Aspects of the Theory of Clitics (Oxford Studies in Theoretical Linguistics)

Aspects of the Theory of Clitics OXFORD STUDIES IN THEORETICAL LINGUISTICS g e n e r a l e d i to r s: Southern Calif...

Author: Stephen R. Anderson

78 downloads 1353 Views 2MB 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

Aspects of the Theory of Clitics

OXFORD STUDIES IN THEORETICAL LINGUISTICS g e n e r a l e d i to r s: Southern California

David Adger, Queen Mary College London; Hagit Borer, University of

a dv i s o ry e d i to r s: Stephen Anderson, Yale University; Gennaro Chierchia, University of Milan; Rose-Marie Dechaine, University of British Columbia; Elan Dresher, University of Toronto; James Higginbotham, University of Southern California; Pat Keating, University of California, Los Angeles; Ruth Kempson, King’s College, University of London; James McCloskey, University of California, Santa Cruz; Gillian Ramchand, University of Tromsø; Maria-Luisa Zubizarreta, University of Southern California published 1 The Syntax of Silence by Jason Merchant 2 Questions and Answers in Embedded Contexts by Utpal Lahiri 3 Phonetics, Phonology, and Cognition edited by Jacques Durand and Bernard Laks 4 At the Syntax-Pragmatics Interface by Lutz Marten 5 The Unaccusativity Puzzle edited by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert 6 Beyond Morphology by Peter Ackema and Ad Neeleman 7 The Logic of Conventional Implicatures by Christopher Potts 8 Paradigms of Phonological Theory edited by Laura Downing, T. Alan Hall, and Renate Raffelsiefen 9 The Verbal Complex in Romance by Paola Monachesi 10 The Syntax of Aspect edited by Nomi Erteschik-Shir and Tova Rapopart 11 Aspects of the Theory of Clitics by Stephen Anderson i n p r e pa r at i o n Phi Syntax: A Theory of Agreement by Susana Bejar Aspect and Reference Time by Olga Borik Prosodic Morphology by Laura Downing Stratal Optimality Theory by Ricardo Bermúdez Otero Tense, Mood, and Aspect edited by Alessandra Giorgi, James Higginbotham, and Fabio Pianesi The Ecology of English Noun-Noun Compounding by Ray Jackendoﬀ A Natural History of Inﬁxation by Alan Chi Lun Yu The Oxford Handbook of Linguistic Interfaces edited by Gillian Ramchand and Charles Reiss [ published in association with the series]

Aspects of the Theory of Clitics STEPHEN R. ANDERSON

1

3

Great Clarendon Street, Oxford OX2 6DP Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With ofﬁces in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Stephen Anderson 2005 The moral rights of the author have been asserted Database right Oxford University Press (maker) First published 2005 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose this same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloguing in Publication Data Data available Typeset by SPI Publisher Services, Pondicherry, India Printed in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn ISBN 0–19–927990–x 978–0–19–927990–6 ISBN 0–19–927991–8 978–0–19–927991–3 (Pbk.) 10 9 8 7 6 5 4 3 2 1

Contents General Preface Acknowledgements

vii viii

1 Introduction

1

2 What is a Clitic?

9

2.1 2.2 2.3 2.4 2.5 2.6

Words, Clitics, and Afﬁxes Case I: An Introduction to Kw akw ’ala Clitics Dimensions of Clitic-hood Case II: How “Simple” are English Auxiliary Clitics? Special Clitics Clitics and Afﬁxes

3 The Phonology of Cliticization 3.1 3.2 3.3 3.4

Prosodic Structure Dimensions of Phonological Cliticization Prosodic Structure and Syntactic Structure Phonological Clitics and Cliticization in English

4 Special Clitics and their Grammar 4.1 Phenomenology 4.2 Special Clitics as the Morphology of Phrases 4.3 Some Examples 5 Theories of Special Clitics 5.1 The Nature of the Problem 5.2 Syntactic Theories of Clitic Placement 6 An Optimal Theory of Clitic Positioning 6.1 6.2 6.3 6.4

Special Clitics as Phrasal Morphology Second Position: Anchors and Domains “Endoclitics” Tagalog Second-Position Clitics

9 14 22 24 31 33 37 37 44 55 64 75 76 82 89 107 108 116 127 127 142 152 165

vi

Contents

7 Verb Second as Alignment 7.1 7.2 7.3 7.4 7.5 7.6

Verb Movement and Second Position Icelandic Kashmiri Breton Surmiran Rumantsch Conclusions

8 Pronominal Clitics 8.1 8.2 8.3 8.4

Pronouns and Agreement Clitics, Agreement, and Doubling Clitic Climbing Subject Clitics

9 Clause Structure and the Grammar of Incorporation 9.1 9.2 9.3 9.4 9.5

Introduction: Two Approaches to Noun Incorporation Fleshing Out the Syntactic and Lexical Accounts Noun Incorporation: Syntax or Lexicon? Denominal Verb Formation in West Greenlandic Conclusion

177 178 183 187 193 204 224 227 228 239 245 249 257 258 264 275 281 286

References

289

Index of Subjects Index of Names Index of Languages

303 311 315

General Preface The theoretical focus of this series is on the interfaces between subcomponents of the human grammatical system and the closely related area of the interfaces ´ between the different subdisciplines of linguistics. The notion of SinterfaceŠ has become central in grammatical theory (for instance, in ChomskyŠs recent Minimalist Program) and in linguistic practice: work on the interfaces between syntax and semantics, syntax and morphology, phonology and phonetics etc. has led to a deeper understanding of particular linguistic phenomena and of the architecture of the linguistic component of the mind/brain. The series covers interfaces between core components of grammar, including syntax/morphology, syntax/semantics, syntax/phonology, syntax/ pragmatics, morphology/phonology, phonology/phonetics, phonetics/speech processing, semantics/pragmatics, intonation/discourse structure as well as issues in the way that the systems of grammar involving these interface areas are acquired and deployed in use (including language acquisition, language dysfunction, and language processing). It demonstrates, we hope, that proper understandings of particular linguistic phenomena, languages, language groups, or inter-language variations all require reference to interfaces. The series is open to work by linguists of all theoretical persuasions and schools of thought. A main requirement is that authors should write so as to understood by colleagues in related subﬁelds of linguistics and by scholars in cognate disciplines. Clitics appear to live at the interfaces between morphology, syntax and prosody. In our eleventh volume, Stephen Anderson addresses the theory of cliticization, and argues that the properties of simple clitics are best captured in terms of independently motivated prosodic structures, couched within an OT framework. He then shows that special clitics can be understood as being the phrasal analogues of morphological afﬁxation, and that an OT account of these can be naturally extended to V2 constructions and pronominal clitics, with important implications for clause structure and the plausibility of head movement. David Adger Hagit Borer

Acknowledgements The work reported here has beneﬁted from the generous support over many years of the National Science Foundation, to whom I have been promising this monograph for all too long. First and foremost, therefore, I would like to acknowledge the support of awards SBR 95--14682 and BCS 98--76456 from the NSF to Yale University. I am extremely grateful for the conﬁdence the NSF, and especially Paul Chapin as director of the Linguistics Program, has shown in my work. A language that ﬁgures prominently in the present book is Kw akw ’ala, on which my research over several years in the 1970s was also supported by the NSF, under award BNS 78--15395 to UCLA, as well as by a grant from the Wenner-Gren Foundation for Anthropological Research. My study of Kwakw ’ala was made possible by the assistance of Sam Henderson, Daisy Moon, and Jim Henderson of Campbell River, and by Tommy and Emma Hunt of Victoria, as well as others in Campbell River, Quadra Island, Alert Bay, and Port Hardy, British Columbia. That work was also facilitated by access to manuscript materials of Franz Boas in the library of the American Philosophical Society, which I was able to make use of thanks to permission from the late Zellig Harris. Another language that ﬁgures in this book, primarily in Chapters 7 and 8, is Surmiran, a form of Rumantsch. My work on this language was ﬁrst supported by NSF award BCS 98--76456 and by grants from the Yale University Social Science Research Fund, and continues under NSF award BCS 04--18410 to Yale University. I am grateful to Petra Uffer for many hours of patient instruction in Surmiran, as well as to Ursus Baltermia and my other friends and consultants in Salouf and Savognin. It is a cliché of prefaces such as this one, that too many people have provided me with help, advice, suggestions, and corrections over the years in pursuing this research for it to be possible to enumerate them all without causing the reader’s eyes to glaze over—poor thanks, indeed. I would, however, like to thank the students who have participated in my seminars on clitics at Yale over a number of years, including especially Lizanne Kaiser and others whose work appears in Kaiser (1997). I would also like to acknowledge the help of Julie Legate, who helped to keep me a bit more honest over the course of a presentation of much of this material in the Fall of 2003 at Yale, and who has provided numerous comments and suggestions on the manuscript. Additional comments and assistance on portions of the manuscript from Geert Booij,

Acknowledgements

ix

Luigi Burzio, Andrew Carstairs-McCarthy, Sandra Chung, Jerry Sadock, and Ida Toivonen have been extremely valuable. Obviously, none of these people should be blamed for my mistakes or assumed to agree with my opinions. Thanks also to Michiya Kawai for his work on the indexes. Much of the actual writing of this book has been done with the generous support of a sabbatical leave from Yale. During that time, I have been a Visiting Scholar in the Department of Linguistics, School of Classics and Linguistics, at the University of Canterbury in Christchurch, New Zealand. I am extremely grateful to the University and the Department for their hospitality and for the generosity with which they have made their facilities available to me, providing me with a delightful atmosphere in which to carry out this work. Particularly pleasant have been frequent conversations with Andrew Carstairs-McCarthy both about matters treated here and about more general issues in morphology and phonology. I must also single out Paul Smolensky. Paul introduced me to the merits of an Optimality Theoretic approach to clitic phenomena in the course of a conversation years ago when I was seeking exactly the opposite: namely, a good argument for ignoring OT in favor of the rule-based analysis I had offered in earlier papers. As the reader will see, I think he was deeply right.

This page intentionally left blank

1 Introduction The literature of (generative) linguistics since the early 1960s is replete with monographs, specialized collections, articles, notes, and other research dealing with the properties of clitics—indeed, entire books (such as Nevis, Joseph, Wanner, and Zwicky 1994) are devoted to nothing but lists of references relevant to the study of clitics. And yet a search of several standard online dictionaries, including Webster’s Unabridged, turns up no mention of the word. The Oxford English Dictionary ﬁrst notes it in 1946, and deﬁnes it thusly: “An enclitic or proclitic.” Not a lot of immediate help there, but at least we have a link to a much older tradition, for the notions of enclitic and proclitic are thoroughly grounded in grammatical description dating back at least to the seventeenth century. The OED’s entry is supported by a quotation from Nida (1949: vii): “Many languages have elements that (1) combine phonologically with words with which they do not form morphological constructions, and (2) do not constitute derivational or inﬂectional formatives . . . The term ‘clitics’ may . . . be employed in this general meaning.” Etymologically, clitic is from Greek klinein ‘to lean,’ and refers to this tendency of clitics to depend on otherwise unrelated material. Our understanding thus far is essentially negative: a clitic is something that is not integrated into the sentence in the way “normal” words are, and/or not integrated into words in the way afﬁxes are. But despite its negative character, this is the basis not just of one but of two nearly independent research traditions. One of these, the older, is associated in linguistics primarily with traditional grammarians and Indo-Europeanists. This is essentially a phonological understanding, on which a (pro- or en-)clitic is a stressless “little” word that lacks independent accent, and that (as a result) depends prosodically on an adjacent word. The OED deﬁnes enclitic, for instance (following Liddell and Scott’s 1843 Greek-English Lexicon) as a word that “ ‘leans its accent on the preceding word’: in Greek grammar the distinctive epithet of those words which

2

Aspects of the Theory of Clitics

have no accent, and which (when phonetic laws permit) cause a secondary accent to be laid on the last syllable of the word which they follow. Hence applied to the analogous Latin particles -que, -ve, -ne, etc., and in mod[ern] use (with extension of sense) to those unemphatic words in other lang[uage]s that are treated in pronunciation as if forming part of the preceding word.” This was certainly what Jakob Wackernagel, the patron saint of clitics, meant in his classic paper (Wackernagel 1892) when he spoke of enclitics and proclitics. Such an interpretation is not at all controverted by the point of that paper, which was the further claim that clitics (deﬁned in this way) in the ancient Indo-European languages—and presumably also in Proto IndoEuropean—may huddle together in some particular location (in that case, in “second position,” after the ﬁrst word of the clause). For example, in the Homeric Greek example in (1.1), the two enclitic elements te ‘and’ and min ‘3sg. acc. pronoun’ both appear immediately after the ﬁrst full word of the sentence.1 (1.1)

polees=te=min e¯ r¯esanto hipp¯ees phoreein many-and-it prayed riders carry And many riders prayed to carry it (Iliad 4.143)

For Wackernagel, the occurrence of these words in this position followed directly from a property of their form: to wit, their lack of an (independent) accent. The fundamental sense of ‘clitic’ on this view, then, is a phonological one. Clitics are little words that lack accent of their own. Modern syntacticians, on the other hand, have their own distinct ideas about the notion of “clitics.” For them, the study of clitics is essentially the study of a particular class of pronominal elements (and others that behave like them), most commonly the Romance (conjunct) pronouns. This conception is particularly associated with a rich tradition of syntactic research originating in the work of Richard Kayne. In Kayne (1975) a number of distinctive properties are identiﬁed with the relevant subset of pronominals, including not only their distinctive position immediately preceding the verb (as opposed to other nominal expressions), but also their inability to be modiﬁed or conjoined. These properties go together with certain phonological and semantic idiosyncrasies, and appear to deﬁne a category of grammatical elements. For Kayne and those who have continued this line of research, the description of clitics is primarily a matter of the special syntax of these pronouns. 1 In examples, I will generally indicate clitics and their direction of attachment in boldface with “=” separating them from their host. This notation is not intended by itself to indicate anything about the speciﬁc phonological or syntactic consequences of such a relation, but only to highlight the items as clitics.

Introduction

3

That notion is extremely widespread. For example, when we consult the index of a recent reference grammar of Catalan (Wheeler, Yates, and Dols 1999) for “clitics,” we ﬁnd “see pronominal clitics.” This is not because Catalan has no other clitic elements of interest (in Chapter 3, we will see a class of unaccented prepositions in this language that have properties quite worth examining), but rather because syntacticians are particularly interested in elements such as those underlined in example (1.2). (1.2)

hi has anat tota sola? Te me n’ 2sg 1sg part there you have gone all alone Did you go off there all alone (on me)?

For researchers of this persuasion, what is remarkable about the underlined elements in (1.2) is not their lack of accent (a phonological property), but rather the fact that (in Catalan, as in most Romance languages) they all pile up immediately before the ﬁnite verb of the clause—a position in which corresponding non-clitic elements would not appear. This sense of “clitic” is thus a (morpho)syntactic one. My goal in this book is to explore both of these senses of clitic, their interrelations, and the implications of their analysis for the theory of grammar more generally. The general perspective will be one that grows out of the theory of A-Morphous Morphology (Anderson 1992), where clitics are regarded as the phrasal analog of (word-level) morphology. In more general terms, theoretically satisfying accounts and explanations will be sought within a view of grammar based primarily on systems of constraints (a version of Optimality Theory) rather than on rules. Much of the analysis on the morphosyntactic side is preﬁgured by that of Anderson (2000c), which is both extended here and complemented by more attention to the phonological side of the problem. The basic architecture of the book is determined by the distinction between the two logically independent senses of the notion of “clitic” introduced above. I begin in Chapter 2 by endeavoring to separate these by distinguishing simple clitics—referring to a purely phonological dimension of clitic-hood— from special clitics in more or less the sense of Zwicky (1977). I consider two systems that might be taken to illustrate simple clitics, whose behavior we can later compare and contrast with that of morphosyntactically unusual special clitics. The ﬁrst of these is a set of clitic elements from the Wakashan language Kw akw ’ala, whose properties (while initially remarkable in some respects) are relatively straightforward. The second case considered here yields a rather more complex story, however. It has become fairly common to cite English contracted auxiliary forms such as the ’s in Nixon’s the one as near canonical instances of simple clitics.

4

Aspects of the Theory of Clitics

Syntacticians, on the other hand, have developed a rich literature on these elements: if they are really so simple, it is remarkable that so many people have received Ph.D.s for dissertations devoted (entirely or in large part) to a discussion of intricate syntactic conditions on their occurrence. I explore these matters with the goal of seeing what a phonologically based analysis must account for if these elements are indeed to be called “simple” clitics. To say that one sense of the term “clitic” is to be identiﬁed with the phonology of the relevant elements does not, of course, actually provide in itself a theory of that dimension. In fact, the phonology of cliticization is closely bound up with the nature and status of the much broader theory of prosodic structure in language, and we can only understand phonological clitics by placing their behavior within this larger context. In Chapter 3, accordingly, I outline a view of prosodic categories and their relations within the framework of Optimality Theory, and use that to arrive at a characterization of the phonology of clitics, especially (but not exclusively) in English. I argue that the properties of English reduced auxiliaries do indeed fall within a phonological account, and thus that there is no obstacle to calling them “simple” clitics in the technical sense, despite the complexities of their behavior. Having given an account of the phonological dimension of cliticization, I then move on in Chapter 4 to the morphosyntax of special clitics. Starting from the classic descriptive generalizations of Zwicky, Klavans, Kaisse, and others (summed up in Anderson 1992) about the locations in which these are found, I will ask what kind of theory might be responsible for getting them there. I conclude on the basis of a number of considerations that morphology and not syntax furnishes the appropriate context for understanding them. Special clitics, that is, constitute the morphology of phrases—a suggestion that has often been made and which I try here to substantiate. Some morphology consists not of afﬁxes but of other sorts of change in shape (“Non-concatenative” morphology). There is a case to be made that the kind of function ﬁlled by clitics is sometimes realized by changes in shape other than the addition of segmentable material to a phrase, and instances of such “Non-concatenative” clitics are proposed and analyzed. And just as word-level morphology can be divided between inﬂection and derivation, there seem to be two corresponding classes of clitics: some that represent grammatical material, such as pronominal arguments; and some that represent more semantic content, such as discourse markers, various adverbials, etc. I discuss the extent to which this distinction among clitics is substantively parallel to that found in morphology. Among the various types of special clitics, by far the most interesting (in the sense of probative) are those that occur in second position within the

Introduction

5

domain to which they are relevant. In Chapter 5 I show why these clitics are so important for the development of an adequate theory of special clitic positioning. Proceeding from a fuller account of the Kw akw ’ala clitics considered earlier, I explore second-position clitics in a variety of languages. I consider a range of possible theories that have been proposed for the positioning of special clitics, and conclude that neither the syntax nor the phonology sufﬁces to describe the grammar of special clitics—a result that is of course entirely in accord with my earlier proposal that the nature of these elements is essentially morphological. I next proceed in Chapter 6 to elaborate a more precise account of the phrasal morphology that is responsible for the appearance of special clitics, within an Optimality Theoretic framework. A limited set of constraint types sufﬁce to describe these (and other possible special clitics) concisely and insightfully, as has also been argued in a series of papers by Géraldine Legendre in recent years as well as in Anderson 2000c. The OT-based account of “Clitics as Phrasal Afﬁxes” is compared with a variety of other theories: those postulating a purely syntactic account of cliticization, and those locating the unusual properties of special clitics in their phonology or in other properties of the interface between syntax and other parts of grammar. The system of second-position clitics in Tagalog provides a complex and nuanced example of a number of the properties discussed to this point. In Chapter 7, I turn from the analysis of clitics per se to an area of syntax that has not generally been taken within the generative tradition to be closely related. Here I pose the question of whether or not it was correct for Jakob Wackernagel, whose classic discussion of second-position clitics in the earliest Indo-European languages has been so often cited (if less often read), to suggest a connection between these clitics and another set of second-position phenomena, the verb-second regularities of Modern German and a number of other languages. I conclude that while Wackernagel’s own notion of the explanatory connection was undoubtedly incorrect, there is indeed a deep link, and the morphosyntactic apparatus used to describe second-position clitics can provide us with an account of verb second in German, Icelandic, Breton, and other languages as well. In the process, we learn some things about the way Optimality Theoretic mechanisms play out in the domain of syntax, where they have been less studied (at least until recently) than in phonology and morphology. Continuing this foray into the syntactic domain, I survey in Chapter 8 some of the substantive syntactic properties of the most widely studied class of clitics, those traditionally analyzed as pronominals. I begin by examining the nature of (predicate-argument) agreement, and comparing it with wellknown phenomena arising in the analysis of special clitics. This requires an

6

Aspects of the Theory of Clitics

elaboration of the analysis of the Morphosyntactic Representations of categories, in order to account for phenomena such as (the presence versus absence versus optionality of) clitic doubling, clitic climbing, and the like. While the bulk of the literature devoted to pronominal clitics focuses on object clitics, some languages (including several spoken in northern Italy and in nearby areas of Switzerland) also have special clitics referring to subjects. I explore their properties, including those of Surmiran and a range of “northern Italian dialects.” I discuss the signiﬁcance of the morphological approach to special clitics for the syntax of functional categories and the proposal that all such categories constitute syntactically autonomous heads (each with its own projection) in syntactic representation. The account of clause structure that results from viewing clitics in the way advocated here has a variety of other wider consequences. One of these involves the plausibility of an approach to incorporation phenomena based on syntactic movement, as developed for instance by Sadock, Baker, and others. If syntactic movement is not involved in incorporation constructions, as I argue, then the foundation for the whole apparatus of Head Movement collapses, and with it much of the plausibility of the elaborately articulated functional structure assumed by syntacticians since the late 1980s. This is a result which is quite in harmony with the picture of clause structure developed in the present work, which assumes that much functional content is present in the form of features on a limited number of structural categories, rather than as a set of functional heads in their own right. In Chapter 9, I explore these matters, and attempt to draw some general morals. From various remarks above it will perhaps be clear that the view of syntactic structure represented in this book does not line up completely with any of the established positions that have dominated discussion over the past decade or so. Apart from occasional turns of phrase, I do not adopt the speciﬁc assumptions that have emerged within the Minimalist Program for syntax, though I think that most of what I say could be transposed into that framework with little substantive alteration. I assume an overall picture involving basegenerated structures and operations of displacement relating these to surface form, subject to conditions including those of Binding Theory and the like in a way reminiscent of much work in the framework of Government and Binding or Principles and Parameters. I do not, however, follow the common development of that framework to include an increasing proliferation of structure based on functional categories. Much of this functional content is associated here with an elaboration of the feature structure of categories drawn from a somewhat more limited set, with considerably less “arborization” as a result. The representations I operate with

Introduction

7

are much more similar to surface forms, at least in their degree of internal articulation, than those to which students of generative grammar are typically introduced today, and involve much less movement within functional structure. In part, these differences reﬂect the fact that for my purposes, much of this functional structure is simply not relevant to the points at issue, and thus can usefully be abbreviated. It should also be clear, however, that there are real differences of principle involved. I assume (and attempt to argue for) the relevance of a constraint-based system of an Optimality Theoretic sort in the development of morphosyntactic structure. This is most prominent in relation to the principles that introduce special clitic elements into sentences, but also plays a role elsewhere (e.g., in producing “verb-second” patterns). This view is of a piece with the Optimality Theoretic interpretation of principles of prosodic structure which forms the basis of the account of phonological clitics offered in Chapter 3. This is probably the longest work published to date based on assumptions from Optimality Theory but which does not include a single constraint tableau. My principal excuse for that is the fact that had I included tableaux to illustrate the arguments made at each point where they would be relevant, the book would be much, much longer. But while I understand the value of these displays for conﬁrming the correctness of complex constraint rankings, I am not convinced of their more general perspicuity. Most of the constraint interactions proposed here are quite straightforward, and I hope they can be appreciated on the basis of the discussion. In any event, I leave the construction of formal tableaux as an exercise for the reader. I cannot hope to argue for all of the details of my (morpho)syntactic assumptions in this book, but I hope that I have made a case for some of the more noticeable deviations from current practice, and that in other instances I have at least made clear what my assumptions are so that others can see whether the differences matter to the real arguments.

This page intentionally left blank

2 What is a Clitic? The modern study of clitics within generative grammar can reasonably be seen as beginning with the survey provided by Zwicky (1977). A number of earlier papers had dealt with the special properties of these elements, especially in terms of their syntax; and after the fashion of the time, Zwicky’s paper was never formally published, but only distributed in mimeographed form. The distinctions proposed there, however, and the observations on which they were based, established a research agenda that has dominated discussion of clitics every since.

2.1 Words, Clitics, and Afﬁxes Zwicky’s classic paper starts from the question of how we distinguish words from afﬁxes, enumerating a number of properties in terms of which these differ. Like many such lists, this one contains some properties that might be regarded as deﬁnitional, mixed in with others that are simply common symptoms of an element’s status. Ordering: Afﬁx order within the word is quite rigid, while word order within phrases can vary. The degree of variation differs from language to language, of course, but virtually all languages allow some alternative word orders corresponding to the same essential meaning. In contrast, any variation we ﬁnd in the order of word-internal afﬁxes is virtually always correlated with a difference in sense. Internal sandhi: Different phonological adjustments may apply within versus across words. The distinction of (word-)internal sandhi from external sandhi goes back (as the etymology of the word sandhi suggests) to the ancient Sanskrit grammarians. It corresponds in large part to the difference between Lexical and Post-lexical phonology in more modern terms (cf. Kaisse and Shaw 1985). Binding: This is Zwicky’s term for the fact that some morphological elements can appear alone, while others only occur in combination. The latter (“Bound Morphemes”) are afﬁxes, while words are free.

10

Aspects of the Theory of Clitics

Construction with afﬁxes: If an element is in construction with an afﬁx, it must be either the base or another afﬁx. Zwicky uses this criterion to support the claim that an element in Madurese which occurs between two identiﬁable afﬁxes in a form whose base can be independently established must itself be an afﬁx. Rule immunity: Syntactic rules, such as deletion under identity, do not affect afﬁxes, since these are proper parts of words. Seen as a criterion for distinguishing words from afﬁxes, this is clearly based on strong Lexicalist assumptions about the relation between morphology and syntax (cf. Anderson (1992) and literature cited there). Accent: Elements that do not bear an independent accent are afﬁxes. If every morphologically discrete element in every language ﬁt nicely into one or the other of (exactly) two patterns with respect to these criteria, we might be content, but of course that is not true, and Zwicky identiﬁes three common types of “mixed” case. These, of course, are three varieties of things we might call clitics. On this account, then, a clitic is a linguistic element which the tests just given do not classify unambiguously as being either a word or an afﬁx. The three patterns identiﬁed in this way are then named special clitics, simple clitics, and bound words. Special clitics are characterized as unaccented bound forms that are variants of free forms (similar, that is, in sound and meaning to some non-clitic word); and which display “special” syntax. This might mean simply “different” syntax from the corresponding free form, as French Je la vois ‘I see her’ versus Je vois la femme de mes rêves ‘I see the woman of my dreams.’ In this case (the syntactician’s prototypical instance of a clitic), the pronoun precedes the ﬁnite verb, while a corresponding non-pronoun is completely excluded from that position, and instead appears post-verbally (where the pronoun, in its turn, is excluded). More dramatically, the syntax of special clitics may involve their appearing in a position which is not in general accessible per se to rules of the syntax at all—e.g., second position, interpreted as immediately following the ﬁrst full phonological word of a phrase or clause. Simple clitics are unaccented variants of free morphemes, which may be phonologically reduced and subordinated to a neighboring word. In terms of their syntax, though, they appear in the same position as one that can be occupied by the corresponding free word. Finally, bound words are forms that are always unaccented and phonologically subordinated to a neighboring word. They are often syntactically associated with a whole phrase, while being phonologically associated with a single word in (or adjacent to) it.

What is a Clitic?

11

Examining this classiﬁcation, we can see that it has some slightly peculiar properties. For example, the difference between special and simple clitics, on the one hand, and bound words, on the other, is based on whether or not a related non-clitic (free) form exists. But what reason do we have to believe that this is really signiﬁcant? That is, assuming that a cluster of properties (not just a single one) can be identiﬁed to separate bound words from the two (other) types of clitic, it is by no means obvious that these should all co-vary depending on whether or not the form in question alternates with another (synonymous but free) form of the same lexical item. Consider the English pronouns which can appear either reduced (If he comes near me, I’ll hit’m [hiRm]) or unreduced (she wanted to meet him " as a clitic by Zwicky’s criteria, in virtue [mijthim]). The former is identiﬁed of the latter’s existence. The assumption made at the outset of Zwicky’s paper, motivating some of his decisions about terminology and categories, is that clitic and non-clitic in cases like that of English object pronouns are related by phonological rule. When we look more closely, though, this turns out to be fairly implausible, at least in the general case. Zwicky and Pullum’s (1983) paper on reduced forms of negation in English, for instance (which I will discuss in more detail below), observes that the clitic forms of English auxiliaries like will and would suggest processes that are otherwise quite unprecedented in English phonology. If these clitics are linked to corresponding non-clitics, then this is almost certainly a fact about the relevant (phonologically complex) lexical entries, not a matter of phonologically predictable variants of simple forms. But why should the structural character of a linguistic form as a “clitic” depend crucially on the contingent matter of whether it happens to be linked with another, non-clitic form in a single (syntactically and semantically, but not phonologically) uniform lexical entry? Macaulay (1987) describes a set of clitic forms in the Otomanguean language Mixtec, and makes this point in greater detail. She shows that in that language there are phonological rules deleting glottal stops, coalescing sequences of identical vowels, and deleting word-initial syllables, in rapid speech. These processes have the effect of reducing disyllabic forms to monosyllables under appropriate conditions, and some forms show an alternation that is driven by these rules. We can see that the variation between disyllabic and monosyllabic forms is a consequence of these rules, because in slower speech the lost syllables are “restored.” But there are other forms for which the ‘short’ form does not correspond to any longer slow-speech form, and in some other cases, the alternation is not something that can be derived from the (independently motivated) phonology.

12

Aspects of the Theory of Clitics

For instance, the third person masculine object pronoun is /ˇcàà/ in its full form, but /–re/ when reduced. The full form of the ﬁrst person familiar pronoun is /ruPu/, for both subjects and objects; the corresponding reduced form /-ri/ can only be used for subjects, however. The reduced form of /ruPu/ as an object is the (phonologically predictable) shape /-ru/. These diverse relations show that the reduced forms are not—or at least not in the general case— derived by the operation of phonological rules. It looks, then, as if the clitic and non-clitic forms in such cases are often lexical alternates. For instance, English would has two lexical shapes: the clitic form /–d/ and the non-clitic /wUd/. The Mixtec third-person masculine object has the forms /ˇcàà/ and /–re/, etc. In each of these cases, the forms in the lexical entry must be accompanied by some statement of their distribution. But in that case, the difference between clitic forms and bound words loses much of its theoretical interest: a bound word is just an unaccented lexical item that does not happen to have any non-clitic lexical alternant—surely not a fundamental difference of status. What about cases where there really is some motivated phonological reduction that relates the clitic to the non-clitic form? This is the case with much (though not all) of the Mixtec lexicon, for example, and perhaps it is also true for English object pronouns. In that case, we want to say the lexical entry provides only one lexical shape, and the phonology has an effect that yields a reduced form (a clitic). Here it appears that we have something which is different both from pure non-clitics and from lexically distributed clitics. But what is it that we have in such cases? Everything depends on what we want to say the nature of a “clitic” is, and in particular, whether the existence of certain rules in the phonology could be relevant to an item’s classiﬁcation. If “clitic” is a (lexical or grammatical) category like noun, verb, etc., it is surely quite anomalous to suggest that a phonological rule could convert something into one. But in fact, there is no reason to believe that “(phonological) clitic” is a primitive, on the order of the lexical categories. Clitic behavior involves (a) the absence of autonomous accent, a property shared by most (though as we will see, not all) afﬁxes; and (b) phonological subordination to another word. In fact, as soon as there came to be a real theory of prosodic structure within which to articulate the problem (well after the appearance of Zwicky’s paper), it became possible to give a coherent deﬁnition of the phonological property that makes something a clitic without treating “clitic” as a lexical or grammatical category. The necessary framework, which we will discuss in more detail in Chapter 3, is grounded in discussions of the so-called “prosodic hierarchy,” for which a classic reference point is the presentation in Selkirk (1984) or Nespor and Vogel (1986).

What is a Clitic?

