Formal Approaches to Function in Grammar: In Honor of Eloise Jelinek (Linguistik Aktuell Linguistics Today, LA 62)

Formal Approaches to Function in Grammar Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (L...

Author: Andrew Carnie | Heidi Harley | Maryann Willie | Eloise Jelinek

24 downloads 757 Views 4MB Size Report

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

DOWNLOAD PDF

Formal Approaches to Function in Grammar

Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

Series Editor Werner Abraham University of Vienna

Advisory Editorial Board Guglielmo Cinque

Ian Roberts

University of Venice

Cambridge University

Günther Grewendorf

Ken Safir

J.W. Goethe-University, Frankfurt

Rutgers University, New Brunswick NJ

Liliane Haegeman

Lisa deMena Travis

University of Lille, France

McGill University

Hubert Haider

Sten Vikner

University of Salzburg

University of Aarhus

Christer Platzack

C. Jan-Wouter Zwart

University of Lund

University of Groningen

Volume 62 Formal Approaches to Function in Grammar: In honor of Eloise Jelinek Edited by Andrew Carnie, Heidi Harley and MaryAnn Willie

Formal Approaches to Function in Grammar In honor of Eloise Jelinek

Edited by

Andrew Carnie Heidi Harley MaryAnn Willie University of Arizona

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Formal approaches to function in grammar : In honor of Eloise Jelinek / edited by Andrew Carnie, Heidi Harley and MaryAnn Willie. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 62) Includes bibliographical references and index. 1. Grammar, Comparative and general. 2. Functionalism (Linguistics) I. Carnie, Andrew, 1969- II. Harley, Heidi. III. Willie, Mary. IV. Jelinek, Eloise. V. Linguistik aktuell ; Bd. 62. P151.F622003 415-dc21 2002038236 isbn 90 272 2785 3 (Eur.) / 1 58811 348 5 (US) (Hb : acid-free paper)

© 2003 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

’Ahéhee’ Eloise. Saad na’alkaah nił bééhózinígíí bik’ehgo nits’¢ø¢ø’dóó ’ídahwiil’¢øø¢’. Díí saad bikáá’ sinilígíí nihił níl˛í˛igo dóó k’é daniidzingo ná ’áadeiilyaa’.

Table of contents

Contributors

ix

Acknowledgments

xi

Introduction: Formalizing Functionalism Andrew Carnie and Heidi Harley Part I The Pronominal Argument Hypothesis On the significance of Eloise Jelinek’s Pronominal Argument Hypothesis Ken Hale

1

11

Categories and pronominal arguments Emmon Bach

45

Doubling by agreement in Slave (Northern Athapaskan) Keren Rice

51

Quasi objects in St’át’imcets: On the (semi-)independence of agreement and Case Henry Davis and Lisa Matthewson Agreement, dislocation, and partial configurationality Mark C. Baker Part I Interfaces Multiple multiple questions Molly Diesing Attitude evaluation in complex NPs Lynn Nichols Topic-Focus articulation and degrees of salience in the Prague Dependency Treebank Petr Sgall, Eva Hajicˇová and Eva Buránˇová Word order and discourse genre in Tohono O’odham Colleen M. Fitzgerald

79

107

135

155

165

179

viii Table of contents

The prosody of interrogative and focus constructions in Navajo Joyce McDonough Subject number agreement, grammaticalization, and transitivity in the Cupeño verb construction Jane H. Hill Lexical irregularity in OT: DOT vs. variable constraint ranking Diana Archangeli Rapid perceptibility as a factor underlying universals of vowel inventories Natasha Warner Part I Foundational issues Argument hierarchies and the mapping principle Eloise Jelinek and Andrew Carnie

191

207

227

245

265

Focus movement and the nature of uninterpretable features Simin Karimi

297

Merge D. Terence Langendoen

307

Phonotactics and probabilistic ranking Michael Hammond

319

Deconstructing functionalist explanations of linguistic universals Thomas G. Bever

333

References

353

Name index

369

Subject index

371

Contributors

Diana Archangeli, University of Arizona Emmon Bach, University of Massachusetts, Amherst Mark Baker, Rutgers University Thomas Bever, University of Arizona Eva Buránˇová, Charles University in Prague Andrew Carnie, University of Arizona Henry Davis, University of British Columbia Molly Diesing, Cornell University Colleen Fitzgerald, Texas Tech University Ken Hale†, Massachusetts Institute of Technology Eva Hajicˇová, Charles University in Prague Heidi Harley, University of Arizona Michael Hammond, University of Arizona Jane Hill, University of Arizona Eloise Jelinek, University of Arizona Simin Karimi, University of Arizona Terry Langendoen, University of Arizona Lisa Matthewson, University of Massachusetts, Amherst Joyce McDonough, University of Rochester Lynn Nichols, University of California, Berkeley Keren Rice, University of Toronto Petr Sgall, Charles University in Prague Natasha Warner, University of Arizona

Acknowledgments

A number of colleagues and friends helped in commenting on and refereeing the papers in this volume. We would like to thank Mark Baker, Luis Barragan, Thomas Bever, Sonya Bird, Noam Chomsky, Megan Crowhurst, Molly Diesing, Richard Demers, Sheila Dooley Collberg, Chris Golston, Robert Franks, Paul Hagstrom, Daniel Hall-Currie, Mike Hammond, Jason Haugen, Simin Karimi, Jaklin Kornfilt, Terry Langendoen, Adrienne Lehrer, Alec Marantz, Joyce McDonough, Martha McGinnis, Fritz Newmeyer, Massimo Piatelli-Palmarini, Norvin Richards, Keren Rice, Leslie Saxon, Laura Siegal, Margeret Speas, Adam Ussishkin, Andy Wedel, Natasha Warner and Mary Zampini for their help. Thanks also to Kent Walker for his help in compiling the references. We would also like to thank Werner Abraham and Kees Vaes for their advice and assistance in bringing this volume to publication. Of course, our most profound thanks go to our honored colleague, Eloise Jelinek, for both her inspiration as a scholar as well as her guidance and friendship.

Introduction Formalizing Functionalism Andrew Carnie and Heidi Harley University of Arizona

1.

Formal vs. Functional

At least since Chomsky (1957, 1965), the field of linguistics has been roughly divided into two camps concerning the philosophical, methodological, and empirical domain of the discipline. Traditional grammarians, following the early work of the Prague School of linguistics, as well as linguists coming from approaches as diverse as computational neural net theory and typological descriptive traditions, identify themselves primarily as functional linguists. The term is meant to evoke the idea that the driving force in linguistic study is a focus on the (communicative) function of grammatical structure. The other approach, which found its origins in the American structuralist school of linguistics (such as the work by Bloomfield) but became solidified in Chomskyan generative grammar and related approaches, is focused on the form and structure of the language, rather than on its communicative (and other) functions. This view is commonly called formalism. The labels functionalist and formalist represent an unfortunate oversimplification of very complex foundational issues. There is very little agreement among linguists about which particular assumptions or methodologies mark one as a functionalist or formalist. There are a number of important works on the question of defining the camps, including Croft (1995) and Newmeyer (1998), and the collection of articles in the volumes edited by Darnell, Moravcsik et al. (1999). We cannot hope to do justice to the complexity of the issue here, but will offer a brief synopsis of some of the leading lines of thought. Part of the problem of defining the terms functional and formal lies in the fact that among practitioners of functionalism there is little agreement as to what characteristics a functional theory must have. On the side of formalist theories (such as Minimalism, HPSG and LFG), there is a greater agreement about the domain of the discipline, but still important foundational disagreements about methodology and implementation. As such, defining what constitutes a formal theory of grammar or a functional one is notoriously difficult.

2

Andrew Carnie and Heidi Harley

Essentially, the formal/functional distinction turns on one’s position along a number of dimensions. A theory’s alignment as functionalist or formalist seems to depend on how many of these dimensions that one takes a more or less functional position on. In no particular order these dimensions include (these dimensions are based on work of Croft (1995) and Newmeyer (1998)): 1. The role of structure in grammatical theory: Less reliance on structure makes one more functionalist.1 2. The role of arbitrariness in grammar: Formalists hold the strict Saussurian view that language, including the grammar, is essentially arbitrary; whereas the most radical of functionalist approaches only see arbitrariness in the lexicon. Less radical functionalists (e.g. Cognitive Grammarians) view arbitrariness in the grammar and the lexicon to be of a like-kind, that is: grammatical arbitrariness is essentially lexical arbitrariness.2 3. The autonomy of syntax:3 Many formalist theories hold that there are a number of grammatical phenomena that allow formal characterization without reference to their semantic or pragmatic function. By contrast, all functionalist approaches view the importance of the semantic or pragmatic function of grammatical phenomena to be so overwhelming in shaping the grammatical form, that speaking of the grammatical form in isolation is meaningless. 4. The diachronic/synchronic distinction: Formalist theories roughly hold that the goal of linguistic theory is to characterize the grammatical system of a speaker at a given moment in time, without reference to the historical pressures that gave rise to that grammatical system. Functionalists, hold that a full characterization of the grammatical system is incomplete without an understanding of the historical events that gave rise to it. 5. The competence/performance distinction:4 Formalist grammarians hold that there is core grammatical knowledge which can be characterized independently of the production/comprehension system that realizes it. Functionalist grammarians, on the whole, argue that the performance system and the competence system are isomorphic. 6. A methodological issue over what constitutes “data” for linguistic study: Formalist grammarians usually, although not uniformly, focus on grammaticality judgements, typological comparison and data from language acquisition. Functionalists often, but not always, exclude grammaticality judgements, focusing instead on statistical corpus analysis. They also include historical and sociological data which is regularly excluded from formalist theories. These broad distinctions among theoretical approaches have given rise to an interesting split in the discipline with respect to particular linguistic phenomena. Some phenomena are treated almost exclusively by functionalist theoreticians, and

Introduction

others exclusively by formalists. For example, with a few notable exceptions, the study of information structure, and the discourse-syntax interface is primarily done in the functionalist traditions. It is this distinction in terms of empirical coverage that we refer to by the “‘Function” in the title of this book. In this volume we gather together papers by well-known authors writing in formalist (primarily Chomskyan) frameworks about many of the phenomena traditionally treated only in functionalist traditions.

2.

Eloise Jelinek and formal functionalism

For the past 20 years, the work of Eloise Jelinek, while she self-identifies as a diehard formalist, has been a driving force in showing formalist linguists how to treat material previously considered only within functionalist traditions. Jelinek’s work on lesser studied languages, including Native American, Semitic and Australian languages, often focussing on phenomena apparently controlled by external factors, such as information structure, has shown other formal linguists how to approach cases of interface between purely “grammatical” (i.e., syntactic, morphological, or phonological) phenomena and the discourse, semantic, phonetic or conceptual forces that drive them. Indeed, the most recent version of the Chomskyan Principles and Parameters approach, Minimalism (Chomsky 1995b, 2001a,b), explicitly claims that cross-linguistic and language-internal variation in the syntactic computational component stems from the interface between that component and other systems such as the conceptual/interpretive system or the auditory/production system. Much of the credit for the change from purely syntactic modeling to detailed consideration of interface conditions, should be attributed to Eloise Jelinek (and, of course, her collaborators and colleagues who investigate the same issues).5 Formal grammar is, for the most part,6 structuralist. On a purely surface level, many of (if not the majority of) the world’s languages exhibit syntactic phenomena that appear problematic for structure-based accounts. Purely structural accounts, such as a phrase structure grammar or a transformational grammar of the EST type, suggest that grammatical and semantic information is read directly off of the structure of the syntactic tree. For example, in the EST, the subject of an English sentence is the daughter of the S node, and sister to the VP. Non-configurational7 languages, such as Warlpiri, allow a multitude of word orders (and by hypothesis a multitude of structures). For such languages, a one to one mapping between structure and grammatical/semantic function seems, at least on the surface, to be severely misguided. This conclusion led many functionalists and formalists (as in the LFG and RG traditions) to abandon structural accounts of word order phenomena. Within the structuralist/formalist traditions, many attempts were made to account for these languages: Hale (1982) proposed that there was a configurationality

3

4


parameter, whereby non-configurational languages didn’t use X-bar structure, but used a non-configurational phrase structure rule instead: S Æ … X*… which allowed free ordering of elements. This type of approach was challenged by the fact that even in non-configurational languages there appear to be some phenomena which are sensitive to hierarchical structure. Ross (1967) proposed that these languages had a stylistic or purely phonological rule which took a fully specified hierarchical structure and “scrambled”8 the elements. Neither of these approaches is terribly satisfying in light of the evidence that word order in non-configurational languages, when examined in a closer light, appears to be determined by discourselevel considerations, such as topic, focus or narrative force. In traditional formal autonomous syntax, such “external” factors are difficult to account for since they essentially involve access to contextual (and thus presumably performance and other extralinguistic) information. Jelinek’s work, however, has shown that an autonomous syntactic component does not necessarily have to operate in a vacuum. Instead, by looking at the interfaces with such domains as discourse information structure, we can distinguish between purely syntactic/grammatical phenomena and those grammatical phenomena that draw information from other modules. Jelinek (1984), a seminal paper in this domain, proposed the Pronominal Argument Hypothesis (PAH), which broke new ground in providing an explanation of why non-configurational languages show some sensitivity to structural hierarchical structural information, but appear to resist an analysis of their surface word orders in terms of structure. The PA hypothesis claims that languages are parameterized into lexical argument languages (LA), which allow full NPs to be arguments of a predicate, and pronominal argument languages (PA) which only allow pronouns (or agreement which represents pronouns) in argument positions. Full lexical NPs in PA languages are not in argument positions, but are adjoined to the structure, similar to left-dislocation structures in English. The surface word order of lexical elements is determined by discourse-configurational considerations (which may or may not also map to particular structural positions in the tree, such as the specifiers of Topic or Focus functional projections). Phenomena that appear to be sensitive to traditional hierarchical “syntactic” constraints make reference to the pronominal arguments. Those that are sensitive to discourse refer to the adjunct NPs. Closely tied to this work on discourse-syntax interaction is Jelinek’s work on the mapping between the syntactic structure, semantic notions such as specificity and definiteness, and grammatical notions traditionally characterized in the functionalist literature9 in terms of accessibility hierarchies (see, for example, Keenan (1976); Keenan and Comrie (1977); Givon (1994) and many others). Because these notions refer primarily to semantic and world knowledge, they are not easily dealt with in traditional autonomous-syntax models. Jelinek’s work, both alone and in collaboration with Molly Diesing, has proposed that such abstract hierarchies correspond, in fact, to hierarchy imposed by the phrase structure.

Introduction

Relational and semantic hierarchies are “mapped” off of the syntactic structure. A number of such hierarchical effects, including person, animacy, specificity, and topic/focus structure result from mapping between specific positions in the syntactic tree and semantic structure. Positions high in the syntactic tree correspond directly to items that are high in the semantic hierarchy. The contribution from Carnie and Jelinek in this volume is an attempt to apply this insight to a wide range of data, considering an interaction of phases and mapping domains. In pursuing these areas, Eloise Jelinek has shown that formal Chomskyan grammar can, with minor modifications, provide explanatory and interesting accounts of phenomena traditionally dealt with only by functionalist theoreticians.

3.

Formal grammars of function

The papers we have collected together in this volume continue in this spirit. They either directly address questions that Jelinek has raised (e.g., the papers of Baker, Hale, Davis & Matthewson, and Bach) or pursue other traditionally “functional” topics in the spirit of a formal approach (e.g. Langendoen, Diesing, Rice, McDonough, Sgall et al., etc.) The papers address both syntactic/semantic/pragmatic questions and phonological/phonetic ones, where the formal/functional divide also exists. Addressing to the relationship between Jelinek’s Pronominal Argument Hypothesis, the structural representation of grammatical relations, and definiteness, Ken Hale presents a non-movement account of the Navajo verb form and nominal incorporation, which he argues is a necessary consequence of the PAH. Emmon Bach explores the consequences of the PAH for composition, adopting a strict semantic type/syntactic category mapping hypothesis. In particular, he suggests that the way in which overt DP arguments are composed with verbal predicates must differ radically in pronominal argument languages, since the verb-word denotes a saturated proposition (or truth value), rather than an argument-taking function. Bach concludes that the PAH provides a framework in which one can characterize this type of parametric variation as variation in the availability of V and VP as categories: these categories are not available in PA languages. Also dealing with parametric variation and the PAH, and perforce adopting a similar hypothesis about composition of dislocated DP arguments, Mark Baker treats the semiconfigurational properties of Kinande, a Bantu language, relating them to similar semi-configurational properties observed for Romance languages by Alexiadou and Agnosopolou (1998). Baker relates these properties to the presence or absence, and the obligatoriness or optionality, of subject and object agreement in Kinnande. He proposes that a particular formulation of the PAH can explain the behavior of agreeing and non-agreeing DPs within a single language. Keren Rice examines the interaction of agreement doubling with the famous y-/b-alternation in the Northern

5

6


Athabaskan language Slave. According to the PAH, pronominal argument languages should always allow co-occurrence of overt DPs with pronominal agreement (the pronominal arguments). Northern Athabaskan languages, which in general do not allow doubling of agreement and overt DP arguments, have been treated as nonPAH languages. Rice demonstrates that in at least some contexts, Slave allows doubling of agreement, and discusses the theoretical consequences of this result for the PAH. In the last paper dealing explicitly with the PAH, Henry Davis and Lisa Matthewson outline a treatment of some non-agreeing objects in St’át’imcets (Straits Salish), which they argue are problematic for a PAH approach. They demonstrate that the interaction of inherent plurality of events and determiners on these objects can be naturally treated as a scope phenomenon. Turning to function more broadly construed as an interface issue, Molly Diesing investigates multiple fronting in wh-questions in Yiddish. This is a very current topic of investigation, and several formal accounts in terms of wh-features have been recently proposed (Richards 1997, Pesetsky 2000, among many others). In the grammar of Yiddish, multiple wh-fronting co-occurs with single-fronting and non-fronting constructions, and Diesing shows that the multiply-fronted construction is strongly conditioned by D(iscourse)-Linking, in the sense of Pesetsky (1987). Lynn Nichols also deals with constraints on extraction, in the context of another currently heated debate about the syntactic realization of Davidson’s event argument. She proposes that the event argument itself can be relativized, which produces a relative clause that has in the past been misanalyzed as a clausal complement to ‘evaluative’ nominals like claim, fact, or theory. Such nominals identify the speaker’s attitude toward the proposition; Nichols shows that the relative-clause hypothesis can explain certain morphological and syntactic facts in English, Burmese and Turkish. Her paper is very much in the Jelinek tradition: she demonstrates the existence of structural consequences of a factor (speaker’s evaluation of reliability) which past formalists would have marginalized as not relevant to the grammar. Eva Buránˇová, Eva Hajicˇová and Petr Sgall propose to investigate the relationship between topic, focus and sentence structure in the Functional Generative Description framework. They divide the syntactic tree into Focus and Topic subtrees, much in the spirit of the mapping hypothesis pioneered by Diesing and Jelinek. In a third paper on focus constructions, looking at phonology/discourse interactions, Joyce McDonough reports the results of an important study of pitch excursions and focus in Navajo, and explores the possibility that Jelinek’s Pronominal Argument Hypothesis can provide the basis for an explanation of the (lack of) intonational contours that appear in Navajo focus constructions. Finally, Colleen Fitzgerald investigates the three-way interface between discourse structure, syntactic structure and rhythmic structure in Tohono O’odham, concluding in particular that rhythmic constraints must also be taken into account in determining surface word order in the language.

Introduction

Morphology/syntax interface questions are addressed by Jane Hill in her paper on the Cupeño verb and the appropriate functional projections in which to situate Cupeño’s valence-affecting affixes. Hill draws heavily on Jelinek’s work on Yaqui (Hiaki, Yoeme) (Jelinek 1996), in which Jelinek posits two valence-affecting functional heads: TransP and VoiceP. Hill also discusses how a formal approach provides a grounding for the historical and functional pressures that gave rise to the system. Diana Archangeli’s paper investigates questions concerning the morphology/phonology interface within Optimality Theory. She examines the morpheme structure constraints operative in Tiv, concluding that a theory which permits direct reference to morpheme classes by OT constraints is empirically and theoretically superior to either a strictly ranked grammar where irregularities are simply encoded as lexically specified variation, or to a ‘co-phonology’ model where the phonological idiosyncrasies of different morpheme classes can be associated with different constraint rankings. In so doing she is able to account for ‘irregular’ variation in a systematic way that has previously escaped formal treatment. Finally, examining the phonology/phonetics interface, Natasha Warner investigates the relationship between the functional dimension of segment perceptibility and typological frequency of distinctive feature. She argues in support of the hypothesis that more quickly perceived distinctions are more likely to show up as grammaticized distinctive features in the grammars of languages. The intuition is that the more quickly perceptible a feature is, the greater its functional utility in speech, and that this pattern is reflected in the typology of grammars. In the final section, we present papers dealing with foundational questions of grammatical structure and operations. The rule-based, grammatical/ungrammatical generation systems of formalist doctrine operate on the assumption that “extralinguistic” factors, such as frequency, should not play a part in the organization of the grammatical system. Hammond, using the formal devices of Optimality Theory, explores the question of how we can incorporate these frequency effects into a formal grammatical theory. In the spirit of the minimalist program, Terry Langendoen presents a new proposal for the formal generation of structured linguistic expressions, formally modifying Chomsky’s proposed family of Merge operations, which are the heart of the generative engine of modern formal linguistic theory. Langendoen’s conclusions about Merge address the long-standing tension between the straightforwardly functional limitations imposed by computational capacity and the formal expressiveness needed to generate the recursive structures of natural language. Also working within the minimalist program, Simin Karimi addresses the foundational issue of the nature of the features whose checking drives movement, in particular examining the theoretical status of so-called ‘uninterpretable’ features, contrasted with interpretable features which drive the movement of wh-phrases. Discussing two types of focus construction in Persian, only one of which seems to involve overt movement, she argues that focus is indeed an instance of the operation

7

8


Move, motivated by the need to check a syntactic feature. Finally, Tom Bever asks the question that formalists avoid and functionalists embrace: given that the formal grammatical systems are the way (formalists say) they are, why are they that way? Bever offers some general constraints on functionalist answers to this question, and broadens the scope of what is usually considered a functional explanation. He considers this alternative version of function-as-explanation from three perspectives: language perception, language acquisition, and the biological sources of linguistic ability.

Notes †

Sadly, Ken Hale passed away in October, 2001 before this volume was completed. He was a driving force behind this volume, encouraging us to put it together and was a great fan of the work of Eloise Jelinek. 1. Although there are many structuralist functionalist theories, such as Dik’s Functionalism or Van Valin’s Role and Reference Grammar. See Noonan (1999) for discussion of the role of structuralism in functionalist theories. 2. See Langacker (1987) on the list-rule fallacy. 3. See Newmeyer (1998) for a discussion of the various usages of this term (meaning variously: autonomy of the grammatical system, autonomy of the competence system, or autonomy of the syntactic system). 4. This question is intimately connected to the autonomy question mentioned in footnote 3 above and the data issue listed in number (6). 5. Such as Ken Hale, Ellen Prince, Molly Diesing, to name only a few. 6. We exclude Categorial Grammar and LFG, since these theories, although formal, don’t take a structuralist approach to the phenomena discussed here. 7. The term “non-configurational” has been variously used to describe free word order or scrambling languages, as well as languages that have fixed word order but may lack the hierarchical effects associated with having a VP (such as Japanese, Navajo and VSO languages, see Speas 1990 for discussion). We are using the term only in the former (free word order) usage here. However, Jelinek’s approach extends to the latter usage as well. 8. Ross’s notion of scrambling and the current conception of the phenomenon are very different, see below and Karimi (this volume) for discussion. 9. Recently, a number of papers in Optimality Theory have also attempted to deal with these questions, see Aissen (1999); Bresnan et al. (2001) for more details.

Part I

The Pronominal Argument Hypothesis

On the significance of Eloise Jelinek’s Pronominal Argument Hypothesis Ken Hale Massachusetts Institute of Technology

Introduction In 1984, Eloise Jelinek proposed a theory of Navajo grammar designed to account for certain observations which indicated that the organization of clauses in the language was “nonconfigurational” in a very particular sense (Jelinek 1984). Her framework has since become widely known, and the name Pronominal Argument (PA) is now attached to a wide range of polysynthetic “head-marking languages” (cf. Baker 1996; Nichols 1986). She has written a number of works developing the PA model in relation not only to Southern Athabaskan but to Salish as well (e.g., Jelinek and Demers 1994). In this essay, I will review certain aspects of the PA hypothesis, with emphasis on its relevance (i) to the grammar of Navajo and (ii) to the grammar of noun incorporation. In past work on the Navajo verb (e.g., Hale 2000b), I assumed that the surface organization of the elements appearing within it was due to a series of movement rules (Head Movement). The verb began its upward journey from a position low in the syntactic structure of a sentence, moving head to head within its extended projection, leaving certain residue (non-nuclear elements). In the first step of this scenario, the verb adjoins to the right edge of the voice marker, traditionally called the Classifier in Athabaskan linguistics. In the second step, the classifier rightadjoins to Mode (an element similar in function to Infl or Tense in more familiar languages). Above Mode in the extended projection one or more Qual(ifiers) (elements with aspectual or “thematic” character) might be present, in which case, Mode right adjoins to the first qualifier, which then right adjoins to the next qualifier, if present, and so on, resulting in a complex word whose surface linear form has the verb on the right-hand edge preceded by “prefixes” corresponding to nuclear elements picked up in the course of cyclic successive head to head movement: [(Qual) [(Qual) [Mode [Voice [Verb]]]]] This constitutes the tightly organized portion of the Navajo verb word known traditionally as the “conjunct” sector. The residue left behind includes (a) particles and

12

Ken Hale

proclitics (phonologically dependent adverbial and quantifier elements) constituting the so-called “disjunct” sector and (b) full phrases corresponding to the overt arguments of the verb and various adjuncts (phrasal adverbs, including postpositional phrases). The arguments of a verb are subject to movement as well, being forced to raise in order (a) to enter into appropriate agreement relations with features of the functional heads assembled in the conjunct sector and (b) to satisfy the EPP. The part of this story having to do with assembling the conjunct sector is highly unsatisfactory, primarily because (a) there is no apparent motivation for Head Movement in the computational system, i.e., in syntax, and (b) right adjunction is unusual in Head Movement and, in this case, is used merely to get the head in the right place. The whole business seems artificial, calling into question its status as a series of processes constrained by genuine principles of syntax. The Pronominal Argument Hypothesis, proposed by Jelinek in 1984, is one among several theories of grammar within the general Principles and Parameter tradition (cf. Chomsky 1981 and subsequent allied literature by many linguists) which have, as an intrinsic feature, the idea that the internal make-up of a complex verb words is less a matter of syntactic movement than of processes of morphophonology operating upon an underlying syntactic base defined by the single operation now called Merge within the Minimalist Tradition (in Chomsky 1959b, 2001a; and see Halle and Marantz 1993 for a theory of morphophonology). In the discussion to follow, I consider certain aspects of the Navajo verb words in the light of the Pronominal Argument (PA) Hypothesis. I extend my coverage to include a brief consideration of the grammar of Noun Incorporation (as reported and analyzed in Baker 1988, 1996), entailing involvement with fully configurational languages as well as the polysynthetic, Discourse Configurational type represented by Navajo. The essential point here will be the idea that the primary adjacency relations holding within complex verb words are due to Merge alone, not to syntactic movement (cf. Bobaljik 1994). In a recent paper (Willie and Jelinek 2000), the status of Navajo as a Discourse Configurational language is convincingly argued for. This aspect of the PA status of Navajo is thoroughly discussed in that paper. Consequently, I will concentrate here on certain purely structural features of the core PA theory, which I will briefly outline in Section 1 (some of which is purloined from Hale, Munro, and Platero 2000).

1.

A sketch of the Pronominal Argument Hypothesis

In a language belonging to the PA type, the person-number morphology internal to a verb word represents the direct arguments of the verb. These elements are not agreement morphology. Instead they are the arguments, pure and simple. In the Navajo verb

On the significance of Eloise Jelinek’s Pronominal Argument Hypothesis

word (or rather, somewhat more accurately, “verb sentence”) cited in (1) below, the prefixes ni- and share, respectively the object and subject of the clause:1 (1) ni-sh-hozh. 2sg-1sg-tickle “I tickle you.” There are no “small pro” elements in this sentence, and if an independent pronoun appeared, as in (2), it would not be an argument but rather it would be a contrastive adjunct: (2) Ni ni-sh-hozh. “I tickle YOU.” The independent pronoun ni ‘you’ is, to be sure, linked to the verb-internal object ni-, but it is not an argument of the verb, any more than the first you is an argument of tickle in the English as for construction in (3): (3) As for you, I’m tickling you. In short, the Navajo independent pronoun ni ‘you’ in (2) is not related to the prefix niin the way an argument is related to agreement morphology. It is the prefix, not the independent pronoun, that represents — alone and fully — the object argument of the verb. The same can be said of a nominal expression, like ’awéé’ ‘baby’ in (4): (4) ’Awéé’ bi-’nii-sh-hóósh. baby 3inch-1sg-tickle “I start to tickle the baby.” This is an inchoative verb form, with the direct object of the verb appearing as bidirectly before the inchoative morphology -’nii- (glossed inch). Here again, the true arguments are represented by the verb-internal person-number morphology, i.e., the third person object pronoun bi- and the first person singular subject pronoun sh-. The nominal ’awéé’ ‘baby’ is an adjunct, not an argument of the verb. Its structural relation to the sentence can be compared to that of the English left-dislocated nominal the baby in (5), where the true object argument is the resumptive pronoun him: (5) The baby, I will start to tickle him/her. The idea, then, is that Navajo is a language in which all of the arguments of a verb are pronouns and, further, the pronouns in question are morphologically dependent (i.e., they are affixes, inflection). The verb word is in reality a complete sentence — a “verb sentence” (VS), although the more conventional (albeit less accurate) expression “verb word” will be occasionally used throughout this discussion. What does all this mean formally? How is the VS composed? How is (1) composed, for example? Of course, the exact make-up of that verb sentence cannot

13

14

Ken Hale

be determined by looking at its superficial form alone, since certain morphophonological processes obscure the full inventory of elements present in it (see Faltz 1998, for an excellent and lucid study of the internal structure of the Navajo verb). A slightly simplified linear representation of the components of (1) is given in (6), each glossed in accordance with traditional usage in Athabaskan linguistics: (6) Obj ni

Mode Ø

Subj sh

Cl ł

Stem ghozh

Two key elements, invisible in (1), are shown here, i.e., the phonologically null imperfective mode prefix complex, and the so-called “classifier” -ł-. These two elements represent categories which are fundamental to, and present in, all fully inflected Navajo verbs. We will assume that a PA language conforms to general principles of argument structure and, therefore, that the structural relations between nuclear elements and their arguments is to a certain extent predetermined. That is to say, a grammatical verb sentence which “converges” is equivalent to a set of binary compositions, each defined by an application of the operation Merge, satisfying all relevant grammatical requirements and achieving full interpretation — such a derivation is said to “converge” at both phonetic and semantic interfaces. A derivation which fails to satisfy the grammatical requirements, or fails to achieve full interpretation, is said to “crash” (cf. Chomsky 1995b: 226, 171). It is almost possible to characterize the internal structure of the Navajo verb sentence by forming binary groupings of the component elements, proceeding from right to left (cf. Hale, Jelinek, and Willie, to appear): (7) [Obj [Mode [Subj [Cl [Stem]]]]] To the extent that this is true, Navajo surface form preserves the grammatical relations which hold among the constituent elements. But (7) is not completely faithful to basic grammatical relations. Before proposing a modified structure for (1), I want to reconsider some of the terminology for naming the component elements of a Navajo verb of this type. Starting from the innermost constituent, the stem, I will assume that element is in fact the lexical verb (V). It is, in reality, a composite of at least two subparts, a root and a verbal host (embodying aspectual inflection). For present purposes, it is sufficient to view the stem simply as the lexical verb and to symbolize it as V. The next element, proceeding leftward, is the traditional “classifier”. This is a voice marker, or transitivity marker, sometimes called a valence marker. More accurately, voice marking is the function of the classifier in its productive uses, i.e., when it is not merely a lexicalized, synchronically inert, component of a verb. I will use the symbol v to represent it, a symbol meant to suggest both “voice” and “verb”.2 Again, like V itself, the voice element is sometimes complex, its inner composition


being a matter beyond the scope of the present discussion. For now, let it be simply v in structural diagrams. Skipping over the subject for the moment, the next nuclear category is mode, so-called in much Athabaskanist usage. I will not depart from this usage, and I will use the label M to represent this category in structural diagrams. It is the primary functional head above the verbal projection, corresponding to I(nfl) and T(ns) in much linguistic literature dealing with more familiar languages. It consists of two aspectual elements, Situation Aspect and Viewpoint Aspect (see Smith 1991, 1996, and Rice 2000c for details). There is a strong dependency relation between M and the verbal component of V. Jointly, these elements realize aspectual categories, primarily, and modality to a lesser extent. In (1), we have the so-called “zero-imperfective”, represented formally by the fact that M is phonologically null (Ø). In actuality, mode and the subject are assembled into a portmanteau (Mode/ Subj, further abbreviated to M/S in future structural diagrams). The internal make up of the Mode/Subj portmanteau is to various degrees obscured by morphophonological processes (cf. Kari 1976; Faltz 1998; Speas 1984, 1990; and McDonough 1996, 2000a). Moving now to the arguments of the verb, the subject and object in (1) are “pronouns”, in keeping with the tenets of PA theory. As Mary Willie has repeatedly pointed out in her work, first, second, and simple third person pronouns are definite in interpretation.3 I will follow the tradition according to which pronouns belong to the determiner category (D), though in diagrams I will refer to them by means of their grammatical function labels subject (S in the portmanteau M/S) and object (Obj). Unlike the other elements found in (1), which are all nuclear, hence X0, pronouns, being arguments, are (trivially) maximal projections. They appear in argument positions — e.g., in Specifier or Complement positions in relation to lexical or functional heads. A preliminary structural diagram for (1) is presented in (8): M/S

(8) M/S sh

v v S

V Obj ni

v ghozh

Obviously, this does not correspond exactly to (7). I have taken the step of representing Mode and Subject together as the portmanteau M/S, and I have simplified the notation for the object to Obj (with the understanding that the object, a pronoun, is categorially a D(eterminer)). But the most important feature in (8) is the position of the object pronoun (ni). This element is not in its surface relative

15

16

Ken Hale

order position. Instead, it is in its “thematic” position — i.e., the position it occupies as the complement (immediate sister) of the verb.4 We must of course account for the fact that surface position of the object is to the left of the mode/subject portmanteau (M/S), as if it were in the specifier position projected by that nuclear element. Technically, this deep-surface disparity could be “corrected” quite simply by raising the object to the specifier position, leaving a trace in its thematic position. But what motivates this movement? The answer could be Case Theory. Jelinek (p.c.) has often claimed that Navajo is an ergative language. If so, then V does not assign case to its object. This circumstance forces the latter to raise out of the verb phrase and into SpecInflP (i.e., SpecMP in Navajo), where it is governed by C (the complementizer), a “case-like” functional element (and hypothetical at this point). By achieving proximity to C, the object “satisfies the Case Filter” (see Bittner 1994, and Bittner and Hale 1996, for details of Case Binding Theory, whose mechanisms and principles are implicated in this idea). On this view, it is the Case Filter that forces the object to raise to SpecM, giving the surface ordering shown in (7). But this is inconsistent with the fundamental spirit of the PA theory, in which there is no motivation (in PA languages) for syntactic movement of the type just suggested. If this is correct, and Navajo is indeed a PA language, then a nonmovement account of the order of elements in the Navajo VS must exist and, more importantly, must fall out without fanfare, so to speak, from the PA theory. In this article, an essentially non-movement, primarily phonological, account of Navajo morpheme order will be explored (built in part upon the phonological theories of Speas 1984, 1990; and McDonough 1996, 2000b). If that effort is successful, then it also has implications for the analysis of noun incorporation, as well as for the significance of Jelinek’s Pronominal Argument framework. Navajo does not itself have noun incorporation, but its Northern Athabaskan cousins do, as do many other languages, of course (cf. Mithun 1984, 1986; Baker 1988). I will suggest that another consequence of the PA theory is a non-movement account of noun incorporation (cf. Rice 2000c) and, in addition, an explanation of the renowned distinction between so-called unaccusatives, transitives, and unergatives in the matter of incorporation from the subject position.

2.

Preverbs, quantifiers, and the verb stem

The transitive VS just exemplified represents a simple and quite common type. Most verbs, however, are somewhat more complex in their internal makeup, containing elements drawn from ten distinct “morpheme order slots” in the templatic display often attributed to the structure (e.g., in Young and Morgan 1987: 37–38). It is unusual to have all of these positions filled at once, but it is in fact possible, as in the VS of (9):


(9) Yisdánídashizhdoołtééł. (~ Yisdáádashizhdoołtééł.) “They (3a = 4th person) will get me back to safety.” The pieces that make up this VS are set out in (10) below, in which the roman numerals stand for the ten positions recognized in the Navajo “template”, and the glosses are chosen from the terminology which figures most prominently in Rice’s recent authoritative book (2000c) on Athabaskan morpheme order. The Navajo morphemes themselves are given in their “basic form”, except that the Iterative/ Reversionary prefix (It/Rev) is cited in its [ní]-allomorph (appearing typically before coronals), and the Mode/Subject complex is cited in the allomorph of the portmanteau (the third person of the gamma situation aspect) appropriate to this particular context:5 (10)6 I II Preverb It/Rev yisdá

III IV V VI VII/VIII IX X Plural Object Deictic Qualifier M/S Voice Stem

ní (< ná) da

sh

zh

d

w

ł

tééł

For the present, it is sufficient to note that (10) represents the surface order of the components of the VS, corresponding to the extended projection of the verb.7 The arboreal representation in (8) is of a verb much simpler in internal structure than (9), of course. The label V corresponds to Stem. The small v corresponds to the traditional “classifier” (cf. Hale 2000b), the various so-called passives of Navajo (cf. Neundorf 2000), and a number of other constructions related to transitivity. I have noted one deep-surface disparity in the Navajo verb, namely the position of the direct object (Obj). There is another such disparity that must be dealt with, involving the category Preverb, exemplified in (9), (10) by yisdá ‘to safety’. Rice (2000c) argues that the Athabaskan preverb forms a constituent with the verb stem (V) in the basic representation of VSs. In Hale (2001), I adopted this view for all preverbs, assigning the preverb a sort of “sister” adjunct relation to the verb in the lexical projection which the verb defines. However, Sharon Hargus and Siri Tuttle (p.c.) question the correctness of this for preverbs in general, relegating this sisterlike relation to Preverb–Verb combinations which are in some sense “truly lexical”, unique combinations and idioms, for example. Following my earlier assumption, I claim that preverbs are adjuncts, for reasons discussed in Hale (2001). However, in relation to their point of attachment, I believe now that productive preverbs are variable, in keeping with their generally adverbial nature. In (9), the most prominent interpretation of the preverb has both the subject (in M/S) and the object in its semantic scope; both the subject and the object correspond to individuals that “will come to be in safety”. I take this to mean that the preverb is adjoined to one or another of the phrasal projections defined by the functional heads Qualifier and

17

18

Ken Hale

M/S, i.e., to a projection which properly includes the subject. Since there is no evidence for one of these over the other, I will assume that the high adjunction site is the projection defined by the highest functional head present in the particular VS at issue. Since Qualifier selects M/S, and not the reverse, Qualifier (abbreviated Qual) is the highest functional head in (9), (10). If the qualifier were absent, then the highest head would be M/S. In any event, the structure of this verb can be diagrammed as follows (with Preverb abbreviated as Pv, Reversionary as Rev, and Plural as Pl): Qual

(11) Pv yisdá

Qual Rev ní

Qual Pl da

Qual M/S

Qual d M/S w

v V

v S Obj sh

V tééS

The deictic, or 4th person subject, is left out of this diagram (see Hale 2001, for a proposal concerning Navajo Deictic Subjects). The adjuncts — i.e., the preverbs and the quantifier elements (the event quantifiers It/Rev and the argument quantifier Pl) — correspond to the disjunct sector of the Navajo VS. In (11), their high position corresponds straightforwardly to their leftward position in the template — the ordering among them, a side issue here, is determined by Rice’s principles of semantic scope (Rice 2000c; cf. Hale 2001). But the adjuncts are not uniformly high in the structure. I believe that Rice is correct in assigning them — sometimes at least — to the lower position (which I will take to be adjunction to the verbal projection vP, comprising the stem (V) and the classifier (v) lexically assigned to it). This possibility seems to be available for productive preverbs like yisdá. Consider, for example, the following illustrative sentence from Young and Morgan (1987: 772): (12) Shideezhí bi’éétsoh taah yíłhe˛e˛zhgo bá yisdástsóós n´t’éé’ taah yígo’. “I fell into the water trying to rescue my sister’s coat when it fell in.”


The relevant verb in this sentence is repeated and segmented in (13): (13) a.

yisdástsóós “I rescue it.” b. I IV VII-VIII IX X Preverb Obj M/S v (Voice) V (Stem) yisdá Ø sh ł tsóós

This is the “surface template”, not the underlying form — the latter is what is at issue. The direct object (Obj) is represented by Ø in this verb form, in accordance with the general rule that a third person (direct) object is not phonologically overt if the subject is a “local” (first or second) person. Thus, it is merely by hypothesis that Obj is assigned to position IV in the template (and likewise, that it is sister to the verb (V) in the underlying structure below). The sole exponent of M/S is the first person subject marker sh, assimilated to [s] by a regular phonological rule. It seems to me that the preverb in (13) does not have the subject in its scope. It is the coat alone that “comes to be in safety” in the hypothetical world of (12). Thus, the d-structure representation of (13) is most likely that depicted in (14): M/S

(14) M/S sh

vP Pv yisdá

vP v S

V Obj Ø

V tsóós

If this is correct, then we have another deep-surface disparity and we must reconcile the order implicated in (14) with the actual surface linear order shown in the template (13b). The recognition of this deep-surface disparity is forced upon us when we consider cases in which the preverb–verb combination is more “idiomatic”, or more clearly of the type commonly characterized as “lexical”. Consider, for example the case of the verb theme ha#chééh ‘start to cry, break into tears’, with preverb ha ‘up and out’ and verb (zero classifier plus stem) chééh ‘cry’, as in (15), with abbreviated template in which #, following tradition, is used to mark the boundary between the disjunct and conjunct sectors: (15) haashchééh ha#sh-Ø-chééh “I start to cry.”

19

20

Ken Hale

In this case, the preverb and the verb are intimately connected, in a manner familiar from particle-verb combinations in Dutch and German, languages which, like Navajo, permit these elements to be separated in the surface forms of sentences (i.e., in the verb-second forms of Dutch and German). This special relation between the preverb and the verb in (15) suggests an underlying representation in which the preverb is in the lower adjunction position, in conformity with the “lexical” association of the preverb and the verb. If correct, the basic syntactic configuration for (15) would entail a deep-surface disparity: M/S

(16) M/S sh

vP Pv ha

vP v

V

It is conceivable, of course, that the issue is not really settled by such cases as this. The question is whether the preverb does or does not have the subject within its scope. The verb of (15) is an unergative verb built on the nominal root cha ‘cry’ (cf. Hale 2000; Hale and Platero 1996), and the configuration assigned in (16) implies that the semantic component supplied by the preverb applies to cha, and not to the event as a whole — the crying breaks out or bursts forth, so to speak. But this is, to be sure, thin ice to skate on. However, that the preverb can have narrow scope in relation to the surface subject is inescapable and we must accept the reality of Rice’s proposal for some cases at least. Consider for example, the causative of ha#chééh, as in the idiomatic expression of (17), from Young and Morgan 1987: 373:8 (17) a.

habiishchxééh ha#b-y-sh-ł-Ø-chxééh “I honk it (make it cry, of car horn).”


Qual

b. Qual y

M/S M/S sh

vP1 vP2

v1 S Pv ha

vP2 Obj(obl)

v2 v2 Ø

V chxééh

The productive causative represented here involves the embedding of the basic third person form haachééh ‘he/she/it starts to cry’ (the structure dominated by v2 in (17b)) as the complement in the structure defined by the transitivizing light verb ł (v1 in (17b)). The original subject of the embedded verb (located in M/S in the original intransitive configuration (16)) appears now as the surface object, overtly, in an oblique case from (hence the gloss Obj(obl)). This derived oblique object is traditionally held to be in a postpositional construction (with a non-overt postposition), accounting for its overt realization as third person b, rather than zero, as a direct object would be. The hypothetical postpositional constituent is assigned alternatively to positions Ia and IV by Young and Morgan (1987: G42, et passim). In addition, the full causative of Navajo always has associated with it a qualifier (Qual) — in this case, the qualifier is what I take to be the canonical causative qualifier y (giving rise here to the long vowel [ii], through regular phonological processes; cf. Hale 2000 for a discussion of the causative construction within the Case Binding framework of Bittner 1994). Once these details of morphophonology are set aside, it is clear that the preverb (ha ‘up and out’, in this case) must be lower than the surface subject (sh) within the derived causative configuration. In the idiomatic use, for example, it is the car, or the car’s horn, which produces the “crying out” here, not the entity corresponding to the subject of the causative construction as a whole. Such causative configurations clearly show that we must contend with the deep-surface disparity implied by Rice’s proposal. In particular, we must account for the consistent leftmost positioning of preverbs in the surface representations of verbs, sometimes at variance with the underlying position which, in some cases at least, must be lower (more “internal”) in the structure than the leftward linear positioning consistently seen in surface forms. The same argument is applicable to certain other causatives, e.g.,

21

22

Ken Hale

those of (18) — in (18a), note the underlying “double classifier”, as expected in the full causative, which entails embedding the complete (unergative) verbal projection (vP2, headed by v2 l, the inner classifier) as the complement of the causative light verb ł (v1, the outer classifier): (18) a.

habiishyeed (Young and Morgan 1987: 373) ha#b-y-sh-ł-l-gheed “I run him/her/it up out.” (idiomatically) “I start it (car).” b. ch’íbidinishdlóóh (Young and Morgan 1987: 281) ch’í#b-d-nish-ł-l-dlóóh “I make him burst out laughing.” c. (configurational structure of (18a)) Qual Qual y

M/S M/S sh

vP1 vP2

v1 S Pv ha

vP2 Obj(obl)

v2 v2 l

V yeed

The reality of the double classifier is “visible” in (18a,c); the inner classifier “protects” the stem initial fricative (y = fronted gamma) from devoicing, otherwise expected after ł, the outer classifier, ultimately adjacent to the inner classifier; see below and, e.g., Hale 1974, and Kari 1976). Even in the productive causative, however, we must recognize the possibility of the alternative higher positioning of the preverb. Consider (19), where, it seems to me, the subject of the causative is within the scope of the preverb na ‘round and about’: (19) nabiishłá na#b-y-sh-ł-Ø-á “I walk him/her (baby) about.” All instances of this causative that I have actually heard, as well as miming by native speakers illustrating its meaning, clearly have both the agent (grammatical subject)


and the “causee” (derived Obj) within the scope of the preverb. The agent manipulates the causee at every step and thus “moves round and about”. It should be said that it is not necessary to resort to the full causative to argue for the lower adjunction site. In many semantically perspicuous cases, the transitive alternant of a labile verb (participating in the so-called causative-inchoative transitivity alternation), clearly has the subject outside the scope of the preverb. Consider the following pair: (20) a.

Jooł naamaas. jooł na#Ø-Ø-maas “The ball rolls around.” b. Mósí jooł neiłmaas. mósí jooł na#y-Ø-ł-maas “The cat rolls the ball around.”

In (20b) it is not the transitive subject (the cat) that “rolls around” but rather the direct object (the ball). On the assumption that the meaning reflects the structure, the verb of (20b) must correspond to the following underlying configuration, in which two instances of deep-surface disparity occur: M/S

(21) M/S Ø

vP Pv na

vP v S

V Obj y

V maas

This is a third person zero-imperfective verb form, hence the M/S constituent is phonologically null (or, at most, the neutral vowel /i/; see Hargus and Tuttle 1997). The object is the overt obviative third person pronoun (3o), as required when both subject and object are third person and the object is not the topic. These pronouns are linked to the overt noun phrases (mósí ‘cat’ and jooł ‘ball’) adjoined to the full VS in accordance with the principles developed in Willie and Jelinek (2000). Before moving on to further discussion of the disparities just noted, I will briefly discuss certain details of the syntactic structures projected by lexical verbs (V). The innermost projection in (21) — that of V, the lexical category — is abbreviated there in a way which obscures an important distinction among verbs, that between unergative and unaccusative verbs. The verb of (21) is labile. The object (Obj) in (21) originates as the specifier of the lexical projection (hence below

23

24

Ken Hale

the classifier v), as depicted in (22) below, corresponding to the intransitive use of the verb, with Ø classifier (i.e., the zero form of v): v

(22) v

V Pronoun

V R

V

The root component R of this verb has the lexical property that it forces V to project a specifier.9 In this intransitive form, the pronominal argument will function as the subject in sentential syntax. In the transitive use, the classifier is ł, as expected, and the pronominal argument will surface as the sentential syntactic object (Obj). The subject of the transitive is a pronominal argument in the specifier of v (not shown in (22). It is an external argument in relation to the lexical projection itself. In the case of an unergative verb like cha ‘cry’, by contrast, the root component is nominal in character and does not force the V to project a specifier.10 Consequently, the pronominal argument will be external to the lexical projection, appearing rather as the specifier of the classifier (v), as shown in (23): v

(23) Pronoun

v v

V R

V

The contrast between (22) and (23) represents the traditional unaccusative/ unergative opposition as it is defined in the theory of argument structure adopted here (cf. Hale and Platero 1996). In what follows, I will continue to abbreviate the subconfiguration [VR V] as simply V in the structural diagrams for labile verbs and non-alternating verbs alike.11 The contrast between the two types resides in the position of the pronominal argument (Pronoun) in relation to v. In labile verbs, whose structure is shown in (22) above, v locally c-commands (and governs) the pronominal argument, given that the latter is the specifier of the complement of v. In non-alternating (i.e., unergative) verbs, the pronominal argument is the specifier of v itself, hence external to the projection of the lexical verb V, as in (23).


3.

Resolving the deep-surface disparities

In the spirit (though not the letter) of Baker (1996), I assume that in a Pronominal Argument language, the case requirements of direct arguments are satisfied in situ. It is part and parcel of the tightly organized syntactic word-like conjunct sector of the Navajo verb that the arguments are licensed in their thematic positions. Hence, there is no motivation for syntactic movement which would reposition pronominal arguments in such a way as to account for their surface positions.12 And there is no motivation at all for movement of adjuncts (the elements of the disjunct sector) — these must be alternatively base generated in the high or low position in any event. How then is the surface ordering to be achieved in the cases of disparity? I believe that the answer lies in the realm of phonology, in particular, in the processes that “spell out” of the elements of the conjunct sector. I adopt the theories of McDonough (2000b) and Speas (1984) according to which the verb is expressed phonologically in the form of a minimal disyllabic skeleton which must be filled out to define the phonetic realization of a Navajo verb. It is precisely this process of “filling out the skeleton” (especially what I will call the “receptor”, corresponding to the left-hand portion of the skeleton) that accounts for the surface arrangement of the verb, its dependents, and the nuclear elements in its extended projection. There is no head-movement in the traditional sense. The processes involved are properly speaking phonological, or so I would like to claim (cf. Hale 2001, for an earlier version of this proposal). The idea is this: The verb stem brings with it a bipartite phonological skeleton whose right-hand half is filled out (in the normal case fully, but partially in a few exceptional cases of a missing onset) by the stem itself, as indicated in (24) by underlining: (24) CVCCVC The CV portion of each of the two syllables must be filled, or “satisfied”. This is obligatory, with minor and essentially irrelevant exceptions. The coda is filled, or not filled, depending on the nature of the elements actually present in the particular VS. In accordance with the Speas-McDonough bipartite theory of the Navajo verb, the right-hand portion of the skeleton is satisfied by the verb stem (V). By contrast, the phonetic expression of the receptor is not fixed for all forms of a verb. Rather it is variously filled in by the phonological features of the elements (nuclear or nonnuclear) which happen to be present in the extended projection of particular verb at issue. The relevant nuclear elements in the conjunct sector are (a) the qualifier (Qual, one or more of which may, or may not, be present in a given verb), (b) the mode/ subj portmanteau (M/S, obligatory but sometimes null phonologically; but see Hargus and Tuttle 1997), and (c) the voice element(s) traditionally called the

25

26

Ken Hale

classifier (v, obligatory, assuming the Ø classifier is taken to be an actual element). The organizing principle among these elements, and the verb stem as well, is selection: Qual selects M/S, and M/S selects the verb (taken to be v+V here). These elements are therefore in a head-complement structural relation which, in Navajo at least, is head-initial. The non-nuclear elements of the Navajo verb are (a) the arguments of the verb (most relevant for this discussion, its object, if there is one), and (b) the adjuncts. The direct object (Obj) is an argument of the verb (V) and therefore appears internal to the lexical projection of V (see below for further detail). In Navajo, the verb takes its direct arguments (specifier, complement) on the left, in keeping with the general head-final character of lexical projections (see footnote 4 above), an arrangement which extends to postpositional and nominal (possessive) projections as well. The adjuncts comprise the disjunct sector of the Navajo VS. The organizing principle among them is (a) adjunction itself, and (b) precedence constrained by leftward semantic scope (cf., Rice 2000c). In structural diagrams, adjunction is symbolized by the repetition of the symbol corresponding to the maximal projection of the phrase to which the adjunct is attached — e.g., written redundantly as vP in (21), for expository convenience alone. Crucially, the node immediately dominating an adjunct is a segment of the node dominating the maximal projection, it is not the maximal projection itself. In this, adjuncts differ crucially from specifiers and complements. Thus, in (21), the adjunct (the preverb na) is not dominated by the maximal projection vP; instead, it is dominated by a segment of that projection. By contrast, the object (y) is dominated by the maximal projection of the verb (V, or VP, a purely notational distinction). Let us consider now the derivation of the verb of (20b). The relevant morphophonological components are set out in (25): (25) Ø

ł na

CVCmaas y

The left-to-right ordering corresponds directly to the arrangement of elements given in (21), but the parts are arranged on two distinct planes, whose purpose here is purely expository, to distinguish heads from non-heads, i.e., nuclear from non-nuclear. The preverb, being an adjunct, and the object, being a complement, are graphically set apart as non-heads. In assuming that the object inflection y is a true direct argument in this Navajo construction, and thus appears as a constituent within the verbal projection, I am following Jelinek’s Pronominal Argument Theory. Satisfaction of the verbal skeleton involves filling in those parts of it which are not already spelled out. With rare exceptions, the right-hand part of the skeleton is satisfied by the stem alone. The left-hand part, or receptor, is filled in variously, depending upon what is phonologically available in the verb at issue. The process is this. First,


the phonetic features of the functional heads (if any are overtly present) are transferred successive cyclically, moving linearly R to L, to the receptor within the verbal skeleton, filling in the onset, vowel nucleus, and coda, if possible. Once the receptor is fully satisfied in this manner, any morphological material left over in the verbal construction (whether nuclear or non-nuclear) is prefixed to the now replete skeleton, proceeding from right to left, until no material is left. If the receptor is not fully satisfied by nuclear material, non-nuclear material is recruited to fill it out. The bipartite skeleton is, so to speak, “opportunistic”. If the nuclear material (i.e., the head plane) does not fully satisfy the skeleton, non-nuclear material is exploited for that purpose, working from right to left, or upward in the structure. In the case at hand, the classifier (light verb v = ł) assumes the coda position in the receptor — that is to say, it is spelled out as the coda in the receptor of (25), yielding the intermediate form (26), in which represents the syntactic position in which v, the transitive voice element, is interpreted at Logical Form (LF). (26) Ø

CVłmaas na y

The voice light verb v itself is not “moved” in the computational part of the derivation of this verb. It remains in situ. But it is spelled out — i.e., instantiated phonologically — within the bipartite skeleton associated with the verb maas ‘roll (of spheroid entity)’. This is a minor adjustment, in fact, given that v and V, qua overt heads, are structurally adjacent. The vocalic nucleus of the receptor is obligatory, and it is satisfied here by insertion of the Navajo neutral vowel [i], corresponding to schwa in many other Athabaskan languages (and, in Navajo, subject to full assimilation to the quality of an adjacent vowel, or to the rounding of an adjacent consonant). The manner in which this “insertion” takes place is open to debate. In Hale (2001) I assumed that it was uniformly supplied by epenthesis — following Speas (1984) and Wright (1984) — and therefore, that it is not phonologically overt in the underlying representation of the so-called zero-imperfective. The alternative, defended strongly by Hargus and Tuttle (1997), is that the vowel nucleus is an actual morpheme, in the form of the neutral vowel — it is, in effect, the overt exponent of the viewpoint aspect component of M/S (quite reasonably termed “tense” by Hargus and Tuttle). Be this as it may, the nucleus of CV is satisfied by this neutral vowel, representing the zero-imperfective: (27)

Ciłmaas na y

In (27), the nuclear material is exhausted at this point, but the skeleton is still incomplete. At this point, we move to the non-nuclear plane, if present. In this case,

27

28

Ken Hale

working right to left (upward in the structure), the object, y, is realized in the onset position of the receptor: (28)

yiłmaas na

Finally, the preverb is prefixed (or perhaps “procliticized”, a distinction I will not always make in this discussion) to the verb form as it now appears. In this particular case, perhaps, nothing special happens. The final form of the verb is in fact achieved at this point, given that all the overt morphemes are in the correct order. Collapsing the expository planar representation, we have (29a), the surface ordering of overt elements, corresponding to the derived morphosyntactic object represented diagrammatically in (29b): (29) a. b.

na

yiłmaas

M/S M/S

v v

VP Pv na

VP Obj

V yiPmaas

The phonological derivation must terminate in a “VS”, however, since it provides the environment for the word-internal phonological processes which give the actual pronunciation: (30) [neiłmaas] ‘3rd (subject) rolls 3rd (object) about’ I must, to some extent, remain silent about the process, or processes, which reduce a morphosyntactic object like (29b), specifically those parts of it which are visible in phonology (thus excluding s), to a prosodic “word”. I will simply assume that such processes exist, above and beyond the processes informally posited here for satisfying the bipartite skeleton itself. I will give a few other examples of verbal derivations within the framework being explored here, beginning with the derivation of (31a), whose underlying structure is diagrammed in (31b): (31) a.

Yisdánídashidoołtééł. (~ Yisdáádashidoołtééł.) “They will get me back to safety.”


Qual

b.

Qual

Pv yisdá

Qual

Rev ní

Qual

Pl da

M/S

Qual d

v

M/S w

V

v S Obj sh

V tééS

This is essentially the same as (9–11) above, omitting the deictic subject (3a, or “4th person”). The informal expository planar representation is as follows (NCL = nuclear, NNCL = non-nuclear): (32) (NCL): (NNCL):

Qual M/S v d w ł Pv Rev Pl yisdá ní da

V CVCtééł Obj sh

This is a structure in which the adjuncts are attached (adjoined) at the higher of the two alternative positions to which they may be assigned, i.e., they are adjoined to the maximal projection determined by the highest nuclear element (Qual, in this instance). We proceed as before, filling the bipartite skeleton first with material from the nuclear array. The classifier (v = ł) assumes the coda position in the receptor, the vocalic nucleus of the receptor is supplied by the neutral vowel (/i/, subject to assimilation), and the onset is supplied by the gamma-Perfective situation aspect prefix (in portmanteau with 3rd person, appearing as w, rounded gamma, in this future verb form; see Hale 2001, for discussion of this matter). This leaves the receptor portion of the skeleton fully satisfied. The remaining nuclear element, the qualifier d, is prefixed to the skeleton and supported there by epenthesis of the neutral vowel /i/, this being no more than a conventional symbol corresponding to an appropriately underspecified nonlow vowel nucleus, realized in Navajo as [i] in the default (unassimilated) situation. At this point we have the hypothetical verb form: (33)

diwiłtééł yisdá ní

da

sh

29

30

Ken Hale

Subsequent rules of phonology effect (a) assimilation of the neutral vowel to the rounding of the adjacent third person situation aspect portmanteau prefix w, and (b) deletion of the intervocalic glide, giving doołtééł ‘3 subject will handle animate object’. The nuclear array is now exhausted. At this point, the non-nuclear elements are prefixed (or procliticized) to the verb as it now stands, working leftward (upwards in the structure), applying appropriate principles of allomorphy and rules of phonology, as required, resulting in the final form (31a). Again, while I am making an actual (though still informal) proposal about the bipartite skeleton, and therefore about a major portion of the conjunct sector, I remain essentially silent on how (33) above is converted into a prosodic word, masking my ignorance in this matter by means of the expressions “prefix” and “procliticize” which, I assume, correspond to processes involved in word formation.13 In any case, the morphologically overt elements in the VS are in the correct linear arrangement in (33). I will conclude these illustrations with the causative verb of (18), repeated here as (34), with the underlying structure in the planar display in (35) and structural diagram in (36) (repeated from (18c)):14 (34) habiishyeed ha#b-y-sh-ł-l-gheed “I run him/her/it up out.” (idiomatically) ‘I start it (car).’ (35) (NCL): (NNCL):

Qual M/S v y sh ł Pv Obj(obl) ha b

v l

V CVCyeed

Qual

(36) Qual y

M/S M/S sh

vP1 v1 S

vP2 Pv ha

vP2 Obj(obl) b

v2 v2 l

V yeed


The true causative of Navajo involves, among other things, the embedding of the maximal projection of a v-phrase (the phrase headed by the voice light verb v, the traditional classifier) as the complement of another v, the transitive light verb ł). This arrangement is symbolized in (36) as the maximal projection of v2 embedded under v1. The basic or underlying position of the subject of a v-projection is specifier in that projection. Since the causative embeds only the projection of v, not its extended projection, the subject of v will not enter into the usual portmanteau relation with Mode (see above where the embedded subject appears autonomously in Specv2. For reasons having to do with aspects of sentential syntax, this subject (i.e., the causee) will appear in an oblique objective form, as it does in (36), where it appears as b, the third person Obj(obl) form, in the final spellout of the sentence (see below for further remarks on Case). Proceeding now with the derivation, the nuclear elements are inserted into the skeleton in an appropriate manner. The vocalic nucleus, as usual, is supplied by the neutral vowel /i/. In this verb, the first three overt heads (l, ł, and sh) all compete for the coda position. Together with the initial C of the stem, this would result in a consonantal sequence exceeding the maximum allowed in Navajo (two consonants: a coda followed by an onset). This is the simple sequence [sh+y], (see references cited in connection with (18) above, as well as in Speas (1984, 1990), McDonough (1996)). At this point, there is just the causative qualifier y in the nuclear plane; hypothetically this assumes the onset position, giving now the intermediate sequence *yishyeed, which will not survive as such, because the causative qualifier is always preceded by overt material triggering rules of phonology which modify the output. In this case, prefixation of Obj(obl) gives hypothetical *biyishyeed which, by intervocalic glide deletion, surfaces as actual biishyeed. To this is prefixed the preverb ha ‘up out’ giving the surface verb form habiishyeed. Notice that the surface position of adjuncts does not vary according to whether they are attached high (above the subject) or low (below the subject). In either case, adjuncts precede the overt nuclear elements in the verbal projection, since these are spelled out in the verbal skeleton in accordance with the proposal being entertained here. Why does the causee argument (embedded subject) appear in the oblique form? Clearly, it cannot surface in the nominative, the case normally assigned to subjects, because the extended projection of v2 is absent. It is, so to speak, trapped within the v-projection and cannot fuse with Mode, as it ordinarily would. That explains the impossibility of the nominative. But why must it be oblique, rather than the normal objective case assigned to direct objects? In the Case Binding theory of Bittner (1994), the causee must necessarily be in a marked structural case, such as the oblique, if it appears in a verbal projection which contains another nominal argument, a so-called “competitor”. In the theory of argument structure developed by Hale and Keyser (1993; and see also Hale and Platero 1996 for relevant discussion of Navajo), unergative verbs are assumed always to contain an inherent object.

31

32

Ken Hale

In the case of verbs like those appearing in (17), (18b), and (19), for example, we must hypothesize that the competitor is a nominal element functioning as the root component of the verb stem — cf. cha ‘crying, weeping’, dlo ‘laughter’, gáál ‘locomotion’, assuming that these items do indeed have nominal bases. Generally, however, this proposal is not merely hypothetical, since most unergatives have an explicit object argument (the indefinite and unspecified object’, glossed 3i in Young and Morgan 1987: 67 et passim), as in (37), for example: (37) a.

’ałhosh (Young and Morgan 1987: 126) ’-Ø-ł-ghosh Obj(3i)-M/S-v-V “He/she/it sleeps/is asleep.” b. bi’iishháásh (Young and Morgan 1987: 215) b-’-y-sh-Ø-ł-ł-gháásh Obj(obl)-Obj(3i)-M/S-v-v-V “I put him/her/it to sleep.”

The underlying configuration of the causative is diagrammed in (38): Qual

(38) Qual y

M/S M/S sh

vP1 v1 S

vP2 Obj(obl) b

v2 V

v2 S Obj ’

V gáásh

The direct object functions as a competitor accounting for the oblique case assigned to the causee (see Bittner 1994, and Bittner and Hale 1996, for detailed discussion). The derivation resulting in the fully spelled out surface form proceeds as in the structurally parallel (36).


4. Noun incorporation Navajo permits incorporation of bare nominals into postpositions. The process is quite restricted, but it is nonetheless quite transparent. Consider the following pair of sentences: (39) a.

łééchøø’í tó y-iih yilwod. dog water 3o-into run-perf “The dog ran into the water.” b. łééchøø’í taa-h yilwod. dog water-into run-perf “The dog ran into water.”

In the first of these, the endpoint reached by the dog is expressed by means of a normal postpositional phrase in which the object is a pronoun (the obviative, 3.o, as expected in a clause in which two third person arguments appear). In accordance with the PA Hypothesis, the pronoun is the true object of the postposition, the overt noun phrase tó ‘water’ being an adjunct (to the PP as a whole, or to the clause as a whole, depending on principles of discourse configurationality of the type developed in Willie and Jelinek 2000). By contrast, in (39b), the object of the postposition is a nominal, in complementary distribution with the pronoun. This is a possibility for a limited range of Navajo postpositions and nominals — a nominal, like a pronoun, may appear in object position. Nothing in principle prevents this; it is not in contradiction to the PA Hypothesis. Rather, it is a matter of parameters — some languages permit nominals to appear as the complement of a lexical head, and some of the languages permitting this are ones which are predominantly PA Languages. The phenomenon belongs to the category of Noun Incorporation (NI), as studied in depth by Baker (1988). The question I would like to address here is this. Is NI in fact a special syntactic process, generated by Head Movement, as opposed to a matter belonging to the morphophonology responsible for converting a concatenation of terminal nodes into a prosodic word? The example just given sheds no light on this, because the apparent incorporation there is nothing other than the product of Merge. The “incorporated” object is simply the complement of the postposition, appearing where expected as a result of Merge. I would like to explore the possibility that this may be true of NI generally; in effect, the possibility that there is no syntactic process of NI. In Navajo, as noted, apparent NI is limited to the relation between a noun and a postposition. Verbs do not participate in the phenomenon, primarily because of Willie’s Generalization, according to which nominal expressions linked to the arguments of a verb are fundamentally definite in Navajo (unless specifically marked as indefinite or nonspecific). This circumstance conflicts with the generally

33

34

Ken Hale

indefinite interpretation assigned to an incorporated nominal (cf. Hale, Jelinek and Willie, to appear). Northern Athabaskan languages do have apparent NI within the verbal projection. However, it is quite obvious that this is merely apparent and that it is not due to a special syntactic process. It is rather simply the result of Merge, since the position of the incorporated nominal is the same as that of the direct object in the underlying representation of verbal projections. If this is so, then it is expected that the surface position of the incorporate will reveal the same deep-surface disparity that the object does. Consider the following example from Slave (Rice 2000: 110; original glossing and hyphenation modified in accordance with the practices used in this paper): (40) ní-yati-dé-ni˛-Ø-’o˛ Pv-word-Qual-M/S-v-V “She blamed.” (lit. “She placed words.”) The incorporate yati ‘word(s)’, like the direct object pronouns in various Navajo examples cited above, is separated from the verb stem by M/S and Qual, as expected if the incorporate is simply the object of the verb and, accordingly, combined with it at Merge. Thus, by hypothesis, the configuration underlying (40) is approximately as follows, with the incorporate the complement of (hence sister to) V: Qual

(41) Pv ní

Qual Qual dé

M/S

M/S ní (aspect, 3rd person) v Ø

v V N yatí

V ’o

Assuming, as I do, that the Slave surface form is derived in a manner closely similar to that proposed for Navajo, the overt nuclear elements (Qual and M/S) are spelled out in an extended bipartite skeleton associated with the verb, and the incorporate will appear linearly to the left of Qual, in apparent contradiction to its actual position in the structure. The deep-surface disparity here is precisely that seen in relation to the surface position of pronominal objects in Navajo (cf. Rice, this volume). And, as expected, Slave shows the same disparity with pronominal objects, in (42), with a pronominal instead of nominal direct object (original glossing modified):


(42) rá-se-re-yi˛-h-t’u Pv-1s-Qual-M/S-v-V “(S)he punched me.” Here, the object pronoun se ‘1st singular object’ appears in the same relative order position as does the incorporate yati ‘word(s)’ in (40). The Athabaskan languages are especially revealing in this regard, because of the peculiarity that the nuclear functional heads are spelled out in a syllabic skeleton associated with the verb and thus appear to intervene between the object and the verb. This is an appearance, nothing more, and there is no evidence whatsoever that the object has moved, in syntax, to its surface leftward position. Nor, apparently, is there evidence that the position of the incorporate is anything different from that of the object — i.e., the position defined by Merge in the definition of the basic syntactic structure. The idea that Athabaskan incorporates are in their basic (Merged) positions, and not “incorporated” by means of a syntactic operation (e.g., Head Movement), is implicit in Rice (2000c: 68–73 et passim), and the idea is explicated in some detail in Tuttle (1996: 113–117) for Salcha Athabaskan.15 For the Northern Athabaskan languages, Rice has proposed that full DP arguments are moved out of VP, out from their “thematic” positions defined by Merge (Rice 1993). By contrast, bare N arguments (so-called incorporates) are in situ, in their base positions within VP. In the Southern Athabaskan languages of the PA type, full DP arguments construed with the verb are adjuncts to the clause, linked to pronouns in the thematic positions defined at Merge. By hypothesis, the pronouns internal to the VS are the only true arguments of the verb. Navajo does permit bare nominal arguments of postpositions to appear in their base positions (i.e., as so-called incorporates), as in (39b) and other examples of its kind. My remarks on incorporation are to be taken as tentative and to apply to nominal direct arguments only, leaving aside for purposes of this discussion other elements which have been described as able to undergo incorporation or Head Movement (e.g., P, V, adverbials). It is because of a peculiarity of the Athabaskan languages that it becomes clear that the position of an object incorporate is possibly none other than the base position of objects in general — i.e., the two share the property that they are separated from the verb in the surface representation. In much of the literature on incorporation, the languages involved belong to a much more common type, in which incorporates are adjacent to the verb stem, while inflectional heads are more peripheral in the structure. Thus, the following Mohawk and Hopi examples represent the more common situation: (43) a.

Wa’-k-hnínu-’ ne ka-nákt-a’. fact-1sg.S-buy-punc NE N.sg.S-bed-nsf “I bought the/a bed.”

35

36

Ken Hale

b. Wa’-ke-nakt-a-hnínu-’. fact-1sg.S-bed-Ø-buy-punc “I bought the/a bed.”

(Baker 1996: 279)

(44) a.

Nu’ i-t taavo-t niina. I this-acc cottontail-acc kill.sg.perf “I killed this cottontail rabbit.” b. Um qa hìita tap-nina? (HDP: 577) you neg what.acc cottontail-kill.sg.perf “Didn’t you kill any cottontail rabbit?”

The (a)-examples represent the case in which the object is not incorporated (i.e., where it is instead an adjunct to the clause, in the polysynthetic Mohawk, or is a fully case-marked object in the configurational language Hopi). In the (b)-variants, on the other hand, the lexical head of the nominal argument, a bare nominal (incorporate) appears adjacent to the verb within the VS (adjacent except for the Mohawk incremental element -a-, irrelevant to the point at issue here). Any inflectional morphemes appear to the left or to the right of the N+V sequence (e.g., Mohawk factual/aorist wa’- and the first person singular subject ke- are prefixes to N+V, while the punctual element -’ is a suffix). Mohawk and Hopi represent the more usual situation, because in these languages, and most languages which allow so-called NI, the verb and incorporated object are adjacent to one another. This contrasts strongly with the Athabaskan cases above, where functional heads are spelled out on the verb in such a way as to intervene, superficially, between the object and the verb — presenting a deepsurface disparity amounting to a surface violation of the Mirror Principle (Baker 1985). The question is whether N-V adjacency of the Mohawk and Hopi type is brought about by syntactic movement or by Merge alone. I propose the latter. As we have seen in the Navajo examples cited, the surface object of a verb may correspond to the true sister of V or to an argument corresponding to an internal specifier. In Hopi verbs of manufacture, for example, a nominal object incorporate appears as the complement of the suffixal verb -ta, as in (45a), diagrammed in (45b): (45) a.

As nu’-nen yanwat kii-ta-ni. (HDP: 139) prt 1sg-if this.way house-ta-fut “If it were me, I would build the house this way.” V

b. N kii

V ta

By hypothesis, the nominal component is the direct object of the verb, being its complement. Because the nominal is bare, it remains internal to the verbal projection.


If the object were a full nominal expression, i.e., a DP, it could not appear as a bare N in this manner, for reasons having to do with case — a determiner associated with an argument is necessarily assigned case, as illustrated in (44) above, where either the entire DP is case-marked, (44a), or the determiner itself is case-marked (hìita ‘what.acc’ in (44b)) while the head N appears as an incorporate. I leave open for present purposes the issue of whether case-marking is in situ, as in Bittner’s Case Binding Theory (Bittner 1994, and Bittner and Hale 1996) or requires Object Shift, i.e., raising of DP, or a part thereof, as in theories of case checking; cf. Chomsky 2001b). In the following example, the surface object is a specifier, projected by V (the ingressive, or inceptive, -va) in response to the needs of the predicative root tuy- (< tuutuya ‘hurt, ache’): (46) a.

Pas Polìi-t nu-y qötö-tuy-va-na-ya. (HDP: 702) very Butterfly.Dancer-pl 1sg-acc head-ache-ingr-NA-pl “The Butterfly Dancers (really) gave me a headache.” V1

b. V2

V1 na

DP D 1s

V2 N qötó-

V3 tuy-

V2 -va

The surface object here is a part-whole expression in which the component corresponding to the whole (i.e., possessor) is marked accusative (nu-y’ 1sg-acc), by virtue of its position in relation to the suffixal transitivizing (causative) verb -na. An alternative would be to mark the entire nominal expression for accusative case — in that variant, the first person singular possessor would appear as the 1sg possessive prefix ’i-, giving the fully inflected form ’i-qötö-y (1sg-head-acc) ‘my head’. In the actual form taken from HDP: 702, however, the lexical nucleus, qötö ‘head’ appears as a bare nominal incorporate. I see no reason to suppose that a syntactic rule of incorporation is responsible for (46). The incorporate is situated in the position to which it would be assigned by Merge — of DP and V2, the composite verb tuy-va (ache-ingressive/inceptive). Any special morphological processes involved here have to do with PF (i.e., spellout) not with the syntax. Now consider the Mohawk verb of (47): (47) a.

Wa’-ka-wír-v’-ne’. (Baker 1996: 293) fact-N.sg.S-baby-fall-punc “The baby fell.”

37

38

Ken Hale

b. T-a’-ka-wís-v’-ne’. (Baker 1996: 213) cis-fact-glass-fall-punc “The glass fell.” By hypothesis, the subject of an unaccusative originates in the specifier position of the verbal projection. In an incorporating polysynthetic language, it is possible for a nominal subject to appear internal to the VS, as it does in the examples of (47). Applied to the PA Hypothesis, the logic is that a direct argument, whether pronominal (the usual case) or a bare nominal, is licensed in situ, hence no raising (for case reasons or to satisfy the EPP) is required, or even possible. In a fully configurational language, like Hopi and most of its Northern UtoAztecan relatives, the subject of an unaccusative must situate itself in a manner which will permit it, or some portion of it, to be case marked (whether this is for reasons of case alone or to satisfy the EPP, or both). Thus an exact replica of the Mohawk situation represented by (47) is impossible in a fully configurational language. Consider in this light the following Hopi examples: (48) a.

… kuktönsi-’at tuy-va. (HDP: 702) … heel-3sg ache-ingressive “… his heel started hurting.” b. nu’ a’ni pono-tuy-va. (HDP: 702) I very stomach-ache-ingressive “I got a painful stomach ache.”

In (48a) the subject (the 3sg.poss form of the noun kuktönsi ‘heel’) is in the nominative case, as expected. No part of the subject nominal is “incorporated”. In (48b), by contrast, the possessor (nu’ ‘I’) appears in the nominative, while the lexical head (pono ‘stomach’) appears as an incorporate. An alternative in which the lexical head appeared incorporated, without a case marked associate, is impossible in Hopi, so far as I know. If the case marked associate is satisfying the EPP, and if the EPP requires raising, then nu’ ‘I’ is raised in (48b). I let the facts stand without further comment. The essential point I wish to make, however, is that, again, I see no reason to assume that the appearance of incorporation here is due to a syntactic rule of NI. The adjacency of pono ‘stomach’ and tuy-va (ache-ingressive) is, I contend, due to Merge alone; Merge of the DP in specifier position of V1, as shown in (49): V1

(49) DP D 1s

V1 N pono

V2 tuy-

V1 -va


I would like now to return very briefly to Navajo and to remark on how the construction termed “inchoative” (Young and Morgan 1987: 187–188) fits into the picture. Consider sentence (4), repeated here as (50): (50) ’Awéé’ bi-’nii-sh-hóósh. baby Obj(obl)-INCH-M/S(sh)-v(ł)-V(ghóósh) baby 3-inch-1sg-tickle “I start to tickle the baby.” The mark of the inchoative is the complex prefix -’nii- consisting of the 3.i indefinite object prefix ’- followed by the qualifiers n- and -ii-. Since the object position is filled, the object of the basic verb (cf. yishhozh ‘I tickle him/her’) must be assigned oblique case (this being traditionally regarded as a null postposition, as mentioned earlier). The overt bi- in (50), in the presence of a local person subject (first singular, in this case), is explained in this way (a third person direct argument would otherwise be itself null in the presence of a local subject). In the case of an unergative verb, the inherent indefinite object would appear as such, overtly, since the 3i object is ’- in both direct and oblique object forms — as in ’i’niishhá˛a´˛ h ‘I start to snore’, in which both 3i object forms appear. Of special interest here is the behavior of unaccusatives. Since there is no inherent object, the sole argument being the inner specifier (ordinarily the subject, as in (51a) below), the question is what happens to this sole argument. Does it appear as a subject or as an oblique object? The answer to this question is evident in (51b): (51) a.

yishdlóóh (Young and Morgan 1987: 781) M/S(sh)-v(d)-V(dlóóh) “I get very cold.” b. shi’niidlóóh (cf. Young and Morgan 1987: 217) Obj(obl)-Obj(’)-M/S(?)-v(d)-V(dlóóh) “I start to get very cold.”

The single argument of the unaccusative shows up as an oblique object, preceding the 3i object brought in by the inchoative morphology. The question now is, what does the question mark in M/S stand for? Is that the trace of movement, is this “eccentric agreement” (as suggested in Hale, Munro, and Platero 2000), or is it an expletive subject? By hypothesis, it cannot be a trace, since the trace would c-command its antecedent (subject being higher than the object in underlying structure). And it cannot be eccentric agreement, because there is no “agreement”, in any usual sense, in a pronominal argument language. But, if we are correct that the future and progressive forms reveal an overt exponent of third person in M/S (i.e., the rounding feature, giving rounded intermediate [w] from the gamma glide of those forms, accounting ultimately for the long rounded vowel [oo] in them), then there must be a third person element, presumably expletive, accounting for the rounded

39

40

Ken Hale

long vowel [óo] (tone irrelevant) in the future forms of such inchoatives — cf. shidí’nóodlóół ‘I will start to get very cold.’

5.

Concluding remarks

It has taken me sixteen years, almost, to fully appreciate the import of the insights Eloise Jelinek brought out in her 1984 paper. For me personally, the significance of her PA Hypothesis has been an entirely new way of looking at complex VSs of the sort found in polysynthetic languages of the Navajo type. As late as Hale 2000b, I steadfastly clung to the idea that such words were the result of Head Movement. However, once I took seriously the idea that the pronominal elements in the Navajo VS were in fact the arguments of the verb, with all the implications of this idea, it became clear to me that another view of the Navajo VS was necessary. This, in turn. cast doubt upon certain uses of Head Movement in fully configurational languages as well, particularly NI, permitting a view of the matter according to which certain adjacencies are due to Merge alone. This forces a distinction between syntactic movement and spellout, whose mechanisms — only partially understood at this point — are responsible for the “transformation” of a syntactic structure into a “prosodic word”.16

Notes 1. The morphological analysis of the verb is simplified here in that it omits reference to the imperfective aspect and the phonologically deleted “classifier” (i.e., voice marker) ordinarily appearing as -ł- in this verb. Similar simplifications will be made in other examples, except where details are especially germane. 2. See for example, Chomsky (1995b) and Kratzer (1996). In Hale and Platero (1996) the Sclassifier was taken to be a phonologically defective “light verb” V, undistinguished notationally from the lexical category verb (V). The element in question is in all probability to be classed with the functional categories rather than with the full lexical category V. 3. There also exist pronominals that receive an indefinite interpretation: the specifically indefinite third person (3i) and the so-called fourth person (Young and Morgan’s 3a, or alternative third person) which is a specifically human pronominal often, but not always, indefinite. 4. I am assuming that the structure projected by the lexical category V is right-headed, while the structures projected by the higher nuclear elements (v, M, etc) are left-headed. The significance of this assumption will become clear presently. 5. The relative order positions I-III belong to the so-called “disjunct” (essentially pro-clitic) sector of the verb, while the phonologically more tightly organized prefix positions IV-IX are said to belong to the “conjunct sector”. It will become clear presently that this division is a significant one in the organization of the Athabaskan verb, as has been recognized at least since the work of Fangkuei Li (1933: 459–60), who used the term “primary prefixes” for the conjunct sector and


“secondary prefixes” for the disjunct sector, noting that the former were subject to much greater “weakening” and “coalescence” than the latter. With a few exceptions, a (monomorphemic) conjunct prefix consists of a coronal consonant ([d, j, s, z, sh, zh, n, ł, l]), an oral glide ([y, w (rounded velar), gh = γ (unrounded velar)], or laryngeal ([’, h]). These function as onset or coda, depending in part on the nature of the segment and in part on context. In the majority of cases, where they are realized as onsets, conjunct prefixes are supported by the Navajo default vowel [i], regarded as either epenthetic or basic, depending on the particular analysis — we assume the former for present purposes. Some conjunct prefixes are composite, assuming the shape CVC, in which V is generally the default [i]; a few have the surface form C-´ (consonant and “floating high tone” docked on an associated neutral vowel); a few appear to contain the long vowel [ii] as their vocalic nucleus (e.g., the semelfactive qualifier and the first person subject), and the optative appears to be realized as the rounded vowel, written [o] (long or short, high or low tone, depending on various factors (cf. Kari 1976)). The neutral vowel is very often assimilated to the features of an adjacent segment, giving the superficial impression that the conjunct sector permits all of the vowel qualities in the Navajo inventory. It is quite generally assumed, however, that, underlyingly, the conjunct sector permits only the default neutral vowel (see Hargus and Tuttle 1997 for discussion and references). The quality of this default vowel in Navajo is [i] when unassimilated. The disjunct sector, by contrast, allows all vowel qualities and all consonant types. Where [’] would be expected to occupy a coda position at the end of a derivation, it is normally absorbed into the adjacent (following) C where it appears as “glottalization”. The notorious d-Classifier never surfaces as a coda, but rather (i) as the so-called d-Effect, or (ii) as an onset to one of the rare V-initial verb stems. 6. The full range of known elements appearing in the Navajo verb is catalogued in Young and Morgan (1987). 7. I will occasionally refer to the verbal extended projection embodied in (10) as the “verb word”, though Sharon Hargus and Siri Tuttle (p.c.) correctly point out that a more appropriate term might be “verbal phonological phrase” or “verbal sentence” (Tuttle’s “greater verb”, 1996) because many elements in position I (Preverb) are in fact separate words phonologically. The term “verb word” must, therefore, be understood in this light. The term “verb” itself will sometimes refer to the entire “verb word”, sometimes to the stem alone (V in tree diagrams), and sometimes to the projection including the voice element together with its verbal complement (v+V, v+VP). 8. The idiom shows “velarization of the onset” of the verb stem (giving -chxééh, beside unmarked -chééh). In this instance, velarization reflects the fact that the stem is being used in an extended sense, rather than in the “ordinary” meaning. This is one of several uses of velarization of the onset in Navajo. 9. The root element in (21) is listed as mááz in Young, Morgan, and Midgette 1992: 396, and it is glossed there as “move by rolling” (of a globular or cylindrical entity). In general, it is assumed in the framework adopted here (cf. Hale and Keyser 1998; Platero and Hale 1996; Hale 2000b) that the root component of labile (alternating) verbs has the lexical property that it forces the verbal head (V) to project a specifier, this being the fundamental characteristic of root elements which are inherently attributive and must appear in a lexical syntactic projection in which it can be attributed (or “predicated”) of a local argument — this latter is supplied by the specifier of V, as in the configuration embodied in (22). 10. In Young, Morgan, and Midgette 1992: 70, cha is entered as a “verb root used nominally”. I assume rather that it is fundamentally nominal, an entity expression, which by virtue of this very characteristic does not force the verbal component of V to project a specifier (cf. Hale and Platero 1996). Its subject is external to the V-projection, appearing instead as the specifier of the voice light verb v. On this view of the matter, the verbal character of cha is secondary, deriving from the verbal component of V when it appears in the construction [VR V], as in (23).

41

42

Ken Hale

11. The verbal component (V) of the {VR V} composite is often phonologically null. It is deleted after “strong roots” (those ending in an oral C). It is retained after “weak roots” (those ending in a vowel or in a laryngeal C, the latter deleting instead). Thus, for example, the V component of the stem in (20) is itself is phonologically null, although the shape of the root reflects temporal, aspectual and mood (TAM) features associated with V, the continuous imperfective, in this case, with long low toned vowel nucleus and devoiced final fricative. In the future and progressive, V is said to be uniformly a suffix of the form -ł. This normally deletes after strong roots, but it has the effect of devoicing an underlying root-final voiced fricative. After weak roots it is retained as -ł. In this regard, d-final roots are “weak”, thus the future and progressive stem of (18) is woł (with deletion of /d/, cf. imperfective gheed [yeed] and perfective ghod). The laryngeal-final root in leeh ‘become’ combines with the future and progressive verbal suffix -ł to give leeł, as expected. This Navajo V is more perspicuous than others — typically, the presence of a particular V is evident not by virtue of the presence of an actual segment but by its effect (or lack thereof) on the final consonant, vowel quality/quantity and tonal prosody of R, these effects being associated with and dependent upon the TAM features of V. See Stanley 1969 and Hardy 1979 for detailed studies of the Navajo verb stem; see Young and Morgan 1987: 167–199 for detailed discussion of the aspectual categories and associated interdependencies within the Navajo verb; and for Athabaskan generally, see Rice 2000c: 283–303 for ample discussion of suffixes, which she regards as derivational, rather than inflectional, in close agreement with our claim that they are the lexical V-heads of the verbal stem (R+V), with R the complement of V. 12. I am sidelining here the question of whether movement might actually be involved in the syntactic derivation of the intransitive alternants of labile verbs (i.e., in the derivation of so-called “inchoative” or “unaccusative” verbs in sentential syntax). The question is this: must the pronominal argument (Pronoun) in the intransitive (22) raise to the specifier position projected by the locally c-commanding v? I do not know the answer to this question, though raising is generally assumed for so-called “configurational” languages. In any event, the pronominal subject of the intransitive must be close enough to Mode to enter into the portmanteau relation with it (giving the composite M/S); if the pronominal subject were raised to SpecvP, the required portmanteau formation would be a perfectly local operation. All of this depends upon how the details of the relevant morphophonological processes work out in this framework, an article of faith on my part at this point, unfortunately. Portmanteau formation seems to be obligatory (under the relevant condition, i.e., that Mode locally c-command and govern the pronoun), since overt morphological elements corresponding to the subject are “carried into” the bipartite skeleton in conjunction with the nuclear element Mode On the other hand, the structure of the causative shows that the external argument (subject, surface causee object) in the embedded clause is a specifier, and not inherently a part of M/S. 13. A possible solution to the problem of “wording up” is hinted at in Hale (2001). The idea is to regard the bipartite skeleton as a minimum that must be satisfied. But it is a minimum which can be augmented with additional CV(C) syllables added to the left, to be filled in by left over material (nuclear or non-nuclear, working right to left. Minor adjustments would have to be made to satisfy special phonological principles of Navajo, above and beyond simple epenthesis of the neutral vowel /i/. The appearance of a coronal consonant often triggers special treatment of certain immediately preceding prefixes (e.g., the optional but preferred contraction noted in (31a), and the fricative alternant zh- of the deictic subject prefix j-, functioning as the coda in (9–10)). The main problem would be to determine when to stop building onto the skeleton, in order to permit some preverbs, for example, to appear as separate words (e.g., the preverb ka ‘sick’ in ka naashá ‘I am sick’ (lit. ‘I walk about sick’). In this construction, and in many comparable cases, the preverb can be separated form the rest of the verb by the polar interrogative enclitic -ísh, as in ka-


ásh naniná ‘are you sick?’ (Paul Platero p.c.). This solution is appealing as well inasmuch as it permits statement the proper allomorphy of elements, such as the special rounded alternant w in the third person of gamma-Perfective and Progressive M/S; the rounded alternant appears in the very specific context CiGi, G the velar glide of the gamma-Perfective/Progressive (rounded to [w] in the third person) and /i/ the neutral vowel, rounded to [o] by assimilation to the rounded glide, which drops intervocalically (cf. Hale 2001)). 14. Jonathan Bobaljik (p.c.) notes correctly that the preverb and oblique object appear in the wrong order here. Since the preverb modifies the event, the oblique object (i.e., “causee”, or underlying subject) should not be in its scope. This is correct. However, according to Young and Morgan (1987: 187), the surface order observed in this form, and others like it, is secondary. Ordinarily, an oblique object appears in the leftmost preverb position (their Position Ia), except when a preverb of the class to which ha-, na-, etc. belong (their Position Ib) also appears, in which case “the environment thus produced causes the pronoun prefix to shift to Position IV” (the standard direct object position). In effect, whenever an oblique object competes for leftmost position with a preverb, it is the preverb that actually appears leftmost. In the absence of competition, the oblique appears in the leftmost (highest) prefixal position — thus preceding the Iterative, for example — as in biná’iishdlí˛˛íh ‘I repeatedly feed him liquid, cause him to drink’ (Young and Morgan 1987D: 215). This principle accounts for the observed surface ordering of the relevant elements in this case, but the principle itself is in need of scrutiny and further research, to say the very least. Bobaljik suggests that the notion “plane” might be exploited here, further subdividing the non-nuclear plane to distinguish adjuncts from arguments. The latter might “spell out” before the former, just as nuclear elements spell out before non-nuclear elements. Heretofore, however, I have regarded the planes as expositional devices only. If this is correct, then direct appeal to the distinction between nuclear, argumental, and adjunct elements would be the appropriate factor. The fact remains, nonetheless, that oblique objects do assume the highest prefixal position when not in competition with a preverb. This itself is not a matter entirely devoid of mystery, however understandable it might be in this particular case. 15. It should be pointed out that incorporates can, in some Athabaskan languages, precede or intervene between the quantifier elements (i.e., the adjunct elements iterative, distributive plural), casting some doubt upon the claim that they are in the same position as pronominal objects (see, e.g., Rice 2000c: 80–81). This requires more study, but it indicates, perhaps, that these particular adjunct elements, as opposed to preverbs (which always precede incorporates), can be adjoined at VP as well as the higher positions (maximal projection of v or Qual) proposed for Navajo. The problem of the predicted strict identity of the position of bare N incorporates and pronominal objects remains, however. But since quantifiers (qua VP adjuncts) and nominal incorporates (qua complements) are not distinctively distant from V (see Hale 2001), it is possible that the two categories might vary in their mutual scopal relations, linearizing accordingly (cf. Rice ibid. et passim). Pronouns, by nature, might not vary in this manner. 16. I cannot leave this topic without one further remark. In earlier work (e.g., Hale 2000a) I suggested that each “phonologically dependent” overt item (prefix or suffix; prepound, postpound) carried with it a phonological skeleton which is “completed” by insertion of the phonological matrix of its complement. Thus, for example, the verbal suffix -en, in English redden, has the form “[ ]en”, in which the brackets stand for the “empty” portion of the skeleton. Insertion of the phonological matrix [red] into this skeleton accounted for the word [redden]. In the present context, this insertion process corresponds to the processes termed “spellout” in this paper (and elsewhere in relevant general linguistic literature).

43

Categories and pronominal arguments Emmon Bach University of Massachusetts, Amherst and SOAS, University of London

Two notes for Eloise Jelinek – –

Do all languages have the same syntactic categories? What is the semantics of the “predicational” element in pronominal- argument languages?

These questions have been brought into prominence by Eloise Jelinek (Jelinek 1984, 1994; Jelinek and Demers 1994). These two notes are about issues that come up when you think about those two questions.

1.

Do all languages have the same syntactic categories?

I will make two assumptions here: i.

Two items belong to the same category if and only if they can stand in all the same environments, saving grammaticality. ii. There is a fixed mapping from syntactic categories to semantic types (kinds of denotations), as in a categorial grammar or Montague Grammar. In fact, I will use a categorial grammar as a framework for discussion here. The pronominal argument hypothesis (Jelinek 1984) says that sentences from (say) English and sentences from (say) Straits Salish that seem to say the same thing differ in a fundamental way. Take this sentence: (1) John walks. The tensed Verb (Phrase) ‘walks’ belongs to a category that can take a term-phrase (TP = DP) like ‘John’ as argument to make a Sentence denoting a truth value (or perhaps a proposition). So the very same structure is instanced in (2). (2) Every man walks. English is not a pronominal argument language.

