifically as a meal. Thus, cross-linguistically ingestive verbs that occur without their patient argument are interpreted with an implicit object. Even in languages in which in transitivity is marked by morphology, such verbs are semantically transitive. For example, in Menomini (Algonquian, cf. Bloomfield, 1946) verbs are classified as intransitive and transitive. The verb meaning “to drink” is morphologically intransitive. According to Bloomfield (1946:94–95) such verbs make sense only in a syntactically transitive frame. He says “some intransitive verbs are used habitually with implied goals thus (…) menuah ‘he drinks (it)’ is intr. [ansitive] in form, but in general makes sense only with a pseudo-object: nepeew menuah ‘he drinks

< previous page

page_10

next page >

< previous page

page_11

next page >

Page 11 some water,’” (emphasis in the original). In Onondaga (Iroquoian), the verb meaning “to eat” can occur without the patient as in the sentence cihá í.weks “the dog eats.” According to Chafe (1970:10) this sentence is ambiguous as “it might mean ‘the dog eats it’ as well.” In languages that use a special marker for unspecified objects, a transitive verb does not have to occur with an object NP. For example, in Pipil (Uto-Aztecan) the prefix ta- on the verb indicates that the object is unspecified. According to Campbell (1985:77) verbs with the prefix ta- “are translated with an object ‘something’ or ‘to be doing’ whatever the action of the verb is, without specifying what the object is.” Hence, it would not be surprising if in languages like Pipil verbs such as ‘to eat’ can occur without an object while maintaining their transitivity. Although the problem of ingestive verbs is noted in some studies, no systematic analysis of the problem has been proposed. A notable exception is Guerssel (1986) who offered an account of the ingestive verbs in Berber. The Lexical Structure of Ingestives Guerssel (1986:6) assumes a framework that recognizes a level of Lexical Conceptual Structure (LCS) that represents the meaning of a verb, and a level of Lexical Structure (LS) that is “the lexical projection of the category verb.” The two representations are related by “a set of linking conventions that associate the variables in LCS to argument positions in LS.” The LCS and LS together are referred to as the Predicate Argument Structure (PAS) of a verb. To account for the problem of ingestive verbs, Guerssel (1986) begins with the assumption that the ingestive verbs have Agent and Patient semantic roles. He argues that the LCS of ttc “eat” contains a clause that identifies the patient variable, as in (30): (29) LCS of ttc ‘eat’ x EAT y, where y is typically FOOD Guerssel proposes that the patient role in the LCS is not obligatorily linked to an argument position in the LS. Thus, depending on whether or not the patient argument is linked, there are two PAS representations for ttc “eat” (Guerssel, 1986:37):

< previous page

page_11

next page >

< previous page

page_12

next page >

Page 12 The PAS representations in (30a) and (30b) are that of the intransitive and transitive “eat,” respectively. Guerssel argues that the causativization rule cannot apply to any transitive PAS, including (30b), but there is no reason why it cannot apply to (30a). The basic idea is that the eat verbs, by virtue of their lexical properties, have a patient role that is not linked into a position in LS. Due to these properties, the eat verbs can behave as intransitive for the purpose of causativization. Guerssel argues that the crucial difference between the eat verbs and other transitive verbs such as hit is that the latter cannot have a PAS like (30a) and as a result cannot be causativized. Guerssel’s (1986) analysis regarding the grammatical function of the arguments in the causativized “eat” (30b) is problematic. Notice that in Berber the Agent of the basic verb is realized as a dative argument. There is no reason why this argument is not realized as the object of the derived verb. To account for this problem, Guerssel (1986:39) invokes the notion of “passive participant”: an argument that is a passive participant in a given activity is mapped onto the object position. Guerssel (1986) stipulates that in (30b), although the Agent argument of the lower verb is a passive participant relative to the external causer, the Patient argument of the lower verb is a “more” passive participant than the Agent argument and as a result it is the Patient that can be mapped onto object position. One problem with Guerssel’s notion of passive participant is that it is not independently determined but is rather evaluated relative to other arguments. Furthermore, it would be difficult to transfer the notion of a passive participant into the analysis of other languages such as Amharic and Malayalam, where it is the causee (not the patient of the lower verb) that is mapped onto the object position. Thus, it would be desirable to derive the effect of passive participant from other independently motivated principles of grammar. In the next section, I will motivate an account of the ingestive verbs on the basis of a more articulated LCS. I will establish that the important property of ingestive verbs is not only the presence of an optional Theme/Patient argument but also the presence of a Goal argument that is co-referential with the Agent. INGESTIVES AS THREE-PLACE PREDICATES To investigate the meaning of the verb “eat,” we need to look at the various components that are present in its LCS. Let us begin with the LCS of eat proposed in Jackendoff (1990). According to Jackendoff (1990:253), the verb eat has a causative LCS with an Agent, an optional Theme/Patient and a Goal argument:

< previous page

page_12

next page >

< previous page

page_13

next page >

Page 13 (31) eat V [Event CAUS ([Thingi], [INCH ([Thing], [Path TO [IN[MOUTH-OF [Thingi]]]])])] The Path argument of INTO is normally conceived of as “self’s mouth,” which is co-indexed with the first argument of CAUSE. Typically, the arguments of CAUSE, INCH, and INTO—Agent, Theme/Patient, and Goal—are mapped onto the subject, object, and indirect object positions, respectively. However, when the Agent and Goal arguments are linked to the same NP, only the higher argument, that is, the Agent, is mapped onto the syntax. In other words, although the eat class of verbs appears to be transitive in the syntax, the verbs are ditransitive in the LCS. Note that, crucially, when the Agent and Goal arguments are co-indexed, it is the higher of the two— within standard assumptions of the Thematic Hierarchy—namely the Agent, that is mapped onto syntax, giving the argument structure of the transitive verb b lla “eat,” with an Agent and a Theme argument. In other words, the Goal argument, (i.e., the argument of the functor PATH) does not project into syntax:

Let us assume that another CAUSE is introduced. Suppose that a- cannot attach to a verb that already has an internal CAUSE. However, given the LCS of the verb b lla “eat” in (32), the possibility of allowing a new CAUSE emerges. Suppose also that the original CAUSE does not project. This option, which is otherwise unavailable with other causative verbs, is made possible by the co-indexation of CAUSE with PATH. In other words, CAUSE can fail to project as it is semantically recoverable from PATH. Thus, the old CAUSE will be displaced by the new CAUSE. This will give the triadic argument structure of the verb a-b lla “feed,” as modeled in (33):

Languages vary in how they realize the LCS in (33). In languages like Amharic, the introduction of the new CAUSE is achieved by a morpho-

< previous page

page_13

next page >

< previous page

page_14

next page >

Page 14 logical causative, the internal CAUSE affix a- . The Tibeto-Burman language Meithei is like Amharic in that a causative suffix is used to derive the causative of the verb ‘eat.’ Thus, čá “eat” → čáh nb “cause to eat” (cf. Chelliah, 1997:110). In English, the LCS in (33) is realized by a suppletive form, the verb feed.3 In Chitimacha, according to Swadesh (1946:318), the causative variant of the verbs meaning “to eat” and “to drink” are formed by suppletion: (34) a. gušt- ‘to eat…’ nokšte ‘to feed…to…’ b. ka.čt- ‘to drink…’ hakte- ‘to give…to…to drink’ The proposed analysis does not imply that all verbs of ingestion will behave in the same way. On the contrary, there will be language-particular lexical gaps. For instance, the verbs eat and drink are conceptually identical except for the specification of the Theme/Patient argument: in the former the Theme/Patient is typically a solid substance (cf. Levin, 1993:213ff), whereas in the latter it is some sort of liquid.4 In both cases, the Theme/Patient can be omitted: John drank beer versus John drank. Nevertheless, although an internal causative can be introduced into the LCS of eat, deriving the lexicalized verb feed, there is an idiosyncratic constraint (i.e., not all ingestive verbs allow the same kind of derivation). Thus, in English we find periphrastic forms with most verbs of ingestion such as give to drink. There is cross-linguistic variation with respect to the productivity of deriving a causative verb through the addition of an internal causative morphology. As we saw earlier in the chapter, the set of ingestive verbs that take the internal causative a- in Amharic include verbs such as b lla “eat,” t t’t’a “drink,” k’ mm s “taste,” among others. Likewise, in Malayalam and Berber, ingestive verbs have causatives that are derived by a productive morphological process. However, recall that Malayalam (Mohanan, 1983) differs from both Amharic and Berber in that the set of ingestives includes verbs of perception and mentation such as the verbs meaning “to see” and “to learn”—metaphorical extensions of the prototypical notion. Note that, even in Amharic, the term “ingestive” is used in a loose sense as it covers verbs of gustation like k’ mm s “taste.” It is interesting to note that in some languages, there is only one abstract verb that can be used with anything that is taken into the body. For instance, in Bengali the verb kha can be used as “eat,” “drink,” “smoke,” or “graze” depending on the identity of the Agent and/or Theme/Patient argument (M. Onishi, p.c.).5 There is also cross-linguistic variation with respect to the case of the Goal argument. As the case assignment of the Goal argument depends on the case resources of individual languages, there may be some varia-

< previous page

page_14

next page >

< previous page

page_15

next page >

Page 15 tion. In Malayalam the Goal argument receives accusative case, whereas in Berber, the Goal argument receives dative case. In languages like English, the goal argument can be expressed either as accusative or dative. Consider the examples in (35), from Carrier and Randall (1992): (35) a. They fed the baby (peas) b. They fed peas to the baby The sentence in (35a) resembles the dative shift structure that is familiar from the verb give . One important difference between the typical dative shift structure and (35a) is that in the former the Theme/Patient argument cannot be left implicit (They gave John * (a present)) . This can be trivially traced back to the LCS of the verb give; namely, unlike the ingestives, the Theme/Patient argument cannot be implicit. I have said that the crucial property of ingestive verbs is that the Agent argument can be co-indexed with the Goal argument. If this assumption is correct, one may wonder whether there are noningestive verbs that exhibit the same lexical property (i.e., meet the requirement that the Agent and Goal be coindexed) and behave in a marked way. Amharic has one such verb, namely l bb s “dress.” Consider the examples in (36): (36) a. aster l bs l bb s -čč A. dress dress. PF-3F ‘Aster dressed in a dress.’ b. l mma asteri n l bs a-l bb s-ati L. A.-ACC dress/cloth CAUS-dress.PF.3M-3FO (lit. ‘Lemma dressed/clothed Aster a dress.’) ‘Lemma dressed/clothed Aster in a dress.’ I assume that the LCS for l bb s “dress” is as in (37)—which is only minimally different from that of the verb b lla “eat”: (37) dress V [Event CAUSE ([Thingi], [INCH ([Thing], [Path TO [ON [BODY-OF [Thingi]]]])])] The LCS of dress is very much like eat except that in the former the Goal argument is not “self’s mouth” but rather “self’s body.” As in the case of the ingestives, the Goal of the verb l bb s “dress/put on/wear” can be different from the Agent argument: X causes Y(clothing) to be on the body of Z. Not surprisingly, within the context of the proposed analysis the verb meaning “to dress” or “to put on” also forms a natural class with the

< previous page

page_15

next page >

< previous page

page_16

next page >

Page 16 ingestive verbs in languages like Hindi-Urdu, which brings us to yet another unusual behavior of ingestive verbs. It has been noted for some time now that ingestive verbs exhibit an unexpected morphosyntactic pattern with respect to the construction known as a resultative participle. Defining a resultative participle is not a straightforward matter, but generally most will agree with Haspelmath (1994:159) that a resultative participle expresses “a state resulting from a previous event.” Typical examples include the abused child, or the wilted dandelion. In many languages the resultative participle is not possible with unergative verbs; thus the equivalents of the run boy or the danced man are ungrammatical. It appears that resultative participles have Passive orientation with transitive verbs but Active orientation with unaccusative or inactive intransitive verbs. What is interesting is that, as pointed out by Haspelmath (1994:159–161), ingestive verbs often have (exceptionally) active orientation (i.e., they behave as if they are intransitive). In Hindi-Urdu the class of transitive verbs that have active resultative participles includes verbs meaning “to eat,” “to drink,” “to see,” “to learn,” “to wear,” and “to put on.” Note that English has the lexicalized form drunken as well as the mental ingestion verb learned (Haspelmath, p.c.), but the process appears to be productive in languages like Hindi-Urdu. According to Haspelmath (1994:161) “what ‘drink’, ‘eat’, ‘learn’, ‘see’, and ‘put on’, ‘wear’ have in common is that the agent is saliently affected by the action.” I suspect that the enriched Lexical Conceptual Structure proposed for the ingestive verbs here may eventually be able to provide a unified account for the resultative facts we find in language like HindiUrdu. The fact that the Agent is saliently affected may be derived from the assumption that it is coindexed with a Path argument, distinct from other transitive verbs. CONCLUSION Ingestive predicates exhibit marked transitivity patterns in many typologically and genetically diverse languages. In many languages an otherwise strict selectional restriction of causativizing morphemes is relaxed just in case the verb in question belongs to the ingestive class. Thus, in Amharic the causative affix a- normally attaches to (unaccusative) intransitives to derive causative predicates. The exception in which a- attaches to a transitive verb occurs with ingestive predicates, such as b lla “eat” -> a-b lla “feed.” Intuitively, ingestive verbs appear to be semantically homogenous. The problem is: how can we account for the marked property of the verbs

< previous page

page_16

next page >

< previous page

page_17

next page >

Page 17 within a restricted theory of argument structure? For example, in Amharic how can we account for the argument structure of this class of verbs without abandoning the otherwise robust generalization that the internal causative morpheme attaches to intransitive predicates. A closer examination of the LCS of the ingestive predicates has revealed that the verbs are actually ditransitive and take an Agent, Theme/Patient, and Goal (cf. Jackendoff, 1990). Distinct from the ditransitivity of other verbs (e.g., give in English) the ditransitivity of ingestive predicates has one special property: the Agent (the first argument of CAUS) and the Goal (the argument of PATH) are linked to a single argument and thus syntactically realized by one argument (the Agent). I have argued that introducing an Agent argument would be possible provided that the former Agent is not syntactically realized. This would be possible because the former Agent is conceptually linked with the Goal argument. The result is a construction where a Goal argument surfaces as the direct object of a ditransitive predicate yielding verbs such as a-b lla in Amharic and feed in English. Thus, the marked behavior of ingestive verbs with respect to transitivity alternation can be accounted for by assuming that the verbs’ LCS specifies a Goal argument that forms a chain with the Agent argument. By virtue of its co-indexation with a Goal argument, the Agent may not project in the syntax. This allows for the introduction of another causer argument thus deriving a special causative of transitive verbs. NOTES The research reported in this chapter is supported partly by a Special Research Grant of the Faculty of Arts and Social Sciences of the University of New South Wales. A shorter version of this chapter was presented at the University of Sydney, and at the 75th annual conference of the Linguistics Society of America. I would like to thank the audiences at these meetings—in particular Jane Simpson and George Aaron—for their critical comments and suggestions. I would also like to thank Martin Haspelmath for pointing out to me the interesting fact related to ingestives and resultative participles. Thanks are also due to Les Bruce and Bernard Comrie for responding to my queries regarding the transitivity of ingestive predicates and to an anonymous reviewer for critical comments and helpful suggestions. Of course, I am responsible for any remaining shortcomings. 1. It is interesting to note that cross-linguistically a causative affix that can attach to a transitive verb can also attach to an intransitive verb, whereas the reverse is not true. There is no language in which a causative morpheme attaches only to transitive verbs (see also Hetzron, 1976:374). 2. Actually, the causative prefix a- attaches only to stative (or unaccusative) intransitives such as k’ ll t’ “melt (intr).” This prefix cannot attach to activity (or unergative) intransitives such as č’ ff r “dance.” In this chapter, I ignore this distinction, as it is not relevant to the main argument (but see Amberber, 1996, 2000).

< previous page

page_17

next page >

< previous page

page_18

next page >

Page 18 3. It should be noted here that the English verb feed, which we assume to be the lexicalized causative of eat, has a different range of usage from the verb eat. As noted in Fellbaum (1990), the verb feed supports a number of compounds such as bottlefeed, breastfeed, spoonfeed, which simply do not occur with the verb eat. This type of meaning extension is typical of lexicalization; recall the famous debate regarding the relationship between kill and cause to die . 4. This is a slight oversimplification. There are other differences between the English verbs eat and drink, which are not relevant for the present discussion. For instance, consider the difference between John had a drink vs. *John had an eat. See Wierzbicka (1982) for a discussion of some interesting differences between the two English verbs. 5. In some languages, it is the manner of eating that seems to be expressed by different verbs. English appears to have a rich inventory of verbs of ingestion that specify the manner of eating. Thus, consider chew, chomp, crunch, gnaw, munch, nibble, pick, peck, sip, slurp, suck (examples from Levin, 1993:214). Some verbs are used to encode “the complete, and usually speedy, consumption of something” (Levin, 1993:215). Such verbs are bolt, gobble, gulp, guzzle, quaff, swallow, swig, wolf. REFERENCES Aikhenvald, A. 2000. Transitivity in Tariana. In R.M.W.Dixon and A.Aikhenvald (eds.) Changing Valency: Case Studies in Transitivity. Cambridge, U.K.: Cambridge University Press, 145–72. Amberber, M. 1996. Transitivity alternation, event-types, and light verbs. Ph.D. dissertation, McGill University. ——. 2000. Valency-changing and valency-encoding devices in Amharic. In R.M.W.Dixon and A.Aikhenvald (eds.) Changing Valency: Case Studies in Transitivity. Cambridge, U.K.: Cambridge University Press, 312–332. Baker, M.C. 1988. Incorporation, A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Bloomfield, L. 1946. Algonquian. In H.Hoijer et al. Linguistic Structures of Native America. New York: Johnson Reprint Inc., 85–129. Carrier, J. and Randall, J. 1992. The argument structure and syntactic structure of resultatives. Linguistic Inquiry 23:173–234. Campbell, L. 1985. The Pipil Language of El Salvador. Berlin: Mouton de Gruyter. Chafe, W.L. 1970. A semantically based sketch of Onondaga. Baltimore: Waverly Press Inc. Chelliah, S.L. 1997. A Grammar of Meithei. Berlin: Mouton de Gruyter. Comrie, B. 1989. Language Universals and Linguistic Typology. 2d ed. Chicago: The University of Chicago Press. Comrie, B. and Polinsky, M. 1993. Causatives and Transitivity. Amsterdam: John Benjamins. Demoz, A. 1964. The meaning of some derived verbal stems in Amharic. Ph.D. dissertation, UCLA. Dixon, R.M.W. 2000. A-constructions and O-constructions in Jarawara. International Journal of American Linguistics 66:22–56. Fellbaum, C. 1990. English verbs as a semantic net. International Journal of Lexicography 3:278–301. Guerssel, M. 1986. On Berber Verbs of Change: A Study of Transitivity Alternations. Lexicon Project Working Papers no.9, Lexicon Project, Center for Cognitive Science, MIT.

< previous page

page_18

next page >

< previous page

page_19

next page >

Page 19 Haspelmath, M. 1993. More on the typology of inchoative/causative verb alternations. In B.Comrie and M.Polinsky (eds.) Causatives and Transitivity. Amsterdam: John Benjamins, 87–120. ——. 1994. Passive participles across languages. In B.Fox and P.J.Hopper (eds). Voice: Form and Function. (Typological Studies in Language 27.) Amsterdam: Benjamins, 151–77. Hetzron, R. 1976. On the Hungarian causative verb and its syntax. In M. Shibatani (ed.) The Grammar of Causative Constructions. Syntax and Semantics 6, 87–120. New York: Academic Press. Jackendoff, R. 1990. Semantic Structure. Cambridge: MIT Press. Kakumasu, J. 1986. Urubu-Kaapor. In D.D.Derbyshire and G.K.Pullum (eds.) Handbook of Amazonian Languages, vol. 1. 326–403. Berlin: Mouton de Gruyter. Kirsner, R. 1976. On the subjectless “pseudo-passive” in Standard Dutch and the semantics of background agents. In N.Li Charles (ed.) Subject and Topic. New York: Academic Press, 385–415. Koehn, E. and Koehn, S. 1986. Apalai. In D.D.Derbyshire and G.K.Pullum (eds.) Handbook of Amazonian Languages, vol. 1. 33–127. Berlin: Mouton de Gruyter. Leslau, W. 1995. Reference Grammar of Amharic. Wiesbaden: Otto Harrassowitz. Levin, B. 1993. English Verb Classes and Alternations. Chicago: University of Chicago Press. Manly, T.M. 1972. Outline of Sre Structure. Honolulu: University of Hawaii Press. Mohanan, K.P. 1983. Move NP or lexical rules? Evidence from Malayalam causativization. In L.Levin et al. (eds.) Papers in Lexical-Functional Grammar. Bloomington: Indiana University Linguistics Club, 47– 111. Swadesh, M. (1946). Chitimacha. In H.Hoijer et al. Linguistic Structures of Native America. New York: Johnson Reprint Inc., 312–36. Wierzbicka, A. 1982. Why can you have a drink when you can’t have an eat. Language 58:753–99.

< previous page

page_19

next page >

< previous page

page_20

next page >

page_20

next page >

Page 20 This page intentionally left blank.

< previous page

< previous page

page_21

next page >

Page 21 2 Explaining Clitic Variation in Spanish José Camacho and Liliana Sánchez Dialects of Spanish display a complicated pattern of variation with respect to the morphological shape of the third person clitics (for a complete list of references, see Fernández-Ordóñez, 1994). Some dialects are more or less stable, whereas others, in contact with other languages, are in constant change. Accounting for clitic variation in these complex situations poses a challenge for Linguistic Theory. In this chapter we will attempt to account for such variation using the framework of Optimality Theory, hence OT (cf. Prince and Smolensky, 1993), which uses universal constraints that are ranked on a languageparticular basis. In OT, linguistic variation stems from the relative position of a given constraint within the ranking. Additionally, a lower-ranked constraint may be violated to satisfy a higher-ranked one. In particular, we will compare the non-contact varieties (etymological and referential dialects) with contact varieties. Contact varieties yield grammars that vary in interesting ways. In some, a unique clitic is the result (but not always the same one for all grammars); in others, the result is two forms in free variation, with wide variation across speakers as to which specific forms are chosen. For these dialects, we will argue that free variation is the consequence of unranked constraints. An important asymmetry will be shown when comparing different contact dialects: whereas the marked form for case (dative) can surface as the unique clitic form, the marked form for gender (feminine) never does. Furthermore, when the feature feminine is present in the lexical representation, it is generally maintained in the surface form. In OT terms, the faithfulness constraint to feminine is highly ranked.

< previous page

page_21

next page >

< previous page

page_22

next page >

Page 22 OVERVIEW OF THE THIRD PERSON CLITIC PARADIGM IN SPANISH Most dialects of Spanish have three forms for third person singular clitics and three for the plural clitics, as shown in (1). Choice of either set of clitics depends on a variety of factors in each dialect, including animacy, case, the count/mass nature of the referent, and gender. (1) Sg. le la lo Pl. les las los Three large groups of grammars have been described with respect to the clitic paradigm: the so-called etymological clitic system (see, for example, Cuervo [1874/1988]; Marcos Marín, 1978, among many others), the referential system (see Klein-Andreu, 1981; Fernández-Ordóñez, 1994, 1999), and the contact varieties (see Fernández-Ordóñez, 1999). The first system divides clitics according to case: le/les are used for singular/plural dative arguments respectively, lo/la for accusative masculine and feminine singular arguments, and los/las for accusative masculine and feminine plural arguments. The second system partitions clitics according to features of the referent (essentially mass/count distinctions), using lo for mass nouns; finally, contact varieties have a simplified system with one or at most two clitics. Our account will analyze variation within these dialects, focusing on contact varieties. Constraint reranking in these varieties yields a simplified paradigm, where most of the clitic forms disappear, resulting in a highly marked dialect. However, the remaining forms are not always the same. Rather, different forms surface in different dialects (for example, le in some dialects and lo in others, but never la ), and certain dialects seem to show systematic free variation between two forms. The chapter is organized as follows. First an overview of the etymological and referential dialects is presented. This is followed by an analysis for these basic dialects. In the last section, we turn to contact dialects and present one-clitic dialects and two-clitic dialects, showing how the solution in each case is different. OVERVIEW OF THE ETYMOLOGICAL AND THE REFERENTIAL DIALECTS The Etymological Dialect The overwhelming factor in the etymological dialect is case, although animacy also plays a role. Le/les is usually a dative clitic, namely a clitic for Indirect Object arguments. Lo(s)/la(s) is generally an accusative clitic,

< previous page

page_22

next page >

< previous page

page_23

next page >

Page 23 linked to Direct Objects, as illustrated in (2)–(3). In (2), Prefers to an Indirect Object a ella “to her”; in (3), lo and la refer to a Direct Object (ella “her” and él “him,” respectively).1 (2) Lei regalé un libro a ellai CLDAT gave a book to her ‘I gave a book to her’ (3) a. Lai saludé a ellai CLACC greeted to her ‘I said hello to her’ b. Loi saludé a éli CLACC greeted to him ‘I said hello to him’ The paradigm for the etymological dialect is summarized in Table 2–1. Referential Dialects Klein-Andreu (1981) first described a set of dialects from central Spain where the distribution of third person clitics is based on very different criteria. Following up on Klein-Andreu’s research, FernándezOrdóñez (1994) has conducted extensive field research in this region and proposes the following system (consistent with Klein-Andreu’s original research): le is used for masculine, count referents, la for feminine, count referents, and lo is used for mass referents, which are always singular. Clearly, the dividing factor in this system is the mass/count distinction, not case or animacy. Fernández-Ordóñez (1994) terms this dialect dialect A. Examples of it, from Fernández-Ordóñez (1999) are presented in (4). (4a) exemplifies a masculine, count referent niño “boy,” which is referred back to by le in the second clause. In (4b), the sheep, which is feminine, count, is referred to by la, even in the context of a dative argument, as in the final VP darla cortes, literally “give it cuts.” Finally, sangre “blood” is referred to by lo, even though it is a feminine noun. TABLE 2–1. Clitic Paradigm for the Etymological Dialect Sg PI M F M F Acc lo la los las Dat le le les les

< previous page

page_23

next page >

< previous page

page_24

next page >

Page 24 (4) a. Al niño le llevaron al hospital to-the boy CL took to-the hospital y le hicieron una radiografía. and CL did an x-ray ‘They took the boy to the hospital and they did an x-ray.’ b. A la oveja hay que esquilarla teniendo to the sheep has that shear-CL having cuidado de no darla cortes. care of not give-CL cuts ‘You have to shear sheep being careful not to cut them.’ c. Según recogías la sangre del cerdo, as picked up the blood of-the pig, lo revolvías, ibas dándolo vueltas. CL mixed turned-CL ‘As you gathered the pig’s blood, you mixed it and you stirred it around.’ Fernández-Ordóñez describes a number of minor variations of this pattern. For example, dialect A’ breaks down mass referents in masculine (le) and feminine (la) , yielding a uniform system where gender is the dominating factor. A third system (dialect B) uses los for plural, masculine referents (instead of les ), once again regardless of case. Dialect C alternates between los and les for masculine, plural referents. A summary of the different dialects is shown in Tables 2–2 through 2–5. For space reasons, we will concentrate in this chapter on dialect A, which contrasts more directly with the etymological dialect. We should TABLE 2–2. Referential Dialect A Sg PI +Count -Count M F M/F M F Acc le la lo les las Dat le la lo les las TABLE 2–3. Referential Dialect A’ Sg PI +/-Count M F M F Acc le la les las Dat le la les las

< previous page

page_24

next page >

< previous page

next page >

page_25

Page 25 TABLE 2–4. Referential Dialect B

+Count

M Acc le Dat le TABLE 2–5. Referential Dialect C +Count

Sg F la la Sg

-Count M/F lo lo

PI M los los

F las las

PI

-Count M F M/F M F Acc le la lo los~les las Dat le la lo los~les las note that dialects B and C present a particular analytical challenge because they involve mixed paradigms: lo stands for mass nouns in the singular, but for masculine nouns in the plural. The constraint rankings we have tried to account for those dialects do not yield consistent grammars. This problem is restricted to the OT framework only to the extent that OT can predict whether a given constraint ranking yields consistent grammars, and also what grammars are possible, given any ranking (the so-called factorial typology, see Prince and Smolensky, 1993). ACCOUNTING FOR THE ETYMOLOGICAL DIALECT AND REFERENTIAL A DIALECTS In this section, we will propose an account for the etymological dialect and the referential dialects A and A’. The first question to settle in the account of these paradigms is the nature of the lexical entries for the clitics. Two options are available: following Grimshaw’s (1997, 1999) approach, it could be assumed that clitics have an underspecified representation in the lexicon (but see note 9). Alternatively, they could have a full specification in the lexicon, and constraints would then determine the output.2 We will assume Grimshaw’s approach for two reasons. First, our attempts to account for the different dialects using full specification yielded contradictory grammars, whereas underspecification did not. Second, there is little evidence to decide precisely which features are relevant for the clitic lexicon. Thus, it could be argued that the partitive/nonpartitive distinction of Italian also applies to clitics in Spanish, although no

< previous page

page_25

next page >

< previous page

page_26

next page >

Page 26 overt counterpart ever surfaces. Although not ruling out that there may be a universal set of clitic features, we do not believe there is enough evidence to posit one at this point. In our approach (as in Grimshaw’s), clitics are specified only for those features in which they contrast with other clitics. Thus, for example, in the etymological clitic system, count/mass does not play any role; therefore lexical entries for clitics in that dialect should not include +/-count but be left unspecified (following Grimshaw, 1997, we will put unspecified features in parentheses). The lexical entries for clitics would be the ones represented in Table 2–6.3 The constraints we will be adopting are from the standard variety in OT: faithfulness and markedness constraints. Faithfulness constraints ensure that inputs and outputs are identical; markedness constraints evaluate how well formed outputs are. Interaction between both types of constraints ensure language particular variation. The higher markedness constraints are ranked, the more variation between input and output there will be (see Kager, 1999, for discussion). In addition, we will assume local conjunction (see Legendre, 1995; Legendre et al., 1998; Smolensky, 1995; Grimshaw, 1997, 1999, for a similar approach to clitics). The idea behind local conjunction of constraints is that a simultaneous violation of two constraints, C1 and C2, is worse than violation of those same constraints individually: C1 & C2 >> C1, C2. Constraint conjunction has proved to be a useful tool in explaining cases where linguistic changes depend on intermediate steps in a derivation (see Kager, 1999). One of the drawbacks of local conjunction is that the number of constraints in GEN could multiply exponentially, since in addition to each constraint, there is a conjunction.4 In the present case, local conjunction can be restricted to the set of features present in the clitic lexicon, confining the exponential growth of combinations somewhat. The following list shows the constraints we have tested, although not all of them are crucial for all of the grammars. (5) Faithfulness Constraints MAX ACC: An accusative case feature in the input must appear in the output DEP ACC: An accusative case feature in the output must appear in the input TABLE 2–6. Lexical Entries for Third Person Singular Clitics (Plurals Are the Same, Except for Number) Etymological System Referential System A La=3 s f acc (An) (Cnt) lo=3 s (G) (C) (An) –Cnt Lo=3 s m acc (An) (Cnt) la=3 s f (C) (An)(Cnt) Le=3 s (G) dat +An (Cnt) le=3 s m (C) (An) +Cnt

< previous page

page_26

next page >

< previous page

page_27

next page >

Page 27 MAX DAT: A dative case feature in the input must appear in the output DEP DAT: A dative case feature in the output must appear in the input MAX AN: An animacy feature in the input must appear in the output DEP AN: An animacy feature in the output must appear in the input MAX F: A feminine feature in the input must appear in the output DEP F: A feminine feature in the output must appear in the input MAX M: A masculine feature in the input must appear in the output DEP M: A masculine feature in the output must appear in the input MAX CNT: A count feature in the input must appear in the output DEP CNT: A count feature in the output must appear in the input (6) Markedness constraints *CNT: No count features *CASE: No case features *ACC: No accusative features *ANIM: No animacy features *PL: No plural features *G: No gender features *G & -CNT: Mass clitics should not bear gender features *G & DAT: Prohibits gender and dative features The Etymological Dialect The etymological dialect highly ranks case and gender faithfulness and also ranks highly markedness constraints for count/mass.5 The proposed constraint-ranking is presented in (7). Sample optimations for some of the inputs and candidates are presented in Tables 2–7 through 2–9. (7) Crucial constraints for the etymological dialect (Figure 2–1) An input like /3 s m acc +An -Cnt/, shown in Table 2–7, will yield an unspecified clitic for all features except for case and gender.6 This is due to highly ranked MAX ACC and also MAX M/F. Likewise, a dative input will yield an output unspecified for everything except for case and gender. Gender underspecification is achieved by ranking *G & DAT above MAX M (or MAX F for the feminine). This is shown in Table 2–8. Referential Dialect A Since mass/count is crucial for this dialect, the constraint ranking will promote faithfulness to count/mass, and markedness to case. Figure 2–1.

< previous page

page_27

next page >

< previous page

page_28

next page >

Page 28 TABLE 2–7. Optimation for /3 s m acc +An –Cnt/ in the Etymological dialect /3 s m acc +An –Cnt/ MAX Acc *CNT *Acc & AN MAX M *AN 3 s m acc (An) (Cnt) 3 s (G) acc (An) (Cnt) *! * 3 s m acc +An (Cnt) *! 3 s m acc (An) –Cnt *! * 3 s m (C) +An (Cnt) *! * TABLE 2–8. Optimation for /3 s m dat +An –Cnt/ in the Etymological dialect /3 s m dat +An –Cnt/ MAX DAT *CNT *G & DAT MAX M *AN * 3 s (G) dat (An) (Cnt) 3 s m dat (An) –Cnt *! * * 3 s m (C) +An (Cnt) *! Conversely, it will demote markedness to count/mass and faithfulness to case. Constraint rankings are presented in (8). (8) Crucial constraint-ranking for Referential Dialect A (Figure 2–2) In this dialect, a /+Cnt/ input will yield a [+Cnt] output, since MAX CNT is highly ranked. Gender faithfulness is also preserved, as shown in Table 2–11. On the other hand, a /-Cnt/ input also comes out as [+Cnt], but gender is not preserved due to *G & -CNT, as shown in Tables 2–9 through 2–12. Case is also not preserved, due to highly ranked *CASE. CONTACT DIALECTS Contact dialects are an interesting case for an OT account. The resulting system usually yields impoverished paradigms, and the question is what course the neutralization takes: Will there be a unique solution Figure 2–2.

< previous page

page_28

next page >

< previous page

page_29

next page >

Page 29 TABLE 2–9. Optimation for /3 s m Acc +An +Cnt/ in Referential Dialect A /3 s m Acc +An +Cnt/ MAX CNT DEP CNT *CASE *AN 3 s m (C) (An) +Cnt 3 s (G) (C) (An) +Cnt 3 s m Acc +An +Cnt *! * 3 s (G) (C) –An –Cnt *! * 3 s m (C) (An) (Cnt) *! 3 s m (C) (An) (Cnt) *! TABLE 2–10. Optimation for /3 s m Acc +An –Cnt/ in Referential Dialect A /3 s m Acc +An –Cnt/ MAX CNT DEP CNT *CASE *AN *G & –CNT 3 s (G) (C) (An) –Cnt 3 s m (C) (An) –Cnt *! 3 s m acc +An +Cnt *! * * 3 s m acc (An) +Cnt *! * 3 s (G) (C) (An) (Cnt) *! 3 s m (C) +An (Cnt) *! * TABLE 2–11. Optimation for /3 s m dat -An -Cnt/ in Referential Dialect A /3 s m dat -An -Cnt/ MAX CNT DEP CNT *CASE *AN *G & -CNT 3 s (G) (C) (An) -Cnt 3 s m (C) (An) -Cnt *! 3 s (G) dat -An -Cnt *! * 3 s (G) dat -An +Cnt *! * * 3 s (G) dat (An) (Cnt) *! * TABLE 2–12. Optimation for /3 s m Acc +An +Cnt/ in Referential Dialect A /3 s f Acc +An +Cnt/ MAX CNT DEP CNT *CASE *AN 3 s f (C) (An) +Cnt 3 s (G) (C) (An) +Cnt 3 s f (C) +An +Cnt * 3 s f Acc +An +Cnt *! * 3 s f (C) (An) -Cnt *! 3 s f (C) (An) (Cnt) *!

< previous page

page_29

MAX M *! *

MAX M *

* MAX M * * * * MAX F *

next page >

< previous page

page_30

next page >

Page 30 in all cases of neutralization, what we could call the UG solution; or will the types of language in contact influence reranking (the transfer solution)? These two questions echo the L2 acquisition debate on transfer versus access to UG (see Schwartz and Sprouse, 1996, for a summary). Since contact languages are, ultimately, contact grammars of individual speakers, the analysis of contact grammars may shed some light on the acquisition issue, and vice versa. We will show that there are several solutions to contact. One is a single-clitic paradigm (which may be le or lo), and another one is to have two clitics in free variation. An interesting contrast will surface: Although there are single-clitic dialects that have le (the majority), and those that have lo, there are no single-clitic dialects that have only la . In other words, assuming that le is primarily dative and lo primarily accusative, neutralization sometimes yields accusative, sometimes dative (regardless of which is more marked), but never the feminine. It is plausible to claim that feminine is the marked gender, for example, based on the fact that in coordination, most languages will resolve conflicting gender between conjuncts by using any gender but feminine (see Corbett, 1983). The three best-known dialects in contact with other languages are Basque Spanish (see Urrutia, 1988; Echenique, 1987; Fernández-Ordóñez, 1994, 1999; Landa, 1995, among others), Andean Spanish (cf. Toscano-Mateus (1953), Pozzi-Escot (1975), Godenzzi (1986), Cerrón-Palomino (1992), FernándezOrdóñez (1994, 1999), among others) and Paraguayan Spanish (Usher de Herreros, 1976; Granda, 1982). Here, we focus on Basque Spanish and Andean Spanish. The three dialects are in contact with languages that lack overt gender distinctions (Basque, Quechua, and Guarani). Rather than present each geographical area by itself, we will group dialects according to similarity of grammars. Single-Clitic Contact Dialects Fernández-Ordóñez (1994) reports a dialect of Basque Spanish where le is the only attested form for all cases. In the accusative, le refers mostly to animate referents, alternating with a null form. This dialect coincides with the one spoken in Lamas, Peru.7 In this dialect the predominant clitic for direct objects is le, in a few cases, lo.8 The two dialects are represented in (9). (9) Single clitic system in Lamas and the Basque Country (Figure 2–3) Is le underspecified for case, or is it inherently dative? In the case of Basque Spanish, we will see that there is another dialect where le contrasts with Ø. For Andean Spanish in contact with Quechua, there are

< previous page

page_30

next page >

< previous page

page_31

next page >

Page 31

Figure 2–3. dialects where only lo surfaces. One possible explanation for the difference between dialects in contact with Quechua that have le and dialects, also in contact with Quechua that have lo, is that case specification is maintained, hence le is inherently dative and lo is inherently accusative in those dialects. The reason why accusative inputs surface as le is because of a highly ranked *ACC constraint, which, together with MAX ACC, yields [dat], as we will see in Table 2–13. The lexical entries for these dialects appear in (10). (10) Clitic Lexicon for Basque Spanish 2 and Lamas Spanish Le = 3 s (G) Dat (An) (Cnt) These two dialects yield highly underspecified clitics by ranking markedness over faithfulness. The constraint rankings are presented in (11). Sample tableaux showing the correct output form are presented in Tables 2–13 and 2–14. (11) Constraint ranking for Lamas Spanish and Basque Spanish 2 (Figure 2–4) Is reranking in these dialects a matter of transfer or a default process? As mentioned earlier, Basque lacks gender specifications, and so does Quechua. Clearly, the contact dialects could have picked lo~la, or le~la, preserving gender, but neither do. This could suggest transfer from Basque and Quechua, where *G is presumably ranked high, but the hypothesis that feminine is cross-linguistically more marked than masculine could also be the determining factor. Let us look at case specifications to see how the solution is different for case than for gender. Quechua lacks third person marking for accusative and dative, which suggests that contact Spanish could favor no case distinctions, as a result Figure 2–4.

< previous page

page_31

next page >

< previous page

page_32

next page >

Page 32 TABLE 2–13. Optimation for /3 s m Acc +An +Cnt/ in Lamas Spanish and Basque Spanish 2 /3 s m Acc +An +Cnt/ *Acc MAX Acc *CNT *G *AN 3 s (G) dat (An) (Cnt) 3 s m dat +An (Cnt) * 3 s m (C) +An (Cnt) *! *! 3 s m (C) (An) +Cnt *! * * 3 s m (C) +An +Cnt *! * * * of transfer. A priori, this lack of distinction should result in case underspecification for clitics, yielding perhaps the most unmarked form (accusative lo). However, what we find is a slightly different picture: The majority of dialects, like the one just described, select le, and some select lo (see later). This is consistent with the lack of case marking for third person in Quechua: The highly ranked *CASE (to be more precise *3 & CASE, because case does appear with other persons) would be transferred. Depending on how MAX DAT and MAX Acc are ranked, the result would sometimes be le and sometimes lo (see later). Returning to gender specification, no dialect ever has only la . And unlike for case, when la appears, it is always feminine. This presents an asymmetrical picture: Case rankings are transferred, but gender rankings are the consequence of universal markedness (perhaps in conjunction with language transfer, although not necessarily). As just mentioned, Escobar (1978) describes a variety of Andean Spanish also in contact with Quechua, where lo is the outcome. This would be the mirror image of Lamas Spanish: The lexical specification for the clitic would be /acc/, and dative inputs will yield accusative outputs by the operation of *DAT and MAX DAT. The situation just presented suggests that gender and case behave in different ways. For the case of gender, universal markedness blocks feminine from surfacing as the unique clitic; however, transfer from TABLE 2–14. Optimation for /3 s m Dat +An +Cnt/ in Lamas Spanish and Basque Spanish 2 /3 s m Dat +An +Cnt/ *Acc MAX DAT *CNT *G *AN 3 s (G) dat (An) (Cnt) 3 s (G) dat +An (Cnt) *! 3 s (G) dat (An) +Cnt *! * 3 s m (C) +An +Cnt *! * * * 3 s m Acc +An (Cnt) *! *

< previous page

page_32

next page >

< previous page

page_33

next page >

Page 33 Quechua ranks *CASE highly. It is unclear what determines whether *DAT or *ACC are highly ranked.9 Dialects with Two Clitics Several dialects have only two clitics in the third person paradigm. These two clitics may involve several combinations: lo~le, lo~la, le~la, le~Ø. Basque Spanish In addition to the Basque Spanish dialect we have just seen, Fernández-Ordóñez (1994, 1999) describes another dialect, which according to her, involves speakers who are at an earlier stage of proficiency in the acquisition of Spanish. In other words, these speakers live in environments where Basque is dominant; hence their grammar is further away from Standard Spanish. This grammar involves a null clitic for the accusative and le for the dative, as shown in (12). The lexicon for this dialect is shown in (13). In Tables 2–15 and 2–16, we present sample tableaux for this dialect. (12) Basque Spanish dialect with two clitics (Figure 2–5) (13) Lexicon for Basque Spanish with two clitics Le = 3 s (G) Dat (An) (Cnt) Ø = 3 (N) (G) Acc (An) (Cnt) The Dialect of San Juan de Miraflores The dialect spoken in one of the districts of Lima, San Juan de Miraflores10 is puzzling. Although the children interviewed did not speak Quechua, their parents did, which makes them contact speakers. Several observations can be made based on it. First, most children have at most two clitics, although not the same combinations: Dialect 1 (see later) has lo and le, dialect 2 has lo and la, and dialect 3 has

Figure 2–5.

< previous page

page_33

next page >

< previous page

page_34

next page >

Page 34 TABLE 2–15. Optimation for /3 s m Acc +An +Cnt/ in the Two-Clitic Basque Grammar /3 s m Acc +An +Cnt/ *MAX ACC *ACC & N *G *CNT 3 (N) (G) acc (An) (Cnt) 3 s (G) acc (An) +Cnt *! 3 (N) m acc +An (Cnt) *! 3 s m acc (An) (Cnt) *! 3 s (G) acc (An) (Cnt) *! 3 s m (C) (An) (Cnt) *! * TABLE 2–16. Optimation for /3 s m Dat +An +Cnt/ in the Two-Clitic Basque Grammar (Dative) /3 s m Dat +An +Cnt/ MAX DAT MAX N *G *AN *CNT 3 s (G) dat (An) (Cnt) 3 s (G) dat +An (Cnt) *! 3 s m dat +An (Cnt) *! 3 (N) m acc (An) +Cnt *! * * 3 s (G) (C) +An (Cnt) *! * 3 (N) (G) (C) (An) (Cnt) *! * le and la . Only three children used all three clitics (we will disregard these speakers below). Second, la always refers to feminine referents in all dialects that use it, although not all feminine referents appear with la . La is not a frequent clitic: Only 7 of 28 children have it. Third, there is no correlation between the type of clitic and any feature we have seen so far (or any other we can imagine). The clitic systems are summarized in (14)–(16). (14) Clitic System for the San Juan de Miraflores dialect 1 (Figure 2–6) (15) Clitic System for the San Juan de Miraflores dialect 2 (Figure 2–7) (16) Clitic System for the San Juan de Miraflores dialect 3 (Figure 2–8) We will argue that this dialect involves free variation. Since the children are exposed to different dialects (in particular, to dialects that have only le, dialects that have all three clitics, and perhaps dialects with only lo), they are aware that there are three forms. However, their own grammar does not map those forms to unique constraint rankings. In other words, clitic forms in these dialects are unspecified, but have two systematic outputs that can be in free variation. For example, in dialect

< previous page

page_34

next page >

< previous page

page_35

next page >

Page 35

Figure 2–6.

Figure 2–7.

Figure 2–8. 3, feminine can surface either as le or as la . For dialect 1, we will assume that both forms lo and le can appear with all kinds of referents, and therefore have exactly the same lexical entries, as illustrated in (17). The unranked constraints appear in (18) and a sample tableau in Table 2–17. In effect, this grammar is similar to that of Lamas and the Basque dialect with one form, except that it has two forms in free variation (i.e., either one can appear in exactly the same grammatical context). (17) Lexical entries for San Juan de Miraflores dialect 1 Lo = 3 s (G) (C) (An) (Cnt) Le = 3 s (G) (C) (An) (Cnt) (18) Unranked constraint for San Juan de Miraflores dialect 1 (Figure 2–9) The unusual aspect of this proposal is that two distinct phonetic forms (lo and le ) do not correlate with any difference in features. A possible

< previous page

page_35

next page >

< previous page

page_36

next page >

Page 36 TABLE 2–17. Optimation for /3 s m Dat +An +Cnt/ in San Juan de Miraflores Dialect 1 /3 s m Dat +An +Cnt/ *CASE *CNT *G *ANIM 3 s (G) (C) (An) (Cnt) 3 s m (C) (An) (Cnt) *! 3 s m (C) +An +Cnt *! * * 3 s (G) dat (An) (Cnt) *! alternative account for this dialect would be to claim that lo and le are specified for case (or some other feature), but tied constraints yield free alternation. Constraint tie has been proposed, for example, by Pesetsky (1998) to account for different grammatical outputs in wh-phrases and complementisers. Thus, in English it is possible to have the person whom I met, the person that I met, or the person I met . Pesetsky argues that this is the result of constraint tie. In the present case, we could suppose that le is marked as dative and lo as accusative, much as in the etymological dialect. Given a /dat/ input, either a [dat] or an [acc] output will win. MAX DAT will be responsible for the first outcome, and *DAT will be responsible for the second one (also yielding the underspecified option [(C)]). If these two constraints are tied in ranking, both outputs will be available. Likewise, given an /acc/ input, MAX ACC will favor [acc] and *ACC will favor [dat]. Crucially, *DAT and *ACC must also be tied. It is not clear whether these two analyses are contradictory. It is possible that constraint tie end up forcing identical lexical specifications, which for a time, would have different phonetic forms. For the second and third dialect, the lexical entries would be the following: (19) Lexical entries for San Juan de Miraflores dialect 2 Lo = 3 s (G) (C) (An) (Cnt) La = 3 s f (C) (An) (Cnt) (20) Lexical entries for San Juan de Miraflores dialect 3 Le = 3 s (G) (C) (An) (Cnt) La = 3 s f (C) (An) (Cnt) In this case, the same markedness constraints would be ranked high, except that MAX F and *G would be tied, as shown in Table 2–18. Figure 2–9.

< previous page

page_36

next page >

< previous page

page_37

next page >

Page 37 TABLE 2–18. Tied Constraints in San Juan de Miraflores Dialects 2 and 3 /3 s f Dat +An +Cnt/ *CASE *CNT *ANIM MAX F *G * 3 s (G) (C) (An) (Cnt) * 3 s f (C) (An) (Cnt) 3 s f (C) +An +Cnt *! * 3 s (G) dat (An) (Cnt) *! CONCLUSIONS The analysis of clitic variation among dialects of Spanish in isolation and dialects of Spanish in contact with another language has shown an important asymmetry with respect to case and gender. Feminine surfaces as a marked, faithful specification, whereas case shows much more variability and permeability to language transfer. As expected, contact dialects tend to have paradigmatically simplified grammars. In contact with Basque, dialects tend to simplify in favor of the dative, which cannot be attributed to transfer, whereas dialects in contact with Quechua can simplify either in favor of the dative or in favor of the accusative. We have analyzed these dialects as indirect consequences of a match between a highly ranked constraint *CASE in two grammars. NOTES We wish to thank the following people for important discussion of the material in this paper: Eric Bakovic, Beto Elías, Jane Grimshaw, Alan Prince, and an anonymous reviewer. We are grateful to Jaime Doherty for facilitating access to the native speakers of Lamas. All errors remain ours. 1. A ella/él are Direct Objects in (3), despite the fact that they have a preposition. This preposition appears with certain Direct Objects, generally animate, although specificity also plays a role. We are glossing over certain syntactic facts that complicate the picture. For example, an impersonal, generic se forces the presence of le for an accusative argument, as in (i) (see Fernández Ramírez, 1964, among others). Furthermore, psychological verbs show a systematic pattern: If the subject is agentive, causative (see (ii)a), lo is preferred, otherwise, le is used (see (ii)b): (i) a. Lo encontramos a él CLACC found to him ‘We found him.’ b. A él, se le encuentra en la to him, CL CLDAT finds at the tienda. store ‘One (usually) finds him at the store.’

< previous page

page_37

next page >

< previous page

page_38

next page >

Page 38 (ii)a. Al príncipe, la bruja lo encantó. to- prince, thewitchCLACCcharmed the ‘The witch cast a spell on the prince.’ b. Al príncipe le encantó la to- prince, CLDATcharmedthe the conversatión de la Cenicienta. conversation of theCinderella ‘The prince was charmed by Cinderella.’ We will not address these issues here. We believe it is related to the argument structure of verbs, an issue that also surfaces in other dialects. 2. Thanks to Eric Bakovic for discussion on this point. 3. N = number (s, p), C = case (dat, acc), G = gender (m or f), An = animacy, Cnt = count. 4. In addition, there are other drawbacks—for example, the fact that two low-ranked constraints A and B could end up being ranked higher than another constraint by virtue of local conjunction. See Kager (1999) for discussion. 5. Constraints were ranked with Bruce Hayes’s “OTSoft” software package, see Hayes (1999). In the presentation of the following tables, we have omitted irrelevant candidates and irrelevant constraints and candidates, although the actual grammar was checked with a much larger set of candidates and the full set of constraints. 6. In the Tables, not all constraints are included, only those relevant. 7. The data presented was gathered in the town of Lamas in 1998–1999. Thirty bilingual children (Spanish-Quechua) between 9 and 12 years old were interviewed. They were asked to narrate a story based on pictures and to complete a sentencematching task. Data are part of a research project conducted by Liliana Sánchez with a research team composed of Milagros Lucero, Katia Murillo, Jaime Navarrete, José Riqueros, and Irma Sánchez in the rural area of Lamas, Peru. The study was funded by Carnegie-Mellon University. 8. Unfortunately, it is not possible to draw clear conclusions with respect to la, based on the study, because there were very few feminine referents in the data. However, from anecdotal evidence, it is clear that le is also used for feminine referents. 9. Notice that the underlying assumption of this reasoning somewhat modifies our initial idea that lexical entries are underspecified. This is because we are now arguing that le is essentially dative and lo accusative. This modification is justified to the extent that the learner will hear input in which le is overwhelmingly associated with three argument environments and lo with two argument environments. 10. The data from San Juan de Miraflores were gathered in the city of Lima, Peru, in 1999. Thirty-six children, ages 9 to 12 were asked to perform the same tasks as in Lamas. The research team, lead by Liliana Sánchez, was formed by Omar Beas, Vidal Carvajal, Milagros Lucero, Jaime Navarrete, José Riqueros, and Irma Sánchez. REFERENCES Cerrón-Palomino, R. 1992. La forja del castellano andino o el penoso camino de la ladinización. In C.Hernández Alonso (ed.) Historia y presente del español de America. Valladolid: Junta de Castilla y León y PABEC.

< previous page

page_38

next page >

< previous page

page_39

next page >

Page 39 Corbett, G. 1983. Resolution rules: Agreement in person, number and gender. In G.Gazdar, E.Klein, and G.Pullum (eds.) Concord and Constituency. Dordrecht: Foris. Cuervo, R.J. 1874/1988. Notas a la Gramática de la lengua castellana de don Andrés Bello. Reprinted in A.Bello Gramática de le lengua castellana destinada al uso de los americanos. Madrid: Arco/Libros. Echenique, M.T. 1987. Historia lingüística vasco-románica. Madrid: Paraninfo. Escobar, A. 1978. Variaciones sociolingüísticas del castellano en el Perú. Lima: Instituto de Estudios Peruanos. Fernández-Ordóñez, I. 1994. Isoglosas internas del castellano. El sistema referencial del pronombre átono de tercera persona. Revista de Filología española 74:71–125. ——. 1999. Leísmo, laísmo y loísmo. In I.Bosque and V.Demonte (eds.) Gramática descriptiva de la lengua española (1317–1393) Madrid: Espasa-Calpe. Fernández Ramírez, S. 1964. Un proceso lingüístico en marcha. Reprinted in S. Fernández Ramírez Gramática española. Vol. 3.2 (pp. 391–405). Madrid: Arco/Libros. Godenzzi, J.C. 1986. Pronombres de objeto directo o indirecto del castellano en Puno. Lexis 10:187–201. Granda, G.de. 1982. Orígen y formación del leísmo en el español de Paraguay. Ensayo de un método. Revista de Filología Española 62:259–83. Grimshaw, J. 1997. The Best clitic: Constraint conflict in morphosyntax. In L. Haegeman (ed.) Elements of Grammar. Dordrecht: Kluwer, 165–96. ——. 1999. Optimal Clitic Positions and the lexicon in romance clitic systems. To appear in G.Legendre, J.Grimshaw, and S.Vikner (eds.). OT Syntax. Cambridge: MIT Press. Available at Rutgers Optimality Archive ROA-374–01100, http://ruccs.rutgers.edu/roa.html Hayes, B. 1999. OT Soft. Software package, http://www.humnet.ucla.edu/humnet/linguistics/people/hayes/otsoft/ Kager, R. 1999. Optimality Theory. Cambridge, MA: Cambridge University Press. Klein-Andreu, F. 1981. Distintos sistemas de empleo de le, la, lo. Perspectiva sincrónica, diacrónica y sociolingüística. Thesaurus 36:35–46. Landa, A. 1995. Conditions on Null Objects in Basque Spanish and their Relation to leismo and Clitic Doubling. Ph. D. dissertation, USC. Legendre, G. 1995. Causee prominence constraints in French and elsewhere. In C.S.Burgess, K.Dziwirek, and D.Gerdts (eds.) Grammatical Relations: Theoretical Approaches to Empirical Questions. Stanford University: CSLI Publications, 291–308. Legendre, G., Smolensky, P., and Wilson, C. 1998. When is less more? Faithfulness and minimal links in Wh-chains. In P.Barbosa, D.Fox, P.Hagstrom, M.McGinnis, and D.Pesetsky (eds.) Is the Best Good Enough? Optimality and Competition in Syntax. Cambridge, MA: MIT Press, 249–89. Marcos M.F. 1978. Estudios sobre el pronombre. Madrid: Gredos. Pesetsky, D. 1998. Some Optimality principles of sentence pronunciation. In P.Barbosa, D.Fox, P.Hagstrom, M.McGinnis and D.Pesetsky (eds.) Is the Best Good Enough? Optimality and Competition in Syntax. Cambridge, MA: MIT Press, 337–83. Pozzi-Escot, I. 1975. Norma cultay normas regionales del castellano en relación con la enseñanza. In R.Avalos de Matos and R.Ravines (eds.) Lingüística e indigenismo moderno en América. Lima: Instituto de Estudios Peruanos, 321–30. Prince, A., and Smolensky, P. 1993. Optimality Theory. RUCCS Technical Report 2. New Brunswick, NJ: Rutgers Center for Cognitive Science.

< previous page

page_39

next page >

< previous page

page_40

next page >

Page 40 Schwartz, B., and Sprouse, R. 1996. L2 Cognitive states and the full transfer/full access model. Second Language Research 12:40–72. Smolensky, P. 1995. On the Internal Structure of the Constraint Component Con of UG. Handout of talk presented at UCLA. Available at ROA-86–0000, Rutgers Optimality Archive ROA: http://ruccs.rutgers.edu/roa.html Toscano-Mateus, H. 1953. El español de Ecuador. Madrid: Consejo Superior de Investigaciones Cientificas. Usher de Herreros, B. 1976. Castellano paraguayo. Notas para una gramática contrastiva del castellanoguaraní. Suplemento antropológico, U.Católica (Asunción) 11:29–123. Urrutia, H. 1988. El español en el País Vasco: peculiaridades morfosintácticas. Letras de Deusto 18:33– 46.

< previous page

page_40

next page >

< previous page

page_41

next page >

Page 41 3 Slavic Passives, Bantu Passives, and Human Cognition Peter F.Kipka It is tempting to think of active-passive pairs such as those in (1) and (2) as the simplest illustrations of the syntax of passivization. (1) a. John observed Mary, b. Mary was observed. (2) a. Tigers will pursue lions, b. Lions will be pursued. Both theoreticians and typologists have made explicit claims to this effect, as the following quotes illustrate: “The traditional characterization of passive as involving a change of object to subject is correct in one important sense: this is the core case of passive.” (Chomsky, 1981:124) “We shall refer to passives like […] John was slapped as ‘basic passives.’” (Keenan, 1985:247) The quoted assertions are made not solely for expository convenience. Chomsky uses his assertion to formulate a Uniformity Principle, while Keenan’s statement is designed to help linguists undertaking fieldwork. The question arises, however, of what we are to make of examples of passivization that do not look like (1b) and (2b). Across the globe, interesting “passive-like constructions” abound:

< previous page

page_41

next page >

< previous page

page_42

next page >

Page 42 (3) a. I consider that phonetics is hard. b. It is considered that phonetics is hard. (4) a. Gotowa em fasolę. cooked (1sg) bean (ACC) ‘I cooked beans.’ b. Gotowano fasolę. ‘Beans were cooked.’ (5) a. wageni wamefikia kijijini strangers arrived village-in ‘The strangers arrived at the village.’ b. kijijini pamefikiliwa na wageni village-in arrived preposition strangers ‘The village was arrived at by the strangers.’ The English data in (3) show passive morphology without promotion of an object to subject position. In (4), Polish morphology shows that a direct object marked with accusative case remains accusative despite passivization of the clause. And in the Swahili sentences in (5), drawn from Vitale (1981:126), a locative (ni suffixed) phrase is promoted to subject position, triggering appropriate agreement on the passivized verb. I will argue that the traditional description of passivization as essentially object-to-subject promotion is akin to such traditional insights as “birds fly” and “the sky is blue.” Such statements seem to be obviously true, until one encounters penguins, ostriches, colorful sunsets, or photographs taken from the moon. Ultimately, such exotic phenomena may give us deeper insights than that which we most frequently observe—surely a commonplace in science and hence something that the cognitive scientist should expect. Ultimately, I claim that careful scrutiny of the details of Slavic and Bantu passives can lead to potentially far-reaching conclusions about the role of prototypicality in human cognition. A FRAMEWORK To set the scene, consider the following pair of English sentences: (6) a. Someone killed three students, b. There were three students killed. The obvious passive counterpart (exhibiting object-to-subject promotion) of (6a) is (7):

< previous page

page_42

next page >

< previous page

page_43

next page >

Page 43 (7) Three students were killed. Question formation (Were three students e killed? where underlining indicates coreference) demonstrates the subject status of three students . By similar reasoning (i.e., Were there e three students killed ?), we see that there behaves as the subject of sentence (6b), even though (6b) contains all the passive morphology of sentence (7). Early transformational work would have related (6b) and (7) by deriving (6b) from (7), using a ThereInsertion Transformation. A deeper, more constrained analysis posits an embedded Small Clause (to be labeled SC, for mnemonic purposes). Slightly adapting the explorations of Haegeman and Guéron (1999:303), we can visualize (6b) as follows: (8) there [ +PAST [be [sc three students [killed e] sc] ]] The verb kill requires a complement—the direct object three students in the active version of the clause in (6a). In (7) the direct object ends up as the surface subject, but in (8) it is moved to the position of specifier (“subject”) of the Small Clause. The auxiliary verb will undergo head-to-head movement to pick up or check the past tense, and the expletive there is inserted into the higher specifier (or subject) position, which cannot remain empty in (the relevant formal register of) English. What is gained by the analysis in (8)? Essentially, we can now account for both (6b) and (7) using identical mechanisms. Note that we require an appropriate lexicon, with specifications including the complement-taking property of kill, the ability of be to undergo head-to-head movement to Tense, the absence of a null expletive, and the expletive status of there . Given such a lexicon, both (6b) and (7) involve displacing the complement; it simply moves further in (7). Crucially, their syntactic structures on this account can be the same. Needless to say, the identical mechanisms responsible for (6b) and (7) can be independently motivated. Small Clauses occur in several English constructions, such as the expressions of incredulity in (9) and as a complement of the preposition with in (10): (9) [scHoward the President of Australia!?] I don’t believe it. [Three students killed!?] I don’t believe it. [Australia broke!?] I don’t believe it. (10) With [Howard the President of Australia], we should move overseas. With [three students killed], we will have to hire more policemen. With [Australia broke], we must go to the U.S.