13

A ﬁrst approximation to a characterization of clitics, along lines we will amplify later, is as follows. Let us say that the phonological substance of a word is organized into a relatively high-level prosodic domain, the “prosodic word.” A constituent of this type is composed of feet, which in turn are composed of syllables, composed of segments, linked to features. Afﬁxes, in contrast to lexical words, consist of segments, or syllables, or even feet; but these are not (generally) organized into a prosodic word. Prosodic words, in turn, are the constituents of phonological phrases, etc. When phonological material does not have enough prosodic structure to be integrated into the prosodic structure of the whole utterance on its own (by virtue of not being organized into a prosodic word), it must be dependent on some adjacent material that can provide the necessary bridge between lower- and higher-level prosodic categories. This sort of incorporation into an adjacent word is just the behavior we associate with clitics (in the phonological sense), which we can thus propose to treat as prosodically deﬁcient forms. Assume that prosodic deﬁciency is not tolerated at the phonetic interface: in order to be pronounced, every bit of phonetic content must be integrated into the prosodic hierarchy of syllable, foot, word, phrase, etc. As a result, where a potential utterance contains some ‘deﬁcient’ material, some principle of adjunction has to operate within the phonology to incorporate it into the structure if it is to be pronounced (and the overall structure is to be wellformed). This usually happens by adjunction into an adjacent constituent of the relevant type: i.e., segments may be adjoined into a neighboring syllable, syllables into a foot, feet into a prosodic word, etc.—though in some instances, as we will see in Chapter 3, the adjunction may not be at the lowest possible level of structure. I will refer to whatever principle(s) a language may have for prosodic incorporation of such stray material as its rule(s) of Stray Adjunction.1 Let us call the adjacent element that serves as the target of this adjunction the clitic’s host. This operation, which I will treat in more detail in Chapter (3), is the phonological form of cliticization.2 By virtue of not projecting structure at the level of the phonological word—a ﬁrst approximation to the deﬁciency diagnostic of simple clitics— such elements also will not have autonomous stress, assuming (as is common) that stress is deﬁned over a word tree, or at least within that domain. Phonological rules that apply internal to words will treat them as part of the same 1 This terminology should not be construed to prejudge such matters as whether rules or constraints are the appropriate way to think about Stray Adjunction, or indeed whether Stray Adjunction is always “adjunction” in formal terms. We simply need a convenient term for the way in which languages deal with prosodically deﬁcient material. 2 The proposal that integration of otherwise deﬁcient material into prosodic structure is what is involved in phonological cliticization has also been made and defended by other authors, including Lahiri, Jongman, and Sereno (1990) and Booij (1996).

14

Aspects of the Theory of Clitics

domain as adjacent material to which they are adjoined, though whether “Lexical” or “Post-lexical” phonology will apply at the boundary between clitics and their hosts, whether the presence of a clitic can affect the location of stress within the host, and other phonological matters will depend on the intricacies of the precise structure which results from the adjunction, and also on the details of the phonologies of particular languages. But this much gives us an overall framework within which we can hope to explore the phonological dimension of cliticization, leaving the morphosyntactic peculiarities that are important to Zwicky’s notion of special clitics to be dealt with in later chapters. Let us now look at some concrete examples of “simple clitics” in more depth.

2.2 Case I: An Introduction to Kw akw ’ala Clitics As an example of a set of clitics whose distinctive properties arise as a consequence of their phonology, I look ﬁrst at the determiner and pronominal clitics of Kw akw ’ala. Although the syntax of these elements is not unusual, their phonology has consequences that are quite striking at ﬁrst glance, and indeed there are aspects of the language’s syntax that also derive from their phonology. In later chapters we will see that this language also displays “special” clitics, but my focus here is on elements whose behavior derives from the fact that they are prosodically deﬁcient in the sense of the preceding section. The name Kw akw ’ala is commonly used to designate a number of distinct but related dialects spoken by a number of distinguishable tribal groups on the northeast coast of Vancouver Island, British Columbia, and in areas of the adjacent mainland (of which the Kwaguì are one, but only one). The area in which Kw akw ’ala is spoken is identiﬁed as “Kwakwaka’wakw” on the map in Figure 2.1, following current local usage. It is a member of the Northern Wakashan family of languages, related to the other Northern Wakashan languages Oowikyala (“Owikeno”), Heiltsuk (Bella Bella), and Haisla; and more distantly, to the languages of the Southern Wakashan family (“Nootka,” “Nitinaht,” and Makah). The time depth of the entire Wakashan family is probably something like 3000 years, and that of its northern branch perhaps 2000 (Bach and Howe 2002). Kw akw ’ala is better known as “Kwakiutl”3 from the name Franz Boas used for the language in his extensive work on it. This includes two full grammars 3 This derives from a nineteenth-century missionary orthography (Hall 1889), and is actually intended to represent the name of the people (the [kw ag,uì]), as opposed to the name of their language, ([kw akw ’ala]). Although this writing system fails to represent several major phonological distinctions in the language, it has a great deal of sentimental appeal to many of its speakers, especially among older people. Writing systems that are more nearly adequate to the phonology of Kw akw ’ala inevitably involve a fair number of diacritics and unusual characters, a feature that speakers dislike since it seems to make their language look strange and exotic.

What is a Clitic?

15

Figure 2.1. Some First Nations languages of British Columbia. “Kwakwaka’wakw” = Kw akw ’ala; other Northern Wakashan languages are Haisla, Heiltsuk, and Oweekeno. Southern Wakashan languages include Nuu-chah-nulth (“Nootka”) and Ditidaht (“Nitinaht”). Reproduced courtesy of the Museum of Anthropology, University of British Columbia. Copyright 1994.

(Boas 1911, 1947) and many volumes of texts, prepared over a period of many years. The description here is based on my own ﬁeldwork during several summers in the 1970s, but I have drawn most of my examples and cited forms from Boas’s work, to facilitate comparison. I have, however, modiﬁed Boas’s transcriptions to reﬂect my own understanding of Kw akw ’ala phonological structure. As noted in Anderson (1985b), Boas employed representations that are extremely close to the surface phonetics, in ways that sometimes obscure their structural interpretation. To appreciate the role of the clitics in Kw akw ’ala morphosyntax, it is necessary to introduce some of the basic features of the language’s syntax (cf. Anderson (1984) and references cited there for further details). The basic clause structure is quite rigidly VSO, and conforms to the schema in (2.1). (2.1) V—Sbj(—x.-Obj)(—s-Obj)(—PP*)

16

Aspects of the Theory of Clitics

The verb always comes ﬁrst in the clause, although the sentence-initial element is often a semantically empty auxiliary verb, with the lexical verb following. The verb is then immediately followed by the subject if this is overt. As discussed in Anderson (1984), a sentence may contain more than one verb. In that case, the subject may immediately follow either the sentence-initial verb—typically, a semantically empty auxiliary—or the last verb in the sequence. Regardless of the position of the subject, remaining arguments and adjuncts follow all verbs, in the sequence to be described below. I abstract away from this limited optionality in the position of the subject in the discussion below. The subject is followed in its turn by one or both of two kinds of object argument phrase. These are called “objects” and “instrumentals” by Boas, though their precise semantics do not always correspond to these descriptive terms, and they should be regarded as formal categories. “Objects” are marked with x. and “instrumentals” with s; they come in that order if both are present. An example containing an overt subject and both types of object argument is given in (2.2). (2.2) y@lkw @mas=ida b@gw an@ma=x.-a ’watsi=s-a gw ax.ňux.w cause hurt-dem man-obj-dem dog-inst-dem stick The man hurt the dog with the stick Non-subject arguments can be supplied by weak pronominal elements, such as the word-ﬁnal =s and =q in the examples in (2.3). (2.3)

a. x.w @sPid=ida b@gw an@ma=x.-a g@nan@ma=s struck-dem man-obj-dem child-instr The man struck the child with it b. x.w @sPid=ida b@gw an@ma=q struck-dem man-obj The man struck him c. x.w @sPida=∅=q=s struck-he-obj-inst He struck him with it

Pronominal forms also exist for the subject, but these behave somewhat differently from the non-subject pronominals. Since the third-person form is typically null phonologically, I will ignore the subject forms for the present. In the examples of (2.2) and (2.3), we have several representative instances of simple clitics. Note ﬁrst that each full nominal expression is preceded by a determiner element: -ida, -x.a, -sa, etc. Although this provides case marking and deictic information about the nominal that follows, it attaches

What is a Clitic?

17

phonologically to the preceding word, regardless of that word’s syntactic afﬁliation. This prosodic attachment shows us that these elements are clitics, and the way it works shows us that Stray Adjunction in Kw akw ’ala operates rigidly to adjoin prosodically weak material to the left, despite the syntactically counter-intuitive nature of the result. As illustrated in (2.3), any of the overt (third-person) nominal arguments may be substituted by a pronominal form. These are also (simple) clitics, and attach to the preceding word. Consistent with the claim that these are ‘simple clitics,’ they appear in the position of the corresponding full nominal argument expression. And like the determiners, they attach to their left. The core of the clause can be followed by any number of adjunct expressions. These take the form of prepositional phrases, although it is worth noting that the language has only a very small number of prepositions. Most PPs employ the same preposition, la4 with the semantics of the adjunct relation supplied by afﬁxes within the verb. As an alternative to a full case-marked nominal expression in argument position, non-subject arguments can be supplied by Adjunct phrases instead, as illustrated in (2.4). (2.4) a. la-ň-@n kw ’ix.Pid-uň y@s-gada kw ix.ayu-k aux-fut-I strike-you with-dem club-dem I’ll strike you with this club b. noìa-∅-s-is kw ix.ayu lax.-is ts’a’ya threaten-he-inst-his club to-his younger brother He threatened his younger brother with his club c. la-’mis-@s ńiqala-ň@-s aňa’n@m gax-@n aux-conn-you name-fut-inst wolf to-me And so you will name me (with) wolf There are no independent clitic forms for ﬁrst-person (singular or plural, inclusive or exclusive) objects. An object argument with ﬁrst-person reference must be supplied by an adjunct (such as gax n ‘me’ in (2.4c)). Further, there are no non-clitic pronouns that occupy argument positions. The only full words with “pronominal” reference are predicates (e.g., nugw aP m ‘it is I’). These are entirely parallel to the language’s ‘wh-words’, which similarly are predicates and not nominals (e.g., ngw a- ‘(to be) who?’, ’mas-‘(to be) what?’). To some extent, Adjuncts alternate with argument nominals, as shown in (2.5). e

e

e

4

It is interesting that this form is homophonous with the verb la ‘go’, and that it is suppletively replaced by the form gax when the object is ﬁrst person. This, in its turn, is homophonous with the verb gax ‘come’. The origin of the form in a serial verb construction seems evident.

18

Aspects of the Theory of Clitics

(2.5)

a. ’nik@-∅-x.-is x.w @nukw say-he-obj-his child He said to his child b. ?’niki-∅ lax.-is x.w @nukw say-he to-his child He said to his child

The same argument cannot be represented simultaneously by an adjunct and an overt element (e.g., a clitic pronoun) in argument position, however; that is, there is no ‘doubling’ of overt pronouns, as shown in (2.6). (2.6)

a. nik@-∅-q say-he-obj He says to him b. *nik@-∅-q lax.-is x.w @nukw say-he-obj to-his child He says to his child

The basic set of clitic pronouns is given in Figure 2.2. Person 1sg 1Incl 1Excl 2nd 3rd

Subject -@n(ň) -@nts [email protected] -@s ∅

Object — — — -uň -q

Instrumental -@n(ň) -@nts [email protected] -us -s

Figure 2.2. Clitic pronouns in Kw akw ’ala

We see that argument positions can be ﬁlled by any of the following: a (case-marked) full nominal expression; a (phonologically) weak pronominal in the same position as a full nominal, attaching phonologically to its left; or else by an empty pro, which I assume to occupy argument positions when the referent of the corresponding argument is speciﬁed by an Adjunct. In this, I follow the line argued for by Baker (1995) with respect to all argument positions in Mohawk. I will return to this analysis in later chapters to explore its consequences in more detail. Notice that the clitic elements introduced above really do form part of a phonological word with the preceding word. This is conﬁrmed by the phonology: at the boundary between a clitic and its host, we get word-internal phonology rather than the phonology which otherwise occurs across word

What is a Clitic?

19

boundaries. One small example of this, cited by Boas (1947), is the rule of consonantal epenthesis given in (2.7), a rule which does not apply to /ìs/ sequences formed across the boundaries between words. (2.7) a. ∅ →t/ì—s b. dug.w ìts < dug.w ł=s ‘it was seen by him’ Even more diagnostic in this regard is the placement of stress. Stress in Kw akw ’ala appears on the ﬁrst full vowel (or non-glottal syllabic resonant) of the word, where the relevant notion of “word” includes a host and any following clitics. Clitics thus display their dependent status by not initiating new stress domains. We could express the phonological behavior of clitics in Kw akw ’ala informally by the rule in (2.8). e

e

(2.8) Stray Adjunction (Kw akw ’ala): Adjoin stray material to a PWord on its left. In Chapter 3, I will reﬁne this description somewhat by recasting it within a framework based on constraints and situating it within a broader account of prosodic categories in general. There is actually a good deal more to say about the Kw akw ’ala determiner system, and I will explore it further in Chapter 4. The clitic nature of determiners in Kw akw ’ala is not simply a matter of the way in which they are pronounced. It has signiﬁcant implications for the range of syntactic possibilities in the language. These follow from the fact that the left edge of a nominal expression in Kw akw ’ala is always marked with a clitic determiner, and this must attach to its left as a consequence of (2.8). From this, it results that a nominal expression cannot be pronounced by itself, since it must always be preceded by a host for its initial clitic. When there is no preceding constituent, the place of such a host must be supplied by an initial prothetic stem, as in (2.9). (2.9) *(yi)=x.ux.da g nan m child (∅)-dem (it’s) that child e

e

A second consequence of the dependent nature of Kw akw ’ala determiners is that no nominal can ever appear in sentence-initial position, as (e.g.) a topic phrase. Again, such an initial phrase would have no preceding host for its determiner clitic, and a prothetic stem would alter the syntax of the construction. It is actually quite unusual for a VSO language like Kw akw ’ala to lack such an initial topic construction.

20

Aspects of the Theory of Clitics

We can compare this situation with that in closely related languages whose clitic systems are different from that of Kw akw ’ala. Both Haisla and Heiltsuk (Bella Bella) lack determiner clitics at the left edge of nominal expressions, although they do have other determiner clitic elements. In Oowiky’ala (Oweekeno, Rivers Inlet), determiner elements are all internal to the nominal, and no determiner clitic appears at the left edge as in Kw akw ’ala. The examples in (2.10) are taken from Boas’s (1928) collection of bilingual texts, and provide both Oowiky’ala and Kw akw ’ala versions for comparison. (2.10)

a. (O) gi sukw a’la hanń@ma=se (K) la’lai ax.Pid@la=x.is hanaň’@ma then took(-his) arrow(-his) Then hei took hisi arrow b. (O) wala’li subotsowi’la hanń@ma=se (K) la’lai dax.Pid=x.a hanaň’@ma=s Then took(-obj) arrow-his Then shei took hisj,∗i arrow

In Heiltsuk (Bella Bella; cf. Rath (1981)), the determiner system is fairly intricate and involves a number of component parts (like Kw akw ’ala, as we shall see in Chapter 4). As the examples in (2.11) illustrate, however, none of these parts is a clitic appearing at the left edge of the nominal. (2.11)

a. p’ála wísm=á=x.i la uxw ňiás=a=x.i work man-det1-det2 on roof–det1-det2 The man worked on the roof b. p’ála p’ác’uá=ya=s wísem=x.i la uxw ňiás=a=x.i work diligent-det1-conn man-det2 on roof–det1-det2 The diligent man worked on the roof p’ác’uá=s wísem=x.i la c. p’ála ’wála=ya=s work really-det1-conn diligent-conn man-det2 on uxw ňiás=a=x.i roof–det1-det2 The really diligent man worked on the roof

Corresponding to this difference in the determiner systems among Wakashan languages, we ﬁnd a syntactic difference: languages without left edge clitic determiners also allow a construction with an initial nominal representing a topic, though otherwise their word order is very similar to what we ﬁnd in Kw akw ’ala. The examples in (2.12) with initial topics are presented

What is a Clitic?

21

for Bella Bella by Boas (1947: 298), who explicitly links them to the absence of initial determiner clitics. (2.12)

a. (gi) pkw ’ala ’m@nuk (then) one-person say (Then) one man said . . . b. h´@lxPainoxw láoňde g.@n´@m-xde=a=se Killer whale removed wife-former-det1-your A killer whale took away your (removed from presence) wife

Haisla, like the other Northern Wakashan languages except Kw akw ’ala, lacks determiner clitics at the left edges of nominals and allows initial topics as illustrated in (2.13), drawn from a published text (Lincoln, Rath, and Windsor (1986)). (2.13)

a. kw a’nalasgułd h’x.w h’a’maka mia-gila-su=si yellow cedar try ﬁrst ﬁsh-make-pass-3agt Yellow cedar wood was the ﬁrst thing he tried to turn into salmon b. hnńm-’wsm sa’wati hs=qids Gala=yads bgw anm=x.i du people-det2 and arrow-still use by-3inv ﬁrst-det1 łk’w =is bow-det The arrow was still used by them, the ﬁrst people, and the bow

Finally, we ﬁnd a very similar situation in Ditidaht (“Nitinaht”), a southern Wakashan language (cf. Klokeid (1976)). In this language, as the examples in (2.14) show, a determiner appears after the ﬁrst word of the nominal and case is indicated by a separate word (Po¯yoqw ‘acc’ in these sentences). (2.14)

a. tl’itchitl=ibt=P Pa John bowatc Po¯ yoqw shoot-past-decl John deer acc John shot a deer b. tl’itchitl=ibt=P Pa John bowatc=P Paq Po¯ yoqw shoot-past-decl John deer-det acc John shot the deer c. tl’itchitl=ibt=P Pa John P¯ıx.=P Paq bowatc Po¯ yoqw shoot-past-decl John big-det deer acc John shot the big deer

Crucially, no left-edge determiner clitic appears in Ditidaht nominals, and accordingly, initial topics are possible as in (2.15).

22

Aspects of the Theory of Clitics

(2.15) [ o¯ yoqw=obt=P Pa [ P¯ıx.=P Paq bowatc ] ] tl’itcitl John acc-past-decl big-det deer shoot John The big deer, John shot (it) The correlation between determiner systems and the possibility of sentence-initial topics make it clear that the absence of this latter construction in Kw akw ’ala must be due to the impossibility of providing a host for the determiner clitics that appear at the left edges of nominals in this language. Although the distinctive properties of these elements are phonological in nature, they have syntactic consequences. I will return to the clitics of Kw akw ’ala in Chapter 4, where additional elements with the properties of “special” clitics will be discussed. There is considerably more to be said about the phonological dimension of cliticization before we move on to those syntactic matters, however. At this point I would like to return to the foundational issues of how to identify clitics, and how to disentangle the diverse senses in which something can be thought of meaningfully as deserving that designation.

2.3 Dimensions of Clitic-hood Let us return to the typology of clitic elements, as presented originally in Zwicky (1977). On that picture, the elements Zwicky called simple clitics are merely one (or more) of the alternative phonological forms of certain lexical items (in particular, the prosodically deﬁcient realization(s) of the item in question). “Bound words” are the limiting case of this: lexical items that only have prosodically deﬁcient phonological forms. And where a free form undergoes phonological reduction to become a clitic, what that means is that the rule in question reduces its prosodic organization (perhaps pruning its “word” node, so as to make it an isolated foot or syllable), with the result that it comes under the same phonological regularities as forms that are lexically deﬁcient. On this account, there are three quite distinct categories of element that display the same prosodic behavior: bound words, which are uniformly deﬁcient in prosody; (simple) clitics, sensu stricto, the prosodically deﬁcient alternants of certain free forms; and free forms which have undergone phonological reduction. But it is far from clear that this distinction is a meaningful one. For instance, we saw above that Kw akw ’ala has no full forms with pronominal reference. Only the clitics of Figure 2.2 serve that purpose in the language. Since these words have no lexical non-clitic alternants, and are not phonologically reduced from free forms, they are (in the classiﬁcation under discussion) bound words. In the related language Heiltsuk (Bella Bella), however, ﬁrst- and second-person forms have both clitic and independent, fullword alternants, as in Figure 2.3 (drawn from Boas (1947:296) and Rath (1981)).

What is a Clitic? Person 1st 1Incl 1Excl 2nd 3rd

Independent form nugw a nugw @nts nugw @ntkw qsu —

Subject -nugw a, -@n -@nts -@ntkw -(@)s(u) ∅

Object -@nňa -@nň@nts -@nň@ntkw -uň -q

23

Instrumental q@sq@ntsq@ntkw -us -s

Figure 2.3. Pronouns in Heiltsuk (Bella Bella)

In the ﬁrst- and second-person cases, the clitic pronouns and full forms are optional variants, as shown by the pairs in (2.16). These examples are cited from Rath (1981), with the orthography used there adjusted slightly to facilitate comparison with other examples in the present discussion. (2.16)

a. a´@mbaya his núgw a /qs= a´@mbaya ahead inst me 1sg-inst ahead one ahead of me one ahead of me b. a´@mbaya his qsú /a´@mbay =us ahead inst you ahead 2-inst one ahead of you one ahead of you

In the third person, however, only the clitic form (as in a´ mbaya=s ‘one ahead of him’) is found. The ﬁrst- and second-person clitic forms in Heiltsuk are truly clitics in Zwicky’s classiﬁcation, since they have full-form variants, but the third-person form is instead a “bound word.” Clearly this distinction does not correspond to the one we want: rather, we want to classify all of the prosodically dependent forms (i.e., all of the pronominals in Kw akw ’ala, and both the proclitic and enclitic forms in Heiltsuk) as clitics, regardless of whether or not they have non-clitic lexical alternants. The most appropriate sense of clitic, then (at least in the present phonological context) seems to be one that is independent of the existence (or lack thereof) of non-clitic alternate forms of the item in question. e

(2.17) Phonological Clitic: A linguistic element whose phonological form is deﬁcient in that it lacks prosodic structure at the level of the (Prosodic) Word. We could then use Zwicky’s term simple clitic to refer to an element whose only claim to clitic-hood is phonological in the sense of (2.17), though I will usually use phonological clitic instead in this book, to highlight the nature of the elements under discussion. Recall that classical usage (e.g., that of Wackernagel) equated clitic status with lack of independent accent, and that property follows directly for

24

Aspects of the Theory of Clitics

elements that satisfy (2.17). Phonologists have assumed at least since the appearance of Liberman and Prince (1977) that stress accent is deﬁnitionally a property of the metrical structure of Prosodic Words, and obviously an element that does not have such structure does not, by itself, project stress. But that analysis also allows us to accommodate one of the types of clitic that Zwicky (1977) considers unusual. In (Modern) Greek, enclitics do not receive stress. Thus, [ðóse] ‘give!’; [ðóse=mu] ‘give me!’ with no stress on the clitic =mu. But when two such enclitics are attached to the same host, a stress appears on the penultimate one, as in [ðóse=mú=to] ‘give it to me!’ The claim that =mu ‘me’ is a clitic seems to be compromised by the fact that it has an accent. The answer to this apparent conundrum is that, while it does not constitute a Prosodic Word in its own right, a clitic does (indeed, must) get incorporated into the larger prosodic structure projected by its host. The relevant stress which appears on the clitic is assigned not to the clitic per se, but rather to the larger word of which it is (now) a part—speciﬁcally, in Modern Greek, to the penultimate syllable. I will discuss prosodic structure and its assignment in more detail in Chapter 3. Assume for now, however, that an appropriate structure has already been assigned to the word [ðóse] ‘give!’ consisting of a word containing a single, left-dominant foot including both of its syllables. Now when just a single additional syllable is incorporated into the word (by whatever principle of Stray Adjunction applies in Modern Greek), as in [ðósemu] ‘give me!’ no new foot is constructed on this added material, since feet in Greek must be minimally bisyllabic, and [mu]σ remains a syllable—part of the Prosodic Word (and thus pronounceable), but not parsed as part of any foot. When two clitics are added, however, as in [ðóse=mú=to], there is now enough material to build a new foot (composed of the two enclitic syllables), and thus to create a new stress. But this foot (and the syllables that compose it) is not a word that projects stress on its own: rather, it is part of a larger Prosodic Word, whose base is the clitics’ host [ðóse].5 There is thus no contradiction between the clitic status of the pronominal elements =mu and =to, on the one hand, and their manifestation with phonetic stress on the other.

2.4 Case II: How “Simple” are English Auxiliary Clitics? In looking at the phonological dimension of cliticization, we have already seen one set of “simple clitics,” in the Kw akw ’ala pronouns and determiners discussed in section 2.2. These illustrate one relatively simple instance in which 5

See also Steriade (1988) for an account of more complex cases of this sort.

What is a Clitic?

25

we might be tempted to attribute distinctive syntax to a set of clitics, but where it turns out that the relevant properties follow from their phonological character alone. What makes the Kw akw ’ala elements of section 2.2 clitics is a purely phonological property, but one with syntactic consequences. The need for a prosodic host entails the restriction that they cannot appear in constructions where they would be initial within a phrase, such as in the position of a sentence-initial topic. Their syntax follows without additional stipulation from their phonology. Before proceeding to a more general theoretical framework for the phonology of cliticization, I look ﬁrst at a somewhat more notorious example, where the “simplicity” of the designation “simple clitic” is rather more seriously in question. This is the case of the reduced forms of a number of auxiliary verbs in English. These have often been suggested to be simple clitics, but an enormous literature attests to the fact that at least superﬁcially, they have distinctive syntactic properties. We need to determine whether, as in the Kw akw ’ala case, these can be derived from the phonology of the items in question. Auxiliary Reduction in English The basic facts are quite straightforward. Several common English auxiliary verbs, including is and has, can appear either in a full form or reduced—in these cases, to a single consonant ’s. (2.18)

a. Fred is the only werewolf I know. b. Fred’s the only werewolf I know. c. Fred has only been a werewolf since last week. d. Fred’s only been a werewolf since last week.

The questions that arise come from the fact that substitution of the reduced form for the full form of the auxiliary is not entirely free. Under some circumstances, only the full form can appear, as illustrated by the sentences in (2.19). (2.19)

a. Do you know what Freddie is/*’s (this year for Halloween)? b. Tommy has been a werewolf more often than Freddie has/*’s (at Halloween).

While not all of the English auxiliary verbs have these reduced forms, quite a number do. The ones in which we will be interested are those in Figure 2.4. They appear in a proper subset of the positions where unreduced forms can appear: that is, an unreduced form can always substitute for a reduced one, but not vice versa. The fact that the reduced auxiliaries appear in positions where the syntax would put corresponding non-clitic forms suggests that they

26

Aspects of the Theory of Clitics Full form is are am has have had will would

Reduced ’s ’re ’m ’s ’ve ’d ’ll ’d

Figure 2.4. English Reduced Auxiliaries

are simple rather than special clitics (in Zwicky’s terminology). On the other hand, the restrictions on where they occur, such as those illustrated in (2.19), have often been associated with the syntax. We need an account both of the possibility of these elements and also of the limitations on their appearance. We also want to keep another consideration in mind. Whatever the correct analysis turns out to be, it has to be one that children can plausibly acquire quite easily. A basic fact about this matter is the observation that transcripts of children’s utterances contain essentially no instances of reduced forms where these are impossible in adult speech. Furthermore, in experiments designed to bias children toward producing inappropriate instances of the reduced forms, they still do not do so.6 This suggests that the correct account must be quite deeply embedded in principles of Universal Grammar that plausibly govern children’s grammars at all stages, rather than based on detailed languageparticular stipulations. Kaisse’s (1985) Analysis A convenient starting point is Kaisse’s (1985) extensive discussion of a range of cliticization phenomena in English, even though Kaisse’s notion of “simple clitics” was evidently somewhat different from the one suggested here. Her description had its origins in an earlier paper entitled “The Syntax of Auxiliary Reduction in English,” and I have proposed that it is characteristic of simple clitics that their syntax has nothing special about it. If these items arise through the intervention of speciﬁcally syntactic effects, they would not be simple clitics in my sense. Although she follows Zwicky (1977) in proceeding from a deﬁnition of clitic that assumes a relation between reduced and full forms (which I suggested in section 2.3 is not really a signiﬁcant property), Kaisse does not depend 6 See Anderson and Lightfoot (2002) and references cited there for discussion and support of these claims.

What is a Clitic?

27

essentially on this. She argues that another distinction is rather more important: among the items that have both a full and a phonologically reduced form, the difference is sometimes attributable to the phonology alone, and sometimes idiosyncratic. For example, don’t displays the shapes [dõ] and [don] which arise from ordinary fast-speech reduction processes. Reduction of a ﬁnal cluster /nt/ to the nasal alone, and further reduction of a ﬁnal nasal to nasalization of a preceding vowel, are processes that apply rather generally in the post-lexical phonology of English, especially in rapid speech. The special form [d˜@] (as in I don’t want to, pronounced as [ad˜@"wan@]), though, appears to be a lexically idiosyncratic variant of don’t, rather than being produced by productive processes. This is suggested by the fact that the phonologically similar auxiliary won’t has forms [wõ] and [won] (in He won’t be there, and she won’t arrrive until later), but no form [w˜@]. Kaisse argues that all of the reduced auxiliaries and modals in Figure 2.4 are lexically speciﬁed, and do not arise through the operation of phonological rules. Since English does not have productive phonological rules that could derive the observed reduced forms of is, are, am, has, have, had, will, would, these variants must all be lexically listed alternative stems. Compare, e.g., would (with the reduced variant ’d) with could, should, whose only variants [k@d] and [š@d] can be derived by fast-speech phonology. This difference points up a signiﬁcant issue in working out a theory of the phonology of cliticization. Given a lexically determined alternation, such as we ﬁnd for instance with has versus ’s, we can study the distribution of the reduced form (’s) and ask what the conditions are under which it can be used or inserted. If the very existence of the reduced form follows from the phonology, though, there are no conditions on where it can be inserted. The full form is inserted everywhere, and then the post-lexical phonology does what it likes on the basis of derived structure. Because of this difference, when we evaluate the conditions on cliticization, we have to be clear in each case that we are actually dealing with a clitic form, and not just with a phonetic variant derived by the phonology from a full form. Kaisse’s view of what is going on in the reduced forms was premised on the notion that the syntax is implicated. Her resolution of the problem was as follows: under certain circumstances, one syntactic element may be cliticized to another in the speciﬁc sense of being syntactically adjoined to it. When that happens, the clitic variant of the lexical item is chosen, and otherwise, the non-clitic variant is used. The basic operation involved in cliticization is thus a syntactic one, and the conditions on the appearance of phonological clitics are to be sought in the properties of the syntactic representation.

28

Aspects of the Theory of Clitics A representative derivation illustrating this view is given in (2.20).

[is]aux [leaving]VP (output of syntax) (2.20) [Jack]NP [[Jack] is]NP [t] [leaving]VP (cliticization) [[Jack] is]NP [t] ["leaving]VP (sentence prosody) [t] ["leaving]VP (allomorph selection) [[Jack]z]NP Jack’s leaving (voice assimilation) This analysis has certain features that are somewhat strange. Centrally, we must ask why we should believe that the syntactic form of a sentence is altered when cliticization occurs. That is, why should we believe that syntactic restructuring is involved in cliticization? One possible argument for that claim would be provided if we found cases where the clitic moves as a unit with its host, and where the corresponding uncliticized sequence did not behave that way. When we construct potential examples, however, that is not what we ﬁnd. (2.21)

a. I think John’s/is at the door. b. Whoi do you think [ei ]’s/is at the door? c. *Who’s do you think at the door?

The failure of such clitic-plus-host combinations to move as a unit could, of course, be due to the fact that cliticization occurs only after all other relevant syntax is completed. To say that, however, is to say that no possible evidence could support the proposal that cliticization involves syntactic restructuring, which obviously eliminates its status as an empirical claim. A different pattern which Kaisse examines is that of sequences of auxiliary plus negation, as in (2.22). (2.22)

a. Won’t Jones be in the ofﬁce tomorrow? b. *Will not Jones be in the ofﬁce tomorrow?