46

Emmon Bach

The closest match for a sentence like (1) in a pronominal argument language is something we can paraphrase as: (3) John, he walks. The difference between the language types comes out in the fact that there is no counterpart to (2) in the pronominal argument language. We can get an intuition about this from an English example: (4) #Every man, he walks. (4) cannot be interpreted as a paraphrase of (2). The hypothesis declares that in languages of the pronominal argument type, there are no items corresponding to the tensed Verb (Phrase) ‘walks’. Similarly, for transitive verbs and their arguments. There are no sentences like (5), compare (6). (5) Every man sees some horse. (6) #Every man, he sees it, some horse. So the syntactic category of the element that contains the main verbal idea is actually S, its semantic type is t, the type of truth values (or in alternative formulations, propositions). More particularly, the verbal element is an open sentence with unbound variables representing the “pronominal arguments” of the hypothesis. In a categorial system the differences can be seen (for the case of the subject term phrase) in these schemata (with TP for Term Phrase): (7) TP TP\S (for Examples 1 and 2) (8) TP S (for Example 4) Since S is not a functor category, there is no way to combine it with the subject by direct functional application. Sentences like (3) are interpreted in a different way: the TP is interpreted as coreferential to the pronominal (variable). Such a relationship is impossible with Example 4, as the term phrase is interpreted as a generalized quantifier, and doesn’t admit an individual referential interpretation. For this account to work it is necessary to exclude mechanisms that might bind the variable by a lambda or a lambda-like operator. This would turn the open sentence into an expression that could be combined with the TP by functional application. The term phrase, as a generalized quantifier, has this categorial figure: TP = t/(e\t). Binding the subject variable with a lambda turns the sentence into an expression of the category e\t, that is a proper argument type for the generalized quantifier. It is further necessary to prevent a bare verbal stem (without a pronominal argument) from combining with items of category e directly. There are two ways this could be done: one is to assume — as in Montague’s PTQ (Montague 1973) —

Categories and pronominal arguments

that all terms, including individual names — come into the syntax as full generalized quantifiers; the other is to assume that bare stems are dealt with in a component or set of rules separate from expressions of the (word-external) syntax proper. The latter course is I believe most consistent with Jelinek’s proposals. It is also consonant with the view discussed below in the second note. The first conclusion that I want to draw here is this: Jelinek’s account for these language differences depends on some such assumptions as those given as (i) and (ii) above. The second conclusion is that this explanation, if correct, establishes a negative answer to the first question at the head of this note: all languages do not share the exact same set of syntactic categories: pronominal argument languages lack the categories of Transitive Verb (Phrase) and tensed Verb (Phrase) of Englishlike — that is, non-pronominal-argument — languages. I believe the second conclusion is easy to see. I will discuss the first. Assumption (i) prevents us from using the fudge factor offered by the use of features. Without it we could say: well, there are two kinds of tensed VP’s, those that have value x for a feature F, and those that have value y for that feature F. So after all, all languages have the same categories, it’s just that some languages have different feature values for some category that is at issue. I believe we can draw a general moral here. Without some fairly explicit spelling out of assumptions like those of (i) and (ii), it is very hard to make clear arguments about the general kind of issues raised by our initial questions.

2.

What is the semantics of the “predicational” element in pronominal argument languages?

I now bring in a different question, again clearly raised in Jelinek’s work, especially in Jelinek 1994, and Jelinek and Demers 1994. Are the categories of word-internal constructions the same as those of word-external constructions? We saw above that this question comes up in ensuring that we get the right consequences for combinations of names and verbal stems in pronominal argument languages. Again, I will here just make a few comments on issues around this question. First, an easy observation: the category of a tensed (inflected) verb like ‘walks’ is different from the category of the bare verb ‘walk’: (7) John can walk. (8) *John can walks. We must admit ‘zero’ elements, so that ‘walk’ in (7) is different from the last word in (9): (9) Jack and Jill walk.

47

48

Emmon Bach

If we dig back deeper into the internal structure of words, especially in languages with lots of action within words, we encounter roots, stems, and affixes, and operations that cannot, in my opinion, be accommodated into the categories of the syntax proper. Whatever the ultimate verdict on the question whether there are languages with no Verb — Noun distinction, as Jelinek has argued (Jelinek 1994; Jelinek and Demers 1994), it is clear that below the level of the inflected word there are roots and stems that are indifferent as to whether they are inflected as Nouns or Verbs. If they belong to some category, then they are something like the Predicational elements claimed for some of the languages of the Pacific Northwest. But some seem to be not even that. Jelinek and Demers cite the triliteral roots of Semitic languages as items that are not even nominal or verbal in character. The Arabic root {k-t-b} connotes the notion of writing but it is not in itself specific in syntactic terms and I suspect that it has no direct model-theoretic interpretation. Its meaning probably resides in the conceptual side of the semantic realms. In categorial terms, it is clear (in the light of Assumption (i) above) that most affixes involved in word formation cannot belong to the same categories as those of the syntax proper. If they did then we would expect them to exhibit the same distributional characteristics as inflected words and phrases. This holds also for compounding, in general. The authors cite examples from Lummi to make the same point. Here let me give a few examples from Wakashan languages In the Northern Wakashan language Haisla (Kitamaat Village, British Columbia), there is a root lhaxw-, which is the base for these words, among many (data from field notes, and Lincoln and Rath 1986. The regular reflexes of xw under conditions of voicing and glottalization induced by various suffixes are w and w’.): (10) a.

lhaxwela ‘tight, fit tightly, stormy, rough water, firmly fastened, wound up tight’ b. lhaxwmala ‘ache, be in pain, hurt, suffer’ c. lhaxwmec’uaqiala ‘(have) headache (and so on for many words for hurting or aching in a particular part of the body)’ d. lhaw’inai ‘hard, difficult, close friend, like, care, love’ e. lhaw’inainukw ‘love (someone), like’ f. lhaw’enem ‘husband’ g. lhawad ‘have husband’

Lincoln and Rath gloss the root like this: (11)

lhaxw- ‘strong, valuable, painful, love, difficult, tight’

This gloss is little more than the union of ideas that can be extracted from the individual items listed (and many more derivative words). There is no free form consisting of just the root, not even one with -a, the most neutral affix that is used for making free forms.

Categories and pronominal arguments

If the ideas just given are right, then we are led to another question, the last one posed in this note: If the meanings and categories of word-internal and wordexternal grammar are incomparable, would we not expect that the syntactic categories of polysynthetic languages and analytic languages might be quite different?

49

Doubling by Agreement in Slave (Northern Athapaskan)* Keren Rice University of Toronto

In comparisons between third person non-subject pronominal systems in Athapaskan languages, differences between the Apachean languages of the American southwest and the northern languages as a whole are often identified. In this article, I focus on one difference, the co-occurrence of noun phrases and the pronominal markers y-/b- as direct objects, objects of postpositions, and possessors. The basic facts as described in the literature are as follows. In the Apachean languages, non-subject noun phrases and pronominals co-occur (e.g., Sandoval and Jelinek 1989, Saxon 1989, Thompson 1996, Willie 1989, 1991, 2000, Willie and Jelinek 2000). In Northern languages in general, on the other hand, a non-subject noun phrase and a pronominal are in complementary distribution, with one or the other being present but not both (e.g., Rice 1989b, Rice and Saxon 2001, forthcoming, Saxon 1986, 1989, Tuttle 1996, but see Gunlongson 2001 on Babine-Witsuwit’en for discussion of a language where both structures are found). These facts have been analyzed in different ways. In influential work by Jelinek and Willie (e.g., Jelinek 1984, 2000a, Sandoval and Jelinek 1989, Willie and Jelinek 2000, Willie 1991, 2000), it has been argued that in Navajo all nominals are adjuncts, and pronominals have the status of arguments. Rice and Saxon (e.g., Rice 2000a,b, Rice and Saxon 1991, 2001, forthcoming, Saxon 1986, 1989, 2001, Saxon and Rice 1993) on the other hand argue that nominals have argument status, and that the pronominals in question represent inflectional agreement; see also Potter (1997) on White Mountain Apache and Speas (1990) and Speas and Yazzie (1996) on Navajo. Rather than entering into this very interesting debate on the status of these elements, I assume the latter framework, that nominals can be in argument positions, and that y-/b- are agreement. I thus refer to the pronominals in question, y-/b-, as inflectional agreement markers rather than as pronouns. My goals are two-fold. First I examine the claim that non-subject non-pronominal elements and inflectional agreement are in complementary distribution in Slave [slevi], a language of northern Canada. I show that, while this statement holds

52

Keren Rice

overall for Slave, constructions exist in which a nominal non-subject and inflectional agreement co-occur, and I propose that in such constructions the nominals in question are dislocated and thus in adjunct rather than argument positions, and the presence of inflectional agreement is expected. This proposal places Slave in the class of languages discussed by Baker (this volume) in which, to paraphrase, a head agrees with an NP if and only if the NP is in a dislocated, adjunct position. Second, I compare the facts of Slave with those of Navajo, and outline possible directions of change. This article should be regarded as part of an ongoing research program (see Rice and Saxon forthcoming), and a part that is in an early stage of development. Much additional data is required to confirm whether the approach proposed here can be sustained.

1.

The Slave facts as generally described

In Slave, non-subject nominals and inflectional agreement generally are in complementary distribution.1 In (1) and (2) direct objects are illustrated. I present only examples with third person subjects. The first clause of (1) has no nominal direct object, and overt inflectional agreement (y-, b-) occurs; the y- form is realized as y, zh, or i- and the b- form as b-, m-, mb-. Which marker is found is determined by factors that are not relevant here; see Rice (1989a,b, 2000a,b) and Rice and Saxon (forthcoming) for discussion of non-third person subjects and for details on the distribution of various third person inflectional agreement forms in Slave. The inflectional agreement markers are bolded, and the verb phrases in which they occur are enclosed in square brackets.2 (1) Slave direct objects: pronominal object semo˛ [zhéhchá] gots’e˛h tudhe thehtsi˛ 1sg.poss-mother 3sg-pf-boil-y.DO and soup 3.S-pf-make “My mother boiled them and made gruel.” (Thom and Blondin Townsend: 40) In the second clause in (1) and in (2), a nominal direct object argument is present, and there is no overt inflectional agreement; if there were, the verb would contain y-. (2) Slave direct objects: nominal object gok’o k’eaehdlá 3pl.poss-neck 3.S-pf-break “He broke their necks.” (Y˛o t’˛óh gondi: 23) The next examples show postpositions with their objects. Again, nominals and inflectional agreement are in complementary distribution. The examples in (3) have

Doubling by Agreement in Slave (Northern Athapaskan)

pronominal oblique objects and inflectional agreement is present. The postposition and its object are in square brackets. (3) Slave oblique objects: pronominal object a. Sáh ozhøø zo˛ móla gha eghálai˛hndá, [megha] Sue ’ah’i˛ one time only white for 1sg.S-pf-work b.OO-for fish 1sg.S-imp-do “I only had to work for a white man, getting fish for him …” (Thom and Blondin-Townsend: 35) b. cho˛ [mezhíi] at’i˛le rain b.OO-inside 3.S-imp-go-neg “The rain didn’t go inside it.” (Thom and Blondin-Townsend: 75) c. etthe˛´ gots’e˛h ek’a k’ála [mezhíi] thela meat and fat still b.OO-inside 3pl-imp-be.located “The meat and fat were still inside there!” (Thom and Blondin-Townsend: 95) With nominal oblique objects, again we find an absence of overt inflectional agreement. (4) Slave oblique objects: nominal object a. [ndéh ts’e˛] gohndíh [land from 1sg.S-imp-live “I live from the land.” (Thom and Blondin-Townsend: 46) b. [nimbáa hodé˛e˛´ ch’oi zhíi] nahthídéh [tent pointed inside 1pl.S-pf-live “We lived in a teepee.” (Thom and Blondin-Townsend: 78) c. [ndéh k’eh] goloøh ˛ ’i˛hSó˛, golo˛, [land on animals pf-be.many moose tsá gots’e˛h sah, dúh lah séé k’ets’e˛ agejá beaver and bear now really surely ? 3pl.S-imp-go “There used to be so many animals on the land, moose, beaver, and bears, but now they have become very scarce.” (Thom and Blondin-Townsend: 105) d. haye néh [tSi˛ t’áh] [shíh tah] [tue ts’é˛] winter when [dog by.means.of [mountain among [lake to athít’i˛ 1pl.S-pf-travel “We travelled by dog sled in the winter to the big lakes around the mountains.” (Thom and Blondin-Townsend: 111)

53

54

Keren Rice

Finally, the same pattern is found in possessives: if the possessor is pronominal, inflectional agreement occurs (5); if it is nominal, inflection agreement is absent (6). (5) Slave possessors: pronominal possessor a. [meli˛é] nátSa b.poss-dog 3.S-pf-go.fast “His/her dog went fast.” (Rice 1989: 207) b. [yembehchiné] náéhndí [y.poss-sled 3.S-pf-buy “S/he bought his/her sled.” (Rice 1989: 208) (6) Slave possessors: nominal possessor a. [Charlie li˛é] t’áh ’aoht’é [ dog by.means.of 1sg.S-opt-go “I want to use Charlie’s dogs.” (Thom and Blondin-Townsend: 229) b. [’amá ts’éré] laidi [mother blanket where “Where is mom’s blanket?” (Rice 1989: 229) In summary, Slave nominals and overt inflection are in complementary distribution. I now turn to some complexities of Slave, showing that the facts are not at first glance so simple as described. The above description represents the canonical case, with nominals and inflection in complementary distribution. However, surface constructions are found in which both a nominal and inflectional agreement are present.

2.

Relative clauses and topicalized structures

Despite what we have seen, Slave nominals can be doubled by agreement. Two constructions which illustrate this are externally-headed relative clauses and topicalized structures, and these are the subject of this section. In these cases, the analysis in Section 1 — the nominals under consideration are arguments and agreement is triggered only by pronominals — is maintained, as the structures here involve displacement contexts where the agreement-triggering DP is not in a local relation with the word showing agreement. In essence, if the DP in question appears outside the monoclausal domain of agreement, complementarity is not enforced. Instead, there is a dependency between the adjunct and a null pronominal in the extraction site, and this pronominal is the agreement trigger. 2.1 Relative clauses Slave has two types of relative clauses, head-internal and head-external. Examples are given in (7), where the head (Sue ‘fish’) and restricting clause are enclosed in brackets.


The externally headed relative clause (7b) is a construction in which nominals and agreement do not appear to be in complementary distribution. (7) direct object as head a. internally headed [seyaa [Sue] Sani˛hdé] s˛íi meghánila [1sg.poss-son [fish 3.S-pf-kill pl comp 1sg.S-pf-give.pl-b.OO “I gave him the fish that my son caught.” (Rice 1989: 1310) b. externally headed [Sue] [seyaa Sayéni˛hdé] s˛íi meghánila [fish [1sg.poss-son 3.S-pf-kill pl-y.DO comp 1sg.S-pf-give.pl-b.OO “I gave him the fish that my son caught.” (Rice 1989: 1313) The differences between these sentences involve word order and the use of inflectional agreement (y-) triggered by the direct object. When the head is internal to the relative clause (7a), as shown by the fact that the understood head is surrounded by elements of the restricting clause, word order within the restricting clause is SOV, the normal word order in Slave, and no overt inflectional agreement element occurs in the verb. In the externally headed structure (7b), in which the understood head precedes its restricting clause, word order is OSV and inflectional agreement (y-) occurs in the verb. The object of a postposition can also form the head of a relative clause. (8) a.

internally headed [neyaa [tthik’íhí] t’áh golo˛ thehk’é] s˛íi [2sg.poss-son [gun by.means.of moose 3.S-pf-shoot comp seghánéneto˛ 2sg.S-pf-give sticklike O-1sg.OO “you gave me the gun that your son shot the moose with.” (Rice 1989b: 1312) b. externally headed [tthik’íhí] [neyaa yet’áh golo˛ thehk’é] [gun [2sg.poss-son y.OO-by.means.of moose 3.S-pf-shoot s˛íi seghánéneto˛ comp 2sg.S-pf-give.sticklike.O-1sg.OO “you gave me the gun that your son shot the moose with.” (Rice 1989b: 1314)

In the internally-headed relative clause (8a), the understood head is surrounded by elements of the restricting clause, while with an external head (8b), the understood head precedes its restricting clause. As with (7), two differences exist between internally and externally headed clauses, word order and absence/presence of inflectional agreement. Overall, when the understood head of the relative clause is internal (7a, 8a), the external head position is empty; when the understood head is external (7b, 8b), a gap

55

56

Keren Rice

occurs in the restricting clause. In the former, the presence of the nominal object within the restricting clause blocks the presence of overt agreement, as in the examples in Section 1 with nominal arguments. In the latter, the gap within the restricting clause is filled by a null pronominal which is coreferential to the external head; this element triggers agreement which is realized on the head, the verb in (7b) and the postposition (8b), as in the examples in Section 1 with pronominal arguments. I refer to this pronoun as a resumptive pronoun without pursuing the theoretical implications of this term. Externally-headed relative clauses involve adjuncts which are outside the monoclausal domain of agreement, and the within-clause pronoun is the agreement trigger. When the head is internal to the relative clause, the relative clause has a structure parallel to that of the examples in (1), (3), (5), where the nominal does not trigger agreement. The domain of agreement includes a complement and its head (direct object and verb, postpositional object and postposition), with agreement found within a maximal projection if the head c-commands its pronominal complement. 2.2 Topicalized structures A second construction exists in Slave in which doubling of a nominal by overt inflection is expected under the hypothesis outlined in Section 1, topicalization.3 Examples of topicalized constituents, in square brackets, are given in (9) and (10). (9) topicalized subject [’edi˛ á] yado˛ [3sg top 3.S-pf-drink-y.DO “That one, s/he drank it.” (Rice 1989b: 1213) (10) topicalized oblique object [ts’éli˛a á] yahti yets’é˛ gondeh [girl top Father y.OO-to 3.S-imp-talk “The little girl, the priest is talking to her.” (Rice 1989b: 1197[modified]) These examples show a pronominal subject (9) and a nominal oblique object (10) dislocated to the left edge of the sentence. In (9) the topicalized constituent can be coreferential only to the subject, as independent pronouns refer only to animates. In both examples, the topicalized constituent is followed by a morpheme yá or á, analyzed by Rice (1989b: 1213) as marking a topicalized constituent. In addition, with the oblique object (10), the nominal object does not directly precede the postposition, as is generally the case. Finally, with the oblique object both a nominal and inflectional agreement (y-) occur. Additional examples, without an overt topic marker, are given in (11) and (12). Note that normal SOV word order is not found — DO S y-V occurs in (11) (normal order would be S DO V) and OO S y-P V order in (12) (usual order is S OO P V).


(11) [gah] tSi˛ ni˛dhá˛a˛´ teyedéhnde [rabbit dog far 3.S-pf-chase-y.DO “It was the rabbit that the dog chased far.” (Rice 1989: 1198) (12) [’elá kéetlah ˛ilé ’és˛íi] mecheekué yek’eh [boat 3.S-pf-go past foc b.poss-follower y.OO-on dékégogi˛hthe 3pl.S-pf-go.into “His followers got into the boat he was in.” (Rice 1989: 1199) In (13), an oblique object which is itself an internally-headed relative clause is topicalized, and b- (realized here as m-) occurs within the sentence. (13) [ezhi˛ dene úzi lah] dúh dze˛e˛ k’ála megáh náhnde [that man 3-name part this day still b.OO-with 1sg.S-imp-live “And the man he named is the man I am still living with today!” (Thom and Blondin-Townsend: 102) Topicalized arguments, like externally-headed relative clauses, require doubling of a nominal by inflectional agreement with non-subjects. Just as the facts of externally headed relative clauses are accounted for structurally, those of topicalization receive a structural account. Consider the position of a topicalized constituent. These are argued by Rice (1989b: 1209) to be adjuncts to the main clause, occurring as a specifier of some projection which is high in the tree; I will call this CP (see the tree in (16), Section 3). This adjunct is coreferential with a resumptive pronoun in the main clause, and it is this pronoun that is the agreement trigger. Just as with externally-headed relative clauses, the pronominal trigger and agreement occur within the same maximal projection, with the nominal in a dislocated non-argument position outside this projection. Topicalized structures are found in which cues from word order and the presence of the topicalization marker are absent. (14) [súdi] yeyi˛’á [candy 3.S-pf-eat-y.DO “It’s candy s/he ate.” (Rice 1989: 1198) (15) [yú] bek’e ’edéht’o˛ [clothes b.OO-on unspecified object-pf-be.handled “The clothes, they’re ironed.” (literally: “unspecified object is handled on them, the clothes”) (Rice 1989b: 1200) Despite the absence of word order evidence, three criteria suggest that the first noun phrase in these examples is topicalized. First is the unexpected presence of pronominal agreement, y- in (14) and b- in (15). Second is their interpretation.

57

58

Keren Rice

Rice (1989b: 1214–1215) points out that one function of topicalization is to place contrastive material at the front of the sentence; in both cases, the initial noun is contrastive, as indicated by the glosses. Third, phonological phrasing suggests this analysis. In topicalized structures with the topic marker, the topicalized constituent is intonationally set off from the remainder of the clause; that is true of topicalized constituents in general. As in previous examples, the nominal is dislocated, appearing in an adjunct position, and the agreement trigger is a resumptive pronoun that is a direct object in (14) and an oblique object in (15). Thus while in both externally-headed relative clauses and topicalized structures doubling of a nominal by agreement is found, these structures do not require revision of the triggering elements in Slave — despite the presence of nominals in these structures, these nominals are not arguments but adjuncts, and agreement is a result of the presence of a pronoun within the maximal projection of the head.

3.

Beyond relative clauses and topicalization

Externally headed relative clauses and topicalization show that surface doubling by inflectional agreement is found in Slave, but is accounted for if the nominal is an adjunct; it is a null pronominal argument which triggers agreement. I now examine other cases of doubling, instances which have, to my knowledge, been overlooked in the literature. The examples presented in this section are drawn from texts, and are of quite low frequency in the texts that I studied.4 Before turning to these examples, however, some discussion of word order in Slave is appropriate. In studies of Slave word order, (Rice 2000a,c) and Rice and Saxon (1991, forthcoming) argue that subjects, direct objects, and oblique objects occur in external and internal positions. Their evidence comes from a variety of sources: idioms, incorporation of nouns into the verb, word order variation, and variation between y- and b- realizations of oblique objects in third person environments. Important here is word order, as I will use it as a diagnostic to show that the nominals under consideration are not in the topic position discussed in Section 2. Abstracting away from non-third person subjects, the tree in (16) represents the overall structure of the Slave clause argued for in the above references. (See Rice and Saxon forthcoming for a more fully articulated structure.)5


CP

(16) DP topic

IP DP external subject

I′ VP

I°

PP VP external oblique object DP VP external direct object DP internal subject

VP

PP V′ internal oblique object DP internal direct object

V

The generalization discussed so far, that if a non-subject DP is in argument position overt inflectional agreement is not found and if a DP is in adjunct position overt inflectional agreement is present, holds as follows. If the DP is in SpecCP, a contrastive topic, agreement occurs; if the DP is a non-subject argument in IP then overt inflectional agreement is not present. (I do not pursue the structure of externally-headed relative clauses.) In the remainder of this section, I examine cases in which third person DPs are not so clearly adjuncts, arguing that if the phrase structure is appropriately enriched, these receive an account under the hypothesis that agreement is found overtly only when the nominal involved is an adjunct. 3.1 Postpositional phrases I begin with a study of postpositional complements, examining several situations in which doubling constructions are found. Consider first a case of doubling where the complement of a postposition is followed by a focus particle. The nominal object is in square brackets, as is the postposition and the inflectional agreement marker. (17) Doubled postpositional objects dúh [ndéh lah] [mets’e˛h] ’ahsíi Sole ˛ now [land part [b.OO-from things many-neg “Now there are fewer things from this land.” (Thom and Blondin-Townsend: 24)

59

60

Keren Rice

In (17) a nominal (ndéh ‘land’) is followed by a focus particle (lah) which has scope over the preceding noun phrase. The noun is an object of the postposition ts’e˛h ‘from’. Note that in addition to the nominal complement, inflectional agreement, in the form b- (bolded), is present on the postposition, contrary to the examples in Section 1. Notice also the word order: Adverb – Postpositional Phrase – Subject – Verb. This sentence has an internal subject (see (16)). It is unlikely that the object of the postposition is topicalized — it is not contrastive in the discourse and it is preceded by an adverb which might be topicalized, as the story contrasts things in the past with things today. The presence of a focus particle does not require the doubling construction, as the following example shows. Here and in similar examples, square brackets enclose the postposition with its nominal complement. (18) á yu’ó˛ dla yundaa á part over there part from.there part [ndu dla ts’é˛] nóodéht’e… [island part to 3.S-pf-go.across.water “From over there, he paddled across to the island …” (Y˛o t’ó˛h gondi: 36) The focus particle dla, a variant of lah in (17), intervenes between the postposition (ts’é˛ ‘to’) and its complement (ndu ‘island’), but does not require the appearance of inflectional agreement on the postposition. In the following examples, we see postpositional complements that are complex in some way. As in (17), the object is doubled by inflectional agreement: (19) [’ahsíi azho] ˛ kíi [mé]ts’edadíle [thing all just [US-pf-touch-b.OO-neg “Everything was untouched.” (literally: “we did not touch anything”) (Thom and Blondin-Townsend: 111) (20) [edi edi˛htS’éh] [meghoh] ˛ segha máhsi [this book [b.OO-about 1sg.OO-for thanks “I am happy about this book.” (literally: “thanks are for me about this book”) (Thom and Blondin-Townsend: 27) (21)6 [k’i tene gots’e˛h tth’á] [met’áh] k’enáets’enedih shu [birch basket and pans [b.OO-with US-imp-wash and ezhi˛h thela there 3.S-imp-pl.be.located “There were also birch bark baskets and pans that you use for washing your face.” (Thom and Blondin-Townsend: 61)


In (19)–(21), the nominal which is doubled by inflection is at the left edge, followed by an inflected postposition. Given their sentence-initial position, one might propose that these nominals are topicalized and thus adjuncts in SpecCP, as discussed in Section 2.2. If so, then the presence of agreement is expected. However, the semantics of topicalization suggest that this is unlikely. In Rice (1989) I propose that topic position houses contrastive information (see Section 2.2) and discourse topics. In (19)–(21), the information introduced in the first clause is non-contrastive and is new, or focussed, not having been mentioned. In the discourse previous to the sentence in (19), the author talks about how the land was, and summarizes with this sentence, expanding but not contrasting. The sentence in (20) comes after the author’s narrative about his life; the book has not been discussed, although it is not contrasted with anything that came before. The sentence in (21) adds to a list of items found inside a tent, where the list began in the previous clause. Prosodic information could potentially sort out structure, as discussed in Section 2.2. Because these examples are drawn from written texts, information about phrasing is not available, although the absence of punctuation suggests that no intonational break occurs. I assume that the examples discussed in this section do not involve topicalization; see below for additional discussion of their semantics. Returning to environments where inflectional agreement doubles a nominal, in (22) a postposition k’eh ‘on’ takes a postpositional phrase as its complement. Topicalization is unlikely since the object represents non-contrastive focus — the author is talking about what she did as her mother was dying; in the previous clause she says that she prayed on her cross. Coming to the end of the rosary does not appear to be contrastive with this.7 (22) [’ehts’o melo˛ gots’é˛] [mek’eh] zhaihti [bead b.poss-end ar-to [b.OO-on 1sg.S-pf-pray “I came to the end of my rosary.” (Thom and Blondin-Townsend: 56) The postposition k’eh ‘on’ takes as object a postpositional phrase; inflectional agreement is not required in this environment, as shown by the example in (23) where again the complement of the postposition ts’´˛e ‘to’ is itself a postpositional phrase, but inflectional agreement is absent. (23) [[ladá yee] ts’é˛] [[table under to “to under the table” (Rice 1989b: 1002) Next consider examples with complement clauses. In these examples, the complementizer is a postposition. It may be inflected, as in (24).

61

62

Keren Rice

(24) [golo˛ oki ˛ Sønidéh gogee setá theti˛] [moose two dead 3pl.OO-between 1sg.poss-father 3.S-imp sleep [meghoh] ˛ náthite [b.OO-about 1sg.S-pf-dream “I dreamt that my father was sleeping between two dead moose.” (Thom and Blondin-Townsend: 39) However, it need not be inflected, as in (25). (Note the intervening adverb kíi as well.) (25) [Ndéh nahets’e˛ hó˛ot’e ˛ nahégedi nidé] [land 1pl.OO-to 3.S-imp-be 3pl.S-imp-say.to.1pl.OO when kíi [gho] ˛ gonihthe˛le just [about 1sg.S-believe-neg “When they tell us the land is ours I don’t believe what they say.” (Thom and Blondin-Townsend: 29) Finally, minimal pairs occur within a text, one of which has an inflected postposition and the other of which does not. The following is from a section of a story where a father gives his son advice about what to do in order to have a vision. The slippers referred to have already been introduced, and the sentence in (26) involves direct discourse, where the father speaks to the son about sleeping with his head on the slippers. (26) [mo ké ch’iláa] [mbek’eh] níatthinit’ah [mother moccasins ragged [b.OO-on 2sg.S-imp-lay.ones.head “You sleep on mother’s raggedy moccasins.” (Moore and Wheelock: 156) Later in the same story, it says, in indirect discourse: (27) a.

[mbeké ch’iláa k’eh] nítthínt’on ayínlá [3-moccasins ragged on 3.S-pf-lay.head 3.S-affect-y.DO “(His father) made him put his head on her ragged/small moccasins.” (Moore and Wheelock: 156) b. [mo ké k’eh] ndatthíe’ón [mother moccasins on 3.S-imp-place.head “He slept on his mother’s moccasins again.” (Moore and Wheelock: 156)

In (27a), doubling is found, but earlier, when the slippers are introduced and later, as the information is repeated, only the nominal is present, without inflectional agreement. The example in (28) involves a nominal complement of a postposition (k’eh) and inflectional agreement. The embers are reintroduced in this sentence after a brief pause


between their first introduction and a change of sentential subject from a wolverine (previous stirrer of the embers) to a man (stirrer of the embers in this clause). (28) thidi [mbek’eh] andats’edla embers [b.OO-on 3.S-pf-do again “He stirred the embers again.” (Moore and Wheelock: 177) As mentioned earlier, although word order in these examples is such that topicalization could be involved, they do not to meet conditions for topicalization either semantically or, based on punctuation, phonologically. While topicalization is not tenable on semantic and phonological grounds or, at least in (17), in terms of word order, an alternative analysis involving dislocation suggests itself. In the brief discussions of the semantics of the examples above, the notion of focus arose on several occasions — the doubled complement is followed by a focus particle or it represents new, non-contrastive information. Based on semantics, one might expect that, in addition to a special position in which topics may occur, there is also a special position in which focused elements can be found. In addition to left-edge topicalized phrases, left-edge focused phrases have been argued for in Athapaskan languages (see Saxon 2001, Rice and Saxon forthcoming, Willie 1991), and elsewhere (e.g., Aissen 1992, Baker this volume). Aissen (1992) in particular argues for three left-edge positions in some Mayan languages, two topic positions and a focus position, hierarchically ordered with the topic positions above focus. If Slave has such a position, and if elements in this phrase count as dislocated, then the hypothesis argued for — that when an item is dislocated, a pronoun is present which triggers agreement — can be maintained. Several facts of Slave are congruent with an analysis that the doubled nominals in question are indeed focused. First, as remarked above, the semantic interpretation of DPs in this position clearly involves focus in a number of the examples — these DPs represent new non-contrastive items in the discourse or may be part of the discourse, but are reintroduced; they are sometimes followed by a focus particle. The full semantics of DPs in specifier position of this phrase, which I will call Focus Phrase, abbreviated FocP, remains to be determined, and many questions arise. For instance, the examples discussed in this section are given in pairs, where one has morphology that suggests that the DP is an adjunct and the other has morphology that suggests it is an argument in a local relationship with its head. What, if any, is the semantic difference between these? This question remains for further research. Second, word order facts are consistent with a left-edge position, although more evidence, namely the position of overt subjects, is needed to see if the hypothesis that the focused element is left-dislocated can be maintained. Word order provides evidence about the ordering of topicalized and focused constituents, which can co-occur. In wh- questions in Slave, the wh- word can optionally be fronted. Rice (1989a) argues that the position to which it is fronted is, in the terms

63

64

Keren Rice

used in this article, the specifier of FocP, and that the question words involved are focused (see Rice and Saxon forthcoming). In (29), the topicalized DP is in square brackets and the focused wh- phrase is underlined. (29) [Sesasóné] judeni˛ gots’e˛ Jim ráyeyéhdí [kicker where ar-from Jim 3.S-pf-buy-y.DO “Where did Jim buy the kicker?” (Rice 1989b: 1168) Notice the word order: Topic Focus Subject Verb. This analysis also accounts for word order in (17) — the initial adverb is a topic, and the following oblique object is in focus. This example also provides one bit of evidence that focused items precede subjects. Third, Baker (this volume) points to another source of evidence for dislocation, non-referential/quantified NPs (citing Cinque 1990, Baker 1996, Alexiadou and Anagnostopoulou 1998). It has been argued that nonreferential indefinite NPs cannot be dislocated, rather dislocated NPs have only a definite or specific reading. See Baker (this volume) for a summary; see Rice (2000c) for discussion of the verbphrase internal placement of indefinites in Athapaskan languages. Looking at the examples in this section, the nominals in question are definite and referential. Note, however, that, based on the examples in Section 2 and many of the examples in this section, it is not true that definite, referential DPs must be dislocated; it is only that it is possible for them to be. In this section we have seen that the generalization that a nominal and agreement are in complementary distribution in Slave can be maintained if the phrase structure is enriched to include FocP. The structures in which both a nominal and agreement are present involve dislocation — an external head of a relative clause or a topicalized or focused constituent in an adjunct position. 3.2 Possessive constructions Possessive constructions provide another circumstance in which doubling can occur. As discussed in Section 1, in the possessive construction generally only a single possessor, either nominal or pronominal, is present. Examples are given in Section 1 ((5)–(6)) of the usual form of the possessive construction: NPpossessor NPpossessed or Pronounpossessor Inflection NPpossessed Double possessive constructions are found. Generally, in inalienable possession (kinship terms, body parts) the doubling construction is used. Some examples are given below, with the possessive construction in brackets and the relevant inflection bolded.


(30) [[Bertha meyaa]] mendaá húle b.poss-child b-eye 3-imp-be.gone “Bertha’s son is blind.” (Rice 1989: 228) (31) [sedené móot’i˛] hálats’ededah hasegenéhto˛ [1sg.poss-husband b.poss-family US-imp-work 3pl.S-pf-teach-1sg.DO “My husband’s family taught me how to work.” (Thom and Blondin-Townsend: 43) (32) [chia bekw’ené] [bird b.poss-leg “bird’s leg” (Rice 1989: 228) (33) [semo˛ medené] náázéh nidé [1sg.poss-mother b.poss-husband 3-imp-hunt when “When my step-father [mother’s husband] went hunting…” (Thom and Blondin-Townsend: 86) There is one condition under which inflectional agreement is regularly missing — when the possessed noun is a body part in a generic construction; see Rice (1989b: 228–229) for details. A few examples are given in (34). (34) a.

medzih tth’ené caribou bone “caribou bones” (Thom and Blondin-Townsend: 40) b. bebí ghú baby tooth “baby teeth” (Rice 1989b: 229)

I do not further discuss this construction. One can imagine that generics have a structure such that the possessor can occur in only one position, perhaps a compound. The facts described above represent the usual state of affairs — a single possessor, nominal or pronominal, occurs in the possessive construction except in inalienable possession, where doubling is found; this is how Slave is described by Rice (1989b). However, examination of texts shows counterexamples in both directions — alienably possessed items where inflection doubles a nominal possessor and inalienably possessed items where no inflection is present. In (35), a nominal possessor and inflection occur in alienable possession. Word order and semantics suggest that the possessor is not dislocated in the sense that we have seen so far — it is neither topicalized or focused. (35) a.

Dehcho ezhi˛ [setá mekó˛e˛´ ] gu˛´ li˛ Big River there [1sg.poss-father b.poss-house ar.OO-imp-exist “At my father’s house at Big River, …” (Thom and Blondin-Townsend: 77)

65

66

Keren Rice

b. Nihshíh Kó˛e˛´ gots’e˛ [zhahti meli˛é] t’áh Fort Rae ar.OO-from [priest b.poss-dog by.means.of ejo˛ níitlah here 3.S-pf-sg,dual-arrive “He came from Fort Rae on the Father’s dogteam.” (Thom and Blondin-Townsend: 102) These sentences have in common that the possessor is human, an important factor in Slave (see Rice 1989b, 2001, Rice and Saxon forthcoming). However human possessors do not require doubling; see (6). In (36), the reverse occurs, inalienable possession with a nominal possessor but no doubling. (36) a.

[semo˛ t’óó] ka ’ehtse [1sg.poss-mother milk for 1sg.S-imp-cry “I used to cry for my mother’s milk.” (Thom and Blondin-Townsend: 19) b. [mbedené chon] eninht’a [b.poss-spouse womb 3.S-pf-cut “He cut open his wife’s womb.” (Moore and Wheelock: 127) c. yedzeé kín kaa eyi [nóghe wú] y.poss-heart in yes that [Wolverine tooth i mani’á that 3.S-pf-be.stuck.in-b.OO “Wolverine’s arrow was stuck in his heart.” (Moore and Wheelock: 189) d. [nóghe ts’íyoan] gheya [Wolverine wife 3sg.S-progr-come “Wolverine’s wife was coming.” (Moore and Wheelock: 181) e. [nóghe yadzáa] sóon ghedhéhxin [wolverine offspring and then 3pl.S-pf-kill “They killed Wolverine’s children.” (Moore and Wheelock: 182)

Both non-doubling and doubling possessive constructions occur. Examples are sparse, so I do not speculate on the function of doubling; I assume discourse conditions exist. In order to understand the possessive construction, it is necessary to address the internal structure of DP. Syntactically, given the hypothesis that if a DP is dislocated, a resumptive pronoun which triggers agreement occurs, possessors should be able to occur in two different positions. First consider the non-doubling structure.8


(37) non-doubling structure DP D

NP DP possessor

N head

Here the possessor is c-commanded by its head, and in the presence of a nominal possessor, agreement is not found. (38) doubling structure DP DP possessor

D′ D

NP DP pronoun

N head

In the doubling construction, the possessor is dislocated, in SpecDP. The resumptive pronoun in the lower DP requires that agreement be present on the head. 3.3 Direct objects I now turn to direct objects, focussing on sentences with on third person subjects. As we have seen, while the claim is usually made in Slave that nominal and pronominal direct objects are in complementary distribution, this is not a surface true generalization. Direct objects also provide examples with both a nominal direct object (in brackets) and inflection (bolded) in the same clause. These examples involve y-. (39) semóo [edhéh gots’e˛h ’ahtS’uh] ts’u chi˛˛i 1sg.poss-aunt [moosehide and babiche tree at base nízhenih’i˛ 3.S-pf-hide-y.DO “My aunt had hidden a moosehide and some babiche in a tree trunk.” (Thom and Blondin-Townsend: 77) Given the word order, the direct object in (39) cannot be in either SpecCP or SpecFocP, but is rather in an expected position for objects, external object (see (16)). In (39), the nominal direct object and inflection are not adjacent. One might think that this is why inflection is present. Rice (2000a) proposes that agreement

67

68

Keren Rice

occurs in Slave if a DP is not governed by and adjacent to the verb; Rice and Saxon (forthcoming) argue that the evidence is inconclusive. Examples like (39) suggest that the adjacency hypothesis might be correct, as a postpositional phrase (‘in a tree trunk’) intervenes between the direct object and the verb. However, parallel examples exist with non-object constituents in immediate preverbal position where no inflectional agreement is present in the verb. [chi So] ˛ See yie athedhah [ducks many ashes in 3.S-pf-put.pl “He put many ducks in the ashes.” (Y˛o t’ó˛h gondi: 25) b.9 [dene náke dechilekuéke náke] kúhdé de’ah [people two worker-pl two that way 3.S-imp-send “He sent two people, two of his workers that way.” (Rice 1989b: 1352)

(40) a.

In these examples, the direct object is separated from the verb by an oblique object (40a) or an adverb (40b), and agreement is absent. In addition, a nominal direct object can be adjacent to a verb and inflectional agreement be present. In the example below, y- is realized as i-; the nominal direct object (i chi ‘the duck’, ayó˛ ‘all’) is in brackets. (41) i t’áh i chi dla i [chi] that with that duck part that duck idedzéh. Nogé dla [ayó˛] 3.S-imp-eat.up-y.DO. Fox part all idedzéh, idedzéh, kayeh’i˛ 3.S-imp-eat.up-y.DO 3.S-imp-eat.up-y.DO 3.S-pf-do “And then he ate up the ducks. Fox ate up all of them; he did that.” (Y˛o t’ó˛h gondi: 26) A similar example is given in (42). (42) k’ach’u [ayó˛] nayidéhdze k’ach’u again [all 3.S-pf-eat.up.again-y.DO again “Once again he ate up everything.” (Y˛o t’ó˛h gondi: 28) In terms of function, as with postpositional and possessive phrases, conditions that determine which structure is chosen are not well-understood. One possibility that holds of the available examples is that the nominals in question are quantified in some way — they are plural or quantificational. Again, additional evidence is required to substantiate this hypothesis, but all of the doubling examples that I have found are congruent with it. In this regard, it is interesting to note that when a human (or humanized) plural non-subject argument is present, plural agreement (go- in the examples below) generally occurs even in the presence of an overt nominal; see Rice (1989b) for discussion.


(43) a.

human plural direct object ’eyi dene yeyaake tehgodéhnde the man y.poss-child-pl 3.S-imp-chased-3pl.DO “The man chased her cubs.” (Rice 1989b: 1020) b. human plural oblique object deyaake gots’e˛ gondeh refl-child-pl pl.OO-to 3.S-imp-talk “She speaks to her children.” (Rice 1989b: 1020) c. ’amá hé ’abá he goli˛é mother and father and 3pl.OO-dog “mom and dad’s dog” (Rice 1989b: 230)

While further work is required, I assume that the doubled construction is possible when the argument in question is plural or quantificational, and that the construction is required if the argument is human plural. Consider the syntactic structure now. Is it possible to maintain the hypothesis that agreement is found together with a nominal only if that nominal is dislocated, and it is a pronominal that is the actual agreement trigger? Some facts suggest that there might be a level of structure between DP and NP, one that I will call a Quantifier Phrase (QP). We have seen that a DP can contain a determiner and a modifier (see note 7). In addition, a determiner can have in its scope a postnominal quantifier or numeral, as in (44). (44) ’eyi ’ekwé˛ honeno˛ the caribou ten “the ten caribou” (Rice 1989b: 1000) Suppose that the DP has the following structure: (45)

DP D′ D

QP Q′ NP

Q

The position of the specifier of QP is supported by word order evidence: quantifiers typically follow the NP, while specifiers of quantifiers precede, yielding intensifiernoun-quantifier order, as in (46). (See Rice 1989b for discussion of word order.)

69

70

Keren Rice

sure tsáwé So˛ wútsi˛ ken˛iwe˛ really fur lots 1pl.S-opt-make 3pl.S-imp-want “They want to get really lots of fur.” (Rice 1989b: 361) b. ’ed˛í tSeh So˛ k’edetSa too.much gas lots 3.S-imp-burn “It burns too much gas.” (Rice 1989b: 359) c. ’ed˛í dene So˛ too.much person many “very many people” (Rice 1989b: 359)

(46) a.

Suppose that the NP were dislocated so that it appeared in SpecQP, with a resumptive pronoun in NP. Such an analysis accounts for the agreement facts illustrated in this section, and provides another dislocation position within DP. A major consequence of the analysis is that the triggering condition on the appearance of agreement — that it is present only under conditions of dislocation — can be maintained, and a unified analysis of triggers is possible. The only time that there appears to be agreement triggered by a nominal, i.e., doubling occurs, is when the nominal is dislocated, or an adjunct. Given the enriched phrase structure proposed, structure for which there is evidence from word order, Slave fits well into Baker’s typology: agreement occurs only when the DP or NP that is doubled is an adjunct and never when it is an argument.

4. Relating Slave and Apachean In Slave, we have seen that a nominal and a pronominal can co-occur when the nominal is dislocated, but no doubling of true arguments occurs in this language. Slave contrasts with Navajo, where a nominal and inflectional agreement cooccur.10 The sentences in (47) illustrate direct objects. Note that the verb does not differ in the two sentences in that in both cases, y- is present. (47) a.

Navajo direct object: nominal object ’ashkii [at’ééd yiyiiStsá˛] boy [girl 3.S-pf-see-y.DO “The boy saw the girl.” (Willie and Jelinek 2000: 262) b. Navajo direct object: pronominal object yiyiiStsá˛ 3.S-pf-see-y.DO “She/he saw her/him.” (Willie and Jelinek 2000: 262)

In (47a), the inflectional agreement marker y- doubles the nominal at’ééd ‘girl’. The examples in (48) show that oblique objects pattern in the same way, with inflectional agreement doubling an oblique object (48a). (48b) is ambiguous; the


reading of interest is the first, with a pronominal oblique, and the bracketing shows this reading. (48) a.

Navajo oblique object: nominal object [tsé yikáá’] neezdá rock y.OO-on 3-pf-sit “He sat down on the rock.” (Willie 1991: 63) b. Navajo oblique object: pronominal object ’ashkii [yikáá’] neezdá boy [y-on 3-pf-sit “The boy sat down on it.” (also “He sat on the boy.”) (Willie 1991: 64)

(48a) shows the doubling of tse ‘rock’ by inflectional agreement. Finally, (49a) shows inflectional agreement with a nominal possessor; in (49b) a pronominal possessor is present. (49) a.

Navajo possessor: nominal object shínaaí [shimá sání yiyéél] 1sg.poss-older brother [1sg.poss-grandmother y.poss-pack yá náiníg˛í y.OO-for 3.S-pf-carry.back “My older brother carried my grandmother’s bundle back for her.” (Young and Morgan 1987: 9/grammar) b. Navajo possessor: pronominal object shimá shizhé’é nádzáago 1sg.poss-mother 1sg.poss-father 3.pf-return-comp [yi’éé’] yich’i˛’ ch’íyiiz’ah [y.poss-clothes y.OO-to 3.S-pf-toss.out-y.DO “When my father got back, my mother tossed his clothes out to him.” (Young and Morgan 1987: 9/grammar)

In (49a), y- doubles the possessor (‘grandmother’). Rice and Saxon (Rice 2000a,b,c, Rice and Saxon 1991, 2001, Saxon 2001, Saxon and Rice 1993) argue that Slave-type and Navajo-type languages are alike in that the nominals in the above examples are arguments. A fundamental difference between Slave-type languages and Navajo-type languages, they propose, has to do with the nature of the agreement trigger (see Rice and Saxon 1991, forthcoming). In Slave, the trigger is pronominal; a nominal does not trigger agreement. In Navajo, on the other hand, either a pronominal or a nominal triggers inflectional agreement (47), (48), (49).11 I now examine the differences between Slave and Navajo in terms of the triggering of agreement and structure. Willie (2000: 381) comments that ‘historical

71

72

Keren Rice

and comparative studies within the Athapaskan family are needed to account for the distinct syntactic developments involving the third-person forms across the language family’, and this discussion is a small step towards this goal. I assume, without argument, that the Slave pattern is conservative, and the Navajo pattern innovative, and I sketch a possible pathway of change under this assumption. This assumption does not seem unreasonable — doubling constructions are reported in a few other northern languages including Babine Witsuwit’en12 (Gunlogson 2001), Kaska (Moore 2001), and Dena’ina (Thompson 1996) — but it requires further study. Assuming that a surface contrast between a doubling and non-doubling construction is conservative, Navajo has not innovated its doubling construction out of nowhere, but the roots of Navajo are present in the north. Here I pursue the question of the structure of Slave and Navajo nominals, showing that Slave has evidence for the DPs which have so far been assumed while Navajo has been analyzed as not having DPs (Speas and Yazzie 1996). Navajo NPs can occur in SpecCP (topic) and SpecFocP (focus); if NP is the highest level of structure available, there are no within-DP positions to which an NP can be dislocated, and thus no contrast is possible. In Apachean, doubling is always found but has no special semantics associated with it. First consider the arguments for DP in Slave. These are of the standard type. Both nouns and determiners can stand on their own, serving as subjects, objects, and so on. This is illustrated in (50) for nouns (50a) and determiners (50b). (50) a.

noun i. ’elá yenehwhe˛ boat 1sg.S-imp-want “I want a/the boat.” ii. ’edi˛tS’é sáé˛chú paper 3.S-pf-give.sg “S/he gave me a/the paper.” iii. tthe t’áh rock by.means.of “with a rock” b. determiners i. deri yenehwhe˛ this 1sg.S-imp-want “I want this.” (Rice 1989b: 255) ii. ’eyi sáé˛chú that 3.S-pf-give.sg-1sg.OO “S/he gave me that.” (Rice 1989b: 256) iii. ’eyi t’áh that because “because of that.” (Rice 1989b: 255)


That the determiner can function as a noun is most obvious in (50bii-iii), where it patterns parallel to an argument in that inflection is not present.13 In the absence of the determiner, an inflectional agreement element would be required (saye˛chú ‘s/he gave it to me’ in (50bii) and bet’áh or yet’áh ‘because of it, by means of it’ (choice of inflectional agreement form depends on context) in (32biii)). If NPs are part of DPs, the ability of a determiner to serve as a noun substitute has an account: DP occupies the relevant positions of subject, object, and so on. In examples like (50a), an NP complement is present and the determiner head is phonologically empty; in (50b), no overt nominal complement is present, but only the head of DP, D. Further, a determiner and a nominal can co-occur (e.g., 20, 36c, 41, 43, 44). Word order facts also suggest the need for DP. Here I ignore post-nominal material, and look just at pre-nominal material. The following possibilities arise: (51) a.

(D) N ’eyi sah that bear “that bear” (Rice 1989: 1000) b. (D) PP N ’eyi sahtú gots’e˛h ts’éli˛ the Bearlake ar-from girl “the girl from Bearlake” (Rice 1989: 1000)

As (51) shows, only a postpositional phrase modifier can occur between the determiner and the noun, suggesting the constituency of these two elements. Turning to Navajo, Speas and Yazzie (1996) present arguments that DP is not present. First, following Young and Morgan (1992), they assume that Navajo demonstratives are composed of a locative plus a nominalizing enclitic, suggesting that they are nouns. Second, again following Young and Morgan (1992), they argue that the Navajo demonstratives ‘seem to be independent phrases in and of themselves rather than syntactic determiners’(pp. 74). The demonstratives can function as adjectives or can stand alone as pronouns (pp. 73). This in itself is not an argument for NP, as it could be equally argued for DP. However, Speas and Yazzie further argue that when a demonstrative and noun co-occur, they do not have to be adjacent. An example is given in (52). (52)

díí sh˛í˛í ’ashkii KinSání-di ’óSta’ this probably boy Flagstaff 3.S-imp-go.to.school “This boy probably goes to school in Flagstaff.” (Speas and Yazzie 73)

In this example, the determiner, díí, is separated from the noun that it modifies by material that could not be considered to be within the same constituent, quite unlike the Slave examples in (51).

73

74

Keren Rice

Speas and Yazzie conclude: ‘Demonstratives, Numerals, (in)definiteness markers and the Universals words in Navajo are not Determiner heads which select NP complements. Rather, they are full phrases in and of themselves, and when they occur with another nominal, the relationship is one of restrictive modification rather than complementation’ (1996: 73–74). Thus, setting aside the possessive construction, Slave and Navajo differ in that Slave has DP, QP, and NP, and can support dislocated items in the specifiers of DP and QP. To complete the typology, evidence is found in Navajo for both topicalized and focused positions. An example of each is given here, although the discussion is not developed. I begin with topicalization. (53) topicalized díí ga’ chidí nizhóní this top car 3.S-imp-good “This is the prettiest car!” (Young and Morgan 1987: 369d) Young and Morgan (1987: 369d) say of the word ga’ that it ‘serves to particularize the noun or pronoun to which it is attached.’ Turning now to focused elements, Willie (1991: 207), Willie and Jelinek (2000), and Rice and Saxon (forthcoming) discuss a process called focus movement by Willie (1991: 207). This construction is illustrated in (54b). ashkii ha’át’íísh yiyiiStsá˛? boy what-Q 3.S-pf-see-y.DO “What did the boy see?” (Schauber 1979: 118) b. ha’át’íísh ashkii yiyiiStsá˛? what-Q boy 3-pf-see-y.DO “What did the boy see?” (Schauber 1979: 117)

(54) a.

Willie and Jelinek propose that the word order in (54b) (OSV rather than SOV) puts the question phrase in focus. Focus movement is not subject to the same restrictions as another better-known Navajo process, subject–object inversion; see Rice and Saxon (forthcoming) on the differences between these. Rice and Saxon (forthcoming) provide examples of other sentences with focus movement. The diagnostics for this are word order and the form of inflection; relevant semantic evidence is still lacking. (55) YistSé ’ayóo naats’o˛ od ˛ léi’ shádí shá sock very 3sg-stretch indef 1sg.poss-older.sister 1sg.OO-for nayiisnii’ 3.S-buy-y.DO “My older sister bought me some very stretchy stockings.” (Young and Morgan 1987: 601)


(56) Tsiighá bee yiilch’íSí ’áSchíní yóó’ ’adeiztá˛a˛´ lá curling iron children 3pl.S-carry.off-y.DO evidential “The children had carried off the curling iron” (Young and Morgan 1987: 219) Slave and Navajo are contrasted in (57) in terms of possible positions for adjuncts; I omit comparison of the possessive constructions here. (57) SpecCP SpecFocP SpecQP

Slave yes yes yes

Navajo yes yes no

The languages share in allowing left-edge topicalized and focused constituents. Slave further allows nominals in SpecQP, a position which does not exist in Navajo. In both languages, nominals can be arguments or adjuncts; Navajo does not have positions other than at the left edge of the clause (SpecCP; SpecFocP) in which adjuncts occur. In the development of a Navajo-type language from a Slave-type language, then, demonstratives, numerals, and so on came to be full phrases in and of themselves, as Speas and Yazzie (1996: 73–74) discuss. With this, contrastive positions for nominals within DP were lost. The basic claim of Rice and Saxon (forthcoming) is maintained: the languages differ in terms of agreement triggers, with any NP triggering agreement in Navajo and only pronominals triggering agreement in Slave. In the loss of contrast between positions, Navajo reflects the morphosyntactic patterning of Slave adjuncts, generalizing the triggers for inflectional agreement.14

5.

Conclusions

My primary goal has been to show that a doubling construction exists in Slave that has not previously been reported, and is well-documented in only one other northern Athapaskan language, Babine-Witsuwit’en (Gunlogson 2001). I have framed this discussion in terms of triggering conditions for agreement and possible positions for adjuncts, showing that the idea that agreement occurs only when the noun which it appears to double is an adjunct can be maintained if the phrase structure is enriched. A secondary goal has been to examine some differences between Slave and Navajo. I assume, after Rice and Saxon (forthcoming), that the primary difference is in triggering conditions, with any NP, pronominal or nominal, triggering y-/bagreement in Navajo, and only pronominals triggering it in Slave. Slave further

75

76

Keren Rice

allows nominals to appear in specifier positions within DP, while Navajo simply has no DP or QP. The chart in (58) summarizes the major differences between Slave and Navajo, and speculates on Proto-Athapaskan. ‘Triggering conditions’ refers to whether any nominal or only pronominals trigger agreement in an appropriate syntactic domain; ‘structures’ refers to the highest level of structure, NP or DP. (58) Triggering conditions Navajo any NP Slave pronominal only Proto-Athapaskan pronominal only

Structures

Structures

NP DP DP ?

CP, FocP CP, FocP, DP, QP CP, FocP, DP, QP (?)

The Proto-Athapaskan reconstruction and the direction of change are, of course, speculative, and research is required to determine the properties of Proto-Athapaskan. This article is a modest addition towards meeting the challenge set by Willie (1990: 381) to, as quoted earlier, ‘account for the distinct syntactic developments involving the third-person forms across the language family’. This study, Gunlagson’s (2001) work on Babine-Witsuwit’en, and hints from other languages (see note 12) suggest that the remarkable differences that divide the Apachean group from their northern relatives find their roots in the north, and did not spring up without precedent in Apachean. Even though the languages differ in their structure, these northern languages and Apachean languages have similar surface constructions that give the appearance of doubling.

Notes * It is a pleasure to dedicate this article to Eloise Jelinek. Eloise has been a tremendous influence on me, and it is difficult to find the words to describe all that she has done. Thank you, Eloise. I would also like to thank Leslie Saxon, Andrew Carnie, Heidi Harley, Mary Willie and an anonymous reviewer for comments on an earlier version of this article. 1. I focus on y- and b-. There are conditions under which co-occurrence of a nominal and inflection is favored; see Section 3.3 for brief discussion. 2. Most of the data in this article is from texts. In some cases, the textual data is inadequate to confirm hypotheses, and additional research is called for. Most examples are from South Slave dialects, although some is from Bearlake and Hare dialects. I use the transcription system found in the source. The following abbreviations are used: b- b-inflection, y- y-inflection, 1 first person, 2 second person, 3 third person, ar areal prefix, u unspecified person, sg singular, pl plural, s subject, do direct object, oo oblique object, poss possessor, imp imperfective, pf perfective, opt optative, progr progressive, top topicalized, foc focus, neg negative, comp complementizer, part particle.


3. This is actually more similar to left-dislocation (e.g., Ross 1967). Note that hanging topics occur in Slave; see Rice 1989b: 1207–1208. 4. To get some sense of this, the examples from Thom and Blondin-Townsend (1987) in this section are all of the sentences that I found in this 120 page book (about half of which is photographs; each page of text contains one column in Slave and one in English translation, yielding perhaps thirty pages of text in total) that I feel sure represent the types of doubling under consideration in this section. 5. The structure in (16) is misleading in detail. In particular, only first and second person subjects occur in the highest subject position in Slave, and third person objects are somewhat lower down, and external objects are in functional projections above VP. However, the basic ordering and hierarchy is reflected in (16); see Rice and Saxon (forthcoming) for detailed discussion and other references. 6. It is possible that this example involves a relative clause, as indicated in the translation. However, shu generally functions as a coordinating conjunction in Slave rather than a complementizer, and I follow this latter analysis. 7. In this example, doubling of a nominal by inflection is found within the possessive phrase ’ehts’o mel˛o, (’ehts’o ‘bead b- ‘3 agreement’ l˛o ‘end’) ‘the end of the rosary’ (literally: the beads’ end). See Section 3.2 for discussion of possessive constructions. 8. Slave NPs are right-headed, as can be seen by the fact that non-verbal modifiers are to the left of the head (e.g., Yellowknife gots’e˛ t’eere [go- ‘there’ -ts’e˛ ‘from’ t’eere ‘girl’] ‘girl from Yellowknife’; tu g˛o ke [tu ‘water’ g˛o ‘for’ ke ‘shoe’] ‘rain boots’). 9. This example, (40b), shows that an adjunct and an argument with the same referent can cooccur, providing additional evidence for there being both adjunct and argument positions. 10. As pointed out by Saxon (1989: 401–402), sometimes postpositions may occur without pronominal objects in Navajo. (i) shádí k’eh yáníSti’ 1sg.poss-older sister according to 2sg.S-imp-talk (ii) dziS-ta’-gi mountain-between-at “between the mountains; inter-mountain” The postpositions in these examples and others in Saxon are preceded by NP objects. In the absence of a nominal object, a pronominal object is required. Young and Morgan (1980: 73, 95) propose that some of these examples involve lexical compounds. Willie (1991:44) recognizes a few postpositions that can ‘can function either as grammatical or lexical postpositions’. 11. Recall that Jelinek and Willie analyze all NPs as adjuncts. Although I am assuming that nominals can occur in argument positions in Navajo, perhaps the most important point is that this is clearly another possible path of development from the more conservative Slave — the contrast between adjunct and argument is lost in Navajo. See the literature cited in the introduction on this debate. 12. Gunlogson (2001) provides detailed discussion of the conditions under which y- occurs in a doubling construction in Babine Witsuwit’en. She argues that a nominal and inflection co-occur if the object is definite. Poser (p.c. 2002) suggests that Carrier appears to have a similar pattern to Babine-Witsuwit’en. Hargus (p.c. 2002) also finds some evidence for doubling constructions in Sekani. In particular, third person plural pronominal prefixes co-occur with plural nouns (see Slave examples in (43)) and doubling is usually present with inalienably possessed nouns. Humanness also appears to play a role.

77

78

Keren Rice

13. Specifically, if something fills a head position (D, N, or Q) within DP, inflectional agreement is absent. 14. Jelinek and Willie would argue that all of the NPs are adjuncts. See the sources given in the introduction for discussion. Note also that if Rice and Saxon are correct in their arguments, doubling is not always interpretable as a diagnostic of adjunct status. It remains to be seen if the difference in triggers — pronominals in Slave and any nominal in Navajo — is able to follow from some more basic properties of the languages; I leave this very interesting question for future research.

Quasi objects in St’át’imcets On the (semi-)independence of Agreement and Case* Henry Davis and Lisa Matthewson University of British Columbia

1.

Introduction

Nowhere has the work of Eloise Jelinek had more impact than in the field of Salishan linguistics, and no one has been more responsible than Eloise for bringing the unique syntactic and semantic properties of the Salish language family to the wider attention of theoretical linguists. In a series of influential articles on the Lummi dialect of Straits Salish (Jelinek 1993, 1994, Jelinek and Demers 1994), she has articulated a model of Salish syntax which brings together a series of puzzling and superficially unrelated phenomena into a coherent, over-arching vision organized by a single, simple parameter: the Pronominal Argument Parameter. The effects of this parameter on Straits Salish syntax are succinctly expressed in the following passage (Jelinek and Demers 1994: 698): In languages with exclusively pronominal arguments, only clitics and affixes occupy argument positions. In Straits Salish, lexical roots do not appear independently; they are always inflected for their arguments, and cannot themselves serve as arguments. As a result, any open-class root appears as the lexical head of its own clause.

These claims are controversial, which is all to the good: the extensive debate which has ensued in the twenty or so years since Eloise began her work on Straits Salish has dramatically improved both the quantity and the quality of work on Salish syntax and semantics. It is therefore in a spirit of respect and affection that we have chosen to discuss here a phenomenon in St’át’imcets (Lillooet Salish) which we think cannot easily be assimilated to the Pronominal Argument Hypothesis, as applied to Interior Salish languages. The phenomenon involves DPs which function as arguments but which are not associated with agreement morphology. These non-agreeing objects (henceforth referred to as quasi objects1) are problematic for the PAH, which

80

Henry Davis and Lisa Matthewson

mandates a one-to-one correspondence between argumenthood and agreement. However, their significance extends beyond the PAH: quasi objects are problematic for any approach which attempts to link Case to agreement in a one-to-one fashion, or to reduce one of these notions to the other. We argue that both Case and agreement are active in St’át’imcets, and are at least partially independent of each other. This is a particularly significant finding for Salish, where the morphological exponence of Case is rudimentary (essentially limited to signaling an oblique/nonoblique distinction), whereas agreement morphology is pervasive and inescapable. Our discussion is structured as follows. After an introduction to basic relevant properties of St’át’imcets syntax in Section 2, we introduce the phenomenon of quasi objects in Section 3. These come in two varieties: (i) theme arguments of formally intransitive predicates, and (ii) theme arguments of (formally monotransitive) ditransitive verbs. In Section 4, we investigate the differing syntactic properties of classes of DPs in St’át’imcets. As shown in (1), four types of DP must be distinguished on the basis of morphological agreement, extraction possibilities, and oblique Case marking: (1)

Subjects, themes of transitives, goals of ditransitives Adjuncts Themes of -cit ditransitives Themes of intransitives, themes of -Vn(’) ditransitives

Agreement

Directly Oblique extractable Marking

yes

yes

no

no no no

no yes no

yes no no

We then discuss the consequences of quasi objects for the PAH. In Sections 5 and 6, we consider the possibility of saving the PAH by appealing to the notion of (semantic) incorporation. If all quasi objects in St’át’imcets were incorporated, then they could be analyzed as predicates or predicate modifiers, rather than argumental DPs. In that case, the correlation between argumenthood and agreement required by the PAH could still hold. However, we show that this analysis cannot account for the clearly non-incorporated status of a subset of quasi objects. Furthermore, we argue that any attempt to rescue the PAH for St’át’imcets in this way would only result in a severe weakening of its predictive power. We conclude that the four-way split shown in (1) can only be accounted for if Case and agreement are both operative in the language, and moreover operate independently of each other. We propose that the ability of a DP to be directly extracted correlates with structural Case, and that a subset of quasi objects have

Quasi objects in St’át’imcets

inherent Case, in the sense of Chomsky (1981, 1986b). We observe that the semantic correlates of inherent Case are diverse (see e.g. Enç 1991, de Hoop 1992, van Geenhoven 1996, Kiparsky 1998). This argues for the autonomy of the Case system as a component of the syntax, rather than as a reflex of any particular semantic operation (such as incorporation). In Section 7, we point out that the relationship between Case and agreement is not completely free. There is a one-way implicational relation between the two notions, such that if a DP co-occurs with agreement, it must have structural Case. We conclude with some tentative remarks on the implications of this finding for contemporary theories of Case and agreement. Let us then begin by providing some brief background on transitivity and agreement in St’át’imcets.

2.

Transitivity and Agreement in St’át’imcets

St’át’imcets is a Northern Interior Salish language spoken in Southwestern British Columbia. Like all other Salish languages, it distinguishes formally transitive from intransitive predicates by special sets of transitivizing and intransitivizing suffixes: only transitivized predicates may occur with object agreement morphology. There are four principal transitivizers in St’át’imcets. We will however limit our discussion in this paper to two of them: the directive or ‘full control’ transitivizer, which has the form -Vn(’)/-Vns (V is generally either schwa or a copy of the root vowel); and the indirective transitivizer -cit. All transitivized predicates are suffixed for the person and number of their object. With directive transitivizers, object suffixes usually relate to the patient argument of the predicate, though there is a significant subset of directives where object suffixes refer to a recipient or goal instead. The object suffix of a predicate transitivized with -cit normally refers to a recipient or goal rather than to a patient, as shown in (2–3) with first and second person objects. The (a) cases contain roots transitivized with the directive transitivizer, the (b) cases the same roots transitivized with the indirective transitivizer. Note that the presence of an independent (emphatic) pronoun is irrelevant to the grammaticality of these examples. tsulh-un’-tsí-lhkan (snúwa)2 point-dir-2sg.obj-1sg.subj (you.sg) “I pointed at you.” b. tsulh-ci-tsí-lhkan (snúwa)3 point-ind-2sg.obj-1sg.subj (you.sg) “I pointed it out to you.”

(2) a.

81

82


(3) a.

legw-en-ts-ás (s7éntsa) ta kúkwpi7-a hide-dir-1sg.obj-3erg (I) det chief-det “The chief hid me.” b. legw-cí-ts-as (s7éntsa) ta kúkwpi7-a hide-dir-1sg.obj-3erg (I) det chief-det “The chief hid it for me/from me.”

Note that while only one object is ever registered by agreement morphology, another is implicitly present in the ditransitive predicates in (2b) and (3b), as evidenced by their English translations. We will return to the significance of these ‘hidden objects’ in Section 3. Aside from transitivizing suffixes, all Salish languages also possess a comparable set of intransitivizing suffixes. There are three main intransitivizers in St’át’imcets, as well as a heterogeneous class of unaffixed intransitive predicates. We will be concerned here with the subset of intransitive predicates which are ‘object-oriented’; these alternate productively with directive transitives and denote activities. They are suffixed either with the active intransitivizer -cal, the middle intransitivizer -Vm, or a Ø-suffixed alternate of the middle. See Davis (1997) for justification of the predicate classification assumed here.4 Examples of these three types of intransitives are given in (4a)–(6a), together with their (directive) transitive counterparts in (4b)–(6b). (4) a.

k’ác-cal-lhkan dry-act-1sg.subj “I dried something, did some drying.” b. k’ác-an’-Ø-lhkan dry-dir-3obj-1sg.subj “I dried it.”

(5) a.

qús-em-lhkan shoot-mid-1sg.subj “I shot at something, did some shooting.” b. qús-en-Ø-lhkan see-dir-3obj-1sg.subj “I shot at him/her/it.”

(6) a.

náq’w(-Ø)-lhkan steal(-Ømid)-1sg.subj “I stole something, did some stealing.” b. náq’w-ens-Ø-kan see-dir-3obj-1sg.subj “I stole it.”

It is ungrammatical to affix any formally intransitive predicate with an object


marker. This means that first and second person objects never occur with intransitives. This is shown in (7) and (8), which contain the same roots as those in (2) and (3), but this time suffixed with intransitivizers. The (a) cases are grammatical examples without object marking; the (b) cases are ungrammatical examples with object suffixes. Again, the presence of an independent pronoun makes no difference to the (un)grammaticality of these examples. (7) a.

tsúlh-cal-lhkan point-act-1sg.subj “I pointed.” b. *tsulh-cal-tsí-lhkan (snúwa) point-act-2sg.obj-1sg.subj (you.sg) leg-úm-lhkan5 hide-act-1sg.subj “I hid (stuff).” b. *leg-úm-ts-as kelh (s7éntsa) hire-mid-1sg.obj-3erg will (I) ta kúkwpi7-a det chief-det

(8) a.

Since third person object agreement is (usually) not overtly marked,6 it is less easy to show that third person object agreement is obligatorily present with a transitivizer, and obligatorily absent without one. However, the interpretation of the null third person object of a transitivized predicate is invariably that of a definite pronoun. This is shown by the discourse in (9): (9a) sets up a discourse antecedent (the coyote) which can be accessed as the object of the (directive) transitive in (9b). (9) a.

wa7 alkst i ucwalmícw-a l-ti lep’cáltn-a prog work det.pl person-det in-det garden-det “Some people were working in the garden.” t’ak káti7 ti nk’yáp-a go deic det coyote-det “A coyote went by.” b. nilh s-qus-en-Ø-ítas pro foc nom-shoot-dir-3obj-3pl.erg pro “So they shot it.”

We assume that the definite pronominal interpretation of the object in (9b) indicates the presence of pro. If we further assume that pro must be morphologically identified (Rizzi 1986a), it follows that a Ø third person object marker must be present in null contexts with transitivized predicates, but not with intransitives. It is more difficult to tell whether Ø object marking is retained when an overt object DP is present. We will assume here that it is, chiefly on the basis of evidence

83

84


from Roberts (1999). Roberts demonstrates the existence of an ‘anti-agreement’ effect with È-extraction in St’át’imcets, whereby certain plural markers which are optionally present (though redundant) when a plural object DP is present become ungrammatical when the object is relativized, focused, or questioned. For an antiagreement effect to exist in special contexts, agreement must exist in the general case: hence, we are led once more to postulate a Ø third person object marker, irrespective of the presence of an overt object DP. So far, then, the generalization is simple: formally transitive predicates always occur with object agreement suffixes (including Ø-suffixes); formally intransitive predicates never do.

3.

Quasi objects

We now come to the main focus of this paper. Surprisingly, in St’át’imcets object DPs may surface quite freely in contexts where they are licensed neither by transitivizers nor by agreement morphology. There are two such contexts: with formally intransitive ‘object-oriented’ predicates; and with formally (mono)transitive ditransitive predicates. We will refer to the former as ‘intransitive objects’,7 and to the latter as ‘(ditransitive) second objects’. Together, these constitute the class of quasi objects. We will start with intransitive objects. Representative examples are given below. (10) mets-cál wi (*e-)ku szenk draw-act pl.imp (*obl-)det circle “Draw a circle, you folks!” (11) wa7 k’úl’-em (*e-)ta tsepalín-a prog make-mid (*obl-)det baby.basket-det ta n-skícez7-a det 1sg.poss-mother-det “My mother is making a baby basket.” (12) náq’w-kan (*ek-)i metslák7-a steal-1sg.subj (*obl-)det.pl pen-det “I steal pens.” While intransitive object constructions exist in other Salish languages, they are generally introduced by an oblique proclitic preposition (typically schwa in Central Salish, and t in the Interior).8 However, in St’át’imcets the oblique marker has disappeared. This is not (at least synchronically) the result of phonological reduction of the oblique preposition (as is commonly found across Salish): In St’át’imcets it is actually ungrammatical to use an oblique preposition with an intransitive


object, as you can see in (10–12). In contrast, while passive agents freely appear with or without an oblique preposition, it may always be restored in careful speech: (13) núk’w7-an-Ø-em ta n-snúk’w7-a help-dir-3obj-pass det 1sg.poss-friend-det (e-)ta sqátsza7-s-a obl-det father-3poss-det “My friend was helped by his father.” We now turn to the second objects of ditransitive predicates. Examples with -cit are given in (14–16). Here, the primary objects (goals) are marked on the predicate by a pronominal suffix, while the second objects (themes) appear as overt DPs. Like intransitive objects, second objects of ditransitives are unregistered by any agreement morphology on the predicate; like intransitive objects, they are always themes; and like intransitive objects, they may not appear with an oblique marker. (Compare the ditransitive object in (16), which may not be oblique marked, with the passive agent in the same example, which may.) (14) t’íq-ci-ts-as ta n-sqátsez7-a arrive-ind-1sg.obj-3erg det 1sg.poss-father-det (*e-)ku sts’wan (*obl-)det dried.salmon “My father brought me some dried salmon.” (15) q’wel-ci-tumúlh-kan kelh (*e-)ta roast-ind-2pl.obj-1sg.subj will (*obl-)det zúmak-a spring.salmon-det “I’ll cook a spring salmon for you folks.” (16) k’ul’-ci-tánemwit (*e-)ki make-ind-3pl.pass.obj (*obl-)coll.det silhts’a7úl-a (e-)ta skicez7-í-ha moccasin-det (obl-)det mother-3pl.poss-det “Moccasins were made for them by their mother.” In the data we have seen so far, intransitive objects and ditransitive second objects behave identically. However, their interpretive possibilities differ. Recall that the null third person object of a transitivized predicate is interpreted as a definite pronoun. The same is true of ditransitive second objects, as shown in (17). (17) a.

wá7-lhkalh alkst l-ta lep’cáltn-a prog-1pl.subj work in-det garden-det “We were working in the garden.”

85

86


t’ak mám’teq káti7 ta nk’yáp-a go walk deic det coyote-det “A coyote went by.” b. tsúlh-cit-Ø-kan pro aylh ta n-sem7ám-a point-ind-3obj-1pl.sub pro then det 1sg.poss-wife-det “I pointed it out to my wife.” Here, even though only the goal (ta nsem7áma) is registered as a (zero) affix on the predicate tsúlhcitkan, a discourse antecedent (the coyote) can still be accessed by the predicate’s unregistered ditransitive object (the theme). Since it has a definite pronominal reference, we assume that the null ditransitive object in (17b) is pro. However, this interpretation is systematically unavailable for intransitive objects, as shown in (18). (18) a.

wa7 alkst i ucwalmícw-a l-ti lep’cáltn-a prog work det.pl person-det in-det garden-det “Some people were working in the garden.” t’ak káti7 ti nk’yáp-a go deic det coyote-det “A coyote went by.” b. !nilh s-qús-em-wit pro foc nom-shoot-mid-3pl pro “They shot.” Consultant A’s comment: “They could be just shooting anywhere, not at the coyote.” Consultant B’s comment: (laughs) “It’s okay, but it means you shot up in the air.”

The definite pronominal interpretation is absent with intransitive predicates even when the predicate, as a lexical property, strongly entails the existence of some theme argument. This is shown in (19)–(21). (19) tsicw áta7 stóh-a kw-s Mary went deic store-det det-nom Mary “Mary went to the store.” wa7 lak láta7 ta leqwáz’-a prog lie deic det dress/blanket-det “There was a blanket or a dress there.” nilh s-tecwp-s foc nom-buy-3sg.poss “She bought something.” Consultant’s comment: “She could have just bought anything.” Did she buy something? “Yes, but not necessarily the blanket.”


(20) wa7 i wa7 záw-em be det.pl prog fish-mid “There were some people fishing.” t’ak káti7 ta xzúm-a zúmak go deic det big-det spring.salmon “A big spring salmon went by.” ??nilh s-7ú7stek-wit foc nom-catch.fish-3pl “So they caught something.” Consultant’s comment: “They might have caught anything like a sturgeon, not necessarily the zúmak.” (21) !kwán-ens-Ø-kan ta xzúm-a ts’úqwaz’, catch-dir-3obj-1sg.subj det big-det fish nilh´ t’u7 n-s-k’ác-cal foc just 1sg.poss-nom-dry-act “I caught a big fish, and then I dried stuff.” NOT “I caught a big fish, and then I dried it.” (transitive required) So far we have seen that in contrast to intransitive objects, the second objects of ditransitive predicates may be null, in which case they have a definite (pronominal) interpretation (pro). In fact, ditransitive second objects may also surface overtly as demonstrative pronouns, again unlike intransitive objects. Compare (22) with (23). (22) tsulh-ci-tani-lhkán ti7 point-ind-3pl.obj-1sg.subj that “I pointed that out to them.” (23) *tsulh-cal-lhkán ti7 point-act-1sg.subj that In sum, quasi objects fall into two types, intransitive objects and ditransitive second objects. Neither are registered on the predicate by agreement morphology; both bear the thematic role of theme or patient; both may surface as overt DPs; and neither may be marked as oblique. Ditransitive second objects but not intransitive objects may surface as demonstrative pronouns or as pro. In the next sections we will build on these facts, exploring the syntactic and semantic properties of quasi objects in more detail.

4. The syntax of quasi objects In this section, we will further investigate the syntax of quasi objects, comparing them to agreement-linked transitive objects on the one hand and true adjuncts on

87

88


the other. We will show that quasi objects display identical distribution and word order behavior to ordinary objects (modulo one independently explained difference).9 However, quasi objects and ordinary objects differ in their extraction possibilities. We will begin with the distribution of overt DPs. Like other Salish languages, St’át’imcets is essentially predicate initial. Subjects may precede the main predicate under certain circumstances, but objects never do (Davis 1999). Post-predicatively, the distribution of DPs is constrained by what has become known, following Gerdts (1988), as The One Nominal Interpretation Constraint (ONI): in an active transitive sentence with two third person arguments, a single overt DP is always interpreted as the object rather than the subject, as shown in (24). (24) xlít-en-as ta kúkwpi7-a invite-dir-3erg det chief-det “S/he called/invited the chief.” *“The chief called/invited him/her.” The ONI follows from the regulation of information structure in discourse. Familiar discourse referents in St’át’imcets are represented by pro, and pro is preferentially mapped onto the subject position. In a transitive sentence with a single overt DP like (24), the subject is always represented by pro and the overt DP is consequently automatically associated with the object position (see Davis 1994). Let us turn to the application of the ONI to quasi object DPs. Intransitive objects are immune to ONI effects: in object-oriented intransitive clauses, a single overt DP may be interpreted either as the subject or the intransitive object, depending on which is pragmatically appropriate. This is shown in (25). (25) a.

wa7 k’ác-cal kw s-Lémya7 prog dry-act det nom-Lémya7 “Lémya7 is drying (something).” b. wa7 k’ác-cal ki stsáqwem-a prog dry-act det saskatoon-det “She is drying saskatoon berries.”

This follows from our claim above that there is never a pro in intransitive object position; in (25a), there is no illegitimate object pro which would render the sentence unacceptable. Ditransitive second objects behave differently: they pattern with transitive objects in satisfying the ONI. This can be seen with the -cit marked ditransitive predicate in (26), where a single DP can either be interpreted as the goal (the primary object) or the theme (the second object), but not as the subject.10


(26) cwíts’-cit-as ta skwékwez7-a hand-ind-3erg det child-det i. “She handed the child to her.” ii. “She handed it to the child.” iii. *“The child handed it to her.” These facts also accord with our analysis so far. In each of (26i) and (26ii) there is an object pro, but since there is also a subject pro, the sentences are correctly predicted to be acceptable (see Davis 1994). Let us turn now to word order facts. In sentences with both a subject and an object DP, there is a dialect split in word order: Upper St’át’imcets (US) has unmarked VOS order, whereas Lower St’át’imcets (LS) prefers VSO. However, the alternate order is easy to elicit in both dialects as long as no ambiguity arises. This is illustrated in (27), from Davis (1999). (27) a.

k’ác-an’-as i stsáqwem-a kw s-Lémya7 dry-dir-3erg det.pl saskatoon-det det nom-Lémya7 “Lémya7 dried some saskatoon berries.” (normal word order in US; possible in LS) b. k’ác-an’-as kw s-Lémya7 i stsáqwem-a dry-dir-3erg det nom-Lémya7 det.pl saskatoon-det “Lémya7 dried some saskatoon berries.” (normal word order in LS; possible in US)

Post-predicative adjunct PPs (including passive agents) are likewise freely ordered with respect to each other and to argument DPs (28). (28) a.

núk’w7-an-Ø-em ta n-snúk’w7-a help-dir-3obj-pass det 1sg.poss-friend-det (e-)ta naplít-a obl-det priest-det “My friend was helped by the priest.” b. núk’w7-an-Ø-em (e-)ta naplít-a help-dir-3obj-pass (obl-)det priest-det ta n-snúk’w7-a det 1sg.poss-friend-det “My friend was helped by the priest.”

Both intransitive and ditransitive second objects behave in exactly the same way as transitive objects. This means that in Upper St’át’imcets, both types of quasi object usually occur immediately after the predicate, preceding the subject, while in Lower St’át’imcets, both types of quasi object usually occur after both the subject and the predicate. And in both dialects, subjects and quasi objects (either ditransitive or intransitive) may exchange positions (i.e., VSO is possible in Upper St’át’imcets and

89

90


VOS is possible in Lower St’át’imcets). This is shown in (29) for intransitive objects and in (30) for ditransitive second objects. (29) a.

cuz’ k’ác-cal ku stsáqwem kw going.to dry-act det saskatoon det s-Lémya7 nom-Lémya7 “Lémya7 is going to dry saskatoons.” (normal word order in US; possible in LS) b. cuz’ k’ác-cal kw s-Lémya7 ku going.to dry-act det nom-Lémya7 det stsáqwem saskatoon “Lémya7 is going to dry saskatoons.” (normal word order in LS; possible in US)

(30) a.

cwíts’-ci-ts-as ta skwékwez7-a ta hand-ind-1sg.obj-3erg det child-det det sésq’wez’-s-a younger.sibling-3poss-det “Her younger sister handed me the child.” (normal word order in US; possible in LS) b. cwíts’-ci-ts-as ta hand-ind-1sg.obj-3erg det sésq’wez’-s-a ta skwékwez7-a younger.sibling-3poss-det det child-det “Her younger sister handed me the child.” (normal word order in LS; possible in US)

The relative order of the primary and secondary objects of ditransitive predicates is free: (31) cwíts’-cit-as ta skwékwez7-a ta hand-ind-3erg det child-det det sésq’wez’-s-a younger.sibling-3poss-det i. “She handed the child to her younger sister.” ii. “She handed her younger sister to the child.” In terms of the distribution of overt DPs, quasi objects behave just like transitive objects. The single observed difference (the failure of intransitive objects to be subject to the ONI) is not a difference in the behavior of overt DPs, but rather follows directly from the absence of pro in intransitive object position.11


We now turn to extraction data. Here, we do find differences between transitive objects and quasi objects. However, we also find differences within the class of quasi objects. Transitive objects may be extracted directly by È-movement (including relative clauses, clefts, and wh-questions). This shown in the wh-questions in (32–33). (32) stam’ ku náq’w-ens-acw what det steal-dir-2sg.erg “What did you steal?” (33) swat ku xlít-en-as kw-s Linda what det call-dir-3erg det-nom Linda “Who did Linda call/invite?” Comparable extraction of intransitive objects, however, always triggers nominalization of the predicate.12 (34) stam’ ku s-náq’w-su what det nom-steal-2sg.poss “What did you steal?” (Literally: “What is your stealing?”) (35) swat ku s-xlít-cal-s s-Linda who det nom-invite-act-3poss nom-Linda Who did Linda call/invite?’ (Literally: ‘Who is Linda’s calling/inviting?’) In contrast, ditransitive second objects do not behave uniformly when extracted. Those marked with the indirective transitivizer -cit allow direct extraction of both goal and theme arguments, as in (36a–b).13 However, those marked with the directive transitivizer -Vn only allow goal arguments to be extracted directly (37a); extraction of the theme of these predicates triggers nominalization of the predicate, in the same way as with intransitive objects (37b). (36) a.

swat ku cuz’ hal’a-cít-al’ap ta who det going.to show-ind-2pl.erg det memew’-láp-a kitten-2pl.poss-det “Who are you going to show your kitten to?” b. stam’ ku cuz’ hal’a-cít-al’ap i what det going.to show-ind-2pl.erg det.pl slalil’tem-láp-a parent-2pl.poss-det “What are you going to show to your parents?”

91

92


(37) a.

swat ku cuz’ um’-en-tánemwit tákem who det going.to give-dir-3pl.pass all izá ku tsális14 these det cherries “Who are we going to give all these cherries to?” b. stam’ ku cuz’ s-7um’-en-tánemwit tákem what det going.to nom-give-dir-3pl.pass all izá ku úcwalmicw these det people “What are we going to give all these people?”

The most straightforward explanation for the distribution of nominalization in object extraction contexts is that it takes place just in case an extracted object is not formally licensed by transitivizing morphology. This is particularly clear with ditransitives, where the indirective transitivizer -cit licenses two extractable objects, whereas the directive transitivizer licenses only one (the goal), even though thematic relations are identical in the two cases, and both transitivizers license a single object agreement.15 It is important to emphasize that the extraction behavior of quasi objects does not correspond to that of any class of adjuncts. There are a variety of adjunct extraction strategies in St’át’imcets, depending on the type of adjunct. With passive agents, extraction takes place directly (38). With temporal and locative adjuncts, extraction involves a quite different construction involving one of the proclitic complementizers lh and i together with conjunctive subject morphology (39–40). (38) swat ku tsew’-en-ém ti k’ét’h-a who det kick-dir-pass det rock-det “Who kicked the rock?” (Literally: “Who was the rock kicked by?”) (39) lhel-nká7 lh-náq’w-ens-acw i from-where comp-steal-tr-2sg.conj det.pl sts’wán-a dried.salmon-det “Where did you steal the dried salmon from?” (40) ikanemasas i-náq’w-ens-acw i when(past) comp-steal-tr-2sg.conj det.pl sts’wán-a dried.salmon-det “When did you steal the dried salmon?” With manner and reason adverbials, the clause from which extraction takes place is nominalized (with or without a preceding determiner, depending on the type of


adjunct), but the nominalization which is employed differs from that used with quasi objects: rather than just the predicate, the whole clause is nominalized, as you can see in the contrast between (43) and (44). In (43), with an extracted manner adjunct, nominalizing morphology (consisting of the nominalizer and associated possessive subject inflection) attaches to a preceding auxiliary, indicating it has scope over the whole clause, but in (44), with an extracted quasi object, nominalizing morphology always remains attached to the main predicate, indicating it has scope only over the predicate itself. See Kroeber (1999: 310) for further evidence for this distinction, found in many Salish languages. (41) skas kw s-náq’w-ens-acw i how det nom-steal-tr-2sg.erg det.pl sts’wán-a dried.salmon-det “How did you steal the dried salmon?” (42) kanem s-náq’w-ens-acw i how.come nom-steal-tr-2sg.erg det.pl sts’wán-a dried.salmon-det “Why did you steal the dried salmon?” (43) skas kw s-cúz’-su náq’w i how det nom-going.to-2sg.poss steal det.pl sts’wán-a dried.salmon-det “How are you going to steal the dried salmon?” (44) stam’ ku cuz’ s-náq’w-su what det going.to nom-steal-2sg.poss “What are you going to steal?” È-extraction data provides evidence for three classes of object: those which are registered by transitivizing and agreement morphology (transitive objects); those licensed by a transitivizer but not agreement morphology (ditransitive second objects with -cit); and those licensed by neither (intransitive objects, and ditransitive objects with a directive transitivizer). None of the three classes behaves like any type of adjunct. The facts are summarized in (45). (45) Extraction Strategies in St’át’imcets Type of extracted constituent direct (agreement marked) object second object of -cit-marked ditransitive passive agent second object of directive ditransitive

Extraction strategy direct direct direct predicate nominalization

93

94


intransitive object manner and reason adjunct locative and temporal adjunct

predicate nominalization clausal nominalization lh-/i- and conjunctive morphology

Putting together the extraction data with what we know about the distribution of object DPs, the following picture emerges. Though there are apparently no differences in terms of distribution or linear order between transitive objects and quasi objects, there are differences in extraction possibilities. These differences are not reducible to either theta-marking or to licensing by agreement. On the other hand, both types of objects behave quite differently from adjuncts.16 In the rest of this paper, we investigate the consequences of these generalizations and propose an account of them. Let us first make the obvious point: the existence of morphologically unregistered objects poses intractable problems for the hypothesis that all arguments must be registered on a predicate by pronominal agreement morphology. Unregistered objects in St’át’imcets are internal arguments, in the sense that they are lexically selected by the predicate; they are direct rather than oblique arguments, since they cannot receive oblique marking; and as far as we can tell they behave exactly like direct objects syntactically, at least in terms of word order (recall that evidence about c-command is difficult to come by). However, they are not marked on the predicate by agreement morphology: therefore, they cannot be pronominal arguments. (See Austin and Bresnan 1998 for similar conclusions on unregistered objects in Warlpiri.) This argument against a PAH analysis of Interior Salish has been questioned by an anonymous reviewer, who asks about the possibility of an incorporation analysis. If quasi objects could be assimilated to a class of ‘incorporating objects’ along the lines of the analyses of anti-passive objects in West Greenlandic proposed by Bittner (1994) and van Geenhoven (1996), then the correlation between argumenthood and agreement morphology could perhaps be maintained. (See also Baker, this volume and Hale, this volume.) In the next two sections we shall provide empirical and conceptual arguments against this move.

5.

Some quasi objects are incorporated

An incorporation analysis of St’át’imcets quasi objects is not without precedent in the literature; see Matthewson (1998), Werle (2000). In fact, we will argue here that a subset of intransitive objects are (semantically) incorporated. However, the remainder of the intransitive objects, as well as all ditransitive second objects, are not. The challenge to a PAH analysis of St’át’imcets therefore remains.


The first piece of evidence for incorporation comes from determiner distribution. St’át’imcets possesses a total of eight determiners, shown in (46). The major division is between ku, a polarity determiner, and all the other determiners, which show plurality and deictic distinctions. See van Eijk (1997), Matthewson (1998, 1999) for discussion of the St’át’imcets determiner system. (46)

singular plural collective

present

absent

remote

ti…a i…a

ni…a nelh…a ki…a

ku…a kwelh…a

ku

The polarity determiner ku has a restricted distribution: it must normally appear inside the c-command domain of an operator such as negation, a question morpheme, or a modal. It is ungrammatical to use ku in a simple declarative sentence, as shown in (47). Where ku does appear, it displays obligatory narrow-scope effects, as shown in (48). (47) *áts’x-en-Ø-as ku sqáycw see-dir-3obj-3erg det man “S/he saw a man.” (48) a.

negation: cw7aoz kw-s áts’x-en-Ø-as ku sqaycw neg det-nom see-dir-3obj-3erg det man “S/he didn’t see any men.” not “There was a man who s/he didn’t see.” b. yes/no question: ats’x-en-Ø-lhkácw ha ku sqaycw see-dir-3obj-2sg.subj ynq det man “Did you see a man/any men?” not “There is a man, did you see him?” c. modal: ats’x-en-Ø-ás k’a ku sqaycw see-dir-3obj-3erg appar det man “There must have been a man who she saw.” not “There was a man who she must have seen.”

Aside from polarity contexts, there is one other place where ku consistently shows up: on intransitive objects. In the examples in (49) (taken from Davis in prep., Chapter 29), the presence of ku is licensed by a non-factual operator, as we would expect.

95

96


(49) a.

future: nás-kan púlh-cal ku qu7 go-1sg.subj boil-act det water “I’m going to boil some water.” b. yes-no question: t’ak-cal-lhkácw ha ku st’ák’el’ go-act-2sg.subj ynq det packed.food “Did you bring lunch along?”

However, unlike agreeing objects, intransitive objects are frequently introduced by ku even in environments lacking a polarity licenser. Examples are provided in (50). (50) a.

ít’-em-lhkan ku s-t’ílim sing-mid-1sg.subj det nom-love.song “I sang a love song.” b. tecwp-kán ku s-7úqwa7 / pukw / t’ec buy-1sg.subj det nom-drink / book / sweet “I bought some drink/a book/some candy.” c. k’ác-cal-lhkan ku sq’wel / sts’úqwaz’ dry-act-1sg.subj det berry / fish “I dried some berries/fish.”

The fact that ku is permitted on the quasi objects in (50) poses a challenge for the analysis of determiners outlined above, since there is no higher licensing operator such as negation, future, a modal or a yes-no question particle. However, the problem disappears if we assume the ku objects in (49–50) are semantically (though not syntactically) incorporated into the predicate, in the sense of van Geenhoven (1996).17 What this means more precisely is as follows. First, we assume that ku itself is semantically vacuous.18 Thus, the interpretation of the ku-phrase is as an NP predicate (we are ignoring intensionality for simplicity’s sake). The verb introduces an existential quantification, as shown in (51). (51) a. [[ tecwp ]] = λP.λx.$y [buy (y) (x) & P (y)] b. [[ tecwp ku pukw kws Mary ]] = $y [buy (y) (Mary) & book (y)] The claim that ku intransitive objects are incorporated in this manner makes various predictions: (52) i. There will be existential entailments in simple sentences. ii. The existential quantifier takes narrow scope with respect to other operators. iii. Discourse referents will be introduced which can be referred back to in subsequent sentences (see van Geenhoven 1996, 1998).