< previous page

page_43

next page >

< previous page

page_44

next page >

Page 44 See Akmajian (1984), Aarts (1992), and Haegeman and Guéron (1999) for further discussion. From a theoretical perspective, it is worth asking what the head of the Small Clause in (8) might be. It is also worth asking what projection the passive morpheme in (8) is the head of. A neat suggestion is that these two questions have the same answer. The Small Clause in (8) simply is the phrasal projection of the passive morpheme. Nouns project noun phrases, verbs project verb phrases, and for consistency, inflectional morphemes like Tense and Agr give rise to functional projections (TP, AgrP). It is plausible that our mnemonic SC is in fact a species of AgrP. This straightforwardly accounts for participle agreement in Romance: (11) Trois étudiants furent [e tués e]. ‘Three students were killed.’ Thus, in the French data in (11), the plural phrase trois étudiants triggers spec-head agreement on both the auxiliary furent and the participle tués, as it is the specifier for furent, and its trace is the specifier of tués . Agreement in Romance Small Clauses where the predicate is an adjective is equally straightforward: (12) Nous considerons [Marie intelligente]. ‘We consider Marie intelligent.’ In (12), the name Marie is the specifier of the Small Clause headed by the feminine agreement suffix that ultimately ends up on the predicative adjective. Returning to English, we note that the only morphological difference between (6b) and (7) is the expletive there, which is drawn directly from the lexicon. Its listing there is independently needed given data such as the following: (13) There was [a loophole in the argument]. There is [no way out]. Thus we have a unified syntactic account of both (6b) and (7), even though only the latter is described as a “prototypical” or “basic” or “core” passive. This sets the scene for our explorations of Slavic and Bantu, where “non-prototypical” passives are even easier to find. I will argue that a sufficiently deep morphosyntactic analysis will account for both the “prototypical” and the “non-prototypical” data. Syntactic operations invoked (e.g., head-to-head movement, spec-head agreement) will all be of the sort that can be justified on the basis of non-passive data. The

< previous page

page_44

next page >

< previous page

page_45

next page >

Page 45 essence of passives, then, will be the passive morpheme. Listed in the lexicon, it has the power to project phrase structure just like nouns, verbs, and other heads. And the fact that it requires a θ-role to be assigned to it will, following Jaeggli (1986), be taken as its defining lexico-semantic property. Needless to say, this is why (6b) and (7) have one phrasal argument fewer than (6a). SLAVIC Having set the scene, I will now propose an analysis of the unusual-looking Polish passive in (4). Polish belongs to the Slavic language family and so, not surprisingly, is inflectionally quite rich and also allows null subjects, which are identified by subject-verb agreement. Polish has three tenses (past, present, future), two aspects (perfective and imperfective), and seven cases. With few exceptions, singular nouns belong to one of three genders (masculine, feminine, neuter), which are signaled also on adjectives, determiners, participles, and past-tense verbs. Polish certainly exhibits straightforward passives that look very much like the “basic” passives of English: (14) a. Lew z apa kaczkę. lion (NOM) caught duck (ACC) ‘The lion caught a/the duck.’ b. Kaczka zosta a z apana. duck (NOM) became caught ‘The duck was caught.’ c. Kaczka zostanie z apana. duck (NOM) become (FUT) caught ‘The duck will be caught.’ (15) a. Policja (za)aresztowa a z odzieja. police (NOM) arrested thief (ACC) ‘The police arrested a/the thief.’ b. Z odziej by aresztowany. thief (NOM) was arrested. ‘The thief was arrested.’ In the active in (14a), the lion is the Agent and the duck is the Patient. In (14b), the passive suffix (here: -n ) appears, triggering assignment to it of the Agent role. Thus the lion need not be mentioned in (14b). If it is, it occurs in an optional prepositional phrase (przez lwa), exactly

< previous page

page_45

next page >

< previous page

page_46

next page >

Page 46 analogous to the optional agentive by -phrases of English. The complement of the verb (kaczk+suffix) undergoes movement leftwards, as shown in (16): (16) Kaczka zosta a [sc e z apana e] . As specifier of the Small Clause, it agrees in gender and number with the past participle. As matrix subject, it is assigned nominative case and agrees in person, number, and gender with the auxiliary. Quite often Polish uses zostać (become) as the auxiliary carrying the tense in a passivized clause. However, być (be) can appear instead, with a less strong implication of aspectual finality. Both these possibilities are illustrated in the preceding examples. The fact that we can choose among auxiliaries suggests that they are not a defining property of the passive. This is precisely our approach, which claims that the properties of passivization follow from the lexical specification of the passive morpheme plus independently justified syntactic principles. It should not come as too much of a surprise that even English allows a choice of passive auxiliary, as shown by the existence of so-called get -passives. Returning to Polish, consider the following paraphrases of (14b) and (17) Z apano kaczkę. catch+suffixes duck (ACC) ‘A/The duck was caught.’ (18) (Za)aresztowano z odzieja. arrest+suffixes thief (ACC) ‘A/The thief was arrested.’ Here the Patients remain in complement position and are marked with accusative case, just as we saw in (4b). One might wonder whether (17) and (18) are really genuine passives. Perhaps, one might argue, they are best treated as impersonal actives along the lines of One will arrest him (note the nonnominative case on the English pronominal). One argument against this hypothesized alternative is the fact that the most apt translation into Polish of One will arrest him is Zaaresztuje się go, which contains the reflexive się and no passive -n suffix, but this brings us to a second, stronger argument. For some verbs, the -n suffix has an idiosyncratic realization as -t. Significantly, whenever this allomorphy occurs in the passive participle, it also occurs in the “impersonal” verb forms illustrated in (17, 18), forms that are invariant (except for aspect) and thus do not exhibit adjective-like agreement:

< previous page

page_46

next page >

< previous page

page_47

next page >

Page 47 (19) gloss passive participle form in (17, 18) sew szyty szyto give dany dano kill zabity zabito open otworzony otworzono close zamknięty zamknięto tear podarty podarto poison zatruty zatruto I conclude that we are seeing n/t allomorphs of one and the same (passive) suffix in both columns of (19). Agreement suffixes, which are placed after the passive morpheme, can now be handled in a uniform manner. Passive participles show three-way gender agreement with (singular) subjects: (20) a. Ch opiec zostd zabrany/wzięty. boy became taken ‘The boy was taken.’ b. Dziewczyna zosta a zabrana/wzięta. girl became taken ‘The girl was taken.’ c. Dziecko zosta o zabrane/wzięte. child became taken ‘The child was taken.’ Further, corresponding to (17, 18), we find: (21) Wzięto/zabrano ch opca/dziewczynę/dziecko. take+suffixes boy/girl/child ‘A/The boy/girl/child was taken.’ The passive verb in (21) ends in -o, which is different from the -y/-a/-e (masculine/feminine/neuter) adjectival agreement seen on the participles in (20). Note that -o appears on weather verbs (e.g., Grzmia o = ‘There was thunder’). I take this invariable -o to be the form of verb agreement compatible with a null expletive subject, henceforth called exp for mnemonic convenience. Given the pro-drop setting of Polish, the availability in the lexicon of a null expletive exp should hardly come as a surprise. We are now in a position to more fully analyze (17, 18). I propose the following structure: (22) [sc exp [z apano [vp e kaczkę] ] ]

< previous page

page_47

next page >

< previous page

page_48

next page >

Page 48 The two-argument verb z apać (to catch) climbs via head-to-head movement to the passive -n, which projects a Small Clause. The complement of the verb is kaczkę, and its Agent role is assigned to the -n (cf. Jaeggli, 1986). The null expletive (exp) is responsible for the -o via spechead agreement. Nothing more need happen. Note in particular that there is no auxiliary and no tense morpheme in (22). This neatly explains why sentences like (17, 18) do not allow three-way tense variation. Full clauses exhibit tenses, but Small Clauses characteristically do not. In Polish, sentences like (17, 18) must be interpreted as anterior to the reference time (effectively past), a property that I take to follow from aspectual properties of past/passive participles. Note the affinity of these participles with perfect aspect in numerous language families (e.g., Germanic, Romance). Note further how The invited guests have arrived (or equivalently, Zaproszeni goście przybyli, in Polish) contrasts with The destitute guests have arrived ( = Biedni goś cie przybyli). The invitation must be anterior to the arriving, while the state of being destitute overlaps with it—correlating with the fact that the inviting is coded participially, but the destitution adjectivally. We will see later that there is ample independent justification for the claim that Small Clauses can be uttered as complete sentences in Polish. But keeping the focus on passivization here, it is worth mentioning that it is simply not an issue that (22) does not involve object-to-subject promotion. In fact, given the absence from (22) of a nominative case assigner (such as Tense), such promotion cannot occur, thus explaining the following ungrammaticality: (23) *Kaczka z apano. duck(NOM) catch+suffixes ‘The duck was caught.’ But if we were to add an auxiliary from the lexicon, the consequent presence of tense together with economy considerations will indeed allow object-to-subject movement, yielding (14b), where the Patient receives nominative case. In general, accusative case is assigned to complements when they need it (cf. I ate them; I ate.). Thus both (22) and (14a) contain accusative-marked complements. Note that the analysis in (22) conforms to the Extended Projection Principle, which requires that all sentences have subjects. In (22) the subject is exp. Note further that hypothesizable alternatives to (22) will require additional mechanisms and are thus dispreferred by virtue of Occam’s Razor. Thus an anonymous reviewer raises the possibility that the sentences in (21) may contain a null Tense with an AspP (perhaps headed by -o) as its complement. This alternative analysis would need to stipulate rather than derive the tense properties of these sentences;

< previous page

page_48

next page >

< previous page

page_49

next page >

Page 49 unlike overt tense, null Tense here must necessarily be past. A further problem is the fact that -o has more affinity with Agr than with Asp, since the relevant aspectual morphology here is prefixal (for further details see Kipka, 1990). All considered, the analysis in (22) seems to be optimal. So far, there is nothing to stop the structure in (22) from occurring with intransitive verbs. This is in fact a highly felicitous prediction, given such data as the following: (24) a. Ziewano. yawn+n+o ‘Yawning occurred.’ b. Śmiano się. laugh+suffixes reflexive ‘Laughing occurred.’ c. Śpiewano. ‘Some singing was done.’ d. Umierano. ‘Some dying/deaths occurred.’ The relevant structure here is simply (25): (25) [sc exp [verb-n-o [VP e] ] ] As before, the passive suffix heads a Small Clause and attracts the verb out of the VP via head-to-head movement. The single θ-role of the verb is assigned to this suffix, just as before, which means there will be no overtly pronounced arguments. The empty expletive subject exp triggers -o agreement exactly as in (22). Even though there can be no object-to-subject promotion (essentially because there is no direct object), given the lexicon of Polish, there is nothing to block the passivization in (24). Thus (24) can be viewed as straightforwardly predicted phenomena rather than quirky “non-prototypical” quasi-passive constructions. In the next section, we will see how the same logic can be applied to Bantu data. BANTU We will focus on passivization in Kikamba, a member of the Bantu language family, that is spoken by more than two million inhabitants of Kenya. Kikamba does not overtly case mark nouns, but it has eight tenses and seventeen noun classes, with subject-verb agreement according to noun class providing clear evidence of subjecthood over and above unmarked SVO ordering. I draw on data from Kioko (1994). Although

< previous page

page_49

next page >

< previous page

page_50

next page >

Page 50 Kikamba is a tone language, I will present examples in the accepted orthography used in schools, which uses a diacritic to indicate vowel quality rather than tone. Thus ĩ is a lower version of i; ũ is a lower version of u . Because the presence of some of the crucial morphemes can be masked by phonologically driven assimilation, reduction, or coalescence processes, I provide underlying shapes of all morphemes immediately above the morpheme-by-morpheme glosses. Kikamba certainly allows passives exhibiting promotion from object to subject, as shown here: (26) a. Kiveti nĩkyakũna kana. kĩ-veti nĩ-kĩ-a-kũn-a ka-ana 7-woman FOC-7-TNS-beat-FV 12-child ‘The woman beat a/ the child.’ b. Kana nĩkakũnwa (nĩ Kĩveti). ka-ana nĩ-ka-a-kũn-w-a (nĩ kĩ-veti) 12-child FOC-12-TNS-beat-PASS-FV (by 7-woman) ‘The child was beaten (by a/ the woman).’ (27) a. Asikalĩ nĩmakwata Kĩveti. a-sikalĩ nĩ-ma-a-kwat-a kĩ-veti 2-soldiers FOC-2-TNS-arrest-FV 7-woman ‘The soldiers arrested a/ the woman.’ b. Kĩveti nĩkyakwatwa. kĩ-veti nĩ-kĩ-a-kwat-w-a 7-woman FOC-7-TNS-arrest-PASS-FV ‘The woman was arrested.’ The passive suffix here is -w. The focus (FOC) prefix can occur on focused subjects, but in statements it most often occurs on the verb. FV stands for Final Vowel, conveying such information as mood and modality, while TNS indicates one of the four available past tenses. Note how the prefixed noun classes (numbered 1 to 17) provide evidence that kĩveti (the woman) is the subject of (26a), while kana (the child) is the subject of (26b). If a third-person subject is not pronounced, its noun class can nonetheless be deduced from verb prefix information. Kikamba does indeed allow null subjects, presumably due to a positive setting of the pro-drop parameter: (28) Nikyooka. nĩ-kĩ-a-ũk-a FOC-7-TNS-come-FV ‘She/it [class 7] came.’

< previous page

page_50

next page >

< previous page

page_51

next page >

Page 51 We might therefore expect to find null expletives in Kikamba, and indeed we can. A subset of Kikamba intransitive verbs (quite possibly the unaccusative subclass) occurs with either thematic or nonthematic subjects. This can be illustrated as follows: (29) a. Kana nĩkaenda. ka-ana nĩ-ka-a-end-a 12-child FOC-12-TNS-go-FV ‘The child went.’ b. Nikwaenda kana. nĩ-kũ-a-end-a ka-ana FOC-17-TNS-go-FV 12-child ‘There went a child.’ Subject-verb agreement with respect to noun class shows that the subject of (29a) is kana, while in (29b) an unpronounced expletive triggers the appearance of the class 17 prefix on the verb. With both a passive morpheme and a null expletive available, we might next predict that (29) can appear in passivized form. This is indeed the case. (30) a. Nĩkwaendwa. nĩ-kũ-a-end-w-a FOC-17-TNS-go-PASS-FV ‘Going occurred.’ b. Nĩkwaendwa (nĩ kana). nĩ-kũ-a-end-w-a (nĩ ka-ana) FOC-17-TNS-go-PASS-FV (by 12-child) ‘The child’s going occurred.’ Sentence (30) contains tense and so is a full clause, not merely a Small Clause. Our theoretical framework leads us to conclude that the verb in (30) has undergone head movement to the passive morpheme and then again to the tense; that is, it climbs to the head of a Small Clause and thence to the tensed inflectional head of TP. Although I have uncovered no direct evidence for the Small Clause in (30), positing it allows us to maintain a simpler, consistent overall approach to the human language faculty. Thus the grammar is simple rather than the analyses; the premises rather than the proofs. Unergative intransitive verbs (i.e., those that are not unaccusative) can also appear with a passive morpheme and an exp subject in Kikamba, as we would expect.

< previous page

page_51

next page >

< previous page Page 52 (31)

page_52

next page >

Kĩveti nĩkyavoya. kĩ-veti nĩ-kĩ-a-voy-a 7-woman FOC-7-TNS-pray-FV ‘The woman prayed.’ b. Nĩkwavoywa. nĩ-kũ-a-voy-w-a FOC-17-TNS-pray-PASS-FV ‘Some praying was done.’ The verb’s sole θ-role is assigned to the passive morpheme in (31b), and head-to-head (-to-head) movement as well as agreement work out exactly as for (30). Kikamba transitive verbs can be passivized in more than one way. If an expletive subject is selected, we obtain Polish-like impersonal passives, illustrated in (32). (32) a. Nĩkwakũnwa kana. nĩ-kũ-a-kũn-w-a ka-ana FOC-17-TNS-beat-PASS-FV 12-child ‘There was a child beaten.’ b. Nĩkwakwatwa Kĩveti. nĩ-kũ-a-kwat-w-a kĩ-veti FOC-17-TNS-arrest-PASS-FV 7-woman ‘There was a woman arrested.’ Of course, the presence of tense in (32) mandates full clausal inflection, with the verb climbing to the passive morpheme and thence to Tense via cyclic head-to-head movement. If, however, no expletive is drawn from the lexicon, the Patient undergoes (cyclic) movement to Spec, yielding sentences such as (26b) and (27b). Here, the key difference between Kikamba and English is the use of auxiliaries (in English the auxiliary carries tense, whereas in Kikamba the lexical verb does so, climbing to a higher head position) and the presence of noun class morphology, but these are orthogonal grammatical dimensions to passivization. A consequence of the noun class system is that Kikamba locatives can belong to a class, signaled by agreement triggered on a verb. Given this morphological property, locatives can be raised to specifier position, yielding active-passive pairs such as the following: (33) a. Kana nĩkaenda ndukanĩ. ka-ana nĩ-ka-a-end-a nduka-nĩ 12-child FOC-12-TNS-go-FV shop-LOC ‘The child went to the shop.’ a.

< previous page

page_52

next page >

< previous page

page_53

next page >

Page 53 b. Ndukanĩ nĩvaendwa (nĩ kana). nduka-nĩ nĩ-va-a-end-w-a (nĩ ka-ana) shop-LOC FOC-16-TNS-go-PASS-FV (by 12-child) ‘The shop was “gone to” (by the child).’ Finally, there is nothing to stop the choice of the null expletive (exp) as a subject, yielding the active in (34a), which corresponds to (33a), as well as the passive in (34b) corresponding to (33b). (34) a. Nĩkwaenda kana ndukanĩ. nĩ-kũ-a-end-a ka-ana nduka-nĩ FOC-17-TNS-go-FV 12-child shop-LOC ‘There went to the shop a child.’ b. Nĩkwaendwa ndukanĩ (nĩ kana). nĩ-kũ-a-end-w-a nduka-nĩ (nĩ ka-ana) FOC-17-TNS-go-PASS-FV shop-LOC (by 12-child) ‘There was “gone to” the shop (by the child).’ At first glance, the surface syntax of Kikamba might look quite different from the surface syntax of English. Yet the key differences are actually best viewed as lexical: Kikamba has noun entries organized into morphologically defined noun classes, whereas English has auxiliaries but no null lexical entry like exp. Thus the fact that (i) both Kikamba and English allow passives like (1b, 2b, 26b, 27b), but that (ii) English does not have passives such as (30, 31b, 34b) is due to (i) syntactic similarities between English and Kikamba and (ii) lexical differences. The same can be said when comparing English with Polish. One might hypothesize quite generally that in any human language with a passive morpheme, that morpheme accepts a θ-role and projects subject-predicate structure (an AgrP-like Small Clause). This yields the “non-canonical subject” property of passives, whereby Agents are backgrounded. The fact that, cross-linguistically, it is easier to find Patients as these “non-canonical subjects” is essentially due to the fact that object-to-subject promotion involves the least additional morphology. To find a construction like (6b) requires an additional lexical morpheme like there ; to find (30) requires exp; to find (33b) requires something like noun classes. Taking this observation to its logical conclusion, we can say that the less morphologically elaborated a passive clause, the closer we get to the essential syntactic heart of the phenomenon of passivization. Thus (24a) (Ziewano, “yawning occurred”), instead of being an odd passive-like construction, actually provides us with some of the most direct evidence available of the Small Clause structure within passive clauses.

< previous page

page_53

next page >

< previous page

page_54

next page >

Page 54 SMALL CLAUSES Recall that the Small Clause-based account was proposed on the basis of constituent order as in (6b) and agreement as in (11). It was mentioned earlier that Small Clauses can occur fairly freely as acceptably complete utterances in Polish. There is even a term in Polish traditional grammar for verbless clauses: równoważniki zdania (“sentence equivalents”). Some examples follow: (35) a. Odjazd tramwaju o czwartej godzinie. departure tram (GEN) about fourth hour ‘The tram leaves at four o’clock.’ b. Kto tam? who there ‘Who is there?’ c. Gdzie brak dyscypliny, tam where deficiency discipline(GEN) there brak porządku. deficiency order(GEN) ‘Where there is insufficient discipline there is insufficient order.’ d. Znowu zima. Anew/again winter ‘Winter has arrived again.’ A Small Clause (i.e., subject plus predicate, with no tense) analysis is straightforward here. Thus in (35a), which is a predication rather than one big nominal, odjazd tramwaju (the tram’s departure) is the subject (i.e., Spec of the Small Clause), while o czwartej godzinie (at 4 o’clock) is the predicate (technically, the phrasal complement of the abstract Small Clause head). Interestingly, this head can be realized overtly and pronounced as to: (36) a. Marek to profesor angielskiego. Mark SChead professor English (GEN) ‘Mark is an English professor.’ b. Chopin to wielki kompozytor. Chopin SChead great composer ‘Chopin is/was a great composer.’ It is perhaps this overt spelling out of the nonverbal Small Clause head that makes these various verbless clauses so prominent and comparatively common in Polish discourse. In line with this, they are not subject to the special intonational and discoursal restrictions of English data such as (9).

< previous page

page_54

next page >

< previous page

page_55

next page >

Page 55 Some caveats are in order, however. Note in particular that what is pronounced as to in Polish has a number of different uses, notably as a determiner for nouns of neuter gender (to dziecko “that child”) and as a pronominal (Zjedz to! ‘Eat that!’). But it should be clear that to is neither determiner nor pronominal in (36), where all nouns are masculine in gender. Note further that I have deliberately not analyzed the to in (36) as a copula, a claim for which the Polish case system provides good evidence. The sentences in (37) show how the complement of a Polish copula, if nominal, must take instrumental case: (37) a. Marek jest profesorem/*profesor. Mark is professor (INSTR)/professor (NOM) ‘Mark is a professor.’ b. Chopin by kompozytorem/*kompozytor. Chopin was composer(INSTR)/composer(NOM) ‘Chopin was a composer.’ The absence—and moreover the ungrammaticality—of instrumental case in (36) argues strongly that to is not simply a variant of a copula. Similar considerations argue against conceivable alternative analyses that invoke an unpronounced copula here. Moreover, the data in (36) behave exactly like Small Clauses with respect to tense. Thus, for instance, (36a) is basically timeless, while general knowledge (the fact that Chopin lived in the nineteenth century) can influence judgments of “anteriority” when discussing (36b). By contrast, Polish copular sentences exhibit the expected three-way morphological tense contrast, with overt marking of one of past or present or future. A to, which is difficult to translate, can actually be added to Polish full clauses, causing some adjustment to their focus-presupposition properties, as the following best approximations to English translations try to show: (38) a. Marek gotuje fasolę. ‘Mark is cooking beans.’ b. To Marek gotuje fasolę. ‘It is Mark who is cooking beans.’ c. Marek to gotuje fasolę. ‘What Mark is doing is cooking the beans.’ d. * Marek gotuje to fasolę. e. * Marek gotuje fasolę to.

< previous page

page_55

next page >

< previous page

page_56

next page >

Page 56 It is striking that the syntactic restrictions evident here in the placement of to can be deduced immediately from our analysis of non-determiner non-pronominal to as Small Clause head. Thus (38b– e) are all Small Clauses, the only legitimate structure for which is (39): (39) [SC Spec [head complement] ] Sentence (38c) fits (39) easily: Marek is the Spec, while the VP (gotuje fasolę) is the (predicative) complement. In (38b), the Spec is exp, and the complement is clausal. Finally, (38d,e) cannot be made to fit the frame in (39) and so are not well formed. In other words, mysterious word-order restrictions on a difficult-to-translate particle fall out naturally once one’s syntactic theory is sufficiently deep. It is worth adding that there are hints of direct evidence for Small Clauses in the literature on Bantu. Thus Myachina (1981:64) gives the following examples of an invariable particle (ni) in Swahili: (40) a. Mimi ni mwanafunzi. I particle pupil ‘I am a pupil.’ b. Mtoto wako ni mzuri sana. child your particle beautiful very ‘Your child is very beautiful.’ c. Mtu huyu ni mwanahewa. man this particle flyer ‘This man is a flyer.’ If the invariable ni of Swahili projects the same structure as the invariable to of Polish (something that is plausible given that they share a copula-like meaning, morphological invariability, and a syntactic position immediately after the subject), the sentences in (40) could well be Bantu Small Clauses. PROTOTYPICALITY We began by recalling the “received wisdom” that passives such as John was slapped are prototypical, while passives of intransitives, those with locative subjects, and those without object-to-subject promotion and the like are somehow special or marked. However, with a sufficiently elaborated view of syntactic structure (including phrasal projections for specifiers and complements of functional heads) and with a characterization of the passive morpheme as accepting a θ-role and projecting a Small

< previous page

page_56

next page >

< previous page

page_57

next page >

Page 57 Clause, we saw that all the cited passive data from Slavic and Bantu could be given a unified account. Contrary to initial appearances, passivization need not be seen as involving prototypicality. I think there are potentially important lessons here for linguistics in particular and for cognitive science in general. It is commonly accepted that some aspects of language show robust prototypicality effects. Thus Aitchison (1994) synthesizes two decades of research on lexical semantics exploring prototypicality in the domain of meaning. More recently, Pinker (1999) summarizes ten years of investigations into irregular and regular inflection. He argues that regular inflection behaves categorically, but irregular inflection (e.g., knew, flew, grew…) exhibits family resemblances, frequency effects, and prototypicality instead. If the source of prototypicality is the way the human mind organizes memory, it should be possible to see such effects across numerous linguistic domains. Sound symbolism seems to be one such domain. Note how clang, clap, and click all pertain to sound, and glow, gleam, and glitter to light. The fact that claw and gland need not be seen as counter-examples follows if we take the cited triplets to form family resemblance sets, instead of embodying a categorical phonetic-semantic correspondence. Similarly, it seems to me that some morphological processes, such as clipping (chem, phys, biol) and blending (brunch, smog) are frequency-dependent, a hallmark of prototypicality. Idioms may be a good example of prototypicality in syntax, if it is true that the more frequent idioms show greater syntactic freedom (e.g., let the cat out of the bag>the cat was let out of the bag>the cat is/was/will be out of the bag versus kick the bucket, ?*the bucket will be kicked ). Comrie (1989) advocates a still broader use of prototypicality in syntactic exploration. Thus he defines relative clauses by way of a prototype, and for him the prototype of subjecthood is the intersection of Agent and Topic. Proceeding in this way, it is tempting to discover prototypicality in passives. Yet I have argued that passives should not be seen in this way, because a more penetrating syntactic perspective yields a deeper, unified explanation. Is it possible to invoke prototypicality appropriately while simultaneously ensuring that it is not over-invoked? Perhaps a simple analogy with the history of mathematics will be enlightening. It is interesting to observe how “non-prototypical” numbers (i.e., those that are not counting numbers) have labels with off-putting connotations: witness negative numbers, irrational numbers, and imaginary complex numbers. All these shady characters were long treated as somewhat dubious and peripheral until a deep-enough extension of the notion of number (beyond counting) became available. Lakatos (1976), dealing primarily with geometry, explicitly talks about “monsters”: unusual entities with odd properties that theorists felt they

< previous page

page_57

next page >

< previous page

page_58

next page >

Page 58 did not have to account for fully, at least until a more abstract mathematics became available. In short, I suspect that pro to typicality is a genuine characteristic of human cognition, but that we should be careful to distinguish between language-internal prototypicality and methodological prototypicality. Irregular inflection, sound symbolism, and lexical semantics illustrate the former, whereas labeling some kinds of passives as non-core illustrates the latter. I see it as a temporary maneuver highlighting quirks (“monsters”) that call out for deeper theories. Because linguistics entails that human cognitive systems(i) study human cognitive systems(ii), prototypicality can arise at level (ii) (language internally) or indeed at level (i) (methodologically). Either way we learn something about human cognition, but at different levels. The former teaches us about language, the latter about linguists’ minds. To a cognitive scientist, both can be of interest. Thus Lakatos’s observations about mathematical cognition apply to linguistics at the methodological or analytical level and are one source of prototypicality, quite distinct from language-internal prototype effects. Hence we can detect in linguistics a feature of human cognition at two different levels (see Kipka, 2000, for a unified model that predicts this state of affairs). If the foregoing discussion is on the right track, prototypicality is one facet, but not the only significant facet, of language: Categorical effects are also waiting to be discovered there. Needless to say it is the categorical (in the form of “proofs”) that typifies published mathematics. The challenge for linguists is to foresee whether it is the categorical side of human cognition or the prototypicality branch or both interacting together (as in the history of mathematics, as opposed to most of the published mathematical literature) that helps us make further discoveries about how language reflects human cognition. REFERENCES Aarts, B. 1992. Small Clauses in English. Amsterdam: De Gruyter. Aitchison, J. 1994. Words in the Mind: An Introduction to the Mental Lexicon. Oxford: Blackwell. Akmajian, A. 1984. Sentence types and the form-function fit. Natural Language and Linguistic Theory 2:1–23. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Comrie, B. 1989. Language Universals and Linguistic Typology. Oxford: Blackwell. Haegeman, L., and Guéron, J. 1999. English Grammar: A Generative Perspective. Oxford: Blackwell. Jaeggli, O. 1986. Passive. Linguistic Inquiry 17:587–622. Keenan, E.L. 1985. Passive in the world’s languages. In Shopen, T. (ed.) Language Typology and Syntactic Description. Cambridge, U.K.: Cambridge University Press, 243–81.

< previous page

page_58

next page >

< previous page

page_59

next page >

Page 59 Kioko, A. 1994. Issues in the syntax of Kikamba: A Bantu language. Ph.D. thesis, Monash University. Kipka, P. 1990. Slavic aspect and its implications. Ph.D. dissertation MIT. ——. 2000. Premises, principles, procedures, prudence: A useful taxonomy of learning objectives. Paper presented at the Sources of Confusion conference, La Trobe University, November 2000. Lakatos, I. 1976. Proofs and Refutations: The Logic of Mathematical Discovery. Cambridge, U.K.: Cambridge University Press. Myachina, E.N. 1981. The Swahili Language: A Descriptive Grammar. London: Routledge and Kegan Paul. Pinker, S. 1999. Words and Rules: The Ingredients of Language. London: Weidenfeld and Nicolson. Vitale, A.J. 1981. Swahili Syntax. Dordrecht: Foris.

< previous page

page_59

next page >

< previous page

page_60

next page >

page_60

next page >

Page 60 This page intentionally left blank.

< previous page

< previous page

page_61

next page >

Page 61 4 The Split VP Hypothesis: Evidence from Language Acquisition Masatoshi Koizumi Young children acquiring English occasionally produce utterances with Object-Verb word order such as “Book read.” Since English is a head initial language, utterances with a preverbal object are not included in the input given to English-learning children. The question then arises as to why they produce such utterances. Assuming the Split VP Hypothesis (Koizumi, 1995), it is argued that preverbal objects obtain when the lower part of a clause up to AGRoP alone happens to be phonologically realized as a result of performance errors. INTRODUCTION Children acquiring English generally use the correct Subject-Verb-Object word order from the earliest stages of multiword speech (Bloom, 1970; Brown, 1973; Pinker, 1984; Valian, 1991; Radford, 1990, among many others; but see Bowerman, 1990, and Tsimpli, 1996, for a different view). Thus, a child at age of 2;1 (i.e., 2 years and 1 month) might produce utterances like “Mommy read book” (SVO), “Mommy read” (SV), and “Read book” (VO), but not utterances like “Read Mommy book” (VSO), for instance. It has been observed, however, that English-acquiring children at around 18 to 30 months of age sometimes produce utterances like those in (1), in which the notional object of a transitive verb occurs in preverbal position, instead of the canonical postverbal position (see, among others, Miller and Ervin, 1964; Brown et al., 1968; Braine, 1976; Bloom, 1970; Radford, 1990).

< previous page

page_61

next page >

< previous page

page_62

next page >

Page 62 (1) Preverbal objects in early child English a. Book read, (for ‘read book’, Susan 1;10), (Miller and Ervin, 1964) b. Balloon throw, (for ‘throw balloon’, Gia 1;7), (Bloom, 1970:86) c. Paper find, (for ‘find paper’, Adam 2;3), (Brown et al., 1968) d. Kendall pick up. (for ‘pick up Kendall’, Kendall 1;11), (Braine, 1976:15) Since the Object-Verb constituent order is not allowed in adult English, this type of utterance is not included in the input given to English-learning children. The question then arises as to why they produce such “exotic” sentences. I argue that the preverbal object in early child English is readily explained if we assume the Split VP Hypothesis (Koizumi, 1993, 1995; see also Johnson, 1991; Travis, 1992; and Fujita, 1993). According to the Split VP Hypothesis, a simple transitive clause contains two verbal heads rather than one, and they are separated by projections of a functional category, AGR. English sentences such as “We can solve the problem,” for instance, have a structure like that in (2), where the arrows indicate overt movement.

Crucially, in English, object undergoes overt raising (overt object shift), and its landing site is lower than the thematic position of subject, which in turn is lower than positions of modal and negation. Suppose now that only the lower portion of the clause up to AGRoP, that is, the portion shown in (3), is phonetically realized. Then, the Object-Verb word order obtains. (3) [AGRoP Objecti [AGRo’ Vj-AGRo [VP tj ti]]] I hypothesize that this is what happens in some children at the stage of early child English (roughly at age 18 to 30 months). When children are performing nonverbal as well as verbal activities at the same time, the full-blown linguistic representation occasionally fails to be fully interpreted by relevant performance systems, due to some limitation in cognitive capacity (e.g., limited information processing resources). When the structure up to AGRoP alone is interpreted by the performance system on the phonetic side, the resultant utterance includes a preverbal object, as shown in (4). (4) [AGRoP booki [AGRo’ readj-AGRo [VP tj ti]]]

< previous page

page_62

next page >

< previous page

page_63

next page >

Page 63 I suggest that utterances like those in (1) are produced in this way. The proposed analysis based on the Split VP Hypothesis accounts not only for the OV word order but also for the other peculiar properties of this type of utterance listed in (5). (5) Characteristics of utterances with a preverbal object in early child English a. Object precedes the verb that selects it. b. Subject is always absent, c. Modals and negation never occur. Sentences with a preverbal object (henceforth OV-sentences) consistently lack the subject constituent, auxiliaries such as will and can, and sentential negators such as not. These elements all belong to the upper part of the clause above AGRoP, which, according to the proposed analysis, is absent from the structure of OV-sentences. This analysis also predicts that preverbal objects are not produced by children acquiring a head-initial language without overt object shift such as French, because, in languages of this type, there is no subpart of a clause whose structure contains (a trace of) a transitive verb following (a trace of) its object. This prediction seems to be correct, as we will see later (see Further Prediction). This chapter has the following organization: After introducing the aspects of the Split VP Hypothesis pertaining to OV-sentences previous analyses of preverbal objects are reviewed, laying out the relevant facts. Next, a novel account of OV-sentences is proposed based on the Split VP Hypothesis. Then child English is compared with child French, providing additional support for the Split VP account of preverbal objects. Finally implications of the proposed analysis for theories of clausal architecture are discussed. THE SPLIT VP HYPOTHESIS In Chomsky (1986), transitive clauses such as John cut the orange were assumed to have a D-structure representation like (6). (6) [IP John I [VP cut the orange]] One of the most influential proposals advanced since then about clausal architecture is the Internal Subject Hypothesis (Fukui and Speas, 1986; Kitagawa, 1986; Kuroda, 1988; Koopman and Sportiche, 1991, among others). According to the Internal Subject Hypothesis, the thematic position of the subject is VP-internal, as shown in (7). (7) [IP I [VP John [cut the orange]]]

< previous page

page_63

next page >

< previous page

page_64

next page >

Page 64 Most empirical evidence for the Internal Subject Hypothesis cited in the literature suggests that in addition to the surface subject position (= Spec, IP), there is another position associated with the subject, and this position is lower than the surface position. Thus Sportiche (1988) argues that the “floating” quantifier all in (8), for example, is associated with the trace of the subject, as indicated here. (8) The childreni can [ [all ti] do it] Nakayama and Koizumi (1991) point out that examples of this sort do not bear out the hypothesis that the base-position of the subject is within the maximal projection of the main verb ( = VP). The examples simply indicate that the subject originates in a position lower than its surface position. Nakayama and Koizumi (1991) then present several pieces of evidence to show that the subject is “external” in the sense that its thematic position is outside VP, as shown in (9). (9) [IP I [XP Subj X [VP V Obj]]] Hale and Keyser (1991, 1993) and Bowers (1993) reached similar conclusions on independent grounds. The clause structure assumed in Chomsky (1995, 1999, 2000) can be regarded as a variant of (9), with X being a light verb (see also Nakayama, 1996, for a different view). Another important hypothesis put forward within the principles and parameters approach is the Split Infl Hypothesis (Pollock, 1989; Chomsky, 1991). Based on detailed observations about distributions of verbs, adverbs, negation, and other constituents in French and English, Pollock (1989) suggested that the socalled Infl is not a single syntactic category, but rather that there are (at least) two inflectional heads, Tense (T) and Agreement (AGR). Chomsky (1991) further argued that a single clause contains two (instead of one) Agreement Phrases, one for subject (AGRsP) and one for object (AGRoP). Chomsky’s (1991) structure is schematically shown in (10). (10) [AGRsP [TP [AGRoP [VP]]]] The Split Infl Hypothesis raises an interesting question regarding the relative ordering of AGRoP and the projection XP in (9): Does AGRoP dominate XP as in (11a) or does XP dominate AGRoP as in (11b)? (11) a. [AGRoP AGR [XP Subj X [VP V Obj]]] b. [XP Subj X [AGRoP AGR [VP V Obj]]]

< previous page

page_64

next page >

< previous page

page_65

next page >

Page 65 Koizumi (1993, 1995) argued in favor of the general structure (11b), and proposed the Split VP Hypothesis, according to which the basic structure of a simple transitive clause is roughly as follows (linear order aside, no movement is represented in [12]).

A single clause contains two verbal categories (represented here as v and V), the higher verb (= v) selecting the external argument, and the lower verb (= V) selecting the internal argument(s).1 According to the particular analysis presented in Koizumi (1993, 1995), English has overt object shift to Spec, AGRoP, and concomitant overt raising of V to AGRo to v. Thus, English transitive clauses assume structures like the following after all overt movements have taken place. (13)a.[AGRsP John1 AGRs [TP T [vP t1 opened [AGRoP the door2 tAGRo [VP tV t2]]]]] b.[AGRsP John1 AGRs [TP T [vP t1 rolled [AGRoP the ball2 tAGRo [VP down the hill [V’ tV t2]]]]]] French differs from English in (at least) two respects: (i) Objects do not undergo overt shift, staying within VP, and (ii) tensed verbs overtly raise up to AGRs. Thus, the French sentence in (14a) has a structure like (14b).

< previous page

page_65

next page >

< previous page

page_66

next page >

Page 66 (14) a. Pierre lit beaucoup de livres. Pierre read lots of books ‘Pierre reads lots of books.’ b. [AGRsP Pierre1 lit [TP tT [vP t1 tv [AGRoP tAGRo [VP tV beaucoup de livres]]]]] Untensed verbs in French are similar to English verbs in that they only raise up to the head of vP, as illustrated here. (15) a. Pierre1 a [vP t1 lu [AGRoP tAGRo [VP tV Pierre has read beaucoup de livres]]] lots of books ‘Pierre has read lots of books.’ b. Ne pas [vP regarder [AGRoP tAGRo [VP tV la télévision]]]… not watch the television ‘Not to watch television…’ The presence of overt object shift in English and its absence thereof in French is responsible for the following contrast: (16) a. Peter has hardly seen Mary, b. *Peter has seen hardly Mary. (17) a. Pierre a á peine vu Marie, b. Pierre a vu á peine Marie. Given these clause structures, the VP-adverb hardly is merged with a projection of the higher verb in (16a), and it is adjoined to AGRoP in (16b), as shown in (18). The VP-adverb á peine “hardly” in (17a) and (17b) is associated with a projection of the higher verb and a projection of the lower verb, respectively, as shown in (19). (18) a. Peter1 has [vP hardly [vP t1 seen [AGRoP Mary2 tAGRo [VP tV t2]]]] b. *Peter1 has [vP t1 seen [AGRoP hardly [AGRoP Mary2 tAGRo [VP tV t2]]]] (19) a. Pierre1 a [vP á peine [vP t1 vu [AGRoP tAGRo [VP tV Marie]]]] b. Pierre1 a [vP t1 vu [AGRoP tAGRo [VP á peine [VP tV Marie]]]] (18a), (19a), and (19b) all conform to the traditional assumption that adverbs modify (and/or are licensed by) projections that are semantically compatible with them; (18b), however, violates this constraint, because the adverb modifies AGRoP, which is a projection of a semanti-

< previous page

page_66

next page >

< previous page

page_67

next page >

Page 67 cally null “pure” functional category. Note that the English example in (16b) and the French sentence in (17b) have the same surface word order, yet they have different structures (as shown in [18b] and [19b]). Because English has overt object shift as well as short verb raising, there is no verbal projection between the surface positions of the verb and its nominal object. On the other hand, in French, verbs undergo raising but objects stay in situ, creating structures in which verbal projections intervene between a verb and its object. These verbal projections may be modified by adverbs. Thus, in a sense, the presence versus absence of overt object shift is responsible for the difference in grammaticality between (16b) and (17b). The existence of two separate verbal heads (v and V) in English and French can be more clearly shown by the following examples, in which vP and VP are modified by different VP-adverbs. (20) a. He had deliberately rolled the ball gently down the hill. (Radford, 1997) b. souvent faire mal ses devoirs frequently do badly one’s homework (c’est stupide.) (that is stupid) ‘To frequently do one’s homework badly (is stupid.)’ (Iatridou, 1990) In the English sentence (20a), deliberately adjoins to vP (or v’), and gently to VP. In the French infinitival example (20b), souvent and mal modify vP (or v’) and VP, respectively. This is schematically shown in (21). (21)a.He1 had [vPdeliberately [vP t1 rolled [AGRoP the ball2 tAGRo [VPgently [VP down the hill [V’ tv t2]]]]]] b.[vPsouvent [vP t1 faire [AGRoP tAGRo [VPmal [VP tv ses devoir]]]]] PREVERBAL OBJECTS A number of researchers have observed that (at least some) English-acquiring children go through a stage at which utterances with a transitive verb and its two arguments (i.e., subject and object) consistently follow the correct Subject-Verb-Object (SVO) order, but utterances consisting solely of a transitive verb and its object (lacking subject) exhibit variability in word order, alternating between OV and VO order (Brown et al., 1968; Bloom, 1970; Braine, 1976; Bowerman, 1990: Radford, 1990, among many others). Brown et al. (1968), for instance, state that according to their study of three children, Adam, Eve, and Sarah (approximately two to three years old) (see also Brown, 1973):

< previous page

page_67

next page >

< previous page

page_68

next page >

Page 68 (…) in most utterances it is clear that the intended subject and object are the constituents found in subject and object position (…) There are in the records a handful of exceptions in which the intended object seems to be in subject position [such as those in (23) below] (…) but these are the only exceptions in thousands of well-ordered sentences, (p. 307) (23) Adam’s (2;3) OV-sentences a. Paper find, b. Paper write. Similarly, Bloom (1970) notes that in examples like (24) by Gia (1;7) “the noun that occurred in sentence-initial position was an object noun …yet object nouns always occurred in sentence-final position in constructions with other nouns” (p. 87). (24) Gia’s (1;7) OV-sentence Balloon throw. Braine (1976) made a similar remark with respect to a corpus of Kendall (1;11), which contains six SVO utterances, seven VO utterances, seven OV utterances, and crucially no utterances with SOV order: “the actor constituent tends to occupy first position when it is present…[but] there is no verb-object formula for composing action phrases” (p. 20). (25) Kendall’s (1;11) OV-sentences a. Kimmy kick, (‘kick Kimmy’) b. Kendall pick up. (‘pick up Kendall’) c. Doggie sew. (‘sew doggie’) Likewise, Bowerman (1990) states: Although word order was generally correct for two-argument strings in Christy’s corpus, single-argument strings were often misordered. In fact, in the two-week period of 1;11 [weeks 1 and 2]—two months after the onset of two argument strings with prototypical agents and patients—the number of incorrectly ordered (OV) strings with prototypical patients was two-thirds that of correctly ordered (VO) strings (12 tokens, 8 verb types vs. 19 tokens, 12 verb types.) (p. 1275) [In the next two-week period, i.e. 1;11 [weeks 3 and 4]], [t]he number of misordered OV strings with proto. verbs drops to one-third that of correctly ordered VO strings, (p. 1272) (26) OV-sentences cited in Bowerman (1990):2 a. Grass cut. b. Milk drink, c. String pull.

< previous page

page_68

next page >

< previous page

page_69

next page >

Page 69 These observations are further supported by the results of our search of CHILDES database (MacWhinney, 2000), shown in (27). (27a) shows the numbers of five types of utterances found in the first file of Adam’s (Adam01.cha), studied by Brown et al. (1968), and (24b) shows the numbers of five types of utterances in the file (04c.cha), studied by Valian (1991), which contains utterances of a child at the age of 1;10.3 (27) a. Adam’s (2;3) utterances b. One of Vali an’s children (1;10) Word order # of utterances Word order # of utterances SVO 38 SVO 40 SOV 0 SOV 0 VO 108 VO 12 OV 5 OV 2 Total 151 Total 54 As shown in (27a), the relevant file of Adam (2;3) includes 151 utterances that clearly contain a transitive verb and its nominal object. Of these 151 sentences, only five are OV-sentences (none of which have overt subject) such as “Paper find,” 108 are VO-sentences (without overt subject) such as “Drive bulldozer,” and 38 are SVO-sentences (with overt subject) such as “Kitty write paper.” There are no utterances that contain both preverbal object and overt subject. Similarly, as shown in (27b), one of Valian’s files in CHILDES, 04c.cha, contains 54 utterances with a transitive verb and its object. Among them, only two are OV-sentences such as “Door push in,” 12 are VO-sentences such as “Take my animal,” and 40 are SVO-sentences such as “I make bear’s home.” It seems safe to conclude from these observations that children at the stage of early child English have already acquired the correct SVO order but that they occasionally produce utterances with a preverbal object, which crucially is lacking a subject. This generalization raises the interesting question of why children produce preverbal objects at all if they know the correct SVO order. Bloom (1970:87) suggests that children at this stage have “learned something about the subject-object order of English sentence constituents” but have “not yet learned the relational or syntactic features of verb forms with respect to nouns.” Similarly, Braine (1976:20) suggests that “the fact that the actor constituent tends to occupy first position when it is present is accounted for by the well-learned actor-action schema,” and that the apparent free word order of utterances with a transitive verb and its object argues that children at this stage have not learned “verbobject formula for composing action phrases.” Although this may be a reasonably adequate description of the relevant aspect of the data they studied, it cannot be considered an explanation of preverbal objects in early child English in general. This is so for the following reasons. First, if children

< previous page

page_69

next page >

< previous page

page_70

next page >

Page 70 at the stage of early child English have not learned the correct order of verb and object, and use OV and VO order interchangeably, we expect that, other things being equal, roughly the same number of utterances with a preverbal object are produced as the utterances with a postverbal object. However, the frequency of OV-sentences is much lower than the frequency of VO-sentences, as we have already seen. Second, if children at the relevant stage are freely using VP with a preverbal object (as well as VP with a postverbal object), it is predicted that they would produce utterances with SOV word order such as “Mommy book read,” which in fact are virtually unobserved, as mentioned previously. Bloom (1970:88) considers an alternative account of preverbal objects, according to which what appears to be a preverbal object is in fact a topic. In this analysis, “Balloon throw” in early child English, for example, corresponds to “The balloon, I throw it” in adult English. The topic analysis is attractive in that it assumes no difference between the grammars of adult English and early child English in the relevant respects except that function words (including pronouns) tend to be dropped in children’s utterances; however, there are several difficulties with it. First of all, it is questionable whether such operations as topicalization and/or left dislocation are already available to children at this stage (Hyams and Wexler, 1993). Second, OV-sentences are not necessarily observed in a context where the referent of the preverbal object has been established as a discourse topic or focus. The OV-sentence in (23b), for example, was produced by Adam when he and his mother were talking about “writing,” rather than “papers,” as shown in (27) (taken from Adam01.cha in CHILDES). (28) ADA: Adam write # hmm? ADA: I write. ADA: I can write. MOT: You can’t write. MOT: He can write but Adam can’t write yet. ADA: Paper write. ADA: Write the [?] paper. Incidentally, it is interesting to note that Adam attempted self-correction after he produced the OVsentence in the preceding conversation. In fact, this is a typical case: Self-correction is frequently observed after utterances with a preverbal object in early child English. The final problem with the topic analysis is that it fails to account for the important fact that sentences with a preverbal object consistently lack subject (see [5b]). The topic analysis wrongly predicts that children who say “Balloon throw” would also say “Balloon Mommy throw” productively.

< previous page

page_70

next page >

< previous page

page_71

next page >

Page 71 Capitalizing on the characteristic (5b) (i.e., absence of subject), Radford (1990:232) argues that sentences like “Book read” are intransitives in the grammar of the children who produce them. Since the “preverbal object” IS subject, according to Radford, there should be no other subject in the same clause. Radford’s analysis then successfully accounts for the characteristic (5b) (i.e., absence of subject) as well as (5a) (i.e., OV order). A problem with his analysis is that it cannot easily accommodate the property (5c) (i.e., absence of modals and negation). Although modals and negation are not completely productively used by children at the stage of early child English, they do sometimes occur. Thus, the same children saying “Book read” occasionally produce utterances with a modal or negation such as “I can take off” and “No Mommy get up.”4 The following examples are taken from Adam01.cha. (29) Adam’s (2;3) utterances with a modal or negation a. No I see truck. (= ‘I don’t see truck’) b. I may give some, c. I can write, d. Must go. According to Radford’s analysis, then, it is expected that children at this stage would produce utterances like “Paper can find” and “No book read,” which are in fact not observed. Another potential problem with his analysis has to do with cross-linguistic difference with respect to the availability of preverbal objects. As we will see in the section, Further Prediction, preverbal objects are not observed in utterances by French-learning children. This is quite unexpected in Radford’s analysis, which mentions no grammatical aspects that can be different in English and French. Finally, if children believe that the preverbal object is subject, it is not clear why they often attempt self-correction after OV-utterances as mentioned previously. A SPLIT VP ACCOUNT I suggest that the preverbal object in early child English is straightforwardly explained if we assume the Split VP Hypothesis proposed by Koizumi (1993, 1995) within a recent framework of generative grammar called the Minimalist Program (Chomsky, 1995, 1999, 2000).5 As we have seen, according to the Split VP Hypothesis a clause contains two verbal heads rather than one, and they are separated by projections of a functional category, AGR. The external argument is merged as a specifier of the upper verb phrase (vP), and the internal arguments are merged within the lower verb phrase (VP). In English, the subject DP and the object DP overtly move to the Spec of AGRsP and the Spec of AGRoP,

< previous page

page_71

next page >

< previous page

page_72

next page >

Page 72 respectively. The main verb V moves to AGRo, which in turn raises to the upper verb, both in the overt syntax. Thus, sentences such as “Mommy read the book” have a structure like (30), where the arrows indicate overt movements.

Note that the object in English undergoes overt object shift, and that its landing site is lower than the original (or “base”) positions of subject, modal, and negation. Assuming, with Chomsky (1995, 1999, 2000), that movement is a copying operation (a copy theory of movement), the sentence will have a structure like (31) after all overt movements take place. (31)[AGRsP Mommy [AGRs’ AGRs [TP T [vP Mommy [v’ read- AGRo -v [AGRoP the book [AGRo’ readAGRo [VP read the book]]]]]]]] Given the conventional assumption that only the highest member among identical copies is pronounced, the lower occurrences of Mommy, the book and read are not phonetically realized, giving rise to the correct surface word order, as shown in (32). (32) [AGRsP Mommy [AGRs’ AGRs [TP T [vP [v’ read- AGRo –v [AGRoP the book [AGRo’ [VP ]]]]]]]] Suppose that for some reason only the lower portion of the clause up to AGRoP (i.e., the portion shown in [33]), rather than the entire clause, is interpreted by the performance system on the phonetic side. Then, the Object-Verb word order obtains, as shown in (34). (33) [AGRoP the book [AGRo’ read-AGRo [VP read the book]]] (34) [AGRoP the book [AGRo’ read-AGRo [VP ]]] I suggest that this is what happens in some children at the stage of early child English. It is well known that “there is a computational bottleneck which prohibits the child from expressing all that she knows” (Crain and Lillo-Martin, 1999:27); see also Bloom (1990, 1993) and references cited there. When young children are performing nonverbal and verbal activities at the same time, the full-blown linguistic representation occasionally fails to be fully interpreted by the relevant performance system, due to the limited resources available to them. Thus, young children might

< previous page

page_72

next page >

< previous page

page_73

next page >

Page 73 say “Mommy read” or “Read book” when they in fact want to say “Mommy read the book.” I hypothesize that OV-sentences obtain when the structure up to AGRoP alone happens to be interpreted by the performance system on the phonetic side. That is, an utterance with OV word order is produced as a result of performance errors caused by “computational bottleneck.” Subject, modal, and negation do not occur in OV-sentences because they all belong to the upper part of a clause higher than AGRoP, which, by hypothesis, is absent from the structure of OV-sentences. As pointed out in the previous section, children often attempt self-correction after OV-sentences, presumably because children are aware that OV-sentences are the result of performance errors. The Split VP-based account of preverbal objects is further supported by utterances with a verb with two internal arguments. Put, for example, takes a Theme DP and a Goal (or Location) PP as its internal arguments. There are six logically possible orderings of put and the internal arguments. Of these six possibilities, only four are actually attested in the corpora of early child English that I have studied.6 This is shown in (35). (35) a. PPGoal V DPTheme : Outside put book. (Adam 2;4) b. V PPGoal DPTheme: Put in chairs breakfast. (Peter 2; 3) c. DPTheme V PPGoal: Telephone put on the table. (Eve 1;10) d. V DPTheme V PPGoal: Put book outside. (Adam 2;4) e. DPTheme PPGoal V: NONE f. PPGoal DPTheme V: NONE Given the Split VP-based account of the preverbal object, the paradigm in (35) is readily explained as follows. As shown in (36), (35a) obtains when only the VP portion of the relevant sentence is pronounced, (35b) when the structure up to AGRo’ is pronounced, (35c) when the structure up to AGRoP is pronounced, and (35d) when the structure up to v’ is pronounced. (36) a. [VP outside [V’ put book]] b. [AGRo’ put-AGR [VP in chairs [V’ breakfast]]] c. [AGRoP telephone [AGRo’ put-AGR [VP on the table [V’ ]]]] d. [V’ put-AGR-v [AGRoP book [AGRo’ [VP outside V’ ]]]] There are two important points in the context of our discussion here. First, under the Split VP-based account, there is no way to produce utterances with the word order in (35e) or (35f): no matter what portion of a clause is pronounced, these orders cannot be obtained. Thus, the word orders in (35e) and (35f) should be absent from early child English. This expectation seems to be in accord with

< previous page

page_73

next page >

< previous page

page_74

next page >

Page 74 the fact, as already mentioned. The second point is that the present analysis predicts that if a structure larger than v’ (e.g., vP, AGRsP) is pronounced, the resultant utterance will assume the same word order as (35d), with or without subject, modal, and/or negation. In other words, utterances like (35a, b, c) should always lack subject, modal, and negation. As far as I know, this prediction, too, is borne out.7 (37) a. My put it back together. (CHI 1;10=the same child as in [27b]) b. No put me in the wastebasket. (Eve 1;10) c. Don’t put it on the table. (Naomi 2;4, cf. Sachs, 1983) In summary, I have suggested, assuming the Split VP Hypothesis, that OV-sentences are produced when only the structure up to AGRoP alone happens to be phonetically realized as a result of performance errors. This analysis accounts not only for their OV word order, but also for the other peculiar properties such as the absence of subject, modal, and negation. FURTHER PREDICTION There is further evidence for the Split VP account of OV-sentences in early child English proposed in the previous section. This analysis crucially assumes that object undergoes overt shift across the verb, both in child English and in adult English. Without overt object shift, object-verb word order would not arise in sentences of head initial languages even at intermediate stages of their derivations. This analysis then predicts that children learning a head-initial language without overt object shift, such as French, do not produce utterances with a preverbal object. As far as I can tell, this prediction seems to be correct. According to our search of the French files in CHILDES database, there is no child utterance containing a non-clitic object before the verb. For example, one of the French files donated by Christian Champaud, GREG07.CHA (which contains utterances of a child named Grégoire [2;3]) includes 8 SVO-sentences, one utterance with Subject clitic-object Verb order (SclV), 26 VO-sentences (without overt subject), and crucially no OV-sentence. The numbers of each type of utterance contained in four files of Champaud and two files of Leveillé (which contain utterances of Philippe) are shown in Table 4.1 (cf. Suppes et al., 1973). Some of their utterances are given in (38) and (39). (38) Grégoire’s (2;5) utterances a. SVO il prend un crayon, he takes a crayon

< previous page

page_74

next page >

< previous page

next page >

page_75

Page 75 TABLE 4–1.

SVO

ScIV

VO

0V

GREG04.CHA (Grégoire [1;11]) 5 0 7 0 GREG06.CHA (Grégoire [2;1]) 8 0 10 0 GREG07.CHA (Grégoire [2;3]) 8 1 26 0 GREG08.CHA (Grégoire [2;5]) 23 10 4 0 Phil01.cha (Philippe [2;1]) 5 2 34 0 Phil 10.cha (Philipper [2;3]) 26 3 25 0 b. ScIV Je le prends pour moi. I it take for myself c. VO pas touché ma photo, not touch my photo d. OV NONE (39) Philippe’s (2;1) utterances a. SVO J’aime les croissants. I love the croissants b. ScIV On le met lá. you it put there c. VO Mettre dans le bol le sucre. put in the bowl the sugar d. OV NONE Note that the contrast between English and French with respect to preverbal objects (i.e., the presence of preverbal objects in child English vs. the absence thereof in child French) is totally unexpected under the previous analyses discussed in the section Preverbal Objects, because none of them crucially refers to grammatical properties of English that are not shared with French. Thus, to the extent that it is real, the contrast constitutes a strong piece of supporting evidence for the Split VP account of the preverbal object in early child English. CLAUSAL ARCHITECTURE The Split VP account of the preverbal object outlined in the section, A Split VP Account, has two crucial assumptions, which are shown in (40).

< previous page

page_75

next page >

< previous page

page_76

next page >

Page 76 (40)a.Objects in English undergo overt raising out of VP (object shift). b.The landing site of object shift is sufficiently lower than the original positions of subject, modal, and negation. The Split VP analysis of adult English proposed in Koizumi (1993, 1995) precisely has these properties, providing us with the right kind of structure with which to account for the preverbal object in early child English. In contrast, most theories of clausal architecture currently popular in the field do not share at least one of these assumptions/claims. They thus fail to accommodate all of the properties listed in (5) and repeated here as (41). (41) Characteristics of utterances with a preverbal object in early child English a. Object precedes the verb that selects it. b. Subject is always absent, c. Modals and negation never occur. Chomsky (1995, 1999, 2000), for instance, assumes a structure like (42) for adult English. (42) [TP wei can [vP ti solve-v [VP tV the problem]]] In his analysis, the object stays in its original position throughout the derivation, and there is no point at which the object precedes the verb. Thus, it is not clear at all why English-learning children should prepose the object across the verb (i.e., why they should produce utterances with a preverbal object). If object shift is incorporated into Chomsky’s structure, we will obtain either of the following (cf. Ura, 2000): (43) a. [TP wei can [XP solve-v-X [vP [the problem]j [v’ ti [v’ tv [VP tV tj ]]]]]] b. [TP wei can [XP solve-v-X [vP ti [v’ [the problem]j [v’ tv [VP tV tj]]]]]] Suppose that vP and the lower portion alone are interpreted at the articulatory-perceptual interface. Then the relevant structures should look like (44). (44) a. [vP the problem [v’ we [v’ solve-v [VP ]]]] b. [vP we [v’ the problem [v’ solve-v [VP ]]]] In these structures, the object precedes the verb, correctly accounting for the characteristic (41a). However, they include the subject constituent as well, which is inconsistent with the characteristic (41b). Thus, Chomsky’s (1995, 1999, 2000) structure plus object shift to one of the Specs of vP

< previous page

page_76

next page >

< previous page

page_77

next page >

Page 77 cannot explain why OV-sentences are found in early child English.8 Similar remarks apply to analyses that posit AGRoP above the thematic position of subject. For example, McCloskey (2000) proposes a structure like (45a). Given this, when the structure up to AGRoP (which is the smallest constituent including the shifted object) is phonetically realized, (a copy of) the subject is necessarily included in it, as shown in (45b). (45) a. we1 can solve [AGRoP [the problem]2 tAGRo [vP t1 tv [VP tV t2]]] b. [AGRoP the problem solve-v-AGRo [vP we [VP ]]] Takano (1998) proposes an analysis of English according to which an accusative object undergoes a kind of object shift within VP (which he calls “short scrambling”) rather than to a VP-external position. On this analysis, John gave a book to Mary, for example, has a structure like (46). (46) [TP Johni T [vP ti [v’ gavej-v [VP a bookk [VP to Mary [V’ tj tk]]]]]] This analysis could account for simple OV-sentences such as “Book read,” assigning to them a structure like (47), on the assumption that they are produced when the lower VP alone is properly interpreted by the performance system on the phonetic side. (47) [VP book [VP read ]] However, slightly more complex examples such as “Telephone put on the table” could not be easily accommodated within Takano’s system, because in his analysis of double complement constructions shown in (46), there is no partial structure (nor intermediate stage of its derivation) with Theme-VerbGoal order, corresponding to “Telephone put on the table.” Worse, if only VP in a double complement construction is phonetically realized, parallel to the case in (47), then the resultant structure will be something like (48). (48) [VP a book [VP to Mary [V’ gave ]]] We would then expect that children producing utterances like “Book read” would also produce utterances with Theme-Goal-Verb order such as “Book (to) Mary give.” However, such utterances are not in fact observed, as we have already seen. Furthermore, since Takano assumes the same analysis for English and French with respect to object shift (his short scrambling of object), the contrast between the two languages discussed in the previous section is totally unexpected. Thus, Takano’s (1998) analysis of clause structure cannot explain the

< previous page

page_77

next page >

< previous page

page_78

next page >

Page 78 preverbal object in early child English and its absence in early child French. CONCLUSION In this chapter, I have discussed utterances like “Book read” in early child English. They have three characteristics that are not found in adult English: 1) object precedes the verb that selects it, 2) subject is always absent, and 3) auxiliaries and negation never occur. If we assume the Split VP Hypothesis (Koizumi, 1993, 1995), they are readily explained as a phonetic realization of a syntactic representation up to AGRoP. Other theories of clausal architecture such as Chomsky’s (2000) cannot straightforwardly account for them. Thus, to the extent that the Split VP-based account of preverbal objects in early child English is on the right track, the core aspects of the Split VP Hypothesis receive substantial empirical support. Needless to say, it is both desirable and necessary for the proposed analysis of preverbal objects to be tested against a wider range of data than has been possible in this chapter, not only in English and French but also in other languages. This will be a task for future research. NOTES 1. In Koizumi (1993, 1995) the higher verb is referred to as Vu (upper verb), and the lower one as V1 (lower verb). Here I adopt a simpler notation, v for the former and V for the latter. 2. The examples in (26) are either Christy’s or Eva’s, but it is not shown in Bowerman (1990) which example is produced by which child. 3. In counting the number of utterances, self-repetitions were excluded. Thus, if a child said “Read book” three times in a row, they were counted as a single utterance of VO word order. The same convention was applied throughout the research reported in this paper. 4. For sentential negation in early child English, see Déprez and Pierce (1993) and the references cited there. 5. For recent discussions of the Split VP Hypothesis, see, among others, Bošković (1997), Harley and Noyer (1997), Lasnik (1999) and Ausin (1999). 6. They include all of the files contained in the following corpora in CHILDES. (i) a. Bloom (1970) and Bloom (1973) (cf. Bloom, 1970, 1973) b. Braine (cf. Braine, 1976) c. Brown (cf. Brown, 1973) d. Sachs (cf. Sachs, 1983) e. Valian (cf. Valian, 1991) 7. Examples (37b) and (37c) lack subjects for reasons independent of preverbal objects. Note that although the Split VP-based account of the preverbal

< previous page

page_78

next page >

< previous page

page_79

next page >

Page 79 object predicts that no utterances with a preverbal object will have an overt subject, it does not predict that all utterances without an overt subject have a preverbal object. For discussions of “subject drop” see Hyams (1986), Radford (1990), Bloom (1990, 1993), Hyams and Wexler (1993), Rizzi (1993/1994, 1994) and Wexler (1994), among many others. 8. Another problem with this analysis is that the status of the category immediately above vP, shown as X in (43), is not clear at all. REFERENCES Ausin, A. 1999. Spelling out multiple spell-out. Paper presented at the 2nd Asian GLOW, Nanzan University. Bloom, L. 1970. Language Development: Form and Function in Emerging Grammars. Cambridge, MA: MIT Press. ——. 1973. One Word at a Time: The Use of Single-Word Utterances before Syntax. The Hague: Mouton. ——. 1990. Subjectless sentences in child language. Linguistic Inquiry 21:491–504. ——. 1993. Grammatical continuity in language development: The case of subjectless sentences. Linguistic Inquiry 24:721–34. Bošković, Z. 1997. Coordination, object shift, and V-movement. Linguistic Inquiry 28:357–65. Bowerman, M. 1990. Mapping thematic roles onto syntactic functions: Are children helped by innate linking rules? Linguistics 28:1253–89. Bowers, J. 1993. The syntax of predication. Linguistic Inquiry 24:591–656. Braine, M. 1976. Children’s First Word Combinations. Monographs of the Society for Research in Child Development 41, Serial no. 164. Brown, R. 1973. A First Language: The Early Stages. Cambridge, MA: Harvard University Press. Brown, R., Cazden, C., and Bellugi, U. 1968. The child’s grammar from I to III. In J.P.Hill (ed.) Minnesota Symposium on Child Development 2. Minneapolis: University of Minnesota Press, 28–73. Chomsky, N. 1986. Barriers. Cambridge, MA: MIT Press. ——. 1991. Some notes on economy of derivation and representation. In Robert Freidin (ed.) Principles and Parameters in Comparative Grammar. Cambridge, MA: MIT Press, 417–54. ——. 1995. The Minimalist Program. Cambridge, MA: MIT Press. ——. 1999. Derivation by Phase. MIT Occasional Papers in Linguistics 18. Cambridge, MA: MIT Working Papers in Linguistics. ——. 2000. Minimalist inquiries: The framework. In R.Martin, D.Michaels, and J.Uriagereka (eds.) Step by Step. Cambridge, MA: MIT Press, 89–155. Crain, S., and Lillo-Martin, D. 1999. An Introduction to Linguistic Theory and Language Acquisition. Maiden: Blackwell. Déprez, V., and Pierce, A. 1993. Negation and functional projections in early grammar. Linguistic Inquiry 24:25–67. Fujita, K. 1993. Object movement and binding at LF. Linguistic Inquiry 24:381–88. Fukui, N., and Speas, M. 1986. Specifiers and projection. In N.Fukui, T.R. Rapoport, and E.Sagy (eds.) MIT Working Papers in Linguistics 8: Papers in Theoretical Linguistics. Cambridge, MA: MIT Working Papers in Linguistics, 128–72. Hale, K., and Keyser, S.J. 1991. Lexicon Project Working Papers 34: On the Syntax of Argument Structure. Cambridge, MA: MIT Working Papers in Linguistics.

< previous page

page_79

next page >

< previous page

page_80

next page >

Page 80 ——. 1993, On argument structure and the lexical expression of syntactic relations. In K.Hale and S.J.Keyser (eds.) The View from Building 20. Cambridge, MA: MIT Press, 53–109. Harley, H., and Noyer, R. 1997. Mixed nominalisations, object shift and short verb movement in English. In Proceedings of NELS 28, University of Massachusetts at Amherst: GLSA. Hyams, N. 1986. Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Hyams, N., and Wexler, K. 1993. On the grammatical basis of null subjects in child language. Linguistic Inquiry 24:421–59. Iatridou, S. 1990. About Agr(P). Linguistic Inquiry 21:551–77. Johnson, K. 1991. Object positions. Natural Language and Linguistic Theory 9:577–636. Kitagawa, Y. 1986. Subjects in Japanese and English. Ph.D. dissertation, University of Massachusetts, Amherst. Koizumi, M. 1993. Object agreement phrases and the Split VP hypothesis. In J. Bobaljik and C.Phillips (eds.) MIT Working Papers in Linguistics 18: Papers on Case and Agreement I. Cambridge, MA: MIT Working Papers in Linguistics, 99–148. ——. 1995. Phrase Structure in Minimalist Syntax. Ph.D. dissertation, MIT. (A revised version published by Hituzi Syobo, Tokyo, 1999) Koopman, H., and Sportiche, D. 1991. The positions of subjects. Lingua 85:211–58. Kuroda, S.-Y. 1988. Whether we agree or not: A comparative syntax of English and Japanese. Linguisticae Investigationes 12:1–47. Lasnik, H. 1999. Pseudogapping Puzzles. In S.Lappin and E.Benmamoun (eds.) Fragments: Studies in Ellipsis and Gapping. Oxford: Oxford University Press, 141–74. MacWhinney, B. 2000. The CHILDES Project: Tools for Analyzing Talk, 3d ed. Hillsdale, NJ: Lawrence Erlbaum. McCloskey, J. 2000. Quantifier float and wh-movement in an Irish English. Linguistic Inquiry 31:57–84. Miller, W.R., and Ervin, S.M. 1964. The development of grammar in child language. In U.Bellugi and R.Brown (eds.) The Acquisition of Language. Monographs of the Society for Research in Child Development 29, 9–34. Nakayama, M. 1996. Acquisition of Japanese Empty Categories. Tokyo: Kurosio Publishers. Nakayama, M., and Koizumi, M. 1991. Remarks on Japanese subjects. Lingua 85:303–19. Pinker, S. 1984. Language Learnability and Language Development. Cambridge, MA: Harvard University Press. Pollock, J.-Y. 1989. Verb movement, UG and the structure of IP. Linguistic Inquiry 20:365–424. Radford, A. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Blackwell. ——. 1997. Syntactic Theory and the Structure of English: A Minimalist Approach. Cambridge, U.K.: Cambridge University Press. Rizzi, L. 1993/1994. Some notes on linguistic theory and language development: The case of root infinitives. Language Acquisition 3.4:371–93. ——. 1994. Early null subjects and root null subjects. In T.Hoekstra and B. Schwartz (eds.) Language Acquisition Studies in Generative Grammar. Amsterdam: John Benjamins, 151–76.