Here a putative host-plus-clitic combination does indeed move as a unit, but the trouble with this example is that, as noted in section 2.6 below, Zwicky and Pullum (1983) have shown convincingly that won’t in such sentences is actually a unitary inﬂected verbal form, and not a host+clitic sequence. As such there is no question that it is a single word,7 while will not is a sequence of two separate categories, not a unit and hence not movable. Thus we have no evidence for the claim that cliticization in English has syntactic consequences. 7 Flagg (2003) argues that Zwicky and Pullum’s analysis is “internally inconsistent,” on the grounds that if has and hasn’t are really just inﬂectionally distinct forms of the same word, the cliticization of has as ’s ought to imply (counterfactually) the corresponding cliticization of hasn’t as *’sn’t. No such conclusion follows, of course, since as we have seen the existence of a reduced form is a lexically idiosyncratic property of particular word forms. Has has a reduced form ’s, while hasn’t has no reduced form listed in the lexicon. No inconsistency is involved.

What is a Clitic?

29

Reduced Auxiliaries and Syntactic Gaps Of course, the lack of evidence in favor of an analysis of auxiliary reduction in English as involving syntactic adjunction still does not rule out the possibility that there are syntactic conditions on the appearance of the reduced forms; and indeed there is quite a robust literature claiming that such is the case. The best-known condition of this sort is the requirement that auxiliaries do not appear in reduced form when followed by a syntactic gap. In sentences such as those in (2.23), the impossibility of a reduced form of the auxiliary appears to be correlated with the presence of an immediately following gap, which may result either from deletion or from displacement of a constituent to another position. (2.23)

a. John is taller than Harry is/*’s [e]. b. John has known Mary longer than Fred has/*’s [e] Martha. c. Who do you think you are/*’r [e]?

In one of the classics of the underground syntactic literature of the 1970s, Bresnan (1978) offered an account of this restriction. She proposed that contrary to appearances, English cliticization is actually obligatory adjunction to the right. The restriction that reduced forms cannot immediately precede a gap would then follow from a plausible constraint to the effect that the reduced auxiliaries cannot adjoin to a phonologically null element (the gap). This position has persisted in the thought of many otherwise right-thinking linguists, despite its completely counter-intuitive nature from a phonological point of view. The phonological interaction of the clitic (assimilation, in this case), after all, is with the item to the left, not the right. This direction of association is supported by the fact that the pronunciation of the clitic varies from [s] to [z] to [@z] as a function of the ﬁnal sound of the preceding word (2.24a); and is completely insensitive to the shape of the word on its right. The variation in shape which we can observe is exactly the same as that seen in the shape of the regular plural ending (spelled (e)s but pronounced in the same three ways as ’s), again as a function of the ﬁnal sound of the preceding word (2.24b). Similarly, while the third-person singular present ending of verbs is always spelled -s, it shows the same pattern of variation in pronunciation (2.24c), as does the ending of possessive forms (2.24d). (2.24)

a. Pat’s ([s]) leaving, Kim’s ([z]) coming in, and Chris’s ([@z]) replacing Jan. b. packs ([s]), pals ([z]), passes ([@z]) c. infects ([s]), cleans ([z]), induces ([@z]) d. Pat’s ([s]), Kim’s ([z]), Chris’s ([@z]) corkscrew

30

Aspects of the Theory of Clitics

Ignoring these patent phonological facts while claiming that the reduced auxiliaries “cliticize” (invisibly) to the right is unfortunately typical of the looseness with which linguists (especially syntacticians) sometimes invoke the notion of “cliticization” as a kind of magic wand, with no requirement that it have any observable correlate. Phonologists, however, have generally been quite uncomfortable with the notion of an operation of rightward cliticization affecting English reduced auxiliaries. Apart from the obvious motivation provided by the facts in (2.24) for seeing auxiliary reduction as an operation which (at least phonologically) associates the reduced form with material to its left, it has also been pointed out that if there were (syntactic) adjunction to the right, we would expect the sequence of (reduced) auxiliary plus following word to act like a unit for purposes of deletion and movement. Examples such as those in (2.25) show that this does not happen. (2.25)

a. John’s been taking his medication every day, but Harry (*/’s been) stashing his under the mattress. b. *‘S been Kobe talking trash again?

Whatever the basis of the relation between auxiliary reduction and a following syntactic gap, it does not appear to favor an analysis on which cliticization is rightward-adjunction in the syntax. In fact, we have no reason to believe that it is a syntactic operation at all, since it has no patently syntactic consequences. This still does not resolve the problems posed by examples like those in (2.23), however. In order to maintain that the reduced auxiliaries are simple clitics in the sense that their only exceptional property is the phonological one referred to in (2.17), I must provide an account of how such apparently syntactic facts as the presence of a following gap can be relevant to their appearance. I cannot attempt to do this, however, until I have developed a much fuller account of prosodic structure, especially in English, and its relation to syntactic structure. This task will be undertaken in Chapter 3 below, where we will see that it is ultimately possible to vindicate the claim that English reduced auxiliaries are indeed simple clitics. First, however, there are two other issues that must be addressed with respect to the overall ontology of clitics and related elements. One of these is to provide at least a sketch of the nature of the second major class of clitic elements, “special clitics,” and the other is to address the outward and visible signs by which clitics can be differentiated from morphological afﬁxes in practical terms.

What is a Clitic?

31

2.5 Special Clitics If “simple” clitics are elements whose only deﬁnitional property is phonological as formulated in (2.17), what should we say about Zwicky’s class of special clitics? I will return in Chapter 4 to an account of their morphosyntactic properties. It is clearly necessary to note, though, that with few exceptions (see below), these elements display the same prosodic dependency as simple clitics, in addition to whatever sets them apart from the rest of the language morphosyntactically. As far as the phonology is concerned, the analysis that follows from this is the same as for the class of simple clitics: the clitic stem is lexically characterized as prosodically deﬁcient, while a possible non-clitic alternant (if one exists) is not. On Zwicky’s (1977) interpretation, special clitics are also related to corresponding non-clitic forms with full-word phonology and ordinary syntax. Surely, however, there is no logical requirement that a special clitic must have a non-clitic alternant, any more than that the same should hold for simple clitics. All of the characteristics of the clitic itself are ideally to be accounted for as ‘local’ properties of that element (assuming of course that we eventually come up with an appropriate account of the morphosyntax, as promised). And indeed, there are many special clitics that do not correspond in any obvious way to non-clitic elements: consider connective particles such as the secondposition clitic =te ‘and’ in the Homeric example (1.1) for instance. Indeed, it is questionable whether there is ever a special relationship (beyond nearsynonymy) between clitics and non-clitics. In particular, there is probably less reason than Zwicky assumed to treat “conjunctive” and “disjunctive” forms (e.g., French me versus moi) as synchronically alternant forms of the same lexical item. I propose, then, to deﬁne a second dimension of clitic-hood as in (2.26), without reference to any property apart from an element’s unusual morphosyntax. (2.26) Morphosyntactic clitic: a linguistic element whose position with respect to the other elements of the phrase or clause follows a distinct set of principles, separate from those of the independently motivated syntax of free elements in the language. As opposed to the case of phonological clitics, this deﬁnition requires considerably more ﬂeshing out. In particular, we need a substantive account of the kinds of positioning principle that may govern the location of morphosyntactic clitics, their place within the overall architecture of a grammar, and their relation to other areas of structure (speciﬁcally, the lexicon, morphology,

32

Aspects of the Theory of Clitics

and syntax). I will address those matters in Chapters 4 and following; for the present, I will simply assume that the cases for which such treatment is required can be identiﬁed. I will in general use Zwicky’s term special clitic for elements designated by (2.26), although it is important to remember that (a) the additional requirement of a matched non-clitic form should be dropped from the original deﬁnition; and (b) most special clitics are phonological clitics, as well: that is, they display the property in (2.17) as well as that in (2.26). This observation, in fact, allows us to accommodate another apparent anomaly in Zwicky’s (1977) system: special clitics that are not prosodically deﬁcient. Examples include standard Italian loro8 ‘to them’ or Tagalog tayo ‘we (dual)’. Whatever is at the heart of the distinctive morphosyntax of special clitics, it is apparently orthogonal to the prosodic property displayed by phonological clitics. The important point to note is that some special clitics, while displaying the relevant special morphosyntax, do not happen to be prosodically weak. These can be presumed to have full prosodic structure, despite displaying morphosyntactic behavior which is (also) associated with prosodically weak items. While the properties in (2.17) and (2.26) often coincide, and morphosyntactic clitics are typically also phonological clitics, these notions are logically separable, and empirically distinct. On the present account, the special class of bound words disappears as an independent category. These are simply elements whose lexical representation is always prosodically weak (or deﬁcient). They can be divided into two classes: those with special morphosyntax, and those without. There is no reason to believe that this distinction is correlated with any other interesting distinguishing property, however. In later work (cf. Zwicky and Pullum 1983), Zwicky also abandons the independent class of “bound words,” though with little discussion. In any event, the simpler ontology adopted here has the merit of being grounded in the theory of grammar, rather than being just a set of descriptive categories. This, of course, is just what Zwicky (1985 and elsewhere) counsels: presystematic tests and descriptive labels eventually have to be cashed out in terms of theoretical constructs. 8 The classiﬁcation of this element as a clitic is a chronic problem: see Cardinaletti (1991) and Nespor (1994) among other sources for discussion. In brief, loro displays some of the properties of other pronominal clitics in Italian, while differing in others. It appears following the verb with which it is associated, for instance, while other clitic pronouns precede the (ﬁnite) verb. Despite this, loro clearly does not display the syntax of a free nominal expression. Ultimately, these differences should be resolved by providing an account of the element’s atypical distribution in terms of its atypical prosody, a matter I will not attempt here.

What is a Clitic?

33

2.6 Clitics and Afﬁxes We now have a reasonable understanding of what clitics are: they are linguistic elements that display prosodically deﬁcient phonology, anomalous morphosyntax, or both. The comparisons to this point have been between clitics and ordinary, non-clitic words, but another contrast to consider is that between clitics and word-internal afﬁxes. After all, if a distinguishing property of (most) clitics is the fact that they form a (prosodic) word with other material, we need to be able to distinguish them from other word-forming elements. The locus classicus for these matters is Zwicky and Pullum’s (1983) examination of the ways in which we distinguish, in practice, between clitics and afﬁxes. They propose a number of tests for separating the two, summarized in (2.27). (2.27)

a. Clitics have a low degree of selection with respect to their hosts; afﬁxes a high degree of selection. b. Afﬁxed words are more likely to have accidental or paradigmatic gaps than host+clitic combinations. c. Afﬁxed words are more likely to have idiosyncratic shapes than host+clitic combinations. d. Afﬁxed words are more likely to have idiosyncratic semantics than host+clitic combinations. e. Syntactic rules can affect afﬁxed words, but not groups of host + clitic(s). f. Clitics, but not afﬁxes, can be attached to material already containing clitics.

As formulated, these points are merely descriptive observations about differences in the behavior of two pre-systematically understood classes of item. Some linguists9 content themselves with lists of behavioral properties of this sort, considering such a more or less comprehensive diagnostic symptomatology to constitute an analysis of a phenomenon. A list like (2.27), however, does not represent an explanation: rather, it lays out what is to be explained. Let us then see if we can ground these differences in the analytic framework I have proposed for studying clitics. Some minimally controversial assumptions about the architecture of a grammar will sufﬁce for that purpose. First, I assume that words are built (including afﬁxation) within the lexical phonology. As a result, afﬁxation processes have access to the form and meaning of stems, and can depend on 9

Not Zwicky and Pullum, I hasten to point out.

34

Aspects of the Theory of Clitics

(and affect) this in idiosyncratic ways. Second, words are combined with one another post-lexically, through the syntax. On the assumptions of the Lexicalist Hypothesis (cf. Anderson (1992)), the syntax does not manipulate or have access to the internal form of words—in the sense of lexical items, which may or may may not be coextensive with the P(rosodic) Words that concern the phonology. After a syntactic structure has been developed,10 it is then subject to the rules of the post-lexical phonology which map it to P(honetic) F(orm). Clitics enter the structure in two ways. Simple clitics, since they are ordinary lexical items that happen to be unusual from a phonological point of view, are combined with other items in the sentence by the ordinary mechanisms of the syntax: hence, post-lexically. Special clitics are not the province of the ordinary syntax—that is what makes them ‘special’—but I assume that they are nonetheless introduced into the structure post-lexically. Speciﬁcally, I will argue in Chapters 4 and 6 below that they are introduced as phonological modiﬁcations of the (post-Spell-Out) shape of phrases, similar to the modiﬁcations made in words by the Word Formation Rules of the morphology. That precise mechanism is not important at this point, however—the only important claim is that special clitics are also introduced post-lexically. From these architectural assumptions, the facts in (2.27) can be derived as theorems. Property (2.27a), for example, tells us that while afﬁxes may be associated only with lexically idiosyncratic classes of bases, clitics attach to any host, depending only on its position in the structure or at most, its basic lexical category.11 This follows immediately from the fact that afﬁxation is a lexical process, and has access to individual word-level properties, while clitics appear only post-lexically, where they have access only to syntactic structure. Essentially the same result holds for (2.27b,c,d). Elements introduced in the lexicon (afﬁxes and other modiﬁcations performed by Word Formation Rules) can behave differently in construction with different lexical items, while elements introduced post-lexically (clitics) have no opportunity to refuse selectively to appear (in the case of gaps) or to make item-speciﬁc modiﬁcations of form or content, since they have no access to the internal properties of the speciﬁc lexical elements with which they combine. Property (2.27e) refers to the fact that while a lexical base is affected by syntactic rules of movement, deletion, etc. together with its afﬁxes as a unit, 10 Possibly the mapping from syntax to PF takes place in “phases” as certain portions of the structure are completed, with the results combining with one another in later phases (Chomsky (2001)). This reﬁnement of the view suggested here is not directly relevant to the matters under discussion in this section, however. 11 Items such as wanna in English represent lexicalized combinations of speciﬁc items, rather than a clitic attaching to speciﬁc host. See section 3.4 for some discussion.

What is a Clitic?

35

movement or deletion of a host does not have corresponding consequences for clitics dependent on it. The fact that the syntax treats afﬁxed words as units follows directly from the Lexicalist Hypothesis, since (on that approach), the syntax has no choice in the matter: it cannot even see the base and the afﬁxes as distinct elements. On the other hand, clitics and their hosts do not in general form constituents, and since these are the currency in which the syntax trades, it has (in general) no opportunity to affect clitic–host combinations in a unitary way. Finally, (2.27f ) also follows from the Lexicalist Hypothesis. Clitics are introduced post-lexically, after lexical items are spelled out, and at this point the internal structure of lexically formed items is no longer visible. As a result, clitics could not possibly be introduced among the afﬁxes of a word. Similarly, lexical afﬁxation is already complete at the point clitics arise, so afﬁxes could not be introduced to a form which “already” hosts a clitic. It is perfectly possible, however, for new clitics to attach to a prosodic word which already hosts other clitics—no architectural considerations bear on this issue, though other considerations (phonological and/or morphosyntactic) may limit the extent of clitic accumulation in a given language. The tests in (2.27) are thus thoroughly grounded in the structure of natural language, and the evidence they provide is correspondingly quite strong. On this basis, Zwicky and Pullum (1983) argue for an initially quite surprising conclusion: English contracted negative forms in -n’t (isn’t, doesn’t, don’t, won’t, etc.) actually represent an inﬂection (restricted to have, be, and auxiliary verbs in Modern English) rather than a clitic form of not. The modern status of -n’t in English appears to have arisen by way of the following stages, on the basis of fairly straightforward reanalysis at each stage. First, in Old English, negation was expressed by the full word combination n¯a wiht ‘no thing’, reduced to nought, and by Middle English, not. This word was presumably an adverb at this stage in the language, and appeared in adverbial positions subject to semantic considerations based on its scope. At this point, however, its semantics had been reduced to that of a logical operator expressing negation. Partly as a consequence of this reduced semantic content, it was subject to de-emphasis and consequent phonological reduction in many instances, leading to a simple clitic form [nt]—a phonological alter" and the fact that (for nant of full word not. Given its semantics, however, reasons based on its logical scope) it generally appeared at the left edge of VP, immediately preceded by an auxiliary element in Infl if one was present, it could be reanalyzed as a special clitic, attached at the left edge of a VP bearing a feature [+Neg]—or at the right edge of the immediately preceding Infl. But a special clitic at the right edge of Infl could easily be reanalyzed, in its turn, as

36

Aspects of the Theory of Clitics

an inﬂectional afﬁx associated with the verbal elements found in that position: have, be, and the auxiliary verbs. Many details remain to be ﬁlled in and substantiated concerning this historical scenario, but it seems plausible. While the conclusion that -n’t represents an inﬂectional afﬁx and not a clitic is somewhat remarkable at ﬁrst glance, the case is actually quite clear on more detailed examination. Along lines similar to those that yield the tests in (2.27), another paper (Zwicky (1985)) proposes a number of additional tests, this time for distinguishing clitics from full words. I will not address these in detail, because most of them follow quite directly from the understanding of clitics articulated above. One exception is the principle that clitics are generally ordered rather rigidly, while words often are not. This is a matter that can only be addressed in the context of a better theory of what it is that is special about special clitics, and I will return to it in later chapters. Interestingly, much of this paper concerns yet another kind of object in traditional descriptive practice that has something in common with clitics: the “particle.” Zwicky dispels the idea that ‘particle’ is an additional word class (like noun and verb), or that particles have no category at all. The conclusion is that some of these elements are just members of heavily restricted word classes (often adverbs), some are simple clitics, and some are special clitics. Attending to the broad class of ‘particles’ highlights the existence of clitics (simple and special) other than the grammatical markers (such as tense markers, pronominals, negation, etc.), which have attracted most of the attention in the clitics literature. Many things that are effectively dismissed descriptively by being designated as ‘particles’ have considerable semantic content. This is true, for example, of the objects Zwicky calls “discourse markers.” These observations suggest that from a descriptive point of view, there are (at least) two somewhat different kinds of clitic element: grammatical markers, and semantically content-ful ones. I will eventually want to draw an analogy between this difference and that between inﬂection and derivation in morphology, and I return to this matter in Chapters 4 and 6. At this point, I have developed a general picture of the kinds of clitics that ought to be accounted for by grammatical theory, and some notions of how that might be done, at least for the phonological dimension of clitic-hood. In the following chapter, I turn to a closer examination of the prosodic bases of phonological cliticization. In the course of that discussion, we will arrive at a resolution of the problem posed by the English auxiliaries discussed above in section 2.4, showing that their unusual properties can indeed be reduced to their nature as phonological clitics.

3 The Phonology of Cliticization Let us proceed on the assumption that the proposal to analyze English reduced auxiliaries as simple (phonological) clitics in Chapter 2 is on target. To take that idea from a gleam in the eye to a genuine analysis, however, we need a more articulated theory of how the prosodic structure of utterances is organized, as well as how that structure is related to syntactic structure. It makes sense to address the nature of prosody ﬁrst, beginning with the set of proposals made by various authors (especially Selkirk (1981, 1984, 1995), as well as Nespor and Vogel (1986)) concerning prosodic organization. Excellent summaries of these issues, including reference to their application to clitics, are to be found in Peperkamp (1997) and Vigário (2003).

3.1 Prosodic Structure Classical Generative Phonology as incarnated in Chomsky and Halle’s (1968) The Sound Pattern of English (SPE) made two broad assumptions about representations, whose repudiation has since been the source of much phonological theorizing. One of these concerned the interface between the syntax and the phonology. The claim that (morpho)syntactic organization is relevant to the operation of the phonology, originating in Chomsky, Halle, and Lukoff (1956), led the phonologists of the 1960s to assume that this connection should be implemented by allowing the rules of the phonology to refer directly to syntactic structure.1 I will return to the appropriate characterization of these interactions in section 3.3 below. First, though, my concern is with another basic assumption of the SPE theory: the claim that phonological representations are matrices of features, whose only internal organization is given by the rows and columns of a matrix with no further hierarchical structure. The division of the representation into units such as syllables, morphemes, words, phrases, etc., was carried out (to the extent it was deemed signiﬁcant) by the introduction of quasi-segmental 1 Since most work of this period assumed that the morphological structure of words was simply the extension of syntactic mechanisms to smaller domains, “syntax” also includes “morphology” here.

38

Aspects of the Theory of Clitics

boundary elements with speciﬁc positions internal to the sequence of segments. Potato, for instance, might be represented as in (3.1), where “D” is the voiced ﬂap characteristic of American English, and “$” represents a syllable boundary. (3.1)

[p@$théj$Do]

Kahn (1976) argued that the boundary-element approach to phonological units larger than the segment (speciﬁcally, the syllable) had a number of deﬁciencies that could be remedied by adopting a looser, Autosegmental picture (Goldsmith 1979). Instead of identifying syllables as stretches of segments between two instances of an appropriate boundary element, Kahn proposed treating them as structural units in their own right, constituting a structural tier linked to segmental units in a potentially many-to-many fashion. On that view, the representation in (3.1) is replaced by that in (3.2). (3.2)

σ

σ

σ

p@

thej

Do

The details of this picture did not remain in place for long, but it served to establish the idea that syllables ought to be treated as structural units—and with it, the possibility that phonological representations have hierarchical internal organization rather than being simply a succession of structurally equivalent units (segments and boundaries). And this, in turn, opened the doors for the more fundamental revisions that lie at the heart of modern views of prosodic organization. If the nature of syllables motivated a hierarchical organization of the string of segments, why stop there? In his 1975 dissertation, Mark Liberman arrived at a vastly superior representation of stress by assuming that the syllables themselves were organized into larger constituents, and that a relation deﬁned on this structure was the right way to interpret the phonetically slippery concept of “stress.” This work is best known from its presentation in Liberman and Prince (1977); the result is a picture on which potato is represented as in (3.3). (3.3)

PWord w s Foot s w σ σ σ p@

thej

Do

The Phonology of Cliticization

39

Actually, the representation in (3.3) is somewhat anachronistic in providing labels such as “PWord” and “Foot” for the constituents above the level of the syllable. Liberman and Prince simply grouped constituents together into higher-order units (recursively), but the provision of a principled inventory of structural types for these units would follow shortly. The Prosodic Hierarchy Telescoping the history somewhat, work of Selkirk (1981, 1984) and Nespor and Vogel (1986) proposed a range of additional constituent types intermediate between segments (or features) and whole utterances. Syllables (σ )2 are seen as organized into feet (Ft), feet into phonological (or prosodic) words (PW(or)d),3 words into phonological phrases (PPh), phonological phrases into intonational phrases (IPh),4 and intonational phrases into whole utterances. Implicit in these theories, and necessary to make them really signiﬁcant, is the assumption that the prosodic organization they impose is an exhaustive one. We can make this explicit as the convention in (3.4). (3.4)

Full Interpretation: In order to be well-formed at PF (i.e., pronounced), phonetic content has to be incorporated into prosodic structure.

That is, stray phonological material not connected ultimately to the root of the prosodic tree is not interpretable, and therefore disallowed. Central to this work is the claim that “Foot,” “PWord,” etc., are not simply labels, but represent substantive categories with properties of their own that distinguish them. Among these properties is the assertion that each constituent type (syllable, foot, word, phrase, etc.) deﬁnes a kind of domain. At least for Nespor and Vogel, phonological rules are expressed with respect to the domain of their application. Thus, where two feet are joined into a word, foot-level rules will apply to each foot separately, while word-level rules will apply to 2 Syllables themselves have an internal organization as well, but those matters do not concern us at this point. 3 This constituent is called the Phonological Word by some authors, and the Prosodic Word by others, without entailing any signiﬁcant difference. The abbreviation “PW(or)d” employed in this book may be interpreted either way, as the reader chooses. As a phonological/prosodic construct, this constituent is not to be identiﬁed with the Morphosyntactic or Lexical unit “Word,” though of course the two are often co-extensive. 4 Whether only a single phrasal constituent type, the PPh, intervenes between the PWd and the IPh has been a matter of some controversy in the literature, and it is possible that more than one type of PPh must be recognized. This matter is not relevant to the concerns of this book, however. One speciﬁc intermediate constituent type, the Clitic Group, will be speciﬁcally discussed and rejected below.

i

i

“Clitics” — 2005/7/26 — 17:22 — page 40 — #48

i

40

i

Aspects of the Theory of Clitics

their combination as a unit. As a result, the specific label given to a constituent formed at a given point has consequences. An example where the correct assignment of prosodic structure is crucial to getting the (non-prosodic) phonology right is provided by Polish, as discussed by Booij and Rubach (1987). Polish has a number of rules which affect the voicing of obstruents, especially in clusters. One of these devoices a fricative following another voiceless obstruent, as formulated5 in (3.5). +Obst −→ [−Voice] / [−Voice] (3.5) Progressive Devoicing: + Cont The domain of the rule in (3.5) is the PWord. Progressive Devoicing applies (lexically) within words, as shown by alternations such as listw+a [l’istf+a] ‘board’; listew+ek [l’istev+ek] (diminutive, gen. pl.) from the stem /list˘ıv/. It does not, however, apply across the boundaries between words within larger constituents such as the Phonological Phrase. Another rule requires that clusters of obstruents agree in voice when formed within the larger unit of the Phonological Phrase. Clusters that would disagree are adjusted regressively, as expressed by the rule in (3.6). +Obst (3.6) Voice Assimilation: [+Obst] −→ [αVoice] / αVoice As a rule of the PPhrase level, Voicing Assimilation applies across word boundaries (though not across phrase boundaries), as illustrated by the examples in (3.7) where the orthographic form indicates the basic value of voicing. (3.7)

a. sklep [bv] warzywny ‘green-grocer’s’ b. krysys [zg] gospodarczy ‘economic crisis’ c. zakaz [sp] postoju ‘no parking’ d. szereg [kk] krzeseł ‘row of chairs’

Yet another rule devoices obstruents at the end of a word as expressed in (3.8), where the “#” element is an ad hoc indication of final position. (3.8)

Final Devoicing: [+Obst] −→ [−Voice] /

#

The effects of Final Devoicing are seen when the word is phrase final, or when the next word begins with a sonorant, as in the examples in (3.9). (3.9)

a. z˙ aba ‘frog (fem. nom. sg.)’; z˙ ab [p] (gen. pl.) b. głazy ‘stones’; głaz [s] lodowcowy ‘glacier stone’

5 The regularities of Polish discussed below are formulated as traditional rules, following the presentation in Booij and Rubach (1987). They could also be expressed by rankings within a constraint system in an Optimality Theoretic framework, though such a reformulation seems unnecessary for my purposes here.

i

i i

i

The Phonology of Cliticization

41

When the following word within the PPhrase begins with an obstruent, whatever effect Final Devoicing might have is of course obliterated by the operation of Voicing Assimilation. Rules (3.5) and (3.8) apply within the PWord domain, while rule (3.6) applies within the PPhrase domain. Each rule must be associated with a particular prosodic constituent type in order to apply correctly, a situation also illustrated for a variety of languages by examples in Selkirk (1980, 1984) and Nespor and Vogel (1986), among many other works in this tradition. In Polish, there is a set of exceptions to the behavior of ﬁnal obstruents as described above. Speciﬁcally, prepositions followed by their complements do not devoice before sonorants, as illustrated in (3.10). (3.10)

a. pod [d] owocem ‘under the fruit’ b. nad [d] rowem ‘over the ditch’ c. bez [z] namysłu ‘without thinking’ d. od [d] mleka ‘from milk’

The explanation of this fact follows from the characterization of rule (3.8) as a regularity of the PWord domain. The generalization is that ﬁnal devoicing is PWord ﬁnal. The prepositions in (3.10) are phonological clitics, and as such do not constitute phonological words. The prosodic structure is thus as given in (3.11). (3.11)

[ [ pod[ owocem]]] PPhr PWd PWd vs. [ [ głaz (→s)][ lodowcovy]] PPhr PWd

PWd

Both Selkirk (1984) and Nespor and Vogel (1986) go beyond the association of different domains with distinct phonological behavior, proposing a set of speciﬁc conditions that came to be known as the prosodic hierarchy condition. One central aspect of this is the claim that in moving up the tree from segments to complete utterances, constituent types are layered in the sense of increasing monotonically in scope. For instance, Feet cannot contain PWords or PPhrases, PWords cannot contain PPhrases, etc. Somewhat more controversially, Nespor and Vogel (1986) claim that the analysis is exhaustive and non-recursive at each step. That is, every bit of phonological content dominated by a constituent of a given type has an analysis as part of a constituent of the immediately lower type. Thus, Phonological Phrases may not contain syllables that are not part of some Foot, which in turn is part of some PWord, etc. As a result, the representation of potato in (3.3) above would be excluded, because it involves a syllable that is analyzed directly as a constituent of a PWord without being part of any Foot.

42

Aspects of the Theory of Clitics

The correctness of this assertion (known as the “Strict Layer Hypothesis,” to which I will return below in section 3.2) is certainly not self-evident, or even, perhaps, all that plausible, and much subsequent work has rejected it. Nonetheless, it serves as quite a useful starting point for discussion, since it is about as restrictive as one could imagine. Whatever we may ﬁnd out about prosodic structure empirically, it can probably be understood as involving potential relaxation of the conditions just discussed. These matters are of direct signiﬁcance for an analysis of phonological clitics, as indeed should already be evident from the Polish facts cited above. Phonologists dealing with the structure of the prosodic hierarchy have generally thought of cliticization in terms of the formation of prosodic constituents. Clitics differ from non-clitics in being combined with other material into a single prosodic unit, (e.g., the PWord in the Polish example) rather than being autonomous units themselves. On this account, much of the phonology of cliticization comes down to the principles of prosodic organization and their consequences. The Clitic Group One of the more controversial aspects of the prosodic hierarchy as described by Nespor and Vogel 1986 (apart from the exhaustivity claim just discussed, to which I will return below) was a particular speciﬁc level of prosodic structure which they assumed. Following a suggestion from Bruce Hayes, they posit a special kind of constituent composed of a sequence of clitics together with their host. This unit, the Clitic Group, was assumed to be intermediate between PWords and PhPhrases. A Clitic Group is constructed by combining a nonclitic word with any suitable adjacent clitics. “Suitable” here refers to the notion that some clitics are lexically speciﬁed to attach in only one direction.6 Otherwise, clitics form a Clitic Group with the host with which they share the most syntactic structure. This looks rather innocuous, but in the overall context of the prosodic hierarchy, it involves several claims that have been questioned by others in subsequent work. First of all, clitics have to be distinguishable from non-clitics at the point prosodic organization is built. And notice that, given the strong form of the prosodic hierarchy assumed here, this distinction cannot be a prosodic one, because clitics have to be PWords themselves by virtue of the exhaustivity condition, a problem ﬁrst noted by Booij (1988). Furthermore, Clitic Groups (like any other prosodic constituent type) are predicted to have idiosyncratic 6

This is also assumed in various forms by Klavans (1985), Zec (1987), Inkelas (1989), and some other authors. I will suggest in later sections of this chapter that this type of lexical speciﬁcation can probably be eliminated.

The Phonology of Cliticization

43

phonology, and not simply that of phrases or other constituents. Finally, this theory provides only one structural way for a clitic to be related to its host, prosodically: as sister PWord within a Clitic Group. To the extent different clitics in different languages relate prosodically to their hosts in different ways, such an analysis is too restrictive. What arguments do Nespor and Vogel offer for the existence of the Clitic Group as a separate prosodic category? In the ﬁrst instance, they must show that Clitic Groups have special phonological properties that cannot be assimilated to those of other constituent types. In this connection, they cite the example of stress in Modern Greek, noted above in section 2.3. They observe that when words are combined into a single word in a compound, the result has only one stress, which may not be the same as either in isolation: [kúklo]+[spíti]→[kuklóspito] ‘doll’s house’; [níxta]+[pulí]→[nixtopúli] ‘night-bird (owl)’. When clitics are added to a host, in contrast, we may get extra stresses. In simple cases, enclitics do not get stress. Thus, the stress pattern of [ðóse] ‘give!’ is preserved when a clitic is added to yield [ðósemu] ‘give me!’ But as we have already seen, when two separate enclitics are added, an additional stress appears as in [ðósemúto] ‘give it to me!’ Nespor and Vogel suggest that this is due to a rule of stress readjustment whose domain is the Clitic Group. This argument does not succeed, however, as pointed out by Zec and Inkelas (1992) and Peperkamp (1997). We need to say that stress in compounds, as in simple words, is governed by some set of lexical principles, building metrical (foot) structure within a PWord. With respect to these rules, compounds count as a single domain of the relevant type. The main stress within such a PWord is associated with a binary (disyllabic) Foot constructed at the right edge of the word, allowing a single following syllable that is not part of that Foot. The generalization that results is that at most two unstressed syllables may follow the main stress within the PWord. Suppose that we have metriﬁed a host word (e.g., [ðóse] ‘give!’) in this way, and we now adjoin some additional material, such as a single clitic (yielding, e.g., [ðósemu] ‘give me!’). This form conforms to the stress regularity of the Modern Greek PWord, since its main stress foot is followed by only one additional syllable. If we add two clitics, however, as in [ðósemúto] ‘give it to me!’ the additional material will cause the generalization “No sequences of unstressed syllables after the main stress Foot” to be violated. What happens, as we have already seen, is that existing foot structure is maintained, but a new Foot is built following the original main stress Foot, since there are now at least two syllables to work with (the minimum for constructing a well-formed Foot in the language).