All of these predictions are correct. For simple sentences, existential entailments ensue; all the sentences in (50) entail the existence of an individual satisfying the object predicate. For example, (50c) cannot be true without there existing some berries or fish. The existential quantification introduced by the incorporating predicate does indeed take narrow scope with respect to other operators in the sentence. This is shown in (52). (52c) is an example of a distributive reading, in which the ku object takes narrow scope with respect to the subject quantified phrase. (52) a.

cw7aoz kw-en-s tsew’-cál ku sqéqy’ecw neg det-1sg.poss-nom kick-act det boy “I don’t kick boys.” not “There is a boy who I don’t kick.” Consultant’s comment: “If there was one boy you did kick, you’d be lying.” b. lh-cúz’-acw múta7 tsew’-cál irr-going.to-2sg.conj again kick-act ku smém’lhats, sek-en-tsín-lhkan kelh det girl beat-dir-2sg.obj-1sg.subj will “If you kick a girl again, I’ll spank you.” not “There is a girl, and if you kick her I’ll spank you.” Consultant’s comment: “If he just kicks one, he still gets spanked.” c. zí7zeg’ i tsícw-a píx-em’ kwámem ku míxalh each det.pl go-det hunt-mid catch det bear “Each of the hunters caught a bear.” (one each)

Finally, just as in Greenlandic, subsequent reference to a discourse referent is possible, as in (53). (53) tecwp-kán ku pukw buy-1sg.subj det book “I bought a booki.” stexw t’u7 áma very just good “Iti was very good.” Based on these facts, we conclude that the quasi objects containing ku are semantically incorporated.19 Summarizing the results of this section, we have argued that when introduced by ku, both intransitive objects and the second objects of ditransitives are semantically incorporated into the predicate. This correctly predicts a range of facts about the interpretation of such objects.20 Now we will turn to the question of whether all quasi objects are incorporated.

97

98


6. Not all quasi objects are incorporated Intransitive objects do not always contain ku; non-polarity DPs are also possible in these positions. Some examples are given in (54) (we are concentrating for simplicity on the singular, present determiner ti…a; plural and absent determiners behave the same in relevant respects).21 (54) a.

máyt-kan ti q’láxan-a make-1sg.subj det fence-det “I made a fence.” b. zúqw-cal ti ts’í7-a k John die-act det deer-det det John “John killed a deer.” c. az’ kw-s Charlie ti káoh-a buy det-nom Charlie det car-det “Charlie bought a car.”

Given the semantics of determiners outlined above, it would be unexpected if semantic incorporation were taking place here. The analysis of the non-polarity determiners proposed by Matthewson (1999), namely that they introduce choice function variables which are existentially closed with widest scope, is not at all compatible with incorporation. In addition, the fact that the non-polarity determiners encode plurality and deictic distinctions which are not neutralized in quasi object positions speaks against an incorporation analysis: kwámem k John ti / i ts’í7-a lhkúnsa catch det John det / det.pl deer-det today “John caught a deer/some deer today.” b. púpen’-lhkan na kwíkws-a pú7y’acw find-1sg.subj det.absent small-det mouse “I found a small mouse (not visible at time of utterance).”

(55) a.

The semantic contrasts illustrated in (54–55) mean that it would not be straightforward to analyze these determiners as semantically vacuous, as we did for incorporated ku.22 Second, recall that the incorporation analysis requires narrow scope for the incorporated objects. ti…a intransitive objects, unlike ku objects, may take wide scope; some examples are given in (56). (56) a.

tákem i sqáyqeycw-a áz’-wit ti púkw-a all det.pl man-det buy-3pl det book-det “All the men bought a book.” (one token)


stexw t’u7 áma very just good “It was really good.” b. lh-cúz’-acw tsew’-cál ti irr-going.to-2sg.conj kick-act det syéy’qts7-a, sek-en-tsín-lhkan kelh girl-det beat-dir-2sg.obj-1sg.subj will “There is a girl, and if you kick her, I’ll spank you.” Accepted in context: There are four girls you don’t care about, and only one girl who you will punish him about. In this context, you can elaborate by adding: …nilh ti skwékwza7-s-a ti …foc det offspring-3sg.poss-det det kúkwpi7-a wa7 qwal’út-min-an chief-det prog talk-appl-1sg.conj …“It’s the daughter of the chief who I’m talking about.” On the basis of these data, we conclude that ti…a intransitive objects, unlike ku intransitive objects, are not semantically incorporated.23 Ditransitive second objects are also not semantically incorporated. The first piece of evidence against an incorporation analysis was already given above in Section 3, where we showed that ditransitive second objects can have definite reference, and in the absence of an overt DP, are realized by pro. In addition, the two pieces of evidence discussed in this section also apply to ditransitive quasi objects. As is shown in (57–58), ditransitive quasi objects can take the full range of deictic and plurality distinctions on their determiners, and they can take wide scope. Finally, as shown in (59), ditransitive second objects can be quantificational. Taken together, these data conclusively rule out an incorporation analysis. (57) plurality/deictic distinctions: a. cuz’ cwíts’-ci-ts-as i going.to give-ind-1sg.obj-3erg det.pl sqwelólao7-cen-s-a ti ts’í7-a deer.hoof-foot-3sg.poss-det det deer-det “He’s going to give me the deer hoofs.” b. táw-en-lhkan kw-s John na sell-dir-1sg.subj det-nom John det.absent cín’-a n-kaoh old-det 1sg.poss-car-det “I sold John my old car.”

99

100 Henry Davis and Lisa Matthewson

(58) wide scope: p’án’t-cit-kan i nkekalhás-a skícza7 return-ind-1sg.subj det.pl three(human)-det mother ta pél’p-a skwékwez7-i det lost-det offspring-3pl.poss “I brought the three mothers the child that they lost.” Consultant’s comment: “Sounds like the three mothers have one child (laughs).” (59) quantificational second objects: nás-cit-kan i sk’wemk’úk’wm’it-a takem go-ind-1sg.subj det.pl children-det all i sáy’si7-ten-a det.pl play-instr-det “I gave the children all their toys.” We have shown that while some quasi objects (namely ku objects of intransitives) are semantically incorporated, other intransitive objects and ditransitive second objects are not. As pointed out above, this situation poses problems for a PAH analysis for St’at’imcets, since according to this hypothesis, the only possibilities for noun phrases is that they should be (a) agreement-linked, (b) true adjuncts, or (c) incorporated. We have shown that a subset of quasi objects do not fall into any of these three categories.

7.

Case, Agreement and Quasi objects

The data from extraction raise a more subtle — and perhaps more interesting — point regarding the licensing of overt DPs. Rather than a bifurcation between agreeing and non agreeing argument DPs, we find a three way split between agreeing DPs, non-agreeing but directly extractable DPs, and non-agreeing, not directly extractable DPs. This clearly indicates that more than one licensing mechanism is at work: not only do we need agreement, but also some other mechanism, lexically tied to the choice of transitivizer (recall that -cit licenses a directly extractable ditransitive second object, whilst the directive transitivizer does not). Moreover, we need to distinguish all three of these cases from that of the various classes of true adjuncts, which are overtly oblique-marked and employ different extraction strategies. The obvious candidate for this second licensing mechanism is Case. We already need Case in St’át’imcets to distinguish oblique-marked from non-oblique DPs, so a further extension is not implausible. The additional distinction we need to make is between directly extracted objects and those which trigger nominalization — a


distinction which we would like to suggest is best captured by the familiar opposition between structural Case and inherent Case (Chomsky 1981, 1986). To be precise, we propose that the subset of quasi objects which trigger nominalization are inherently Case-marked, whereas those which directly extract have structural Case. We further assume that in St’át’imcets inherently Case-marked objects may not be extracted directly, but must first be ‘promoted’ to a structural Case position, from which extraction may take place. The promotion operation involves predicate nominalization, which takes the theme argument of an intransitive or ditransitive verb and makes it into the external argument of a derived nominal, while the verb’s external argument is realized as the possessor of the nominal.24 Since the external argument of the nominal is marked with structural Case, extraction may then take place directly. It is important to emphasize that under this conception, not all non-agreeing objects bear inherent Case. As we have, seen, some ditransitive themes extract directly, even though they are not agreement-marked. We will assume that these arguments bear structural Case, assigned by -cit, in contrast to the themes of ditransitives with the directive transitivizer, which bear inherent Case. This entails that -cit has the special property of licensing two structural (object) Cases — a property instantiated elsewhere in languages with ‘symmetrical’ as opposed to ‘asymmetrical’ ditransitives, as explored extensively in the literature on Bantu, for example. We end up with the following typology of Case and agreement for objects in St’át’imcets: (60)

structural Case

+agreement

− agreement

agreeing (primary) objects

second objects of -cit marked ditransitives second objects of directive marked ditransitives, intransitive objects adjuncts

inherent Case

–

oblique Case

–

Two further points need to be emphasized. First, whereas in many frameworks, structural Case is a reflex of the agreement system (see e.g. Chomsky 2000), in the model proposed here the two are partially independent. This is because agreementlinked objects always bear structural Case, but the converse is not true: the class of directly extractable quasi objects have structural Case but are not agreement-linked. Second, we treat the Case system as an autonomous syntactic component, in the sense that no particular semantic interpretation is associated with inherent Case (or, more generally, with quasi objects). In this respect, our view contrasts with that

101


of de Hoop (1992), who proposes a theory of weak Case to handle many types of quasi object, including ‘caseless’ objects in Turkish and Hindi, objects marked with instrumental case in West Greenlandic, and objects marked with partitive case in Finnish. For de Hoop, objects with weak Case have syntactic properties similar to inherently Case-marked objects (in particular, they are associated with particular thematic roles, and generally fail to trigger agreement), but also have a uniform semantic interpretation as incorporated objects. However, as we have shown in St’át’imcets, quasi objects do not always show the properties of semantically incorporated elements. More generally, cross-linguistically quasi objects show a variety of semantic profiles that do not all fit neatly into any theory of semantic incorporation. For example, Enç (1991) claims that the relevant semantic dimension for Turkish quasi objects is a specific/non-specific contrast, while in his careful treatment of Finnish partitive case, Kiparsky (1998) concludes that the relevant dimension is aspectual. For this reason, we reject the weak Case approach in favor of one which maintains the syntactic autonomy of the Case system. Nevertheless, we do not wish to maintain a position of radical autonomy, either with respect to the independence of Case and agreement or to the relation between quasi objects and semantic interpretation. In both cases, in fact, we find a strong but one-way correlation: between agreement and structural Case, on the one hand, and between semantic incorporation and inherent Case, on the other. As far as the agreement/Case connection is concerned, while we have argued that we need both Case and agreement to explain the distribution and behavior of DPs in St’át’imcets, it should be obvious from the table in (60) that the two are not entirely independent. In fact, there is a clear one way implication, stated in (61): (61) A DP may be agreement-marked only if it bears structural Case Where the correlation breaks down is in the opposite direction: there are structurally Case-marked DPs which are not agreement-marked.25 Turning to the syntax-semantics connection, a one-way correlation can also be observed in the interpretive properties of objects in St’át’imcets: while agreementmarked objects always show the typical semantic properties of ‘strong’ objects (wide scope, existential force), the semantic properties of quasi objects are more diffuse, as we have seen. The one-way correlation is given in (62): (62) An incorporated argument must bear inherent Case. It is obviously well beyond the scope of this paper to give a full account of the observed one-way correlations between agreement and Case on the one hand and Case and semantic interpretation on the other. It should be clear, however, that in both areas, our findings have non-trivial theoretical implications. For example, the relation between Case and agreement reported here is incompatible with theories which regard the two as one and the same — or at least, as two sides of the same

Quasi objects in St’át’imcets 103

coin — as well as with theories which regard them as completely independent. Amongst the former we should probably include versions of the Minimalist Program which view structural Case as a reflex of the probe-agree relation (Chomsky 2000a); the latter, on the other hand, would include earlier versions of Minimalism (e.g., Chomsky 1995b), where Case and agreement features may be separately specified as ‘strong’ or ‘weak’. As for the connection between semantic incorporation and quasi objects, we have already rejected the strong mapping between the two advocated by de Hoop: but clearly, we do not wish to give up the correlation altogether. The inability of structurally Case-marked objects to be semantically incorporated suggests that structural Case forces those objects to vacate the VP by LF, though by hypothesis this is not due to attraction by agreement features or projections, but to some property of structural Case itself. One possibility is that all structurally Case-marked DPs must QR; another is that they are subject to a version of the mapping hypothesis (Diesing 1992). We will not attempt to choose between these possibilities here.

8. Conclusion In this paper we have discussed a phenomenon in St’át’imcets which poses serious problems for a Pronominal Argument Hypothesis analysis applied to interior Salish. Formally intransitive predicates, which obligatorily lack object agreement marking, and ditransitive predicates, which license a single object agreement linked to the goal argument, nevertheless allow their theme arguments to be realized as overt DP objects. The existence of arguments which fail to correspond to pronominal agreement is something which is unaccounted for by the PAH. We investigated the possibility that the PAH can be rescued by appealing to an incorporation analysis of quasi objects, and concluded that some, but crucially not all, quasi objects are semantically incorporated. In the second half of the paper we investigated further theoretical consequences of quasi objects. We showed that when agreement, extraction possibilities and oblique marking are taken into account, there are four distinct classes of overt DPs in St’át’imcets. We argued that this four-way split can only be accounted for if Case and agreement are independent notions which are both operative in the language. We suggested that a DP can only be directly extracted if it bears structural Case, and that a subset of quasi objects have inherent Case. Finally, we pointed out that the relationship between Case and agreement is not completely free. There is a one-way implicational relation between the two notions, such that if a DP co-occurs with agreement, it must have structural Case.


Abbreviations act = active intransitivizer, appar = apparently, appl = applicative, caus = causative, conj = conjunctive, deic = deictic, demon= demonstrative, det = determiner, dir = directive transitivizer, erg = ergative, foc = focus, irr = irrealis, mid = middle intransitivizer, neg = negation, nom = nominalizer, obj = object, obl = oblique, pass = passive, pl = plural, poss = possessive, prog = progressive, quot = quotative, sg = singular, sta = stative, subj = subject, ynq = yes/no question.

Notes * We are very grateful to our St’át’imcets consultants Beverley Frank, Gertrude Ned, Laura Thevarge and Rose Whitley for their insights and generosity with their language. Thanks to the audience at a University of British Columbia research seminar, and to Heidi Harley, Andrew Carnie, Mary Willie and two anonymous reviewers for helpful comments and feedback. This work is supported in part by SSHRC grant #410–98–1597. 1. Thanks to an anonymous reviewer for suggesting this term. 2. Examples are given in the van Eijk orthography used by speakers and teachers of the language: see van Eijk (1997). 3. The /t/ at the end of -cit deletes immediately before a following coronal stop or affricate. 4. We will refer to all three subcases as ‘active intransitives’. Ø-marked middles correspond to the traditional Salish class of ‘control intransitives’; see Thompson (1985). See Davis (1997) for a more or less complete list of these verbs in St’át’imcets. 5. The rounded uvular resonant gw is still phonetically present in (8), just as in (2); the w is not written before a following rounded vowel, by orthographic convention. 6. The exceptions are certain third person plural object forms, where number is optionally marked overtly. 7. Van Eijk (1997) refers to intransitive predicates that take DP objects as ‘with object’ predicates. 8. Gerdts (1988) calls the parallel construction in Halkomelem an ‘anti-passive’. We reserve the term ‘anti-passive’ for constructions involving a transitive predicate which is de-transitivized. That is not the case with the Salish intransitive object constructions, which show no trace of transitivizing morphology. 9. In this respect, quasi objects differ from the ‘pseudo-incorporated’ objects in Niuean discussed by Massam (2001), which differ crucially in word order possibilities from (absolutive) casemarked objects. There are also differences in interpretation: whereas pseudo-incorporated objects in Niuean show the general profile of semantically incorporated objects, this is true of only a subset of quasi objects in St’át’imcets. 10. These findings differ from those of van Eijk (1997:229), who states that a single post-predicative DP is always interpreted as theme (ditransitive object) with ditransitive predicates. 11. The reader may be wondering about other ways of testing for asymmetries between agreeing and quasi objects, notably binding and quantifier scope. It turns out that for a variety of independent reasons, it is extremely difficult to elicit such data, and even when it is possible, judgments are highly variable. Condition A binding effects are absent because there are no free

Quasi objects in St’át’imcets 105

standing reflexive or reciprocal pronouns. Condition C effects with non-quantificational DPs are absent or very limited (see Demirdache 1997). As far as quantifier scope is concerned, there are (for independent reasons to do with determiner semantics) no scope interactions between any pair of DPs, unless the potential narrow-scope DP is either a polarity item or contains a bound pronoun (see Matthewson 1998, 1999). However, even then there are difficulties. Quantified intransitive objects are usually avoided altogether, making it almost impossible to test quantifier binding in intransitive object configurations. For these reasons, we have not included binding or quantifier scope data here, though we certainly do not rule out the possibility that such data will eventually prove valuable to the issue at hand, once all confounding variables have been eliminated. 12. There are three different levels of nominalization in St’át’imcets, all marked by s-. We call the type exhibited here predicate nominalization, since it creates a derived nominal predicate, rather than a noun or a nominalized clause. 13. Direct extraction of theme arguments selected by an indirective transitivizer is a St’át’imcets innovation; in most Salish languages, the theme of ditransitive predicates triggers nominalization when extracted, regardless of which transitivizer is employed (see Kroeber 1999: Chapter 6). Moreover, though contemporary speakers uniformly reject nominalization of the predicate when the theme argument of an indirective is extracted, van Eijk (1997) reports that his consultants (a generation older than ours) did apply nominalization under these circumstances: the innovation, in other words, may be quite recent. It is also worth noting that our consultants sometimes fail to nominalize ditransitive predicates suffixed with the directive (-Vn) transitivizer, though they always accept the nominalized variant as grammatical. All this points to an incipient change from an ‘asymmetrical’ to a ‘symmetrical’ object system, whose logical endpoint is a grammar where either goal or theme argument may trigger object agreement. Strikingly, this precisely describes the grammar of one of our consultants, for whom a sentence like (i) has two possible interpretations: (i) legw-ci-tumúlh-kan kelh i cmán’-lap-a hide-ind-2pl.obj-1sg.subj will det.pl enemy-2pl.poss-det “I will hide your enemies from you.” “I will hide you from your enemies.” However, for another five fluent speakers, only the first interpretation is possible, indicating that their grammars are in an intermediate stage of the transition from an asymmetrical to a symmetrical object system, where ditransitive objects are directly Case-marked but still fail to agree. 14. Note that in Interior Salish languages there is no first person plural transitive subject marker; passive is used instead. 15. It could be argued that direct extraction of the theme argument in St’át’imcets ditransitives with -cit is in fact licensed by an extra, phonologically null, agreement morpheme, thereby maintaining the generalization that all and only agreement-licensed objects may be extracted directly. Similar moves have been made by Jelinek (1984) for Warlpiri, and Baker (1996) for Mohawk when faced with the predicament of unregistered objects. We do not make this move, for the following reasons. First, since first and second person ditransitive themes are not registered on the predicate by (overt) object suffixes, the relevant agreement would have to be confined to third person contexts. While it might be argued that with predicates of transfer like “give”, “sell”, or “lend”, the theme is generally inanimate and thus has to be third person, there are predicates such as “show” and “point out” whose themes can easily be first or second person — yet they still cannot be represented by first or second person agreement morphology on the predicate. Second, since (overt) third person plural marking is also ungrammatical for ditransitive themes, the relevant agreement would have to be confined to singular contexts. We do not find the postulation of a paradigm consisting of a single Ø marked third person singular agreement morpheme convincing.


16. See also Mattina (1996:44–50) for discussion of similar facts in Okanagan, a Southern Interior Salish language. Mattina shows that there are objects (which she calls ‘second objects’) which are not cross-referenced on the head, and are not oblique-marked, but which nevertheless are extractable in clefts. 17. Werle (2000) argues that ku objects in St’át’imcets are all semantically incorporated. However, the details of his proposal are slightly different; he adopts Dayal’s (1999) analysis of incorporation. 18. This is not a new idea (Hamida Demirdache, p.c., Davis 1999). 19. Note that adjacency at the surface is not required by the semantic incorporation analysis. 20. As noted above, intransitive objects differ from transitive objects in that the former allow ku even without a licensing operator. However, the question still remains as to whether transitive objects which do contain ku (licensed by an appropriate operator) are also incorporated. Werle (2000) suggests that they are; Matthewson (1998, 1999) does not. For the purposes of this paper, we assume that ku transitive objects are not incorporated. 21. There is variation, both within and across speakers, with respect to the grammaticality and interpretation of ti…a objects in intransitive constructions. Some speakers actually reject the sentences in (54); others accept them sometimes, and still others accept them most of the time. 22. It would still be possible for an incorporation analysis to work if we were only dealing with number distinctions, since we could say that a singular NP predicate (e.g. ‘woman’) becomes pluralized, but remains a predicate (‘women’), and is then incorporated. However, this is much less likely with the deictic distinction illustrated in (55). 23. The scopal properties of quasi objects containing non-polarity determiners are not always the same as those of ordinary DPs containing the same determiners. In particular, sometimes narrowscope readings are possible for non-polarity quasi objects, unlike for ordinary argument DPs. This is a puzzle to be addressed in future research. 24. We remain agnostic about the exact formulation of this operation, which is well-known in Salish but not well understood. Hukari (1977) suggests that it is a lexical (noun-creating) rule, but there are a couple of reasons to treat it as a syntactic operation instead: first, the possessor (the agent) of a predicate nominalization is obligatory, unlike the possessor of an ordinary noun; and second, predicate nominalization is used to create an ‘escape hatch’ for long-range È-movement out of subordinate clauses: to the extent that È-movement is a syntactic process in St’át’imcets, so is predicate nominalization. See Kroeber (1999: 313) for other arguments against the lexical analysis. 25. This one-way correlation can also be observed clearly in the process of historical change from the standard Salish ditransitive asymmetric object system to the innovated St’át’imcets symmetrical object system, which follows the trajectory in (i): (i) loss of oblique marking for ditransitive themes Æ loss of predicate nominalization with extracted ditransitive themes Æ object agreement with either goal or theme It is the intermediate stages which are of interest to us here, since they effectively create the class of quasi objects — those with non-oblique Case but without agreement. In our terms, nonoblique marked themes which trigger nominalization have inherent Case; those which do not have structural Case. Diachronically, then, we find the following one-way implicational relation: (ii)

oblique Case Æ inherent Case Æ structural Case Æ agreement

Agreement, dislocation, and partial configurationality* Mark C. Baker Rutgers University

1.

Introduction

Jelinek’s seminal (1984) article introduced the Pronominal Argument Hypothesis into Principles and Parameters-style theories. In brief, her idea was that some languages have obligatory pronominal agreements/clitics that count as the arguments of verbs and other predicators. Full NPs in such languages are thus never themselves arguments; when present at all they have the status of optional adjuncts of some kind. This view, as developed and extended by Jelinek herself and others, has proved to be quite a successful way of accounting for the properties of fully nonconfigurational, uniformly head-marking languages within generative theory. Baker (1996), for example, provides a whole-scale analysis of the major features of Mohawk grammar in these terms. The main ingredients of the analysis have also been replicated for languages such as Southern Tiwa, Nahuatl, Mayali, Chukchee (Baker 1996), Cree (Reinholtz and Russell 1994), and Mapudungun (Baker 2000, in progress). Jelinek’s Pronominal Argument Hypothesis can be factored into two logically distinct claims, identified in (1). (1a) is the Pronominal Argument Hypothesis proper, and (1b) provides the crucial link between it and the nonconfigurational syntactic behavior. (1) a.

In language X, pronominal elements (clitics or agreements) must be related to each argument position in the clause. b. In the presence of a pronominal element related to an argument position Y, an overt NP Z cannot occupy position Y at S-structure. Z can only be an adjunct to some constituent from which it binds Y.

I refer to the syntactic configuration described in (1b) as dislocation, a phenomenon found to some extent even in English (That book, I really like it). Following Cinque (1990) and Baker (1996) I assume that the dislocated NPs are base-generated in adjoined positions, rather than arriving there by movement, although the evidence

108 Mark C. Baker

that distinguishes the two views is slender and not particularly relevant to the main issue I want to address (but see note 4). While questions remain about both aspects of (1), my concern in this article is to investigate (1b) more closely. There is little doubt that something like (1a) holds in languages like Mohawk. There is also a good deal of syntactic evidence that all overt NPs in this language are in fact dislocated (Baker 1996: Ch. 2, 3). It is not, however, certain that this second property is caused by the first, as (1b) would have it. It is true that the dislocation of NPs correlates with the presence of agreement almost perfectly in Mohawk, since both are present in virtually every structure. But the empirical evidence for very implausible correlations is just as strong. One could, for example, say that the presence of vowels within the verb causes dislocation in Mohawk. There would be no empirical evidence against such a claim, since every verb has vowels as well as agreement markers. Nor is it entirely clear exactly what it is about agreement that forces NPs to be dislocated. Data from other languages have been used to call the theoretical connection between agreement and dislocation posited in (1b) into question. On the one hand, Austin and Bresnan (1996) argue against Jelinek (1984) on the grounds that there are Australian languages like Jiwarli that have the same nonconfigurational effects Jelinek observed for Warlpiri, but the pronominal clitics that are found in Warlpiri are absent in Jiwarli. Austin and Bresnan go on to claim that agreement and nonconfigurationality are independent of each other in the Australian languages. On the other hand, Indo-European languages like Spanish and Greek clearly do have agreement with subjects, but this agreement does not necessarily force dislocation of the subject (Alexiadou and Anagnostopoulou 1998). Therefore, it is not clear that agreement is either necessary or sufficient to cause dislocation and hence nonconfigurationality. The Bantu languages provide an excellent laboratory to study these questions. Descriptively speaking, they fall somewhere between fully nonconfigurational languages like Mohawk and “ordinary” configurational languages like Spanish or Greek. Subject agreement is required in most Bantu languages, but object agreement/clitics are optional, as shown in (2).1 (2) Njuchi zi-na-(wa)-lum-a alenje. (Chichewa) bees sm-past-(om)-bit-fv hunters “The bees stung the hunters.” Bresnan and Mchombo (1987) show that when the Object Marker (OM) is present, the NP associated with the direct object theta-role can be freely omitted, freely ordered, and appears outside the VP; Mchombo (2001) adds the information that discontinuous constituents are permitted in just this case. Chichewa with the OM is thus syntactically like Mohawk. When the OM is omitted, however, a direct object NP is required, it appears inside the VP, right-adjacent to the verb, and

Agreement, dislocation, and partial configurationality 109

discontinuous constituents are not permitted, just as in English. A Mohawk-like mode and a Spanish-like mode thus seem to co-exist in Chichewa and the other Bantu languages. There are also certain “inversion” constructions in which thematic subjects are not agreed with, so the effects of agreement on subjects can also be studied. (1a) clearly does not hold for Bantu languages the way it does for the languages Jelinek originally studied, which makes it easier to investigate the dynamics of (1b). To capitalize on this opportunity, I consider in this article the relationship between agreement and dislocation in the Bantu language Kinande, building on Bresnan and Mchombo’s work on Chichewa and Kinyalolo’s (1991) work on Kilega, among others. By comparing Kinande, which has optional agreement, to languages like Mohawk with obligatory agreement, I argue that the biconditional in (3) (a strong version of (1b)) is valid for an interesting range of languages. (3) A verb X agrees with an NP Y if and only if Y is in a dislocated, adjunct position. Section 2 looks in particular at the effects of object agreement in Kinande, and Section 3 looks at the effects of subject agreement. Section 4 then compares Kinande, in which (3) holds, with the Indo-European languages studied by Alexiadou and Anagnostopoulou (1998) (henceforth A&A), in which it does not. I propose a parameter that distinguishes the kind of agreement that causes dislocation from the kind of agreement that does not in a way that is independently motivated. Section 5 considers briefly where Salishan languages and Athapaskan languages — two families that Jelinek herself has worked on, with controversial results — fit into the typology that emerges from Section 4. Section 6 concludes the discussion.

2.

Agreement and objects

I begin my defense of (3) by considering objects in Kinande. As in Chichewa, transitive verbs in Kinande may or may not bear an OM, as shown in (4). (4) a.

N-a-gul-a eritunda. 1sg.S-T-buy-fv fruit.5 “I bought a fruit.” b. Eritunda, n-a-ri-gul-a. fruit.5 1sg.S-T-om5-buy-fv “The fruit, I bought it.”

Conceivably one might say that there is phonologically null object agreement even in (4a) — a logical possibility that Jelinek (1984) explicitly allows. If so, then object

110

Mark C. Baker

agreement is really obligatory in Kinande after all. Evidence that this is not the case comes from pro-drop. When an overt OM is present, no NP object is required, as one would expect. If there is no overt OM, however, omitting the overt NP object forces the verb to have an intransitive interpretation. This is shown by the contrast in (5). (5) a.

Ng-u-li-gul-a ebitsungu, Kambale if-2sg.S-npst-buy-fv potatoes.8 Kambale a-luandi-bi-kuk-a. 1.S-will-om8-cook-fv “If you buy potatoes, Kambale will cook them.” # b. Ng-u-li-gul-a ebitsungu, Kambale a-luandi-kuk-a. if-2sg.S-npst-buy-fv potatoes.8 Kambale 1.S-will-cook-fv “If you buy potatoes, Kambale will do the cooking.”

Assuming that OMs either are pronominals themselves or license null pronouns (i.e. pro), (5) provides evidence that there is no Ø OM in Kinande, and hence that the OM is optional.2 With this established, we can investigate how the presence of the OM correlates with evidence that the associated NP is dislocated in Kinande. There are two major syntactic sources of evidence as to whether an NP is dislocated or not: evidence from word order, and evidence from nonreferential/quantified NPs (see Cinque (1990), Baker (1996), and A&A, among others).3 Consider first word order. Dislocated objects are adjoined at the periphery of a clause, whereas undislocated objects are inside VP, typically adjacent to the verb. (4) already shows that “object” NPs appear in different locations in Kinande depending on whether object agreement is present or not. In the absence of an OM, the direct object must immediately follow the verb (as in Chichewa). When an OM is present, the direct object — if there at all — must come at the front of the sentence, set off from it by an intonation break. It is impossible to have a postverbal object together with an OM, or a preverbal object without one: (6) a.

N-a-(*ri)-gul-a eritunda. 1sg.S-T-om5-buy-fv fruit.5 “I bought a fruit.” b. Eritunda, n-a-*(ri)-gul-a. fruit.5 1sg.S-T-om5-buy-fv “The fruit, I bought it.”

Kinande differs slightly from Chichewa on this point, in that the equivalent of (6a) is grammatical in Chichewa (see (2)). But Bresnan and Mchombo (1987) show that even in Chichewa the object is not in the VP when the OM is present. The difference between Kinande and Chichewa can be described by saying that Chichewa

Agreement, dislocation, and partial configurationality

permits both right dislocation and left dislocation, whereas Kinande allows only left dislocation. This is independently motivated by the fact that agreed-with subjects also cannot follow the verb in Kinande (see (7)), whereas this is perfectly possible in Chichewa. (7) *Mo-a-sat-ire omukali. aff-1.S/T-dance-ext woman “The woman danced.” For this reason, even with the OM present, word order in Kinande more restricted than it is in Mohawk or Chichewa. It is nonetheless true that all and only agreedwith objects show the word order characteristic of dislocation in Kinande, in accordance with (3). Evidence from nonreferential, quantified NPs also supports (3). It is known that nonreferential indefinite NPs cannot be dislocated in a language like Italian (Rizzi 1986b); Cinque 1990); rather dislocated NPs have only a definite or specific reading. Kinande nouns usually begin with an augment vowel that matches the vowel of its noun class prefix. Under the scope of negation, however, and in certain other contexts this vowel can be omitted. This augmentless NP has a narrow scope, nonspecific indefinite reading (Progovac 1993). Thus, we have the following contrast: (8) a.

Yohani si-a-nzire o-mu-kali. John neg-1.S/T-like aug-cl1-woman “John does not like the woman.” b. Yohani si-a-nzire mu-kali. John neg-1.S/T-like cl1-woman “John does not like a(ny) woman.”

That this special nonreferential form can be in object position when there is no object agreement indicates that this is not a dislocated position. The augmentless noun cannot be used together with an OM, however, as expected:4 (9) a.

Omukali mo-a-teta-gul-a ki-ndu. woman.1 aff-1.S-neg/past-buy-fv cl7-thing “The woman didn’t buy anything.” b. *(E)-ki-ndu, omukali mo-a-teta-ki-gul-a. (aug-)cl7-thing woman aff-1.S-neg/past-om7-buy-fv “Anything, the woman didn’t buy it.”

Kinande differs in this respect from Mohawk, which has no truly nonreferential NPs. Wh-phrases in Kinande have a similar distribution to augmentless noun phrases, as expected given that they too are nonreferential, quantified expressions. An object wh-phrase can appear in-situ, immediately following the verb when there

111

112

Mark C. Baker

is no OM ((10a)), but it cannot appear in a dislocated position with an object marker present ((10b)). (10) a.

Kambale a-gul-a ebihi? (also OK in Chichewa) Kambale 1.S/T-buy-fv what.8 “What did Kambale buy?” b. *Ebihi, Kambale a-bi-gul-a? what.8 Kambale 1.S/T-om8-buy-fv “What did Kambale buy?”

Rizzi (1986b) subsumes the ungrammaticality of the Italian equivalent of (10b) to the Weak Crossover Condition, which restricts when pronouns can act as variables bound by quantified expressions. Again, an obligatory agreement language like Mohawk is different: since dislocation is required in that language, there is no wh-in-situ (Baker 1995); true wh-movement must always apply in such languages. Nonreferential NPs of various kinds are thus possible in true object position in Kinande, when there is no OM, but not in dislocated position when there is an OM. Over all then, we have evidence that agreement and dislocation go hand in hand for objects in Kinande. True polysynthetic languages like Mohawk are consistent with this, in that they always have agreement and always have dislocation. The generalization in (3) is thus supported in this domain.

3.

Agreement and Subjects

Next let us consider the relationship of agreement and dislocation for subjects in Kinande. Unlike object agreement, subject agreement is in a sense obligatory in Kinande. There is no alternative form of (11) that lacks the subject prefix ba-, for example. (11) Abakali *(ba)-[a]-gul-a eritunda. woman.2 2.S-T-buy-fv fruit.5 “The woman bought a fruit.” Kinande does not seem very different from a polysynthetic language like Mohawk in this respect. There are, however, constructions in which the Kinande verb does not agree with the thematic subject. Rather, a variety of nominal expressions can be moved into the Specifier of TP in Kinande, and whatever is moved there triggers “subject” agreement. (12b) illustrates one such case, the so-called “subject–object reversal” construction, in which the direct object is preposed to clause-initial position. The verb comes between the preposed object and the thematic subject, and shows “subject” agreement only with the former. These sentences are best in a generic


tense (they have a proverbial flavor) and they express contrastive focus on the thematic subject. (12) a.

Omukali mo-a-seny-ire olukwi (Normal) woman.1 aff-1.S/T-chop-ext wood.11 (lw’-omo-mbasa). lk11-loc.18-axe.9 “The woman chopped wood (with an axe).” b. Olukwi si-lu-li-seny-a bakali wood.11 neg-11.S-pres-chop-fv women.2 (omo-mbasa). (SO-reversal) loc.18-axe.9 “WOMEN do not chop wood (with an axe).”

Similar reversal constructions have been analyzed in other Bantu languages, including Kinyarwanda (Kimenyi 1980; Ura 2000), Kilega (Kinyalolo 1991) and Kirundi (Ndayiragije 1999). OVS orders are also possible in verb-second languages like German, of course, but in German the verb does not agree with the fronted object. This is our first hint that agreement in Bantu languages is fundmentally different from agreement in Indo-European languages. A second kind of inversion is found with unaccusative intransitive verbs, including the passives of transitives. A locative expression can come before such verbs, triggering “subject” agreement, while the NP that would otherwise be the subject comes after the verb and is not agreed with. The Chichewa version of this locative inversion construction is studied in detail by Bresnan and Kanerva (1989).5 (13) ?Omo-mulongo mw-a-hik-a omukali. loc.18-village.3 18.S-T-arrive-fv woman.1 “At the village arrived a woman.” Locative inversion exists also in English (Bresnan 1994), but in English the verb agrees with the postverbal theme, not the preverbal locative. This is the second hint that agreement in Bantu is fundamentally different from agreement in IndoEuropean languages — leads that I follow up on in the next section. Kinande also has a kind of impersonal construction that can be used with unergative verbs, shown in (14). Here the intransitive verb comes before the agentive subject and does not agree with it. There is no obvious Specifier of TP, but the verb has a pleonastic locative affix ha- in the subject agreement slot.6 (14) Mo-ha-teta-sat-a mukali (omo-soko). aff-there-neg/past-dance-fv woman.1 loc.18-market “No woman danced in the market.” Thus while it is true that all finite verbs bear “subject agreement” in Kinande, there

113

114

Mark C. Baker

are nevertheless a variety of erstwhile subjects that are not agreed with. We can therefore investigate the relationship between agreement and dislocation for putative subjects as well. Consider first evidence for dislocation that comes from word order. In normal, SVO constructions like (11), the subject appears at the left edge of the TP. It cannot appear after the verb, as shown in (15) (see also (7)). (15) *A-gul-a omukali eritunda. 1.S/T-buy-fv woman.1 fruit.5 “The woman bought a fruit.” This is consistent with the view that agreed-with subjects are dislocated in Kinande, and are left-adjoined to TP. In contrast, the unagreed-with thematic subjects in (12), (13), and (14) are not peripheral to the clause; rather, they come between the verb and a locative or instrumental PP. This clearly indicates that they are not dislocated. Presumably, they sit in a specifier position — either SpecvP or the specifier of some functional head lower than T. As such, they come before VP-adverbs but after the verb that has been raised to T. Agreed-with subjects cannot, however, occupy this internal position. This then supports the biconditional in (3). Sample structures are given in (16). (16) a.

Ordinary sentence: Agentive subject is agreed with and occupies a peripheral position ((12a)). TP

TP

NPi woman

NP

T′

proi Agri+T

vP v′

NP ti

v chop

VP NP wood

V′ V

PP with-axe


b. SO-reversal sentence: Agentive subject is not agreed with and occupies an argument position ((12b)). TP

TP

NPi wood

NP

T′

proi Agri+T

vP NP women

v′ v chop

VP NP ti

V′ V

PP with-axe

The situation is complicated somewhat by evidence that dislocated, preverbal subjects in Kinande do not appear in exactly the same positions as left-dislocated objects do. A left-dislocated object is set apart from the rest of the clause by a clear intonation break, whereas a preverbal subject is not. Moreover, a dislocated object must come before a preverbal subject in a sentence that has both, never the other way around: (17) a.

Eritunda, omukali a-ri-gul-a. fruit.5 woman.1 1.S/T-om5-buy-fv “The fruit, the woman bought it.” b. *Omukali, eritunda, a-ri-gul-a. woman.1 fruit.5 1.S/T-om5-buy-fv “The woman, the fruit, she bought it.”

Kinande is more restricted in these respects than Chichewa, Greek, Italian, or Mohawk; in all of these other languages the equivalent of (17b) is possible. A dislocated object can also (marginally) come before a focused NP, but never after it, between the focus particle and the verb. Preverbal subjects show the opposite distribution: they naturally appear between the focus particle and the verb, but cannot come before the focused NP. The contrast is shown in (18).

115

116

Mark C. Baker

(18) a. *Ehilanga hyo omukali, n-a-mu-h-a. peanuts.19 foc.19 woman.1 1sg.S-T-om1-give-fv “It’s peanuts that the woman, I gave to her.” b. Ehilanga hyo Kambale a-h-a omukali. peanuts.19 foc.19 Kambale 1.S/T-give-fv woman.1 “It’s peanuts that Kambale gave to the woman.” These facts suggest that, although both agreed-with subjects and agreed-with objects are dislocated, they are adjoined to different positions. Dislocated subjects may adjoin to TP, below the Focus head, whereas dislocated objects may only adjoin to the very highest projection of the clause, presumably CP.7 The structure of (17a) is thus (19). CP

(19)

CP

NPk fruit

C Ø

TP TP

NPi woman

NP

T′

proi Agri+T

vP NP

v′

ti Agrk+v buy

VP NP prok

V V

(PP)

If a focus projection were included too, as in (18), it would come above TP as the complement of C in such a structure. Further evidence shows that even the dislocated subject’s right to adjoin to TP is not absolute. The Kinande verb always raises at least as high as T, as shown by the fact that it comes before the verb-phrase-internal subject in (12b)/(16b) and (14). Sometimes, however, it apparently raises even higher. In (20b) it has raised to unite with the affirmative element mo-. This mo- is in strict complementary distribution with the focus particle kyo seen in (20a), so I assume that it belongs to the same


category, which I call the Focus/Polarity head. In (20c), the verb has raised through T and Foc/Pol all the way up to C, thereby combining with the monosyllabic complementizer nga- ‘if ’. (20) a.

Ekihi kyo Marya a-ka-bul-a what.7 foc.7 Maria 1.S-pres-wonder-fv nga-kyo Yosefu a-gul-a? if-foc.7 Joseph 1.S/T-buy-fv “What does Maria wonder if Joseph bought?” b. Mo-ba-nyi-bw-ire ba-ti Kambale aff-2.S-1sg.O-tell-ext 2-that Kambale mo-a-gul-ire eritunda. aff-1.S/T-buy-ext fruit.5. “They told me that Kambale bought fruit.” c. Marya a-ka-bul-a Yosefu Maria 1.S-pres-wonder-fv Joseph nga-mo-a-gul-ire amatunda. if-aff-1.S/T-buy-ext fruit.6 “Maria wonders if Joseph bought fruits.”

If the dislocated subject could always adjoin to TP, then it should be able to appear after the verb in (20b) and (20c), given that the verb has moved on past T. But this word order is not possible. The ungrammaticality of such examples confirms that the subject is truly dislocated and does not sit in the specifier of TP or some other designated functional category that licenses its case (see Kinyalolo (1991) for a similar argument in Kilega). Apparently, the subject adjoins to TP if the verb is in T, to Foc/PolP if the verb is in Foc/Pol, and to CP if the verb is in C.8 The descriptive generalization can be stated as in (21). (21) An NP can be adjoined to YP only if YP contains the head of the chain of the verb that theta-marks the pronoun bound by NP. This generalization also applies to dislocated objects; it explains why the NP ‘fruit’ cannot adjoin to vP in a structure like (19), producing a Subject–Verb–Object–PP order in which the verb agrees with the object. Why then can the dislocated subject avail itself of the opportunity to adjoin to TP or Foc/PolP, while the dislocated object must adjoin higher, to CP? I assume that this is because the dislocated subject binds the highest pro in the basic clause, whereas the object does not. Because of this structural fact, the clause can be interpreted as a predicate of the subject, but not of the object, along the lines originally laid out by Williams (1980): (22) An NP adjoined to YP is licensed by predication only if NP is coindexed with the highest nominal in YP and Y the head of YP contains the verb.

117

118

Mark C. Baker

According to Williams, two kinds of pronominals can make a clause into a predicate: a PRO in the highest subject position of a nonfinite TP, and a wh-operator in the highest Specifier of a CP. The first possibility gives rise to infinitival relatives like a mani [PROi to fix the sink], the second to ordinary tensed relative clauses like the mani [whoi [I met ti yesterday]. Kinande sentences with preverbal subjects are, I claim, similar to the infinitival relatives, except that the “predicate variable” is the pro subject of a tensed clause, rather than the PRO subject of a nonfinite clause. A dislocated object cannot be licensed in this way, because it does not correspond to the highest position in the clause — unless operator movement has taken place, as in clefts like (18b). Dislocated subjects, then, can be licensed in a variety of relatively low positions by a kind of predication, whereas dislocated objects can only adjoined to the expression as a whole, where they are licensed not by predication but only as a discourse topic.9 I turn next to evidence of dislocation that comes from the distribution of nonreferential and quantified NPs. Recall that Kinande has special noun forms that have no initial augment vowel and appear under the scope of negation and certain other operators. These augmentless nouns can occur in the ordinary direct object position, but they cannot be left-dislocated (see (9)). (23a) shows that augmentless nouns also cannot be agreed-with, preverbal subjects. This is just what we expected if such subjects are dislocated. (23b) shows how the intended meaning is normally expressed in Kinande, with the erstwhile subject functioning as the complement of an existential verb. (23) a. *Mu-kali mo-a-teta-gul-a eritunda. cl1-woman aff-1.S/T-neg/past-buy-fv fruit.5 “No woman bought a fruit.” b. Si-ha-li n’-omukali n’-omuyima oyo neg-there-be by-woman.1 by-one.1 that.1 u-a-gul-a eritunda. 1.Swh-T-buy-fv fruit.5 “There is not a single woman who bought a fruit.” Augmentless nouns can be subjects in the inversion constructions, however. Indeed, the subject must not have an augment in SO-reversal sentences ((24a)) and impersonal inversion ((24b)): (24) a.

Olukwi si-lu-li-seny-a (*a-)ba-kali. wood.11 neg-11.S-pres-chop-fv (aug)-cl2-women “WOMEN do not chop wood.” b. Mo-ha-sat-ire (*o)-mu-kali muyima. aff-there-dance-ext (aug)-cl1-woman.1 one “Only one woman danced.”


Locative inversion is also possible — indeed, improved — when no augment vowel is prefixed to the postverbal subject: (25) Omo-mulongo mw-a-hik-a (?o-)mu-kali. loc.18-village 18.S-T-arrive-fv (aug)-cl1-woman.1 “At the village arrived a woman.” Nonreferential subjects are thus possible if and only if they are not agreed with, in accordance with (3). Wh-expressions are expected to have a similar distribution. An in-situ wh-phrase can be in the postverbal object position, but not in the dislocated object position (see (10)). In the same way, I predict that wh-in-situ should be bad for preverbal, agreed-with subjects, but possible for postverbal, unagreed with subjects. The first part of the prediction is clearly true: interrogative subjects must normally be clefted in Kinande: (26) a. *(Iyo)ndi a-gul-a eritunda? who 1.S/T-buy-fv fruit.5 “Who bought a fruit?” b. Iyondi yo u-a-gul-a eritunda? who foc.1 1.Swh-T-buy-fv fruit.5 “Who (is it that) bought fruit?” The second part of the prediction also seems true, although some of the sentences are less than perfect: (27) a. ?Olukwi si-lu-li-seny-a ndi? wood.11 neg-11.S-pres-chop-fv who “WHO does not chop wood?” (SO-reversal) b. ?Mo-ha-teta-sat-a ndi? aff-there-neg/past-dance-fv who “Who didn’t dance?” (Impersonal inversion) c. Omo-mulongo mw-a-hik-a ndi? loc.18-village 18.S-T-arrive-fv who “Who arrived at the village?” (Locative Inversion) I assume that these structures are possible as far as the syntax of dislocation is concerned, and the uncertain, intermediate status of (27a) and (27b) comes from a tension between the pragmatics of asking a question and the very specific discourse presuppositions associated with SO-reversal and impersonal inversion.10 Another way to tell whether an NP is dislocated or not has to do with the scope/ specificity of the NP. Dislocated NPs can have indefinite readings, but they must be specific (Cinque 1990: ch. 2). A consequence of this is that they act like they have wide scope with respect to other quantified NPs. Consider, for example, the

119

120 Mark C. Baker

contrast in (28). The postverbal object in (28a) can be understood as a nonspecific indefinite, taking narrow scope with respect to the quantified subject. This sentence has the normal meaning in which each man bought a different fruit. In contrast, the dislocated object in (28b) must take wide scope with respect to the quantified subject; this sentence can only have the implausible reading that there is a single fruit that every woman bought. (28) a.

Obuli mundu mo-a-gul-ire eritunda. every man.1 aff-1.S/T-buy-ext fruit.5 “Every man bought a fruit.” (can be different fruits) b. #Eritunda, obuli mukali mo-a-li-gul-ire. fruit.5 every woman.1 aff-1.S/T-om5-buy-ext “A fruit, every woman bought it.” (only one fruit bought)

(29) shows the results of applying this test to the preverbal subject position. The agreed-with subject must take wide scope with respect to a quantified object; the sentence implies that there is a woman who bought every fruit. Kinande differs in this respect from English, but is like Spanish and Greek, the pro-drop languages studied by A&A. (29) Omukali a-gul-a obuli ritunda. woman.1 1.S/T-buy-fv every fruit “A (single) woman bought every fruit.” Overall, then, Kinande provides particularly good evidence for Jelinek’s (1984) and Baker’s (1996) claim that dislocation (and hence nonconfigurational syntax) is inherently related to the presence of agreement in some languages. Neither direct objects nor thematic subjects need to be agreed with in Kinande, so minimal pairs with and without agreement can be constructed. Those arguments that are agreedwith consistently show the behavior of dislocated NPs, both with respect to word order and quantificational properties, whereas unagreed-with arguments do not. The optional agreement in Kinande thus has the same syntactic consequences as the obligatory agreement found in polysynthetic languages like Mohawk. Agreement does cause dislocation over a range of languages, including both polysynthetic and nonpolysynthetic ones.

4. An agreement parameter The biconditional in (3) is not a universal of human language, however. As impressive as it is for Mohawk, Kinande, and other languages, it does not hold for Indo-European (IE) languages. A&A show that agreed-with preverbal subjects in pro-drop IE languages like Spanish and Greek act like they are dislocated. My


Kinande data replicates their results in this respect. But Spanish and Greek also allow the VSO order (or VOS) shown in (30b) for Greek. The subject is still agreed with in this order, but it shows no sign of being dislocated: it is clearly clauseinternal, and it can receive a narrow scope indefinite reading. (30) a.

Kapios fititis stihiothetise kathe arthro. (SVO) some student filed every article “Some student filed every article.” ($ > " only) b. Stihiothetise kapios fititis kathe arthro. (VSO) filed some student every article “Some student filed every article.” ($ > " (preferred) or " > $)

The equivalent of (30b) is clearly bad in Kinande (see (15)) and in Chichewa (Bresnan and Mchombo 1987). The similarities and differences between the two language families can be summarized as in (31). (31)

Preverbal subjects Postverbal subjects

IE (Spanish, Greek)

Bantu (Kinande, Chichewa)

Trigger agreement Are dislocated Trigger agreement In argument position

Trigger agreement Are dislocated Don’t trigger agreement In argument position

This raises the question of why agreement systematically causes dislocation in the Bantu languages, but not in the IE languages. Some kind of parameter must be at work here, and identifying it should help to clarify the relationship Jelinek discerned between agreement and nonconfigurationality. There is an independently motivated difference between agreement in IE languages and Bantu languages that I mentioned in passing in Section 3. This difference, stated in (32), may be another manifestation of the same parameter.11 (32) a. Tense agrees with the nominative NP in Indo-European. b. Tense agrees with its specifier in Bantu. Locative inversion (and there-insertion) sentences in English provide the classic illustration of (32a): the finite Tense on the verb agrees not with the preverbal PP, but with the postverbal NP that depends on it for nominative case. (33) a.

On the table stands the trophy Chris won at the debate tournament. b. On the table stand the trophies Chris won as a debater.

121

122 Mark C. Baker

This is a robust property of IE languages: verbs agree with a nominative NP somewhere in VP rather than with an obliquely case-marked structural subject all the way from Icelandic in the extreme northwest to Hindi in the extreme southeast. Bantu languages also have locative inversion and there-insertion constructions with much the same properties (see Bresnan (1994)), but throughout the Bantu area the verb agrees with the locative structural subject, not with the “nominative” postverbal subject. This could be seen in (13) (repeated here as (34a)) and (34b) in Kinande; it also holds true for Chichewa, Sesotho, Kilega, and others. Omo-mulongo mw-a-hik-a (?o)-mu-kali. loc.18-village.3 18.S-T-arrive-fv aug-cl1-woman “At the village arrived a woman.” b. Oko-mesa kw-a-hir-aw-a ehilanga. loc.17-table 17.S-T-put-pass-fv peanuts.19 “On the table were put peanuts.”

(34) a.

The effects of the parameter in (32) can also be seen in auxiliary constructions. Many complex tenses are found in IE languages, and the rule is that person-number agreement appears on the highest auxiliary, but not on the nonfinite participle. (Adjective-like gender-plus-number agreement may appear on a participle under complex conditions that vary from one language to the next; I assume this is a different matter and put it aside.) (35) a. I am eating fish. b. She has eaten fish. This too is valid from Iceland to India. In contrast, complex tenses in the Bantu languages show full person-number-gender agreement on both the participle and the auxiliary. (36) shows this for Kinande; see Kinyalolo (1991) for detailed discussion of Kilega and Swahili.12 Abakali ba-bya ba-ka-gul-a amatunda. women.2 2.S-were 2.S-pcpl-buy-fv fruits.6 “The women were buying fruits.” b. Tú-lwé tú-ká-ly-a. 1pl.S-leave 1pl.S-pcpl-eat-fv “We were eating.”

(36) a.

We may suppose that the theory of NP-movement requires that the thematic subject move successive cyclically from its theta-position, through the subject of the participle projection, to the subject of TP in both language families, as sketched in (37). (37) [TP proi T+Aux [PcplP ti Pcpl+eatk [vP ti tk (fruits)]]]


The difference in agreement then follows directly from (32). There is a case relationship between the subject and T, but not between the subject and Pcpl in (35). Therefore, in IE languages where agreement is keyed to the checking of case, only the tensed auxiliary shows agreement. In contrast, agreement in Bantu is keyed to specifier relationships and the specifiers of both T and Pcpl contain a link in the chain of the thematic subject. Therefore, the Bantu value of parameter (32) implies that agreement appears on both verbs, as in (36). There are other constructions, more peculiar to Kinande, that also illustrate (32). The focus particle in cleft-like constructions agrees with the focused NP in its specifier, for instance; such agreement is not usually found in IE languages: (38) Eritunda ry-o n-a-h-a omukali. fruit.5 5-foc 1sg.S-T-give-fv woman.1 “It’s a fruit that I gave to a woman.” There is also a vP-internal particle that comes between an object and some other complement. This particle agrees with the object in its specifier position, even though case comes from the verb (see Baker and Collins (in preparation)). (39) Mo-n-a-hir-ire okugulu k’-omo-kihuna. aff-1sg.S-T-put-ext leg.15 lk.15-loc.18-hole.7 “I put the leg in the hole.” Particles appearing in approximately the same position in English do not show agreement with the object that precedes them (cf. I put the book/books back in the box). We also saw in Section 3 that objects fronted past the thematic subject trigger tenserelated agreement on the verb in second position in Kinande (see (12b)), whereas agreement is with the nominative subject in comparable constructions in German. It is easy to see why this robust parameter is relevant to the dislocation question. At stake is why a VSO word order is possible with agreement in the IE pro-drop languages, but not in the Bantu ones. The relevant configuration is something like (40). (40) … T·AgriÒ+verbk … [vP NPi tk…] In IE languages, this structure is conceivable (subject to language-specific requirements), because it is legitimate in these languages for Tense to agree with something that is not in its specifier as long as there is a case-checking relation. The same parameter value that allows “downward” agreement in locative inversion also allows agreeing VSO orders with no (overt) specifier of TP. The structure in (40) is not, however, well-formed in Bantu; Bantu Tense can only agree with something that is in its specifier. One might try fixing the structure by putting a pro in SpecTP, as in (41). Then the agreement is legitimate, but the Theta Criterion is violated, since there are two argumental NPs (pro and the overt NP) associated with a single theta-role.

123

124 Mark C. Baker

(41) *[proi T·AgriÒ+ verbk … [vP NPi tk…]] (41) is however possible if there is no binding relation between pro and the thematic subject, and hence no agreement between the verb and the thematic subject. This happens either if some other element moves to SpecTP (as in the SO-reversal and locative inversion constructions), or if SpecTP is filled with a null locative expletive, which triggers the special agreement marker ha-. This goes part of the way toward explaining why agreed-with NPs must be dislocated in Bantu; specifically, it explains why an undislocated NP cannot appear lower than the agreement-bearing head, a configuration that is allowed in IE languages. To complete the analysis, I must also explain why an overt NP cannot be in the specifier of the agreeing head in Bantu. The representation in (42a) must also be ruled out, and only the one in (42b) permitted. (42) a. *[TP NPi T·AgriÒ+Verb … [vP ti…]] b. [TP NPi [TP proi T·AgriÒ+Verb … [vP ti…]]] The Bantu languages do not differ in this respect from the IE ones that have pro-drop, according to A&A, although (42a) is possible in non-pro-drop IE languages like English and French. Why would this be? In standard Minimalist accounts, T has two features: an EPP feature that attracts a specifier and a nominative case feature that checks the case of some NP. Suppose that these are the two basic features of T. T may also have Agr features, but these must be parasitic on one of its two primary features. IE languages choose to package Agr with the nominative case features, and Bantu languages package it with the EPP feature, as expressed in (43a,b). So far, this is nothing more than a restatement of the parameter in (32) in more technical terms. But suppose that we also add the statement in (43c). (43) a. Agreement is not a distinct feature of a head. b. It must be packaged with an EPP feature or a Case feature. c. If Agr is packaged with one feature, it checks the head’s other feature. In IE languages like Spanish or Greek, Agr is packaged with the nominative case feature. (43c) therefore implies that it checks the EPP feature of T. This means that (42a) is impossible in Greek and Spanish, because nothing triggers movement of the subject NP out of vP to SpecTP: (44) [NPi T +verb [ti…]] * unless NPi = Ø (nom) EPP

(no feature triggers movement)

NOM +Agri

This is nothing more than a restatement of A&A’s analysis of IE pro-drop languages:


the agreement features on the verb satisfy the EPP feature of tense, making A-movement to the subject position impossible. Consider now the Bantu side of the parameter. The same result holds, but for a different reason. In Bantu, Agr is packaged with the EPP feature. Therefore, according to (43c) it checks the nominative case feature of Tense. An NP can be attracted to SpecTP in these languages, but it must be an NP that has no Case feature to check. In other words, it must be an empty category, pro. (45) [NPi T +verb [ ti…]] *unless subject has no case (nom) EPPi + Agr

(i.e., unless it is pro or trace)

NOM

Thus, the representation in (42a) is also bad in Bantu. The structure in (42b), in contrast, is possible in pro-drop languages of both the IE and Bantu families. Dislocated NPs are exempt from case-checking, which applies only to NPs in A-positions (see Baker (1996) and references cited there). If they have a case-form at all, it can be a default one not necessarily the same as the case of the theta-position it binds (cf. Him, I think he is really smart). Such dislocated NPs are therefore possible no matter how the nominative case feature of Tense is checked. My theory technically implies that the subject pro is in SpecTP and caseless in Bantu and that it is in SpecvP and checks nominative case in Spanish and Greek. But these theoretical differences are not visible in the surface form of SVO sentences, which look the same in both language families. Finally, there is the question of why (42a) is possible in non-pro-drop IE languages like English. For such languages, I assume that the few agreement endings that there are (mostly just –s in the present tense, plus the irregular forms of be) are either syntactically inert (A&A: 523) or are entirely absent in the syntax and added in the postsyntactic morphology, as proposed by Halle and Marantz (1993). Neither the EPP feature nor the Case feature is taken out of the equation by Agr in such a language, and the same overt NP can check both. This completes my discussion of the basic cases. A prediction that emerges from this analysis is that the unagreed-with postverbal subjects in Kinande must not receive nominative Case, in contrast to their IE counterparts. This is because the Case feature is automatically absorbed by the agreement in Kinande. I suggest that this explains the fact, noted above, that postverbal subjects must usually be augmentless, nonreferential NPs in Kinande. This is clearly true for SO-reversal and impersonal inversion, as shown again in (46). (46) a. ?Obwabu bu-ka-hek-a (*a)-ba-kali. drink.14 14.S-hab-carry-fv aug-cl2-women “WOMEN carry drink (not men).” (SO reversal)

125

126 Mark C. Baker

b. Mo-ha-sat-ire (*o)-mukali muyima. aff-there-dance-ext aug-cl1-woman one.1 “Only one woman danced.” (Impersonal inversion) This restriction follows if the augmentless NPs are NPs that do not have structural Case: they either bear a special partitive case (Belletti 1988) or they are nonreferential NPs rather than DPs and as such do not need to have Case at all. The one inversion construction that (marginally) allows an augment vowel on the postverbal NP is locative inversion. But locative inversion obeys another well-known restriction: the postverbal “subject” must be the theme argument of an unaccusative verb, not the agent argument of an unergative verb, as shown in (47) (compare Bresnan and Kanerva (1989) on Chichewa). (47) a.

Omo-mulongo mw-a-hik-a (?o-)mu-kali. loc.18-village.3 18.S-T-arrive-fv aug-cl1-woman “At the village arrived a woman.” b. *Oko-ririma kw-a-lim-a omukali. loc.17-field.5 17.S-T-cultivate-fv woman.1 “In the field cultivated a woman.”

Because it is a theme, generated in VP, the postverbal NP in (47a) can get the inherent accusative case found also in double object constructions (cf. Baker (1988)). Agents generated in vP cannot get this Case, so (47b) is bad. We thus correctly account for the fact that …VS… constructions are much more restricted in Kinande than in IE languages like Greek and Spanish, even when the difference in agreement is controlled for.13 These ideas can also be applied to object agreements/clitics in Bantu and the IE languages. Some IE varieties allow an NP in argument position to be doubled by an object clitic, including Greek ((48a), from Artemis Alexiadou, personal communication) and River Plate Spanish (Jaeggli 1982). However, simple clitic-doubling seems not to be possible in Bantu languages: agreed-with objects (if possible at all) are always dislocated. O Janis de ti theori ti Maria eksipni. det John not her considers det Maria intelligent “John does not consider Mary intelligent.” b. *N-a-ri-gul-a eritunda. (Kinande) 1sg.S-T-om5-buy-fv fruit.5 “I bought a fruit.”

(48) a.

This can be seen as a straightforward consequence of the parameter in (43). The object clitic in both language families appears on some head lower than T, say for concreteness v. Then the substructure of interest is (49):


(49) [vP (proi) om/cli+v [VP V NPi ]] In an IE language, it is possible (subject to language-particular requirements) for this agreement to be related directly to an NP in argument position, lower than the object clitic. In a Bantu language, however, this is impossible. There must be a pro in SpecvP (or the equivalent) in order to trigger the object agreement in Bantu. If there is also a full NP in the argument position, the Theta Criterion is violated. A structure like (49) is only possible in Bantu if a pro in SpecvP binds a trace inside VP. If there is an overt “object” NP at all, it must be dislocated, adjoined to the CP as a whole.14 Thus, the same principles apply to subject agreement and to object agreement in so far as dislocation is concerned — a very attractive feature of the current theory.15

5.

Agreement and dislocation in other languages

Whenever a new parameter is proposed, it creates curiosity about how it applies to other languages and language families. Jelinek exemplifies this enterprising spirit, having pursued the Pronominal Argument Hypothesis not only for Warlpiri (the primary language considered in her first article) but also for many other languages, including Salish and Athapaskan languages. As a result of the research she has inspired, I can close this article with informed conjectures about how my agreement parameter applies to St’át’imcets and Slave, based on the information in Davis and Matthewson (this volume) and Rice (this volume). It seems clear that St’át’imcets is like IE languages with respect to the parameters of agreement and dislocation discussed here, not like Kinande or Mohawk. One obvious sign of this is that VSO word orders in which the verb agrees with the subject (and the object is not right-dislocated) are the norm in this language: (50) Cuz’ k’ác-cal ta n-sísq7-a ku sts’úqwaz’. going.to dry-act det 1sg.P-uncle-det det fish “My uncle is going to dry fish.” In this respect, St’át’imcets is clearly like Greek and Spanish (see (30b)) and not like Kinande or Chichewa (see (15)). Further confirmation comes from the fact that subjects in St’át’imcets can contain the nonreferential, narrow-scope indefinite determiner ku: (51) Cw7aoz kw-s ít’-em ku smúlhats. neg det-nom sing-mid det woman “No woman sang.” Ku-NPs in St’át’imcets are semantically very similar to the augmentless NPs found

128 Mark C. Baker

in Kinande. Nevertheless, the grammaticality of (51) contrasts with the ungrammaticality of the comparable Kinande sentence in (23a). This reconfirms that agreed-with subjects need not be dislocated in St’át’imcets. I thus agree with Davis and Matthewson that the Interior Salish languages are not Pronominal Argument Languages in the sense of (1), although I believe Jelinek was right to posit a Pronominal Argument Parameter as part of UG. In contrast, the Northern Athapaskan language Slave seems very much like Kinande in the relevant respects. Since this is a head-final language, one cannot expect to find VSO orders that will show whether the subject is dislocated or not. But for direct objects it is clear that agreement goes along with dislocation: Gah tSi˛ ni˛dháá teyedéhnde. rabbit dog far 3.S-pf-chase-yDO “The rabbit, the dog chased it far.” (Agr and Dislocation) b. Semo˛ zhéhchá gots’e˛h tudhe thehtsi˛. 1sg.P-mother 3.S-pf-boil-yDO and soup 3S-pf-make “My mother boiled them and made gruel.” (no Agr, no dislocation)

(52) a.

Rice herself (1989: 1209) analyzes sentences like (52a) as left dislocations of the object. Indeed, there are several similarities of detail between Slave and Kinande. In both languages object agreement is optional from the point of view of verbal morphology but highly conditioned by the syntax. Object agreement is required when the object is pro-dropped (see the first conjunct of (52a) for Slave, and (5) for Kinande) or dislocated, and it is (usually) forbidden when the object is not dislocated but adjacent to the verb.16 Slave is also like Kinande in that the dislocated object must come before the subject, giving an OSV order (not SOV), and is set off from the sentence by an intonation break. This suggests that object NPs adjoin only to CP in Slave, whereas subjects (assuming that they too are dislocated by agreement) adjoin to IP. Thus not only does (43b) have its Bantu setting in Slave, but (21) and (22) hold as well. Slave is another good example of a partially nonconfigurational language, in which the (1a) part of the Pronominal Argument Hypothesis does not hold, which serves to highlight the fact that the (1b)/(3) part holds in spades. I conclude that the agreement parameter proposed in Section 4 applies in a meaningful way to languages other than the Bantu and IE languages that originally motivated it.

6. Conclusion In this article, I have argued that the correlation between agreeing with a nominal and having that nominal be dislocated is very close in some languages. This applies


not only to the purely polysynthetic, nonconfigurational languages studied by Jelinek and Baker, but also to partially configurational languages like the Bantu languages. Since agreement is optional in Bantu languages, one can see all the more clearly that it forces dislocation. This vindicates an essential theoretical component of Jelinek’s vision of a pronominal argument language. I also compared agreement in Bantu languages with agreement in IndoEuropean languages, to see why agreement always causes dislocation in one group but not the other. I related this to the independently motivated fact that agreement is keyed to EPP features in Bantu, but to case features in IE languages. Since agreement in Bantu is not directly related to Case features, it absorbs those Case features. The consequence of this is that agreed-with arguments in Bantu can only be Case-less pros and any overt NP must be dislocated. In this way differences in dislocation can be related to other differences in the syntax of …VS… constructions and in the behavior of auxiliary constructions in a satisfying way. Salish languages seem to be like IE languages in these respects, whereas Slave seems to be very much like Kinande. Fully polysynthetic languages like Mohawk apparently all have Kinande/Slave-style, EPP-related type of agreement, which causes dislocation. This fact could be worked into the statement of the Polysynthesis Parameter itself, pending further investigation. It is very exciting that over the last 20 years it has become more and more feasible to discover subtle and interesting parameters that distinguish (say) Kinande, Slave, and Mohawk on the one hand from St’át’imcets, Greek, and Navajo on the other. This is true in part because better data is becoming available on this range of languages, and in part because there is increasing awareness about how the principles of generative linguistics can be applied to them, even though they look at first quite foreign to the fields that first spawned those principles. Both the empirical and theoretical advances can be attributed in no small part to the research, stimulation, and personal example of Eloise Jelinek.

Notes * A preliminary version of this paper was presented at the Utrecht workshop on the Role of Agreement in Argument Structure, August-September 2001. I thank all the participants of that workshop, and especially Eloise Jelinek, Artemis Alexiadou, Henry Davis, Jonathan Bobaljik, Phil Branigan, for their comments and suggestions. I also thank Chris Collins, both for specific comments and the general stimulation of working together on the syntax of African languages. I thank Heidi Harley, Andrew Carnie, and an anonymous reviewer for their comments and suggestions on an earlier draft of this article. My greatest thanks go to Philip Mutaka: without his interest, influence, and native Kinande judgments this paper would have been completely impossible. Its remaining faults are entirely my responsibility.

130 Mark C. Baker

1. The following abbreviations are used in the glosses of the examples: 1sg.S, 1st singular subject marker; 1sg.P, 1st singular possessor; 1pl.S, 1st plural subject marker; act, active voice; aff, affirmative; aug, augment vowel; cl, noun class prefix; det, determiner; ext, extended mood marker; foc, focus particle; fv, final vowel (perhaps a mood marker); hab, habitual tense; lk, linker particle; loc, locative gender prefix; mid, middle voice; neg, negative; npast, nonpast tense; pcpl, participle; om, object marker; pres, present tense; sm, subject marker; T, tense prefix. Arabic numerals between 1 and 19 are used to represent the various gender/number noun classes in Bantu languages; they can be appended to the gloss of a noun (e.g. fruit.5), to a subject agreement (5.S), to an object prefix (om5), and to certain other elements that agree with nouns in their environment (e.g., focus particles). 2. See Baker (in preparation) for a comparison between Kinande and Mapudungun, a language that does have a null OM as well as an overt one. The difference is induced by the fact that overall Mapudungun has the morphosyntactic properties of a polysynthetic language (noun incorporation, restricted causatives, nominalized clauses, etc.) but Kinande does not. 3. Phonological and/or pragmatic evidence is sometimes used to tell if an NP is dislocated. For example, dislocated NPs are sometimes set off from the rest of the clause by an intonation break, and they are sometimes associated with topic or contrastive focus interpretations (see Rice (this volume) on Slave). The Pronominal Argument Hypothesis is committed to there being cases of syntactic dislocation that do not have these nonsyntactic properties, however. NPs in Mohawk, for example, need not be set off intonationally, and they can have any discourse function. I show below that agreed-with objects in Kinande have these phonological and pragmatic characteristics of dislocation, but agreed-with subjects do not. All this implies that there are natural interconnections between the various components of language, but they are not fully deterministic. 4. Several reviewers point out that, although the definiteness of the agreed-with object is consistent with its being clitic-left-dislocated in the sense of Cinque (1990), it is also consistent with the claim that the object has undergone “object shift” via movement. This process too produces a definite interpretation of the object (Diesing 1992). Nevertheless, it is clear that the object agreement cannot itself be the trigger for such an object shift, attracting the object to its specifier, as one reviewer suggests. The fronted object must come before the subject; it is thus either adjoined to TP or it is in the specifier of some functional head higher than T. The OM, on the other hand, is adjacent to the verb root and inside of the tense marker and subject agreement. This suggests that whatever head houses the OM is lower than T (I assume it is adjoined to v). The fronted object is therefore much higher in the structure than the head that putatively causes its fronting. This confirms my belief that dislocation is a better account of these facts than object shift. The complete structure is drawn in (19) below. 5. This example is better without an augment vowel on the theme NP; see Section 4 below for discussion. There is a second type of locative inversion in Kinande, in which locative gender marker appears not on the preverbal NP but as an enclitic on the verb. This Kinande-specific type of locative alternation (which is preferred in many contexts) has the same properties as the crosslinguistically more common type illustrated in the text in the relevant respects. 6. Both SO-reversal and impersonal inversion are subject to a transitivity restriction in Kinande: they are impossible if an object as well as a subject follows the verb. See Collins and Branigan (1997) and Collins (1997) for analysis of a similar restriction on quotative inversion in English. SO-reversal in Kilega and Kirundi, in contrast, is not restricted in this way. 7. Alternatively, one could identify the head of the projection where the dislocated object appears as a Topic head, as in Rizzi (1997).


8. The alternative, suggested by an anonymous reviewer, would be to say that the subject is always generated in TP, but that Focus/Polarity and complementizer heads lower to T in (20b) and (20c). I have no conclusive evidence against this, but assume that head-lowering analyses are the last resort, especially for a language like Kinande, which is already known to employ V-raising to T, rather than T lowering (see (16b)). Note that the attachment of Foc/Pol and C to the inflected verb cannot be simply a matter of PF cliticization under string-adjacency because the overt subject would intervene between the particle and the verb. Also unlike these alternatives, (21) does independent work in my analysis, explaining why dislocated objects cannot adjoin to vP. 9. That an NP adjoined to TP must be licensed by predication is evidently a parameter, distinguishing Kinande from languages like Chichewa, Mohawk, Italian and Greek, all of which allow S-O-V orders with an object clitic/agreement on the verb. The exact nature of this parameter deserves closer consideration than I can give it here. Presumably dislocated subjects can adjoin to CP instead of TP as long as they are discourse topics. In most cases, it is hard to tell whether the dislocated subject is adjoined to TP or to CP. The unacceptability of (i) as contrasted with (18b) could be interpreted as evidence that subjects cannot adjoin to CP. (i) *Kambale, ehilanga hyo a-h-a omukali. Kambale, peanuts.19 foc.19 1.S/T-give-fv woman.1 “As for Kambale, it’s peanuts that he gave to the woman.” This sentence is judged to be “too artificial”. I assume that it is ruled out as some kind of garden path effect or discourse clash, rather than as a violation of syntactic principles. (Note that even dislocated objects adjoined to Focus Phrases are considered marginal.) I still guess that dislocated subjects can adjoin to CP even though I have not constructed the perfect sentence to prove it. 10. A related concern is that (27a,b) might be echo questions, not true constituent questions, in which case their grammaticality may not be relevant. The discourse situations in which inversion is used are so restricted that it is hard to tell the two readings apart. 11. This parameter has emerged out of joint work on verb phrase structure in African languages done with Chris Collins; I think him for inspiration and extensive discussion of this topic. 12. These structures are somewhat problematic for one aspect of Jelinek’s (1984) theory: her claim that the pronominal prefixes on the verbs are literally the arguments of the verb. These sentences have two subject prefixes, but only one relevant theta role, creating a potential Theta Criterion violation. The problem does not arise if these prefixes are simply agreements, which are not themselves arguments but have the capacity to license a null pro argument, as assumed in Baker (1996), Kinyalolo (1991), and in (36) below. 13. Still to be investigated is how these considerations apply to languages like Chichewa and Kilega, which also have inversion constructions but which do not show Kinande’s contrast between augmented and nonaugmented nouns. 14. See Section 3 for discussion of why the dislocated object cannot adjoin to vP or (in Kinande) to TP, but only to CP. Another notable difference between object clitics in IE languages versus object clitics in Bantu shows up in periphrastic tenses. In IE languages, object clitics typically attach to the inflected auxiliary (e.g. French: Je l’ai mangé ‘I it-have eaten’), whereas in Bantu languages they attach to the participle: (i) Tu-a-(*ri-)by-a tu-ka-(ri)-ly-a. 1pl.S-T-(om5)-be-fv 1pl.S-pcpl-om5-eat-fv “We were eating it.”

131

132

Mark C. Baker

It seems very plausible to see this as another consequence of the parameter in (43), which allows certain nonlocal agreement relations to exist in IE but not in Bantu. I do not pursue the details of such an account here, for reasons of space. 15. There is still one irreducible difference between the two, however, which is that T must bear an EPP feature (and hence, in Bantu, an agreement), whereas vP may or may not bear such a feature. I have no special insight on why this is so. (I thank Henry Davis for discussion helping to clarify this point.) 16. Rice (this volume) discusses a few marginal exceptions to this general pattern, conjecturing that they helped lead to historical change in Southern Athapaskan, but these do not invalidate (52) as revealing the basic parameter setting for Slave. I leave open the question of how these ideas apply to Navajo, in which agreement is obligatory. The obvious possibility is that Navajo is an IE-like language, with agreement not causing dislocation, as Rice assumes. But it could also be that Jelinek is right to analyze Navajo as a pronominal argument language, and the difference is that (for some reason) dislocated NPs can adjoin to a wider range of categories in Navajo than in Slave or Kinande, including PP and vP.

Part II

Interfaces

Multiple multiple questions* Molly Diesing Cornell University

1.

Introduction

The syntax of multiple questions and multiple fronting has received a good deal of attention, with most works focusing on Slavic and Romanian (see Richards 1997, Boškovic´ 1998, and Pesetsky 2000). Yiddish not only stands out among the Germanic languages in allowing multiple fronting (as in Slavic languages like Bulgarian) but it also allows single fronting as an option in multiple questions (as in English and German). Furthermore, these two strategies for question formation differ in their syntactic properties, most notably with regard to the ‘Superiority Effect’. In this paper I argue that the syntactic contrasts between the two types of questions are explainable under a somewhat different view of superiority than is usually assumed in the literature. The two forms of multiple questions also provide evidence against LF movement of wh-phrases, and thereby lend support to an analysis along the lines of Reinhart’s (1998) treatment of wh-in-situ using choice functions. This analysis requires a re-assessment of the role that discourse-functional properties such as ‘D(iscourse)-linking’ (Pesetsky 1987) play in the syntax of superiority and the semantics of multiple questions. Specifically, choice functions serve as the formal means by which contextual effects of the discourse are encoded in the grammar — thereby yielding a formal approach to a functional phenomenon. The outline of the paper is as follows: I first present the basic facts regarding multiple questions in Yiddish, I then examine the situations in which multiple fronting is not possible, this is followed by a presentation of the superiority data, and an explanation of the syntax of superiority in Yiddish, finally I examine the implications the syntax of multiple questions in Yiddish hold for the treatment of the semantics of multiple questions.

2.

Yiddish multiple questions: A first glance

Multiple questions in Southeastern Yiddish can take multiple forms. The first option consists of fronting of a single wh-phrase to the position immediately preceding the inflected verb (Yiddish is a verb-second language):

136 Molly Diesing

(1) a.

ver vet vuhin geyn mit aykh? who will where go with you? “Who will go where with you?” b. ver hot vos gekoyft? who has what bought “Who bought what?” c. ver geyt vuhin? who goes where “Who is going where?”

One fact worth noting here is that the object wh-in-situ appears immediately to the left of the (non-finite) main verb (when there is one), despite the fact that the normal VP order is VO. While I will not deal with this fact in detail here, a possible explanation lies in the fact that Yiddish has a preverbal focus position below the subject position (Diesing 1997), which can (and even must) host a single in-situ wh-phrase. This of course involves a somewhat extended meaning of the term ‘insitu’: I use it here to mean a wh-phrase which has not been moved to a specifier position above the subject (that is, a non-fronted wh-phrase). The second question strategy involves multiple fronting of the wh-phrases, in these cases the fronted wh-phrases all precede the position of the inflected verb:1 (2) a.

ver vuhin vet geyn mit aykh? (Mark 1978: 380) who where will go with you “Who will go where with you?” b. ver vos hot gekoyft? who what has bought “Who bought what?” c. ver vuhin geyt? (Zaretski 1929) who where goes “Who is going where?”

To the best of my knowledge, in these examples the two orders are synonymous. Thus, it appears that the multiple fronting possibility is truly optional; neither option is associated with any special meaning. In particular, the wh-in-situ in the single fronting cases need not be discourse-linked (D-linked) — the potential answers to these questions are not necessarily limited to sets of entities previously defined in the discourse (Pesetsky 1987). This observation contrasts with what has been claimed for multiple fronting languages like Bulgarian (Pesetsky 2000), in which the in-situ option is restricted to D-linked wh-phrases. One question that arises is that of where the multiply-fronted wh-phrases move to: to CP (as argued for Bulgarian by Rudin and others), or to IP (as claimed for Polish). Evidence bearing on this question can be seen in multiple fronting in embedded questions:

Multiple multiple questions

hot zi ni(sh)t gekent farshteyn [ver es shlogt zikh mit vemen] has she not been-able understand who exp hits self with whom b. hot zi ni(sh)t gekent farshteyn [ver mit vemen es shlogt zikh] has she not been-able understand who with whom exp hits self “(So) she couldn’t understand who was fighting with whom.” (Jacobs et al. 1994: 414) c. Lomir geyn, [ver vuhin es geyt]. let’s go who where exp goes “Let’s go where(ever) who(ever) is going.” (Zaretski 1929)

(3) a.

What is essential to note here is that the expletive is required when the subject is extracted in an embedded question, leaving a gap in the subject position (Birnbaum 1978, Travis 1984, Diesing 1990). The expletive does not appear if a non-subject (in this example, the adverb frier ‘earlier’) has been topicalized, occupying the SpecIP position (see Diesing 1990 for arguments justifying this placement of non-subject topics): (4) zi iz gekumen zen [ver frier vet kontshen] (Diesing 1990: 50) she is come see who earlier will finish “She has come to see who would finish earlier.” Thus, the expletive does not appear when there is no gap in SpecIP, but when such a gap exists, the expletive is obligatory. If the fronted wh-phrases were able to land in SpecIP (as in Rudin’s analysis of Polish), this obligatory appearance of the expletive in the examples in (3) would be totally unexpected. The appearance of the expletive in the multiple fronting cases demonstrates that the multiply fronted wh-phrases are not adjoined to IP, but rather are attached higher up, at the CP level, as in Bulgarian. Consistent with this conclusion, the fronted wh-phrases form a unit which cannot be broken up (i.e. by a parenthetical), a property which Yiddish shares with Bulgarian: (5) *ver, nokh dayn meynung, vuhin vet geyn? who after your opinion where will go While it is clear that Yiddish does genuinely display a form of multiple fronting as an option in multiple questions, there are cases where the multiple fronting is not possible. Before proceeding to the discussion of superiority effects in Yiddish multiple questions, I illustrate these cases in the next section.

3.

Constraints on Multiple Fronting

An interesting property of multiple fronting constructions in Yiddish is that multiple fronting is not possible with certain types of wh-phrases. Interestingly,

137

138

Molly Diesing

D-linking emerges as a factor which blocks multiple fronting. As noted by Hoge (2000) ‘inherently D-linked’ Wh-phrases (such as ‘which N’, ‘whose N’, see the discussion in Comorovski 1996) are only grammatical in single-fronting questions, the multiple fronting counterparts are ungrammatical: (6) a.

voser farshlog hot ver kritikirt? (Hoge 2000: 216) which proposal has who criticized “Who criticized which proposal?” b. *ver voser farshlog hot kritikirt? who which proposal has criticized c. *voser farshlog ver hot kritikirt? which proposal who has criticized

(7) a.

vemens hitl hot ver avekgeganvet? whose hat has who away-stolen “Who stole whose cap?” b. *ver vemens hitl hot avekgeganvet? who whose hat has away-stolen c. *vemens hitl ver hot avekgeganvet? whose hat who has away-stolen

At first blush, this may not seem so terribly surprising, since in some multiple fronting languages there is at least a preference for leaving D-linked wh-phrases insitu (Pesetsky 1987, fn. 32). But those examples (in Polish and Romanian) all involve two D-linked wh-phrases, one of which is fronted. As far as I know, the Yiddish restriction is the only such case which applies to questions containing a single D-linked wh-phrase.2 This restriction will turn out to follow quite straightforwardly from the analysis of the superiority facts discussed in the next section. A second restriction on multiple fronting involves adjuncts. Certain adjuncts — namely ‘why’ and ‘how’ — cannot front in a multiple fronting question, but do front in single fronting constructions (with blatant disregard for superiority — a fact which will be discussed further below):3 (8) a.

vi azoy hot ver gezungen? how has who sung “Who sang how?” b. ver hot vi azoy gezungen? who has how sung “Who sang how?” c. *ver vi azoy hot gezungen? who how has sung d. *vi azoy ver hot gezungen? how who has sung

Multiple multiple questions 139

(9) a.

farvos hot ver ongeklungen der mamen? why has who called the mother “Who called the mother why?” b. ver hot farvos ongeklungen der mamen? who has why called the mother? “Who called the mother why?” c. *ver farvos hot ongeklungen der mamen? who why has called the mother d. *farvos ver hot ongeklungen der mamen? why who has called the mother

This is rather surprising, since the adjuncts ‘why’ and ‘how’ resist being left in-situ in multiple questions in a number of other languages (see Reinhart 1998, Hornstein 1995, among others). I will return to this observation in the discussion of semantic implications in Section 6.

4. Superiority in Yiddish The two types of multiple questions show rather different behavior with respect to superiority. Multiple questions formed by fronting a single wh-phrase do not show superiority effects, patterning in this regard like German (Haider 1986). Extracting a subject over an object is grammatical (as is the reverse pattern): (10) a.

veri hot ti vos gekoyft? who has what bought b. vosi hot ver ti gekoyft? what has who bought “Who bought what?”

This behavior is quite general; no superiority effects are seen in interactions between object wh-phrases with wh-adverbials. (11) a.

vos hot Maks vi azoy/farvos geshpilt? (Hoge 2000) what has Max how/why played b. vi azoy/farvos hot Maks vos geshpilt? how/why has Max what played “How/why did Max play what?”

Nor does a superiority effect arise in multiple questions involving a subject wh-phrase and a wh-adverbial: (12) a.

ver hot vi azoy/farvos geshpilt pyane? who has how/why played piano

140 Molly Diesing

b. vi azoy/farvos hot ver geshpilt pyane? how/why has who played piano “Who played the piano how/why?” This absence of superiority effects is not limited to the cases which might be reduced to ECP effects. No ‘pure’ superiority effects are seen either. (13) a.

vemen hot der lerer t vos geheysn leyenen? whom has the teacher what ordered read b. vos hot der lerer vemen geheysn leyenen t? what has the teacher whom ordered read “Who did the teacher tell to read what?”

Interestingly, questions involving multiple fronting show an entirely different pattern in that superiority effects do in fact arise. Again, multiple fronting in Yiddish appears to behave like Bulgarian in this respect.4 Thus, we see that while the subject wh-phrase can precede the direct object wh-phrase in the initial cluster of wh-phrases, the reverse order is strongly ungrammatical: (14) a.

ver vos hot gekoyft? who what has bought “Who bought what?” b. *vos ver hot gekoyft? what who has bought c. ikh veys nit ver vos es hot gekoyft. I know not who what ES has bought. “I don’t know who bought what.” d. *ikh veys nit vos ver es hot gekoyft. I know not what who ES has bought.

Similar contrasts are seen in questions involving a subject and an indirect object, regardless of whether the indirect object was marked simply with the dative case (ab examples), or with a preposition (c-d examples): (15) a.

ver vemen hot gegebn khanike-gelt? (Hoge 2000) who whom has given Hanukkah-money “Who gave whom Hanukkah money?” b. *vemen ver hot gegebn khanike-gelt? whom who has given Hanukkah-money c. ver tsu vemen hot geshikt leshonetoyves? who to whom has sent Rosh Hashanah cards “Who sent Rosh Hashanah cards to whom?” d. *tsu vemen ver hot geshikt leshonetoyves? to whom who has sent Rosh Hashanah cards


Finally, questioning of double objects shows that the indirect object is superior to the direct object:5 (16) a.

vemen vos hot Maks gegebn? whom what has Max given “What did Max give (to) whom?” b. *vos vemen hot Maks gegebn? what whom has Max given

To give a brief summary, Yiddish has both single and multiple wh-fronting in both main and embedded clauses. The appearance of a subject expletive in embedded questions involving subject extractions makes it clear that the multiply fronted wh-phrases are in SpecCP — or at least some position higher than topics. Although both the single and multiple fronting questions occur as free variants in many cases, multiple fronting is restricted in that not all wh-phrases are able to appear in multiple fronting questions. Specifically, the wh-adverbials vi azoy ‘how’ and farvos ‘why’ and inherently D-linked wh-phrases cannot occur in multiple- fronting constructions. Multiple fronting questions are also distinguished syntactically from single fronting questions in that they show superiority effects, while single fronting questions do not. Finally, it is important to note that D-linking does not override superiority in multiple fronting constructions (as it has been shown to do in other languages, Pesetsky 1987), since D-linked wh-phrases are not even allowed in these cases.

5.

Superiority syntax

In this section I attempt to explain the peculiar behavior with respect to superiority displayed by Yiddish multiple questions.6 The facts presented in the previous sections pose a number of puzzles, which I will address in turn. The first of these is: Why does Yiddish show superiority effects only with multiple fronting? In tackling this question, I will assume without argument that superiority effects in wh-questions are a manifestation of the ‘Attract Closest’ condition (Chomsky 1995a) which applies to phrasal movement more generally:7 (17) Attract Closest: α can attract K only if there is no legitimate operation Move α targeting K, where α is closer to K. As far as the specific mechanics of multiple fronting (and multiple specifiers more generally) are concerned, I follow Richards (1997) in deriving multiple-fronting through a ‘tucking-in’ operation. For example, in a multiple question involving a subject wh-phrase, the subject wh-phrase is the one closest to CP, therefore, as required by Attract Closest, it moves first. The remaining wh-phrase(s) move afterwards and ‘tuck-in’ under the first-moved wh-phrase, leaving the subject

141

142 Molly Diesing

wh-phrase highest in the tree.8 Applied to the Yiddish example, the ‘Attract Closest’ approach appears to derive the superiority effect in the multiple fronting cases, but there is still the problem of why superiority effects are not seen with single fronting. One previous solution to the absence of superiority effects (applied for example, to languages such as German), is to allow the wh-phrases to undergo scrambling prior to wh-movement (perhaps adjoining to VP, or landing in the specifier of an inflectional head, here indicated as ?P), and then move the wh-phrases (or features, in a single-fronting language). Such an account is found in Richards (1997, p. 90 ff.). The scrambling has the effect of bringing a ‘lower’ wh-phrase closer to CP prior to wh-movement, in effect reordering the superiority relations, or even voiding them by rendering the scrambled phrases ‘equidistant’ (but see Hoge 2000 for criticisms of this approach). Since Yiddish does allow scrambling (Diesing 1997), a solution of this sort might well work for Yiddish single fronting: (18) a.

vos hot ver gekoyft? what has who bought “What did who buy?” b. [CP vosj hot[?P tj [VP ver tj gekoyft]]]

Crucially, this strategy exploits the fact that scrambling in Yiddish itself does not show superiority effects (Diesing 1997). The end result is that scrambling moves the object wh-phrase to a position where it is now ‘superior’ to the subject wh-phrase, which allows subsequent wh-movement to proceed without a ‘real’ superiority violation. In other words, superiority gets pre-empted by scrambling. But it is clear that this proposal is not without problems when applied to the full array of Yiddish data. The main difficulty is that in the multiple fronting cases, the pre-wh-movement scrambling must be blocked somehow, so that the superiority effects will surface. One possibility is that there might be some notion of derivational economy that can achieve this. A comparison of two candidate derivations will illustrate how this might work. Consider first the derivation which yields a grammatical result: (19) a.

ver vos hot gekoyft? who what has bought “Who bought what?” b. [CP veri vosj hot [ ti tj gekoyft]]

In this case no scrambling occurs before the wh-phrases are moved to CP; no extraneous operations take place. Now, the derivation of the illicit order via scrambling would look like this: (20) a. *vos ver hot gekoyft? what who has bought b. *[CP vosj veri hot [ tj ti tj gekoyft]]


This derivation clearly has to be ruled out, or else we would see no superiority effect. If we compare the two derivations globally (comparing the cost of the two derivations side-by-side), it’s easy to see that there is a sense in which the ‘bad’ derivation is less economical — it takes more steps. But global economy metrics are notoriously problematic in that they carry a very high computational cost (see Collins 1997 for more detailed arguments against global economy). In any case this attempt at explanation still doesn’t tell us why the same measure of economy wouldn’t apply to the single-fronting cases. This then is the real problem: What mechanism will allow scrambling in the single fronting cases, but not in the multiple fronting derivations? On the way to answering this question, I will first consider the issue of what it is that enables multiple fronting in Yiddish in the first place. Here the literature on multiple fronting (and multiple specifiers in general) offers a number of options. One class of explanations utilizes the idea that CPs come in multiple ‘flavors’. In this case Yiddish would be seen as choosing between two types of CPs: one with the feature [+multiple specifiers] and the other bearing the feature [−multiple specifiers] (cf. Richards 1997 and Pesetsky 2000). A second type of explanation follows the work of Luigi Rizzi. Along these lines, we could suppose that the ‘fine structure’ of the Yiddish periphery contains two inflectional heads, one which allows multiple specifiers and the other which doesn’t (cf. Rizzi 1997 TopP and FocP). An analysis of multiple fronting in Slavic of this sort can be found in Grohmann (2000). Because of space limitations, I won’t attempt to motivate a systematic choice between these two approaches, and will simply pursue the [±multiple specifiers] strategy.9 Spelled out in somewhat more detail, Yiddish freely chooses between [±multiple] CPs. The choice of a [+multiple] CP allows multiple instances of phrasal movement, enabling multiple fronting of wh-phrases. The [−multiple] CP allows only one wh-phrase to move. In this case, the remaining in-situ phrases will check features by either feature movement (as in Chomsky 1995a, Chapter 4) or Agree (Chomsky 2000). So far this is just a minor modification of proposals made by David Pesetsky (2000) to account for multiple fronting in Slavic. To derive the Yiddish superiority facts, I add in one additional constraint on [+multiple] CPs: (21) In Yiddish, movement to a [+multiple] CP must originate from an A-position. This constraint has the effect of blocking A-bar scrambling before multiple-wh-movement, thereby ‘enforcing’ superiority. Since [−multiple] CPs are not subject to this constraint, pre-wh-movement scrambling can occur in these cases, yielding the apparent immunity to superiority.10 This constraint has an additional desirable effect in that it also prevents adjuncts from participating in multiple fronting constructions, thereby solving the third puzzle which arises in Yiddish multiple fronting constructions, which is why adjuncts can’t appear in multiple

144 Molly Diesing

fronting constructions. The answer is simply that they don’t appear in A-positions, and therefore are not eligible for multiple fronting, given the constraint in (21). Taking a somewhat more speculative turn, it is possible that the constraint in (21) may also provide an answer to the question of why D-linked wh-phrases cannot participate in multiple fronting constructions. It is known that [+specific] NPs in Yiddish must scramble out of the VP (Diesing 1997). Furthermore, in my earlier work I have shown that this scrambling in Yiddish has properties of A-barmovement. If this constraint also applies to D-linked wh-phrases (which have been claimed to be [+specific] in the sense of involving reference to a set of individuals previously established in the discourse, see Kiss 1993 and Comorovski 1996), then the fact that they move to an A-bar position before wh-movement takes place, makes it impossible for them to subsequently move to a [+multiple] C-Spec, given the constraint in (21). It remains to be shown that the ‘specificity’ seen in D-linked wh-phrases is in fact associated with the same sort of scrambling behavior attested in [+specific] non-question noun phrases (as Irene Heim has pointed out to me, this seems somewhat at odds with the idea of wh-phrases being in focus), so I will leave this at the purely conjectural level. Here, in a nutshell, is the proposed account of the Yiddish superiority facts. Yiddish allows both [+multiple] and [−multiple] C-Specs. The [±multiple] feature has been argued to be the locus of cross-linguistic variation in superiority behavior (Pesetsky 2000). Yiddish displays the option of being able to choose freely between the two types of CP specifiers. Yiddish also allows scrambling of noun phrases; this scrambling (which happens prior to wh-movement) is what enables single fronting wh-questions to apparently evade the effects of superiority. This option is not available in the multiple fronting cases because the [+multiple] CP specifier is subject to an additional restriction in that movement to a [+multiple] C-Spec must be from an A-position, while movement to a [−multiple] C-Spec is unrestricted. This asymmetry between the two types of specifiers in Yiddish is what produces the asymmetric superiority effects. A consequence of this analysis is that it raises the possibility of a rather different view of superiority in Yiddish multiple-fronting questions. What we see here is that the ‘superiority effect’ seen in these cases appears to boil down to a requirement that the fronted wh-phrases and the A-positions have to be isomorphic — a constraint that is something quite different from the structurally-based ‘Attract Closest’ accounts of superiority. A number of possibilities arise here. The first is that perhaps all superiority effects are simply due to an ‘A-position isomorphism constraint’, which imposes a stricter condition on multiple fronting (resulting in the banning of adjuncts and D-linked wh-phrases). Another is that the constraint on multiple fronting in Yiddish is merely a ‘pseudo-superiority effect’ resulting from the A-position isomorphism constraint, but that ‘true superiority’ (as seen in English multiple questions) does arise from the structural effects of Attract Closest.