< previous page

page_80

next page >

< previous page

page_81

next page >

Page 81 Sachs, J. 1983. Talking about the there and then: The emergence of displaced reference in parent-child discourse. In K.E.Nelson (ed.) Children’s Language, Vol. 4. Hillsdale, NJ: Lawrence Erlbaum. Sportiche, D. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19:425–49. Suppes, P., Smith, R., and Leveillé, M. 1973. The French syntax of a child’s noun phrases. Archives de Psychologie 42:207–69. Takano, Y. 1998. Object shift and scrambling. Natural Language and Linguistic Theory 16:817–89. Travis, L. 1992. Derived objects, inner aspect, and the structure of VP. Ms. McGill University (Presented at NELS 22, University of Delaware). Tsimpli, I.-M. 1996. The Prefunctional Stage of First Language Acquisition: A Cross-linguistic Study. New York: Garland Publishing. Ura, H. 2000. Checking Theory and Grammatical Functions in Universal Grammar. Oxford: Oxford University Press. Valian, V. 1991. Syntactic subjects in the early speech of American and Italian children. Cognition 40:21–81. Wexler, K. 1994. Optional infinitives, head movement and the economy of derivations in child grammar. In D.Lightfoot and N.Hornstein (eds.) Verb Movement. Cambridge, U.K.: Cambridge University Press, 305–50.

< previous page

page_81

next page >

< previous page

page_82

next page >

page_82

next page >

Page 82 This page intentionally left blank.

< previous page

< previous page

page_83

next page >

Page 83 5 Syntactic Constraints in a “Free Word Order” Language Mary Laughren The Warlpiri1 language of central Australia has gained considerable notoriety in the syntactic literature as a language in which the ordering of both argument and adjunct nominal and postpositional phrases is “free,” and in which there is only indirect evidence from surface syntactic structure for a verb phrase (VP) constituent (Hale, 1983; Laughren, 1989; Nash, 1986; Simpson, 1991). Furthermore, the expressions embedded within a single noun phrase (NP) in English, may be distributed as independent, although semantically related, constituents throughout a clause in Warlpiri. Thus Warlpiri appears, on the surface at least, to contrast very strongly with languages such as English in which the order of nominal and prepositional phrases relative to the verb is highly constrained; for example, the difference in the relationship between the verb and the “object” NP as opposed to the verb and the “subject” NP is directly reflected in the surface syntactic relationship between the verb and these NPs. Variations in word order in English sentences have been shown to be best accounted for as derived, subject to recognized constraints, from an underlying structure in which the subject NP c-commands the object NP.2 (See Speas, 1990, for a thorough discussion of the theory of phrase structure in a cross-linguistic perspective.) The freedom of word order for which Warlpiri, like many other Australian languages, is famous is illustrated in (1).3 Word order appears to play no part in the identification of a subject-object contrast in Warlpiri. Rather it is the case-marking on the nominal constituents that signals their grammatical function as shown in (1) and (2)

< previous page

page_83

next page >

< previous page

page_84

next page >

Page 84 in which these functions are contrasted: overtly marked Ergative case on the transitive subject NP and unmarked Nominative case on the Object NP. (1) a. Yankirri-rli maju-manu yakajirri. Emu-ERG bad-made berries:NOM b. Yakajirri maju-manu yankirri-rli. c. Yankirri-rli yakajirri maju-manu. d. Maju-manu yakajirri yankirri-rli. e. Maju-manu yankirri-rli yakajirri. f. Yakajirri yankirri-rli maju-manu. ‘Emu spoilt the berries.’ (lit. ‘Emu bad-made (the) berries’) (2) a. Yakajirri-rli maju-manu yankirri. berry-ERG bad-made emu:NOM b. Yankirri maju-manu yakajirri-rli. ‘The berries hurt Emu.’ The phenomenon of discontinuous or split nominal constituents is illustrated in (3). In (3a) both nominal phrases referring to the same entity are adjacent to each other, the modifying phrase immediately following the modified phrase.4 In (3b), on the other hand, these phrases are separated by intervening constituents. Again the coreferential function of both phrases is formally marked by identical case: ergative case on the components of the transitive subject phrase (yakajirri-rli and maju-ngku )5 and unmarked nominative case on those of the transitive object phrase (yankirri and wita). (3) a. Yakajirri-rli maju-ngku yankirri wita berry-ERG bad-ERG emu:NOM small:NOM maju-manu. bad-made ‘The bad berries hurt the little emu.’ b. Yakajirri-rli yankirri maju-manu wita berry-ERG emu:NOM bad-made small:NOM maju-ngku.6 bad-ERG ‘The bad berries hurt the little emu.’ Another feature of Warlpiri syntax that contrasts with English is that argument NPs are not obligatorily expressed—they are only optional. This aspect of the grammar, illustrated in (4), has been widely discussed

< previous page

page_84

next page >

< previous page

page_85

next page >

Page 85 (Austin and Bresnan, 1996; Hale, 1983; Jelinek, 1984; Nash, 1986; Pensalfini [to appear]; Simpson, 1991, among others). (4) Maju-manu. bad-made ‘(He/she/it) spoilt (him/her/it).’ These contrasts in the “surface” syntactic relations between the phrasal constituents of English and Warlpiri finite clauses raise a number of obvious questions to which a range of answers have been offered in the literature. The most radical approach would be to entertain the hypothesis that the language faculty of speakers of English-type languages differs from that of speakers of Warlpiri-type languages. To put it another way, are these languages so different because they are the product of different language acquisition devices? The answer to this question must surely be no, since Warlpiri children exposed to English during the critical period acquire native-speaker competence in English, and the children of English-speaking parents appropriately exposed to Warlpiri acquire that language with the same ease as the children of Warlpiri parents.7 We also know that languages derived from a common ancestor language can differ in the way that English and Warlpiri do.8 Thus these differences would seem to reflect minor variations within one or more of the linguistic modules or subcomponents of the morphosyntactic system and/or in the relationships between them. On the face of it, the surface syntactic structure of the Warlpiri finite clause would appear to be less constrained than that of the English counterpart. However, if language is constrained by the nature of human linguistic capacity (the so-called “language acquisition device”), then no language can operate outside those constraints. We can discover very little about linguistic systems by studying those areas in which a particular language appears to display random behavior, such as free word order, or optionality. It is the study of constraints on observable linguistic behavior that provides us with deeper insights into the grammar of a particular language and consequently into language in general. The aim of this chapter is to examine a number of constraints on the surface structure of Warlpiri finite clauses and show how these can be accounted for if we assume that the syntactic relationships underlying the Warlpiri clause are not fundamentally different from those underlying the finite clauses of languages such as English in which the surface word order is so much more rigid. My particular focus here is on the complex of functional categories that underlie a word-like element referred to as the “auxiliary” (AUX) whose position within the clause is anything but free and on the interactions between the AUX and the other constituents of a finite clause.

< previous page

page_85

next page >

< previous page

page_86

next page >

Page 86 In the second section, I present the basic structure of the Warlpiri AUX, concentrating on the morphemes that convey modal, temporal, and aspectual values and show how these interact with inflectional properties marked by verbal suffixes. The next section examines the constraints on the position of AUX in the clause and also on that of other clausal constituents relative to the AUX. I discuss three main proposals that have been offered to account for these data and present my own proposal. In so doing I examine in some detail the composition of complex verbs and the role of the directional enclitics in the verbal complex. Next I describe the interactions between negation and the AUX and the particular syntactic constraints that can be observed in the presence of negation. Then, the syntactic properties of the AUX categories in the Warlpiri finite clause are compared with those in the auxiliary complex in some genetically related Ngumpin-Yapa languages: Gurindji, Mudburra, and Walmajarri. This discussion focuses on the interaction of the shared negative word kula with the “auxiliary” complex in these languages, and the place of attachment of the pronominal enclitics. These conclusions are reviewed and summarized in the last section. COMPOSITION OF WARLPIRI AUX Most analyses of the Warlpiri AUX characterize it as being composed of three distinct categorical constituents whose surface form is template-like (Nash, 1986; Simpson and Withgott, 1986). These three AUX categories have been variously labeled, but there is a fairly general consensus that the leftmost category is a “complementizer” type category (C), which precedes an aspectual category first dubbed “auxiliary base” (BASE) by Hale (1967), and that the rightmost category is pronominal (PRON). The pronominal category is realized by forms that mark the person and number features of both subject and non-subject constituents. The AUX characterizes finite clauses that contrast with non-finite clauses that lack AUX (Hale, 1982; Simpson and Bresnan, 1983; Laughren, 1988). Finite clauses are of two types: verbal and non-verbal. It is in the former that the “full” AUX is found, while in non-verbal clauses the AUX is “defective,” being only overtly represented by PRON. This is also the case where the verb is inflected as “presentational present.” Clauses containing an imperative verb have an even more “defective” AUX in that it consists only of PRON minus the forms that mark the subject person features, as the imperative itself obligatorily has second person (addressee) subject reference. As we will see, there is a selective relationship between the non-PRON AUX forms and the form of verbal inflection expressed as a verbal suffix.

< previous page

page_86

next page >

< previous page

page_87

next page >

Page 87 Auxiliary Bound Pronouns The AUX in (5a) is made up of a morpheme ka belonging to the BASE category and a pronominal morpheme =rna marking the first person subject. The absence of overt number marking is interpreted as singular reference. In (5b) the auxiliary contains the plural subject enclitic =lu in addition to the first person subject enclitic, so that the reference is first person plural exclusive (of addressee). In (5c) the absence of overt person marking is interpreted as third person, while the plural subject enclitic =lu indicates plural reference. (In [5] and all following examples, AUX is bolded.) (5) a. Wangka-mi ka=rna Yurntumu-wardingki. speak-NPAST CENTr-1.s Y-habitant:NOM ‘I, a Yuendumu person, am speaking.’ b. Wangka-mi ka=rna=lu Yurntumu-wardingki-patu. speak-NPAST CENTR-1S=PL.S Y-habitant-PL:NOM ‘We Yuendumu people are speaking.’ c. Wangka-mi ka=lu Yurntumu-wardingki-patu. speak-NPAST CENTR-PL.S Y-habitant-PL:NOM ‘The Yuendumu people are speaking.’ AUX may also contain non-subject person and number morphemes signaling the appropriate characteristics of the referent of a direct object, indirect object, or “dative” adjunct (beneficiary, possessor, maleficiary, etc.). Non-subject auxiliary pronominals are featured in (6). For convenience all non-subject forms are glossed “NS.” (6) a. Wangka-mi ka=rna=ngku Yurntumu-wardingki. speak-NPAST CENTR-1.s=2.ns Y-habitant:NOM ‘I, a Yuendumu person, am speaking to you.’ b. Wangka-mi ka=rna=ngku=pala Yurntumu-wardingki. speak-NPAST CENTR-1.S=1.ns=dual Y-habitant:NOM ‘I, a Yuendumu person, am speaking to you two.’ c. Wangka-mi ka=rna=ngku=lu Yurntumu-wardingki. speak-NPAST CENTR-1.S=2NS=PL.S Y-habitant:NOM ‘We Yuendumu people are speaking to you.’ The bound pronouns are always encliticized (in strict order) to the other AUX elements if present, otherwise to the immediate pre-AUX constituent as shown in (7).9 The presence of the AUX base form ka in (5) and (6) contrasts with its absence in (7) where the act of speaking

< previous page

page_87

next page >

< previous page

page_88

next page >

Page 88 referred to is viewed as potentially taking place subsequent to the utterance10 and carries modal nuances. (7) a. Wangka-mi=li =ngalpa Yurntumu-wardingki-patu. speak-NPAST=PL.s =1INCL.PL.NS Y-habitant-PL:NOM ‘The Yuendumu people may/shall speak to us.’11 b. Yurntumu-wardingki-patu=lu=ngalpa wangka-mi. Y-habitant-PL:NOM=PL.S=1 INCL.PL.NS speak-NPAST ‘The Yuendumu people may/shall speak to us.’ In the remaining sections, I will have little further to say about the pronominal12 components of AUX, which I will simply refer to as PRON, or how they are to be represented in the syntactic structure underlying the finite clause. My focus will be on the interrelationship between the categories that express negation, modal, temporal, and aspectual values within the AUX and the syntactic relationships between AUX and the pre-AUX constituents. Relationship Between the Auxiliary Categories and Verbal Inflections In the English verbal complex, the order of verb types13 is strictly ordered: modal, perfect “have,” progressive “be,” passive “be,” lexical verb (e.g., “He will have been being interviewed.”). Tense (and subject person and/or number agreement if non-modal) is marked on the first verb in the string. The form in which each verb is realized is determined by the preceding verb: base form after modal, past participle after perfect “have,” gerundive participle after progressive “be,” past participle after passive “be.” This type of highly regulated syntactic structure is reflected also in the Warlpiri AUX and verb system in finite clauses. The BASE14 ka, illustrated in (5) and (6) is found only in conjunction with a verb bearing a nonpast (NPAST) suffix; thus there is a dependency between the BASE form and the form of the verbal inflection: ka (V)-NPAST. The BASE -lpa can be used only in conjunction with a verb bearing either the past or irrealis suffix as illustrated in (8): lpa V-PAST or V-IRREALIS.15 (8) a. Wangka-ja=lu. speak-PAST=(3)PL.S ‘They spoke.’ b. Wangka-ja=lpa=lu, speak-PAST=CENTR=(3) PL.S ‘They were speaking.’

< previous page

page_88

next page >

< previous page

page_89

next page >

Page 89 c. Wangka-yarla=lu. speak-IRR=(3) PL.S ‘They should have spoken.’ d. Wangka-yarla=lpa=lu speak-IRR=CENTR=(3) PL.S ‘They should speak.’ These two monosyllabic BASE forms, lpa and ka, fail to constitute a phonological word in Warlpiri, which minimally requires a foot containing two vowels. Furthermore, the consonant cluster in =lpa is not permitted at the “left edge” of a phonological word. Where ka hosts a pronominal enclitic, the AUX may be clause initial, thus creating a marked emphatic structure in which contrastive focus is placed on the predicate, e.g., Ka-rna ya-ni. (CENTR-1.S go-NPAST) “I am going.”16 The possible combinations of AUX base forms and verbal inflections are summarized in Table 5–1. Augmented AUX An AUX base (BASE), ka, -lpa (or null), can be augmented by a member of another category, usually taken to be the complementizer (C), but which will be provisionally labeled AUG (auxiliary “augment”). Members of this category must immediately precede the BASE; they express both temporal and modal contrasts. Unlike the monosyllabic BASE forms, AUG forms, with one exception, are disyllabic, and thus have the required prosodic characteristics of a phonological word. AUG must always immediately precede the BASE (no other form may intervene between the AUG and BASE). As shown in Table 5–2, not all combinations of AUG and BASE are permitted, nor can all verbal TABLE 5–1. AUX BASE and Compatible Verb Inflections AUX BASE VERB INFLECTION MEANING ka non-past ‘present’

ø ø ø ø ø ø

non-past past irrealis future imperative presentative

‘immediate future’ ‘past perfective’ ‘past counterfactual’ ‘future’ ‘imperative’ ‘presentative present’

lpa lpa

past irrealis

‘past imperfective’ ‘present counterfactual’

< previous page

page_89

next page >

< previous page

next page >

page_90

Page 90 TABLE 5–2. AUX Augment, AUX Base, and Compatible Verbal Inflections AUG BASE Verb Inflection Meaning kuja ø past ‘past’ Ipa past ‘past imperfective’ ka present ‘present’

kaji

ø ø ø ø ka lpa lpa

past present future irrealis present past irrealis

‘if/when/must past’ ‘if/must future’ ‘if/must future’ ‘if counterfactual’ ‘potential’ if/when past ‘imperfective’ ‘if hypothetical’

yungu ~ yi~ yinga ~ yingi

ø ø ø ø ka lpa lpa

past present irrealis future non-past past irrealis

‘past’ ‘present/future’ ‘desire counterfactual’ ‘future’ ‘present/immediate future’ ‘past continuous’ ‘desire future’

kala kala

ø/lpa ka

past present

‘customary past’ ‘potential’

kapu ~ kapi

ø present/future ‘future’ ø irrealis ‘counterfactual’ inflections combine with all AUG forms even where the BASE form is null. For example, only the “past” and “non-past” verbal inflections co-occur with the AUGs kuja and kala although these AUGs may combine with both overt BASE forms, ka and -lpa .17 The “future” AUG kapu selects the null BASE but combines with all verb inflections that combine with an overt AUG or BASE except the “past” form. The infinitival, nomic, presentational present, and imperative verb inflections, which may not combine with an overt BASE, cannot combine with an overt AUG either.18 The AUG forms kaji and yungu are clearly the least restrictive, combining with both overt BASE forms and with all verbal inflections that may combine with an overt AUG or BASE. The templatic morphological

< previous page

page_90

next page >

< previous page

page_91

next page >

Page 91 structure of the full AUX is illustrated by the examples in (9), which also provide a sample of the semantic nuances expressed by different combinations of AUG, BASE, and verb inflection. (9) a. Wati kaji=li ya-nu. man:NOM KAJI=PL.S go-PAST ‘The men must have gone.’ b. Kaji=lpa=lu wati ya-nu. KAJI=CENTR=PL.S man:NOM go-PAST ‘When/As the men were going.’ c. Kaji=lpa=lu wati ya-ntarla, KAJI=CENTR=PL.S man:NOM go-IRR kaji=ka-lu=nganpa kuyu luwa-rni. KAJI=CENTR=PL.S=1EX.PL.NS game:NOM shoot-NPAST ‘If the men were to go hunting, they would shoot us (some) game.’ d. Kala=lu wati ya-nu wirlinyi, KALA=PL.S man:NOM go-PAST hunting:nom kala=lu=nganpa kuyu luwa-rnu. KALA=PL.S=1EX.PL.NS game:NOM shoot-PAST ‘When the men used to go hunting, they used to shoot us game.’ Figure 5–1 presents a summary of some of the ways in which the AUX components have been categorized and labeled. I will return to the problem of categorization of the AUX morphemes after examining their syntactic behavior in some detail, taking into account a number of previous analyses, particularly Austin and Bresnan (1996), Brunson (1988), and Hale (1967, 1968, 1973, and 1983). I have omitted from the AUG forms listed in Table 5–2 the negative kula, which has been classified in previous studies as a “negative”

Figure 5–1. Categorization of Warlpiri AUX morphemes.

< previous page

page_91

next page >

< previous page

page_92

next page >

Page 92 COMP[lementizer] because of its distribution; it may not co-occur with another AUG or C form, and it occupies the same position in the AUX template. In the section on Negative AUX, I will revisit this morpheme that imposes more restrictive syntactic behavior on pre-AUX constituents than any of the other AUG morphemes. As illustrated by (9) and the preceding examples with overt AUX morphemes, AUX is typically realized in the “second” or so-called Wackernagel’s position (9a), or in the clause initial (9b-d) position. The underlying representation of Warlpiri AUX and its position in the clausal structure will be further analyzed in the next section. SYNTACTIC CONSTRAINTS ON THE POSITION OF AUX Any AUX that contains an overt AUG, including the monosyllabic form yi (provided yi hosts other AUX morphemes), may occupy the clause initial position. As stated previously, an AUX consisting of the BASE ka may also occupy the initial position as long as it hosts an overt pronominal.19 All other AUX forms must occupy a position that we will provisionally characterize as clause second position, as illustrated in (5) to (8). The AUX containing an overt AUG may also occupy this position.20 Thus in those cases where AUX can only be realized in second position, as in (8), the motivation for placing it there rather than in initial position would appear to be phonological. Where the AUX may occupy the initial position, some other motivation must account for its realization in second position. The examples in (10) illustrate the positioning of an AUX that contains the AUG element kala, which in conjunction with the “past” inflection on the verb, gives a usitative remote past value; this AUX may occupy either the initial or second position. (10) a. Kala=lu warru-pu-ngu yapa-patu-rlu kuyu. PAST-(3)PL.S around-kill-PAST person-PL-ERG animal:NOM b. Yapa-patu-rlu kala=lu warru-pu-ngu kuyu. person-PL-ERG PAST=(3)PL.S around-kill-PAST animal:NOM c. Kuyu kala=lu warru-pu-ngu yapa-patu-rlu. animal:NOM PAST=(3)PL.S around-kill-PAST person-PL-ERG d. Warru-pu-ngu kala=lu kuyu around-kill-PAST PAST=(3)PL.S animal:NOM yapa-patu-rlu. person-PL-ERG ‘The people used to kill animals all over.’

< previous page

page_92

next page >

< previous page

page_93

next page >

Page 93 The three constituents following the AUX kala=lu in (10a), namely the complex verb warru-pungu, and the case phrases21 (KP) yapa-patu-rlu and kuyu, can be placed in any order relative to each other. Any one of these constituents can occupy the pre-AUX position as shown in (10b-d), but no more than one. The first task, of course, is to account for why AUX is restricted to one of these two positions, given the otherwise “free” order of other clausal constituents. In so doing, one must establish how these surface positions must be represented in the underlying system of structural relationships. Does AUX always occupy the same underlying position? If so, how should this position be characterized? If not, how do we account for the constancy of its surface location? Hale (1967 and 1968) was the first to propose an underlying structure for the Warlpiri finite clause to accommodate the observations he made about the placement of AUX. He analyzed the clause (S) as composed of categories AUX -NP-VP, which accounted for the clause initial placement of AUX. To allow for one pre-AUX constituent, Hale proposed a movement rule (partially motivated by the phonological component) that placed a phrasal constituent before the AUX to serve as a host to which AUX could encliticize. He followed a similar approach in Hale (1973). In later works, Hale continued to address this question, accounting for the free word order and apparently “flat” constituent structure by delinking surface or phonetic structure from a deeper syntactic or lexical structure (his W* proposal),22 while accounting for the positioning of AUX by stipulating it in the phrase structure of the finite clause headed by V: VP—> AUX XP* V XP* (Hale, 1982 and 1983). What this approach fails to explain is (1) what motivates the pre-AUX positioning of a constituent when there is no phonological motivation for movement of a post AUX constituent to the pre-AUX position, and (2) what prevents more than one phrasal constituent moving to the pre-AUX position. Examples of relevant data are given in (9) and (10) where AUX contains an AUG morpheme. Another problem for Hale’s (1967) and (1968) proposals, to which we will return later (see The Verb in Pre-Aux Position), is how to exclude the VP, as opposed to an NP (or KP), from moving to the pre-AUX position, assuming that VP dominates both the verb and an object NP.23 This problem is solved in Hale’s (1983) representation of the Warlpiri finite clause where VP represents the clause and immediately dominates both V and the NPs. However, this result is achieved at a high theoretical cost, as pointed out by Brunson (1988). My approach to answering these questions will start by examining the nature of the constituents that may occupy the immediate pre-AUX position and to ask what they have in common and if the motivation for occupying this position is the same in all cases.

< previous page

page_93

next page >

< previous page

page_94

next page >

Page 94 Pre-AUX Position As noted both by Austin and Bresnan and by Brunson (following the documentation of Warlpiri already cited by Hale et al. and also Swartz [1991]), the pre-AUX position is a focus position as illustrated by the examples in (11): interrogative nominative case phrase (KP) in (11a), a nominative KP presenting “new” information in (11b) (for example, in reply to [11a]), an interrogative nominative case phrase (KP) in the scope of the propositional particle mayi in (11c), an interrogative verb in (11d), and a verb expressing “new” information in (11e), solicited, for example, by (11d). (11) a. Ngana-patu ka=lu wangka-mi? who-PL:NOM CENTR-(3) PL.S speak-NPAST ‘Which ones are speaking?’ b. Yurntumu-wardingki-patu ka=lu wangka-mi. Y-habitant-PL:NOM CENTR-(3) PL.S speak-NPAST ‘Yuendumu people are speaking.’ c. Ngana-patu mayi ka=lu wangka-mi. who-PL:NOM NOT:KNOW CENTR-(3) PL.S speak-NPAST ‘I don’t know which ones are speaking.’ d. Nyarrpa-jarri-mi ka=lu Yurntumu-wardingki-patu? how-INCH-NPAST CENTR-(3) PL.S Y-inhabitant-PL:NOM ‘What are the Yuendumu people doing?’ e. Wangka-mi ka=lu Yurntumu-wardingki-patu=ju. speak-NPAST CENTR-(3) PL.S Y-inhabitant-PL:NOM=TOP ‘The Yuendumu people are speaking.’ (i.e., Speaking are the Yuendumu people.) The KP in Pre-AUX Position No type of KP is excluded from the pre-AUX position. For example, it may be occupied by a KP containing coordinated KPs (12a), juxtaposed coordinated KPs (12b), a co-referential sequence of KPs (12c), a KP containing multiple NPs (12d), or a KP containing a VP headed by an infinitival (or nominalized) verb (12e). Thus the immediate pre-AUX position cannot be easily characterized in phonological terms but can be easily characterized in syntactic terms as a position that can be filled by an XP (on condition that no syntactic constraint is violated by its being in that position).

< previous page

page_94

next page >

< previous page

page_95

next page >

Page 95 (12) a. Wati-ngki manu karnta-ngku ka=lu ngarri-rni. man-ERG and woman-ERG CENTR=PL.S tell-NPAST ‘Men and/or women tell (him off).’ b. Wati-ngki, karnta-ngku ka=lu ngarri-rni. man-ERG, woman-ERG CENTR=PL.S tell-NPAST ‘Men and women tell (him off).’ c. Ngulya-ngka jinta-ngka ka=lu paka-rni. burrow-LOC one-LOC CENTR=PL.S hit-NPAST ‘They kill (them) in the one burrow.’ d. Ngulya jinta-ngka ka=lu paka-rni. burrow one-LOC CENTR=PL.S hit-NPAST ‘They kill (them) in one burrow.’ e. Nalija(-ku-ngarnti-rli) jinta-kari(-ki-ngarnti-rli) tea (-DAT-PRIOR-ERG) one-other (-DAT-PRIOR-ERG) purra-nja-ku-ngarnti-rli cook-INF-DAT-PRIOR-ERG ka=lu ngawu-ngawu nalija warru-yinti-rni […] CENTR=PL.S bad:NOM tea:NOM around-pour-NPAST ‘Before making another (billy of) tea, they pour out the (old) bitter tea…’ Similar observations led both Austin and Bresnan (1996) and Brunson (1988) to the conclusion that the pre-AUX position is a SPEC position. This accounts for why the constituent realized in this position is a phrasal constituent, and not a head. Furthermore, to account for the clause initial occurrence of this constituent, it must be in the SPEC of a phrasal category projected by a functional head, this phrase defining the clause. Since it is AUX that immediately follows the clause initial XP, it must be AUX, or at least a component of AUX, which heads the clause. Austin and Bresnan (1996) propose a representation of the Warlpiri finite clause that reflects elements of Hale’s earlier formulations (Hale, 1967, 1968, 1973, and 1983). They argue that Warlpiri AUX is to be equated with the functional category I [nflection] that heads an IP taking as its complement the category S. In their account, S dominates an unordered sequence of constituents with a “lexical” head category: verb, NPs, etc. Thus their S is equivalent to Hale’s (1983) clause structure except that AUX is outside S in Austin and Bresnan’s account, being equated with IP.24 Their IP may or may not have a SPEC[ifier]. Without a specifier, the clause is AUX-initial; with a specifier, AUX is in second position. This proposal is schematized in Figure 5–2.

< previous page

page_95

next page >

< previous page

page_96

next page >

Page 96

Figure 5–2. Main Finite Clause structure (compare Austin & Bresnan (1996:224 [12] & 225 [13]). Arguing against Hale’s (1983) weakening of the Government and Binding approach to syntactic representation, which maintained structure preservation at all levels of representation expressed in terms of a “projection principle” (Chomsky, 1981), Brunson (1988) proposes a syntactic structure in which the pre-AUX position is the SPEC of CP. This has the advantage of accounting for why any NP (not just the subject NP) can be realized as its SPEC—a result of A-bar movement—and also for the motivation for occupying that position: It is the site of focus. Austin and Bresnan (1996:228), on the other hand, explicitly reject the possibility that AUX equates with COMP. Brunson analyzes AUX as composed of multiple heads of distinct functional categories, C and I, corresponding to our AUG and BASE categories. To account for the impossibility of having a phrasal category in the SPEC of I (intervening between C and I) as opposed to CP, Brunson proposes that the maximal projection of I is I’, this node only dominating its head and a COMP, as opposed to CP (=C”) which dominates both a SPEC and a COMP. This analysis is schematized in Figure 5–3. While all the proposals reviewed so far can account for the pre-AUX placement of a KP, none can account for the pre-AUX placement of the verb since the shared assumption is that the verb represents a head category Xº and not a phrasal XP category. Therefore the verb cannot

Figure 5–3. Partial structure of Warlpiri finite clause after Brunson (1988:16).

< previous page

page_96

next page >

< previous page

page_97

next page >

Page 97 raise to the SPEC of CP. Hale’s (1967, 1968) formulations of clause structure, which include a VP dominating both a verb and its complement NP, and likewise Brunson’s VP of the same type, pose a different problem: how to exclude the VP from the SPEC of CP. This problem is avoided by Hale’s (1983) representation of clause structure as an unordered string in which the verb and subject and object KPs (or NPs) are all sisters, as in the structure dominated by S proposed by Austin and Bresnan (1996). Before considering the solutions to these problems proposed by these authors, I will revisit sentences with the verb in pre-AUX position. The Verb in Pre-AUX Position As seen already in (5) through (8), in which an inflected form of the verb wangka “speak, talk, say” is found in clause initial position immediately followed by AUX, the pre-AUX position may be occupied by the verb whether the AUX is composed of an overt BASE, an overt BASE+PRON, or just PRON. The verb may also occupy the pre-AUX position when the AUX contains an overt AUG as seen in (10d) repeated here for convenience as (13a). Although the verb warru-pu-ngu in (13a) is transitive, it may not occur in pre-AUX position along with its “object” kuyu “animal” as shown by (13b and c). (13) a. Warru-pu-ngu kala=lu kuyu around-kill-PAST past=PL.S animal:NOM yapa-patu-rlu. person-PL-ERG ‘The people used to kill animals all over.’ b. * Warru-pu-ngu kuyu kala=lu. around-kill-PAST animal:NOM past=PL.S c. *Kuyu warru-pungu kala=lu. As discussed previously, Hale’s earliest approach to the problem of the pre-AUX position was to assume that AUX would surface as clause initial, reflecting its underlying “deep” structure position, unless there was some motivation for movement of another clausal constituent to that position. It was assumed that the motivation was phonological. Hence this movement rule was part of the phonological, rather than the syntactic component. It moved a phonologically defined constituent to a morpho-phonologically defined position, where it provided a “host” for the enclitic AUX. The verb, like the other components found in the pre-AUX position, belonged to the same class of phonological objects,

< previous page

page_97

next page >

< previous page

page_98

next page >

Page 98 presumably a phonological phrase, given the complexity of the permitted pre-AUX constituents as illustrated in (12). It appeared then that the verb formed a single “phrase,” whereas the verb plus a complement NP/KP would comprise two “phrases.” Hale does not attempt to define the phonological properties of the pre-AUX constituent, nor does Laughren (1989:322), who proposed a similar account of the distribution of constituents in the pre-AUX position. The fact that any single KP, or the finite verb, may occupy the pre-AUX position in a clause containing the sort of AUX that may occupy the clause initial position indicates that the motivation for filling this position cannot be solely phonological. As seen previously, there is a clear semantic motivation since the pre-AUX position is associated with focus, and this position must therefore be defined in syntactic terms. This approach is taken by both Brunson (1988) and Austin and Bresnan (1996). For Austin and Bresnan, the exclusion of the verb with its object from the pre-AUX position follows from their characterization of the sister relationship between the V and the other clausal constituents dominated by S. There is no VP in their constituent structure representation of the Warlpiri clause.25 Brunson, on the other hand, assumes the existence of a D-Structure VP dominating both V and a complement NP. How then to exclude this VP from moving to the pre-AUX position identified as SPEC of CP? Recall that in English, the entire VP headed by an uninflected V may move to a pre-subject position, SPEC of C, while the inflected auxiliary verb is in the post-subject position, as in (i)a in note 23. Brunson explains the absence of the Warlpiri VP in pre-AUX position by arguing that V in Warlpiri projects a degenerate V’ level constituent (dominating a complement, but no specifier) as its maximal projection (the same structure she proposes for IP). Such a constituent cannot move to SPEC of C, which only allows X” maximal projections (in other words, an XP dominating both complement and specifier). Positing two types of phrases, one with a SPEC and the other without a SPEC, is not a desirable move (although this position is also adopted by Austin and Bresnan, who allow that the SPEC of IP is optional) because it introduces additional unconstrained variation into our syntactic model. How does one know/learn which categories project which type of phrase? Can language variation be represented this way? Can categories that project a SPEC in one language be considered the same as those that do not in another language? Can a language change over time in such a way that a “defective” category comes to project a full XP, or vice versa? For both Brunson and Austin and Bresnan, who assume that the Warlpiri verb is a head X° category and not an XP, the underlying

< previous page

page_98

next page >

< previous page

page_99

next page >

Page 99 position or function of the verb in the pre-AUX position must be characterized differently from that of KP constituents. Brunson analyzes the verb in pre-AUX position as the incorporation of V into C, thus a case of syntactic head-to-head movement or raising via I (Brunson 1988:56) familiar from analyses of both Romance and Germanic languages. What is not explained is what blocks a KP from filling the SPEC of CP, given that it is still “available” under this account.26 Austin and Bresnan, on the other hand, propose a phonologically motivated movement rule by which an enclitic AUX requiring a host moves down to encliticize to the first phonological word dominated by S as illustrated in Figure 5–4. This proposal fails to account for the pre-AUX placement of the verb in the absence of any obvious phonological motivation, also a problem for Hale’s earlier account. It also predicts surface structures that are not permitted, as recognized by Austin and Bresnan (1996:227). Giving this phonological rule the status of a rule of last resort, which is what Austin and Bresnan invoke to overcome these problems, does not provide the solution, since it predicts that verb-initial clauses would be found only where the only surface constituents were the verb and the AUX, as the syntactic rule of XP movement to the SPEC of the clausal phrase should be available in all the other cases. In both analyses, then, a KP in pre-AUX position is accounted for differently from a verb in pre-AUX position. In both accounts, the pre-AUX KP occupies the SPEC of the functional category that heads the clause: C in Brunson’s account, I in Austin and Bresnan’s account. Brunson accounts for both the preAUX positioning of the verb (raised to C via I) and of an NP (movement to SPEC of CP) as a syntactic operation. In contrast, Austin and Bresnan propose a syntactic account for the positioning of the KP in the pre-AUX position (movement to SPEC of IP), while they analyze the verb in pre-AUX position as the result of a phonological process.27

Figure 5–4. Auxiliary lowering rule (compare Austin & Bresnan [1996:226 (14)]).

< previous page

page_99

next page >

< previous page

page_100

next page >

Page 100 Austin and Bresnan also invoke their phonological “lowering” rule (Figure 5–4) to account for the preAUX placement of a constituent of a complex verb referred to in the Warlpirist literature as a “preverb” (PV). As I will show, Austin and Bresnan’s phonological solution fails to account for the complexity of the syntactic relationship between the components of a complex verb and the AUX, to be explored in the next section. Brunson does not address the problem of how to represent AUX-straddling in the clausal syntactic structure. AUX Straddling By AUX-straddling I mean sequences of the type PV-AUX-Verb or Verb-AUX-PV, where the verb may be morphologically “simple” or “complex.”28 The AUX cannot be straddled if it contains an overt AUG. These AUX-straddling structures are illustrated in (14). The PV yarda in pre-AUX position in (14a) contrasts with the presence of the entire complex verb, PV-verb yarda -wangkami in (15a), and verb-PV kulpamirra pina in (15b), which reverses the word order of (15a).29 In the AUX-straddling example in (14b) the complex verb turnu-mani is in pre-AUX position, while the associated PV muku “all” immediately follows the AUX. Except for (15a) in which the PV yarda and verb wangkami constitute a prosodic unit that may be characterized as a compound word (or single prosodic phrase), the PVs and verbs in (14) and (15) constitute distinct words (or phrases). (14) a. Yarda ka=lu=nyanu wangka-mi. [PV-AUX-Verb] more CENTR=3PL.S=ANAPH talk-NPAST ‘They are talking to each other again.’ b. Turnu-ma-ni ka=jana muku […] [Verb-AUX-PV] gather-CAUSE-NPAST CENTR=3PL.NS all ‘It gathers them all (up).’ (15) a. Yarda-wangka-mi ka=lu=nyanu. . [PV-Verb-AUX] more-talk-NPAST CENTR=3PL.S=ANAPH ‘They are talking to each other again.’ b. Kulpa-mi-rra pina kaji=ka=rna. [Verb-PV-AUX] return-NPAST-THITHER back POSS=CENTR-1SG.S ‘I might go back again.’ The assumption that AUX is projected as a functional category and that the immediate pre-AUX is associated with its SPEC leads to the further assumption that only XP constituents may occupy this position. Assuming also that the verbal constituents that appear in the pre-AUX

< previous page

page_100

next page >

< previous page

page_101

next page >

Page 101 position also occupy this SPEC position, then it must be the case that they too are XPs rather than X°. This includes the PV, the inflected V, and the complex constituent composed of PV(s) + inflected V. On the other hand, if it can be shown that a PV or verb immediately preceding AUX is not in its SPEC then, of course, our claim of phrasal status for the Warlpiri verb would have to be argued on different evidence. Apart from their syntactic distribution, there is other evidence that the components of a complex verb, PV and V, are somewhat independent. One such piece of evidence comes from the behavior of regressive vowel harmony. As Nash (1982) shows, the regressive rounding harmony triggered by the high rounded vowel u in the past tense inflection of certain verb classes, extends only to the edge of the verb root. Thus the domain of regressive harmony in a complex verb excludes all preverbs. In (16) the harmony domain is included in square brackets. (16) a. pirri-[kuju-rnu] b. pirri-[kiji-rni] scatter- [throw-PAST] scatter- [throw-NPAST] PV [Vroot-Vinfl] If we assume that the inflectional suffix represents the head of a functional category that I will refer to as verb-tense (Vt), then the relationship of the verb root to Vt can be characterized as one of SPEC to head. This structure that we will label VtP (tensed verb phrase) is then analogous to a KP in which the NP immediately preceding K forms a harmony domain with K. The PV (or PVs) are adjoined to the structure dominating the inflected verb, just as additional NPs can be adjoined to the structure dominating a KP.30 This assumes that PVs are also XPs that may occupy a SPEC position, or adjoin to another XP. I have labeled them PVP (preverb phrases). More than one PVP can be adjoined to an inflected VP (Figure 5–5a); similarly more than one NP can be adjoined to a KP (Figure 5–5b).

Figure 5–5a. Structure of complex inflected finite verb.

< previous page

page_101

next page >

< previous page

page_102

next page >

Page 102

Figure 5–5b. Complex KP dominating adjoined NP. However, the ability of the PVP to move out of the finite/tensed verb phrase to a pre-AUX position is not shared by the adjoined NP in the KP structure shown in Figure 5–5b. Neither is the ability of the PVP to occupy the post-verbal position, including the post-AUX position, shared by the adjoined NP. For example, the adjoined NP kurdu ‘child’ in Figure 5–5b, cannot be placed in front of AUX unless the entire KP is in pre-AUX position, nor can it be placed after the case-marked NP wita-ngku . Similar constraints hold of a verb inflected as non-finite or nomic. While Verb-INF must be case-marked, the case-marking cannot be extended to the associated PV (or PVs). Thus the non-finite structure appears to be that of a KP in which only the rightmost constituent (Verb-INF) is case-marked. Relevant data are given in (17) in which the PV is underlined. (17) a. Yarda-ya-ninja-ku again go-INF-DAT PV-V-INF-K ‘to go again’ b. *ya-ninja-ku yarda. c. *Yarda-ku ya-ninja-ku. d. *ya-ninja-ku yarda-ku This reveals a formal distinction between a PV and an NP; the former cannot be case-marked by the case on the accompanying non-finite verb, while an NP can be case-marked by the same case as the NP with which it forms a complex KP31 as schematized in Figure 5–6. Verb and Directional Enclitics Another difference between the behavior of the components of a finite as opposed to non-finite verbal complex is in the placement of a direc-

< previous page

page_102

next page >

< previous page

page_103

next page >

Page 103

Figure 5–6. Complex KP dominating adjoined KP. tional deictic enclitic from a small set: =rni “hither,” =rra “thither,” =mpa “across, past,” =yi “continuous” (Hale, 1986). A directional enclitic attaches only to a constituent of the verbal complex, so it must be considered as a verbal category, one I will represent as a functional head DIR[ectional]. In finite clauses, either the outermost PVP or the inflected verb may host the enclitic as shown in (18).32 In the single prime examples in (18) the AUX base -lpa is “straddled” by the PVP yarda and the verb, either of which may host DIR, whereas in the double prime examples, the entire verbal complex is preposed to AUX irrespective of which component hosts DIR, or their relative order. (The PV is underlined in [18].) (18) a. Yarda=rni ya-nu. a’. Yarda=rni=lpa yanu. again=DIR go-PAST a”. Yarda=rni yanu=lpa. b. Yarda-ya-nu=rnu. b’. Yarda=lpa yanu=rnu. again-go-PAST=dir b”. Yarda-yanu=rnu=lpa. c. Ya-nu=rnu yarda. c’. Yanu=rnu=lpa yarda. go-PAST-dir again c”. Yanu=rnu yarda=lpa.33 ‘He came again.’ I argue that DIR projects a phrasal category that governs the inflected verb phrase34 (VtP). Its SPEC must be filled by the VtP or a phrasal category immediately dominated by the highest VtP node. Graphic representations of the structures underlying (18a–c) are given in Fig. 5–7. AUX-straddling occurs when the PVP or VtP in the SPEC of DIRP raises to occupy the SPEC of CP, leaving a remnant DIRP in its wake. The additional assumption that must be made is that the DIR form encliticises to the phrase in SPEC of DIRP. The rarity of tokens of the c. structure in Figure 5–7 in which the remnant VtP dominates a trace of the embedded VtP is to be expected.

< previous page

page_103

next page >

< previous page

page_104

next page >

Page 104

Figure 5–7. Structure of Directional Phrase (DIRP). Unlike the examples of movement of a KP or the full verbal complex (DIRP) to SPEC of CP, which appears to be a “long distance” movement, AUX straddling clearly involves a more local movement, since the constituent following AUX must also be part of the verbal complex. The SPEC of DIRP is the launch site for movement to SPEC of CP, but DIRP must be in the SPEC of ASP for this move to be allowed. Furthermore, it is allowed only if AUX can raise into the empty head of CP. It may well be that DIRP can move to the SPEC of ASP only if AS has raised into C. The incorporation of AS into an overt C would not allow AS to C-command ASP and thus licence the presence of DIRP in SPEC of ASP.

< previous page

page_104

next page >

< previous page

page_105

next page >

Page 105 Before moving from SPEC of ASP to SPEC of CP, DIR encliticizes to the left most edge of the constituent in its SPEC (as do other heads). These syntactic relationships are schematized in Figure 5–8. While the non-straddling construction with the full verbal complex in pre-AUX position can be represented as the long distance movement of DIRP into the SPEC of CP (whether from SPEC of ASP or elsewhere needs to be determined35), the movement of the phrase in the SPEC of DIRP to SPEC of CP assumes that DIRP dominated by ASP does not constitute a barrier to movement. We can only assume that this movement is licensed because the AS category may be projected in C, and hence governs into the ASP. In non-finite clauses with a complex verb, there is a very strong preference for only the PVP to host DIR as shown in (19), not the inflected verb (19b).36 This preference occurs even though a non-finite verb may host DIR when it functions as a PV to another verb. In the absence of a PV, DIR is occasionally encliticized to an INF verb.37 (19) a. Yarda=rni ya-ninja-ku. again=DIR go-INF-K ‘to come again’ b. ?*Yarda-ya-ninja=rni-ki. again go-INF=DIR-K However, the PV (yarda) or the PV=DIR (yarda=rni) accompanying a non-finite verb may not straddle the AUX, as shown by (20a). This

Figure 5–8. AUX straddling structure.

< previous page

page_105

next page >

< previous page

page_106

next page >

Page 106 lends further weight to the claim that the PV-Verb-INF complex forms an XP that is governed by K. Although the uppermost PVP may move out of the KP to the SPEC of the Directional Phrase (DIRP), it can move no further, thus preventing AUX straddling. Movement out of the KP is not possible, although the KP itself can be found in the pre-AUX position as shown in (20c). The inability of yarda=rni to precede AUX in (20a), contrary to (20b) with a finite verb, provides a further challenge to Austin and Bresnan’s AUX lowering rule, since yarda=rni has the status of an independent phonological word (or indeed phrase) in both (20a) and (20b). (20) a. *Yarda(=rni) ka=lu ya-ninja-ku […] again (=DIR) AUX go-INF-K b. Yarda(=rni) ka=lu ya-ni. again (=DIR) AUX go-NPAST ‘They are going/coming again.’ c. Yarda(=rni) ya-ninja-ku ka=lu ngampurrpa nyina-mi. again (=DIR) go-INF-K AUX desirous be-NPAST ‘They want to go/come again.’ The close relationship between the PV and Verb-INF, reflected in their strict ordering and the fact that only the Verb-INF (the rightmost constituent) can be case-marked, parallels that between the kinpropositus NP-DAT and the NP expressing the kin term in kin nominal expressions of the type: Jakamarra-ku jaja-nyanu-rlu (Jakamarra-DAT granny-ANAPH-ERG) “Jakamarra’s granny.” This KP may occupy the pre-AUX position only if there is no variation in its internal word order and if no element, including AUX, intervenes between the dative-marked constituent and the case-marked kin NP.38 Although a dative case-marked KP may freely occur in pre-AUX position, it cannot do so if embedded within an NP within a KP. Similarly PVs may straddle AUX, but not if embedded within a nominalized verbal expression embedded within a KP. Thus the failure of the AUX straddling example (20a) must be treated in like fashion to the failure of AUX straddling in both (21b) and (22b) in which AUX splits a complex KP, where K is marked only on the final component of the complex constituent headed by K. The dative KP Jakamarra-ku in Fig. 5–9 cannot be extracted from the dominating NP. (21) a. Jakamarra-ku jaja-nyanu-rlu ka=ju paka-rni. J-DAT MM-ANAPH-ERG CENTR=1SG.NS hit-NPAST “Jakamarra’s granny hits me.’

< previous page

page_106

next page >

< previous page

page_107

next page >

Page 107

Figure 5–9. KP dominating complex NP with embedded KP. b. *Jakamarra-ku ka=ju jaja-nyanu=rlu paka-rni. J-DAT CENTR=1SG.NS MM-ANAPH-SERG hit-NPAST ≠ Jakamarra’s granny hits me. These structures are compared in Figure 5–9 and Figure 5–10. The NP immediately dominated by KP in Figure 5–5, Figure 5–9, and Figure 5–1039 cannot move out from the KP, nor can any sub-constituent of that NP. Thus the first NP in Figure 5–5, cannot be preposed to AUX, unless the KP of which it is part is also in that position. Thus the AUX straddling structure in (22b) is not possible for the same reason that (20a) and (21b) are invalid. NP is a barrier to extraction.

Figure 5–10. KP dominating complex non-finite verb.

< previous page

page_107

next page >

< previous page

page_108

next page >

Page 108 (22) a. Ngulya jinta-ngka ka=lu paka-rni. (same as [15d]) burrow one-LOC CENTR=PL.S hit-NPAST ‘They kill (them) in one burrow.’ b. *Ngulya ka=lu jinta-ngka paka-rni. burrow CENTR=PL.S one-LOC hit-NPAST As shown in (12c), repeated for convenience as (23a), a complex KP made up of adjoined KPs with the same case value and the same grammatical function may occupy the pre-AUX position. The structure underlying this constituent was shown in Figure 5–6. The relationship between the adjoined KPs appears to be much freer than between a PVP and a finite VtP, since the KPs may be discontinuous and may occupy any position within the clause as illustrated in (23b–c). Recall that AUX straddling requires part of the verbal complex in the immediate pre- and post-AUX positions, whereas discontinuous KPs are not subject to this constraint as evidenced by (23c), nor are they subject to the “empty” C condition required for AUX straddling. This is shown by (26b) in which AUX is composed of the AUG form kaji, BASE ka, and PRON -lu. (23) a. Ngulya-ngka jinta-ngka ka=lu paka-rni. burrow-LOC one-LOC CENTR=PL.S hit-NPAST ‘They kill (them) in the one burrow.’ b. Ngulya-ngka kaji=ka=lu jinta-ngka paka-rni. burrow-LOC POS=CENTR=PL.S one-LOC hit-NPAST ‘They kill (them) in one burrow.’ c. Ngulya-ngka kaji=ka=lu paka-rni jinta-ngka. burrow-LOC POS=CENTR=PL.S hit-NPAST one-LOC ‘They kill (them) in one burrow.’ Assuming that the AUX base is a functional category that projects an Aspectual Phrase (ASP) that dominates the finite verbal complex (DIRP),40 while the AUG category is the functional category C that subcategories for ASP,41 then as Brunson (1988) argued, the pre-AUX position in which an XP is in focus would be SPEC of CP. The fact that AUX-straddling is only possible in the absence of an overt C, raises a number of possibilities. One is that such clauses are ASPs and not CPs and that the part of the verb complex preceding the AUX base is a phrasal category (VtP or PVP) that has raised to the SPEC of ASP leaving a remnant phrase in COMP of ASP. However, the pre-AUX verbal constituent is still attributed a focus interpretation, which indicates that the position it occupies is SPEC of CP, rather than SPEC of ASP. Another possibility would be that the pre-AUX verbal constituent is an X° (PV or

< previous page

page_108

next page >

< previous page

page_109

next page >

Page 109 V) category that raises to the empty C. Unanswered, however, would be the question of why the SPEC of C must remain empty. Acceptance of the first solution also leaves unexplained the mechanism by which the C category fails to be projected and how a clause that is underlying an ASP differs from one that is a CP. My proposal, which allows an XP to move from SPEC of DIRP, where DIRP is situated in SPEC of ASP, to the higher SPEC of CP, accounts for the data without leaving this residue of outstanding problems to be solved. To account for the absence of any embedded KP constituent within the finite VtP (except within a PVP), I must assume that overt argument expressions are not internal to VtP. Although, as we have seen, an overt Object nominal that is not case-marked may be embedded inside a non-finite verb phrase, on condition it occupy the position immediately preceding the verb, case-marked expressions, including the nominative object (or subject), are realized only outside the VP, presumably because they can be licensed for case only by moving to a VP external position. If we allow the finite verb in Warlpiri to be represented by a phrasal category, then nothing special needs to be said about its realization in pre-AUX position taken to be SPEC of CP, which any XP in focus may occupy provided there is no subjacency violation. Similarly, assuming that preverbs are adjoined to the VP in the way that an NP may be adjoined to a KP, the preverb category must also be a phrasal category—not surprising given that most preverbs also function as nominals. Hence the PVP may also occupy a SPEC position, including the SPEC of CP, under certain conditions spelt out in this section. Representing both the inflected verb and accompanying preverbs as XPs and allowing that they, like KPs, may move to SPEC of CP account for the fact that a verb or preverb in the immediate pre-AUX position is interpreted as in focus. If, however, non-focused XPs are found in the immediate pre-AUX position, as suggested by (24), we need to consider the possibility that the pre-AUX position may be represented as the SPEC of more than one category. In (24) it is the topic “spear-thrower” that occupies the pre-AUX position, while the interrogative phrase “where” is found in post-AUX position, although it is clearly the focused constituent. (24) Pikirri=ji=npa nyarrpara-rla spearthrower:NOM=TOPIC=2SG.S where-LOC warungka-ma-nu-rnu? forget-cause-PAST-hither ‘Where did you forget the spearthrower on your way here?’ [HN:0047] I will return to a discussion of the issues raised by (24) in the section Negative Aux, having laid the groundwork in the next section where I

< previous page

page_109

next page >

< previous page

page_110

next page >

Page 110 briefly examine finite clauses in which prepositional particles occupy the clause-initial position rather than any of the phrasal categories so far seen in that position, namely KP, DIRP, VtP, or PVP. Propositional Particles in Pre-AUX Position A prepositional or evidential particle (Laughren, 1982) with both semantic and syntactic scope over the clause, may occupy a pre-AUX position as in (25). (The particle is underlined in [25] and [26].) (25) Kari [ka=lu wangka-mi]. perceptually:evident CENTR=PL.S speak-NPAST ‘(I) can see/hear (that) they are speaking.’ A survey of the distribution of propositional particles indicates that these operate within a functional projection that may take the CP as their complement, rather than occupying the HEAD or SPEC of CP, since they may precede the C (AUG) as in (26a and b) and also a constituent occupying the SPEC of CP as in (26c and d). (26) a. Kula-nganta [kapu=npa=ju contrary-to-expectation FUT=2SG.S=1SG.NS yu-ngkarla]. give-IRREALIS ‘I thought [you would have given (it) to me (but you didn’t)].’ b. Kula-nganta [kaji=npa nyuntu pantu-rnu]. contrary-to-expectation POSS=2SG.S 2SG spear-PAST ‘I thought (wrongly) [that you must have speared it].’ c. Kari [wiyarrpa-rlu kala=ka=npa=nyanu perceptually evident poor_thing-ERG POT=CENTRd=2SG.S-ANAPH ngarrpangarrpa-ma-ni]. lie-CAUS-NPAST ‘I can see that [you are liable to be telling lies].’ d. Kari-nganta [miyi-wangu ka=rna=lu=jana fact food-WITHOUT CENTR=1.S=PL.S=PL.NS yarnunjuku nyina]. hungry sit:NPAST “Isn’t it obvious that [we are waiting for them (here) hungry without any food].” e. Kari-nganta=rna kuyu-jarra yampi-ja-rni. fact =1.S meat-two leave-PAST=DIR ‘The fact is I left two animals (I speared) (back there) and came here.’

< previous page

page_110

next page >

< previous page

page_111

next page >

Page 111 The positioning of the propositional/evidential particles in (26) shows that there is more than one preAUX constituent. The higher constituent is projected by the functional category we will symbolize as F,42 which selects the CP as its complement, as shown in Figure 5–11 and which licenses the FOCUS interpretation on the constituent in the SPEC of CP. While the SPEC of CP is associated with focus, the SPEC of the higher constituent FP is associated with a topic function. Thus in (24) the initial KP pikirri, to which the topic marker =ji is encliticized, is the already established topic of conversation. It is projected in the SPEC of non-overt F, while the interrogative nyarrpara-rla “where-LOC” is projected in the SPEC of CP, c-commanded by F.AUX (only instantiated by PRON in this example) raises to the highest functional category position, to license both TOPIC and FOCUS functions. However, unlike the situation where an overt category is projected in C, to which the overt AS form must incorporate, (AS having already incorporated PRON), a particle such as kari does not require that the AUX categories incorporate into it, for example (26c), although it may host AUX morphemes as in (26e). The AUX may raise into the unfilled F, as in (24), or it may incorporate into the filled F constituent as in (26b and e), thus licensing a TOPIC as well as a FOCUS function. Thus in (26b) it is the post-AUX constituent nyuntu “you” that is in focus, since it occupies the SPEC of CP while the AUX has incorporated

Figure 5–11. Locus of Topic and Focus functions.

< previous page

page_111

next page >

< previous page

page_112

next page >

Page 112 into the higher constituent in F, forming a complex of propositional particle and AUX. There is one morpheme, however, that must occupy the position F, which C-commands SPEC of CP, although, like AUG morphemes, it obligatorily hosts AUX (BASE (=AS) and PRON) and is incompatible with an AUG. This morpheme is the negative kula . Because it is in complementary distribution with the morphemes pre-theoretically classified as AUG, which project in C, this morpheme has been generally classified as a “negative complementizer” in the Warlpirist literature. It has a number of features, however, that distinguish it from the other “complementizer” forms listed as AUG in Table 5–2. NEGATIVE AUX Finite clauses may be negated in Warlpiri by introducing the negative morpheme kula to the clause. Kula must precede the constituents in its scope. In verbal clauses kula appears to occupy the AUG or C position. Like other AUG elements listed in Table 5–2, kula may combine with any AUX base (or AS form), but it cannot combine with an AUG form (projected in C), as shown in (27). (27) a. Kaji=ka=rna ya-ni. KAJI=CENTR=1SG.S go-NPAST ‘I might go.’ b. *Kula=kaji=ka=rna ya-ni. NEG=KAJI=CENTR=1.S go-NPAST ≠’I might not go/be going.’ c. Kula=lpa=rna ya-ntarla. NEG=CENTR=1.S go-IRR ‘I can’t go.’ In clauses with kula, the distribution of pre-AUX phrases is far more restricted than it is in non-negative clauses and the pre-AUX position with kula is not a focus position. In (28a) ngaju “I” is a topic, while yani “go” is the focused constituent. In (28c), on the other hand, ngaju is in focus. Thus the focused constituent immediately follows NEG-AUX in (28a and c) while it precedes the non-NEG-AUX in (29b). While a KP may be realized in the pre-NEG-AUX position where it receives a topic interpretation as in (28a), a verb may not occupy this position, as shown by (28b). As exemplified by (29b), a verb may occupy the pre-AUX position and receive a focus interpretation provided the AUX does not contain the negative kula .

< previous page

page_112

next page >

< previous page

page_113

next page >

Page 113 (28) a. (Ngaju) kula=ka=rna ya-ni…. (Focused verb (I:NOM) NEG-CENTR-1SG.S go-NPAST ‘I’m not going/ don’t go.’ b. *Ya-ni kula=ka=rna (ngaju). c. Kula=ka=rna ngaju ya-ni. (Focused pronoun) NEG-CENTR-1SG.S I:NOM go-NPAST ‘ I’m not going/I don’t go.’ (29) a. Kaji=ka=rna ya-ni. POSS=CENTR=1SG.S go-NPAST ‘I might go.’ b. Ya-ni kaji=ka=rna. (Focused verb) ‘Go I might.’ These data indicate that kula must occupy a higher position than C, possibly equivalent to the F position (Figure 5–11), so that it must always C-command the focus position, which is the SPEC of CP. This predicts, then, that interrogative phrases may not occupy a pre-NEG-AUX position, since they must occupy the focus position C-commanded by kula . This prediction is borne out. Interrogative KPs may not occupy the pre-NEG-AUX position. Negative kula is an operator that must always occupy a syntactic position where it C-commands the other functional categories and their projections that constitute the AUX, including the quantifier/interrogative phrase, which is only licensed in SPEC of CP. Thus the relationship between the interrogative KP nyarrpara-kurra and kula in (30a) cannot be interpreted. In (30b) where kula C-commands the interrogative KP (and verb) the relationship is interpretable without an interrogative meaning as indicated by the English gloss. As seen in (28a), a non-quantifier KP may occupy the pre-NEG-AUX position thus outside the scope of negation where it may receive a topic or neutral interpretation. (30) a. *Nyarrpara-kurra kula=ka=rna ya-ni? b. Kula=ka=rna nyarrpara-kurra ya-ni. NEG-CENTR-1SG.S where-ALLAT go-NPAST ‘I’m not going anywhere.’ (≠ Where are you not going?) In contrast with (30), a quantifier phrase such as nyarrpara-kurra may occupy the pre- or post-AUX position where the AUX is headed by a non-negative C. Only the interrogative interpretation is available when the quantifier phrase precedes AUX as in (31b). In (31a) it is interpreted

< previous page

page_113

next page >

< previous page

page_114

next page >

Page 114 within the scope of the modalized assertion, but given (24) the interrogative interpretation is also available in the post-AUX position. (31) a. Kaji=ka=rna nyarrpara-kurra ya-ni. POSS=CENTR=1SG.S where-ALLAT go-NPAST ‘I might go somewhere.’ (or ‘Where might I go?’) b. Nyarrpara-kurra kaji=ka=rna ya-ni? where-ALLAT POSS=CENTR=1SG.S go-NPAST ‘Where am I likely to go to?’ (≠ ‘I might go somewhere.’) I claim that in both (31a) and (31b), the quantifier phrase occupies the same position (SPEC of CP), while AUX occupies different positions: in (31a) AUX has raised from C to F (Figure 5–11), while in (31b) it remains in C. This explains two observations we have made: first, that the quantifier phrase nyarrpara-kurra is the focused constituent in both (31a) and (31b), since in both sentences it occupies the SPEC of CP where focus is licensed; second, that the AUG kaji may be replaced by negative kula in (31a), this is not the case in (31b), since kula must be able to raise to F and hence precede the focused quantifier. Another property of kula that distinguishes it from the AUG morphemes listed in Table 5–2 is that kula may also negate a proposition expressed by a finite nominal clause, as illustrated in (32a). PRON is the only AUX category (apart from kula ) found in nominal clauses.43 It would seem then that, unlike the AUG and BASE components of AUX, which are dependent on the inflectional properties of the finite verb, kula is more independent of these “verbal” categories. Note also that just as the verbal predicate in a finite clause may not be pre-posed to kula (28b), the nominal predicate is also excluded from this position (32b).44 (32) a. Kula nyanungu. NEG him:NOM ‘It’s not him/her/it.’ b. *Nyanungu kula. ≠ ‘That’s not him.’ I conclude then, that kula occupies a higher position than C and that it governs the CP. It functions as a clausal operator and hence must not be C-commanded by an element that must be within its scope, such as a verbal or nominal predicate, a quantifier phrase, or the other functional categories that constitute AUX. The phrasal categories that may occupy its SPEC position are outside the range of the operator. To account for

< previous page

page_114

next page >

< previous page

page_115

next page >

Page 115 the fact that kula can only be used where C is unfilled by an AUG, I propose that kula is first projected in C and then obligatorily raised to F to satisfy its scope requirements. At this point let us return to a consideration of (24), repeated here for convenience as (33), in which the topic of conversation, pikirri “spearthrower” is in the pre-AUX position, while the interrogative KP nyarrpara-rla “where-at” is in focus in the immediate post-AUX position. In (33) the “operator” position F (which kula occupies in negative finite clauses, for example [28a] and [30b]) is not filled (nor are C or AS). PRON raises to the highest AUX position, which is F, via the AS and C positions. The topic KP pikirri=ji is in the SPEC of FP, as is ngaju “I/me” in (28), while the focused interrogative KP is in the SPEC of CP. (33) Pikirri=ji=npa nyarrpara-rla spearthrower:NOM=TOPIC=2SG.S where-LOC warungka-ma-nu-rnu? forget-cause-PAST-HITHER ‘Where did you forget the spearthrower on your way here?’ [HN:0047] Negation and Imperative The Warlpiri imperative verb form does not combine with an overt C or AS, or with subject personmarking forms in PRON. The PRON components of AUX, which are expressed in the sentences in (34), are the subject number enclitics (=lu “plural”) and the non-subject bound pronoun (=jana ‘3.PL’). As in nominal finite clauses, the “defective” AUX consisting only of PRON elements is hosted by the clause initial phrase, which may be the imperative verb (34a), but not necessarily so (34b). The imperative verb does not combine with the negative operator kula (34c). To express a negative command, the main verb is nominalized inside a negative (or “privative”) expression embedded in a matrix clause with an imperative verb (34d-e).45 The commands in (34) are addressed to more than one person. (34) a. Paka-ka=lu=jana maliki-patu. hit-IMP=PL.S=3PL.NS dog-pl:NOM ‘Hit the dogs’ b. Maliki-patu=lu=jana paka-ka. dog-pl:NOM=PL.S=PL.NS hit-IMP ‘Hit the dogs’ c. *Kula=lu=jana paka-ka! ≠ ‘Don’t hit them!’