44

Aspects of the Theory of Clitics

On this analysis, no special property needs to be assigned to the Clitic Group that would distinguish it from the PWord. The behavior of host-plusclitic sequences can be seen as due to the interaction of the lexical rules of stress (which apply to simple words and to compounds, both of which are domains of type PWord) with a post-lexical rule that applies when phonological words with unstressed sequences arise in the syntax (hence post-lexically). No distinct constituent type need be recognized for this account to go through. The Modern Greek situation can be contrasted with what happens in some Italian dialects (and Latin), where the basic stress pattern takes the presence of a clitic into account. For example, Lucanian vínn ‘sell!’; v nníll ‘sell it!’; vinn míll ‘sell me it!’ display stress that remains consistently penultimate in the form including all clitics. Here we simply say that the basic stress rule is post-lexical, and no metrical structure is built lexically. Again the behavior of host-plus-clitic groups can be described without assuming a distinct constituent type, on the assumption that prosodic structure can be built lexically, post-lexically, or both. The most interesting theory of the phonological behavior of clitics is surely one that says as little as possible speciﬁcally about these items, while getting their properties to fall out from other factors. In that sense, eliminating the Clitic Group is a desirable move. The examples Nespor and Vogel provide, as well as those offered by Hayes (1989) and Vogel (1990), seem to be eliminable if we recognize a difference between lexical and post-lexical phonology. Other potential examples of unique properties of the Clitic Group disappear as a result of the fact that within a PWord, the boundaries of included PWords can be visible to rules, allowing us to distinguish among [[X] [Y]], [X [Y]] and [X Y]. Since positive arguments in favor of the Clitic Group as a prosodic category are lacking, and the other restrictive claims mentioned above that result from positing this category within the overall system turn out not to be correct (as argued by Zec and Inkelas 1992), most authors have abandoned it as a part of the prosodic hierarchy.7 e

e

e

e

e

3.2 Dimensions of Phonological Cliticization Let us assume that the phonology of cliticization is to be understood within the following scenario: a lexically interpreted surface structure is interpreted by a hierarchical structure of prosodic units within which lexical and functional material appears. That structure is built in a way that reﬂects, at least in 7 See Booij (1988, 1996) for further discussion of reasons to reject the Clitic Group as part of the prosodic hierarchy.

The Phonology of Cliticization

45

part, the syntactic structure, and serves as the input to the post-lexical phonology.8 A framework for talking about the phonology of cliticization, based on the combination of Lexical Phonology and Prosodic Phonology, would go as follows: (3.12)

a. Lexical principles9 determine the phonological shapes of lexical items (members of lexical categories). This includes assigning prosodic structure to them, up to the level of the PWord, as argued by Booij (1988). b. Post-lexical principles govern the relation between syntactic (S-)structure and the prosodic organization of the whole utterance. This includes building structure above the level of the PWord, and also incorporating material that (for whatever reason) is not already part of a PWord, so as to satisfy Full Interpretation (3.4). c. Post-lexical phonological adjustments apply to the resultant structure.

If we think of (phonological) cliticization in these phonological terms, it becomes a matter of describing the ways in which clitics are incorporated into prosodic structure along with non-clitic elements, together with the phonological adjustments that may apply to the resulting structures. In these terms, the questions that must be answered include those in (3.13). (3.13)

• How do we distinguish clitics from non-clitics? • For a given clitic, how is its host to be identiﬁed? • What structure results from the combination of a clitic with its nonclitic host?

With respect to the ﬁrst of these questions, I have already suggested that the best theory would be one on which there is no speciﬁc property such as “[±Clitic]” unique to clitics, but rather where clitics are distinguished from non-clitics in terms of some independently available dimension within the phonology. The proposal of section 2.3 in the previous chapter follows exactly this program, within the architecture of (3.12) above. Non-clitic lexical words are assigned prosodic structure up to the level of the PWord within the Lexical Phonology. (Phonological) clitics, in contrast, are “prosodically deﬁcient” in the sense that they do not have PWord structure at the point they are I will return to the nature and content of this relation in section 3.3 below. This formulation is intended to be neutral between the expression of such principles as rewriting rules or as a ranked set of constraints within Optimality Theory. 8 9

46

Aspects of the Theory of Clitics

introduced into S-structure. To satisfy Full Interpretation, they must be incorporated into PWords or some other constituent in the prosodic structure, together with the associated lexical material, in the post-lexical phonology.10 For the present, I will assume that a unique host can be identiﬁed for a given clitic, a matter to which I return in section 3.3. For any given hostplus-clitic combination, there are essentially four formal possibilities as to the prosodic structure that might result from combining them, as Selkirk (1995) points out.11 (3.14)

a. PWord Clitic:

b. Free Clitic:

PPh PWd

PWd

Host

Clitic

PPh PWd Host

Clitic

c. Internal Clitic:

PPh PWd Host

d. Afﬁxal Clitic:

Clitic

PPh PWd PWd Host

Clitic

The ﬁrst of these, the PWord clitic, is the case where the clitic is assigned independent status as a PWord in the phonology. A Free clitic is associated (as a non-PWord) with its host within the PPhrase. An Internal clitic is incorporated into the same PWord as its host, as opposed to an Afﬁxal clitic, which is adjoined to the host to form a recursive PWord structure. While all of these structures are logically and formally possible, however, it remains to be shown which of them are actually instantiated in natural languages, and how to distinguish among them in the phonology of a particular language to the extent that more than one possibility is admitted by Universal Grammar. 10 This is far from an original proposal. Previous work that treats the properties of phonological clitics in essentially the same way includes Berendsen (1986); Selkirk (1986, 1995); Zec (1988); and Inkelas (1989), among others. 11 These possibilities are of course independent of the linear order of the clitic with respect to its host. For the sake of concreteness, I represent the speciﬁc case of enclitics here.

The Phonology of Cliticization

47

The Strict Layer Hypothesis Two of the formal possibilities in (3.13) above violate the Strict Layer Hypothesis, as this was informally described in section 3.1. Free clitics involve structures in which some of the material dominated by a PPhrase is not part of a PWord, while Afﬁxal clitics involve recursion of PWords. The empirical question of whether such structures are ever motivated thus bears on the plausibility of the Strict Layer Hypothesis itself. Inkelas (1989) argued that this claim, as it appeared originally in work such as Selkirk (1984) and Nespor and Vogel (1986), should not be treated as a monolithic notion, but rather broken down into a number of logically independent claims. Some of these are inherent in the nature of the prosodic hierarchy itself. That theory of representations claims not only that prosodic constituent types are drawn from a limited universal inventory, but also that (unlike, say, the constituent types that appear in syntactic representations, such as DP, NP, VP, etc.) there is a speciﬁc ordering relation deﬁned over them. (3.15)

σ < Foot < PWord < PPhrase < IntPhrase < Utterance

Associating positions on this ordering with consecutive integers, we could express the basic nature of the prosodic hierarchy as involving two fundamental requirements. (3.16)

Layeredness: No C i dominates a C j where j > i (e.g., no Foot contains a PWord) Headedness (ﬁrst approximation): Every C i directly dominates some C i−1 (e.g., every PWord contains a Foot)

The Strict Layering Hypothesis can be expressed as the claim that representations also meet two other requirements. (3.17)

Exhaustivity: No C i directly dominates a C j where j < i − 1 (e.g., no PWord directly dominates a σ ) Non-recursivity: No C i directly dominates another C i (e.g., no PWord contains another PWord; adjunction structures do not exist)

In order to maintain its logical independence from Non-recursivity, the formulation of Headedness in (3.16) should be replaced by the following. (3.18)

Headedness: Every C i directly dominates some C j where j ≥ i − 1

Layeredness and Headedness are inherent in the notion of the prosodic hierarchy, and will not be questioned here. The requirements in (3.17), however,

48

Aspects of the Theory of Clitics

make substantive claims about the range of prosodic structures found in the languages of the world, and as such, are subject to empirical conﬁrmation. Evidence suggests, in fact, that they are violated in some instances, and this calls for some revision of the theory. The Prosodic Hierarchy as a Set of Constraints If the conditions (3.17) were always observed, then there could be only two possible outcomes of the post-lexical combination of a phonological clitic and its host: either a PWord is built over the clitic, and the two become part of the same PPhrase, or else the clitic is incorporated into an adjacent element of the next level up (a syllable into a foot, or a foot into a PWord). These are the “PWord clitics” and “Internal clitics,” respectively, of (3.14a,c). A language in which Free clitics (3.14b) appeared would violate Exhaustivity, since a syllable (or Foot) would be directly dominated by PPhrase; while a language with Afﬁxal clitics (3.14d) would violate Non-recursivity (and perhaps Exhaustivity as well, if the clitic consisted only of a syllable). If these structures are motivated in some language(s), the theory must be amended so as not to impose the Strict Layering requirements (3.17) as absolute conditions on prosodic structure. English Function Words Selkirk (1995) argues that both of these problematic structures occur in English. She analyzes the phonology of “function words,” typically monosyllabic prepositions, determiners, auxiliary verbs, and personal pronouns. These have full forms that appear, with stress, under some conditions: in isolation, when focused (Bettina can speak, but refuses to), or in most instances when phrase ﬁnal (I can [kn] eat more than Sara can [kæn, *kn]). " They also have weak forms that occur" in non-phrase-ﬁnal positions. Selkirk analyzes the strong forms as cases where PWord structure is built over the item in question, either to support the phonetics of focus or because ends of phrases are preferably also ends of words. In the remaining cases, the function words enter the structure as prosodically deﬁcient items—phonological clitics in the sense of (2.17). They must thus be integrated into prosodic structure in some way to satisfy Full Interpretation. Consider ﬁrst the cases of weak monosyllabic function words that are not phrase ﬁnal. Since these do not bear stress, they cannot be PWords. They cannot even be organized into feet, because the resulting structure would produce additional secondary stresses. In a phrase such as o˘r f˘or cónferences the two initial syllables are without stress, showing that neither is the head of a foot. We can conclude a fortiori that neither alone, nor the two taken together constitute a PWord, and thus the PWord clitic analysis is excluded. But then how are they grouped together with the lexical word conferences into a PPhrase? They cannot have been incorporated into the PWord [kánf@r@ns@z] as Internal clitics

The Phonology of Cliticization

49

either, because English excludes sequences of two or more unstressed syllables at the beginning of a PWord. But a series of non-ﬁnal phonological clitics can give rise to much longer unstressed stretches, as in o˘r f˘or a˘ c˘onvérsion. The stress patterns found with English function words in non-phrase-ﬁnal position thus argue that they are attached to the PPhrase but are not part of a PWord. This is conﬁrmed by other phenomena as well. PWord initial voiceless stops are aspirated, even when the syllable they initiate is unstressed, as in Canadian [kh@˘ néjdi@n], but this is not true of PPhrase initial function words, as in can aid (orphans) [knéjd]. It is necessary to conclude that function words that are non-ﬁnal in their "phrases are phonological clitics of the type identiﬁed as Free clitics, with a structure like (3.19), the mirror image of (3.14b). (3.19)

PPhr PWd

kn eid " They must thus violate Exhaustivity, at least at the level of the PPhrase. English also contains structures involving function words structured as Afﬁxal clitics (3.14d), Selkirk argues. Following the analysis of John McCarthy (1991) for his own Eastern Massachusetts dialect, she shows that intrusive r appears in this form of English after a PWord-ﬁnal low vowel, when the following word begins with a vowel. The one class of weak function words that do not receive stress (and PWord status) when appearing in phrase-ﬁnal position are weak object pronouns, as in hit’m again! In a sequence of verb plus unstressed object pronoun, the verb itself retains its status as a PWord, as shown by the fact that intrusive r appears at its right edge under appropriate conditions: sawr us, withdraw-r it. But the pronoun also behaves as if it were at the right edge of a PWord, in sequences such as I saw ya-r and asked about it. The structure of verb plus weak object pronoun is thus evidently as in (3.20), violating Non-recursivity at the level of the PWord. (3.20)

PPh PWd PWd saw

us

The conclusion that the motivated structures for phonological clitic-plushost combinations in English involve violations of the principles of Exhaustivity and Non-recursivity is by now quite well supported for a variety of other languages as well. One well-known case involves a set of Neo-Štokavian

50

Aspects of the Theory of Clitics

dialects of Serbo-Croatian analyzed by Zec (1993). The properties of tonal accent in these three dialects provide a minimal triplet for phonological clitic structures. The same unstressed preposition u ‘(in)to’ combines with a following PWord such as graad ‘city’ or glaavu ‘head’ as an Internal clitic in one dialect (Eastern Herzegovina), a Free clitic in another (Belgrade), and an Afﬁxal clitic in a third (Šrem, Maˇcva). We will see other examples of these structures later in this book. Constraint Systems We might take these examples to suggest that Strict Layering is just a “tendency,” or a matter of “markedness,” but surely that is just a cop-out. Or we might take them to show that the principles in (3.17) are just wrong. But throwing the baby out with the bathwater in that way is not satisfactory either. Selkirk suggests in each case that what is actually going on is subtler than this: the requirements in (3.17) are a part of the picture, all right, but sometimes their demands are overruled by others. Increasingly, work in phonology, morphology, and syntax shows that principles of grammar are not absolute. Rather, they are violable, true to the extent that some more important principle does not require them to be violated. This is a familiar enough notion in life . . . “Thou shalt not steal,” and we do not, at least not until we need to in order to feed our family. This is the core notion of Optimality Theory (or “OT”), which forms the basis of several recent analyses of cliticization phenomena (including Selkirk’s). The basic idea of OT is that the structure of a grammar comprises a system of ranked but violable constraints rather than a set of ordered rewriting rules. Rules and constraints provide alternative algorithms for answering the question: given the abstract (input) representation of a form, how do we determine the overt shape of its realization? The rule-based answer says: start with the input, and apply a sequence of systematic modiﬁcations to it, resulting eventually in the output. This format allows us to express a great deal, but each such modiﬁcation has an all-or-nothing character to it. Either a given rule applies or it does not. Of course, another rule may apply later and obscure what this rule has done, but either it applied or it did not. Within OT, things are quite different. For any given input, the output could in principle differ from it in any arbitrary way. Therefore, again in principle, any well-formed structure within the vocabulary of phonological expressions might be the output corresponding to the given input. Taking that as the starting point, we could assume that the problem is to provide a set of “triage” conditions—the constraints—for sorting out all of these (logically) possible outputs and arriving at the most suitable: the optimal one.

The Phonology of Cliticization

51

In those terms, we can say that the constraints may (indeed, must) be ranked. This ranking establishes a balance between two contradictory inﬂuences on the output form. On the one hand, the output should look as much like the input as possible (Faithfulness), but on the other hand, the output should also conform to general principles of Markedness. The relative rankings of speciﬁc Faithfulness and Markedness constraints determine the way these conﬂicts are resolved in the choice of the optimal output form. Some things in the theory of OT as it has evolved in phonology (and more recently, in syntax) are essential: notably, the description of an input–output relation as based on a ranked set of violable constraints evaluated simultaneously. Other matters are contingent, part of the speciﬁc implementation of this idea that has been widely explored in phonology. One of these is the claim of universality for constraints, a proposal which sometimes seems intended primarily to avoid the question of how speciﬁc constraints might be learned. There is clearly a general component to constraint systems, but whether everything but the ranking can be ascribed to Universal Grammar is much less obvious. A second, even more controversial claim is the principle that there is only a single step between input and output, mediated by a single ranked constraint system. Some fairly common and well-established situations in phonology, such as opaque interactions among regularities, pose very serious problems for this idea, and have led to rather convoluted supplements to the theory that have not gained wide acceptance. There are also reasons to believe that basic insights of the cyclic organization of phonology and morphology as this developed within the theory of Lexical Phonology are quite real and should be preserved, although that grammatical architecture is quite at odds with the one-step conversion process assumed in much of the OT literature. In this book, I adopt a modiﬁed version of OT, along lines developed in recent work of Kiparsky (2000, forthcoming); see also Booij (1997). Speciﬁcally, I assume a cyclic organization of the Lexical Phonology of words, where constraints operate to instantiate the phonology of a given level of the Lexical Phonology, as well as that of the Post-lexical phonology, in such a way that there is more than one constraint-satisfaction operation in the phonology as a whole. The analysis of phonological clitics is almost exclusively a matter of the Post-lexical constraint system, on this view. For particular items, several issues must be resolved. First, we must establish whether the item’s prosodic deﬁciency is inherent or derived by rule. I assume (with Selkirk, for instance,

52

Aspects of the Theory of Clitics

among others) that items that do not belong to lexical categories are prosodically deﬁcient, at least to the extent of not having inherent PWord structure. This follows from their not having passed through the parts of the Lexical Phonology that create such structure. Perhaps, however, some other principles (e.g., fast speech reductions) can deprive even lexical material of some of its prosodic organization. Second, we must establish the way in which phonologically clitic material is to be incorporated into the prosodic structure of neighboring material so as to satisfy Full Interpretation (3.4). At what level of prosodic structure (syllable, Foot, PWord or PPhrase) is the item attached, and what is the resulting structure? Whatever this is, it must be consistent with the prosodic hierarchy, based on the ranked categories of (3.15). Furthermore, as we discussed above, no structure in any language appears to violate the additional requirements in (3.21), and so presumably the only candidates submitted to the constraint system for evaluation conform to these. (3.21)

Layeredness: No category dominates a higher-level category. Headedness: Every category directly dominates (at least) one element no more than one level below it on the hierarchy.

The additional conditions of the Strict Layering Hypothesis can, as we have seen, be violated. Furthermore, violation may be ‘local’ in the sense that a language violating, say, Exhaustivity at the PPhrase level may nonetheless conform to this principle at other levels, such as the PWord. The relevant principles, therefore, need to be formulated as families of constraints, varying over the categories of the hierarchy as in (3.22). (3.22)

Exhaustivity(Ci ): Every element of category Ci is exhaustively composed of elements of category Ci−1 . NonRecursive(Ci ): No element of category Ci directly dominates another instance of Ci .

A description of the phonology of Stray Adjunction (in the terminology of Chapter 2) includes a ranking of these constraints (perhaps with respect to others), as well as a resolution of such other matters as the choice of a host on the left or the right, to which I return in section 3.3 below. An Example: Italian Dialects As an example of how we might describe various modes of phonological cliticization, let us review the descriptions of a set of three Italian dialects discussed by Peperkamp (1997). These provide a nice contrasting set, differing minimally in the way clitics are incorporated into prosodic structure which can be described in terms of varied rankings of

The Phonology of Cliticization

53

the constraints seen thus far. The ﬁrst of them is Neapolitan, as illustrated in (3.23). (3.23)

Neapolitan: Verb Imperative do fá tell cónta comb péttina

Imperative + ‘it’ fáll@ cóntal@ péttinal@

Imperative + ‘you’ + ‘it’ fattíll@ cóntatíll@ péttinatíll@

We can assume that PWords are built lexically over the host verbs, and then prosodically deﬁcient clitics are added post-lexically. Note that when clitics are added, the ﬁrst stress does not change except in one case (fattíll ), where we can say that the new stress appearing on the clitic sequence has the effect of suppressing the original stem stress to avoid Clash (a sequence of two adjacent stresses). Peperkamp shows that we can describe this system by saying that the clitic material is adjoined to the existing prosodic word, without modifying its structure, as in (3.24). e

(3.24)

a.

PPh

b.

PPh

PWd

PWd

PWd

PWd

F. . .

σ

F. . .

lex

cl

lex

F cl1

cl2

I assume that a single clitic is a single syllable, and not a Foot; two clitics, however, provide enough material to constitute a Foot, and thus to introduce an additional stress. Peperkamp’s discussion suggests that there are aspects of formal suppletion that require the treatment of the two-clitic sequence as a single unit, which is eligible to be a Foot. Alternatively, we could asume simply that the two monosyllabic units are introduced together, and subsequently organized into a Foot. The choice makes no difference. We can describe this system as follows. Full Interpretation, Headedness, and Layeredness are all undominated well-formedness conditions on the candidates that are to be compared, so they play no part in the ranking. It is also the case that prosodic structure assigned lexically is generally preserved, so we assume a high-ranking Faithfulness constraint to this effect. (3.25)

Prosodic Faithfulness: Prosodic structure in the input should be preserved in the output.

In the case of a monosyllabic stem followed by two clitics, however, the need to avoid successive stressed syllables is more important, so the stress on the

54

Aspects of the Theory of Clitics

stem is lost as a result of the domination of Prosodic Faithfulness by another constraint (3.26).12 (3.26)

*Clash: Sequences of two consecutive stressed syllables are disallowed.

The prosodically deﬁcient material (i.e., the clitics) must be incorporated into the structure somewhere, and there are not a lot of choices available. Incorporation into a Foot would violate well-formedness conditions on feet, as well as Faithfulness to existing prosodic structure. Incorporation into the existing PWord would also violate faithfulness. Incorporation at the PPhrase level would violate Exhaustivity(PPh). The Afﬁxal clitic structures we actually ﬁnd indicate that Exhaustivity(PPh) outranks NonRecursivity(PWd): that is, building a recursive PWord preserves the existing prosodic structure, and avoids having lower-level constituents (syllables, Feet) directly dominated by a PPhrase. The overall constraint ranking for Neapolitan is as in (3.27). (3.27)

*Clash Prosodic Faithfulness Exhaustivity(PPh) NonRecursive(PWd)

Now let us compare the Neapolitan approach to Stray Adjunction with that employed in another dialect, Lucanian. (3.28)

Lucanian: a. vínn ‘sell’; v nníll ‘sell it’ e

e

e

b. rammíll ‘give me it’; mannat míll ‘send me it’ e

e

e

We see in (3.28a) that the addition of a clitic in this language causes stress to shift rightward.13 Apparently a binary trochaic foot is constructed over the last two syllables of the form, including both stem and any following clitics. The forms in (3.28b) with two clitics have this foot constructed entirely over clitic material. In this language, Stray Adjunction produces Internal clitics, sacriﬁcing Faithfulness to maintain the Strict Layering constraints. The resulting structure for a form with two clitics is as in (3.29). (3.29)

PPh PWd Ft mannat@ mí

ll@

12 The fact that it is the ﬁrst, rather than the second of two adjacent stresses that is lost must be resolved by other aspects of the prosodic phonology of Neapolitan not considered here. 13 Stress shift is responsible for the vowel alternation in these forms, with stressed [´ı] corresponding to unstressed [˘@].

The Phonology of Cliticization

55

The constraint ranking necessary to obtain this result is (3.30). (3.30)

NonRecursive(PWd),Exhaustivity(PWd) Prosodic Faithfulness

Let us ﬁnally compare the situation in (standard) Italian, illustrated in (3.31). (3.31)

Standard Italian: a. pórta ‘bring’, pórtami ‘bring me’ b. pórtamelo ‘bring me it’, teléfonamelo ‘telephone it to me’

Here the addition of a clitic does not alter the lexically assigned stress, suggesting that Faithfulness is highly ranked. Even when two clitics are added, as in (3.31b), the stress is not altered, and apparently no new stress is assigned even though two syllables of additional material would support the construction of a new Foot if this material were within the PWord. Apparently, then, Stray Adjunction in Standard Italian produces Free clitics by attachment to the PPhrase, as in (3.32). (3.32)

PPh PWd pórta me

lo

The required ranking is that of (3.33). (3.33)

NonRecursive(PWd),Exhaustivity(PWd),Prosodic Faithfulness Exhaustivity(PPh)

Stray Adjunction in these three Italian dialects is thus based on different rankings of the prosodic constraints, yielding three different structural types of clitic as a reﬂection of these differences in their post-lexical phonology.

3.3 Prosodic Structure and Syntactic Structure In discussing the operation of phonological cliticization above, I have assumed in each case that the appropriate host for a given clitic is apparent, and that the problem is to determine how that clitic and its host relate to one another in the structure resulting from Stray Adjunction. When a clitic appears between two prosodically complete potential hosts, however, it is necessary to establish the direction in which incorporation will take place. In some cases this is a purely phonological matter, but in others it depends (at least apparently) on the syntactic structure within which the material in question appears. I turn

56

Aspects of the Theory of Clitics

now to the issue of how syntactic and phonological representations interact to determine these matters. As I noted at the beginning of this chapter, “classical” Generative Phonology assumed that the relation between syntactic structure and phonology was a basically simple one: phonological rules referred directly to syntactic form, and could, at least in principle, make use of any and all information they might ﬁnd there. A complication recognized at least as early as Chomsky and Halle (1968), however, was that the structures provided by the syntax might not be completely appropriate for the phonology. For instance, syntactic organization tends to involve greater degrees of embedding as sentences get longer, but the division of sentences into phrases is comparatively ﬂat. Indeed, the phrasing suggested by the syntax may not be the same as that suggested by the phonology. The classic example of this is a sentence like This is the cat that chased the rat that ate the cheese . . . , where syntactic phrasing groups nominal heads together with the following relative clause but the phrase boundaries in pronunciation come between the head and the relative clause. (3.34)

Syntax: [ This is [ the cat [ that chased [ the rat [ that ate CP

DP

CP

DP

[ the cheese [ that . . . ]]]]]]] DP

Phonology: [

CP

CP

PPh

ate the cheese] [

This is the cat][

PPh

that . . . ]

PPh

that chased the rat][

PPh

that

To accommodate such divergences, SPE assumed that a set of Readjustment Rules operated before any of the phonological rules proper, to massage the structure in various ways so as to make it appropriate for phonological interpretation. Though this notion was never pursued in any detail in the literature of the period, the principal lesson to be drawn from the necessity of such readjustments is surely that the organization of an utterance for syntactic purposes on the one hand, and for phonological ones on the other, is simply not the same. The development of prosodic phonology, as sketched in the sections above, brings this point into greater relief. If substantial hierarchical organization has to be imposed on a phonological representation for its own purposes, the need for the phonology also to consult another similar but non-identical and logically independent hierarchical structure (that provided by the syntax) is not obvious, and risks substantial redundancy where there is not actual conﬂict. In fact, the approach that has emerged (and which I will follow here) assumes that the role of syntactic representations in phonology is limited to providing information that may be consulted (along with other, purely

The Phonology of Cliticization

57

phonological factors) in the construction of prosodic structure. The rules of the phonology per se only have access to the prosodic organization, and not (directly) to the syntax. Let us proceed, then, on the assumption that the only reference made in the phonology to categories and structures of the syntax is by the principles that construct prosodic organization. In the context of an OT approach to phonology, the relevant principles are Alignment constraints, which specify preferred relations between two distinct analyses of the same material. The general form of such a constraint is as in (3.35). (3.35) Align(Cati ,Edger/l ,Catj ,Edger/l ) A speciﬁc instance of this family of constraints is to be interpreted as specifying that “For any instance of Cati , its Right/Left edge should be aligned with the Right/Left edge of an instance of Catj .” Perhaps the simplest instantiation of this system is the description of the fact that lexical words project PWords in the phonology. We can express the preference for this structure in terms of the ranking of the two constraints in (3.36). (3.36)

a. Align(LexWord,L,PWord,L) b. Align(LexWord,R,PWord,R)

These say that, for any stretch of material that constitutes a lexical word, its left and right edges should be aligned with the left and right edges, respectively, of a PWord. Various other factors, however, especially the need to incorporate phonological clitics into the structure, may lead to circumstances in which this alignment is not perfectly satisﬁed. These are described in terms of other constraints that outrank one or the other of those in (3.36). In terms of higher-level categories, we want prosodic structure to reﬂect syntactic structure, in the absence of overriding factors. We can accomplish this, to a ﬁrst approximation, by a similar set of constraints. (3.37)

a. Align(XP,L,PPhrase,L) b. Align(XP,R,PPhrase,R)

The constraints in (3.37) say that, ceteris paribus, syntactic phrases (at least maximal projections) should project PPhrases in the prosodic structure. Left to their own devices, these constraints would produce much the same recursive embedding in the phonology as that of the syntax, but we already know from examples like (3.34) that a rather ﬂatter structure should be preferred. In fact, however, this will follow (to a ﬁrst approximation, at least) from the presence in the grammar of the independently necessary constraint

58

Aspects of the Theory of Clitics

NonRecursive(PPh) in (3.22). Unless other factors intervene, this constraint will prefer the non-recursive structure in (3.38a) to the recursive one in (3.38b). (3.38)

a. [

the cat] [

that chased the rat]

b. [

the cat [

that chased the rat]]

PPh PPh

PPh

PPh

Now let us see how this apparatus helps us describe Stray Adjunction, the incorporation of prosodically deﬁcient material into existing structure. In many languages, Stray Adjunction prefers to follow syntactic structure, in the sense that stray material is incorporated into the neighboring constituent with which it bears the closest syntactic relation. Recall the case of weak prepositions in Polish, for example, which form a recursive PWord with the following material that constitutes (part of) their object, as in (3.11). We need to ensure that a new PWord is not introduced to project nonlexical prosodically deﬁcient material. Assume that some form of a constraint *Struct (“Do not build structure unless necessary”), highly ranked, produces this effect for function words (where (3.36) cannot apply). Full Interpretation still requires the preposition to be incorporated into prosodic structure in some way. In this case (3.37a) requires the left edge of a PPhrase to coincide with the left edge of the PP, thus including the preposition in that PPhrase. The choices are to incorporate the stray syllable either directly into the PPhrase, or else into the following PWord, recursively (as an Afﬁxal clitic) or not (as an Internal clitic). The constraint ranking in (3.39) will produce Afﬁxal clitics as required for Polish. (3.39)

*Struct,Prosodic Faithfulness,Exhaustivity(PPh) NonRecursive(PWd)

For the case in which clitic attachment follows the direction of syntactic afﬁnity, then, we can say that the language prefers to have the edges of syntactic phrases coincide with the edges of PPhrases, as mandated by (3.37). As a result, clitics attach either as proclitics or as enclitics because the other direction would be counter-syntactic. For instance, Romance clitics are proclitic when pre-verbal, and enclitic when post-verbal, because (as I will argue in later chapters) their closest syntactic afﬁliation is with the verb. Consistent adjunction of a phrase-initial item to its left would require us to violate the constraint (3.37a), and in languages such as Kw akw ’ala, that is just what happens. We saw in section 2.2 that phonological clitics in that language systematically adjoin to material on their left, usually resulting in a violation of the similarity between syntactic and prosodic structure found in other

The Phonology of Cliticization

59

languages such as Polish. It must therefore be the case that in Kw akw ’ala some other constraint outranks one or both of those in (3.37). Suppose we want to force clitics always to associate to their left, as in Kw akw ’ala. In the present system, there is no way to say that directly. But if a clitic attaches in the counter-syntactic direction, this must be because some other constraint is thereby satisﬁed, a constraint that (in this language) outranks (3.37a). What is gained by having the clitic attach to its left? A possibility is that this avoids having the PPhrase begin with the clitic, rather than with lexical material. On that view, we can describe Kw akw ’ala by the ranking in (3.40). (3.40)

Align(PPhrase,L,LexWord,L) Align(XP,L,PPhrase,L)

That is, in this language it is more important to keep non-lexical material out of the left periphery of the PPhrase than it is to have syntactic and prosodic phrasing be isomorphic. The high ranking of a constraint requiring prosodic constituents to begin with lexical material can probably be connected to the fact that Kw akw ’ala is a language that has essentially no preﬁxes: the only exceptions to this are reduplicative elements whose content can be argued to be lexical. We can relate this situation to that in other languages where similar alignment constraints are important. In Catalan, for example, the class of ‘atonic’ prepositions including a, amb, de, en that play a role in assigning case to nominals constitute a set of simple clitics. According to Wheeler, Yates, and Dols (1999: 262), “[w]hen a ﬁnite complement clause, beginning with que ‘that’, is the object of a weak [i.e., atonic] preposition [. . . ], the weak preposition is dropped” as in the examples of (3.41). (3.41)

a. Estem contents (*de) que hagis vingut ‘We are pleased that you have come’ b. Estem contents *(de) la teva vinguda ‘We are pleased at your coming’ c. Els hem d’acostumar (*a) que no hi vagin ‘We must get them used to not going there’ d. S’ha acostumat *(a) no anar-hi ‘They have got used to not going there’