Yiddish is presumably also subject to the strictures of Attract Closest, but in singlefronting questions this is masked by the intervention of scrambling. I will leave the investigation of these possibilities to further research.

6. What does the Yiddish LF look like? In this section I consider the question of the semantic interpretation of multiple questions in light of the syntactic discussion above. Multiple wh-fronting languages have frequently been described as ‘wearing their LF on their sleeves’, in that the overt syntax produces what presumably is structurally equivalent to the LF of a single fronting language (with the implication that these latter languages undergo some form of corresponding multiple-fronting at LF, see Pesetsky 2000 for an overview of the possibilities). If this view of the relationship between multiple fronting and single fronting languages is correct, Yiddish expresses both options: the flaunting of LF with overt multiple fronting, and the more abstract and concealed LF seen in single fronting. But are Yiddish multiple-fronting questions simply ‘LF laid bare’? In other words, do the single fronting multiple questions in Yiddish undergo a subsequent LF fronting of the in-situ wh-phrases? Another way of stating the question is to ask if there is any evidence of LF movement of the wh-in-situ. The classic form such evidence takes is the superiority effect. But, as we have seen, single-fronting questions do not show superiority effects — A-bar scrambling provides an ‘escape hatch’. Furthermore, if the [±multiple specifiers] story in the previous section is correct, it’s not clear whether LF movement of the in-situ wh-phrases would even be possible, given that it is not clear that there is any syntactic motivation for such phrasal movement in terms of checking theory. On these syntactic grounds it appears that we must conclude (contra Kiss 1986, 1992, among others) that multiple fronting in Yiddish is not simply ‘visible LF movement’, as the two types of multiple questions have different (syntactic) properties. In fact, Yiddish multiple questions might even be construed as an argument against the LF movement of wh-in-situ, since the two options for question formation do not show the expected syntactic parallels. But is there any semantic motivation for LF movement of wh-phrases? I turn now to the implications of the discussion above for the interpretation of multiple questions in Yiddish. The analysis given so far implies that the existence of the two syntactic strategies for multiple questions means that Yiddish must construct two possible types of LFs for wh-questions. One way of doing this would be to follow ideas in Reinhart (1998) concerning the interpretation of wh-in-situ. Before outlining the picture for Yiddish, I will briefly review the relevant features of Reinhart’s results. Reinhart starts from the minimalist viewpoint, assuming that LF movement of in situ wh-phrases is in fact ruled out by economy (that is, generalized

145

146 Molly Diesing

pied-piping occurs only when it is ‘audible’ in the output, Chomsky 1995, p. 265). The end result is that syntactic wh-in-situ must also be interpreted in situ in some fashion. Thus, in an example involving two wh-phrases in a single fronting language like English, the scope of the in situ wh-phrase must be assigned without movement, perhaps using a version of ‘unselective binding’ via an existential quantifier in the form of an abstract ‘Q-morpheme’ (Pesetsky 1987).11 Reinhart illustrates the in situ interpretation as follows: (22) a. Who bought which book. b. for which ·x, yÒ, x bought y and book y (Reinhart 1998, (9)) What is essential to note here is that the restriction on the variable y remains unmoved, and is bound by the existential quantifier nonetheless. Reinhart argues that this leads to a serious problem of interpretation in sentences such as: (23) Who will be offended if we invite which philosopher? In this case, the unmoved wh-phrase occurs in an if-clause. Reinhart gives the resulting logical form as: for which ·x, yÒ, if we invite y and y is a philosopher, then x will be offended. b. {P | $·x, yÒ & P = ^((we invite y and y is a philosopher) Æ (x will be offended)) & true(P)} c. Lucie will be offended if we invite Donald Duck. (Reinhart 1998, (16))

(24) a.

The representation given by in situ interpretation is given in (24 a), and (24 b) gives the semantics of the question in terms of a set of propositions denoting the true answers to the question, along the lines of Karttunen (1977). Given this, (24c) ought to be one of the possible answers to (23), ‘since if Donald Duck is not a philosopher, it must be true of him that if he was a philosopher and we invited him, Lucie would be offended’ (Reinhart 1998, p. 37). The source of the problem is the fact that the restriction for the variable y is in the antecedent clause of an implication; therefore, anything that is not a philosopher can be a value for y. This is clearly not what the question in (23) really means. I will henceforth refer to this problem as ‘the Donald Duck problem’.12 As a solution to the problem of how to assign wide scope to wh-phrases which otherwise appear to remain in situ (and thereby derive the correct reading for (23)), Reinhart employs existential quantification over choice functions. A choice function is simply a function which applies to a non-empty set and yields an arbitrary member of that set. Applied to example (23) above, this strategy yields:13


(25) a. Who will be offended if we invite which philosopher? b. for which ·x, f Ò, if we invite f(philosopher), x will be offended. c. {P | $·x, fÒ (CH(f) & P = ^((we invite f(philosopher)) Æ (x will be offended)) & true(P))} Here the in-situ wh-phrase is interpreted as a choice function applied to the set of philosophers. The function variable f can be existentially bound from arbitrarily far away, and thus is able to be given wide scope. Since the choice function applies only to the set of philosophers (which presumably does not include Donald Duck), the Donald Duck problem does not arise, despite the fact that the restriction occurs in an if-clause.14 This gives us the correct reading: ‘for which pair of individuals x and y, where y is a philosopher, and if we invite y, x will be offended?’. Note that the wh-phrase that has moved is not interpreted as a choice function. Reinhart claims that the moved wh-phrase in fact cannot take the form of a choice function, since it is no longer in an argument position and cannot serve as an argument of the form f(person), unless it is reconstructed. The moved wh-phrase is simply interpreted as an existential quantifier, resulting in a heterogeneous interpretation for multiple questions — the two wh-phrases are interpreted by different mechanisms. Turning back to Yiddish, the interpretation of the multiple fronting questions turns out to be quite straightforward. Since all of the wh-phrases do in fact front in the syntax, they can be simply interpreted ‘as is’, as quantifiers. There is no LF movement necessary, and since no restrictions are left in situ, Reinhart’s Donald Duck problem doesn’t arise. (26) a.

ver vos hot gekoyft? who what has bought What did who buy? b. for which ·x, yÒ, (person (x) and thing (y)), (x bought y) c. {P | $x $y (person(x) and thing(y) & P=^(x bought y) & true(P))}

In the case of the multiple questions involving syntactic fronting of a single wh-phrase, the moved wh-phrase is interpreted as an existential quantifier, as in the English examples. The remaining wh-in-situ is interpreted with a choice function, in the manner outlined by Reinhart. Here again there is no LF movement. Consequently there is also no worry about the superiority effect arising at LF either. Thus, for both of the possible forms of a double question, both orders receive a wellformed interpretation: (27) a.

vos hot ver gekoyft? what has who bought “What did who buy?” b. for which ·x, f Ò, ((thing(x)) and (f(person) bought x)

148 Molly Diesing

c.

{P | $ ·x, f Ò, (CH(f) and thing(x) & P = ^(f(person) bought x) & true(P))}

(28) a.

ver hot vos gekoyft? who has what bought “Who bought what?” b. for which ·x, f Ò, (person(x)) and (x bought f(thing)) c. {P | $ ·x, fÒ, (CH(f) and person(x) & P = ^(x bought f(thing) & true(P))}

Thus, wh-movement can be reduced to something motivated on purely syntactic grounds, since phrasal movement is not required for a well-formed interpretation. This is an idea that has been proposed by Chomsky (1995a), in which wh-movement serves to mark a construction as interrogative — in other words, it has purely a clausetyping function. On this scenario the difference between multiple fronting and single fronting languages is also purely syntactic — it arises from the value of the [±multiple specifiers] feature. As pointed out by Stechow (1996), this may in fact be one of the strongest arguments for the use of choice functions in wh-questions, in that they make a purely syntactic view of wh-movement semantically viable. There are still a number of loose ends in the area of interpretation. One of these concerns the behavior of adjuncts. Recall that in Yiddish the adjuncts ‘how’ and ‘why’ cannot appear in multiple fronting constructions, but can quite legitimately occur in single fronting questions, even remaining in situ: (29) a.

vi azoy hot ver gezungen? how has who sung “Who sang how?” b. ver hot vi azoy gezungen? who has how sung “Who sang how?” c. *ver vi azoy hot gezungen? who how has sung d. *vi azoy ver hot gezungen? how who has sung

(30) a.

farvos hot ver ongeklungen der mamen? why has who called the mother “Who called the mother why?” b. ver hot farvos ongeklungen der mamen? who has why called the mother? “Who called the mother why?” c. *ver farvos hot ongeklungen der mamen? who why has called the mother


d. *farvos ver hot ongeklungen der mamen? why who has called the mother The examples in (29b) and (30b) show that it is quite possible for wh-adverbials to remain in situ. This, however, is not the case in English: (31) a. *Who fainted when you behaved how? b. *Who sneezed why? In an account which does not make use of LF movement, these facts cannot be reduced to the ECP (as in Huang 1982), and they do not clearly reduce to the classic structural notion of superiority, since the moved subject wh-phrase is in fact superior to the in-situ adjunct. Reinhart notes that this problem in English is limited to purely adverbial wh-phrases — replacing how with (in) what way gives a grammatical result: (32) Who fainted when you behaved in what way? Reinhart explains the English judgments by appealing to the idea that wh-adjuncts are semantically different from wh-NPs in that they do not denote functions ranging over a set of individuals, but instead denote functions ranging over higherorder entities (following ideas in Szabolcsi and Zwarts 1993). Reinhart claims that if this is the case, the wh-adjuncts cannot be interpreted as choice functions, which select an individual from a set, not some higher-order entity. In short, the denotation of wh-adverbials is not suitable for a choice function interpretation (see Hornstein 1995 p. 147 ff. for a similar claim). Thus, wh-adjuncts are interpretable only in SpecCP. In (32), the adjunct is in fact an NP, so it can be interpreted in situ. Taking a closer look at this explanation of the English facts, one question that might be raised is whether the NP-adjuncts themselves actually denote sets of individuals, or whether they too might range over higher order entities.15 In other words, it is not clear to me that the contrast between (31a) and (32) is in fact due to a semantic difference between the two in situ wh-phrases, or whether the difference is simply the syntactic difference between NP and adverb. Taking the view that it is actually a syntactic contrast points the way to an alternative explanation: the restriction might simply derive from the fact that the choice function interpretation is limited to argument positions. In other words, adjuncts are prohibited from having a choice function interpretation in English in the same way that the choice function interpretation is blocked for moved wh-phrases, the prerequisite of being in an argument position is not met. The Yiddish facts raise a more significant problem in that Yiddish wh-adverbials not only can remain in situ, but also cannot be multiply fronted (to receive a quantificational interpretation). Thus, it must be the case that they can receive a choice function interpretation. If the choice function interpretation is restricted

150 Molly Diesing

semantically in the manner that Reinhart suggests, this is totally unexpected.16 Taking the syntactic approach to restricting choice functions would require that Yiddish does not require that choice functions be in argument positions. In other words, the restriction would be allowed to vary parametrically from language to language. As such, the restriction would amount to an ‘interface condition’ (a condition on translating the syntactic representation into an interpretation) which would be arguably more in the spirit of the Minimalist Program than a semantic parameter which would allow adjuncts in Yiddish to denote a different sort of semantic entity than in English. The issues raised by the interpretation of multiple questions in Yiddish point to the need to look more closely at what constraints need to be formulated in regard to the deployment of choice functions in natural language semantics. Choice functions are an attractive, but extremely powerful tool. Investigations of their application to the interpretation of indefinites has yielded a range of possible approaches, from the relatively unrestricted use seen in Winter (1997) and Reinhart (1997), to the more constrained theories of Kratzer (1998) and Matthewson (1999) (the latter based on the crosslinguistically rather unusual behavior of indefinites in St’át’imcets).17 It is reasonable to suppose that choice functions in questions may need to be constrained as well, perhaps with interesting cross-linguistic implications.

7.

Final remarks

In summary, Yiddish allows both single and multiple fronting in multiple questions, with the multiple fronting cases showing superiority effects while their single fronting counterparts do not. Additionally, adjuncts and D-linked wh-phrases are unable to undergo multiple fronting. I have proposed an explanation of these syntactic facts based on the idea that Yiddish allows both [+multiple] and [−multiple] C-Specs. This [±multiple] feature has been argued to be the locus of crosslinguistic variation in superiority behavior (Pesetsky 2000). The variability in Yiddish multiple question constructions results from the possibility of being able to choose freely between the two types of CP specifiers. The lack of superiority effects in single fronting questions is linked to the fact that Yiddish also allows scrambling of noun phrases; this scrambling (which happens prior to wh-movement) is what enables single fronting wh-questions to apparently evade the effects of superiority. This option is not available in the multiple fronting cases because the [+multiple] CP specifier is subject to an additional restriction in that movement to a [+multiple] C-Spec must be from an A-position, while movement to a [−multiple] C-Spec is unrestricted:


(33) Yiddish Multiple-Fronting Constraint: In Yiddish, movement to a [+multiple] CP must originate from an A-position. This asymmetry between the two types of specifiers in Yiddish is what produces the asymmetric superiority effects. One question that arises is whether the restriction on multiple fronting in (33) is language-particular, or a universal property of multiple fronting. Is the Yiddish Multiple-Fronting Constraint simply The Multiple-Fronting Constraint? I don’t have a definitive answer to this question, but instead offer some preliminary speculations. A piece of evidence in favor of (33) having universal force is the fact that the languages which front multiple wh-phrases to CP all display clear superiority effects (Rudin 1988). What is less clear is the behavior of adjuncts and D-linked wh-phrases. As I noted in footnote 4, Rudin has claimed that adjuncts cannot undergo multiple fronting in Bulgarian, but Boškovic´ (1998) and Grewendorf (2001) present evidence to the contrary. As far as D-linked wh-phrases are concerned, there is a preference for leaving these in-situ in at least some of the other multiple-fronting languages. A factor that interferes with a direct comparison regarding the behavior of adjuncts and D-linked wh-phrases in the languages which front multiple wh-phrases to CP is the fact that Yiddish appears to be alone among them in also having a robust single-fronting alternative. It may be that it is this possibility that allows for the strict enforcement of the Multiple Fronting Constraint in Yiddish, as it provides a grammatical alternative for questions involving adjuncts and D-linked wh-phrases. Thus, the strong form of the constraint (movement to a [+multiple] CP only from an A-position) might be expected to be attested only in languages which allow for a free choice between [±multiple] CPs, with a weaker form requiring only that the wh-phrases be moved to the [+multiple] CP from their base position (corresponding to the classical superiority judgments without the Yiddish restrictions on adjuncts and D-linked phrases) holding in the ‘pure’ [+multiple] CP languages like Bulgarian. Clearly further research is need in the area of typological variation in multiple fronting, but my hope is that this brief examination of the Yiddish facts opens up some new avenues of inquiry.

Notes * It gives me great pleasure to dedicate this chapter to Eloise Jelinek, who has been an exemplary colleague, an inspiring collaborator, and a very good friend to me over the years. An earlier version of this paper was presented at the SULA Conference at the University of Massachusetts at Amherst in May 2001. I am grateful for the comments and suggestions I received there, as well as conversations and correspondence with Chris Collins, Zoltán Gendler Szabó, and Kerstin Hoge, and comments from Andrew Carnie, Heidi Harley and two anonymous reviewers. Finally, I’d also like to thank my consultants, Aaron Lerner, Rose Rosen, and Harry Zuckman (sadly, all now deceased).

151

152

Molly Diesing

1. Hoge (2000) reports three dialect variants with respect to multiple fronting: i. multiple fronting is possible, but not obligatory in both direct and indirect questions. ii. multiple fronting is only possible in indirect questions iii. multiple fronting is not possible at all The speakers I have worked with have been mainly of type (i), one consultant was of type (ii). 2. Irene Heim (p.c.) raised the question of whether this is in fact an effect of D-linking, or whether perhaps ‘heaviness’ may in fact be at stake. Testing heavier wh-phrases which are not clearly D-linked yields somewhat unclear results: multiple fronting constructions containing wh-phrases such as vifl bikher ‘how many books’ were accepted by some speakers, while wh-phrases such as vos far bikher ‘what kind of books’ were less acceptable. Interestingly, one speaker (brought to my attention by Kirsten Hoge) has the exact opposite judgments. In any case, the clearest cases of ungrammaticality are those involving D-linked wh-phrases. 3. This may also be true of Bulgarian (see Rudin 1988, fn. 21), though there is contradictory evidence on this point (see e.g. Boškovic´ 1998, Grewendorf 2001, fn. 4). 4. Yiddish wh-movement does obey islands, unlike Bulgarian. This means that Yiddish doesn’t fit neatly into Rudin’s (1988) typology. While this classification of multiple fronting languages has served as a guide for a number of researchers (such as Richards 1997), Yiddish is not the only multiple fronting language with a poor fit — see for example Golden (1995) on Slovene. 5. In double object constructions where a prepositional indirect object appears, the reverse pattern appears, the direct object is superior to the indirect object: (i) a.

vos tsu vemen hot Maks geshikt? what to whom has Max sent “What did Max send to whom?” b. *tsu vemen vos hot Maks geshikt? to whom what has Max sent

6. Conversations with Chris Collins were instrumental in shaping my thoughts regarding this section. 7. This is not to suggest that this matter is entirely uncontroversial. See for example Hornstein (1995), Reinhart (1998), Hoge (2000) and Grewendorf (2001) for consideration of alternative accounts of superiority. 8. Richards also notes a ‘Minimal Compliance’ effect with superiority, in that superiority only needs to be satisfied by the first wh-phrase moved. Subsequent wh-phrases may move in any order. Testing this effect requires multiple questions involving at least three wh-phrases, and my consultants tended not to like multiple questions with more than two wh-phrases, so I have not been able to determine whether this effect is seen in Yiddish wh-movement. 9. The two approaches may well make different predictions, see Hoge (2000) for a discussion of the differences as they apply to multiple questions in German. As far as the Yiddish data are concerned, while I do not deal with the matter of dialect variation with respect to multiple wh-fronting in Yiddish in this paper (see footnote 2 above), I suspect that the [±multiple specifiers] approach may provide a more natural means of accounting for these differences than the Rizzi-inspired split CP analysis. 10. Essential to this line of explanation is that the focus movement mentioned in Section 1 not be a case of A-bar movement. More research is needed here, but this movement is in fact quite restricted (only one phrase can occupy this focus position), and does not appear to display any of the traditional properties of A-bar movement (see Diesing 1997 for further discussion). In any case, this shorter leftward movement may well correspond to cases of ‘short scrambling’ which


have in fact been claimed to be A-movement, in contrast to A-bar, longer-distance scrambling (see vanden Wyngaerd 1989, Mahajan 1990, Bailyn 2002, among others). 11. References to unselective binding in this context often compare it to the interpretation procedure in Heim (1982). Presumably what it meant here is Heim’s analysis in her Chapter 2, but this is substantially different from what Pesetsky and others seem to mean by ‘unselective binding’ in that LF movement (in the form of the NP-Prefixing operation) is crucially involved. 12. Note that this argument depends on an analysis of the conditional as material implication. 13. As Reinhart notes, things are a bit more complicated than this, since the extensionality of the choice functions must be insured. See Reinhart (1998) Section 2.3.2. 14. A potential problem for a choice function analysis of questions pointed out to me by Zoltán Gendler Szabó is that not all question answers are chosen individuals: Q: A:

Which boy will read which book? Max will read any book on puffins.

This may not be an insoluble problem, since there is the possibility of intensional choice functions; see Stechow (1996), who discusses a similar example raised by Irene Heim in her 1994 lecture notes (I have not myself seen the Heim notes). 15. An alternative suggested to me by Heidi Harley is that this contrast may result from presuppositions associated with adverbial PPs like ‘in what way’ which do not arise with true adjuncts like ‘why’. 16. Yiddish is not the only language that raises this particular problem; as is well known, German also allows adverbial adjunct wh-phrases to remain in situ, unlike English. 17. Another sort of question that remains to be investigated is the fact that (as far as I know) there are no anaphoric instances of choice functions.

153

Attitude evaluation in complex NPs* Lynn Nichols University of California, Berkeley

1.

Syntactic and semantic properties of complex NPs

The study of propositional attitudes has largely been concerned with their semantic properties. This study makes the argument that the semantic properties of attitudes may provide an explanation for aspects of the syntactic and morphosyntactic structure of certain complex NP constructions in English and cross-linguistically. In other words, the semantic properties of propositional attitude evaluations play more than an interpretive role in the grammar, they determine properties of formal structure. 1.1 The structure of complex NPs The type of complex NP at issue is illustrated in (1a,b); these have commonly been referred to as NP complement constructions, based on the assumption that the subordinate clause receives a thematic role from the noun, paralleling the thematic properties of the clause in the related verbal construction, (1c). (1) a. Sonia made the claim that she had bought the lottery ticket. b. John ignored Sonia’s claim that she had bought the lottery ticket. c. Sonia claimed that she had bought the lottery ticket. Traditionally contrasting with this type of subordinate clause, relative subordinate clauses are adjuncts with respect to the head nominal. In (2), the NP (DP) the claim corresponds to an argument position inside the subordinate clause, while the claim in (1a) ostensibly does not. (2) John ignored the claim that Sonia had made. It is therefore interesting that while extraction out of a theta-marked complement clause is ordinarily possible (excluding factive cases), (3a), extraction of a wh element out of a complement to a complex NP is not possible, (3b). The ungrammaticality of wh extraction out of complex NP clauses is traditionally argued to be the result of a subjacency violation; that is, movement out of the complement would cross more than one bounding node (NP and IP; Chomsky 1986a).

156 Lynn Nichols

(3) a. Which ticketi did John think that Sonia had misplaced ti? b. *Which ticketi did Sonia deny the claim that she had misplaced ti? Several types of evidence suggest that the standard view of this type of complex NP is inaccurate. Specifically, there may be no thematic relationship between the head N and the associated subordinate clause. This evidence is discussed in detail in Nichols (2001); I briefly summarize here. First, not all nouns in this construction derive from verbs (story, idea, fact); it is unclear why (and unlikely that) this latter set of basic nouns would be able to assign a thematic role only in this particular context.1 According to Baker (2000), the defining property of elements of the category N is that they are referential index (theta role) receivers and are unable to assign referential indices (theta roles) themselves. The latter is a property only of elements of the category V. Furthermore, the standard argument that (deverbal) nouns may (optionally) assign theta roles is in actuality based on the stipulation that the complement theta role is carried over from the verb. There is no direct evidence for this position in the case of subordinate clauses, and, moreover, (as I will argue below) there are other ways in which instances of apparent theta assignment by nouns may be explained. We might also consider the case of inalienable possessors: though these do appear to bear a theta role associated with the head N, this theta role may be licensed in the DP by additional functional structure, not the N itself; note that alienable possessors, which would not have this functional structure, behave in some languages (e.g., Turkish) like attributive elements (Balkiz Öztürk, p.c.). Second, consider the patterning of referential NP movement. Rizzi (1990) argued that referential NP movement is licensed by binding, consequently referential NP movement is possible under certain circumstances where adjunct movement would be prohibited (i.e., long-distance movement and movement over intervening wh-elements). Referential NP movement is excluded, however, out of an adjunct, a domain that is not theta-marked by an element in the domain that the NP is moving to. Thus movement of the NP which uncle out of the relative clause adjunct in (4a) or the adverbial adjunct in (4b) is ungrammatical. (4) a. *Which unclei did John give away the lottery ticket that Sonia had bought for ti? b. *Which unclei did John stay at home since he didn’t want to visit ti? This behavior of referential NPs suggests the hypothesis that the ungrammaticality of referential wh extraction in (3b) may be due to the fact that the subordinate clause in (3b) is a type of adjunct clause. If the subordinate clause in (5a) below is an adjunct, then a further reasonable hypothesis is that this is a relative clause adjunct, with a structure similar to that of the relative clause in (5b,c). In (5b). the nominal head corresponds to a gap in the subordinate clause associated with an argument position (here direct object). In (5a) there is no nominal argument

Attitude evaluation in complex NPs

position gap, although there is another argument available in the subordinate clause for relativization. This is the event argument (or event variable; Davidson 1967, Higginbotham 1985, Parsons 1990, Rothstein 1995), in (6). (5) a. The claim [that Sonia had misplaced the lottery ticket] b. The train [that John missed] c. [DP The traini [CP Øi [that [IP John missed ti]]]] (6) Sonia bought a lottery ticket in Rhode Island. $e[buying(e) & Agent(Sonia)(e) & Theme(lottery ticket)(e) & in(RI)(e)] (6) can be paraphrased as ‘There is an event which is a buying whose agent is Sonia whose theme is a lottery ticket and which takes place in Rhode Island’. The subordinate clause associated with the noun claim in (5a) may be represented as a relativization of this event argument and therefore as a canonical relative adjunct,2 as in (7): (7) [DP The claimi [CP Øi [that [IP Sonia [ei [had bought the lottery ticket]]]]]] Interestingly, we can observe that this type of relative adjunct does not occur with all nouns but is limited to a certain semantic class. This limitation is explained by the nature of this relative adjunct itself, specifically, by the coindexation between the lexical head noun and the event argument of the subordinate clause: the formation of this type of relative clause occurs only with that subclass of nominals whose lexical meaning contains an eventuality descriptor. Thus even though the subordinate clauses in (8a,b) each contain an event argument, these relativizations are ungrammatical because trail and studiousness are not possible descriptors of an eventuality. (8) a. *The trail that John climbed Mt. Moosilauke b. *The studiousness that the children all completed their homework Likewise the nouns cause, event and issue do not allow finite clause adjuncts (9)–(11), though they do allow other kinds of modifying adjuncts such as PPs or subordinate relative clauses that do not involve event argument relativization, (12). (9) a. The reason that Sonia left town. b. *The cause that Sonia left town. (10) a. The fact that the circus came to town b. *The event that the circus came to town. (11) a. The idea that Sonia would be able to do such a thing. b. *The issue that Sonia would be able to do such a thing. (12) a. The cause of his leaving town b. The event that pleased us most

157

158

Lynn Nichols

c. The event of the circus coming to town d. The issue that everyone wanted to discuss e. The issue of John’s taking part in the protests 1.2 The absence of a semantic restriction in complex NPs Recent work by Aoun & Li (2002) on Chinese relative clause structure bears further on this issue. Aoun & Li argue that Chinese NP-relative clauses such as that in (13) have the structure in (14). This structure is different from the operator structure assumed earlier for English relative clauses. In Chinese, the relativized NP itself is argued to move from inside the relative phrase to a position external to the clause; the relative clause is an adjunct to the externalized NP. (13) ta kan-guo de na/yi-ben shu he read-asp DE that/one-cl book “(the/a) book that he has read” (Aoun & Li 2002) (14) [NP [CP […[IP ti ]…] ] [Head NP]i ] This analysis makes a very interesting prediction in light of the arguments just made for the structure of complex nominals. The form of Chinese relative clauses in (14) predicts that clauses in which the event argument is relativized cannot be formed in Chinese, since the event argument itself, unlike NP arguments, is unable to move (cf. Rothstein 1995 for arguments). The semantic class of nouns occurring in complex nominals is restricted in those languages in which the noun is coindexed with an event argument, but in Chinese we should expect to find no semantic restriction on the class of nouns occurring in complex nominals. This prediction appears to be fulfilled. Consider the following clause + de + N construction from Mandarin Chinese in which the head noun is not associated with an NP argument position in the subordinate clause. Both event descriptor nouns such as keneng ‘possibility’ and shi ‘fact’ (15a,b) as well as non-descriptor nouns such as yuanyin ‘cause’ or shi ‘event’ (15c,d) are found in this construction. (15) a.

ta chucheng de keneng he leave.town DE possibility “The possibility of his leaving town” b. ta shi zhongguo ren de shi he be Chinese person DE fact “The fact/issue of his being Chinese” c. ta chucheng de yuanyin he leave.town DE cause “The cause of his leaving town”

Attitude evaluation in complex NPs 159

d. ta chucheng de shi he leave.town DE event “The event of his leaving town” The fact that both descriptor and non-descriptor nouns occur in this nominal construction, in contrast to English, supports the conclusion that the associated clauses are not relativized clauses involving the event argument. (The fact that an NP occurs in the clause external position in the examples in (15) also suggests that the event argument does not occupy in that position.) The examples in (15) are instead comparable to the English gerund formations in (12a,c,e) and involve the adjunction of a gerund-like clause in which there is no coindexation between some internal event variable and some external element. Consequently there is no transfer of lexical properties from the head noun to the event variable. In contrast, the fact that there is a semantic restriction in English on the class of nouns that may occur in this complex NP construction (cf. (9)–(11)), supports the idea that (9a), (10a) and (11a) involve the sort of relativized event argument structure proposed. 1.3 Attitude evaluation in nouns We can be even more specific about what is meant by the property of being an ‘event descriptor’. Examination of the (abbreviated) list of this class of nouns in (16) indicates that the English nouns compatible with this type of event-relativization all contain some sort of attitude evaluation. These nouns therefore have properties similar to those of propositional attitude predicates, namely that of specifying an evaluation set of worlds for an associated proposition (Hintikka 1969). Thus rather than the traditional complex nominal for the constructions in (1a-c), the term attitude nominal more accurately characterizes their semantic and syntactic properties; attitude noun characterizes the head of these nominals. While some of the attitude nouns in (16) are basic nouns (fact, hunch, story, rumor) and others are derived from verbs (discovery, knowledge, assertion, belief), I assume that the two types of nouns do not behave differently with respect to the inability to assign a theta role. Despite deriving from verbs, the deverbal nouns are just that — nouns — and do not have a theta role to assign, following the criteria of Baker (2000). (16) Attitude Nouns Factive fact regret discovery knowledge realization

Non-Factive rumor fear claim opinion notion hunch story assertion belief hope

160 Lynn Nichols

Like the complement of attitude predicates, the subordinate clause associated with an attitude noun has either factive or non-factive interpretation: in the case of factive clauses, the truth of the proposition is presupposed by the speaker, while in non-factive clauses, the speaker makes no commitment to the truth or the falsity of the proposition (Kiparsky & Kiparsky 1971). In other work (Nichols 2001) I have argued that in the non-factive cases, the subordinate clause gets its interpretation from the head noun (transmitted by the coindexation between head N and event argument);3 factive nominals, on the other hand, get their presuppositional interpretation from the discourse common ground, the default evaluation set. In the remainder of this paper I wish to discuss certain crosslinguistic morphosyntactic evidence for the presence of an attitude evaluation in the subordinate clause associated with these nouns. An attitude evaluation expresses a relationship between a particular individual and a proposition (Montague; Hintikka 1969). It is this relationship that distinguishes both factive and non-factive attitudes such as regret and believe from factive non-attitudes such as happen and occur.4 We should therefore be able to find morphosyntactic evidence that even factive attitude nominals contain an evaluative semantic component. In what follows, I will argue for the presence of an attitude evaluation in the subordinate clause associated with an attitude nominal, demonstrate that this atitude evaluation is a semantic component of both in non-factive and factive attitude nominal constructions, and finally, argue that the absence of a restriction to attitude noun heads in certain languages indicates that complex nominals in such languages are not formed via relativization of the event argument.

2.

Evaluative morphology in clauses associated with nouns

The behavior of the subjunctive in English provides a useful starting point. Giorgi & Pianesi (1997) argue that the subjunctive is used to mark the divergence between an evaluation set of worlds (= set of propositions) and the set of presuppositions making up the conversational common ground. In other words, the presence of the subjunctive marks the presence of some attitude evaluation. In this light, note that in English the subjunctive form of the verb is obligatory in certain cases (and a possible choice in others cases) in the subordinate clause associated with attitude nominals.5 (17) a. The decision that John be given a promotion b. *The decision that John was/is/will be given a promotion (18) a. Nancy’s request that Joe be allowed to leave b. *Nancy’s request that Joe was/is/will be allowed to leave Further data from Turkish and Burmese is particularly informative. In both of these languages, attitude as well as non-attitude nouns may be associated with a subordinate

Attitude evaluation in complex NPs

clause. As was the case for English as illustrated in (9)–(12), however, the forms of these two categories of subordinate clauses are different. Most crucially, the subordinate clause associated with the attitude noun bears additional evaluator morphology that the non-attitude nominal clause does not. Taking the case of Turkish first, many types of subordinate clauses in Turkish are form via nominalization. The examples in (19) illustrate three of these nominalizers, which pattern in two groups. Öztürk (2001) argues on the basis of event nominal derivation that the nominalizer -Is¸ attaches to the clause not very far above VP, hence these nominalizations lack certain functional projections of the clause. The nominalizers -DIK and -mA attach further up in the clause structure and not only include but likely themselves represent this additional functional structure. Based on their distribution in the new data below, -DIK and -mA might in addition be called attitude evaluation morphology: they appear only in (19b-d), i.e., inside the subordinate clauses associated with the attitude nominals gerçeg ‘fact’ and söylenti ‘rumor’.6,7 Neither evaluator nominalizer appears in the clause associated with the non-attitude noun durum ‘case, situation’ in (19a); -Is¸ is found here instead. Sirk-in s¸ehr-e gel-is¸-i durum-u-nda circus-gen town-dat come-Is¸-3sg case-3sg-loc “The case/situation/event of the circus coming to town” b. Ali-nin s¸ehir-den git-tig-i gerçeg-i A.-gen town-abl go-DIK-3sg fact-3sg “The fact that Ali left town; The fact of Ali’s leaving town” c. Ali-nin s¸ehir-den git-tig-i söylenti-si A.-gen town-abl go-DIK-3sg rumor-3sg “The rumor that Ali left town; The rumor of Ali’s leaving town” d. Ali-nin s¸ehir-den git-me-si-nin neden A.-gen town-abl go-mA-3sg-gen reason “The reason that Ali left town; The reason of Ali’s leaving town”

(19) a.

Thus there are differences in the morphological marking of nominalized subordinate clauses in Turkish that may plausibly be accounted for by the semantic properties of the head noun. Even in a nominalizing language different types of nominalization are required for subordinate clauses associated with attitude vs. non-attitude nominals. Burmese presents a case similar to Turkish in having additional evaluator morphology inside the attitude nominal subordinate clause. Moreover, attitude nominal subordinate clauses in Burmese are formally similar to relative adjuncts and unlike finite subordination (Burmese explicitly contrasts these two types of subordinate clauses). First, note that the subordinate clauses associated with factive attitude noun Á6thi ‘knowledge’ (20a) as well as with non-factive kauláhalá ‘rumor’ (20b) contain the attitude verb hsou ‘say’. ((21) illustrates a main clause usage of

161

162 Lynn Nichols

this verb.) In contrast, the nominalization in (20c) involving a non-evaluative head noun does not (and may not) contain hsou. The evaluative property of hsou may be further illustrated in its evidential usage in factive contexts to convey that a proposition, while presupposed true by the speaker, is information gained from another source, cf. (22a) vs. (22b); (22b) is the protasis of an epistemic conditional. Evidentials fall among the class of attitude evaluators. (Data in (20–22) from author’s field notes.) [[thu chàn tha te] [hsou té] Á6thi Á6 [[he rich realis [eval.] rel.realis nom-know “The knowledge [of the fact that] he is rich” b. [[thu nei m6 kàun hpù] [hsou té] kauláhalá [[he stay neg good neg [eval.] rel.realis rumor “The rumor that he is/was ill” c. [gyou hna‘ lòun tai‘ té] [planet two classifier hit rel.realis ‘ 6-hpyi‘-66-pye‘ nom-happen-nom-destroy “The event [creation-destruction] of the two planets colliding”

(20) a.

(21) hsou pa òun say polite hort “Please tell me more” (22) a.

thu sa mèi bwè cá tó 3sg.exam drop since “Since she failed the exam” b. thu sa mèi bwè cá te hsou yin 3sg.exam drop realis say if “If she failed the exam [and we know she did]”

Additionally, the subordinate clause itself in these complex nominal constructions appears in the form of a relative clause, with relative suffix -té or -mé; compare the canonical relativized subject clause in (23a). The relative clause forms in the attitude nominals in (20a,b) explicitly contrast with finite complementation which makes use of the complementizer lóu, (23b), and nominalized complementation with nominalizing suffix -hta/-da, (23c). hou thòu thwà té we‘ thà hìn that rancid go rel pork curry “That pork curry which has turned rancid” (Soe 1999) b. thu lou‘ hcin te lóu yu hsá te she do want realis comp consider realis “I consider/assume that she wants to do (it)” (Soe 1999)

(23) a.

Attitude evaluation in complex NPs 163

c.

thu ku nyi pèi nain nta kou mìn thí là he help give can nom obj.you know Q “Do you know that he can help?” (Soe 1999)

If we are indeed dealing not simply with complex nominals but with a class of attitude nominals, then the complex distribution of nominalization types in Burmese and Turkish falls into place in a principled manner. Without the notion that these complex nominals contain an evaluative relative adjunct in their subordinate clause, the morphological patterning in (19) and (20) above is difficult to explain.

3.

Concluding remark

I have argued here for a new look at certain complex nominal constructions, specifically that these are better characterized as attitude nominal constructions. Following the proposal that these constructions involve the relativization of the event argument, I have presented morphosyntactic evidence that the proposition in the subordinate clause in these constructions is associated with an attitude evaluation. It is hoped that the characterization of attitude nominals as such will open up this class of phenomena as an important area for further crosslinguistic study.

Notes * I am extremely grateful to be able to dedicate this research to Eloise Jelinek, who has always been an immense source of inspiration. 1. A reviewer suggests that this argument may not hold in a model of derivation such as Distributed Morphology where the lexicon consists of roots and syntactic category is determined by syntactic position. I assume following Baker (2000), however, that basic semantic content distinguishes categories like nouns and verbs (the ability to bear a referential index (= theta role) and the ability to assign one, respectively). 2. I leave open here the issue of whether the relative operator moves from the position of the event argument (Rothstein 1995 claims that the event argument does not take part in syntactic chain formation) or the event argument is bound in situ. 3. Unlike predicates the head N itself cannot function as an operator over worlds (see discussion in Nichols 2001); coindexation between N and the relative operator corresponding to the event argument essentially turns the relative operator into an operator over worlds. 4. The noun fact is rightly included among the class of attitude nouns, since it establishes a presuppositional evaluation relative to some individual. Consider the following two dialogues, where denial of the presupposition associated with fact by the hearer is grammatical (i) but denial of the presupposition associated with the non-attitude happen is odd (ii). (i) A: B:

The fact that you don’t like me really bothers me But I do like you!

164 Lynn Nichols

(ii)

A: It just so happened that he won the lottery by buying only a single ticket. B: #But he didn’t buy a ticket!

5. The subjunctive per se is not found in English factive attitude nominals; according to Giorgi & Pianesi (1997) languages differ as to the degree of divergence from the common ground the subjunctive marks and apparently emotional/evaluative factive contexts fall within the indicative domain in English. It is the case, however, that at least some sort of modal marking may found in such contexts (Kiparsky & Kiparsky 1971, Farkas 1992), cf. our regret that John should do such a thing. 6. The precise distinction between -DIK and -mA is unclear. The traditional grammar terms of Factive Nominal for -DIK forms (Kornfilt 1997) is not quite right, since it occurs with both factive & non-factive evaluations, cf. [19b,c]. 7. The nouns reason and cause appear in most languages to have similar lexical content and differ precisely along the evaluator dimension: reason indicates a subjective evaluation as to cause, cause indicates an objective evaluation. It is often difficult for speakers to tease out their intuitions for the two cases, thus a more detailed consideration of the semantics of (19d) may reveal that neden is more akin to ‘cause’ and, consequently, a potential alternative interpretation of the data in (19) is that (19a) is a true gerund, -DIK marks subordinate clauses (such as (19b,c)) containing attitude evaluations, and -mA marks subordinate clauses (such as (19d)) not subject to attitude evaluation. Despite this uncertainty, the general point in the discussion still holds: under either interpretation of the data, Turkish appears to overtly mark clauses with regard to whether they are evaluated or non-evaluated.

Topic-Focus articulation and degrees of salience in the Prague Dependency Treebank* Petr Sgall, Eva Hajicˇová and Eva Buránˇová Charles University

1.

Introductory remarks

Since Eloise Jelinek has been interested in the issues of negation, focus and information structure, to the research of which she has contributed substantially, we want to use this nice occasion and present here partial results of an analysis of the TopicFocus articulation (TFA) of Czech and of the impact of these results on inquiries into coreference in coherent discourse. The aim of our paper thus is (i) to discuss the analysis of the TFA in a framework based on dependency syntax, illustrating the analyses by examples taken from a large corpus of Czech, and (ii) to examine a specification of at least some of the aspects of discourse cohesion, using a scale of degrees of activation of the elements of the stock of information shared by the speaker and the hearer(s). In Czech linguistics, the TFA has been systematically explored thanks to the classical Prague School of Functional and Structural Linguistics. As reflecting the ‘given — new’ strategy in discourse, the TFA has been considered to belong to the main objects of linguistic study. Continuing the widely known results gained by the Prague School since the 1920s, the explicit linguistic descriptive framework characterized in detail in Sgall et al. (1986), Hajicˇová (1993), Hajicˇová E., Partee B. and P. Sgall (1998) allows for the possibility of describing the TFA not only as concerning the intrinsic dynamics of the process of communication, patterned in the utterance (sentence occurrence), but also as constituting the structure of the sentence itself, i.e. grammar. Within this framework, the TFA is understood as one of the basic aspects of (underlying) sentence structure, which characterizes the sentence as a unit of the interactive system of language; the TFA thus is seen as a manifestation of the sentence being anchored in the context. The Praguian framework, described in detail in the quoted monographs, can be characterized quite briefly by the following two principles: (i) it stresses the opposition of unmarked (primary, prototypical, although not always most frequent)

166 Petr Sgall, Eva Hajicˇová and Eva Buránˇová

items of all levels and their marked (secondary, peripheral) counterparts, which only occur under specific contextual or other conditions; (ii) it understands the basic syntactic relation to consist in that of syntactic dependency, i.e. in the relation between a head word and its dependents (arguments and adjuncts). Let us recall that the underlying (tectogrammatical) representations (TRs) of sentences prototypically are dependency trees (only sentences containing, in addition, other than dependency relations, i.e. especially coordinated constructions, require representations of a more complex character, namely networks of more than two dimensions). In the ‘endocentric’ constructions, such as go slowly or old man, the head is specified as that item which cannot be deleted without a loss of the well-formedness of the construction, and ‘exocentric’ constructions can be similarly classified on the level of word classes, e.g. with the verb being present both in constructions which include an object and those which do not. The framework does not work with the concept of ‘constituent’. This makes it easier to account for the fact that, as our examples below document, most different combinations of sentence parts may belong either to Topic or to Focus. In the TRs, the left-to-right order of nodes, i.e. the underlying word order (also called the scale of ‘communicative dynamism’) starts with Topic proper and proceeds to Focus proper (the most dynamic part of the sentence, usually the bearer of the intonation center). The interplay of word order and of specific features of sentence prosody is the means by which the underlying word order is expressed (a more detailed characterization of the TRs is presented in Section 3 below).

2.

Topic-Focus articulation of the sentence (TFA)

The TFA is semantically relevant, as the following examples show: (1) a. I work on my dissertation on Sundays. b. On Sundays, I work on my dissertation. (2) a. We went by car to a lake. b. We went to a lake by car. (3) a. They moved from Chicago to Boston. b. They moved to Boston from Chicago. The semantic basis of the TFA may be seen in the relation of aboutness: a primary declarative sentence asserts that its Focus holds about its Topic. The Topic primarily consists of ‘contextually bound’ items (referring to what presystemically is called “given“ information), and the Focus contains ‘contextually non-bound’ (“new“) elements. Thus, on its preferred reading (and with the normal intonation, i.e., with the intonation center at the end of the sentence), sentence (1a) asserts about the

Topic-Focus articulation and degrees of salience in the Prague Dependency Treebank 167

speaker’s work on her/his dissertation that this takes place on Sundays, while (1b) asserts about Sundays that the speaker spends them working at her/his dissertation. In (2a) there are (at least) two possibilities: either it is asserted about a group including the speaker that they went by car to a lake, or it is asserted about their way by car that its goal was a lake. In (2b) to a lake is included in the Topic on all readings, and it is asserted that the transport took place by car. In a similar way, also with (3a) the two readings are present, which is not the case with (3b). The assignment of the TFA features is based on operational criteria such as the question test, according to which e.g. in (3b) from Chicago is understood as the Focus, since this part of the sentence is the counterpart of the interrogative element in the question (3c), which can be answered by (3b). The rest of the sentence, the content of which is “known“ from the question, is its Topic. (3) c.

3.

From where did they move to Boston?

TFA in the Prague Dependency Treebank (PDT)

After having been discussed and theoretically elaborated for several decades, these issues now are analyzed computationally in the Prague Dependency Treebank (PDT), based on the Czech National Corpus (CNC, containing hundreds of millions of word occurrences in journalistic fiction and other texts). The PDT research, which thus serves for checking and enriching the chosen description, comprises three layers of annotation: i.the morphemic (POS) layer with about 2000 tags for the highly inflectional Czech language, assigned by a stochastic automatic tagger (Hajicˇ and Hladká 1997, Böhmová and Hajicˇová 1999), with a success rate of more than 95%; on this layer, the Czech counterparts of the word tokens in sentence (1b) (O nedeˇlích pracuju na své disertaci) are characterized (with many simplifications), as: o ‘on’ — nedeˇlích ‘Sundays’ — pracuju ‘I-work’ — na — své — disertaci —

preposition, taking the Locative case noun, feminine, plural, Locative verb, 1st person, singular, present, indicative, imperfective preposition, taking the Locative possessive reflexive pronoun, singular, feminine, Locative noun, feminine, singular, Locative

ii.a layer of ‘analytic’ (“surface”) syntax (Hajicˇ 1998): approx. 100 000 Czech sentences have been annotated by a semi-automatic procedure; the result can be schematically characterized as the dependency tree in Figure 1. iii.the underlying syntactic level: tectogrammatical tree structures (TGTSs),


pracuju.Pred o.Prep

na.Prep

nedelích ¦ .Adv

disertaci.Obj své.Attr

Figure 1.

characterized below; among the main differences from the analytical layer, there is the treatment of function words as parts of complex word forms (preposition groups, periphrastical verb forms, etc.) and the more subtle classification of objects, adverbials and attributes; up to now, in an experimental phase, four samples of 50 sentences (of running texts) each have been annotated in full detail (the so-called ‘model collection’), and 20 000 sentences in what concerns syntactic structure itself (‘large collection’); out of this number, 2000 sentences have already received a treatment of their TFA). The TGTSs, which are based on the results of the long-term research in dependency based syntax in Prague (more details on the motivation were presented esp. in Sgall et al. 1986, as for TFA, see now esp. Hajicˇová et al. 1998), are much simpler than most of the constituency based structure layers of annotation (some of which do contain dependency based notions, distinguishing heads from their dependents, or modifiers, specifiers, arguments, adjuncts, etc., so that it may be asked whether the constituency based notions still are necessary). The TGTSs do not contain any nonterminal symbols; each of their nodes is labelled by a complex symbol composed of a lexical and a morphological part (values of morphological categories such as number, tense, modalities), and each edge is labelled by the symbol indicating a syntactic relation (i.e. the type of the dependent node, see point (iv) below). The main properties of TGTSs can be summarized as follows (note that TGTSs differ from TRs just in the technical handling of coordinating conjunctions, mentioned in point (i) below): i.

only autosemantic (lexical) words have nodes of their own; function words, as far as semantically relevant, are reflected by parts of complex node labels (with the exception of coordinating conjunctions); ii. nodes are added in case of deletions on the surface level; iii. the condition of projectivity is met (i.e., no crossing of edges is allowed), and the underlying word order often differs from the surface order also in other respects (e.g. in the order of an adjective and its head noun); iv. tectogrammatical functions (‘functors’) such as Actor/Bearer, Patient, Addressee, Origin, Effect, different kinds of Circumstantials are assigned;


v.

basic features of the TFA are introduced by means of one of the following three values assigned to every lexical (autosemantic) occurrence: t for ‘contextually bound’, CB (prototypically in Topic, T), c for ‘contrastive (part of) Topic’, f (‘non-bound’, NB, typically in Focus, F)

Let us illustrate the TFA values assignment by an example (with John and Ann bearing rising contrastive stresses and letter stressed with the typical sentence final falling pitch contour): (4) (What are the kids doing?) John.c is writing.f homework.f. and Ann.c is writing.t a letter.f. In unmarked cases, the main verb (V) and its direct dependents following it belong to Focus, they carry index f; the items preceding V carry t or c. In marked (nonprototypical) cases, the verb can be CB, i.e. in the Topic (if it has been mentioned or can be directly inferred from the context), or the Focus may precede the verb; usually the intonation centre (sentence stress) then marks the Focus, occupying a marked position. Dependents of nouns primarily are NB, i.e. they carry index f, even if they belong to the Topic of the sentence together with their head noun. In the underlying order, NB dependents follow and CB dependents precede their heads. Let us present the description of the TFA in PDT by the tectogrammatical analysis of a sample of sentences contained there, to illustrate how this approach makes it possible to analyze also complex sentences as for their TFA patterns, with neither Topic nor Focus corresponding to a single constitutent (argument or adjunct). In (5¢), which is a highly simplified linearized TGTS of (5), every dependent item is enclosed in a pair of parentheses; syntactic subscripts of the parentheses are left out here, for the sake of transparency, as well as subscripts indicating morphological values, with the exception of the two which correspond to function words, i.e. to the preposition v ‘in’ (connected with tomto, which is the Locative of tento ‘this’), expressing the syntactic value or ‘functor’ of Temp, and to the auxiliary musí ‘must’, expressing the morphological value of Necess(ity); Figure 2. presents the respective tree structure, in which three parts of each node label are specified, namely the lexical value, the syntactic function (with ACT for Actor/Bearer, RSTR for Restrictive, MANN for Manner, and OBJ for Objective), and the TFA value (we do not indicate the morphological values, such as those of Number, Tense, Modality); although the subject noun group is a name, we understand the adjunct in it to be NB, since it semantically specifies its head noun and the specification has not been mentioned in the preceding co-text. (5) Cˇeské radiokomunikace musí v tomto roce rychle splatit Czech Radiocommunications have in this year quickly to-pay


dluh televizním diváku˚m. debt (to) TV viewers. “This year, Czech Radiocommunications have to quickly pay their debt to the TV viewers” splatit f radiokomunikace t.ACT ceský ¦ f.RSTR

v_roce t.WHEN

rychle f.MANN

tento t.RSTR

dluh f.OBJ divák f.BEN televizní f.RSTR

Figure 2.

The (highly simplified) linearized form of the TGTS: (5¢) (radiokomunikace.t (Cˇeské.f)) ((tomto.Temp.t) rok.t) splatit.Necess.f (rychle.f) (dluh.f (diváku˚m.f (televizním.f))) The possibility of such a one-to-one linearization of the dependency tree in the form of a well parenthesized string of complex symbols is of fundamental importance. On the one hand, it may be maintained that a relatively natural image of the sentence structure, as internalized by speakers, comes close to the pattern based on rooted trees; in fact, sentence structure is more complex, since the combinations of the relations of dependency and of coordination require more dimensions than the two that are proper to the dependency tree. On the other hand, the strong restrictions of ‘projectivity’ (with no two edges crossing each other) and of a similarly limited repertoire of relationships between dependency and coordination (as well as apposition or parenthesis) lead to the possibility of the one-to-one linearization, the parenthesized strings of which come close to propositional calculus. This points to the possibility of describing the core of sentence structure (without non-prototypical features and subsystems such as coordination, secondary positions of focus sensitive operators, movements concerning wh- items, irregularities of morphemics) as not substantially surpassing what often is understood by logicians as common human mental abilities. Thus, it appears worth a further discussion whether also the internalization of the core of the mother tongue could not be explained on the basis of such common abilities, without postulating a complex framework of innate features.

Topic-Focus articulation and degrees of salience in the Prague Dependency Treebank

To illustrate the issues of tectogrammatical annotation, we add further examples of sentences from PDT in the next two sections.

4. Focus sensitive particles Let us examine first an example with a focus sensitive particle in its primary (prototypical) position: (6)

Pražská mateˇjská pout’ má již cˇtyrˇsetletou tradici. Prague Matthew Fair has already 400-year tradition. “The Prague St. Matthew Fair already has a 400-year tradition”

The linearized TGTS (with many simplifications): (6¢) (pout’.t (pražská.f) (mateˇjská.f)) má.t (již.f) (tradici.f (cˇtyrˇsetletou.f)) Czech differs from English in that the position of the focus sensitive particle již ‘already’ in (6) directly reflects the boundary between Topic and Focus. With the order již má (‘already has’), the main verb would be included in the Focus. However, with the order present in our example, the verb má ‘has’ is understood as contextually bound (which is quite possible with such a semantically light verb, although in general the main verb in the unmarked case belongs to the Focus). Thus, (6) is an example of the primary position of an operator such as already, which covers the whole Focus of a sentence, as may be illustrated by semantically more specific examples, cf. (7) and (8): (7) a. Jim was looking only for a swimming pool. b. Jim was only looking for a swimming pool. (8) a.

Jerry came to the seminar not to listen to the lecture (but to meet Jane there.) b. Jerry did not come to the seminar to listen to the lecture.

It is possible to paraphrase (7a) by ‘Jim was looking for nothing else than for a swimming pool’ and (8a) by ‘Jerry came to the seminar for another aim than to listen to the lecture.’ Only in the (b) examples the verb itself can be understood as being negated, so that a paraphrase with ‘What Jerry did not do is to come to…’ is possible.

5.

Further specific cases

Our next example from PDT illustrates the (marked, i.e., non-prototypical) presence of contextually bound (CB) items (contrastive or not) within Focus:

171


(9) Prˇiznám se, že já osobneˇ to dost prožívám. I-admit that I personally it intensively live-through. “I admit that I personally live this through quite intensively” (9¢) (já.t) (gen.t) prˇiznám-se.f ((já.c (osobneˇ.f)) (to.t) prožívám.f (dost.f)) In the TGTS (9¢) the deleted subject pronoun has been restored; it is CB and constitutes the Topic (the values of its grammatemes are expressed, on the morphemic level, by the agreeing personal ending of the verb). Another node has been added for the General Addressee of prˇiznám se ‘I admit’. The verb together with the embedded clause constitute the Focus. The subject of this clause, expressed by the pronoun in its strong form, is a contrastive CB item, and together with the CB pronoun to ‘it’ it belongs to the Focus, since both the pronouns depend on an item in Focus different from the main verb (namely to the embedded verb). The NB adverb osobneˇ ‘personally’ is understood in PDT to depend on já (‘I’). Note that it is a general rule in Czech that the weak pronominal forms (such as ho ‘him.accus’, mu ‘him.dat’, teˇ ‘you.dat’, ti ‘you.accus’, or the zero form of the Nominative, ‘pro-drop’) always are CB. The order of items within Topic can be illustrated by (10): (10) Dnes už si však bez neˇho svoji práci Today already refl however without him their-refl work nedovedou prˇedstavit. they-cannot imagine. “Nowadays, however, they cannot IMAGINE their work without him” (10¢) (dnes.t) (však.t) (oni.t) (bez-neˇho.t) ((svoji.t) práci.t) (už.f) (neg.f) prˇedstavit-si.Possib.f In Czech, the word order is “free“ enough (i.e., is flexible enough to reflect the scale of communicative dynamism, the underlying word order, without many movement rules) to be understood as the main means expressing the underlying order of the items within the Topic of a sentence. If, following the founder of the Prague TFA research, Vilém Mathesius, we speak of ‘Topic proper’ and ‘Focus proper’ as the two extreme parts of the sentence (i.e., of its underlying representation), with other parts of Topic and Focus occupying intermediate positions, we may see Topic proper as the least dynamic part of the sentence (referring to “what the sentence is about”, and Focus proper as the most dynamic one. In (10), then, we would say that dnes ‘today’ is the Topic proper, with the zero subject (Actor, the strong form of which is oni ‘they’), the group bez neˇho ‘without him’, and the object svoji práci ‘their-refl work’ all occurring as “accompanying members of the Topic”. It is necessary to acknowledge that the specific positions of už ‘already’, si (a reflexive particle lexically belonging to the verb) and však ‘however’ are determined by the


character of these words as clitics. The operator of negation, which we understand as one of the focus sensitive operators, has the form of the verb prefix ne- in Czech.

6. Coreference in discourse This way of describing the Topic and Focus of a sentence in a perspicuous manner, without using means which would be unnecessarily complex, may be useful also for the description of certain aspects of the foundations of discourse patterns, namely of coreference in a connected text. The basic question in this domain may be looked for in the factors relevant for the addressee’s identification of the referents of expressions such as definite noun groups or pronouns. If the discourse contains items such as the boy, the table, or he, how can the hearer/reader find out which boy, which table, or who is meant? An examination of this question, i.e. of how the referent is identified, has been an object of a longer discussion, see esp. Hajicˇová and Vrbová (1982), Hajicˇová (1993), the results of which might serve to enrich the theories of discourse structure formulated by H. Kamp and others. Especially the following two points are relevant: i.

it is certain that the ‘iota inversum’ operator (which is sometimes used in formal semantics and indicates the singularity of its argument) does not offer the proper ground for specifying the referent of a definite noun group, since in the prototypical case more than one boy or table, etc., are included even in the narrow part of the ‘universe of discourse’ (its part that the speaker assumes to be easily accessible for the hearer, the ‘scene’), or in what often is called the stock of knowledge (information) shared by the speaker and the hearer; ii. also the often accepted assumption that a definite noun always has an antecedent in the preceding verbal co-text is not fully substantiated. Let us present an example of a simple discourse segment: (11) In the library he entered the reading room, took some books from the shelf and put them on the single desk that was free, not knowing that it was reserved for you. It may be assumed that the antecedents of he and of the library are present in the preceding co-text (although not necessarily in the preceding sentence token). The referent of the reading room may be determined on the basis of the associative links to the word library (a kind of accommodation). The presence of the presupposition that the (every?) library has a single reading room probably is not necessary (although it belongs to the pragmatic background of this utterance that the possible existence of other reading rooms at the library is backgrounded). This is similar with the shelf, which also exhibits an associative link to the library, without a

173


presupposition that the room contains a single shelf (although there is also an associative link between the shelf and some books; in fact, the sentence does not strictly entail that all the books came from the same shelf). The group some books, being indefinite (‘specifying’), introduces a new referent, not identified, although serving as a starting point for further coreference; in this way, a new member of the scene is established. The pronouns them and it do have their antecedents in the cotext, and the group the single desk that was free constitutes an explicit individual description, again with an associative link to the library. The pronoun you is an example of expressions referring to entities which can be mentioned in a discourse without any specific co-textual antecedent, since they either are directly connected to the pragmatic background of the utterance (as I, you, now, here) or belong to the set of entities easily accessible to speakers sharing a certain cultural background (Shakespeare, Paris, Churchill) or technical domain. Our question is, however, what is the finite mechanism the addressee may use to identify the referents in individual utterances of a discourse. The concept used as the basis for answering this question is that of the hierarchy of salience, introduced by D. Lewis and discussed in the publications quoted above; recently see also Krahmer (1998), Krahmer and Theune (1999). The degrees of salience are understood as relevant for the reference potential of referring expressions in the subsequent utterances in a discourse, and thus also for the connectedness of the discourse. The hierarchy (partial ordering) of salience degrees is being modified by the flow of discourse in a certain way, which we want to capture in annotating the utterances included in PDT. Before we discuss the attributes of the contextual anchoring of word tokens used to this purpose, let us just remark that we believe there is a possibility of enriching H. Kamp’s concept of discourse referent so that the referents display degrees of salience. Thus, in (11) above, the situational context of the utterance makes a certain library (and similarly the referents of he and them) much more salient than other ‘competing’ referents. Items such as you (Shakespeare, New York, and so on) refer to entities enjoying a high degree of salience (in the given group of speakers, or generally) in general, without strict temporal limitations. Let us now illustrate how the hierarchy of the degrees of salience in a discourse can be captured by the descriptive means of PDT. In the prototypical case, a new discourse referent emerges as corresponding to a lexical occurrence that carries the value f of the attribute TFA. Further items referring to the same referent without longer interruptions in the discourse carry the value t, with referents determined by their degrees of salience; a contrastive item in Topic, chosen from a set of alternatives, gets c, although it refers to an entity that is not “given“ in the literal sense. The relationships of individual lexical occurrences to their coreferential antecedents are indicated in PDT by specific attributes for coreferential links:


– – –

–

COREF (for ‘coreference’) — the lexical value of the antecedent, CORNUM (for ‘coreference-number’) — the serial number of the antecedent, CORSNT (for ‘coreference-sentence’) — value NIL, if the antecedent in the same sentence; otherwise, the value is PREVi (with i being a natural number) for the case in which the antecedent is in the i-th preceding sentence, ANTEC — value equal to the functor of the antecedent with grammatical coreference: relative clauses, reflexive pronouns (and particles) such as se (‘-self ’), the relation of control.

As has been discussed since the beginning of the 1980s, certain basic rules determining the degrees of (reduction of) salience can be assumed. In a schematic way, with n(r) indicating that the referent r has the salience of degree n (0£n), we in fact measure the reduction of salience (the higher n(r) is, the lower is the salience of r):1 i. if r is expressed by a noun (group) or pronoun carrying f, then n(r) fi 0(r); ii. if r is expressed by a noun (group) or pronoun carrying t or c, then n(r) fi 1(r); iii. if n(r) fi m(r) in sentence S, then m+2(q) obtains for every referent q that is not itself referred to in S, but is immediately associated with the item r present here; iv. if r neither is included in S, nor refers to an associated object, then n(r) fi n + 2(r) if r was referred to by an NB item only, in the preceding co-text; if r was referred to by a CB item, then n(r) fi n + 1(r).2 As example (12) illustrates, if the degree of salience of a referent r that can be expressed by an item k is higher (by 2 or more) than that of its possible competitors, then k is interpreted by the hearer/reader as referring to r (we add indices indicating the degrees of salience to the relevant nouns and pronouns): (12) Bill.1 met his cousin.0 at the airport yesterday. (13) a. He.1 looked very tired after the long flight. b. He.1 was just looking at the list of arrivals there. (14) He.1 started to explain that… If (12) is followed in a discourse by (13a), then he may be (preferably, rather than necessarily) interpreted as coreferential with cousin; however, if (13b) is uttered after (12), then he may refer to Bill as well as to cousin. This shows that the difference between the degrees 0 and 1 with cousin and Bill (assigned in (12)) is not sufficient for a safe choice of reference. (Note that both in (13a) and (13b) the subject has value 1, constituting the Topic in either case.) The preference of cousin as the antecedent of (13a) is determined by inferencing based on the lexical setting of this sentence, rather than on the degrees of salience. On the other hand, if (14) follows after (12) and (13a), then he in (14) can only be coreferential with the referent of he in (13a), since the salience of the other referent has now been lowered to 2, according to rule (iv).

175


This shows that the degrees of salience are relevant for the choice of reference, although the minimal difference between the degrees 0 and 1 does not give a reliable basis for the choice and can be outweighed by inferences based on contextual or other knowledge. This (with an impact of certain features of S. Kuno’s ‘empathy’, or “the speaker’s viewpoint”) is the case of the choice in (13)(a), and this view is confirmed also by examples from PDT, such as the following one: In the segment of text chosen from PDT, the utterance (5) presented above, and reproduced here as (15), is followed by (16): (15) Cˇeské radiokomunikace musí v tomto roce rychle Czech Radiocommunications have in this year quickly splatit dluh televizním diváku˚m. to-pay debt TV viewers. “This year, Czech Radiocommunications have to quickly pay their debt to the TV viewers.” (15¢) (radiokomunikace.t (Cˇeské.f)) ((tento.t) rok.t) splatit.necess.f (rychle.f) (dluh.f (diváku˚m.f (televizním.f))) (16) Jejich vysílacˇe dosud pokrývají signálem programu Their transmitters up-to-now cover by-signal of-program CˇT 2 méneˇ než polovinu území republiky. CˇT 2 less than half of-territory of-Republic “Their transmitters hitherto only cover less than a half of the territory of the Republic.” (16¢) ((jejich.t) vysílacˇe.t) (dosud.t) pokrývají.f (signálem.f (programu.f (CˇT.t (2.f)))) (méneˇ.f (než-polovinu.f (území.f (republiky.t)))) The reference of the pronoun jejich (their) in (16) as following (15) by itself is indistinct; the rules (i)–(iv) allow the pronoun to refer either to Czech Radiocommunications or to the TV viewers, since the referent of the former expression exhibits salience of degree 1 after (15), in which it occurred as CB (t), and the referent of the latter expression has degree 0 (occurring in (15) as NB, with f). Only inferencing based on knowledge helps then to establish that Czech Radiocommunications, rather than their viewers, are referred to, since they possess transmitters, while viewers normally do not have transmitters at their disposal. If whole texts are examined on this basis, then it may be possible to study much more systematically issues such as those of “topics of texts” or of their individual segments; text segmentation itself may be established with the help of the method outlined. Such inquiries can also bring results advantageous for information retrieval or data mining (distinguishing between texts in which a given topic is actually treated and those in which it is just occasionally mentioned), for question answering, and so on.


We are well aware that the contextual factors discussed here are not the only ones relevant for the salience degrees, and that also other factors than salience degrees are to be investigated as possibly having an impact on the reference potential in discourses of different kinds. Among these factors, there are different sources of inferencing based on contextual and other knowledge, such as the situation of the discourse, domain knowledge and cultural background, as well as specific textual patterns (with episodes, poetic effects, and so on). Factors of these and further kinds can be studied on the basis of the salience degrees that are typical for basic discourse situations. In any case, we may conclude that it is useful for a theory of discourse semantics to reflect the degrees of salience. This makes it possible to distinguish the reference potential of referring expressions and thus the connectedness of the discourse. An explicit, formal representation of the semantics of discourse can be enriched by taking into account the degrees of salience of individual discourse referents in different time points of the discourse.

Notes * We are grateful to the editors of the present volume, as well as to two anonymous reviewers, for their comments, which have been highly valuable for us. The work reported on in this paper has been carried out under the projects GACR 405/96/K214 and MSMT LN00A063. 1. This appears to be necessary, since the only fixed degree of salience that can be observed is that of maximal salience, which is exhibited by an item occurring in the Focus of the utterance under examination. It would not be appropriate to work with an arbitrary value for this maximal degree, from which the rate of salience reduction would be abstracted. 2. These tentative rules, formulated here with a slight modification, have been presented at several occasions for the aims of a further discussion. However, although they were first published twenty years ago, they still wait for a broader discussion, for systematic testing and evaluation, as well as for enrichments and more precise formulations. Such issues may find new opportunities now, when a large set of annotated examples from continuous texts in PDT is available and when a comparison with the centering theory is possible (i.e. with the first theory that, as far as we know, is aimed at similar tasks). Further steps made in these two directions within the research on PDT will be discussed in our forthcoming contributions.

Word order and discourse genre in Tohono O’odham* Colleen M. Fitzgerald Texas Tech University

1.

Introduction

Tohono O’odham word order has received a fair amount of attention, from a variety of perspectives, including descriptive (Saxton 1982), pedagogical (Zepeda 1988), formal (Hale et al. 1977, Hale 1992), functional (Payne 1987, 1992) and historical (Langacker 1977). One subject of controversy includes the issue of whether O’odham has a basic word order, and what type of data should be used to answer this question. The use of discourse data, such as narrative texts, is often one focus of research in functional frameworks (such as Givón 1979 and Chafe 1980). Here we use functional data for formal purposes. The assumption made in this paper is that data from an O’odham narrative provides the primary information about basic word order, as in the discourse-based approach of Payne (1987, 1992). The treatment here deals with a comparison of word order in autobiographical narrative and poetry, and offers a formal analysis of these discourse phenomena in the framework of Optimality Theory. Both prose and poetry contain the same basic types of patterns; the empirical difference comes in how frequently these patterns occur in the two genres. Recent work links frequency to the idea that well-formedness is gradient; this is modeled by using constraint ranking in Hayes and MacEachern (1998), Golston (1998) and Golston and Riad (2000). Frequency of type occurrence correlates with the degree of constraint satisfaction. More frequently occurring types better satisfy higher ranked constraints, while less frequently occurring types incur more constraint violations and on higher ranked constraints. Well-formedness is gradient, not necessarily absolute. I claim that the difference in frequency patterns in prose and poetry can be accounted for by re-ranking the syntactic and prosodic constraints. In the prose genre, syntactic constraints are ranked higher.1 In poetry, prosodic constraints are ranked more highly. The upshot is that prose and poetry can be characterized by different constraint rankings, which results in different patterns of occurrence for each genre.

180 Colleen M. Fitzgerald

This adds to a considerable body of work in Optimality Theory showing that poetic meter involves ranking prosodic constraints above other components (i.e., morphology, syntax) in the grammar (Golston 1995, Fitzgerald 1994, 1997, 2001b, 2002, Rice 1997a,b, Golston and Riad 2000). The specific claim made in this paper is that poetry involves a case of the domination of syntactic constraints. An interesting outcome of this analysis is that it also suggests that there is a considerable amount of rhythmic consistency in Tohono O’odham, as the same trochaic rhythm emerges in modern poetry (this paper), as in other domains such as lexical stress patterns (Fitzgerald 1997, 2001a), traditional songs (Fitzgerald 1998), and cliticization (Fitzgerald 1994). The organization of this paper is as follows. Section 2 briefly discusses the relevant prosodic facts of Tohono O’odham. The subsequent two sections present the frequency patterns for how utterances begin in autobiographical narrative and contemporary poetry. While both prose and poetry contain the same types of initial patterns, the difference between the two genres lies in type frequency. Certain patterns are more common in prose and less common in poetry, and vice versa. This can be accounted for by different rankings of syntax and prosody in the different genres, as presented in an Optimality Theoretic analysis.

2.

Background on O’odham rhythm

O’odham words display trochaic rhythm, as they always begin with a stressed syllable, and words longer than two syllables surface with an alternating strong– weak pattern (Fitzgerald 1997). In contrast, function words are typically unstressed monosyllables, and may even appear as a single segment. (Acute accents indicate primary stress and grave accents indicate secondary stress.) (1) a.

Content Words: begin STRONG–(WEAK)… kí˜ “house” tád» ai “roadrunner” mú-msi:ò “musicians (Sp)” páko’òla “Pascola dancer (Sp)” pá˜do-:à-kam “someone who owns a duck (Sp)” há-haiwàñ-:a-kàm “one who owns cattle” b. Function Words: begin WEAK–(WEAK)… ’o 3rd person imperfective auxiliary ’at 3rd person perfective auxiliary ’am locative specifier s» evidential

Word order and discourse genre in Tohono O’odham

The rhythmic patterns of the content words as presented above are governed primarily by the three metrical constraints in (2).2 (2) Phonological constraints3 a. *Clash b. Trochee (RhType=T) c. *Lapse

No adjacent strong beats on the grid. Feet have initial prominence. No adjacent weak beats on the grid.

Each of these constraints is independently needed. *Clash favors the avoidance of strong–strong patterns, Trochee prefers to avoid weak–strong patterns, and *Lapse militates against weak–weak patterns. These constraints predict an alternating strong–weak pattern in words. This is illustrated by the tableau in (3), although this does not give us ranking arguments. (3) Trochaic lexical rhythm /tad» ai/ ‘roadrunner’ *Clash

Trochee

*Lapse

a. strong–weak: (tád» ai) b. weak–weak: (tad» ai)

*!

c. weak–strong: (tad» ái) d. strong–strong: (tà)(d» ái)

*! *!

These basic rhythmic constraints will be shown below to interact with syntactic constraints, with different interactions for each genre. In the case of the prose genre, prosodic constraints are active, yet subordinate to the syntax. The reverse ranking, where syntactic constraints are subordinate to the prosody, is shown to be the appropriate ranking for the poetic genre. The different word orders of prose and poetry are accounted for by different rankings of syntactic and prosodic constraints.

3.

Autobiographical narrative

In this section, I outline the details of word order in prose, and sketch out the basics of an analysis. The basic point of this section is to demonstrate the types of initial patterns that are found in prose with regard to grammatical characteristics, that is, whether the initial word is a content word (lexical head) or a function word (auxiliaries, specifiers, etc.). In terms of prosodic characteristics, content words are also those words that begin with a stress, while function words do not receive a lexical stress. This section also lays out the distribution of stress in the first two syllables of each line. Both the syntax and the prosody are relevant to determining word order, although syntactic considerations are more important than prosodic ones in prose.

181


In 1909, Juan Dolores wrote down his autobiographical reminiscences using a system he was trained on by A. Kroeber. This text and Dolores’ interlinear translation are combined with a morphological analysis and free translation (not given here) in Mathiot (1991). Two sample lines4 from this text appear in (4). (4) Two lines of autobiography from Mathiot (1991: lines 116–117) a. Tcid haweho sap haptewua vac pitac dhema tekakeda. ÁC- dŸ a voho s-Áap hab t-vua vabŸs pi taŸs[m-a-]dŸ hema t-gagda Ÿ “We, true as I said, good were doing but not long[that] someone sold us.” b. AmÁi semai ke Mr Thompson matchastewua ÁAm Ái s-mai g Mr Thompson m-ac has t-vua “When learned the Mr. Thompson what we were doing,” ˇ itcaceta am oitak wuy. c. adkhema geÁe am key dhek Áad- d-kï Ÿ hema geÁe Áam kei ÁdŸ heg Ái t-Ÿsasda Ÿ Áam Áoidag vui. “he had one grown up appointed as he drove us there farm toward.” The lines are fairly typical of the entire text, in which nearly 81% of all discourse units begin with a function word, rather than a content word. The overall preference is heavily skewed in favor of multiple function words in the utterance initial position. This information is displayed in Table 1, which categorizes lines according to the gramamatical status of the initial two words.5 Table 1.Grammatical classification of line-initial sequences in autobiography Initial Grammatical sequence

Actual # of lines

% of lines

Function Word# Function Word Function Word# Content Word Content Word# Function Word

198 23 21

72.5% 8.4% 7.7%

Total

242

88.6%

The skewed distribution in favor of initial function words (and in particular, of an initial auxiliary) is general characteristic of O’odham texts (J. Hill, p.c.). In the Dolores narrative, 165 of the 273 lines begin with an auxiliary or copula verb. A cursory examination of another prose genre, myths, gives similar results. The first set of myths from Saxton and Saxton (1973: 1–10) shows that function words begin 86% of the first 98 lines. Auxiliary usage is also comparable. Myth lines begin with auxiliaries 68% of the time, only slightly higher than the 60% rate used by Dolores. Content words begin only 14 of the lines in the myths, a number that is inflated by the presence of 8 embedded song lines.6

Word order and discourse genre in Tohono O’odham 183

An additional factor is that the word order patterns are influenced by the fact that O’odham allows the preposing of various constituents, although Payne (1987, 1992) and Hale (1992) offer differing accounts for this. These two sets of facts suggest something like these constraints exert a strong pressure over O’odham. (5) Syntax a. Aux Align (Auxiliary, Left, Complementizer, Left)7 b. Stay (Economy of Movement) Trace is not allowed. The constraints in (5) are obviously not the entire story for O’odham syntax (just as the rhythmic constraints in (2) were not complete). In (6), the influence of these constraints over word order is shown in a tableau. The example utterance demonstrates possible permutations of a clause with a possessive phrase. The first candidate, (6a) satisfies all constraints. Candidates (6b–e) also exhibit allowable word order variation, where the verb moves out of final position and the auxiliary gets displaced from initial position. These all trigger violations of constraints. Displacement of the auxiliary out of initial position so that it appears as the third constituent leads to excessive violation of Aux. Note the allowable word orders nearly always correspond to some degree of constraint violation (6b–e), but the unattested (6f) shows that there is a point at which constraint violation leads to unacceptability. (6) Word order variation in /No g húsi kí: géÁej/ ‘Is Joe’s house large?’ [No [[g húsi]NP kí:]NP géÁej]VP]S?8 int.aux det Joe house big

Aux

Stay

a. No g húsi kí: géÁej? b. No géÁej g húsi kí:?

*

c. No géÁej g kí:j g húsi

**

d. GéÁej no g húsi kí:?

*

*

e. Húsi kí: no géÁej?

*

*

f. Húsi kí: géÁej no?

**!

The tableau in (6) is like the one in (3), in that an input is paired with one (or more) optimal output candidates. If we instead treat these examples in (6) as type variation, then we predict certain frequency patterns for each type. The type schematized in (6a) should be most frequent. Note that cases with two initial function words do occur most frequently in Table 1. The next frequently occurring type would be the one that maintains an initial auxiliary but exhibits extraposition of the lexical verb from final position, as in (6b) and (6c); the similar rate of occurrence for initial iambs is also borne out by Table 2. The other two attested


examples in (6) consist of initial content words, and should probably occur the least frequently. Actual rates vary based on what item follows the initial content word. Initial sequences as in (6d) occur at rates similar to those where function words precede content words, while initial sequences with two initial content words, (6e), are almost nonexistent in the Dolores corpus. Clauses that have an initial content word (noun or verb) followed by a second position auxiliary are equally wellformed syntactic constructions (Zepeda 1988, Hale et al. 1977, Payne 1982). Since (6d) and (6e) are equivalent in syntactic terms, yet occur at different frequency rates, some other factor must play a role. There is a prosodic difference; the less frequently occurring (6e) has two adjacent stressed syllables (a violation of the *Clash constraint). While syntactic constraints obviously play a strong role, they cannot be the only influence on word order. Table 2 gives the results of the phonological patterns of the prose. The phonological patterns mirror the grammatical results, with threefourths of all lines beginning with a weak–weak sequence. Table 2.Phonological classification of line-initial sequences (prose) Initial Phonological sequence

Actual # of lines

% of lines

weak–weak weak–strong strong–weak strong–strong

205 36 24 2

75.0% 13.2% 8.8% .8%

Total

267

97.8%

Examples with adjacent stresses, or stress clash, constitute approximately 1% of all lines. These cases only occur when there are adjacent content words. But iambic sequences begin 13.2% of all lines. The near absence of line-initial cases of stress clash is provocative, especially given the syntactic acceptability of the relevant constructions. If prosodic constraints like *Clash can affect prose, even when ranked below syntactic constraints, prosodic considerations would influence word order. Below, tableau (7) demonstrates how the syntactic and prosodic constraints can be ranked in such a way as to correlate with frequency types in the Dolores narrative. The introduction of the prosodic constraints shows that the best-formed syntactic candidates have prosodic violations (7a-b), and the best-formed prosodic candidate (7c) has syntactic violations. If the prosodic constraints were to rank below the syntactic constraints, then the violation on *Clash would play a role in determining word order frequency. Evaluation in (7) compares four different types of lines that occur in the Dolores corpus.

Word order and discourse genre in Tohono O’odham

(7) Frequency of line types from Dolores Syntax

Prosody

Aux Stay *Clash Trochee *Lapse *

a. (ÁTam o) híhi hemu b. (ÑidŸ só) Ÿ sa Ÿ k pi

*

c. (Jéwed) Ÿ Áat Áújig

*

*

d. (TásŸ Áói)dam Áam

*

*

*

Finally, these patterns and their corresponding constraint violations show that wellformedness is gradient, not absolute (Hayes and MacEachern 1998, Golston 1998, Golston and Riad 2000). In other words, when the comparison is made across types (rather than as an input/output pairing), the candidates exhibit a range of wellformedness that varies according to the constraint violations. More frequently occurring types better satisfy the more highly ranked constraints, while less frequently occurring types incur more highly ranked violations, as in (7d).

4. Contemporary poetry In this section, I describe the grammatical and phonological patterns of Zepeda’s poetry. Zepeda’s poetry is four times more likely to begin with an unambiguous trochaic sequence than Dolores’ prose. Without the use of rhythmic constraints, this is a difficult asymmetry to explain. This section shows that the different frequency patterns in poetry can be accounted for by re-ranking prosodic constraints above syntactic constraints. Syntax is still important, and still plays a crucial role in lines. Both syntax and prosody play vital roles in accounting for prose and poetry, but the two genres differ in the ranking of each component of the grammar. The second corpus consists of modern poetry by Ofelia Zepeda (1995, 1997) written in O’odham and in English. Only lines that are written in O’odham are analyzed in this paper. An example of a bilingual poem is given in (8), where the English could be considered a translation of the O’odham lines. (8) “Na:nko Ma:s Cewagæı /Cloud Song” (Zepeda 1997: 17) Ce:daghim Áo Áab wu:sanhim. Ÿ To:tahim Áo Áab wu:sañhim. Ÿ Cuckhuhim Áo Áab him Wepeghim Áo Áabai him.

185


Greenly they emerge. In colors of blue they emerge. Whitely they emerge. In colors of black they are coming. Reddening, they are right here. The two collections of Zepeda’s poetry total 165 O’odham lines,9 and the beginnings of these lines can be classified in terms of grammatical or phonological patterns. Table 3 gives the grammatical patterns of poetic lines. Function words begin 59.4% of all lines, with nearly half of all lines beginning with two function words. The overall percentage of line-initial content words is 36.4%, which quadruples the rate found in prose. Table 3.Grammatical classification of line-initial sequences in poetry Initial Grammatical sequence

Actual # of lines

% of lines

Function Word # Function Word Content word # Function Word Function Word # Content word Content Word # Content word

73 60 25 7

44.2% 36.4% 15.2% 4.2%

Total

165

100%

The phonological profile of these sequences presents different results than the prose patterns. Table 4 shows that nearly half of Zepeda’s lines begin with a weak– weak sequence, and one-third of the lines begin with a trochaic sequence. Table 4.Phonological classification of line-initial sequences in poetry Initial phonological sequence

Actual # of lines

% of lines

weak–weak strong–weak weak–strong strong–strong

78 64 20 3

47.3% 38.8% 12.1% 1.8%

Total

165

100%

The trochaic rhythm in the poetry comes in two ways. First, adjacent stressed syllables begin the smallest number of lines despite the fact that such sequences are syntactically acceptable. The claim here is that it is the phonological nature of such sequences that make them problematic. Stress clash is a marked phenomenon cross-linguistically and in Tohono O’odham (Fitzgerald 1997, 2001a,b). Second, in


comparing the presence of unambiguous trochaic and iambic sequences, we see a preference for the former: strong–weak sequences are twice as likely to occur as weak– strong. The prevalence of weak–weak sequences neither favors nor disfavors trochaic rhythm, and the fact that this type is the most frequent suggests that the word order of poetry is not exclusively determined by rhythmic constraints.10 This is handled below by allowing *Lapse and Aux to be unranked with respect to each other.11 Prose and poetry vary from each other in the overall distribution of trochaic characteristics, so the variations among the genres can be thought of as a way to differentiate genre type. Representative poetic lines in (9) are given in order from a more frequent type to a less frequent type; this method of evaluation (across types) shows how the gradient results can be linked to constraint ranking. The line beginnings in (9) instantiate the initial portions of the four different prosodic line types from in Table 4. (9) Constraint hierarchy predictions for poetry line types Prosody

Syntax

*Clash Trochee *Lapse Aux Stay a. (mat o) sa Ÿ mú: g mádaj Ÿ

*

b. (Ñéid Áo) mat am má:si c. (An mé)dad Ÿ c g móÁoj d. (S-áp) (Áú:wæı ) g móÁo

*

*

*

*

* *

Note that if we treat *Lapse and Aux as unranked with respect to each other, (9b) is predicted to occur less frequently by reason of its violation on Stay. This correlates with the fact that the weak–weak line type is more frequent than a strong–weak type; strong–weak types will always violate Stay and Aux even while this type satisfies all the prosodic constraints.

5.

Conclusion

The striking descriptive difference between poetry and prose comes in the contrast between the distribution of canonical trochees and stress lapses. Strong–weak sequences begin 9% of Dolores’ lines and 36% of Zepeda’s lines, while weak–weak sequences begin three-fourths of Dolores’ lines and almost half of Zepeda’s lines. Perhaps just as striking is what does not differ; the genres have similar rates for the other two initial rhythmic sequences, iambs and adjacent stresses. Both genres examined here systematically avoid initial sequences of two content words, despite its syntactic acceptability. I claim that this can be accounted for if


rhythmic constraints play a role in determining well-formedness among different syntactic options. This position is not novel. For example, Golston (1995: 343) argues that “all clear cases of prosodic influence on syntax in the literature involve pairs of syntactic constructions, one of which is ruled out by a prosodic constraint.” Some of his examples are given in (10): (10) Antihomophony effects in syntax (Golston 1995: 344) a. [ The [video [of Macbeth]]] and [The [[Macbeth] video]] b. but: [The [video [of The dead]]] and *[The [[The dead] video]] The argument made here is that prosody shapes the word order of both prose and poetry. Prosodic constraints are subsidiary to syntactic ones in prose (but still have an effect), while prosodic constraints are ranked higher than syntactic constraints in poetry. The different rankings of the syntactic and prosodic components coincide with the frequency of occurrence for the different line-types of each genre. This paper has demonstrated that in terms of rhythmic properties, modern poetry is split between strong–weak and weak–weak, favoring weak–weak, while autobiographical narrative overwhelmingly shows weak–weak. Weak–weak sequences occur in O’odham contexts with a more rigid trochaic rhythm, such as in traditional songs (Fitzgerald 1998). A larger context for the analysis here is the notion of rhythmic consistency in Tohono O’odham, where different domains (lexical words, cliticization, meter, discourse, etc.) reflect varying degrees of the same trochaic rhythm. In closing, this paper described the variation in word order in two discoursebased patterns. I have proposed that rhythmic constraints play a role in accounting for both the prose and the poetry. By varying the ranking given to the prosodic constraints by genre, each genre is differentiated in terms of the frequency of its surface rhythmic patterns. This was formalized in terms of Optimality Theory, in which gradient acceptability is a consequence of ranked, violable constraints. In this case, I have shown that Tohono O’odham allows rhythmic constraints to interact with syntactic constraints to shape discourse patterns, resulting in gradient properties of rhythm across the different genre types.

Notes * This paper is dedicated to Eloise Jelinek, a PC user and a syntactician, from someone who is a Mac user and a phonologist. I think that my only “formal” collaboration with Eloise related to computers. She influenced me nonetheless; her research on Native American languages served to attract graduate students with similar interests (including me) to the University of Arizona. The material presented here has benefited from a presentation at the University of Oregon and discussions with Jean-Pierre Koenig, Madeleine Mathiot, Karin Michelson and David Zubin. Thanks also to Andrew Carnie, Megan Crowhurst, Heidi Harley and an anonymous reviewer for comments on an earlier written version. None of the above is responsible for any errors.


1. It should be noted that Golston (1995) and Golston and Riad (2000) argue further that prose is always characterized by the relationship where syntactic constraints dominate prosody. 2. This is an oversimplification for our purposes here; a more complete account can be found in Fitzgerald (1997, 2001a). 3. These constraints are in Kager (1999), as is the syntactic constraint in (5b). 4. Throughout this paper, I use the terms “line” and “discourse unit”, rather than the term “clause”. Discourse and poetry provide some problematic aspects to using clauses, as these genres are not always composed of complete and discrete clauses. 5. Some lines have been omitted from the two tables below because they show different patterns, i.e., a formulaic beginning “thus x said”, an English word, or a one syllable line. 6. These numbers are cited here to show that the properties of the Dolores text are more likely characteristic of narrative prose forms in O’odham, rather than being limited to a single author. The Dolores text is the focus here because it was written by a native speaker and thus reflects his line breaks. The same property holds of Zepeda’s poetry. 7. Thanks to A. Carnie for suggesting this particular constraint. For more details on the alignment of clitics in the syntactic domain of Optimality Theory, see Bresnan (2000a). 8. This example and the first three output candidates are taken from Hale and Selkirk (1987:156), who claim that the verb-final example is basic. Candidate (b) extraposes the matrix verb and candidates (c) extraposes two elements (verb, genitive). There are also two constructed examples, (d) and (e) which are treated as grammatical because such a word order is grammatical in other moods. In actuality, question auxiliaries must be clause-initial. I abstract away from the issue of mood and modality here, as well as the behavior of the determiner [g] (see Fitzgerald 1994 for discussion of the latter). 9. Poems that appear in both volumes are only counted once. 10. The poetic meter of traditional songs permits only two of the four logical rhythmic possibilities, weak–weak or strong–weak (Fitzgerald 1998). Traditional songs invoke a more rigid meter (strong is always followed by weak); stress lapses are less offensive than stress clashes or iambs. 11. Another solution would be to allow Aux to dominate *Lapse; the effect of this would be the interleaving of prosodic constraints with syntactic ones, as is argued to be the case in Fitzgerald (1994). In the interest of space, this solution is not pursued here.

The prosody of interrogative and focus constructions in Navajo* Joyce McDonough University of Rochester

1.

Introduction

1.1 Prosody and argument structure I consider in this paper the question of the tone prosody of Navajo and the relationship of this prosody to its syntactic features. I take as a starting point observations by native speaking linguists (Willie p.c., Austin-Garrison p.c.) that Navajo has no tonal intonation. Jelinek (1989) Hale, Jelinek & Willie (2001) claim that the lack of focus intonation is predicted by the features of Navajo’s argument structure. The main section of the paper is an instrumental documentation of the tonal contours of various constructions in Navajo to support this prediction. There are no systematic pitch perturbations that differentiate these constructions. The evidence lays open the question of the relationship of the tonal prosody to syntactic features that generate the pattern and to the parameters of intonation, accentuation and metrical structure. While the interdependence of syntax and prosody is well known (Selkirk 1984, Beckman 1986, Venditti et al. 2001), and a type of dependence where the syntax determines features of accentuation has been discussed in the literature on prosody (Bollinger 1986, Ladd 1996, Halle and Vergnaud 1987), systematic demonstrations of this dependence are exceptional, and the dependence is supported by mainly theoretical arguments, with few instrumental studies to back it up. The second part of this paper examines the relationship between syntactic and prosodic typology in Navajo, and some predictions it makes about the kind of variation found among other languages of the Athabaskan family 1.2 Intonation and focus The distinction between declaratives and interrogative and focus constructions is among the most universal characteristics of intonation. We normally associate a fall in pitch with the end of declaratives indicating finality, a rise for interrogatives indicating continuation, and the presence of local pitch excursions on focused words.1

192 Joyce McDonough

Beckman (1986) Pierrehumbert & Beckman (1989), Ladd (1996), Laniran (1992) and others have adopted models where there is no fundamental distinction between tones associated to lexical items in tone languages and those assigned by postlexical constraints, such as the association of a pitch accent to a focused word in an utterance or the association of a rising or H edge tone in a interrogative construction. Both kinds of tones, lexical and postlexical, are indicated by the presence of a tonal specification on a particular syllable or at a boundary, with this consisting of a string of tones, serving as an input to the implementation algorithm. Ladd (1996) points out that in this theory (AM or autosegmental metrical), the phonology distinguishes between core tones and edge (‘boundary’) tones, and not between lexical and postlexical specifications. The phonology provides a string of tones to the implementation system, and all languages are equally open to intonational modifications, differing only in the density of their tonal specification at the phonological level. Languages with lexical tone are as likely as those without it to mark intonation by pitch excursions, and to have a distinction between core and boundary tones. This viewpoint increases the appeal of intonational universals. However several studies have indicated that despite their pervasiveness these general properties of intonation are not universal (Connell & Ladd 1990, Laniran 1992). De Jong and McDonough (1992) and McDonough (1999) presented evidence that while Navajo is a tone language in that every syllable has a tonal target, there is little evidence for the presence of boundary tones. In addition, the evidence from studies of Navajo is in contrast results from other Athabaskan languages, such as reports on Minto (Northern Athabaskan) by Tuttle and Kaisse (1996), on Tanana and Salcha (Northern Athabaskan) by Tuttle (1994, 1998) and the work of Hargus (1986) on Sekani (Northern Athabaskan). These studies, which are not instrumental studies, report the existence of stress and metrical structure in Northern Athabaskan. McDonough (1996, 2000a,b) has argued that metrically based structure has no presence in Navajo, and phenomena explained as metrical in Northern Athabaskan are the result of exclusively morphological imperatives in Navajo. In the first section of this paper, I present data from in instrumental study of Navajo focus and yes/no questions (YNQ’s) for evidence of tonal intonation or a distinction between core and edge tones. I use AM theory, where the surface pitch contours (the pitch trace) are records of the implementation algorithm of a language’s phonological specification of H and L tones; including implementation of any tonal intonation. I then relate these results to recent work on the syntax of topic and focus positions in Navajo (Hale, Jelinek and Willie to appear) which distinguishes Navajo from other Athabaskan language in terms of the pronominal argument hypothesis. Insofar as this correlation between intonational structure, focus intonation and syntactic typology is correct, then both a language’s use of intonation and, arguably,

The prosody of interrogative and focus constructions in Navajo 193

the distinction between core and edge tones may be associated to its syntactic typology. Since tonal structure often governs the realization of phonemic contrasts (Fougeron and Keating 1997, Klatt 1976, Culter et al. 1997) this implicates a confluence of syntactic features and phonetic detail. 1.3 Tone in Navajo Many of the Athabaskan languages have been characterized as tone languages, although with a great deal of variation among the types of tone systems. Comparative work on Athabaskan such as Leer 1979, Krauss & Leer 1976, propose that the reconstructed proto-language had no tonal contrasts; these arose from the incorporation of segmental material from the suffixes into the stem. While this is controversial (Kingston 1990, McDonough 2000b), it points to a basic fact about Athabaskan; tone often plays a significant role in the lexicon. Navajo has been classified as a tone language with a tone specified for each syllable (Sapir and Hoijer 1967, Young and Morgan 1987). There is an asymmetric distribution of tonal contrasts among the morphemes, with L tone being the default and the most common tone (the use of a default tone, either H or L, is typical in the family). Characterizations of the tonal system, such as the description of the ‘underlying tonelessness’ of several morphemes in the word and the variety of the reports on tone, stress and metrical structure across the family, undermine the notion that tone has a purely contrastive function in Athabaskan. The instrumental study of McDonough (1999) provided evidence that there are tonal targets on each syllable of the word in Navajo, as reported by Young and Morgan 1987. It was further argued that morphological structure puts distinct pressure on the function of tone according to the domain a morpheme is in; this pressure shapes the role of tone in the lexicon, and the tone typology. Only in the stem domain does tone function contrastively — significantly the rightmost content morpheme in the verb in an SOV language. (The stem domain is the rightmost of three domains in the verb: disjunct-conjunct-stem respectively.) In the conjunct domain, tonal contrast is neutralized; the morphemes in this domain were characterized as ‘underlyingly toneless’, arguably due to the epenthetic alternations many of the vowels exhibit as well as the widespread neutralization of segmental contrasts that occur in that domain. Both of these facts reflect the syntactic type of the morphemes in the conjunct domain as functional or inflectional, a class of morphemes characteristically associated with reduced phoneme inventories and simple syllable structure. Because of the cross-Athabaskan similarity of morphological structure, this fact about the interaction of the function of tone (as contrastive or not) and the morphology is likely to be true of Athabaskan in general and likely to affect the tonal typology. Nevertheless, despite the vagaries of Athabaskan inflectional morphology, tonal contours in Navajo indicated tonal targets were present in

194 Joyce McDonough

every syllable in the verb, and the default specification for these morphemes is L. This being the case, Navajo has the tonal density of a tone language, if not the contrastive tone function of one. This depiction of the tonal typology is much in keeping with Sapir’s (1925) comment that tone in Athabaskan is phonetically simple but morphologically complex, due to the various roles tone plays in the lexicon. For the purpose of this investigation, I accept the hypothesis that Navajo has the tonal density of a tone language, with a tone target specified on each syllable.