< previous page

page_115

next page >

< previous page

page_116

next page >

Page 116 d. Paka-rninja-wangu-rlu=lu=jana yampi-ya! hit-INF-PRIV-ERG=PL.S=PL.NS leave-IMP ‘Don’t hit them!’ (lit. ‘Hitting-without them leave’) e. Yampi-ya=lu=jana paka-rninja-wangu-rlu! leave-IMP=PL.S=PL.NS hit-INF-PRIV-ERG ‘Don’t hit them!’ (lit. ‘Leave them hitting-without’) The incompatibility of the imperative (IMP) and negative kula recalls that of AUG and kula. If we assume that there is a non-overt IMP operator that licenses the realization of IMP on the verb, which is projected in C, then its incompatibility with kula derives from the same source as the *kula -AUG incompatibility; that is, the negative operator in C prevents the projection of another head to that position, whether AUG or the covert imperative operator.46 I am assuming that the PRON elements raise into AS and then incorporate into C in the standard way. In imperative clauses, the pre-AUX position retains the focus function and may be occupied by the imperative verb (e.g., [34a] and [34e]). Unlike kula, IMP does not raise to F. These syntactic constraints associated with the kula “negative” are syntactic rather than purely semantic. This is shown by the fact that the most commonly used forms of negation in contemporary Warlpiri derive from English “no” (> nuu) and “not” (> nati). Unlike kula, these forms occupy the initial AUX position and may combine with an overt AUG (although not with all AUG forms) as in (35a). They may also combine with the imperative verb form to create a negative imperative as in (35b). (35) a. Nuu=kala=lu=jana maliki paka-rnu. NEG=PAST=PL.S=PL.O dog:NOM hit-PAST ‘They used not kill (=hit) dogs.’ b. Nati=li=jana maliki-patu paka-ka! NEG=PL.S=PL.O dog-PL:NOM hit-IMP ‘Don’t hit the dogs.’ What nati and nuu have in common with kula, apart from their semantic similarity, is that they must Ccommand the CP and are always immediately preposed to AUX. Where they differ from kula is that they allow the head position of the CP they govern to be filled. In other words nati and nuu are projected directly to F, whereas kula is first projected to C and then raised to F. Where nuu or nati heads AUX, quantifier KPs may only occupy the post-AUX position, confirming these negative morphemes in F function. As expected, the verb may not precede nuu or nati. To summarize then, the Warlpiri AUX reflects an underlying structure of three functional categories: C, AS, and PRON.47 For reasons we have

< previous page

page_116

next page >

< previous page

page_117

next page >

Page 117 not pursued here, each head must incorporate into the one that C-commands its phrasal projection, or may raise into an “empty” head. The CP is governed by an additional functional category F, to which the introduced negatives nuu and nati must project, along with prepositional particles with clausal scope. The negative kula must always raise from C to F to C-command the FOCUS position in SPEC of CP and the predicate it negates. The imperative operator IMP is also projected as a non-overt C, which blocks its combination with kula but not with the “borrowed” negatives, as IMP does not raise to F. Where F is “empty,” AUX may raise into that position, thus licensing the topic function of the constituent in SPEC of FP, while permitting the FOCUS function of the constituent in the SPEC of CP. In the next section I briefly compare the behavior of Warlpiri kula with its cognate in two relatively closely related Ngumpin languages, Gurindji, and Mudburra,48 drawing on work by Patrick McConvell (especially McConvell, 1996, and personal communications). This study provides some additional justification for the approach to the grammar of Warlpiri AUX presented so far. A COMPARATIVE OVERVIEW In this section I argue that the cross-linguistic differences that exist in the syntactic behavior of the negative kula in relation to other AUX categories and the finite verb provide further evidence in support of the approach taken to the representation of the syntactic constraints on the ordering of these categories in Warlpiri. In each of the languages under investigation, cognate PRON forms encliticize to functional categories. This complex typically occupies the clause initial or clause second position, but a very limited number of other possibilities also exist. The features that I focus on in this section are: (1) the interaction between AUX and IMP, (2) the interaction between negative kula and IMP, (3) the interaction between negative kula and components of AUX, and (4) the nature of the categories to which PRON encliticizes. The Imperative and AUX Warlpiri and Mudburra are distinguished from the other Ngumpin-Yapa languages in not requiring PRON forms to obligatorily encliticize to an imperative verb, although this option is available in Mudburra. In all Ngumpin-Yapa languages the only AUX elements that co-occur with an imperative verb are the PRON elements minus the subject person forms, as illustrated for Warlpiri in previous sections. In all the Ngumpin-Yapa languages in which clausal negation is marked by kula —which

< previous page

page_117

next page >

< previous page

page_118

next page >

Page 118 includes Gurindji and Mudburra in addition to Warlpiri—this negative morpheme may not coexist with the imperative. I accounted for this incompatibility in the Warlpiri case by proposing that kula is projected in C, which is also the position in which the non-overt IMP operator is projected. At first glance a similar explanation can be extended to the other Ngumpin-Yapa languages with negative kula . For example, the obligatory encliticization of PRON to the imperative (and also hortative) verb in Gurindji can be explained if we assume that the imperative verb is required to move to the SPEC of C associated with the imperative operator to check the features associated with IMP. In Mudburra, unlike Gurindji, this “movement” is optional (McConvell, 1996). In Warlpiri, the verb only moves at LF to check this feature. However, if the imperative verb occupies the SPEC of CP, then one would expect it to occupy the clause initial position. This appears, however, not to be the case in Gurindji, as shown in (36a) in which the “object” KP karnti “wood:NOM” is in the preverbal position, although the alternative order is also allowed (36b). (36) Gurindji a. Karnti jayi-ngka=yi. (McConvell 1996:307 [15]) wood:NOM give-IMP=1SG.O ‘Give me a stick.’ b. Jayi-ngka=yi karnti. ‘Give me a stick.’ This can be accommodated within the framework proposed so far to account for the syntax of functional categories that underlie the Warlpiri AUX. The “object” KP karnti in (36a) may occupy the SPEC of FP. We would predict then that it could not receive a focus interpretation in that position but that it is the imperative verb that is focused.49 We would also predict that the imperative verb may be clause initial (as in [36b]), as the focus position (SPEC of CP) is the typical clause initial position for all construction types except where there is an overt form projected as F, as exemplified by Warlpiri kula and propositional particles with clausal scope. Placement of Kula in Relation to AUX The Gurindji AUX complex may consist of an element such as ngu=, which obligatorily hosts the enclitic pronouns (PRON) if present. A modal element =nga, typically expressing doubt, may follow the PRON elements, as in (37).50 There appears to be no category in Gurindji that corresponds exactly to the Warlpiri AS (or BASE); the aspectual values

< previous page

page_118

next page >

< previous page

page_119

next page >

Page 119 expressed by Warlpiri ka and lpa are expressed by verbal inflections in Gurindji. Unlike Warlpiri, in which C (other than kula ) and AS may only be filled in a clause with an overt finite verb, Gurindji and some other Ngumpin languages permit overt non-pronominal AUX morphemes in clauses with non-verbal predicates as in (37a). While Gurindji ngu like Warlpiri ka and lpa obligatorily host PRON, ngu is typically used in clause initial position (37a and b), whereas clause initial ka or lpa is highly marked in Warlpiri. Gurindji ngu may also be preceded by a KP or a verb (37c). (37) Gurindji a. Ngu=rna=nga tampang. NGU=1SG.S=DOUBT dead ‘I might be dead.’ b. Ngu=rna=nga ya-ni. NGU=1SG.S=DOUBT go-PAST ‘I might have gone.’ c. Ya-ni ngu-rna-nga. go-PAST NGU=1SG.S=DOUBT ‘I might have gone.’ McConvell (1996:307) shows that ngu, like Warlpiri AUG and BASE forms, cannot coexist with the imperative verb. This is illustrated by (38). (38) Gurindji *Karnti ngu=yi jayi-ngka. (McConvell, 1996:307 [15]) wood:NOM BASE=1SG.O give-IMP ≠ ‘Give me a stick.’ This result would be predicted on the assumption that ngu= is projected as C and that in imperative clauses a modal operator is projected in the C position, as proposed for Warlpiri, thus excluding ngu- (or any like category) from coexistence with IMP.51 The incompatibility of kula and ngu, and also of kula and IMP, would also suggest that these forms are projected in the same functional category. Unlike Warlpiri kula, however, which is always the first element in the AUX complex, Gurindji kula may be found in a number of different positions relative to PRON. While kula is incompatible with the imperative verb in all Ngumpin-Yapa languages in which it operates as the clausal negator, languages such as Gurindji, unlike Warlpiri, do permit kula in interrogative clauses with a quantifier phrase, as shown in (39a). Furthermore, a member of the category labeled as COMP by McConvell (1996) may also combine with kula where the COMP precedes kula and hosts PRON, as exemplified by nyamu in (39b).

< previous page

page_119

next page >

< previous page

page_120

next page >

Page 120 (39) Gurindji a. Ngana=lu kula ya-ni? (McConvell, 1996:3115 [30b]) who:NOM=PL.S NEG go-PAST ‘Who (they) did not go?’ b. Nyamu=lu kula ya-ni… (McConvell, 1996:3115 [30a]) COMP=PL.S NEG go-PAST ‘Those who have not gone…’ (lit. ‘that they not went’) On the other hand, ngu may not replace kula in the examples in (39). If present, ngu obligatorily hosts PRON (see [41b] and associated discussion). On the other hand, kula may not host PRON in (39). McConvell accounts for the constructions in (39) by assuming that kula remains in I while PRON encliticizes to ngana in the SPEC of CP in (39a) where C is empty, but to the COMP nyamu in (39b). In each case, PRON is dominated by C. If kula is projected in a lower functional category than C, then what is this category, and how do we explain the incompatibility between kula and IMP in Gurindji, as seen previously in Warlpiri? If we assume that Gurindji kula and the dubitative monosyllabic =nga (not found in Warlpiri) belong to a functional category that is lower than C, which is associated with M[ood], then it would seem only natural that IMP is also projected in this position, but must raise to C.52 Recall that Warlpiri also has a functional category we have represented as lower than C (i.e., AS). This category is not found in Gurindji. Unlike Gurindji M, however, Warlpiri AS must always raise to C or incorporate in C and must always host PRON.53 Like Warlpiri, Gurindji has also acquired a negative morpheme with clausal scope from a variety of Aboriginal English; it is numu (<no more) . This morpheme, like the Warlpiri nati and nuu, is projected in a higher node than both IMP and kula and hence is compatible with the imperative projected in C, and thus has the imperative verb and associated predicate in its scope. However, unlike Warlpiri nuu or nati, which must host AUX when used with the IMP, Gurindji numu does not host PRON; the imperative verb does, as (40) shows. (40) Gurindji a. Numu ya-nta=lu. NEG go-IMP=PL.S ‘Don’t go!’ b. *Numu=lu ya-nta. In Gurindji, then, kula is projected in a position lower than C (equivalent to AS in the structure proposed to account for Warlpiri AUX in

< previous page

page_120

next page >

< previous page

page_121

next page >

Page 121 Figure 5–11). It must raise into an empty head such as C or even F where it may host PRON, but it cannot incorporate into another head. Just as IMP blocks other functional AUX categories (other than PRON) in Warlpiri, the same occurs in Gurindji and the other related languages. Negative kula may not be in the scope of IMP or assertive ngu, but in Gurindji (unlike Warlpiri) it may be in the scope of a COMP such as nyamu and the covert form that licenses a quantifier KP as shown by (39a and b). That Gurindji kula may raise through “empty” C to F is shown by constructions that parallel the Warlpiri example in (30b) in which the focused quantifier KP is in the scope of kula . Gurindji kula must also Ccommand the predicate in its scope, as it does in Warlpiri, so that it cannot be preceded by a finite verb (cf. [30a]). What then is the position to which ngu is projected? This morpheme does not surface in the presence of an imperative verb (38) or the negative kula. Recall that the Warlpiri aspectual forms ka and lpa do coexist with kula, which suggests that ngu= is not projected to the same syntactic position as the Warlpiri AS. Further evidence about where ngu is projected is provided by a development in “young people’s” Gurindji, considered incorrect by older speakers (60+), which is that ngu= is included in the non-negative equivalent of (39b) and that ngu= and not nyamu (or its equivalents) hosts the enclitic PRON (McConvell, personal communication). This is illustrated by (41b). The older speakers’ variant is given as (41a). (41) Gurindji a. Nyamu=lu ya-ni. (Older speakers) COMP=PL.S go-PAST ‘Those who went…’ b. Nyamu ngu=lu ya-ni. (Younger speakers) F COMP=PL.S go-PAST ‘Those who have gone…’ Ngu differs from kula, however; while ngu in (41b) must host PRON, kula in (39) may not. This difference between kula and ngu is expected if kula is projected in a lower functional head than ngu and if kula cannot incorporate into another overt head. Also indicated is that in Gurindji, PRON must project in C, and that the presence of kula in the lower functional category does not block this, any more than does the presence of the dubitative nga, which always follows PRON. The difference between (41a) and (41b) can be accounted for if we assume that for the younger speakers, nyamu, like the negative numu, is projected directly to the F position, which C-commands CP. PRON incorporates into C where it is hosted by ngu and no further incorporation

< previous page

page_121

next page >

< previous page

page_122

next page >

Page 122 occurs.54 For the older speakers, nyamu is projected in C, thus blocking ngu, and hence hosts PRON. CONCLUSION I have tried to show in this chapter that it is profitable to pursue a close investigation of observable constraints on the syntactic structure of a language such as Warlpiri from the viewpoint of theories developed to explain the underlying phrase structure of languages such as English, where the match between the hypothesized underlying structure and the observable surface strings is more readily apparent than in Warlpiri. I have argued that AUX is an instantiation of a number of functional categories, two of which, C and AS, are “verbal,”55 and two of which are merely properties of a finite clause, the clausal operator that projects in F (including negative kula ) and the pronominal enclitics (PRON). The composition of AUX can be accounted for as the surface reflex of the successive raising and/or incorporation of these functional heads. Minor differences between related languages with cognate forms can also be accounted for in these syntactic terms as demonstrated in the preceding section. The distribution of constituents in the pre-AUX position (as well as the clause initial realization of AUX) can be explained straightforwardly assuming that the “highest” functional category in the AUX complex is either C or F. It is F that governs into the CP and that “licenses” the focus interpretation on the constituent in the SPEC of CP. The phrase in the SPEC of FP may be interpreted as a topic in the discourse structure, or as a relatively neutral element. Crucially, this constituent is “outside” the scope of the operators such as negative kula, which are immediately dominated by F. The ability of the verbal complex, or part of it, to occupy the pre-AUX position is accounted for in a straightforward way by assuming that the inflected verb and its PV constituents are phrasal categories. The inability of the object NP (or KP) to occupy the pre-AUX position along with the verb results from the “movement” of the NP out of the VP to a position where it can be assigned case.56 Finally, the additional syntactic constraints observed when the NEG element kula is present, as well as its place in the AUX complex, have been accounted for. In Warlpiri, kula is an operator that must Ccommand all constituents in its scope; hence it must raise to the highest functional category from C, at the same time excluding the instantiation of other forms in C, including the null “imperative” and “interrogative” operators. We have briefly sketched an outline of how we might extend this analysis of Warlpiri AUX to other languages of the Ngumpin-Yapa

< previous page

page_122

next page >

< previous page

page_123

next page >

Page 123 group where cognate forms display some interesting variation in their syntactic behavior. This variation can be captured in the syntactic framework constructed to account for constraints in both the formation of AUX and its relationships with other categories within the Warlpiri finite clause. NOTES This chapter develops some of the ideas presented in Laughren (2000). In writing this chapter I have benefited from discussion with Patrick McConvell, who has generously shared with me his knowledge of Ngumpin languages, and also with Ken Hale and David Nash, who identified important errors of fact in an earlier draft, and with Julie Legate, who also commented on an earlier draft. I am also grateful to an anonymous reviewer for some very insightful observations. None of these is in any way responsible for remaining errors of fact or analysis presented herein. Research presented in this paper was partially supported by Australian Research Council grant No. 00/ARCS213, while the writing was enabled by the University of Queensland Special Studies Program 2000. 1. The traditional Warlpiri-speaking area is situated in the Tanami Desert area of the Northern Territory of Australia. Warlpiri belongs to a group of Pama-Nyungan languages known as the Ngumpin-Yapa group (Laughren and McConvell, 1996; O’Grady et al., 1966). Warlpiri occupies the southeastern part of the Ngumpin-Yapa area. The Warlpiri data presented herein is drawn from a variety of sources including Ken Hale’s Warlpiri fieldnotes (principally from Hale’s 1959–60 and 1966–67 field trips), which are indicated by “HN” and a page number, my own fieldnotes, and Warlpiri written and oral texts were collected between 1975 and 1999. 2. Whether what have traditionally been called “noun phrases” (NP) are projections of a functional category such as “determiner” forming “determiner phrases” (DP) or “case phrases” (KP) is not relevant to the argument being made here. The essential point is that the referring expressions that relate to a predicate’s arguments constitute a syntactic phrase. 3. The following abbreviations are used in this chapter: 1 = first person, 2 = second person, 3 = third person, ALLAT = allative case, AS = aspect, ASP = aspect phrase, AUG = auxiliary augment, AUX = auxiliary, CENTR = central coincidence, C = complementizer, COMP = complement, DAT = dative case, DESID = desiderative, DIR = directional, DU = dual, ERG = ergative case, PUT = future, IMPF = imperfective aspect, INCL = inclusive, INF = infinitive, IRR = irrealis, K = case, KP = case phrase, LOC = locative case, NEC = negative, NOM = nominative case, NPAST = non-past, NS = nonsubject, PL = plural, POSS = possibility, POT = potential, PV = preverb, PVP = preverb phrase, S = subject, SG = singular, and SPEC = specifier. 4. While this tends to be the preferred ordering in Warlpiri, the reverse order in which the modifying phrase precedes the modified one is also attested. 5. The ergative case allomorphs are phonologically determined. 6. While the words in (3a) form three phonological phrases that can be shown to correspond to three phrasal constituents (i.e., yakajirri-rli maju-ngku, yankirri wita, and maju-manu), the words in (3b) each constitute a phonological and syntactic phrase. Both post-verbal words, wita and maju-ngku, are typically separated from the verb and from each other with a pause and a “new

< previous page

page_123

next page >

< previous page

page_124

next page >

Page 124 phrase” intonation, partially resembling the English “afterthought” intonational pattern. 7. In the Australian context many Aboriginal children have been brought up in English-speaking environments—and still are—so that their first, and in most instances only, language is Australian English. During my time among Warlpiri people, I observed that a number of children of European parents who interacted with Warlpiri people from an early age, especially those who had a Warlpiri nanny and who played throughout their first years with Warlpiri-speaking children, acquired nativespeaker-like competence in Warlpiri, as well as English. 8. Within the Australian context, for example, the Arandic languages spoken to the immediate east of Warlpiri do not manifest the same degree of word order freedom or the “splitting” of nominal phrases that is found in Warlpiri, although these languages are genetically related, albeit distantly. There are many people who have grown up fully bilingual in Warlpiri and an Arandic language. 9. The phonological variation in the high vowel of the plural subject enclitic =lu/li is due to a process of progressive vowel harmony. 10. See Hale (1986) for details of the semantic distinctions between BASE forms. 11. (7a) and (7b) may also be translated as “Let the Yuendumu people speak to us.” 12. For full details of the Warlpiri system of bound pronouns see Hale (1973), Nash (1986), Simpson (1991), and Simpson and Withgott (1986). There has been an extensive debate in the literature about the nature of the relationship between the AUX pronouns and the co-referring KPs within a clause (see especially Hale, 1973 and 1983: Jelinek, 1984: McConvell, 1996; and Simpson, 1991). For comparative studies of pronominal clitics within the Ngumpin-Yapa languages, see McConvell (1980 and 1996) and Nash (1996). 13. What is common to the elements classified as “verbs” in English is that they may be inflected for tense, marking a past/non-past distinction (except for a few modal verbs such as must). 14. The term “base” follows Hale’s usage (Hale, 1967, 1968, 1973). 15. I have followed Hale’s (1986) characterization of the aspectual contrast between the overt BASE forms ka and -lpa and their absence as one of “central” versus “terminal” coincidence, a semantic distinction that Hale identifies as grammaticalized in a number of different morphological paradigms in Warlpiri. I do not mark the non-overt BASE as “terminal,” treating it as the default aspectual value. 16. However, in connected speech one finds examples of clause initial ka including examples of this monosyllabic form without enclitic pronominals. Some examples are found in Hale’s Warlpiri fieldnotes (HN) (e.g., on pages 1386, 1602, 1756), although these can be characterized as being clause internal if the preceding expressions, although separated from ka by a pause, are analyzed as occupying the clause initial position. 17. There is some dialectal variation in the selection of BASE morphemes in the presence of the AUG kala. In western and southern dialects -lpa is not used with kala while in eastern dialects, particularly in Lander River Warlpiri, there is an aspectual contrast marked by the presence or absence of -lpa in conjunction with kala and the past tense verb form. 18. Unlike the infinitive and nomic verb forms, which do not combine with any AUX component, the “presentational present” and “imperative” forms combine with PRON.

< previous page

page_124

next page >

< previous page

page_125

next page >

Page 125 19. I have also recorded examples of clauses starting with the BASE =lpa in spontaneous speech; however, these occurrences are rare and must be treated as somewhat marginal, although they warrant further attention. They typically appear after a pause or hesitation in connected speech. Some examples of clause initial -lpa in Hale’s transcriptions of fluent speech can be found in Hale’s Warlpiri fieldnotes on the following pages: HN 280, 868, 196, 883, 887, 888, 889, 890, 1107, 1151 (this features monosyllabic -lpa ). These examples by no means exhaust those found in this corpus. 20. It is always possible to extrapose topicalized phrases by moving them to the left of the left edge of the clause. Such phrases acquire a special intonational melody and are typically followed by a real or virtual pause. Furthermore, these phrases are often repeated within the clause. These cases are not relevant to our discussion here. 21. I am assuming that all case-marked phrases are a phrasal projection of the functional category case (K) in Warlpiri. The case-marked phrase, typically an NP, occupies the SPEC position within the KP. I also treat the unmarked nominative phrase as a KP. 22. W represents a “word” while * indicates any number of occurrences of the category to which it is attached, including zero. 23. This feature of Warlpiri appears to contrast with English, in which a constituent containing an uninflected lexical verb and its object (as well as all subconstituents of the VP) may be realized in a pre-subject position, usually represented as in the SPEC of CP, as illustrated in (i)a. (i) a. (They say) he hates carrots, (and) hate carrots he does, b. *Hates carrots he. c. Carrots he hates/does hate, d. *Hate he does carrots. It is important to note that the inflectional features of tense and subject-verb agreement must be overtly realized on the auxiliary verb “do” and not on the lexical verb. Unlike modern English, Warlpiri allows a fully inflected verb to precede AUX. However Warlpiri is like English in that it allows the verb’s object phrase to precede the AUX (e.g., “Carrots he hates.”) while not allowing both the inflected verb and its object to precede AUX (Laughren, 1989) (e.g., “*Hates carrots he [does].”). In English, only a phrasal category may occupy this pre-subject position, so the ungrammaticality of (i)d. would indicate that the verb does not constitute a phrasal (XP) but a head (X°) category. 24. In some respects, Austin and Bresnan’s proposed structure is akin to that proposed for Hungarian by E.Kiss (1981 and in later work) to account for the “free order” of lexical constituents, in contrast with the strict ordering of positions at the left edge of the clause associated with various discourse functions such as focus and topic, subject to constraints on operator scope (e.g., E. Kiss, 1984). Nordlinger (1998) also argues for a non-configurational structure for the core of the clause in another Australian language, Wambaya. 25. In explanatory terms this is a costly assumption given the overwhelming evidence of subject-object asymmetries in Warlpiri (Laughren, 1989, 1992; Simpson, 1991; Simpson and Bresnan, 1983; among others) as in other languages such as English where the asymmetry has clearer surface phrase structure reflexes. Furthermore, it is difficult to explain the KP constituent formed by the non-finite verb and its “object” as shown in pre-AUX position in (12e) if we assume such a “flat” underlying structure.

< previous page

page_125

next page >

< previous page

page_126

next page >

Page 126 26. Compare French Ou allez-vous? or early English Where goest thou? with Warlpiri *Nyarrpara yani ka=npa? (Where go-NPAST AUX). 27. Legate (2001 and personal communication) has suggested that movement of XPs to SPEC of CP and the raising of the inflected verb (or PV) to C could both be motivated by the need to “check” the focus feature. Assuming that this feature can only be checked once, then either the V or an XP can occupy the pre-AUX position, but not both. The consequences of this proposal certainly deserve to be examined further in the light of the data presented in this chapter. Here, I will propose an alternative solution. 28. Nash (1982) provides the most extensive description of the Warlpiri verb and classification of the forms that have been put into a super syntactic category dubbed “preverb.” See also Simpson (1991:111–120) for an extensive discussion of the relationship between PVs, the verbal stem, and the AUX. In my discussion of the syntactic behavior of preverbs in this chapter, I gloss over the finer distinctions that Nash correctly recognizes between PV types. 29. This order is extremely rare in pre-AUX position in our data base. 30. It is almost certainly the case that a null functional head must be introduced to allow the attachment of PVs to the inflected verb, because, as Nash (1982) shows, different classes of PV attach in different orders—there are inner and outer PVs. These facts can be best explained in terms of functional heads that select a complement phrase and that provide a SPEC position for a type of PV. Similarly, a null functional head could underly the semantic relationship between the component NPs of a complex KP shown in Figure 5–5b. I will not proceed any further with this line of inquiry here, since it is not essential to my argument. Nor is it a problem for my proposal; however, it deserves further research. 31. This derives not from any distinct lexical property, but from their distinct syntactic functions. 32. Although a complex verb may contain more than one PVP, only the outermost PVP can raise to SPEC of DIRP, as it occupies the SPEC position that must serve as the “escape hatch” from VtP to DIRP. Lower PVPs are thus blocked from raising because the (highest) SPEC within the maximal projection immediately dominating them is filled. 33. (18c”) is somewhat rare. 34. The verbal category can be further extended by another functional category that we will call APPL[icative], which licenses reference to another participant by a dative case-marked expression. The nature of the relationship between this participant and the predicate-argument structure of the lexical verb (including all PVs contained in DIRP) may be specified by a member of the small class of “dative-adjunct” preverbs (Nash, 1982). These most outer PVs do not host DIR. As far as I have been able to ascertain, there is no equivalent of this class of PV in the other Ngumpin-Yapa languages. The DIR category is instantiated in most of the Ngumpin-Yapa languages, although a reanalysis of the forms as “tense” or “aspect” morphemes has occurred in some languages. I have not included the “dative adjunct preverbs” in this study, but they could be included in away that is compatible with the analysis of the verbal complex and its interaction with AUX proposed here. 35. This line of inquiry will not be pursued here, but the dependencies between the verbal inflections and the AUX categories suggest that these reflect a relationship that could be best represented in structural terms, as a SPEC to HEAD relationship. 36. Like other PVs apart from Dative Adjunct PVs, non-finite verbs functioning as a PV may host DIR, e.g., [PVPya-ninja] =rni ya-ni ([PVPgo-INF] = DIR go-NPAST) “go along.”

< previous page

page_126

next page >

< previous page

page_127

next page >

Page 127 37. David Nash brought some examples from entries in the Warlpiri-English dictionary (Laughren et al., 2001) to my attention, including the following: Manu kuja-ka=rlipa=jana yirri-para yapa-kari ya-ninja-rra-kurra ngurra-kurra, yangka kujapurda-kurra wapa-nja-ya-ninja-rra-kurra, manu yangka parnka-nja-kurra warnta-warnta-kurra —paniya-jarra-rlu. “Or when we follow other people with our eyes as they go off home, like as they are walking away (from us) or running past (us).” [Definition of nya-nyi “look, see”]. The relevant infinitive form that includes the directional -rra “thither” is ya-ninja-ira-kurra (go-INF-DIRALLAT). It is a simple verb in its first occurrence but the head verb of a complex verb structure in which the PV is another infinitival verb in its second occurrence [wapa-nja-]ya-ninja-rra-kurra ([PVP walk-INF]-go-INF-DIR-ALLAT). The DIR encliticizes to the non-PV infinitive in this case. I have found no example of enclisis of DIR to an infinitival verb, where the host INF is neither in PV function nor associated with a non-INF preverb. 38. In the relatively closely related Mudburra language, “possessive” noun phrases are constituted by encliticizing an AUX base to which PRON is encliticized to the dative-marked constituent, which must precede the “possessed” constituent (McConvell, 1996:316 [31]). Such structures clearly parallel AUX straddling by Warlpiri verbs. 39. The structure in Figure 5–10 features an empty N node, which is responsible for the conversion of the verbal category DIRP to a nominal one. Where there is no adjoined PVP, the ViP (infinitive verb phrase) may raise into the SPEC of DIRP and host DIR; otherwise there is a very strong preference for the PVP immediately dominated by the highest ViP node to move into SPEC of DIRP rather than the ViP constituent. 40. Or the higher projection of the “applied” category that licenses adjunct dative expressions. 41. C is typically shown as having IP as its complement. The nature of the IP is not well understood, I being used as a cover for many different functional categories—mood, tense, aspect, person, number, and so on, as discussed by Pollock (1989) and subsequent writers. Here I am presenting a conservative analysis in which AS realizes the traditional INFL category since both ka and lpa clearly have aspectual values. However, the modal values associated with these, as well as with the AUG forms and the various interactions of these with the verbal suffixes, must be accounted for in a fuller description of the functional categories underlying the Warlpiri finite clause. 42. This may be taken to be a projection of a category associated with illocutionary force that dominates the core syntactic structure of a finite clause. 43. (32a) does not exemplify a gapped or elliptic structure. An analogous elliptic clause with a missing verb such as Kaji ngurrju, “if (it had been) good,” is not available in Warlpiri. 44. Kula cannot negate a non-finite clause, irrespective of the word class of the predicate. 45. Similarly an interrogative phrase can be associated with a negated predicate only if the predicate is nominalized and then embedded in the privative negative as in (ii). (ii) Ngana ka nyina ya-ninja-wangu? who:NOM CENTR be:NPAST go-INF-PRIV ‘Who is not going?’ (Lit. ‘Who is staying without going?’)

< previous page

page_127

next page >

< previous page

page_128

next page >

Page 128 46. IMP’s incompatibility with an overt BASE expressing “central” or “imperfective” aspect is another feature of Warlpiri IMP, which requires an explanation not attempted here. 47. The PRON component of AUX, which has not been investigated in this chapter, must be decomposed into person and number categories, and the relationship between these features and the grammatical relations of referring expressions must be represented in an adequate syntactic model. A sketch of how this might work and the relationship between the components of AUX in Ngumpin languages closely related to Warlpiri is proposed in McConvell (1996). 48. Gurindji and Mudburra are eastern Ngumpin languages spoken to the north of the Warlpiri speaking area—Gurindji to the northwest and Mudburra to the northeast. 49. Patrick McConvell (personal communication) claims that the interpretation of the verb’s object karnti in (36a) is neutral, which is also true of the pre- kula KP in Warlpiri negative clauses. 50. This “modal” category, which features dubitative -nga, is widespread in northern and eastern Ngumpin-Yapa languages but is absent from both Warlpiri and Walmajarri. It is also found in Mangala, classified as a Marrngu language (O’Grady et al., 1966), spoken to the west of the Ngumpin-Yapa-speaking area, which has an AUX constituent similar to that of the Ngumpin-Yapa languages. Neither Warlpiri nor Walmajarri allows any encliticization to PRON elements in AUX. 51. The distribution of ngu may also be compatible with its projection as I, as argued by McConvell (1996), but I will not pursue this possibility here. 52. It may well be the case that IMP in Warlpiri (and also Mudburra) is also first projected in the functional category governed by C and then obligatorily raises into C. 53. In Walmajarri, the dubitative AUX morpheme rta is preceded by the C form and is always followed by PRON (Hudson, 1978); hence it behaves like Warlpiri AS in that it must incorporate into C. In Mudburra, kula is always the AUX initial form, as in Warlpiri, and may be followed by another nonPRON form such as pa, which is the immediate host of PRON. This is illustrated in (iii). (iii) Kula=pa=rna jalkaji lap warnta-rnirra. NEG=PA=1s spearthrower pick+up get-PAST (McConvell, 1996:310 [25]) ‘I did not pick up a spearthrower.’ 54. This analysis predicts that the F negative numu could be used in conjunction with ngu as in Numu ngu=lu ya-ni “They didn’t go.” I have not been able to check out this hypothesis. 55. These are verbal in the sense that they can be used only in a finite clause with a verbal predicate and that there are constraints on the coexistence of C, AS, and verbal inflectional forms. 56. In this chapter I have not discussed the evidence that supports this assumption, but see some discussion in Laughren (1992). REFERENCES Austin, P.K., and Bresnan, J. 1996. Non-configurationality in Australian aboriginal languages. Natural Language & Linguistic Theory 14:215–68.

< previous page

page_128

next page >

< previous page

page_129

next page >

Page 129 Brunson, B. 1988. A Processing Model for Warlpiri Syntax and Implications for Linguistic Theory. M.A. Thesis, University of Toronto. (Technical report CSRI-208. Computer Science Department, University of Toronto.) Chomsky, N. 1981. Lectures on Government and Binding: The Pisa Lectures. Dordrecht: Foris. E.Kiss, K. 1981. Structural relations in Hungarian, a “Free” Word Order language. Linguistic Inquiry 12:381–416. ——. 1984. The order and scope of operators in the Hungarian sentence. In W.Abraham and S.de Mey (eds.) Groninger Arbeiten zur Germanistishen Linguistik 24. Groningen: Rijksuniversiteit Groningen Germanistisch Instituut, 82–126. Hale, K.L. 1967. Preliminary Remarks on Walbiri Grammar: I. Ms. Cambridge, MA: MIT. ——. 1968. Preliminary Remarks on Walbiri Grammar: II. Ms. Cambridge: MIT. ——. 1973. Person marking in Walbiri. In S.R.Anderson and P.Kiparsky (eds.) A Festschrift for Morris Halle. New York: Holt, Rinehart & Winston, 308–44. ——. 1982. Some essential features of Warlpiri verbal clauses. In S.Swartz (ed.) Papers in Warlpiri Grammar: In Memory of Lothar Jagst. Work Papers of SIL-AAB. Darwin: SIL-AAB, 217–313. ——. 1983. Warlpiri and the grammar of non-configurational languages. Natural Language & Linguistic Theory 1.1:5–47. ——. 1986. Notes on world view and semantic categories: Some Warlpiri examples. In P.Muysken and H.van Riemsdijk (eds.) Features and Projections. Dordrecht: Foris. Hudson, J. 1978. The Core of Walmatjari Grammar. Canberra: AIAS, 233–54. Jelinek, E. 1984. Empty categories, case, and configurationality. Natural Language & Linguistic Theory 2:39–76. Laughren, M. 1982. A preliminary description of propositional particles in Warlpiri. In S.Swartz (ed.) Papers in Warlpiri Grammar: In Memory of Lothar Jagst. Work Papers of SIL-AAB. Darwin: SIL-AAB, 129–63. ——. 1988. Lexical representation of Warlpiri verbs. In W.Wilkins (ed.) Thematic Relations, Syntax and Semantics, 21. New York: Academic Press, Inc., 215–42. ——. 1989. The configurationality parameter and Warlpiri. In L.Maracz and P.Muysken (eds.) Configurationality: The Typology of Asymmetries. Dordrecht: Foris, 319–53. ——. 1992. Secondary predication as a diagnostic of underlying structure in Pama-Nyungan languages. In I.M.Roca (ed.) Thematic Structure: Its role in Grammar. (Linguistic models series.). Berlin, New York: Foris/Walter de Gruyter, 199–246. ——. 2000. Constraints on the pre-auxiliary position in Warlpiri and the nature of the auxiliary . In J.Henderson (ed.) Proceedings of the 1999 Conference of the Australian Linguistic Society. URL at http://www.arts.uwa.edu.au/LingWWW/als99/proceedings. Laughren, M., Hale, K.L., and the Warlpiri Lexicography Group. 2001. Warlpiri-English Encyclopaedic Dictionary. Typescript. The University of Queensland. Laughren, M. and McConvell, P. 1996. The r:rl correspondence in Ngumpin-Yapa languages and the prehistory of Western Pama-Nyungan. Paper presented at the Australian Linguistic Institute Workshop “Where’s the Western Desert language come from?” Australian National University, Canberra. Legate, J.A. 2001. Warlpiri datives: A High Applicative, a Low Applicative, and an “Applicative.” Paper presented at APPL Fest, MIT, Cambridge, MA. To appear in Proceedings of Appl Fest, MIT Working Papers in Linguistics.

< previous page

page_129

next page >

< previous page

page_130

next page >

Page 130 McConvell, P. 1980. Hierarchical variation in pronominal clitic attachment in the eastern Ngumbin languages. In B.Rigsby and P.Sutton (eds.) Papers in Australian Linguistics No. 13: Contributions to Australian Linguistics, Pacific Linguistics Series A, No. 59. Canberra: ANU, 31–117. ——. 1996. The Functions of Split-Wackernagel Clitic Systems: Pronominal Clitics in the Ngumpin Languages (Pama-Nyungan family, Northern Australia). In A.L.Halpern and A.M.Zwicky (eds.) Approaching Second: Seccond Position Clitics and Related Phenomena. Stanford: CSLI Publications, 299–331. Nash, D.G. 1982. Warlpiri verb roots and preverbs. In S.Swartz (ed.) Papers in Warlpiri Grammar: In Memory of Lothar Jagst. Work Papers of SIL-AAB. Darwin: SIL-AAB, 65–191. ——. 1986. Topics in Warlpiri Grammar. Outstanding Dissertations in Linguistics. New York: Garland Publishing. ——. 1996. Pronominal clitic variation in the Yapa languages: Some historical speculations. In W.McGregor (ed.) Studies inKimberley Languages in Honour of Howard Coate. München: Lincom Europa, 117–38. Nordlinger, R. 1998. Constructive Case. Stanford: CSLI Publications. O’Grady, G.N., Voegelin, K., and Voegelin, F. 1966. Languages of the World: Indo-Pacific Fascicle Six. Anthropological Linguistics 8.2:1–197. Pensalfini, R. (to appear). Towards a typology of configurationality. Natural Language and Linguistic Theory. Pollock, J.-Y. 1989. Verb movement, universal grammar and the structure of IP. Linguistic Inquiry 20:365–424. Simpson, J.H. 1991. Warlpiri Morpho-Syntax: A lexicalist approach. Dordrecht, Boston, London: Kluwer Academic Publishers. Simpson, J.H., and Bresnan, J. 1983. Control and Obviation in Warlpiri. Natural Language and Linguistic Theory 1.1:49–64. Simpson, J.H., and Withgott, M. 1986. Pronominal clitic clusters and templates. In H.Borer (ed.) The Syntax of Pronominal Clitics, Syntax and Semantics, 19. Orlando: Academic Press, 149–74. Speas, M. 1990. Phrase Structure in Natural Language. Studies in Natural Language and Linguistic Theory. Dordrecht, Boston, London: Kluwer Academic Publishers. Swartz, S.M. 1991. Constraints on Zero Anaphora and Word Order in Warlpiri Narrative Text. Darwin: SIL-AAIB Occasional Papers No. 1.

< previous page

page_130

next page >

< previous page

page_131

next page >

Page 131 6 On the Range and Variety of Cases Assigned by Adpositions Alan R.Libert Adpositions (Ps) show a considerable amount of variation, both within and across languages, with respect to the cases they assign. To my knowledge, this variation has net been surveyed, nor has it been fully accounted for by generative theory. In this chapter I present a classification of case marking by Ps and in addition make a few brief remarks on how such case marking might be accounted for. This will not be a comprehensive account, as only a fraction of the large number of relevant languages will be discussed, but it should illustrate the great variety that occurs in this area of the grammar and provide a framework in which to place other variants, should they be found. There are two general possibilities for adpositional case assignment in a language, each of which has subpossibilities—Type I: all Ps in the language assign the same case (I include here languages where there are exceptions, i.e., where all but a few Ps assign the same case); and Type II: More than one case can be assigned by Ps. Table 6–1 lists languages that belong to the two types. Given that I allow for Type I languages to have exceptions, the line between the two types may sometimes be a matter of judgment. However, it seems better not to allow a small number of exceptions to remove a language from Type I if the great majority of its Ps assign the same case. Let us look at each broad category and the subtypes that occur.

< previous page

page_131

next page >

< previous page

page_132

Page 132 TABLE 6–1 Possibilities for Adopositional Case Assignment Type Characteristics I Ia Ib Ic Id Ie If Ig II

most/all Ps assign same case most/all Ps assign a case that is only used for this purpose most/all Ps assign accusative most/all Ps assign dative most/all Ps assign genitive most/all Ps assign nominative most/all Ps assign ergative most/all Ps assign locative more than one case assigned by Ps different Ps assign different cases the same P can assign different cases

next page > Possible Examples of Languages Punjabi, Catalan Modern English, Bulgarian Yiddish Akkadian Mansi, Kanashi Sindhi Chukchee

Latin, German, Turkish Latin, German (semantic split) Turkish (non-semantic split) TYPE I—LANGUAGES WITH ONE (MAIN) ADPOSITIONAL CASE Within the set of languages that basically have only one case marked on prepositional objects (POs), there is variation—the case used for this purpose differs. One must be careful with the names of cases, since, for example, the dative of one language may be different from the dative of another language. They may have a somewhat different range of functions, and so it may not be clear that we are dealing with the same case (in fact one might wonder whether we can ever really say we are dealing with the same case if there is any difference at all in function). In addition each case might be said to take its “meaning” in part from its paradigmatic opposition to other cases in the language, which will of course depend on which other cases exist in that language. Here I am not considering languages such as Chinese, which have no surface case marking at all; I am relying on surface case as evidence for abstract case, and such evidence is lacking in these languages, although one may still posit abstract cases in them for various reasons (e.g., to maintain universal case properties of language). I shall only assume the existence of different cases in a language when there is some surface indication of it. However, I follow the principle that if there is any paradigm or lexeme that distinguishes a particular case, one should

< previous page

page_132

next page >

< previous page

page_133

next page >

Page 133 assume that all lexemes in the language will bear that case in the same environment.1 Both nouns and pronouns in English will bear accusative case in this sense. There are at least a few languages that have a surface case used only for complements of adpositions, but in most of the languages of Type I, the adpositional case is used for some other function (s). I shall now examine the different kinds of Type I languages. Type Ia—Special/Exclusive Prepositional Case to All or Most POs Among languages with what we might call an adpositional case, used mainly or exclusively on complements of adpositions, and that is the only case assigned by prepositions,2 may be Punjabi, as indicated by Shackle (1972:38): “When nouns are followed by postpositions, they must be put into what is called the oblique case. The other case, used without postpositions, is called the direct case.” An example is given here. (1) k mr-e yc room-OBL in ‘in the room’ (ibid.) This may also be true of Hindi and Urdu. With at least some of these languages it is not clear whether postpositions or secondary cases are involved (which is often the case, v. infra for secondary cases). I believe that this situation may also exist in some Romance languages, though not described in these terms. In Catalan, as Yates (1975:129) states, “[Strong pronouns] are invariably governed by a preposition.”3 If these “strong pronouns” only occur in this environment, then they could be seen as representing a particular case of the pronominal paradigm, instead of or in addition to the strong/weak distinction of pronouns, that is: Are they only in one case when they are strong, or are they always strong when in one case? In fact, it is interesting that two works on Catalan differ in their classification of the personal pronouns: For Yates (1975) the strong pronouns include only the prepositional forms, and not subject forms such as jo “I,” bearing in mind that these two forms are usually identical; while in Gili (1967) this group of pronouns includes both types (i.e., both jo and the prepositional mi ). Presumably then for Gili the difference between the subject forms and the prepositional forms would be a case or caselike difference, while Yates might conceive of the difference in terms of strong and weak forms. The same sort of situation may hold in Spanish and Portuguese, though not in French, since in French the “disjunctive” pronouns (e.g., the first person singular moi ) can be used in “strong” contexts when they are not prepositional

< previous page

page_133

next page >

< previous page

page_134

next page >

Page 134 complements, so there it is a strong/weak distinction rather than one of case. At least one Portuguese-based creole may belong to this subtype. Green (1988:448) says that “Crioulo distinguishes up to five sets of forms [of personal pronouns]: disjunctive or emphatic subject, neutral subject, post-verbal clitic, post-prepositional, and possessive.” Type Ib—Accusative (or Accusative/Dative) Assigned to All or Most POs Let us now turn to languages where the same case appears on POs and on direct objects. If we take the case of English direct objects to be the accusative (though noting that this case is also used for indirect objects) and assume that this case is assigned to all NPs with this function, although only pronouns show it, then this case appears to be superficially identical to the case of adpositional complements, as shown in (2). (2) a. I saw him. b. I stole a book from him. However, presumably this case is assigned in a different way, and/or by a different category. Perhaps examination of some phenomena will indicate that they are different cases at some level, but in the absence of compelling evidence to the contrary, I would assume that the case assigned by English Ps is accusative and is a structural case, since all NPs in the same structural position, complement of P, are assigned the same case, accusative. The idea that English prepositional objects bear structural case was mentioned in Chomsky (1981:172): “I have tentatively been adopting Kayne’s proposal that the Case system has in part been lost in English even at the level of abstract Case, so that prepositions assign objective rather than oblique case.” (Thus his earlier statement [ibid.:170] that “NP is oblique if governed by P” would not apply to Modern English4). Other languages with the accusative as the only or main prepositional case are modern Dutch, Afrikaans, Frisian, Swedish, modern Greek, and Bulgarian. One might wonder whether P and V have something in common in English that they do not have in common in Punjabi, since the two heads assign the same case in the former language but not in the latter. Another approach is to ask what properties these languages share that are not shared by languages of Type Ia, and then to see whether this could be connected with a similarity/difference in the features of the heads involved. One property is that all languages I have found with the accusative as the only adpositional case have a

< previous page

page_134

next page >

< previous page

page_135

next page >

Page 135 double object construction in which both the indirect object and the direct object, or whatever the two objects are, have surface accusative marking.5 In Bulgarian this is true to a very limited extent, with at least two verbs under certain circumstances, including pitam “to ask.” In Greek, at least as late as what Jannaris (1897) calls the “3rd Neohellenic or Modern period” (1450–1880) there was a “double accusative,” “at least partially” (ibid.:332), although, “for the most part, however, the personal or remoter accusative is now expressed by the datival genitive” (ibid.). In the other Type Ib languages that I have found, which are all Germanic, there is more freedom with double objects. With the exception of clitic pronouns, none of these languages has a morphological dative case either, but this is also true of the Type Ia languages. On the other hand, as far as I know, few, if any, of the languages of that type have a double object construction.6 Therefore, it could be argued that in the Type Ib languages, POs, direct objects, and indirect objects can be assigned the same case, and that the accusative is what is assigned by [-N] categories, using the category features, for example, in Chomsky (1981) (although the rarity of double objects in Bulgarian and modern Greek considerably weakens the case for this, at least for these languages). In fact, one could say that what accusative means in these languages is “assigned by a [-N] category,” while what the adpositional case seems to mean in Type Ia languages is “assigned by a [-N, -V] category.”7 Type Ic—Dative Assigned to All or Most POs In Yiddish all prepositions take the dative case. There is a similar situation in at least some forms of East Franconian (German): Rowley (1990:401) mentions “the widespread tendency to use all prepositions exclusively with the dative.” Following the previous discussion, in such languages and dialects we would look for properties that P has in common with verbs that assign dative (e.g., Yiddish helf’n “to help”) or for more general properties shared by Type Ic languages, and not by other types of languages. Note that we would consider only languages and dialects that have distinct dative and accusative case marking (in at least one paradigm) to be of this subtype; if they are not distinct, then the language or dialect in question would be of Type Ib. Type Id—Genitive Assigned to All or Most POs In several languages, the genitive is the case assigned by all or most adpositions. These include the Semitic languages Akkadian and Arabic, the Cushitic languages Bilin and Khamtanga, the Uralic language North-

< previous page

page_135

next page >

< previous page

page_136

next page >

Page 136 ern Saami, and the Sino-Tibetan language Ladakhi. The following is an example from Akkadian. (3) ina bīt-i(m) in house-GEN ‘in the/a house’ (Riemschneider 1977:23) From a descriptive point of view, to some extent this can be explained by the fact that in at least many of the languages of this subtype I have mentioned, all or most of the adpositions are derived from nouns (or are still really nouns). One may note the remark by Blake (1994:167) about a postposition of Tamil (which does not belong to Type I since some of its postpositions assign other cases): “It betrays its nominal origin in governing a complement in the genitive.” In fact, it may not be clear whether all the languages in question really have adpositions, or whether these elements have remained as nouns, and not evolved into adpositions. Here, as in some other areas, the line between an adposition and some other type of item may not be clear. In an article on Bilin, Palmer (1967:208) says, “Prepositions are to be defined as a subclass of noun because their forms always occur with a preceding genitival form.” One may also note the remark of Appleyard (1987:258) about Khamtanga: “Finer degrees of relations are expressed by postpositions, many of which are themselves demonstrably case forms of nouns […]; postpositions are preceded by the noun in the possessive case.” As for Akkadian and Arabic, Gray (1934/1971:74) says, “The majority of Semitic postpositions are nominal in origin.” Evidence that Ladakhi postpositions are nouns, but that they are not typical nouns, is given by Koshal (1979:80): “The post-positional items are in some sense nouns as they take case suffixes and the entire construction is related to the main noun phrase by a genitive case marker. […] However, they are in a sense special types of elements as they can be declined only for three cases dative, ablative, and genitive, and not for others.” We can then say that what P has in common with N in these languages is some nominal feature(s). Possibly the standard features given in Chomsky (1981) for Ps, [-N, -V] would not apply to Type Id languages, and for that matter adpositions that assign genitive in Type II languages; rather they should be [+N, -V].8 What is not explained on this account is why in other languages that have Ps derived from Ns, and that have a genitive case, POs are marked not with the genitive, but with some other case. Perhaps in all such languages the Ps have changed too much from their original type to act as Ns in this respect, although one would like to use some independent test for the N-hood or P-hood of these elements.

< previous page

page_136

next page >

< previous page

page_137

next page >

Page 137 At least one Type Id language, Arabic, has double accusative objects, which indicates that the presence of this construction in a language is not a sufficient condition for membership in Type Ic. Type Ie—Nominative Assigned to All or Most POs There are at least a few languages in which it seems that the nominative is the case assigned by most or all adpositions. Again, we are concerned here with languages with overt case systems, not languages where there would be no alternative to having an object with no overt case marking as an adpositional complement. The languages that may belong to this class are Udmurt and Mansi (both Uralic), Kanashi (Gyarung-Mishmi branch of Tibeto-Burman), and Romanian; however, we shall presently see some facts perhaps indicating that none of these languages is really of this type. In (4) is a possible example of an Udmurt postposition with a nominative object. (4) ukno ul-am window under-myINESS ‘under my window’ (Csúcs, 1998:293) There are several issues involving this example. An interesting feature is that the marker indicating the person of the possessor of the object denoted by the noun is attached to the postposition rather than the noun.9 What is problematic is that it is the inessive form of this affix; if there were no possessive marker, then the form that would occur here would presumably be the inessive case form of ul, namely ul-ïn. Since this word is declined, one might think that it is not a true adposition, but rather a noun (as with some postpositions of the Type Id languages). Another point is that in Udmurt only definite direct objects bear overt accusative case marking; one could thus say that the “nominative” (i.e., the form with no overt case marking) marks both subjects and some direct objects, and so should not be labeled the nominative, or at least one must be aware that it is rather different from the nominative of, for example, Latin. This means that Udmurt may not be what we might want to call a “pure” Type Ie language, that is, a language in which adpositions take objects that clearly are nominative in the traditional sense of the term, a case marking subjects and not objects. In Romanian, which has a very limited case system, the nominative is assigned to both subjects and direct objects as well as POs (while the oblique is borne by indirect objects and possessors), so it is even less like the canonical nominative than the nominative of Udmurt. One might wonder how extensive the case systems of Type Ie languages are in general. Do they use the nominative for POs because they lack

< previous page

page_137

next page >

< previous page

page_138

next page >

Page 138 cases that are more usually used for this reason? There are some languages with extensive case systems but without genitives, and one might think that some Type Ie languages are basically Type Id languages, but with the absence of an overt genitive.10 This seems to be the situation with Mansi, which does not have a genitive in its inventory of cases. Udmurt does have an extensive case system, with fifteen cases, including a genitive. Therefore we would not consider it a Type Id language without a genitive, but we have just seen that it is not a pure Type Ie language either. Kanashi’s inventory is smaller, but with six cases, including a genitive, it is larger than that of many Indo-European languages. A complication in Kanashi is that it appears to be a split ergative language. The general rule apparently is that in the present tense transitive subjects are nominative and objects are accusative/dative while in the past tense transitive subjects are ergative and direct objects nominative; intransitive subjects are always nominative. What is more, the accusative/dative affix is at least sometimes optional, and in fact less common than the nominative (=caseless) form for marking verbal objects: Sharma (1992:342) says that “in most cases of direct or indirect objects it has been noticed that the speakers of Kanashi do not care to use any object markers for them.” In fact, the genitive and ablative forms apparently are also sometimes identical to the nominative. There thus appear to be very few, if any, pure Type Ie languages. I do not know of any properties that the putative (impure) Type Ie languages mentioned share and that distinguish them from the other Type I languages, or other languages in general. Following the line of thought outlined previously, one might expect something that P and INFL (or Tense) in these languages share. Or, one might think that INFL/Tense is involved not only in clauses in such languages but also in PPs. However, although at least one language, Malagasy, does have Ps that “show tense distinctions” (Lisa Travis, personal communication), as far as I know this is not true of the Type Ie languages. If one believes that AGR (or the AGR part of INFL) is responsible for nominative case assignment to subjects, then one might expect that Ps, which have overt agreement, would assign nominative as well. This seems to hold in the Type II language Hungarian, in which some Ps bear agreement whereas others do not, and the former have nominative objects whereas the latter do not. Another possibility is to explore the possible connection between the fact that POs can be nominative and that direct objects in Romanian, Udmurt, and Kanashi are or can be nominative. A further area for investigation would be to see what the other uses of the nominative are in these languages. In particular, if the nominative is a sort of default case, then perhaps their Ps are defective in some way and lack the ability to assign any case, and so their objects receive nominative to avoid violating the Case Filter. There are of course many

< previous page

page_138

next page >

< previous page

page_139

next page >

Page 139 other languages that have nominative as the default case, but in which Ps do not assign nominative case. One could also look at the properties of Ps in Type II languages that appear to assign nominative case, to see whether they have anything in common with the Ps of Type Ie languages that causes them to assign this case, or not assign any case. Indeed this might generally be a fruitful approach, that is, to compare the Ps of Type II languages with those assigning the same cases in Type I languages (e.g., do those Ps of Type II languages that assign genitive case have something in common with the Ps of Type Ic languages?). I shall now discuss an issue that may add some other languages to this subtype, although it conceivably could also remove some languages from it (and other subtypes). Excursus on Secondary Cases and the Adposition/Case Distinction As mentioned previously, the distinction between adpositions and nouns is not entirely clear. Another difficult boundary is that between adpositions (usually postpositions) and case markers. If certain apparent case suffixes are in fact adpositions, then we would have to say that they take nominative complements. On the other hand, there are some apparent case suffixes that attach not to caseless stems, but to stems already bearing some case marking. These are called secondary case by Mel’čuk (1986). One of the examples he gives is from Lezgian, of which a partial paradigm is given in (5). (5) base: lam ‘donkey’ ergative: lam-ra genitive: lam-ra-n dative: lam-ra-z (ibid.:63) Here the genitive case marker, -n, is attached not to the nominal root, but to a stem already bearing the ergative case affix -ra; the same is true of the dative suffix -z. The relevance of such cases here is that if these “secondary cases” are in fact adpositions, they may represent other possibilities of adpositional case assignment, perhaps making us expand our account if we want to have a comprehensive description of case assignment by Ps in the world’s languages. On the other hand, if some apparent adpositions are really case suffixes, then some presumed examples of case marking by adpositions may not be such, perhaps narrowing the range of possibilities involved. Thus, when talking about the relatively rare phenomenon of Ps assigning nominative case, we would want to make sure that the putative Ps are really Ps, and not simply case markers, in which case it would not be unusual for them to take a nominative “complement.” The terminology

< previous page

page_139

next page >

< previous page

page_140

next page >

Page 140 of some authors is not always helpful in this respect, and can lead to misunderstandings. For example, Sharma (1992) states that in the Gyarung-Mishmi language Chhitkuli postpositions take caseless forms (which are phonetically equivalent to the nominative/accusative). However, some of these “postpositions” appear to be case suffixes, at least as they are transcribed by Sharma, while he says (pp. 227–28) that “most of the case markers belong to the class of postpositions.” Of course orthographic or transcriptional facts may be superficial and misleading, and for that matter even phonological and morphological facts may not be good indicators of an element’s status as affix or separate element, especially with the current trend toward positing functional elements, both affixal and non-affixal, as separate syntactic categories; of course one’s definition of adposition and/or criteria for membership in this category are crucial.11 Assuming one can agree on criteria for distinguishing adpositions from case markers, they would then have to be applied to many apparent instances of case markers or adpositions. We should therefore be cautious before assuming that a language is of Type Ie without further investigation. One ought to bear in mind with respect to Type Ie, and indeed every other type, that some apparent members of the class may not be such, although it might be a lengthy process to determine this. I shall not spend much time on the description of patterns of case assignment by secondary cases, beyond making some general remarks, although as noted, if these can be shown to be adpositions, then they would have to be considered in a full account of adpositional case assignment. As with adpositions we could divide relevant languages into two basic groups, those in which all secondary cases are attached to a stem with the same case (Type I languages with respect to secondary cases), and those where secondary cases are attached to stems with different primary cases (Type II languages in this respect).12 The situation will be complicated, and at least one language, Dargwa, has tertiary cases (to use Mel’čuk’s 1986 term), the markers of which are attached to stems bearing a secondary case marker (and a primary case marker). Tokharian A, in which the oblique case is the one “assigned” by the secondary cases, would be of the first type. In Estonian the genitive is the main case required by secondary cases, though not the only one, as the partitive is also required under certain circumstances; it would apparently be a member of the second type. In the instances just described, there is no obvious semantic reason why particular primary cases are required. However, there are also situations in which one could say a case affix can attach to a choice of other case affixes to give a difference in meaning. Mel’čuk (1986) sees these as involving what he calls markers of compound cases, markers that

< previous page

page_140

next page >

< previous page

page_141

next page >

Page 141 can be decomposed into several meaningful elements, and in his terms Lezgian also has compound cases, in addition to its secondary cases. This is illustrated in (6), from Mel’čuk (ibid.:64): (6) base: vax ‘sister’ ergative: vax-a postessive: vax-a-qh ‘[being] behind the sister’ postlative: vax-a-qh -di ‘[moving] to behind the sister’ subessive: vax-a-k ‘[being] under the sister’ sublative: vax-a-k-di ‘[moving] to under the sister’ Here we can see that the postessive and the subessive are secondary cases, because they are attached to an ergative form rather than the bare form, and they are simple cases, since they can not be broken into smaller meaningful elements. According to Mel’čuk the postlative and the sublative are compound cases (as well as being secondary cases), as they can be seen to consist of two elements, a lative part and a locational part. Looking at it another way, we could say that the lative affix attaches to a postessive stem to indicate one meaning, and to a subessive stem to give another, just as German in means “in” and “into” with dative and accusative complements, respectively. Type If—Ergative Assigned to All or Most POs In Sindhi POs “are usually in the oblique case” (Yegorova, 1971:120). This case has one other use, “to express the agent in the ergative construction” (ibid.:40). These two uses are illustrated here. (7) a. rast-ē tē street-OBL along ‘along the street’ (ibid.:40) b. chokarī-a khatu likhyo girl-OBL letter wrote ‘the girl wrote a letter’ (ibid.) Sindhi then would seem to be an example of a language where the ergative (which Yegorova calls the oblique) is the main case assigned by adpositions. To account for this type of language, we should look for properties shared by P and whatever category assigns ergative case. Type Ig—Locative Assigned to All or Most POs There is at least one language where the locative is the main case borne by POs. In the ChukotkoKamchatkan language Chukchee postpositions “[u]sually take Loc[ative] case” (Spencer, 1999, chapter 3, section 2).

< previous page

page_141

next page >

< previous page

page_142

next page >

Page 142 Following are examples of the Chukchee locative in its locational and prepositional object functions. (8) a. memylte kupre-k gakwalenat seals net-LOC got.caught ‘The seals got caught in the net’ (ibid.:ch. 3, section 1.2.3) b. ŋinqej nynnnyŋyttyqin ytlyg-yk reen boy fished father-LOC with ‘The boy fished with the father’ (ibid.:ch. 3, section 2) It is unclear how and whether this can be explained in a principled way. A possible clue comes from Spencer (ibid.): “Postpositions [are] mainly derived from adverbs.” Perhaps they share some feature with this class of words that is connected with their case assigning properties. However, the locative of Chukchee is rather different from the locative of, for example, Turkish in that it is not only used for indicating location. The locative is also used to indicate the object of comparison with comparative forms and is one of the cases that can mark the direct object in the antipassive. In addition, Chukchee does not have a genitive case; some possessors are in the absolutive case, but the notion of possession can also apparently be expressed by the locative: (9) a. Ekke-k ujŋe milger son-LOC NEC gun ‘The son hasn’t got a gun’ (ibid.:ch. 8, section 1.1) b. muryk amqyn-ŋawysqat-ety us.LOC EACH-woman-ALL ‘to each of our women’ (ibid.:ch.10, section 1.2) We can thus say that Chukchee is not a pure Type Ig language, in which a case is used only for locations and POs. Other Type I Languages The Iranian language Yagnob has only two cases, the absolute and the oblique, but it has a complex distribution of them. According to Comrie (1981:170), the latter “is used for indirect and prepositional objects, for possessors in the possessive construction […], for subjects in the ergative construction and for definite objects in the nominative-accusative construction.” Assuming that Comrie means that all POs bear this case, one could classify this as a Type If language, but not a pure one, since the ergative case has other functions than the usual ergative, or for that matter as an impure Type Ic or Id language. However, this would perhaps

< previous page

page_142

next page >

< previous page

page_143

next page >

Page 143 be misleading; here is a good instance of the problems sometimes encountered when attempting to compare cases cross-linguistically. It will be interesting to see whether languages with a similar case system also use the oblique to mark POs. (Kanashi is somewhat similar, but uses the caseless/nominative form, as we have seen.) In such languages, P may have some feature in common with whatever assigns case to the other elements marked with oblique. Ossetic, which is related to Yagnob, has nine cases. The case referred to as the oblique by Comrie (1981:171) and the genitive by Abaev (1964) is borne by definite direct objects, possessors, and it is also “the usual case for prepositional government” (ibid.: 18), though “in rare instances” (ibid.:35) a different case is assigned. If we consider this case to be genitive, then the reason why it is the adpositional case in this language is the same as in (at least some of) the Type Id languages, the categorical status of its “postpositions”: Abaev (ibid.:33) says, “On the morphological level, postpositions cannot be separated from substantives, and their inclusion in a special category is justified only from the syntactic-functional point of view.” There are some exceptions to this, that is, some but not all adpositions are nominal in origin. We have seen that Type I languages use a variety of cases as their adpositional case, but some pure types are rare or non-existent. For example, I know of no languages in which the instrumental is the only or main case assigned by Ps. TYPE II—LANGUAGES WITH MORE THAN ONE ADPOSITIONAL CASE We now turn to languages with more complex systems of adpositional case assignment. These languages are not limited to a single adpositional case, or one main one with some exceptions. This type includes the best known “case languages,” ancient Greek, Latin, Russian, and German, as well as many other languages of various families: languages from different branches of Indo-European such as Albanian, Latvian, Gothic, Middle Dutch, Old English, Sanskrit, and Czech; Uralic languages such as Finnish and Hungarian; Altaic languages including Turkish; and the language isolate Basque. The Dravidian family is also represented here (Tamil), as are the Cushitic (Awngi) and Caucasian (Lezgian) families. In fact, one might have the impression that languages with a fairly extensive case system will make use of their variety of cases in marking POs, and it is true that the Type II languages may generally have larger case inventories than the Type I languages. This is not always true, since, as we have seen, the (impure) Type Ie languages Udmurt and Kanashi both have fairly large case systems, while the Type Id language Ladakhi

< previous page

page_143

next page >

< previous page

page_144

next page >

Page 144 has seven. On the other hand, the Type Ia and Ib languages I have found, such as English, are relatively poor in cases. In languages of this type, it is fairly clear that some sort of lexical case must be involved: At least some individual Ps will be listed in the lexicon as assigning a particular case. Type II has more variations than Type I. On the one hand, particular Ps might each assign only one case, but this case may differ from P to P. This is illustrated here with Turkish postpositional phrases. (10) a. bu vaziyet-e göre this situation-DAT according to ‘in view of this situation’ (Lewis, 1988:88) b. bugün-den evvel today-ABL before ‘before today’ (ibid.:89) On the other hand, a particular P might assign different cases depending on some factor. One factor is the meaning to be conveyed, another is the category of the complement. Many languages exploit both of these possibilities: For example, Latin, Classical Greek, Turkish, and German all have Ps that differ in which case(s) they assign; and some individual Ps in each of these languages can assign different cases depending on the meaning or another factor. Space limitations do not permit a detailed analysis or even listing of all the variations of Type II languages that occur. I shall simply briefly deal with the following issues: (1) general possibilities and tendencies of case choice, (2) semantic factors determining case choice, and (3) non-semantic factors determining case choice. General Possibilities and Tendencies Among Type II languages, we find cases assigned to POs that are not used, or rarely used, for marking POs in Type I languages. For example, as far as I know there is no sub-class of Type I language where the ablative is the sole or main case assigned by Ps, while Latin has prepositions assigning this case. This is perhaps to be expected, given the wider possibilities of case assignment by Ps among Type II languages, and the restricted case inventories of many Type I languages; few of them have an ablative case. However, we can note some general tendencies. As noted previously, care must be taken with the names of cases, since, for example, the dative of one language may be somewhat different from the dative of another language, and it is possible that the difference could play a role in whether the case is assigned to POs. Nevertheless, the general content

< previous page

page_144

next page >

< previous page

page_145

next page >

Page 145 of labels such as genitive and dative (cases of possession and indirect objects, respectively) is fairly stable across different descriptions of languages, and I shall continue to use them. There is one case that I do not ever recall having seen marked on POs in either Type I or Type II languages, the vocative, while three other cases, the nominative, ergative, and absolutive, are rare in this function. The absence of the vocative here is not at all surprising; there have been those who would deny that it is even a case (for example, Hjelmslev, 1935). In any event, its canonical function, marking addressees, is quite different from that of the other cases. Blake (1994:9) says that vocatives “are unlike other cases in that they do not mark the relation of dependents to heads”; one would therefore not expect them to also have the function of marking POs, or objects of verbs either. One might think of the nominative and ergative as typically marking subjects, and this is a major use of the absolutive, in addition to marking objects. There may be an intuition that it would be unusual for these cases to be assigned to POs, even among “errors” in case marking due to language change. There could be something behind this intuition, since in many languages there are no nominative POs, but as we have seen, in languages of Types Ie and If (most) Ps assign the nominative and ergative, respectively. In some Type II languages the nominative is assigned by some Ps, whereas other Ps assign other cases. For example, there are two Albanian prepositions nga “from” and te/tek “near” that assign nominative (a fact that Demiraj [1998:492] calls “striking”), while other prepositions in the language assign accusative or dative-ablative. This is apparently because, as mentioned by Buchholz and Fiedler (1987:215), these prepositions used to be conjunctions, introducing subordinate clauses; hence they were complementizers. Perhaps the PPs they head still have an abstract INFL/Tense node (i.e., are still similar to CPs), and this element assigns nominative case. Lezgian postpositions assign a variety of cases, including the genitive and superelative, but the absolutive can also appear on POs. For example, Haspelmath (1993:222) says that patal “for” “inexplicably takes an Absolutive argument.” This postposition is not derived from a complementizer, so again the explanation given for the two Albanian postpositions will not work here. Some cases may not often be marked on POs, because they are rare among the world’s languages and/or because they exist in languages with extensive case systems and relatively small inventories of adpositions, or because they are already adposidonal in meaning (although here again we have the problem of distinguishing between case markers and adpositions). Thus although the superelative is assigned by some postpositions in Lezgian, as is the inelative, this will be uncommon cross-linguistically.