Of course we could describe this situation (as Wheeler, Yates, and Dols do implicitly) by a rule: “delete unstressed prepositions before que.” This formulation is descriptively accurate, but not particularly illuminating. We could improve on it by relating the disappearance of the (semantically empty) clitic

60

Aspects of the Theory of Clitics

prepositions to a constraint requiring the left edge of a clause (CP) to coincide with the beginning of a prosodic word, assuming that NonRecursive(PWord) would force these phonological clitics to attach directly to the PPhrase as Free clitics. If it is in fact more important for CPs to begin at a word boundary than for the empty prepositions that serve to assign case to be overtly realized, we could get the facts of deletion by ranking the alignment constraint higher than a presumed constraint Max(Pfnc) requiring semantically empty structural elements to appear on the surface. (3.42)

NonRecursive(PWord),Align(CP,L,PWord,L) Max(Pfnc)

We still need to embed this description in a fuller account of Catalan phonology, including constraints that enforce Stray Adjunction in this language in a way that otherwise aligns prosodic and syntactic structure, among other facts, but the approach seems to hold some promise. In a study which forms an important part of the background of this book, Klavans (1985) proposed a set of parameters for clitics, including one specifying the direction of their attachment. For Klavans, this directionality parameter was to be speciﬁed individually as a property of particular clitics, rather than being reducible to general properties of the language. The view presented here, in contrast, treats direction of attachment as a matter that follows from the overall prosodic properties of the language, and which is not available for lexical speciﬁcation with respect to individual items. In fact, the literature provides very few instances of directional attachment of clitics which are plausibly item-speciﬁc, and even fewer explicit arguments for the necessity of such description. Nespor and Vogel (1986: chapter 5) argue such a case for some pronominal clitics in Modern Greek. They show that clitics like mu ‘me’ and to ‘it’ are proclitic to a following verb in examples like (3.43)

o ðáskalos mu= to= ípe art teacher me it said The teacher said it to me

In contrast, when used possessively, these clitics appear as enclitics to a previous word. This is illustrated in (3.44), including (3.44a), which forms a minimal pair with (3.43). (3.44)

a. o ðáskalòs =mu to= ípe art teacher me it said My teacher said it

The Phonology of Cliticization

61

b. to prósfatò =mu árTro art recent 1sg article my recent article =mu esTímata c. ta iposiníðità art unconscious 1sg feelings my unconscious feelings d. o pio ﬁlóðoksòs =mu fílos art most ambitious 1sg friend my most ambitious friend They then suggest that in their possessive use, clitics like mu must be speciﬁed as attaching only to their left, as enclitics. This does not follow, however, within the present framework. There is a clear difference between the environment of the clitic in (3.43), on the one hand, and in (3.44) on the other. In (3.43), the Verb Phrase “me it said” surely constitutes a PPhrase separate from the subject “the teacher.” A clitic such as mu in this example, introduced as a special clitic associated with the verb ípe ‘said’ could only attach to its left by crossing the boundary between these two PPhrases. Within the DPs of (3.44), however, there is no reason to assume such a boundary between PPhrases, and thus nothing to prevent the clitic from attaching to its left as an enclitic. Suppose we say, then, that in Modern Greek (as described in Nespor and Vogel 1986) prosodically deﬁcient elements attach to a preceding PWord as enclitics where this is consistent with PPhrase boundaries projected from the syntax, but as proclitics to a following PWord otherwise. There is then no need to say anything about the direction of attachment of any particular clitic on an item-speciﬁc basis. The motivation for Nespor and Vogel’s (1986) wish to specify the possessive markers as obligatorily enclitic may be to force them into a non-initial position within the DP, by requiring them to have a appropriate host on their left. There is no need to do that, though, if we simply treat them as special clitics introduced in second position within the DP, within the framework to be developed below in Chapter 6. From example (3.44d), it appears that “second position” in this case means “following the ﬁrst full PWord,” since initial unstressed elements (articles, and the intensiﬁer pio ‘most’) do not serve as hosts. None of this, however, requires us to specify the clitics themselves for the direction of their prosodic incorporation. In the course of a detailed description of the clitics of Dutch, Booij (1995, 1996) has argued that a very limited set of weak pronouns in that language require an item-speciﬁc indication of direction of attachment. He ﬁrst establishes

62

Aspects of the Theory of Clitics

that the basic direction of prosodic integration of Dutch clitics is leftward, as enclitics to a preceding host. This is true for a wide range of pronouns, determiners, and other weak elements, which he argues are incorporated as Internal clitics into the preceding PWord, adducing a substantial number of phonological phenomena in support of this analysis. The leftward attachment of clitics is attributed to general principles of Dutch prosody, and is entirely consistent with the position espoused here. Booij then notes that when clitics appear in a position where there is no possible host on their left (particularly in sentence-initial position), they attach instead as proclitics to the PWord on their right. The resulting structure, in which the clitic is adjoined to its host as an Afﬁxal clitic, is then shown to be entirely parallel to the structure of weak derivational afﬁxes in the language. Since the same clitic may attach either leftward (in the general case) or rightward (when necessary), he notes that these elements could not be lexically speciﬁed for a direction of attachment. Rather, that question must be determined on the basis of the way general properties of Dutch prosody affect particular structures, without regard to the identity of the individual element. There are two Dutch pronominal clitics, however, for which Booij suggests that a lexical speciﬁcation of direction of attachment is necessary: the weak form ie ‘he’, [i], and its allomorph [di]. These, he argues, can only attach leftward. When no host is available on the left, they cannot be used, unlike other Dutch clitics. (3.45)

a. Komt-ie? ‘does he come?’ [ kOm][ ti] b. c. d. e.

σ

σ

*Ie komt ‘he comes’ Kan-die ‘can he’ *Die kan ‘he can’ Het gaat wel goed ‘it goes well’ [ @t][ xat] σ

σ

This argument is not decisive, however. In fact, [i] and [di] differ from the rest of the Dutch clitics in another way: they are the only ones whose weak form contains a full vowel, rather than a schwa. Syllables containing schwa are not eligible to constitute PWords by themselves in Dutch, while syllables containing a full vowel can stand by themselves, even when vowel initial like [i]. This suggests that while most of the phonological clitics of Dutch—those whose vowel is schwa—consist only of a stray syllable, [i] and [di], in contrast, constitute a stray foot. On that basis, the idiosyncrasy of [i] and [di] ﬁnds an explanation consistent with the position taken here. As Booij shows, a proclitic form integrates with its host as an Afﬁxal clitic, as in the structure (3.46) for de keer ‘the turn’.

The Phonology of Cliticization (3.46)

63

PWord PWord

σ

Ft σ

d@

ker

If a clitic constituting a complete foot such as [i] or [di] were placed in the same way within this structure, the resulting initial foot would create a new initial stress within the larger PWord, violating the Dutch stress pattern or (if the original stress on the host were deleted) producing a stress pattern unfaithful to that of the input host. It seems reasonable to attribute the ill-formedness of proclitic [i] and [di] to such factors, rather than to a lexical speciﬁcation for direction of attachment. Booij points out (personal communication) that there are preﬁxed words in Dutch, such as overkomen ‘to happen to somebody’ in which the preﬁx (here, over-) consists of a Foot, and this Foot bears (secondary) stress. This construction is a lexical item, however, and so the prosodic structure involved is created by the lexical phonology, not the post-lexical phonology as would be the case with the integration of a clitic such as ie into prosodic structure. If we say that a new secondary stress in a Foot preceding and adjoined to the PWord with main stress cannot be created post-lexically (though the same conﬁguration created lexically is preserved by high-ranking Prosodic Faithfulness), we can ensure that initial ie will be blocked without having to stipulate its direction of attachment as a lexical property of the clitic. I conclude that the Dutch data do not prevent us from dispensing with Klavans’s parameter of (item-speciﬁc) directionality. Quite generally, the choice of proclitic or enclitic attachment can be made to follow from more general principles of a language’s prosodic organization. Sometimes the direction of cliticization is partially obscured by postlexical adjustments of prosodic structure. An example of this is furnished by Gallo-Romance (Old French) facts cited by Jacobs 1993. Here a clitic between the subject and the verb is generally enclitic to the preceding word, but proclitic if the following verb is vowel initial. Jacobs argues that at this point in history, the left edges of PWords coincided with the left edges of lexical words (similar to Kw akw ’ala). As a result, there was normally a PWord boundary before the Verb, causing the clitic to associate leftwards so as to produce the usual enclitic structure. On the other hand, the language also ranked highly the constraint Onset requiring syllables to have onsets. In the post-lexical phonology, this outranked the PWord–LexWord alignment constraint. When a lexical verb began with a

64

Aspects of the Theory of Clitics

vowel, the constraint violation in its initial syllable could be repaired by shifting the clitic across the PWord boundary so as to provide the requisite initial onset. Examples are given in (3.47). (3.47)

a. b. c. d. e.

Jot (= jo+te) vi ‘I saw you’ jo t’aim (= te+aim) ‘I love you’ . . . qu’il parloient einsi . . . ‘that they were speaking thus’ qu j’en recevroie tel cop ‘that I would receive such a blow from it’ Si com j’es livres sui lisant ‘Just as I am reading in books’

The post-lexical constraint ranking required to achieve this result is as in (3.48). (3.48)

OnsetAlign(PWord,L,LexWord,L),Prosodic Faithfulness

Along similar lines, Chamicuro (Parker 1999) shows a constraint (or constraint complex) requiring PWords to end in a light syllable. In this language, the (prosodically deﬁcient) article usually attaches to the right as a proclitic, yielding a prosodic structure in harmony with that of the syntax. This situation is disturbed, however, when the preceding word ends in a heavy syllable and the article itself is light. Under those circumstances the clitic article shifts to its left prosodically, becoming an enclitic on the preceding word and thus satisfying the ﬁnal light syllable requirement on that PWord. Developing the relevant constraints to express the prosodic requirements on PWords in Chamicuro would take us too far aﬁeld here, and I do not propose to formulate a complete analysis that would achieve the observed interaction between prosodic and syntactic factors, but the general lines should be clear within the system of this book.

3.4 Phonological Clitics and Cliticization in English Let us now return to the case of contracted auxiliary verbs in English, which was raised in section 2.4 of the previous chapter. Recall that I proposed to follow Kaisse in treating items like ’s, ’d as prosodically deﬁcient lexical variants of full word forms (is, has, had, would, etc.). The problem, if we are to maintain a treatment of them as simple clitics, is to account for the distribution of these elements in terms of the phonology alone in the face of apparently syntactic restrictions on their occurrence. Within the theory adopted here, it is not possible to allow the phonology to consult the syntactic representation directly (to discover the presence of a following gap, for instance): the only way

The Phonology of Cliticization

65

syntactic form can inﬂuence the phonology is through its role in constructing the prosodic representation. We need to account for the following apparent paradox. On the one hand, the reduced auxiliaries display an obvious phonological dependency on the material to their left, and no sensitivity to that on their right. On the other hand, the presence of a syntactic gap on the right precludes the use of the reduced form of the auxiliary, and requires the full (non-clitic) form, while left syntactic context is irrelevant. Finally, whatever is going on, it has to be something that follows largely from general principles, since children learn how to use the reduced auxiliaries quickly, without making mistakes and without being provided with negative data. If we reject an account on which cliticization is actually a syntactic operation, how could syntactic effects (like the presence of a gap) show up as conditions on cliticization? Let us take a hint from Kaisse’s proposal above in section 2.4, on which the choice of allomorph (i.e., determination of the possibility of the clitic form) is made after sentence prosody has been assigned. That suggests that sentence prosody might be what is really relevant. The appearance of a syntactic condition on clitic auxiliaries could be just the reﬂex of the effects of syntactic form on sentence prosody. A simple possibility along those lines is suggested by the core cases in which clitic forms of the auxiliary are impossible, such as (3.49). (3.49)

a. John is happier with his marriage than his wife is/*’s [e]. b. Claire has published more books than Fred has/*’s [e] articles. c. What Harry is/*’s [e] is a master of the story-teller’s art.

In the phrasing of these and similar sentences with gaps, it can be seen that in each case the potentially clitic auxiliary appears at the end of a phonological phrase. We might then note that a phrase-ﬁnal verb must bear the nuclear stress of the phrase. The clitic forms of the auxiliaries, however, are prosodically ineligible to be stressed, due to their lack of phonological content (they consist of a single segment, not even a syllable). As a result cliticization, construed as choice of a clitic allomorph, will be blocked. The relevance of the gap is that its ﬁller would otherwise follow the item in question within the phrase, and the absence of that material causes the auxiliary to be ﬁnal in its phrase. While promising, this approach is probably insufﬁcient. One problem is that it does not connect these facts with the other thing we know clearly about the contracted auxiliary forms: phonologically, they attach to a preceding word. Another is that in constructions like (3.50), the degree of stress on

66

Aspects of the Theory of Clitics

the full auxiliary verb is quite low, but that does not improve the acceptability of the clitic form. (3.50)

a. Do you know what that is/*’s [e] in the tree ahead of us? b. Pat’s happier than is/*’s [e] his brother-in-law.

It does not look as if accent per se will furnish the required explanation, though that does not mean prosodic structure is irrelevant. Recall that our problem is to account for the fact that not only are sentences like (3.51a) impossible in English, but language learners know this without ever having been explicitly told not to use reduced auxiliary forms in such positions. (3.51)

a. *Tim’s happier than Kim’s. b. Tim’s happier than Kim is.

What differentiates the unacceptable (3.51a) from its well-formed neartwin (3.51b)? The only apparent difference is in the ﬁnal phrase: syntactically [VP’s [e]], in (3.51a), versus [VPis [e]] in (3.51b).14 While reﬂecting no distinction of syntactic structure or meaning, this does have one consequence. If each of the syntactic phrases in these examples is reﬂected by a PPhrase in the prosodic structure, the content of that PPhrase contains only a single phonological unit. In (3.51a) that unit is a clitic (’s), while in (3.51b) it is the full form is. As noted in section 3.2 above, Selkirk (1995) has established that (with the exception of post-verbal object pronouns) function words in English (including auxiliaries) are Free clitics, attaching directly to the PPhrase rather than having a PWord build over them. The cases she considered were those in which the function word consists of a full syllable. The contracted auxiliaries, in fact, are the only instances in which a function word consists of only a segment, not a syllable, and their phonological behavior shows that they must be integrated into prosodic structure in a different way. Let us assume that (in English, at least) stray segments can only be incorporated as parts of a syllable, not directly as constituents of a Foot, PWord, PPhrase, or higher level category. But then if Prosodic Faithfulness is ranked highly enough to require the preservation of existing syllables at the point where the clitic auxiliaries are under consideration, there is only one way to incorporate a stray segment such as one of these forms: by adjunction, as a syllabic afﬁx to the syllable immediately to its left. The complete prosodic structure of Tim’s leaving, then, is: 14 I represent this phrase as a VP, but the category is not relevant. All that matters is the claim that the predicate of a sentence constitutes a phrase.

67

The Phonology of Cliticization (3.52)

IntPhrase PPhrase PPhrase PWord

PWord

Ft σ

Ft

σ

tIm

z

σ

σ

lij

viN

We know that adjoined structures at the level of the syllable are possible. English syllables have a structure which is rather tightly constrained internally; but much that is valid about the sequences of segments that are possible in codas, and about relations between nucleus and coda, is violated by syllables containing the “syllabic afﬁxes” /z/ and /d/. Final clusters such as that in texts [tEksts] occur only under these circumstances, and it seems reasonable to regard the syllabic afﬁxes as adjoined to the basic syllable, rather than incorporated into it. The adjoined structure is the one assigned to a number of diverse constructions in English: regular plurals and past tenses, as well as the thirdperson singular present of regular verbs, all formed in the morphology; possessives such as Rick’s ofﬁce, formed in the syntax, and sequences of a host word plus reduced auxiliary clitic. All display the same relation to normal syllable structure, and all display the same phonological modiﬁcations of the sequences that result (cf. (2.24) in section 2.4 above). Adjunction of the stray segment represented by a clitic auxiliary to the preceding syllable as an afﬁx is thus possible. Given other constraints, including Prosodic Faithfulness and the impossibility of syllable-initial afﬁxes, this is indeed the only possible way for these elements to satisfy Full Interpretation. It is also exactly what we expect, given the clear evidence which has already been noted that the reduced auxiliaries depend phonologically on material to their left. But now notice another consequence of this manner of incorporating the clitic auxiliaries into prosodic structure. When the auxiliary would otherwise be PPhrase-initial, adjunction to a preceding syllable effectively shifts it out of the PPhrase to which it would otherwise belong, since the preceding syllable belongs to a distinct PPhrase, and a single syllable cannot be shared by more than one higher-level constituent. This in turn suggests an approach to the ungrammaticality of (3.51a). In this sentence, the ﬁnal PPhrase which is motivated on syntactic grounds, corresponding to the syntactic phrase [ ’s [e]], would have the structure [ z]. VP

PPh

68

Aspects of the Theory of Clitics

But after the restructuring required to incorporate the stray element /z/, this PPhrase would have no phonological content at all. We can thus suggest that what blocks the use of clitic auxiliaries in such contexts is the fact that they would lead to violations of (3.53), a constraint which is arguably a universal well-formedness condition on phonological representations. (3.53)

*[

PPh

∅]: Phonetically empty PPhrases are disallowed.

If (3.53) is to serve as an explanation for the cases in which contracted auxiliaries are not possible, it must be established that the prosodic structure of the relevant sentence types contains a PPhrase whose only phonetic content is a reduced auxiliary, an element that is necessarily aligned out of its original PPhrase as a result of Stray Adjunction. This, in turn, requires a full account of the principles that construct prosodic representation on the basis of syntactic structure in English. I do not attempt a comprehensive description of this sort here, but rely instead on the proposals in Selkirk 1995. To a large extent, prosodic and syntactic structure are similar. In particular, Selkirk argues that the edges of lexical words and of PWords preferably coincide. (3.54)

a. LWdCon: Lexical word boundaries should be PWord boundaries. i. Align(LexWord,L,PWord,L) ii. Align(LexWord,R,PWord,R) b. PWdCon: PWord boundaries should be lexical word boundaries. i. Align(PWord,L,LexWord,L) ii. Align(PWord,R,LexWord,R)

Similarly, the edges of syntactic maximal projections of lexical categories preferably coincide with the edges of PPhrases. (3.55)

LPhCon: Boundaries of lexical maximal projections should be PPhrase boundaries. a. Align(LexMax,L,PPhrase,L) b. Align(LexMax,R,PPhrase,R)

Furthermore, at least the right edge of a PPhrase is preferably aligned with the right edge of a PWord. (3.56)

Align(PPhrase,R,PWord,R): A PPhrase should end with a PWord.

The Phonology of Cliticization

69

These constraints do not by themselves determine the way in which nonlexical items, or “function words,” are incorporated into prosodic structure, except through their interaction with the constraints of the prosodic hierarchy more generally and the condition of Full Interpretation. The result is that PPhrase ﬁnal function words (such as the particle at in What did you look at (last time)) have a PWord built over them, while function words that precede the lexical head of a phrase are incorporated as Free clitics, attached directly to the PPhrase. Selkirk shows that the correct structures follow from the ranking in (3.57) of these constraints. (3.57)

LPhCon, Align(PPhrase,R,PWord,R) LWdCon, NonRecursive(PWord) PWdCon, ExhaustivePPh

PPhrases on this analysis are generally built on the basis of the lexical material comprising a syntactic XMax. This PPhrase may also include following function words belonging to the same item, like the at of look at, over which PWords are constructed as just noted. The lexical material is preceded within the PPhrase by Free clitic function words belonging to the same or an including syntactic phrase until the right edge of another PPhrase is reached. Consider now some of the sentence types from which contracted auxiliaries are excluded. (3.58) wh-movement: I wonder where the concert is/*’s [e] on Wednesday. Gapping: Fred is tired of Spinoza, just as Mary is/*’s [e] of Schopenhauer. Adj-preposing: Tired as he is/*’s [e] of his job at the car wash, Fred refuses to look for something better. In each of these sentences, it is clear that a PPhrase boundary occurs just before the auxiliary is. This auxiliary, in turn, is immediately followed by a syntactic gap, and it is exactly the claim of the standard account of contracted auxiliaries that they are excluded when followed by a syntactic gap. But attention to the prosodic structure shows another property of these structures: in each case, a new PPhrase clearly begins in pronunciation immediately after the gap. We could say, then, that the conditions which exclude contracted auxiliaries do not refer directly to the syntactic gap per se, but rather depend on a PPhrase boundary induced by the structure in which it occurs. This, in turn, might be made to follow from a highly ranked constraint such as (3.59).

70 (3.59)

Aspects of the Theory of Clitics Align([e],R,PPhrase,L): Syntactic gaps are followed by a PPhrase boundary.

This constraint encourages the construction of a PPhrase in each of the sentences in (3.58), with its left edge immediately following the (post-auxiliary) gap.15 If the full form of the auxiliary is chosen, it will project a PWord, as with other phrase-ﬁnal function words. If the contracted form is chosen, however, Stray Adjunction forces it into the preceding PPhrase, producing a violation of (3.53) and rendering the sentence unacceptable. If this were the whole story, the resulting analysis would be a fairly minor variation on the traditional picture, which relates the ill-formedness of a contracted auxiliary to the presence of a following syntactic gap. That is not the case, however. My claim is that the factor which excludes the reduced auxiliaries is not the gap itself, but rather the prosodic structure which it induces. And this is conﬁrmed by other facts. Pullum and Zwicky (1998) note that some constructions containing gaps block the appearance of reduced auxiliaries, despite the fact that the gaps in question do not immediately follow the position of the auxiliary element, as in (3.60). (3.60)

Comparative Subdeletion: Marie’s Pierre is/*’s an [e] engineer.

a

better

scientist

than

Subject-Aux Inversion: Marie’s better known than is/*’s her husband [e]. In these cases, we can note that the material following the auxiliary element is set off by the left edge of a new PPhrase. In the case of Subject-Aux Inversion, this phrasing follows from the general principles we have discussed. It is not obvious that the same is true for Comparative Subdeletion, but regardless, it is clear that this phrasing is associated with the construction. As a consequence, in each of these structures the auxiliary element will ﬁnd itself alone in a PPhrase beginning at its left edge, such that its phonological attachment to the preceding word will produce a violation of (3.53). Only the selection of the full form of the auxiliary (over which a PWord can be built, given its PPhrase ﬁnal position) will allow the structure to be well-formed. 15 This is similar in effect to a proposal of Zec and Inkelas (1990) to the effect that a dislocation site in the syntax initiates a new prosodic phrase in the phonology. It should be noted, however, that a number of authors, including Nespor and Vogel (1986) and Truckenbrodt (1999), have argued that phonologically empty syntactic categories are not visible to the processes that construct prosodic structure.

The Phonology of Cliticization

71

Indeed, contracted auxiliaries are excluded even in structures with no (relevant) gap at all, but where violations of the condition (3.53) arise. Bresnan (1978) observed that a following parenthetical expression as in (3.61a) blocks contracted auxiliaries. (3.61)

a. John is/*’s, my dear, a bastard. b. John, my dear is/’s a bastard.

Bresnan attributed this to the role of the parenthetical in blocking a supposed cliticization to the right, but in fact the clear phrasal break around parentheticals sufﬁces to explain the phenomenon. Whether the contracted form is also blocked following the parenthetical depends on prosody: if pronounced with a clear break before the following phrase, (3.61b) can only have the full form of the auxiliary, but if no such break is present, many speakers accept the contracted form as well. This makes clear the dependence of the contracted auxiliary on prosodic, rather than syntactic conditions. The same is true of the construction in (3.62) called “Rejoinder Emphasis” by Pullum and Zwicky (1998). (3.62)

Fred is/*’s too going to ﬁx it.

I conclude (with Pullum and Zwicky 1998) that there is no special rule of Auxiliary Reduction in English. Indeed, there is no syntactic condition at all per se that governs the appearance of clitic auxiliaries.16 The role of the syntax in this area is limited to the effect of syntactic constituency in the construction of prosodic structure. All we need to say about the auxiliaries is that each of these have two forms, either of which can in principle be chosen freely. If the full form is chosen, it will behave either as a ﬁnal or a non-ﬁnal function word, depending on its prosodic position. If the reduced (non-syllabic) form is chosen, it will necessarily undergo reattachment as an afﬁx to a preceding syllable, resulting in a violation of (3.53) where this results in a phonetically empty PPhrase. The reduced forms of English auxiliaries are thus simple or phonological clitics in these sense of this book. 16 Kaisse (1985) argues for additional conditions on the relation between the auxiliary and preceding material, speciﬁcally the “Government Condition” which stipulates that auxiliaries may cliticize only onto a constituent that they govern. This would block sentences such as Speaking tonight’s our star reporter to which Kaisse assigns a star, but which I and nearly all speakers with whom I have discussed the matter ﬁnd perfectly acceptable. If some such condition nevertheless exists, I would interpret it as governing the formation under appropriate syntactic circumstances of prosodic constituents at higher levels, such as the IntPhrase, whose presence blocks the incorporation via Stray Adjunction of the clitic auxiliary into the preceding PPhrase, resulting in its stranding in a prosodically ill-formed position.

72

Aspects of the Theory of Clitics

Let us now return to the observation that children acquire the use of the contracted auxiliary forms in English rapidly and essentially without error. On the traditional account, this is quite mysterious. If “Auxiliary Reduction” is a process speciﬁc to English, depending on a comprehensive analysis of rather subtle syntactic factors, including as a central component the role of syntactic elements which are not manifested overtly in sentence form, it is hard to see how the learner could arrive at the correct analysis as directly as seems to be the case. My claim is that the phonology of these clitics depends only indirectly on syntactic form, and only to the extent this is a signiﬁcant determinant of prosodic form. Syntactic phrasing is not always faithfully reﬂected in phonological phrasing, but neither is the relation between the two arbitrary and unconstrained. Children are sensitive at a very early age (around eight to ten months)17 to the phonological organization of utterances into phrases, and there is no doubt that they use this structure as a crucial key to discovering the syntactic organization of sentences and phrases. In order to do this, they must assume that syntactic phrases correspond largely to phonological ones. It seems clear that the prosodic structure is available to the learner before the syntactic form. On the prosodic analysis, there is no mystery to the rapidity with which use of the clitic auxiliaries is acquired. All the child needs to learn is the fact that certain function words have optional variants which are prosodically deﬁcient (i.e., phonological clitics). This information is directly available in surface forms, since children hear both The tiger is in that cage and The tiger’s in that cage and have no reason not to treat them as optional variants. They must learn the phonology of forms such as the plural of nouns, the possessive, and the third-person singular present of verbs, and that knowledge provides them with the phonological apparatus to account for the phonetic forms of the clitic auxiliaries. They must also gain access to the prosodic structures within which these regularities play out, but as noted above, this information is probably in place before a complete syntactic interpretation is arrived at. No language-speciﬁc conditions, syntactic or otherwise, govern the contracted auxiliaries per se, once it is determined that they are phonological clitics. An understanding of the signiﬁcance of prosodic structure can probably lead to the resolution of another chronic puzzle in English, the analysis of contracted forms such as wanna, gonna, hafta, and a few others. These represent combinations of want, going, have, etc., with the to of a following inﬁnitive. 17

See de Boysson-Bardies (1999) for a review of the literature establishing this.

The Phonology of Cliticization

73

It is widely assumed that the presence of a syntactic gap intervening between the base word and to is responsible for blocking the “Contraction” in cases such as those illustrated in (3.63). (3.63)

a. New Haven is the place I want to/wanna go next. b. Fred is the guy I want [e] to/*wanna go next. c. Community college is the only school he’s going to/gonna get into. d. New Haven is the next place he’s going [e] to/*gonna ﬁnd out whether he likes the east coast. e. This is the money I have to/hafta give to my sister. f. This is all the money I have [e] to/*hafta get through the week.

The apparent role of a gap in preventing contraction has caused numerous authors to attempt to assimilate this phenomenon in some way to that of the contracted auxiliaries. Arguably, however, it is not the presence of a gap per se that blocks the use of wanna, gonna, hafta in these examples, but rather a fact about prosodic structure. Notice that in pronunciation, there is a clear PPhrase boundary preceding the to in each of the sentences where the contracted forms are impossible. No such boundary is necessary, however, in the cases where contracted forms can be used. Since this is not a book about the syntax of complementation, I will not propose explicit structures from which this intonational difference should be derived, but there is little doubt that structural distinctions are present to which it can be attributed. Suppose, therefore, that rather than being the result of a “Contraction” process, forms like wanna, etc. are lexicalized portmanteaux—in this respect, comparable to French au (=a+le) and du (=de+le). Each of the English contracted forms consists of two syllables, forming a trochaic foot. If they have this prosodic structure already at the point where they are inserted, it is clear why their use will be blocked when a PPhrase boundary should intervene between the syntactic positions of want, going, have, etc., and that of to. The prosodic hierarchy prevents a single foot from belonging simultaneously to two distinct constituents at a higher level, such as the PPhrase. As in the case of the reduced auxiliaries, no reference to a condition on syntactic structure is necessary here at all, but only an appreciation of the way such structure may inﬂuence the formation of prosodic representations. Prosodic structure, then, plays an important part in governing the wellformedness of sentences. For present purposes, the most signiﬁcant aspect of this component of linguistic form is the fact that phonological (or “simple”) clitics, material without the prosodic structure of normal lexical items, must

74

Aspects of the Theory of Clitics

be integrated into the prosodic hierarchy through Stray Adjunction and the phonological adjustments that may arise as a consequence. The study of these effects across a variety of languages forms the content of one of the two distinct dimensions of “clitic” phenomena to which this book is devoted.

4 Special Clitics and their Grammar Let us return to Zwicky’s proposed class of Special clitics as introduced in section 2.1. These were deﬁned as unaccented bound forms that are variants of free forms (similar in sound and meaning) and that display “special” syntax. I will focus in this chapter on just what this last point might mean. What, exactly, is special about the distribution of special clitics? The “special” syntax of a clitic might simply be “different” syntax from the corresponding free form, as in the French sentences in (4.1). Assuming the underlined expressions in these sentences are intended to refer to the same individual, when that reference is made by a full DP, this phrase follows the ﬁnite verb, while the corresponding argument appears pre-verbally when it is a clitic pronoun. (4.1) a. Je vois l’homme dont je t’ai parlé I see the man of whom I to you have spoken I see the man I told you about vois b. Je le I him see I see him Alternatively, it might be something more unusual such as obligatory appearance in second position, as in the Ngiyambaa examples in (4.2). (4.2) a. Nunhi=lu=na Nadhi gave- 3erg- 3abs me- obl He gave it to me Na:nhi Nadhu b. Ninu: I- nom you- obl saw I saw you In Ngiyambaa (Donaldson 1980), bound pronouns are all located together in a ﬁxed sequence in second position within the clause, although the order of other words in the clause (including non-bound pronouns) is free. The placement of the pronouns in (4.2a) is thus the only one available, while their

76

Aspects of the Theory of Clitics

order relative to each other and to the verb in (4.2b) is only one of several possibilities. “Second position” is a notion that we will explore later in much more detail in this book. Signiﬁcantly, it may not be available at all to the syntax to the extent the relevant sense of “second” is based on phonological constituents (such as PWords) rather than syntactic phrases.