2.

Data

2.1 Navajo interrogatives and focus constructions I examined two sets of constructions in Navajo; Yes/No questions (YNQ’s) and Focus constructions. In both sets, I contrasted them to a default declarative construction. In Navajo YNQ’s are marked by one or both of two particles: initial da and -’ísh (ásh) after the noun. In the data below, the first sentence is a declarative statement, followed by three YNQ’s: one with each particle, and one with both. To see if constructions differ in the tones at the end of utterances depending on their tones, the two sets of sentences below are constructions that end with H and a L tone respectively. (Low tone is unmarked, high tone is marked with a grave accent.) (1) Statement vs. YNQ’s, High tone final: Statement: nich’iyø’ hóló˛ “You have food.” your-food 3-exists Yes/no: da’ nich’iyø’ísh hól˛ó “Do you have food?” Q your-food-Q 3-exists Yes/no: da’ nich’iyø’ hóló˛ “Do you have food?” Q your-food 3-exists Yes/no: nich’iyø’ísh hóló˛ “Do you have food?” my-food-Q 3-exists (2) Low tone final: Statement: nimá hoozdohdi bighan. your-mother Tuscon 3-lives “Your mother lives in Tucson.” Yes/no: da’ nimáásh hoozdohdi bighan?’ Q your-mother-Q Tuscon 3-lives “Does your mother live in Tucson?” Yes/no: Nimáásh hoozdohdi bighan?’ your-mother-Q Tuscon 3-lives “Does your mother live in Tucson?”


Yes/no:

da’ nimá’ hoozdohdi bighan?’ Q your-mother Tuscon 3-lives “Does your mother live in Tucson?”

We expect a perturbation in the realization of the f0 of the final tones in the utterances to mark the intonation differences between declarative/interrogative constructions in some manner consistent with a falling/finality and rising/continuance hypothesis. Contrastive focus in Navajo can be marked with the ga’ particle after the focused noun (Young and Morgan 1987, Willie 1991, 2000, Willie and Jelinek 2000, Hale, Jelinek and Willie 2001), in constructions that are equivalent to a cleft, as you can see in the constructions below. Again two sets of constructions were used, H and L tone final. (3) Focus constructions: High tone final Statement: shich’iyø’ hóló˛ “I have food.” my-food 3-exists Focus: shich’iyø ga’ hóló˛ “It’s food that I have.” my-food-F 3-exists (4) Low tone final Statement: sitsilí hoozdohdi bighan little.brother Tuscon 3-lives “My little brother lives in Tuscon.” Focus: sitsilí ga’ hoozdohdi bighan little.brother-F Tuscon 3-lives “It’s my little brother who lives in Tucson.” We expect a local f0 perturbation associated with the focused word in the focus constructions. If there is intonational focus, then it ought to appear as consistent, systematic and measurable differences in the implementation of the f0 targets between these two utterances, preferably on or around the focused noun. Focus construction in Navajo raise special problems regarding semantic constraints on syntactic structure and cross-linguistic variation. It has been convincingly argued that the pronouns, not the NP’s, in Navajo are in argument positions. NP’s cannot be focused syntactically, because the NP’s aren’t in argument position, though they do mark a topic/focus ordering in information structure. As stated in a recent paper by Hale et al. (2001): “Intonation is not used to mark Contrastive Focus in Navajo, as it is in English”. This is an empirical issue with clear predictions for intonational phonology: there are no local or global f0 perturbations associated with focus constructions or YNQ’s. To investigate the claim that there is no focus intonation in Navajo, I used sets of sentences listed in (3) and (4) consisting of pairs with and without the ga’ particle. Two sets of focus constructions are reported on in this paper: H and L tone final.

196 Joyce McDonough

2.2 Methods A list of utterances exhibiting the desired constructions were prepared in consultation with Mary Willie at the University of Arizona and Ms. Martha Austin-Garrison of Dine College, Shiprock, New Mexico in January 2001. These utterances were then recorded as careful speech. Periods of silence (1000ms) were spliced into the sound track after each item, resulting in an instrument suitable for prompting oral responses.2 Native bilingual Navajo speakers listened and were asked to repeat the utterances after the individual prompts. 16 speakers were recorded in this way. For the purpose of this study, 7 speakers were selected from the data. The high (f0 maxima) or low (f0 minima) f0 turning points on each syllable were measured from the pitch contours of the utterances in (1)–(4). In the case of a steady state contour, a mid-point was measured. Figure 1 below demonstrates a pitch contour and gives examples of the alignment of f0 minima, maxima and a steady state. The late alignment in the syllable of the f0 minima and maxima is characteristic of the data in all instrumental studies on Navajo tone.

2500

0

d a n

i

m á

h

oo

z d o d i b i

gh a n

500

250

F0

Hz

5000

0

Figure 1.A characteristic spectrogram and pitch trace of nimá hoozdohdi bighan. ‘Your mother lives in Tucson’.

In addition to the core tones associated to the syllables, pitch perturbations at edges were recorded if they were found after the last vowel. These perturbations usually consisted of f0 traces in postvocalic, utterance-final sonorants. These are marked ‘final t’ in the charts below. If a speaker paused during the recitation, the pitch at the edge of the pause was recorded (see Figure 5 and discussion). Though speakers were not asked to speak carefully, most speakers tended to use careful speech because of the deliberate nature of the prompts. In this paper, the AM model as elucidated in Ladd (1996) is assumed. The lexical items with tonal specifications are submitted to the phonology, where word


formation processes and phonological constraints modify the lexical representations, including the tonal specifications, resulting in a phonological tone pattern. I assume the output of this process is a word, a verb form marked for tone. Thus the lexical output for the verb form yídzi˛˛is ‘I drag it’ would have a H and L tone associated to the first and second syllable respectively. This output is submitted to phonetic implementation processes. Here, basic pitch rules assign a frequency scale (f0 targets) and time alignment (position in utterance) to the tonal targets, basically assigning an f0 value to the H and L tones, and aligning them to the text. The tone targets can be seen in the surface contours as local f0 turning points. The targets are interpolated by context dependent implementation mechanisms such as grid lines (which set the pitch range), any boundary tones, and any adjustments or constraints which increase or decrease the basic realization of f0 (such effects as upstep, declination or downstep). I follow Gårding and Bruce (1982), Beckman and Pierrehumbert (1986), Ladd (1996) and others in defining surface pitch contours as the sum of the factors that interpolate these f0 targets. Implementation also includes adjustments to pitch at the utterance level, such as possible sentence level intonation patterns, like pitch range expansion. Thus the surface pitch contours that I examine are records of the fundamental frequency of the voice as it rises and falls to meet the high and low targets which have been specified by the phonology, plus the full set of pitch implementation rules. Crucially, any intonational tone marking will be observable as systematic perturbations in the surface contours associated with the appropriate constructions. 2.3 Results 2.3.1 Interrogatives and focus constructions Here I consider the question of whether there are f0 contours and implementation rules associated with YNQ’s and Focus constructions in Navajo; whether there is systematic evidence for the presence of edge tones; and whether the focus particle is associated with phrasal domain or local effects. The claim is: due to properties of the syntactic typology of Navajo, no intonation will be found. Figures 2 and 3 are graphs of the f0 targets of the syllables in the statement versus interrogative constructions representing H (as in 1) and L (2) tone final sentences sets respectively. If there are pitch perturbations in the intonation contours of YNQ’s versus statements, we expect to see them in the differences in the way the f0 is implemented in statements versus the interrogatives in either or both the H and L tone final utterances. In particular, I was looking for localized pitch perturbations at the right edge of the word, such as but not limited to a falling f0 for the declaratives and rising for the interrogatives. Or we might suppose that the Q particles themselves carry either recognizable f0 prominence, or that they cause perturbations on the f0 of nearby syllables.

198 Joyce McDonough 300 YNQ and statement

Fundamental Frequency

250

200

150

circle = YNQ square = statement

100 da’ ni Q-particle

ch’i

yø’

ísh hó Q-particle

ló

t

50

Figure 2.A graph of the f0 peaks and valleys for the yes/no versus statement utterances with final high tone. The individual F0 points are plotted, with the averages for statements and questions.

For each sentence set ((1) and (2)) there are7 speakers, with 4 sentences in each set. One of these sentences is a declarative statement; the other three are variations of YNQ’s, as discussed in the last section. Thus each graph represents 28 utterances over 7 speakers: 7 declaratives and 21 YNQ’s utterances. The question words/ morphemes in the utterances are the initial da, and the ’ish (’ash) after the first noun. The declarative utterances have neither of these morphemes; the YNQ’s have one of the two, or both. Thus for the non-question words and morphemes (the statements) there are 28 f0 tokens for each syllable target. For the question words/ morphemes, da, and ’ísh (’ásh), there are 14 data points each. These are plotted in the charts as data points on each syllable. These data point graphs were then overlaid with the average f0’s for the statement and the YNQ’s separately. The black circles represent the averages f0’s for each syllable of the YNQ’s, the squares represent averages for the statements. Points marked ‘final t’ represent an f0 measurement taken after the last vowel in an utterance. Pitch traces after the last vowel tended to show up in final words that ended in sonorants, such as bighan. The data is striking. In these utterances, there is very little, if any, evidence for a fall or rise in pitch at the end of the sentences in either the YNQ’s or the statements. As we can see from the graphs, the declarative and YNQ’s have near identical contours, regardless of whether the utterance is H or L final. The two averages (circles and


300 YNQ versus Statement

Fundamental frequency

250

circle = YNQ square = statement

200

150

100

da’ ni Q-particle

má-ásh hooz Q-particle

doh

di

bi

ghan

t

50

Figure 3.A graph of the f0 peaks and valleys for the yes/no versus statement utterances with final L tone.

squares) fall nearly on top on one another. Furthermore, little evidence of any perturbations in f0 associated to either of the YNQ particles. No systematic patterns of intonation can be construed for these utterances. Of particular interest is category final tone ‘t’, since perturbations in the f0 of this position may indicate the existence of edge tones. In a pattern characteristic of all the data, we find very little f0 movement from the preceding point, which can be taken as indicating a lack of edge tones. In the second set of examples (Figures 4 and 5) are data from the focus construction sets. The focus particle is ga’. In these utterances we are looking for local pitch perturbations on or around the focused NP, which could be used to indicate a difference between the plain and focused constructions. The perturbations may be indicated in the f0 value of the focused item, a local effect, or they may be more global, affecting the contour of the utterance following the pitch contour. Focus might be indicated by the presence of another tone, or in the alignment of the tone. We are looking for any significant differences in the pitch contour between the two type utterances. As with the previous set, this graph represents a graph of the individual data points overlaid with a plot of the averages for the statement (square) versus focus (circle) constructions. The statement versus focus constructions consisted of two sentences, one focus and one statement, for seven speakers. Thus there are approximately fourteen data points for each target, with seven data points for the ga’ particle. As we can see, as with the previous it is hard to construct any evidence of f0 perturbation associated with focus constructions, the f0 contours for the Focus and statements fall on top of each other. As Hale et al. predict there is no focus intonation in Navajo.

200 Joyce McDonough 300

Focus versus Statement

Fundamental frequency

250

200 150

circle = Focus square = Statement

100

50

0

shi

ch’i

yø

ga’ hó Focus particle

1

2

3

4

ló

5

t

6

7

8

Figure 4.A plot of the f0 targets for the focus versus statement constructions and averages, H tone final utterances. The ‘ga’ morpheme is the focus marker. 300 Focus versus Statement 250 Fundamental frequency

circle = Foucs square = statement

200

150 si

100

tsi

lí

ga’

hooz

doh

di

bi

ghan

t

Focus particle 50

0

11

Figure 5.A plot of the f0 targets for the focus versus statement constructions and averages, L tone final utterances.

As with the YNQ’s above in Figures 2 and 3, there are no systematic final tone differences that could be used to indicate a phrasal boundary marking.


2.3.2 Phrasal tones Prosodic organization is primarily cued by the prosodic phrasing at boundaries, and the tone types used to mark these boundaries. In addition to the utterance final phrasal boundary, one place to look for evidence of prosodic marking at boundaries is at the juncture of an NP and a verb. In our data, I did not control for prosodic grouping, I simply asked our subjects to repeat what they’d heard presented to them. In the data there are several instances where a subject paused at what could be reasonably proposed as a natural boundary likely to be marked by the prosody, at the edge of an NP followed by a verb. The example we see below is indicative of the pattern found in our speakers’ data. The data in Figure 6 and 7 come from two interrogative utterances of one of the female speakers (from (1)). The utterances are da nich’iyø’ hól˛ó and nich’iyø’ ísh hól˛ó, “Is this your food?: two instances of a yes/no question with single interrogative particles da’ and ’ísh, respectively. In the first one, the speaker paused 291ms after the noun nich’ya˛’ ‘your food’, before beginning the verb hól˛ó. The f0 measurement was taken at the last f0 minima in the noun, which carries a L tone specification and falls at the right edge before the pause (Figure 6). In the second (Figure 7), there was no pause after the NP nich’ya˛’. In addition, the NP in the second utterance is suffixed by the question particle -ísh. Therefore, the comparison in these figures is between a L tone aligned to a syllable before a 293 ms pause (nich’ya˛’ -LLL#) and a tone on the same lexical item that occurs inside a word boundary (nich’iyø’ísh — LLLH#). A 291ms pause represents a strong boundary with enough time to stop and reset articulatory parameters. If there is prosodic phrasing marking a NP juncture, this is the likely place to find it. There should be some evidence of the implementation of the tone at this point in the two utterances; it ought to be realized as differences in the f0 values of these two tokens. The order of the repetition was as follows: auditory prompt — first utterance — auditory prompt — second utterance, clearly a case of non-causal speech. I have presented a waveform, pitch trace and spectrogram for both of these utterances to demonstrate this data (Figure 6 and 7). The L toned ya’ syllable is the syllable in question. It is closed by a glottal stop in both utterances, as we can see in the spectrogram and waveforms. The f0 value was measured at the lowest edge of the fall on this syllable, the f0 minima. The f0 value for the first utterance (Figure 6) was 189Hz, for the second (Figure 7) 191 Hz, a difference of 2 Hz. A phrasal tone such as we find in English would result in some pitch perturbation associated with this juncture, possibly raising the pitch after the f0 minima to indicate continuance. Insofar as these examples are representative of what a native Navajo speaker produces at this type boundary, this is quite clear evidence for the failure of phrasal prosody to be marked by edge tone at an internal juncture in Navajo.

202 Joyce McDonough

d

a’

n i

ch’

i

y ø’

h ó

P

ó

500

< pause > 291 ms

250

2500

0

F0

Hz

5000

0

f0 = 189

Figure 6.Spectrogram and pitch trace demonstrating an f0 minima on a low toned syllable before a 291 ms pause.

ni

ch’

i

y ø’

í

h ó P

ó

500 250

2500 0

sh

f0 = 191

F0

Hz

5000

0

Figure 7.Spectrogram and pitch trace demonstrating f0 minima on the low toned syllable in Figure 6 when it is inside a morpheme boundary.

2.4 Summary No evidence of any pitch perturbations or special pitch-related implementation rules associated with interrogatives or focus constructions is found in this data. No pitch perturbations associated with edge tones at phrasal boundaries are found in this data. There is no differences in the implementation of the L tone at the edge of a potential phrase boundary (NP Pred) and the L tone inside the word (Figures 6 and 7).


Because edge tones are associated to levels in the prosodic hierarchy rather than syllables, insofar as this data is representative of a native Navajo speaker, there is no evidence that prosodic levels in Navajo are systematically defined by tonal prosody. There is no apparent distinction between core and edge tones in this data. This begs the questions: what determines whether a language does or does not implement intonation distinctions? Is there a parametric variation between languages that do and do not have a distinction between core and edge tones? What is the relationship between argument structure and accentuation that predicted this pattern in Navajo?

3.

Discussion

3.1 Issues In this section I present a brief discussion and outline of the issues that this data raises, and submit some ideas for future work. The issues examined in this paper concern the role of structure and parametric variation in accounting for prosodic phenomena with claims to universality such as the presence or absence of intonation and tonal accentuation to mark semantic contrasts. We have seen that there is no clear evidence for intonation or accentuation contrasts between statements and yes/no questions, and no evidence of pitch perturbations on focused nouns in focus constructions. We also have little evidence that marking of phrasal boundaries by tonal contours exists. It is impossible to prove the lack of the presence of phenomena, especially in the face of arguments for its universality. However, the data we have seen shows that, if there is tonal marking in Navajo of this sort, it is not obvious or overtly systematic, and we might look elsewhere for an explanation of the patterns we do find. There is argument to be made that such explanations do exist, and, in fact, they predict the types of phenomena exhibited by Navajo. 3.2 Accent and arguments The argument here hinges on the relationship between accentuation, argument structure and hypothesized parameters of accent placement. It has been observed in the literature that many accentuation patterns depend on a distinction between arguments and predicates (Bollinger 1986, Halliday 1967, Halle and Vergnaud 1987, Ladd 1996, others). Some of these observed patterns include the differentiation of definite and indefinite NP’s, accent differences on yes/no questions, focus constructions and intonation in general (see Ladd 1996, Chapter 6 for a discussion of these facts). Bollinger (1986) has noticed that there is general pattern cross linguistically for accent to appear on arguments over predicates.

204 Joyce McDonough

As Navajo has been argued to be a pronominal argument (PA) language, there are no NPs in argument positions (Willie 1991, Jelinek 1984, 1989, 1997, Willie and Jelinek 2000, Hale et al. 2001). The arguments are part of the predicate. Navajo does not have a shifting accent. There is no focus accent in Navajo. We’ve seen the data supports this hypothesis. In conjunction with these patterns of accent placement, a parameter of variation has been proposed by several researchers. It’s been proposed that there are two kinds of languages, those that have accent on the rightmost content word in a sentence and those where it may appear earlier in the sentence. We refer to this parameter as [+rightmost] after Halle and Vergnaud (1987) and following Ladd (1996). A full discussion of these ideas is well beyond the scope of this paper and I refer the reader to the literature on this subject. What is important to the present discussion, however, is the idea that in rightmost languages, accent is structurally determined (occurring on the rightmost content word), and is independent of the argument/ predicate distinction for its placement. This predicts that in PA languages like Navajo, if there is any ‘accent’ in the language, it’s not shiftable, it is not used to focus particular NP’s, and it is realized on the rightmost content morpheme by default. 3.3 Prominence and SOV languages Turn for a moment to the notion ‘prominence’ as a phonetic phenomena. Although the realization of accentuation is language specific, it is commonly accepted that a word carrying stress or accent exhibits phonetic characteristics commonly associated with prominence. These are generally some combination of the amplification of duration, intensity and local pitch perturbances such as a pitch range expansion on the prominent word. If a bundle of accentual phenomena is tied to the differentiation of arguments and predicates, the prediction is that languages that are both SOV and PA may not have accentuation, intonation or edge tones. Thus the prediction works both from the syntax to the prosody and vice versa. With respect to the [rightmost] parameter, in [+rightmost] languages, a default accent specification falls on the rightmost content word in a sentence. This accent is not tied to an argument/predicate distinction and is not used to flag pragmatic information. Relevant to the present discussion, in an SOV language that is also a PA language, pragmatic use of accent is barred because NP’s aren’t in argument positions. The rightmost content word is the V; the default specification falls on the verb. If [rightmost] is a parameter of variation, we predict the verb to carry the [rightmost] accent and to exhibit the phonetic properties of prominence. Navajo, an SOV-PA language, presents an interesting case for testing this hypothesis. The instrumental data presented in part bears this view out. Though it may or may not be possible for a [+rightmost] language to have metrical structure,


it is not possible for accent to be assigned on any but the rightmost morpheme. Navajo can only be a [+rightmost] language. Furthermore, given the prefixal nature of the Navajo verb, we may predict that the rightmost content morpheme in the verb, the verb stem, which carries rich semantic and aspectual information, is likely to be phonetically prominent. This seems to be the case. Instrumental studies indicate that the verb stem has many of the properties commonly associated to prominence: longer durations, pitch range expansion, semantic salience (McDonough et al. 1993, McDonough 1999). Recent work on the phonetic structure of Navajo has shown that the verb stem may have up to two to three times the duration of the conjunct domain (McDonough to appear). The stem is the only place in the word where the full set of phonemic contrasts occurs. As Hale et al. have pointed out, several Northern Athabaskan languages (such as Slavey, Witsuwit’en) are not PA languages. In these languages, we thus expect to find evidence of topic/focus and interrogative implementation rules, word formation rules sensitive to foot structure (which Navajo does not) and edge tones to mark phrasal boundaries. We expect these languages to be open to the development of stress systems, tonal intonation, and the deliminative use of pitch. All studies indicate that this is so. The details of these relationships and the details of the patterns of accent and metric structure and syntactic type beg investigation backed by the empirical work of instrumental analysis.

4. Conclusion In this view, a language’s intonational features such as boundary tones, focus and accent placement, and its phonetic implementation system are tied to its general typology by a cluster of structural relationships around the argument/predicate distinction and the prosodic hierarchy. Laying out the particulars and full predictions of this theory is beyond the scope of this paper and awaits further work, including instrumental work and the modeling that a theory of parameters requires. But the theory indicates that varying one parameter, such as the language’s predicate type, results in systematic changes in accentuation because these changes are inherently tied to its argument structure. Navajo’s SOV-PA status predicts its lack of intonation, forces it to default to [+rightmost] — argued to predict the prominence of the verb stem. Navajo’s lack of metrical structure is at least related to lack of the play of the prosodic hierarchy in the grammar. This also implicates a regulated relationship between structural features and phonetic implementation algorithms. Furthermore, we may predict this stem prominence, which is a striking feature of Navajo verb, may not exist in related languages with more accent type systems, and with different argument structures. The variation across the Athabaskan languages, such as the pitch accent type system Kaisse and Tuttle report on in

206 Joyce McDonough

Minto, may lend insight into the development of accent systems and into the development of the language’s phonetic structures, as these changes relate to changes in the status of the argument structure. Because all the Athabaskan languages are closely related and have similar syntactic features, and because they exhibit systematic differences in their argument structure and, by report, systematic differences in their prosodic structure, this family of languages provides an important testing ground for these hypotheses, backed by instrumental studies an undertaking that serve to further our understanding of human grammar.

Notes * This paper is dedicated to the faculty and staff at Dine College in Shiprock, New Mexico, especially to Martha Austin-Garrison and Anthony Goldtooth, whose interest in the study of Navajo phonetics continues to make it happen. I benefited from conversations on this topic with Eloise Jelinek and Mary Willie and Ken Hale, who first explained to me the connection between the intonation patterns and the syntax, though the ideas as they are laid out in this paper are my own responsibility. I thank the members of the seminar on Navajo morphophonemics at the University of Edinburgh, Spring 2001, especially Bob Ladd, Alice Turk, two anonymous reviewers and the members of the Department of Linguistics & Center for the Language Sciences at the University of Rochester, especially Greg Carlson, Katherine Crosswhite, Bruce Peng, Jeff Runner and Michael Tanenhaus. I’m also grateful for support from the Rockefeller Foundation’s Bellagio Study and Conference Center. This work is supported by a grant from the NSF. 1. These phenomena have usually been associated with systems that have metric stress, such as English, and but they have also been shown to operate in languages with pitch accent (Japanese, Poser 1984, Pierrehumbert & Beckman 1986, Serbo-Croatian, Lehiste and Ivic 1986) and lexical tone (Thai, Abramson 1962, Hausa, Lindau 1986). 2. Since Navajo is an oral culture, I considered oral prompts more appropriate than text.)

Subject number agreement, grammaticalization, and transitivity in the Cupeño verb construction* Jane H. Hill University of Arizona

1.

Introduction

Cupeño is a member of the Cupan group (Bright and Hill 1967) of the Takic subfamily of Northern Uto-Aztecan.1 Along with its close sisters, Luiseño and Cahuilla, the language is notable for multiple, indeed exuberant, marking of person and number (cf. Steele 1990, Seiler 1977 for Luiseño and Cahuilla respectively). This essay will explore the marking of subject number, the most richly developed dimension of the person-number (PN)-marking system, focussing on the thematic verb construction. The system of grammatical number in Cupeño distinguishes singular vs. plural. In the verb construction subject number can be marked in five different ways: (a) On the pronominal prefix of past-tense verbs by the plural -m, (b) by tense suffixes that are suppletive for number, (c) by aspect suffixes that are suppletive for number, (d) on the thematic suffix of the -in verb class, which takes the form -in “singular” vs. -men “plural”, and (e) by suppletion for number of the verb root in eight verbs. Since in transitive verbs the suppletive verbs agree with their object, and since -in is a transitivizing suffix, only four of these subject– number markers can appear in a single verb construction. Jacobs (1975) and Heath (1998) have shown that much of this number agreement marking results from the grammaticalization of light verbs, some of them themselves suppletive for subject number. Heath (1998) used this system to exemplify what he calls “hermit crab” grammaticalization. In this type of grammaticalization, an affix that has begun to erode below the threshold of phonetic salience (the meaning and structural slot constitute the “hermit crab”) is replaced by a new “shell”, a grammaticalized element that resembles it phonetically and semantically. Heath argued that these phonetic and semantic resemblances are necessary to the grammaticalization process. My extension of Heath’s proposal is based on work by Jelinek (1998, 2001) and Jelinek and Escalante (2000). Jelinek argues that in Yaqui, a sister language of Cupeño in the Uto-Aztecan family, we must recognize a system

208 Jane H. Hill

of strong transitivity, where transitivity and voice are functional heads, independent of the lexical head of the clause. I argue that for hermit crab grammaticalization to occur, there must not only be phonetic and semantic similarity between the grammaticalizing element and the marker that it replaces. There must also be an appropriate syntactic site. Jelinek’s model of the Yaqui verb provides a set of structural sites for certain grammaticalizing elements, and no others, thus providing constraints on grammaticalization that are absent from Heath’s model. The application of Jelinek’s work to the problem of the Cupeño verb thus contributes to the development of both functionalist theory, grammaticalization, and formal theory, the theory of functional categories in the verb, by permitting us to see the ways in which universal grammar constrains the shape of language change, and the ways in which historical events actualize categories made available by UG.

2.

Subject number in the Cupeño verb construction

Marks of tense, aspect, and modality in Cupeño are distributed across both the auxiliary second-position clitics and the verb construction itself. Markers of number occur on independent nouns and pronouns, in the auxiliary complex, and in the verb construction. Here I restrict the discussion to the verb construction. Cupeño has three major verb classes, named after their thematic suffixes: the zero (0) class, the -in class, and the -yax class. The 0 class is unmarked. It includes verbs for many basic bodily processes and human activities (from the perspective of Cupeño culture). Many of these verbs are “unergatives”, with subjects that are in Agent or agentive thematic roles, and objects not permitted, e.g. ’ísaxw “to sing a type of men’s song”, muu “to shoot a bow”, wíw “to make acorn mush”, ’ásrqetu “to menstruate”. The -in class, marked with a suffix of that shape (in- with singular, -men with plural subjects) is basically a transitive class. The -yax class, marked with that suffix, is basically intransitive and includes many unaccusative and stative verbs. While Hill (1969) proposed that the -in and -yax thematic suffixes distinguish volitional and non-volitional verbs respectively, Jacobs (1975) argued that the distinction is fundamentally one of transitivity. While a few -in verbs do not take objects, and a few -yax verbs are attested with objects, Jacobs’ distinction does work for most cases. A few verb roots occur in all three classes; many occur in both -in and -yax classes as transitive-unaccusative pairs, e.g. (1) a.

chásr-in “polish something” b. chásr-yax “something shines”

The Cupeño verb construction 209

(2) a.

púve-in “make something round” b. púve-yax “something is spherical”

In addition to thematic class, verb constructions are marked for aspect and tense. Perfective aspect is unmarked. Imperfective verbs in non-future tenses are marked with the suppletive aspect suffixes -qa, agreeing with singular subject, and -we, agreeing with plural subject. The future imperfective suffixes are -nash, agreeing with singular subject, and -wene, agreeing with plural subject. These suffixes follow the thematic suffix. Past-tense verbs are identifiable because they require a PN prefix crossreferencing the subject or agent (henceforth, “subject”). Verb constructions in nonpast tenses do not accept these prefixes. Thus past perfective verbs are marked only by the thematic class suffix and the PN subject prefix. Future perfective verbs are marked only for thematic class. In addition to requiring the subject PN prefix, the past tense of the imperfective is marked by the suffixes -l, agreeing with singular subject, and -n, agreeing with plural subject. These suffixes follow the imperfective suffixes, yielding -qa-l and -we-n.2 Number marking in the verb construction that cross-references the subject thus depends on verb tense and aspect. Future perfective verbs (except for those with suppletive roots) have no subject number marking at all, since they have neither subject PN prefixes or imperfective suffixes. All other verb constructions are marked for subject number at least once. In the past imperfective subject number is marked up to four times in verbs of the -in class, since the -in thematic suffix has a special form -men cross-referencing plural subject. Examples and details are presented below. 2.1 Subject number marking in the future tense The simplest Cupeño verb construction is the perfective of the future tense. Since the verb is non-past, it has no subject prefix, and since it is perfective, it has no aspect suffix. Such verbs are marked only with thematic suffixes (and with derivational suffixes such as causative or inchoative). An example is given in (3), with the suppletive root chíx “to die of several beings, to kill several beings” (the singular is qaáwi, seen in (4)).3 Note that this sentence has no subject marked at all (both “null subject” and “null object” sentences are possible in Cupeño). (3) Patá’a-nim ’axwé-ch-im-i chíx-ni all-pl that-npn-pl-ob die-cause (I) will kill all these. (Coyote and Wolf 024)

210 Jane H. Hill

While (3) is acceptable, it is more common for a subject marker to be present somewhere in the sentence, either as an independent noun or pronoun, or, most commonly, as a PN clitic in the auxiliary complex (which usually follows the first word), shown in (4). (4) tukumáy=ne=pe qaáwi tomorrow=1sg=irr die.sg “Tomorrow I will die” (RN Creation 064) The future imperfective suffix is suppletive: -nash “future imperfective singular” and -wene “future imperfective plural”. -wene is homophonous with the plural of the customary aspect. These are illustrated in (5). (5) a.

tukumáy=né=pe ne-má-’aw nengú-nash tomorrow=1sg=irr 1sg-hand-in hold-fut.imp.sg. “Tomorrow I will hold it in my hand” (2 13 107) b. tukumáy=che=pe che’-má-’aw nengú-wene tomorrow=1pl=irr 1pl-hand-in hold-fut.imp.pl. “Tomorrow we will hold it in our hands” (2 13 108)

2.2 Subject number marking in the past tense Past-tense verbs must be marked with prefixes for subject person and number. The subject prefixes are shown in Table 1. On possessed nouns, the same prefixes crossreference the person and number of the possessor. Table 1.Subject pronominal affixes

1 2 3

Singular

Plural

ne’epe-

chem’empem-

Examples of a past and non-past verb construction, showing the presence and absence of the prefix, are given in (6). túku=’ep pem-chíx yesterday=r 3pl-die.pl “Yesterday they died” b. tukumáy m=el=pe chíx tomorrow pl=3pl.abs=irr die.pl “Tomorrow they will die.”

(6) a.

The Cupeño verb construction

While subject prefixes are restricted to the past tense, any transitive verb can appear with an optional object prefix. The object prefixes are shown in Table 2. Table 2.Object pronominal-argument prefixes

1 2 3

Singular

Plural

niÁipi-

chimiÁimimi-

A verb construction marked with a subject prefix and an object prefix is shown in (7). Note that the root max “give” is a “stressless” root (Hill and Hill 1968). With such roots, stress shifts to the imperfective suffix, if one is present, or to the subject prefix, if no suffix is present. Stress never shifts to the object prefix. The implications of this fact are discussed in 3.2.1.2. (7) túku=’ep ’i-ché’-max yesterday-r 2sg.ob-1pl-give “yesterday we gave it to you” (2 43 480) In verbs of the -in and -yax classes, the subject “prefix” shifts to a position following the root and before the thematic suffix. Barragan (2001) suggests that this position reflects the fact that the thematic suffix, a grammaticalized light verb, is structurally the functional head, v (in the proposal of Chomsky 1995a; in the analysis from Jelinek 1998 to be discussed below, the relevant functional head is tran). Thus the prefix occurs just before it, and the verb root itself is adjoined to this construction, originating as V. I will return to the details of this structure below in 3.2. The position of the PN marker with thematic verbs is illustrated in (8), with the verb hásri “to go” (which is attested only in the intransitive -yax class) and with pem- “3pl”. (8) a.

túku=’ep hásri-pem-yax yesterday-r go-3pl-yax “Yesterday they went off” b. túku=’ep wíchax-pe-n yesterday-r throw-3sg-in “Yesterday he threw it.”

With past-tense verbs in the transitive -in thematic class, plural number is marked by a special past plural form of the thematic suffix, which appears as -men instead of -in; this is seen in (9) with the verb wíchax “to throw, drop”, and the subject PN marker chem- “1pl”. Note that the sequence /mm/ becomes /’m/ by regular rule.

211

212 Jane H. Hill

(9) túku-’ep wíchax-che’-men temá-t’a-yka yesterday-r throw-1pl-in.pl ground-acc-to “Yesterday we threw it to the ground.” (3 37 94) There is an exception to the above pattern with the thematic suffix -in. Many 0-class verbs can be made transitive by adding the causative suffix -ni “cause X to do V” (A very few 0-class verbs form causatives with -in, in which case the regular pattern described above appears). Such causative verbs are then further suffixed with -in. However, in this case -in does not attract the subject PN marker. Instead, this remains as a prefix on the verb root (where it cross-references the person and number of A). Nor does the special plural form of -in appear. Examples are seen in (10): (10) a.

túku=’ep pulínyi-ch-i pe-’ásr-ni-n (*’ásr-ni-pe-n) yesterday-r baby-npn-ob 3sg-bathe-caus-in “Yesterday he baptized the baby” (8 31 193) b. túku-’ep ’iví-y chem-’elél’i-chi-ni-n yesterday-r this-ob 1pl-bad-inch-caus-in “Yesterday we spoiled this” (6 7 251) (*’elél’i-chi-ni-che’-men)

2.3 Subject number marking on non-future imperfectives Non-future imperfective verbs are distinguished from future-tense verbs because they appear with the non-future set of imperfective suffixes, -qa “singular” and -we “plural”. In the past imperfective, a subject prefix will also be present, and, in addition, the past-tense suffixes, -l “singular” and -n “plural” appear. Examples are shown in (11). (11) a.

túku=’ep mi-wíchax-ne-n-qa-l yesterday=r 3pl.ob-throw-1sg-in-nf.imp.sg-pst.sg temá-t’a-yka ground-acc-to “Yesterday I was throwing them to the ground.” (3 37 94) b. túku=’ep mi-wíchax-che’-men-we-n yesterday-r 3pl.ob-throw-1pl-in.pl-nf.imp.pl-pst.pl temá-t’a-yka ground-acc-to “Yesterday we were throwing them to the ground.” (3 37 95)

For the present tense no subject prefix is present, and the non-future imperfective suffixes appear, as shown in (12).


(12) a.

’é’e=sre=t=pe ’í-qi tútuvchi-qa pém-eyik you-dub-2/3abs-irr 2sg-self tell-nf.imp.sg 3pl-to “Probably it is you yourself who are telling them” (Faye KP 129–163) b. ’é-mene=sre=l=pe nénen-we 2sg-with=dub=3pl.abs=irr go.along-nf.imp.pl “They are probably going along with you” (Faye Creation 2 72)

2.4 Subject marking on the customary aspect The customary suffixes are -na “singular” and -wene “plural”. They often appear with the non-instantiative clitic (so-called because it refers to the designated event “in general”, not to any particular instantiation) =qwe, which can often be glossed as “can”. Note that the plural of the customary is homophonous with the plural of the future imperative (shown in (5)b). The customary suffixes are illustrated in (13). (13) a.

’atíre qwe=l mélen naxáni-sh kwew-kwáw-ya-na very can-2/3abs much man-npn dup-shout-yax-cus.sg “The man is too noisy.” (FN 4 107 94) b. mé=qwe=sh qáy hísh mi-chákwi-n-wene and=can=1pl.abs not thing 3pl.ob-grab-in-cus.pl “And we grab at nothing” (every time we try to catch something) (Faye Creation 069).

A chart of all the imperfective suffixes appears in Table 3. Table 3.Object pronominal-argument prefixes Non-future

sing pl

Future/irrealis

Present

Past

Customary

-qa -we

-qal -wen

-na -wene

-nash -wene

2.5 “Stative” aspect: An apparent plural imperfective aspect suffix with

semantically singular subjects of statives Number agreement in Cupeño is complicated by the fact that with stative verbs, the suffixes -we, -we-n, and -wene, which normally appear agree with plural subjects, appear with singular subjects. This behavior is shown in the following examples.

213

214 Jane H. Hill

(14) ’axwá-’aw kúpa-’aw pe-mí-yax-we-n that-at Cupa-at 3sg-be-yax-imp.pl-past.imp.pl pe’áw’ay yút-ax-we. mountain stand-yax-imp.pl “There at Cupa there used to be a mountain standing.” (Eagle I 001) (15) Mu-ku’ut pé’ pé-ye pé-yax, “Wím-yax-wene.” and-rep det 3sg-mother 3sg-say heavy-yax-cus.pl “And it is said his mother said, ‘It is heavy’.” (Faye KP 105 83) In (14) there is only one mountain. In (15) the discourse subject of the customary verb is a single filled carrying net. However, in both sentences the aspect suffix looks like a plural form. While most stative verbs exhibit the -yax thematic suffix, there is also a small group of statives in the 0 class, such as ’áyew “to like”, kík’isr “to be nothing there”, néne “to be going around, travelling”, and téw “grow (of plants)”. These verbs are referred to as “stative” because they exhibit the non-future affix suffix -we with singular subjects. A stative verb in the 0 class is illustrated in (16). The verb root ’áyew with imperfective suffix -qa and singular subject means “want”. Here, with a plural suffix, but a singular subject encoded as the independent pronoun né’ “I”, it is interpreted as stative “like”: (16) né’ ’áyew-we ’á-yka ’e-’ú’la-qali-ve 1sg like-imp.pl there-to 2sg-sew-ds.sg-subr “I like your sewing” (Faye 2–6–27 5) Historically the stative and the plural imperfective suffixes were distinct. A passivestative suffix *-wa can be reconstructed for Proto-Uto-Aztecan (Heath 1998). In Cupeño this passive has fallen together phonologically with the imperfective plural, which, as will be discussed below, comes from an old position verb, reconstructed for Proto-Uto-Aztecan as *-wt˙nt˙ “stand”. However, in sychronic perspective the two suffixes are phonologically identical. The only justification for distinguishing them is that the stative suffix does not distinguish singular and plural number. 2.6 Subject number marking in imperatives and the immediate future Imperatives and the immediate future exhibit a distinctive pattern of agreement with subject number, marking the plural with the suffix -m, which also appears on nouns. While these constructions are not relevant to the present paper, I review them in order to complete the descriptive overview of number in verb constructions. 2.6.1 Imperatives The details of the phonological derivation of imperatives are beyond the scope of


this paper. In brief summary, in singular imperatives verb themes of the -in and -yax class lose their final consonant and are followed by a suffixed glottal stop, often followed in turn by an echo vowel. The same pattern is seen in vowel-final 0-theme verbs. In consonant-final 0-theme verbs, we observe metathesis of the glottal stop with the final consonant, along with certain other complications. Plural imperatives add -(V)m to the verb theme, as shown in (17b). This is the same as the usual suffix for noun plurals, but imperatives are not nominalizations, since they are never marked with “non-possessed-noun” (npn) suffixes.4 (17) a.

háwi-’! sing-impv.sg “Sing, you!”. b. háw-in-em! sing-in-pl. “Sing, you all!”

2.6.2 The “immediate future” The immediate future is a tense that has developed in Cupeño from a nominalization that meant something like “the one(s) whose purpose is to do X”. Some examples of the construction still have this meaning. Immediate future verbs are formed by adding the npn suffix -t to the present-tense singular form. If the subject is plural, the noun plural suffix -m is added, but the present-tense suffix (the truncated form of the non-future imperfective) remains singular, e.g. (18) a.

háw-i-qa-t sing-in-imp.sg-t “Singular is going to sing” b. háw-i-qa-ti-m sing-in-imp.sg-t-pl “Plural are going to sing”

2.7 Verb-root suppletion for number Eight suppletive verb roots agree with the subject in the case of intransitive verbs, and with the object in the case of transitive verbs. Dixon (1994) suggests that this pattern is universal where verbs are suppletive for number. These verbs are listed in (19). (19) a. chúlup (singular), súlul (plural) “to go in, push in” b. híw (singular), qá (plural, non-future); máx (plural, future) “to be in a place, dwell, sit) (note that this verb also suppletes for tense) c. méq (singular), chíxnin (plural) “kill” (note that this is the causative of chíx “die, plural subject”).

215

216 Jane H. Hill

d. e. f. g. h.

púlish (singular), múyaq (plural) “to go out” qá (singular), wé (plural) “be there” qaáw (singular), chíx (plural) “die, be sick” xálew (singular), yévev (plural) “fall” yá’ (singular), ngén (plural) “to run”

An example of the agreement pattern can be seen with the verb chúlup–súlul. In example (20a), chúlup, appearing in the -yax class as an intransitive, agrees with a singular subject. But in (20b), chúlup, in the -in class and transitive, has a plural subject, but the verb root agrees with its singular object. All of the suppletive verbs except for (19h), yáÁ (singular), ngén (plural) “to run”, are unaccusative verbs in the intransitive variant. Thus the agreement pattern may simply be with underlying objects, which have been promoted to subjects in the case of the unaccusatives (like (20a), but remain subjects in the case of transitives (like (20b)).5 (20) a.

chúlup-yax=pe. go.in-yax=fut “He may come in” (Faye Notes 017) b. túm súpluwet ’ishmiví-y chúlup-pe’-men-weni even.if one thing-ob go.in-3pl-in.pl-ds.pl “even if they put in only one thing…” (Faye Images 249.7)

3.

The limits of subject number marking in the verb construction.

When we examine all of the possibilities for subject–number agreement in the Cupeño verb construction, we find the following: There are five sites where subject number can be marked: Subject PN prefixes, the thematic suffix -in/-men, the imperfective aspect suffixes, the past-tense suffixes on imperfectives, and the suppletive verb roots. However, there are never more than four such markers in any verb construction. The maximally-marked construction, with four marks for subject number, is the past imperfective plural of -in class verbs. Examples given above of this construction are repeated in (21): (21) a.

wíchax-ne-n-qa-l 3pl.ob-throw-1sg-in-nf.imp.sg-pst.sg “I was throwing it.” b. wichax-pe’-men-we-n 3pl.ob-throw-1pl-in.pl-nf.imp.pl-pst.pl “They were throwing it.”

Even where the verb construction includes one of the number-suppleting verb roots, the number of marks for subject number remains no more than four. This is because


when suppletive verbs appear are in the -in class, which is possible for chúlup-súlul “put in”, púlish-múyaq “to take out”, xálew-yévev “to drop”, and chíx-ni-in “cause (pl.ob) to die”, they are transitive, and the agreement of the root is with the object. 3.1 Formal and functional accounts of number-marking constraints

in the verb construction I present here two accounts of the limits on number marking in the Cupeño verb construction. The first is a language-specific historical account, developed by Jacobs (1975) and Heath (1998), that explains the limits on number marking as the result of the grammaticalization of light verbs into thematic and imperfective suffixes. This account is entirely plausible as a description of the historical processes through which the Cupeño verb construction has developed. However, it does not provide any constraints on grammaticalization that would explain the limits mentioned in Section 3.0. The second is a formal syntactic account, that explains the limits on number marking as the outcome of general constraints on the syntax of verb constructions. The syntactic account exploits Jelinek’s (1998, 2001) and Jelinek and Escalante’s (2000) proposed structure for “strong transitivity” in Yaqui, a relative of Cupeño in the Uto-Aztecan language family. 3.1.1 The historical morphology of the Cupeño verb Heath (1998) and Jacobs (1975) proposed that many of the thematic and tenseaspect suffixes in the Cupeño verb are grammaticalized light verbs. Under this suggestion, multiple number marking in the Cupeño verb construction is the result of the grammaticalization as aspect suffixes of number-suppleting stance and motion verbs. According to the Heath and Jacobs accounts, the history of these grammaticalizations, and related syntactic changes, can be reconstructed as follows. First, Jacobs (1975) argued that a person-marked possessed deverbal noun underwent phonological change and was refunctionalized as a past tense. Thus a form like Proto-Cupan (22a) became a form like (22b), by a very regular historical process that reduces and then deletes final vowels. (22) a. *ne-max-’i (reconstructed form) 1sg-give-psd “my giving” b. né-max 1sg-give “I gave.” The requirement that the subject of past-tense verbs be marked by a prefix (the same prefix that marks the possessor of possessed nouns) then derives from the origin of these verbs in possessive constructions.

218 Jane H. Hill

Second, a series of grammaticalizations took place by a process that Heath (1998) called “hermit crab” grammaticalization, named for a type of crab that lives in a suitable shell cast off by some other animal. When the hermit crab outgrows its shell, it abandons it and seeks a larger one. In this type of grammaticalization, a grammatical element erodes phonologically, creating a problem for perception. The slot and associated meaning — the hermit crab — where this grammatical element resides attracts a new “shell” in the form of a fuller phonological form which is felt to resemble the original grammatical element, because of some features of shared phonology and semantics. The new “shell” then encodes the old meaning, and begins to erode in its own turn. The difference between Heath’s notion of “hermit crabs” and standard accounts of grammaticalization (e.g. Hopper and Traugott 1993) is that we require not only a suitable candidate for grammaticalization — the new shell — but an organism — the old meaning and position slot — that can inhabit it. The first example of hermit crab grammaticalization relevant to our history involves the marking of transitivity. Heath (1998) pointed out that Proto-UtoAztecan (PUA) contrasted transitive *-a and intransitive *-i. This distinction was reversed in the ancestral Cupan language by the following process: a.

PUA had a very productive causative suffix *-ina. In the Cupan languages, this suffix (which may never have been a full verb) replaces the PUA transitive *-a, probably because the latter suffix was threatened by the general tendency toward loss of word-final vowels. By the regular loss of the final vowel, *-ina becomes -in, the thematic suffix for transitives b. Proto-Northern-Uto-Aztecan (PNUA) had a mediopassive suffix *-kG. By regular sound change this appears in Cupan as *-ax. By regular rule, the suffixfinal /x/ is lost before /q, k/, the first element of some very common suffixes. Since the “transitivizing” slot has been filled by -in, -a(x) can be reinterpreted as stative-intransitive instead of mediopassive. That is, the PUA -i/-a opposition of transitivity is retained, but reversed! c. Cupan yax originates as a light verb meaning “be, say” (Jacobs 1975). Because it is a “light verb” of very general meaning (and also, I believe, because it is among the class of roots that bears no inherent stress (Hill and Hill 1968)), it is an excellent candidate for grammaticalization. It provides a suitable new shell for the stative-intransitive meaning and slot, eroded by regular phonological processes from intransitive *-ax to -a. yax replaces -a as a new “shell” for the hermit-crab intransitivity slot. The association with the light verb remains, though, so that yax in its new suffixal role brings with it its PN prefixes. By analogy, the transitivizing -in (which, recall, was probably never a full verb) acquires PN prefixes as well.


The second set of hermit-crab grammaticalizations involve the imperfective aspect suffixes. PUA had an agentive suffix *-qa, which shows up in Cupan as a nominalizing agentive suffix that takes absolutives (singular -qa-t, plural -qa-t-im). In Cupeño this construction acquires a new function as an immediate future, discussed in 2.6.2 above. Once -qa was reinterpreted as a tense-aspect marker, it was possible to abandon the nominalizing absolutive suffix -t, yielding -qa. The regular vowel loss and spirantizations reshaped this element to -ax, and finally to -a. The new shell for this hermit crab was another light verb, PNUA *qal, “to be in a place, dwell”, derived from PUA *katG “sit”. Like yax, *qal, which still exists as a lexical verb in Cupeño, lacks inherent stress. The lexical verb qal is restricted to singular subjects. When it becomes a new “shell” for eroded -qa, qal brings with it this agreement pattern. It retains the “imperfective” implicature of the agentive which is of course reinforced by the association with the imperfective Aktionsart of a verb meaning “sit, dwell, be there”. In Cupeño, it comes to exist alongside the ancestor of -qa, now the immediate future, still appearing in the immediate future construction -qa-t(-im). Since this new imperfective suffix -qal is singular, a second light verb grammaticalizes to cross-reference plural subjects. This is PUA *wGnG “stand”, appearing in Cupeño as the stressless lexical verb root wen “lie, put”, which also has light-verb function, “be there” of plural subjects, in alternation with qal with singular subjects. The final consonants of -qal, -wen are lost, by the regular processes mentioned above. In the past tense they were protected for a while by the old possessed nominalizing suffix -’i, until this was lost in its own turn. This, however, led to the reinterpretation of the original final consonants -l, -n as past-tense suffixes for the imperfective aspect (see note 2). As mentioned in the discussion of stative verbs in 2.5, the new imperfective plural suffix -we falls together with the old PUA passive suffix *-wa. *-wa becomes -we in Cupan by regular rule, looking exactly like -we from *- wGnG, and yields an “irregularity”: in statives, -we can occur with singular as well as plural subjects. The future tense develops when PUA *-ni “future” erodes phonologically and eventually disappears, yielding the “naked” future perfective forms seen in Cupeño. By analogy with past tenses, which have acquired aspect distinctions by the processes discussed above, an imperfective develops for the future. The source for the new imperfective suffix is provided in Cupan by yet another high-frequency verb referring to a stance, násh “sit”, which shares the /n/ of the old future suffix. By analogy with the non-future aspect markers, this future imperfective is interpreted as a singular. Jacobs (1975) suggests that a plural future imperfective -wene probably develops by suffixing to the non-perfective aspect -we the very common motion verb ném “go around” (with good imperfective Aktionsart), which erodes to -ne. While there is much of interest in this account from the perspective of grammaticalization, there are several problems with it. First, Jacobs’ account of the origins

220 Jane H. Hill

of the Cupeño subject prefixes as possessive prefixes neglects the fact that subject pronominal prefixes are widespread in Uto-Aztecan languages. Given this, it would have been appropriate to consider the possibility that subject prefixes might be original, but have been lost in Cupeño on non-past-tense verbs. Second, number marking in tense-affix suffixes appears in other U-A languages. For instance, such suffixes are found in Nahuatl. Thus number marking in these suffixes may reflect an ancient pattern. Nor does the Jacobs-Heath account provide an explanation of -men, the plural form of -in “transitivizer”. One difficulty with the theoretical framework of grammaticalization adopted by Jacobs (1975) and Heath (1998) is that it is relatively unconstrained. Heath (1998) has attempted to constrain grammaticalization theory by suggesting that grammaticalizations are more likely when these provide “shells” for phonologicallyeroded meaning slots. However, the kinds of historical accidents that Heath (1998) invokes — the phonological erosion of an affix, the presence of a phonologically similar full lexical item with a somewhat bleached semantics — can happen again and again, but we do not encounter in human languages all the kinds of verbs that such historical accidents might produce. What is missing in his account is any constraint on the inventory of possible structural slots or on their agreement behavior. The only explanation for the particular inventory of affix positions in the verb is historical: original slot-meaning combinations appear at some point, and persist because of phonological reinforcement by hermit-crab grammaticalization. In the case of agreement, there are a number of derivational suffixes in Cupeño that do not exhibit subject–number agreement. Examples include -ni “causative”, discussed in Section 2.2 (examples (10a,b)), -luw “to go to do”, -ngiy “going away doing”, -mi’aw “arriving doing”, -vichu “desiderative”, etc. That this is a serious question is shown by the fact that in the sister Uto-Aztecan language Nahuatl, the purposive motion suffixes -ti/-tiw “to go to do” and -kí/-kiw “to come to do” do display subject number agreement. 3.1.2 A syntactic account: Cupeño as a strong transitivity language Recent work by Jelinek (1998, 2001) and Jelinek and Escalante (2000) on Yaqui, a Uto-Aztecan language distantly related to Cupeño spoken in southern Arizona and in Sonora, Mexico, provides the kinds of constraints on slot inventory and agreement pattern that are needed to complement the historical account proposed by Jacobs and Heath. Jelinek argues that Yaqui is a “strong transitivity” language. By this is meant that the underlying structure of the Yaqui verb includes the functional heads, Tran and Voice. Jelinek also treats aspect, overtly marked in Yaqui. as a functional head Asp. With the additional functional projection TnsP, overtly marked in Cupeño, there are four sites that can accept marking for subject number. For reasons of space, I treat here only transitive sentences. Figure 1 shows Jelinek’s (1998: 205) proposed structure for the Yaqui Transitive Active clause.

The Cupeño verb construction 221 a.

AspP Asp′

Peoi VoiceP ti

Asp -k Voice′

TranP Maria

b.

Voice Active Tran′

V +Tran ’ani-a Peo Maria-ta ’ani-a-k Pete Mary-acc help-tr-perf “Pete helped Mary”

Figure 1.Yaqui Transitive Active Clause.

Like Yaqui, Cupeño has overt markers of transitivity: the thematic suffixes -in “transitive” and -yax “intransitive”. It has an overt marker of voice in the form of the stative -we where subject number agreement does not hold (discussed in 2.5). In addition, it has overt markers for aspect in the form of the imperfective suffixes, and for tense. On this morphological evidence, it seems likely that the array of functional projections that Jelinek advances for Yaqui are also required for Cupeño. Recently Bowers (2002) has argued that a functional head for transitivity (his Tr) must be recognized in UG, replacing “v” of Chomsky 1995b. Bowers cites morphological marking of transitivity in San languages (Collins 2000, cited in Bowers 2002) as one line of evidence for Tr; the Uto-Aztecan languages provide additional examples. Following Jelinek’s proposal for Yaqui, the structure of a Cupeño transitive verb in the past imperfective active is diagrammed in Figure 2. Note that, like Yaqui, Cupeño is verb-final. As pointed out above, the past imperfective is the verb construction that has the maximum number of subject–number agreement marks. The verb construction diagrammed, misúlulche’menwen “we were pushing them in”, exhibits the plural root of the suppletive verb chúlup~súlul. In this case, the root cross-references the number of the object. In Figure 2 I follow Jelinek (1998) in showing the object as originating at specifier of TranP (although, for the specifics of the morphology, see the discussion below in 3.1.2.2). For Bowers, the object would originate at Comp of VP and raise by “Short Object Movement” to specifier of his TrP (=TranP), satisfying the

222 Jane H. Hill AgrSP

a.

AgrS′

proi

AgrS chem-

TP T′

ti AspP

(IV) T -n

Asp′

ti VoiceP

(III) Asp [S, P] -we

Voice′

S ti TranP

(II) Voice [S, P] -Ø

Tran′

O miVP

(I) Tran (+) [S, P] -men V [S, P] súlul

b.

Mi-súlul-che’-men-we-n 3pl.ob-push.in-1pl-in.pl-nf.imp.pl-past.pl “We were pushing them in”

Figure 2.Past Imperfective Transitive Verb.

EPP-feature of Tr. (Bowers 2002: 187). This treatment of the object could be the start of an explanation for the “agreement” between the suppletive verb root súlul and object number; this matter is left for further research. Following Jelinek (1998), I show the subject as originating at the specifier of VoiceP. This subject raises to the specifier of AspP and TP and then to the specifier of AgrS to check EPP features (Collins 1997). In Bowers’ (2002) treatment, the subject originates at the specifier of PrP (“predication” phrase), and raises to TnsP (=TP). It may be that Bowers’ PrP is equivalent to Jelinek’s VoiceP (it occupies the same structural position, between TranP and TnsP), but exploration of this point is beyond the scope of this paper.


3.1.2.1 Functional heads and raising. The head-to-head raisings required for derivation of the past imperfective active construction are shown as romannumbered arrows in Figure 2. These are described below: I.Tran to Voice. Bowers (2002) maintains that Tran has φ features. In Cupeño, of course, these are morphologically evident in the alternation between -in “singular subject transitive” and -men “plural subject transitive”. Voice will also have such features, by Spec–Head agreement with the subject originating at Spec of Voice P. Tran raises to Voice and checks its φ features, as shown: (22) ... [VoiceP Voice [TranP -men [VP ...]]] II.Tran, Voice to Asp: the merged complex, Tran, Voice has φ features. Asp also has Subject φ features. Voice (active voice has a zero morphological realization) checks its φ features at Asp, as shown: (23) ... [AspP Asp [VoiceP -meni, -Ø [TranP ti [VP ...]]]] III.Tran, Voice, Asp to T: the merged complex Tran, Voice, Asp retains φ features from Asp. T has subject φ features by Spec–Head agreement, and can check φ features of Asp, as shown: (24) ... [TP T [AspP -meni, -Øj, -we [VoiceP ti, tj [TranP ti [VP ...]]]]] IV.Tran, Voice, Asp, T to AgrS: this merged complex retains φ features from T. These check at AgrS. (25) ... [AgrSP AgrS [TP -men-Ø-we-n [AspP ti, tj, tk [VoiceP ti, tj [TranP ti [VP ...]]]]]] After the movement shown in (25), the linear morphological sequence che’-menwe-n is the resulting phonological string. In summary, unlike the historical-morphological explanation of Cupeño number marking, this syntactic account makes a prediction: We will never encounter more than four sites for subject number agreement in a polysynthetic verb. TranP, VoiceP, AspP, and TP would seem to exhaust the inventory of relevant functional projections. Of course complex aspects and tenses do appear (as inventoried, for instance, in Comrie 1985). However, this account predicts that the markers of such complex tenses will not share PN or case feature markings in a single polysynthetic verb. A remaining problem is the treatment of the verb root súlul. The lexical item diagrammed in Figure 2 is mi-súlul-che’-men-we-n “we were pushing them in”. The

224 Jane H. Hill

conventional treatment of the relationship between V in VP and the “light” verb in the functional head v (here Tran) is that V should raise to check its transitivity features at Tran. However, this would yield the ungrammatical sequence *mi-chemsúlul-men-we-n. Heidi Harley (p.c., June 11, 2001) suggests that a possible treatment which would yield the correct sequence would be to delay the merger of V, súlul, comparable to treatments of English do-support, merging súlul by linear adjacency (see also Hale this volume), rather than structural adjacency (Bobaljik 1994), with the output of the head-head raising shown in (24). Barragan (2001) adopts a version of this approach, raising v to AgrS to check features at AgrS. The verb root, súlul, does not move because, as Barragan notes, “Thematic class verbs have an available lexical head (the -in or -yax morpheme) which causes movement to bypass the root in favor of the thematic suffix. Zero class verbs have no available head at v, and because of this, attract the root for morphological support.” This accounts for the fact that in zero class verbs, the subject marker is prefixed to the verb root. 3.1.2.2 The position of the object prefix. A final question is the initial position of the object prefix mi- “3plob”. Within Uto-Aztecan, this peculiarity is specific to Cupeño. For instance, in Nahuatl, another Uto-Aztecan language with strong transitivity, number marking in the tense-aspect suffixes, and both subject and object prefixes, the subject prefixes always precede the object prefixes, as would be expected if the object originates in VP, raising to Spec of TranP, and the subject originates in VoiceP, raising to Spec of TP. In Cupeño, there is some evidence that the object prefix originates as part of the Focus complex (Kiss 1995) and cliticizes to the verb construction. In brief, the lines of evidence for this are as follows. First, there is a focus marker -i, homophonous with the (optional) object suffix on nouns. -i also marks the object forms of PN prefixes in the verb construction, as shown in Table 2 above. Second, object prefixes in the verb construction are optional (as are object suffixes on nouns). They occur mainly on verb constructions in “peak” or climactic components of discourse. Third, object prefixes never exhibit stress when they appear with unstressed verb stems, even when, as in the case of the future perfect or immediate future constructions, no other stress-accepting affix is present. Both subject prefixes and aspect suffixes do accept stress when affixed to unstressed verb roots (if a subject prefix is present, it will have primary stress. In non-pasts, the aspect suffix will have primary stress). Fourth, I have identified hesitation forms where the break occurs between the object prefix and the remainder of the verb construction, as in (25). Dysfluencies at other points in the verb construction do not seem to occur (as apparently is typical in polysynthetic verb constructions, much as dysfluencies interrupting “words” are unusual in isolating languages).


(25) Me ’áye ’atáx’a-m mi-, mi-kwáw-pe’-men-we-n and then person-pl 3pl.o, 3pl.o-call-3pl-in.pl-nf.imp.pl-pst.pl “And then they called them, them, the people.” (Burning 012) Another example, where the speaker abandons a verb construction in favor of an expression using a preposition, is seen in (26): (26) Me ’áye mi, ’áye mi, ’áye pé’-men hásri-pem-yax-we-n. and then 3pl.o, then 3pl.o, then 3pl-with go-3pl-yax-nf.imp.pl-pst.pl “And then them, then them, then they went with them.” (Burning 015)

4. Conclusion The Cupeño verb construction can exhibit agreement with subject number in inflectional suffixes marking transitivity, aspect, and tense, as well as in PN subject affixes. On the approach outlined here, informed by Jelinek’s analysis of “strong transitivity” in Uto-Aztecan languages, multiple number marking in the Cupeño verb construction can be accounted for and constrained. The hypothesis of strong transitivity constrains the theory of “hermit crab grammaticalization” advanced by Heath (1998) by predicting what types of elements will grammaticalize as affixes in the verb construction. This restricts the theory of grammaticalization beyond Heath’s suggestions about phonological and semantic congruity between the grammaticalizing element and its target slot, by requiring the availability of a syntactic position as well. Adapting Jelinek’s work to the analysis of Cupeño not only permits us to deeply explore the structure of a very interesting language, but it shows areas in which functional and formal theories of language can contribute complementary accounts of historical and synchronic processes.

Notes * I would like to thank Luis Barragan, Andrew Carnie, and Heidi Harley for comments on earlier drafts of this manuscript. Ken Hale, who attended the meetings of the Arizona Uto-Aztecan Workshop in Spring 2001, provided comments and encouragement at the initial presentation of this material, as did Eloise Jelinek. My recent descriptive work on Cupeño was undertaken with the support of a fellowship at the Research Centre for Linguistic Typology, LaTrobe University, Bundoora, Victoria, Australia in Fall 2000. Alexandra Aikhenvald and R. M. W. Dixon provided extensive comments on the initial descriptive efforts and their typological implications. All errors of course remain my own responsibility. 1. Before 1902 Cupeño was spoken only in two villages, Kupa and Wilaqalpa, in northeastern San Diego County. Materials on the language come from nine speakers. Paul-Louis Faye, who

226 Jane H. Hill

collected texts and grammatical materials in three brief field trips in 1919–1920, 1920–1921, and 1927, worked with Domingo Moro, Carolina Nolasquez, Salvadora Valenzuela, Carolina Welmas, and a “Mrs. Chavez”. My field work in 1962 (for 3 months) and 1963 (for a month) was mainly with Roscinda Nolasquez of Pala, California, but I collected a few texts from Frances Bosley of Morongo and songs and some lexical material from James Brittian of Hemet. Roderick Jacobs, working in the late 1960’s, studied with Roscinda Nolasquez and Cyrillo Welmas (the latter a speaker from Wilaqalpa; the other speakers are all descended from families who lived at Kupa). In addition to these materials, there are scattered forms published by Kroeber (1907), and in anthropological studies such as those of Strong (1929) and Gifford (1918, 1922). J. P. Harrington collected some material on the language (some forms appear in Harrington 1933). Thus material on Cupeño constitutes a relatively small closed corpus, and many questions that contemporary linguists wish to ask about the language can never be answered. 2. Historically, qal and wen are light verb roots. Barragan (2001) has developed arguments for treating the synchronic sequences as single morphemes, -qal and -wen, in which case the present imperfective suffixes, -qa, -we would be treated as a case of tense-marking by truncation. 3. In the examples, I gloss object pronominals with ob, e.g. 2sg.ob “Second person singular object”, but gloss subject pronominals only with person and number, e.g. 1pl “First person plural (subject). Imperfective verbs are marked as imp, but perfective verbs are unmarked. Similarly, there is no gloss “future” or “past”: if the subject pronominal prefix is present, the verb is in the past tense. If no subject marker or other affixal material is present (other than future imperfectives and thematic suffixes), the verb is in the future tense. Sources can be retrieved from the parenthetical material at the end of the examples. If the source is from Faye’s field notes from the 1920’s, this is indicated as “Faye…”. Elicited sentences from my own field notes from 1962 and 1963 have notebook number, page, and example number. Material from taped text performances from my own field sessions are indicated with text titles and sentence numbers. The sentences are given exactly as performed unless otherwise indicated. All examples from the 1960’s in this paper are from the speech of Roscinda Nolasquez unless otherwise indicated. Abbreviations are as follows: abs absolute case, acc accusative (a frozen case marker with a few nouns), caus causative, cond conditional, cus customary, ds different subject, dub dubitative, dur imperfective, erg ergative case, fut future irrealis, i -i subordinator (simultaneous action), in -in theme-class suffix, inch inchoative, irr irrealis, NF non-future, NI non-instantiative, npn non-possessed noun, ob object case, pl plural, pro independent pronoun. psd possessed-noun suffix, pst past, r realis, rep reportative, sh -sh absolutive nominalizer, ss same subject, subir irrealis subordinator, subr realis subordinator, t -t absolutive nominalizer, yax -yax theme-class suffix The practical orthography of Cupeño uses the following symbols in non-canonical values: ·’Ò glottal stop, ·´vÒ primary stress, ·eÒ mid-central vowel, ·srÒ retroflex s. 4. A confusing usage in Uto-Aztecan philology is that non-possessed nouns are suffixed with one of a series of consonants (in Cupeño these are -t, -l, -sh). These non-possessed consonants are known to Uto-Aztecanists as “absolutive” suffixes. Unfortunately, unique among Uto-Aztecan languages, Cupeño distinguishes an “absolutive” from an “ergative” case in the PN clitics. The “absolutive” non-possessed noun suffix is not a case marker at all. For this reason, the noun suffix usually referred to as “absolutive” is here called “non-possessed noun” suffix (npn). 5. I owe this observation to Heidi Harley.

Lexical irregularity in OT DOT vs. Variable Constraint Ranking* Diana Archangeli University of Arizona

1.

Introduction

Some phonological patterns hold of monomorphemic forms but do not necessarily hold of polymorphemic forms. For example, in Japanese, two voiced obstruents rarely co-occur within a single morpheme. Yet when two morphemes are combined, forms with multiple voiced obstruents are possible: kuzu+kago ‘wastebasket’, geta+bako ‘footwear case’, naga+gutsu ‘long shoes, boots’ (Itô & Mester 1986). Under the view that a grammar encodes generalizations about forms in a language, these morpheme-internal patterns are part of a grammar; they are commonly known as morpheme structure constraints (MSCs).1 Despite such constraints, there are anomalous lexical items, ones that do not conform to the MSCs that generally hold in the language. These anomalies challenge us to find a principled means of formally characterizing both the MSCs and the exceptions to them. Prior to the advent of Optimality Theory (Prince and Smolensky 1993, McCarthy and Prince 1993a,b), MSCs tended to be characterized either by rules, by constraints, or by a combination of rules and constraints, holding of a particular language. Exceptions were characterized by one of two means: prespecification or “co-phonologies”. The idea of prespecification is to lexically specify (“prespecify”) certain features and thereby prevent the application of the relevant rules/constraints. The idea of co-phonologies is to propose a phonological system that characterizes the exceptional case(s) and that operates in tandem with the phonology of the regular cases. Under rule-based accounts, the co-phonology view is exemplified by a set of rules whose application is limited to certain morphemes (Chomsky & Halle 1968, Kisseberth 1970, Kiparsky 1982, etc.). With the advent of Optimality Theory, the possibilities shift somewhat. The notions of co-phonologies and of prespecification translate fairly readily into the Optimality modal. With co-phonologies (e.g. Kisseberth 1993, Kirchner 1993, Itô and Mester 1993, 1995), MSCs are characterized by constraints, which are motivated by alternations. Exceptions are accounted for through different constraint

228 Diana Archangeli

rankings for different morphemes. With prespecification, (e.g. Inkelas, Orgun and Zoll 1997), MSCs are not characterized in the constraint hierarchy at all. Patterns and exceptions are simply part of the lexical representation with no attempt at characterizing general patterns. (Alternations alone are expressed by constraints and may hold of morphemes.) A third possibility under Optimality Theory is Direct OT (DOT), a model proposed in Golston (1996). Under DOT, MSCs are expressed by constraints whose domain may be limited by morphological classes. Exceptions are characterized by specifying unexpected necessary constraint violations in the lexical entry. Here, I demonstrate that the lexical regularities and irregularities of vowel distribution in Tiv (Abraham 1940a,b) verbs are readily characterized by DOT, but present serious challenges to both the co-phonology view and prespecification. An intriguing aspect of this project is that, although the analysis itself is grounded in a formalist model, it centers on an explanation of types of occurring exceptions, whose relevance is often emphasized in functionalist approaches. To make this argument, it is necessary to understand the general distribution of vowels in Tiv verbs. I provide a sketch here; for further details and a pre-OT analysis, see Pulleyblank (1988) or Archangeli and Pulleyblank (1994).

2.

A sketch of the distribution of vowels in Tiv verbs

There are six vowels in Tiv, two high [i–u], two mid [e–o], and two low [a–f], with a round and an unround member of each pair. Having six vowels and no restrictions on their distribution, in a CVCV form,2 we expect 36 possible vowel sequences. Instead, in verbs we find that ten combinations are common, the others rare or non-occurring. (Nouns do not follow this pattern.) (1) Vowel distribution in (C)VC2V(C) verbs notes: C2 is not [h]; V1 may be long; number of examples found is given in each cell; robust patterns are shaded; n = @ 500 V2 Æ V1 Ø i

i

e

a

37

33

48

e

102

a

85

f

1

2

o u

36

33

f

o

u

1 48

5

31

4 45

Lexical irregularity in OT 229

There are three general patterns here.3 First is full identity of V1 and V2 when V1 is not low: [i…i], [e…e], [o…o], and [u…u]. Second, when V1 is low, V2 is mid, but agrees in rounding, [a…e] and [f…o]. Finally, there are also a significant number of examples where V1 is high [i, u] and V2 is either mid or low [e, a]: [i…e], [i…a], [u…e], and [u…a]. I treat the first two classes here, and the remaining cases in Sections 3–5. Let us consider first the forms with complete identity between the two vowels, which occur only when the first vowel is not low. (2) Case 1: Identity in (C)VCV sequences if V1 is not low i

e

a

f

o

u

i

tíndi

‘to send’

e

bènde

‘to touch’

o

dóndo

‘to adjoin’

u

hùndu

‘to muddy’

To account for these patterns, we might posit the set of constraints in (3): faithfulness to [+high] and [+round] is critical; [+high] is well-aligned to both the left and right of the verb, as is [+round]. Otherwise, the features for a mid vowel are preferred.4 (3) Constraints (following McCarthy & Prince 1993a,b, 1995) a. FaithHi, Rd = MaxHi, MaxRd, DepHi, DepRd (standard definitions for [+high] and [+round]) b. AlignHiL: Align([High], L, verb, L) c. AlignHiR: Align([High], R, verb, R) d. AlignRdL: Align([Round], L, verb, L) e. AlignRdR: Align([Round], R, verb, R) f. VisE = avoid [+High], avoid [+Round], Avoid [+Low]: net effect: prefer the features that create [e] At this point, the only critical ranking is that VisE is subordinate to all other constraints noted above. The tableaux in (4) and (5) illustrate the interaction of these constraints. With the feature [+high], as in [míshi] ‘to sprinkle’ (4), [+high] may not be lost, due to FaithHi (candidate a), and [+high] must align (and so link) to both the right and left edges of the form (violated in candidates b-c). The preferred candidate, with two [i]s, has a double violation of VisE, but nonetheless is the best of the options.


(4) [míshi] ‘to sprinkle’ m´vshv [+hi]

FaithHi,Rd

a. méshe

AlHi

AlRd

VisE

*!

MaxF & DepF [-hi]

b. míshe

*!

i

[-hi]

c. méshi

*!

i

[-hi]

d. míshi

ii

With the input feature [+round], as in [dóndo] ‘to adjoin’ in (5), loss of [+round] or addition of [+high] creates a faithfulness violation (candidates a, b). Where [+round] is not aligned at both edges, fatal violations are also accrued (candidates c,d). Again, the winning candidate, (e), violates only VisE, due to faithfulness to and alignment of the feature [+round]. (5) [dóndo] ‘to adjoin’ d´vndv [rd]

FaithHi,Rd

a. dénde

*!

b. dúndu

*!

AlHi

AlRd

VisE

MaxF & DepF [-hi, -rd]

uu

c. déndo

*!

o

[-hi, -rd]

d. dónde

*!

o

[-hi, -rd]

oo

[-hi]

e. dóndo

The effect when both [+high] and [+round] are present is straightforward. Neither feature could be lost, due to the high-ranking of faithfulness to [+high] and [+round], and both features would necessarily align at both the right and left edges of the form, thereby creating a (C)uCu word like hùndu ‘to muddy’. A form with neither [+high] nor [+round] in the input would render moot the undominated constraints. The unmarked vowel, due to VisE, would then emerge, deriving the class of (C)eCe verbs like bènde ‘to touch’. The constraint hierarchy sketched above accounts for the most common patterns when the initial vowel is not low. Let us turn now to the cases where that initial vowel is low. In such forms, the second vowel is invariably mid, and agrees in rounding with the first vowel.

Lexical irregularity in OT

(6) Case 2: With a low V1, V2 is mid and agrees in rounding i

e

f

a

o

u

a

sàase

‘to comb hair’

f

k´fndo

‘to go uphill, to rise’

These forms motivate a high-ranked faithfulness to [+low] (7a), along with alignment of [+low] to the left of a form, but not to the right edge (7b). (See footnote 4 for definitions of Max, Dep, and Align.) (7) Additional constraints a. FaithLo = MaxLo, DepLo (standard definitions for [+low]) b. AlignLowLeft (AlLo): Align([Low], L, verb, L) The effect of these additional constraints is illustrated by the tableaux in (8) and (9). (Candidates violating the high-ranked FaithHi and FaithRd are omitted.) With only the feature [+low] in the input, illustrated in (8), loss of [+low] violates FaithLo (candidate a). If [+low] is not linked at the left edge, AlLo is violated (b). Otherwise, the unmarked VisE takes over, preferring the features resulting in [e] to any other. Thus, sàase (d) wins over *sàasa (c).5 (8) [sàase] ‘to comb hair’ s´vvsv [lo] a. sèese b. sèesa

FaithHi,Rd, AlLo Lo,Rd

AlHi AlRd VisE MaxF DepF

*! *!

a

c. sàasa

aaa!

d. sàase

aa

The analysis of a form with both [+low] and [+round] is again straightforward. As (9) shows, such forms reveal the combined effect of the relevant constraints — FaithLo, FaithRd, and AlLo, AlignRd.

231


(9) [k´fndo] ‘to go uphill, to rise’ k´vndv [lo][rd]

FaithHi, Lo,Rd

a. kóndo

*!

AlLo

AlHi

AlRd

VisE

MaxF DepF

oo

b. kóndf

*!

of

c. k´fnde

*!

f

d. k´fndo

fo

e. k´fndf

ff!

Key here is that VisE refers to the features that together create the segment [e], not specifically to the segment [e]. Thus, a form with two low vowels, such as *k´fndf (9e) is worse than a form with one mid vowel and one low vowel, such as the winning k´fndo (9d). To review, the most common patterns in Tiv (C)VCV verbs are expressed through the interaction of undominated faithfulness to [+high], [+low], and [+round] in conjunction with the bidirectional alignment of both [+high] and [+round] and the leftward alignment of [+low]. All of these constraints outrank the preference for [−high], [−low], and [−round], expressed by VisE, which in turn outranks faithfulness to any other feature specification. As noted in footnote 3, this analysis extends to the patterns found with longer verbs. This is not the whole story, however. We now examine the major exceptions in the language, and see that this motivates viewing the rightward alignment of [+round] as restricted by vowel height properties. 3.

A major class of exceptions

As noted in (1), there is a significant class of (C)VCV verbs with an initial high vowel and either [e] or [a] as the second vowel. We consider the cases with [e] here.6 Trivocalic forms are included, too: here, only the final vowel is mid. (10) Case 3: A major class of exceptions: hi-(hi)-[e] forms i

e

a

f

o

u

i

[pírem] ‘to be knock-kneed’ [ímbise] ‘to jam into’ *[iCeCe]

u

[kùse] [sùre]

‘to feel ill’ ‘to farm’


In such cases, [+round] does not align to the right. It appears that these forms are exceptional in two ways: they violate both AlignHiR and AlignRdR to the maximum extent possible. This interpretation leads us to expect equally or more robust classes of exceptions that violate only one of these constraints, u…o…o fully violating AlignHiR and u…i…i fully violating AlignRdR. Neither occurs. Forms like ímbise support the view that AlignHiR fails in some forms. If we express AlignRdR in terms of [+high], then the forms in (10) are exceptions solely to AlignHiR. The effect is achieved by aligning [+round] to the right edge of [+high], ranked above the general rightward alignment of [+round]. Under this view, there is no exception to AlignRdR. (11) Rightward round alignment constraints a. AlignRdR-HiR: Align([+Round], Rt; [+hi], Rt): Align the right edge of every [+round] to the right edge of some [+high]. b. AlignRdR: Align([+Round], Rt; Word, Rt) With these constraints, if both [+high] and [+round] are present, the preferred representation aligns [+round] to the rightmost extent of [+high]. Where [+high] is not present, [+round] is well-aligned to the right edge of the word, emergence of the unmarked. The tableaux in (12) and (13) illustrate. With [+high] and [+round], when rightward [+high] alignment is lexically suspended (indicated by the “*L” in (12)), the preferred form has [+round] and [+high] with co-extensive domains (12c). Where [+round] is well-aligned to the right-edge of the word, it is not well-aligned to the right-edge of [+high] and so is eliminated (12b). (12) [kùse] ‘to feel ill’ k`vsv [rd][hi]

FaithHi, Lo,Rd

a. kèse

*!*

AlLo

AlHi

AlRd Hi

b. kùso

*L

*!

c. kùse

*L

AlRd

VisE

uo *

u

By contrast, when there is no [+high] in the form as in (13), AlignRdHi is irrelevant, and the effect of AlRd emerges (13c), not (13b).


(13) [k´fndo] ‘to go uphill, to rise’ k´vndv [lo][rd]

FaithHi, Lo,Rd

a. kóndo

*!

AlLo

AlHi

AlRd Hi

b. k´fnde c. k´fndo

AlRd

VisE oo

*!

f fo

There are certain advantages to this analysis of rightward round alignment. First, the hi-mid and hi-low patterns emerge in verbs regardless of whether [round] is present. Thus, it is necessary to allow for exceptions in which rightward [high] alignment is suspended independently of [+round]. Under the present analysis, the suspension of rightward [high] alignment is sufficient to account for these classes of cases. Were one to posit that rightward round alignment is also suspended, the fact that the two coincide is unexplained. We would expect the non-occurring cases in which rightward [round] alignment alone is suspended (*CuCi, *CoCe) and we would also expect rightward [high] to be suspended while rightward [round] alignment holds (*CuCo). None of these cases occur. Second, there are numerous “height-dependent round harmonies” in the languages of the world. (See Hong 1994; see Kaun 1995 for an Optimality view.) The present analysis is consistent with the existence of such cases, as well as allowing general rightward [round] alignment. To summarize, the basic Tiv verb patterns have been accounted for by ranking faithfulness to and alignment of [+high], [+round], and [+low] above a general preference for [−high], [−low], [−back] vowels. The robust class of exceptions, in which the initial vowel is high and the final vowel is low or mid, is used to motivate the rightward alignment of [+round] to [+high], preferred over the more general rightward alignment of [+round]. With this background, we are now in position to explore the characterization of exceptions to the general patterns.

4. The formal treatment of exceptional lexical items In this section, we turn our attention to the formal implications of the exceptional patterns. To begin, we consider the analysis of the exceptional pattern discussed in Section 3, the imperfect rightward alignment of [+high], in terms of three different OT models of lexical exceptions, Direct Optimality Theory (Golston 1996), Cophonologies (Kisseberth 1993, Kirchner 1993, Itô and Mester 1993, 1995), and Prespecification (Inkelas, Orgun, and Zoll 1997).


4.1 Direct Optimality Theory (DOT) Under Direct Optimality Theory (DOT) (Golston 1996), exceptions are expressed by encoding the necessary violation of a specific constraint in the lexical representation of the exceptional form. That is, part of the lexical representation of an exceptional item is the information that some constraint C must be violated. As we will see with the analysis of Tiv, a variety of different constraints contribute to the exceptions in a single language.7 Candidates accrue violations in two ways in DOT. First, violations in excess of the necessary violations (if any) are accrued normally. Second, the absence of the necessary (lexical) violation counts as a violation. Under DOT, then, lexical items that are AlHiR exceptions (as discussed in Section 3) are characterized by required violations of AlHiR, a gradiently evaluated constraint.8 This is represented directly in the input form as *AlHiR. Necessary violations are marked with *L, for a lexically required violation. Such violations do not count against a candidate because they are lexically required. Consider the tableau in (14). Where [+high] links at the left edge, but does not link to any other vowels, the candidate accrues both a required and a “real” — and fatal — violation of AlHiR, shown in (14a). A completely well-aligned form, counter to the demands of the lexical entry, will receive a fatal “inverse violation” of AlHiR, *I, as in (14b).9 (Candidates violating AlHiL are not considered in (14).) (14) [ímbise] ‘to jam into’ (under DOT) v´ mbvsv [+hi]; *AlHiR

AlHiL

AlHiR

VisE

a. [ímbese]

*L*!

i

b. [ímbisi]

*I!

iii

c. [ímbise]

*L

ii

Notes: *L is a required violation so does not count against the form *I is an inverse violation, indicating the absence of a required violation

The winning candidate, (14c) has the excused — and therefore non-fatal — violation of AlHiR. Were AlHiR simply suspended for this form, candidate (14a) would win; were AlHiR to hold normally, candidate (14b) would be selected. In this manner, the DOT model provides an explanation for both the general exceptions to MSCs in Tiv, such as the example above, as well as for the more parochial exceptions. As expected, virtually every critical alignment constraint in Tiv has exceptions; exception classes with fewer members correlate to forms that lexically list exceptions to multiple constraints.


4.2 The Co-phonology Hypothesis Under the co-phonology hypothesis (Kisseberth 1993, Kirchner 1993, Itô and Mester 1993, 1995), exceptional forms are characterized by ranking constraints differently for different lexical items. As seen above, the critical constraints are AlHiR and VisE for a form like [ímbise] ‘to jam into’. The tableau in (15) offers the co-phonology alternative, namely the reranking of this pair of critical constraints (shown by the boxed columns). If the constraint VisE outranks AlHiR, then the preferred candidate is one in which [+high] satisfies AlHiL and otherwise maximally satisfies VisE (and maximally violates AlHiR), namely the incorrect (15b). Without reranking, the preferred form maximally satisfies AlHiR (and maximally violates VisE), (15c). Neither ranking produces the attested (15a), with only a single violation of the gradient AlHiR. (15) [ímbise] ‘to jam into’ (under the Co-phonology Hypothesis) v´ mbvsv [+hi], VisE » AlHiR

AlHiL

VisE

AlHiR

( ) a. [ímbise]

ii!

*

i

**

b. [ímbese]

( ) c. [ímbisi]

ii!i

The correct (15a) cannot be selected by the simple reranking of these two constraints because the gradient AlHiR is only violated once in the optimal form. Reranking removes the gradience of alignment, and creates a categorical effect instead. Efforts to salvage the co-phonologies hypothesis in Tiv might include introducing additional (and otherwise unnecessary) constraints or introducing some other mechanism besides reranking. Each of these strategies suffers from the same problem: unmotivated formalism must be added. From this demonstration we arrive at our second conclusion, that the cophonologies hypothesis fails to account for certain exceptions to MSCs in Tiv. This adds to the arguments against co-phonologies advanced in Kisseberth (1993), Kirchner (1993), Itô and Mester (1993, 1995). 4.3 Prespecification Hypothesis Under the prespecification hypothesis (Inkelas, Orgun, & Zoll 1997), constraints account for alternations but do not account for “static” patterns: these are treated formally as coincidental. Only alternations are expressed through the constraint hierarchy. Where no alternation exists, exceptions and static subregularities are accounted for by prespecification (i.e. including the relevant phonological structure


in the lexical representation). In Tiv, the distribution of vowels in verbs might be considered a “static” pattern and so encoded directly in the lexical representation, i.e. by prespecification. There are some conceptual disadvantages with this approach. First, the prespecification hypothesis denies the significance of general patterns in accounting for exceptions, thereby predicting a much wider range of idiosyncratic variation in morphemes than languages exhibit. Second, the prespecification hypothesis allows co-phonologies when motivated by alternations, as defined in the Alternation Criterion, yet we have already seen that co-phonologies are inadequate. (16) Alternation Criterion: Co-phonologies exist only if they induce alternations. Finally, the prespecification hypothesis makes no predictions about the types of variation to be expected in lexical items, yet variation among the exceptions can be extremely limited, as in the case of Tiv. Tiv also provides empirical arguments against this hypothesis. For example, there is a morphologically induced Ø/V alternation in Tiv that shows the same patterns as already seen with monomorphemic CVCV verbs. The additional vowel found in the continuous noun form follows exactly the same pattern of vowel distribution already noted. (17) Continuous noun formation adds [−Vn] to C-final roots General pattern

verb root

continuous noun

Gloss

i…i e…e a…e f…o o…o u…u

mìr yèm yáv k´fr kór búgh

mírin yémen yáven k´fron kóron búghun

‘brew’ ‘go away’ ‘sleep’ ‘seize’ ‘sew’ ‘open’

Under the prespecification hypothesis, this alternation licenses the constraints introduced above in Section 2: Faithfulness to [−high], [−round] and [−low] are subordinate to all constraints governing the distribution of [+high], [+round], and [+low]. But the effect of the low-ranking of half of the faithfulness constraints is to eliminate the prespecification analysis of an exceptional, non-alternating form like [ímbise].


(18) [ímbise] ‘to jam into’ (under Prespecification) ímbise a. [ímbese]

b. [ímbisi]

(

) c. [ímbise]

AlHiL

AlHiR

VisE

**!

i iii

*!

ii

This exceptional form cannot be rescued by a co-phonology either: By hypothesis, co-phonologies are motivated only by alternations yet this form exhibits no alternations. If this particular aspect of prespecification were abandoned, then the model would predict no limitations at all on exceptional forms: Every lexical item could in principle have its own constraint hierarchy. 4.4 Summary In this section, we explored three models of accounting for lexical exceptions, Direct OT, Co-phonologies, and Prespecification. DOT accounts easily for the exceptional patterns considered thus far while co-phonologies and prespecification make false predictions. Both predict the non-existence of the attested class of exceptions. We turn now to further exceptions to the general Tiv verb patterns, to see that the occuring exceptions conform to patterns readily expressed under DOT, and are completely unexpected under co-phonologies and/or prespecification.

5.

Attested and unattested exceptions

In this section, we examine exceptional forms and pair them with comparable, but unattested exceptions. The contrast between existing and unattested exceptions conforms to the predictions of DOT, but is accidental under the other two models. 5.1 More on *AlHiR In Sections 3 and 4, we examined [i…(i)…e] forms, accounting for them under DOT through the necessary violation of AlHiR. In this section, we look at comparable cases, in which the feature [+low] and/or the feature [+round] are present. The prediction for inputs with [+high] and [+round] and a necessary violation of AlHiR is the surface form [u…(u)…e], not [u…(o)…o]. Because the rightward alignment of [+round] has both a narrow version (AlRdHi) and the (lower ranked) general version (AlRdR), in those forms where [+high] fails to spread rightwards, so too does [+round].


(19) [sùre] ‘to farm’ s`vrv [+hi],[+rd]; *AlHiR

AlHiR

a. [sùro]

*L

b. [sùre]

*L

c. [sùru]

*I!

AlRdL

AlRdHi

AlRdR

*! *

The unattested [u…o] form would require suspension of both AlHiR and AlRdHi. Although it is in principle possible for a lexical representation to crucially encode violations of two constraints, the absence of such forms suggests that such encoding comes at cost under DOT. This result contrasts with the predictions of prespecification and co-phonologies, each of which finds the complete absence of *[u…o] forms to be coincidental: their cost is no more significant than the cost of the attested [u…e] forms. We turn now to cases with both [+high] and [+low] in the input representation. The fact that [+high] surfaces on the initial vowel and [+low] on the final vowel suggests that AlHiL and AlHiR both outrank AlLoL. Under this view, for [+low] to surface at all, AlHiR must be violated, exactly as seen with the [i…e] and [u…e] cases. (20) Forms with [+high], [+low] and *AlHiR [+high, +low] yína “be less than” ndíal “lick” [+high, +low, +round] búna “to boil grain for beer” mùra “to swell into a lump” As seen in (21a), the candidate with the standard [u…u] pattern fails because it has no necessary violation of AlHiR. Where there is the necessary violation of AlHiR, [+round] is limited to the initial high vowel, as in (21c), due to AlRdHi.10 (21) [búna] ‘to boil grain for beer’ b´vnv [+hi],[+lo],[+rd]; *AlHiR

AlHiL

AlHiR

a. [búnu]

*I

b. [búnf]

*L

c. [búna]

*L

AlRdHi

AlRdR

*! *


Once again, co-phonologies and prespecification make no predictions about limits on the types of exceptions that can occur. 5.2 *AlHiL Under DOT, other constraints may be necessarily violated. Tiv exhibits examples of a required violation of AlHiL. We anticipate forms with a low or mid initial vowel and a high final vowel. These are, in fact, attested.11 (22) Forms with *AlHiL a. sóghur “to shake thing to cause contents to fall out” tsóghul “to grow up (of a person, animal)” b. ny`fghul “to turn shoulders aside; to shake from side to side” s´fngur “to gulp with grief” Requiring a violation of AlHiL means that the [+high] feature will not be linked at the left edge, leaving this position available for [+low] if present (compare (24) with [+low] and (23), which has no [+low]). Since neither [round] constraint is necessarily violated, [+round] will align at the left edge; its right edge is delimited by AlRdHi and/or AlRdR (both constraints have the same effect when the final vowel is [+high]). In both tableaux, candidate (a) fails because of the unnecessary violation of AlRdHi while candidate (c) fails to include the necessary violation of AlHiL. (23) [sóghur] ‘to shake thing to cause contents to fall out’ s´vghvr [+hi],[+rd]; *AlHiL

AlHiL

a. [sóghir]

*L

b. [sóghur]

*L

c. [súghur]

*I!

AlRdL

AlRdHi

AlRdR

*!

*

AlRdHi

AlRdR

*!

*

(24) [s´fngur] ‘to gulp with grief ’ s´vngvr [+hi],[+lo],[+rd]; *AlHiL

AlHiL

a. [s´fngir]

*L

b. [s´fngur]

*L

c. [súngur]

*I!

AlRdL

In both cases, a nonhigh-high sequence results. Under prespecification and under co-phonologies, either the existing patterns are impossible, or we expect a lot more variation among the exceptions.


The other three types of exceptions, [a…o], [a…a], and [f…e], succumb to a comparable analysis. The first involves the necessary violation of AlRdL, the second of VisE, and the third of AlRdR. This is shown in the following abbreviated tableaux, where the only contender is the form expected given the critical features. (Note that in each of these cases, the closest contender is an attested form, ruled out by the inverse violation of the lexically identified constraint.) (25) Analysis of the remaining exceptional forms mgbághom ‘approach’ v…v [+lo],[+rd]; *AlRdL

AlRdL

AlRdHi

AlRdR

VisE

a. f…o

*I!

fo

b. a…o

*L

ao

kpàma ‘annoy’ v…v [+lo]; *VisE

AlRdL

AlRdHi

AlRd

VisE

c. a…e

aeI!

d. a…a

aaL

k´fse ‘abrade lightly with the fingernail and gently squeeze the pus out’ v…v [+lo],[+rd]; *AlRdR

AlRdL

AlRdHi

AlRdR

VisE

e. f…o

*I!

fo

f. f…e

*L

f

5.3 Comparison: DOT vs. prespecification for Tiv As shown in this section, there are strict limits on exceptional verb forms in Tiv. The more robust exceptional patterns have been considered here. However, even when the more unusual exceptions are considered, they all fall within the range of exceptions predicted by DOT. This is contrary to the results with co-phonologies and with prespecification, which make no predictions whatsoever.