< previous page

page_145

next page >

< previous page

page_146

next page >

Page 146 One point of interest is which cases are not assigned to POs, other than those just mentioned, within a language that assigns a variety of cases to such NPs. Some languages with fairly rich case systems, such as Russian and Slovak, use all of their non-nominative cases to mark POs. This is not true of all Type II languages, however. In Sanskrit adpositions assign genitive, accusative, ablative, locative, and instrumental, but not dative: What is special about the dative in this language (or about the category or categories that assign it)? The words “in this language” are important, because certainly the dative in some other languages can be assigned by adpositions. There is at least one other Indo-European language with the dative that does not use it to mark POs, namely Latin, although this gap is less striking in Latin, since, with one exception, it uses only two cases to mark POs. In Lithuanian dative POs occur only “in certain fixed phrases” (Dambriunas et al., 1966:296). If there is a reason behind such gaps, we might find it by comparing the datives of Sanskrit and Latin with those of Russian and Slovak. Or, just as we looked for shared properties of languages generally or always using one particular case to mark POs, we can look for shared properties of languages not using certain cases in this way: For example, is there some difference in the way indirect objects behave in Sanskrit and Russian? It may well turn out that this is a random gap, but before asserting this, we should attempt to attribute it to other factors. Another such gap involves the locative. Turkish Ps can assign the absolute (which marks subjects and indefinite direct objects), the genitive, the dative, or the ablative.13 Somewhat similarly, in another Turkic language, Kirghiz, Ps can govern every case but the locative and the objective (= accusative). Here I suspect that the gap is largely due to semantic reasons: The locative already has some semantic content, some of the content that would be added to an NP by adding an adposition in some languages. This may not be the whole explanation, however, given that, for example, some Vedic Ps can assign locative and given that the ablative appears to also be a more semantic case and yet is assigned by some Turkish and Kirghiz Ps. There is a significant difference between the locative of these Turkic languages and the case sometimes referred to as the locative of Russian (also known as the prepositional case): The latter is assigned only by prepositions and never appears under other circumstances, while the former are never assigned by Ps. This might indicate that the Russian locative has less semantic content than that of Turkish. We might also look for implicational hierarchies among Type II languages: For example, “if a language has the following five cases, and three of them are assigned to POs, then they will be the following three.” Such hierarchies will be elaborate, if they exist at all, since as we have seen, if a language uses only one case to mark POs, it is difficult or

< previous page

page_146

next page >

< previous page

page_147

next page >

Page 147 impossible to determine a priori which one it will be. Perhaps after looking at features of the language such as the possession of double object constructions, as discussed previously, one might be able to predict the case used, although it is not yet clear whether that is so. However, one tendency is that among many Type II languages, the accusative will be among the cases assigned by Ps. This holds for all of the Indo-European Type II languages I have mentioned and for Awngi. However, it does not seem to hold for modern Turkish, with the exception of the “obsolescent” adposition mentioned in note 13, or for Turkmen, Kirghiz, or Chagatay. Another Turkic language, Uigur, does have two postpositions that assign the accusative; I would guess this is due to the fact that they were derived from gerundives. In Orkhon Turkic some postpositions assign accusative to certain types of words. I know of no adpositions in Uralic languages that assign accusative, but note that some languages of this family do not have an (overt) accusative. In languages where different Ps assign different cases, but where each P assigns only one case, we can look for properties these particular Ps share with other case-assigning elements in the language. This is what we did with respect to Type I languages, but there we were looking at Ps as a class in a language. For example, in a language where some Ps assign genitive and others accusative, is there some property that the former share with Ns and some other property that the latter share with Vs? Of course which case is assigned might have something to do with the semantics of the cases and adpositions, but other factors may also come into play. An obvious one to investigate is the etymological origin of the Ps: Some Ps are derived from nouns, and others from verbs: In a language that has both kinds of Ps, is there a tendency for the former to assign genitive and the latter accusative? Even though no one case may predominate with respect to marking POs in Type II languages, one may wonder whether there is a default case for this purpose in at least some such languages. If so, then the burden of lexical marking will be lessened, as only Ps that assign a non-default case will require lexical marking. However, it may not be a simple matter to determine what the default case is in a language, and perhaps even whether there is one. Beard (1995:236) cites two works in which it is asserted that the default case in German is the dative, Emonds (1985:227) and Bierwisch (1988:39). Beard takes issue with this conclusion, saying that “by one measure Dative is the default case for Ps and by another, Genitive is” (ibid.). It is common for one P to assign more than one case, due to different factors. Many Ps assign two cases, and some (e.g., in ancient greek) assign three. I know of only one P in any language that assigns more than three: The Vedic preposition parás “beyond” has complements in the accusative, ablative, locative, and instrumental cases. Part of the reason for this

< previous page

page_147

next page >

< previous page

page_148

next page >

Page 148 limit is presumably because to exceed it, a language would have to have a relatively large case system —that is, at least four cases, and at least five if the nominative is not an adpositional case in the language.14 Adpositions, like many other kinds of words, can be borrowed, although this happens less often with this class than with nouns. What happens to the case assignment when an adposition is borrowed? Sometimes it would have to change, if it assigned a case in the source language which did not exist in the borrowing language. I leave this subject for further research along with the question of what happens to adpositional case marking as the case system of a language changes (often in the direction of having fewer cases). Some of the Type I languages are descended from Type II languages—for example, Catalan from Latin; is there a particular point in the decline of a language’s case system that it will tend to change from one type to the other? And, when there are diachronic changes in PO case marking, say between Old English and Modern English, does that mean that something about (some) Ps has changed, that is, have they lost (or gained) some feature? Semantic Factors As noted, in some languages particular Ps will take more than one case depending on the meaning intended. Are there any generalizations that can be made here? In several languages a P will take accusative when motion to some location is meant, and another case when location is meant. For example, the Latin preposition in takes the accusative when meaning “into” and the ablative when meaning “in,” as shown below. (11) a. i-bo in Pirae-um go-FUT.1sg in Piraeus-ACC ‘I will go into the Piraeus’ b. in urb-e Rom-a in city-ABL Rome-ABL ‘in the city of Rome’ (Kennedy, 1962:145) As mentioned earlier, the German preposition in behaves similarly (the cases involved being accusative and dative), and the same sort of situation also holds in Russian. However, this does not seem to be a general property of Type II languages; I do not know of its occurrence outside of Indo-European. There is at least one semantically oriented type of case-marking split that occurs in more than one family. The Mongolian postposition ömnö “before” assigns genitive when used with a spatial meaning, and ablative when used with a temporal meaning, except that it “even in its temporal

< previous page

page_148

next page >

< previous page

page_149

next page >

Page 149 meaning sometimes takes the Genitive Case, […]. In modern Khalka Mongolian, the Genitive Case is used more often with postpositions than formerly, thus tending to replace other cases” (Rincine, 1952:47, quoted in Buck, 1955:129). This case-meaning correlation does not hold consistently even in Mongolian, but it is an intriguing coincidence that in the unrelated language Lezgian there are several postpositions that assign genitive when indicating a spatial meaning, but superelative when used temporally—for example, wilik “before.” However, given the lack of universality of this phenomenon, I assume that such case marking will have to be marked lexically, and not attributed to general semantic properties of cases. It would appear to be difficult to make a principled account of all the splits in adpositional case marking due to meaning. For example, it is not surprising that several ancient Greek prepositions assign accusative objects when meaning “to.” What is more difficult to account for is the fact that metá means “with” when assigning genitive and “after” when assigning accusative, or that diá means ‘through’ with the genitive and “on account of with the accusative. In any event, even if there are semantic regularities, much of this semantically-related case assignment will have to be explained by lexical marking of adpositions. Non-Semantic Splits in Adpositional Case Assignment Some Ps assign different cases for reasons that appear not to be semantic. There are four Turkish postpositions that occur with the absolute of nouns but the genitive of some personal, interrogative, and demonstrative pronouns. These are gibi “like,” için “for,” ile “with,” and kadar “as much as.” The pronouns of the types mentioned to which the genitive is not assigned include ne “what” and those with the plural ending -lar . The situation is illustrated in (12). (12) a. sen-in gibi you-GEN like b. on-lar gibi they-PL like c. bülbül gibi nightingale (ABSOLUTE) like (Lewis, 1988:86) This would appear difficult to account for, especially since it is not all pronouns that are assigned genitive.15 Kornfilt (1984:61, 225–27) has discussed this question, but I am not convinced that her solution will explain all the complexities involved here. Van Schaaik’s (1998)

< previous page

page_149

next page >

< previous page

page_150

next page >

Page 150 diachronically oriented remarks on gibi may point toward a general answer: This word was originally a noun, and so took a genitive complement, but there then occurred in this context the “loss of the genitive case marker, except for highly frequent words such as personal pronouns and demonstratives.” In the theoretical terms discussed earlier, we could say that such Ps, being derived from nouns, bore the feature [+N], and so assigned genitive case to their complements. However, this [+N] feature has been eroded in some way, as the elements have become less nounlike, and so generally genitive case can no longer be assigned. The absolute case is then assigned either by the P, or through some default case assignment mechanism. The instances where genitive is still assigned are like fixed phrases, and so the genitive marking is retained. Presumably something like this may happen in other languages when words evolve from nouns into adpositions, and it would be interesting to see whether there is a similar split among their objects with respect to case assignment. I shall mention one other non-semantic split. Latvian prepositions assign various cases to singular NPs, but only dative/instrumental16 to plural complements (compare the rarity of the dative on POs in the only other living Baltic language, Lithuanian). This split, though not as complicated as the one in Turkish, will still be difficult to account for. TOWARD A MINIMALIST ACCOUNT OF ADPOSITIONAL CASE Up to this point I have been speaking of case assignment—that is, I have been thinking in terms of the case theory of Government-Binding Theory. One might wonder what a Minimalist view of adpositional case would be like. The latter framework has not dealt much with this kind of case. Some authors (e.g., de Jong, 1994) have posited or discussed a category AgrPP, parallel to AgrSP and AgrOP. One would assume that the case of at least some adpositional objects (POs) is checked by means of spec-head agreement with the Agr head (to which the P has adjoined). The question then is which instances of adpositional case marking can be accounted for in this way. It seems to me that this mechanism will be able to deal with languages in which all POs bear the same case, and one may still be able to make generalizations in terms of categorical features—for example, for the Type Ib languages, if a [-N] head adjoins to an Agr head, it will check accusative case. Type II languages will be more difficult to account for. As noted, presumably Ps will need to be lexically marked for assigning particular cases, with the possible exception of those Ps assigning a default adpositional case, if there is one. Perhaps POs in such languages do not get their case by moving to the spec of

< previous page

page_150

next page >

< previous page

page_151

next page >

Page 151 AgrPP, but rather get their case in situ, or perhaps their cases are also checked through spec-head agreement. The question is how similar such cases are to the clearly structural cases that are checked in the specifiers of AgrSP and AgrOP, and the answer to this is not yet clear. CONCLUSION We have seen that there is a wide range of possibilities among the world’s languages for the assignment of case to adpositional objects: A variety of cases can be assigned by adpositions, either as the only adpositional case of a language, or as one of several possibilities within that language. In addition, various factors in different languages determine which case a PO will bear: the particular adposition involved, the meaning intended, and certain properties of the PO itself. To account for this complex array of data, some semantic and lexical factors will have to be invoked. However, I have pointed out directions one might take in an attempt to explain some adpositional case assignment in a principled way. NOTES For comments, discussion, and/or proofreading, I thank Mengistu Amberber, George Horn, Christo Moskovsky, Jerry Sadock, Ruth Stockdale, and an anonymous reviewer. The responsibility for any errors is of course mine. 1. Such a principle is put forth by Comrie (1986:91–92), who says, “If a case distinction is made formally in any nominal, then that same case distinction exists for all nominals.” 2. Russian has a “prepositional” (or locative) case; however, other cases in the language can be assigned to prepositional complements, so Russian does not belong to this sub-class of Type I languages, or to Type I at all. 3. The strong forms are the same as the nominative in all but the first person singular and the third person reflexive. However, as noted, this is enough to justify positing another case or form, and saying that an abstract difference (if it is a case difference) also exists in nominal paradigms. There is one situation in which a preposition will take a different sort of pronoun: “When a preposition governs two or more co-ordinated pronouns, one of which would be mi, this takes the form of jo” (Gili, 1967:44), jo being the nominative form of the first person singular pronoun. 4. However, he later (ibid.:293) limits the scope of this: “We may modify Kayne’s suggestion slightly, assuming that the case system is lost in English only within VP, whereas PPs that are immediate constituents of S have inherent Case-marking.” 5. I do not include secondary predicate type constructions or small clauses under the heading of double object constructions. 6. Some varieties of Portuguese and Portuguese-based creoles may be counter-examples to this: de Mello (1996:176) gives an example of a double

< previous page

page_151

next page >

< previous page

page_152

next page >

Page 152 object construction in Brazilian Vernacular Portuguese (BVP) (i.e., where the indirect object is not preceded by a P), which does not happen in the standard language; on the next page she gives such an example from Guinea-Bissau Creole, while at least one BVP variety seems to have the prepositional form mim (ibid.:295), and as mentioned previously, at least one Portuguese-based creole has prepositional pronouns. However, I do not know whether there is any variety that has both prepositional pronouns and double objects, and in any case, Portuguese and some other Romance languages, as noted, may not in fact possess prepositional cases of pronouns, but rather strong forms. 7. Another analysis of Type Ib languages is that there is a basic opposition between (finite) subjects and most other kinds of arguments and non-arguments, and this is reflected in some feature in the relevant heads: INFL or T bears the (case) feature [+subject] while P and V bear the feature [subject]. In languages of Type Ia, there may be a different feature in the heads: P bears a feature such as [+oblique], while T and V bear the feature [-oblique]. 8. Ouhalla (1988) argues that prepositions are [+N]. Perhaps this value of Ps can differ from language to language, or among Ps within a language, accounting for some differences in the cases they assign. 9. This is not the same situation as when an adposition bears a marker to agree in person with its object. 10. Note the remark by Abondolo (1998:23): “In languages closer to the core [of Uralic], which have no genitive case, such postpositions most frequently take the nominative”. 11. See, for example, Blake (1994, Section 1.2.5), on the questions of case markers versus adpositions. 12. Secondary cases should not be confused with multiple case marking due to agreement: In several languages some nouns (e.g., those denoting possessors) may bear more than one case marker, one to indicate their own function, and another to agree in case with another element, such as the head of the NP, somewhat like case agreement of adjectives in Latin. 13. In addition, one postposition, mütaakip “following, after,” which Lewis (1988:85) calls “obsolescent” and which was borrowed from Arabic, takes the accusative. 14. One could ask questions similar to those raised here with respect to verbs, and compare the answers for the two word classes. For example, other than the obvious accusative, what cases are and are not assigned by verbs in a language? Some verbs assign different cases in different situations, sometimes, as with adpositions, associated with different meanings. What is the maximum number of different cases that are thus assigned by any one verb? 15. Complicating the picture still further is the point that even the pronouns in question can be assigned absolute here, “colloquially” as Lewis (1988:85) says. 16. The dative plural and instrumental plural have the same forms in Latvian. REFERENCES Abaev, V.I. 1964. A Grammatical Sketch of Ossetic (edited by H.H.Paper, translated by S.P.Hill). Publications of the Indiana University Research Center in Anthropology, Folklore, and Linguistics 35. Abondolo, D. 1998. Introduction. In D.Abondolo (ed.) The Uralic Languages. London: Routledge.

< previous page

page_152

next page >

< previous page

page_153

next page >

Page 153 Appleyard, D.L. 1987. A Grammatical Sketch of Khamtanga—I. Bulletin of the School of Oriental and African Studies 50.2:241–66. Beard, R. 1995. Lexeme-Morpheme Base Morphology. Albany: State University of New York Press. Bierwisch, M. 1988. On the grammar of local prepositions. In M.Bierwisch, W. Motsch, and I.Zimmerman (eds.) Syntax, Semantik und Lexikon (Studia Grammatika 29). Blake, B. 1994. Case. Cambridge, U.K.: Cambridge University Press. Brecht, R.D., and Levine, J.S. (eds.) 1986. Case in Slavic. Columbus, OH: Slavica. Buchholz, O., and Fiedler, W. 1987. Albanische Grammatik. Leipzig: VEB Verlag Enzyklopädie. Buck, F.H. 1955. Comparative Study of Postpositions in Mongolian Dialects and the Written Language. Cambridge, MA: Harvard University Press. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Comrie, B. 1981. The Languages of the Soviet Union. Cambridge: Cambridge University Press. ——. 1986. On delimiting cases. In R.D.Brecht and J.S.Levine (eds.) Case in Slavic. Columbus: Slavica Csúcs, S. 1998. Udmurt. In D.Abondolo (ed.) The Uralic Languages. London: Routledge. Dambriunas, L., Klimas, A., and Schmalstieg, W.R. 1966. Introduction to Modern Lithuanian. New York: Darbininkas. Demiraj, S. 1998. Albanian. In A.G.Ramat and P.Ramat (eds.) The Indo-European Languages. London: Routledge. Emonds, J. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris. Gili, J. 1967. Introductory Catalan Grammar. Oxford: Dolphin Book Co. Gray, L.H. 1934/1971. Introduction to Semitic Comparative Linguistics. Amsterdam: Philo Press. Green, J.N. 1988. Romance Creoles. In M.Harris and N.Vincent (eds.) The Romance Languages. London: Routledge. Haspelmath, M. 1993. A Grammar of Lezgian. Berlin: Mouton de Gruyter. Hjelmslev, L. 1935. La catégorie des cas: Etude de grammaire générale. Copenhagen: I.Munksgaard. Jannaris, A.N. 1897. Historical Greek Grammar. London: Macmillan. de Jong, J.J. 1994. Local binding and the minimalist framework. Language and Cognition 4:11–124. Kennedy, B.H. 1962. The Revised Latin Primer (edited and further revised by J. Mountford). London: Longman. Kornfilt, J. 1984. Case Marking, Agreement, and Empty Categories in Turkish. Ph.D. Dissertation, Harvard University. Koshal, S. (Edited by B.G.Misra). 1979. Ladakhi Grammar. Delhi: Motilal Banarsidass. Lewis, G.L. 1988. Turkish Grammar. Oxford: Oxford University Press. Mel’čuk, I.A. 1986. Toward a definintion of case. In R.D.Brecht and J.S.Levine (eds.) Case in Slavic. Columbus, OH: Slavica. de Mello, H.R. 1996. The Genesis and Development of Brazilian Vernacular Portuguese. Ph.D. Dissertation, City University of New York. Ouhalla, J. 1988. The Syntax of Head Movement: A Study of Berber, Ph.D. Dissertation, University College London. Palmer, F.R. 1967. “Bilin.” Lingua 17:200–9. Riemschneider, K. 1977. An Akkadian Grammar (translated by T.A.Caldwell, J.N. Oswalt, and J.F.X.Sheehan). Milwaukee: Marquette University Press.

< previous page

page_153

next page >

< previous page

page_154

next page >

Page 154 Rincine, A.R. 1952. Ucebnik Mongol’skogo Yazika. Moscow: Nauka. Rowley, A.R. 1990. East Franconian. In C.V.J.Russ (ed.) The Dialects of Modern German. London: Routledge. van Schaaik, G. 1998. On the usage of Gibi. In L.Johanson, in cooperation with E.A.Csato, V.Locke, A.Menz, and D.Winterling (eds.) The Mainz Meeting. Proceedings of the Seventh International Conference on Turkish Linguistics. Wiesbaden: Harrassowitz. Shackle, C. 1972. Punjabi. London: Teach Yourself Books. Sharma, D.D. 1992. Tribal Languages of Himachal Pradesh (Part Two). New Delhi: Mittal Publications. Spencer, A. 1999. Chukchee. URL at http://privatewww.essex.ac.uk/~spena/Chuk-chee, version current on April 28, 2000. Yates, A. 1975. Catalan. London: Teach Yourself Books. Yegorova, R.P. 1971. The Sindhi Language (translated by E.H.Tsipan). Moscow: Nauka.

< previous page

page_154

next page >

< previous page

page_155

next page >

Page 155 7 Optimality and Three Western Austronesian Case Systems Anna Maclachlan The aim of this paper is to apply the Optimality Theoretic typology of case systems proposed by Legendre et al. (1993) to three representative Western Austronesian languages: Malagasy, Chamorro, and Tagalog. The Legendre et al. typology, as it relates to ergative and accusative systems, is tested and extended in the process, but its assumptions are maintained. The typology is tested, since applying it to new languages in some detail raises problems for the proposal. These problems are documented here, but are mostly not solved. Through its application to new languages, the typology is extended in three distinct ways. First, there is a sentence type, the circumstantial topic sentence, which is central to all three Western Austronesian languages that the proposal is shown to accommodate. Second, Tagalog has a significantly different set of output sentence types as compared with the other two languages and indeed as compared with languages of the world. The most common transitive sentence type never succeeds as an optimal candidate, and three related uncommon sentence types that have not been considered in Legendre et al. (1993) do surface in Tagalog. Third, Tagalog belongs to a language type, the Philippine type, which can be included within the Legendre et al. typology with only minor extensions to the constraint set. This is significant since for many typologies of case systems, the Philippine type presents a challenge, being neither accusative nor ergative. For example, in Dixon’s (1994) extensive characterization of ergative versus accusative systems, Philippine-type languages are simply deemed difficult to characterize and sidestepped.

< previous page

page_155

next page >

< previous page

page_156

next page >

Page 156 BACKGROUND: CASE SYSTEM TYPOLOGY, OPTIMALITY AND AUSTRONESIAN Ergative and accusative systems differ in the following way. In an accusative system, the agent of a basic transitive sentence patterns with the sole argument in an intransitive sentence. In an ergative system, it is the patient in a basic transitive sentence that patterns with the sole argument of an intransitive. An additional system is one in which patient-like intransitive arguments pattern with transitive patients and agent-like ones pattern with transitive agents. This system, known as activestative, is not considered in this chapter. It will suffice to show that intransitives behave uniformly in the languages under consideration. Optimality Theory and Case Systems: Legendre et al. (1993) The notion of two arguments patterning together is considered by Legendre et al. (1993) in terms of abstract case. That is, arguments show the same overt case marking or trigger the same form of agreement, or exhibit the same relative word order. Their abstract cases are simply labeled C1, C2, and C4. C1 is defined as the abstract case on the agent argument and C2 as that on the patient in a basic transitive sentence. These are the two core cases. C4 refers to all other lower cases. The inputs to their optimality theory (OT) analysis consist of combinations of agent and patient arguments without case, and the output candidates are those arguments with the various possible abstract case combinations. The arguments in the input are also distinguished as having high or low prominence. I will follow the annotation in Legendre et al. (1993) in which high prominence is indicated with upper case and low prominence with lowercase. For example, a passive input is one in which the agent (a) is low prominence and the patient (P) is high prominence, and hence is referred to as aP. Other inputs in their typology are A, an agent-like intransitive argument; P, a patient-like intransitive argument; AP, a basic transitive in which both arguments are high in prominence; and an antipassive Ap. Outputs to consider for the passive are a1P2, a2P1, a4P1, and a4P2. Other outputs such as a1P1 are not considered, since the same case cannot appear twice in the same output in general and still others, such as a1P4 are not considered, since they are too far outside the realm of possibility, given the constraints. The eight constraints proposed by Legendre et al (1993) in (1) are simple and general but also violable and conflicting, as they should be in an OT approach. It is in ranking the constraints differently in different languages that the conflicts come to be resolved with different results.

< previous page

page_156

next page >

< previous page

page_157

next page >

Page 157 (1) The Universal Constraints Governing Abstract Case: Legendre et al. (1993) a. A = C1: Agents receive abstract case C1. b. P = C2: Patients receive abstract case C2. c. A ≠ C2: Agents do not receive abstract case C2. d. P ≠ C1: Patients do not receive abstract case C1. e. α ≠ C4: Core arguments (agents and patients) do not receive abstract case C4. f. α = C2: Some argument is case-marked C2. g. X = C1: High-prominence arguments receive abstract case C1. h. x ≠ C12: Low-prominence arguments are not core case-marked (C1 or C2). These constraints capture basic intuitive generalizations such as “agents are usually realized as subjects” and “high-prominence arguments are usually core case-marked, whereas low-prominence arguments are not.” Examples of how these constraints interact will be given while each individual language is considered in the next section. Next we can situate this OT approach within a larger context. The Legendre et al. (1993) article was an early attempt to apply OT in the domain of syntax. It proposes simple constraints that do not rely on syntactic assumptions from any particular framework. Many applications of OT within syntax since then have involved grafting OT principles onto a pre-existing set of non-OT principles from some syntactic framework. The Legendre et al. (1993) Approach Compared with Other Approaches As an example from case systems in particular, Woolford (2001) proposes to maintain Chomskyan case licensing principles as inviolable but to apply violable OT constraints in order to choose among the particular cases that could be licensed on a given argument. Under her approach, all languages have case markedness constraints of the form *accusative and *ergative. All languages also have faithfulness constraints that require that lexically specified inherent case features on verbs be realized in the output. The input consists of a verb with case features and caseless arguments, and the output consists of the verb with its case features checked and the arguments with their surface cases. In an accusative language, *ergative is ranked above all faithfulness constraints so ergative case never surfaces. In an ergative language, *ergative is ranked below a faithfulness constraint that is contextualized to apply only in transitive sentences. This ensures that ergative case surfaces on transitive agents but not elsewhere. Under such an approach, the OT constraints seem to play a minor role. Inputs specified with ergative subject features lead to outputs containing ergative subjects. For this reason, I have

< previous page

page_157

next page >

< previous page

page_158

next page >

Page 158 preferred to explore further the more theory-neutral constraints conceived in Legendre et al. (1993). One drawback of not relating the constraints to particular structures is that the structural effects associated with a given case system cannot be easily explored. Consider some approaches to the typology of case systems that have not been based on OT. For example, under recent Chomskyan approaches, including Murasugi (1992), Campana (1992), and Maclachlan (1996), languages differ based on differences in the availability of structural positions or in restrictions on (overt or covert) movement between positions. Since the approach of Legendre et al. (1993) is not tied to syntactic structure, it is more akin to linking approaches in which arguments are ordered according to a thematic hierarchy, and a language will have ordered case frames or grammatical relation realizations to which they link. This is coupled with the notion that special lexically determined cases can be prelinked and therefore override the general tendencies. This type of approach is espoused in general in Yip et al. (1987) and in Carrier-Duncan (1985) specifically for Tagalog, a language in which such linking is registered by way of distinctive verbal morphology. Another linking approach is taken within Head-Driven Phrase Structure Grammar. Manning and Sag (1998) argue for separation but linking between two distinct lists associated with the subcategorized arguments of a head: (1) the argument structure list of the head, which corresponds to argument roles; and (2) the valency list associated with that head (usually thought of as a concatenation of lists), which corresponds to the surface grammatical relations in which arguments of that verb can be realized. They are then able to characterize the distinction between accusative and ergative languages and indeed between these and Western Austronesian languages (like Tagalog) in terms of different linking patterns between the two lists. In accusative languages, the first argument in the argument structure list (usually an agent) is linked to the subject surface grammatical relation. In ergative languages, the second argument (usually a patient) is linked to the surface subject. It is especially relevant here that in Western Austronesian languages, either the first or the second argument can be linked to the surface subject. The insight of Manning and Sag (1998), that in some languages either linking pattern is possible, is captured under OT assumptions in the subsection, Tagalog. Finally, it is worth noting that Sells (2001) takes an OT approach to linking in a reworking of the Legendre et al. (1993) typology. While his focus is on developing the typology to better handle grammatical voice systems, my focus is on how their typology handles case systems. He, too, is interested in accommodating Philippine-type languages within the typology as we will see in more detail in the subsection, The Sells Analysis of Philippine-type Languages.

< previous page

page_158

next page >

< previous page

page_159

next page >

Page 159 Western Austronesian Languages The Western Austronesian languages presented in this chapter are representative of three case patterns: Malagasy has an accusative pattern, Chamorro has a split ergative pattern, and Tagalog has a Philippine-type pattern. This last pattern is not restricted to languages spoken in the Philippines, so the term is unfortunate in being too narrow. Manning and Sag (1998) refer to the languages having this pattern as Western Austronesian, but this is too broad for the present purpose since languages like Malagasy and Chamorro are also Western Austronesian but do not exhibit the pattern. The circumstantial topic (CT) is a sentence type that is not commonly found in the languages of the world but is a prevalent sentence type among Western Austronesian languages, including the three languages described in this chapter. It is a sentence type in which (1) an optional participant that would normally be realized in some oblique case appears core case-marked and (2) there is CT morphology on the verb. For example, in a Malagasy circumstantial topic sentence, we will see in the subsection, Circumstantial Topic in Malagasy, that the circumstantial participant is not expressed as a prepositional phrase but is nominative, while the agent is marked, as it would be in a passive. The Legendre et al. (1993) OT system, which was devised without consideration of this additional sentence type, can nevertheless accommodate it under each of the case system types considered here. Consider an intuitive constraint that is adhered to in most world languages: “a non-core argument cannot have high prominence.” As a result, inputs with a high-prominence circumstantial participant cannot usually be parsed. In Austronesian languages, however, there is a mechanism for making circumstantial participants into core participants, namely the CT sentence, and so these inputs can be optimal. Applicative sentence types found in other world languages also avoid violations of this constraint and are not addressed by Legendre et al. (1993). The difference between CT sentences and applicatives in terms of the Legendre et al. approach is that the circumstantial participant ends up in C1 in the former and C2 in the latter. As CT sentences have an additional argument, they require an extension of Legendre et al. (1993) similar to extensions required of other sentence types where there is an additional argument, including three place verbs and causatives. When a verb takes an agent, a patient, and a goal, using G to refer to the goal, the relevant input is APG. In such sentences there is the possibility of marking either the patient or the goal or indeed both with C2. These correspond to A1P2G4, A1P4G2, and A1P2G2 outputs, respectively (with prominence variations). To exemplify

< previous page

page_159

next page >

< previous page

page_160

next page >

Page 160 this further, causatives include an additional agent argument and hence have input AAP. Outputs vary across languages, being A1A2P4, A1A4P2, or A1A2P2 (again with prominence variations). The additional argument in the CT sentence is not subcategorized by the verb; rather it is an adjunct— call it X—that functions as an argument with respect to case marking and syntactic behavior. It has been promoted to core status by derivation. Since it has high prominence in the sentence, our input will contain an additional uppercase X. As for the other arguments in the input, it is not obvious what their prominence status should be. In other words, the CT sentence could be based on a transitive (APX), a passive (aPX), or an antipassive (ApX); or it could be that the relative prominence of X is so high that the input is apX. Interestingly, we find that the three different languages we consider differ in their optimal CT outputs. For each language, based on the ranking of case constraints established for sentences involving agents and patients only and based on differing case patterns exhibited in the CT sentences, we must conclude that a different one of these prominence patterns was the input. THREE WESTERN AUSTRONESIAN CASE SYSTEMS IN OT We start with Malagasy, the language that is most straightforwardly handled in the OT approach of Legendre et al. (1993).1 Then we consider the two other languages in turn that are also amenable to the approach but less easily so, Chamorro and Tagalog. They are less amenable because certain assumptions must necessarily be made of a language to characterize it with respect to other languages. These assumptions are debatable and indeed are debated in the literature. For Chamorro, the debate centers primarily on what case marking the agent and patient arguments have, while for Tagalog it centers primarily on what sentence type is to be taken as the basic transitive sentence. Both are crucial to deciding whether a language is ergative or accusative, since it is on the basis of examining the case marking on agents and patients and comparing intransitives with the basic transitive sentence that the distinction is made. Malagasy Accusative Language Tableaux We begin by considering Malagasy, spoken in Madagascar, which is a good exemplar of an accusative language. A basic transitive sentence like (2) illustrates the agent in nominative case, and the patient in accusative case. The Malagasy data are from Pearson (1999) and Keenan (1995).

< previous page

page_160

next page >

< previous page

page_161

next page >

Page 161 (2) Malagasy: Basic Transitives (Agent Topic) a. Mamono ny akoho amin’ny antsy ny mpamboly. AT.kill DET chicken with’ DET knife DET farmer ‘The farmer kills the chicken with a knife.’ b. Nividy ity lobaka ity hoan’dRasoa Rabe. AT.bought this shirt this for’ GEN Rasoa Rabe ‘Rabe bought this shirt for Rasoa.’ In Malagasy, nominative and accusative case marking is not overt on noun phrases, but rather it is indicated by word order and is also apparent in the pronominal system. The nominative case, C1, corresponds to the noun phrase in sentence final position, ny mpamboly in (2a) and Rabe in (2b), while the accusative case, C2, corresponds to the noun phrase in sentence medial position preceding prepositional phrases, ny akoho in (2a) and ity lobaka ity in (2b). Intransitive sentences, such as those in (3a) and (3b), all use C1, nominative case, on the sole argument, regardless of whether that argument is thematically a patient (thus a P input) or an agent (thus an A input). Notice that the arguments are in sentence final position after a prepositional phrase. The outputs for these intransitives are thus A1 and P1, respectively. (3) Malagasy: Intransitives a. Matory ao anatin’ny ala ny gidro. AT.sleep there inside’ DET woods DET lemur ‘The lemur is sleeping in the woods.’ b. Mandeha ao anatin’ny ala ny gidro. AT.went there inside’ DET woods DET lemur ‘The lemur is going into the woods.’ All the forms considered so far carry AT morphology. This contrasts with PT verbal marking on a passive as in (4) (where the verb roots are vono “kill” and vidy “buy”). (4) Malagasy: Passives (Patient Topic) a. Vonoin’ny mpamboly amin’ny antsy ny akoho. PT.kill’DET farmer with’ DET knife DET chicken ‘The chicken is killed by the farmer with a knife.’ b. Novidin-dRabe hoan’dRasoa ity lobaka ity. PT.buy-GEN Rabe for’ GEN Rasoa this shirt this ‘The shirt was bought for Rasoa by Rabe.’

< previous page

page_161

next page >

< previous page

page_162

next page >

Page 162 In these examples, the sentence-final patient is nominative, C1. The agent in these examples is in the genitive case, which we will consider to be an instance of C4 and revisit later. On a common noun like ny mpamboly in (4a), the genitive does not appear overtly, but this noun phrase nevertheless forms a unit with the verb, indicated in the orthography with an apostrophe (n-bonding in Keenan’s 1995 terms). The genitive case is evident on proper names as in dRabe in (4b) and in the pronominal system. As in many typical accusative languages, the output for a passive input aP is a4P1. The output for an antipassive input Ap is A1p4. In an accusative language, this amounts to alternations between transitive and intransitive forms of a given verb. For example, I believe him versus I believe in him or he ate the fruit versus he ate at the fruit or even he ate. Just as antipassive alternations such as these are unmarked on the verb in English, they are also unmarked in Malagasy. The tableaux that illustrate A1, P1, a4P1, and A1p4 as optimal candidates are given in Figure 7–1. These tableaux are similar to those given in Legendre et al. (1993) for typical accusative languages except that only the most relevant portion of the constraint ranking: X = C1 >> x ≠ C12 >> α = C2 is provided. The remaining constraints are not included, as their ranking is not distinguished by our examples. The A1 and P1 outputs are selected by virtue of the high ranking of X = C1 with respect to α = C2, which these outputs violate. The passive output a4P1 and the antipassive output A1p4 are selected over the other candidates due to both the x = C1 and the x ≠ C12 constraints, as illustrated in the tableaux. The constraint violation that rules out a candidate is annotated with an exclamation point. Once the optimal candidate is chosen, the lower ranked constraints may be violated. This is indicated by shading those portions of the tableaux, following standard OT practice. Thus for A and P inputs, the first constraint is sufficient; and for the aP and Ap inputs, the first two constraints in the ranking are sufficient. Violations of α = C2 by the optimal candidates are irrelevant in these tableaux, but we will see that the constraint does play a role in a subsequent tableau. The A1P2 output never violates any of the constraints as discussed by Legendre et al. (1993) and therefore is universally selected on a basic transitive input of AP. For this reason it is not included in any of the tableaux. Circumstantial Topic in Malagasy Next consider the examples of circumstantial topic sentences given in (5). The CT verbal morphology contrasts with the PT and AT morphol-

< previous page

page_162

next page >

< previous page

page_163

next page >

Page 163

Figure 7–1. Malagasy Case Pattern Tableaux. ogy we have seen and consists of a prefix and a suffix with phonological adjustments. (5) Malagasy: Circumstantial Topic a. Amonoan’ny mpamboly ny akoho ny antsy. CT.kill’DET farmer DET chicken DET knife ‘The knife is used by the farmer to kill the chicken.’ b. Nividianan-dRabe ity lobaka ity Rasoa. CT.buy-GEN Rabe this shirt this Rasoa ‘Rasoa was bought this shirt by Rabe.’ In these sentences the agent appears genitively marked, thus in C4. The circumstantial participant, an instrument, ny antsy, or beneficiary, Rasoa, appears in sentence final position and thus in nominative case, C1, and the patient is in accusative case, C2. If the input for the sentence is aPX, then the output will be a4P2X1 under the ranking of constraints we have already applied. The tableau demonstrating this is in Figure 7–2, where there are 13 potential candidates and three constraints are sufficient to select a4P2X1 as optimal among them. For other potential

< previous page

page_163

next page >

< previous page

page_164

next page >

Page 164

Figure 7–2. Malagasy Circumstantial Topic Tableau. prominence combinations in the input, apX, APX, and ApX, the established ranking would result in unobserved case patterns for circumstantial sentences in Malagasy, assuming faithfulness to prominence information in the input. Specifically, apX would result in a4p4X1, APX in A1P2X4, and ApX in either A1p4X2 or A2p4X1. The reason English, which has the same ranking of constraints, does not have a4P2X1 is that it has no CT sentence type with which to express it. Problems with the Account of Malagasy Although Malagasy, as documented here, fits squarely into the accusative case system type, there are some problems with this classification. Keenan (1995) argues for the inadequacy of a relational grammar approach to Malagasy, and some of the same arguments could be levied against the present OT analysis. In summary, these problems are as follows. First, the genitive case should be viewed as a core case, not a lower case like C4, as it is assumed to be here. Having no “demoted” agent in the passive makes Malagasy more like a Philippine-type language, yet the analysis here does not express this. Malagasy is not adequately analyzed as a Philippine-type language like Tagalog is in the section on it, since Malagasy’s AT sentence is readily

< previous page

page_164

next page >

< previous page

page_165

next page >

Page 165 identifiable as a basic transitive, while the PT sentence is derived. Second, the rich verbal morphology cannot be correlated with any aspect of the OT analysis and yet is central. For example, CT verbs carry both PT and AT morphology in Malagasy, and all verb forms carry topic morphology, including basic transitive sentences. These are significant morphosyntactic facts about Malagasy that distinguish it from other languages including Chamorro and Tagalog, and yet they are accidental under the OT approach taken here. The OT approach taken in Sells (2001) directly addresses this problem with Legendre et al. (1993) by having his constraints predict which forms will be morphologically marked. The main prediction made under his assumptions for the present analysis of Malagasy would be that AT sentences are unmarked and PT sentences are marked. This is not the case, but AT sentences are arguably less marked than PT sentences. Chamorro Next, we turn to Chamorro. This Western Austronesian language, spoken in Guam and Saipan, is our exemplar of an ergative Case Pattern. Indeed, Legendre et al. (1993) list it among their many sample languages as ergative with a4P2 passive and A2p4 antipassive outputs, based on Cooreman (1988). Campana (1992) provides a case-checking analysis of Chamorro, and he too finds it to be an ergative language. However, the point made by Chung (1998), that the evidence for Chamorro ergativity is conflicting and complicated, is well taken. Some of the complexities she raises are addressed in the next subsection. Nevertheless, I maintain that there is even more scant evidence that Chamorro is accusative. As complex as it is, the system is best characterized as an ergative language using the tools provided in this OT approach. Basic Sentence Types in Chamorro A basic transitive sentence is given in (6), where the verb forms bear no topic markers. The Chamorro data are taken from Cooreman (1988) and Topping (1973). (6) Chamorro: Basic Transitives (No Topic) a. Si Pedro ha li’e’ i palao’an. UNM Pedro E3SG see UNM.DET woman ‘Pedro saw the woman.’ b. Ha- patek yo’ si Juan. E3SG-hit AlSG UNM John ‘John hit me.’

< previous page

page_165

next page >

< previous page

page_166

next page >

Page 166 c.

Hu li’e’ gue’. E1SG see A3SG ‘I saw him.’ In Chamorro, abstract case is not registered via word order, which can vary, but rather by agreement or by case, which is visible in pronominal forms but not on full NPs. In (6a) there is agreement with the agent and none with the patient. (6b) and (6c) give some of the pronominal forms. If these sentences are compared with intransitive sentences like those in (7), it is clear that the pronominal form of the sole argument of the intransitive matches the form of the patient in the transitive, exhibiting an ergative pattern. (7) Chamorro: Intransitives (Agent Topic) a. Mu nangu yo’ gi tasi. SGswim A1SG LOC sea ‘I swam in the ocean.’ b. Humanao gue’ giya Guam. SG.go A3SG LOC Guam ‘He went to Guam.’ Consistent with this, notice that the pronouns are labeled ergative and absolutive by Cooreman (1988). Thus C1 is associated with triggering agreement or an ergative pronominal, while C2 is associated with an absolutive pronominal. The intransitives in (7a) and (7b) correspond to A1 and P1 outputs, respectively; hence there is no variation in the case of the sole argument across intransitives. Next consider the aP examples in (8). In these aP sentences in Chamorro, there is no agreement with the agent, or any other argument. The agent is marked with oblique case, as Pedro in (8a) and nu guiya in (8b), and hence is a4. The patient is in the absolutive case, hence is P2. I have not labeled this passive since in Chamorro there is an additional sentence type with that label, namely, sentences where the verb is marked with ma- and where the agent is omitted (Topping 1973, p. 257). (8) Chamorro: aP (Patient Topic) a. L-in-i’e’ i palao’an as Pedro. PT-see UNM.DET woman OBL Pedro ‘The woman was seen by Pedro.’ b. L-in-i’e’ yo’ nu guiya. PT-see A1SG OBL EMPH3SG ‘I was seen by him.’

< previous page

page_166

next page >

< previous page

page_167

next page >

Page 167 The antipassive in Chamorro, which is indicated on the verb with the man- prefix, is associated with more than one case pattern, as discussed extensively in Cooreman (1988). Consider the four examples in (9) in which the agent is consistently absolutive A2, but the patient has many different forms. (9) Chamorro: Antipassives (man-) a. Man-hongge hao nu i lahi. APASS-believe A2SG OBL DET man ‘You believe in the man.’ b. Mamaisen gue’ gi patgon nu i kuestion. APASS.ask A3SG LOC child OBL the question ‘He asked the question of the child.’ c. Para baihumam-bisita gi espitat. IRR IRR1SGAPASS-visit LOC hospital ‘I’m going to visit somebody in the hospital.’ d. Mangonne’ guihan i peskadot. APASS.catch fish the fisherman ‘The fisherman caught a fish/fish.’ In (9a) the patient is marked with oblique case. In (9b) it is marked with a different oblique case (referred to in the literature as locative case). In (9c) the patient is omitted completely, and it has an indefinite reading, just as in the English antipassive example He ate. The last example, (9d), also has an indefinite patient, but one that appears in the sentence and is unmarked for case. Because the patient in such sentences must be indefinite, we cannot test the case by substituting a pronoun. The patient in these sentences is not absolutive, since there is already an absolutive in the sentence. It is not ergative, since ergative is associated with agent arguments. Under the assumptions of Legendre et al. (1993) we must consider the patient in (9d) to be an instance of p4 since there is no other way to include such sentences. Thus we will consider all these to be instances of A2p4 and recognize that this is an uncomfortable move in the case of (9d). Agent Topic in Chamorro Before moving on to the OT analysis, there is yet another sentence type to consider in Chamorro: the AT sentence. Notice that our AP sentence in (6) carried no topic marking, whereas in Malagasy the AP sentence carried AT morphology. Chamorro has, in addition to its basic transitive sentence, an AT sentence that is probably also best aligned with the AP sentence type. Examples of AT sentences are given in (10), which can be compared with the unmarked sentences in (6).

< previous page

page_167

next page >

< previous page

page_168

next page >

Page 168 (10) Chamorro: Agent Topic a. Si Juan s-um-angan i estoria. UNM Juan AT-tell UNM.DET story ‘Juan told the story.’ b. Guiya l-um-i’e’ i palao’an. EMPH3SG AT-see UNM.DET woman ‘He saw the woman.’ In (10a) there is no ergative agreement unlike in (6). Furthermore, if the agent is pronominal, as in (10b), it is not ergative, but it is not absolutive either. The pronoun is in the case form known as emphatic. We can take the same approach as we did above for p4 in antipassives and treat the emphatic as a different instantiation of a case, this time C1 case. However, this may be yet another uncomfortable move: While C4 is a catch-all case, C1 is not meant to be. Split Ergativity in Chamorro Chamorro, like Malagasy, has elements of a Philippine-type system but is not considered to be one here. In both these languages, there is a basic transitive sentence type that aligns with AP, whereas in Tagalog, a completely Philippine-type language, we will see that there is not. One final complication to consider is that of split ergativity. As is common in ergative languages, Chamorro has a split system in which one subpart of the language operates as an accusative language. In Chamorro, this subpart corresponds to sentences that are in the irrealis mood. All the examples we have considered, with the exception of (9c), have been in moods other than the irrealis. In the irrealis mood, there is no ergative agreement or ergative pronominal forms; rather, agreement and pronominal form follow a different paradigm and vary with the agent in transitives and the sole argument of intransitives, thereby exhibiting an accusative pattern. Legendre et al. (1993) suggest that splits of this sort might be handled by introducing a more specific version of α = C2 that refers to the required feature. In the case of Chamorro, the specific version of the constraint would be α[–irr] = C2, meaning that when the input sentence is not irrealis, the constraint is adhered to, when the input is irrealis the constraint is ignored. The interaction of the constraints results in the emergence of the accusative pattern when the input is irrealis, since α[–irr] = C2 dominates X = C1 and the ergative pattern emerging otherwise. The tableaux presented in the next subsection illustrate this. Ergative Language Tableaux As an ergative language (outside the irrealis mood), Chamorro has the ranking x ≠ C12 << α[−irr] =C2 << x ≠ C1. This ranking leads to A2 and P2 outputs for the intransitives since α = C2 is highly ranke,d as shown in

< previous page

page_168

next page >

< previous page

page_169

next page >

Page 169 the tableaux in Figure 7–3. These tableaux are analogous to those provided by Legendre et al. (1993) for other typically ergative languages like Eskimo. As an accusative language in the irrealis mood the optimal output candidates are utterly different since the ranking becomes simply x ≠ C12 >> X = C1. The winning candidates are the typically accusative ones: A1, P1, a4P1, and A1P4 (Figure 7–4). Circumstantial Topic in Chamorro Circumstantial topic sentences in Chamorro are signaled by the –i verbal suffix. The agent is ergative in both CT examples in (11) and is thus C1. The patient carries oblique case marking, and thus has abstract case C4. Finally, the circumstantial participant (in [11a] si Juan, a recipient, and in [11b] i guafi, a purpose) is unmarked for case, and thus has absolutive abstract case, C2. (11) Chamorro: Circumstantial Topic a. Si Pedro ha sangan-i si Juan ni estoria. UNM Pedro E3SG tell-CT UNM Juan OBL story ‘Pedro told the story to Juan.’

Figure 7–3. Chamorro Case Pattern Tableaux: Not in the Irrealis.

< previous page

page_169

next page >

< previous page

page_170

next page >

Page 170

Figure 7–4. Chamorro Case Pattern Tableaux: Irrealis. b. Ha utut-i i guafi ni hayu. E3SG cut-CT UNM.DET fire OBL wood ‘He cut the wood for the fire.’ For the input ApX, A1p4X2 is indeed the output candidate selected under the same ranking of constraints established previously with the additional constraints X = C1 and A = C1 coming into play, as the tableau in Figure 7–5 illustrates. Just as with Malagasy, our established constraints together with one particular prominence combination in the input, ApX this time, results in the observed circumstantial sentence case pattern, A1p4X2. Other prominence combinations do not: apX results in a4p4X1, APX results in A1P2X4, and aPX results in a4P2X1. (See the subsection, The Recent Perfective and LowProminence Inputs.) A Problematic Account of Ergative Languages A problem that will remain unresolved here in the Legendre et al. (1993) analysis of an ergative language is that ergative is deemed C1 and absolutive is deemed C2. This means that ergative arguments are considered to be primary and subject-like, whereas absolutive arguments are

< previous page

page_170

next page >

< previous page

page_171

next page >

Page 171

Figure 7–5. Chamorro Circumstantial Topic Tableau. secondary and object-like. This is not so in ergative languages. Analyses from various frameworks, including Murasugi (1992) and Manning and Sag (1998), to name just two, have made the opposite assumption, namely, that absolutive, like nominative, is primary and that ergative, like accusative, is a lower case secondary to it. It becomes clearer that this is a better way to go when syntactic effects are considered. Making such a change to the Legendre et al. (1993) typology would require completely rethinking the constraint set. Tagalog Tagalog, a major language of the Philippines, is a language that is an exemplar of neither ergative nor accusative case patterns. Although it is not a pattern described by Legendre et al. (1993), who do not address where relatively rare Philippine-type languages would fit within their OT approach, Tagalog’s case pattern is readily characterizable with the consideration of an additional constraint regarding prominence that is as intuitive as those already proposed by Legendre et al. (1993). Exceptionally, Tagalog never allows more than one participant to be prominent per sentence. This accounts for the fact that Tagalog fails to have a typical output, as we will see in the next subsection. Contrast this with

< previous page

page_171

next page >

< previous page

page_172

next page >

Page 172 Chamorro, which, as we have seen, has some additional sentence types to consider but which is not lacking any of the output types. In that situation, it was a matter of deciding how the additional output types might be aligned with the input sentences laid out in Legendre et al. (1993). Here is a brief overview of how Tagalog differs from the languages we have considered so far. One claim of this chapter is that in Tagalog, the A1P2 output, thought to be universal, never succeeds. A further claim is that Tagalog has additional outputs not found in the other languages corresponding to a, p, and ap inputs. Another reason for Tagalog’s differing pattern is that it has more than two types of C2 available and these can co-occur. Two Basic Transitives in Tagalog Tagalog is shown to exhibit a hybrid of the ergative and accusative case patterns in Maclachlan (1996), who proposes a case-checking analysis of Tagalog’s hybrid nature. Under that Principles and Parameters analysis, the special nature of the Tagalog case system derives from the exceptional availability of an additional case position. In addition to the usual single case checked in the highest functional projection (corresponding to nominative or absolutive case), there are not one but two other cases available: an ergative case associated with agents and an inherent accusative case associated with patients. We will see how this view can be recast in OT terms. In Tagalog, case markers are overt and distinctive and word order is variable. The pair of transitive sentences in (12) shows the ergative case ng on the agent when the patient is marked with ang (nominative or absolutive case), and the accusative case ng on the patient when the agent is ang marked. I will label these neutrally as NG and ANG accordingly. The Tagalog data are taken from Maclachlan (1996). (12) Tagalog: Basic Transitives (PT and AT) a. lulutu-in ng lalaki ang karne. will.cook-PT NG man ANG meat ‘The man will cook the meat.’ b. mag-luluto ang lalaki ng karne. AT.PAG-will.cook ANG man NG karne ‘The man will cook meat.’ Sentence (12a) is the patient topic sentence, and the verb bears the suffix -in, whereas sentence (12b) is the agent topic sentence, and the verb bears the mag- prefix. I maintain that these two sentence types in Tagalog are both basic transitive sentence types with no oblique case marking, but I also maintain that neither of these sentences has A1P2 as its output.

< previous page

page_172

next page >

< previous page

page_173

next page >

Page 173 Unlike in Chamorro, there is no other transitive sentence type that lacks topic marking on the verb to take as the A1P2 sentence. Unlike in Malagasy, where the AT sentence was the basic transitive, neither of the two sentences bearing topic marking in (12) could plausibly be taken to be the A1P2 basic transitive sentence over the other. Certainly there have been approaches that force one of the two sentences to be a derived sentence type or more marked so as to fit Tagalog into the ergative or accusative classifications, but there is little evidence for doing so (see the discussion in Maclachlan, 1996). Having two basic transitive sentence types makes Tagalog a good exemplar of the Philippine-type system that is neither an ergative nor an accusative system. If both sentences in (12) are A1P2 sentences, then there is an immediate problem in deciding how to classify the language. The first step in categorizing Tagalog as having one of these systems cannot be taken: Determining whether any argument bears C1 or C2 is based on examining the A1P2 sentence. If both sentences in (12) are A1P2, then there would be no consistent C1 and C2. According to (12a), C1 would be realized as ng marking and C2 would be realized as ang marking, whereas according to (12b) the opposite would hold. Rather than assume that both these sentences are A1P2 sentences, I propose that what places Tagalog into a language type apart is that its AP input never results in an A1P2 output. Each of the sentences in (12) is claimed instead to have one high-prominence argument and one-low prominence argument. The input for the AT sentence (12a) is aP, and the input for the PT sentence (12b) is Ap. The aP input does not result in a passive since the output is neither a4P1 nor a4P2, but rather it is a2P1. Exactly parallel to this, the Ap input does not result in outputs A1p4 or A2p4 corresponding to antipassives, but rather the output is A1p2. Now it is possible to identify C1 and C2 cases consistently. C1 is indicated with ang marking, as on the patient in the PT sentence and the agent in the AT sentence, and C2 is indicated with ng marking, as on the agent in the PT sentence and the patient in the AT sentence. Note that the opposite assumption, identifying C1 with NG and C2 with ANG, cannot hold since the A2p1 output is impossible. I propose that the constraint responsible for the failure of the A1P2 output to be optimal is the following: “At most one participant may be prominent in any sentence.” This constraint is not highly ranked in most languages, but it is supreme in Tagalog. One could say it is the “genius” of the language. In OT terms, the restriction is ranked above faithfulness to prominence. On an AP input, only a2P1 or A1p2 can succeed. As a result, for Tagalog speakers, there is a forced choice of highprominence participant in every utterance, unlike in most languages. Put another way, to express basic transitive propositions, there is a choice between

< previous page

page_173

next page >

< previous page

page_174

next page >

Page 174 PT (or a2P1) and AT (or A1p2) forms that cannot be avoided. It is a choice that is different in nature from the more familiar choice between a symmetric prominence transitive sentence and asymmetric prominence passives or antipassives, made by speakers of ergative or accusative languages. Namely, the latter is a choice between a symmetric and an asymmetric prominence input, both of which maintain faithfulness to prominence information: The speaker decides if the participants are roughly equal in prominence or if either A or P is significantly less prominent. The former is a choice between two asymmetric prominence outputs. When the input is asymmetric, the output is faithful, but when the participants are roughly equal in prominence, the only choice is to be unfaithful to this prominence information—the speaker decides which participant will be highlighted. In a similar vein, Sells (2001) suggests that English Ap input is actually expressed as the unfaithful A1P2 output since that is the best possible output for such an input in English. This special characteristic of Tagalog has been variously expressed to me by native speakers of the language. It is captured in terms of OT here as an AP input being coerced into an aP or Ap output due to a highly ranked prominence constraint. Philippine-Type Language Tableaux Before presenting the tableaux for Tagalog, we need to make a further observation. Namely, for the aP and Ap inputs (or AP coercions to them), there are additional output candidates to consider in Tagalog, given that it has two morphologically similar, but functionally distinct, C2 cases: ergative and accusative. In opposition to the identification of ergative as C1 in Chamorro, it is identified as C2 here. Although Legendre et al. (1993) suggest that Universal Grammar normally prevents two instances of a core case from occurring in the same sentence, they do mention that the Lakhota language might violate this restriction. Woolford (2001) places language-specific and case-specific restrictions on the co-occurrence of two identical cases in her OT analysis and observes that such restrictions are analogous to OCP restrictions in phonology. I suspect that the restriction is also violated in languages that allow double accusative marking, like Kinyarwanda, or double nominative marking, like Korean; but such case patterns are beyond the scope of the present chapter, the Woolford (2001) chapter, and the Legendre et al. (1993) article. In the next two subsections, we will see two types of example sentences in Tagalog where both ergative case and accusative case can co-occur: CT sentences and recent perfective transitive sentences. The Tagalog case pattern outputs for intransitive, passive, and antipassive inputs (A, P, aP, and Ap) can be selected as optimal on the basis of two of the constraints proposed in Legendre et al (1993). We will see

< previous page

page_174

next page >

< previous page

page_175

next page >

Page 175 that the high ranking of X = C1 and α ≠ C4 is sufficient. The α ≠ C4 constraint is ranked lowest in all three major rankings considered in Legendre et al. (1993) so as to never have an influence in selecting among candidates in those language types. Grammars with α ≠ C4 highly ranked generate languages that have no passive (neither a4P1 nor a4P2) and no antipassive (neither A1p4 nor A2p4), examples of which are mentioned in Legendre et al. (1993). Turning to other basic sentences in Tagalog, the single argument in intransitive sentences with inputs A or P is uniformly C1. That is, it is marked with the ang case, as the examples in (13) show. (13) Tagalog: Intransitives a. lumakad ang lalaki. AT.walked ANG man ‘The man walked.’ b. ma-tutulog si Ben. PT-will.sleep ANG Ben ‘Ben will sleep.’ There is a correlation between the choice of topic morphology AT, as in (13a), versus PT, as in (13b), and whether the intransitive argument is thematically an agent or a patient. This is apparent on verbs that can be in either A or P sentences and therefore take either type of morphology. For example, the verb upo “to sit” can alternate between um-upo (AT-sat) and ma-upo (PT-sat), where the meaning differs in the degree of agency of the sitter. Thus the difference between A1 and P1 intransitives is monitored in the language, but not in terms of abstract case. Outputs A1 and P1 are ensured if X = C1 is highly ranked. Next consider the aP and Ap inputs. As we have discussed, examples of these are given in (12), and their outputs are a2P1 and A1p2. These are indeed the optimal candidates if α ≠ C4 is highly ranked. The tableaux in Figure 7–6 illustrate this, where the relative order of the two constraints does not matter. Note that candidates with both ergative and accusative cases appearing—namely, a2P2 and A2p2—are included among the candidate sets but are not selected as optimal. Circumstantial Topic in Tagalog Like in Malagasy and Chamorro, there are circumstantial sentences where a circumstantial participant that would normally appear as an oblique is not oblique in Tagalog. Unlike in Malagasy and Chamorro, there are many different types of nonoblique circumstantials distinguished by the shape of the CT morphology on the Tagalog verb. Examples are given in (14).

< previous page

page_175

next page >

< previous page

page_176

next page >

Page 176

Figure 7–6. Tagalog Case Pattern Tableaux. (14) Tagalog: Circumstantial Topics a. i-pag-luluto ng lalaki ng karne ang asawa. CT-PAG-will.cook NG man NG meat ANG spouse ‘The man will cook meat for his wife.’ b. binilhan ng bata ng bigas ang palenke. bought.CT NG child NG rice ANG market ‘The market is where the child bought rice.’ c. ipang-bibili ng bata ng bigas ang pera. CT-will.buy NG child NG rice ANG money ‘The money will be used by the child to buy rice.’ The CT morpheme in (14a) is i- and the beneficiary noun, ang asawa, is in C1. In (14b) it is -an and the location noun, ang palengke, is in C1, while in (14c) it is ipang- and the means noun, ang pera, is in C1. The agent in these sentences appears in ergative case, while the patient appears in

< previous page

page_176

next page >

< previous page

page_177

next page >

Page 177 accusative case, both of which we have deemed to be C2. With an input apX, the optimal output, using the ranking of the two constraints so far established, is a2p2X1. This is illustrated in the tableaux in Figure 7–7 where the candidate set includes six more candidates than those in Figure 7–2 and Figure 7– 5 since for Tagalog, candidates with two instances of C2 must also be considered. Again, there is one prominence pattern in the input that results in the observed a2p2X1 case pattern. The Recent Perfective and Low-Prominence Inputs Next we turn to a set of data that represents an extension of the case patterns we have considered thus far. We have seen that an AP output is never optimal in Tagalog, and now we see that there are a series of inputs

Figure 7–7. Tagalog Circumstantial Topic Tableaux.

< previous page

page_177

next page >

< previous page

page_178

next page >

Page 178 that generally fail in other language types, but succeed here. Namely, I propose that inputs of the form a, p, and ap can be optimal in Tagalog. These occur only in one particular aspectual form, known as the recent perfective (see Schachter and Otanes, 1976). Examples are shown in (15), where (15a) is an intransitive with a thematic agent argument, (15b) is an intransitive with a thematic patient argument, and (15c) is a transitive sentence. (15) Tagalog: Recent Perfective (low prominence) a. kalalakad lang ng bata. RP.walk just NG child ‘The child just walked.’ b. katutulog lang ng bata. RP.sleep just NG child ‘The child just slept. c. kabibili lang ng cloth ni Pedro. RP.buy just NG cloth NG Pedro ‘Pedro just bought cloth.’ While verbal forms are rarely seen without some topic marking in Tagalog, exceptionally recent perfective verb forms never carry topic morphology. Concomitant with this, no noun bears abstract case C1 in these sentences. All the nouns in (15) are C2. As X = C1 is a highly ranked case constraint in Tagalog, these sentences must not involve high-prominence inputs. I propose that the inputs a, p, and ap are violations of another intuitive prominence constraint: “At least one argument is high prominence.” This constraint is rarely violated in languages but is exceptionally ranked low in recent past sentences. Just as the α[–irr] = C2 constraint proposed for Chamorro is contextualized for irrealis mood, the preceding constraint is contextualized for recent perfective aspect. This does not explain adequately the connection between this particular verbal aspect and prominence status, but no theory I have seen can do so. Unfortunately, it is also impossible to show independently that these arguments have lower prominence than arguments in other sentence types. Prominence is taken to be a relative measure within a sentence, not as an absolute measure. We are simply led to these conclusions based on the case pattern exhibited as it compares to the case pattern in other sentence types of the language. The sentences in (15) are thus taken to have a, p, and ap inputs, which are never optimal outside the recent perfective. The grammar selects the attested a2, p2, and a2p2 outputs as optimal if we admit one finer distinction to the constraint system of Legendre et al. (1993). Namely, it

< previous page

page_178

next page >

< previous page

page_179

next page >

Page 179 appears that the two constraints referred to together as x ≠ C12 are dissociated in Tagalog. The constraints are that low-prominence arguments do not appear in C1 or C2 and these could be expressed as x ≠ C1 and x ≠ C2. Now with x ≠ C1 ranked higher than x ≠ C2, we can see in the tableaux in Figure 7–8 that the observed optimal outputs for recent perfective are selected. The x ≠ C2 constraint is presented in the final column for consideration. If the conjoined constraint x ≠ C12 is used, the last two columns would be conflated under it and no output would be selected as optimal. The other constraints are indeterminate or select incorrect outputs and are taken to be ranked below the three highest constraints proposed here. To recap, Tagalog as a hybrid language can be analyzed in terms of the OT constraints of Legendre et al. (1993) as a language with A1P2 and a2P1 and no A1P2 outputs. Furthermore, it is a language that permits two C2 cases per phrase, and it has some rare output types over and above those considered by Legendre et al. (1993). The relative rarity of Philippine-type languages in OT terms is most likely due to the instability associated with no single winning AP output. Some sentence form, likely either aP or Ap, will come to be reanalyzed in the language as being the AP candidate, and the other will come to be reanalyzed as a passive or antipassive, and concomitantly the language will rerank its constraints.

Figure 7–8. Tagalog Recent Perfective Aspect Tableaux.

< previous page

page_179

next page >

< previous page

page_180

next page >

Page 180 The Sells Analysis of Philippine-Type Languages Sells’ (2001) analysis of Philippine-type languages in OT shares with the analysis proposed here the insight that neither AT nor PT sentences are symmetric in prominence, and that neither of these should be considered to involve demotion. His analysis differs from mine in that he proposes an additional level of prominence to handle Philippine-type languages. He considers AT and PT sentences to involve promotion to super prominence of an agent and patient, respectively. Where I have achieved asymmetry in AT and PT sentence types by aligning them with Ap and aP, Sells achieves this asymmetry by aligning them with new types A′P and AP′ (where X′ stands for an even higher level of prominence than X). It is difficult to get evidence for relative prominence of arguments, as Sells points out. He brings up textual evidence from Payne (1994) on Cebuano that suggests that in that language, PT sentences appear to be less textually marked than AT sentences. Sells ties this to the fact that Cebuano PT sentences are also morphologically unmarked and centers his account around these two facts. However, the evidence in Tagalog is equivocal on both these counts (see Maclachlan, 1996). Furthermore, there is no textual evidence to be had regarding the CT or recent perfective sentences types that would help with the current analysis. CONCLUSIONS Now that we have considered all three languages, we can compare the optimal outputs of these languages and what types of sentences those outputs correspond to. This comparison is summarized in Figure 7–9. We note that while in most world languages, including in Malagasy and Chamorro, an AP input corresponds to an optimal A1P2 output, no such symmetric prominence output can be optimal in Tagalog. A further difference is evidenced in the optimality of three low-prominence outputs that Tagalog does not share with the other two languages. We observe further that while all three languages have PT, AT, and CT equivalents, these all differ in terms of the inputs they align with and the outputs they correspond to. PT structures align with aP inputs in all three languages, but their outputs correspond to three distinct passive types outlined in Legendre et al. (1993). In their terminology, Malagasy’s PT is a passive1, Chamorro’s PT is a passive2 and Tagalog’s PT is a reversal, so called because the usual role associations of agent with C1 and patient with C2 are reversed. Similarly, all three antipassive patterns documented by Legendre et al. (1993) are instantiated. Malagasy has antipassive1, Chamorro has antipassive2, and Tagalog has no antipassive. The alignment of AT morphology with input types in the different languages is

< previous page

page_180

next page >

< previous page Page 181 input type

page_181

next page >

Malagasy Chamorro Tagalog topic output topic output topic output intransitive A AT A1 (AT) A2 AT A1 unaccusative P AT P1 (AT) P2 PT P1 transitive AP AT A1P2 Ø or AT A1P1 * * passive aP PT a4P1 PT a4P2 PT a2P1 antipassive Ap AT A1p4 manA2p4 AT A1p2 circumstantial αpX CT a4P2X1 CT A1p4X2 CT a2p2X1 low intransitive a * * * * Ø a2 low unaccusative P * * * * Ø p2 low transitive ap * * * * Ø a2p2 Figure 7–9. Optimal Outputs Compared. especially variable. In Malagasy, AT aligns with both AP and Ap inputs, and it also aligns with all intransitives. In Chamorro, AT aligns with the AP input even though this requires postulating two different realizations of C1 in complementary distribution. It may also be the case that intransitives and antipassives bear AT morphology, but this morphology has been reanalyzed in the language (this morphology is -um- in [7b] and man- in the examples in [9]). Unlike in the other two languages, Chamorro has an AP sentence that is none of PT, AT, or CT. In Tagalog, AT aligns with Ap and with agentive intransitives. Finally, CT aligns with slightly different inputs in the three languages. Although they all have a high-prominence circumstantial participant, the relative prominence of the A and P arguments varies: aPX, ApX, and apX. These particular prominence combinations for CT sentences in the input were arrived at (1) on the basis of knowing the case marking pattern in the CT output, (2) by assuming faithfulness to prominence information, and (3) on the basis of patterns observed in transitive sentences. Presumably, this process is similar to that followed by language learners faced with the same data. While inputs have differed in prominence information, it has been possible to deduce an appropriate input for each language to arrive at each of the three distinct attested outputs: a4P2X1, A1p4X2, and a2p2X1. Malagasy’s CT is built on a passive, Chamorro’s CT is built on an antipassive, and Tagalog’s CT adheres to the exceptional constraint that governs Tagalog by which only one participant can have high prominence per sentence.