4.1 Phenomenology A useful initial step toward understanding the morphosyntactic dimension of clitic structure would be a survey of the phenomena, in the form of a catalog of the varieties of special distribution which clitics may show. After some initial clariﬁcation of terminology, I turn to such a survey, as a prelude to the question of where to locate the description of special clitics within a grammar. Special Clitics as a Linguistic Category Recalling the discussion in Chapter 2, there are some components which should probably be factored out of Zwicky’s deﬁnition. For one thing, to the extent special clitics are unaccented, this can be treated as following from the fact that they may be phonological clitics as well, independently of their distinctive distribution. I propose, then, that rather than treating simple clitics and special clitics as two distinct and non-overlapping categories, we regard the phonological and the morphosyntactic dimensions of clitic behavior as separable. In fact, most special clitics are also phonological clitics, but the two properties are logically quite distinct and neither necessarily implies the other. Furthermore, not all special clitics are unaccented, as we have seen. For instance, in Lucanian (as discussed in section 3.2), stress may appear on a pronominal special clitic because in this language stress is always penultimate within the PWord, and clitic pronominals are incorporated into this constituent—where they may receive stress. It is by no means obvious that all special clitics are also phonological clitics, though. Recall that the basis of phonological clitic behavior is prosodic deﬁciency (failure to constitute a PWord, in particular). Where a clitic consists of isolated segments, or a single light syllable (in a language whose minimal PWord consists of two syllables or two moras), such a diagnosis is obvious. In some instances, though, a special clitic may contain enough material to constitute a foot, and as such, may display stress in all or most of its occurrences. Examples of this type include Tagalog tayo ‘we (dual)’ (about which I will have more to say in Chapter 6) and Italian loro. Even in such a case, however, it is still difﬁcult to say with certainty that the element under investigation does or does not constitute a PWord. For instance, in Seediq (an Austronesian language of Taiwan; cf. Holmer 1996), stress is

Special Clitics and their Grammar

77

uniformly penultimate. Members of a set of clitic pronouns occur in second position within the clause, as illustrated by saku ‘2sg.sbj;1sg.obj’ in (4.3). (4.3) Wada =saku ini qtayi pret 2sg→1sg neg see You didn’t see me Some of these clitics, like saku, are bisyllabic; others include sami ‘1pl.excl.sbj’; simu ‘2pl.sbj’; misu ‘1sg.sbj; 2sg.obj’, etc. These bisyllabic units get stressed quite regularly. Furthermore, according to Holmer, the word preceding one of them undergoes ‘end of word’ phonology. This certainly suggests that the clitic constitutes a new word, but in fact it is consistent either with a two-word PPhrase (e.g., [ [ wada][ saku]]) or with a recursive PPh

PWd

PWd

PWord structure (e.g., [ [ wada]saku]). To distinguish between these PWd PWd possibilities (and thus to establish the prosodic status of the clitic), we need evidence from distinctive sandhi at the boundary between two PWords. An example of the appropriate sort is apparently furnished by Warlpiri. In this language, as is well known, auxiliary elements and pronominal clitics cluster together in second position within the clause. In addition to the auxiliary bases (ka and lpa as well as a hypothetical phonologically null base), complementizers can also appear in this position. Given the contrast between the generally free word order of Warlpiri sentences and the rigidity of ordering of clitics, auxiliaries and complementizers, all of these elements are undoubtedly to be analyzed as special clitics. While the auxiliary bases are monosyllabic, and thus prosodically deﬁcient with respect to the minimally bisyllabic Warlpiri PWord, the complementizers (with the exception of yi ‘for, since’) are bisyllabic, and thus at least candidates for independent PWord status. This status is supported by a fact noted by Legate (forthcoming): vowel harmony, a PWord-internal process assimilating the backness of high vowels in successive syllables, does not apply between the sentence-initial stem and a complementizer. An example is given in (4.4), where the vowels that are underlined should harmonize in backness if they were part of the same PWord. kuja=lpa=lu liwanja-paju-rnu, . . . (4.4) Nganayi whatchamacallit decl.comp-pst.imperf-3pl.sbj ﬁsh- call- pst That thing they called ﬁsh . . . Further analysis of the precise interaction between prosodic structure and vowel harmony in Warlpiri is still necessary before such examples can be securely interpreted as instances of prosodically autonomous special clitics. I will

78

Aspects of the Theory of Clitics

assume, though, that it is possible for a special clitic to constitute a PWord in its own right, and thus that special clitics are not necessarily phonological clitics as well. Another thing to be factored out is the requirement that special clitics be “variants of free forms.” Recall that Zwicky was led to posit a class of “bound words” which are just like special clitics, except that they are not variants of any free form. That suggests that a clitic may or may not alternate with a free form, but this is a lexical fact about particular elements rather than a deﬁning property of a linguistically signiﬁcant class of items. In French, for instance, we have (clitic or ‘conjunct’) me, te, etc. but nonclitic (or ‘disjunct’) moi, toi. The latter appear in ordinary DP positions; the former as clitics attached to the ﬁnite verb of the clause. On Zwicky’s deﬁnition, it is the existence of the disjunct pronouns that authorizes us to call the conjunct forms special clitics, since we can treat the latter as weak variants of the former. I suggest, though, that it is a mistake to treat corresponding members of these two sets as alternate forms of the same lexical item. It is true that they are generally in complementary distribution, but this is due to the general impossibility in French of clitics that “double” or have the same reference as a full DP expression.1 This is true not only for disjunct pronouns, but for any other DP as well, as illustrated in (4.5).2 (4.5)

a. J’ai vu toi qui sortait de la chambre ‘I saw you coming out of the room’ b. *Je t’ai vu toi c. *Je le vois le garçon d. *Je le lui donne le livre au garçon

The conjunct and disjunct forms, that is, are two separate lexical items that happen to have the same reference, rather than one single item. Their mutual exclusivity within a single sentence results from the broader regularity that, for reasons to be explored later in connection with clitic doubling more generally, French pronominal clitics are in complementary distribution with full DPs.

1 Some forms of colloquial French are much freer than the standard language in allowing clitic pronouns to double full DP expressions, including disjunct pronouns. This does not affect the point being made here. 2 If the full DPs in these examples are set off intonationally, as right-dislocated elements, they are perfectly compatible with a coreferring clitic, but this is not directly relevant to the impossibility of clitic doubling within the clause in (standard) French.

Special Clitics and their Grammar

79

A more complete account of this (which I will offer in Chapter 8) is based on principles of anaphora, not syntactic movement, but need not occupy us at this point. A Typology of Special Clitics What remains of Zwicky’s deﬁning properties of special clitics, then, is their characteristic special positioning. A special clitic is thus a “little word” whose syntax is not assimilable to that of full words that might seem to be syntactically parallel. To develop a theory of special clitics, we need to survey the range of such “special syntax” that we might ﬁnd. In work that has served as the classic foundation for all theories of special clitics since Zwicky’s original proposals, Klavans (1982, 1985) provides us with a descriptive typology. She ﬁrst notes that any given clitic can be seen as located with respect to some domain with which it is (syntactically and semantically) associated. In practice, this yields three general sorts of clitic. First, there are sentence clitics, located with respect to an entire clause. This includes auxiliary elements, many sorts of pronominals, particles marking discourse or illocutionary force, and many other sorts of clitics taking scope over the whole sentence. A second type is speciﬁcally associated with nominal expressions (NP/DP clitics such as case markers, determiners, or possessives as in some Balkan and Uralic languages). Finally, we have clitics which can be associated with phrases of any type, as markers of emphasis, constituent negation, interrogation, or other similar operators. The syntax of any particular special clitic is thus characterized in part by the domain within which it is located (CP/IP, DP/NP, XP). Relative to its domain, Klavans proposed two further parameters that determine where the clitic is located. The ﬁrst of these, which I will call Anchoring instead of Klavans’s term Dominance, characterizes a clitic as oriented with respect either to the ﬁrst or to the last element within the domain. The second positional parameter then speciﬁes whether the clitic is placed before or after the anchoring element. These parameters accommodate four kinds of clitics in terms of their position within a domain: initial and second position (before/after the ﬁrst element); pre-ﬁnal and ﬁnal position (before/after the last element). Three of these types are abundantly attested in natural languages: initial clitics, such as the Kw akw ’ala determiners we saw in Chapter 2, second-position clitics which will be discussed in much more detail below, and ﬁnal-position clitics such as the English possessive marker ’s. From considerations of symmetry within the system, we would also expect pre-ﬁnal or penultimate position clitics, but in fact it is hard to ﬁnd any real examples of this type. Klavans proposes (Kugu) Nganhcara (Smith and

80

Aspects of the Theory of Clitics

Johnson 2000) pronominals as a possible instance, but there are two problems with that analysis. First, in some sentences, the clitics in question are actually ﬁnal, rather than pre-ﬁnal. Secondly, in those sentences with pre-ﬁnal clitics, they always immediately precede the verb (which is itself in ﬁnal position). It is therefore difﬁcult to say that the positional regularity here is “before the last element of the clause” rather than “before the verb.” It is not unreasonable to ask, in fact, whether there are any genuine examples of speciﬁcally pre-ﬁnal clitics, a question which I will note again below. Are these in fact the only possibilities for clitic positioning? Sometimes it appears that rather than being anchored to the ﬁrst or last element within their domain, clitics are anchored by the head element within that domain. Romance pronominals, for example, are clearly anchored by the ﬁnite verb of their clause (if there is one; alternatively, by the inﬁnitive or other main verb in some types of non-ﬁnite clauses). The anchoring verb is clearly the head of its VP (and when bearing Tense, Agreement, etc., of IP as well). For a nonpronoun example, Finnish –kin ‘also’ (cf. Nevis 1986) is a sentential clitic that attaches to the right of the ﬁrst tensed verb of the clause (as head of IP). (4.6) Kalle on=kin ostanut auton Karl is- also bought car Karl also bought a car Kaisse (1985), Nevis (1986), Zwicky (1987), and Anderson (1992), among others, have thus suggested the availability of a third value for the Anchoring parameter, one which would allow for orientation with respect to the head of the relevant phrasal domain. There is another way to look at these clitics, however. Instead of seeing them as placed within the phrasal domain, and anchored by its head, we could regard them as placed within a smaller domain: that of the head constituent itself (V or perhaps I).3 On this picture, the Domain parameter can be set to any category that is either a maximal phrasal projection or a lexical category which projects such a phrase. There would then be no need to allow a third value for the Anchoring parameter. In some languages, multiple possibilities for the positioning of a given clitic may be instantiated. In most of the Romance languages, for example, pronominal clitics generally appear initially within the domain of the ﬁnite verb, but under some circumstances (with non-ﬁnite verbs, imperatives, or 3 Note that the syntactic domain V as head of VP is not the same as the lexical item (a verb) which appears within that domain. Saying that a clitic appears initial or ﬁnal within the domain of V is thus not to be confused with treating it as an inﬂectional preﬁx or sufﬁx on the verb. This difference will be signiﬁcant in some instances to be discussed later.

Special Clitics and their Grammar

81

in Portuguese, sometimes with ﬁnite verbs) they may instead appear ﬁnally within this verbal domain. Similar facts obtain in some South Slavic languages. To the extent the conditions for initial versus ﬁnal appearance can be formulated exactly, this complication does not impugn the overall typology of clitic positioning, however. With respect to second-position clitics, an important point that will occupy us at some length below is the fact that “second position” can have more than one meaning. It often means “following the ﬁrst syntactic daughter of the constituent deﬁning the domain,” but in some languages it may instead mean “following the ﬁrst PWord, PPhrase, etc. within the phonological realization of the domain.” The best-known example of this type is the clitic system of some forms of Serbo-Croatian, as described by Browne (1974), which will be the subject of closer attention in Chapter 5. This possibility corresponds to the fact that (at least in descriptive terms) the anchoring element may be interpreted either as a syntactic or as a prosodic object. All of the possible placements for clitics considered so far assume that clitics are placed at the periphery of some anchoring element. Some examples have been proposed, however, in which so called “endoclitics” (in the terminology of Zwicky 1977) appear internal to a word. Putative examples include Pashto (Tegey 1977; van der Leeuw 1995a; Roberts 1997) á-de-xist ‘you were buying it’; Udi (Harris 2002) a-ne-q’-sa ‘she takes’ (root /aq’/); and Portuguese dar-no-lo-á ‘he will give it to us’. These cases will be examined more closely in Chapter 6, and I will ignore the issues they might pose for the moment. A ﬁnal parameter of individual clitics within Klavans’s theory is the direction of their phonological attachment (proclitic or enclitic). One might expect a clitic to take as its host the element with respect to which it is anchored, or perhaps the element with which it is most closely associated syntactically, but in fact clitics may attach phonologically to an element which does not even form a part of the syntactic domain in which the clitic is positioned. We have already seen this with respect to Kw akw ’ala determiners and pronouns (a case also cited by Klavans). A closely similar example is provided by Yagua (Payne and Payne 1990), as in (4.7). In this language, Object DPs are immediately preceded by a coreferring clitic pronominal which attaches to the word that precedes it, regardless of the syntactic relation (if any) which that word may bear to the object. e

(4.7) sa- púúchiy Pauro rooriy- v11mu=níí Anita 3sg.sbj- lead/carry Paul house- inside- 3sg.obj Anita Paul leads/carries Anita inside a/the house Another example is provided by Northern Vogul, as in (4.8) from Nevis (1990).

82

Aspects of the Theory of Clitics

´ Xani (4.8) Xum jot=ke åleGe˙ m naurem man with- if I live child clings If I live with a man, the child belongs to me Here the clitic ke ‘if ’ is located immediately before the verb, but attaches to its left. Similarly, in Kugu Nganhcara (Smith and Johnson 2000), pronominal clitics precede the verb but attach to the preceding word, as in (4.9). (4.9) ngaya ku’an hingkurum ka’im=ngkurum kala- ng 1sg.nom dog 2sg.abl neg- 2sg.abl take- 1sg I didn’t take your dog These and a number of other examples of the same sort are discussed and analyzed by Cysouw (forthcoming). According to Klavans, they motivate an additional parameter of individual clitics, their phonological liaison. In section 3.3 above, however, I argued that the direction of phonological attachment is not a lexical property of individual clitics, but rather follows from the way(s) in which the principles of Stray Adjunction operate in a language with respect to its prosodic structure. I will therefore assume that we can dispense with any such parameter as an item-speciﬁc property within the typology of special clitics. In summary, an exhaustive typology of special clitic positions is given by the parameters in (4.10). (4.10) A clitic is located a. within the Domain of some syntactic constituent (X0 or Xmax for some value of X); b. by reference to the first versus last daughter constituent of that domain (interpreted either syntactically or prosodically); and c. preceding or following this anchor point.

4.2 Special Clitics as the Morphology of Phrases The typology of special clitics that I have just presented seems to be an adequate taxonomy of what we ﬁnd in the languages of the world, but it does not go very far beyond simple observation. That is, it does not really tell us how special clitics are to be accounted for in an explicit grammar. It does not even tell us directly where to look for such an account: in the syntax (as syntacticians assume virtually without discussion), in the phonology (where we have already

Special Clitics and their Grammar

83

found all that is apparently necessary to describe simple clitics), or perhaps somewhere else. Certainly the most popular picture is the syntactic one, according to which special clitics are introduced into sentences in the same way as other (lexical) items, and then displaced from where we might expect to ﬁnd syntactically corresponding words to their surface position. On that view, the parameters of the preceding section become parameters that characterize a set of syntactic rules. A priori, though, that is not a very attractive position. For one thing, as I have already noted and will discuss below in more detail, second-position clitics may come literally after the ﬁrst word in some languages; and this (phonological) notion is not in general available to the syntax (which deals in phrasal constituents). For another, this kind of “syntax” is quite incompatible with the way syntactic movement is otherwise characterized. The notion of a set of special syntactic rules of the sort implied by this typology is clearly a relic of the more general conception of syntax as grounded in a collection of construction-speciﬁc rules (“Passive,” “Dative Movement,” “Subject-to-Object Raising,” etc.). Most modern theories of syntax4 deplore such rules and attempt to derive the properties of individual constructions from more general properties of structure and displacement operations. This might, of course, turn out to be impossible, but our ﬁrst impression surely ought to be that if a syntactic theory of special clitics must take this form, we should consider the possibility that some other part of the grammar might be better suited to the purpose. The alternative which I will pursue in this book is the proposal that clitics are not syntactic objects at all, but morphological ones: overt morphological markers of the morphosyntactic properties of phrases. Clitics on this view are phrasal afﬁxes. Later chapters will be devoted to ﬁlling out and justifying this view, but to establish its a priori attractiveness, let us compare the typology of special clitics developed above with the typology of morphological markers realizing the morphosyntactic properties of words: primarily (but not exclusively) the typology of afﬁxes. Most afﬁxes are either preﬁxes or sufﬁxes of course, but there are also inﬁxes. Examination of these (Moravcsik 1977; Anderson 1992) reveals that we 4 There are exceptions, of course, of which the most obvious is the framework of “Construction Grammar” (Goldberg 1995). Various other syntactic theories also admit at least a limited amount of parochial, construction-speciﬁc structure in syntax, without going so far as to suggest that such formations constitute the core of grammatical knowledge. I take it for granted that anything that can be reduced to general principles ought to be, and that rules which directly stipulate the properties of a construction are an admission of failure in this endeavor—perhaps necessary, but surely not desirable.

84

Aspects of the Theory of Clitics

ﬁnd morphological markers inﬁxed not just anywhere in a word, but only in a restricted range of positions. Summarizing a number of different surveys, we can say that inﬁxes either follow the ﬁrst element of some prosodic type (segment, syllable, etc.) within the word whose properties they mark, or else they precede the last element of such a type in the word. This of course looks rather like what we saw with special clitics, and essentially the same set of parameters can characterize both types of element. Preﬁxes go before the ﬁrst element, sufﬁxes after the last, and the two sorts of inﬁx are located after the ﬁrst or before the last. Among afﬁxes, there is an analog to the issue of the domain within which clitics are placed. Recall that some special clitics are located within the domain of the head of a phrase (e.g., Romance conjunct pronominals). The same can also be true for afﬁxes. It is suggested in Anderson (1992) that words sometimes have internal structure, as with Icelandic [[kalla]st] ‘to be called s.t.’ or Georgian [mo[k’lav]] ‘(will) kill’. In such a case the afﬁx may be located as a preﬁx (or sufﬁx) to the internal head of the word, as in [[köll- um]st] ‘we are called s.t.’ or [mo[v- k’lav]] ‘I will kill (him/her/them)’. Pursuing the parallel between special clitics and word-level morphology, we can note that among the various types of inﬁx, those located in pre-ﬁnal position are comparatively quite rare. In this they are rather like penultimate clitics, which may exist, but as a distinctly unusual type. This similarity is of course only a tendency (assuming penultimate clitics exist at all), but it is interesting to note that it takes the same form in both areas of grammar. A striking similarity between the two classes of phenomena is in the matter of ordering. Within a word, the order of morphological markers is virtually always strictly determined and invariable, holding meaning constant. Of course, the passive of a causative will typically differ in the order of markers from the causative of a passive: this is the basis of the “Mirror Principle” proposed by Baker (1985). But for given content, no known language allows the order of afﬁxes to vary freely. In contrast, practically all languages allow for some freedom in the ordering of constituents or even individual words while preserving semantic content. When we look at clitics, we ﬁnd that their order is largely ﬁxed, like that of morphological afﬁxes, and essentially never free, like that of syntactic units. The contrast can be stark within an individual language. In Warlpiri, for example, word order is extremely free, but the one constant is the fact that a set of special clitics (including auxiliary elements and pronominals) must come in second position, and must conform to a strict internal template. In this respect, special clitics clearly behave much more like morphology than like syntax.

Special Clitics and their Grammar

85

There are cases in which the same clitic may appear in different positions with respect to its anchor without affecting the semantics. A well-known case of this sort is furnished by (European) Portuguese, as illustrated in (4.11) from Barbosa (1996). (4.11)

a. Só o Pedro o viu only art Pedro him= saw Only Pedro saw him b. *Só o Pedro viu-o c. Viu- o só o Pedro saw=him only art Pedro Only Pedro saw him d. *O viu só o Pedro

Other alternations in order have already been mentioned above in other Romance and Balkan languages, where clitics often precede the ﬁnite verb but follow non-ﬁnite and imperative forms. The alternation in position which we ﬁnd in these cases is nothing like free word order or other “Scrambling” phenomena, however. Rather, the order which is strictly required under one set of circumstances is replaced by another under a complementary set of conditions. In the Portuguese case, pronominals appear in enclitic position when the ﬁnite verb is initial within its clause (or within a constituent of some other speciﬁc type: see Barbosa 1996; Galves and Sandalo 2004; and section 6.2 below for discussion). Otherwise, the pronominals appear proclitic to the ﬁnite verb. No optionality is involved, as the ungrammatical forms in (4.11) show. Parallel to these cases are examples in word-level morphology where the same afﬁx may show up either as a preﬁx or as a sufﬁx, depending on speciﬁc factors. The most widely cited example of this is from Afar (Fulmer 1990), where several afﬁxes show up as sufﬁxes when the verb stem begins with a consonant or the low vowel [A], but as preﬁxes when the verb stem begins with a non-low vowel. Other movable afﬁxes are described by Nevis and Joseph (1992), who note several cases in a footnote, and by Noyer (1994). What is notable is that in each case, the position of the afﬁx is not at all free. Rather, it depends on factors such as the shape of the stem to which it attaches, or on some other morphological category which is copresent. For instance, in Huave as described by Noyer (1994), a theme vowel appears as a sufﬁx to the stem in detransitivized forms but otherwise as a preﬁx. Both clitics and word-internal afﬁxes only display alternations in order under clearly speciﬁable conditions,

86

Aspects of the Theory of Clitics

as opposed to the much greater freedom of word (or constituent) order found in many languages. I have suggested that post-initial inﬁxes are analogous to second-position clitics, both occurring after the ﬁrst element of some speciﬁed type within the domain whose properties they specify. In general, the anchoring element for word-internal inﬁxes is prosodically deﬁned: a consonant, syllable nucleus, syllable, foot, etc. In at least one case, however the parallel with clitics is even more striking. Nevis and Joseph (1992) argue that a reﬂexive/passive afﬁx in Balto-Slavic (Lithuanian –s(i)) occurs precisely in a word-internal second position, as illustrated in (4.12). (4.12) No preﬁx: m¯atymasi-s ‘seeing each other’; kalbúo-si ‘I converse’ One preﬁx: pa-si-m¯atymas ‘a date, meeting’; be-si-kalbant ‘while conversing’ Two preﬁxes: su-si-pa-žìnti ‘to become acquainted with’; pri-sipa-žìnti ‘to confess, avow’ Nevis and Joseph call –s(i) a “Wackernagel afﬁx,” and establish that it is part of the same PWord with the stem and other afﬁxes. I suggest that this is indeed a case of “sufﬁx to the ﬁrst element in the word.” Assuming that the preﬁxes after which –s(i) appears are structure-building, like the Georgian and Icelandic afﬁxes discussed above, the element that counts as “ﬁrst” is a structural one. This account is consistent with dialectal variation which they mention in whether the reﬂexive marker, when it comes after the unpreﬁxed stem, comes before or after inﬂectional sufﬁxes. The similarity between this afﬁx and second-position clitics is thus unusually striking. Just as not all morphological markers are concatenative afﬁxes, there are also clitics that are processual (or “non-concatenative”). I will discuss below the analysis of the “deﬁnitive accent” in some Polynesian languages. In Anderson 1992, I also cited in this connection the marking of (some) accusative objects in Welsh by soft mutation (a non-concatenative morphological process) at the left edge of the nominal phrase. Another interesting example is furnished by some languages of the Algonquian family. Algonquian languages such as Potawatomi display person-marking preﬁxes at the left edge of the verbal phrase. Though these have often been regarded as inﬂection markers on the verb (and were so treated in Anderson 1992), Hockett (1948) showed that they should actually be analyzed as phraseinitial clitics, a view also adopted by Halle and Marantz (1993). In their positioning, the Potawatomi personal pronoun clitics are interestingly similar to another element in Algonquian morphosyntax which appears in several

Special Clitics and their Grammar

87

languages of the family: a replacive vowel change affecting the leftmost vowel in the same verbal projection as that to which the personal proclitics are attached. This “initial change” (to adopt the terminology of Bloomﬁeld 1946 and elsewhere) functions as a sort of complementizer in marking certain subordinate constructions, as in (4.13). Vowels affected by initial change (or related to others that are) are underlined. (4.13)

Potawatomi: a. Pe ki mpot prt past died He died mpot b. ka past (< ki) died the one who died Menomini:

a. ne=pa:pam- nato:nE:hok me=going about- he seeks me He goes about seeking me b. (emeq) pE:pam- esiat (yonder) going about- he goes It is over yonder that he goes about

Fox:

kano:n- ehka a. e:shki ﬁrst (
Non-segmental apophonic relationships such as the initial change illustrated in (4.13) can serve functions which might in other languages be ﬁlled by a phrasal clitic: here, that of a complementizer. If we think of clitics as the phrase-level analog of word-level morphology, it is natural to expect such phenomena, on a par with comparable markers in the form of words. Interestingly, both sorts of clitic here—the initial change and the person markers—are located at the left edge of the same projection of V (VP or V). Reduplication is a type of formal relationship between forms that occurs in many languages as a marker of morphological categories at the word level. An example where this same formal device marks a category at the phrase level is furnished by Chamorro (Chung 2003). In this language, a predicate phrase XP of any one of several types is marked for Progressive aspect by reduplication

88

Aspects of the Theory of Clitics

of the CV portion of the primary stressed syllable5 in the leftmost PWord within the phrase. This can affect a word of any category, and is not conﬁned to the head of the phrase. Examples in which this reduplication affects words of various types are cited in (4.14) with the reduplicant underlined. (4.14)

a. Ni hagu mismu ti un- li’i’ amanu un- hanánagui even you self not agr- see where? agr- go.to.prog Even you yourself do not see where you are going b. Ti atrasásao not agr.late.prog He’s never late c. Á’aguaguat =ha’ na patgun =gui’ esta pa’gu naughty.prog emph linker child he until now He’s still a naughty child Juan si Mr Perez? d. Ma’estrótro- nña =ha’ si teacher.prog- agr emph art Juan art Mr Perex Is Mr Perez still Juan’s teacher? si nana- ña e. Ginin= i= gima’- yúyu’us from the house- God.prog art mother- agr Her mother was coming from the church. f. Falak6 = i= chächagu’ na guma’ agr.go.to.imper the far.prog linker house (Keep on) going to the farthest house!

Whether reduplication constitutes afﬁxal or non-concatenative morphology at the word level is a representational issue which I will not address here. Regardless of one’s views on that matter, the Chamorro example provides us with a further analog at the phrase level to a morphological type which is well established in the analysis of words. It looks, then, as if there are a number of substantive parallels between the morphological markers we ﬁnd in the structure of words, on the one hand, and special clitics within phrases, on the other. Two further points tending in the 5

That is, the stressed vowel and either (a) the entire onset or else (b) the last consonant of the onset, but not the coda: e.g., tristi ‘sad’ gives (for some speakers) triristi ‘(still) sad’ and (for other speakers) tritristi; planta ‘set (the table)’ gives plalanta ‘set (prog)’ for some speakers and plaplanta for other speakers; kanta ‘sing’ gives kakanta ‘sing (prog)’ but not *kankanta; maigu’ ‘sleep’ (orthographic i here is the glide [j]) gives mamaigu’ ‘sleep (prog)’ but not *maimaigu’ (Sandra Chung, personal communication). 6 This verb is one of a small class of verbs in Chamorro that are prosodically deﬁcient, and thus appear as proclitics attached to the following PWord.

Special Clitics and their Grammar

89

same direction are also worth noting. First, both special clitics and afﬁxes are generally deﬁcient in prosodic structure, though in both cases there are a few instances with structure at least up to the level of the foot. In this, both distinguish themselves from ordinary lexical items, which typically project a PWord (in most languages). Second, it seems useful to recognize a distinction between ‘inﬂectional’ (grammatical, functional) markers and ‘derivational’ ones both for clitics and for word-level morphology. Somewhat remarkably, this distinction (which in both cases can be grounded in the architecture of grammar) correlates with a difference in ordering, with inﬂectional elements coming “outside of ” derivational ones. I will return to this matter in Chapter 6. In contrast to these similarities between special clitics and morphology, clitics distinguish themselves from units manipulated by the syntax in terms of their possible positions of occurrence. It was precisely the unusual placement of these elements, indeed, that led to the positing of a class of special clitics in the ﬁrst place. These factors suggest the plausibility of treating special clitics as something other than syntactic objects manipulated by syntactic rule, and in particular, as a kind of morphological object (as suggested by the expression “phrasal afﬁx”). The account to be developed in later chapters will generalize this result by treating special clitics not as lexical items inserted and moved around within the syntax, but rather as phonological material inserted (like afﬁxes) into the phonological content of a phrase. Special clitics express the morphosyntactic properties of a phrase (the content of its functional categories, more or less), or else indicate (in the “derivational” case) modiﬁcations to the semantics, discourse properties, and the like, of the phrase. It remains, of course, to construct an explicit theory on these lines, the task of Chapters 5 and 6.

4.3 Some Examples Before I return to the theory of special clitics, though, I explore some concrete examples of special clitics in a bit more detail. I will consider three cases: the ’s that marks nominal possession in English, a kind of “poster child” of special clitics; a potential example of a non-afﬁxal, or processual determiner element in some languages of Polynesia; and ﬁnally, the special clitic system of Kw akw ’ala, expanding on the description of simple clitics in that language in Chapter 2. Case I: The English Possessive Virtually every discussion of clitic phenomena involves at least a passing reference to the possessive marker ’s in English. Some scholars simply take it as selfevident that ’s is attached as a clitic to the right edge of a nominal expression,

90

Aspects of the Theory of Clitics

and it is easy to see how immediately appealing this analysis is. As opposed to languages that mark possession by inﬂecting the head of the nominal in an appropriate way (usually called “genitive case”), the possessive marker in English attaches freely to heads and non-heads alike, provided only that their host be located at the right edge of the phrase as in (4.15). (4.15)

a. Fred’s taste in wallpaper is appalling. b. The man in the hall’s taste in wallpaper is appalling. c. Every man I know’s taste in wallpaper is appalling. d. That brother-in-law of mine that I was telling you about’s taste in wallpaper is appalling. e. Even that attractive young man who is trying to ﬂirt with you’s taste in wallpaper is appalling.

In its fondness for this “group genitive,” modern English is somewhat unusual among the languages of the world. Assuming that the right account of clitics is one that treats them as phrasal morphology, it seems straightforward to describe the English possessive in terms of a feature—let us call it [Poss]—which is assigned to DPs. The distribution of this feature is also straightforward: it is assigned (as what Anderson 1985a describes as a “conﬁgurational” property) to a DP that appears as the determiner of another DP. We can then say that a rule of phrasal morphology introduces a phonological marker /z/ at the right edge of a string of phonological material whose syntactic analysis is [DP,Poss]. The resulting string undergoes appropriate phonology, so that /z/ appears phonetically as [s], [z], or [1z] depending on the shape of its host, in a way parallel to other elements in English (the regular plural, the 3sg pres form of verbs, and the reduced auxiliaries) with the same shape. I am not concerned at this point with the mechanics of the process by which the marker is introduced into the representation, though it is worth highlighting the assumption that phrasal categories (like DP) have an internal feature structure that can include elements like [Poss] which are not simply markers of the type and bar level of the phrase. If this were all there were to the English genitive, we could pass quickly on to other matters, but a fairly abundant literature (including, but not at all limited to, Allen 1997; Carstairs 1987; Stemberger 1981; Stump 2001; Zwicky 1987) attests to the fact that there may indeed be more to say. In fact, some DPs with the feature [Poss] are not realized simply as the basic phrase with an appended /z/. That is the case, notably, for DPs consisting simply of a pronoun: we say my lunch, your lunch, his lunch, etc., rather than *me’s, *you’s, *him’s, etc., lunch. Pronouns are in fact the only forms in English

Special Clitics and their Grammar

91

that have a distinct possessive form, and since this is unpredictable (suppletive), it is presumably listed in their lexical entries. Let us assume that pronouns originate not as the heads of NPs, but rather as determiners without NP complements, heading DPs—an analysis originating in Postal (1966) which has returned to vogue in recent years. Then the exceptional possessives are exactly those DPs with the feature [Poss] which consist of nothing but a determiner. Assume further that a principle of English grammar requires the feature [Poss] to be inherited by D from the DP to which it is originally assigned precisely when that DP consists of nothing but the D (perhaps D and DP collapse as the same category in this case). The feature complex [D,Poss] can be interpreted by an item from the lexicon when this is a pronoun—and only then—and I assume that some principle of disjunction blocks the Possessive clitic from being introduced phrasally when the feature [Poss] is already realized by a lexical form. Observe that other DPs consisting only of a (non-pronoun) determiner do not generally take the possessive /z/ either: *These’s illustrations are more competently drawn than those’s, *Of the books I lent you, two’s/some’s/many’s covers were soiled when you brought them back, etc.7 This suggests that a DP consisting only of a D is not in fact a determiner phrase at all, but only a word, as some work within the Minimalist Program might suggest. As a result, the rule introducing the possessive clitic does not apply to it, and the only way the feature [Poss] can be realized in this case is if a lexically listed form exists to serve that function. Non-pronoun determiners have defective paradigms in the sense that they have no listed possessive forms. Note also that the possessive pronouns have special forms when they serve as determiners of a DP containing a null NP complement: I want mine, yours, his, etc. These are presumably lexically listed and associated with the speciﬁc environment [ D [ ∅]]. DP

NP

The fact that [Poss] is realized sometimes by the form of an individual word (for pronouns) might be seen as evidence against the view that it is a phrasal afﬁx, but in my view this is actually interesting evidence in favor of this position. It is signiﬁcant that the same feature shows up sometimes as a determinant of the morphology of a word and otherwise as a marker clearly appended to the right edge of a phrase. This argues that although a given property may be realized in more than one way, that realization is still a matter of morphology regardless of whether it takes place at the word or phrase level.