(26) Comparison of DOT and other models of lexical exceptions Tiv verb patterns

DOT

evidence in tableau…

u…e, not *u…o

predicted *AlHiR

(19)

u…a, not *u…f

predicted *AlHiR

(21)

o…u, not *o…i

predicted *AlHiL

(23)

f…u, not *f…i

predicted *AlHiL

(24)

Prespecification & co-phonologies

accidental that one type of exception occurs but not the other: both types are expected

6. Conclusion Sections 2 and 3 demonstrated that the limited vowel distribution in verbs in Tiv is characterizable under OT as the high-ranking of faithfulness for [+high], [+round], and [+low], coupled with alignment constraints restricting their distribution. These constraints outrank the markedness constraints preferring the features [−high], [−low], and [−round], subsumed under VisE, as well as faithfulness constraints to these feature specifications. Lexical exceptions to these general patterns were explored in the remainder of this work, comparing three hypotheses about how to represent exceptions: Direct OT, Co-phonologies, and Prespecificaton. While DOT successfully accounts for the types of both occurring and non-occurring exceptions, co-phonologies and prespecification predict either no exceptions or completely free vowel patterns. Cophonologies predict non-occurring forms and fail to allow the occurring forms while prespecification fails to make predictions. Although not all exceptions in Tiv have been explored here, were we to examine further forms we would find that not every posited constraint has a corresponding exceptional form. If this is the correct analysis of Tiv, then the question arises of whether the even more limited distribution of exceptions is accidental or whether there is a yet to be determined component of the DOT hypothesis, identifying the set of constraints to which there can be exceptions. An alternative approach, implied by Chris Golston, p.c., is that this mismatch is an indication that the analysis is incorrect, and that we can appeal to the existence of exceptions to a constraint as evidence that we have identified a necessary constraint, in short, that the exception proves the rule. This remains as an intriguing area for further research.


Nonetheless, Direct OT does not suffer from the flaws of unconstrained cophonologies, for a variety of reasons. First, the distribution of exceptions is principled under DOT: Exceptions are possible only if the necessary violation would not otherwise occur, and exceptions can only be encoded in terms of independently necessary constraints. Further, only quirky violations of constraints are represented lexically, not those due to the interaction of constraint hierarchy and input, thereby limiting the types of exceptions we can expect. Finally, under DOT, as under OT, constraints are universal, again limiting the types of exceptions predicted by the model. DOT makes predictions where prespecification fails to because under DOT, exceptions are represented as lexically specified violations of constraints which would otherwise fail to be violated by the form. Thus, exceptions are tied directly to the constraints that are critical (high-ranked) in the language. Under prespecification, exceptions are represented by specifications whose survival in the output is due to faithfulness alone. Faith must outrank other constraints to ensure the survival of these specifications, so anything is possible. In short, DOT allows the representation of MSCs and exceptions to MSCs, while the Co-phonology Hypothesis fails to express exceptions and the Prespecification Hypothesis fails to express the generalizations. At the same time, DOT does not lead to the unconstrained proliferation of spurious generalizations.

Notes * I dedicate this work to Eloise Jelinek, who has served as role model to me both in how to conduct research and how to conduct oneself professionally throughout my years at the University of Arizona. Thanks to the audiences at SWOT IV, Tucson 1998 and at the Linguistic Society of America annual meeting in 1999 for useful discussion of some of the ideas contained here. Thanks especially to Donca Steriade for pointing out the significance of Inkelas, Orgun, and Zoll (1997), to Chris Golston for very helpful comments, to the anonymous reviewer, and to the patient and helpful editors of this volume. 1. Earlier versions of this project appeared as Archangeli (1998) and (1999). 2. The full picture of bi-vocalic verbs in Tiv must distinguish two classes, (C)V(h)V verbs and (C)VC2V verbs, where C2 is any consonant other than [h], the sole laryngeal consonant of Tiv. In the (C)V(h)V class, the overwhelming tendency is for the two vowels to be identical, as in s´ff ‘to sting, to prick’ and s`fhf ‘to blow up the fire’. See the references cited for the full picture. For the remainder of this work, I use (C)VCV to refer only to the class of verbs with a nonlaryngeal intervocalic consonant. Digraph consonants, such as gh and nd are single segments. 3. There are two additional notes. First, along with (C)V(V)hV forms, monosyllabic verbs ((C)V(V)(C)) and trisyllabic verbs ((C)VCVCV) are omitted in this table. The account offered here extends to both shorter and longer forms. Second, what is treated here as an [i…a] pattern surfaces in most contexts as an [e…a] pattern. I do not provide an explanation for this fact here. 4. Max and Dep are given their standard definition: MaxHi, for example, stands for “Every input [+high] has a corresponding output [+high]” while DepRd stands for “Every output [+round] has


a corresponding input [+round]”. Align(X,L;Y,L) stands for “Align the left edge of every X to the left edge of some Y”. 5. I assume that the same constraint(s) that will account for VhV cases (recall from footnote 2 that the two Vs are identical) will also account for the identity of VV in a bimoraic syllable. 6. The [i…a] and [u…a] forms are addressed in Section 5.1. 7. Golston (1996) argues that all properties of representations should be encoded as exceptions to constraints. I have used more familiar input representations here, for example an input feature [+high] as opposed to a lexically necessary violation of a constraint prohibiting the feature [+high]. The cover constraint VisE is one of the types of constraints necessary in a full-fledged DOT analysis, where the presence of a feature in a surface form is indicated by a lexically necessary violation of a constraint on feature quality. 8. “Gradient” evaluation of constraint C means that n+1 violations of C are worse than n violations of C. 9. Golston (1996) uses the symbols R and A for necessary violations of a Root constraint and an Affix constraint respectively. ·RÒ and ·AÒ are used for inverse violations of Root and Affix constraints. The Tiv data do not require this distinction, hence the different formalism. 10. As pointed out to me by Heidi Harley, forms without [+round] (like yína ‘be less than’) also show the interaction of FaithLo » VisE. 11. These forms all have a [+round] feature too; I have no explanation for the absence of [e…i] and [a…i] forms. Recall that “gh” is a digraph for a single segment, as is “ts” and “ng”.

Rapid perceptibility as a factor underlying universals of vowel inventories* Natasha Warner University of Arizona and Max Planck Institute for Psycholinguistics

1.

Introduction

Although the vowel inventories of the languages of the world differ widely, some features distinguish vowel phonemes in many languages, while others are used by relatively few languages. For example, a very large proportion of the languages of the world distinguish at least three heights of vowels (using the features [high] and [low]), and also distinguish front from back vowels (Liljencrants and Lindblom 1972, Ladefoged and Maddieson 1996), as in the universally most common vowel system, /i, e, a, o, u/. However, much smaller numbers of languages make distinctive use of features such as nasalization, non-modal phonation types (e.g., creaky or breathy voice), advanced tongue root, pharyngealization, or length. This study tests a possible explanation for why some contrasts are more common than others. In so doing, it addresses the relationship between formal (in this case featural) and functional (in this case perceptual) aspects of vowel inventories. Stevens has argued many times (Stevens 1971, 1980, 1985, Stevens and Blumstein 1981) for the importance of brief regions around points of abrupt acoustic change for perception of speech (cf. also Furui (1986) and Liu (1996)). Stevens and Keyser (1989) and Stevens (1980) suggest that the features which are most often used distinctively in the world’s languages are the ones having cues located at these points of abrupt acoustic change. They identify the features [continuant], [sonorant], and [coronal] as the primary distinctive features based partly on this acoustic criterion. Lang and Ohala (1996) take this idea further, hypothesizing that for vowels as well as consonants, the cross-linguistically most common contrasts are those which listeners can distinguish based on a very brief portion of the signal (such as 40 ms or less). For example, all or nearly all languages distinguish stops from nasals, thus using the feature [sonorant], which Stevens argues depends on an abrupt change in amplitude. Cues such as the stop burst, or lack thereof, should be rapidly perceptible. However, relatively few languages make distinctive use of secondary articulations, such as palatalization, pharyngealization, or uvularization. These are expected to require a longer portion of the signal to distinguish.

246 Natasha Warner

Lang and Ohala (1996) tested this hypothesis for the vowels of North American English by collecting confusion data for gated vowels (vowels of which only the initial portion is presented). They performed hierarchical clustering analyses to determine what contrasts were salient when listeners were allowed to hear various portions of the vowel. (Both the experimental method and the statistical analysis will be explained in detail below, as the same methods are used in the current study.) They found that the vowels were grouped perceptually into clusters around the five universally most common vowels /i, e, a, o, u/ when listeners heard only the first 50 ms of the vowels. Tekieli and Cullinan (1979) also found that vowel height and frontness/backness could be perceived relatively well early in the vowel (within the first 10 ms), but that the tense/lax distinction required a longer portion of the signal to perceive. However, in both of these studies, the main cross-linguistically less common distinctive feature is tenseness. It is important to investigate additional distinctions. In the current study, I use the Dutch vowel system to test the hypothesis that universally common distinctions can be perceived rapidly. Dutch is particularly appropriate because its vowel system includes several less common distinctions beyond those which have been tested before (Lang and Ohala 1996). Standard Dutch includes several front rounded vowels, and can be analyzed as having a fourway height distinction. It also has a length or tenseness distinction similar to that in English (that is, with considerable difference in vowel quality as well as a durational difference). This paper presents a confusion study for gated versions of all the Dutch vowels. Hierarchical clustering analysis and multidimensional scaling analysis are used to test the hypothesis. Thus, this paper uses experimental methods (a perception study) and statistical analyses to test a proposed functional explanation for a formal construct. Distinctive features, a formal phenomenon, are used differently by various languages. The hypothesis that the most commonly used ones are those which offer listeners perceptual cues early in the signal is the functional explanation to be tested.

2.

Methods

2.1 Materials One male native speaker of Standard Dutch produced several tokens each of the 15 non-reduced vowels and diphthongs of Standard Dutch in an /hVt/ environment, which has relatively little influence on the quality of the beginning of the vowel. Recording was done on a DAT tape in a sound-treated recording booth. The Dutch vowel inventory appears in Table 1. The analysis given here is that of Booij (1995), but with the vowels which only occur in recent loanwords omitted. The transcription

Rapid perceptibility as a factor underlying universals of vowel inventories 247

is that of Gussenhoven (1992). The high vowels are transcribed without length marks because they are of approximately the same duration as the short vowels in most environments, although they pattern phonologically with the long vowels. There is also a reduced vowel, [6], which occurs only without stress. There is some question as to which short vowels are counterparts of which long vowels, but on grounds of phonetic locations in the vowel space, duration, phonological alternations, and historical origins, the long high vowels are generally assumed to be without short counterparts. /I, e˜/ are considered upper-mid, and /7/ lower-mid. Since /a˜, "/ are low vowels, this means that Dutch may be analyzed as having four distinctive vowel heights, although it does not have four vowels distinguished solely by height. The transcription of vowels which form tense/lax or long/short pairs reflects the fact that both duration and quality differ in these vowels, as in English. Table 1.The Dutch vowel system.

high upper mid lower mid low diphthongs

front unrounded

front rounded

back

i I, e˜ 7

y y, ø˜

7i

œy

u o˜ f ", a˜ %u

The vowel system of the speaker for the experiment, as measured from the tokens chosen for use as stimuli, is shown in Figure 1. The first, second, and third formants of each item were measured at three points in time, one-quarter, half, and three-quarters through the duration of the vowel. (Only the first and second formants are shown in the figure.) Several important points are evident from this figure. First, the long mid vowels are phonetically quite diphthongal (as is shown by the length of the lines connecting the measurement time points), although they pattern phonologically with monophthongs. Second, this figure demonstrates one of the phonetic reasons for considering /I, y/ to be upper-mid, and not high, vowels: they have approximately the same height in the vowel space as the beginning points of the long mid vowels. Third, the front rounded vowels are relatively far back in the vowel space, appearing more central than front. Finally, most of the change in the diphthongs /7i, œy, %u/ is late in the diphthong: they have very little change between the first two measurement points, but a rapid change between the measurement points at half and threequarters of the duration of the diphthong. These observations have been noted in the literature on Dutch vowels before (Pols, Tromp and Plomp 1973, Pols 1977,

248 Natasha Warner

F2 (Hz) 2100

1900

1700

1500

1300

1100

900

[y]

[i]

[y]

[6]

[ø˜]

500

200

300

[u]

[e˜] [I]

700

400

[o˜] 500 [7]

F1(Hz)

2300

[%u]

[œy]

600 ["] [7i]

[a˜]

700

800

Figure 1.The Dutch vowels, measured from the stimuli. Smallest plot symbol: location in the formant space at one quarter of the vowel’s duration. Middle symbol: half the duration. Largest symbol: three quarters of the duration.

1979, Koopmans-van Beinum 1980), but they are important to interpretation of the results of the current experiment. 2.2 Stimuli One token of each /hVt/ item with minimal overlap of the frication and the vowel was chosen for use in the experiment. Each item was final-gated at 9 points, located at 20 ms intervals: the shortest gate included only the /h/ (up to onset of voicing), the next allowed the listener to hear from the beginning of the syllable up to 20 ms after the onset of voicing, the next up to 40 ms after onset of voicing, and so on. The longest gate ended 160 ms after onset of voicing (which was the end of the shortest vowels). All stimuli were gated to white noise, with a 5 ms transition period during which the signal was ramped down and the noise ramped up. The noise then continued, producing a total stimulus duration of 455 ms. Three additional stimuli were also generated from each /hVt/ syllable for use in a training test. These were gated in the same way as the test items, but with gate end points at 30 ms before the end of the vowel, at the end of the vowel, and after the /t/ burst. These stimuli had a total duration of 750 ms. 2.3 Subjects and testing 58 native speakers of Dutch participated in the experiment. Subjects were tested individually in a sound treated booth, and heard over headphones first the 45


training stimuli (described above), followed by the 135 test stimuli. Each subject heard the test stimuli in a different random order (not blocked by gate). Each stimulus was presented only once to each subject. Subjects identified the vowels by using a mouse to click on one of 15 buttons on a computer screen. The buttons were labeled with the orthographic representation of the Dutch vowels, and also had a real word illustrating the vowel next to them on the screen. The data from 9 subjects was excluded because they either identified less than 90% of all training stimuli correctly, or they identified all three training stimulus gates of a particular vowel incorrectly. Since the training stimuli allowed listeners to hear (nearly) the entire vowel, or even the entire CVC syllable, consistent failure to identify training stimuli correctly suggests a difference in the vowel system attributable to dialect. A confusion matrix (frequency of each type of response to each type of stimulus) was tabulated from the responses of the remaining 49 listeners, and was used as the input for the analyses described below.

3.

Results

3.1 Hierarchical clustering analysis A hierarchical clustering analysis groups stimuli into clusters based on their similarity. The results are presented in terms of the abstract distance at which two stimuli or clusters of stimuli combine to form a (larger) cluster. The analysis begins with each vowel as a separate unit (no clusters), and then groups the most similar vowels together into a cluster, and then the next most similar together, and so on, until all of the vowels form one large cluster. What is of interest is which vowels group together to form sub-clusters. The order in which various items combine into clusters is shown in a dendogram. For a given gate point, if the vowels which are distinguished by some phonological feature combine into a cluster at a small distance, that indicates that those vowels are rather similar to each other, and thus that the phonological feature distinguishing them is not very perceptible at that gate. The hypothesis that less common distinctive features require a relatively long portion of the signal to be perceived predicts that vowels which are distinguished by an uncommon feature will form a cluster at early gates, because at those gates they will be perceived as similar. Particularly, there should be clusters around the areas of the most commonly distinguished vowels, perhaps around the Dutch vowels /i, e˜, a˜, o˜, u/. The long and short vowel pairs should cluster together, and the front rounded vowels should be included in nearby clusters (for example, with /y/ in the same cluster as either /i/ or /u/).

250 Natasha Warner

3.1.1 The four large clusters A hierarchical clustering analysis was carried out for each gate point.1 The results of this analysis for the gate ending 40 ms after the onset of voicing are shown in Figure 2. At this gate, the vowels fall into four large clusters, consisting of the low vowels (plus the back diphthong) /a˜, ", %u/, the back non-low vowels /o˜, f, u/, the front rounded vowels /y, y, ø˜/, and the front unrounded vowels (plus the front rounded diphthong) /i, I, e˜, 7, 7i, œy/.2 The grouping into the four large clusters low, back non-low, front rounded, and front unrounded is typical of the hierarchical clustering results for most gates, although it is less clear for the very late gates, where there are few confusions. The very first gate, which allowed listeners to hear only the /h/ of the /hVt/ syllable, also shows a somewhat different pattern, as shown in Figure 3. However, even at this gate, the low, non-low back, and front rounded clusters appear. (Although the fricative /h/ can contain much information about an upcoming vowel, gating the stimuli to noise made it difficult to use the information in the frication noise.) 0

5

10

15

20

25

y ø˜ y i i 7 7i e˜ œy " %u a˜ f o˜ u

Figure 2.Dendogram showing combination of clusters for the 40 ms gate. The horizontal dimension shows the normalized distance at which each pair of stimuli or clusters combine.

The fact that the front rounded diphthong /œy/ falls into the front unrounded cluster rather than the front rounded one may seem anomalous. This is true of all gates except the 0 ms, 20 ms, and 160 ms gates. (At the 20 ms gate, /œy/ clusters with the low vowels. At the 0 ms and 160 ms gates, it does form a cluster with the front rounded vowels, but at great distance, and with some front unrounded vowels

Rapid perceptibility as a factor underlying universals of vowel inventories

0

5

10

15

20

25

œy %u 7 7i y ø˜ y " a˜ f o˜ u i e˜ i


included in the same cluster.) The grouping of /œy/ with the front unrounded vowels reflects its frequent misperception as /7/. Although /œy/ is rounded, rounding has less acoustic effect on low vowels, and its beginning point is rather low. Furthermore, there is no low or lower-mid front rounded monophthong in the Dutch system for /œy/ to be misperceived as. 3.1.2 Results for vowel height At the 40 ms gate (Figure 2), within each of the large non-low clusters, there are smaller clusters separating the mid vowels from the high vowels. Thus, the non-low back cluster contains smaller clusters, one consisting of /o˜, f/ and the other only of /u/. The front rounded cluster has sub-clusters of /y/ and /y, ø˜/, and the front unrounded cluster contains one smaller cluster of /i, I/ and another of /7, e˜, 7i/ (along with /œy/). This arrangement of the mid and high vowels within the larger clusters is apparent at each gate from 20 ms to 80 ms (after which the long high vowels are too rarely misperceived to form such clusters). These smaller clusters of mid and high vowels within the larger clusters mean that each high vowel is more similar to its corresponding mid vowel than it is to the other high vowels in the system. Each mid vowel is also more similar to its corresponding high vowel than it is to other mid vowels. That is, the front/back and rounded/unrounded distinctions are clearer from the first 20–80 ms of the vowel than the high/mid distinction is. This is not the grouping predicted by the hypothesis: the front rounded vowels

251

252 Natasha Warner

0

5

10

15

20

25

" %u a˜ f o˜ u y ø˜ y I e˜ 7 7i œy i


form a cluster by themselves, and have little perceptual similarity with either the front unrounded or the back vowels. The 60 ms gate also has smaller high and mid clusters within the larger non-low clusters, but a slightly different pattern appears within the front unrounded cluster, as shown in Figure 4. There are three subclusters within the front unrounded cluster, one consisting only of the vowel /i/, one containing /e˜, I/, and another containing /7, 7i, œy/. Thus, instead of clusters of high and mid vowels, there are clusters of high, upper-mid, and lower-mid vowels. The 80 ms gate shows the same pattern, and at later gates, /e˜, I/ and /7, 7i/ continue to form upper-mid and lowermid clusters, although /i/ is rarely misperceived. The change from just high and mid clusters at the 40 ms gate to high, upper-mid, and lower-mid clusters at later gates shows that listeners are only able to distinguish high from mid (and low) vowels based on the information in the first 40 ms of the vowel, but they are able to distinguish upper-mid from lower-mid vowels based on the information in the first 60 ms. Thus, as predicted by the hypothesis, listeners require more of the signal to distinguish four vowel heights than three. 3.1.3 Results for vowel length In the dendogram representation of a hierarchical clustering analysis, the distances at which stimuli combine are normalized, so one cannot compare them across


different gates. One can only compare the composition of the clusters, not the distances at which they form, across gates. However the non-normalized distance coefficients, which represent the average dissimilarity of all pairs of stimuli in the clusters being combined, can be compared across gates to determine whether a particular set of vowels becomes more or less similar over time. This provides a useful way to examine the role of the long/short vowel distinction. The long/short vowel pairs form clusters at nearly all gates (although /I/ varies as to whether it forms a pair with /i/ or /e˜/), so the composition of the clusters alone does not tell us how listeners’ use of this distinction changes over time. The non-normalized distance coefficients for the formation of long/short vowel clusters at the 40 ms, 100 ms, and 140 ms gates appear in Table 2. These coefficients show that for all four long/short vowel pairs, the distance at which they join the cluster increases from earlier to later gates. Thus, although the vowels in long/short pairs are very similar to each other, the perceptual distance between them does increase as listeners hear more of the vowel. The distinction between the long and short vowels becomes clearer as more of the vowel is heard. Table 2.Non-normalized distance coefficients for clustering of long and short vowels at several gates Vowel pair or cluster 40 ms gate

100 ms gate

140 ms gate

y – ø˜ f – o˜ ", %u – a˜ e˜ – I

3.28 7.05 7.97 3.38

6.71 8.91 9.70 (160 ms gate)3 7.76

2.63 5.27 6.07 not a direct cluster

3.2 Multidimensional scaling analysis A hierarchical clustering analysis shows which vowels are perceptually most similar, but it does not show explicitly what acoustic characteristics listeners use in distinguishing the vowels. A multidimensional scaling analysis4 was performed on the same confusion matrix data to address this question. This analysis incorporates the confusion data from several gates simultaneously, to determine what perceptual dimensions account for listeners’ confusions overall. However, the analysis becomes inaccurate if confusion matrices with extremely few confusions are included, so only the first six gates were used. (See van der Kamp and Pols 1971 and Wright 1986 for other examples of the application of this method to the perception of vowels.) A multidimensional scaling analysis uses similarity data, in this case based on confusions, to place the items (vowels) in a multidimensional space which represents

254 Natasha Warner

the perceptual dimensions listeners use in perceiving the vowels. One can then examine the dimensions the analysis produces to determine what acoustic characteristics they correspond to. The analysis produces solutions using various numbers of dimensions. Solutions using more dimensions will always fit the data better, because they allow more ways for variation in each vowel to be accounted for. However, it is unlikely that human listeners actively classify vowels on, say, six or more dimensions. If the solution with three dimensions provides nearly as good a fit to the data as the solutions with four or more dimensions, one can conclude that the solution with three dimensions is at least a sufficient representation of the perceptual space used by human listeners, and one can then turn to determining what acoustic characteristics those three dimensions represent. In this case, the three dimensional solution provides a good fit to the data (Table 3). There are relatively gradual decreases in the average R2 (goodness of fit of the model) between the six dimensional and the three dimensional solutions, but a large decrease between the three dimensional and the two dimensional solutions, indicating that a two-dimensional solution does not model listeners’ perception well. Table 3.Average R2 for the multidimensional scaling analysis using the first six gates, for solutions with various numbers of dimensions Number of dimensions 6 5 4 3 2

Average R2 across gates .89 .86 .79 .74 .56

Figures 5 and 6 show the stimulus locations as calculated by the multidimensional scaling analysis for the three dimensional solution. Figure 5 shows the locations of the vowels in the first and second dimensions, and Figure 6 shows the first and third dimensions. In Figure 5, the vowels seem to be arranged approximately in the shape of the vowel space, with /i, I/ having high values for both dimensions, /u/ having a low value for dimension one but a high value for dimension two, and /a˜, ", %u/ having somewhat low values for dimension one and very low values for dimension two. Thus, from a visual inspection of the figure, dimension one seems to reflect frontness/backness of vowels, and dimension two to reflect vowel height. However, the front rounded vowels are lower in dimension two than one would expect. The correlation of the values on dimension one with the second formant measured at one quarter through the vowel is r = .86 (p < .001). The correlation of the second dimension with the first formant measured at one quarter


2.0 i 1.5

u f

Dimension 2

1.0 .5

e˜ o˜

0.0

y

7 ø˜

y

-.5

a˜

-1.5

-1.0

7i

œy

"

-1.0 -1.5 -2.0

i

%u

-.5

0.0

.5

1.0

1.5

Dimension 1

Figure 5.Dimensions 1 and 2 of the stimulus configuration derived by the multidimensional scaling analysis.

through the vowel is r = −.79 (p < .001).5 As Figure 6 shows, the third dimension primarily seems to separate the front rounded vowels (except the front rounded diphthong) from all the other vowels. This dimension might reflect rounding, although the low values for /u, o˜, %u/ are problematic. The correlation of values on this dimension with the third formant is r = −.76 (p < .002). Multidimensional scaling analyses were also carried out using the data from the first 4 or 5 gates instead of the first 6, as the number of confusions at the sixth gate might already be too small to be useful. Results were similar to the results with six gates. Since the four-dimensional solution in the analysis with six gates provides a somewhat better fit to the data than the three-dimensional solution, and it is reasonable to suppose that listeners might use four dimensions in perceiving vowels, that solution was also examined, but the dimensions were not readily interpretable (i.e., they did not map onto any known characteristics of vowels). Therefore, the three-dimensional solution to the analysis with data from 6 gates, above, will be adopted. Van der Kamp and Pols (1971) and Pols et al. (1969) have also applied multidimensional scaling analysis to perceptual data on Dutch vowels. They used either a single period excised from vowels or resynthesized completely steady state vowels, rather than natural gated vowels, as stimuli. Both found that a three dimensional

256 Natasha Warner

2.0

y

y

1.5

ø˜

Dimension 3

1.0 .5

ɔ

0.0 -.5 -1.0 -2.0

U

œy

%u

o˜ " -1.5

a˜ -1.0

I -.5

0.0

.5

i

7i 1.0

e˜

7 1.5

Dimension 1

Figure 6.Dimensions 1 and 3 of the stimulus configuration derived by the multidimensional scaling analysis.

solution was the best analysis for the data, as in the current study. Pols et al. (1969) also found that two dimensions were related to the first two formants, but that the third dimension was difficult to interpret, although in their data it did not appear to be related to rounding.

4. Discussion The hierarchical clustering analysis and the multidimensional scaling analysis show good agreement as to what distinctive features are perceptually most important for the Dutch vowels. The hierarchical clustering analysis has major divisions into back, front rounded, and front unrounded vowels, as well as low and non-low vowels, with subordinate divisions into high and mid vowels. The back — front rounded — front unrounded division corresponds to the multidimensional scaling analysis frontness/backness dimension. The division into low and non-low vowels, and subordinately into high and mid, corresponds to the height dimension of the multidimensional scaling analysis. The fact that the front rounded vowels form a high-level cluster by themselves and do not cluster together with other vowels mirrors the way that the third dimension of the multidimensional scaling analysis separates front rounded vowels from all others.


4.1 Front rounding: The feature for which the hypothesis is not supported The results regarding the perception of front rounded vowels contradict the hypothesis that cross-linguistically uncommon features require a longer portion of the signal to distinguish than common features do. This is particularly clear from the hierarchical clustering analyses at the early gates, in the fact that the front rounded vowels cluster together instead of forming clusters with their corresponding front unrounded vowels, for example. This leads to the question of why, if front rounding can be distinguished so reliably based even on the first 20 ms of a vowel, it is not used distinctively in more of the world’s languages. In particular, since front rounding is more clear than the difference between high and mid vowels from the early portions of the vowel, why is it that far more languages distinguish high from mid vowels than front rounded from front unrounded and back vowels? There are several possible explanations for this. First, the perception of front rounding may be more affected by consonantal environment than the high/mid distinction is. In this study, all vowels were recorded in the environment /hVt/ because Lang and Ohala (1996) found that responses at the crucial early gates were strongly influenced by consonantal context if /d/ or /g/ was used as the initial consonant instead. Some earlier studies excised vowels from a variety of contexts, though. Pols (1977) recorded the Dutch vowels in all possible CVC environments, excised the vowels from the syllables, and presented them to listeners in isolation for identification. His results (1977: 110) show a large number of confusions between front rounded vowels and both front unrounded and back vowels. In order to compare these results to those of the current study, I performed a hierarchical clustering analysis on the data from Pols’ (1977) study, and found that in his data, the front rounded vowels form clusters with front unrounded vowels instead of with each other. /7/ forms a cluster with the mid front rounded vowels, and /y/ forms a cluster with /i, e˜, I/ (Figure 7). Koopmans-van Beinum (1980) presented 12 Dutch vowels recorded in isolation, excised from isolated words, and excised from connected conversational speech to listeners for identification. The last condition, of course, produced many confusions, with more confusions of the front rounding distinction than the high/ mid distinction, except for the back high and mid vowels. Thus, it may be that when consonantal influence on the vowels varies, the high/mid distinction is more perceptible than front roundedness. Consonantal effects on the vowel may be more detrimental to the perception of front rounding than to the perception of height.6 A second possible explanation for the discrepancy between the ready perceptibility of front rounding in this experiment and its cross-linguistic low frequency lies in the acoustic cues for vowels. Vowels, at least when stressed, are inherently rather long segments. Crystal and House (1982, 1988) find that the average duration of English vowels is longer than the average duration of all types of consonants except

258 Natasha Warner

0

5

10

15

20

25

f o˜ " 7 y ø˜ I e˜ i y u 7i œy %u a˜

Figure 7.Dendogram for a hierarchical clustering analysis of the confusion data from Pols’ study (1977).

affricates. Since vowels are inherently rather long, they may be able to bear relatively slow cues. Some, or even many, of the ways to make perceptual distinctions in speech may depend on rapid changes in the signal (fast cues). However, there are also aspects of the signal which provide potential perceptual cues, but which do not involve rapid changes. Formant frequency changes and duration differences are two types of cues which probably do not involve perceptibility over a short time window. This account does not deny the importance of rapid perceptibility for some distinctions, but suggests that at least for vowels, both rapidly perceptible distinctions and others may be cross-linguistically common, because vowels are long enough to realize both slower and faster cues. Finally, an examination of cross-linguistic tendencies among languages with more than the canonical five vowels may provide a partial explanation. In a fivevowel system, it is clear why front rounding is not used distinctively. As shown in Figure 1, front rounded vowels are far from the periphery of the vowel system. If a five-vowel system included one or more front rounded vowels, the vowels would not be positioned maximally far from each other in the vowel space. As much work by Lindblom has shown, such a system would be unlikely (Liljencrants and Lindblom 1972, Lindblom 1986). However, among languages with vowel systems of seven vowels or larger, it is very common to have two non-peripheral vowels, whether they are front rounded or central. Lindblom (1986) and Liljencrants and Lindblom (1972) find that there are two predominant types of vowel systems for languages with seven vowels, shown in (1).


(1) i

u e

o 7

i e

f

G/y 6/ø a

u o

a Languages with seven-vowel systems usually either have an upper-mid/lower-mid distinction or have two central or front rounded vowels, and neither of these patterns clearly predominates. Thus, when separation of the vowels within the vowel space does not dictate otherwise, a front rounding distinction may not be cross-linguistically uncommon. However, this does not explain why front rounding is less common than the high/mid distinction, even though it is more readily perceptible even over a very short time window. 4.2 Features for which the hypothesis is supported Unlike front rounding, both the four-level height distinction and the long/short distinction support the hypothesis that less common features require a longer duration of the signal to perceive. The change in the hierarchical clustering analysis for the front unrounded vowels from the 40 ms gate to the 60 ms gate is particularly convincing: at 40 ms, there is a high vowel cluster /i, I/ and a mid vowel cluster /e˜, 7, 7i/ (also including /œy/), as well as a low vowel cluster consisting of /a˜, ", %u/. However, at 60 ms, there are high, upper-mid, lower-mid, and low vowel clusters. Thus, as the differences among /i, I, e˜, 7/ become clearer, listeners begin to distinguish four vowel heights instead of three. The universally very common three-level height distinction can be perceived based on cues in the very beginning of the vowel, but perception of the less common four-level height distinction requires a longer portion of the signal. Although there is some question as to which vowels form long/short pairs in Dutch, the results discussed in Section 3.3.4 above demonstrate that the length distinction is more perceptible at late gates than at early ones. This might seem to be an obvious consequence of truncating segments which are differentiated by duration, but in Dutch, as in English, there are substantial differences in vowel quality as well as in duration between long and short members of a pair. Some researchers consider this a tenseness distinction rather than a length distinction (Gussenhoven 1992, van Oostendorp 1995). Thus, there are potential cues to this distinction even early in the signal, but they are not strong enough to make this distinction readily perceptible at early points in the vowel. The results for long and short vowels in Dutch replicate Lang and Ohala’s (1996) results for English vowels: they also found that long/short (or tense/lax) vowel pairs tended to form clusters, and that these clusters were stronger at earlier gates. Although a length or tenseness distinction is not unusual cross-linguistically, it is certainly less common than the

260 Natasha Warner

features used in the system /i, e, a, o, u/. These results for Dutch confirm that it takes longer to perceive the less common length distinction than to perceive the more common features.

5.

Conclusions

The results support the hypothesis that some distinctions appear in many languages because they are readily perceptible from cues in a brief time window around an abrupt acoustic change. However, some distinctions which are perceptible based on a short time window may not be cross-linguistically common because of other influences on the phonological system, such as separation of vowels in the vowel space or a differential effect of surrounding context. Thus, the choice of distinctive features in languages’ inventories is related to rapid perceptibility, but this is not the only factor which contributes to universals of vowel systems. An interesting direction for future research might be the exploration of a yet wider variety of vocalic distinctive features, using several languages. The current paper extends past work on English, particularly on the tense/lax distinction, to a four-way height distinction and a front rounding distinction, and finds support for the hypothesis for some but not all features. Future work on timing of perception of distinctively nasalized vowels, vowels with creaky or breathy voice, vowels with advanced tongue root, as well as perhaps back unrounded vowels, would elucidate which types of cross-linguistically less common features require a long portion of the signal to perceive, and which, like front rounding, are immediately perceptible. Work on a wider variety of languages and distinctive features might also allow one to examine speed of perception of very uncommon vs. somewhat uncommon vs. very common distinctions. The current work divides vocalic distinctive features only into those which are used in a typical 5-vowel system and those which are less common than that. However, distinctive use of non-modal voicing for vowels is probably even less common than distinctively nasalized vowels, for example, and work on a wider variety of languages would allow one to determine whether these finer distinctions are also related to perceptual factors. The current work relates formal properties of the vocalic system, the phonological features which distinguish vowels, to a functional motivation, rapid perceptibility of acoustic cues. The results confirm that at least in the area of vowel inventories, formal and functional aspects of language are closely related.


Notes * I would like to thank John Ohala, Anne Cutler, Allard Jongman, Carlos Gussenhoven, Terry Nearey, and Roel Smits for discussion of this research and of statistical methods, as well as two anonymous reviewers for their helpful comments. I would also like to thank Niels Janssen for his effort in running the experiment, and John Nagengast for technical assistance. Errors are, of course, my own. 1. The average linkage between groups method was used, with a chi-squared dissimilarity measure (Norušis & SPSS, 1994). 2. The method of hierarchical clustering (Norušis & SPSS, 1994) involves first looking for the two items which are most similar to each other and combining them into a cluster. It then looks for the next most similar items, and either adds one item to the first cluster, or combines two items into a separate cluster. This process continues until all items have been combined into a single large cluster. Thus, the process necessarily involves all items being combined into one cluster comprised of rather dissimilar items, and also involves each item appearing by itself at the initial stage of the analysis. What is of interest is which items combine into clusters of intermediate size. Although all hierarchical clustering analyses will show each vowel starting out as separate, and all vowels being combined into one large cluster, the composition of the clusters in between these stages, and the order in which vowels join those intermediate clusters, can vary. 3. At the 40 ms and 100 ms gates, /a˜/ combines with /"/ only after it has already formed a cluster with /%u/. However, at the 140 ms gate, where there are very few confusions, /a˜/ forms a cluster more closely with /I, e˜/ than with /", %u/, and this cluster is therefore not representative of the distance at which the long and short vowel combine. Therefore, the value for the cluster /a˜, "/ at the 160 ms gate is given here. 4. The INDSCAL method, chi-squared dissimilarity measure (Norušis & SPSS, 1994) was used. 5. F1 may not be an exact reflection of what listeners perceive as vowel height, but it is generally accepted that F1 varies inversely with vowel height. Similarly, F2 alone may not be the exact perceptual cue for frontness, but it is clearly related to it. Rounding is more difficult to map directly to an acoustic characteristic, but rounding does tend to lower F3. See Johnson (1997) and Stevens (1998) for general discussion of the relationship between formant frequencies and vowel quality. 6. This effect of consonantal environment might seem to imply that perception of front roundedness would differ across languages depending on consonant inventory. However, the major effect of consonantal environment is probably determined by place of articulation, and since most languages have at least labial, coronal, and dorsal consonants, there may not be much cross-linguistic difference along these lines in degree of influence on perceptibility of front rounding.

Part III

Foundational issues

Argument hierarchies and the mapping principle* Eloise Jelinek‡ and Andrew Carnie University of Arizona

1.

Introduction

As is well known, in many languages — such as Lummi and Dyirbal — we find a split in case marking, where third person and other non-local arguments take an Ergative/Absolutive marking, and local persons (1st and 2nd persons) take a Nominative/Accusative pattern. Argument hierarchies are used by linguists to describe situations like these where syntactic phenomena interact with arguments of various types (along various dimensions) in distinct manners. The topic of argument hierarchies has, until recently, been mainly confined to the domain of typology and language description. In a series of influential articles in the Functional Optimality Theory framework (FOT), Judith Aissen, Joan Bresnan and their colleagues have attempted a formal account of argument hierarchies.1 These accounts fully “syntacticize” the phenomenon, in that these phenomena are treated as purely morphosyntactic notions (as encoded in the variable output constraint rankings). In this paper, we would like to claim that such an approach lacks an explanatory basis. We would like to claim instead that, following the work of Jelinek (1993), Meinunger (1999), and to a certain extent Isaak (2000), the effects of argument hierarchies emerge from a formally encoded correspondence between syntactic prominence and semantic/pragmatic prominence. Syntactic effects of semantic argument hierarchies are a direct result of the mapping between syntactic and semantic structure, and are not purely syntactic phenomena as FOT accounts would lead us to believe. We encode this in what Carnie (in prep) calls Jelinek’s Hypothesis: (1) Argument hierarchies are the direct morphosyntactic registration of some aspects of the presuppositionality scale. In proposing such an analysis, we provide a semantic grounding to a wide variety of syntactic phenomena, including ergative splits, object shift, differential object marking, dative/accusative alternations, clitic placement, voice alternations — all of

266 Eloise Jelinek and Andrew Carnie

which seem to be sensitive to presuppositionality. For reasons of space we don’t attempt a direct refutation of FOT accounts here. Instead we offer only positive evidence in favor of a mapping approach to argument hierarchies. To see the discussions of the merits and flaws of FOT, see the debate between Newmeyer and Bresnan and Aissen in the pages of Natural Language and Linguistic Theory (2002), the paper by Tom Bever in this volume and Carnie (in prep). 1.1 The mapping principle Argument hierarchies come in a number of different types cross-linguistically. We find hierarchies based on definiteness/animacy/person, etc. Despite their surface diversity, argument hierarchies are all alike in the following two attributes: a.

There is some ranking of arguments according to a scale of presuppositionality. More local, more specific, more definite, and more animate arguments are more likely to be presupposed by the speakers in discourse. Such elements outrank non-local, non-specific, less animate etc., elements. On the hierarchy, these elements are more likely to be asserted than presupposed.2 b. There is some syntactic manifestation of the hierarchy: such as case and/or voice alternations which place an argument in subject or object position, to order arguments in accord with presuppositionality. There is crosslinguistic variation in what particular features on the presuppositionality scale a language elects to grammaticize, just as some languages choose to grammaticize gender and others don’t. The crucial notion here, however, is that no matter which of the features determining presuppositionality a particular language uses, there is a predictable correspondence between the information structure status and the syntactic realization of the arguments. Diesing (1992) developed a deceptively simple proposal, where there is a direct mapping (encoded in a semantic mapping principle) between syntactic constituent structure (at some level of representation) and semantic structures.3 We take the mapping principle to be the primary means by which a grammar encodes a presuppositionality scale. The view advocated by Diesing assumes the Kamp (1981)/ Heim (1982) approach to the interpretation of nominals (see also Krifka 1991; Lambrecht 1994). The sentence (or proposition) is divided up into three parts: (a) a quantifier which asserts the number of entities participating in the action or state; (b) a restrictor, which asserts the presupposed information about the participants; this roughly correlates with the TP or IP portion of the clause; and (c) a nuclear scope, corresponding to the VP, which asserts what is true of the entities and provides the new information to the clause. A very simplistic example is given in (2): (2) (Quantifierx,y Every

[Restrictor (x, y)] [person

[Nuclear Scope (x, y)]) [loves cookies

Argument hierarchies and the mapping principle 267

These tripartite structures are derived directly from the syntax via the mapping principle (3) as schematized in (4). (3) The Mapping Principle (Diesing 1992) a. By LF, the material from IP [TP] and above maps into the restriction on some operator. b. The material from VP maps into the nuclear scope. TP

(4)

VP

restrictive clause nuclear scope

On a more formal level, only variables are allowed in the nuclear scope. Oversimplifying Diesing’s system greatly, variables can be of two sorts: (i) the traces of NPs that have moved out of the VP; (ii) a non-quantificational, non-presuppositional NP, which is bound by Existential Closure (i.e., will be as a default taken to mean “there is an X”.) In terms of the syntax, what this means is that (at LF) quantificational (presuppositional) NPs (such as specifics or definites) cannot be inside the VP (they have to move to create a variable), only non-presuppositional ones (like non-specific indefinites) can appear there. To see how this works, let us consider the well known facts from object scrambling in German (data from Diesing and Jelinek 1995:128, 130). (5) a.

… weil ich nicht eine einzige Katze gestreichelt habe … since I not a single cat petted have “Since I have not petted a single cat” b. … weil ich eine einzige Katze nicht gestreichelt habe … since I a single cat not petted have “Since I have not petted a single cat c. *?… weil ich nicht die Katze streichle … since I not the cat pet “Since I do not pet the cat” d. … weil ich die Katze nicht streichle … since I the cat not pet “… Since I do not pet the cat”

Let us assume that negation marks the left edge of VP, and VP is the domain of existential closure here. Definite objects and indefinites with a specific reading appear to the left of negation; non-specific indefinites appear to the right. This gives us the outward appearance of a specificity hierarchy: (6) Elements to the left of X on the following hierarchy must appear to the left of negation: Definite < strongly quantified < specific indefinite < X < non-specific


The VP can only contain unbound variables, so other NPs must scramble4 out of the nuclear scope (VP) to a higher position in the tree. The net result of this is that elements high on the relational hierarchy correspond to elements that are relatively high in the syntactic tree. We use the terms topic and focus informally5 as means of describing presuppositionality. We follow the writings of Partee, Heim, Diesing et al. in identifying topic/ focus structure as the information structure of the sentence. Old (presuppositional) information in the sentence is topical, familiar and backgrounded. New information in the sentence is unfamiliar and focused. Across languages, most of the time, there is a strong tie between subjects and topics, vs. objects and focus. But alas, this is not always the case; we see topicalized objects and focused subjects — marked by intonation and movement in English, and with Topic and Focus particles, movement, and affixes in other languages, forcing us to include Topic and Focus in our accounts of morphosyntax. In addition, we see thetic sentences with surface pleonastic or indefinite subjects; at LF these are existential sentences with all contentful (non-pleonastic) arguments in the nuclear scope. The particular version of the mapping principle we adopt is that proposed in Carnie (in prep). This version follows Percus (1993) in relativizing the mapping principle to the particular argument involved. More specifically, the domain of existential closure (nuclear scope) for argument Y is the VP or vP that introduces Y. This is constructed within a version of phasing theory. The particular view of phases that we will advocate here is different than that proposed by Chomsky (2001a). Chomsky proposes that phases are, in essence, propositional; they consist of a predicate and its arguments (vP), or a temporal and force operator (TP or CP). Carnie offers an alternate view of phase. In his approach, phases minimally consist of all of the following: (7) a. a predicative element (v or V) b. a single argument (NP) c. a temporal operator that locates the predicate and argument in time and space (Asp or T) For a simple transitive clause, then, the first phase of a sentence consists of a lexical predicate which expresses an event or state (V), any internal arguments, and the Asp head.


(8)

AspP Asp′ Asp

VP V

O

The link between aspect, aktionsarten and objects is well known (see among many others Tenny 1994), and is expressed in this phase. The second phase in a transitive clause contains the external argument introducing light predicate (v), the subject NP, and Tense: (9)

TP T′ T

vP S

v′ v

...

As should be obvious from combining the two trees in (8) and (9), this particular view requires the split VP or inner aspect approach of Travis (1994) and Koizumi (1995) (see also Harley 1995 and Carnie 1995). When it comes to the goal arguments of ditransitives a third phase exists, corresponding to the lowest vP which introduces the goal/source argument (Larson 1988). The nature of the spatio-temporal operator in that phase is more mysterious, but may consist of a special end-point (telicity) or aktionsart operator. While remaining truly agnostic about the nature of such an operator, we use EndP as a heuristic label in this paper, without truly attributing any real significance to the name. Carnie (in prep) shows that the importance of the phasing approach is seen in the way in which the semantic interpretations of the phases do not interact in terms of the mapping principle. Take for example the following English sentence: (10) There was some linguist helping the boy (when I walked into the room). We follow Chomsky (2001b) in assuming that there is no covert movement. The object NP is a definite, as such it is mapped into the restrictor. But notice that the structurally higher subject is clearly non-specific, nonpresuppositional and focal, thus mapped onto the nuclear scope. Crucially, at least on the surface, the nuclear scope of the VP seems to contain the definite, which should be mapped onto the


restrictor. More formally, the VP should not be allowed to contain a non-variable, such as the strongly quantified object. The phasing approach eliminates this problem (as well as providing a new solution for phenomena previously analyzed as involving LF movement). The first phase consists of the aspect marking, the lexical verb and the definite object. (11) [AspP Prog [VP help [the boy]]] The object (overtly: Johnson 1991) shifts out of the VP, creating a variable (trace). We leave aside here the problem of the order of the object and participle, noting that it can be constructed through a variety of head movements (V Æ Asp Æ vP). (12) [AspP [the boy]i Prog [VP help ti]] This maps to a partial (phase-specific) LF, where the object is construed outside the domain of existential closure, and is thus mapped to the restrictor. This partial LF, like the rest of the phase, is unavailable to the derivation at the next level. The structure of this LF is partial only in the sense that it doesn’t represent the LF for the whole sentence, it is a complete well formed LF (consisting of a quantifier and its nuclear scope and restrictor) for that phase. Let us call the structure in (12) = Z. The next phase contains Z, but may not refer to nor interpret its contents: (13) [TP There T [vP some linguist [v’ v [Z]]]] Here some linguist is interpreted existentially, internal to the nuclear scope. The fact that the boy is also contained in this constituent is irrelevant, since it is in a phase that has already received an interpretation. The interpretation of nominals in lower phases need not interfere with the mapping of nominals at higher levels. Nonetheless, there are clear situations where the phases interact. For example, we will see that shifted goals with ditransitive verbs appear shifted in the higher phase as well (they appear to the left of both shifted and unshifted objects). We attribute this to Case. Similarly, cases of split ergativity clearly require situations where objects move into higher phases to get absolutive/nominative case (see Carnie in prep. for more discussion). Voice alternations also seem to be driven by the interaction of the interpretative requirements on various arguments, as will be discussed below.

2.

Differential object marking: Object shift and clitic movement

We start our exemplification of the mapping approach to hierarchies with a brief discussion of the best known effects: object shift (also called differential object placement or DOP), differential object marking (DOM), and clitic raising. These topics are widely treated in the literature and we present them here both for


completeness and because they present the simplest examples of the mapping approach to hierarchies. 2.1 Object shift/scrambling6 We have already seen one example of object shift in this paper, that of the positioning of specific DPs in German, seen in example (5) above. We will not belabor the point here, only pointing out that objects under the analysis we have given here shift from inside the VP on their phase to the specifier of AspP to get out of the nuclear scope defined by VP. Similar accounts can be given to the phenomena of object shift in the Scandinavian languages (see Thráinsson 2000 for a survey, and Diesing and Jelinek 1995 for extensive discussion), VOS/VSO alternations in Austronesian (Rackowski and Travis 2000), and Mayan (Carnie in prep), and verb/ object ordering in Scots Gaelic (Adger 1997), and particle shift in English (Diesing and Jelinek 1995) to name only a few. 2.2 Differential object marking The phenomenon of differential object marking occurs when a distinct case form is used for two different types of objects based on some argument hierarchy, as seen in the Turkish example below (data from Enç 1991): (14) a.

Ali bir kitab-G aldi Ali one book-acc bought “A book is such that Ali bought it.” b. Ali bir kitap aldi Ali one book bought “Ali bought some book or another.”

Overt accusative marking correlates to specificity/presuppositionality. This is further supported by the fact that strongly quantified (and thus presumably strongly presuppositional) NPs must be marked with accusative case (data taken from Diesing 1992): (15) a.

Ali her kitab-G okudu Ali every book-acc read “Ali read every book” b. *Ali her kitap okudu

Diesing’s analysis of these facts is easily translated into the variation of her system presented here:


(16) ... [AspP Acc [VP Ø V]] ∃

Overt accusative case marks situations where the object has shifted out of the VP. A similar account can be given for Persian -râ marking and Nez Perce accusative/ objective alternations (Cash Cash and Carnie 2002). Although it may be coincidental, all of the cases we have mentioned show an ambiguity between the shifted and unshifted positions (because the verb is final, or the arguments can further scramble), so Case marking is used to indicate the positional shift caused by the mapping. DOM is the focus of much of the research in the Functional Optimality Theory framework. In that literature, emphasis is placed on the morphology/syntax interface. Morphologically marked cases correspond to marked argument positioning. We take issue with the idea that this is essentially a morphological phenomenon, given that it is clearly related to phenomena that are expressed purely syntactically, such as object shift and scrambling, to name a few. The mapping principle approach can account for a wider variety of cases, as seen below. 2.3 Clitic raising Diesing and Jelinek (1995) claim that relative scope in Egyptian Arabic (EA) is fixed overtly at s-structure. Evidence for this comes from the fact that existential indefinites are excluded from the specifier of IP, as are non-presupposed readings of indefinites. With this in mind, consider the case of clitic object pronouns in EA. Jelinek and Diesing show that these elements are real arguments and are not merely agreement morphology. However, these clitic pronouns are tightly bound to the inflected verb. Indeed they even appear internal to the sentential negation morphology ma…š. Full DP objects, by contrast, cannot (17b,d) (data from Diesing and Jelinek 1995: 144). (17) a.

ma-biyibii’-hum-š neg-selling-them-neg “he isn’t selling them” b. *ma-biyibii’-š humma neg-selling-neg they c. ma-biyibii’-š burtu’aan. neg-selling-neg oranges “he is not selling oranges” d. *ma-biyibii’ burtu’aan-š neg-selling oranges-neg

Diesing and Jelinek’s analysis of the distinction between full NPs and object clitic pronouns is that pronouns are inherently presupposed (they require a referent to be


fixed before they are used), as such they are incompatible with existential force, and must be moved out of the VP. One mechanism for doing this is by cliticizing to the tensed verb — which itself is external to the VP (due to head movement). The pronoun thus raises out of the nuclear scope.

3.

Dative/Accusative alternations7

We now turn to the slightly more complicated situations that arise with alternations in the deepest phase of the structure. We consider Dative/Accusative alternations in a variety of languages, including but not limited to the phenomenon known as dative shift/movement. After examining a range of data we return to a formal account in terms of the mapping principle, where we implement the intuition of Basilico (1998) of a topic/focus alternation in ditransitive alternations in terms of our phasal mapping. 3.1 Yaqui Yaqui (also known as Hiaki and Yoeme), is an Uto-Aztecan language spoken in Sonora, Mexico and Arizona. Yaqui has SOV word order, and NPs have overt structural case. However, dative movement is never “optional”, that is, discourse controlled. In Yaqui, the constraints on the distribution of the dative are lexical; it is the verb that determines the distribution of the dative. In English, there are also lexical constraints on the verb with respect to dative alternations. These constraints appear to be related to the manner of the action, and crucially to the affectedness of the goal. (14) a. b. c. d. e. f.

They donated the books to the library/committee. *They donated the library/committee the books. He confided his hopes to his friend. *He confided his friend his hopes. They communicated their concerns to the president. *They communicated the president their concerns.

Most Yaqui ditransitive verbs select an acc/dat array, but there is a small closed class of verbs that permit two acc marked objects. Compare the following two examples: (15) ’aapo Huan-tau ’uka vachi-ta maka-k he John-dat det.acc corn-acc give-perf “He gave John the corn.”

acc/dat

(16) ’aapo Huan-ta ’uka vachi-ta miika-k acc/acc he John-acc det.acc corn-acc give.food-perf “He gave John the corn (as a gift).”


There is a clear difference in the meaning of the verbs shown in (15) and (16). The verb in (16), miika, is employed when food is given “with good heart”, to “feed” someone in a culturally recognized situation. Those Yaqui verbs that require the acc/acc argument array have animate goals, that are strongly affected by the action of the agent. Other verbs that select an acc/acc argument array include: (17) ’aapo ’enchi ’uka ’etehui-ta mahta-k he you.acc det.acc story-acc teach-perf “He taught you the story.”

acc/acc

(18) ’aapo ’enchi ’uka kava’i-ta rewwa-k acc/acc he you.acc det.acc horse-acc borrow-perf “He borrowed the horse from you.” (19) ’aapo ’enchi ’uka tomi-ta ’u’aa-k acc/acc he you.acc det.acc money-acc take away-perf “He took the money away from you.” Verbs that select an acc/dat pattern include:8 (20) ’aapo ’eu ’uka toto’i-ta mana-k acc/dat he you.dat det.acc chicken-acc serve-perf “He served the chicken to you.” (as of a waiter) (21) ’aapo ’uka laapis-ta neu bwise-k he det.acc pencil-acc me.dat hand-perf “He handed the pencil to me.”

acc/dat

(22) ’uka miisi-ta=ne Maria-ta-u toha-k det.acc cat-acc=I Mary-dat bring-perf “I brought the cat to Mary.”

acc/dat

Additional evidence for the affectedness of the acc goals in Yaqui is provided by the fact that with the double accusative verbs, both arguments are obligatory; neither can be dropped. However, with acc/dat verbs, the more peripheral or unaffected dative argument can typically be omitted. (23) a.

Huan Peo-ta ’uka vachi-ta miika-k acc/acc John Pete-acc det.acc corn-acc give-perf “John gave Pete the corn.” b. *Huan ’uka vachi-ta miika-k c. *Huan Peo-ta miika-k

(24) a.

Huan Peo-ta-u ’uka vachi-ta nenka-k “John sold the corn to Pete.” b. Huan ’uka vachi-ta nenka-k “John sold the corn.” c. *Huan Peo-ta-u nenka-k

acc/dat


In Yaqui then, we see a difference in the marking of goal (and other second object) arguments in terms of their affectedness. More affected, and thus more presupposed arguments are given acc case, less affected more asserted arguments are given dat case. 3.2 Dative movement in English We use the term “dative movement” informally to refer to sentence pairs of the following kind: (25) a. I gave a book to Mary. b. I gave Mary a book. In the days of transformational grammar, it was more or less assumed that this kind of alternation between sentences with the same truth value was a matter of free choice on the part of the speaker. But dative movement is not a matter of free variation. Notice that these sentences occur in very different discourse contexts. This is an expression of Information Structure; the subject is backgrounded or topical, unstressed, and the VP is the focus, the new information in the sentence. The question/answer context is very revealing about Information Structure across languages. (26) Q: WHO did you give a +book to? A: I gave a book to +Mary. (27) Q: WHAT did you give +Mary? A: I gave Mary a +book. (Capital letters mark contrastive focus on the wh-word. The + sign in marks the normal or default intonation contour of the English sentence, where the stress peak occurs on the VP or last DP in the sentence.) If we reverse the question contexts here, as in: (28) Q: WHO did you give a +book to? A:##I gave Mary a +book. (29) Q: WHAT did you give +Mary? A:##I gave a book to +Mary. The result is very odd, to say the least. The dative alternation sentences have the same truth values — the same thing happened in the world — but one or the other construction is appropriate in context. The speaker is not aware of selecting between sentences with or without dative movement; she simply produces the sentence that fits best in the discourse in order to maintain topicality.


In English, we mark Information Structure via stress and intonation, so we tend to think of it as “post-syntactic”. We can even “override” the effects of dative movement by placing contrastive stress on an argument: (30) a. I gave a BOOK to Mary. b. I gave MARY a book. But in many languages, Information Structure is marked in the morphosyntax, and has to be taken into account in the analysis of the syntax proper. Important work by David Basilico (1998) points to the analogy between (i) ordering the subject and object in terms of topicality, and (ii) ordering the two objects in a dative construction in terms of topicality. Basilico claims that in a dative construction, the first object is higher in topicality than the second object. In other words, there is a topic/ focus structure within the vP itself. This is entirely consistent with the approach we posit here, except we attribute topic/focus interpretation with tree mapping rather than particular functional categories. With dative movement, the goal argument is “advanced” to a position above the theme at a “light verb” projection (see Section 3.5 below for discussion). Across languages, an important generalization concerning dative movement may be made: Only animate goals are advanced to topic positions. A restriction of this kind was noted by Bresnan (1982) and Oehrle (1976). Compare: (31) a. b. c. d.

The lawyer sent a letter to +Ellen. The lawyer sent Ellen a +letter. The lawyer sent a letter to +Dallas. *The lawyer sent Dallas a +letter.

When the goal is an inanimate destination, dative movement is excluded. This contrast brings out a important feature: Animate Goals are affected in a way that inanimates are not. The link between the features of affectedness and animacy are shown in the following: (32) Q: What happened to +Ellen? A: The lawyer sent Ellen a +letter. (33) Q: What happened to +Dallas? A: *The lawyer sent Dallas a +letter. Let us assume, along with Harley (1995), that to give X to Y causes X to possess Y. As a result of the action described in (32), Ellen is caused to possess a letter; but as a result of (33), the city of Dallas is not caused to possess a letter, unless by “Dallas” we are referring to some group of people, a committee or an institution. Animates have certain properties that inanimates do not, including being voluntary possessors or “holders” to use Kratzer’s terminology.


Note that the implication does not go the other way in English: animate goals are not obligatorily shifted. Topic continuity in discourse is the controlling factor, and animates tend to be topical, established in the discourse, while inanimates are more often new information. There is a cluster of features that fall together here; definiteness, person, and animacy. Animates tend to be definite, inanimates more often indefinite. First and second person are exclusively animate, while only some third persons are. Languages elect to grammaticalize discourse probabilities of the kind that control topic continuity.9 3.3 Datives in Lummi In the English and Yaqui examples above, we find the following pattern emerging: (34) In dative hierarchies, higher-ranked goal arguments have only structural case, and lower-ranked goals have only an inherent (oblique) case. In hierarchies of this kind, “productive” dative movement is excluded; the case assignment of goal arguments follow strictly from their rank in the hierarchy. Arguments that differ in rank also differ in structural position, and thus differ in their Case options. Now let us consider some examples of hierarchies in which dative movement is excluded for particular argument classes defined with respect to person or definiteness. In some hierarchies, first/second person goal arguments are obligatorily “advanced” to acc case marking, and are never oblique; in contrast, third person goal arguments, a class which includes indefinites, are expressed only as obliques. There may be additional ranking internal to the category of third person with respect to the features of definiteness, animacy, humanness, volitionality, etc. Lummi excludes productive dative movement; the speaker cannot place focus alternatively on the theme or the goal according to discourse factors. Animate goals are obligatorily “advanced” to the status of direct object. The item exchanged may optionally be identified by adding an oblique nominal. In (35), the root ’o]6s “give” appears with the auxiliary or “light” verb -t, one of a small closed set of transitivizers: (35) ’o]6s-t-s=l6’=Ø (’6 c6 scˇeen6xw) give-trans-3erg=past=3abs (obl det salmon “He gifted them ([with] a/the salmon).” Nominal adjuncts function as “anti-topics” (Lambrecht 1994). Without a nominal, the sentence is interpreted as having a definite third person pronominal object10 — the absolutive, which is the only null pronoun in any paradigm. (36) ’o]6s-t-Ø=l6’=sxw (c6 xwl6mi) give-trans-3abs=past=2sgnom (det Lummi “You gifted them, (the Lummi).”


Lummi has no free-standing pronouns with which the pronominal arguments could agree, ruling out the possibility that they constitute agreement. There are no oblique pronouns or pronominal objects of prepositions in the language (Jelinek 1997). The single oblique marker occurs only with nominals. Inanimate destinations are oblique nominals: (37) ye’=l6’=Ø (’6 c6 swi’iScˇ) go=past=3abs (obl det lake “He went (to the lake).” Inanimate destinations cannot be “advanced”. Pronominal arguments are backgrounded, and oblique nominals are given focus. In sum, these constraints produce the following effect in Lummi: Pronouns, presuppositional and topical, are direct arguments with grammatical case; Full nominals (DPs), non-presupposed and focal, are adjuncts that may have oblique case. 3.4 Datives in Navajo In Navajo as in Yaqui and Lummi, there is no “productive” dative movement. Oblique arguments in Navajo are marked with postpositions. Willie (1991) divides Navajo postpositions into two classes, the grammatical vs. the lexical. Grammatical postpositions include the animate goal marker -aa and the benefactive -á. Lexical postpositions include góó which indicates inanimate goals. You will notice that both góó and -aa mark goals. They differ, however in a number of important properties. For example, góó can only be used to mark inanimates: (38) KinSání-góó déyá Flagstaff-to 1sg.nom.go.fut “I will go to Flagstaff” (39) *’awéé’-góó jooS yíSmáás baby-to ball 3acc.3nom.rolling [“She is rolling the ball to the baby.”] To indicate the intended meaning of (39), one uses the goal marker -aa: (40) (’awéé’) jooS [yaa yílmáás]. (baby ball [3-to 3acc.3nom.rolling “She is rolling the ball to him (the baby).” The distribution and phonology of these two postpositions is very different. -góó is affixed to the DP it modifies. The object of -góó is not optional (although the whole góó phrase itself is optional). -aa, by contrast, falls within the phonological domain of the verb complex (see Hale this volume for discussion); in particular it follows the famous yi-/bi- pronouns (e.g., the y in yaa in (40)). Any DPs with -aa are optional, just like full NP subject and object arguments. The object of -aa is


obligatorily pronominal. This, we claim, is key. Being pronominal, the argument marked by -aa is presuppositional. We take -aa to be the rough equivalent of the accusative11 case marking found on presuppositional goals in Yaqui, Lummi and English. In many ways, they behave like other structurally case marked arguments. For example, they participate in the direct/inverse voice alternation. (41) a.

yich’i˛’ yáSti’ 3-to 3nom-speaking “He is speaking to her.” b. bich’i˛’ yáSti’ 3-to 3nom-speaking “He is being spoken to by her.”

Direct

Inverse

We thus have the identical situation to that of Lummi: we have structural case on pronouns; and inherent case on DPs. In sum, in languages where information structure is marked overtly in the morphosyntax, there is no “optional” dative movement; there are syntactic constraints on the distribution of the dative. In Lummi, goal pronouns are direct objects; only nominals may be marked oblique. In Navajo the structural cases, the accusative plus the postpositional dative and benefactive, appear with pronominal arguments, and there is a class of lexical postpositions that derive oblique nominals. 3.5 A Mapping analysis In what follows we sketch a phase-geometric mapping explanation for the general pattern seen in the data above. We first sketch the broad theoretical approach, and then return to the differences among the languages. Let us start with the case of a presupposed goal (i.e., dative movement). We work first with the most deeply embedded phase, which consists of the lexical verb, the goal argument and the end point spaciotemporal operator “End”. The goal cannot remain in the nuclear scope since it does not constitute a variable. It thus raises to the specifier of EndP. As such it is mapped to the restrictor of the phase. (42) Phase 1: EndP End′ End

VP V

nuclear scope Goal


This can’t be the end of the story, however. In English, for example, advanced goals appear before the theme argument. (43) The linguist gave the consultant a present In first phase, the theme isn’t even present. We claim that the solution to this problem lies in case licensing. We claim that EndP is not capable of case licensing any arguments. The goal then, in (42) is not case-marked. In order to receive case, we claim that it moves to a case checking position in the next higher phase. Fortunately, by moving the NP to the specifier of EndP, we have placed the goal at the phase edge, allowing the next phase to access it. (43) Phase 2

AspP

ACC case AspP

Asp

vP Theme

v′ v

EndP Goal

...

Notice that the interpretation of the nominal is fixed in the first phase, so subsequent movement for PF phenomena like Case will not affect the semantics. In order to get the correct word order in a language like English, we also need to claim that the verb undergoes short verb movement (Johnson 1991) (V Æ End Æ v1 Æ Asp Æ v2 ) although this need not occur in every language. Needless to say, Asp needs to be able to license multiple acc cases, since the theme argument also gets acc. This kind of situation is not unique. For example, the phenomenon of multiple wh-movement in languages like Serbo-Croatian and other south Slavic languages causes a similar problem for feature checking (see Karimi, this volume, for discussion). Whatever the solution to the problem of multiple wh-movement is, we could apply it here. Let us turn now to the cases where the goal is asserted, rather than presupposed. Unlike (42), the goal here can remain inside the nuclear scope, where it is interpreted under existential closure.


(44) Phase 1:

EndP End′ End

VP V

Nuclear scope Goal

Again, case checking is required. But here the goal is not accessible to the next phase (it is not on the phase edge). The only way to avoid crashing is to mark the goal with an inherent case. Were we writing this paper in the late 1980s or early 1990s, we’d say that this inherent case marking occurred under government from End (lexicalized as to or from). Interestingly, recently (in Beyond Explanatory Adequacy, 2001a) Chomsky has introduced checking (via Agree) under “local c-command”. As far as we can tell, this is the same thing as government. At the risk of being slaves to fashion, we tentatively point to this as an explanation for the inherent case marking on asserted goals. Let us turn now to the differences among the various languages discussed above. The broad strokes of the analysis remain the same. English is the most transparent case. It corresponds directly to the description above. The account of Yaqui is closely related. The only difference is that in Yaqui, mapping to the restrictor has been grammaticized to include only strongly affected objects, and is as such lexically limited. The cases of Navajo and Lummi are a little more complex. Both these languages are pronominal argument languages (Willie and Jelinek 2000). Recall that pronominals are inherently12 presuppositional, as such they cannot be left in the nuclear scope of the lexical verb. Both Navajo and Lummi use a device remarkably similar to that used by Egyptian Arabic. The arguments “incorporate” into the verb and raise13 with it out of the nuclear scope. There is an important difference between languages like English and languages like Lummi and Navajo. In English we can override discourse considerations in the choice of dative movement or not by using contrastive stress. In Lummi and Navajo, dative alternations are obligatory, they cannot be overridden by contrastive stress — this is a property of pronominal argument languages in general (Jelinek 2000b).

4. Mapping the highest phase: The role of voice and case splits We now turn to mapping effects in the highest (subject oriented) phase. Here we will see the effects of the mapping in terms of split case/agreement marking and voice alternations.


4.1 Topicality Agreement in Oromo Consider the following data from the Cushitic language Oromo from Ethiopia. In this language, topicality is directly encoded in the agreement system of the verb (data from from Clamons et al. 1999). Verbs in Oromo show agreement with presupposed/ topical subjects (45), but not with focal/non-presuppositional NPs (46). (45) intal-t-ií-n hoolaa bit-t-e girl-fem-subj-top sheep buy-fem-past “The girl bought a sheep” (46) intala takka-á hoolaa bit-e girl one-fem-subj sheep buy-past “A girl bought the sheep” Let us take morphological agreement to be a local relation between a specifier and head. Further let’s assume that Oromo verbs are in the highest head position (TP). Subjects in SpecTP will trigger agreement (47a). Low subjects in vP will not (47b): (47) a.

b.

TP Topic

t

---

T′ vP

TP

vP

T+V-agr v′

T′

Subj

T+V v′

In (45) the subject is presupposed, so must raise out of the vP, where it triggers gender agreement with the verb in T. In (46), the existence of girl is asserted, thus does not raise out of vP, and thus does not trigger agreement 4.2 Navajo animacy hierarchies Navajo is well known for the complex associated animacy hierarchy in NPs. However, first and second person, as well as the “fourth” (other third) person are off the animacy scale and when a sentence contains one of these arguments along with a third, the higher ranked arguments are topical and the third person is in focus. ashkii yiiSts´ø boy 3obj-1sg.subj-saw “I (Topic) saw +the boy.” b. ashkii shiiSts´ø boy 1sg.obj-3subj-saw “+The boy saw me (Topic).”

(48) a.


The animacy hierarchy varies to some extent across speakers, as the language now has fewer and fewer fully fluent speakers. A typical hierarchy has the ranks: Adult human > child > large animal > small animal > inanimate. The hierarchy works like this: if there is more than one DP in a sentence, they are always ordered with respect to the hierarchy. Regardless of their grammatical relations, they must always be ordered with respect to the animacy scale. Compare: (49) asdz´øø´ at’ééd aSk’ésdisí yeiní’´ø direct womani girlj candyk 3j-to-3objk-3subji-gave “The woman gave the candy to the +girl.” (50) asdz´øø´ at’ééd aSk’ésdisí beiní’a inverse womani girlj candyk 3j-to-3objk-3subji-gave “The woman was given the candy by the +girl.” The change in the initial verbal prefix is an instance of the yi-/bi- alternation, which makes the lower ranked patient (the woman) topical, while the higher ranked agent (the girl) is in focus. The bi- pronoun marks a topicalized patient. This is the inverse voice, which like the passive makes the patient topical. However, unlike the passive, it is a transitive with two direct arguments (there is no by-phrase). The passive is just the best available translation. The DPs never change order; the verbal morphology marks the voice change. If the sentence contains only DPs that are equal in rank on the scale, then the alternation can be used to mark discourse topicality; that is, the speaker can maintain an established topic. (51) ashkii at’eed yizts’os ˛ boy girl 3-3-kissed “The boy kissed the +girl.”

direct

(52) ashkii at’eed bizts’os ˛ boy girl 3-3-kissed “The boy was kissed by the +girl.”

inverse

Willie (1991) showed that the default reading of nominals in Navajo is definite; they may receive indefinite readings in existential contexts, and the like. It is possible to derive an indefinite NP by the use of a particle léi’. (53) a.

ashkii léi’ at’ééd yizts’os ˛ boy part girl 3-3-kissed “A boy kissed the girl.” b. ashkii at’ééd léi’ yizts’os ˛ boy girl part 3-3-kissed “The boy kissed a girl.”

direct

direct

However, if the inverse form is used, the topical NP cannot be made indefinite.


(54) a.

ashkii at’ééd léi’ bizts’os ˛ inverse boy girl part 3-3-kissed “The boy was kissed by a girl.” b. *ashkii léi’ at’ééd bizts’os ˛ inverse boy part girl 3-3-kissed [*“A boy was kissed by the girl.”]

We suggest that the topmost NP in the inverse has moved up to the restrictor topic position in TP, and therefore cannot be made indefinite by existential closure. In the direct construction, the initial NP has been subjected to existential closure, and the definite NP raises (possibly at LF), leaving a trace. The Navajo animacy hierarchy and voice alternation are central to Navajo syntax, and have been the source of numerous disputes between Athabaskanists. The reason for this is that these aspects of Navajo grammar represent unique developments within the far-flung Athabaskan family, and the cognates of the pronouns used in the voice alternations have developed into many other uses in various relatives of Navajo, giving rise to many different syntactic developments. But there appears to be some version of a hierarchy in most of the languages in the family. 4.3 Lummi ergative splits14 Lummi is an example of a language that elects to pay attention to the upper reaches of the presuppositionality scale. It employs an ergative split of the most common variety (1, 2 > 3) to exclude 3 > 1, 2. First and second person are nom/acc in case, and third person is erg/abs. It is very instructive to lead bilingual Lummi speakers through a paradigm such as the following. (55) a.

b.

c.

d. e.

n6p-t-o]6S=sxw advise-trans-1pl.acc=2sg.nom “You advised us.” n6p-t-Ø=sxw advise-trans-3abs=2sg.nom “You advised him.” n6p-t-s=Ø advise-trans-3erg=3abs “He advised him.” *n6p-t-s-o]6S or *n6p-t-o]6S-s [*“He advised us”] n6p-t-]=S advise-trans-passive=1pl.nom “We were advised.”

2nom > 1acc

2nom > 3abs

3erg > 3abs *3erg > 1acc

1nom


Speakers produce the example sentences comfortably until they are asked to say (55d) “He advised us”. Then they stop, look surprised and uneasy, and then if they are good consultants, after a while may say something like “Well, we don’t say it that way. You might say [(55e)], but it’s not really the same, is it?” Jelinek (1993) extends the mapping principle to account for this kind of person split. The analysis she gives is very much in the spirit of the analyses by Abraham (1996), Dubois (1987), Delancy (1981), which give a discourse basis to split ergativity; however, Jelinek formalizes these intuitive characterizations in terms of the mapping principle. She claims that nominative local persons (1st and 2nd person) are inherently presuppositional, and thus must rise out of the domain of existential closure. Ergative non-local persons, by contrast, remain VP internal. The case split involves two intersecting grammatical properties. Jelinek assumes, following Murasugi (1993), Levin (1981), and Bok-Bennema (1991) that ergative case is a lexical case. Further, like Murasugi, she claims that the ergative is a VP internal case. VPs, you will recall, define the domain of existential closure. Local NPs are thus disallowed from this position, since they are presuppositional. So no local NP would ever take ergative case. The unavailability of the reading in (e) with an ergative third person and a 1/2 object follows from a simple fact about Lummi morphology. Lummi only allows a single internal argument to be realized in the verbal morphology, and both ergatives and accusatives are VP internal.15 There is a morphological template in Lummi: V + (transitivity) + (voice) + (internal argument) = (external argument clitic). For example, double object constructions are simply impossible in Lummi (as seen above in Section 3), because there is only one morphological slot for internal arguments in the verb complex. The unavailability of (e) follows from the fact that we have both an ergative and an accusative pronoun competing for a single morphological position. Let us translate Jelinek’s account into the phasing approach advocated here. The basic intuition that ergative case marking represents a lexically case marked internal argument construed within the nuclear scope is retained. The difference between the Jelinek’s account and the one we present here is notational (and we hope without negative empirical consequences). The crucial distinction is made in terms of the nature of the little ‘v’ category. First let us consider a 1/2 subject, which would get nominative case and is interpreted topically in the restrictor. The subject raises out of the VP that introduces it, on the topmost phase then. It takes nominative case in the specifier of TP. Such a structure looks like (56):


TP

(56) Nom

T′ T

nuclear scope

vP 1/2

v′ ...

v1/2

The little v in this construction (v1/2) selects only for 1/2 person subject and selects for a AspP that can assign accusative case: (57)

...

vP AspP

v1/2 [1/2, +Asp1]

Asp′ VP

Asp1 [acc] object

V

Contrast this with a third person subject, which is interpreted focally.16 This pronoun is selected for by a different little v (v3). The lexical entry for v3 differs in that it (1) assigns a lexical case to its specifier, and (2) selects for a defective Asp2, which cannot assign acc case. The subject, construed internal to the nuclear scope, does not raise, and takes the inherently assigned ergative case marking. (58)

TP T′ T

vP 3-erg

nuclear scope v′

v3 [erg]

...

On the lower phase, the object is Caseless and cannot receive Case from the defective Asp2. In order to be licensed then, it must raise into the next phase, where


it gets absolutive (= nominative) case marking in the specifier of TP (presumably stopping in the specifier of Asp2P for generalized EPP reasons, thus placing it at the phase edge — we abstract away from this intermediate step) (59)

TP T′ T [abs]

vP 3-erg

v′

v3 [3, erg, +Asp2]

Asp2P Asp2′ VP

Asp2 V

Obj

Again, these trees should not be taken as the surface form, as further fronting of the verb and the pronominal argument status of the language further obscures the surface form. What is crucial, however, is that ergative and accusative cases compete for a morphological slot (as encoded by the selection of different types of Asp head by the various kinds of v) and that there is a mapping relation between the information structure status of the pronoun and its syntactic position and case. The availability of the passive voice here reflects yet a third type of v (vpass), which does not assign an external argument or lexical case, but selects for Asp2P,17 so the object (of any kind) can raise to the specifier of TP for case checking. 4.4 Word order in Kirundi18 Ndayiragije (1999) presents a fascinating paradigm of data from Kirundi, a Bantu language. We propose that the simplest analysis of this data falls out from a mapping approach to word order, where the phenomenon is intermediate between a split-case pattern (in that we appear to get distinct agreement patterns depending upon the presuppositionality of the subject) and inverse marking (the fact that the grammatical relations are reversed is indicated by a verbal prefix). Consider the following data, which concern contrastive focus on the object.


(60) a.

Abâna ba-á-ra-nyôye amatá children 3pl-past-af-drink.perf milk “Children drank milk” b. Abâna ba-á-nyôye amatá children 3pl-past-drink.perf milk “Children drank milk (not water)”

SVO

SVO

Neutral information structure in Kirundi requires the presence of a special antifocus marker (glossed af). This particle is disallowed in all contrastive focus constructions (like 60b). An interesting alternation occurs when a subject, rather than an object is contrastively focused, this is seen in the pair of sentences in (11) (61) a.

Petero a-á-ra-guze ibitabo Peter 3s-pst-af-buy.perf books “Peter bought books” b. Ibitabo bi-á-guze Petero books 3pl-pst-buy.perf Peter “Peter (not John) bought books.”

SVO

OVS

A word order alternation occurs here: the contrastively focused subject is final, and the object is in initial position and most importantly, the verb agrees with the object. In order to account for this pattern we appeal to the information structure of these constructions. We assume, following Rooth19 (1999) (and contra Kiss 1998) that contrastive focus, like information focus introduces new information, even though it does so from a presupposed set. As such, contrastively focused elements are compatible with the nuclear scope of the clause. Evidence for this is the fact that, in Kirundi, pronominal objects (inherently presupposed) are incompatible with contrastive focus (62).20 (62) a.

Abâna ba-á-ra-ya-nyôye. children 3p-pst-af-cl-drink.perf “Children drank it.” b. *Abâna ba-á-Ø-ya-nyôye. children 3p-pst-f-cl-drink.perf “Children drank it.”

On the other hand, contrastive focus marking (Ø) is obligatory in negatives (63) and wh-questions (64), both of which are typically associated with novel information. (63) a.

Abâna nti-ba-á-nyôye amatá. children neg-3p-pst-drink.perf milk “children didn’t drink milk.”


b. abâna ba-á-ra-nyôye amatá. children 3p-pst-af-drink.perf milk “children drank milk.” c. *abâna nti-ba-á-ra-nyôye amatá. children neg-3p-pst-af-drink.perf milk (64) a.

Abâna ba-á-Ø-nyôye iki? children 3p-pst-f-drink.perf what “What did children drink?” b. *Abâna ba-á-ra-nyôye iki? children 3p-pst-af-drink.perf what “What did children drink?”

The crux of our analysis here should be obvious from our treatment of Lummi and other languages. Novel, focal subjects remain within the nuclear scope (as indicated by their post verbal position); the object raises to satisfy the prototypical subject properties of the clause (as indicated by its initial position and the fact it triggers agreement). However, the analysis is complicated by certain word order facts. Indeed, as Ndayiragije shows, there is extensive evidence that there is movement of contrastively marked arguments to the right edge of the clause. Consider first the following evidence from adverb placement. When the anti-focus marker — ra is present, the object must be adjacent to the verb, no intervening adverb can occur (65a&b). With contrastive focus constructions however, the contrastively focused object appears finally (65c) (65) a.

Yohani a-á-ra-oógeje imiduga néezá. John 3s-pst-af-wash.perf cars well “John washed cars well.” b. *Yohani a-á-ra-oógeje néezá imiduga. John 3s-pst-af-wash.perf well cars c. Yohani a-á-Ø-oógeje néezá imiduga. John 3s-pst-f-wash.perf well cars “John washed cars well (not trucks).”

Related evidence comes from embedded clauses: Objects without contrastive focus must appear adjacent to the verb, to the left of embedded clauses, focused objects appear to the right of such clauses. (66) a.

pro tu-á-ra-rungitse abânai [CP PROi kuryâma]. pro 1p-pst-af-send.perf children inf-sleep “We sent children to sleep.” b. *pro tu-á-ra-rungitse [CP PROi kuryâma] abânai. pro 1p-pst-af-send.perf inf-sleep children


c.

pro tu-á-Ø-rungitse [CP PROi kuryâma] abânai. pro 1p-pst-f-send.perf inf-sleep children “We sent to sleep children (not adults).”

Ndayiragije presents these faces as evidence that the contrastively focused element has shifted to the right. This evidence must be reconciled with the interpretive requirement that contrastive focal arguments remain in the nuclear scope. We adopt here part of the analysis of Ndayiragije, who posits a functional category for contrastively focused elements. He calls this position FOC. Due to the potential confusion of such a functional category with informational focus, we relabel it Cont (for contrastive focus). This position immediately dominates the vP. In order to account for the strictly final position of contrastively focused elements, Ndayiragije posits a rightwards specifier for this position. Cont has the dual function of marking the NPs as contrastively focused, and inherently case marking them. Let us see how this would work with a contrastively focused subject. We must make the following assumptions. Cont selects for a particular kind of v — one which introduces external arguments that are to be contrasted.21 This v is realized as a Ø and we call it vcont. vcont selects for Asp2. Selectional considerations on the merge operation will only allow the following combination of functional and lexical heads if Cont is present (both phases are shown in the tree below). (67)

TP T′ T

ContP A

Cont′

highest phase nuclear scope

vcontP

Cont [+vcont] agent [+Cont]

vcont′

vcont [+cont subj, +Asp2]

Asp2P Asp2′

B

VP

Asp2 V

theme


Crucially, as a slight departure from the above, nuclear scope extends to ContP. The agent raises to SpecContP (movement A in (67)) inherently checking its case. Since the selected-for Asp2 does not assign accusative case, the theme moves to the specifier of TP, where it gets nominative case and triggers verbal agreement (we leave aside the interpretation of the theme here, which may or may not involve construal within its own nuclear scope). The verb raises through all the relevant functional heads, giving us the correct OVS order. Consider now a contrastively focused object. Here the relevant phase-specific nuclear scope is a lower ContP (which like its higher phase counterpart is an inherent case assigner). TP

(67)

T′ T B

vP v′

agent v

AspP Asp′ Asp

lower phase nuclear scope

ContP A

Cont′ Cont

VP V

theme

Here the theme moves to the specifier of ContP to be marked as contrastive (appearing clause finally) and be inherently case marked A . The subject raises to the specifier of TP, where it triggers subject agreement B . Again the verb raises through all the functional categories, to pick up the relevant morphemes. Finally let’s consider the case of the antifocus sentence. In this form, neither phase has a Cont. As such the special antifocus v (-ra) is used. This v selects for an accusative assigning Asp (68), which in turn cannot select for Cont:


TP

(68)

T′

B

T [nom]

vP v′

agent v -ra af

AspP Asp′ Asp [acc]

VP V

theme A

This gives us straightforward nominative/accusative ordering with neutral interpretation.

5.

Some speculations on language variation in hierarchies

In this paper, we’ve presented a view of hierarchies based on the syntax semantics mapping, and we have liberally applied the terms “presuppositional” and “topical” without being specific about what these terms mean. Indeed, the careful reader will note that our discussion of Arabic has all pronouns as being “presuppositional” but our account of Lummi has only 1/2 persons as “presuppositional”. We would like to offer some very vague speculations on why this might be the case. Two issues are at stake: On one hand there appears to be a universal scale of presuppositionality. That is, for example, you never find a language that has third persons ranked above 1/2 persons. On the other hand, there is widespread variation in what counts as “presuppositional” or “topical”. For example, VSO/VOS word order alternations in the closely related Mayan languages of Tzeltzal and K’ichee’ vary in precisely which features drive the alternation: Tzeltzal makes use of animacy, K’ichee’ uses specificity. Why should the feature of animacy enter into the hierarchies? Notice that the distribution of this feature is precisely the same as that of definiteness/specificity: that is, all first and second person arguments have the properties of both definiteness/specificity and animacy, while only some third persons do. All arguments that are either inanimate or indefinite are exclusively third person. Of course, there are


also third persons that are animate and definite; that is, the features of person and definiteness/animacy do not coincide, but there is class inclusion. (1) all first and second persons are definite and animate; (2) some but not all third persons are definite and animate; (3) all indefinites and inanimates are third person. This codistribution of these properties make it likely that a particular grammar might use to express either one or the other in an argument hierarchy, and this is what we find. Also, a language may choose to group all pronominal referents together, as apposed to NPs, whereas another language might lump all third persons with indefinites/inanimates, since all indefinites/inanimates are third person. Returning to the larger question of language variation, let us first consider the model of grammar we are working in here — roughly that of Chomsky (1995). This framework holds that there are essentially two components to the grammar (leaving aside morphology and phonology): (a) a lexicon, which holds all language specific idiosyncratic information and (b) a grammar, which is universal (i.e. there are no parameters per se). The grammar contains the rules Merge and Move/Attract/Agree and a set of interface constraints, such as the principle of full interpretation and the mapping principle, which govern convergence. These parts of the system are the “Linguistic” components of the human language capacity. But these linguistic portions do not exist in a vacuum. They interact with other kinds of cognition. For example, the syntax generates strings which are submitted to Articulatory/Perceptual and Conceptual/Intensional Interfaces. Needless to say, however, these AP and CI interfaces also have an effect on the lexicon. As we develop, we use our general cognitive skills to create form/meaning correspondences relating the world to concepts and concepts to words. These correspondences are further complicated by the fact that we also rely on the linguistic input of our caregivers. We thus see the effects of historical change, language acquisition, and general cognitive skills in the shape of lexical entries. Variation among languages lies here. Which features a language chooses to grammaticize as “presuppositional” are an effect of all of these factors. The apparent universal side of the equation is much trickier, and we can only guess at its origins. Part of it may reside in the grammatical mapping principle — that is, hierarchy sensitive syntactic phenomena are universally determined by mapping rules. Part of it may come from extra-linguistic sources, such as general cognition. To give a parallel, we all perceive the differences among colors, knowing what is a more “blue” color and what is a more “green” color, but languages vary in if and how this cognitive/perceptual issue is lexicalized. We may well have a general cognitive sense of “what’s new” and “what’s old” but languages differ in how they lexicalize this into formal features in the lexicon.


6. Conclusion In this very cursory treatment of some very complex data, we have reviewed the Mapping Principle, and noted that the hierarchies always conform to it. The hierarchies often grammaticalize discourse tendencies and produce overgeneralizations, for example: by excluding all sentences where 3 > 1/2, Lummi excludes all sentences where Indefinites > Definites. Nevertheless we consider the wide variety accounted for in this paper good support for the conjecture that there is a correlation between syntactic prominence and prominence on the presuppositionality scale. The hierarchies are interesting “imperfections” in particular languages that produce variation within and across language families, and represent the grammaticalization of certain highly frequent discourse tendencies.

Notes * We would like to thank Emmon Bach, Luis Barragan, Andrew Barss, David Basilico, Tom Bever, Sheila Dooley Collberg, Molly Diesing, Dick Demers, Ken Hale, Heidi Harley, Jason Haugen, Simin Karimi, Terry Langendoen, Barbara Partee, Montserrat Sanz, Mary Willie, Robert Young, four anonymous reviewers, and the audience at the Spring 2002 Arizona Proseminar in Syntax, for their help and counsel. Maria Amarillas, Frances Delgado and Maria Morales helped us with the Yaqui examples and analysis. ‡

.Editors’ note: It may at first blush seem odd to find a paper by a volume honoree in a festschrift. However, we’d like you to know that we invited Eloise Jelinek to contribute to this volume without telling her that the book was in her honor. We knew the paper she would write would be very appropriate to the theme, and felt it would be nice to include something by our favorite colleague. This paper received the same peer review and editing process that all the other papers in the volume did. When Eloise found out that the book was in her honor, she vehemently wanted to pull this paper. We wouldn’t let her, since we think it is a excellent contribution to the volume theme. Eloise would like everyone to know that she did not write this paper in her own honor. However, Andrew Carnie would like everyone to know that it was a honor to work with Eloise and his small portion of the paper is dedicated to her. [AC, HH, MW] 1. Aissen (1999, 2000). 2. The precise means of encoding “more animate, more topical, more specific” etc. is discussed below in Section 5. 3. In many ways, this kind of approach mirrors formally the intuitions of many functional accounts of hierarchies and ergativity: DeLancy’s (1981), given in terms of figure/ground viewpoint and attention flow mechanisms (a.k.a. voice marking and case marking) and DuBois’s (1987) discourse based account. 4. We leave aside the question of whether this scrambling (see Karimi, this voulme), or indeed any of the movement posited in this paper, is motivated by purely formal syntactic features (as would be consistent with Chomsky 1995) or by semantic considerations such as effability (see Adger 1997).


5. The terms ‘topic’ and ‘focus’ have had various uses in the linguistic literature; they have been used to refer to discourse phenomena, often in quantitative studies. The term ‘topic’ has been used to refer to a sentential adjunct. The terms have also been used with reference to the information structure of the sentence; we follow this usage, as in work by Partee (1991), Heim (1982), and Diesing (1992), as well as work by Sasse (1987), von Fintel (1989) and others, in the analysis of the thetic/categorical contrast. In this tradition, that part of the sentence that is familiar and presuppositional is classed as topical, established in the discourse, while what is new information in the context of the sentence belongs to the focus. Basilico extends the thetic/categorical contrast to the topicalization of goal arguments within the object array in dative movement. 6. We collapse object shift and object scrambling into one category here, although it is by no means clear that they form a unitary phenomenon, see Thráinsson (2000) for discussion. 7. Parts of this section are based on Jelinek (2000a) 8. We have only been able to find a single minimal pair, where both an acc/dat and an acc/acc object array appear. There is a derived construction with the root vit- “see” plus the causative suffix -tua. With an acc/acc object array, this derived form has the meaning “show”. (i) ’aapo ’enchi ’uka kava’i-ta vit-tua-k he you.acc det.acc horse-acc see-caus-perf “He showed you the horse.”

acc/acc

The derived form vit-tua also occurs with a dat goal argument: (ii) ’aapo ’eu ’uka kava’i-ta vit-tua-k he you.dat det.acc horse-acc see-caus-perf “He sent you the horse.”

acc/dat

In view of the contrasting glosses, it appears that we are dealing with two distinct derived verbs here, rather than the alternation seen in “dative movement” in English, where the meaning of the verb itself does not change. We noted above that the Yaqui “double accusative” verbs assign the property of being strongly affected to their animate goals. An acc (animate) goal is generally more affected than a dat goal, which may be inanimate. In Yaqui this strongly affected property is also seen with benefactive (or “malefactive”) objects in the language. The Yaqui applicative (benefactive) suffix -ria adds an acc benefactee argument: (iii) ’aapo ’enchi bwiik-ria-k he you.acc sing-bene-perf “He sang for you.” Hale and Keyser (1993) show that in other Uto-Aztecan languages, some causative constructions receive a benefactive interpretation. The causative and benefactive suffixes do not co-occur. We suggest that the Yaqui “double accusative” construction, with a strongly affected goal, should be classed as a variety of the benefactive. Tying this together with other ditransitives, we note that Harley (1995) argues that a causative projection is universal in the structure of verbs corresponding to “give”. 9. Note also, that like Yaqui, English has lexical restrictions on dative movement. Latinate verbs, and verbs like whisper, do not allow the alternation. 10. Note that the interpretation here must be a definite presupposed entity. Indefinite readings for these pronouns (such as “They gifted someone.”) are not possible with just the bare pronouns and no nominal adjuncts. 11. Although confusingly, this postposition is called the dative in most Athabaskanist linguistics. Nomenclature is irrelevant to the basic point we are making, these elements are structurally case marked, no matter what we call the case. In this regard they stand in stark contrast to lexical postpositions, which we claim actually mark oblique cases.


12. In the next section we will claim that some pronominals are more presuppositional than others. This will be discussed in detail there. 13. We are assuming here that Navajo is a verb raising language. 14. The discussion in this section is based partly on Jelinek (1993). 15. The observant reader may have noted that our account here appears to be inconsistent with the account of Arabic discussed above. In that 3rd person pronouns raise obligatorily in Arabic because they have to be mapped to the restrictor (pronouns receiving a presupposed reading). Yet here we claim that third pronouns are mapped to the nuclear scope. An account of the difference might lie in the fact that Arabic and Lummi differ in their statuses as Pronominal Argument languages. We, however, believe that a simpler solution exists. We have consistently used the terms topic and focus here. This is because we wish to argue that what is important is the relative presuppositionality of the arguments, what is asserted and what is “older” information. Topic and focus then range over degrees of presuppositionality. 16. Although not necessarily entirely non-presuppositionally, as pronouns usually have a discourse or sentential referent. They are simply less presuppositional than the highly topical 1/2 person. 17. The analysis here essentially stipulates Burzio’s generalization. The reason for the stipulation is a topic for future research. 18. The data in this section is taken exclusively from Ndayiragije (1999). Ndayiragije offers a formal account in terms of the minimalist program. We adopt only parts of his account here. Nevertheless, we benefited greatly from his careful presentation of the data, and both his descriptive and theoretical characterizations of the facts, even if we don’t adopt all the details of his analysis. 19. Our thanks to Molly Diesing for very helpful discussion of these issues. 20. Obviously this isn’t true for English, pronouns can be contrastively stressed (I kissed HIM). We have no idea why these should be the case, except to note that this is primarily acceptable only in a deictic sense of the pronoun rather than an anaphoric usage. 21. A simpler solution may be that Cont is a v,. We leave this as an open possibility.

Focus movement and the nature of uninterpretable features* Simin Karimi University of Arizona

1.