< previous page

input

page_181

next page >

< previous page

page_182

next page >

Page 182 The application of an existing OT typology in an Austronesian context has brought up certain problems that remain unsolved (which is not to say that they cannot be solved). One problem was calling a case C4 when really it is on a par with other core cases. This arose in Malagasy with respect to genitive case and in Chamorro with respect to one subtype of antipassive (see [9d]), which appeared to have two core case-marked arguments where other types of antipassive in (9) bearing the same antipassive morphology clearly have just one. Another problem was the fact that the ubiquitous verbal morphology in Western Austronesian languages is not correlated with anything in the OT analysis. This was noted for Malagasy following observations by Keenan (1995) but is equally true in the other languages. For example, the approach as it stands cannot distinguish between unmarked (6) and AT transitive forms (10) in Chamorro, classing them both as AP even though they differ in case pattern. A final problem raised but left unresolved is that of patients not being object-like in ergative languages but treated as such here. On the positive side, some potentially problematic issues raised by this application of an existing OT analysis to Western Austronesian languages were resolved. Circumstantial topic sentences, a feature of Western Austronesian languages that is relatively uncommon, were handled for all three languages. The aspectual ergative split in Chamorro proved to be easy to handle along the lines briefly suggested in Legendre et al. (1993). The three sentence types in Tagalog in which C2 was the preferred case were readily analyzed as rare outputs in which all arguments were low in prominence. Finally, and perhaps most impressively, the elusive Philippine-type language represented by Tagalog was incorporated into the Legendre et al. (1993) ergative/accusative typology almost exclusively using their constraint set. Thus their typology was easily extended in the necessary ways to new sentence types and new language types. The necessary extensions were found to be the following faithfulness to prominence information constraints: 1. A non-core argument cannot have high prominence. 2. At most one participant may be prominent in any sentence. 3. At least one argument is high prominence. Let us say that constraint (1) is almost universally ranked high, but all three Western Austronesian languages considered here have a mechanism to circumvent it by making a non-core argument into a core argument, namely the CT sentence. Constraint (2) is almost universally ranked low in languages of the world but is high in Tagalog, setting this language and other Philippine-type languages apart from Malagasy and Chamorro in the relevant respects. Finally, constraint (3) is almost

< previous page

page_182

next page >

< previous page

page_183

next page >

Page 183 universally highly ranked but is violated in just the recent perfective aspect in Tagalog. NOTE 1. In presenting the Austronesian data in the next few sections, I have kept as close as possible to the glossing given in the original sources but have standardized the topic marking as follows. I have used patient topic (PT), agent topic (AT), and circumstantial topic (CT) throughout where Pearson (1999) uses accusative topic, nominative topic, and circumstantial topic; Keenan (1995) uses active verb, passive verb, and circumstantial verb; Topping (1973) uses Goal focus, Actor focus, and Referential focus; and Maclachlan (1996) uses patient topic, agent topic, and various more specific names for circumstantial topic (e.g., benefactive topic and locative topic). Other abbreiviations used in this chapter include: DET: determiner, GEN: genitive case, UNM: unmarked, A: absolutive, E: ergative, I: first person, 3: third person, SG: singular, EMPH: emphatic case, APASS: antipassive, IRR: irrealis, and RP: recent perfective. REFERENCES Campana, M. 1992. A Movement Theory of Ergativity. Ph.D. dissertation, McGill University. Carrier-Duncan, J. 1985. Linking of thematic roles in derivational word formation. Linguistic Inquiry 16:1–34. Chung, S. 1998. The Design of Agreement: Evidence from Chamorro. Chicago: University of Chicago Press. Cooreman, A. 1988. The Antipassive in Chamorro: Variations on the theme of transitivity. In Masayoshi Shibatani (ed.) Passive and Voice. Amsterdam: John Benjamins, 561–93. Dixon, R.M.W. 1994. Ergativity. Cambridge, U.K.: Cambridge University Press. Keenan, E. 1995. Predicate argument structure in Malagasy. In C.Burgess, K. Dziwirek, and D.Gerdts (eds.) Grammatical Relations: Theoretical Approaches to Empirical Questions. Palo Alto: CSLI, 171–216. Legendre, G., Raymond, W., and Smolensky, P. 1993. An Optimality-Theoretic typology of Case and grammatical voice systems. In Proceedings of the 19th Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, University of California, Berkeley, 464–78. Maclachlan, A. 1996. Aspects of Ergativity in Tagalog. Ph.D. dissertation, McGill University . Manning, C., and Sag, I. 1998. Argument structure, valence and binding. Nordic Journal of Linguistics 21:107–44. Murasugi, K.G. 1992. Crossing and Nested Paths: NP Movement in Accusative and Ergative Languages. Ph.D. dissertation, Massachusetts Institute of Technology. Payne, T. 1994. The pragmatics of voice in a Philippine language: Actor-focus and goal-focus in Cebuano narrative. In T.Givon (ed.) Voice and Inversion, Amsterdam: John Benjamins. Pearson, M. 1999. Tense-Marking on Malagasy PPs: An Argument/Non-argument Asymmetry. Paper presented at AFLA VI, Toronto.

< previous page

page_183

next page >

< previous page

page_184

next page >

Page 184 Schachter, P., and Otanes, F.T. 1976. Tagalog Reference Grammar. Berkeley: University of California Press. Sells, P. 2001. Form and function in the typology of grammatical voice systems. In G.Legendre, J.Grimshaw, and S.Vikner (eds.) Optimality-Theoretic Syntax, Cambridge: MIT Press. Topping, D. 1973. Chamorro Reference Grammar. Honolulu: The University Press of Hawaii. Woolford, E. 2001. Case patterns. In G.Legendre, J.Grimshaw, and S.Vikner (eds.) Optimality-Theoretic Syntax, Cambridge, MA: MIT Press. Yip, M., Mailing, J., and Jackendoff, R. 1987. Case in Tiers. Language 63:217–50.

< previous page

page_184

next page >

< previous page

page_185

next page >

Page 185 Affixes, Clitics, and Bantu Morphosyntax Sam Mchombo Conventional analyses of the morphological structure of the verb in Chichewa (and other Bantu languages) have the verb comprising a verb root (VR) to which such verbal extensions as the causative, applicative, reciprocal, passive, and so on, are suffixed, and to which prefixes are added. The latter encode information pertaining to agreement with the subject and object(s) of the verb, tense/aspect, negation, modality, and so on. The morphology of suffixation and prefixation to the VR is so well established as to require no further comment. Consider the following example from Chichewa: (1) Mkángo u-da-ómb-án-íts-á alenje ndí asodzi. 3-lion 3sM-pst-hit-recip-caus-fV 2-hunter and 2-fisherman ‘The lion made the hunters and the fishermen hit each other.’1 In this, the VR -omb- “hit” supports the extensions -an- “recip” and -its- “caus,” as well as the prefixes u “subject marker” and da “past tense.” The question of whether the “prefixed” material and the extensions have comparable status as affixes does, however, merit discussion. In this chapter I examine the affix status of these morphological elements, arguing for the analysis of the prefixal elements as “clitics” in Chichewa. The chapter advocates a specific conception of the morphological structure of the verb in Chichewa, a conception argued for in previous work (Mchombo, 1999a). The claim will be that the proposed structural organization of the verb contributes toward an explanation of the formal and functional differences between the extensions and the “prefixes.”

< previous page

page_185

next page >

< previous page

page_186

next page >

Page 186 MORPHOLEXICAL VERSUS MORPHOSYNTACTIC PROCESSES The chapter has both narrow and broad objectives. The narrow objective is to determine the status of pre-Verb Stem and post-Verb Root elements in the morphological structure of the verb in Chichewa (extendable to other Bantu languages; cf. Hyman, 1991). The broad objectives relate to the distinction between morpholexical and morphosyntactic processes and the support that the morphological analysis of the verb in Bantu lends to the architecture of Universal Grammar (UG) advocated within the theory of Lexical-Functional Grammar (LFG). The study will begin with the narrow concerns, relating to the distinction between verbal suffixes (extensions) and the verbal prefixes. I will prejudice the issue by referring to the elements prefixed to the verb stem as “clitics.” The ensuing account will highlight the substantive issues underlying this terminological preference. Clitics are generally characterized as independent syntactic constituents that appear phonologically as part of a host word, an attachment characterized by Marantz as “morphological merger,” in which “an independent syntactic constituent shows up phonologically as part of a derived word” (Marantz, 1988:253). As independent syntactic constituents, clitic nodes are (base-) generated in structural positions away from the host to which they get attached, but lack “phonological independence.” In other words, clitics are syntactic words but not phonologically independent units, perhaps due to failure to satisfy minimality conditions (Jackendoff, p.c.). Obvious examples of cliticization are evident in Chichewa, as shown in the following data: (2) a. Mbidzí izi ndi zá 10-zebra 10prox.dem be 10SM-assoc.mark nzēru. 10-intelligence ‘These zebras are intelligent.’ b. Mbidzízi ndi zá nzēru. 10-zebra clitic c cop 10SM-assoc.mark 10-intelligence ‘These zebras (the ones already introduced) are intelligent.’ (3) Mbidzí izo zí-má-dy-á 10-zebra 10-dist.dem 10SM-pstprog-eat-fv maūngu. 6-pumpkin ‘Those zebras were eating pumpkins.’

< previous page

page_186

next page >

< previous page

page_187

next page >

Page 187 (4) Mkángo u-ma-fúná ku-séwéra ndí mbîdzi koma 3-lion 3SM-pstpro-want inf-play with 10-zebra but mbidzízo zi-na-tháwa chifukwá chá mântha. 10-zebra clt 10SM-pst-escape 7-reason 7SM-asso.cmark 6-fear ‘The lion wanted to play with the zebras but those zebras run away because of fear.’ (5) a. Mikángó i-sanu iyo i-na-lówá m’phanga. 4-lion 4SM-five 4distdem 4SM-pst-enter 16–5-cave ‘Those five lions entered the cave.’ b. Mikángó i-sanu-yo i-na-lówá m’phanga. 4-lion 4SM-five-clt 4SM-pst-enter 1 6–5-cave ‘Those five lions (the ones just referred to) entered the cave.’ (6) a. Mikángó isanu yóchénjéra iyo 4-lion 4-five 4SM-clever 4distdem i-na-lówá m’phanga. 4SM-pst-enter 16–5-cave ‘Those five clever lions entered the cave.’ b. Mikángó isanu yóchénjéra-yo i-na-lówá m’phanga. 4-lion 4-five 4SM-clever-clt 4SM-pst-enter 16–5-cave ‘Those five clever lions (the ones just referred to) entered the cave.’ In these examples, the syntactically independent demonstratives, izi “these,” izo “those,” for Class 10 nouns, as well as iyi “these,” and iyo “those” for Class 4 nouns, have reduced counterparts, zi, zo, yi, and yo, with a more anaphoric usage, which do not appear as independent words. Within a Noun Phrase (NP) configuration the reduced forms appear attached to the immediately preceding full word. These reduced forms, whether attached to NPs or to other syntactic classes, in classical suffixation manner, are called enclitics (cf. Halpern, 1998; Poulos, 1990). As syntactically independent but phonologically bound elements, “clitics involve a mismatch between the bracketing or structure motivated on semantic and syntactic grounds and the bracketing or structure motivated on phonological grounds.” (Marantz, 1988). Such mismatches are evident in Chichewa. Consider these data from textual material: (7) Nkhani ya tsikuliyi idawonjezera kusafunika kwa mkuluyu mumtima mwa Sautso. ‘This story of this day increased the undesirability of this guy to Sautso.’

< previous page

page_187

next page >

< previous page

page_188

next page >

Page 188 In this passage, the word in boldface, “tsikuliyi,” has two clitics, li and yi, reduced forms of the demonstratives ili, and iyi “this” for Classes 5 and 9, respectively. The NP structure underlying this word is: [[Nkhani [ya [tsiku ili]]] iyi] [[9-story [9SM-assoc.mark 5-day 5proxdem] 9proxdem] ‘This story of this day’ Although the clitic yi, by class agreement, is associated with nkhani “story,” it is attached to the full word on its left, which also hosts the other clitic, li, which, by agreement too, goes with tsiku “day.” Following are more instances of cliticization in Chichewa, typical of Bantu languages: (8) Mikángó isanu yóchénjéráyo inalówánsó mu ndênde ndipó ikusákásákábé afísi amené ánábá nyama. ‘Those five clever lions also entered the jail and they are still searching for the hyenas which stole the meat.’ The clitic nso “too, also, again, as well” can be attached to a number of hosts, as exemplified by the following: (9) Mkángo ndi nyama yá mthengo kománsó físi nayénso ndi nyama yá mthengonso. ‘The lion is a wild animal but also the hyena, it, too, is a wild animal as well.’ (10) Kalulu wabwera ndí bowa, iwénso wabweránso ndí zókházókházo! ‘The hare has brought mushrooms and you too have also come with that same stuff!’ The clitic be “still,” while most commonly attached to verbs, can also appear with nominal hosts. Note the following: (11) Mtsogoleri akufufuzabe zá momwé ángáthandizíré asilikári. ‘The leader is still investigating ways in which s/he can assist the soldiers.’ (12) Ngakhálé munyinyirike, iyé tsópano ndi mtsogoleri wánúbe. ‘Although you may grumble, s/he still remains your leader.’ VERBAL SUFFIXATION Bantu languages abound in verbal suffixes, bound morphemes that, in general, are involved in the determination of expressible NP argu-

< previous page

page_188

next page >

< previous page

page_189

next page >

Page 189 ments within the sentence. The suffixes, also known as extensions (Dembetembe, 1987; Dlayedwa in prep.; Du Plessis and Visser, 1992; Guthrie, 1962; Katupha, 1991; Mchombo, 1993a; Ngunga, 1997; Rugemalira, 1993; Satyo, 1985), include the morphology for encoding, inter alia, the causative, applicative, stative (or neuter), reciprocal, passive, as well as reversive, contactive and, positional (these latter far less productive now and, in many cases, fused to their host roots). These, together with the verb root, and terminated by the final vowel (fv) [a], constitute the Verb Stem (VS). Constructions involving some of these are given here: (13) Mkángo u-na-thyól-á mpánda. 3-lion 3SM-pst-break-fv 3-fence ‘The lion broke the fence.’ (14) Mkángo unathyóléts-á mbidzí mpánda. 3-lion 3SM-pst-break-caus-fv 10-zebra 3-fence ‘The lion made the zebras break the fence.’ (15) Mkángo u-na-thyól-ér-á mbidzí mpánda. 3-lion 3SM-pst-break-appl-fv 10-zebra 3-fence ‘The lion broke the fence for the zebras.’ (16) Mkángo u-na-thyól-éts-ér-á mbidzí mpánda kwá alenje. 3-lion 3SM-pst-break-caus-appl-fv 10-zebra 3-fence by 2-hunter ‘The lion made the hunters break the fence for the zebras.’ (17) Mbîdzi zi-na-thyól-éts-ér-édw-á mpánda kwá 10-zebra a 10SM-pst-break-caus-appl-pass-fv 3-fence by alenje (ndí mkángo). 2-hunter (by 3-lion) ‘The zebras got the fence broken for (them) by the hunters at the instigation of the lion.’ In these examples, the verb thyola “break” has been extended to form the causative thyoletsa “make break,” the applicative thyolera “break for,” the applicative of a causative thyoletsera “get something broken for someone,” and the passive of the applicativised causative thyoletseredwa “have something broken for one by.” There are morphotactic constraints on the verb stem in Bantu languages (Hyman and Mchombo, 1992; Machobane, 1993; Ngunga, 1997), but those are not reviewed in this study. Verbal extensions are involved in the determination of the argument structure of the predicate (Abasheikh, 1978; Bokamba, 1981; Hoffman, 1991; Machobane, 1989; Mchombo, 1997b; Ngunga, 1997; Satyo, 1985; Simango, 1995). When the clitics noted above appear together with the

< previous page

page_189

next page >

< previous page

page_190

next page >

Page 190 verbal extensions, the clitics are attached outside of the final vowel. This is demonstrated in (18)–(22) below: (18) Mkángou-ku-thyól-á-nsó mipando. 3-lion 3SM-pres-break-fv-too 4-chair ‘The lion is breaking the chairs too (as well).’ (19) Mkángou-ku-thyól-á-bé mipando. 3-lion 3SM-pres-break-fv-still 4-chair ‘The lion is still breaking the chairs.’ (20) Mkángou-ku-thyól-éts-ér-á-bé kalulú mipando kwá 3-lion 3SM-pres-break-caus-appl-fv- la-hare 4-chair by still físi. la-hyena ‘The lion is still making the hyena break the chairs for the hare.’ (21) Mipando i-ku-thyól-édw-á-nso. 4-chair 4SM-pres-break-pass-fv-too (as well) ‘The chairs are getting broken too.’ (22) Njovu zi-ku-thyól-éts-ér-án-á-nso mipando. 10-elephant 10SM-pres-break-caus-appl-recip-fv-too 4-chair ‘The elephants are also getting the chairs broken for each other.’ The relative order of the extensions and the clitics is significant. The extensions appear to be more intimately connected to the host VR, indicated by the fact that the VS is the domain of a number of linguistic processes whose influence does not extend to the suffixed clitics. For instance, there is vowel harmony in Chichewa (Mtenje, 1985), Luganda (Katamba, 1984), and many other Bantu languages, but its domain is exclusively the VS. Further, there are nominalizations that, again, only target the VS (Mchombo, 1993b, 1999a). Consider the following: (23) kónda ‘love’ kóndána ‘love each other’ chikondano ‘mutual love’ onetsera ‘demonstrate’ chionetsero ‘exhibition’ kodza ‘urinate’ kodzera ‘urinate with’ chikodzero ‘bladder’ ongola ‘straighten’ ongolera ‘straighten with’ chiongolero ‘steering wheel’ tenga ‘take’ tengana ‘take each other’ mténgáno ‘death pact’

< previous page

page_190

next page >

< previous page

page_191

next page >

Page 191 ononga ‘damage’ onongéka ‘get damaged’ chionongeko ‘destruction’ senda ‘skin’ sendédwa ‘be skinned’ kasendedwe ‘manner of skinning’ da ‘hate’ dana ‘hate each other’ mdani ‘enemy’ In addition, in Chichewa, verbal reduplication, like vowel harmony, is confined to the VS, excluding the suffixed clitics. This is illustrated here, with the relevant reduplications in boldface: (24) Chinkokomo chija chidakúlírákulirábe. ‘The rumbling still steadily increased in loudness.’ (25) Ona zinthu zikunka zíkuyípíráyipiratu. ‘Look, things are progressively worsening.’ (26) Agnes ndí máyí aké ómwe adasángálala ndí mbíríyi pozíndíkira kutí mwína kupyólérá mu zókámbákambazi zinthu zíkhózá kudzángothérá momwémo basi. ‘Agnes and her mother were happy with this spreading story recognizing that probably through these incessant allegations things might indeed lead to the desired conclusion.’ (27) Masó áwo ákupónyáponyábe apa ndí apo m’chipíríngú chá anthu pa dépotípo, ine ndidafíká pomwé pánálí atáte. ‘Their eyes still darting back and forth, checking every which way in that crowd at the depot, I approached the place where my father was.’ In these examples the enclitics be in chidakúlírákulirábe and ákupónyáponyábe, tu in zíkuyípíráyipiratu, as well as zi in zókámbákambazi, are excluded from reduplication. The clitics are, thus, treated comparably to the material prefixed to the VS, but differently from the verbal extensions, which include the reciprocal -an- which, like other extensions, is subject to processes whose domain is the VS. The reciprocal participates in reduplication, as shown next, the relevant word in boldface: (28) Máká ádámúnyánsá adalí máséwéro áke omakóndá kugwíránágwirana. ‘What repulsed her particularly was his casual manner of wanting them to fondle each other.’ In brief, the VS constitutes the domain of significant linguistic processes that do not apply to material outside of that domain.

< previous page

page_191

next page >

< previous page

page_192

next page >

Page 192 VERBAL PREFIXATION The Verbal Root in Bantu supports a number of extensions and a number of “prefixes.” The morphemes that appear as verbal prefixes in Chichewa include the morphology for Tense/Aspect, Subject and Object (Agreement) Markers, Directional morphemes, Conditional marker, Modal elements, as well as markers for Negation (NEG). Other languages may also include a focus particle. A rather complex instance of the morphemic organization is exemplified by the following expression: (29) Sí-ú-kú-ká-ngo-zí-thyóktseránso mipando. Neg-3SM-pres-go-just-10OM-break-caus-appl-too 4-chair ‘It is not just going to have the chairs broken for them as well (too).’ The prefixed material is exempt from the types of nominalization indicated earlier,2 as well as from vowel harmony, reduplication, or involvement with argument structure changing.3 In Chichewa the VS has lexical integrity, constituting a significant subdomain in the morphological structure of the verb. In previous work the structural representation of the verb in Chichewa has been proposed as shown in Figure 8–1.4 Considerations internal to the functioning of linguistic processes and of factors external to, but interacting with or having effects on, the language faculty seem to support such a structural representation. Consider the internal factors. It has been argued that the verbal extensions are categorically verbs. The arguments rely, in part, on the analysis of affixes as heads (Marantz, 1984) and, in part, on the endocentricity of the word (Mchombo, 1999b). The analysis derives extra credence from diachrony. According to Givón, verbal extensions in Bantu originated as verbs that participated in serial verb constructions, common in Kwa languages of West Africa, to which the Bantu languages are related. In Bantu these verbs under-went morphologization, reducing them to affixes. Givón points out, further, that some of the verbal prefixes may also have originated as verbs. To account for the constituency of the VS, he advances the hypothesis that their conversion to prefixes, or their morphologization, occurred at a much later stage. Givón hypothesizes that “[T]here has obviously occurred a considerable time lag between the conversion of the main verbs onto verb suffixes and that of the other main verbs into verb prefixes” (Givón, 1971:157). The diachronic account is intended to explain involvement of the verbal suffixes in lexical derivation, deriving the VS. The prefixal material, on the other hand, does not participate in lexical derivation. The diachronic account of the differential behavior of verbal suffixes and verbal prefixes is intended to provide a basis for the empirical fact that there is motivation to the verbal organization

< previous page

page_192

next page >

< previous page

page_193

next page >

Page 193

Figure 8–1. given previously, at least in its identification of the VS as a significant subdomain. The proposal made here is that the prefixal material is not affixal in a sense comparable to that applicable to verbal extensions. Rather, the prefixal elements are clitics, categorically non-verbal, and more involved in the morphosyntactic organization as opposed to the morpholexical aspects characterizing the extensions. AFFIXES VERSUS CLITICS Affixes are bound morphemes, constituents of words that occur attached to some host that may itself be free or bound. Haverkort notes that “affixes exhibit a high degree of selection with respect to their stems” (Haverkort, 1993:115). Clitics, on the other hand, while syntactically independent, are phonologically bound. They appear “prosodically bound to an adjacent word” (Halpern, 1998:104). Further, they exhibit a low degree of selection with respect to their hosts. The property of being bound that characterizes affixes is relevant to the study of pre-verb stem elements and verbal suffixes (extensions) in Chichewa. The extensions only select verb roots, are morphologically bound, and, at the level of phonology, do not satisfy the basic syllable structure or phonotactics of the language. The causative, applicative, passive, stative, and recipro-

< previous page

page_193

next page >

< previous page

page_194

next page >

Page 194 cal extensions in Chichewa are (ignoring variants determined by vowel harmony), respectively, -its-, -il-, -idw-, -ik-, and -an- . Notably, all these have a phonological structure of -VC-, departing from the basic syllable structure of the language. In a study of syllable structure of Chichewa, Mtenje argued that “Chichewa syllables are essentially of a CV nature” (1980:2). In this work Mtenje argues for the functional unity or conspiracy of phonological rules in Chichewa, which, while independently required, appear to be “primarily motivated by the need to destroy vowel sequences and to reserve the canonical syllable type (CV)” (ibid., 18). The Verb Stem in Chichewa, courtesy of the final vowel, satisfies the syllable structure constraints of the language, but still constitutes a domain with notable nonisomorphism between morphological and phonological structure. Interestingly, the clitics that appear after the final vowel, attached to the verb stem, invariably have the required phonological structure of CV. Thus, the enclitics be, tu, di, nso, and so on, all conform to the syllable structure requirements of the language. In this regard, the elements prefixed to the Verb Stem, morphemes that have to do with such “functional projections” as NEG, SM, Tense/Aspects, OM, and so on, share with the enclitics this conformity to the syllable structure constraints of the language. They all have CV organization. The similarity in phonological structure could, arguably, be accidental and of no material import. However, the prefixal material and the enclitics also share the property of being, distributionally, outside the VS and, inter alia, not involved in argument structure changing, as well as the host of linguistic processes whose domain is the VS. In brief, the similarity may derive from shared fundamental properties. The conclusion drawn is that the prefixal material in the verb structure, treated as functional projections within Principles and Parameters Theory (Chomsky, 1991; Chomsky and Lasnik, 1993; Laka, 1994) lying outside the lexical domain (Pollock, 1988), are clitics, oriented more toward syntactic than lexical organization. In brief, the VS in Chichewa constitutes a structural domain separating morpholexical from morphosyntactic operations (cf. Sadler and Spencer, 1998, for further comment on morpholexical versus morphosyntactic distinction). This structural analysis of Chichewa has also been argued for within an autosegmental and Lexical Phonology-theoretic framework (Mtenje, 1985). CLITICS AND INFLECTIONAL MORPHOLOGY The position adopted here is reminiscent of the traditional distinction between inflectional and derivational morphology (cf. Anderson, 1988), apparently supporting the split morphology hypothesis (Perlmutter, 1988). The specialization of the VS to lexical processes while the morphology outside of that domain is oriented toward the non-

< previous page

page_194

next page >

< previous page

page_195

next page >

Page 195 lexical aspects of linguistic structure is underscored by the tendency within various theoretical frameworks to, at the minimum, acknowledge the special status of the VS. For instance, Givón, pursuing his analysis within the framework of the “Standard Theory” version of Transformational Grammar (Chomsky, 1965), derived the VS through applications of the transformational rule of predicate raising, a rule that affected the verb of an embedded sentence and “raised and adjoined to the one from the higher sentence—which then becomes, in Bantu, a suffix. This format is at present controversial, but regardless of one’s position with regard to it, I would like to claim that the semantic facts seem to suggest a verbal origin of the Bantu suffixes” (op. cit., 152). The rule of predicate raising was eventually analyzed as precyclic, and precyclic transformations were claimed to be presyntactic (Newmeyer, 1974, 1975). The VS was, thus, not syntactically derived. Of course Givón also attributed the integrity of the VS to aspects of diachrony, as noted previously. The analysis offered by Givón has echoes in work within PPT. Baker proposes to account for the creation of the VS through the process of Incorporation, an aspect of Head Movement, constrained by the Head Movement Constraint (Baker, 1988; Marantz, 1988; Sproat, 1988; Travis, 1984). In the incorporation analysis there is either (head-) movement of the affixal heads of Causative, Applicative, and so on, to the predicate of the higher or matrix clause, as in the case of aplicativization or, as with the causative, it may itself be the higher (affixal) predicate CAUSE (its- in Chichewa) to which the predicate of the embedded proposition is moved (cf. Abasheikh, 1978). Either way the derivation involves predicate raising. The incorporation analysis of the formation of VS, much like the predicate raising approach of Givón, of which it is a variant in relevant respects, reduces to the recognition of the VS as a domain of lexical processes. Note that the rule of predicate raising is not implicated in the morphologization of the material outside the VS.5 In the Lexical Syntax approach of Hale and Keyser (1992), extended to Chichewa (and Bantu) by Hoffman, the VS is the domain of 1-syntax (lexical-syntax) processes. In Lexical Phonology and Morphology-theoretic approaches or Level-ordered morphology (Kiparsky 1982a, 1982b; Mohanan 1986; Mtenje 1986a), the VS constitutes the domain of level 1 rules, with the clitics coming in at level 2 (or latter) processes. This included the analysis adopted by Klavans (1983), whose distinction between lexical clitics and postlexical clitics, interacting in a manner consistent with the stipulations and organization of level-ordered morphology, reduces to no more than terminological innovation or preference. In brief, within every framework, various processes converge on a unit comparable to the VS in Chichewa, giving it a distinctive character that argues for its uniqueness or lexical integrity. The inescapable con-

< previous page

page_195

next page >

< previous page

page_196

next page >

Page 196 clusion is that the prefixal material is more involved in the morphosyntactic organization of the expression. In fact, Baker remarks, with regard to Mohawk, that “pronominal morphology is clearly a sub-type of inflectional morphology, whereas noun incorporation is more similar to derivational morphology” (Baker, 1990:27). Does this suggest that clitics should be reduced to inflectional morphemes? The identification of clitics with inflectional morphemes is a vexed question resolved in part by the theories of inflectional morphology adopted and in part by the specific views about the nature of clitics in linguistic theory. Everett (1989) argues that all clitics and inflectional affixes are syntactically the same type of element, differing only in their requirements for morphological or phonological support. A reviewer pointed out that clitic status is based on the clitics’ “freedom in what syntactic category they attach to since their relationship to their host is not determined by syntactic dependence.” Comparable arguments, the reviewer points out further, cannot be made for the morphemes discussed here. The conclusion supported is that the morphemes in question remain inflectional, thereby sustaining the tradition in African linguistic scholarship, which has normally recognized inflectional and derivational morphology in the Bantu verbal morphosyntax. The reviewer specifically points out that “[u]nless there is empirical reason to call the Chichewa subject marker, object marker, etc., proclitic, I would recommend avoiding calling them so.” While this issue will remain unresolved here, it should be noted that the reviewer’s objections are easily undermined by the application of the label clitic to comparable pronominal arguments (which, in fact, the SM and OM are in Chichewa and many other Bantu languages [cf. Bresnan and Mchombo, 1987; Demuth and Johnson, 1989]) in, for instance, Romance languages (cf. Jaeggli, 1982; Montalbetti, 1981). Their restricted distribution, occurring within the verbal organization, derives from their status as arguments of a predicator, invoking ideas of locality. In Chichewa the question of whether prefixal elements are best analyzed as clitics or as inflectional morphemes thus reduces to terminological preference (Roberge, 1990), purported departure from tradition notwithstanding. The real issue is that the morphological structure of the verb in Chichewa makes for a principled distinction between the domain characterized by the VS and the larger structure in which the VS is embedded. This work maintains that the elements prosodically associated with, but not contained within, the VS, are clitics. Those that attach to, or form a prosodic unit with, a host on their right are proclitics. These include the SM and OM, elements analyzed as incorporated pronominal arguments in Bantu (Bentley, 1994; Bresnan and Mchombo, 1986, 1987; Demuth and Johnson, 1989; Reach, 1995).This would appear to assimilate these languages to the

< previous page

page_196

next page >

< previous page

page_197

next page >

Page 197 pronominal argument hypothesis, thereby accounting for aspects of word order (Jelinek, 1984; Neale, 1996). The argument status of the OM is underscored by its complementarity with overt nonclitic arguments in some Bantu languages, such as in Gikuyu (Bergvall, 1985; Mugane, 1997). In Chichewa, and Setswana, although the OM can co-occur with an overt nonclitic argument (clitic doubling), the nonclitic argument is prosodically marked, by means of, inter alia, tone, as outside the Verb Phrase (VP) configuration. In other words, the nonclitic argument is licensed by discourse structure rather than syntactic organization, in the narrow sense, which excludes such discourse notions as Topic or Focus (cf. Bresnan and Mchombo, 1987; Demuth and Johnson, 1989). Assuming the correctness of this analysis, does the fact that affixation is involved in the derivation of VS, but cliticization holds of the morphology outside of the VS in Chichewa, have any bearing on the architecture of UG? Further, is there external evidence bearing on the distinction? ON THE ARCHITECTURE OF UNIVERSAL GRAMMAR The facts reviewed in this work are susceptible to different analyses, depending on the theoretical model adopted and the machinery provided by the specific theory of grammar. This chapter advocates the architecture of UG given within the theory of LFG. The basic view advanced within LFG is that there is factorization of syntactic representation of sentences into parallel structures, each with its primitives, organizing rules, and principles. The parallel structures include the overt organization of the sentences, the constituent or categorical structure (c-s), modeled by phrase structure and/or word structure, capturing the configurationality/nonconfigurationality parameter; the functional structure (f-s), which spells out the internal organization of language required for semantic interpretation, normally represented in the format of an attribute-value matrix, and argument structure (a-s). Each level models different dimensions of grammatical structure: category, function, and role. Further, “each level has its own distinctive prominence relations, characterizing the logical subject, the functional subject, and the structural subject. The levels are associated by principles of functional correspondence (also called ‘linking’ or mapping principles)” (Bresnan, 1994:74). Argument structure provides the interface between lexical semantics and syntactic structure, encoding lexical information about the number of arguments and their syntactic types, and their hierarchical organization necessary for the mapping to syntactic structure. F-s provides the interface between that and c-s. Figure 8–2 shows a

< previous page

page_197

next page >

< previous page

page_198

next page >

Page 198 simplified version of the structural representation of an English sentence, where the PREDICATOR “read” has a particular argument structure and is the locus of linking of thematic roles with grammatical functions. Within the LFG conception, a grammar defines a mapping μ from c-structure nodes to f-structures, where μ is mathematically a function, mapping each c-structure node onto a unique f-structure (for technical details about the architecture of LFG, see Bresnan, 2001; Nordlinger, 1998, and references cited there). The significant aspect of the proposed architecture of UG, ignoring details about discourse structure (ds), semantic structure (σ-s), as well as prosodic structure (p-s), their linkage with the other informational structures, and, ultimately, with the overt representation of c-s, is that it provides for a separate set of principles for the linking of a-s to grammatical functions, and the linking of f-s with c-s. Crucially, in Chichewa, the linking of lexical semantics with argument structure and of argument structure with functional roles involves processes with morphological reflexes, which affect argument structure. The processes hinted at here are lexically realized by the extensions, which, together with the basic predicator, constitute the VS node. Information relating to tense, pronominal arguments (SM and OM), and negation, is specified in the f-s, and realized in the c-s as verbal proclitics. The overt verbal structure thus provides for variable encoding of these informational structures, and various linguistic processes converge on these differences. The differences in the processes that apply within the different components and across them seem to set the theory of LFG apart from other theories such as the Principles and Parameters Theory. In the latter, the lexical base and the functional projections are indeed separate. However, their

Figure 8–2.

< previous page

page_198

next page >

< previous page

page_199

next page >

Page 199 uniform representation in X’-theoretic configuration, with Move-α applying across them, does not make for explanation of the phenomena to be noted next. Clitics, excluded from the linkage of argument structure with functional structure, appear in the realization of morphosyntactic information in the f-s. The proposed architecture of UG argues for modularity in the language faculty, a conception of cognitive architecture that is supported by other considerations, to which we proceed. Corroborative external evidence for such a conception of the language faculty comes from cases of acquired language deficits as well as language change. ACQUIRED LANGUAGE DEFICIT One area where the structure of language is revealing with respect to the organization of the cognitive systems that underlie lexical production and comprehension is that of acquired language deficits, when a speaker suffers an impairment to the language faculty occasioned by a stroke, aphasia, and so on. The interest in this phenomenon centers around the insights it may provide on the normal functioning of the apparatus. As noted by Badecker and Caramaza, “[B]y identifying the constraints that acquired deficits appear to respect, one can hope to infer the character of the normal system that would impose such constraints.” (Badecker and Caramazza, 1998:390). It could also provide motivation for specific proposals with regard to the architecture of UG. In the domain of lexical structure or morphological functioning, there is a distinct possibility that evidence from aphasia may have bearing on “the relevance of morphological productivity to the issue of compositionality, processing differences relating to the inflectional/derivational distinction, and the contribution of morphological processing to the comprehension and production of sentences” (ibid., 391). The evidence from acquired language deficits requires judicious interpretation and may not be readily available for a variety of reasons. First, there is the issue of ethics governing research using human subjects; second, subjects for study are not readily available; third, there is the question of the sophistication of linguistic research and of studies of human cognition in general, and of evidence from African languages in particular. While acknowledging the difficulties of deriving evidence for affixation versus cliticization in Chichewa from studies of aphasia or acquired language deficit, there are two cases of individuals known to the author that seem relevant. The first, a female, identified here simply as MM displayed certain patterns of language impairment at the age of around 7, when mastery of language is complete (except for increase in vocabu-

< previous page

page_199

next page >

< previous page

page_200

next page >

Page 200 lary). Specifically, her language was, basically, telegraphic. She produced sentences of the following type: (30) khóza-tu néna-tu iné-tu apa. can-clt tell-clt me-clt here (ndikhozatu kunenatu inetu apa) ‘I can report (tell on you) (about what is going on).’ (31) pita chitsíme ine. go water-well me (ndikupítá ku chitsíme (ine)) ‘I am going to the well.’ The second member, a male, identified here as LN, was much older. In 1961 he suffered a stroke that impaired his speech, effectively inducing aphasia. The recovery process did not restore his language ability completely, leaving him with speech patterns that also turned out to be largely telegraphic. He produced expressions such as this: (32) funíká kuchézá bwinobwíno. want-stat Inf-chat well ‘You need to have a good conversation.’ This is representative of his speech, outside of formulaic expressions or when he couldn’t get by with one-word sentences. In both cases, the language impairment or deficit reveals a pattern. The problems of production seemed to be more acute with, and almost exclusively centered around, pre-VS material, the proclitics. Affixation was not comparably affected. In this respect they manifested problems noted in patients speaking other languages. Thus, Badecker and Caramazza studied an Italian-speaking patient, named FS, who made morphological substitution errors but the errors “predominantly affected the inflectional specification of a word (gender and number markers for nouns and adjectives; and tense, aspect, person, and number specifications of verbs). Derivational morphology was virtually unaffected” (1998:401). In fact, it has been noted in various studies of agrammatisms that patients of acquired language deficits tend to have difficulties with inflectional morphology. Such selective impairment, largely specialized to the verbal proclitics in the case of MM and LN, while leaving the domain of affixation practically intact, should be suggestive or indicative of the organization of the language faculty in human cognition. It could, thus, be relevant to the evaluation of, and choice made between, competing conceptions of the architecture of UG.

< previous page

page_200

next page >

< previous page

page_201

next page >

Page 201 LANGUAGE CHANGE According to Givón’s hypothesis the integrity of the VS in Bantu derives from the earlier morphologization of certain verbs into suffixes, followed by subsequent morphologization of those verbs that became verbal prefixes. The hypothesis provides little, if any, insight into the features that made these groups of verbs natural classes, variably affected by the morphologization process. It is, nonetheless, a matter of fact that the suffixes are involved in the determination of argument structure, whereas the preverbal proclitics are not. There is also need to explain the phonological differences between them. The significant consideration is that Bantu verbal morphosyntax appears to reflect aspects of diachrony in the structural representation. The verbal suffixes are distinct from the proclitics in form and function; they differ in their connection with the VR, connections established at different stages in the historical development. The verbal suffixes, realizing argument structure, constitute part of the VS. If the affixes and the clitics reflect aspects of diachrony, they might also shed light on aspects of language change. The modular organization suggests the possibility of selective change. Two examples appear to be relevant. First, argument structure changing suffixes in Bantu are subdivided into those that increase the number of arguments, such as the causative and the applicative (Hoffman, 1991; Machobane, 1989), those that reduce the number of expressible arguments, such as the passive, stative, reciprocal (cf. Dlayedwa in prep.; Mchombo, 1993a, 1999a, 1999b), and the largely nonproductive affixes that leave the argument structure unchanged, such as the reversive. Guthrie (1962) designated these as O+, O-, and O extensions, respectively. (O+ indicates increase of grammatical objects, O- loss of a grammatical object, and O leaves the grammatical objects unchanged.) In the linking theory of LFG (Alsina, 1993) argument structure reducing affixes suppress an argument, affecting the syntactic realization of the remaining arguments. In some studies a more abstract lexical conceptual structure (lcs) underlies the argument structure that is then linked to f-s or notational variants of that level of representation. Within such proposals the difference between the passive, which allows for an implicit agent argument, and the stative, which does not permit the construal of an agent at all, is handled by locating stativization in the lcs to a-s mapping, and passive in the a-s to f-s mapping (Bature, 1991; Dubinsky and Simango, 1996; Jackendoff, 1990). While argument reduction is distinct from argument increasing, such differences in the nature of argument reduction have to be accommodated, too. Lunda-Ndembu, a language spoken in the Northwestern Province of Zambia (belonging to a cluster of Bantu languages on the Zambia-CongoAngola border that include Kimbundu, Kaonde, Lovale, Lunda),

< previous page

page_201

next page >

< previous page

page_202

next page >

Page 202 displays an interesting pattern of language change. In this language the argument structure reducing affixes, the passive, stative, reciprocal, have been lost, retaining only restricted vestigial remains (Kawasha, p.c.). Other constructions, such as the impersonal or reflexive, have been adapted to express passivity or reciprocalization (cf. Givón and Kawasha, 1999; Kawasha, 1999). The language change is non-arbitrary, affecting an identifiable component of the a-s. Second, there is the study of Kituba, an “endogenous creole” spoken in the Democratic Republic of Congo (formerly Zaire). This language, derived primarily from other Bantu languages, is remarkable for its less agglutinating nature in its verbal morphosyntax. The salient aspects of Kituba’s departure from the canons of Bantu morphosyntax are “the absence of a subject-agreement prefix on the verb, the lack of object pronominal prefixes, and the absolute reliance on independent and morphologically invariant pronouns, as well as on their syntactic positions to determine the subject and objects in a sentence. Aside from derivational extensions, the verb form varies only according to tense” (Mufwene, 1997:175). It is noted further, that “in losing the pronominal object prefix…Kituba has also lost the reflexive” (ibid., 177). The study of Kituba indicates that in its development, as a corollary to Givón’s later morphologization of clitics in Bantu, Kituba simply failed to get them morphologized. Thus, Kituba diverges from the features typical of Bantu morphosyntax in that it appears to be more isolating. Again, this divergence is manifested in the clitic domain, with the verbal extensions largely intact within the verb stem. Kituba has “selected the free-pronoun option for anaphoric objects over the incorporated alternative” (Mufwene 1994:81), but this does not extend to affixes involved in lexical derivation. Grammatical theory should provide for explanation of such facts. The variable behavior between pre-VS elements (clitics) and postVR extensions (affixes) provides more than circumstantial evidence for the separability of morpholexical from morphosyntactic operations or, to put it differently, derivational from inflectional morphology. These aspects of language change provide motivation for the modularity of the language faculty characterized by the LFG conception of UG. LANGUAGE ACQUISITION Studies of Bantu language acquisition are sparse, and evidence from acquisition may not be unequivocal. The available case studies of Bantu language acquisition have focused on acquisition of nominal and, to a lesser extent, verbal morphology, as well as to the acquisition of tonal patterns. The relation between tone assignment and tone realization, and morphosyntax in Bantu is significant. For instance, Mtenje (1986b, 1988)

< previous page

page_202

next page >

< previous page

page_203

next page >

Page 203 noted differences between extensions and clitics with regard to tone patterning in Chichewa. Briefly, extensions are unspecified for tone. They inherit the tone of the VR to which they are suffixed, giving the VS the appearance of an accentual system. Clitics, on the other hand, may be associated with specific tone melodies. In Chichewa, tone provides phonological cues for some grammatical configurations, for instance a relativized NP configuration, or the VP (cf. Bresnan and Mchombo, 1986, 1987; Kanerva, 1990), just as stress assignment or penultimate lengthening may signal the domain of word boundary (cf. Demuth, 1992). The claim that clitics are associated with tone melodies drawn from a tone lexicon is incorporated into a study of the acquisition of negation in Chichewa. Chimombo and Mtenje (1989) build on the idea of tone lexicon of Mtenje (1988) in a study of the acquisition of negation in Chichewa. They claim that the idea of a tone lexicon provides morphosyntactic elements, such as negation and tense, with the tone melodies that they can select, and argue that the case studies offer additional support for the postulation of tone patterns that are assigned to entire morphosyntactic domains. This study of the acquisition of negation and tone patterns is of interest in that the tone lexicon is accessed by lexical items and the clitics, but not by the suffixes, whose tone pattern is largely determined by that of the VR.6 Once again, the affixes behave differently from the clitics. The implications for acquisition studies reduce to a delimitation of the elements of Bantu morphosyntax that access the tone lexicon. The affix/clitic distinction is also evident from other acquisition studies. In a study of the acquisition of Sesotho, Demuth (1992) claims that the progression in the acquisition of the verbal system is from the lexical stem through rudimentary morphological marking, accommodating the obligatory SM (given the pro-drop nature of Bantu languages), which are acquired earlier than the optional OM. The development of the fully inflected adult form is manifested later. Demuth summarizes the observations as follows: Lack of early inflection > shadow vowel > well-formed morphemes.7 The developmental stages that are correlated with the emergence of the various preverbal clitics do not correlate with the development of the verbal suffixes. Demuth, while correctly noting that the study of how Sesotho verbal extensions are acquired is still in its infancy, still claims that it is evident that “some verb forms are not analysed as verb+extension initially, but are rather acquired as lexical wholes, and probably even treated as such by adults” (Demuth, 1992:604ff). This is suggestive of different status accorded to the affixes and the clitics, a difference that has implications for acquisition strategies.

< previous page

page_203

next page >

< previous page

page_204

next page >

Page 204 PARSING STRATEGIES FOR BANTU The analysis provided thus far has implications for processing of Chichewa sentences. The affixes, as part of lexical structure, are involved in argument structure and the assignment of thematic role information, relevant to syntactic realization of the arguments (for critical comments, see Ravin, 1990). Clitics, on the other hand, are excluded from those functions (Mchombo, 1999a). Studies of parsing of Bantu languages have yet to get off the ground (see Mchombo, 2000), but the indications are that the affix/clitic distinction will prove relevant to information processing. Consider the following: (33) a. a-ku-phíká chitumbúwa anyáni. 2SM-pres-cook 7-pancake 2-baboon ‘The baboons are cooking a pancake.’ b. chitumbúwa chi-na-phík-ík-a. 7-pancake 7SM-past-cook-stat-fv ‘The pancake got cooked.’ In sentence (33a) the NPs anyáni “baboons” and chitumbúwa “pancake” are identified with the grammatical functions of subject and object, and the thematic roles of agent and theme, respectively. This is facilitated in part by word order and in part by lexical aspects of the predicate phika “cook.” However, note that the identification of the NP anyáni “baboons” with the subject function is mediated by its anaphoric relation with the subject marker “a,” a clitic, in the verbal structure. In sentence (33b), on the other hand, the clitic “chi,” sharing the Φ-features of the NP chitumbúwa “pancake,” links that NP with the subject function. However, the stative extension -ik-is relevant to the assignment of the role of theme to the NP (cf. Mchombo, 1993a, 1997a, 1997b, 1999a, 2000). The separability of thematic role from grammatical function and their correlation with different overt forms in the verbal morphosyntax must have consequences for the computational procedure. Work in this area may be only beginning, but it already signals the relevance of the distinction between affixes and clitics in these languages. If language processing draws directly on knowledge of grammar, close correspondences between the organization of the linguistic system and the organization of the processing system should be expected (Fodor, 1978, 1990; Frazier, 1988, 1990; Halvorsen, 1988). It is instructive in this regard that the architecture of the theory of LFG provides for elegant description and explanation of the various facets of Bantu noted here through parallel representation of the informational structures.

< previous page

page_204

next page >

< previous page

page_205

next page >

Page 205 CONCLUSION This chapter has focused on aspects of Bantu verbal morphosyntax, noting the difference in form and function between the verbal extensions (affixes) and the clitics added to the verb stem. This motivates a specific conception of the structure of the verbal unit in Bantu, given in Figure 8–1. The structure provides for a principled separation of derivational from inflectional morphology and for principled explanations of, inter alia, aspects of language acquisition, language deficit, language change, and language processing. It has been argued that the theory of LFG, providing for parallel representation of linguistic information, linked in principled ways, not only captures the modularity evident in Bantu linguistic organization but also provides the best means of capturing the various issues noted previously. The modularity of language fits in with studies of the structure of the mind, central to research in cognitive science. NOTES The ideas explored here were previously presented at Yale University, University of Wisconsin, Madison, and at the University of Illinois, Urbana-Champaign. I owe a special debt of gratitude to Alex Alsina, Mark Baker, Isabel Barreras, Elabbas Benmamoun, Eyamba Bokamba, Joan Bresnan, Mary Dalrymple, Zodwa Dlayedwa, Larry Hyman, Ray Jackendoff, Peter Ihionu, Andreas Kathol, Boniface Kawasha, Pascal Kishindo, A1 Mtenje, Stanley Peters, Antonia Folarin Schleicher, Salikoko Mufwene, John Mugane, and Stephen Neale, for intellectual stimulation as well as insights into some of the data. I am also grateful to the participants in those presentations and to the anonymous reviewer for very helpful comments. None of the individuals mentioned here necessarily agrees with or endorses the ideas expressed here, responsibility for which rests entirely with me. This work grew out of a project supported in part by a grant from the Committee on Research at the University of California, Berkeley and, in part, by the National Science Foundation grant #SBR-93–19371. 1. The following abbreviatory convention will be adopted: Appl = applicative Assoc = associative marker Caus = causative clt = clitic Distdem = distal demonstrative f.v. = final vowel Inf = infinitive marker Pass = passive Pres = present tense proxdem = proximal demonstrative Pst = past tense pstprog = past progressive Recip-reciprocal Stat = stative 2. Mugane claims that in Gikuyu the OM can participate in such nominalizations, although the data remain inconclusive with regard to the productivity of such nominalizations. In Chichewa nominalizations involving the OM are not possible.

< previous page

page_205

next page >

< previous page

page_206

next page >

Page 206 3. The reflexive has been analyzed in some studies as a valence reducing morpheme (cf. Grimshaw, 1990). Her ideas are extended to Bantu by Matsinhe, who treats the reflexive in Tsonga as a derivational morpheme, essentially on par with the verbal extensions in Bantu (cf. Matsinhe, 1994). This analysis is at odds with the approach argued for here and elsewhere (cf. Mchombo, 1993b). 4. Myers has claimed that, on morphological and phonological grounds, there is evidence for what he has termed the “inflectional stem hypothesis.” According to this hypothesis, Bantu languages have an INFL node comprising the prefixal material from the subject marker up to but excluding the object marker. In other words, the inflectional stem is a sister to the macrostem (cf. Myers, 1998). I will not adopt this hypothesis here, retaining the binary branching structure that Myers calls the “prefixation hypothesis” and that he rejects as incorrect for Bantu morphosyntactic organization. There certainly does not appear to be obvious syntactic evidence for the inflectional stem hypothesis, so the issue will be left open. 5. Of course within the Principles and Parameters Theory the verb raises to the tense to have its features checked just as other categories raise to have their case or number features checked. However, such movement, forced by the feature-checking mechanism, remains distinct from predicate raising of the variety discussed here. 6. In Chichewa the stative and, for some speakers, the passive, are associated with high tones. The complication with these is that they are also the extensions that, in general, tend to occur in the penultimate syllable, normally the locus for stress. 7. The “shadow vowel” indicated here was the vowel /a/ or /i/, apparently used to fulfill the function of the first person subject clitic in the early speech of children acquiring Sesotho. REFERENCES Abasheikh, M.I. 1978. The Grammar of Chimwini Causatives. Ph.D. dissertation, University of Illinois at Urbana-Champaign. Alsina, A. 1993. Predicate Composition: A Theory of Syntactic Function Alternations. Ph.D. Dissertation, Stanford University. Anderson, S.R. 1988. Inflection. In M.Hammond and M.Noonan (eds.) Theoretical Morphology. Approaches in Modern Linguistics. San Diego: Academic Press. Badecker, W., and Caramazza, A. 1998. Morphology and aphasia. In A.Spencer and A.Zwicky (eds.) The Handbook of Morphology. Oxford: Blackwell. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. ——. 1990. Pronominal Inflection and the Morphology-Syntax Interface. Paper presented at 26th Annual Meeting of the Chicago Linguistics Society, at Chicago. Bature, A. 1991. Thematic Arguments and Semantic Roles in Hausa. Ph.D. Dissertation, Stanford University. Bentley, M.E. 1994. The Syntactic Effects of Animacy in Bantu Languages. Ph.D. dissertation, Indiana University. Bergvall, V. 1985. A typology of empty categories for Kikuyu and Swahili. In G. Dimmendaal (ed.) Current Approaches to African Linguistics. Dordercht: Foris.

< previous page

page_206

next page >

< previous page

page_207

next page >

Page 207 Bokamba, E. 1981. Aspects of Bantu Syntax. Unpublished manuscript, University of Illinois at UrbanaChampaign. Bresnan, J., 1994. Locative inversion and the architecture of universal grammar. Language 70.1:72–131. ——. 2001. Lexical-Functional Syntax. Oxford: Blackwell Publishers. Bresnan, J., and Mchombo, S.A. 1986. Grammatical and Anaphoric Agreement. Paper presented at Papers from the Parasession on Pragmatics and Grammatical Theory, at University of Chicago. ——. 1987. Topic, Pronoun and Agreement in Chichewa. Language 63.4:741–82. Chimombo, M., and Mtenje, A. 1989. Interaction of tone, syntax and semantics in the acquisition of Chichewa negation. Studies in African Linguistics 20.2:1–26. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. ——. 1991. Some notes on economy of derivation and representation. In R. Freidin (ed.) Principles and Parameters in Comparative Grammar. Cambridge, MA: MIT Press. Chomsky, N., and Lasnik, H. 1993. Principles and parameters theory. In J. Jacobs, A.V.Stechow, W.Sternefeld and T.Vennemann (eds.) Syntax: An International Handbook of Contemporary Research. Berlin: Walter de Gruyter. Dembetembe, N.C. 1987. A Linguistic Study of the Verb in Korekore. Harare: University of Zimbabwe. Demuth, K. 1992. The Acquisition of Sesotho. In D.I.Slobin (ed.) The Cross-linguistic Study of Language Acquisition. Hillsdale, NJ, and London: Lawrence Erlbaum Associates, Publishers. Demuth, K., and Johnson, M. 1989. Interaction between discourse functions and agreement in Setswana. Journal of African Languages and Linguistics 11:21–35. Dlayedwa, C.Z., in prep. Morphology and Argument Structure Reducing Affixes in Xhosa. Ph.D. Dissertation, Linguistics, University of Essex, Colchester. Du Plessis, J.A., and Visser, M. 1992. Xhosa Syntax. Pretoria: Via Afrika. Dubinsky, S., and Simango, S.R. 1996. Passive and stative in Chichewa: Evidence for modular distinctions in grammar. Language 72.4:749–81. Everett, D.L. 1989. Clitic doubling, reflexives, and word order alternations in Yagua. Language 65.2:339– 72. Fodor, J.D. 1978. Parsing strategies and constraints on transformations. Linguistic Inquiry 9:427–73. ——. 1990. Thematic roles and modularity: Comments on the chapters by Frazier and Tanenhaus et al. In G.T.M.Altmann (ed.) Cognitive Models of Speech Processing. Psycholinguistic and Computational Perspectives. Cambridge: MIT Press. Frazier, L. 1988. Grammar and language processing. In F.J.Newmeyer (ed.) Linguistics: The Cambridge Survey. Linguistic Theory: Extensions and Implications. Cambridge, U.K.: Cambridge University Press. ——. 1990. Exploring the Architecture of the Language-Processing System. In G.T.M.Altmann (ed.) Cognitive Models of Speech Processing. Psycholinguistic and Computational Perspectives. Cambridge, MA: MIT Press. Givón, T. 1971. On the verbal origin of the Bantu verb suffixes. Studies in African Linguistics 2.2:145– 63. Givón, T., and Kawasha, B. 1999. Indiscrete Grammatical Relations: The Lunda-Ndembu Passive. University of Oregon. Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press.

< previous page

page_207

next page >

< previous page

page_208

next page >

Page 208 Guthrie, M. 1962. The status of radical extensions in Bantu languages. Journal of African Languages 1:202–20. Hale, K., and Keyser, S.J. 1992. The syntactic character of thematic structure. In I.M.Roca (ed.) Thematic Structure. Its Role in Grammar. New York: Foris. Halpern, A.L. 1998. Clitics. In A.Spencer and A.Zwicky (eds.) The Handbook of Morphology. Oxford: Blackwell. Halvorsen, P-K. 1988. Computer applications of linguistic theory. In F.J.Newmeyer (ed.) Linguistics: The Cambridge Survey. Linguistic Theory: Extensions and Implications. Cambridge, U.K.: Cambridge University Press. Haverkort, M. 1993. Clitics and parametrization. Case studies in the interaction of head movement phenomena. Ph.D. dissertation, Katholieke Universiteit Brabant. Hoffman, M. 1991. The syntax of argument-structure-changing morphology. Ph.D. dissertation, Massachusetts Institute of Technology. Hyman, L. 1991. Conceptual issues in the comparative study of the Bantu Verb Stem. In S.S.Mufwene and L.Moshi (eds.) Topics in African Linguistics. Amsterdam/Philadelphia: John Benjamins Publishing Co. Hyman, L., and Mchombo, S.A. 1992. Morphotactic constraints in the Chichewa verb stem. Paper presented at Papers from the Parasession on the Place of Morphology in a Grammar, at Berkeley, California. Jackendoff, R.S. 1990. Semantic Structures. Cambridge, MA: MIT Press. Jaeggli, O. 1982. Topics in Romance Syntax. Dordrecht: Foris Publications. Jelinek, E. 1984. Empty categories, case, and configurationality. Natural Language & Linguistic Theory 2:39–76. Kanerva, J.M. 1990. Focus and phrasing in Chichewa phonology. In J.Hankamer (ed.) Outstanding Dissertations in Linguistics. New York: Garland. Katamba, F. 1984. A nonlinear analysis of vowel harmony in Luganda. Journal of Linguistics 20:257–75. Katupha, J.M. 1991. The grammar of Emakhuwa verbal extensions. Ph.D. dissertation, School of Oriental & African Studies, University of London. Kawasha, B. 1999. Some Aspects of Lunda Grammar. M.A., Linguistics, University of Oregon, Eugene. Keach, C.N. 1995. Subject and object markers as agreement and pronoun incorporation in Swahili. In A.Akinlabi (ed.) Theoretical Approaches to African Linguistics. Trenton: Africa World Press, Inc. Kiparsky, P. 1982a. Lexical morphology and phonology. In T.L.S.O.Korea (ed.) Linguistics in the Morning Calm. Seoul: Hanshin Publishing Company. ——. 1982b. Word-Formation and the Lexicon. Massachusetts Institute of Technology: Unpublished paper. Klavans, J.L. 1983. The Morphology of Cliticization. Paper presented at Chicago Linguistic Society, at Chicago. Laka, I. 1994. On the Syntax of Negation. New York: Garland Publishing. Machobane, M. 1989. Some Restrictions on the Sesotho Transitivizing Morphemes. Ph.D. dissertation, McGill University. ——. 1993. The ordering restriction between the Sesotho applicative and causative suffixes. South African Journal of African Languages 13:129–37. Marantz, A. 1984. On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. ——. 1988. Clitics, morphological merger, and the mapping to phonological structure. In M.Hammond and M.Noonan (eds.) Theoretical Morphology. Approaches in Modern Linguistics. San Diego: Academic Pres.

< previous page

page_208

next page >

< previous page

page_209

next page >

Page 209 Matsinhe, S. 1994. The status of verbal affixes in Bantu languages with special reference to Tsonga: Problems and possibilities. South African Journal of African Languages 14.4:163–76. Mchombo, S.A. 1993a. A formal analysis of the stative construction in Bantu. Journal of African Languages and Linguistics 14:5–28. ——. 1993b. On the binding of the reflexive and the reciprocal in Chichewa. In S.A.Mchombo (ed.) Theoretical Aspects of Bantu Grammar. Stanford: Center for the Study of Language and Information. ——. 1997a. Chichewa: A morphological sketch. In A.Spencer and A.Zwicky (eds.) Handbook of Morphology. Oxford: Blackwell. ——. 1997b. Contributions of African languages to generative grammar. In R.K. Herbert (ed.) African Linguistics at the Crossroads. Papers from Kwaluseni. Koln: Rudiger Koppe Verlag. ——. 1999a. Argument structure and verbal morphology in Chichewa. Malilime. Malawian Journal of Linguistics 1:57–75. ——. 1999b. Quantification and verb morphology: The case of reciprocals in African languages. Linguistic Analysis 29. ——. 2000. Choppin’ up Chichewa: Theoretical Approaches to Parsing an Agglutinative Language. Paper presented at 31st Annual Conference on African Linguistics, Boston University. Mohanan, K.P. 1986. The Theory of Lexical Phonology. Dordrecht: Reidel. Montalbetti, M. 1981. Consistency and Clitics. Unpublished paper, MIT. Mtenje, A.D. 1980. Aspects of Chichewa derivational phonology and syllable structure constraints. Master of Arts, Southern Illinois University, Carbondale. ——. 1985. Arguments for an Autosegmental Analysis of Chichewa Vowel Harmony. Lingua 66:21–52. ——. 1986a. Remarks on extra-harmonicality in Bantu harmonic systems. Sheffield Working Papers in Language and Linguistics, University of Sheffield 3:62–76. ——. 1986b. Issues in the Nonlinear Phonology of Chichewa. Ph.D. dissertation, University College London. ——. 1988. On tone and transfer in Chichewa reduplication. Linguistics 26:125–55. Mufwene, S. 1994. Restructuring, Feature Selection, and Markedness: From Kimanyanga to Kituba. Paper presented at Twentieth Annual Meeting of the Berkeley Linguistic Society: Special session on historical issues in African linguistics, at Berkeley. ——. 1997. Kitúba. In S.G.Thomason (ed.) Contact Languages: A Wider Perspective. Amsterdam/Philadelphia: John Benjamins, B.V. 173–208. Mugane, J. 1997. A Paradigmatic Grammar of Gikuyu. Stanford: CSLI Publications. Myers, S. 1998. AUX in Bantu morphology and phonology. In L.M.Hyman and C.W.Kisseberth (eds.) Theoretical Aspects of Bantu Tone. Stanford, CA: CSLI Publications. Neale, S. 1996. Prolegomena to a “variable-driven” syntax: Reflections on logical form and Chomsky’s minimalist program: Stanford University. Newmeyer, F. 1974. The Precyclic Nature of Predicate Raising. Bloomington: Indiana University Linguistics Club. ——. 1975. The Position of Incorporation Transformations in Grammar. Paper presented at Berkeley Linguistics Society at Berkeley. Ngunga, A. 1997. Lexical Phonology and Morphology of Ciyao. Ph.D. dissertation, Linguistics, University of California at Berkeley.

< previous page

page_209

next page >

< previous page

page_210

next page >

Page 210 Nordlinger, R. 1998. Constructive Case. Evidence from Australian Languages, Dissertations in Linguistics. Stanford: CSLI Publications. Perlmutter, D.M. 1988. The split morphology hypothesis: Evidence from Yiddish. In M.Hammond and M.Noonan (eds.) Theoretical Morphology. Approaches in Modern Linguistics. San Diego: Academic Press. Pollock, J-Y. 1988. Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20.3:365–424. Poulos, G. 1990. A Linguistic Analysis of Venda. Pretoria: Via Afrika. Ravin, Y. 1990. Lexical Semantics without Thematic Roles. Oxford: Clarendon Press. Roberge, Y. 1990. The Syntactic Recoverability of Null Arguments. Kingston: McGill-Queen’s University Press. Rugemalira, J.M. 1993. Runyambo Verb Extensions and Constraints on Predicate Structure. Ph.D. dissertation, University of California at Berkeley. Sadler, L., and Spencer, A. 1998. Morphology and argument structure. In A. Spencer and A.Zwicky (eds.) The Handbook of Morphology. Oxford: Blackwell. Satyo, S.C. 1985. Topics in Xhosa Verbal Extensions. Ph.D. dissertation, University of South Africa, Pretoria, South Africa. Simango, R. 1995. The Syntax of Bantu Double Object Constructions. Ph.D. dissertation, University of South Carolina. Sproat, R. 1988. Bracketing paradoxes, cliticization and other topics: The mapping between syntactic and phonological structure. In M.E.A.Everaert (ed.) Morphology and Modularity. Dordrecht: Foris. Travis, L. 1984. Parameters and Effects of Word Order. Ph.D. dissertation, Massachusetts Insitute of Technology.

< previous page

page_210

next page >

< previous page

page_211

next page >

Page 211 9 Two Types of Wh -In-Situ Masanori Nakamura In languages like English, the [+WH] Comp must host a wh-phrase in its specifier position at Spell-Out, as shown here: (1) a. *John bought what? (non-echo question) b. What did John buy? (2) a. *Mary asked Bill John bought what, b. Mary asked Bill what John bought. In the ill-formed (a) examples the wh-phrase what has failed to move to an appropriate position, that is, [+WH] Spec of CP. The (b) examples are well formed since wh-movement has taken place in overt syntax. Since Huang’s (1982) influential work, much attention has been paid to the syntax of wh-in-situ in the generative literature. Consider the following examples from Japanese:1 (3) a. Hana-ga nani-o kaimasita ka? Hana-NOM what-ACC bought Q? ‘What did Hana buy?’ b. Hana-ga Ai-ni [Emi-ga nani-o katta Hana-NOM Ai-DAT Emi-NOM what-ACC bought ka] tazuneta. Q asked ‘Hana asked Ai what Emi bought.’