7 But see . . . one’s cover was soiled. I have no explanation for this, apart from an appeal to the notion that since one also serves as a pronoun, it might have a listed genitive like other pronouns.

92

Aspects of the Theory of Clitics

For those who would prefer to avoid an analysis involving phrasal afﬁxes, the alternative to such an account of the English possessive has been to treat it as an inﬂectional element that is conﬁned to an element at the right edge of a DP. On this view, ’s is a word-level afﬁx, capable of occurring on any word of the language (or at least any word that can end a DP), realizing an “edge feature.” Aside from the conceptual difﬁculties posed by that notion, note that ’s can in fact be realized on a DP-ﬁnal pronoun as in (4.16). (4.16)

a. The person who told me’s name must remain a secret. b. I don’t know how appetizing your dinner is, but some of the person who cooked mine’s hairs are ﬂoating in my soup. c. Fred’s mother’s walk-in closet alone is bigger than mine’s entire apartment.

If [Poss] were regarded as a feature assigned to the word that occurs at the right edge of a DP, further apparatus would be needed to ensure the correct forms in these cases. There is a further set of environments in which the possessive ’s is not phonologically realized. When ’s is added to a noun which is already marked for plurality with /z/, only one /z/ appears, and not two: the three boys’ [bojz] caps, not the three boys’s [bojz1z] caps. As has been widely noted, this cannot be due to a general loss of possessive marking with plurals, since plural formations other than /z/ are perfectly compatible with the marker of the possessive: the three children’s/women’s/deer’s/mice’s etc., feet. Beyond the combination of the /z/ plural and the possessive, there is some disagreement about the exact facts here. Zwicky (1987) says that /z/ is lost after any word that ends in an inﬂectional /z/. For him, this is true not only when the possessive is added directly to a regular plural (the two boys’ books) but also when it appears sufﬁxed to a word that happens to end in any other sufﬁxal /z/, as in (4.17). (4.17)

a. anyone who likes kids’ (/*kids’s) ideas b. people attacked by cats’ (/*cats’s) reactions to them c. anyone who hurries’ (/*hurries’s) ideas d. everyone at Harry’s (/*Harry’s’s) ideas e. a friend of my two kids’ (/*kids’s/*kids’s’s) ideas

For Zwicky, then, any sequence of two (or more) /z/ afﬁxes is reduced to a single /z/. An instance of /z/ is not, however, lost following a stem that ends phonetically in the same way as one of the allomorphs of /z/, as shown in (4.18).

Special Clitics and their Grammar (4.18)

93

a. the fuzz’s old cars; at Buzz’s b. the bus’s doors; at Cass’s c. the terrace’s tiling; at Thomas’s

These facts show that the reductions in (4.17) cannot be due to simple phonetic degemination. In fact, however, there is a fairly direct way to formulate the necessary simpliﬁcation of sequences of /z/s. Recall the proposal I made in section 3.4 of the preceding chapter, in the context of the reduced auxiliary ’s, for the phonology of afﬁxed /z/ (and /d/). I suggested that when /z/ is added to a form, it is introduced not as a daughter of the syllable to which it is attached, but rather as an adjoined syllabic afﬁx. The proposed structure of, e.g., dogs is thus as in (4.19).

σ

(4.19)

σ dOg z As a result, sufﬁxal /z/ is structurally distinct from stem-ﬁnal /z/, and there is no difﬁculty in formulating a regularity that reduces two identical syllabic afﬁxes to a single one. This is still not the end of the story, however. Zwicky (1987: 140, n. 6) notes that “POSS is occasionally suppressed in speech (as it regularly is in writing, according to at least some style sheets) after proper names ending in /s z/: Jones’, Nevis’, Jeeves’.” This can be accommodated if we assume that the ﬁnal segment of these names is treated by some speakers (or writers) as a syllabic afﬁx, rather than as part of the stem. Those who say (or write) Jeeves’ rather than Jeeves’s, that is, represent the name lexically as [ [ ˇjijv]z]. As a result, the addition σ σ of a further afﬁxal /z/ invokes the reduction process just discussed. Speakers who say Jeeves’s, on the other hand, have a lexical representation consisting of a single unafﬁxed syllable [ ˇjijvz]. σ Carstairs-McCarthy (1995) rejects Zwicky’s claim that the multiple afﬁxations of /z/ in (4.17) are ungrammatical. He ﬁnds them at most questionable, and in some instances completely acceptable. This does not mean, however, that he does not have a process reducing multiple instances of /z/: *the two boys’s caps is just as bad for him as it is for Zwicky (and virtually all other speakers of English). It is only when /z/ is added directly to a regular plural noun that afﬁx reduction must take place for Carstairs-McCarthy: elsewhere, it is at most optional.

94

Aspects of the Theory of Clitics

A proposal for describing the grammar of a speaker for whom the multiply afﬁxed forms in (4.17) are acceptable might go as follows. We must recognize that while the /z/ afﬁxes are added in a structurally adjoined position phonologically, they must eventually be incorporated into simple syllables by the time they are produced phonetically. Assume, therefore, a post-lexical process of Syllable-afﬁx Incorporation. If this process is allowed to apply cyclically, it might already have applied on an earlier cycle when another afﬁx is introduced later, with the result that the earlier afﬁx no longer occupies an adjoined position, and the principle reducing multiple afﬁxes is no longer applicable. In Zwicky’s dialect, Syllable-afﬁx Incorporation always applies post-cyclically, at the last stage of phonetic interpretation. As a result, syllable afﬁxes are always identiﬁable as such for him, and always undergo reduction when more than one is present on the same word form. For Carstairs-McCarthy (and many other speakers), however, the facts are more complex. A possessive sufﬁx added directly to a plural noun will always undergo reduction, because there is no intervening cycle on which the plural could have undergone Syllable-afﬁx Incorporation so as to lose its identity as an afﬁx. When the possessive is added at the end of a longer DP, however, so that it no longer appears on the head of the NP, intervening cycles may have incorporated the afﬁx of an included plural noun so that the reduction of multiple afﬁxes is no longer applicable. There is obviously some work to be done to ﬂesh out this suggestion fully, but it seems to show some promise for dealing with the range of variation we ﬁnd. I conclude, therefore, that the possessive ’s can be treated as a phrasal afﬁx, introduced at the right edge of DPs bearing a feature [Poss]. The complicating factors which have been cited in the literature can in general be treated as the result of (a) the transfer of the [Poss] feature to a pronoun D head when this is the only content of the DP; and (b) phonological processes reducing multiple syllabic afﬁxes of the same phonological shape to a single instance.8 Case II: Some Polynesian Deﬁniteness Markers In Anderson (1992), I suggested that the ‘deﬁnitive accent’ in Tongan constitutes an example of a non-afﬁxal clitic: a process which serves as a phrasal marker comparable to other clitics. Closer examination shows the facts to be more complex than I thought, but a case still seems to exist for treating this 8 I still have no account for the observation (Zwicky 1987: 140, n. 6) that “[f]or many speakers, PL+POSS is unacceptable if POSS is not located on the head of its NP: *our fathers-in-law’s, *the queens of England’s, *the men I mentioned’s. I agree to some extent with this judgment, attributed by Zwicky to Kruisinga (1932) and partially invoked by Carstairs-McCarthy (1995) as well, but I cannot see how to make it follow from anything else in the present analysis.

Special Clitics and their Grammar

95

(or perhaps a cognate of it found in another language, Rotuman) as a processual special clitic. The deﬁnitive accent is described carefully by Churchward (1953); its semantics and usage are analyzed by Aitchison (2001) within the framework of HPSG. Other relevant literature will be noted below. In Tongan, the normal location of stress is on (the syllable containing) the penultimate mora. Deﬁniteness9 is marked by a stress shift to the ﬁnal mora of the entire DP. The examples in (4.20) show that the stress shift occurs only on the last word of the DP. (4.20)

a. kuo maumau ’a [e s¯alioté ] perf broke abs art cart- def The cart is broken b. kuo maumau ’a [e s¯aliote ’a Feletí ] perf broke abs art cart gen Fred- def Fred’s cart is broken c. kuo maumau ’a [e s¯aliote ’a Feleti mo Sioné ] perf broke abs art cart gen Fred and John- def Fred and John’s cart is broken d. te u ’alu ki [he fale kuo nau fakataha aí ] fut I go to art house past they assemble in- it- def I will go to the house in which they have assembled e. na’a ke ’alu ki [he fakataha lahí ] aneaﬁ past you go to art meeting big- def yesterday Did you go to the big meeting yesterday?

The phrasal nature of this process is quite clear: it occurs at the right edge of a DP without regard to the syntactic role of its “host.” Furthermore, the deﬁnitive accent is clearly analogous to other elements in Tongan that are uncontroversially clitics: the demonstratives in the ﬁrst column of Figure 4.1. The demonstrative clitics ni ‘this’ and na ‘that’ are also introduced at the right edges of DPs which they determine. Their (phonological) clitic status is shown by the fact that (a) being monomoraic, they do not meet the language’s minimal PWord constraint, and thus must be stray syllables; and (b) they are evidently incorporated into the preceding PWord as Internal clitics, since they 9 This dimension is distinct from that marked by articles appearing initial within the Tongan DP. Condax (1989: 426) formulates Churchward’s description as follows: “Thus e/he is ‘speciﬁc’ or ‘referential’ in meaning, and within that range the deﬁnitive accent marks noun phrases as deﬁnite, while normal noun phrases are indeﬁnite.”

96

Aspects of the Theory of Clitics Clitic —

ni this na that

Pronoun ¯e this that eni this ena that

Locative h¯e here there heni here hena there

Adverb peh¯e like this like that peheni like this pehena like that

Figure 4.1. Tongan demonstratives

cause the accent on that word to shift one mora to the right (from penult to ultima): he falé ni ‘this house’, not *he fále ni. Semantically as well as formally, it is tempting to see the deﬁnitive accent as ﬁlling the gap in the top row of the leftmost column in Figure 4.1. Indeed, Clark (1974) argues precisely that the deﬁnitive accent represents the reﬂex of an original demonstrative *a that would ﬁll that gap. Apparently, this element came to be assimilated to the quality of the preceding (word-ﬁnal) vowel, and the resulting long vowel was then shortened—after having attracted the stress. The reﬂex of *a in Modern Tongan is thus a DP-ﬁnal clitic consisting of a processual shift of the accent from penultimate to ﬁnal mora. These facts would not constitute an instance of a non-afﬁxal special clitic if it were possible to maintain something like the “edge inﬂection” analysis examined (and rejected) above for the English possessive marker. And in fact such an account is explored in the context of Lexical Phonology by Poser (1985), who argues for that conclusion on the basis of a claim that the change involved is deeply embedded in the phonology of the language. The deﬁnitive accent involves a shift of stress which treats underlying long vowels as sequences of two identical moras. Poser assumes that a lexical rule reduces such sequences to a single long vowel at some point. This rule is argued to be lexical since it requires access to properties of speciﬁc lexical items, and not simply to phonological form. If the deﬁnitive accent rule precedes this lexical rule, it must itself be lexical. There are several problems with this line of reasoning, though. First, there is no reason to assume a rule collapsing sequences of identical (short) vowels into a unitary long vowel, if long vowels are represented simply as two segmental positions linked to the same vowel melody, the standard assumption today (though not as generally at the time Poser wrote). The putative lexical rule of vowel coalescence, then, is no rule at all: it is simply the phonetic interpretation assigned to a sequence of vowel moras linked to the same melody when these are syllabiﬁed together.

Special Clitics and their Grammar

97

Some of Poser’s arguments for the lexical nature of processes necessarily following the deﬁnitive accent shift are dealt with in detail in Anderson (1992: 212–15), and need not be repeated. One that is not dealt with there is his observation that sequences of identical vowel segments do not coalesce—that is, in the terms of the present discussion, are not syllabiﬁed together—when they are formed in certain ways. Speciﬁcally, Poser notes that vowel sequences formed by reduplication and by the preﬁxation of faka- do not undergo coalescence, as illustrated in (4.21). (4.21)

a. b. c. d.

ongoongo ‘news’; ongoongoa ‘famous’ piko ‘crooked’; angaangapiko ‘somewhat crooked’ ava ‘open’; fakaava ‘to open’; fakaava’i ‘to open completely’ ako ‘school’; fakaako ‘education’

The explanation for this fact is not far to seek, however. Churchward (1953: 5–6) makes it clear that in addition to the standard word stress in penultimate position,10 extra stresses appear in several classes of words. Among these are words with preﬁxes of two or more moras (such as faka- ), compounds, and fully reduplicated words. I assume that these extra stresses represent the fact that the components of such words are independently footed, with a bimoraic trochee formed at the right edge of each part. When, e.g., fakaava ‘to open’ is formed, then, the ﬁnal vowel of the preﬁx faka- constitutes the weak mora in such a foot, and the stem initial /a/ which follows is the strong mora in another foot. The requirements of the prosodic hierarchy prevent moras belonging to distinct feet from forming a single syllable, and these two as are thus not interpreted as a unitary long vowel. Even when main stress on the whole word shifts one mora to the right in fakaava’i ‘to open completely’ the two adjacent as cannot be syllabiﬁed together, due to the pre-existing foot structure. Similar reasoning in the other cases noted by Poser results in their elimination as evidence for a lexical process of vowel coalescence, and a fortiori for the lexical nature of the deﬁnitive accent shift. As a result, there is nothing to prevent an analysis on which this accent shift occurs in the post-lexical phonology, as should be the case if it is a modiﬁcation at the phrase level. All of this assumes, of course, that the deﬁnitive accent is marked as I have been assuming: by a shift in the location of the stress from the penultimate mora to the ﬁnal. This is clearly implied in Churchward’s description of, e.g., he falé ‘the house’ as ending in a stressed short vowel, as opposed to he fale akó ‘the school house’ where stress falls on the ﬁrst syllable of fale. But questions 10

Or ﬁnally, of course, in words with the deﬁnitive accent or an added clitic.

98

Aspects of the Theory of Clitics

have been raised in the subsequent literature which appear to undermine this account of the facts. Recall that historically, the deﬁnitive accent appears to originate in an earlier clitic demonstrative *a which shifts stress one mora to the right, assimilates to the ﬁnal vowel, and then disappears. But what if this element has not, in fact, disappeared in the modern language? We might then assume that the deﬁnitive accent is represented by a single unspeciﬁed vocalic mora. When this mora is added to a (phrase-ﬁnal) word, it results in a shift of the stress onto a ﬁnal syllable, which now contains two moras. Assuming the new mora links to the melody of the basic ﬁnal vowel, this syllable ought to be long, and stressed (like all ﬁnal long vowels). On this account we would still have a phrasal special clitic for the deﬁnitive accent, but since it has a segmentable form (an empty vocalic mora), it would no longer serve as an example of a non-afﬁxal clitic. Distinguishing the two analyses at issue here would seem to be a simple matter: either the ﬁnal vowel is short, in which case we have a non-afﬁxal clitic, or else it is long, in which case we have a straightforward phrasal afﬁx not much different from the English possessive. Churchward’s description is fairly explicit in favoring the ﬁrst interpretation, but Taumoefolau (2003), a native speaker, argues that the ﬁnal vowel on which the deﬁnitive accent falls in Tongan is geminated, or long. Measurements by Anderson and Otsuka (2002) do ﬁnd some lengthening in these vowels. Furthermore, the related language Pukapukan has a very similar marker for deﬁniteness, and Salisbury (2002) reports that the stressed ﬁnal vowels are lengthened in this language. These accounts would seem to favor the “empty mora afﬁx” version of the Tongan deﬁnitive accent over Churchward’s. The most careful set of phonetic measurements available, though, seems to suggest that neither account is quite right: the ﬁnal stressed vowels in words with the deﬁnitive accent are shown by Condax (1989) to be longer than other (stressed) short vowels, but shorter than other (stressed) long vowels. Furthermore, there are other differences between words with ﬁnal deﬁnitive accent and words with ﬁnal long vowels. One of these is the fact that the penultimate syllable (which would have primary stress in the absence of deﬁnitive accent) displays a secondary stress, yielding a stress clash (or sequence of two consecutive stressed syllables) that does not occur otherwise in the language and whose exact phonetic properties are therefore difﬁcult to project independently. In addition, there are pitch contours associated with deﬁnitive accented words that are similarly isolated within Tongan phonology. Condax concludes that the deﬁnitive accent consists neither in a simple shift of stress (as implied by much of the literature, based on Churchward’s

Special Clitics and their Grammar

99

description) nor a simple addition of a mora at the right edge of the word. There is, rather, a unique re-organization of the word’s prosodic pattern which serves to mark deﬁniteness. In the absence of a full analysis of the facts as Condax reports them, it is hazardous to venture a categorial interpretation. Nonetheless, it seems plausible to suggest that the deﬁnitive accent in Tongan is marked by a process that applies at the right edge of phrase-ﬁnal words, rather than by the addition of speciﬁable segmental material (an afﬁx). As such, it does seem to constitute a non-afﬁxal special clitic after all. Careful observation shows that the deﬁnitive accent in Pukapukan, as well, involves more than the addition of an extra mora. After a thorough discussion of the phonetic differences between phrases with and without the accent, including comparisons with the Tongan facts cited above, Salisbury (2002: 42) concludes that “the deﬁnitive accent is a grammatical morpheme which is not realized as a phoneme, but as a prosodic element on the ﬁnal vowel of a phrase. Phonetically, it adds the length of one mora to the ﬁnal vowel and changes the pitch contour, but the resultant length is not phonemically equivalent to that of a long vowel.” And there is more. Tongan is not the only language that Churchward worked on and in which he found phenomena of the sort we have just been discussing. In his Tongan Grammar (Churchward 1953: 268) he suggests that “[f]undamentally the distinction between the ordinary form of a Tongan word and its deﬁnitive form corresponds to the distinction between the two ‘phases’ of Rotuman words” and refers us to his grammar of that language11 (Churchward 1940), for further elaboration. The basis of the parallel, as he puts it (Churchward 1953: 269), is that “Rotuman [. . . ] takes the original form of the word as deﬁnite, and then, by weakening it, makes it indeﬁnite, while Tongan takes the original form as indeﬁnite (or only semi-deﬁnite), and then, by strengthening it, makes it deﬁnite. In both cases, however, the result is the same: the stronger form of the word is comparatively deﬁnite, the weaker form comparatively indeﬁnite.” Recent phonological analyses of the difference between the phases in Rotuman (e.g., McCarthy 2000) make this analogy even more compelling by analyzing the incomplete phase as resulting from an accent shift applied to the complete phase form.

11 The position of Rotuman within Austronesian is a matter of some discussion, but the language is not related in a uniquely close way to Tongan and/or Pukapukan. Schmidt (1999) summarizes previous work, arguing that Rotuman is part of a group including some Fijian dialects which is coordinate with Polynesian as a whole. It is thus relatively separated historically from Tongan and Pukapukan.

100

Aspects of the Theory of Clitics

The differences between the “complete” and “incomplete phase” forms in Rotuman have been the object of much discussion in the phonological literature in connection with the appearance of metathesis, perhaps morphologically conditioned, in some of the relevant forms. There are several distinct phonological changes involved in different sets of forms, but the choice of one or the other is completely determined by the shape of the stem involved. Rehearsing the full range of facts would take us much too far aﬁeld, but fortunately the phonological speciﬁcs are not crucial for this discussion.12 What matters is (a) that the incomplete phase is clearly derived from the complete phase, and (b) a signiﬁcant distinction between the two appears only at the right edge of a full DP. Non-ﬁnal words within the DP all appear in the incomplete phase, as do words preceding a speciﬁc set of sufﬁxes and clitics. Examples can be found in the works cited above; it is uncontroversial that the phase distinction appears only on the ﬁnal word of the phrase, and so this is an excellent candidate for the status of a special clitic. Hale and Kissock (1998) propose, however, that there is no independent marker of “complete phase” (or of “incomplete phase”) at all. On their account, one set of afﬁxes and clitics conditions the appearance of incomplete phase forms, while another set conditions the appearance of complete phase forms. The latter set includes, somewhat suspiciously, a marker –∅ ‘deﬁnite plural/locative’ which contributes an abstract mora that never appears in surface forms.13 The distinction between the two phases, according to Hale and Kissock, is nothing more than a complex but phonologically predictable concomitant of the difference between the two sets of afﬁx. McCarthy (2000) follows essentially the same analysis, though differing with Hale and Kissock on other points. However appealing a purely phonological analysis of the Rotuman phase distinction may be, Kurisu (2002; see also Kurisu 2001) shows that it does not completely account for the facts. At least some afﬁxes and clitics require a phase form of a preceding word which is not predictable from its phonological shape, and this includes one category with no formal marker (the deﬁnite plural/locative –∅ of Hale and Kissock’s and McCarthy’s analysis). As a result,

12 See McCarthy (1995, 2000) and Kurisu (2002) for summaries of the literature and speciﬁc formal proposals for analyzing the phonology of the Rotuman incomplete phase forms. 13 The parallel between Rotuman and Tongan originally noted by Churchward rests on the fact that a complete phase form with no overt marker serves as deﬁnite in contrast with a similarly unmarked indeﬁnite form in the incomplete phase. For Hale and Kissock this corresponds to the absence versus the presence of this –∅.

Special Clitics and their Grammar

101

Kurisu concludes that the alternation is triggered (at least in part) by morphological factors. But since its realization is at the right edge of a phrasal constituent, this must be phrasal morphology: a special clitic. And if this reasoning is correct, the resulting special clitic appears to provide us with yet another non-afﬁxal instance. Case III: Kw akw ’ala Special Clitics A ﬁnal and somewhat more elaborate example involving an entire system of special clitics is furnished by Kw akw ’ala. Much, but not all, of the complexity of this system is located in the distinctions made by its determiners, whose analysis was begun in Chapter 2 and to which I now return. The pronominal simple clitics already mentioned distinguish only grammatical function and person, but Kw akw ’ala has a much richer set of distinctions that are carried by its determiners and pronouns. In addition to the basic properties previously illustrated, the weak pronominals also display deictic distinctions. In principle, at least, the form of every pronominal indicates whether its referent is near the speaker, near the addressee, or distant from the speech situation, as well as whether it is visible or not. Some dialects recorded by Boas also indicate whether the referent still exists at the time of the speech event or not, but these distinctions were not present in forms of the language which I was able to record in the late 1970s. As an example of the kind of distinction this system makes possible, the form –quP in (4.22) makes explicit the fact that thing forgotten is presently near the addressee, but not visible to the speaker. (4.22)

la- x.@nt=@n ň’@lsPola=quP aux-evid- I leave and forget- OBJ(2ND.INV) Evidently I left and forgot about it (near 2nd p., invis.)

A full chart of the third-person clitic demonstrative pronouns showing these distinctions is given in Figure 4.2. Not only pronouns show this range of distinctions; full nominals do as well. All of the deictic distinctions made by the pronouns in Figure 4.2 also

Subj: Obj: Inst:

1st.vis -k -q@k -s@k

1st.inv -gaP -x.gaP -sgaP

2nd.vis -ux. -qw -sux.

2nd.inv -uP -quP -suP

3rd.vis -iq -q -s

3rd.inv -iP -qi -si

Figure 4.2. Kw akw ’ala (third- person) clitic demonstrative pronouns

102

Aspects of the Theory of Clitics

appear in full nominal phrases. In these, however, the complete formal expression of the relevant categories is spread across two separate structural elements, as shown in the sentences in (4.23). A fuller chart of these determiner elements is given in Figure 4.3. Notice that the two separate elements in a full determiner combine to give the full deictic category of the nominal expression. (4.23)

qaq@s@l- is- ň lax.=gada a. la- ň=@n aux-fut- I go along- beach- fut to- dem(1st) aňi’nagw is=@k country- dem(1st.vis) I shall go along the beach to this inland country (near 1st person, vis.) b. dux.‘wida- s=x.-ux.da gukw =ix. see- you- obj-dem(2nd) house- dem(2nd.vis) Do you see this house (near 2nd person, vis.)? wis=aq’ c. ka’yadzoxiňa=x.-ux. drive away- obj-dem(2nd) boy- dem(2nd.inv) Drive away this boy (near 2nd person, invisible)! dux.‘waň@la laqw @layug.w a=i=x.a d. la’laPi aux(evid) saw Laqw @layug.w a- dem(3rd.inv)- obj(3rd) w guk =i house- dem(3rd.inv) Then, it is said, Laqw @layug.w a (name) saw the house (neither L nor house visible)

Complicating matters still further, there is yet another dimension of information which is carried by determiners: that of possessor person. Where a nominal is possessed, the determiner indicates whether the possessor is ﬁrst, second, or third person. For third-person possessors, a further distinction is made based on whether the possessor is identical with the subject of the clause Pre- nominal: Subj: - ga(da) - ux(da) - i(da) Obj: - x.ga(da) - xw a, - x.ux.(da) - x.(a) Inst: - sga(da) - sa, - sux(da) - s(a) Post- nominal: -k - ga - ix. - aq’ -∅ - a/i 1st.vis 1st.inv 2nd.vis 2nd.inv 3rd.vis 3rd.inv Figure 4.3. Kw akw ’ala DP demonstrative clitics (unpossessed)

Special Clitics and their Grammar

103

or not. Examples of determiners cumulating the deictic and possessive distinctions are given in (4.24). A chart of the system of demonstrative clitics which appear with possessed nominals is given in Figure 4.4. (4.24)

a. Pix’m=is wałd@m=os good- your word- dem(2nd poss of 3rd.vis) Your word is good gagima- xd=a@n b. g.w @nał gax.- @n lax.=@n pay to- me to (for)- my loan- past-dem(1st poss of 3rd.inv) l- o ň to- you Pay me for what I loaned to you (mine, distant, invisible) c. dux.’wid- ∅=x.us xw @nukw =ix. see- she- her child- dem(3rd poss of 2nd.vis, poss=subject) (Let her) see her child (hers, near you, visible) d. la- ’mis=@n g.@g- [email protected] aux- so- I wife- have- instr-dem(2nd.vis) k’idała=x.s princess- dem(3rd poss of 2nd.vis, poss=subject) And so I have this their princess (theirs, near you) as wife

How are we to describe these determiner elements? The “pre-nominal” component is straightforward: it is a clitic, appearing at the left edge of the nominal. We could treat it as we have to this point, as an item appearing in a structural determiner position which is initial within the nominal. Alternatively, we could describe it as a left-edge special clitic, but the choice between

Demonstrative of 1ST.INV 2ND.VIS 2ND.INV 3RD.VIS 3RD.INV Possessor 1ST.VIS 1ST (SG,INCL,EXCL), PRE- N - g@n, - g@nts, - [email protected] - @n, - @nts, - @nuPx.w 2ND, PRE- N - gas - us, - x.s, - ux.s - is 3RD, POSS=SBJ, PRE- N - ga - ux. -i (all above) post- N - g- ga- q/x.- - q’∅ - a(followed by INSTR PRO) 3RD, POSS=SBJ, PRE- N - gas - us - is 3RD, POSS=SBJ, POST- N - k - gaP - q, - ix. - q’, - aq’ ∅ -a

Figure 4.4. Kw akw ’ala DP demonstrative clitics (possessed)

104

Aspects of the Theory of Clitics

these two alternatives is not clear and not relevant to our immediate concerns, so we will continue to assume it is a simple (phonological) clitic occupying the D position. But what of the second, “post-nominal” component of these complex determiners? In the sentences we have seen to this point, we might describe that as appearing either (a) at the right edge of the nominal expression, or (b) as following the head noun, since all of the nominals in those sentences consisted exactly of a single word, the head. When we consider nominals containing more than one word, though, it becomes clear that neither of these accounts of the “post-nominal determiners” is adequate. The sentences in (4.25) illustrate this. (4.25)

b@gw an@m=x.a g@nan@m a. m@x’id=ida ’walas=i hit- dem(3rd) big- dem(3rd.inv) man- obj(dem) child The big man hit the child g@nan@m b. (vs.) m@x’id=ida b@gw an@m=a=x.a hit- dem(3rd) man- dem(3rd.inv)- obj(dem) child The man hit the child c. le næ’nakw lax.=is h@s=aq gukw aux goes home to- self ’s(dem) own-dem(2nd.inv) house She goes home to her own house k’is=ix. ał’@m wałdema d. g.ilagaP la=xw a come to- dem(2nd) not- dem(2nd.vis) new word Come to this not new speech

In fact, rather than being “right-edge” or “post-head” clitics, these elements are actually located in second position within the nominal, following the ﬁrst independent word—regardless of that word’s function. Examples like (4.25d) show that it is indeed the ﬁrst word of the DP (and not, for example, the ﬁrst syntactic constituent of the DP, which in this case would be the two-word AdjP k’is ał m ‘not new’) that serves as the anchor for the second determiner clitic. These second elements of the Determiner system are clearly special and not (just) simple clitics, since they are positioned in a way that has no obvious basis in the syntax of the DP. If we assume that the ‘left-edge’ Determiner element occupies the D position, the other element is a second-position clitic within the complement of D. Note that these structural consequences are the same regardless of whether we regard the nominals as basically DPs or as NPs, as shown in (4.26). e

Special Clitics and their Grammar (4.26) a.

DP NP D1 Word D2 X

b.

105

NP

N D1 Word D2 X

Special clitics of this type pose particularly serious challenges to standard theories of clitic placement. The notion of ‘second position’ that they are based on requires reference to a ‘ﬁrst position’ that is ﬁlled by the ﬁrst PWord of the domain (or in some languages another prosodic constituent such as the PPhrase: see the discussion of Chamorro in Chapter 5 below). These are phonological notions, and as such are generally assumed not to be accessible to syntactic processes. In the following chapter I will discuss a range of approaches that have been taken to such second-position phenomena within various frameworks, and offer reasons to believe that neither the syntax nor the phonology can provide the relevant apparatus. Kw akw ’ala also displays other clitic elements, each with its own interest for the overall analysis of the language, but these do not present new sorts of problem for a general theory. Some of them are arguably simple clitics. For example, a Complementizer such as (i)x. ‘that’ in example (4.27) appears at the left edge of the complement clause and like other simple clitics in Kw akw ’ala, attaches to a host on its left. In fact, this host will always be a word that is part of the matrix clause, as in (4.27). (4.27)

dux.’waň@la=q=ix.=s w@nq@la=ida xw @p’a deep- dem hole saw- it- comp-3sbj He saw that the hole was deep

Other grammatical elements in Kw akw ’ala are special clitics, such as the Subject pronominals which appear in second position within the CP. Within main clauses, these will come after the clause-initial verb, but in subordinate clauses ‘second position’ will be immediately after the Complementizer and thus before the verb. Both possibilities are illustrated by the sentence in example (4.28). (4.28) qa=suP g.w ag.w ixsPala la=qw q@=ida ’n@mukw =i if- you talk about on- it to- dem one- dem(3rd.inv) b@gw an@m . . . man ... la- ň=@@n kw ix.Pid@- ň=uň aux- fut- I club- fut- you If you talk about this to (any) one man. . . I’ll club you

106

Aspects of the Theory of Clitics

The domain here is different from what we have already seen in DP special clitics, but the parameters affecting these elements’ positioning are the same. In this chapter, I have surveyed a range of special clitic phenomena which appear to represent what we ﬁnd in the languages of the world. Although I have argued that close parallels with word-level morphology suggest that it is appropriate to view special clitics as the morphology of phrases, and to describe them by essentially morphological mechanisms, I have not yet shown that alternative, more traditional treatments within the syntax are not available. In Chapter 5, I address that issue, as the foundation for providing a more explicit account of what such a “morphological” theory of special clitics might look like in Chapter 6.