Introduction

Focus has been discussed in recent literature with respect to a variety of its properties (Kiss 1998, Zubizarreta 1998, Hale, Jelinek, Willie (to appear), among others). This phenomenon, which is often characterized as “optional” or “discourse-based”, is most often dealt with in the functionalist tradition. However it raises special questions when examined in the light of a formalist syntactic approach, such as the Minimalist Program (MP). For example, is there any evidence indicating that focus movement is feature driven? Is it subject to any principles governing the operation Move such as the Minimal Link Condition (MLC) (Chomsky 1995a)? Second, if focus is a feature driven operation representing an instance of Move, how can we account for the optionality we observe with respect to this movement in some languages? More precisely, how can a feature only optionally trigger movement? Within the Minimalist Program, the feature triggering the operation Move is considered to have an uninterpretable nature with no semantic or phonological content (Chomsky 1995a and work thereafter, Pesetsky and Torrego (2001), among others). Thus if focus movement is feature driven, the feature triggering this operation must be of an uninterpretable nature. The issue of the existence of uninterpretable features is itself problematic for a theory whose goal is to go beyond the level of explanatory adequacy (Chomsky 2001a). This paper is an effort to address the question of the functional motivation of movement, focus in this case, and its formalization within a feature based system. Since uninterpretable features are considered to be the triggering force for Move, the nature of these features is examined, and an alternative analysis to Chomsky’s (1995a) proposal is suggested. Finally, the optionality of focus, a functionally motivated movement, is discussed, and a solution is proposed within a theory that prohibits the existence of optional operations.

298 Simin Karimi

2.

Focus movement in Persian

Movement that represents discourse functions is not a new idea, and goes back to early seventies (Jackendoff 1973, Culicover and Rochemont 1983, and Rochemont 1986). Some authors consider focus to be an instance of feature triggered Move (Bailyn 2002, Kiss (to appear), among others). Persian exhibits two types of focus. One type appears within the VP, and receives an interpretation as denoting new information. The second type requires a heavy stress, motivates an optional movement out of VP, and expresses contrastive interpretation (or identificational interpretation in the sense of Kiss 1998). In this paper, I concentrate only on the second type. Persian lacks obligatory wh-movement. When stressed and indicating a contrastive interpretation wh-phrases may optionally move. Two wh-phases may undergo contrastive or identificational focus movement in the same simple clause. I have shown elsewhere (Karimi 1999) that the dislocated focused elements are subject to several conditions, indicating that focus movement must be triggered by an uninterpretable focus feature. A summary of that discussion appears below. First, movement of two focused items is subject to an adjacency condition, as illustrated by the contrast in (1). (1) a.

KIi bâ KIj pro fekr-mi-kon-i [CP ti tj be-raghs-e] WHO with WHO thought-prog-do-2sg subj-dance-3sg Lit: “It is WHO with WHO you think will dance?’ b. ??KIi emruz bâ KIj pro fekr-mi-kon-i [CP ti tj be-raghs-e] KIi today (Karimi 1999)

The two wh-phrases are adjacent in (1a) while they are separated by the adverb emruz ‘today’ in (1b). The ungrammaticality of (1b) suggests that the dislocated wh-phrases must occupy two specifiers of the same head. Thus (2) illustrates the structure of (1a), and accounts for the ungrammaticality of (1b). That is, the adverb cannot intervene between two specifiers of the same head. (2) [FocP XPi [XPk [Foc′ Foc [yp ... ti ... tk ... ]]]]

Second, the contrast between (3a) and (3b) below implies that MLC must be obeyed when two elements bearing the same feature compete for the same position. Otherwise, the derivation crashes. (3) a.

[faghat be Kimea]i man ti se tâ KETAB dâd-am [only to Kimea I three-part book gave-1sg “It was only to Kimea that I gave three BOOKs.” (I gave other people other things.)

Focus movement and the nature of uninterpretable features 299

b. *se tâ KETAB man faghat be KIMEA ti dâd-am. Intended meaning: “It was three BOOKS that I gave only to KIMEA” In (3a), we have the only-phrase and the stressed direct object, both requiring identificational focus.1 The only-phrase has moved while the direct object has remained in situ. By contrast, the direct object has moved in (3b), crossing the only-phrase in a higher position. The ill-formedness of (3b) indicates that focus movement must be subject to MLC.2 Similarly, in a double wh-construction, one of the wh-phrases may remain in situ. In such a case, the MLC must be obeyed, as the contrast between (4a) and (4b) indicates.3 (4) a.

KIi pro fekr mi-kon-i [CP ti bâ KI be-raghs-e] WHO thought prog-do-2sg with WHO subj-dance-3sg “WHO is it you think will dance with who?” b. *bâ KIj pro fekr mi-kon-i [CP KIi tj be-raghs-e]

The sentence in (4b) is grammatical if the wh-phrase in situ is not stressed. In that case, it is interpreted as an indefinite DP with no quantificational force (similar to someone in English): with whom is it you think that someone will dance? The sentences in (3a) and (4a) indicate that focus movement is optional in Persian. This operation is optional in a number of other languages as well, such as Rumanian, Italian, and Catalan.4 I will come back to this issue in Section 5. Third, the surface order of two dislocated wh-phrases is also subject to a restriction, as illustrated by the contrast between (1a), repeated in (5a), and (5b) (see also the structure in (2)). (5) a.

KIi bâ KIj pro fekr-mi-kon-i [CP ti tj be-raghs-e]

WHO with WHO thought-prog-do-2sg subj-dance-3sg Lit: “It is WHO with WHO you think will dance?” b. *bâ KIj KIi pro fekr-mi-kon-i [CP ti tj be-raghs-e]

This contrast is compatible with Richards’ (1997, 2002) “tucking-in” condition: moveable elements have to move into the closest (i.e., the lowest) specifier of a head with multiple specifiers. In (5a), the lower wh-phrase has been tucked-in into the lower specifier of the same head, satisfying the tucking-in requirement. The ungrammaticality of the sentence in (5b) is due to the fact that the lower wh-phrase has not been tucked-in into the lower specifier. The fact that focus movement is subject to conditions governing Move suggests that this movement is triggered by a feature. If this conclusion is on the right track,

300 Simin Karimi

the feature triggering the movement must be an uninterpretable one, as proposed by Chomsky (1995a) and others.

3.

The nature of uninterpretable features

Chomsky (1995a: 277) proposes that certain features enter into interpretation at Logical Form (LF) while others are uninterpretable and must be eliminated for convergence. This distinction is manifested in terms of [±interpretable] features. Categorial features and φ-features of nouns are interpretable features — they have to be visible at LF for the purpose of interpretation. Other features, including the Case features of T and V, are uninterpretable, and will cause the derivation to crash if they are not checked and eliminated by LF. Interpretable features do not need to be checked, and remain unchanged even if they are checked. Chomsky (1995a: 281–3) suggests that the F feature of the target, rather than the F¢ feature of the moved element, is always uninterpretable. If the features of the moved element and the target were both interpretable, many unmotivated operations would be allowed: (6) ___ T seems [that Mary is the winner] The categorial and φ-features of Mary have been checked in the embedded clause. But since these features are interpretable, they are unchanged. Chomsky considers the feature representing the Extended Projection Principle (EPP) as an interpretable feature of D and thus is not eliminated after checking. Therefore, a DP should be able to enter into multiple checking of EPP. Consequently, Mary could raise to the specifier position of the matrix sentence to check the EPP feature of the matrix T. This operation will yield the ill-formed string in (7):5 (7) *Maryi seems [that ti is the winner] The ungrammaticality of (7) is explained if the Case feature is uninterpretable, and is deleted after checking. That is, the moved DP, although satisfying EPP in the matrix clause, cannot satisfy the Case feature of the matrix T. Thus the uninterpretable Case feature of T remains unchecked, causing the derivation to crash. Within earlier versions of the theory of Principles and Parameters, the sentence in (8) would be ruled out by the Case Filter, an isolated condition which was purely stipulative. (8) *It seems John to be intelligent. Within the Minimalist Program, (8) is ruled out since the Case feature of John remains unchecked, and thus this sentence becomes subject to the Case Filter. Given the uninterpretable Case feature the Case Filter becomes an unnecessary


condition. The ungrammaticality of (8) follows from the general principle of Full Interpretation (FI). That is, the uninterpretable Case feature, if not checked and eliminated, will not be interpretable at LF, and the derivation will crash. The same situation holds for other types of features that trigger movement. In the case of whconstructions, C must have an uninterpretable feature which needs to be checked and eliminated by LF. Thus overt wh-movement is triggered by the presence of this feature.6 The same situation should hold for focus movement, if this operation is feature driven as was claimed to be the case in the previous section. There are two problems with this view: a.

If the feature of the target is uninterpretable, and is eliminated after checking, how can we explain multiple overt wh-movements in languages that allow such operations? Similarly, how are multiple focus movements such as the one in (1a) justified? In both cases, the uninterpretable feature of the target would be eliminated after the first checking, leaving the other wh or focus phrases in situ, contrary to facts. b. Uninterpretable features lack semantic and phonological content. How is the existence of such elements justified within a theory that seeks explanatory adequacy? First I discuss (a). Lasnik (1999) suggests that if F¢, the feature of the moved element, were responsible for movement, we could explain the facts in languages that exhibit multiple wh-movements. Consider the following data taken from Serbo-Croatian. (9) a.

Ko šta gdje kupuje? who what where buys “Who buys what where?” b. *Ko kupuje šta gdje? c. *Ko šta kupuje gdje? d. *Ko gdje kupuje šta? (Boškovic´ 1997, per Lasnik 1999)

The difference between Serbo-Croatian and English could be attributed to a parametric difference with respect to the number of specifiers of CP each language allows. The same parametric difference holds between languages that allow more than one scrambled focus, and those that do not. The difference boils down to the number of specifier positions a functional head is allowed to have in a given language.7 Lasnik’s claim contradicts Chomsky (1995a: 281–283) who states that the triggering feature is in the head of the target position, not in the element that moves. This proposal is based on the assumption that F can be checked more than once without being eliminated. This approach accounts for LF movement of wh-phrases in languages like English where multiple overt wh-movement is not

302 Simin Karimi

allowed: The wh-feature of the wh-phrase in situ moves covertly and is checked before the uninterpretable F of the target is eliminated. Although this analysis seems to explain the facts, we still face the conceptual problem with respect to uninterpretable features: the explanatory justification for their existance? Chomsky (2000a) recognizes this problem. He suggests that the language faculty may be close to perfect, but nevertheless, two potential imperfections exist: uninterpretable features and the ‘displacement property’. Chomsky states, however, that the displacement property is not an imperfection since it is motivated by interpretive requirements. That is, displacement has external motivation in terms of semantic interpretation and perhaps even processing. Uninterpretable features motivate the implementation of the displacement property in natural language. Chomsky concludes that “all of this falls into place if uninterpretable features are the mechanism for displacement, perhaps even an optimal mechanism” (Chomsky 2001a: 14). The existence of the uninterpretable features is justified on the assumption that they motivate the displacement operations in language.8 We can ask two questions about this approach. First, is overt displacement a necessary condition for semantic interpretation in natural language? Let us take a look at some cases. Overt wh-movement, although obligatory in English type languages, does not apply in Japanese type languages. Similarly, contrastive focus involves (optional) displacement in some languages (cf. Persian), but not in others (cf. English). Movement triggered by Case is not a semantically motivated displacement, and is not a necessary operation cross-linguistically: Nominative Case triggers movement in English, but not in Persian, and it is not clear whether Accusative Case triggers displacement in any language. Finally, the EPP feature is responsible for DP-movement in English type languages, and V movement in Romance and Greek type languages (Alexiadou and Anagnostopoulou 1998). However, it does not trigger displacement in Persian type languages (Karimi, under revision). Given these facts, it is not clear that overt movement is a necessary requirement in language, and thus does not justify the imperfect nature of the uninterpretable features. The second question regarding the uninterpretable features is this: where do they come from? In Chomsky’s system they must exist in the lexicon since they have to be distinguished from the interpretable features: uninterpretable features appear without value since their value is redundant, and is determined by Agree. Thus ‘match’ in these cases is ‘non-distinctness’ rather than ‘identity’ (Chomsky 2001a: 13). If the uninterpretable feature F exists in the lexicon without a value, does this mean that it can be matched with any kind of interpretable feature (e.g. the uninterpretable Case-F with the interpretable wh-F¢ in a wh-phrase)? This points to another problem: if the uninterpretable feature F lacks semantic and phonological content as well as any kind of ‘value’, why does it exist? Can the displacement property of language be satisfied in a different fashion? This issue is addressed in the next section.


4. An alternative proposal Chomsky (2001a) states that the Interface Condition (IC) requires that all surface features be interpretable. If this is true, we could satisfy this requirement with the following proposal: (10) a. F and F¢ are both interpretable. b. F motivates movement. c. Convergence is possible only if F/F¢ are visible at LF. (11) Visibility Principle9 F/F¢ are visible if locally checked. Note that (10b) represents Chomsky’s idea that F, the feature of the target, triggers the displacement. The properties in (10) together with the principle in (11) account for wh-movement in English type and Serbo-Croatian type languages. Since F and F¢ are both interpretable, they are not eliminated after checking, and thus multiple wh-movement is possible. The difference between the two types of languages is reduced to a parametric distinction with respect to the number of specifiers each language allows (see Section 3). Multiple focus movement is accounted for in the same fashion (cf. (1a)). The ill-formedness of (7), however, cannot be explained by this proposal since nothing prevents the DP Mary from moving into the specifier position of the matrix TP to check its Case once more. There is a solution to this problem. Case has been traditionally considered a property that does not contribute to interpretation, and thus is irrelevant to LF. However, it motivates movement, and therefore, is relevant to the Phonological Form (PF), and must be interpreted at that level.10 If this line of argumentation is on the right track, we can impose the following restriction: (12) PF features become inactive after checking. We also need to modify the condition in (10c) to include PF features. (10) c¢. Convergence is possible only if F/F¢ are visible at LF and PF. Given (12), the ill-formedness of (7) is explained: although the Case feature of the matrix T needs to be checked, and therefore, may force the DP Mary to move, the Case feature of the latter is inactive. Thus this sentence violates (10c¢) since there remains a feature (that is, the Case feature of the matrix T) that cannot be interpreted at PF. The ungrammaticality of (8), repeated below in (13), is explained the same way. (13) *It seems John to be intelligent. The Case feature of the DP John remains unchecked, and therefore, is not visible and thus cannot be interpreted at PF.

304 Simin Karimi

5.

Back to focus and optionality

We saw in Section 2 that contrastive/identificational focus can optionally move. This issue creates a problem for a feature based theory: If displacement is triggered by features, how can it be optional? In other words, how can a feature optionally trigger movement? Note that this problem exists regardless of the nature of F: if it is uninterpretable, it has to be checked and eliminated. If it is interpretable, as suggested in the previous section, it needs to be checked in order to become visible for convergence. Thus optionality provides a problem for both systems. Contrastive focus is usually heavily stressed. It could be argued that focus has a set of three, rather than two, features: F, F¢, and Fs, the latter a PF feature representing stress. F¢ represents the semantic property of contrastive focus, and thus is an LF feature. Let us assume that stress, rather than movement, is the primary feature required for contrastive focus; thus selection of F becomes optional. If F is not selected from the lexicon, F¢ (the LF feature) and Fs (the PF feature) are checked against each other, and there is no movement. If F is selected, the movement becomes obligatory. Feature driven movement thus remains obligatory. Optionality is reduced to the selection of F from the lexicon. Let us see how this three-way relation works in terms of checking. If F is not selected, F¢ and Fs, the features of the head of DP, undergo Agreement in situ. This is illustrated in (14). (14) [DP [D¢ [D F¢ Fs ]]] If F is selected, it triggers the movement of DP into the specifier of FocP, as in (15). (15) [FocP DP F′ FS [Foc′ F [... t]]]

In (15), F and F¢ are checked in a specifier-head relationship. Note that even though F¢ is already checked against Fs, it will still be able to be checked against F since it is the LF feature of focus, and therefore, does not become inactive after checking. In addition to the fact that the proposal advanced in this section removes the imperfect property of the uninterpretable features from the grammar, it is also supported by an empirical consideration. In this system, Fs feature triggers a phonological reflex, that is stress. Something that Chomsky’s and Lasnik’s approaches have no explanation for. 6. Conclusion I argued in this paper that focus movement, a functionally motivated displacement, is a feature driven operation, as evidence by the fact that it is subject to a number of


principles and conditions that govern the displacement property in language. If focus movement is feature driven, the feature triggering it must be uninterpretable, as assumed within the MP framework. However, the existence of uninterpretable features is a source of imperfection’, and is not justified within a theory whose goal is to reach the level of explanatory adequacy and beyond. Thus I proposed in this paper that (a) all features are interpretable, (b) the principle of Full Interpretation is not only relevant at LF, but also at PF, and (c) PF interpretable features, unlike LF features, may be checked only once. Allowing the PF features to be checked only once (cf. (12)) prevents the grammar from generating ungrammatical strings such as the one in (7). I leave a thorough examination of this proposal regarding all instances of PF features to future research.

Notes * I would like to thank Andrew Carnie, Heidi Harley, and an anonymous reviewer for their insightful and valuable comments. I am also indebted to Andrew Barss, Sheila Dooley Collberg, Eloise Jelinek, Terry Langendoen, Massimo P. Palmarini, Raffaella Folli, and Rudy Troike, as well as students in the proseminar of Spring 2002 for their helpful and encouraging input. All shortcomings are solely mine. 1. See Kiss 1998 for a discussion of only-phrases as an instance of identificational focus. 2. Note that (3b) is grammatical with the interpretation all I did was giving three books to Kimea, where faghat ‘only’ modifies the verb. 3. Similar constraint is observed in Japanese. The following data are from Takahashi (1993), cited by Richards (1997). (i) a.

John-ga [Bill-ga dare-ni [Mary-ga nani-o tabeta to itta to] John-nom [Bill-nom who-dat [Mary-nom what-acc ate that said that omotteriu no? thinks q “Who does John think that Bill told t that Mary ate what?” b. Dare-ni John-ga [Bill-ga t [Mary-ga nani-o tabeta to] itta to] omotteriu no c. *Nani-o John-ga [Bill-ga dare-ni [Mary-ga t tabeta to] itta to] omotteriu no

4. Kiss (1998) states that the identificational feature triggers obligatory overt movement in some languages, for example, Greek. This movement is also obligatory in Modern Standard Arabic if XP is not morphologically marked for focus. 5. Chomsky (1995a) suggests that the EPP-feature of D is divorced from Case. Furthermore, every T has an EPP feature, but only control infinitives have a Case feature. It is not obvious, however, why the EPP feature has to be interpretable within Chomsky’s framework. That is, it is not clear what constitutes the semantic content of the EPP feature. I do not pursue this question in this paper. 6. Cyclic wh-movement is triggered by the EPP feature on the intermediate C(s). Only the final C has a wh-feature which is checked and deleted when the wh-phrase moves into its specifier. 7. See also Pesetsky (2000) who divides languages into three categories with respect to the number of specifiers of C: those with no specifier of C (Japanese and Korean), those with one specifier of C (such as German), and those with more than one specifier of C (such as English and Bulgarian). The

306 Simin Karimi

difference between English and Bulgarian is that the former allows only one overt wh-movement, and the rest of the wh-phrases move at LF, while the latter allows multiple overt wh-movements. 8. See Langendoen (2001) for a review of Chomsky (2000b), and a brief discussion of the same issue examined in this section. 9. The principle in (11) is obviously distinct from the traditional Visibility Condition that links Theta marking to Case assignment (Chomsky 1981: 96). 10. Haeberli (2001) suggests that morphological Case is interpreted at PF. This proposal can be extended to structural Case since the latter is responsible for a number of displacements in language.

Merge* D. Terence Langendoen University of Arizona

1.

Introduction: The minimalist program

The minimalist program is an effort to discover the degree to which the human language faculty is determined by sensorimotor and conceptual-intentional “interface conditions” together with considerations of “virtual conceptual necessity”, in particular by “general considerations of simplicity, elegance, and economy”. If it is entirely determined by these factors, then the human language faculty is a “perfect system, meeting external constraints as well as can be done” (Chomsky 1995a: 385–386). Chomsky calls the assumption of linguistic perfection “the strong minimalist thesis SMT”, and contends that SMT should guide all linguistic theorizing until shown incorrect (2001a: 3). 1.1 The centrality of the operation merge Chomsky also assumes that a language is a system whose “generative engine”, called “narrow syntax”, constructs a derivation for each choice of “lexical array” (2001a: 4). Presumably other arrangements consistent with SMT are possible, but the exploration of what these might be is not at issue here. I am concerned only with the specific question of what operations narrow syntax makes use of in constructing a derivation from a lexical array, and whether their existence is consistent with SMT. Chomsky asserts that narrow syntax “has one operation that comes ‘free’, in that it is required in some form for any recursive system: the operation Merge…” The condition that language is a recursive system is imposed by the conceptual-intentional interface. Merge is therefore “free”, a consequence of general principles, because recursion is impossible without it. Moreover, any other operation in narrow syntax besides merge “requires empirical motivation, and is a prima facie departure from SMT” (2001a: 6). 1.2 A minimalist approach to a functional phenomenon A major concern of the minimalist program is the reduction of the computational

308 D. Terence Langendoen

load in carrying out a derivation. A natural extension of that concern is the reduction of the complexity of the generated objects themselves, such as their degree of embedding, without sacrificing expressive power. Syntactic transformations, as they were first formulated in generative grammar, to some extent had the property of reducing the structural complexity of the generated objects (Miller and Chomsky 1963, Langendoen 1970). Finally, I consider a minimalist update of this idea in which a specific form of merge reduces the structural complexity of the objects it generates.

2.

Forms of merge

Merge is not a single operation, but a family of operations. To belong to the merge family, an operation must be able to yield an infinite set of objects from a finite basis. What type of merge operations are appropriate for narrow syntax depends on the nature of the infinite sets of objects it is required to create. 2.1 Set merge Chomsky identifies the simplest form that merge can take as the formation of twomember sets: Applied to two objects α and β, Merge forms the new object γ. What is γ? γ must be constituted somehow from the two items α and β … The simplest object constructed from α and β is the set {α, β}, so we take γ to be at least this set. (1995a: 396) [Merge] takes two elements α, β already constructed and creates a new one consisting of the two; in the simplest case {α, β}. (2001a: 6)

Chomsky calls this form of merge “set merge” (2001a: 18). Further, set merge is “external” if α and β are separate objects, and “internal” (essentially “move”) if one is part of the other, e.g. if β is part of α, in which case β is said to be a “copy” of its occurrence in α. He contends that external set merge builds “argument structure”, whereas internal set merge builds the structures required for “scopal and discourserelated properties” (2001a: 9). In (1), I show the result of externally merging c with the result of externally merging a and b.1 (1) S(S(a, b), c) = {{a, b}, c} The derivation can also be diagrammed as a tree as in (2), where the root and internal node labels are sets, not categories. Since the relation S is symmetric, the order of nodes in the tree is not significant.2

Merge 309

(2) Tree diagram of the derivation in (1) S({a, b}, c} = {{a, b}, c}

S(a, b) = {a, b} a

b

c

To distinguish external from internal set merge, I write the result of the latter as {α/β, β} if β is part of α, and {α, β/α} if α is part of β. In (3), I show the result of internally merging b with the result of externally merging a and b. In the corresponding tree diagram (4), the b that is the sister of {a, b} is the copy of the b that is the sister of a, and b/b is the “trace” of the original b which is bound by its copy. (3) S(S(a, b), b) = {{a, b}/b, b} = {{a, b/b}, b} (4) Tree diagram of the derivation in (3) S({a, b}, b} = {{a, b/b}, b}

S(a, b) = {a, b} a

b

b

2.2 Pair merge Chomsky (2001a) maintains that a more complex form of merge is required to generate adjoined structures; he calls this operation “pair merge”: But it is an empirical fact that there is also an asymmetric operation of adjunction, which takes two objects β and α and forms the ordered pair ·α, βÒ, α adjoined to β. Set-merge and pair-merge are descendants of substitution and adjunction in earlier theories. Given the basic properties of adjunction, we might intuitively think of α as attached to β on a separate plane, with β retaining all its properties on the “primary plane”, the simple structure. (2001a: 18)3

He also argues that the presence of pair merge in narrow syntax is motivated by a requirement of the conceptual-intentional interface, so that its existence is consistent with SMT: [R]ichness of expressive power requires an operation of predicate composition: that is not provided by set-Merge… But it is the essential semantic contribution of pair merge. (2001a: 18)


I represent the pair merge of β adjoined to α as P(α, β) = ·α, βÒ.4 The discussion of the use of pair merge for generating adjunct structures continues below. 2.3 Inadequacy of set merge to represent argument structure When the objects α and β are merged, one of them is the head. But the result of their set merge doesn’t indicate which one, since {α, β} = {β, α}. If α is lexical and β phrasal (i.e., if β but not α is of the form {γ, δ}), then from the configuration {α, β} = {α, {γ, δ}}, one might infer that α is the head and β its complement (Comp). However if both are lexical, one must further specify which is the head. Similarly, if α = {γ,{δ,η}} and β is lexical, then either β is a specifier (Spec) of γ, or α is the Comp of β, and one must further specify which. Chomsky (1995a) solves this problem by proposing a variety of merge that “labels” the result of the operation. The passage from Chomsky (1995a) quoted in Section 2 continues as follows: Does this [= set merge] suffice? Output [= interface] conditions dictate otherwise… γ must therefore at least (and we assume at most) be of the form {δ, {α, β}}, where δ identifies the relevant properties of γ; call δ the label of γ… If α projects [is the head], then γ = {α, {α, β}}. (1995a: 397)

Since the construct {α, {α, β}} is the standard way of representing the ordered pair ·α, βÒ set-theoretically (Enderton 1972: 6), the use of labels to represent argument structure conflicts with the use of pair merge to represent adjunct structure. Further, Collins (2001) argues against the use of labels in narrow syntax, showing that elements combine properly under set merge using information independently required in the lexical array. However, Collins’s arguments do not address the inadequacy of the output of set merge to represent argument structure. Another form of merge must be found to do so. 2.4 List merge The simplest and most elegant and economic way of representing argument structure is by listing the predicate and its arguments in a sequence. For example, Mostowski (1979) proposes that while a predicate in general is a function on infinite sequences of objects, every propositional function has finite “support”. Koslow remarks: It is difficult to say whether this practice was inspired by the “observation” that a natural language … has verbs with a natural number of places or whether the influence is just the reverse, from logical practice to the analysis of [natural language] verbs and predicates. (1992: 181)

Merge

Whatever the direction of influence, the requirement of the conceptual-intentional interface for argument structure, together with “general considerations of simplicity, elegance, and economy” dictates that the proper representation of argument structure in narrow syntax should be in the form of a list, an ordered sequence of objects. In keeping with logical practice, let us suppose that the predicate occurs first in each list, followed its arguments in sequence, first its Comp and then its Specs. Pair merge can represent the first step in constructing such lists, since if α is a predicate and β its Comp (its first argument), then P(α, β) = ·α, βÒ. However, if γ is the first Spec of α (its second argument), then pair merge gives the wrong result on the next step, since P(·α, βÒ, γ) = ··α, βÒ, γÒ, not the desired ·α, β, γÒ. To get the desired result, I propose a type of merge called “list merge” L, according to which L(α, β) = P(α, β) = ·α, βÒ if α is atomic, and L(·α, …Ò, β) = ·α, …, βÒ, an “ordered n-tuple” (Enderton 1972: 6), a list of n items.5 In the ordered triple ·α, β, γÒ, α is the head (the predicate), β is its Comp (its first argument), and γ its Spec (its second argument). Further application of list merge yields additional Specs, precisely as Chomsky (1995a: 432; 2001a: 6) argues for. For example L(·α, β, γÒ, δ) = ·α, β, γ, δÒ, an “ordered quadruple” in which γ and δ are Specs of α. Finally, since the output of list merge is less structured than that of pair merge, list merge must, like set merge, be considered conceptually simpler than pair merge.6 2.5 Another inadequacy of set merge According to Chomsky, application of internal merge of β with α, where β is part of α, must result in the creation of a new Spec: [D]isplacement [of β] from within α [must] be to the edge of α, yielding a new Spec. (2001a: 9)

To get this result using set merge, it must be explicitly stipulated that β is not the head of the resulting structure.7 However with list merge, the result follows immediately from the fact that the internal list merge of β with α can only be expressed as L(·α, …β… Ò, β)=·α, …β/β, … βÒ, and in that structure β is a Spec of α. 2.6 A possible objection to list merge Suppose we grant that in the structure (2) generated by set merge, a is the head, b is its Comp, and c is its Spec. In that structure, c asymmetrically c-commands b, and that relation can be used in an account of why an anaphor occurring in the position of b can be bound by an antecedent occurring in the position of c, whereas an anaphor occurring in the position of c cannot be bound by an antecedent occurring in the position of b. On the other hand, in the structure (5) of the syntactic object

311

312

D. Terence Langendoen

·a, b, cÒ generated by list merge, b and c c-command each other, and some other explanation for the asymmetry of the antecedent-anaphor relation is required. (5)

Tree diagram of the structure of the ordered triple ·a, b, cÒ 〈a, b, c〉 a

b

c

In fact, two different explanations are possible. One is that the asymmetry of the antecedent-anaphor relation is based on the positions of b and c in the list, rather than on c-command. The other is that in the tree diagram of the derivation of the ordered triple shown in (6), c does asymmetrically c-command b. Therefore, there can be no objection based on the c-command relation to the use of list merge to account for argument structure. (6)

Tree diagram of the derivation of the ordered triple ·a, b, cÒ L(〈a, b〉, c) = 〈a, b, c〉 L(a, b) = 〈a, b〉 a

3.

b

c

Using pair merge for generating adjunct structures

Using list merge for generating argument structure leaves pair merge available for generating adjunct structure, as Chomsky (2001a:15) proposed; see Section 2.2. The essential property of adjunction is the fact that if β is adjoined to α, then α as a whole is the head of the structure, not the head of α. That is, the adjunction of β to α = ·γ, δ, …Ò is ··γ, δ, …Ò, βÒ, where the analysis of α as a list prevents β from being construed as a Spec of γ, the head of α. To obtain this structure however, we need only apply pair merge, since P(·γ, δ, …Ò, β) = ··γ, δ, …Ò, βÒ. The contrast between adjoining β to α = ·γ, δ, …Ò and adding β as a Spec of γ, the head of α, can be seen by comparing (7) and (8), and the resulting trees in (9) and (10), where the ordering of nodes is significant, unlike in (2) and (4). (7)

P(α, β) = P(·γ, δ, …Ò, β) = ··γ, δ, …Ò, βÒ

(8)

L(α, β) = L(·γ, δ, …Ò, β) = ·γ, δ, …, βÒ

Merge

(9) Tree diagram of the pair merge of α = ·γ, δ, …Ò with β; i.e. the adjunction of β to α 〈〈g, d, ... 〉, b〉

〈g, d, ... 〉 g

...

d

b

(10) Tree diagram of the list merge of α = ·γ, δ, …Ò with β; i.e., the addition of β as a specifier of γ, the head of α 〈g, d, ..., b〉

g

...

d

b

Subsequent addition of elements to an adjunct structure are carried out by applications of list merge as in (11).8 As a result, adjuncts can be added to a head, in principle without limit, just as Specs can. The formula in (11) shows “high” attachment of the second adjunct ζ. On the other hand, the adjunction of ζ to β = ·η, θ, …Ò results in “low” attachment, as in (12).9 (11) L(P(·γ, δ, …Ò, ·η, θ, …Ò), ζ) = L(··γ, δ, …Ò, ·η, θ, …ÒÒ, ζ) = ··γ, δ, …Ò, ·η, θ, …Ò, ζÒ 〈〈g, d, ...〉, 〈h, q, ...〉, z〉

〈g, d, ...〉 g

〈h, q, ...〉 ...

d

h

z ...

q

(12) P(·γ, δ, …Ò, P(·η, θ, …Ò, ζ)) = P(·γ, δ, …Ò, ··η, θ, …Ò, ζÒ) = ··γ, δ, …Ò, ··η, θ, …Ò, ζÒÒ 〈〈g, d, ...〉, 〈〈h, q〉, ...〉, z〉〉

〈g, d, ...〉 g

d

〈〈h, θ, ...〉, ζ〉 〈h, q, ...〉

... h

q

z ...

3.1 Analysis of coordinate structure using pair merge Coordinate structure is distinct from argument structure, so it should not be constructed entirely by list merge (i.e., it is not the “functional projection” of

313


coordination). Asyndetic coordination (coordination without the use of a coordinating particle) can be handled by the combination of pair and list merge as described above. Similarly, the analysis of syndetic coordination requires the use of pair merge of the first member to the result of the list merge of the coordinating particle with the second member. So for example the structure of the coordinate phrase Y and Z is given by (13), and of the coordinate phrase X, Y and Z in (14); in both structures ·and, ZÒ is the head. (13) Structure of the coordinate phrase Y and Z 〈〈and, Z〉, Y〉 〈and, Z〉 and

Z

Y

(14) Structure of the coordinate phrase X, Y and Z 〈〈and, Z〉, Y, X〉 〈and, Z〉 and

Z

Y

X

3.2 Adjunction and internal merge Since internal merge is “freely available”, we may expect to find applications of internal pair merge in the formation of adjunct structures; its absence, in Chomsky’s words, “would be an imperfection” (2001a: 8). Chomsky (2001a) does not consider the possibility of internal pair merge, but only what happens when internal set merge (our internal list merge) applies to an adjunct structure as a whole. In fact, two types of internal pair merges are possible: (i) “internal head pair merge”, in which β is adjoined to α, where α = ·γ, δ, …Ò is part of β, as in (15); and (ii) “internal adjunct pair merge”, in which β is adjoined to α, where β = ·γ, δÒ is part of α, as in (16).10 (15) P(·γ, δ, …Ò, β/·γ, δ, …Ò) = ··γ, δ, …Ò, β/·γ, δ, …ÒÒ (16) P(α/·γ, δ, …Ò, ·γ, δ, …Ò) = ·α/·γ, δ, …Ò, ·γ, δ, …ÒÒ 3.2.1 Internal head pair merge Application of internal head pair merge (adjunction of β to α, where α is part of β) results in the “readjustment” of structure, as in the application of “readjustment rules” (Langendoen 1975). For the example in (15), let β be the structure in (17), which is the argument structure of they thought that you liked them, and α = ·that,

Merge

·liked, them, youÒÒ (so that γ = that and δ = ·liked, them, youÒ). Application of internal head pair merge in (18) results in the “readjusted” structure in which ·thought, t1, theyÒ (where t1 = α/α is the trace of ·that, ·liked, them, youÒÒ) is adjoined to ·that, ·liked, them, youÒÒ, and which is realized as the “pseudocoordinate” pattern they thought t1| that you liked them, where the vertical bar indicates an intonation break. 〈thought, 〈that, 〈liked, them, you〉〉, they〉

(17)

〈that, 〈liked, them, you〉〉

thought

that

〈liked, them, you〉 liked

(18)

they

them

you

〈〈that, 〈liked, them, you〉〉, 〈thought, t1, they〉〉 〈that, 〈liked, them, you〉〉 that

〈thought, t1, they〉


them

thought

t1

they

you

Although ·thought, t1, theyÒ is an adjunct in (18), it can be the locus of further external list merge operations. For example, subsequent external list merges of the complementizer that, the verb say, and the pronoun you results in the structure (19). To this structure, internal head list merge applies, attaching ·said, t1, youÒ as a second adjunct to ·that, ·liked, them, youÒÒ, resulting in (20), where t2 is the trace of ·that, ·thought, t1, theyÒÒ. The realization of this structure is the pseudocoordinate pattern you said t2| that they thought t1| that you liked them. As a result, the degree 4 nesting of the nonreadjusted argument structure (21) is reduced to degree 2. With each additional recursion, the degree of nesting of the nonreadjusted structures increases by 2, but that of the readjusted structures remains at 2. Such structures, unlike those generated by external list merge alone, can be processed by a finite-state device, thus providing a minimalist account of a functional phenomenon, as promised above. (19) ··that, ·liked, them, youÒÒ, ·said, ·that, ·thought, t1, theyÒÒ, youÒÒ

315


〈〈that, 〈liked, them, you〉〉, 〈that, 〈thought, t1, they〉〉, 〈said, t2, you〉〉

(20)

〈that, 〈liked, them, you〉〉 that

〈that, 〈thought, t1, they〉〉


them

you

that

〈said, t2, you〉

〈thought, t1, they〉 thought

t1

said

t2

you

they

〈said, 〈that, 〈thought, 〈that, 〈liked, them, you〉〉, they〉〉, you〉

(21) said

〈that, 〈thought, 〈that, 〈liked, them, you〉〉, they〉〉

you

〈thought, 〈that, 〈liked, them, you〉〉, they〉

that thought

〈that, 〈liked, them, you〉〉 that

they


them

you

3.2.2 Internal adjunct pair merge Internal adjunct pair merge (adjunction of β to α, where β is part of α) is an alternative mode of analysis for A-bar movements, instead of substitution (movement to the Spec of a higher “functional projection”). For example, given the argument structure of who did you like in (22), simple application of the internal adjunct pair merge in (23) (i.e., without any functional projections) results in the adjunction of who to the main clause, where tw is the trace of who, and where I ignore for ease of exposition the effects of subject–auxiliary inversion and do-support. (22) ·liked, who, youÒ (23) P(·liked, who, youÒ, who) = ··liked, tw, youÒ, whoÒ If such “wh-movement” is attempted from within a multiply embedded complement clause, as in (24), it may be supposed that the target and the goal occurrences of who are separated by too many list edges (a kind of “subjacency violation”), requiring intermediate applications of such movement. An alternative is that the structure of the first argument in (24) first undergoes internal adjunct merge as in (25), so that the target and the goal are separated by at most two list edges. Then the application of internal adjunct pair merge of who results in the structure (26), whose realization is who did you say t2| that they thought t1| that you liked tw. (24) *P(·said, ·that, ·thought, ·that, ·liked, who, youÒÒ, theyÒÒ, youÒ, who)

Merge

(25) P(··that, ·liked, who, youÒÒ, ·that, ·thought, t1, theyÒÒ, ·said, t2, youÒÒ, who) 〈〈〈that, 〈liked, t w, you〉〉, 〈that, 〈thought, t1, they〉〉, 〈said, t2, you〉〉, who〉

(26)

〈that, 〈liked, t w, you〉〉, 〈that, 〈thought, t1, they〉〉, 〈said, t2, you〉 〈that, 〈thought, t1, they〉〉

〈that, 〈liked, t w, you〉〉 〈liked, t w, you〉

that

liked

tw

that

you

〈said, t2, you〉

〈thought, t1, they〉 thought

t1

who

said

t2

you

they

This use of internal adjunct pair merge to generate “long distance” A-bar movements suggests that all such cases require the prior application of internal head pair merge, so as to flatten the structure sufficiently to avoid subjacency violations, but further consideration of this possibility lies outside the scope of this paper.

4. Conclusion I propose that recursive operations are ranked according to their “simplicity, elegance, and economy”, with set merge ranked highest, list merge second, and pair merge third. Then if set merge were adequate for the purposes of narrow syntax (i.e. if it provided the basis for interpretation of linguistic objects at the interfaces), SMT would compel us to use it. However, set merge is not adequate for any such purpose, so that list merge emerges as the next candidate. I find that list merge is adequate for some of the purposes that Chomsky (2001a) proposes for set merge, specifically accounting for argument structure. However, list merge (like set merge) is inadequate for adjunct structures, so that pair merge must be adopted for those cases. The presence of pair merge in narrow syntax then makes available a new range of mechanisms for narrow syntax, internal head and adjunct pair merge, whose explanatory power looks quite promising.

Notes * I thank Andrew Carnie, Noam Chomsky, Arnold Koslow and a mystery reviewer for helpful comments on earlier versions of this article, and Heidi Harley and Massimo Piattelli-Palmarini for their steadfast encouragement of this work. 1. The objects a, b, and c are not necessarily atomic (lexical). 2. In addition, the operator S is commutative.

317

318

D. Terence Langendoen

3. Technically, the pair merge relation is antisymmetric, not asymmetric, since ·α,βÒ = ·β,αÒ if and only if α = β. In addition, the pair merge operation is noncommutative. Pair merge is comparable to the LIST operator of Common LISP, since (list ’α ’β) = (α β), which is the ordered pair ·α, βÒ in LISP notation; see Touretzky (1997: 164). 4. The adjunct and the head are in the opposite order from Chomsky’s definition above, for reasons that will become clear below. 5. List merge is comparable to the Common LISP operator APPEND, since (append ’(α …) ’β)= (α … β), which is the LISP equivalent of ·α, …, βÒ. 6. A similar argument shows that the list-merge counterpart to set merge, call it “union” merge U, where U({α,β},γ)={α,β,γ}, is conceptually simpler than set merge; see also Chomsky (2001a:14, n. 50). 7. Chomsky (2001: 9) derives this requirement from what he calls the “extension condition” on set merge. No such condition is required for list merge. 8. If the order in which adjuncts are added is not linguistically significant, then additional adjuncts should be added by union merge; see note 6. 9. The structure in (11) represents the interpretation of a phrase like the box behind the chest next to the crate in which the box is both behind the chest and next to the crate, whereas the structure in (12) represents the interpretation of that phrase in which the box is behind the chest and the chest is next to the crate. 10. Internal pair merge, unlike internal list merge, is capable of copying into the head position, since pair merge is not an append-type operation. On the other hand, head-to-head movement, the adjunction of a lower head to a higher head, is only possible if the lower head is a list and not an atom. But lower list heads can only be constructed by adjunction in the first place, which means that head-to-head movement cannot be part of narrow syntax, as Chomsky (1995b: Ch. 4) has proposed; see also Chomsky (2001a: 23, n. 69).

Phonotactics and probabilistic ranking* Michael Hammond University of Arizona

1.

Introduction

Coleman & Pierrehumbert (1997) and Frisch et al. (2000) show that the wellformedness of nonsense words is a function of the frequencies of their parts. For example, the well-formedness rating assigned to the string [bnIk] by native speakers is a function of the frequency of [bn] as an onset multiplied by the frequency of [Ik] as a rhyme. This is a serious challenge to orthodox linguistic theory, which maintains that the well-formedness of a form correlates with whether the form violates one of the rules or constraints of the language. In the case of [bnIk], its ill-formedness should stem from the fact that [bn] violates a constraint in English on the structure of onsets, ruling out stop-nasal onsets. In this paper, I propose a revision of Optimality Theory whereby the ranking of constraints is probabilistic. Thus every pair of constraints, A and B, can be assigned a probability that the first constraint A outranks the second constraint B. Thus P(A»B) = n, n ≥ 0, n £ 1. This provides for a natural expression of the generalization above: the well-formedness of a nonsense form follows from the combined likelihood of rerankings of markedness constraints referring to the various parts of the form. This is an important result because it shows how a traditionally “functionalist” notion like frequency can be accommodated in a “formalist” approach like Optimality Theory. The organization of this paper is as follows. First, I review the structure of Optimality Theory. Second, the evidence that the well-formedness of a nonsense word is a function of the probabilities of its prosodic constituents is considered. Third, I outline the theory of probabilistic ranking proposed by Boersma (1997). Fourth, I develop an alternative conceptualization of probabilistic ranking and develop some of its mathematical properties. Fifth, I show how probabilistic ranking automatically accounts for the nonsense word effect. Finally, I show how the results of Anttila (1995) follow automatically on the formalism developed.

320 Michael Hammond

2.

Optimality Theory

Optimality Theory (Prince & Smolensky 1993, McCarthy & Prince 1993b; henceforth OT) maintains that phonological constraints interact solely via the mechanism of strict ranking. A number of other claims are typically associated with the theory as well. First, phonological derivations proceed in a single step. Second, phonological generalizations are enforced by constraints. Third, those constraints are universal; all linguistically significant typological variation is accounted for by language-specific differences in constraint ranking. Consider the set of possible syllable-initial consonants in a language. English, for example, allows certain consonants syllable-initially, but not others, e.g. [s] and [k] as in sick and kick, but not []] viz. *[]Ik]. In addition, certain sequences are allowed, but not others, e.g., [sk] as in ski, but not [ks] viz. *[ksi]. These sorts of generalizations are treated in terms of cross-linguistic variation in constraint ranking. The usual story goes as follows. Assume there are constraints against every possible consonant as an onset: *Onset/[k], *Onset/[s], *Onset/[]], etc. Assume in addition that there are constraints against altering a phonological input: Faith.1 Generalizations about onset structure are then captured by the ranking of the Onset family of constraints with respect to the Faith constraint.2 Hence, the fact that [t] is a legal onset consonant of English follows from the fact that the Faith constraint outranks the *Onset/[t] constraint. The fact that []] is not a legal onset of English follows from the fact that *Onset/[]] outranks Faith. (1) *Onset/[]] » Faith » *Onset/[t] Imagine an input form /tæp/, which can be pronounced with the onset intact, e.g., [tæp], or in some other way that avoids [t] as an onset, e.g., [æp], [sæp], [æt.æp], etc. Ranking Faith above *Onset/[t] means that violations of Faith are worse than violations of *Onset/[t]. Hence, it’s better to pronounce the form as is. Now consider a hypothetical input form /]æp/. Again, this form could be pronounced as is: []æp], violating *Onset/[]]. Alternatively, some other pronunciation where []] is not pronounced in the onset could obtain: [æp], [næp], [æ].æp], etc. Since Faith is outranked by *Onset/[]], violations of the latter are worse than violations of the former. Hence, any pronunciation without []] in onset position is preferred to []æp].3 These relationships are normally depicted in a constraint tableau. In this representation, the input form is given in the upper lefthand corner. Constraints are given in the order of their ranking along the top of the diagram and competing candidate representations along the lefthand side. Consider the constraint tableau for /]æp/ below.

Phonotactics and probabilistic ranking

(2) /]æp/ []æp]

*Onset/[]]

Faith

*Onset/[t]

*!

[æp] etc.

*

Here, the winning candidate is designated with the pointing finger and constraint violations are indicated with asterisks. The effects of strict ranking are indicated with the exclamation point and cell shading. The /]æp/ candidate violates the highest-ranking constraint *Onset/[]], a fatal violation since this is the highest constraint and there is at least one other candidate that does not earn a similar violation. This is indicated with the exclamation point. At this point, the remaining constraints are irrelevant and the second candidate wins. That the other constraints are irrelevant is indicated with shading. Compare this with the tableau for /tæp/ below. (3) /tæp/

*Onset/[]]

Faith

[tæp] [æp] etc.

*Onset/[t] *

*!

Here, Faith outranks *Onset/[t]. Therefore the first candidate wins because it escapes the Faith violation which the second candidate earns. The fact that the first candidate violates *Onset/[t] is irrelevant, because the issue is decided by the higher-ranking Faith constraint. The basic idea then is that all cross-linguistic variation should be describable in terms of variation in constraint ranking. All else being equal, for the three constraints considered here, we would then expect all rankings to be possible. The number of possible rankings for n constraints is n!.4 In the case at hand, we should expect 3! = 6 different rankings. (4) 1. 2. 3. 4. 5. 6.

*Onset/[t] » Faith » *Onset/[]] *Onset/[t] » *Onset/[]] » Faith Faith » *Onset/[]] » *Onset/[t] Faith » *Onset/[t] » *Onset/[]] *Onset/[]] » Faith » *Onset/[t] *Onset/[]] » *Onset/[t] » Faith

There are two ways in which the factorial typology is constrained. First, cases where the two markedness constraints are adjacent in the ranking are indistinguishable; the ranking of the markedness constraints only makes a difference with respect to Faith.5 The six rankings above therefore reduce to four.

321

322 Michael Hammond

(5) 1. 2. 3. 4.

*Onset/[t] » Faith » *Onset/[]] *Onset/[]] » Faith » *Onset/[t] {*Onset/[t], *Onset/[]]} » Faith Faith » {*Onset/[t], *Onset/[]]}

Second, certain constraints are universally ranked. For example, the *Onset/[]] constraint is universally ranked over the *Onset/[t] constraint. This follows from the Sonority Hierarchy, which holds that more sonorant onsets are dispreferred.6 (6) 1. Faith » *Onset/[]] » *Onset/[t] 2. *Onset/[]] » Faith » *Onset/[t] 3. *Onset/[]] » *Onset/[t] » Faith The claim of the model is that there should be three kinds of languages distinguished in terms of how they treat these consonants. In the first type, any sort of onset consonant is allowed, e.g., Cantonese Chinese. In the second type, [t] is allowed as an onset, but []] is not, e.g., English. Finally, the third type of language precludes onsets of either type.

3.

Experiments

Coleman & Pierrehumbert argue that the grammaticality of a nonsense form is a function not of the phonological well-formedness of that form, but of the frequency of its parts. Thus, a nonsense word like [blIk] is a function of the frequency or probability of the onset [bl] and the frequency or probability of the rhyme [Ik]. They establish this by performing an experiment using nonsense words. They presented their nonsense forms to naive subjects and collected the judgments of those subjects.7 Specifically, subjects were asked to grade forms on a seven-step scale for how well-formed they were. Coleman & Pierrehumbert show that those judgments correlate with the probability of a form in terms of three measures: overall probability, log probability, and worst-part probability. All of these measures are calculated by first considering the probabilities of individual prosodic constituents. This is a straightforward process whereby we consider how many times some prosodic constituent occurs in some sufficiently large sample.8 For example, if we are interested in how many times [f] occurs as a monosyllabic onset in English, we first find out how many monosyllabic words occur in some reasonable sample of English text or speech. Call this number N. We then determine how many of those words begin with [f]. Call this number n. The probability of [f] as a monosyllabic onset is then n/N. We’ll write this equation as follows. (7)

P(Ons/[f]) = n/N

Phonotactics and probabilistic ranking 323

Overall probability is calculated by multiplying the probabilities of the prosodic constituents of a form together. On this measure, the well-formedness of [blIk] correlates with P(Ons/[bl]) × P(Rhm/[Ik]). Log probability is calculated by taking the natural logarithm of the overall probability of a form. On this measure, the wellformedness of [blIk] correlates with log(P(Ons/[bl]) × P(Rhm/[Ik])).9 Finally, worst-part probability is simply the probability of the least frequent prosodic constituent of the form. The fact that all three correlations hold at the same time should not be surprising as all three measures make use of overlapping information. Coleman & Pierrehumbert quite rightly point out that the correlations with overall and log probability are a problem for orthodox phonological theory which holds that the grammaticality of a form is i) a function of the grammaticality of its components, and ii) that there is no role for frequency/probability in the model. One might object that the fact that there is a correlation with these probabilities doesn’t show that the grammar must be constructed on any sort of probabilistic basis. For example, one might counter that these probabilistic factors play a role in the historical evolution of a grammar, but not necessarily in the structure of the grammar itself. This is certainly true. The model I propose in the next section uses probabilistic weightings that are derived from the acquisition process. In addition, however, the model itself incorporates these probabilities indirectly in the ranking of constraints.10 Another possible objection is that such an experiment should have no bearing on phonological theory. After all, the task that Coleman & Pierrehumbert use is not really the same thing as asking subjects if a form is grammatical. One might argue that while traditional grammaticality judgments can be modeled by phonological theory, the kinds of judgments that Coleman & Pierrehumbert collected should be accounted for by some other theory. Frisch et al.’s study addresses this question directly. They show that the same correlations with probability hold even when the nonsense word judgments solicited from subjects are the usual ones, where subjects are simply asked whether a form is grammatical. The preceding objection cannot be sustained because the same effects obtain in the most orthodox of linguistic tasks. There is, then, a problem here for phonological theory.

4. Probabilistic ranking The effects presented above can be captured by augmenting OT with probabilistic ranking. In this section, I describe Boersma’s (1997) approach and the problems it faces. I present an alternative in the following section.11 Boersma proposes that constraints are ranked along a continuous numerical scale. Thus, for example, if constraint A has a ranking value of n and constraint B

324 Michael Hammond

has a ranking value of m, where mn, then constraint A is said to outrank constraint B. All constraints are associated with some ranking value between 0 and 100. Probabilistic ranking arises from the fact that the actual ranking of a constraint varies around its specified ranking. This variation — or disharmony — is described by the following equation: (8) disharmony = r + s × z The term r is the numerical ranking of the constraint. The s term refers to the ‘ranking spreading’ of constraints, or how much noise there is in the system. Finally, z is a Gaussian random variable with a mean of 0 and a standard deviation of 1. A random variable with a Gaussian distribution will take on values as in Figure 1.12 0.4 0.3 0.2 0.1

–3 –2 –1

0

1

2

3

Figure 1.A Gaussian random variable.

Given two constraints A and B, with ranking values r1 and r2, the distribution of their difference is given as: (9) disharmony1 − disharmony2 = r1 − r2 + s × (z1 − z2) Computing the probability that A and B will exhibit some particular ranking with this model involves a little calculus. This is given as:

 r −r (10) P(disharmony > disharmony ) = 1/ 2×(1 − erf  1 2 x 1 2 1 2  2  sx 2 (Boersma 1997; p. 331).13

  ) 

Loosely, the equation calculates the likelihood that the difference between the absolute rankings of the two constraints (r1 and r2) is small enough — and the noise in the system (s) is large enough — that the constraints can be reranked. Basically,


this equation tells us what the chances are that two overlapping distributions like in Figure 2 will result in a constraint inversion.

82

83

84

85

86

87

Figure 2.Overlapping Constraints.

There are several problems with this approach. First, the approach predicts that it should be impossible to get categorical rankings. That is it should be impossible to find two constraints that are always ranked in some specified order. The reason for this is that a normal curve never really reaches 0; it extends off to either side of its peak infinitely, approaching, but never reaching 0. Boersma & Hayes (2001) address this issue, noting that for two constraints sufficiently far apart, the likelihood of a reranking is astronomically small. While this is true, the theory does make such distinctions and Boersma & Hayes’s claim would entail that those distinctions have no empirical consequences. A second related problem is that the theory predicts a difference between ranking relationships that are close and ranking relationships that are distant, e.g., if the ranking of some constraint A is 10 and the ranking of some constraint B is 90 or 100. Assuming that all rankings have a normal distribution, there should be an empirical difference here, but as Boersma & Hayes note, it is far too small to be testable. Finally, a third problem is a theoretical one. The model requires a certain amount of mathematical complexity. The model developed in the following section gets by with a much simpler — much more elegant — mathematical model, however.

5.

Probabilistic phonology

In this section, I outline an alternative formalization of variable ranking. This formalization is conceptually and mathematically simpler, and will be easier to deal with in the following sections.14

326 Michael Hammond

Let’s assume that the constraints of Universal Grammar have a specific and total ordering in any language we might look at. (11) In any language, all constraints are completely ordered. This entails several things. First, empirically and theoretically, it is impossible for constraints to “tie”. Any two constraints must be ranked in one way or the other.15 This can be formalized as follows: (12) "A"B(A » B ⁄ B » A) This, of course, leaves open the possibility that there may be situations where we simply don’t know what the ranking of two constraints is. Second, this entails that if two constraints A and B are not ranked A » B, then they must be ranked B » A, and vice versa: if they are ranked A » B, then they are not ranked B » A. (13) "A"B(A » B ´ ¬(B » A)) We can use these to express the full ranking of a language. Every pair of constraints will exhibit one of two ranking relationships. For example, if Universal Grammar contained only the constraints A, B, C, D, and in the language in question, these were ranked A » B » C » D, then this can be expressed (more cumbersomely) as follows. (14) A » B, A » C, A » D, B » C, B » D, C » D Ranking relationships are normally understood transitively. To capture this aspect of a statement like A » B » C » D, we must add a principle to this effect. If A outranks B, and B outranks C, then A outranks C: (15). (15) "A"B"C((A » B) Ÿ (B » C) Æ A » C) Let’s now consider how to add probability to the mix. The basic idea is that all ranking statements have a probability associated with them. (16) "A"B(P(A » B) = n, n ≥ 0, n £ 1) For example, if there is a 25% chance that A » B, then P(A » B) = .25. Because of principle (12), it follows that there is a 75% chance of B » A: P(B » A) = .75. We can recast (12) in probabilistic terms as follows: (17) (P(A » B) = n) ´ (P(B » A) = 1 − n) If a ranking is certain, then it has a probability of 1. If a ranking is impossible, it has a probability of 0. This results in a somewhat different characterization of the total ranking of a language. In probabilistic terms, A » B » C » D can be expressed (even more cumbersomely) as follows.


(18) P(A » B) = 1 P(A » C) = 1 P(A » D) = 1 P(B » C) = 1 P(B » D) = 1 P(C » D) = 1 The following statements can be added by principle (17). (19) P(B » A) = 0 P(C » A) = 0 P(D » A) = 0 P(C » B) = 0 P(D » B) = 0 P(D » C) = 0 The effect of adding probability is that a grammar may not exhibit a single ranking, but rather a set of rankings that altogether exhibit a probability distribution, i.e. the sum of the probabilities of all possible rankings for a language will equal 1. Finally, let’s consider how transitivity (15) is accommodated in this system. It turns out that there are several versions of transitivity to consider. The most limited kind is trivial. We can show that the probability of A » B » C is the product of the probabilities of A » B and B » C. (20) (P(A » B) = n) Ÿ (P(B » C) = m) Æ P(A » B » C) = mn This follows from a general principle of probability theory whereby the probability of two independent events co-occurring is the product of their independent probabilities. It is rather more difficult to calculate the probability of A » C directly. To do this, we must decompose the probabilities of A » B and B » C further. Thus, the probability of A » B is really the probability that one of three rankings might hold: A » B » C, A » C » B, or C » A » B. If these three rankings have the respective probabilities x, y, and z, and A » B has the probability n, then by normal laws of probability — since these are not independent events — n = x + y + z. (21) P(A » B) = P(A » B » C) + P(A » C » B) + P(C » A » B) The second antecedent to the transitivity statement can be similarly decomposed: (22) P(B » C) =P(A » B » C) + P(B » A » C) + P(B » C » A) Given these equivalences, the probability of A » C is: (23) P(A » C) =P(B » A » C) + P(A » B » C) + P(A » C » B)

328 Michael Hammond

6. Deriving the effects The preceding two sections have introduced a fair amount of math. In this section, we show some of the payoff of this math: the effects described by Coleman & Pierrehumbert and by Frisch et al. can be derived from probabilistic ranking of constraints. The key observation from the experimental work cited is that the wellformedness of a nonsense form is a function of its parts. Let’s look at a hypothetical example to try to understand how this works. The values used here are completely made up. Consider a nonsense form [bnIk]. Let’s assume that P(*Onset/[bn] » Faith) = .9 and that P(*Rhyme/[Ik] » Faith) = .2 What we’re interested in is what the chances are that either or both of these constraints outrank Faith. The equation here is exactly the same one you’d use if you wanted to figure out what the chances are of throwing at least one heads given two coins. Assuming A and B are independent events, probability theory tells us that the probability of one or both happening is P(A) + P(B)− (P(A) × P(B)). This comes out to .75 for the coin example. The probability of throwing heads with one coin is .5, the probability of throwing heads with the other coin is also .5, and the probability of throwing two heads is .25. Hence .5 + .5 − (.5 × .5) = .75. Graphically: (24) coin 1 tails heads tails heads

coin 2 tails tails heads heads

In 3 out of 4 possible cases, you can throw at least one heads. Returning to the syllable structure example, it follows that the probability of illformedness for [bnIk] is .92. That is, [bnIk] is ill-formed just in case one or both of the syllable structure constraints outranks Faith (.9 + .2 − (.9 × .2)). Thus, on the model developed here, the fact that judgments correlate with multiple properties of the nonce forms follows from the fact that there are multiple markedness constraints and any of those constraints can outrank Faith.16 The machinery above gives us i) gradience, and ii) the fact that probabilities “add up”. It doesn’t give us an account of why constraint rankings should correlate in any way with the frequency of occurrence of some pattern. How do we account for the fact that the severity of a constraint correlates inversely with the frequency of the pattern? Assume — following orthodox OT — that there are constraints against every possible sequence of consonants as an onset.17 To get this, let’s assume the following:


(25) Constraint Reranking: Whenever a speaker hears a form that would violate some markedness constraint in his/her language, that constraint is reranked probabilistically with respect to the highest ranked Faith constraint. On this view, the probability that some constraint might outrank Faith is a function of how frequently a speaker hears forms that might violate that constraint.19 The fact that [phr] is a better onset in English than [vj] would then follow from the fact that English speakers hear the former more often than the latter. Hence: P(*Onset/[vj] » Faith) > P(*Onset/[phr] » Faith).19 Note that the principle above has several important consequences. First, the judgments that speakers produce on some set of nonsense forms are not a simple function of the set of words in that speaker’s lexicon or grammar. They are also a function of the relative frequency of those words. Second, this means that two superficially identical speakers of a language might exhibit different judgments by virtue of their differing experience.20 Finally, it is important to point out that the model does not reduce phonology to word learning. The model subscribes to the usual nativist position in OT that the constraints are finite and universal. Hence, while word frequency can dictate all sorts of possible rankings, the grammars that result are still defined by the universals of human language. Put more bluntly, frequency effects are confined to the possible rankings of constraints. The set of possible constraints itself is given by Universal Grammar. Another possible position would be that the set of constraints isn’t given by Universal Grammar per se, but by phonetic experience. If one adopts such a view then probabilistic ranking still does not reduce to word learning. Word learning still only gives the possible rankings; the set of available constraints still comes from a different domain, in this case, phonetic experience.

7.

Combination of constraints

There is an additional advantage of the approach developed here: we can also derive the results developed in Anttila (1995). Anttila shows how certain cases of variation can be described in terms of variable constraint ranking. Anttila’s analysis treats cases of variation in the Finnish inflectional system. He shows that the rate of occurrence of any form correlate with the number of rankings that produce that form as an optimal output. This follows directly on the approach developed here. Consider a hypothetical case where *Onset/[bn] can be variably ranked with respect to MaxIO-[b] and MaxIO-[n]. (These constraints are instances of Faith constraints; they are the ones that penalize outputs where the relevant segment or

330 Michael Hammond

segment class does not surface.) There are 6 possible rankings of these constraints. For a hypothetical input form [bnIk], considering only the outputs [bnIk], [nIk], and [bIk], these different rankings select the following outputs. (26) *Onset/[bn] » MaxIO-[b] » MaxIO-[n]: *Onset/[bn] » MaxIO-[n] » MaxIO-[b]: MaxIO-[b] » MaxIO-[n] » *Onset/[bn]: MaxIO-[b] » *Onset/[bn] » MaxIO-[n]: MaxIO-[n] » MaxIO-[b] » *Onset/[bn]: MaxIO-[n] » *Onset/[bn] » MaxIO-[b]:

[bIk] [nIk] [bnIk] [bIk] [bnIk] [nIk]

Assuming all rankings are equally probable, then each output should occur 1/3 of the time, e.g., P([bIk]) = .33, etc. On the approach developed here, this follows automatically. If, in the case at hand, each ranking has an equally likely chance of occurring, then each occurs with P = .166 (i.e. each has a 1/6 chance of occurring). Since the rankings are mutually exclusive (and not independent), the chance of any two of them occurring is the sum of their respective probabilities: 1/6 + 1/6. Hence P([bIk]) = .33.

8. Conclusion In this paper, I’ve argued for a particular formalization of probabilistic ranking in Optimality Theory. In particular, I’ve proposed that every pair of constraints can be joined by a ranking statement and that that statement can be associated with a probability value. The normal mathematics of probability theory then apply. This system is simpler than the alternative developed by Boersma & Hayes (2001). We’ve used this system to establish two important results. First, the experimental result whereby the well-formedness of a nonsense form is a function of its prosodic constituents follows. The ungrammaticality of a form is the product of the independent likelihoods that any markedness constraint against the components of the form might outrank the highest faithfulness constraint in the system. Second, we’ve established that Anttila’s result that the frequency of occurrence of a form correlates with the number of rankings that result in that form being optimal also follows from the theory. A number of questions remain unanswered, but I hope to have established the initial plausibility of this approach to phonotactic frequency. We’ve seen that an explicit and formal approach is fully capable of treating these sorts of facts. Rather than comprising a threat to phonological theory, the experimental results presented are amenable to direct analysis under the approach developed here.

Phonotactics and probabilistic ranking

Notes * Thanks to Rachel Hayes, Terry Langendoen, Diane Ohala, Natasha Warner, two anonymous reviewers, and the editors for useful discussion. All errors are my own. 1. There are actually a number of ways to alter a phonological input and a number of constraints against those different manipulations. It is convenient to begin with the simplification that there is a single faithfulness constraint. 2. The basic story here is due to Prince & Smolensky (1993). 3. Precisely which pronunciation is preferred depends on the ranking of the set of constraints we’ve collapsed under Faith. 4. The value of n! is n × n−1 × … × 1. 5. This is not a general fact about markedness constraints, but only about markedness constraints which can’t apply to a single representation. 6. See Ohala (1996) and Hammond (2000b) for more discussion. 7. The authors only considered nonsense words in their experiment; no actual words were tested. 8. It’s not at all obvious how large such a sample should be. The relevant statistical question is how big a sample of a language is necessary for it to be “representative” of the language. Another question is why we should compute these probabilities over actual prosodic units, rather than other strings of segments. For example, the well-formedness of a nonsense form like [bnIk] might have more to do with the probability of [bnI] and [k], or of [b], [n], [I], and [k]. Likewise, we might entertain the possibility of computing these probabilities over discontinuous spans. I leave these questions for future research. 9. Basically, log probability scales the numbers differently, keeping the overall ordering intact. 10. Thanks to Andrew Carnie for useful discussion here. 11. Some of the ideas of probabilistic ranking are anticipated in pre-OT work by Goldsmith (1993). 12. Another way of saying this is that the distribution of such a variable is “normal”. 13. “erf” stands for error function. 14. This section proposes asimple logical formalization of OT. For a more detailed proposal about the logic of OT, see Hammond (2000a). 15. As an empirical matter, this may be impossible to verify completely. What of the set of constraints that have no effect in some language? Presumably, these are ranked so low that their violations have no effect. How are they ranked with respect to each other? The principle given demands that they be ranked, but does not specify what this ranking is or how it is arrived at. 16. Notice, incidentally, that this claim does not change if one replaces Faith with multiple faithfulness constraints. The demonstration in the text proceeds with respect to the highest ranked faithfulness constraint, whatever that may be. 17. See Hammond (2000b) for an example of this. 18. This result is achieved directly under the Gradual Learning Algorithm of Boersma & Hayes (2001). 19. That constraint ranking should correlate with frequency of occurrence is a claim made in earlier work. Hammond (1999) argues that the likelihood of the Rhythm Rule in phrases with disyllabic modifiers, like thírtèen mén, is a partial function of the lexical frequency of the modifier.

331

332 Michael Hammond

20. In fact, one can imagine an experimental paradigm where a subject’s linguistic experience is manipulated directly, e.g., by repetition of items with particular phonotactic properties, and then tested along the lines discussed in the text.

Deconstructing functionalist explanations of linguistic universals* Thomas G. Bever University of Arizona Still glides the Stream, and shall forever glide; The Form remains, the Function never dies. (Wordsworth, 1820)

1.

Introduction

The more we learn about what syntax is, the less we know about why it is the way it is. “Functionalist” interpretations offer potential explanations for the peculiarities of syntactic structure and characteristically blend what they are with why they exist. In this paper, I review a few examples of such explanations, with several morals in mind. First, it is critical that functional explanations of linguistic structures have independent theoretical and empirical support; second, functional explanations can expand beyond the traditional applications, which are most often based on the communicative needs and semantic habits of speakers. Third, we may be quite surprised to realize that what starts out as a cognitive explanation of a linguistic property may reverse, and show us how a linguistic property might explain the cognitive property. Functionalism has become a controversial topic in linguistic theory (see e.g., Newmeyer 1998, for an excellent review). Around it swirl disputes over whether grammar is a formally independent mental structure, and if so, whether it is computationally arbitrary. The functionalist answers to such questions are characteristically, (1) that grammatical structures either are an ephemeron, or (2) that they are caused and shaped historically by the functional role of language. This controversy has been attached to larger issues in cognitive science, such as the notion that the mind is organized in modules — if grammar is independent of other systems, it is a quintessential example of a module. If it is interdependent and not selfdefined, then one of the strongest prima facie cases of modularity falls aside. I start with some general consideration of functionalist explanations, drawn from examination of a very simple communication system. Then, I review three

334 Thomas G. Bever

cases, which exemplify functional constraints on linguistic structures of different kinds; one from language perception, one from the biological underpinnings of language ability, and one from language learning. All three involve an aspect of how sentences are composed of ordered sequences that express hierarchical structures. In each case, I will question whether the particular linguistic property is the result of extra-linguistic constraints, or conversely, whether the computational requirement of mapping serial form and hierarchical structure unique to language itself underlies more general constraints on human cognition. My ambitious goal is to convince even the most dogmatic nativist formalist linguo-centrist linguist that eclectic consideration of functionalist interpretations of linguistic universals can be illuminating in surprising and helpful ways.

2.

Meaning and language go to the dogs

Language has a number of apparent functions. These include communication of meanings to others and representation of meanings to oneself. These roles set constraints on the nature of language and the useful structure of meanings. Which is logically prior, meaning or language? We ask this because we want to have some idea which kind of system is setting the groundwork constraints. This question is not a complete “chicken-and-egg” conundrum; we do have some evidence about communication which suggests that potentially shared meanings are independent of language. An example of this is how what I can communicate with my poodle, Cassie, determines our mutual language (Limber 1996). Cassie and I don’t have much to say to each other about most of the world, we simply view it together: her primary fascination is other dogs, lesser animals, and the smells they have left behind, all of little interest to me. She clearly has theories about the world that enter some domains I can only dimly sense (e.g., visual aspect of other dogs), and others I cannot sense at all (myriad smells in the air, grass, trees). Conversely, her perceptual system limits her ability to discriminate human artifacts — even though it would help her greatly, for example, in getting into the right car in a parking lot. But we do have certain fundamental concepts that we share: water, food, really good food, walk, car, stay, sit, down, (find) Addie (a puppyhood co-puppy), bone, puppy, (find) LouAnn (a former co-owner), (get) down, (get) up, bark, (go) out, (goto) my-office, (go) in, (get) off, (go in) house, (get) on, (getthe) ball, no, ok. This may seem like a lot for a non-circus dog, but one must remember that she is a poodle. The questions at hand are: is our communication system a language? How is it constrained by our mutual meaning systems? I will take the unconventional position (for linguists), that Cassie and I do share a language, albeit one that is so heavily constrained by minimal cognitive overlap, that it remains very simple. The

Deconstructing functionalist explanations of linguistic universals 335

fundamental fact is that we do communicate effectively about those things that we actually share semantically. If we shared more semantically, I see no principled upper limit on the number of different “words” we could use. If, for example, I could differentiate smells the way Cassie can, this would enrich our discussions enormously, at least from her standpoint. The same would be true if she could differentiate cars — our life in parking lots would be a great deal more communicative and easier, at least from my standpoint. In the long theoretical run, it will matter a great deal what notion of “semantics” Cassie and I share. One might argue that it is not “intensions” or propositional attitudes, but only extensions, a.k.a. referents. The fact that we share variable but appropriate behaviors in relation to each concept suggests some kind of shared propositional representation. But I will not argue that here, because it certainly requires a lot of demonstration. The essential point is that whatever the true nature of our shared concepts, they are the only things that we can convincingly communicate about. Of course, it may be that our communication system is ALSO constrained by Cassie’s (and my) computational limitations. In general, she does not appear to respond to combinations of words, and so does not exhibit much of what we usually call “syntax”, the system that governs how word elaborations and combinations are formed. An exception is the negative operator, “no”, which can be combined with actions or expectations she has. She can be told to stop barking, or that she is NOT going in the car, or that she is NOT going to get a treat. A critical (informal) experiment on the productivity of “no” involves a neutral action, such as “sit”: she will refrain from sitting, if “sit” and “no” are presented together. (Though she does not look happy about the situation.) Thus, “no” does not merely mean to her that something bad will happen. But, it is not clear that these are syntactic combinations bound into larger units of meaning, as opposed to a serial presentation of information in which each element modifies what it is adjacent to. In practical terms, does word order affect Cassie’s interpretation of “no”? In fact, it does: in Cassie’s “syntax”, “no” has to follow what it negates, not precede it: thus, if I say, “no, bark”, she responds to “no” by looking guilty, but continuing to bark. If I say, “bark, no”, then she stops barking. If I say “no, car” she clearly expects (after briefly looking crestfallen) a ride in the car: but, if I say “car, no”, she gets the disappointing picture right away. Thus, she evidences understanding of the frame, “X, no” with enough statistical regularity to satisfy those experimental psychologists who allegedly teach dolphins, sea-lions and chimps to talk. So, we might surmise that Cassie has a syntactic rule: “no” must follow what it modifies. Now the question is: can we explain Cassie’s syntactic constraint that “no” follows its argument by reference to another aspect of Cassie’s cognition. One can surmise that “no” needs an already established mental representation to apply to, so it works out best if I elicit a representation in her before she hears the negation that

336 Thomas G. Bever

applies to it. (An alternative interpretation is that dogs use Polish notation (in which operators follow their arguments), or at least poodles do.) In this sense, she is not combining “no” in a syntactic frame at all; rather, she is responding serially to representations as she hears them. On this view, Cassie cannot hold operators such as negation in mind without something to operate on, but objects or actionideas can be held in mind on their own. This is not an unintuitive idea, but, to go through in a profound way, we need other evidence that dogs (or poodles, anyway) think serially and allow the contents of a current representation to modify a prior one. However, let us assume that such evidence is forthcoming (certainly, most dog owners would attest to it). Then we would have an example of a functional explanation of what might otherwise be taken to be a syntactic process. Cassie has to hear “no” after the idea it applies to, because she could not hold it in mind if it preceded the idea it applies to. Further evidence for this kind of explanation comes from Cassie’s treatment of the general “find” instruction, “where is X?” In this case, the operator must precede the argument. If I say, “puppy, where?”, Cassie merely looks puzzled; but if I say, “where puppy?”, she goes to the window, looks out and barks. This pattern, too, generalizes across goals. If we think of this in terms of the underlying semantic operations, we can make sense of why it is different from the negative operator. The “where is” or “find” instruction can be acted on before hearing the argument: indeed, if I say just “Where?…” Cassie looks alert and may start running around, waiting for the orienting information. Thus, “where is” has an immediate enactable referent. Another way of putting this is that “where” is a modal modification of the entire utterance, as opposed to an operator directly on the argument. Here too, rather than postulating a syntactic frame, we can interpret Cassie’s understanding in terms of how the utterance functions. The reader may think all of this is a reductio ad canem, on the topic of functional explanations of ‘syntax’, and there may be something to that. I am personally convinced of what I claim about Cassie, but I would agree that it requires more scientific justification and breadth to be taken as serious Dog Science. We can, however, derive some principled morals on how to think about functionalist explanations for syntactic processes. If one is uncomfortable with my reasoning above, it could be because I did not develop an independent demonstration of the idea that Cassie cannot hold a semantic operator in mind, but must apply it to an already established representation. That of course, can be tested experimentally, for example with directional or negation operators presented before the objects to which they apply in a standard animal learning paradigm. The same argument and response goes for the locative imperative. Until the principles are verified in general, it rests on intuition alone, and the appearance of being post hoc. I also did not demonstrate the natural priority of shared semantic concepts in what we communicate about, I merely asserted it. And, of course, I did not prove that Cassie’s sensitivity to word


order is compositionally syntactic in any sense except superficially. So, to be fully satisfactory, a functionalist explanation is most convincing when it has the following characteristics: a. The property to be explained is truly syntactic. b. There is independent evidence for the explanatory functional principle beyond its effect on language. Most important, the result of this consideration is to make clear that a functional explanation may not be adequate when merely referring to “meaning” as determining form. In this case, the putative functional explanation for apparent constraints on form, have to do with how the meaning is processed, not just what it is. But there is an even more profound question. I started with the assumption that Cassie and I share certain meanings in our mutual cognition, and that this limits and focuses the kind of language we share. However, how can we be sure that the causal link is not in the reverse direction? That is, is our apparent shared cognition itself shaped by our shared symbolic system? The ultimate answer to this will depend on far more serious study of canine cognition, and more success in understanding human cognition that we currently have. At the moment, we do not have a strong theoretical basis to justify the intuition that dogs and humans start with shared cognition and build language around it. And, indeed, it may turn out to be the other way.

3.

Case studies in human/human language

Let’s see how these principles guide us in considering a range of alleged functional explanations of syntactic universals. 3.1 Me first! The order of thought, and the thought of order It is often noted that languages tend to present agents prior to patients. Some languages appear to require this grammatically under certain circumstances. Recently, Bresnan and Aissen (2002) and Bresnan et al. (2001) have attempted to integrate a constraint within a stochastic optimality-theoretic framework. On this view, Agent Æ Patient is a linguistic principle, embedded as a constraint, which will appear as grammatical in certain languages, if it is ranked high enough in the overall ranking of constraints. If it is not ranked very high, it can still have occasional effects, because of a probabilistic penumbra, which occasionally elevates it to a high rank. This treatment appears to bridge with one principle, the fact that speakers tend to prefer sentences with Agent Æ Patient order and the generally observed fact that if a language grammaticizes order, that will be the order required.

338 Thomas G. Bever

Is this truly a linguistic principle, i.e., a universally present formal constraint on possible languages, or can it be explained by reference to some aspect of how language functions and is learned? In the middle of the last century, a noted pioneer cognitive psychologist, Jerome Bruner observed that languages seem to prefer sentence structures in which the “agent” precedes the “object”. He suggested that this grows out of early language learning, especially language production: children have an idea, and they express it in serial parts that correspond to the temporal order of the idea’s conception. Bruner reasoned that the child starts with sentences about his or her own plans, desires, etc. The child realizes that its own intentions precede its actions. Through this, the child discovers intentions more generally, and the knowledge that they definitionally precede their corresponding actions. Agents must be animate, verbs tend to be activated intentionally and objects tend to be inanimate: thus, agents will tend to appear before objects by virtue of the serial stages of the underlying thought processes. This is an appealing idea, but it did not have independent empirical support. However, it has the virtue of making some predictions about when the Agent Æ Patient order might be less optimal or violated, and when it would be the strongest. In particular, when the Agent is animate and the patient is inanimate, the restriction should be weakest: intermediate when both are animate or both are inanimate, and strongest when the Agent is inanimate and the Patient is animate. Furthermore, since intentions are discovered first by hypothesis, it is reasonable to expect that if there are cases in which animates are ordered in relation to each other, 1st person will have the highest rank: since utterances are generally to a second person, that will be the next highest rank, with third (often non-present) person the lowest rank. Of course, we would be advised to have some independent empirical support for Bruner’s thesis, intuitively appealing as it is. In fact, some of the underlying assumptions about the hypothesis do not have support. In particular, Bruner assumed that the Agent Æ Patient order was fundamental to the earliest stages of language, since it was based on a theory of how the infant thinks. This predicts that the order should appear in the earliest stages of language use, and build up from there. There is some evidence, going back to Lois Bloom (1970), that early utterances tend to maintain Agent Æ Verb, and Verb–Patient order. But of course, in English, that is the canonical order anyway, so it is ambiguous as to whether the child’s basis for the order is the order of thought, or what s/he is hearing. A different way to approach the development of the order is to examine children’s comprehension strategies. In a series of studies, we did just that (Bever, Mehler, Valian and Epstein, reported in Bever, T. G. 1970). We tracked the child’s comprehension of reversible actives, passives, subject clefts and object clefts (we developed a version of the “puppet acting out” task, in which children are the puppeteers). We found something very surprising: 2 year olds comprehend all of


the sentence types better than chance, indeed they understand object-clefts as well as they understand actives! But 3–4 year olds understand passives and object clefts systematically worse than chance, while comprehension of actives and subject clefts approaches perfection. We concluded that this shows that the child learns the regularity that Agents precede Patients (in English), rather than creating it by virtue of any alleged order of thought. Similarly, we tested the development of the child’s sensitivity to semantic constraints, using sentences like “the dog ate the cookie” vs. “the cookie was eaten by the dog” etc. Even more surprising than the prior result, we found that two year old children were only slightly sensitive to the semantic constraints: they often correctly acted out anomalous sentences (e.g., by having the cookie bounce all over the dog, and announcing as one child did, “now the doggie inside cookie, that silly!”). Again it was older children who became selectively dependent on semantic reversibility. The generalization that Agents are animate, and Patients inanimate, is not available to the youngest children, it is learned as a strategy for sentence comprehension. We later tested a variety of sentence orders and inflections on languages that have a range of each (notably Italian, Serbo-Croatian and Turkish) (Slobin and Bever 1982). We found that in each language, children develop comprehension strategies that are tuned to the contingencies in their language. Most notably, this meant that in Turkish, there was little attention to word order, but great attention to the presence and location of the object inflection marker. Most important, these strategies were learned, not the initial stage of comprehension. The acquisition of such strategies is part of building up the processes of everyday sentence comprehension (and probably production). There is considerable research with adults showing that an early step in comprehension depends on such strategies, quite independent of the immediately following syntactic assignment of a derivation (Townsend and Bever 2001). Thus, in the end, the principles governing the usual relation between Agency, order and animacy are not part of the initial grammatical foundation of language. They are learned as part of learning to use language, as convenient (and probably necessary) shortcuts through syntactic computation. This does not change the validity of the claim that the principles have an extra-linguistic basis, it simply clarifies how the interaction between language behavior and the principles work. It does, however, give us empirical support to reject the claim that the principles are a direct part of the innate basis for language. Rather, on our view, they emerge out of the acquisition of the processes of language use, which automatically constrain the kinds of languages that can be learned. The acquisition of strategies also plays a critical role in a dialectical model of language acquisition (Townsend and Bever 2001). On this model, acquisition (logically) alternates between two kinds of acquisition processes: the abductive discovery of structural principles with limited domain, and the extension of that

340 Thomas G. Bever

domain by the induction of extendable regularities in that domain. In the example given, this cashes out in English as: –

–

–

The development of simple declarative transitive sentence frames, based primarily on early acquisition of the verb + patient constituent, with agent status assigned by default to the other NP. Thus this is limited to short sentences with clearly locatable arguments. The development of NVN the strategy based on the simple declarative order. This extends the domain of sentence production/understanding to longer utterances, with embeddings (themselves subject to the NVN constraint). The development of exceptions to the NVN strategy as marked morphologically or in other ways. For example, object-first clefts and passives, which are systematically misunderstood during the development of the NVN strategy, become differentiated by virtue of their exhibited structural features.

Consider the implications of this treatment for Aissen and Bresnan’s ideas about the continuity of a discourse tendency and grammaticization for Agent Æ Patient (Bresnan and Aissen, 2002, Bresnan 2001a,b). There are two aspects to their ideas. On the one hand, they recognize a tendency across many languages (which certainly has been noted before), even when the order constraint is not a grammatical one. On the other hand, they offer stochastic optimality theory as a unifier of tendencies and grammatical requirements as a function of how important the constraint is in a given language. I leave to others the critique of how well and consistently optimality theory does the job of describing tendencies and grammatical restrictions (see, e.g., Newmeyer 2002a,b). The issue here is the ontology of grammar that it represents. The virtue cited by Aissen and Bresnan is that the OT treatment is unified — one principle has statistical or grammatical effects depending on how highly it is ranked. Occam’s razor is invoked. This is fine if one has as a goal merely to describe the statistical and categorical facts with the least amount of ink. But if one aspires to propose that the grammar (or at least its architecture) is a model of human knowledge, then matters become bewildering. If I know that a given constraint has a specific frequency distribution that overlaps with other constraints, how does this cash out in my grammaticality intuitions? Why are our intuitions for specific sentences fairly constant, as opposed to stochastically wandering? Aside from the explanatory virtues of the functionalist interpretation of the ordering priority of agents, we would like to have some “new” facts explained by some of the same general principles. That is, a functionalist interpretation of a phenomenon (like any theory) does best when it leads to other phenomena. It appears that the general notion of an ordering hierarchy involving animacy, and centrality to a sentence has some currency in various ways across languages. The general idea is that more highly ranked phrases must precede less highly ranked phrases


in all sentences, regardless of their functional role: special markers occur when the patient is the more highly ranked phrase. To quote Jelinek (personal communication) “Navajo has an elaborate set of restriction on Agent/Patient order that are manifested in an Animacy Hierarchy and the famous Active/Inverse Voice alternation (the yi-/bi alternation). If the Patient is higher than the Agent on the animacy hierarchy, then it must come first, and the verb is in the Inverse voice. For example, you need to say the equivalent of The boy was kicked by the horse. using the Inverse, which is comparable to a Passive, except that it is a transitive, with two direct arguments (no “by” phrase). This kind of a hierarchy is also manifested in many Native American languages. For example, in Lummi (Salish) you can say I hit him but you can’t say he hit me you have to say I was hit (by him) using the Passive. The rule is 1st, 2nd, 3rd person, NP. This kind of Agent hierarchy is very common.”

This is part of a complex hierarchy for Navajo, which orders NPs: Adult human > child > animal > inanimate. Finally Willie (Willie and Jelinek 2000)), also noted that the hierarchy includes the constraint that topicalized definite nouns must precede indefinites. Jelinek points out that such hierarchies occur in many languages, even English has some such constraints for indirect objects and beneficiaries: only animate beneficiaries can occur prior to the benefit. (1) a. b. c. d.

He mixed a drink for his friend He mixed his friend a drink He mixed some gasoline for his racecar. *He mixed his racecar some gasoline

Put in more general terms, she writes: Despite the surface diversity [in many different languages], argument hierarchies are alike in the following two traits: 1. There is some ranking of arguments according to a scale of presuppositionality. Some subdomain of the following ranking is observed (Jelinek 2000). 1,2 > 3 > NPs > rankings within NPS according to definiteness, animacy, etc. 2. There is some syntactic manifestation of the hierarchy [which] … orders arguments in accord with the presuppositionality scale.

342 Thomas G. Bever

Finally, she sums up her overall generalization concerning ordering hierarchies: Argument hierarchies are the morphosyntactic registration of … the Presuppositionality Scale … in particular languages … [they] represent the grammaticalization of certain highly frequent discourse tendencies.

If Jelinek is right, the alignment of an intuitive order hierarchy and Agent/Patient word order is part of the grammar of certain languages, while only preferred in others. Thus, one might argue that the fact that the constraint can be grammatical shows that it is truly part of linguistic universals, as Bresnan proposes. But, it is more complicated than just requiring agents to precede patients linearly. In certain languages, agents can follow patients, but then must be marked if that results in a linear order that violates the hierarchy. In both expressions, invoking the universal would not explain why the Agency/ordering hierarchy exists, nor why it has a characteristic order. The processing hypothesis offers a true explanation of the whole phenomenon. On this view, languages could have other kinds of constraints on Agency, and other orders of Agency hierarchies: but the language learning child naturally filters out those orders and constraints that do not conform to the order and processes of attributing the presupposed parts of sentences. We can generalize this argument to the fuller order hierarchy which Jelinek summarizes. The sometimes-grammatical constraint that persons 1 > 2 > 3, animate > inanimate, definite > indefinite, are interpretable as a linguistic codification of conceptual hierarchies, which can differ in richness between cultures, but which have common themes. The surface order constraints differ from the Agent > Patient constraint, in the sense that it is the conceptual content of the phrases that governs the order rather than their functional syntactic role. However, the deeper consistency is that the priority of thought has a reflex in the precedence of the corresponding linguistic form. Jelinek’s way of characterizing the basis for the hierarchy as “presuppositional” is particularly felicitous for our explanation: basically it proposes that the more presupposed an element is, by virtue of being a default in actions, the earlier it must appear in a sentence. The hypothesis that the mapping from presupposition to linguistic constraint via discourse and perceptual strategies, leads to a rich research program. The idea is that the presuppositional hierarchy based on categories and/or discourse markers, reflects salience in conceptual access during speaking and listing. Salience can is reflected in surface order, which is exactly what happens. The mechanism for learning this in childhood is essentially automatic for conceptual categories and basic for syntactically marked presuppositionality (topic, focus, definiteness). For conceptual category based hierarchies it simply falls out of the conceptual system, apparently universal, possibly innate in an interesting sense. For syntactic categories, the idea predicts that surface order will convey topic/focus/presuppositionality by initial prominence, and secondarily special inflectional markings.


To some extent all this is testable, which is why I am writing this. First it suggests that in English, sentences that violate /presuppositional/ order will be more complex. Aside from the Agent > Patient ordering constraint, very little research bears on hierarchies Æ order principles. A few studies i.e., (Gibson, E. 1998), involve the idea that “I” and then “you/s/he” are presuppositionally built into sentences, hence reducing complexity. Other studies (Frazier, L., and Clifton, Jr., C. 1996), suggest that mapping pronouns onto discourse-presupposed elements is actually easier than mapping them onto sentential antecedents not yet in the discourse representation. So there are a number of things to look at in English: e.g., are the examples (2,3a) easier than (2,3b), and (2,3c) easier than (2,3d) (2) a. b. c. d.

The boy hurt a girl A boy hurt the girl The girl was hurt by a boy A girl was hurt by the boy

(these have to be tested with and without contexts because the use of “the” without a context is itself a bit odd). (3) a. b. c. d.

A/The boy crushed a/the rock A/The rock crushed a/the boy A/The boy was crushed by a/the rock A/The rock was crushed by a/the boy

Note that much prior research would suggest that (2,3d) should always be easier than (2,3c), since (2,3d) violates the NVN strategy, and there is some research to that effect. However, for purposes of testing the presuppositionality hierarchy it should be tested with the indefinite article which has almost never been done. Testing it with the definite article may obliterate presuppositionality at least in English, because the article is so pronounced. Furthermore, what is really at issue is the difference between (2,3a) and (2,3c) vs. that between (2,3b) and (2,3d). Finally, no one to my knowledge has contrasted sentences like (2,3a) vs. (2,3b). It is easy to generalize the paradigm to the full hierarchy e.g. (4a,b,c). (4) a. You hit him, vs. He hit you b. I hit you vs. you hit me etc. c. The nurse loves the infant vs. The infant loves the nurse (perhaps following Navajo, the differences may depend on being in the “inverse” construction, i.e. passive).That is the first kind of thing to look at in English (or Korean, Russian, Chinese or Spanish). The other aspect of the model involves acquisition of the constraints as perceptual strategies. Again, little has been done on this directly. But it is sure easy to test for those who like to have kids act out short sentences. If my hunch is correct, four year olds are at a choice point in their

344 Thomas G. Bever

treatment of agency and presuppositionality hierarchies: they can link them, so that the first item is generally taken to be the agent (as in English), while allowing odd surface orders when they are marked as in the passive: or they can impose the expected order as a surface order, regardless of the agency and use special markers when the agency is out of register with the order. The choice seems to be whether the hierarchy attaches to agency itself or to surface sequence constraints. A conservative formalist might argue that this choice is a parameter that the child must learn: if parameter setting is a useful model of acquisition, so be it (but vide infra). What I am concerned to show is that the substantive content of the constraint is possibly conditioned by the way language functions for the young language learning child. If this is supported experimentally, it will also support the interpretation of the hierarchy of order constraints as indeed due to how language functions. That is, it will stand as a clear case in which we can explain an apparently formal linguistic property as truly caused by how language behavior is assembled during cognitive development. 3.2 Up a tree! Linking elements in embedded and dominant fields Movement has been a feature of transformational grammars since Harris. Movement occurs in the service of linking sites that a particular constituent has sequentially occupied during a derivation. It is striking that when only one of the exemplars of the constituent is pronounced (the typical case) it is the least embedded, the “highest” available in the constituent tree. This is the case regardless of whether the movement itself is conceived as “up” the tree (by far the most common architecture in the last fifty years), or “down” the tree (Richards 1997, Philips 1996). Why? One can certainly imagine computational derivations in which the more embedded position is the pronounced one, which would have the same expressive power. Why does it turn out that the most salient site for a constituent that has one is the least embedded? Is there a functional explanation? My answer to this may require some initial suspension of disbelief typical of drama in general. In this case, I consider the neurological foundations of humans’ ability to discover or create language. It is well known that the brain areas devoted to language in humans evolved as visual areas, as reflected in the localization characteristics of our near simian relatives. That is, the computational properties of the “language” areas in the human brain were shaped originally for the needs of vision in apes. Suppose that the existence of movement as a mechanism within syntax recruits the mechanisms of vision originally related to the perception of real movement. Such co-opting of old functions for new ones is the watchword of evolution, and given the areas that language took over, it is not a wild speculation, only an adventurous one. We can make it less adventurous by actually exploring the


question of whether real movement between sites in a visual scene makes the less embedded site more salient. In studies of the perception of actual movement, this question translates into whether the movement is easier to perceive when it ends up in the less embedded site, compared with the reverse. Our experimental studies of this suggest that the answer is yes, movement is easier to perceive from more to less embedded parts of visual arrays (Lachter, Weidenbacher. and Bever in prep). We test this by presenting people with stimuli spaced so that they would ordinarily create illusory movement as in the classic gestalt phi phenomenon (a typical example is the perception of movement between alternating lights at a railroad crossing). We have found that perceived movement out from a square within a larger square is easier to perceive than movement in the opposite direction. I won’t give all the appropriate details and controls here, except to note that it is not the case that static visual stimuli are more salient in less embedded parts of a scene: rather, the relative salience of the less embedded position depends on there having been movement to it from a more embedded position. Real movement is easier to perceive visually from more to less embedded parts of a representation. If language has co-opted computational mechanisms originally adapted to the perception of vision, then they may determine aspects of linguistic computations that are otherwise arbitrary from the standpoint of syntax. The result is that syntactic movement, like that of a cat, is up the tree, not down the tree. A similar question arises as in the causation of order hierarchies: is the salience of the less embedded exemplar of a constituent due to principles of the behavioral organization of movement, or is that a more general mental principle which lends salience to all phenomena that are less embedded? Most specifically, is it the result of why hierarchies and movement exist at all, as suggested by Kayne (1994) and Moro (1997). In essence, their proposals are that the role of linguistic movement is to create asymmetry in otherwise symmetric strings, to insure a specific manifest serial order that reflects an hierarchical asymmetry: that is, in a binary sequence [a, b], there is no structural basis for order, but if one of them is moved to an immediately higher tree, then a mapping rule which gives precedence to the less embedded member of a tree, will result in a specific output order, thus guaranteeing recoverability of the underlying relations. On this view, it is the computational properties of language which dictate the salience of less embedded constituents. As in other cases, it is not easy to find a direct test to differentiate these hypotheses. In this case, we might turn to animals, since they can learn serial order, and study whether the order becomes fixed when a serial subset is treated as a constituent. That is, suppose we train an animal that, a–b or b–a are acceptable orders of a and b, and x–y or y–x are acceptable orders of y and x. (If it helps the reader envision how this would be done, imagine a monkey trained to point to different points on a compass rose, in a particular sequence.) At this point all the

346 Thomas G. Bever

animal has learned is that a, b and x, y are cooccurrence pairs. Now, suppose we embed x, y within a, b, so the possible sequences are: a, x, y, b a, y, x, b b, x, y, a b, y, x, a The critical question is: does the animal pick a particular order of x, y, now that it is embedded within a, b and hence a “constituent”. Of course, this simultaneously tests whether the animal can learn structurally dependent contingencies at all (some recent studies suggest that it can), and whether serial order becomes fixed as soon as a sequence becomes recognizable as a constituent. With Fraser Wilson and Nick LaMendola, I hope to be carrying out such studies with monkeys, and we shall see. A positive outcome will tend to argue that the order constraint results from an hierarchical structure, be it in language, perception or other behaviors. We shall see. 3.3 Aha! The eccentric mapping of meaning and form makes grammar

fun to learn — or is it the other way around? One purpose of language is to construct and convey meaning. In that case, why is the mapping between syntax and semantics so complex? Would it not be most functional if categories and units of meaning mapped univocally onto corresponding categories and units of form? Why does form jumble up the surface presentation of meaning, so that it is often impossible to extract it, except at the level of sentence, or even sometimes, discourse? Following Chomsky and Miller (1963), we can view the problem of the meaning-form mapping as a problem of mapping between a hierarchical unordered multidimensional structure and a linear ordered sequence. The representation of meaning contains complex information, which has to be condensed into a linear string of symbols. Recursion virtually guarantees that the linear string will contain elements ‘related at a distance’ as opposed to combined or adjacent. Thus, there must be some mismatch between the linear groupings available in sentences and the semantic structures they express. On this view, much of syntax lays down restrictions on distantly related elements so that the relevant meaning can be recovered. Thus, the first answer to the meaning/form mismatch is based on the incommensurability of the metalanguage of the two domains. Put in rampant functionalist terms, the purpose of syntax is to constrain serial and lexicalized forms so that they are reliably related to meanings. On this view, the answer to the semantic/ syntax mismatch conundrum is that it cannot be avoided. But clearly it can be avoided for isolated phenomena. For example, there is no channel or communicative reason why semantic gender could not reliably predict


syntactic gender in those languages that have it. Yet this is rarely the case, if ever. Characteristically, syntactic gender must be learned as an arbitrary feature of each noun — there may seem to be some vague semantic field associated with limited subsets of a syntactic gender, but nothing systematic and always replete with exceptions. The corresponding phenomenon exists for verb conjugation classes: sometimes they can be identified phonologically, but rarely if ever, semantically. These are clear cases where nothing would be lost in expressive power if there were strict regularity, i.e., between semantic and syntactic categories: so why aren’t there? An operational way of putting this is to ask: why don’t language-learning children re-regularize the semantic/syntactic category lineup, disregarding their parent’s perversity? Of course, for a while they do, that is, there are reports from various languages that young children sometimes pick one syntactic gender as the default for a new word, but also sometimes assign its syntactic gender based on some theory about its meaning. But that does not stick. The language learning process may be the culprit. We all know and regularly genuflect to the concept that language is innate: that is, the child must contribute many — perhaps all — of the interesting structural properties inherent to the semantic/syntax mapping, because by virtue of the reasoning above, the syntactic universals are not transparent in the language experiences of the child. All well and good: the child has the innate capacity to invent grammar. But where is the motive? It won’t do to sit back and assert that the ‘communicative urge’ is sufficient: children communicate very well with ungrammatical utterances, indeed often without much in the way of utterances at all. A Markovian 8th order of approximation to English would produce and process most any sentence the child might need. Why does the child work and work away, to arrive at the kind of abstract grammar that children learn? Look at it from the child’s standpoint — why does s/he work so hard to get it exactly right? There probably are convergent reasons, including social identification, communication, the need to feel superior (over littler kids), among others. But I think there is another overriding internal motive: language is a puzzle that is fun to solve. More than any other indigenous skill, it affords one “aha” discovery after another, over an extended period, at least 8 years in my estimation. What other activity can beat that? This argument presupposes several claims: a. People like to solve problems of a particular kind. b. Language presents that kind of problem. We know that people like to solve puzzles, but what constitutes a really satisfying problem? It is hard to discover that by looking at real problems, even those instantiated in games, because real problems bring their own constraints and goals to the situation, thus muddying any analysis. A different way to approach the issue is to examine what makes an aesthetically satisfying experience. Aesthetic preferences —

348 Thomas G. Bever

by definition in some cases — are free from any external goals, except the one of providing enjoyment. In this, we follow the original reason to study aesthetics voiced by the founder of such investigations — Gustav Fechner: such study reveals how the mind works in free action (im Leerlauf). The fine arts are complicated because we pay for them, so it is more helpful to examine “folk aesthetics”, entities that people prefer naturally. Consider for example, the golden section rectangle, which has the ratio of 1.62+ to one. Fechner documented that it is preferred in many ways, despite the fact that from the standpoint of obvious material functions (e.g., in construction blocks, or windows) it is an infelicitous ratio. The explanation lays in a convergence of both general aesthetic theories, framed originally by Aristotle: a.