< previous page

page_211

next page >

< previous page

page_212

next page >

Page 212 (3a) and (3b) are the Japanese counterparts of (1b) and (2b), respectively. Japanese is an SOV language. We can see that the object wh-phrase in (3) remains in its base position. Thus, Japanese contrasts with English in terms of the obligatory application of wh-movement. Interestingly, the syntactic behavior of wh-phrases is not uniform across wh-in-situ languages of the world. It seems that relevant languages fall into two major groups. One group is represented by Japanese. The other is represented by Iraqi Arabic (hereafter “IA”). Observe (4a–b) from IA (Wahba, 1991:255): (4) a. Mona shaafat meno? Mona saw who ‘Who did Mona see?’ b. Mona se lat Ali [Ro a ishiarat sheno]? Mona asked Ali Ro a bought what ‘Mona asked Ali what Ro a bought.’ As in (3a-b), (4a-b) involve no application of overt wh-movement, indicating that IA allows wh-in-situ.2 As has been noted, however, there are certain differences between Japanese-type languages and IAtype languages. Compare (5) and (6) (the latter from Wahba, 1991): (5) Hana-ga [Ai-ga nani-o katta to] Hana-NOM Ai-NOM what-ACC bought COMP omoimasita ka? thought Q ‘What did Hana think (that) Ai bought?’ (6) *Mona tsawwarat [Ali ishtara sheno]? Mona thought Ali bought what (‘What did Mona think Ali bought?’) The wh-phrase within the finite embedded clause in Japanese (5) can be construed with the matrix [+Q] Comp, whereas its equivalent in IA (6) cannot be.3 This difference calls for an explanation. Furthermore, as we will see, there are other systematic differences between the two types of languages. The purpose of this chapter is to address the question of how the parametric differences in question can adequately be captured.4 Many of the previous analyses assume explicitly or implicitly that in-situ whelements are licensed basically in the same way cross-linguistically (cf. Huang, 1982; Nishigauchi, 1990; Ouhalla, 1996). In what follows, four kinds of licensing mechanisms found in the literature are tested against a set of relevant data. It is argued that the two types of wh-in-situ are in

< previous page

page_212

next page >

< previous page

page_213

next page >

Page 213 fact subject to separate modes of licensing: unselective binding (Tsai, 1994; Reinhart, 1998) in the case of Japanese and feature movement (Chomsky, 1995) in the case of IA. It is also argued that the lexical properties of wh-in-situ fully determine how it is licensed. The organization of the chapter is as follows. The first section touches on the lexical characteristics of the two types of wh-in-situ, setting the stage for the succeeding discussions. The next section surveys their syntactic properties in terms of locality. This is followed by a comparison of existing analyses of wh-in-situ based on the results of the previous section. The final section elaborates on the present analysis of IA-type wh-in-situ from the viewpoint of feature attraction. It is followed by brief concluding remarks. LEXICAL PROPERTIES One of the basic tenets of the current principles-and-parameters approach to syntax is that the computational system of the human language is invariant across languages and observable crosslinguistic variations derive solely from differences in the morphological properties of the lexicon (see Chomsky, 1995, and references cited there). With this background in mind, let us briefly consider the lexical characteristics of the two types of wh-in-situ. As has been pointed out (see Ouhalla, 1996, among others), there are noticeable lexical differences between Japanese wh-elements and their IA equivalents. First, there is a difference in terms of the quantificational force of wh-phrases. Compare (3) with the following Japanese example: (7) Hana-ga nani-o katte mo, boku-wa. Hana-NOM what-ACC buy MO I-TOP odoroka nai be surprised not ‘No matter what Hana buys, I would not be surprised.’ In (3) where the wh-phrase nani “what” is associated with the question marker, it is interpreted as an existential expression, whereas in (7) where it is associated with the morpheme mo, it is interpreted as a universal one. This indicates that the Japanese wh-words themselves are devoid of a quantificational force. The point can also be made with the examples in (8) where the morphemes ka and mo are attached directly to the (8) a. Hana-ga dare-ka-o mita. Hana-NOM what-KA-ACC saw ‘Hana saw someone.’

< previous page

page_213

next page >

< previous page

page_214

next page >

Page 214 b. Hana-ga dare-mo-o mita. Hana-NOM what-MO-ACC saw ‘Hana saw everyone.’ Thus, there are reasons to believe that Japanese wh-phrases are indefinites: They lack a quantificational force of their own, and their force is determined by an appropriate quantificational element (Kuroda, 1965; Nishigauchi, 1990; Watanabe, 1992; see also Lewis, 1975; Kamp, 1981; Heim, 1982, for related phenomena). On the other hand, as observed by Ouhalla (1996), IA wh-words do have inherent (existential) quantificational forces: No examples parallel to those in (7) and (8) are found in IA. Second, there is a difference in the morphology of wh-phrases. Ouhalla (1996) points out that those in IA are made up of the wh-morpheme and the pronominal element -o, as shown next. (9) a. men-o who-him b. sheno-o what-it The morpheme -o can be used as an object clitic pronoun, as exemplified in (10). (10) a. Shuft-o. saw-I-him/it ‘I saw him/it.’ b. Ishtarit-o. bought-I-it ‘I bought it.’ Following Ouhalla (1996), I assume that the pronominal element -o occupies the D position of a DP and that the wh-element is an NP occupying the sister position of the D. The structure for (9a), for example, is assumed to be as follows:

< previous page

page_214

next page >

< previous page

page_215

next page >

Page 215 The surface order of the morphemes in (9a) is obtained by N-to-D head movement. As a result, the D acquires the [+WH] feature. Let us refer to the kind of wh-phrase depicted in (11) as “DP wh-phrase.” Japanese wh-phrases, on the other hand, do not contain any pronominal elements. As mentioned earlier, they are indefinites. Thus the structure of the Japanese counterpart of (9a) can be represented simply as follows: (12) differs from (11) in that it lacks a DP projection altogether. The Japanese-type wh-phrase is referred to here as “NP wh-phrase.” SYNTACTIC PROPERTIES Let us now turn to discussion of the syntactic properties of the two types of wh-in-situ. The general picture that emerges is that DP wh-in-shu obeys stricter locality conditions than NP wh-in-situ. First, as already illustrated in (5) and (6), DP wh-in-shu is subject to a tense condition (hereafter, TC), whereas NP wh-in-situ is not. The TC states that an in-situ wh-phrase and its scope position cannot be separated by a finite clause. Thus, in ill-formed (6) the in-situ wh-phrase within the finite embedded clause cannot be related to the matrix [+WH] C. In well-formed (4a–b), no finite clausal boundary intervenes between the in-situ wh-phrase and the [+WH] C. Consistent with the condition is the fact that the DP wh-word can appear in an infinitival embedded clause and take the matrix scope, as in (13) (Wahba, 1991). (13) Mona itmannat [tishtiri sheno]? Mona hoped to-buy what ‘What did Mona hope to buy?’ As has been documented in the literature (Mahajan, 1990; Ouhalla, 1996), the same condition is at work in Hindi.5 Compare (14a) and (14b), both taken from Mahajan (1990). (14) a. *Raam-ne socca [ki Mohan-ne kis-ko dekhaa]? Ram-ERG thought that Mohan-ERG who saw (‘Who did Ram think Mohan saw?’) b. Raam-ne [Mohan-ko kise dekhne ke liya] kahaa? Ram-ERG Mohan-ERG whom to-see for told ‘Who did Ram tell Mohan to look at?’

< previous page

page_215

next page >

< previous page

page_216

next page >

Page 216 (14a) is ruled out as a violation of the TC.6 (14b) is grammatical, since the embedded clause containing the wh-phrase is not finite. Second, DP wh-in-situ exhibits CED (Condition on Extraction Domain; Huang, 1982) effects, whereas NP wh-in-situ does not. The CED, which bans extraction out of a domain not “properly governed,” subsumes such conditions as the Subject Condition, the Adjunct Condition, and the Complex NP Constraint (CNPC).7 Let us consider the following Japanese examples: (15) a. Hana-ga [Ai-ga sono hon-o nakusita atoni] okotte Hana-NOM Ai-NOM the book-ACC lost after angrily kaetta. left ‘Hana left angrily after Ai lost the book.’ b. Hana-ga [sono hon-o katta onnanoko]-o sitteiru. Hana-NOM the book-ACC bought girl-ACC know ‘Hana knows the girl who bought the book.’ (15a) contains an adjunct clause, and (15b) a relative clause. The following examples of scrambling, based on (15a–b), show that the Adjunct Condition and the CNPC are operative in Japanese (see Saito, 1985, among numerous others):8 (16) a. *Sono honi-o Hana-ga [Ai-ga ti nakusita atoni] the book-ACC Hana-NOM Ai-NOM lost after okotte kaetta. angrily left (‘Hana left angrily after Ai lost the book.’) b. *Sono honi-o Hana-ga [ti katta onnanoko]-o sitteiru. the book-ACC Hana-NOM bought girl-ACC know (‘Hana knows the girl who bought the book.’) Let us now see whether an in-situ object wh-phrase patterns with the scrambled object in (16). Relevant examples are given in (17). (17) a. Hana-ga [Ai-ga nani-o nakusita atoni] okotte Hana-NOM Ai-NOM what-ACC lost after angrily kaerimasita ka? left Q Lit. ‘What did Hana leave angrily after Ai lost?’

< previous page

page_216

next page >

< previous page

page_217

next page >

Page 217 b. Hana-ga [nani-o katta onnanoko]-o Hana-NOM what-ACC bought girl-ACC sitteimasu ka? know Q Lit. ‘What does Hana know the girl who bought?’ Above, the in-situ wh-phrase is successfully associated with the [+Q] Comp outside the adjunct clause (17a) and the Complex NP (17b). The grammaticality of (17) demonstrates that Japanese-type NP whin-situ does not obey the CED. Shifting our attention to DP wh-in-situ, we realize that testing whether it is subject to the CED is difficult because it is constrained independently by the TC, which imposes strict locality. Consider the following IA example with a Complex NP (Wahba, 1991): (18) *Mona ’urfut [il-bint illi ishtarat sheno]? Mona knew the-girl who bought what (Lit. ‘What did Mona know the girl who bought?’) In (18) the wh-phrase appears within the Complex NP whose head is il-bint “the girl.” The illformedness of (18) may be taken to indicate that DP wh-in-situ obeys the CED. A note of caution, however, is in order. As Ouhalla (1996, fn. 15) notes, (18) violates the TC. Thus, an example with a nontensed relative clause would be more appropriate. But to construct such examples without recourse to nominalization seems difficult in IA. Fortunately, there is more convincing empirical evidence that DP wh-in-situ is in fact subject to the CED. Let us draw relevant examples from Hindi. (19) exemplifies the canonical control structure in Hindi (Mahajan, 1990:160). (19) a. Raam-ne [siitaa-ko dekhnaa]caahaa. Ram-ERG Sita to-see wanted ‘Ram wanted to see Sita.’ b. Raam-ne [kis-ko dekhnaa]caahaa. Ram-ERG who to-see wanted ‘Who did Ram want to see?’ In (19) the embedded clause appears to the left of the matrix verb. The minimal difference between (19a) and (19b) concerns the presence or absence of the wh-word. Notice that (19b) with the in-situ wh-word does not violate the TC, since the embedded clause is nonfinite. An interesting contrast emerges when the embedded clause in (19) undergoes rightward extraposition. Observe the following examples (Mahajan, 1990:162):

< previous page

page_217

next page >

< previous page

page_218

next page >

Page 218 (20) a.

Raam-ne caahaa [siitaa-ko dekhnaa]. Ram-ERG wanted Sita to-see ‘Ram wanted to see Sita’ b. *Raam-ne caahaa [kis-ko dekhnaa]. Ram-ERG wanted who to-see (‘Who did Ram want to see?’) (20a) shows that extraposition is in principle possible in Hindi. Curiously, (20b) where the extraposed clause contains the wh-phrase is ill-formed. Since the TC is irrelevant here, the ill-formedness of (20b) must be attributed to some other condition. I suggest that the condition involved is the CED: As is wellknown, extraposition puts a constituent in a position that is not “properly governed.” The following pair of examples is a case in point: (21) a. Whoi did you buy [a famous picture of ti] yesterday? b. *Whoi did you buy yesterday [a famous picture of ti]? (21a) is acceptable, for the wh-phrase has been extracted out of the “properly governed” constituent in object position. (21b) is unacceptable in violation of the CED, for the extraposition has removed the constituent from object position. It is easy to see that the contrast between (19b) and (20b) mimics that between (21a) and (21b). In this light, I take examples like (20b) to show that DP wh-in-situ exhibits CED effects.9 Finally, both NP wh-in-situ and DP wh-in-situ exhibit apparent wh-island effects. Consider the Japanese example in (22), where the embedded C as well as the matrix one is occupied by the Q-marker. (22) Hana-wa [dare-ni nani-o ageta ka] Hana-TOP who-DAT what-ACC gave Q wasuremasita ka? forgot Q ‘Did Hana forget what she gave to whom?’ (22) permits only one interpretation, on which both of the wh-phrases take the embedded scope. In other words, neither dare “who” nor nani “what” can take the matrix scope. This appears to imply that the wh-island Condition is part of the Japanese grammar.10 DP wh-in-situ is also subject to the wh-island Condition. This can be illustrated by the following IA example (Wahba, 1991:260): (23) *Nasat Mona [li-meno tinti sheno]? forgot Mona to-whom to-give what (‘What did Mona forget to give to whom?’)

< previous page

page_218

next page >

< previous page

page_219

next page >

Page 219 Phrasal wh-movement of the kind found in English is optional in IA (Wahba, 1991; Ouhalla, 1996). In (23) the dative wh-phrase li-meno “to whom” has moved into the [+WH] Spec of CP of the clausal complement, whereas the object wh-phrase sheno “what” remains in-situ. The latter cannot be construed as taking scope over the whole sentence, hence the ungrammaticality of (23). The syntactic behavior of the two types of wh-in-situ can be summarized as in Table 9–1. Now the task is to explain why they behave the way they do. TOWARD A NONUNITARY ACCOUNT Since Huang (1982), various kinds of analyses have been put forth to come to grips with the syntax of wh-in-situ. There exist two major approaches: the movement approach and the binding approach. The former includes such analyses as those defended by Huang (1982), Lasnik and Saito (1984), Watanabe (1992), among others. Huang (1982) and Lasnik and Saito (1984) propose that in languages like Japanese, the movement of wh-in-situ to its scope position takes place at LF. Under this proposal, the only relevant difference between English and Japanese lies in the level at which they perform whmovement. Watanabe (1992), on the other hand, argues that wh-movement does take place in overt syntax in Japanese. Under his analysis, what moves is an invisible operator associated with wh-in-situ. The latter has been advocated by Tsai (1994), Ouhalla (1996), Reinhart (1998), and others. On this approach, no movement is posited to capture the behavior of wh-in-situ. Ouhalla (1996) presents a modified theory of Generalized Binding (Aoun, 1986). Tsai (1994) and Reinhart (1998), on the other hand, claim that the right way of explaining the syntax of wh-in-situ is through unselective binding. I would like to show now that no matter which analysis one wishes to adopt, no single analysis would be able to cover both types of wh-in-situ without invoking ad hoc assumptions and stipulations. It is argued that DP wh-in-situ should be accounted for in terms of null operator (NO) movement (cf. Basilico, 1998) and NP wh-in-situ in terms of unselective binding (cf. Nishigauchi, 1990). TABLE 9–1. Syntactic Behavior of Two Types of Wh -In-Situ NP wh DP wh Tense effects No Yes CED effects No Yes Wh -island effects Yes Yes

< previous page

page_219

next page >

< previous page

page_220

next page >

Page 220 Let us take up the four kinds of analyses in turn. LF Movement In its simplest form, the LF movement theory predicts that LF wh-movement should behave just like overt wh-movement, which is free from the TC but constrained by the CED and the Wh -island Condition, as shown here: (24) a. Whoi does John believe (that) Mary kissed ti? b. *Whoi did you leave without meeting ti? c. ?*What do you wonder who bought ti? The prediction, however, is correct only partially. In particular, the theory leaves unexplained the presence of tense effects in the case of DP wh-in-situ (see [6]) and the absence of CED effects in the case of NP wh-in-situ (see [17]). To deal with the difficulties, it must resort to ad hoc stipulations. For instance, Wahba (1991), without giving any independent motivation, assumes that TP headed by finite T as well as CP acts as a bounding node for Subjacency in IA. Nishigauchi (1990) and others argue that the lack of CED effects in Japanese (17) is due to the possibility of LF pied-piping: The in-situ whphrase moves only within the island, which in its entirety raises into the [+WH] Comp. Stechow (1996), however, has demonstrated that none of the arguments for LF pied-piping presented by Nishigauchi (1990) is tenable. Therefore, there are reasons to believe that the LF movement analysis should be rejected. Null Operator Movement The NO analysis predicts that the wh-in-situ construction should share the same set of syntactic characteristics with the NO construction of the more familiar type, exemplified in (25). (25) John is easy to please t. (25) has been analyzed as involving the kind of derivation in (26), where the NO generated in object position undergoes wh-movement (see Browning, 1987, for extensive discussion). (26) John is easy [CP Op i to please ti] Consider the following tough-movement examples pertaining to the TC, the CED, and the Wh -island Condition (Browning, 1987):

< previous page

page_220

next page >

< previous page

page_221

next page >

Page 221 (27) a. *John is easy to believe (that) Mary kissed t. b. *John is easy for us to describe [a plan to assassinate t.] c. *The job was dangerous enough (for us) to wonder [whether to offer t to John]. (27a) shows that NO movement, unlike regular wh-movement, is subject to the TC. (27b-c) demonstrate that it exhibits CED and wh-island effects. The parallel between the DP wh-in-situ construction and the tough-construction indicates that the former involves NO movement. Thus I suggest that Watanabe’s (1992) analysis is basically correct for DP wh-in-situ but, crucially, not for NP wh-in-situ, which the analysis was originally designed to explain (cf. Basilico, 1998). The problematic aspects of the analysis center around the unexpected lack of tense and CED effects in the case of NP wh-in-situ. It has to make ad hoc stipulations to deal with relevant data such as (5) and (17). For instance, in his discussion of the CNPC, Watanabe (1992) stipulates that the NO is generated outside the complex NP, thereby avoiding crossing a barrier. It is not clear at all, however, how the association between the NO and the in-situ wh-phrase inside the complex NP is established. In brief, the NO analysis is valid for IA-type languages, but not for Japanese-type ones. Generalized Binding The fact that DP wh-in-situ is subject to the “Tensed-S Condition” invites an account based on the notion of Generalized Binding (Aoun, 1986). Ouhalla (1996) argues precisely for such an account. Recognizing the lexical differences between the NP and the DP wh-phrases alluded to in the previous section, Ouhalla (1996:686–87) puts forth the following binding conditions: (29)a.A bare (refl-/wh-) anaphor need not have a local antecedent in the minimal finite clause in which it occurs. b.A compound (refl-/wh-) anaphor must have a local antecedent in the minimal finite clause in which it occurs. Here “bare wh-anaphor” corresponds to “NP wh-in-situ” in our terms and “compound wh-anaphor” to “DP wh-in-situ.” Ouhalla claims that (28b) accounts for wh-in-situ in IA. For example, (6) is ruled out as a violation of (28b): The Comp in the embedded clause lacks the feature [+WH] and thus the whphrase is not bound by an appropriate antecedent in the minimal finite clause. (4a–b), on the other hand, are well-formed, since

< previous page

page_221

next page >

< previous page

page_222

next page >

Page 222 they satisfy (28b): in (4a) the wh-phrase is bound by the root [+WH] Comp, and in (4b) it is bound by the [+WH] embedded Comp selected by the matrix verb. Nonfinite embedded clauses like the one in (13) are transparent for the purpose of binding. To rule out other ill-formed IA examples, Ouhalla introduces (29). (29) A compound (ref-/wh-) anaphor must be bound to the nearest potential antecedent. The CNPC violation in (18) and the Wh -island Condition violation in (23) are excluded because the whphrase is not bound by the nearest antecedent, that is, the [+WH] Comp within the relative clause in the case of (18) and the [+WH] Comp of the embedded clause in the case of (23).11 There are, nonetheless, problems with the account. Note that the binding conditions are “generalized” in that they cover both reflexive anaphors and wh-anaphors. Thus, it is expected, for instance, that a bare reflexive anaphor like Japanese zibun “self” and NP wh-in-situ should share the same distributional properties. In addition, a compound reflexive anaphor like Japanese kanojo-zisin “herself” and DP whin-situ should behave in the same way. The following pair of examples establishes that zibun is a bare anaphor, which is able to be bound long distance across a finite clausal boundary and that kanojo-zisin is a compound anaphor, which must find its antecedent in a local domain (see Katada, 1991, among numerous others). (30) a. Hanai-ga [Aij-ga zibuni/j-o kiratteiru to] omotteiru. Hana-NOM Ai-NOM self-ACC hate COMP think Lit. ‘Hana thinks Ai hates self.’ b. Hanai-ga [Aij-ga kanojo-zisini*/j-o kiratteiruto] Hana-NOM Ai-NOM her-self-ACC hate COMP omotteiru. think ‘Hana thinks Ai hates herself.’ Under the generalized binding analysis, the fact that DP wh-in-situ is sensitive to the CED poses a problem. This is because the compound reflexive anaphor is insensitive to the CED, as shown in (31). (31) Ai-ga sono sigoto-o [kanojo-zisin-no kojintekina riyuu Ai-NOM the job-ACC her-self-GEN personal reason de] yameta. for quit Lit. ‘Ai quit the job for herself s personal reason.’

< previous page

page_222

next page >

< previous page

page_223

next page >

Page 223 (31) is fully grammatical despite the fact that the compound anaphor is contained in the adjunct phrase. Ouhalla’s treatment of wh-island effects poses another problem. For one thing, the statement in (29) is incorrect, as shown here: (32) Hanai-ga Aij-ni kanojo-zisini/j-no ayamari-o Hana-NOM Ai-DAT her-self-GEN mistake-ACC sitekisita. pointed out Lit. ‘Hana pointed out herself’s mistake to Ai.’ (32) is ambiguous: The anaphor can refer to either the nominative argument or the dative one. Under (29), it is wrongly expected to permit only one interpretation where the anaphor is bound to the nearest potential antecedent, that is, the dative-marked nominal Ai. For another thing, wh-island effects on NP wh-in-situ are mysterious. As exemplified in (30a), the bare anaphor need not be bound to the nearest potential antecedent. NP wh-in-situ, in contrast, must be, as we saw in (22): The matrix [+Q] Comp cannot bind the wh-phrases because of the intervening embedded [+Q] Comp. In a nutshell, generalized binding accounts for neither NP wh-in-situ nor DP wh-in-situ.12 Unselective Binding At the heart of the unselective binding approach is the idea that in-situ wh-arguments are indefinites, which are not endowed with any quantificational force and behave like free variables to be bound by a certain element (Heim, 1982). The approach turns out to be very successful with NP wh-phrases, which, as we saw earlier, are indefinite expressions without any quantificational force of their own. It has been pointed out that unselective binding is insensitive to syntactic islands (see Heim, 1982; Nishigauchi, 1990). Observe the following: (33) a. [A man who owns a donkey] always likes it. b. [A man who knows [that a donkey is forthcoming]] usually buys it. In (33) the indefinite NP a donkey is c-commanded and unselectively bound by the adverbs of quantification (Lewis, 1975), always in (33a) and usually in (33b), at an appropriate level of representation. Note that the NP is inside a complex NP. The well-formedness of (33) is expected since the relation between the binder and the bindee is not established by movement. (33) also shows that finite clausal boundaries do not consti-

< previous page

page_223

next page >

< previous page

page_224

next page >

Page 224 tute a barrier for unselective binding. Thus, the claim that NP wh-in-situ is licensed by unselective binding immediately accounts for its insensitivity to the TC and the CED. It is not the case, however, that unselective binding is totally free from locality effects. Heim (1982) and Nishigauchi (1990) among others demonstrate that it is subject to a version of minimality. Nishigauchi (1990:143–45), for example, illustrates the minimality effects using the following Japanese examples: (34) a. [Honi-o saigo-made yonda hito]-wa taitei book-ACC end-to read person-TOP usually sorei-o utte simatta. it-ACC sold ended ‘A person who read a book to the end usually ended up selling it.’ b. *[Hon i-o kanarazu saigo-made yonda hito]-wa book-ACC always end-to read person-TOP taitei sorei-o utte simatta. usually it-ACC sold ended (‘A person who always read a book to the end usually ended up selling it.’) As in the case of (33), (34a) allows for the “donkey interpretation,” indicated by the coindexation of hon “book” and sore “it.” There the adverb of quantification taitei “usually” c-commands and unselectively binds both the bare NP and the pronoun. (34b) differs minimally from (34a) in that another adverb of quantification kanarazu “always” has been inserted into a position that c-commands only the bare NP. The illformedness of the sentence shows that the quantificational force of the bare NP is determined by the closer binder kanarazu, rendering the “donkey interpretation” impossible. The generalization is that indefinites are unselectively bound by the closest binder possible. It turns out that the wh-island effects on NP wh-in-situ we have seen in connection to (22) are merely minimality effects in disguise (see Nishigauchi, 1990, among others). The fact is that the closest unselective binder determines the scope and the quantificational force of a wh-phrase in Japanese.13 Thus, even if we eliminate one of the wh-phrases in (22), as in (35), the other remaining wh-phrase must take scope over the embedded clause. (35) Hana-wa [Ai-ni nani-o ageta ka] Hana-TOP Ai-DAT what-ACC gave Q wasuremasita ka? forgot Q ‘Did Hana forget what she gave to Ai?’

< previous page

page_224

next page >

< previous page

page_225

next page >

Page 225 This is so since the Q-element ka in the embedded Comp is closer to the wh-phrase than that in the matrix Comp. The ungrammaticality of examples like (36) is amenable to the same line of analysis. (36) ?*Hana-wa [Ai-ni nani-o ageta kadooka] Hana-TOP Ai-DAT what-ACC gave whether wasuremasita ka? forgot Q (‘What did Hana forget whether she gave to Ai?’) Given that the first instance of ka in kadooka “whether” is a Q-element (Watanabe, 1992), (36) is ruled out by minimality: kadooka qualifies as the closest potential unselective binder for the wh-phrase and thus blocks binding from outside the embedded clause, though it is arguably incompatible with the feature [+WH].14 Despite its success with NP wh-in-situ, the unselective binding approach does not extend to DP wh-insitu. In particular, it is unable to explain the fact that they are susceptible to the TC and the CED. Summary The preceding discussion leads us to conclude that no unified analysis of the syntax of wh-in-situ utilizing a single licensing mechanism is possible: DP wh-in-situ uses NO movement, whereas NP wh-insitu uses unselective binding. The next section elaborates on the former from the viewpoint of feature movement (F-movement). We have nothing more to say about the latter here (see Tsai, 1994; Reinhart, 1998; and references cited therein). NULL OPERATOR MOVEMENT AS FEATURE MOVEMENT It was observed earlier that the syntactic behavior of an in-situ DP wh-phrase resembles that of a NO. This indicates that some form of pre-Spell-Out movement is involved in the DP wh-in-situ construction. In fact, we are forced into this thesis if we hypothesize that interrogative Comp is universally “strong” in the sense that it has to perform feature checking in overt syntax (Watanabe, 1992; Chomsky, 1995, 2000). Examples like Japanese (3a) are in accord with the hypothesis since the Q-marker is assumed to check the features of C (see Cheng, 1991). Examples like IA (4a), where the [+WH] C contains no Qmarker, must involve (invisible) movement before Spell-Out. Adapting Watanabe’s (1992) proposal, originally made for Japanese wh-words, let us assume that the DP wh-phrase in (11) can have the following alternative structure:

< previous page

page_225

next page >

< previous page

page_226

next page >

Page 226

(37), unlike (11), hosts a NO in Spec of DP, licensed by the D endowed with the [+WH] feature from the N under Spec-head agreement. I assume that structures like (11) without a NO are used when category movement of the whole wh-phrase takes place, as in the case of li-meno “to whom” in IA (23) (see [49] for another example).15 Given (37), the similarities between the DP wh-in-situ construction and the NO construction come as no surprise: They both involve pre-Spell-Out movement of a NO. Now the question is: What is this movement really like? In his discussion of IA, Basilico (1998) argues that the NO movement in question is a species of adjunct movement. His argument is based on the observation that the NO movement seems to behave in the same way as the extraction of the French wh-determiner combien “how many” out of a DP. In French, two options are available to form a combien question: One is to pied pipe the entire wh-DP, as in (38a), and the other is to extract only combien, leaving behind other elements within the DP, as in (39b) (see Obenauer, 1984; Rizzi, 1990). (38) a. [DP Combien des livres] a-t-il consulté t? how many of-the books has-he consulted ‘How many books has he consulted?’ b. Combieni a-t-il consulté [DP ti des livres]? how many has-he consulted of-the books ‘How many books has he consulted?’ Basilico maintains that the extraction in (38b) is the visible/audible equivalent of the NO movement. Following Rizzi (1990), he treats the extracted phrase in (38b) as an adjunct in the sense that it lacks a referential index. This treatment is allegedly supported by the following examples: (39) a. ?Combien des problèmes sais-tu [comment how many of-the problems know-you how a resoudre t t]? to solve ‘How many problems do you know how to solve?’

< previous page

page_226

next page >

< previous page

page_227

next page >

Page 227 b. *Combieni how many

sais-tu [comment a resoudre [DP ti know-you how to solve des problèmes] t] ? of-the problems (‘How many problems do you know how to solve?’) (39a-b) both involve extraction out of a wh-island. (39a), where the whole argument DP has been extracted, results in a slight deviance, probably a mild Subjacency violation. (39b), where only part of the DP, that is, combien has been extracted, is markedly worse. Under Basilico’s view, the contrast represents an argument/adjunct asymmetry of the familiar sort. If NO movement is analogous to movement of combien, the kind of contrast in (39) should be observable with DP wh-in-situ as well. This prediction seems to be borne out by the following pair ([40b] is repeated from [23]): (40) a. ?? Sheno nasat Mona [li-meno tinti]? what forgot Mona to-whom to-give ‘What did Mona forget to whom to give?’ b. *Nasat Mona [li-meno tinti sheno]? forgot Mona to-whom to-give what (‘What did Mona forget to give to whom?’) In (40a) the wh-phrase sheno “what” has moved overtly out of an island, whereas in (40b) it stays insitu. We can see the difference in grammaticality in (39) replicated in (40).16 Based on (40), then, Basilico (1998) concludes that a NO behaves like an adjunct. Although Basilico’s (1998) analysis is illuminating in many respects, it is still problematic since it fails to explain the discrepancy between the in-situ DP wh-phrase and the operator/adjunct in terms of extraction out of tensed clauses. Observe the following example (Obenauer 1984:181): (41) Combieni dis-tu qu’il a invité [DP ti de filles]? how many say-you that he has invited of girls ‘How many girls do you say he invited?’ The full grammaticality of (41) demonstrates that extraction of combien, like that of an adjunct but unlike that of a NO (see IA [6], English [28a]), is free from the TC. This casts serious doubt on the correctness of Basilico’s proposal. To deal with the tense effects on wh-in-situ in IA, Basilico (1998) essentially stipulates that tensed embedded clauses in the language undergo raising into the projection of AGRo (analogous to raising of

< previous page

page_227

next page >

< previous page

page_228

next page >

Page 228 object for structural Case checking) and hence barriers for CED-related reasons. Since adjunct movement out of tensed clausal complements is possible in languages like French and English, Basilico would have to introduce an ad hoc parameter that makes a distinction between the two groups of languages (cf. Basilico, 1998:336, fn. 37). Even if such a distinction could be made, the similarities between the DP wh-in-situ construction and the tough-construction would be completely missed.17 Therefore, we need another way of characterizing NO movement. I suggest, following Takahashi (1997), that NO movement should be characterized as F-movement rather than (phrasal) adjunct movement. Chomsky (1995) argues that the default case of movement is F-movement. This is a natural consequence of the theory of feature checking, where movement is triggered exclusively by the need to delete uninterpretable features under such strictly local relations as Spec-head and head-head configurations. Chomsky also argues, based on considerations bearing on bare phrase structure and Last Resort, that feature raising is adjunction to a head. Chomsky (1995:265) suggests that only PF convergence forces anything beyond features to raise. Following this suggestion, Takahashi (1997) maintains that since NOs lack phonological content and are free from PF-crash, their movement is realized as F-movement rather than category movement.18 Under the F-movement analysis, (25) involves the following derivation instead of that in (26) (FF stands for formal feature): (43) John is easy [CP FF[Op]-C to please [DPtFF[Op ]]] In (42) only the formal features of the NO have been raised. The raised features are adjoined to the head C of the clausal complement. Similarly, the IA example in (4a) involves the following derivation (with English words): (43) [CP FF[Op]-C [Mona saw [DPtFF[Op ] who]]] As in the case of (42), the features of the NO have raised and entered into a checking relation with the root C. Following Chomsky (1995), let us assume (44) as the definition of the transformational operation Attract:19 (44) Attract K attracts F iff F is the closest feature that can enter into a checking relation with a sublabel of K. The notion of closeness used in (44) is defined as follows (Chomsky, 1995:358):

< previous page

page_228

next page >

< previous page

page_229

next page >

Page 229 (45) Closeness: β is closer to the target K than α if β c-commands α. I suggest that the F-movement analysis offers a new way to deal with the tense effects observed in the case of DP wh-in-situ and tough-struc tures. It has been observed in the literature that there is a local selectional relationship between T and C (see for instance Chung and McCloskey, 1987, for Irish complementizers bearing tense features). Furthermore, some authors have suggested that the COMP position is where tense operators, if any, must appear (see Stowell, 1981; Besten, 1983; among others). Within the minimalist approach, the matching that holds between T and C can most naturally be interpreted as a result of feature checking. The intuitive idea is that in certain environments, the features of an NO cannot be attracted to C because there exists a closer feature on T that can enter into a checking relation with the C. Let us try to implement this idea. First, let us be more specific about the feature specifications of the elements at issue, that is, C, T, and an NO. I assume that an NO in (37) and in general bears at least two kinds of features, a [+WH] feature, which comes from the N part, and a [+OP] (operator) feature, which is associated with the D part. They are both interpretable. The [+OP] feature is what makes a wh-phrase an operator: Without it, a wh-phrase is an indefinite, as in Japanese. I assume further that finite T, just like its nominal counterpart D in (37) (see Enç, 1987, and Hornstein, 1990, among others for the parallel between D and T), is marked with an interpretable [+OP] feature. Four kinds of features associated with complementizers should be distinguished here. First, the [+Q] feature, which is interpretable, is found on C with interrogative force. Second, the [+WH] feature is used to mark the scope of wh-elements. Third, C has the [+OP] feature when it is used as an intermediate C in successive cyclic F-movement, which is a reasonable conjecture in light of the hypothesis that pre-Spell-Out movement is typically motivated by D-related features (Chomsky, 1995), or when it selects finite T. Fourth, the [+EPP] feature on C triggers category movement rather than F-movement (Chomsky, 1995, 2000). The last three features are all uninterpretable and hence are able to trigger attraction. With this much assumed, consider the following ill-formed structures of IA (6) and English (27a) with the relevant features indicated: (46)a.*[CP C[+Q, +WH,+OP] Mona thought [CP C[+OP] [TP T[+QP] Ali bought [DP Op[+WH,+OP]what]]]] b.*John is easy [CP C[+WH, +OP] to believe [CP C[+OP] [TP Mary T[+QP] kissed Op[+WH,+OP]]]

< previous page

page_229

next page >

< previous page

page_230

next page >

Page 230 What is wrong with these derivations concerns the impossibility of the F-movement of the NO to the lowest C: The finite T with the [+OP] feature counts as the closest element that can enter into a checking relation with the [+OP] C. One might think that the features of the NO could successfully be attracted by the [+WH] C once the uninterpretable [+OP] feature of the lowest C is checked off by the finite T. The putative attraction, however, is banned due to the general locality condition on movement (cf. Rizzi’s, 1990) Relativised Minimality or Chomsky’s (2000) Phase-Impenetrability Condition. Thus, the uninterpretable features [+WH, +OP] on C cannot be appropriately erased. Under the present analysis, the TC follows from the characteristics of Attract. Let us ask now why the IA examples in (4) are well formed. Consider their structures here: (47) a. [CP C[+Q, +WH,+OP] [TP T[+OP] Mona saw [DP Op[+WH,+op]who]]] b. Mona asked Ali [CP C[+WH,+OP] [TP T[+OP] Ro a bought [DP Op[+WH,+op]what]]] As in (46), there is a finite T that c-commands the NO in (47). Nonetheless, the attraction is possible thanks to the [+WH] feature on the C: if the computational system accesses this particular feature, the closest element that can enter into a checking relation with the C is the NO. This option is not open in the case of (46), where the lowest Comp lacks the [+WH] feature.20 This account also covers the facts surrounding phrasal wh-movement. For instance, consider the following structural representation for English (24a): (48)[CP C[+Q, +WH,+OP,+EPP] [TP John T[+OP] believe [CP C[+OP,+EPP] [TP Mary T[+OP] kissed who]]]] In (48) the EPP feature plays a crucial role. If the computational system accesses the feature, the finite T becomes virtually ignorable for the purpose of the attraction of the wh-phrase. This is why category wh-movement is not affected by the presence of finite T in (24a). It was pointed out in connection to (23) that IA has optional wh-movement. In the present context, this means that an EPP feature can optionally be added to C in IA. It is expected then that, as in (24a), phrasal wh-movement can take place out of a finite clausal complement in (6). This expectation is fulfilled (Wahba, 1991): (49) Shenoi tsawwarit Mona [Ali ishtara ti]? what thought Mona Ali bought ‘What did Mona think Ali bought?’

< previous page

page_230

next page >

< previous page

page_231

next page >

Page 231 (49) differs from (6) in that the former uses the category movement strategy. It is well formed for the same reason that (24a) is. Our proposal based on Attract extends naturally to the wh-island cases like those in English (27c) and IA (40b). The structure for (40b), for example, is given here: (50)*[CP C[+Q, +WH,+OP] [forgot Mona [CP to-whom[+WH,+OP] C [to-give [DP Op[+WH,+OP]what]]]]] In (50) it is impossible for the features of the NO to be attracted by the matrix C: No matter which feature is accessed, the [+WH] feature or the [+OP] one, the overtly moved wh-phrase in the intermediate Spec of CP acts as a closest potential checker for the matrix C. Thus, (40b) is illegitimate due to the definition of Attract. This explains why it induces strong ungrammaticality compared to (40a), which can probably be derived without wh-movement of sheno “what:” The wh-phrase may be merged into the matrix Spec of CP directly from the numeration and (marginally) bind an empty category in its θ-position (Rizzi, 1990). As has been argued (Lasnik, 1995; contra Chomsky, 1995), a set of formal features cannot be a legitimate binder in the relevant sense, and thus the binding strategy is not available in the case of (40b). Turning now to the CED effects on DP wh-in-situ, such as the one illustrated in Hindi (20b), the Fmovement analysis correctly captures them, given the assumption that F-movement, just like category movement, is subject to the CED. As Takahashi (1997) demonstrates, this assumption is justified by the fact that NO movement cannot take place out of subjects in English, as shown in (52) (see Browning, 1987, and references cited therein). (51) a. *John is easy to believe [t kissed Mary]. b. *John is easy to believe [t to have kissed Mary]. In this respect, NO movement contrasts sharply with movement of wh-phrases, as exemplified in (52). (52) a. Whoi do you believe [ti kissed Mary]? b. Whoi do you believe [ti to have kissed Mary]? The seemingly peculiar condition on NO movement can be accounted for in terms of the CED. Consider the following derivations for (51):

< previous page

page_231

next page >

< previous page

page_232

next page >

Page 232 (53) a. *John is easy [CP FF[Op]-C [PRO to believe [[DPtFF[Op]] kissed Mary]]] b. *John is easy [CP FF[Op]-C [PRO to believe [[DPtFF[Op]] to have kissed Mary]]] As is well known, subject position in English is not “properly governed” (see for instance Lasnik and Saito, 1984). In (53) the formal features of the NO have been extracted out of the DP not “properly governed,” leading to a CED violation.21 Then the ill-formedness of (51) is treated on a par with that of (54) in violation of the Subject Condition: (54) a. *Whoi did you believe [[DP a picture of ti] was on sale]? b. *Whoi did you believe [[DP a picture of ti] to be on sale]? In (54) the wh-phrase who has been extracted out of the subject DP, violating the CED. (52) is fine since the entire category of the wh-phrase has moved, which circumvents a CED violation.22 To summarize, this section has argued based on Takahashi (1997) that the movement of a NO out of DP wh-in-situ takes the form of F movement. The F movement analysis, in conjunction with the theory of Attract and the CED (or whatever replaces it in the minimalist theory; see Uriagereka, 1999), captures the locality conditions on DP wh-in-situ. CONCLUSION It has been argued that the two types of wh-in-situ—NP wh-in-situ found in Japanese-type languages and DP wh-in-situ found in Iraqi Arabic-type languages—require separate accounts utilizing distinct mechanisms available in Universal Grammar. The former, being indefinite, is licensed through unselective binding. The latter, on the other hand, needs to undergo feature movement. To the extent that the present discussion is on the right track, the lexical traits of an in-situ wh-phrase determine which mode of licensing it uses. This implies that the principles-and-parameters hypothesis is well motivated, that there is only one computational system of human language, and that linguistic variation is limited to variation in the lexicon (Chomsky, 1995, among others). Furthermore, the present analysis implies that the feature-based theory of syntactic dependencies, like the one advocated in the minimalist approach, gains support: Attract has proved useful in accounting for the syntax of one type of wh-in-situ (cf. Bošković, 1998). NOTES I would like to express my sincere gratitude to the editors of this volume, Mengistu Amberber and Peter Collins, for their valuable assistance. I would also

< previous page

page_232

next page >

< previous page

page_233

next page >

Page 233 like to thank an anonymous reviewer for a number of very helpful clarifications. Naturally, any errors are strictly my own. The work reported here was supported by a 2000–2002 Grant-in-Aid for Encouragement of Young Scientists #12710288 from the Japan Society for the Promotion of Science, for which I am grateful. 1. The following abbreviations are used here: ACC-accusative COMP-complementzser DAT-dative ERG-ergative GEN-genitive NOM-nominative Q-question marker QP-question particle TOP-topic 2. Overt wh-movement is optional in IA (Wahba, 1991; Ouhalla, 1996). Thus, (i) here is grammatical alongside (4a). (i) Meno Mona shaafat? who Mona saw ‘Who did Mona see?’ 3. One may wonder how the proposition that (6) should express is expressed in IA. (6) becomes well formed if the wh-phrase undergoes optional wh-movement, as in (49) later. See note 20 for another strategy for expressing the proposition in question. 4. It should be borne in mind that here we concentrate on nominal wh- phrases like those in (3)–(6). It is in their behavior that the two types of languages diverge most: In-situ adjunct wh-phrases behave more or less in the same way across languages (although there seem to be complications). See Tsai (1994), where it is argued along the lines suggested by Huang (1982) that adjunct wh-insitu must uniformly undergo covert movement. 5. Anoop Mahajan (personal communication) informs me that Hindi nominal wh-phrases are less transparent in terms of morphology than their IA counterparts. In particular, they do not seem to be composed of a wh-morpheme and a pronominal morpheme. Nonetheless, I assume that they are DP wh-phrases. It would be natural to say that the DP status of nominal wh-words is the default, which children assume in the absence of counterevidence. The NP status of wh-words is the marked case: Ample positive evidence, related, for example, to quantificational forces and negative polarity, is available to children acquiring NP wh-phrases. More generally, it can be suggested that nominal operators are universally DPs (see the discussion of null operators later). 6. See note 9. 7. It is beyond the scope of this chapter to discuss how the CED should be adequately recast in the current framework. For a minimalist reformulation of the CED compatible with the present analysis, see Uriagereka (1999). I will continue to use the term “proper government” for the sake of descriptive convenience. 8. Saito (1985) shows that Japanese does not obey the Subject Condition. 9. As Mahajan (1990) notes, finite clausal complements, unlike nonfinite ones, must undergo extraposition in Hindi. Then (14a) doubly violates the TC and the CED. On the basis of data like (14a), one may conjecture that the TC is somehow reducible to the CED. In fact, Basilico (1998) pursues precisely this line of analysis for IA. His account, however, is not free from problems. 10. According to Huang (1982), (i), a Chinese equivalent of (22), permits the interpretations where one of the two wh-phrases takes the matrix scope. (i) Ni xiang-zhida [shei mai-le sheme]? you wonder who bought what ‘You wonder who bought what?’

< previous page

page_233

next page >

< previous page

page_234

next page >

Page 234 If Huang’s judgment is correct, (i) would threaten a unified account of Japanese and Chinese wh-in-situ in spite of the fact that Chinese wh-phrases are like their Japanese counterparts in relevant respects (see Ouhalla, 1996). Nishigauchi (1990:32, fn. 13), however, notes that none of his Chinese informants accepted the matrix interpretations: For them the wh-phrases in (i) take scope only over the embedded clause. As pointed out by Nishigauchi (1990), focus may play an important role in allowing the matrix readings in (i) (and even in [22]). 11. As noted earlier, (18) violates (28b) as well. 12. For more criticism of Ouhalla’s (1996) analysis of IA, see Basilico (1998) who offers additional cases where the similarities between reflexive anaphors and wh-anaphors break down. 13. As mentioned in note 10, there seem to be interfering factors. See Nishigauchi (1990). 14. (36) improves significantly if the wh-phrase escapes the c-command domain of kadooka via scrambling, as shown here: (i)Nanii-o Hana-wa [Ai-ni agetakadooka] ti what-ACC Hana- Aigave whether TOP DAT wasuremasitaka? forgot Q, ‘What did Hana forget whether she gave to Ai?’ This state of affairs is expected under the unselective binding approach if the scrambled wh-phrase is allowed to remain in its surface position. The analysis extends naturally to paradigms of the following sort discussed by Hoji (1985): (ii)a.?*Dare-mo- dare-o syootaisimasitaka? ga who-MO- who-ACC invited Q NOM (‘Who did everyone invite?’) b. Darei-o daremo-ga syootaisimasitaka? ti who-ACC who-MO- invited Q NOM ‘Who did everyone invite?’ In (iia) the Q-marker is unable to bind the whphrase due to the intervening element daremo “everyone” containing a potential (but inappropriate) binder (the morpheme mo) (see [7] and [8b]). As in the case of (36), (iia) can be “rescued” by scrambling the wh-phrase out of the c-command domain of the nominative quantifier ([iib]). Watanabe’s (1992) analysis of (36) as a Subjacency violation clearly fails to capture the resemblance between (36) and (iia). 15. Thus the optionality of overt wh-movement in IA is only apparent. No economy considerations apply to, say, (4a) and (i) in note 2, since they arise from different numerations. 16. The difference between (39a) and (40a) in the degree of deviance, if real, may be attributable to the fact that the wh-phrase in (39a) is D(iscourse)-linked in the sense of Pesetsky (1987), whereas that in (40a) is not. 17. The defect is in fact much more severe than it appears, since (a certain type of) wh-in-situ in French and English, just like that in IA, is constrained by the TC. Bošković (1998) shows

that French wh-in-situ exhibits tense effects.

< previous page

page_234

next page >

< previous page

page_235

next page >

Page 235 (i) a

Tu as vu qui? you have seen whom ‘Who did you see?’ b. *Jean et Pierre croient que Marie a vu Jean and Pierre believe that Marie has seen qui? whom (‘Who do Jean and Pierre believe that Marie saw?’) This is totally unexpected under Basilico’s (1998) analysis (recall the grammaticality of French [41]). Under the present account, French wh-in-situ is assumed to be of the DP type and involves NO movement. Bošković argues that it involves F-movement, a position we will argue for shortly, though his theory and ours differ in details. Even in English, finite T defines an opaque domain for wh-in-situ within a pied-piped constituent, as shown in (ii) (Nanni and Stillings, 1978). (ii)a.The elegant parties, [to be admitted to one of which] was a privilege, had usually been held at Delmonico’s. b.*They bought a car, [that their son might drive which] was a surprise to them. Given the strict locality imposed on feature checking, it must be that in (iia) the relevant feature of the in-situ wh-phrase is somehow “transferred” to the head of the pied-piped constituent. Within the minimalist framework, there is no place for feature percolation. Then we are led to assume that F-movement out of the wh-phrase to C takes place overtly. Again, facts like (iib) go against Basilico’s proposal. What remains to be explained is the ungrammatically of IA (iii) (Wahba, 1991). (iii) *Leeshi tsawwarit Mona [Ali masha ti]? why thought Mona Ali left (‘Why did Mona think Ali left?’) Adjunct extraction, unlike argument extraction (see [49]), out of tensed clauses is prohibited in IA. I must leave the matter open here, largely because tense effects on phrasal wh-movement belong to one of the most poorly understood areas. In fact, the reverse situation holds in the northern German dialect described by McDaniel (1989) where extraction of nominal wh-phrases out of indicative clausal complements is banned, whereas corresponding extraction of adverbial and prepositional wh-phrases is fine. 18. Nakamura (1999) argues that an NO is nothing but a bundle of formal features, and its movement is not constrained by global economy (cf. Takahashi’s [1997] discussion of PRO). What matters for the present purpose is the assumption that NO movement is F-movement. 19. Chomsky (1995:297) uses an if rather than an iff in his original definition. (44) follows a suggestion made by an anonymous reviewer.

< previous page

page_235

next page >

< previous page

page_236

next page >

Page 236 20. The impossibility of the embedded C in (46a) bearing a [+WH] feature is evidenced by the illformedness of (i): (i) *Mona tsawwarat [shenoi Ali ishtara ti]? Mona thought what Ali bought (‘What did Mona think Ali bought?’) Recall that phrasal wh-movement is optional in IA. If the C in question could be specified as [+WH], (i) would be grammatical. It is worth pointing out that the text analysis readily explains IA wh-questions involving the whscope marker sheno “what” or its contracted form sh- . Consider (ii) (Wahba, 1991): (ii) a. Sh-tsawwarit Mona [Ali ishtara sheno]? QP-thought Mona Ali bought what ‘What did Mona think Ali bought?’ b. Sh-tsawwarit Mona [shenoi Ali ishtara ti]? QP-thought Mona what Ali bought ‘What did Mona think Ali bought?’ As shown in (iia), ill-formed (46a) can be “rescued” by using the wh-scope marker in the matrix clause (see Mahajan, 1990; Dayal, 1996, for parallel facts in Hindi). Let us assume with Mahajan (1996) and others that the marker is a “ wh-expletive” associated with a CP whose head C has the feature [+WH] (see also McDaniel, 1989). Then the well-formedness of (iia) is on a par with that of (47b); in both the embedded C is specified as [+WH]. The [+WH] status of the clausal complement in (iia) is supported by (iib) with the wh-movement. 21. (53a) is not ruled out by the TC, now reduced to the definition of Attract, since the subject DP in Spec of TP c-commands the finite T. 22. A question arises as to data like (i) from Hindi, which contain an in-situ subject wh-phrase in a finite embedded clause (Mahajan, 1990:128). (i) *Raam-ne socca [ki kOn aayaa hE]? Ram-ERG thought who has come (‘Who did Ram think has come?’) In particular, is (i) excluded by the definition of Attract (as in the case of IA (6) and English [27a]) or by the CED (as in the case of English [51a])? There are reasons to believe that it is excluded by the former. Consider (ii), in which the Hindi question particle kyaa “what” has been added to (i) (Mahajan 1990:129). (ii) Raam-ne kyaa socca [ki kOn aayaa hE]. Ram-ERG KYAA thought who has come ‘Who did Ram think has come?’ Just like its counterpart in IA (see note 20), kyaa marks the embedded C with a [+WH] feature, making the feature attraction out of the wh-phrase possible. Then the well-formedness of (ii) implies that subject position is “properly

< previous page

page_236

next page >

< previous page

page_237

next page >

Page 237 governed” in Hindi and that the ill-formedness of (i) must be attributed to Attract. It is plausible that structural subjects in the language (unlike those in English) do not undergo overt raising into Spec of TP and hence remain c-commanded by T. This assumption is consonant with their “properly governed” status (see Mahajan, 1990). REFERENCES Aoun, J. 1986. Generalized Binding: The Syntax and Logical Form of Wh-Interrogatives. Dordrecht: Foris. Basilico, D. 1998. Wh-movement in Iraqi Arabic and Slave. The Linguistic Review 15:301–39. Besten, H.den. 1983. On the interaction of root transformations and lexical deletive rules. In W.Abraham (ed.) On the Formal Syntax of Westgermania. Amsterdam: John Benjamins, 47–131. Bošković, Z. 1998. LF movement and the Minimalist Program. In Proceedings of NELS 28. Amherst: GLSA, University of Massachusetts, 43–57. Browning, M. 1987. Null Operator Constructions. Ph.D. dissertation, MIT. Cheng, L.L.-S. 1991. On the Typology of Wh-Questions. Ph.D. dissertation, MIT. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. ——. 2000. Minimalist inquiries: The framework. In R.Martin., D.Michaels, and J.Uriagereka (eds.) Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press, 89–155. Chung, S., and McCloskey, J. 1987. Government, barriers, and small clauses in Modern Irish. Linguistic Inquiry 18:173–238. Dayal, V. 1996. Locality in WH Quantification. Dordrecht: Kluwer. Enç, M. 1987. Anchoring conditions for tense. Linguistic Inquiry 18:633–58. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Hoji.H. 1985. Logical Form Constraints and Configurational Structures in Japanese. Ph.D. dissertation, University of Washington, Seattle. Hornstein, N. 1990. As Time Goes By. Cambridge, MA: MIT Press. Huang, C.-T.J. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D. dissertation, MIT. Kamp, J.A.W. 1981. A theory of truth and semantic representation. In J. Groenendijk, T.Janssen, and M.Stokhof (eds.) Formal Methods in the Study of Language. Amsterdam: Mathematical Centre, 277–321. Katada, F. 1991. The LF representation of anaphors. Linguistic Inquiry 22:287–313. Kuroda, S.-Y. 1965. Generative Grammatical Studies in the Japanese Language. Ph.D. dissertation, MIT. Lasnik, H. 1995. Last resort and attract F. In Proceedings of the Sixth Annual Meetings of the Formal Linguistics Society of Mid-America. Bloomington: Indiana University Linguistics Club, 62–81. Lasnik, H., and Saito, M. 1984. On the nature of proper government. Linguistic Inquiry 15:235–89. Lewis, D. 1975. Adverbs of quantification. In E.Keenan (ed.) Formal Semantics of Natural Language. Cambridge, MA: Cambridge University Press, 3–15. Mahajan, A. 1990. The A/A-bar Distinction and Movement Theory. Ph.D. dissertation, MIT. ——. 1996. Wh-expletives and the syntax of partial wh-movement. In U.Lutz and G.Müller (eds.) Papers on wh-Scope Marking. Arbeitspapiere des

< previous page

page_237

next page >

< previous page

page_238

next page >

Page 238 Sonderforschungsbereichs 340, July 1996, Universität Stuttgart and Universität Tübingen, 163–77. McDaniel, D. 1989. Partial and multiple wh-movement. Natural Language and Linguistic Theory 7:565– 604. Nakamura, M. 1999. On feature movement. Ms., Senshu University, Kawasaki, Kanagawa. Nanni, D., and Stillings, J. 1978. Three remarks on pied piping. Linguistic Inquiry 9:310–18. Nishigauchi, T. 1990. Quantification in the theory of grammar. Dordrecht: Kluwer. Obenauer, H.-G. 1984. On the identification of empty categories. The Linguistic Review 4:153–202. Ouhalla, J. 1996. Remarks on binding properties of wh-pronouns. Linguistic Inquiry 27:676–707. Pesetsky, D. 1987. Wh-in-situ: Movement and unselective binding. In E.J.Reuland and A.G.B.ter Meulen (eds.) The Representation of (In)definiteness. Cambridge, MA: MIT Press, 98–129. Reinhart, T. 1998. Wh -in-situ in the framework of the Minimalist Program. Natural Language Semantics 6:29–56. Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT Press. Saito, M. 1985. Some Asymmetries in Japanese and Their Theoretical Implications. Ph.D. dissertation, MIT. Stechow, von A. 1996. Against LF pied-piping. Natural Language Semantics 4:57–110. Stowell, T. 1981. Origins of Phrase Structure. Ph.D. Dissertation, MIT. Takahashi, D. 1997. Move-F and null operator movement. The Linguistic Review 14:181–96. Tsai, W.-T.D. 1994. On Economizing the Theory of A-bar Dependencies. Ph.D. dissertation, MIT. Uriagereka, J. 1999. Multiple spell-out. In S.Epstein and N.Hornstein (eds.) Working Minimalism. Cambridge, MA: MIT Press, 251–82. Wahba, W. 1991. LF movement in Iraqi Arabic. In C.-T.J.Huang and R.May (eds.) Logical Structure and Linguistic Structure: Cross-linguistic Perspectives. Dordrecht: Kluwer, 253–76. Watanabe, A. 1992. Wh-in-situ, subjacency and chain formation. MIT Occasional Papers in Linguistics 2, MIT.

< previous page

page_238

next page >

< previous page

page_239

next page >

Page 239 10 Vowel Place Contrasts Keren Rice An important goal in phonology is to achieve an understanding of the mental representation of speech and the sounds that make it up. There has long been agreement that sounds should not be viewed as atoms, but are composed of more basic primitives. Many debates about the nature of these primitives, or features, are now at the forefront of phonological research. Consider just a few: What is the phonetic basis for features? Is it primarily articulatory, acoustic, or perceptual? Can these different perspectives be captured with a single multifaceted feature set, or must different feature sets be required to account for each of the aspects of sound? What is the role of distinctiveness in the grammar: is it just distinctive features, whatever their foundations, that are required, or must non-distinctive features play a role in grammar as well? In this chapter, I focus on place in vowel systems, examining the features required to account for the phonology of vowel place. The representation of vowel place has been a major topic in recent phonological research (e.g., Clements, 1991; Clements and Hume, 1995; Halle, 1995; Harris and Lindsey, 1995, 2000; van der Hulst, 1996; Hume, 1992; Kaye et al., 1985; Ní Chiosáin and Padgett, 1997; Odden, 1991; Sagey, 1986; Selkirk, 1992). I address two issues. The first concerns the phonological features that mark place in vowels. Most recent North American work assumes the need for three features: one marking labiality, one velarity, and one palatality. Such an analysis is assumed by Chomsky and Halle (1968), using the binary features [±back] (palatality, velarity) and [±round] (labiality); it is carried over into work by Sagey (1986), who assumes that unary features Labial (labiality), Coronal (palatality), and

< previous page

page_239

next page >

< previous page

page_240

next page >

Page 240 Dorsal (velarity) mark vowel place, and, with somewhat different definitions of the features, in work by Clements (1991), Clements and Hume (1995), and Hume (1992). Steriade (1995) states unequivocally that features of labiality, velarity, and palatality are necessary phonologically, and treats velarity and palatality as equipollent. I argue that only two place features are needed for vowels phonologically, one marking palatality (Coronal) and one non-centrality.1 I refer to the latter feature as Peripheral, a feature that is similar to the Jakobson et al., (1969) [grave]. (Peripheral does not refer to the vowels at edges of the vowel space, namely [i a u], but to what is realized as labiality and dorsality.) In other words, Labial, marking labiality, and Dorsal, marking velarity, do not play a role in vowels phonologically, but instead compose a single feature, Peripheral. Vowel place is thus fully described with the phonological features in (1a), with two distinctive features for place, rather than those in (1b), with three distinctive place features. (1) Phonological features: a. Proposed b. Clements 1991, Clements and Hume 1995 Coronal Coronal Peripheral Labial Dorsal The features in (1a), I argue, allow for the set of phonological place systems found in languages of the world, providing exactly the major category distinctions that occur. In this sense I return to the view espoused by Jakobson et al. (1969) and Jakobson and Halle (1956), that distinctive features are cover terms for possibly different phonetic realizations, creating some level of abstraction in the phonological system. More specifically, in the absence of an opposition between, say, / / and /u/ in any language system, these phones should be treated as phonetic variants of the same phonological feature. The proposed features Coronal and Peripheral can occur individually, combined, or not at all to define vowel place systems. While the features in (1a) provide an account of phonological facts of language, they do not allow for the detailed articulatory and acoustic facts, and it is necessary to address the phonetics of vowel systems as well. In the second part of the chapter, I examine the relationship between the phonological representation and its phonetic realization, concentrating on Peripheral. While the three-way place distinction is not necessary phonologically, different peripheral vowels occur nondistinctively. I argue that Labial and Dorsal do enter into vowel representations: They may optionally serve as phonetic implementation features, features that serve to implement backness and roundness, leading to a greater distance between the peripheral vowel and the other vowels of the system, an aid to

< previous page

page_240

next page >

< previous page

page_241

next page >

Page 241 perception (e.g., Stevens et al., 1986; Stevens and Keyser, 1989; Lindblom, 1986, 1990a, 1990b; Flemming, 1995). EVIDENCE FOR PERIPHERAL I now turn to arguments that Labial and Dorsal (or equivalent features) in fact do not play independent roles in the phonology. In particular, I argue for the feature structure in (2a) rather than those in (2b) or (2c) as the depiction of the full phonological place structure of vowels.

The Motivation for Features What motivates the introduction of a feature? There is general agreement on certain criteria: (1) features must be phonetically motivated, (2) they must capture natural classes, and (3) they must permit any two sounds that contrast to be represented. The criterion of capturing contrasts is a complex one. It is generally agreed that features should capture contrasts that exist in any language. In determining the set of features, however, what is it that one chooses to examine? Two hypotheses come to mind. First, it is possible to take the set of contrasts found in all languages and determine the features required based on this set. Second, it could be that to determine the set of required features, one examines the contrasts within a particular language and those that are never found within a language, and bases the distinctive feature system on this, as proposed by Jakobson et al. (1969). Suppose, for the sake of argument, that both [f] and [Φ] exist cross-linguistically, but never enter into a phonological contrast in a single language. Under the first hypothesis, phonological features are required to describe a bilabial/labiodental contrast, while under the second, this fine degree of features is not needed since the sounds are not distinctive within any language; they are both appropriately labial, with the differences between them involving phonetic spellout.

< previous page

page_241

next page >

< previous page

page_242

next page >

Page 242 Much recent work on vowel place features assumes the first hypothesis. For instance Clements (1991), in an examination of vowel features, argues for the need for three features—labial, coronal, and dorsal —to capture vowel place contrasts. Clements examines the vowels found in the languages of the world and shows that to differentiate them, three features are required. Using high vowels to represent the various places, he proposes the phonological representations in (3). (3) i y u labial + + + coronal + + dorsal + + Clements shows that contrasts between pairs of these vowels occur in languages; to take some of the more unusual contrasts, in the Swedish dialect of Finland, the contrast /y/, / /, /u/ occurs; in Nimboran (Austronesian), both / / and / / are found. Interestingly, however, Clements does not cite languages where the two central vowels contrast, nor languages where the two back vowels contrast. Thus his claim is basically that the phonological representation of a vowel is determinable from its phonetic realization. Lass (1984), working within a different framework, makes a claim that results in a similarly large possible phonemic vowel inventory. Lass argues for two principles. The first, the Phonemic Availability Principle, is based on the notion of contrast. (4) Phonemic Availability Principle: Any pair of qualities that contrast phonemically in some language are primitive. (Lass, 102) The term “primitive” refers to “irreducible, prohibited from, ‘counting as a variety/type’ of something else” (102). This principle is designed to limit the number of possible basic vowel types. Pairwise comparisons allow Lass to establish not only the well-accepted contrasts (e.g., [i]:[u], [y]:[u]) but also some that he claims might be disputed, such as the high vowels, [ ]:[ ] and [ ]:[u]. Under this principle, the basic phonemes are established on the basis of single language systems, consistent with the first hypothesis discussed previously. To return to the example of labial fricatives, if /f/ and /Φ/ are never contrastive in any language, they need not have a distinctive feature that distinguishes them. Having introduced a principle that requires that features be established on the basis of contrasts within particular languages, Lass goes on to discuss phonetic distinctions like [a]/[æ], pointing out that he knows of no languages in which these two phones clearly contrast. One might then expect that they would be represented by a single primitive, or

< previous page

page_242

next page >

< previous page

page_243

next page >

Page 243 feature. However, Lass argues that this is not correct. Because this distinction can differentiate dialects, and signal regional and chronological differences, and be audibly distinct, he proposes that the phones [a]/[æ] should be taken to represent a primitive distinction. This is not accounted for by the Phonemic Availability Principle, and he thus introduces a second principle, the Dialect Distinguishability Principle, given in (5). (5) Dialect Distinguishability Principle: Any pair of qualities capable of consistently signaling a dialect difference within one language, and therefore available as speaker choices, are primitive. (Lass, 103) This principle requires that one take dialect differences into account. To return to the bilabial-labiodental fricative example, if one dialect of a language uses [f] where another uses [Φ], this would be enough to necessitate a feature to distinguish a bilabial from a labiodental even if they never contrast phonologically within a language. With respect to vowels, the combination of the Phonemic Availability Principle and the Dialect Distinguishability Principle leads Lass to propose a system with five degrees of backness (rounding omitted): front, centralized front, central, centralized back, and back. The full system that Lass proposes is shown in (6), where F abbreviates front, B back, R round, U unround, and C centralized. (6) FU FR FCU FCR CU CR BCU BCR BU BR i y ï ÿ u ū u While the phonemes distinguished under the Phoneme Availability Principle capture contrasts in some language, the Dialect Distinguishability Principle adds to this set. The introduction of this second principle leads to a large set of features (one could capture these contrasts using the features front [or coronal], back [or dorsal], round [or labial], and central). Having introduced these features, or primitives in Lass’s terms, we are now left with the following dilemma. If primitives are given by both the Phonemic Availability Principle and the Dialect Distinguishability Principle, what limits the contrasts within any particular dialect to those given by the first of these principles? Lass is concerned with audibility: If two sounds can be distinguished within a language, even if they do not contrast within a dialect, phonological features must exist to capture this. The proposal that I present rests instead on categoricity: Categories are distinguished rather than the details of what can be perceived or produced. To study this, I now examine occurring inventories in languages.

< previous page

page_243

next page >

< previous page

page_244

next page >

Page 244 Contrasts in Inventories: Evidence for Peripheral When phonological vowel place is examined cross-linguistically, several points stand out. First, no language or dialect is reported to have more than four distinctive places of articulation at any single vowel height. Second, no language is reported to have contrasting central vowels at a particular height. A third point, that no language/dialect has contrasting back vowels at a particular height, is less immediately apparent, and I return to this next. I begin by examining the number of places of articulation found cross-linguistically, drawing on a representative sample from Maddieson (1984). The inventories in Maddieson represent surface realizations, without allophony. A few notes are in order. First, I take all examples from high vowels; other heights are similar in their patterning. Second, I abstract away from several factors including length, tongue root features, tenseness, nasalization, pharyngealization, and the like. An examination of inventories shows the following. (Note: F = front, U = unround, R = round, B = back, C = central). (7) See Table 10–1. Table 10–1 illustrates the maximal place inventory at any particular height within a language. The overall generalizations are summarized in (8) in phonological terms. (8)•The maximal number of phonological oppositions for place at a particular height within a language is four. •Within the front vowels, a language may have a phonological opposition between an unround and a round vowel at a particular height. •Within the central vowels, no single language has a phonological opposition between an unround and round vowel at a particular height. The back vowels appear to pattern together with the front vowels, as some of the languages listed have both an unround and a round back vowel. However, when the languages listed as having two back vowels are examined more carefully, and from a phonological perspective, this generalization appears to be wrong; instead the following generalization holds. (9)•Within the back vowels, no single language has a phonological opposition between an unround and round vowel at a particular height. A number of languages are reported by Maddieson (1984) to have both a phonetic back unround and a back round vowel. I discuss a few of these,

< previous page

page_244

next page >

< previous page

page_245

next page >

Page 245 TABLE 10–1.