The Goldilocks Principle: objects must have just the right amount of complexity, not too much, not too little. b. Conflict resolution: an object must stimulate some kind of representational conflict, and afford a resolution of it by accessing a new integrative concept. The golden mean ratio affords an interesting kind of complexity, being the mathematical limit to both geometric and linear progressions. In perceptual terms, this means that if a square of the short side of the rectangle is subtracted from its area, what remains is another rectangle with exactly the same ratio (rotated 90 degrees). There is considerable evidence that the perception of simple figures (and perhaps everything) involves analysis into simple canonical figures such as squares. Thus, the interesting complexity of the golden mean rectangle is that it can be analyzed by successive subtraction of squares, always leaving the same figure as residue, but the analysis is never entirely complete. The claim that this is an aesthetically optimal level of processing complexity requires independent verification, so at the moment it is a reasonable speculation only. We have, however, investigated some implications of the claim that the golden mean rectangle affords a resolvable representational conflict. Consider first the square-subtraction method for a rectangle with a ratio of 1.5 to one: The two squares overlap to create three vertical rectangles of .5 to one: thus, the figure can be “resolved” in two dimensions by the propagation of a single figure (itself resolvable into two canonical squares of .5). Most important it does not matter for this analysis, which square is chosen ‘first’, the overlap collapses into a simple resolution either way. The golden mean rectangle does not afford such a resolution: overlapping squares on the short side define a rectangle of .42 to one, which does not afford any direct resolution of the entire figure into a single two-dimensional figure. This sets up a conflict — should the left or right square be the one subtracted? This sets up the representational conflict. The conflict can be resolved by accessing a different level of representation, namely by analyzing the two squares as overlapping in three


dimensions. Thus, on this interpretation, the golden mean rectangle stimulates a situation that meets Aristotle’s conflict-resolution model of aesthetic objects. It stimulates a conflict that is resolvable by accessing a different level of representational analysis. This makes a neat prediction: the golden mean rectangle should stimulate more perception of depth (at least in its center) than other rectangle shapes. First, we noted that various schools of art that allow variation in frame shape, tend to converge on the golden mean rectangle for landscapes, compared with interior scenes: perhaps artists already know that the golden mean rectangle enhances depth. We have also studied the effect of frame shape on depth perception experimentally. We use standard depth illusions, and find that they are enhanced within the golden mean frame, compared with other shapes (Nicholas and Bever, ms). So, there is one experimental confirmation of the general theory of aesthetics, at least in regard to conflict resolution. It also accords with intuitions about many experiences of explicit problem solving. In a classic case study of “insight”, elucidated by the Gestalt Psychologist Wertheimer, people solve the problem of how to x-ray tumors without killing healthy tissue, with the following stages: a. the x ray gun will kill the healthy tissue if it is strong enough to kill the tumor b. weakening the gun will allow the tumor to survive (conflict) c. AHA, point the weakened gun at the tumor from different points each time and accumulate radiation of it, without killing any particular healthy tissue. That is, accessing the idea of pointing the gun from different places (a new dimension) resolves the conflict. People just like to do that. Perhaps, conflict resolution by accessing unexpected levels of representation is innate. To return to language in particular. We can now interpret the child’s tolerance of the persistent mismatch between semantic and syntactic units and categories: a language with such mismatches presents an ongoing series of problems to solve. The problems are fun to solve just because of the conflict between the different apparent representations — surface form and meaning units. Abstract syntax is the level of computation, which solves the constantly appearing conflicts between meaning and surface form. So, the function of the mismatch is that it makes multileveled grammar fun to learn, and therefore what is learned. But now we can turn the argument on its head. Return to Chomsky and Miller’s observation that the problem that language sets is mapping a complex hierarchical semantic representation onto a serially ordered string. This necessarily involves a “compression” of the hierarchical information, (or decompression of the serial information), thus guaranteeing some kind of multi-staged derivation when mapping one kind of representation onto the other. The arguments I made concerning the child’s intrinsic motivational structure to learn abstract systems, requires the language learning child to have “fun” learning the kind of derivational

350 Thomas G. Bever

sequences that languages must have. Result? In order for languages to be learned, humans must find multi-leveled structures fun to learn. Result of that? Humans find many problems and objects more satisfying when they exhibit multileveled properties. Thus, on this view, the characteristic human problem solving and aesthetic preference systems indeed have strong formal relations to language learning, but the causal direction is from language to cognition, rather than the reverse. This may turn out to be an unexpected kind of Whorfian result: a profound property of human cognition may be the result of a mental proclivity whose initial role is to guarantee the learnability of language in particular. On either causal view, we can explain why the semantic/syntax mapping is irregular, even in those cases where making it regular would not impair the expressive power of language. Children are pattern learners, with ever more evidence that this starts virtually at birth (e.g., Gomez and Gerken 1999; Saffran, Aslin and Newport 1996). Suppose that nominal semantic and syntactic gender, or lexical verb semantics and syntactic subcategorization were in perfect register. The child would arrive at an easy pattern, tending to confirm the idea that semantics and syntax have simple mappings onto each other. Yet, that is critically false when it comes to non-lexical compositional forms. The child could be prematurely mislead into an acquisitional cul-de-sac, expecting all semantic/syntax mapping to be similarly transparent. On this view, the existence of exceptions is critical to stimulating the right kind of learning. This hypothesis, too, is subject to some testing, perhaps using “artificial grammar” or pattern learning in infants and adults. The theoretical question is easy, if the experimental design is hard: do learners actually learn structurally multileveled systems faster when the data are judiciously sprinkled with exceptions to their general rules?

4. Conclusion. What is Functionalist Investigation for? I have tried to demonstrate how far reaching functional explanations of linguistic structures might be. Of course, showing that a linguistic universal might be explained by an extra-linguistic function, does not mean that it is not embedded as part of the innate structures that directly prefigure language. But it does open the question. Thus, the priority of canonical deep and surface word orders may not be innate, but rather the result of the early stages of language use by children; the fact that less embedded positions are more salient, does not mean that this is part of the innate structures for language: rather the biological substrate for language forces links between ever more salient sites to work upward and not downward. Finally, the mismatch between semantic and syntactic categories may not be linguistically innate, but an historically winnowed properties of extant grammars, because they were the ones children want to learn for general reasons.

Deconstructing functionalist explanations of linguistic universals

Why should a linguist care about these abstruse potential extra-linguistic sources for linguistic structures? I cannot dictate an answer, but I do know that the truth shall set you free. In this case, free from being required to embed every linguistic universal theoretically as derived from a linguistic structure. For example, the proposal that canonical word order is a universal linguistic constraint within a stochastic ordering of universal constraints, presupposes and supports a particular architecture for syntax — stochastic optimality theory. Indeed, the apparent merging of grammaticized requirement and statistical preference for canonical orders is a strong argument in favor of the architecture that allows such merging. But if canonical construction forms are themselves learned from the surrounding language, and perhaps shaped by the order of thought, then the claim that the constraint is embedded as a central linguistic universal is gratuitous, and potentially quite wrong. In times of yore in language science (when Eloise was starting out her training), there was disagreement about the ontology of grammars: on the one hand, ‘hocus pocus’ linguists argued that grammars are merely efficient organizations of linguistic data; ‘god’s truth’ linguists believed that there is a correct grammar. Most today are probably in the latter camp. But if getting it right only involves the correct description of an abstraction, it would matter only to those who hold linguistics to be a branch of mathematics: that is, ‘hocus pocus’ and ‘god’s truth’ linguists might end up agreeing that whoever is right, there are no implications for human biology or cognition. Today, the biological basis for language looms gigantically in the background of linguistic investigations. None of us knows where linguistic science is headed, but many suspect that it will include considerations of the specific genetic foundations of language capacity: Indeed, speculations about the relation between linguistic universals and genetic substrate are finding respectability — premature in my opinion, but the goal is noble (Baker 2001, Jackendoff 2002). This makes it important in the concrete, to ascribe linguistic universals to their actual cause: we do not want to be looking for the genetic source for extra-linguistic features of the mind, when we focus on the unique genetic source for language. Conversely, we do not want to be looking for the genetic source for general cognitive features of the mind, if they are actually extensions of linguistic capacities. Put more softly, we want to understand the potential relations between linguistic and general cognitive properties, when exploring the biological and ontogenetic bases for either one.

Note * My thanks first to Eloise, who is an inspiration to everyone around her as a scholar with quiet steel inside two velvet gloves. My thanks also to Massimo Piatelli-Palmarini for politely reviewing an earlier (and much more primitive version). Finally, special thanks to Heidi Harley, who in one stroke made me realize the importance of considering the possibility that apparent functionalist explanations of language may actually reveal deep linguistic explanations of human cognition.

351

References

Abraham, R. 1940a. A Dictionary of the Tiv Language. Hertford: Stephen Austin and Sons. Abraham, R. 1940b. The Principles of Tiv. Hertford: Stephen Austin and Sons. Abraham, W. 1996 “The aspect-case typology correlation: Perfectivity triggering split ergativity.” Folia Linguistica 30: 5–34. Abramson, A. 1962. “The vowels and tones of Standard Thai: Acoustical measurements and experiments.” International Journal of American Linguistics 28(2). Adger, D. 1997. “Deriving the mapping hypothesis” in Studies on Universal Grammar and Typological Variation, A. Alexiadou and T. A. Hall (eds.), 109–104. Amsterdam/Philadelphia: John Benjamins. Aissen, J. 1992. “Topic and focus in Mayan.” Language 68: 43–80. Aissen, J. 1999. “Markedness and subject choice in optimality theory.” Natural Language and Linguistic Theory 17: 673–711. Aissen, J. 2000. “Differential object marking: Iconicity vs. economy.” ms. U. of California, Santa Cruz. Alexiadou, A. and Anagnostopoulou, E. 1998. “Parametrizing AGR: word order, V-movement and EPP-checking.” Natural Language and Linguistic Theory 16: 491–539. Anttila, A. 1995. “Deriving variation from grammar: a study of Finnish genitives.” ROA #63. Aoun, J. and Li, A. 2002. The Structure of Chinese Relative Clauses. Ms. U. of Southern California. Archangeli, D. 1998. “Exceptions to Morpheme Structure Constraints in OT.” Paper presented at the Southwest Workshop on Optimality Theory IV, U. of Arizona. Archangeli, D. 1999. “Lexical Irregularity in OT: DOT vs. Variable Constraint Ranking”, Paper presented at the Linguistics Society of America Annual Meeting, January 1999. Archangeli, D. and Pulleyblank, D. 1994. Grounded Phonology. Cambridge: MIT Press. Austin, P. and Bresnan, J. 1996. “Non-configurationality in Australian aboriginal languages.” Natural Language and Linguistic Theory 14: 215–268. Bailyn, J. 2002. “On scrambling: A reply to Boškovic´ and Takahashi.” Linguistic Inquiry 32 (4): 635–658. Baker, M. 1985. “The mirror principle and morphosyntactic explanation.” Linguistic Inquiry 16: 373–415. Baker, M. 1988. Incorporation: a theory of grammatical function changing. Chicago: U. of Chicago Press. Baker, M. 1995. “On the absence of certain quantifiers in Mohawk.” In Quantification in natural languages, E. Bach, E. Jelinek, A. Kratzer and B. Partee (eds), 21–58. Dordrecht: Kluwer. Baker, M. 1996. The Polysynthesis Parameter. Oxford: Oxford U. Press. Baker, M. 2000. Categories and Category Systems. ms, Rutgers. Barragan, L. 2001. “Cupeño verb morphology and the theory of contextual allomorphy.” ms., U. of Arizona. Basilico, D. 1998. “Object position and predication form.” Natural Language And Linguistic Theory 16: 541–595.

354 References

Beckman, M. 1986. Stress and Non-Stress Accent. Dordrecht: Foris. Beckman, M., and Pierrehumbert, J. 1986. “Intonational structure in Japanese and English.” In Phonology Yearbook 3: 255–309. Belletti, A. 1988. “The case of unaccusatives.” Linguistic Inquiry 19: 1–34. Bever, T. G. 1970. “The cognitive basis for linguistic structures.” In Cognition and Language Development, R. Hayes (ed.), 277–360. New York: Wiley & Sons, Inc. Birnbaum, S. 1978. Yiddish: A Survey and a Grammar. Toronto: U. of Toronto Press. Bittner, M. 1994. Case, Scope and Binding. Dordrecht: Kluwer. Bittner, M. and K. Hale. 1996. “The structural determination of case and agreement.: Linguistic Inquiry 27: 1–68. Bloom, L. 1970. Language Development: Form and Function in Emerging Grammars. Cambridge: MIT Press. Bobaljik, J. 1994. “What does adjacency do?” MIT Working Papers in Linguistics 22: 1–32. Boersma, P. 1997. “How we learn variation, optionality, and probability”, ms. ROA#221. Boersma, P., and Hayes, B. 2001. “Empirical tests of the Gradual Learning Algorithm.” Linguistic Inquiry 32: 45–86. Böhmová, A., and E. Hajicˇová. 1999. “The Prague Dependency Tree Bank I: How much of the underlying syntactic structure can be tagged automatically?.” The Prague Bulletin if Mathematical Linguistics 71, 5–12. Bok-Bennema, R. 1991. Case and Agreement in Inuit. Dordrecht: Foris. Bollinger, D. 1986. Intonation and Its Parts. Stanford: Stanford U. Press. Booij, G. 1995. The Phonology of Dutch. Oxford: Clarendon Press. Boškovic´, Ž. 1998. “Wh-phrases and wh-movement in Slavic.” ms., U. of Connecticut. Bowers, J. 2002. “Transitivity.” Linguistic Inquiry 33: 183–224. Bresnan, J. 1982. The Mental Representation of Grammatical Relations. MIT Press. Bresnan, J. 1994. “Locative inversion and the architecture of Universal Grammar.” Language 70: 72–131. Bresnan, J. 2000a. “Explaining morphosyntactic competition.” In Handbook of Contemporary Syntactic Theory, M. Baltin and C. Collins (eds). 11–44. Oxford: Blackwell. Bresnan, J. 2001b. The emergence of the unmarked pronoun, in Legendre et al. (eds.), pp.113–142. Bresnan, J. and Aissen, J. 2002 “Optimality and functionality: Objections and refutations.” Natural Language and Linguistic Theory 20: 81–94. Bresnan, J, Dingare, S and Manning, C. 2001. “Soft constraints mirror hard constraints: Voice and person in English and Lummi”. in Proceedings of the LFG01 Conference, M. Butt and T. Holloway King (eds.) http://csli-publications.stanford.edu/LFG/6/lfg01-toc.html. Bresnan, J. and Kanerva, J. 1989. “Locative inversion in Chichewa: a case study of factorization in grammar.” Linguistic Inquiry 20: 1–50. Bresnan, J. and Mchombo, S 1987. “Topic, pronoun, and agreement in Chichewa.” Language 63: 741–782. Bright, W. and Hill, J. 1967. Linguistic history of the Cupeño. In Studies in Southwestern Ethnolinguistics, D. H. Hymes and W. Biddle, (eds.), 352–391. The Hague: Mouton and Company. Carnie, A. 1995 Head Movement and Non-Verbal Predication. Ph.D. dissertation, MIT. Carnie, A. in prep. “Phase Geometric Hierarchies.” Ms. U. of Arizona. Cash Cash, P. and Carnie, A. 2002. Nez Perce Case: a note on Woolford 1997. ms. U. of Arizona. Chafe, W. 1980. “The deployment of consciousness in the production of a narrative.” In The Pear Stories, W. Chafe (ed), 9–50. Norwood, NJ: Ablex. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge: MIT Press.

References

Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1986a. Barriers. Cambridge: MIT Press. Chomsky, N. 1986b. Knowledge of Language. New York: Praeger. Chomsky, N. 1995a. “Bare phrase structure.” In Government and Binding Theory and the Minimalist Program, G. Webelhuth (ed), 383–439. Oxford: Blackwell. Chomsky, N. 1995b. The Minimalist Program. Cambridge: MIT Press. Chomsky, N. 2000a. “Minimalist inquiries: the framework.” In Step by Step, R. Martin, D. Michaels, and J. Uriagereka (eds), 89–155. Cambridge: MIT Press. Chomsky, N. 2000b. New Horizons in the Study of Language and Mind. Cambridge: Cambridge U. Press. Chomsky, N. 2001a. Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20. Cambridge, MA: MITWPL Chomsky, N. 2001b. “Derivation by phase.” In Ken Hale: A Life in Language, M. Kenstowicz (ed), 1–52. Cambridge, MA: MIT Press. pp. 1–52 Chomsky, N. and Halle M. 1968. The Sound Pattern of English. New York: Harper and Row. Chomsky, N. and Miller, G. 1963. “Introduction to the formal analysis of natural languages.” In Handbook of Mathematical Psychology D. Luce, R. Bush, and E. Galanter (eds.), 269–321. New York: Wiley Cinque, G. 1990. Types of A-bar Dependencies. Cambridge: MIT Press. Clamons, R., Mulkern, A., Sanders, G., and Stenson, N. 1999. “The limits of formal grammar: Pragmatic motivation in Oromo grammar.” in Formalism and Functionalism in Linguistics. Volume 2: Case Studies. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley (eds.) pp 59–76. Amsterdam/Philadelphia: John Benjamins. Coleman, J., and J. Pierrehumbert. 1997. “Stochastic phonological grammars and acceptability.” In Computational Phonology: Third Meeting of the ACL Special Interest Group in Computational Phonology, J. Coleman (ed), 49–56. Somerset: Association for Computational Linguistics. Collins, C and P Branigan. 1997. “Quotative Inversion.” Natural Language and Linguistic Theory 15: 1–41. Collins, C. 2000. “VP internal structure in Ju’huan and Hoan.” ms. Cornell U. Collins, C. 2001. Eliminating labels. MIT Occasional Papers in Linguistics 20. Comorovski, I. 1996. Interrogative Phrases and the Syntax-Semantics Interface. Dordrecht: Kluwer. Comrie, B. 1985. Tense. Cambridge U. Press. Connell, B. and D. R. Ladd 1990. “Aspects of pitch recognition in Yoruba.” Phonology 7: 1–29. Croft, W. 1995. “Autonomy and functionalist linguistics.” Language 71: 490–532. Crystal, T. H. and House, A. S. 1982. “Segmental durations in connected speech signals: preliminary results.” Journal of the Acoustical Society of America 72: 705–716. Crystal, T. H. and House, A. S. 1988. “Segmental durations in connected speech signals: current results.” Journal of the Acoustical Society of America 83: 1553–1573. Culicover, P and Rochemont, M. 1983. “Stress and focus in English.” Language 59: 123–165. Cutler, A, Dahan, D and van. Donslaar, W 1997. “Prosody in the comprehension of spoken language: a literature review.” Language and Speech 40: 141–201. Darnell, M., Moravcsik, E., Newmeyer, F., Noonan, M., Wheatley, K. (eds.) 1999. Functionalism and Formalism in Linguistics. Amsterdam/Philadelphia: John Benjamins (two volumes). Davidson, D. 1967. “The logical form of action sentences”. In The Logic of Decision and Action, N. Rescher (ed.), 81–95. Pittsburgh: U. of Pittsburgh Press. Davis, H. 1994. “Tali-ho!” In Papers for the 29th International Conference on Salish and Neighboring Languages, 117–144. Pablo, Montana: Salish Kootenai College.

355

356 References

Davis, H. 1997. “Deep unaccusativity and zero syntax in St’át’imcets.” In Theoretical Issues at the Morphology-Syntax Interface, Supplements to the International Journal of Basque Linguistics and Philology XL, A. Mendikoetxea and M. Uribe-Etxebarria (eds), 55–96. Bilbao: Universidad dal Pais Basco. Davis, H. 1999. “Word order and configurationality in St’át’imcets.” In Papers for the 34th International Conference on Salish and Neighboring Languages, M. Q. Hinkson (ed), 61–82. Kamloops, B. C.: Simon Fraser U. SCES/SFU Program. Davis, H. in prep. A Teaching Grammar of Upper St’át’imcets. Dayal, V. 1999. “Bare NP’s, reference to kinds, and incorporation.” Proceedings of SALT 9: 34–51. de Hoop, H. 1992. Case Configuration and Noun Phrase Interpretation. Ph.D. dissertation, U. of Groningen. de Jong, K., and McDonough, J. 1993. “Tone in Navajo.” In U. of California Working Papers in Phonetics 84: 165–182. DeLancy, S. 1981. “An interpretation of split ergativity and related patterns”. Language 57: 626–657. Demirdache, H. 1997. “Condition C.” In Atomism and Binding, H. Bennis, P. Pica and J. Rooryck (eds). 51–88 Dordrecht: Foris. Diesing, M. 1990. “Verb movement and the subject position in Yiddish.” Natural Language and Linguistic Theory 8, 41–79. Diesing, M. 1992. Indefinites. Cambridge: MIT Press. Diesing, M. 1997. “Yiddish VP order and the typology of object movement in Germanic.” Natural Language and Linguistic Theory 15, 369–427. Diesing, M. and Jelinek, E. 1995. “Distributing Arguments.” Natural Language Semantics 3: 123–167. Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge U. Press. DuBois, J. 1987. “The Discourse Basis of Ergativity.” Language 63: 805–855. Eijk, J. van 1997. The Lillooet Language: Phonology, Morphology, Syntax. Vancouver: UBC Press. Enç, M. 1991. “The semantics of specificity.” Linguistic Inquiry 22: 1–25. Enderton, H. 1972. A Mathematical Introduction to Logic. New York: Academic Press. Faltz, L. 1998. The Navajo Verb. Albuquerque: U. of New Mexico Press. Farkas, D. 1992. “On the semantics of subjunctive complements.” In Romance Languages and Modern Linguistic Theory, P. Hirschbühler (ed.), 69–105. Amsterdam/Philadelphia: John Benjamins. Fintel, K. von 1989. “Theticity in Generative Grammar.” Papers on Quantification. Report on NSF Grant 871999. Principal Investigators: E. Bach, A. Kratzer and B.Partee. U. of Massachusetts at Amherst. Fitzgerald, C. 1994. “Prosody drives the syntax”. In Proceedings of BLS 20: 173–183. Fitzgerald, C. 1997. O’odham Rhythms. Ph.D. Dissertation, U. of Arizona. Fitzgerald, C. 1998. “The meter of Tohono O’odham songs.” International Journal of American Linguistics 64: 1–36. Fitzgerald, C. 2001a. “The morpheme-to-stress principle in Tohono O’odham.” Linguistics 39: 941–972. Fitzgerald, C. 2001b. Across-the-Board Effects in Tohono O’odham Meter. Ms., U. at Buffalo, SUNY. Fitzgerald, C. 2002. “Syntactic inversions in English meter: implications for Optimality Theory.” Paper presented at the 2002 Toulouse Conference on English Phonology, Toulouse, France. Fougeron, C. and Keating, P. 1997. “Articulatory strengthening at edges of prosodic domains.” Journal of the Acoustical Society of America, 101: 3728–3740.

References 357

Frisch, S., Large, N., and Pisoni, D. 2000. “Perception of wordlikeness: effects of segment probability and length on the processing of nonwords.” Journal of Memory and Language 42: 481–496. Furui, S. 1986. “On the role of spectral transition for speech perception.” Journal of the Acoustical Society of America, 80: 1016–1025. Gårding, E. and Bruce, G. 1982. “A presentation of the Lund model of Swedish intonation.” Lund Working Papers 21: 69–75. Geenhoven, V. van 1996. Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects of West Greenlandic Noun Incorporation. Ph.D. dissertation, Universität Tübingen. Geenhoven, V. van 1998. “On the argument structure of some noun incorporating verbs in West Greenlandic.” In The Projection of Arguments: Lexical and Compositional Factors, M. Butt and W. Geuder (eds), 225–263. Stanford: CSLI Publications. Gerdts, D. 1988. Object and Absolutive in Halkomelem Salish. New York: Garland. Gibson, E. 1998. “Linguistic complexity: Locality of syntactic dependencies.” Cognition 68: 1–76. Gifford, W. 1918. “Clans and moieties in Southern California.” University of California Publications in American Archaeology and Ethnology 14: 155–219. Gifford, W. 1922. “California kinship terminologies.” University of California Publications in American Archaeology and Ethnology 18: 1–285. Giorgi, A. and F. Pianesi. 1997. Tense and Aspect. Oxford: Oxford U. Press. Givón, T. 1979. On understanding grammar. New York: Academic Press. Givón, T. 1994. Voice and Inversion. Amsterdam/Philadelphia: John Benjamins. Golden, M. 1995. “Multiple wh-questions in Slovene.” In Annual Workshop on Formal Approaches to Slavic Linguistics: The Cornell Meeting. W. Brown, E. Dornisch, N. Kondrashova, and D. Zec (eds ) 240–266. Ann Arbor: Slavica Publishers. Goldsmith, J. 1993. “Harmonic phonology.” In The last phonological rule: Reflections on constraints and derivations, J. Goldsmith (ed), 21–60. Chicago: U. of Chicago Press. Golston, C. 1995. “Syntax outranks phonology.” Phonology 12 (3): 343–368. Golston, C. 1996. “Direct Optimality Theory: Representation as pure markedness.” Language 72: 713–748. Golston, C. 1998. “Constraint-based metrics.” Natural Language and Linguistic Theory 16: 719–770. Golston, C. and T. Riad. 2000. “The phonology of Classical Greek meter.” Linguistics 38: 99–167. Gomez, R. L., and Gerken, L. 1999 “Artificial grammar learning by 1-year-olds leads to specific and abstract knowledge.” Cognition 70: 109–135. Grewendorf, G. 2001. “Multiple wh-fronting.” Linguistic Inquiry 32: 87–122. Grohmann, K. 2000. Prolific Peripheries: A Radical View From The Left. Ph.D. dissertation, U. of Maryland. Gunlogson, C. 2001. “Third-Person object prefixes in Babine-Witsuwit’en.” International Journal of American Linguistics. 67: 365–396. Gussenhoven, C. 1992. “Illustrations of the IPA: Dutch.” Journal of the International Phonetic Association 22: 45–47. Haeberli, E. 2001. “Deriving syntactic effects of morphological case by eliminating abstract case.” Lingua 111: 279–313 Haider, H. 1986. “Affect α: A reply to Lasnik and Saito, ‘On the nature of proper government’.” Linguistic Inquiry 17: 113–125. Hajicˇ, J. 1998. “Building a syntactically annotated corpus: The Prague Dependency Treebank.” In Issues of Valency and Meaning: Studies in Honour of Jarmila Panevová, E. Hajicˇová (ed), 106–132. Prague: Karolinum.

358 References

Hajicˇ, J. and Hladká, B. 1997. “Probabilistic and rule-based tagger of an inflective language: A comparison.” In: Proceedings of the Fifth Conference on Applied Natural Language Processing, 111–118. Hajicˇová, E. 1993. Issues of sentence structure and discourse patterns. Prague: Charles U. Hajicˇová, E. and Vrbová, J. 1982. “On the role of the hierarchy of activation in the process of natural language understanding.” In COLING 82. J. Horecký (ed.), 107–113. Amsterdam: North Holland. Hajicˇová, E., Partee, B. H. and Sgall, P.1998. Topic-focus Articulation, Tripartite Structures, and Semantic Content. Dordrecht: Kluwer. Hale, K. 1974. “Counterexamples and explanations in Navajo linguistics: Phonology.” Diné Bizaad Náníl’i˛˛ih/Navajo Language Review, l, 175–203. Hale, K. 1982. “Preliminary remarks on configurationality.” Proceedings of NELS 12: 86–96. Hale, K. 1992. “Basic word order in two ‘free word order’ Languages.” In Pragmatics of Word Order Flexibility, D. L. Payne (ed), 63–82. Amsterdam/Philadelphia: John Benjamins Publishing Company. Hale, K. 2000b. “Remarks on the syntax of the Navajo verb.” Working Papers on Endangered and Less Familiar Languages 3: 55–97. Hale, K. 2001. “Navajo verb stem position and the bipartite structure of the Navajo conjunct sector.” Linguistic Inquiry 32.4: 678–693. Hale, K. and Keyser, S. J. 1993. “On argument structure and the lexical expression of syntactic relations.” In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. K. Hale and S. Keyser, (eds.) pp 53–109. Cambridge: MIT Press. Hale, K. and Keyser, S. J. 1998. “The basic elements of argument structure.” MIT Working Papers in Linguistics 35: 73–118. Hale, K., Jeanne, L., and Platero, P. 1977. “Three cases of overgeneration.” In Formal syntax, P. Culicover, T. Wasow, and & A. Akmajian (eds), 279–419. New York: Academic Press. Hale, K., Jelinek, E., and Willie, M. to appear. “Topic and focus movement in Navajo.” In Word Order and Scrambling, S. Karimi (ed). Oxford: Blackwell. Hale, K., P. Munro, and P. Platero. 2000. “The Navajo Inchoative Verb Form.” Working Papers on Endangered and Less Familiar Languages 3: 97–116. Hale, K. and Platero, P. 1996. “Navajo reflections of a general theory of lexical argument structure.” In Athabaskan Language Studies: Essays in Honor of Robert W. Young, E. Jelinek, S. Midgette, K. Rice, and L. Saxon (eds), 1–13. Albuquerque: U. of New Mexico Press. Hale, K. and Selkirk, E. 1987. “Government and tonal phrasing in Papago.” Phonology Yearbook 4: 151–183. Halle, M. and Marantz, A. 1993. Distributed morphology and the pieces of inflection. In The view from Building 20, K. Hale and S. J. Keyser (eds), 111–176. Cambridge: MIT Press. Halle, M. and Vergnaud, J-R. 1987. An Essay on Stress. Cambridge: MIT Press. Halliday, M. A. K. 1967. Intonation and Grammar in British English. The Hague, Mouton. Hammond, M. 1999. “Lexical frequency and rhythm.” In Functionalism and Formalism in Linguistics, M. Darnell et al. (eds), 329–358. Amsterdam/Philadelphia: John Benjamins. Hammond, M. 2000a. “The logic of OT.” Western Conference on Linguistics 28: 146–162. Hammond, M. 2000b. The Phonology of English. Oxford: Oxford U. Press. Hardy, F. 1979. Navajo Aspectual Stem Variation. Ph.D. Dissertation. Albuquerque: U. of New Mexico Hargus, S., and S., Tuttle. 1997. “Augmentation as affixation in the Athabaskan languages.” Phonology 14: 177–220. Harley, H. 1995 Subjects Events and Licensing. Ph.D. Dissertation, MIT.

References 359

Harrington, J. P. 1933. “Annotations.” In Chingichnish, Fr. Geronimo Boscana (ed.), translated by Alfred Robinson. Santa Ana, CA: Fine Arts Press. Hayes, B. and MacEachern, M. 1998. “Quatrain form in English folk verse.” Language 74: 473–507. Hayes, R. 2001. “Interlanguage development as variable, weighted constraint rankings.” ms. U. of Arizona. Heath, J. 1998. “Hermit crabs: Formal renewal of morphology by phonologically mediated affix substitution.” Language 74: 728–750. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, U. of Massachusetts at Amherst. Higginbotham, J. 1985. “On semantics.” Linguistic Inquiry 16: 547–593. Hill, J. 1969. “Volitional and non-volitional verbs in Cupeño” in Papers from the Fifth Regional Meeting of the Chicago Linguistics Society. pp 348–356. Hill, J. and Hill. K. 1968. Stress in the Cupan languages. International Journal of American Linguistics 34: 233–241. Hintikka, J. 1969. “Semantics for propositional attitudes.” In PhilosophicalLogic, J. Davis, D. Hockney, and W. Wilson (eds), 21–45. Dordrecht: Reidel. Hoge, K. 2000. Superiority. Ph.D. dissertation, U. of Oxford. Hong, S-H. 1994. Issues in Round Harmony: Grounding, Identity and their Interaction. Ph.D. dissertation, U. of Arizona. Hopi Dictionary Project (HDP). 1998. Hopìikwa lavàytutuveni/Hopi dictionary: A Hopi-English dictionary of the Third Mesa dialect. Tucson: U. of Arizona Press. Hornstein, Norbert. 1995. Logical Form: From GB to Minimalism. Blackwell Publishers. Huang, C.-T. J. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D. dissertation, MIT. Hukari, T. 1977. “A comparison of attributive constructions in two Coast Salish languages.” Glossa 11: 48–74. Inkelas, S, Orgun, O. and Zoll, C. 1997. “The implications of lexical exceptions for the nature of grammar.” In Derivations and Constraints in Phonology, I Roca (ed.), 393–418. Oxford, Clarendon Press. Isaak, A. 2000. Split Case Marking and Prominence Relations. Ph.D. Dissertation. U. of Massachusetts, Amherst. Itô, J. and Mester, A. 1986. “The phonology of voicing in Japanese: Theoretical consequences for morphological accessibility.” Linguistic Inquiry 17: 49–73. Itô, J. and Mester, A. 1993. Japanese Phonology: Constraint Domains and Structure Preservation. LRC-93–06. UC Santa Cruz. Itô, J. and Mester, A. 1995. “Japanese phonology.” In The Handbook of Phonological Theory, J. Goldsmith (ed), 817–838. Oxford: Blackwell. Jackendoff, R. 1973. Semantic Interpretation in Generative Grammar. Cambridge: MIT Press. Jackendoff, R. 2002. Foundations of Language. Oxford: Oxford U. Press. Jacobs, N., Prince, E. and van der Auwera, J. 1994. “Yiddish.” In The Germanic languages, E. König and J. van der Auwera (eds), 388–419. London: Routledge. Jacobs, R. 1975. Syntactic Change: A Cupan Case Study. U. of California Publications in Linguistics 79. Berkeley: U. of California Press. Jaeggli, O. 1982. Topics in Romance Syntax. Dordrecht: Foris. Jelinek, E. 1984. “Empty categories, case, and configurationality.” Natural Language and Linguistic Theory 2: 39–76. Jelinek, E. 1989. “Argument type in Athabaskan: Evidence from noun incorporation”, ms. U. of Arizona.

360 References

Jelinek, E. 1993. “Ergative ‘splits’ and argument type”. In MIT Working Papers in Linguistics 18: 15–42. Jelinek, E. 1994. “Quantification in Straits Salish.” In Quantification in Natural Language, E. Bach, A. Kratzer, E. Jelinek and B. Partee (eds), 487–540. Dordrecht: Kluwer. Jelinek, E. 1997. “Prepositions in Northern Straits Salish and the noun/verb question.” Salish Languages and Linguistics. E. Czaykowska-Higgins and M. Dale Kinkade, (eds.) pp. 325–346. The Hague: Mouton. Jelinek, E. 1998. Voice and transitivity as functional projections in Yaqui. In The Projection of Arguments, Lexical and Compositional Factors, M. Butt and W. Geuder, (eds.), 195–224. Stanford: CSLI Publications. Jelinek, E. 2000a “Datives and Argument Hierarchies”. Working Papers in Endangered and Less Familiar Languages 1: 51–70. Jelinek, E. 2000b “Predicate Raising in Lummi, North Straits Salish.” In The Syntax of Verb Initial Languages. A. Carnie and E. Guilfoyle, (eds.) pp. 213–234. Oxford: Oxford U. Press. Jelinek, E. 2001. “Voice and transitivity as functional projections in Yaqui.” Paper presented to the Uto-Aztecan Workshop, Departments of Linguistics and Anthropology, U. of Arizona, Tucson, Arizona, April 2001. Jelinek, E. and Demers, R. 1994. “Predicates and pronominal arguments in Straits Salish.” Language 70: 697–736. Jelinek, E. and Escalante, F. 2000. “Unaccusative and unergative verbs in Yaqui.” In Uto-Aztecan, Structural, Temporal, and Geographic Perspectives: Papers in Memory of Wick R. Miller, E. H. Casad and T. L. Willett, (eds.), 171–82. Hermosillo: Universidad de Sonora, Division de Humanidades y Bellas Artes. Johnson, K. 1991. “Object positions.” Natural Language and Linguistic Theory 9: 577–636. Johnson, K. 1997. Acoustic and Auditory Phonetics. Oxford: Blackwell. Kager, R. 1999. Optimality Theory. Cambridge: Cambridge U. Press. Kamp, J. A. W. 1981. “A theory of truth and semantic representation.” In Formal Methods in the Study of Language. J. Groenendyck, T. Janssen, and M. Stokhof (eds.) pp 277–321. Amsterdam: Mathematical Centre. Kamp, L. J. Th. van der and Pols, L. C. W. 1971. “Perceptual analysis from confusions between vowels.” Acta Psychologica 35: 64–77. Kari, J. 1976. Navajo Prefix Phonology. New York: Garland. Karimi, S. 1999. “Is Scrambling as strange as we think it is.” In MIT Working Papers in Linguistics 33: 159–190. Karimi, S. under revision. A Minimalist Approach to Scrambling: Evidence from Persian. ms. U. of Arizona. Karttunen, L. 1977. “The syntax and semantics of questions.” Linguistics and Philosophy 1: 1–44. Kaun, A. 1995. The Typology of Rounding Harmony: An Optimality Theoretic Approach. Ph.D. dissertation, UCLA. Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge: MIT Press. Keenan, E. 1976. “Towards a universal definition of ‘subject’”. Subject and Topic. C. Li and S. Thompson (eds.), 303–333. New York, Academic Press. Keenan, E. and Comrie, B. 1977. “Noun phrase accessibility and Universal Grammar.” Linguistic Inquiry 8: 63–100. Kimenyi, A. 1980. A Relational Grammar of Kinyarwanda. Berkeley: U. of California Press. Kingston, J. 1990. “Articulatory binding.” In Papers in Laboratory Phonology 1: Between the Grammar and Physics of Speech, J. Kingston and M. E. Beckman (eds), 406–434. Cambridge: Cambridge U. Press.

References 361

Kinyalolo, K. 1991. Syntactic Dependencies and the Spec–Head Agreement Hypothesis in Kilega. Ph.D. dissertation, UCLA. Kiparsky, P. 1982. “Lexical morphology and phonology.” In Linguistics in the Morning Calm, I.-S. Yang (ed), 3–91. Seoul: Hanshin. Kiparsky, P. 1998. “Partitive Case and Aspect.” In The Projection of Arguments: Lexical and Compositional Factors, M. Butt and W. Geuder (eds), 265–307. Stanford, CA: CSLI Publications. Kiparsky, P. and C. Kiparsky. 1971. “Fact.” In Semantics. D. Steinberg and L. Jakobovits (eds.), 345–369. Cambridge: Cambridge U. Press. Kirchner, R. 1993. “Turkish vowel disharmony in Optimality Theory.” Ms., UCLA. Kiss, K. 1986. “The order and scope of operators in the Hungarian sentence.” In Topic, Focus, and Configurationality, W. Abraham and S. de Mey (eds), 181–214. Amsterdam/Philadelphia: John Benjamins. Kiss, K. 1992. “Logical structure in syntactic structure: the case of Hungarian.” In Logical Structure and Linguistic Structure, C.-T. J. Huang and R. May (eds), 111–147. Dordrecht: Kluwer. Kiss, K. 1993. “Wh-movement and specificity.” Natural Language and Linguistic Theory 11: 85–120. Kiss, K. 1995. Discourse-Configurational Languages. New York: Oxford U. Press. Kiss, K. 1998. “Identificational vs. information focus.” Language 74: 245–273. Kiss, K. To appear. “Argument scrambling, operator movement, and topic movement in Hungarian.” In Word Order and Scrambling, S. Karimi (ed). Oxford: Blackwell. Kisseberth, C. 1970. “The treatment of exceptions.” Papers in Linguistics 2: 44–58. Kisseberth, C. 1993. “Optimal domains: a theory of Bantu tone: A case study from Isixhosa.” Paper presented at Rutgers Optimality Workshop 1. Rutgers U. Klatt, D. H. 1976. “Linguistics uses of segmental duration in English: Acoustic and perceptual evidence.” Journal of the Acoustical Society of America 59: 1208–21. Koizumi, M. 1995. Phrase Structure in Minimalist Syntax. Ph.D. dissertation, MIT. Koopmans-van Beinum, F. J. 1980. Vowel Contrast Reduction: An Acoustic and Perceptual Study of Dutch Vowels in Various Speech Conditions. U. of Amsterdam dissertation, Academische Pers. Kornfilt, J. 1997. Turkish. New York: Routledge Koslow, A. 1992. A Structuralist Theory of Logic. Cambridge: Cambridge U. Press. Krahmer, E. and M. Theune 1999. “Efficient generation of descriptions in context.” In Proceedings of the Workshop on the Generation of Nominal Expressions, R. Kibble and K. van Deemter (eds). Utrecht, the Netherlands: U. of Utrecht. Krahmer, Emiel. 1998. Presupposition and anaphora. CSLI Lecture Notes 89. Stanford: CSLI. Kratzer, A. 1996. “Severing the external argument from its verb.” In Phrase Structure and the Lexicon, J. Rooryck and L. Zaring (eds), 109–137. Dordrecht: Kluwer. Kratzer, A. 1998. “Scope or pseudo-scope? Are there wide-scope indefinites?” In Events in Grammar, S. Rothstein (ed), 163–196. Dordrecht: Kluwer. Krauss, M and Leer J. 1976. Proto-Athbaskan-Eyak *y and the Na-Dene Sonorants. Alaska Native Language Center, Fairbanks Krifka, M. 1991, “A compositional semantics for multiple focus constructions.” Cornell Working Papers in Linguistics 10: 127–158. Kroeber, P. 1907. “Shoshonean dialects of California.” University of California Publications in American Archaeology and Ethnology 4. Berkeley: U. of California. Kroeber, P. 1999. The Salish Language Family: Reconstructing Syntax. Bloomington: U. of Nebraska Press. Ladd, D. R. 1996. Intonational Phonology. Cambridge U. Press. Ladefoged, P. and Maddieson, I. 1996. The Sounds of the World’s Languages. Oxford: Blackwell.

362 References

Lambrecht, K. 1994. Information Structure and Sentence Form. Cambridge: Cambridge U. Press. Lang, C. and Ohala, J. J. 1996. “Temporal cues for vowels and universals of vowel inventories.” In Proceedings of the Fourth International Conference on Spoken Language Processing, October 3–6, 1996, Philadelphia, 434–437. Newark: U. of Delaware. Langacker, R. W. 1977. Studies in Uto-Aztecan Grammar Volume 1: An Overview of Uto-Aztecan Grammar. Arlington: SIL Publications in Linguistics. Langacker, R. W. 1987. Foundations of Cognitive Grammar: Volume 1: Theoretical Prerequisites. Stanford: Stanford U. Press. Langendoen, D. T. 1970. “The accessibility of deep structure.” In Readings in English Transformational Grammar, R. A. Jacobs and P. S. Rosenbaum (eds), 99–104. Waltham: Ginn and Company. Langendoen, D. T. 1975. “Finite-state parsing of phrase-structure languages and the status of readjustment rules in grammar.” Linguistic Inquiry 6: 533–554. Langendoen, D.T. 2001. “Review of New Horizons in the Study of Language and Mind.” Language 77: 583–585. Laniran, Y. 1992. Intonation in Tone Languages: the Phonetic Implementation of Tone in Yoruba. Ph.D. dissertation, Cornell U. Larson, R. 1988. “On the double object construction” Linguistic Inquiry 22: 103–139. Lasnik, H. 1999. “On feature strength: Three minimalist approaches to overt movement.” Linguistic Inquiry 30: 197–217. Leer, J. 1979. Proto-Athabaskan Verb Stem Variation, Part One: Phonology: Alaskan Native Language Center Research Papers Lehiste, I. and P. Ivic 1986. Word and Sentence Prosody in Serbo-Croatian. Cambridge: MIT Press. Levin, B. 1983. On the Nature of Ergativity. Ph.D. dissertation, MIT. Li, F. 1933. “Chipewyan consonants.” Bulletin of the Institute of History and Philology of the Academia Sinica. Supplementary Vol. 1: 429–67. Liljencrants, J. and Lindblom, B. 1972. “Numerical simulation of vowel quality systems: the role of perceptual contrast.” Language 48: 839–862. Limber, J. 1996. “Interspecies communication, consciousness, and the cognitive verb gap.” Paper presented at “Toward a Scientific Consciousness II”, Tucson, AZ. Lincoln, N. J. and J. C. Rath. 1986. Phonology, Dictionary and Listing of Roots and Lexical Derivates of the Haisla Language of Kitlope and Kitimaat [sic]. (2 vols.) Ottawa: National Museum of Man Mercury Series. Canadian Ethnology Service: Paper No. 103. Lindau, M. 1985. “Testing a model of intonation in a tone language.” ULCA Working Papers in Phonetics 61: 26–39. Lindblom, B. 1986. “Phonetic universals in vowel systems.” In Experimental phonology, J. J. Ohala and J. J. Jaeger (eds), 13–44. Orlando: Academic Press. Liu, S. 1996. “Landmark detection for distinctive feature-based speech recognition.” Journal of the Acoustical Society of America 100 (5): 3417–3430. Mahajan, A. 1990. On the A–A¢ distinction. Ph.D. dissertation, MIT. Mark, Y. 1978. Gramatik Fun der Yidisher Shprakh. New York: YIVO Institute for Jewish Research. Massam, D. 2001. “Pseudo noun incorporation in Niuean.” Natural Language and Linguistic Theory 19: 153–197. Mathiot, M. 1991. “The Reminiscences of Juan Dolores, an Early O’odham Linguist.” Anthropological Linguistics 33: 233–315. Matthewson, L. 1998. Determiner Systems and Quantificational Strategies: Evidence from Salish. The Hague: Holland Academic Graphics. Matthewson, L. 1999. “On the interpretation of wide scope indefinites.” Natural Language Semantics 7: 79–134.

References 363

Mattina, N. 1996. Aspect and Category in Okanagan Word Formation. Ph.D. dissertation, Simon Fraser U. McCarthy, J. and A. Prince. 1993a. “Generalized alignment.” In Yearbook of Morphology 1993, G. Booij & J. van Marle (eds), 79–153. Dordrecht: Kluwer. McCarthy, J. and A. Prince. 1993b. “Prosodic morphology I: constraint interaction and satisfaction.” ms, UMass, Amherst and Rutgers U. McCarthy, J. and A. Prince. 1995. “Faithfulness and reduplicative identity.” UMass Occasional Papers in Linguistics 18: 249–384. McChombo, S. 2001. “Effects of head-marking on constituent order in Chichewa.” Paper presented at the LFG2001 conference, Hong Kong. McDonough, J. 1996. “Epenthesis in Navajo.” In Athabaskan language studies: Essays in honor of Robert W. Young, E. Jelinek, S. Midgette, K. Rice, and L. Saxon (eds), 235–257. Albuquerque: U. of New Mexico Press. McDonough, J. 1999. “Tone in Navajo.” Anthropological Linguistics 41: 503–540. McDonough, J. 2000a. “Athabaskan redux: Against the position class as a morphological category”. In Morphological Analysis in Comparison, O. Pfeiffer, W. Dressler, M. Pochtrager & J.Rennison Dressler, (eds.), 155–78. Amsterdam/Philadelphia: John Benjamins. McDonough, J. 2000b. “On the bipartite model of the Athabaskan verb.” In The Athabaskan Languages: Perspectives on a Native American Language Family, T. B. Fernald and P. R. Platero (eds.), 139–166. Oxford: Oxford U. Press. McDonough, J., Ladefoged, P., and George, H. 1993. “Navajo vowels and universal phonetic tendencies”. In UCLA Working Papers in Phonetics 84: 143–150. Meinunger, A. 2000. “Topicality and Agreement”. In Formalism and Functionalism in Linguistics. Volume 2: Case Studies. M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley (eds.) pp 203–220. Amsterdam/Philadelphia: John Benjamins. Miller, G., and Chomsky, N. 1963. “Finitary models of language users.” In Handbook of Mathematical Psychology, Volume 2, R. D. Luce, R. R. Bush and E. Galanter (eds), 419–491. New York: John Wiley and Sons. Mithun, M. 1984. “The evolution of noun incorporation.” Language 60: 847–95. Mithun, M. 1986. “On the nature of noun incorporation.” Language 62: 32–38. Montague, R. 1973. “The proper treatment of quantification in ordinary English.” In Formal Philosophy, R. Thomason (ed) pp. 247–270. New Haven: Yale U. Press. Moore, P. 2001. Point of View in Kaska Historical Narratives. Ph.D. dissertation, U. of Indiana. Moore, P. and A. Wheelock (eds.). 1990. Wolverine Myths and Visions. Dene Traditions from Northern Alberta. Lincoln: U. of Nebraska Press. Moro, A. 1997. The Raising of Predicates. Cambridge: Cambridge U. Press. Mostowski, A. 1979. “On a generalization of quantifiers.” In Foundational Studies: Selected Works, Volume II. Amsterdam: North-Holland. Murasugi, K. 1992. Crossing and Nested Paths: NP Movement in Accusative and Ergative Languages. Ph.D. Dissertation, MIT. Ndayiragije, J. 1999. “Checking economy.” Linguistic Inquiry 30: 399–444. Neundorf, A. 2000. “The Navajo simple passive.” Working Papers on Endangered and Less Familiar Languages 3, 117–128. Newmeyer, F. 1998. Language Form and Language Function. Cambridge: MIT Press. Newmeyer, F. 2002a. “Optimality And Functionality: A Critique of Functionally-Based Optimality-Theoretic Syntax.” Natural Language and Linguistic Theory 20: 43–80. Newmeyer. F. 2002b. “A Rejoinder to Bresnan and Aissen” Natural Language and Linguistic Theory 20: 97–99

364 References

Nicholas, C. D., & Bever, T. G. (In Prep). “The aesthetics of visual-form: The golden section proportion enhances depth perception.” ms. U. of Arizona. Nichols, J. 1986. “Head-marking and dependent-marking grammar.” Language 62: 56–119. Nichols, L. 2001. The Syntax and Morphosyntax of Propositional Attitude Complementation. ms, Harvard U. Noonan, M. 1999. “Non-structuralist syntax.” Functionalism and Formalism in Linguistics, vol.1. M. Darnell, E. A. Moravcsik, F. Newmeyer, M. Noonan and K. M. Wheatley (eds.), 1–36. Amsterdam/Philadelphia: John Benjamins. Norušis, M. J., and SPSS Inc. 1994. SPSS professional statistics 6.1. Chicago: SPSS Inc. Oehrle, R. 1976. The Grammatical Status of the English Dative Alternation. Ph.D. Dissertation, MIT. Ohala, D. 1996. Clusters and Constraints in Phonological Acquisition. Ph.D. Dissertation, U. of Arizona. Oostendorp, M. van. 1995. Vowel Quality and Phonological Projection. Ph.D. Dissertation, Catholic U. of Brabant. Öztürk, B. 2001. “Nominalization in Turkish.” ms, Harvard U. Parsons, T. 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge: MIT Press. Partee, B. 1991. “Topic, focus and quantification.” Proceedings of SALT 1: 159–188 Payne, D. 1987. “Information structuring in Papago narrative discourse.” Language 63: 783–804. Payne, D. 1992. “Nonidentifiable information and pragmatic order rules in ‘O’odham.” In Pragmatics of Word Order Flexibility, D. L. Payne (ed), 137–166. Amsterdam/Philadelphia: John Benjamins. Percus, O. 1993. “Mapping indefinites.” Generals paper, ms. MIT. Pesetsky, D. 1987. “Wh-in-situ: movement and unselective binding.” In The Representation of (In)definiteness, E. Reuland and A. ter Meulen (eds), 98–129. Cambridge: MIT Press. Pesetsky, D. 2000. Phrasal Movement and its Kin. Cambridge: MIT Press. Pesetsky, D. and Torrego, E. 2001 “T-to-C movement: Causes and consequences.” In Ken Hale: A Life in Language, M. Kenstowicz (ed). 355–426. Cambridge: MIT Press Philips, C. 1996. Order and Structure. Ph.D. dissertation, MIT. Pierrehumbert, J., and Beckman, M. 1988. Japanese Tone Structure. Cambridge: MIT Press. Pols, L. C. W. 1977. Spectral Analysis and Identification of Dutch Vowels in Monosyllabic Words. Soesterberg: Institute for Perception TNO. Pols, L. C. W. 1979. “Some spectral and perceptual measurements on Dutch diphthongs.” In Anniversaries in Phonetics: Studia Gratulatoria Dedicated to Hendrik Mol, 270–285. Amsterdam: Institute of Phonetic Sciences. Pols, L. C. W., Tromp, H. R. C. and Plomp, R. 1973. “Frequency analysis of Dutch vowels from 50 male speakers.” Journal of the Acoustical Society of America 53: 1093–1101. Pols, L. C. W., van der Kamp, L. J.Th. and Plomp, R. 1969. “Perceptual and physical space of vowel sounds.” Journal of the Acoustical Society of America 46: 458–467. Poser, W. 1984. The Phonetics and Phonology of Tone and Intonation in Japanese. Ph.D. dissertation, MIT. Potter, B. 1997. Wh/Indefinites and the Structure of the Clause in Western Apache. Ph.D. dissertation, UCLA. Prince, A. and Smolensky, P. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. Ms, Rutgers U. and U. of Colorado. Progovac, L. 1993. “Non-augmented NPs in Kinande as negative polarity items.” In Theoretical Aspects of Bantu Grammar 1, S. Mchombo (ed), 257–270. Stanford: CSLI.

References 365

Pulleyblank, D. 1988. “Underspecification, the Feature Hierarchy and Tiv Vowels.” Phonology 5: 299–326. Rackowski, A. and Travis, L. deM. 2000. “V-initial languages: X or XP movement and adverbial placement.” in The Syntax of Verb Initial Languages. A. Carnie and E. Guilfoyle (eds.) pp. 117–141. Oxford: Oxford U. Press. Reinhart, T. 1997. “Quantifier scope: How labor is divided between QR and choice functions.” Linguistics and Philosophy 20, 335–397. Reinhart, T. 1998. “Wh-in-situ in the framework of the Minimalist Program.” Natural Language Semantics 6, 29–56. Reinholtz, C. and K. Russell. 1994. “Quantified NPs in pronominal argument languages: evidence from Swampy Cree.” Proceedings of NELS 25.1: 389–403. Rice, C. 1997a. “Phonology outranks morphology: Chaucerian past participles and the grammar of poetry.” Paper presented at HILP 3. Amsterdam. Ms., U. of Tromsø. Rice, C. 1997b. “Ranking components: the grammar of poetry.” In Phonology in Progress — Progress in Phonology, G. Booij and J. van de Weijer (eds), 321–332. The Hague: Holland Academic Graphics. Rice, K. 1989a. “The phonology of Fort Nelson Slave stem tone: syntactic implications.” In Athabaskan Linguistics: Current Perspectives on a Language Family, E.-D. Cook and K. Rice (eds), 265–316. The Hague: Mouton. Rice, K. 1989b. A Grammar of Slave. Berlin: Mouton de Gruyter. Rice, K. 1993. “The structure of the Slave verb” in Studies in Lexical Phonology, S. Hargus and E. Kaisse (eds.), New York: Academic Press, pp. 145–172. Rice, K. 2000a. “Another look at the Athapaskan y-/b- pronouns: evidence from Slave for b- as a case marker.” Working Papers in Endangered and Less Familiar Languages 1: 109–128. Rice, K. 2000b. “Monadic verbs and argument structure in Ahtna, Navajo, and Slave.” In The Athabaskan Language. Perspectives on a Native American Language Family. Theodore Fernald and Paul Platero (eds.). 167–199. Oxford: Oxford U. Press. Rice, K. 2000c. Morpheme Order and Semantic Scope. Word formation in the Athapaskan verb. Cambridge: Cambridge U. Press. Rice, K. and L. Saxon. 1991. “A structural analysis of *y in Athapaskan.” Paper presented at the Athapaskan Conference, U. of California at Santa Cruz, Santa Cruz, California. July 1991. Rice, K. and L. Saxon. 2001. “The y-/b- pronouns in Athapaskan languages: perspectives on content.” Talk presented at WAIL/SSILA. U. of California, Berkeley. July 2001. Rice, K. and L. Saxon. forthcoming. “Comparative Athapaskan syntax: arguments and projections.” In The OUP Handbook of Comparative Syntax, G. Cinque and R. Kayne (eds.) Oxford U. Press. Richards, N. 1997 What Moves Where in Which Language? Ph.D. dissertation, MIT. Richards, N. 2002 Movement in Language. Oxford: Oxford U. Press. Rizzi, L. 1986a. “Null objects in Italian and the theory of pro.” Linguistic Inquiry 17: 501–557. Rizzi, L. 1986b. “On the status of subject clitics in Romance.” In Studies in Romance Linguistics, O. Jaeggli and C. Silva-Corvalan (eds), 391–420. Dordrecht: Foris. Rizzi, L. 1990. Relativized Minimality. Cambridge: MIT Press. Rizzi, L. 1997. “The fine structure of the left periphery.” In Elements of Grammar, L. Haegeman (ed), 281–337. Dordrecht: Kluwer Academic Publishers Roberts, T. 1999. “Grammatical relations and ergativity in St’át’imcets (Lillooet Salish).” International Journal of American Linguistics 65: 275–302. Rochemont, M. 1986. Focus in Generative Grammar. Philadelphia: John Benjamin. Rooth, M. 1999. “Association with focus or association with presupposition?” In P. Bosch and R. van der Sandt (eds.), Focus, Cambridge: Cambridge U. Press. pp. 232–244.

366 References

Ross, J. R. 1967. Constraints on Variables in Syntax. Ph.D. dissertation: MIT Rothstein, S. 1995. “Adverbial quantification over events.” Natural Language Semantics 3: 1–33. Rudin, C. 1988. “On multiple questions and multiple WH fronting.” Natural Language and Linguistic Theory 6, 445–502. Saffran, J. R., Aslin, R. N., and Newport, E. L. 1996. “Statistical learning by 8 month old infants.” Science 274: 1926–1928. Sandoval, M. and E. Jelinek. 1989. “The bi-construction and pronominal arguments in Apachean.” In Athapaskan Linguistics. Current Perspectives on a Language Family. E-D Cook and K. Rice (eds.), 335–377. Berlin: Mouton de Gruyter. Sapir, E. 1925. “Pitch accent in Sarcee.” Journal de la Societé des Americanistes de Paris, 17: 185–205. Sapir, E. and Hoijer, H. 1967. The Phonology and Morphology of the Navajo Language. U. of California, Berkeley. Sasse, H.-J. 1987. “The thetic/categorical distinction revisited.” Linguistics 25.511–580. Saxon, L. 1986. The Syntax of Pronouns in Dogrib (Athapaskan): Some Theoretical Consequences. Ph.D. dissertation, UCSD. Saxon, L. 1989. “Lexical versus syntactic projection. The configurationality of Slave.” In Athapaskan Linguistics: Current Perspectives on a Language Family. E-D Cook and K. Rice (eds.), 379–406. Berlin: Mouton de Gruyter. Saxon, L. 2001. “On two OSV constructions in Navajo: beyond Subject–Object Inversion.” To appear in Proceedings of the Workshop on Structure and Constituency in the Languages of the Americas. Vancouver: UBC Working Papers in Linguistics. Saxon, L. and K. Rice. 1993. “On subject–verb constituency: Evidence from Athapaskan languages.” Proceedings of the West Coast Conference on Formal Linguistics 11: 434–450. Saxton, D and Saxton, L. 1973. O’otham Hoho’ok A’agitha: Legends and Lore of the Papago and Pima Indians. Tucson: U. of Arizona Press. Saxton, D. 1982. “Papago”. In Studies in Uto-Aztecan Grammar 3, R. Langacker (ed), 92–266. Arlington: SIL Publications in Linguistics. Seiler, H.-J. 1977. Cahuilla Grammar. Banning, CA: Malki Museum Press. Selkirk, E. P. 1984. Phonology and Syntax: The Relation between Sound and Structure. Cambridge: MIT Press. Sgall, P., E. Hajicˇová, and J. Panevová. 1986. The Meaning of the Sentence in its Semantic and Pragmatic Aspects. J. L. Mey (ed), Dordrecht: Reidel — Prague: Academia. Slobin, D. I., and Bever, T. G. 1982. “Children use canonical sentence schemas: A crosslinguistic study of word order and inflections.” Cognition 12, pp. 229–265. Smith, C. 1991. The Parameter of Aspect. Dordrecht: Reidel. Smith, C. 1996. “Aspectual categories of Navajo.” International Journal of American Linguistics 62: 227–263. Soe, M. 1999. A Grammar of Burmese. Ph.D. Dissertation, U. of Oregon. Speas, M. 1984. “Navajo prefixes and word structure typology.” In MIT Working Papers in Linguistics 7: 86–109. Speas, M. 1990. Phrase Structure in Natural Language. Dordrecht: Kluwer Academic Publishers Speas, M. and E. Parsons Yazzie. 1996. “Quantifiers and the position of noun phrases in Navajo.” In, Athabaskan Language Studies. Essays in Honor of Robert W. Young. E. Jelinek, S. Midgette, K. Rice, and L. Saxon (eds.), 35–80. Albuquerque: U. of New Mexico Press. Stanley, R. 1969. The Phonology of the Navajo Verb. Ph.D. Dissertation MIT. Stechow, A. von. 1996. “Some remarks on choice functions and LF-movement.” In Proceedings of the Konstanz Workshop ‘Reference and Anaphorical Relations’, K. von Heusinger and U. Egli (eds), 1–23. Konstanz: Fachgruppe Sprachwissenschaft, Universität Konstanz.

References 367

Steele, S. 1990. Agreement and Anti-agreement in Luiseño. Dordrecht: Foris. Stevens, K. N. 1971. “The role of rapid spectrum changes in the production and perception of speech.” In Form and Substance. Phonetic and Linguistic Papers Presented to Eli FischerJørensen, L. L. Hammerich, R. Jakobson and E. Zwirner (eds), 95–101. Kobenhavn: Akademisk Forlag. Stevens, K. N. 1980. “Discussion.” In Proceedings of the Ninth International Congress of Phonetic Sciences Vol. 3, 185–186. Copenhagen: Institute of Phonetics, U. of Copenhagen. Stevens, K. N. 1985. “Evidence for the role of acoustic boundaries in the perception of speech sounds.” In Phonetic Linguistics: Essays in Honor of Peter Ladefoged, V. A. Fromkin (ed), 243–255. Orlando: Academic Press. Stevens, K. N. 1998. Acoustic phonetics. Cambridge: MIT Press. Stevens, K. N., and Blumstein, S. E. 1981. “The search for invariant acoustic correlates of phonetic features.” In Perspectives on the Study of Speech, P. D. Eimas and J. Miller (eds), 1–38. Hillsdale, NJ: Erlbaum. Stevens, K. N., and Keyser, S. J. 1989. “Primary features and their enhancement in consonants.” Language 65: 81–106. Strong, W. D. 1929. Aboriginal Society in Southern California. Reprinted 1972, with an introduction by Lowell John Bean. Banning, CA: Malki Museum Press. Szabolcsi, A. and Zwarts, F. 1993. “Weak islands and an algebraic semantics for scope taking.” Natural Language Semantics 1: 235–285. Takahashi, D. 1993. “Movement of wh-phrases in Japanese.” Natural Language and Linguistic Theory 11: 655–678. Tekieli, M. E. and Cullinan, W. L. 1979. “The perception of temporally segmented vowels and consonant-vowel syllables.” Journal of Speech and Hearing Research 22: 88–102. Tenny, C. 1994. Aspectual Roles and the Syntax Semantics Interface. Dordrecht: Reidel. Thom, M. and E. Blondin-Townsend (eds.). 1987. Nahecho Keh/Our Elders. Slavey Research Project, Government of the Northwest Territories. Thompson, C. 1996. “The history and function of the yi-/bi- alternation in Athabaskan.” In Athabaskan Language Studies: Essays in Honor of Robert W. Young. In E. Jelinek, S. Midgette, K. Rice, and L. Saxon (eds.), 81–100. Albuquerque: U. of New Mexico Press. Thompson, L. C. 1985. “Control in Salish grammar.” In Relational Typology, Trends in Linguistics (Studies and Monographs 28), F. Plank (ed), 391–428. The Hague: Mouton. Thráinsson, H. 2000. “Object Shift and Scrambling” in Handbook of Contemporary Syntactic Theory. M. Baltin and C. Collins (eds.) 148–202. Oxford: Blackwell. Touretzky, D. 1997. Common LISP: A Gentle Introduction to Symbolic Computation. Redwood City, CA: Benjamin/Cummings Townsend, D. J., and Bever, T. G. 2001. Sentence Comprehension. Cambridge: MIT Press. Travis, L. de M. 1984. Parameters and Effects of Word Order Variation. Ph.D. dissertation, MIT. Travis, L. de-M. 1994. “Inner Aspect” ms. McGill U. Tuttle, S. 1994. “The peg prefix, the foot and word minimum: evidence from Salcha”, ms. U. of Washington, Seattle. Tuttle, S. 1996. “Direct objects in Salcha Athabaskan.” In Athabaskan Language Studies: Essays in Honor of Robert W. Young, E. Jelinek, S. Midgette, K. Rice, and L. Saxon (eds), 101–121. Albuquerque: U. of New Mexico Press. Tuttle, S. 1998. Metrical and Tonal Structures in Tanana Athabaskan. Ph.D. dissertation, U. of Washington. Tuttle, S., and Kaisse, E. 1996. “Tonal Variation in Minto Athabaskan.” Papers from the Regional Meetings of the Chicago Linguistic Society 32: 347–361.

368 References

Ura, H. 2000. Checking Theory and Grammatical Functions in Universal Grammar. New York: Oxford U. Press. Venditti, J., Stone, M., Nanda, P. and Tepper, P. 2001. Toward an Account of Accented Pronoun Interpretation in Discourse Context: Evidence from Eye-tracking. Technical Report, Rutgers Center for Cognitive Science. Werle, A. 2000. “Semantic incorporation in Lillooet.” In The Thirty-Fifth International Conference on Salish and Neighbouring Languages, UBC Working Papers in Linguistics, S. Gessner and S. Oh (eds), 219–225. Vancouver: U. of British Columbia. Williams, E. 1980. “Predication.” Linguistic Inquiry 11: 203–238. Willie, M. 1991. Pronouns and Obviation in Navajo. Ph.D. dissertation, U. of Arizona. Willie, M. 1989. “Why there is nothing missing in Navajo relative clauses.” In E-D Cook and K. Rice (eds.), Athapaskan Linguistics. Current Perspectives on a Language Family. 407–437 Berlin: Mouton de Gruyter. Willie, M. 2000. “The inverse voice and possessive yi-/bi- in Navajo.” International Journal of American Linguistics 66. 360–382. Willie, M., and E. Jelinek. 2000. “Navajo as a Discourse Configurational Language.” In The Athabaskan Languages: Perspectives on a Native American Language Family, T. Fernald and P. Platero (eds), 252–287. Oxford: Oxford U. Press. Winter, Y. 1997. “Choice functions and the scopal semantics of indefinites.” Linguistics and Philosophy 20: 399–467. Wright, J. T. 1986. “The behavior of nasalized vowels in the perceptual vowel space.” In Experimental Phonology, J. J. Ohala and J. J. Jaeger (eds), 45–65. Orlando: Academic Press. Wright, M. 1984. “The CV skeleton and mapping in Navajo verb phonology.” Proceedings of the North-Eastern Linguistic Society Meeting 14: 461–477. Wyngaerd, G. van den. 1989. “Object shift as an A-movement rule.” MIT Working Papers in Linguistics 11: 256–271. Y˛o t’˛óh gondi/Slave stories. 1970. Calgary, Alberta: Summer Institute of Linguistics, Inc Young, R. W. and W. P. Morgan. 1980. The Navajo Language: A Grammar and Colloquial Dictionary. Albuquerque: U. of New Mexico Press. (2nd edition 1989) Young, R. W., W. P. Morgan and S. Midgette. 1992. Analytical Lexicon of Navajo. Albuquerque: U. of New Mexico Press. Young, R., and W.P. Morgan. 1987. The Navajo Language. Albuquerque: U. of New Mexico Press. Zaretski, A. 1929. Yidishe Gramatik. Vilna: Vilner Farlag. Zepeda, O. 1988. A Papago Grammar. Tucson: U. of Arizona Press. Zepeda, O. 1995. Ocean Power: Poems from the Desert. Tucson: U. of Arizona Press. Zepeda, O. 1997. JewedŸ ’I-Hoi: Earth Movements. Tucson: Kore Press. Zubizarreta, M-L. 1998. Prosody, Focus, and Word Order. Cambridge: MIT Press.

Name index

A Aissen, J. 8, 265 Alexiadou, A. 108, 120, 124–127 Anagnostopoulou, E. 108, 120, 124–127 Anttila, A. 319, 329–330 Archangeli, D. 227–244

Deising, M. 4–5, ff130, 135–153, 266–267, 272–273 Delancy, S. 294

B Bach, E. 5, 45–49 Baker, M. 5, 11–12, 16, 25, 52, 108–132 Barragan, L. ff226 Basilico, D. 273–277 Beckman, M. 191, 196 Bever, T. 5, 333–353 Bittner, M. 16, 37 Bobaljik, J. ff43 Boersma, P. 319, 323–325, 330, ff331 Boskovic, Z. 135, 301 Bowers, J. 222 Branigan, P. ff130 Bresnan, J. 108, 110, 113, 122, 265 Buránˇová, E. 5, 165–177

F Fitzgerald, C. 179–189 Frisch, S. 319, 328 Furui, S. 245

C Carnie, A. 1–8, 265–296, ff331 Chomsky, N. 12, 141, 293, 300, 302–305, 307–318 Cinque, G. 107, 111, 119 Coleman, J 319, 322–323, 328 Collins, C. ff130–131, 310 Croft, W. 1–3 D Davis, H. 5, 79–106, 127, ff132 de Hoop, H. 102–103

E Enderton, H. 310

G Gerdts, D. 88 Goldsmith, J. ff331 Golston, C. 179–189, 227–244 Grohmann, K. 153 Gunlongson, C. 50 Gussenhoven, C. 247 H Hajicˇová, E. 6, 165–177 Hale, K. 11–43, 179–189, 191–206 Halle, M. 125 Hammond, M. 7, 319–332 Harley, H. 1–8 Hayes, B. 325, 330, ff331 Hayes, R. ff331 Heath, J. 207, 217–220 Hill, J. 7, 207–226 Hoge, K. 138, 140–142 Hukari, T. ff.106 I Inkeles, S. 234, 236–238

370 Name index

Isaak, A. 265 Itô, J. 234, 236 J Jelinek, E. 3–5, 11–132, 220–225, 265–296 K Karimi, S. 7, 297–306 Keyser, S. J. 245 Kimenyi, A. 113 Kinerva, J. 113, 126 Kinyalolo, K. 109, 113, 117, 122, ff131 Kirchner, R. 234. 236 Kiss, K. 288, 298 Kisseberth, C. 234, 236 Koslow, A. 310, ff317 L Ladd, D. 192, 196 Ladefoged, P. 245 Langendoen, D. T. 5, 307–318, ff331 Lasnik, H. 301 Lindblom, B. 245 M Maddieson, I. 245 Marantz, A. 125 Matthewson, L. 5, 79–106, 127 McCarthy, J. 320 McDonough, J. 5, 15, 25, 191–206 Mchombo, S. 108, 110 Meinunger, A. 265 Mester, A. 234, 236 Mostowski, A. 310 Mutaka, P. 129 N Ndayiragije, J. 113, 287–292, ff296 Newmeyer, F. 1–3 Nichols, L. 155–164

O Ohala, D. ff331 Ohala, J. 245 P Percus, O. 268 Pesetsky, D. 6, 135, 141, 144, ff305 Pierrehumbert, J. 192, 196, 319, 322–323, 328 Pols, L. 247 Prince, A. 320, ff331 Progovac, L. 111 R Reinhart, T. 135, 139, 146–148 Rice, K. 5, 15–18, 26, 34–35, ff43, 50–78, 127–128, 141, ff132 Richards, N. 6, 299 Rizzi, L. 111–112, ff130 S Saxon, L. 50–78 Sgall, P. 5, 165–177 Smith, C. 15 Smolensky, P. 320, ff331 Speas, M. 15–16, 25, 50 Stevens, K. 245 T Touretzky, D. ff318 Travis, L. 269, 271 W Warner, N. 7, 245–261, ff331 Willie, M. 15, 50, 191–206 Z Zepeda, O. 179, 185–179 Zoll, C. 234, 236–238

Subject index

A A-bar movement (È-movement) 316 accent 203 accusative case 126 active 82 adjunct 52, 87–88, 94, 310, 313 adjunction 26, 116–118, 128, 312, 317 aesthetics 348 affectedness 273 agent-first order 338 agreement 51–78, 79–132, 207–226 anaphor 311 animacy hierarchies 282–284, 341–343 animal language 334–337 antecedent 311 anti-agreement 84 anti-focus 287–292 Apachean 70–77 applicative ff295 Arabic 272, ff296 argument hierarchies 265–295 argument structure 191–206, 310–313, 316–317 argument 52 asyndetic coordination 314 Athapaskan 11–47, 51–78 attachment, see low attachment or high attachment attitude evaluation 155–164 attitude nominals 6, 159–162 Attract Closest 141, 149 Australian languages 108 Austronesian 271 auxiliary constructions 122, ff131

B Babine Witsuwit’en 72, 76 Bantu languages 107–132 bipartite phonological skeleton 25–43 Burmese 6, 161–162 C canonical order 338 Cantonese 322 Case Binding 16, 31 case features 129 case split 281–292 Case, inherent 101 Case, structural 80–81, 107–132, 303 Case, weak 102 Categorial Grammar ff8, 45 causative 21–23, 31 c-command 311–312 Chichewa 108–110, 121, 126, ff131 choice functions 135, 146–148 cleft 123 clitics 107, 126, 272 complement 155, 310–311 Complementizer (C) 117, 315 complex NPs 155–164 computational limitation 335 conceptual-intentional interface 293, 309, 311 configurationality 108–132 conjunction, see also coordinate structure constraint ranking 227–244, 337 constraint Tableau 320 contrastive focus, see focus

372 Subject index

coordinate structure 313 coordination, see also conjunction, asyndetic coordination or syndetic coordination co-phonologies 236 coreference 173–177 Croatian, see Serbo-Croatian Cupeño 7, 207–226 Czech 6, 165–177 D dative shift 273–281 dependency syntax 165–177 derivation 307–308, 312 determiner 95–96, 98, Differential Object Marking (DOM) 271 Direct Optimality Theory (DOT) 227–244 directives 81–84 discourse genre 179–189 Discourse linking (D-linking) 135, 138 discourse 173–177 disharmony 324 dislocation 52, 108–132 ditransitive 82 D-linking, see Discourse linking Donald Duck problem 146–147 Dutch 245–261 E economy 142 end point 279 English 113, 125, 271, 275–277, ff295, 322 EPP, see Extended Projection Principle ergativity 284–287 event variable 157–159 evolution 344 exceptional forms 227–244 expletive 39, 124 Extended Projection Principle (EPP) 124–125, 129, ff132 300, ff305 Extension Condition ff318 external list merge, see list merge external pair merge, see pair merge external set merge, see set merge

extraction 80, 88, 91–94, ff105 F factorial typology 321 Farsi, see Persian features, phonetic 259 features, uninterpretable 7, 297–306 finite-state device 315 Finnish 329 Focus head 116–117 focus particle 115, 123 focus 115, ff131, 165–177, 191–206, 268, 287–292, 297–306 formalism 1–8 French 124 front rounded vowels 246 Full Interpretation (FI) 301 Functional Optimality Framework (FOT) 265 functional projection 313, 316 functionalism 1–8, 333–352 G generalized quantifier 46 German 123 golden section rectangle 348–350 Gradual Learning Algorithm ff331 grammaticality 322–323 grammaticization 292–296 Greek 120–121, 125–127, ff305 H Haisla 48 head to head movement 11–12, 33, 40ff318 head 310, 312–313 head-marking languages 107 hermit crab grammaticalization 207–226 Hiaki, see Yaqui hierarchical clustering analysis 246–253 high attachment 313 Hindi 122 historical morphology 217–220 Hopi 35–38

Subject index 373

I Icelandic 122 identificational focus, see focus impersonal construction 113, 125 inchoative 23 incorporation 12, 16, 33–40, 94–100 indefinite NPs 111, 119, 121 indirectives 81 Indo-European languages 108, 113, 120–121 information structure 265–296 inherent case, see Case innate features 170 Interface Condition (IC) 303 internal list merge, see list merge internal pair merge, see pair merge internal set merge, see set merge interrogatives 191–206 intonation break 115, 128, ff130, 315 intonation 191–206 intranstive 82, 95 inversion 118 J Japanese ff305 Jiwarli 108 K Kilega 109, ff131 Kinande 107–132 Kirundi 287–292 Korean ff305 L label 310 labile verb 23–24 lambda operator 46 language acquisition 339, 347 language of thought 337 language variation 292–296 latinate verbs ff295 left dislocated 13 LF, see Logical form Lillooet Salish, see St’át’imcets

LISP (common) ff318 list merge 310–311 locative inversion 113, 118, 122, 124–126, ff130 Logical Form (LF) 145–150, 300, long distance movement 317 low attachment 313 Lummi 79, 277–278, 284–287 M mapping principle 6, 265–295 Mapudungan ff130 Mayan 271 meaning 334–337 merge, see also set merge, pair merge, union merge, list merge 7, 12, 14, 33, 307–318 middle 82 Minimalist Program 6, 297–318 modularity 333 Mohawk 35–36, 107, 109, 112, 129 Montague Grammar 45 multidimensional scaling analysis 246, 253–256 multiple questions, see questions, multiple multiple specifiers 143 N names 47–48 narratives 179–189 narrow syntax 307, 309, 317, ff318 Navajo 11–43, 50, 74–75, ff132, 191–206, 278–279, 282–284, 341 nesting 315 nominalization 93–94 nominative case 124–125 non-configurational 3–4, 107 non-referential NPs 112, 125–126 nonsense words 319, 322, noun incorporation, see incorporation NP-movement 122 nuclear scope 266–296 number 207–226

374 Subject index

O object agreement 111 object marking 81–83 object shift ff130, 271 objectives 81 objects 67–70, 109–110, 118 oblique 31–32, 35, 84–85, 101, 106 One Nominal Interpretation (ONI) 88 ONI, see One Nominal Intrepretation operator 171 Optimality Theory 7, 179–189, 227–244, 265, 319–332, 337, 340 optionality 297–306 Oromo 282 OSV order 128 OVS order 287–292 P PAH, see Pronominal Argument Hypothesis pair merge 307–318 Papago, see Tohono O’odham parameter 122–123, 127, ff131, 344 partial configurationality 108–132 participles 122 partitive case 125 passive 85, 93 perceived movement 345 Persian 7, 297–306 person 285 PF, see Phonological Form phase 265–295 Phonetic Form, see Phonological Form phonetics 191–206 Phonological Form (PF) 303 phonotactics 319–332 phrase structure 307–318 pleonastic 113, see also expletive poetry 179–189 Polysynthesis Parameter 129 polysynthetic languages 120, 128 possessive constructions 64–67 post-positional phrases 59–64 Prague Dependency Treebank 6, 165–177 Prague School 165–177

predicate 117–118, 310 predication 47 prespecification 236–237 presupposition 265–296, 341–342 preverb 17–43 Principle of Full Interpretation, see Full Interpretation pro 117–118, 125 probabilistic ranking 319–332 pro-drop 110, 123–124, 128 Pronominal Argument Hypothesis (PAH) 3–132 pronominal argument 205 pronominal 118 propositional attitudes 155–164 prosody 191–206 pseudo-coordinate 315 pseudo-incorporation ff.104 Q quantified NPs 68–70, 111–112, 119–120 quasi objects 79–106 questions, multiple 135–153, 301 R ranking spreading 324 ranking, strict 320 ranking, variable 227–244, 324 rapid perceptibility 245–261 readjustment rules 314 recursion 307, 315, 317, 346 relative clauses 54–56 representational conflict 348 restrictor 266–296 reversal construction 113 rhythm 179–189 right adjunction 12 S Salcha Athabaskan 35 salience 174 Salish 79–106, 127–128 scope 97–100, 119–121, 127 Scots Gaelic 271 scrambling 141, 271

Subject index 375

semantic constraints 339 semantic incorporation, see incorporation semantic types 45–49 sentence comprehension 339 Serbian, see Serbo-Croatian Serbo-Croatian 301 serial response 336 set merge 307–318 Slave 34–35, 51–78, 128 Sonority Hierarchy 322 SO-reversal, see subject-object reversal SOV 204–205 Spanish 120–121, 125–127 specifier 310–312 St’át’imcets 6, 79–106, 127–128 Straits Salish, see Lummi stress 179–189, 203 strict ranking, see ranking strong minimalist thesis (SMT) 307, 309 structural Case, see Case Subjacency Condition 316–317 subject agreement 112–113 subject-object reversal 115, 118, 124–125, ff130, subjects 111, 113, 116–119 substitution 316 superiority effect 135, 139–145 syndetic coordination 314 syntactic categories 45–49 T Tense (T) 121, 123–124 Theta Criterion 123, ff131 Tiv 7, 227–244 Tohono O’odham 6, 179–189 tone 193–194 topic 165–177, 268 Topic head ff130 topicalization 56–58 Topic-Focus articulation 6, 165–177 trace 316 transitivity 81–84, 207–226 trochees 181 truth value 45

tucking in 141, 299 Turkish 161, 271–272 U UG, see Universal Grammar unaccusative 39, 126 unergative 20, 24, 113, 126 uninterpretable features, see features union merge ff318 Universal Grammar (UG) 326, 329 universals 245–261, 351 V variable ranking, see ranking verb raising ff296 Visibility Principle 303 voice 281–292 VOS order 90, 121 vowel features 227–244, 259 vowel height 227–244, 251–252 vowel inventories 227–244, 245–261 vowel length 252–253 vP ff132 VSO order 90, 121, 123, 128 W Warlpiri 127 weak Case, see Case Weak Crossover Condition 112 well-formedness 319 wh-adverbials 139 wh-in-situ 145 wh-movement 302, 316 wh-phrases 111 word order, see also OVS, OSV, SVO, SOV, VSO, VOS 58, 110, 113, 179–189 Y Yaqui 7, 220–221, 273, ff295 Yiddish Multiple Fronting Constraint 141 Yiddish 6, 135–153 Yoeme, see Yaqui Z zero elements 47

In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles have been published thus far, or are scheduled for publication: 1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene Malige-Klappenbach. 1980. 2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. 3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. 4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. 5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barriers. 1990. 6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. 7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. 8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. 9. GELDEREN, Elly van: The Rise of Functional Categories. 1993. 10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics. 1995. 11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995. 12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal Ideas. Linguistic studies in the minimalist framework. 1996. 13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and Typological Variation. 1997. 14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds): Materials on Left Dislocation. 1997. 15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and pro-drop. 1999. 16. LIU, FENG-HSI: Scope and Specificity. 1997. 17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward Movement. 1997. 18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. 19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. 20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. 21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. 22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Movement in the Determiner Phrase. 1998. 23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. 24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999. 25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999.

26. ACKEMA, Peter: Issues in Morphosyntax. 1999. ° 27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. 28. HERMANS, Ben and Marc van OOSTENDORP (eds): The Derivational Residue in Phonological Optimality Theory. 1999. 29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 1999. 30. BEUKEMA, Frits and Marcel den DIKKEN (eds): Clitic Phenomena in European Languages. 2000. 31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000. 32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds): The Syntax of Relative Clauses. 2000. 33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000. 34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. 35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO. 2000. 36. GERLACH, Birgit and Janet GRIJZENHOUT (eds): Clitics in Phonology, Morphology and Syntax. 2000. 37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds): Wh-Scope Marking. 2000. 38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000. 39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and Interpretability. 2000. 40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton van der WOUDEN (eds): Perspectives on Negation and Polarity Items. 2001. 41. ZELLER, Jochen : Particle Verbs and Local Domains. 2001. 42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and ergativity. 2001. 43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001. 44. TAYLAN, Eser E. (ed.): The Verb in Turkish. 2002. 45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds): Issues in Formal German(ic) Typology. 2002. 46. PANAGIOTIDIS, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. 47. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eds): Modality and its Interaction with the Verbal System. 2002. 48. ALEXIADOU, Artemis, Elena ANAGNOSTOPOULOU, Sjef BARBIERS and HansMartin GAERTNER (eds): Dimensions of Movement. From features to remnants. 2002 49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. 2002. 50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. 51. GERLACH, Birgit: Clitics between Syntax and Lexicon. 2002. 52. SIMON, Horst J. and Heike WIESE (eds): Pronouns – Grammar and Representation. 2002.

53. ZWART, C. Jan-Wouter and Werner ABRAHAM (eds): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26-27, 2000)(Workshop). 2002. 54. BAPTISTA, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2002. 55. COENE, M. and Yves D'HULST (eds): From NP to DP. Volume 1: The syntax and semantics of noun phrases. n.y.p. 56. COENE, M. and Yves D'HULST (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. n.y.p. 57. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. n.y.p. 58. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. n.y.p. 59. DEHÉ, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. 60. TRIPS, Carola: From OV to VO in Early Middle English. 2002. 61. SCHWABE, Kerstin and Susanne WINKLER (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. 62. CARNIE, Andrew, Heidi HARLEY and Mary WILLIE (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003.

Formal Approaches to Function in Grammar: In Honor of Eloise Jelinek (Linguistik Aktuell Linguistics Today, LA 62)

Causatives in Minimalism (Linguistik Aktuell Linguistics Today)

Continuity and Change in Grammar (Linguistik Aktuell Linguistics Today)

Theoretical Approaches to Universals (Linguistik Aktuell Linguistics Today)

Diachronic Clues to Synchronic Grammar (Linguistik Aktuell Linguistics Today)

Diachronic Clues to Synchronic Grammar (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

Current Issues in Generative Hebrew Linguistics (Linguistik Aktuell Linguistics Today)

Experimental Pragmatics Semantics (Linguistik Aktuell Linguistics Today)

Multiple Wh-Fronting (Linguistik Aktuell Linguistics Today)

Parentheticals (Linguistik Aktuell Linguistics Today, 106)

Argument Structure (Linguistik Aktuell Linguistics Today)

Phase Theory (Linguistik Aktuell Linguistics Today)

The Emergence of Order in Syntax (Linguistik Aktuell Linguistics Today)

Cyclical Change (Linguistik Aktuell Linguistics Today)

Quantifier Scope in German (Linguistik Aktuell Linguistics Today)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

The Linguistics Enterprise: From knowledge of language to knowledge in linguistics (Linguistik Aktuell Linguistics Today)

Time and Again: Theoretical Perspectives on Formal Linguistics: In Honor of D. Terrence Langendoen (Linguistik Aktuell Linguistics Today)

Grammaticalization As Economy (Linguistik Aktuell Linguistics Today, LA 71)

The Function of Function Words And Functional Categories (Linguistik Aktuell Linguistics Today)

The Derivation of Anaphoric Relations (Linguistik Aktuell Linguistics Today)

The Copy Theory of Movement (Linguistik Aktuell Linguistics Today)

The Structure of Stative Verbs (Linguistik Aktuell Linguistics Today)

Movement Theory of Control (Linguistik Aktuell Linguistics Today)

Syntatic Aspects of Topic and Comment (Linguistik Aktuell Linguistics Today)

Adjectives: Formal Analyses in Syntax and Semantics (Linguistik Aktuell Linguistics Today)

The Grammar of Repetition: Nupe grammar at the syntax-phonology interface (Linguistik Aktuell Linguistics Today)

Imperative Clauses in Generative Grammar: Studies in honour of Frits Beukema (Linguistik Aktuell Linguistics Today, LA 103)

Theoretical and Crosslinguistic Approaches to the Semantics of Aspect (Linguistik Aktuell Linguistics Today)

Formal Approaches to Function in Grammar: In Honor of Eloise Jelinek (Linguistik Aktuell Linguistics Today, LA 62)

Causatives in Minimalism (Linguistik Aktuell Linguistics Today)

Continuity and Change in Grammar (Linguistik Aktuell Linguistics Today)

Theoretical Approaches to Universals (Linguistik Aktuell Linguistics Today)

Diachronic Clues to Synchronic Grammar (Linguistik Aktuell Linguistics Today)

Diachronic Clues to Synchronic Grammar (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

Current Issues in Generative Hebrew Linguistics (Linguistik Aktuell Linguistics Today)

Experimental Pragmatics Semantics (Linguistik Aktuell Linguistics Today)

Multiple Wh-Fronting (Linguistik Aktuell Linguistics Today)

Parentheticals (Linguistik Aktuell Linguistics Today, 106)

Argument Structure (Linguistik Aktuell Linguistics Today)

Phase Theory (Linguistik Aktuell Linguistics Today)

The Emergence of Order in Syntax (Linguistik Aktuell Linguistics Today)

Cyclical Change (Linguistik Aktuell Linguistics Today)

Quantifier Scope in German (Linguistik Aktuell Linguistics Today)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

The Linguistics Enterprise: From knowledge of language to knowledge in linguistics (Linguistik Aktuell Linguistics Today)

Time and Again: Theoretical Perspectives on Formal Linguistics: In Honor of D. Terrence Langendoen (Linguistik Aktuell Linguistics Today)

Grammaticalization As Economy (Linguistik Aktuell Linguistics Today, LA 71)

The Function of Function Words And Functional Categories (Linguistik Aktuell Linguistics Today)

The Derivation of Anaphoric Relations (Linguistik Aktuell Linguistics Today)

The Copy Theory of Movement (Linguistik Aktuell Linguistics Today)

The Structure of Stative Verbs (Linguistik Aktuell Linguistics Today)

Movement Theory of Control (Linguistik Aktuell Linguistics Today)

Syntatic Aspects of Topic and Comment (Linguistik Aktuell Linguistics Today)

Adjectives: Formal Analyses in Syntax and Semantics (Linguistik Aktuell Linguistics Today)

The Grammar of Repetition: Nupe grammar at the syntax-phonology interface (Linguistik Aktuell Linguistics Today)

Imperative Clauses in Generative Grammar: Studies in honour of Frits Beukema (Linguistik Aktuell Linguistics Today, LA 103)

Theoretical and Crosslinguistic Approaches to the Semantics of Aspect (Linguistik Aktuell Linguistics Today)

Recommend Documents