FU FR cu CR BU BR Single place of articulation Adygh Two places of articulation Chipewyan, Ainu, Aleut, Greek, Komi, Moro, Bambara i u Jaqaru, Nungghbuyu, Japanese, Alawa, Nez Perce, Adzera, Ocaina i Three places of articulation Amahuaca, Lappish, Kashmiri, Kanuri, Saek, Cham, Rukai i u Tabassaran, Kyuri, Albanian, French, Cheremis, lai i y u Nemboi, Tsou, Woleian, Yay i u Nimboran, Nambakaengo i Vietnamese, Longchow, Dafla, Ao, Ostyak i u Four places of articulation Tavgy, Tuva i y u Osmanli, Chuvash, Turkish, Yakut, Korean i y u Swedish, Norwegian i y u Five, six places of articulation No examples found suggesting that they are really central-back contrasts phonologically. By “central” I mean the following: The vowel is without a specified place of articulation phonologically. The revisions are thus based on facts of phonological patterning rather than on phonetic realizations, consistent with the goal of examining phonological features. First, consider Korean. While Maddieson (1984) reports that it has this contrast phonetically (based on Martin, 1951; Cho, 1967; Martin and Lee, 1969), in more recent work Korean is said to have a phonological central unround rather than a back unround vowel based on assumptions about harmony targets (e.g., Cho and Iverson, 1997). Vietnamese, too, appears to be better analyzed as having a central-back contrast rather than a contrast within the back vowels phonologically. In this case, this claim is based on vowel neutralization in some Vietnamese dialects. Pham (1998) shows that both short front vowels and short back round vowels centralize in certain environments, surfacing as the vowel notated

< previous page

page_245

next page >

< previous page

page_246

next page >

Page 246 with the high back unround vowel symbol, [ ]. She argues that this vowel is phonologically central rather than back, based on its phonological patterning as a neutralization target. Similar evidence can be adduced in languages such as Turkish and Yakut. In these languages, vowel harmony is generally assumed to target the vowels that are written with the symbols [ ] and [a], and based on feature systems with just the feature [back] (Chomsky and Halle, 1968), it is assumed that these vowels are back. However, more recent work treats this vowel as phonologically central, notating it with [ ] (e.g., Clements and Hume, 1995, 290, Ní Chiosáin and Padgett, 1997). Thus, phonological patterning suggests that in the languages that are said to contrast /u/ and /[ ]/, the phonological contrast is in fact between a back vowel and a central vowel. This allows for the revision of two rows in (10) Table 10–2 in phonological rather than phonetic terms. (10) TABLE 10–2. Revision (to [7]) FU FR CU CR BU BR Vietnamese, Longchow, Dala, Ao, Ostyak i u Osmanli, Chuvash, Turkish, Yakut, Korean i y u The generalizations given in (8) can be amended as follows: (11)•The maximal number of oppositions for place of articulation at a particular height within a language is four. •Within the front vowels, a language may have a phonological opposition between an unround and a round vowel at a particular height. •Within the central vowels, no single language has a phonological opposition between an unround and round vowel at a particular height. •Within the back vowels, no single language has a phonological opposition between an unround and round vowel at a particular height. Contrasts in Vowel Systems: Evidence for Peripheral 1 I now turn to an account of these facts, comparing two hypotheses, a three-feature model (Coronal, Labial, and Dorsal are required) and a two-feature model (Coronal and Peripheral). Note that the arguments against the three-feature model also hold against the four-feature hypothesis required by Lass.

< previous page

page_246

next page >

< previous page

page_247

next page >

Page 247 Clements (1991), Hume (1992), and Clements and Hume (1995), among others, argue that three place features, Labial, Coronal, and Dorsal, are required to capture the place of articulation distinctions found in vowels. They further assume that two of these features, Coronal and Dorsal, cannot co-occur, and I simply accept this assumption. Under this hypothesis, with the co-occurrence of Coronal and Dorsal ruled out by constraint, the following feature combinations are possible (Table 10–3). (12) TABLE 10–3. front unround: front round: Coronal central unround: central back unround: back round: Labial Coronal and Labial no feature round: Labial Dorsal and Dorsal i y u This model predicts that the round/unround contrast is available within a language for front, central, and back vowels. It thus accurately predicts the places that occur when all languages are taken into account by providing a distinct representation for each vowel that can contrast within this large inventory. However, when the actual systems in (7) and (10) are considered, this model is problematic: It predicts that one might find a single language with up to six places of articulation at a particular height. The problem then with a system that utilizes three (or more) features to capture vowel places is that, although it correctly predicts the range of vowel place qualities found in language, it does not say anything directly about the range of vowel qualities allowed in a particular phonological system, either language or dialect. The two-place model offers a way of dealing with this. Consider its predictions: The combinatorial possibilities are as in (13). (13) Coronal Coronal/Peripheral no features Peripheral i y u This model predicts that in any particular language it would be possible to have a front unround vowel, a front round vowel, a single central vowel, and a single back vowel, for a maximum of four possible places of articulation. As shown in (7) and (10), this is exactly what is found. Rather than treat the difference between, say, the back unround and the back round vowel as a matter for the phonology, as, for instance, Clements (1991) does, under this model the difference between these must fall to the phonetic component: The claim is that no language has these

< previous page

page_247

next page >

< previous page

page_248

next page >

Page 248 vowels distinctively as argued in the section The Phonetic Realization of Peripheral Vowels, below. One might attempt to maintain a three-feature model and deal with the fact that it overgenerates possible inventories through constraints such as the complexity constraints proposed in Prince and Smolensky (1993). Clements and Hume (1995) introduce a constraint, *Coronal/Dorsal, which prohibits the co-occurrence of these two features. Consider, however, the gaps in (7): No more than one central vowel and no more than one back vowel (of the same height) can occur contrastively in a phonological inventory. The constraints required to rule out such inventories do not prohibit the co-occurrence of two features on a single segment, and thus are not constraints against complexity. Rather the constraints must ban two central vowels or two back vowels from co-occurring within a system, while allowing the co-occurrence of two front vowels. One might imagine constraining the phonological inventory to the particular four places that are found in a different way, through the minimal distance constraints proposed by Flemming (1995). Flemming proposes that auditory features are required in phonology in addition to articulatory features, and he bases place of articulation features on F2. Could an auditory feature set be more successful than the three-place hypothesis in allowing for /i/-/y/ oppositions, while at the same time ruling out / /-/ / and / /-/u/ oppositions? Catford (1988:162) provides the following F2 values for the high vowels. (14) TABLE 10–4. unround vowel F2 round vowel F2 i 2400 y 2100 1600 1350 1390 u 595 Let us consider the consequences if F2 values are taken into account in terms of phonological representations. Recall that it is only within the front vowels that a two-way opposition between round and unround is possible phonologically; such a distinction does not exist phonologically at the other major places of articulation. Table 10–4 (14) shows that if F2 distances are relevant to the phonology, this is not what one would expect, as the front vowels are in fact closer to one another in terms of F2 distance than the back vowels are. One might expect, then, based on F2 distances alone, that a back round-back unround distinction would be the optimal contrast within a place. Auditory features (based on F2) coupled with distance constraints do not produce the appropriate phonological contrasts, just as the three-place feature hypothesis does not. Thus, assuming a larger feature inventory, whether it is the three features

< previous page

page_248

next page >

< previous page

page_249

next page >

Page 249 proposed by Clements and Hume (1995) and others or the auditory features of Flemming (1995), the constraints required to rule out the unattested phonological oppositions rest on shaky ground. Whatever constraint system is used, it is necessary to have a motivated feature system. This particular work contributes to feature theory, a theory that stands independently of issues of constraints and representations. I do not pursue a constraint-based analysis here, but instead examine an account that grounds the asymmetric patterning of front vowels at least partially in the feature system available to vowels. Contrasts in Secondary Articulations: Evidence for Peripheral 2 The two-place model receives additional support from secondary articulations. Consider the predictions made by the two theories with respect to secondary articulations. (15) Secondary articulations: predictions of the Labial/Coronal/Dorsal model A single language can show distinct labialization, coronalization (palatalization), and velarization (ignoring possible combinations of these). (16) Secondary articulations: predictions of the Coronal/Peripheral model A single language can have distinct coronalization and labialization or velarization (or both combined), but cannot have both labialization and velarization distinctively (again, ignoring possible combinations). When secondary articulations are examined, the following generalization can be made. (17) No language has both distinctive labialization and velarization. (Hyman, 1975; Jakobson et al., 1969; Maddieson, 1984). Some possible combinations of secondary articulations are given in (18) Table 10–5. A comma means that the language illustrates independently the secondary articulations named; a slash indicates that these are combined in that language. For instance, “labialization, velarization” means that these are distinct secondary articulations, while “labialization/velarization” indicates that they are non-distinct. Data are from Maddieson (1984). (18) See Table 10–5 on next page. Table 10–5 in (18) reveals that within a language, a single consonant can be labialized and velarized simultaneously, but a single phone cannot

< previous page

page_249

next page >

< previous page

page_250

next page >

Page 250 TABLE 10–5. Secondary Articulations coronalization Russian labials, dentals, velars; Irish velars labialization Chipewyan velars velarization ? labialization/velarization Ponapean labials, Menominee coronalization, labialization Nambakaengo labials, dentals, velars coronalization, velarization Irish dentals coronalization, labialization/velarization Irish labials, Nupe *labialization, velarization *coronalization, labialization, velarization occur distinctively in both a labialized and a velarized form. This is strikingly different from coronalization, which appears independently. Under the three-place model, there is no reason to expect this asymmetry: Labial, Coronal, and Dorsal should all be able to function independently as secondary articulations, and three secondary articulations should be possible with a given phone within a language. Under the two-place model, the facts are predicted: two phonological secondary articulations are possible, Coronal and Peripheral. The Peripheral secondary articulation is realizable in different ways (see the section below, The Phonetic Realization of Peripheral Vowels). No further simple secondary articulations are predictable. Summary The survey of occurring places for high vowels and secondary articulations provides evidence for the two-place model. While the three-place model offers a good account of the phonetics of languages, it over-generates existing phonological patterns. The two-place model shows a fit between predicted places in any single language and actual contrasts, accounting for the absence of certain types of contrasts. What remains for the two-place model is to account for how, for instance, a vowel that is Peripheral can be realized as back unround in one language and as back round in another. Before addressing this issue, I examine some other sources of evidence for the two-place model. Evidence from Markedness Considerations: Evidence for Peripheral 3 An assumption that is often made in phonology is that phonological markedness is structurally represented: The more complex a structure,

< previous page

page_250

next page >

< previous page

page_251

next page >

Page 251 the more marked the sound that it represents; see, for example, Causley (1999), Dresher and van der Hulst (1999), Harris and Lindsey (1995); Rice (1996), Rice and Avery (1993 and 1995), Rose (1993), and Walker (1993). In the following discussion, I assume that it is worthwhile to pursue a theory that allows markedness to be represented structurally, and consider the consequences for each of the two theories of place under this assumption. I assume the following complexity scale (see Dresher and van der Hulst [1999], Rice and Avery [1993], Walker [1993] and others for further discussion): (19) a. X b. X c. X | /\ Y YZ The representation in (19a) is the least complex; (19b) is more complex than (19a), but less than (19c), and (19c) is the most complex. (A language may demand a minimal degree of complexity, ruling out [19a]; see, for example, Causley [1999], Goldsmith [1985], Rice [1999] for some discussion. I abstract away from this here.) Suppose that increasing structure correlates with increasing markedness, with the presence of the more complex structure implying the presence of the less complex one (modulo the type of minimality constraint mentioned earlier). Implication is often used as a diagnostic for markedness (e.g., Battistella, 1990; Greenberg, 1966; Hamilton, 1996; Trask, 1996): If the presence of x implies the presence of y, then y is less marked than x. One would expect that front round vowels should be more complex than both front unround and back round vowels, given the implicational relationship. The markedness predictions made by the three-place model are rather surprising under this assumption. Consider the representations of back round and back unround vowels in (12). Back round vowels involve two features, Labial and Dorsal, and are thus structurally more complex than back unround vowels, which involve only the single feature Dorsal. Back round vowels are also structurally more complex than central round vowels, which involve the single feature Labial. Front round vowels and back round vowels are of equal complexity. The markedness scale predicted here is the following: front unround = back unround = central round > front round=back round (“>” indicates less marked). This implicational scale suggests that the presence of back round vowels implies several other vowel places, but as (7) illustrates, this is not borne out. The two-place model makes the right predictions: Front round vowels imply the existence of both front unround and back round vowels since front round vowels are more complex than either front unround vowels or back vowels, having two features rather than one.

< previous page

page_251

next page >

< previous page

page_252

next page >

Page 252 Markedness is often looked at in a second way: It is said to correlate with frequency, with more frequent segments being less marked and less frequent segments more marked. Part of the basis for Kean’s (1975) claim that coronals are the unmarked consonant follows from the fact that almost every language contains a coronal stop (see Battistella, 1990; Greenberg, 1966; Hamilton, 1996; and Paradis and Prunet, 1991, among others, for markedness arguments based on frequency). When frequencies of high vowels of different places of articulation are examined, the difference between the frequencies of the front unround vowel and the back round vowel compared with the others is striking (Maddieson, 1984). (20) FU FR CU CR BU BR 452 29 55 10 31 417 The predictions of the three-place model based on statistical markedness are not found since the back round vowel occurs in significantly more languages than do the others, which would be less marked on a structural basis, despite their greater markedness based on implication and frequency. The two-place model has back round and unround vowels sharing a phonological representation and thus of the same category. It is not meaningful to speak of their different frequencies phonologically; rather these vowels differ in implementation features. These distinctions in the back vowels then should simply be ignored for phonological purposes. This theory makes accurate predictions concerning back round and front round vowels: Back round vowels require a single feature (Peripheral) as do front unround vowels, while front round vowels require two features, Coronal and Peripheral. The difference in frequency relates directly to markedness as measured by complexity. Given the three-place model, one might envisage a solution: In addition to structural markedness, an independent markedness scale could be used; such an approach is proposed for consonantal place by Clements (1990). A two-place model collapses markedness scales and features, with both expressed structurally; one could imagine both being expressed through hierarchies, as in Optimality Theory (e.g., Prince and Smolensky, 1993). The vowel representations in the three-place model require the independent existence of structure and a markedness scale. Scales, however, can mask certain problems. For instance, they are often stated based on segments rather than features. As an example, Ní Chiosáin and Padgett (1997) propose the scale in (21) for vowel place. (21) *i, *u>* Notice that the use of phonetic symbols obscures the difference between the first two vowels: *i is the equivalent of *Coronal while *u is the

< previous page

page_252

next page >

< previous page

page_253

next page >

Page 253 counterpart of *Labial, Dorsal. This latter is a type of complexity constraint, and one would expect that it would indicate a segment that is more marked than the elements that compose it, *Labial and *Dorsal. As discussed previously, based on implication and frequency, *Labial Dorsal must be universally less complex than its individual components *Labial and *Dorsal. Evidence from the Back/Round Relationship: Evidence for Peripheral 4 The next reason to reconsider a three-place model comes from the relationship between Labial and Dorsal. It is often noted that these features are intimately related. The features proposed in Sound Pattern of English (Chomsky and Halle, 1968), [back] and [round], are linked, with one often redundant from the other (e.g., Chomsky and Halle, 1968; Schane, 1973). Considerable attention has been paid to the directionality of the relationship: Is [back] predictable from [round] or vice versa? A non-hierarchical notational system forces this topic to be discussed as part of redundancy, markedness, and linking rules. With the three-place model, the insight that a close relationship between the two features exists is not captured directly; instead, [back] is encompassed by Dorsal and [round] by Labial, and these features do not form a constituent. However, in the two-place model, they are a constituent; in fact they are a single feature. This argument can be viewed as one for constituency rather than for specification. One could imagine the structure in (22) for vowels.

Dorsal and Labial are dependents of Peripheral, so it is not surprising to find a special relationship between them. However, predictions about possible inventories remain problematic, even if the relationship problem receives a solution. In addition, the markedness problems are equivalent to those of the three-place model: Markedness facts do not follow directly from the representation, but must be encoded in a separate way.

< previous page

page_253

next page >

< previous page

page_254

next page >

Page 254 Evidence from Phonological Processes and Natural Classes While theory-internal evidence from inventories, markedness, and the relationship between Labial and Dorsal provides support for the two-place model, many phonological arguments have been advanced for the features [back] and [round], or Dorsal and Labial, as separate features. The question that I address next is whether phonological evidence exists for Dorsal and Labial functioning independently of one another. This is not predicted by the two-place model, where phonologically only the single feature Peripheral is available. Could it be the case that while contrasts can be captured with only two place features, a third place feature must be brought into play in order to capture phonological patterning? I argue that reference to peripherality alone is sufficient to account for phonological processes. Two kinds of languages might present problems to this hypothesis. First are languages in which Dorsal and Labial have been argued to be independent. And second are languages where back vowels and velar consonants appear to form a natural class. Since back vowels are not specially marked by Dorsal, as in Clements’ model, one would not predict any special phonological patterning of vowels with these consonants. Round Implies Back, Back Implies Round I begin with some cases that have been analyzed under the three-place model, but which have clear equivalent treatments in the two-place model. Clements and Hume (1995), following Itô (1984), argue that to account for vowel dissimilation in Ainu, Coronal and Dorsal are required. Ainu has the root vowel inventory in (23). (23) ieaou There are some Ainu suffixes consisting of a vowel. In the regular case, this vowel is a copy of the root vowel. In a group of roots with non-low vowels, the suffix vowel is a high vowel with the opposite value of the feature [back] from the root vowel, as schematized in (24). (24)CVC+V, where +V is a high vowel with the opposite value of [back] from root vowel, e.g., ket-u ‘to rub’, pok-i ‘to lower’ Itô, using the feature [back], argues that in this class of roots, the suffix vowel has the opposite value for backness from the root vowel. Clements and Hume, using the privative Labial, Coronal, and Dorsal rather than the binary [back] and [round], propose that Coronal and Dorsal are

< previous page

page_254

next page >

< previous page

page_255

next page >

Page 255 required, and that if the root vowel is Coronal, then the suffix vowel will be Dorsal, and vice versa. However, Clements and Hume (1995) do not discuss why Coronal and Dorsal are the relevant features rather than Coronal and Labial. Assuming (as does Clements [1991]) that the central vowel /a/ is unmarked for frontness or backness, labiality and velarity are predictable one from another in Ainu, and no way exists within the language to determine which feature is primary: One could just as well say that a coronal root vowel takes a labial suffix vowel and vice versa. The two-place theory accounts for these data in a straightforward way: The only features available are Coronal and Peripheral. An analysis along the following lines is possible. The vowels in Ainu have the phonological place representations in (25). (25) i, e u, o a Coronal Peripheral Following Clements and Hume (1995:291), if the Obligatory Contour Principle (OCP) disallows like features and an antigemination constraint disallows geminates in this class, the forms with dissimilated vowels arise. Clements and Hume’s basic insight into what controls dissimilation can be maintained with the two-place model, and thus provides no evidence for a three-place model, as Clements and Hume themselves show. In Ainu-type inventories, Dorsal is sometimes called upon to differentiate [i] from [u]; at other times Labial is used to distinguish these vowels. In such languages, it is difficult to find phonological evidence for either feature being primary since one is predictable from the other (see the subsection Evidence from the Back/Round Relationship Evidence for Peripheral 4). In the three-place model, it is arbitrary whether one, the other, or both are chosen. Only two features are available in the two-place model, and only two features are required. Thus the two-place model is always able to account for such cases. Back Harmony Without Round Harmony The second type of case concerns languages with back harmony but no round harmony, such as Finnish and Hungarian. This language type is expected under the three-place hypothesis: The feature Dorsal spreads, giving back harmony, but Labial does not; and thus there is no round harmony. This is problematic under the two-place hypothesis because Dorsal and Labial are not independent features, and thus it is not possible to assimilate to one and not the other. A solution to these problems can be found by reanalyzing the harmony trigger. While it is often proposed that velarity, or back, must spread in

< previous page

page_255

next page >

< previous page

page_256

next page >

Page 256 these languages or that both palatality and velarity must spread (e.g., Steriade, 1987, on Finnish and Hungarian; Farkas and Beddor, 1987, on Hungarian), alternative analyses have been proposed where only palatality spreads; see, for example, Goldsmith (1985) and Rose (1993) on Finnish and Goldsmith (1985) on Hungarian. Back Harmony and Round Harmony under Separate Conditions A strong test of the two-place model comes from languages that are argued to require independent back and round harmony. If a language has both, it is impossible to use Peripheral instead of Labial and Dorsal since it is not possible to distinguish the two effects. In this section I briefly examine one such language, Turkish, arguing after Walker (1993) that Turkish has two harmony processes, Coronal harmony (in place of back harmony) and Peripheral harmony; there is no Dorsal harmony. Given this, two features suffice. Turkish has the phonological vowel inventory in (26). (26) FU FR C BU i ū u e ō a o Recall that the vowel that I have notated with / / is simply unmarked for place of articulation phonologically; it is realized phonetically as a back unrounded vowel. The basic facts of Turkish harmony are shown in (27), abstracting away from round (peripheral) harmony, and showing forms with central and front vowels. If a stem has a front vowel, it takes a front vowel suffix; if it has a central vowel, it takes a central vowel suffix, and if it has a back vowel, it takes a back vowel suffix.2 (27) ip ip-[i] ‘rope’ kz k z-[ ] ‘girl’ ek ek-[i] ‘joint’ tak tak-[ ] ‘arch’ An analysis of Turkish harmony is available under the two-place model, where the vowels have the place features in (28). (28) front vowel central vowel back vowel Coronal Peripheral Front-back harmony is accomplished through the spreading of Coronal and rounding harmony through the spreading of Peripheral. The regular forms are derived as in (29).

< previous page

page_256

next page >

< previous page Page 257 (29) V-Place V-Place versus

page_257 V-Place V-Place

next page > V-Place V-Place

Coronal Peripheral Coronal spreads, front no spreading, central Peripheral spreads, vowel results vowel results back round vowel results While this presentation is sketchy, the important conclusion can be drawn that Turkish can be analyzed as having front harmony (coronal) and round harmony (peripheral). Back (dorsal) harmony is not necessary to account for the assimilation facts. Summary The three- and two-place models make different predictions about the kinds of vowel harmony that might exist. These are summarized in Table 10–6 (30), where all attested patterns are captured in the two place column. I follow the representations in (12), and for each process list a language or two that illustrates that harmony. (30) See Table 10–6 on the next page. Many of these languages have already been referred to in the course of this chapter. Arapaho (Algonquian) is one that has not been; it has a harmony process that rounds a stem-final mid front vowel in the presence of a following mid back round vowel; see Goddard (1974) and Picard (1980) for details. Turkish exhibits front harmony alone with low vowel targets, and both front and round (Peripheral) harmony with high vowel targets. The three-place model predicts assimilation systems that do not appear to be attested; for instance, I have found no languages that must be analyzed as having Coronal and Labial assimilation without Dorsal assimilation, nor any that appear to have separate Labial and Dorsal assimilation, or just one of these. Just as this model overgenerates inventories, it also appears to overgenerate types of occurring harmony systems. The two-place model, on the other hand, predicts exactly three types of harmony systems: Coronal, Peripheral, and Coronal and Peripheral. This system thus generates the attested cases in languages. Consonant/Vowel Interactions Several of Clements’ arguments for dorsal in vowels are based on interaction between consonants and vowels. He argues that if certain

< previous page

page_257

next page >

< previous page Page 258 TABLE 10–6. Three-Place Hypothesis Coronal harmony: Finnish Labial harmony: ? Dorsal harmony: ?

page_258

next page >

Two-Place Hypothesis Coronal harmony: Finnish, Turkish (low vowel targets) Peripheral harmony: Arapaho (not discussed) Coronal, Peripheral harmony: Turkish (high vowel targets)

Coronal, Labial harmony: ? Labial, Dorsal harmony: Arapaho Coronal, Dorsal harmony: Turkish (low vowel targets) Coronal, Labial, Dorsal harmony: Turkish (high vowel targets) vowels have the feature Dorsal, it is possible to account for their patterning with Dorsal consonants. More generally, he predicts the following patterns. (31) front vowel coronal consonant round vowel labial consonant back vowel dorsal consonant The two-place model predicts the following classes. (32) front vowel coronal consonant peripheral vowel labial, dorsal consonant I examine briefly three cases that Clements addresses, Maxacalí, Palestinian Arabic, and the historical development of French. Maxacalí Clements (1991) argues from epenthesis in Maxacalí (Gudschinsky et al., 1970) that dorsal consonants and back vowels form a natural class. In Maxacalí, a consonant is inserted after a vowel under certain conditions that are not of relevance here. What is important is the relationship between the vowel and the inserted consonant. This is illustrated in (33). (33) if the first vowel is: then C is: i j (palatal glide) o w ,a e none

< previous page

page_258

next page >

< previous page

page_259

next page >

Page 259 Clements proposes the following analysis. The vowels have the features in (34). (34) front vowel central vowel back vowel Coronal Dorsal Labial Dorsal The [j] glide is achieved by the spreading of Coronal and the [w] by the spreading of Labial and Dorsal. The glide results from the spreading of Dorsal. Clements does not offer an account of the patterning of /e/ (nor will I). Assuming the vowel system in (35), the place features associated with the vowels under the two-place hypothesis are given in (36), again ignoring [e]. (35) i eao (36) i o ,a Coronal Peripheral The vowel features spread to the epenthetic consonant, giving [j] after Coronal and [w] after Peripheral. The remaining vowels have no place dependent to spread, and a velar glide results. I suggest, following Trigo (1988) and Rice (1996), that the glide is not a phonological dorsal, but is a placeless segment; it is velar only by phonetic interpretation. Palestinian Arabic Clements (1991) provides an argument from Palestinian Arabic for Dorsal on back vowels. In Palestinian Arabic, if a root contains an emphatic or uvular consonant, a different stem vowel is found in the imperfective than with other consonants: While [i] is generally found, [u] occurs with an emphatic or uvular. Clements suggests that the class of emphatic consonants, uvular consonants, and the vowel [u] can be captured through the use of Dorsal. However, since the consonantal class in question excludes dorsal consonants, the prototypical consonants marked by the feature Dorsal, it is not clear that this is evidence for Dorsal in the vowel. Dorsal Consonants and Back Vowels: Development of French Another case cited by Clements (1991) as evidence for Dorsal on back vowels is from the loss of consonants in the historical development of French. A labial or velar obstruent deleted intervocalically when one of the surrounding vowels was /u/ or /o/. Clements argues for two processes: a dorsal consonant deleted in the environment of a dorsal vowel

< previous page

page_259

next page >

< previous page

page_260

next page >

Page 260 (u, o) and a labial consonant in the environment of a labial vowel (u, o). Note that Peripheral defines the class of labial and dorsal consonants and the vowels /u, o/. An alternative is that a peripheral consonant is deleted in the environment of a peripheral vowel. Summary I have proposed that the arguments adduced for vowels being marked for labiality and velarity must both be reinterpreted to avoid the over-generation of possible vowel systems in languages and to account for occurring harmony patterns to the exclusion of non-occurring ones. If the proposed analysis is correct—that is, if Peripheral is sufficient and Labial and Dorsal are unnecessary—a major question remains. Under the proposed analysis, there is not a necessary relationship between the details of phonetic implementation and the phonological patterns of distribution and alternation. How then do the different phonetic vowel qualities that are phonologically peripheral come to be differentiated? In the next section, I examine the general question of relating the phonological representation to the range of possible phonetic realizations. THE PHONETIC REALIZATION OF PERIPHERAL VOWELS In the previous section I provided phonological evidence that vowels with the feature Peripheral are not further specified for labiality or dorsality, but peripherality is itself a basic category, along with coronality. I thus argued for a type of representational economy in the phonology, where phonological contrasts are represented without phonetic detail. In this section, I describe how this is played out in the phonetic realization of vowels, focusing on the phonetic realization of the phonological feature Peripheral. This feature has phonetic correlates, referring articulatorily to something at the periphery, or edges, of the oral cavity. It may be back and round, back only, round only, or simply peripheral (i.e., not front or central). Perceptually the feature Peripheral is measured by F2, indicated by a low F2 value. Harris and Lindsey (1995, 2000) argue that peripheral vowels (their element U) are well defined by an auditory spectral image, or minimal energy above the middle of the frequency band containing the most significant information relating to vocalic contrasts. Thus auditorily, Peripheral indicates a range of F2 values or a spectral auditory image rather than a point. Given the economy required by the phonological system, it is necessary to account for this variability in the phonetics of phonologically peripheral vowels. In this section I briefly develop a model to account for the relation

< previous page

page_260

next page >

< previous page

page_261

next page >

Page 261 between the more abstract phonological representation and its implementation. In particular, I examine the range of phonetic variability allowed to a vowel that is phonologically marked by the feature Peripheral. I argue for the following point: (37)A non-contrastive feature may function as an implementation feature to enhance the vowel, allowing for easier differentiation of sounds by increasing the perceptual saliency of a contrast. First consider evidence for Labial and Dorsal being implementation features. This comes from the phonetic side of the inventories discussed in the first section. Recall that phonologically only a single back vowel is found in an inventory. This vowel may be round or not, indicating a need for a distinction between these vowels phonetically. Similarly the secondary articulation of peripherality varies both across languages and within a language, it may be labiality alone, velarity alone, or both. Again, further distinctions are required phonetically rather than phonologically. Within-language variability reinforces this point, and I report on some of this later. In Yimas (Papuan, Foley [1991]), the vowel characterized as /u/ is generally back and round, but it is unround before a labial consonant. These vowels share a phonological representation (38); they differ phonetically in terms of implementation features (39). (38) Phonological representation: Peripheral (39) Some possible phonetic realizations:

In Kannada (Dravidian, Schiffman [1983]), /u/ is back round or back and slightly unround (this pronunciation is reported to be stigmatized). Again, the representations in (38) and (39) are relevant. In Diyari (Australian, Austin [1981]), /u/ is slightly more central and less rounded than the cardinal vowel 8; it may approach the high back unround vowel (abstracting away from the finest level of phonetic de-

< previous page

page_261

next page >

< previous page

page_262

next page >

Page 262 tail). In Gooniyandi (Australian, McGregor [1990]) the vowel /u/ is reported to have allophones [u ]. These vary in place and height, with the allophones for place being represented as in (39). I have given examples of vowels with enhancement by Labial and Dorsal ([u]) and enhancement by Dorsal alone ([ ]). Two other phonetic realizations of a representation with Peripheral are expected, one with just Labial and the other with no implementation features. In American English (de Jong, 1993), the vowel /u/ in a word like “kook” has been shown to have three realizations, depending on dialect: [u], [ vowel with rounding, but without particular fronting; I suggest that this is the vowel that results if only front rounded vowel (y); this is a vowel with rounding, but not fronting. The final predicted vowel is one with no implementation features beyond Peripheral. The vowel is an extremely lax non-front, noncentral vowel, but without features of backness or roundness. The variability in realization of Peripheral has universal properties: It can be implemented as it is, with labiality (lip rounding, F2 lowering), with dorsality (tongue backing, F2 lowering), or with both labiality and dorsality. It also has language-particular aspects: A language chooses how it is to be implemented, if conditions exist on its implementation (e.g., optionality, domains, interaction with other features). The single category thus can be implemented in a variety of ways, differing in articulatory and auditory detail. The statement in (37) not only asserts that non-contrastive features can be used as implementation features; it also provides a reason why such features might be required, namely as enhancement features to render a contrast more salient along the lines proposed by Stevens and Keyser (1989). Consider the fact that as a phonetic realization of peripherality, [u] is by far the most common, being the most reported. This is not surprising if implementation features function to increase the saliency of a contrast, one would expect to find the fully implemented version frequently as it is perceptually the most distinct from the other vowels. It should be kept in mind, however, that in reporting vowel inventories, the symbol used for the phoneme is often the one found in metrically strong or other salient positions, with other allophones regarded as reduced in some way. A survey of the variability of the peripheral vowel in a full range of domains may reveal a greater degree of variability than is often suggested. Before leaving the discussion of the phonetic realization of phonological peripherally, I examine briefly just what it means to enhance a contrast. One of the reasons for the mismatch between the economical phonological and rich phonetic representations is enhancement, which,

< previous page

page_262

next page >

< previous page

page_263

next page >

Page 263 as we have seen, allows for easier differentiation of sounds by increasing the perceptual saliency of a contrast. Flemming (1995) pursues the theme of enhancement in a formal way, proposing minimal distance constraints that militate against two vowels being too close perceptually. By such a hypothesis, inventories most easily support an [i-u] contrast, as these are the most distinct perceptually, but disfavor a contrast such as [ - ] because the vowels are too close together to form a good contrast. This hypothesis, which finds its basis in dispersion theory (e.g., Lindblom 1986, 1990a, 1990b), suggests that vowels disperse to the edges of the vowel triangle. Enhancement of perceptual saliency through vowel dispersion is clearly at work in language, given the preponderance of languages reported to have an [i-u] contrast when there are only two high vowels, as discussed previously. However, dispersion is not the only factor involved. As the inventories in (7) show, two-place languages like Japanese, with [i], do not fit this characterization as these vowels are not maximally distant from each other, but nevertheless [ ] is enhanced in having the feature Dorsal present. The role of distance in increasing the saliency of a contrast is not to be underrated, but it is not requisite for enhancement that maximal distances be obtained. Enhancement of a peripheral vowel, whether by labial alone, dorsal alone, or both, serves to increase the F2 distance from the front vowel, rendering the contrast perceptually more distinct.3 CONCLUSIONS AND CONSEQUENCES Summary I have proposed the following: 1. In vowels the features Coronal and Peripheral serve to mark place phonologically. 2. Labial and Dorsal are implementation features for peripheral vowels. These features play no role in the phonology proper but function to increase the perceptual salience of a contrast. 3. A close correlation is found between phonological specification and possible phonetic variability. Consequences for the Phonetics/Phonology Relationship In recent phonological work, it has been proposed that the phonetic representation of the vowel determines its phonological representation under most conditions; see, for example, Steriade (1995). This chapter argues against that position: The proposed analysis suggests that Labial and Dorsal are absent phonologically, although they are generally present phonetically. If this conclusion is accepted, some degree of abstract-

< previous page

page_263

next page >

< previous page

page_264

next page >

Page 264 ness from the details of articulation and perception/audition must be allowed in phonological representations; they distinguish major categories that are perceived, but not details. The phonology thus deals in oppositions, with sparse, economical representations; the phonetics in details that enhance perceptual saliency. While the phonological and phonetic representations are not isomorphic, the analysis is nevertheless true to a principle argued for by proponents of a what-you-see-is-what-you-get theory: The representation, be it phonological or phonetic, is always fully interpretable (see, for example, Kaye et al., 1985; Harris and Lindsey, 1995; and Steriade, 1995, for different instantiations of this principle). Rather than require isomorphism, the theory must tolerate some ambiguity. First, there is phonetic ambiguity: A single phonological representation may have more than one articulatory and perceptual realization. (40) illustrates this for peripheral vowels. (40) phonological phonetic And second, not discussed here, there is phonological ambiguity: A single phonetic realization may have more than one phonological representation. As remarked earlier, the relationship between the phonology and phonetics is highly constrained but, at the same time, requires a degree of abstractness, where the full range of properties required phonetically is not present phonologically. Similarities/Differences Between Consonant and Vowel Features I would like to end with a brief discussion of similarities and differences between consonants and vowels. Clements (1991), Clements and Hume (1995), Sagey (1986), and others argue that three features are required for place of articulation in consonants, Coronal, Labial, and Dorsal, and I have assumed this, with the additional assumption that Labial and Dorsal form a constituent, Peripheral (e.g., Rice, 1994, 1996; Iverson et al., 1999). Consider a feature like Coronal in consonants. It is generally assumed that if a contrast between, say, apical and laminal is not distinctive in a language, this feature is simply not distinctive in that language. A feature may be used contrastively in a language, in which case it is phonologically present; it may be non-contrastive in a language, in which case it is phonologically absent but may be present phonetically, both in terms of articulation and audition. I have argued that the features available to vowels are also phonologically

< previous page

page_264

next page >

< previous page

page_265

next page >

Page 265 activated by contrasts. In the absence of a contrast, a feature does not play a phonological role. In vowels, contrasts between peripheral vowels are simply not found within a system. Labial and Dorsal serve only as implementation features for vowels since they play no contrastive role in this system. Thus while I have assumed that consonants and vowels share their set of place features, I have argued that they differ in the number of features required phonologically: Vowels require only Coronal and Peripheral, but consonants require three place features, Coronal and Peripheral, which dominate Labial and Dorsal. Place is more fully developed phonologically in consonants than in vowels. Why might this be?4 To re-phrase the question, why is it that more place contrasts are allowed in the consonant system than in the vowel system? It is well known that a correlation exists between the number of allowed places and sonorancy in consonants: Less sonorant consonants tend to show a greater range of possible places, while more sonorant consonants tend to allow a lesser range of places of articulation. For example, most languages have stops at the three basic places of articulation, but liquids at only a single place. The facts are similar in vowels: The more sonorant vowels (low vowels) tend to show less of a range of places of articulation than the less sonorant vowels (high vowels), as in (41); see Rice and Avery, 1993, for discussion. (41) FU FR CU CR BU BR high 452 29 55 10 31 417 mid 425 32 100 8 19 448 low 81 0 392 1 13 36 (Maddieson, 1984) How is this relevant to consonants and vowels? It is not surprising that vowels, which are more sonorant than consonants, are phonologically more restricted in the possible places of articulation, allowing only a subset of those found in consonants. Following Hamilton (1996), Prince and Smolensky (1993), Rice and Avery (1991, 1993), and others, segments are constrained in complexity, where complexity is an overall measure of place, sonority, and stricture. Increased sonority in consonants implies decreased options in terms of place. The fact that vowels allow fewer places of articulation than consonants can be seen as a direct consequence of their higher sonority. SUMMARY I have challenged some common assumptions about phonological representations, arguing that features often thought to belong to under-

< previous page

page_265

next page >

< previous page

page_266

next page >

Page 266 lying representations of vowels play no role phonologically. These features can be present phonetically, as implementation features. The model receives interesting support from phonetic variability, with the range of variability relating to phonological specification. Overall, the facts discussed suggest the need for relatively abstract, or economical, phonological representations, representations that are determinable based on knowledge of contrasts, or distinctive categories, in the system. Whether the inventory is basic or whether it emerges from constraint ranking, the features that play a phonological role in vowel place are highly limited. NOTES This work was supported by Social Science and Humanities Research Grant 410–99–1309. Thank you to Peter Avery, Elan Dresher, Carrie Dyck, Heather Goad, Daniel Hall, Larry Hyman, Bill Idsardi, Sharon Inkelas, Glyne Piggott, and Milan Rezac, and an anaymousw reviewer for discussion about the ideas in this paper; all errors are, of course, mine. 1. The model proposed here is closest in concept to particle (e.g., Schane, 1984) and element (e.g., Kaye, Lowenstamm, and Vergnaud, 1985; Harris, 1990; Harris and Lindsey, 1995) theories. 2. I do not discuss the exceptions to vowel harmony here; see Clements and Hume (1995), Clements and Sezer (1982), Orgun (1993), and Walker (1993), among others, for discussion. 3. The fact that the central vowel is often realized as the back unround [Í] phonetically (as in Turkish and Vietnamese; see [7]) is perhaps not too surprising. In a three-place system with the vowels [i u], [i] is closer to the central vowel than [u] is in terms of F2, while in the [i u] system, this is reversed, with [u] closer to the back unround vowel than [i] is in terms of Φ’: In the former cases, the distance between [i] and [ ] is 800, and that between [ ] and [u] 1015, while in [i u], the distance between the first two is 1010 and the latter two 795. (See [14] for the F2 values.) “Ideal” distances do not appear to be available, and these two are essentially equivalent to each other in terms of three-place systems. 4. This can be captured in Optimality Theory by not having faithfulness constraints on the features Labial and Dorsal for vowels, but having faithfulness constraints for Coronal and Peripheral. A phonetic module is required to introduce Labial and Dorsal; measures of global complexity are required to explain why there is an asymmetry between consonants and vowels in terms of features controlled by faithfulness. REFERENCES Austin, P. 1981. A Grammar of Diyari, South Australia. Cambridge, U.K.: Cambridge University Press. Battistella, E. 1990. Markedness: The Evaluative Superstructure of Language. Albany: State University of New York Press.

< previous page

page_266

next page >

< previous page

page_267

next page >

Page 267 Catford, J.C. 1988. A Practical Introduction to Phonetics. Oxford: Clarendon Press. Causley, T. 1999. Complexity and Markedness in Optimality Theory. Ph.D. dissertation, University of Toronto. Cho, S.-B. 1967. A Phonological Study of Korean (Acta Universitatis Upsaliensis, Studia Uralica et Altaica Upsaliensia 2). Uppsala: Almqvist and Wiksells. Cho, Y.-m. Yu, and Iverson, G. 1997. Korean phonology in the late twentieth century. Language Research 33.4:687–735. Chomsky, A.N., and Halle, M. 1968. Sound Pattern of English. New York: Harper and Row. Clements, G.N. 1990. The role of the sonority cycle in core syllabification. In J. Kingston and M.Beckman (eds.). Papers in Laboratory Phonology I. Between the Grammar and Physics of Speech. Cambridge, U.K.: Cambridge University Press, 283–334. ——. 1991. Place of articulation in consonants and vowels: A unified theory. In G.N.Clements and E.Hume (eds.) Working Papers of the Cornell Phonetics Laboratory, vol. 5. 77–123. Clements, G.N., and Hume, E. 1995. The internal organization of speech sounds. In J.Goldsmith (ed.) The Handbook of Phonological Theory. London: Blackwell, 245–306. Clements, G.N., and Sezer, E. 1982. Vowel and consonant disharmony in Turkish. In H.van der Hulst and N.Smith (eds.) The Structure of Phonological Representations II. Dordrecht: Foris, 213–55. de Jong, K. 1993. On the status of redundant features: The case of backing and rounding in American English. Paper presented at the Laboratory Phonology Conference. Dresher, B.E., and van der Hulst, H. 1999. Head-dependent asymmetries in phonology: Dependency and variability. Phonology 15:317–52. Farkas, D., and Beddor, P. 1987. Privative and equipollent backness in Hungarian. In A.Bosch, B.Need, and E.Schiller (eds.) Papers from the 23rd Regional Meeting of the Chicago Linguistic Society. Part two: Parasession on Autosegmental and Metrical Phonology. 90–105. University of Chicago. Flemming, E.S. 1995. Auditory Representations in Phonology. Ph.D. Dissertation, UCLA. Foley, W. 1991. The Yimas language of New Guinea. Stanford: Stanford University Press. Goddard, I. 1974. An outline of the historical phonology of Arapaho and Atsina, IJAL 40.2:102–16. Goldsmith, J. 1985. Vowel harmony in Khalkha Mongolian, Yaka, Finnish, and Hungarian. Phonology Yearbook 2:253–75. Greenberg, J. 1966. Language Universals, with Special Reference to Feature Hierarchies. The Hague: Mouton. Gudschinsky, S., Popovich, H., and Popovich, F. 1970. Native reactions and phonetic similarity in Maxacalí phonology. Language 46:77–88. Halle, M. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26:1–46. Hamilton, P. 1996. Phonetic Constraints and Markedness in the Phonotactics of Australian Aboriginal Languages. Ph.D. Dissertation, University of Toronto. Harris, J. 1990. Segmental complexity and phonological government. Phonology 7:255–300. Harris, J., and Lindsey, G. 1995. The elements of phonological representation. In J.Durand and F.Katamba (eds.) Frontiers of Phonology: Atoms, Structures and Derivation. Harlow, Essex: Longman, 34–79.

< previous page

page_267

next page >

< previous page

page_268

next page >

Page 268 ——. 2000. Vowel patterns in mind and sound. In N.Burton-Roberts, P.Carr, and G.Docherty (eds.). Phonological Knowledge: Conceptual and Empirical Issues. 185–205. Oxford: Oxford University Press. Hulst, H.van der 1996. Radical CV phonology: The segment-syllable connection. In J.Durand and B.Laks (eds.) Current Trends in Phonology: Models and Methods. Salford: ESRL, 333–61. Hume, E. 1992. Front Vowels, Coronal Consonants, and Their Interaction in Nonlinear Phonology. Ph.D. Dissertation, Cornell University. Hyman, L. 1975. Phonology. Theory and Analysis. New York: Holt, Rinehart and Winston. Itô, J. 1984. Melodic dissimilation in Ainu. Linguistic Inquiry 15:505–13. Iverson, G., Davis, G., and Salmons, J. 1999. Peripherality and markedness in the spread of the High German consonant shift. Beiträge zur Geschichte der deutschen Sprache und Literatur 121.2:1–24. Jakobson, R., Fant, G., and Halle, M. 1969. Preliminaries to Speech Analysis: The Distinctive Features and Their Correlates. Cambridge, MA: MIT Press (first published 1951). Jackobson, R., and Halle, M. 1956. Fundamentals of Language. The Hague: Mouton. Kaye, J., Lowenstamm, J., and Vergnaud, J-R. 1985. The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2:305–28. Kean, M.L. 1975. The Theory of Markedness in Generative Grammar. Ph.D. Dissertation, MIT. Lass, R. 1984. Vowel system universals and typology: Prologue to theory. Phonology Yearbook 1:75– 112. Lindblom, B. 1986. Phonetic universals in vowel systems. In J.Ohala and J. Jaeger (eds.) Experimental Phonology. Orlando: Academic Press. ——. 1990a. Phonetic content in phonology. Perilus 11:65–100. ——. 1990b. On the notion “possible speech sound.” Journal of Phonetics 19:135–52. Maddieson, I. 1984. Patterns of Sounds. Cambridge: Cambridge University Press. Martin, S. 1951. Korean phonemics. Language 27:519–33. Martin, S., and Lee, Y-S.C. 1969. Beginning Korean. New Haven, CT: Yale University Press. McGregor, W. 1990. A Functional Grammar of Gooniyandi. Amsterdam: John Benjamins. Ní Chiosáin, M., and Padgett, J. 1997. Markedness, Segment Realization, and Locality in Spreading. Report no. LRC-97–01 Linguistics Research Center, University of California at Santa Cruz. Odden, D. 1991. Vowel geometry. Phonology 8:261–89. Orgun, O. 1993. Alterable geminates and the release node. Ms, University of California, Berkeley. Paradis, C., and Prunet, J-F. 1991. Asymmetry and visibility in consonant articulations. In C.Paradis and J-F Prunet (eds.) The Special Status of Coronals. Internal and External Evidence. Phonetics and Phonology 2 (pages 1–28). San Diego: Academic Press. Pham, H. 1998. Coronal-velar relationship in Vietnamese dialects: A prosodic account. Asia Pacific Language Research 1. (http://asiapacific.webjump.com/). Picard, M. 1980. Vowel assimilation and morphophonemic rules. In R.Vago (ed.) Proceedings of the CUNY Linguistics Conference in Vowel Harmony, May 1977. Issues in Vowel Harmony. Amsterdam: J.Benjamins.

< previous page

page_268

next page >

< previous page

page_269

next page >

Page 269 Prince, A., and Smolensky, P. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. TR-2, Rutgers University Cognitive Science Center. To appear, MIT Press. Rice, K. 1996. Default variability: The coronal-velar relationship. Natural Language and Linguistic Theory 14:493–543. ——. 1994. Peripheral in consonants. Canadian Journal of Linguistics 39:191–216. ——. 1999. Featural markedness in phonology: Variation. Part 1. GLOT 4.7:3–6. Part 2. GLOT 4–:3–7. Rice, K., and Avery, P. 1991. On the relationship between coronality and laterality. In C.Paradis and J-F Prunet (eds.) The Special Status of Coronals. Internal and External Evidence. Phonetics and Phonology, vol. 2. San Diego: Academic Press, 101–24. ——. 1993. Segmental complexity and the structure of inventories. In C.Dyck (ed.) Toronto Working Papers in Linguistics 12:131–54. ——. 1995. Variability in a deterministic model of language acquisition: A theory of segmental elaboration. In J.Archibald (ed.) Phonological Acquisition and Phonological Theory. Hillsdale, NJ: Erlbaum, 23–42. Rose, S. 1993. Coronality and vocalic underspecification. In C.Dyck (ed.) Toronto Working Papers in Linguistics 12:155–77. Sagey, E. 1986. The Representation of Features and Relations in Non-linear Phonology. Ph.D. Dissertation, MIT. Schane, S. 1973. [back] and [round]. In S.R.Anderson and P.Kiparsky (eds.) A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston, 174–84. ——. 1984. The fundamentals of Particle Theory. Phonology Yearbook 1:129–55. Schiffman, H.F. 1983. A Reference Grammar of Spoken Kannada. Seattle: University of Washington Press. Selkirk, E.O. 1992. Labial Relations. Ms, University of Massachusetts, Amherst. Steriade, D. 1987. Redundant values. In A.Bosch, B.Need, and E.Schiller (eds). Papers from the 23 Annual Regional Meeting of the Chicago Linguistic Society. Part 2. Parasession on Autosegmental and Metrical Phonology. 339–62. ——. 1995. Markedness and underspecification. In J.Goldsmith (ed.) The Handbook of Phonological Theory. London: Blackwell, 114–74. Stevens, K., and Keyser, S.J. 1989. Primary features and their enhancement in consonants. Language 65:81–106. Stevens, K., Keyser, S.J., and Kawasaki, H. 1986. Toward a phonetic and phonological theory of redundant features. In J.S.Perkell and D.H.Klatt (eds.) Invariance and Variability in Speech Processes. Hillsdale, NJ: Lawrence Erlbaum , 426–49. Trask, R.L. 1996. A Dictionary of Phonetics and Phonology. London, New York: Routledge. Trigo, R.L. 1988. On the Phonological Derivation and Behavior of Nasal Glides. Ph.D. Dissertation, MIT. Walker, R. 1993. A vowel feature hierarchy for contrastive specification. In C. Dyck (ed.) Toronto Working Papers in Linguistics 12:178–97.

< previous page

page_269

next page >

< previous page

page_270

next page >

page_270

next page >

Page 270 This page intentionally left blank.

< previous page

< previous page

page_271

next page >

Page 271 Author Index Abasheikh, M., 195 Aikhenvald, A., 4, 7 Aitchison, J., 57 Alsina, A., 201 Anderson, S., 194 Aoun, J., 219, 221 Austin, P., 85, 91, 94–97, 100, 106 Baker, M., 4, 196 Bloom, L., 68, 69, 70 Bowerman, M., 68 Bowers, J., 64 Braine, M., 68, 69 Bresnan, J., 85, 86, 91, 94–100, 106, 197 Brown, R., 67–68 Browning, M., 220, 231 Brunson, B., 91, 93, 94–99, 108 Campana, M., 158, 165 Carrier-Duncan, J., 158 Catford, J., 248 Chimombo, M., 203 Chomsky, N., 41, 64, 71, 76–77, 78, 93, 134, 136, 194, 195, 246, 253 Clements, G. 240, 242, 247–48, 254–55, 257–58, 259–60 Comrie, B., 1, 9, 57 Cooreman, A., 165, 166, 167 Corbett, G., 30 Crain, S., 72 Demoz, A., 3–4 Demuth, K., 203 Dixon, R.M.W., 4, 7–8, 155 E.Kiss, K., 125 Escobar, A., 32 Everett, D., 196 Fernández-Ordóñez, I., 21, 23, 24, 30, 33 Flemming, E., 248, 263 Givón, T., 192 Grimshaw, J., 25, 26 Guerssel, M., 4–6, 11–12 Hale, K., 64, 83, 85, 86, 91, 93, 94–100, 195 Halle, M., 240, 246, 253 Halpern, A., 193 Harris, J., 260 Haspelmath, M., 2, 16 Haverkort, M., 193 Huang, C., 212, 219 Hume, E., 240, 254–55 Hyams, N., 70

< previous page

page_271

next page >

< previous page

page_272

Page 272 Jackendoff, R., 12, 17 Jakobson, R., 240, 241 Jelinek, E., 85 Kager, R., 26 Katamba, F., 190 Kean, M., 252 Keenan, E., 41, 160, 164, 182–83 Keyser, S., 64, 195 Klein-Andreu, F., 23 Koizumi, M., 62, 64, 65, 71, 76 Lakatos, I., 57–58 Lass, R., 242–43 Laughren, M., 83, 86, 110 Legendre, G. et al. 155–60, 165, 167–172, 174–175, 178 Leslau, W., 3 Levin, B., 14 Lillo-Martin, D., 72 Lindblom, B., 263 Lindsey, G., 260 Maclachlan, A., 158, 172, 173, 180, 183 Maddieson, I., 244–45, 252 Mahajan, A., 215, 217 Manning, C., 159, 171 Marantz, A., 186, 192 McCloskey, J., 77 McConvell, P., 117 Mchombo, S., 185, 190, 204 Mel’čuk, I.A., 139 Mohanan, K., 4–5, 14 Mtenje, A., 190, 194, 203 Mufwene, S., 202 Murasugi, K., 158, 171 Nakayama, M., 64 Nash, D., 83, 85, 86, 101 Newmeyer, F., 195 Nishigauchi, T., 212, 214, 219–220, 224 Nordlinger, R., 125 Obenauer, H., 226, 227 Otanes, F., 178 Ouhalla, J., 212–215, 217, 219, 221 223 Payne, T., 180 Pearson, M., 160, 183 Pensalfini, R., 85 Perlmutter, D., 195 Pesetsky, D., 36 Pinker, S., 57 Pollock, J., 64 Prince, A., 21, 25, 248 Radford, A., 71 Reinhart, T., 213, 219, 225 Sag, I., 158–59, 171 Schachter, P., 178 Schwartz, B., 30 Sells, P., 158, 165, 180 Simpson, J., 83, 85, 86 Smolensky, P., 21, 25, 248 Speas, M., 83 Sprouse, R., 30 Steriade, D., 240, 263

next page >

Swartz, S., 94 Takahashi, D., 228, 231–32 Takano, Y., 77–78 Topping, D., 165, 166, 183 Wahba, W., 212, 217–19, 230 Watanabe, A., 214, 219, 221, 225 Wexler, K., 70 Withgott, M., 86 Woolford, E., 157, 174 Yip, M., 158

< previous page

page_272

next page >

< previous page

page_273

next page >

Page 273 Index of Languages and Language Families Afrikaans, 134 Afroasiatic languages, 5 Ainu, 254, 255 Akkadian, 132, 135, 136 Albanian, 143, 145 Amharic, 2–3, 9, 10, 14, 15, 16, 17 Ancient Greek, 143, 147, 149 Apalai, 8–9 Arabic, 135, 136, 137; Iraqi, 212, 214, 218, 229–31; Palestinian, 259 Arapaho, 257 Awngi, 143, 147 Bantu languages, 185–210 Basque, 30–31, 33–37, 143 Bengali, 14 Berber, 2, 4, 5–6, 8, 11, 12 Bilin, 135, 136 Bulgarian, 132, 134, 135 Catalan, 132, 133, 148 Chagatay, 147 Chamorro, 155, 159, 165–71, 180–82 Chhitkuli, 140 Chichewa, 4, 185–88, 190–92, 197, 199, 203–204 Chinese, 233–234 n. Chitimacha, 14 Chukchee, 132, 141–42 Classical Greek, 144 Crioulo, 134 Czech, 143 Dargwa, 140 Diyari, 261 Dutch, 134 East Franconian, 135 English, 9, 14, 16–17, 42, 65, 67, 73, 132, 134, 148, 211, 218, 220–21, 229–30, 262 Estonian, 140 Finnish, 143, 255–256 French, 63, 65–67, 74–75, 133–34, 226–27, 244, 259 Frisian, 134 German, 132, 143–44, 147 Germanic languages, 135 Gikuyu, 197, 205 Gooniyandi, 261–62 Gothic, 143 Greek, 134–35 Gurindji, 86, 117–22 Hindi, 133, 215, 217–18. See also Hindi-Urdu

< previous page

page_273

next page >

< previous page

page_274

Page 274 Hindi-Urdu, 16. See also Urdu Hungarian, 138, 143, 255–56 Japanese, 211–12, 213–14, 216–17, 218, 222–23, 224–25, 263 Jarawara, 4, 7–9 Kanashi, 132, 137, 138 Kannada, 261 Khamtanga, 135–36 Kikamba, 49–53 Kinyarwanda, 174 Kirghiz, 146–47 Korean, 174, 245 Kwa, 192 Ladakhi, 136 Lakhota, 174 Latin, 132, 143–44, 146, 148 Latvian, 150 Lezgian, 141, 143, 145, 149 Lithuanian, 146 Luganda, 190 Lunda-Ndembu, 201 Malagasy, 155, 159, 160–65, 180–82 Malayalam, 2, 4–5, 9, 12, 14, 15 Mansi, 132, 137–38 Maxacali, 258–59 Meithei, 14 Menomini, 10 Middle Dutch, 143 Mohawk, 196 Mongolian, 148–9 Mudburra, 86, 117–18, 128 Ngumpin languages, 117, 119 Ngumpin-yapa, 86, 117–19, 122–23, 128 Nimboran, 242 Northern Saami, 136 Old English, 143, 148 Onondaga, 11 Orkhon Turkic, 147 Ossetic, 143 Pama-Nyungan, 123 Philippine-type languages, 155, 158, 168, 171, 173–74, 180, 182 Pipil, 11 Polish, 42, 45–49, 54–55 Portugese, 133 Punjabi, 132–33 Quechua, 30–33 Romance languages, 133, 196 Romanian, 137 Russian, 143, 146, 148 Sanskrit, 143, 146 Sesotho, 203 Setswana, 197 Sharma, 140 Sindhi, 132, 141 Slovak, 146 Spanish, 21–40, 133 Sre, 9 Swahili, 42, 56

next page >

Swedish, 134, 242 Tagalog, 155, 159, 172–82 Tamil, 136, 143 Tariana, 4, 7, 9 Tokharian A, 140 Tsonga, 206 Turkish, 132, 143–44, 146–47, 149–50, 246, 256 Turkmen, 147 Udmurt, 137–38 Uigur, 147 Urdu, 133. See also Hindi-Urdu Urubu-Kaapor, 8 Vedic, 146–47 Vietnamese, 245–46 Walmajarri, 86, 128 Warlpiri, 83–130 Western Austronesian languages, 155–84 Yagnob, 142 Yakut, 246 Yiddish, 132 Yimas, 261

< previous page

page_274

next page >

< previous page

page_275

next page >

Page 275 Subject Index Absolutive, 145, 166, 167, 168, 170. See also Case Abstractness, 263–64 Accusative, 4, 134–35, 147, 163, 172, 174; languages, 158, 160, 162, 173; systems, 155; inherent case, 172. See also Case Acquired language deficit, 199–200 Adposition, 131–54 Affixes, 185–210 Agent, 13 Agrammatism, 200 Agreement, 62, 64, 166 AgrPP, 150 Ambitransitivity, 10–12 Antipassive, 155, 162, 173–74, 179, 180, 182 Aphasia, 200 Applicative, 159, 189, 201 Argument, external, 65, 71; internal, 65, 71 Argument structure, 13 Aspect, 88, 108–109, 118, 121, 124, 128; recent perfective, 177–79 Attract, 228 Auxiliary/AUX, 83–130; augmented, 89–92; lowering, 99–100; straddling, 100–12; negative, 112–15 Case, 21–40, 83–84, 155–56, 166; instrumental, 4; secondary, 139–41; compound, 140–41; tertiary cases, 140; vocative, 145; checking, 172; genitive, 135–37, 162, 164; core, 182; case phrase, 93, 94–97, 122; locative, 141–42, 146; dative, 15, 135, 146. See also Accusative, Ergative, Nominative Causative, 5–10, 159–60, 189, 201 CHILDES database, 69 Clausal architecture, 75–78 Clitic, 21–40, 185–210 Cliticization, 186 Complementizer/COMP, 86, 91–92, 96 Complexity, 251, 253 Computational bottleneck, 72–73 Condition on Extraction Domain, 216–18, 231–32 Constraint, 26–27, 36; marking, 162; markedness, 157 Contact dialects, 28–37 Copying, 72 Dative shift, 15 Demotion, 180

< previous page

page_275

next page >

< previous page

page_276

Page 276 Dialect Distinguishability Principle, 243 Directional phrase, 104, 106 Discontinuous constituent, 84, 108 Dispersion theory, 263 Ditransitive, 1, 17 Double object construction, 135 Early child English, 62, 69, 72–75 Enclitic, 87–88, 115, 117, 187–88; directional, 102–110 Ergative, 141, 145, 158, 165, 169–70, 173; split, 182; systems, 155. See also Case Etymological dialect, 22–23, 27 Evidential particle, 110–111 Expletive (subject), 47, 51 Extension, 189 Faithfulness constraint, 26–27, 157, 173, 181–82 Feature movement, 213, 220, 225–32 Features, 241–43 Floating quantifier, 64 Focus, 111–12, 114, 118, 197 Frequency, 252 Generalized binding, 221–23 Goal, 12, 13, 17, 159. See also The-matic (role). Head Movement Constraint, 195 Head-Driven Phrase Structure Grammar, 158 Head-initial language, 63, 74 Imperative, 86, 115–22 Incorporation, 195 Indefinite, 167 Ingestive verbs/predicates, 1–19 Internal Subject Hypothesis, 63–64 Intransitive, 71, 156, 161 Language acquisition, 61–81, 202–204 Language change, 201 Level ordered morphology, 195 Lexical Conceptual Structure, 11–12, 14, 16, 17 Lexical Functional Grammar, 186, 197–98 Lexical Phonology, 195 Lexical Syntax, 195 Linking approach, 158 Markedness, 26, 250–53 Mathematical cognition, 57–58 Minimal Distance Constraint, 263 Minimalist Program, 7 Modal, 128; operator, 119 Mood, irrealis, 168, 170 Morphological, merger, 186; processes, 186–88 Morphologization, 201 Morphosyntax, 185–210 Natural classes, 254–57 Negation, 116, 120, 203 Nominal clause, 114 Nominative, 137–39, 145, 162–63. See also Case Null operator, 219–20, 225–32

next page >

Object shift, 65–67, 72, 76–77 Obligatory Contour Principle, 255 Operator, 113–14, 116, 122 Optimality Theory, 21, 25, 155–84 Parsing strategies, 204 Passive, 41–59, 156, 162, 164, 173–74, 179, 180 Performance error, 73–74 Phonemic Availability Principle, 242–43 Phonological processes, 254–57 Postverbal object, 70 Predicate Argument Structure, 11–12 Predicate raising, 195 Preverb, 100 Preverbal object, 62–63, 67–71, 76 Principles and Parameters Theory, 194, 198 Pro-drop, 203 Prominence, 160, 173–74, 180–81; low prominence, 177–80; super prominence, 180

< previous page

page_276

next page >

< previous page

page_277

next page >

Page 277 Pronominal, forms, 87, 166; strong pronouns, 133 Prepositional particle, 110–12 Prototype, 56–58 Referential dialect, 23–25, 27–28 Relational Grammar, 164 Resultative participle, 16 Secondary articulation, 249–50 Self-correction, 70–71, 73 Small clause, 43–44, 54–56 Split ergativity, 168 Split Infl Hypothesis, 64 Split Morphology Hypothesis, 195 Split VP Hypothesis, 62, 64–67, 73–76, 78 Standard Theory, of Transformational Grammar, 195 Suppletive, 9 Tense, 64 Thematic, hierarchy, 13; role, 204 Theme/patient, 12–14 Three-place predicate, 12 Topic, 70, 111–112, 122, 197; circumstantial, 159, 162, 169, 171, 175–77 Topicalisation, 70 Transitivity, 1–19. See also Unaccusative, Unergative, Verb Unaccusative, 5, 16. See also Transitivity Underspecification, 25, 32 Unergative, 5. See also Transitivity Universal grammar, 186, 197–99 Unselective binding, 219, 223–25 Verb, raising, 67; root, 185; stem, 194; transitive, 156. See also Transitivity Verbal suffixation, 189–92; prefixation, 192–93; reduplication, 191 Vowel, harmony, 101, 124, 190, 255–57 Vowel place, 239–69

Wh -in-situ, 211–38 Wh -island, 218 Wh -movement, 211y19612, 220 Word order, 61, 67, 73

< previous page

page_277

next page >

< previous page

page_278

next page >

page_278

next page >

Page 278 This page intentionally left blank.

< previous page

< previous page

page_279

next page >

Page 279 About the Editors and Contributors MENGISTU AMBERBER is a lecturer in the Linguistics Department at the University of New South Wales. His main research interests include morphosyntactic theory (with particular reference to generative grammar) , the semantics-syntax interface and linguistic typology. JOSÉ CAMACHO is an assistant professor of Spanish and Portuguese and Linguistics at Rutgers University. His research interests are Spanish Syntax, dialectal variation in Spanish, and Amazonian linguistics, in particular the study of Capanahua, a Pano language from Peru. PETER COLLINS is an associate professor of linguistics and Head of the Linguistics Department at the University of New South Wales. His main areas of interest are grammatical theory and description, corpus linguistics and Australian English. His most recent book is English in Australia (with David Blair). PETER F.KIPKA, who lectures on syntax and phonetics in the School of Human Communication Sciences at La Trobe University, is researching the extent to which the architecture of the human language faculty casts light on cognition, information processing and skill acquisition. MASATOSHI KOIZUMI is an associate professor of linguistics at Tohoku University. His research interests are in syntactic theory and language

< previous page

page_279

next page >

< previous page

page_280

Page 280 acquisition. He is author of Phrase Structure in Minimalist Syntax and coauthor of Bun-no Kozo (Clausal Architecture). MARY LAUGHREN is a lecturer in linguistics at the University of Queensland. Her research interests are in Australian Aboriginal languages and the interface between syntax and semantics. She is a coauthor of A Learner’s Guide to Warlpiri and a co-editor of Forty Years On: Ken Hale and Australian languages . ALAN R.LIBERT is a lecturer in linguistics in the University of Newcastle, Australia. His research interests include case theory and artificial languages. He is the author of A Priori Artificial Languages. ANNA MACLACHLAN has worked in both theoretical and computational syntax. These interests are reflected in her doctoral work at McGill University and her postdoctoral work at the natural language technology group of Macquarie University’s computing department. She is currently a senior computational linguist at Discern Communications in Menlo Park, California. SAM MCHOMBO is an associate professor of linguistics at the University of California, Berkeley. His research interests are: African linguistic structure (especially Bantu morphosyntax), syntactic theory, and (Southern) African politics. He is the editor of Theoretical Aspects of Bantu Grammar (Stanford: Center for the Study of Language and Information, 1993). MASANORI NAKAMURA is an associate professor at Senshu University, Japan. His research interests are in comparative syntax, locality principles, and grammatical functions. He has published in such professional journals as Linguistic Inquiry, The Linguistic Review, and Lingua . KEREN RICE is a professor of linguistics at the University of Toronto. Her research interests are in phonological representations and the phonology and morphology of Athapaskan languages. Her most recent book is Morpheme order and semantic scope: Word formation in the Athapaskan verb. LILIANA SÁNCHEZ is an assistant professor of Spanish and Portuguese at Rutgers University. Her research interests are Spanish Syntax, Generative Second Language Acquisition and Bilingualism. Her work focuses on Quechua-Spanish Bilingualism.

< previous page

page_280