5 Theories of Special Clitics Up to this point we have seen a representative range of phenomena that fall under the rubric “special clitic,” and we have also seen some interesting analogies between these elements and the formal markers that appear within words as a consequence of their morphological properties. However interesting such an analogy may be, though, it does not by itself constitute a theory of special clitics. And indeed, most of the existing literature assumes that a real theory of these phenomena should be located within the domain of syntax—or, for a few authors, the phonology. The task of this chapter is to ﬂesh out the suggestion that both of these views are mistaken, and that neither the constructs and mechanisms of the syntax nor those of the phonology are fully appropriate to the description of special clitics. Instead, the possibility raised in Chapter 4 that these constitute the morphology of phrases is more than a suggestive metaphor, and the formal devices of a theory of word structure should be recruited for this purpose as well. I begin by showing something of the nature of the problem, in the form of an apparent conﬂict between the fundamental terms of the generalizations we saw in Chapter 4 and those of the core components of a grammar as standardly construed, the syntax and the phonology. These issues come into focus in connection with second-position clitics,1 in particular, and the reasons for this will be developed. I will then review a careful survey by Boškovi´c of the range of theories we might entertain to deal with these matters within the syntax, and conclude with him that this task cannot be conﬁded to the syntax alone. I diverge from Boškovi´c, however, on the appropriate response to this and, in Chapter 6, I will develop the outlines of a morphological view, based on mechanisms of Optimality Theory.

1 Penultimate-position clitics ought to pose an exactly symmetrical range of issues, in principle, but as we have seen there are few if any examples of this type to examine. I will therefore conﬁne my attention to the second-position case.

108

Aspects of the Theory of Clitics

5.1 The Nature of the Problem If we want to explore the question of what kind of theory is best suited to describe the distribution of special clitics, it is clear that only a limited subset of possible clitic types is likely to be relevant. Of the four possible varieties of special clitics discussed in Chapter 4, only the second-position type (or the penultimate type, if any such examples exist) will be of probative value. That is because essentially any sort of theory should have no difﬁculty locating elements in initial or ﬁnal position within a domain. Second (or penultimate) position, on the other hand, is somewhat more theoretically challenging. When Clitics are Second, What Comes First? This is the most basic question: what constitutes the ﬁrst (or anchoring) element that a second-position clitic follows, in terms of the typology in (4.10)? I have already noted that it may be necessary to allow this to be a prosodic constituent (PWord, PPhrase, possibly even IntPhrase), whether or not this corresponds to a syntactically unitary (and recognizable) constituent. This will prove to be a central issue in developing an adequate theory of special clitics. It is apparent that Wackernagel’s (1892) idea of “his” position was “after the ﬁrst word,” and not “after the ﬁrst (syntactic) constituent.” The examples he cites quite generally involve clitics that appear after exactly one clause-initial word, rather than after a phrasal constituent. Although he himself undoubtedly interpreted these in terms of (what we would see today as) a prosodic unit rather than a syntactic one, that is not as obvious as it might seem. Wackernagel’s citations are primarily from Homeric Greek, Vedic Sanskrit, and early Latin; and all of these (as well as their later stages which provide some other examples) allow rather free word order. If we treat this ﬂexibility as resulting from a syntactic operation of “Scrambling,” it follows that these languages must allow phrases to be broken up into smaller constituents rather freely. In consequence, it is next to impossible to ﬁnd evidence against the claim that the single word preceding a second-position clitic in Wackernagel’s examples is not also a syntactic constituent. The absence of examples with multi-word constituents preceding the clitic, even when these might be expected sentence initially in an unmarked order, could be seen as at least suggestive: it implies that the break-up of constituents into their minimal “scramble-able” units must have been obligatory, even in the absence of reordering. Nonetheless, the evidential value of Wackernagel’s examples for the point at issue is much less than it might be, in light of other aspects of the structure of the languages from which they are drawn.

Theories of Special Clitics

109

Somewhat better evidence is provided by Hittite,2 a language that does not appear to have presented much variation in word order of the sort described by “Scrambling.” Hittite has generally rather ﬁxed word order, and there is no evidence in this language for the existence of scrambling processes that would presuppose a break-up of phrasal constituents into their component words. Nonetheless, we ﬁnd that Hittite clitics generally follow the ﬁrst full word of the clause, even when this involves locating them inside phrases of the sort that are not easily broken up even in scrambling languages, as in (5.1). (5.1)

a. kun=wa=za DUMU-an da nu=kan É.ŠÀ-ni this=quot=refl baby-acc take conn=dir house-loc anda it . . . in go . . . Take this baby and go into the house . . . (KUB XXIV 7 IV 45–6) D IM-unni a¯ ssus e¯ sta b. nepisas=as=sta heaven-gen=he=ptc Weather-God-dat dear-nom he-was He was dear to the Weather-God of Heaven (StBoT 18.2)

c. Lupakkin=ma=kan Uzalman=na INA Amka paraa Lupakkis=quot=ptc Uzalaas=and into Amka forth naista he-sent He sent L. and U. forth to Amka (JCS 10, p. 94) The material preceding Hittite second-position clitics is not in general consistent with the claim that it should consist of exactly the ﬁrst syntactic daughter constituent of the clause. Another system in which “second-position” placement of clitics is deﬁned by an initial word rather than an initial phrase is West Greenlandic. Sadock (2003: 61) says that “[c]ertain clitics are positioned syntactically before a syntactic phrase and are sufﬁxed to the ﬁrst word of that phrase. This property of clitics provides a test for wordhood” (my emphasis). Among the relevant clitics are the conjunctional elements =lu ‘and’, =luunniit ‘or’, and =li ‘but’, each of which is “positioned syntactically between phrases [and] will show up as a sufﬁx to the ﬁrst word of the phrase it precedes” (ibid). An example in which this entails placement within an initial phrasal constituent is provided by (5.2) from the text in Sadock (2003: 71). 2 The Hittite material available to us today was still undeciphered when Wackernagel wrote in 1892, of course. His interests were in the early history of Indo-European, and the Indo-European afﬁliation of Hittite was not even suggested until nearly a quarter of a century later. It is hardly surprising that he does not cite this evidence, although it provides interesting conﬁrmation of his proposals.

110

(5.2)

Aspects of the Theory of Clitics

Inuil=li tamaani nunaqartut iluanni person.dir.pl-but here.loc land.have.ing.dir.pl midst.loc.3p.s pinngoriartorpoq erinineq ... thing.become.more.and.more.ind.3s long.for.nom.abs.sg . . . [But] amongst the Inuit living here there arose yearning . . .

Another example (provided by Sadock, personal communication, from Lynge 1978) is the underlined instance of =lu ‘and’ in the second conjunct of the sentence in (5.3). (5.3)

276-inik ammassattortoq ... . . . oqartarpor=lu. . . . . . say.hab.ind.3s-and. . . 276-inst.pl sardine.eat.part.3s . . . ilivitsut marluk nipisal=lu blowﬁsh.dir.p-and whole.dir.p two . . . and she often said that he ate 276 sardines . . . and two whole blowﬁsh

In the modern syntactic literature, the claim that “second position” might be deﬁned in some instances by a prosodic unit (e.g., a word) that did not constitute the ﬁrst syntactic daughter of the clause ﬁrst appeared in work by Wayles Browne (1974, 1975).3 Browne argued that in Serbo-Croatian, either the ﬁrst full word or the ﬁrst full constituent of the clause can precede the clitic(s), as in (5.4). (5.4)

a. i. Moja c´ e mladja sestra do´ci u my fut younger sister come on My younger sister will come on Tuesday ii. Moja mladja sestra c´ e do´ci u my younger sister fut come on My younger sister will come on Tuesday b. i. Lav je Tolstoi veliki ruski pisac Leo is Tolstoi great Russian writer Leo Tolstoy is a great Russian writer ii. Lav Tolstoi je veliki ruski pisac Leo Tolstoi is great Russian writer Leo Tolstoy is a great Russian writer

utorak Tuesday utorak Tuesday

3 See also the extensive reviews in Sch¨ utze (1994) and Boˇskovi´c (2001), as well as literature cited in those sources.

Theories of Special Clitics

111

Some native speaker linguists have questioned the acceptability of the data cited by Browne, but it seems that (as so often in such discussions) what is at stake is more a difference among dialects than a categorical distinction. In general, and simplifying somewhat, dialects spoken in the western part of the Serbo-Croatian area (grossly, “Croatian” dialects) admit clitics after the ﬁrst word, while dialects more to the east (grossly, “Serbian”) only allow them after the ﬁrst full syntactic constituent. Thus, the (i) examples in (5.4) are much less acceptable than the (ii) examples for these eastern speakers, who prefer clitics to be located after a full syntactic phrase (and not, for instance, after the ﬁrst word of a two-word proper name). This sort of variation is quite interesting, of course, and it is important to be able to characterize it formally. It does not, however, directly impugn the claim that some languages allow clitics to be placed after an anchor whose unitary status is prosodic in nature rather than syntactic. There is no doubt that systems exist where the initial anchor is deﬁned in syntactic terms: what is at stake is whether the prosodic anchor type occurs as well. If this possibility is instantiated in even a single dialect of a single language, it must still be accommodated within the terms of Universal Grammar, thus posing a problem for theories that assume a uniformly syntactic solution to the problem of second-position clitic placement. DP Second-Position Clitics The examples of second-position clitics we have seen thus far all involve clitics whose domain is the clause, but of course some languages have secondposition clitics within other domains. For example, in section 4.3 I presented the set of second-determiner elements in Kw akw ’ala, clitics that are placed immediately after the initial word of the DP. Somewhat better-known cases of second-position elements within nominals, though, are the postposed determiners found in several Balkan languages, including Albanian, Bulgarian, Macedonian, and Romanian. I illustrate this type from Bulgarian, a language in which deﬁnite articles appear immediately after the ﬁrst word of the nominal as in (5.5). (5.5)

a. knigi=te ‘the books’ b. interesni=te knigi ‘the interesting books’ c. mnogo=to interesni knigi ‘the many interesting books’

There have been two basic analyses offered for such cases. One account, associated particularly with Halpern (1992a, 1992b, 1995) is Phonological in character, and based on a proposed operation of Prosodic Inversion. This view

112

Aspects of the Theory of Clitics

assumes that the clitic is “basically” positioned at the left edge of its domain, by syntactically trivial mechanisms. Subsequently, inversion of the clitic with a following word occurs somewhere in the mapping between syntactic structure and Phonetic Form so as to locate the clitic in second position. Such a post-syntactic manipulation of word order is assumed to be forced in order to satisfy prosodic requirements of the clitic. If this element is subcategorized to require that it attach to a PWord on its left within the phrase that constitutes its domain, that will only be possible (given its syntactically phrase-initial position) if it moves so as to assure that there is (at least) a single phonological word preceding it within that domain. The notion that grammatical elements can be subcategorized for a speciﬁc prosodic environment in which they may (or must) appear is a widely held conception, developed most explicitly perhaps by Inkelas (1989). It would surely be preferable, however, if we could derive such restrictions from more general principles. I have already suggested in Chapter 3 that similar itemspeciﬁc stipulations about the direction of attachment can (and should) be replaced by a full account of general principles of Stray Adjunction within a language. To the extent possible, it would be desirable to maintain a similar position with respect to the broader notion of “Prosodic Subcategorization.” In that case, however, the analysis of second-position phenomena in terms of Prosodic Inversion would not be possible. Rather than identifying particular items with respect to the prosodic afﬁliation they should exhibit, I assume that these facts should follow from an identiﬁcation of the item’s prosodic category (syllable, Foot, PWord, etc.) together with the principles of Stray Adjunction that govern the language as a whole. Regardless of that consideration, though, the Prosodic Inversion analysis cannot accommodate the Kw akw ’ala second determiners. Assuming that these clitics are initially generated in the left periphery of the DP, they would already be preceded by material that could support left attachment, without requiring any inversion. We can tell that the material preceding the left edge of the nominal is an appropriate host for clitics, because in fact the phrase-initial determiner elements consistently attach in just that way. There could not, then, be a prosodic motivation for their repositioning. We will see some further problems for the phonological view of secondposition clitic placement below. In any event, most linguists have not adopted that position, but have rather preferred to see second-position clitics as located within the syntax, by syntactic mechanisms. Nearly all of the discussion of this matter has concerned clitics located within a clausal domain (such as those of Serbo-Croatian), but before passing on to a more detailed examination of syntactic theories of such clitics, I will consider brieﬂy the way such a theory plays

Theories of Special Clitics

113

out in the particular case of DP-internal clitics, like the Kw akw ’ala determiners or the Balkan determiners. On the syntactic view, the syntax generates the clitic in the head D position within a DP. A subsequent operation of displacement must then raise precisely the ﬁrst word of the embedded NP to SpecDP, as illustrated (for Bulgarian) in (5.6). (5.6)

DP D

SpecDP D mnogoi

NP

=to [ei] interesni knigi

The syntactic account is motivated by a theory-internal assumption that the syntax must be the locus of description for such facts, so it is perhaps appropriate that it raises some general theory-internal problems even within Balkan languages such as Bulgarian, Macedonian, Albanian, and Romanian, for which it has been most widely discussed. One such problem is the fact that the displacement in (5.6) crucially involves a single word rather than a complete phrase. As such, it must be the kind of displacement known as “Head Movement,” rather than normal phrasal movement. But the target of the displacement in this case is not a head position, but phrasal: SpecDP. To the extent Head Movement is assumed to have properties distinct from those of ordinary phrasal movement, this presents a conceptual anomaly. Secondly, we can ask what the motivation is for the displacement in (5.6). Apparently, this is driven only by the needs of the clitic determiner (its presumed prosodic requirements), and not by those of the word that moves. But within at least one version of the sort of theory that is at issue here, movement is only supposed to be driven by the needs of the item that moves, rather than by the resulting conﬁguration. The movement in (5.6) would present a conceptual problem for any such account. Additional problems for this analysis arise in attempting to extend it to the case of the Kw akw ’ala second determiners. First, it is unclear where these should be generated within the DP. There is already a clitic in phrase-initial position (the initial determiner element), so where is the second one introduced? Must we say that somehow DPs have two heads in such a language? The only

114

Aspects of the Theory of Clitics

other motivated head position in sight is that of N, the head of NP, within DP, but that is surely not available for the second of two elements jointly realizing the content of D. In an analysis whose ontology of linguistic elements is based resolutely on the traditional morpheme, this is difﬁcult to resolve, although the analogy suggested in Chapter 4 between clitics and morphology allows us to regard it simply as one more instance of “multiple exponence” (Anderson 2001). Furthermore, just as in the phonological account, there is no motivation (syntactic or prosodic) for the necessary displacement of the initial nondeterminer word of the DP to the left of the second determiner element. This word need not move for its own purposes; and even if we allow for “altruistic movement” there is no prosodic requirement that would go unmet if it remained in its basic position, as noted above in connection with the Prosodic Inversion account. I conclude that an analysis of these second-position clitics as resulting from movement within either the syntax or the phonology cannot be maintained. Other Prosodically deﬁned Clitic Types Kw akw ’ala, Hittite, West Greenlandic, and (perhaps some forms of) SerboCroatian provide us with examples in which the anchoring element relative to which a clitic appears in second position is a prosodic unit, the PWord, and not a syntactic one. In principle, at least, if the PWord can constitute such an anchor, it should be possible for some language to determine the notion of second position by reference to some other prosodic constituent type.4 Exactly such a system is described and analyzed in detail by Chung (2003). Chamorro has a set of weak pronominals that can represent direct objects and certain intransitive subjects.5 These elements are prosodically deﬁcient, and must attach phonologically to a preceding PWord. Clearly, then, they have the properties of phonological clitics, as Chung shows. Furthermore, they can occur in positions where corresponding non-clitic DPs cannot, as illustrated by the pair of sentences in (5.7). (5.7)

a. Kao patgon-ña =hao ädyu na ma’estra? Q child-agr you that linker teacher Are you the child of that teacher?

4 Indeed, Halpern (1992b, 1995) entertains this possibility in connection with the operation of Prosodic Inversion, though he does not develop it in detail. 5 Transitive subjects and the subjects of irrealis verbs and adjectives are cross-referenced by agreement material in Chamorro, and it is a general principle of the language that pronominals indexed by agreement must be phonologically null. This complementarity between agreement and the possibility of a weak pronominal is interestingly consistent with the position I will develop in Chapter 8, according to which pronominal special clitics are themselves a form of agreement material.

Theories of Special Clitics

115

b. *Kao patgon-ña si Dolores ädyu na ma’estra? Q child-agr art Dolores that linker teacher (Is Dolores the child of that teacher?) The weak pronouns, then, have the properties of pronominal special clitics. Assuming that the question-marking particle kao in (5.7a) is either a proclitic attached to the following word or outside the domain within which the clitic appears, such a sentence would suggest that these elements are located in second position, anchored by the ﬁrst PWord of their domain. This is consistent with many other examples, such as (5.8). (5.8)

Fahani =yu’ =fan gä’-hu ga’lagu buy.for me please pet-agr dog Please buy me a pet dog

However, in many other examples such as (5.9) it appears the clitics follow an entire syntactic phrase, rather than a single PWord. (5.9)

a. Man-maleffa na [mansiudadanu-n Amerikanu] =hit lokkui’ agr-forget comp citizens-linker American we also They forget that we are American citizens too b. [Más yä-hu na taotao] =hao most wh[obj].like-agr linker person you You’re the person I like most c. [Ginin San Roque na songsung] =yu’ from San Roque linker village I I’m from San Roque village

Furthermore, in other examples such as (5.10) the material preceding the clitic is more than a single PWord but not a syntactically complete phrase, either. (5.10) Taotao San Roque =yu’ na songsung person.linker San Roque I linker village I’m a person of San Roque village Chung shows in detail that the material preceding the clitic need not constitute an entire syntactic constituent. This is true in a great many examples, such as (5.7a) and the sentences in (5.11). (5.11)

a. I más amku’ =gui’ na chi’lu-hu palao’an the most old she linker sibling-agr female She’s my oldest sister

116

Aspects of the Theory of Clitics gima’ ni ginin ädyu =yu’ i b. Malak agr.go.to that I the house Comp imperf =yu’ staba s-um-ásaga agr-stay.prog I used.to I went to that house where I used to live

How are we to reconcile these apparently conﬂicting facts about the placement of Chamorro pronominal clitics? Chung argues that the correct (and exceptionless) generalization is the formulation in (5.12). (5.12) Pronominal clitics are located immediately after the ﬁrst PPhrase within the IntPhrase with which they are associated. In support of this, she develops a set of speciﬁc principles that relate syntactic structure to prosodic structure in Chamorro, showing that the PPhrases they project are precisely the units that serve as initial anchors for the clitics. Indeed, where these principles provide more than one option for phrasing, a corresponding option exists for the placement of a clitic. The details underlying this analysis are not immediately relevant here, and the reader is referred to Chung (2003) for much additional discussion. What matters to us is the conclusion that in this language, the prosodic constituent PPhrase serves as the anchor for a set of second-position clitics, showing that the PWord is not the only prosodic type that can serve this purpose.6 These prosodic units are constructed on the basis of syntactic structure, but do not correspond directly to units that the syntax can manipulate autonomously.

5.2 Syntactic Theories of Clitic Placement The examples discussed above pose serious problems for any approach to the positioning of special clitics that allocates this task entirely to the syntax. The most extensive survey of the adequacy of syntactic theories is that of Boškovi´c (2000, 2001), a discussion that concentrates on Serbo-Croatian and related languages of the Balkans while addressing the relevant issues in a rather general fashion. I turn now to a review of Boškovi´c’s arguments. Assuming that the available options for a theory of second-position clitics reduce to a balance between syntactic and phonological mechanisms, Boškovi´c distinguishes four general classes of theory:

6 Aissen (1992) also argues that in some Mayan languages, including Tzotzil and Jacaltec, there are clitics whose location is speciﬁed as following an IntPhrase.

Theories of Special Clitics

117

Strong Syntax: Clitic positioning is fully determined by the syntax. Strong Phonology: Phonology is fully responsible for placing clitics in second position. The syntax generates clitics in normal argument positions. They are then moved into second position as necessary in the phonology. Weak Syntax: Most movement of clitics is syntactic. However, Prosodic Inversion may intervene in the phonology to satisfy the requirement of a clitic for an appropriate host. Weak Phonology: Movement of clitics takes place in the syntax, and involves a considerable amount of freedom of positioning. The role of phonology is passive, ﬁltering out certain syntactically well-formed sentences that violate phonological requirements of the clitics. Accounts that involve movement of syntactically speciﬁed items within the phonology are undesirable a priori, according to Boškovi´c, and most analysts who have discussed the matter agree. We have already seen some reasons to believe such phonological movement is an unsatisfactory answer to the problem of second-position clitics, in the discussion of Kw akw ’ala DP clitics in section 5.1. With reference speciﬁcally to Serbo-Croatian, the language that has been at the center of discussion in this regard, Boškovi´c concludes that there is no persuasive evidence in favor of phonological movement (including Prosodic Inversion), and some evidence against it. One such argument comes from the fact that clitic placement seems to be sensitive to syntactic information of a sort that ought not to be available in the phonology. Placement of clitics in clitic-climbing constructions depends on the syntactic/semantic relation between the matrix verb and a complement, as suggested by the examples in (5.13). (5.13)

a. Milan kaže da =ga vidi Milan says that him sees Milan says that he sees him b. *Milan =ga kaže da vidi c. Milan želi da =ga vidi Milan wants that him sees Milan wants to see him d. ?Milan =ga želi da vidi

I will return to the analysis of clitic climbing in Chapter 8, but there is no reason to believe it is based on properties that are plausibly present in Phonetic Form.

118

Aspects of the Theory of Clitics

A related fact arguing against Prosodic Inversion as the source of postinitial positioning of clitics is that there are some circumstances in SerboCroatian where we ought to be able to invert a clitic with a clause-initial word, but for some reason this yields incorrect results, and some other movement must be invoked. This is illustrated in (5.14). (5.14)

a. Prema Mileni =su Milan i Jovan išli toward Milena aux Milan and Jovan walked Milan and Jovan walked toward Milena b. *Prema =su Mileni Milan i Jovan išli toward aux Milena Milan and Jovan walked Milan and Jovan walked toward Milena

Unlike most prepositions in Serbo-Croatian, prema bears stress and is presumably a full prosodic word, so if the clitics are generated in initial position and then undergo Prosodic Inversion around a following word, that should lead to placement after the preposition in this case. Boškovi´c relates the positioning of the clitic in (5.14) to another property of the phrase prema Mileni ‘toward Milena’ in Serbo-Croatian: while word order in the language is generally quite free, such a phrase must remain as a unit and cannot be separated through Scrambling. I will return later to the importance of this observation, but for present purposes its signiﬁcance is due to its interaction with the putative rule inverting a clitic with a following word so as to satisfy prosodic requirements. Apparently the generalization is: “Prosodic Inversion” can locate a clitic after an initial word if and only if that word constitutes a syntactic unit that can move independently of what follows it. Since prema and its object form an inseparable unit for the purposes of the syntax, this condition is not satisﬁed. But that means that whatever is responsible for the placement of the clitic, it cannot be something that is purely phonological in character, and so we are not really dealing with “Prosodic” inversion in the general case of second-position clitics. On the premise (introduced in previous chapters) that access to syntactic structure in the phonology can in general be reduced to access to the prosodic structure induced by syntactic phrasing, I conclude (with Boškovi´c) that there is no advantage to be gained from having movement that takes place in the phonology. Apparently, that excludes both ‘Strong Phonology’ and ‘Weak Syntax’ positions within Boškovi´c’s typology, and leaves us with accounts in which clitics are placed entirely by syntactic mechanisms. Focusing on second-position clitics, two major purely syntactic theories (and one minor one) are to be found in the literature.

Theories of Special Clitics

119

One of these assumes that clitics are initially placed in phrase-initial position. Subsequent syntactic operations then have the consequence of moving or adjoining exactly one constituent to the left of the clitics. Alternatively, we might assume that clitics are placed (either initially or as the result of movement) in a head position quite high within the structure. The Spec position associated with this head is then ﬁlled (again, either basically or via movement) with exactly one phrase. A third possible view, suggested by Franks (2000; see also Franks and Progovac 1999), can be seen as a variant of these. On this picture, it is assumed that the clitics are attached to the verb, which then moves (crucially, by a copying operation rather than simple movement) into second position. The copy of the verb, while in second position, is not however pronounced, though the copies of the clitics are. Correspondingly, the copy of the verb in its original is pronounced, while the instances of the clitics in this position are not. This approach, frankly, strikes me as simply bizarre, especially given that there is no evidence for the claim that verb second is operative in every language with second-position clitics. Boškovi´c (2001), nonetheless, ﬁnds advantages for this view within the range of theories based exclusively on syntactic mechanisms. It is not necessary to deal directly with these matters, however, since this view of second-position clitics has the same general consequences as the other two, as these are explored by Boškovi´c. Any theory describing the placement of clitics in second position by such purely syntactic mechanisms makes several predictions, some of which are explored by Boškovi´c: • Clitics cluster in the same position syntactically; • This position is esentially ﬁxed for all constructions; and • It is located high enough in the tree that there is no space for more than one constituent to precede it. Thus, there ought to be some consistent syntactically motivated position that clitics occupy. On one of the variants just described, this will be the position of an adjoined phrase on the left periphery, preceded by another adjoined phrase. On another, it will be a head position, perhaps head of CP, TP, or some other functional projection, and preceded by the speciﬁer of that category. The key to all of these analyses is the presence of a syntactically consistent target for clitic placement. All have in common the (implicit or explicit) claim that “second position” is only an accidental, epiphenomenal consequence of the presence of clitics in this structurally deﬁned location. Syntactic accounts have generally attempted to argue that second-position clitics do indeed appear in some such uniform position in the structure of the

120

Aspects of the Theory of Clitics

clause. Phonology-based accounts make a similar claim. Halpern (1992b, 1995) argues that second-position clitics are structurally initial in their clause (and either come to be preceded by another constituent, or else undergo phonological, but not syntactic movement). Schütze (1994), within a syntactic account, argues that Serbo-Croatian second-position clitics appear in C. Others have advocated other functional heads, such as T. Boškovi´c argues that there is no consistent structural position which is occupied by Serbo-Croatian second-position clitics. He establishes this by showing that the assumption of such a position leads to a contradiction. First, he observes that in Serbo-Croatian, participles may precede (or follow) VP-adverbs. In contrast, sentence adverbs always precede (and never follow) a participle. These points are illustrated in (5.15). (5.15)

a. Jovan=je zaboravio potpuno Petra Jovan aux forgotten completely Petar Jovan forgot Petar completely Petra b. Jovan=je nesumnjivo istukao Jovan aux undoubtedly beaten Petar Jovan undoubtedly beat Petar c. *Jovan=je istukao nesumnjivo Petra Jovan aux beaten undoubtedly Petar *Jovan beat Petar undoubtedly

From this he concludes that sentence adverbs are located higher in the structure of the clause than VP adverbs; and that a participle can move from below the location of VP adverbs to a position between the two, but no higher. Consider what this evidence tells us about the position of clitics, assuming these occupy some speciﬁc structural position. The clitics in the examples of (5.15) are to the left of (and thus, ex hypothesi, higher than) participles and both kinds of adverb. This shows that the designated position for clitics must be located above that of sentence adverbs (or VP-Adverbs, a fortiori) and participles. But now compare the previous examples with (5.16). pravilno Mariji (5.16) Odgovorio=je answered aux correctly Marija only (He) answered Marija correctly not (He) was correct to answer Marija In terms of the location of the clitic je, nothing has changed, in at least one sense: like the clitics in the other Serbo-Croatian examples above, this one is in second position. The adverb pravilno ‘correctly’ is potentially interpretable as

Theories of Special Clitics

121

modifying either the VP or the entire sentence, but since in (5.16) the participle odgovorio ‘answered’ precedes it, we predict (correctly) that it must here be interpreted only as a VP adverbial. The participle also precedes the clitic, and thus, on the assumptions about how word order is related to syntactic structure being made here, it occupies a higher position in the tree than that in which clitics occur. But this contradicts our earlier conclusion that the highest position for participles is lower in the tree than that for clitics, at least if we assume that the position occupied by clitics is structurally uniform.7 Boškovi´c concludes from these and a variety of other facts that the syntax by itself cannot accomplish all that it has to. The notion of “second position” does not have a natural reconstruction in purely syntactic terms, because it does not correspond to any structurally unitary position. Nonetheless, exactly “second position” seems to be the notion that is relevant for placing clitics in Serbo-Croatian. Something other than a theory of the “Strong Syntax” sort must therefore be required. Within the spectrum of theories Boškovi´c considers, the remaining possibility is a “Weak Phonology” view, and it is such an analysis that he presents and defends. In support of the claim that the phonology is involved, he argues that the very domain within which clitics are placed in this language requires reference to the phonology. That is because various constructions that induce unusual phrasal organization also result in unusual clitic placement, as illustrated in (5.17). (5.17)

a. Sa Petrom Petrovi´cem srela =se samo Milena with Peter Petrovi´c met refl only Milena With Peter Petrovi´c, only Milena met b. Znaˇci da, kao što rekoh, oni =c´ e sutra do´ci means that as said they aux tomorrow arrive It means that, as I said, they will arrive tomorrow c. Ja, tvoja mama, obe´cala =sam =ti sladoled I your mother promised aux 2sg ice cream I, your mother, promised you an ice cream

Initial topics, inserted parentheticals and appositives all induce IntPhrase boundaries at their edges, and under these circumstances we ﬁnd that the notion of “second position” also shifts. Aside from providing further arguments against a purely syntactic view of clitic placement (for which the presence of 7

Laughren (2002) and Legate (forthcoming) provide arguments that second-position clitics in Warlpiri also do not occupy a structurally uniform position. Legate nevertheless proposes an analysis of the placement of these elements based on syntactic movement alone.

122

Aspects of the Theory of Clitics

these intonational elements ought not to matter), these facts suggest that clitics are located in second position within a domain that does not necessarily have a syntactically unitary status. They are also, of course, strongly reminiscent of the English constructions considered in Chapter 3, where special phrasing (with contrastive stress, parentheticals, gapping, etc.) affected the possible occurrence of clitic auxiliaries in ways that do not make sense in terms of the syntax alone. Boškovi´c concludes that the real generalization is that in (5.18). (5.18) Serbo-Croatian clitics occur in the second position of their intonational phrase. He proposes to achieve that effect by allowing the syntax to manipulate the order of words (including clitics) fairly freely, and then ﬁltering the result in terms of stipulated requirements governing the position of clitics in Phonetic Form. The statement in (5.18), roughly, is thus imposed as an output constraint, and any ordering of elements that includes clitics not conforming to it is blocked as ill-formed. Boškovi´c argues that the syntax must play a central role in this, because a purely prosodic deﬁnition of the initial anchor preceding the clitics does not sufﬁce. In particular, while word order in Serbo-Croatian is fairly free, there are some constructions that cannot be separated, and must move as syntactic units. When one of these occurs initially, the clitics must follow the entire structure, and not just the ﬁrst PWord. These include the structure in (5.14), and others such as those in (5.19). (5.19)

a. i. Tvoja majka i Petar =su otišli your mother and Petar aux left Your mother and Petar left ii. *Tvoja =su majka i Petar otišli b. i. Roditelji uspešnih studenata =su =se razišli parents successful students aux self dispersed Parents of successful students dispersed ii. *Roditelji =su =se uspešnih studenata razišli

Interestingly, speakers differ somewhat as to the separability of these constructions, and it appears to be the case that if and only if a given speaker allows one or another of them to be split up, that speaker also allows clitics to occur after the ﬁrst PWord of the structure, and not (only) after the entire syntactic unit. A similar fact can be cited with respect to sentences in which a two-part proper name occurs, such as Browne’s example (5.4b). In such DPs, inﬂection

Theories of Special Clitics

123

may occur either on only the last part of the name, or independently on both parts. The distinction is not evident in (5.4b), where the nominative has no overt inﬂectional marker, but can be seen in other cases. Crucially, if and only if both parts of the name show independent inﬂection, they can be formally separated—and it is exactly in this case that clitics can appear after the ﬁrst part of the name, as well. (5.20)

a. i. Lava Tolstoja cˇ itam Leo.acc Tolstoy.acc read.1sg Leo Tolstoy, I read ii. Lav Tolstoja cˇ itam Leo Tolstoy.acc read.1sg Leo Tolstoy, I read b. i. Lava =sam Tolstoja cˇ itala Leo.acc aux Tolstoy.acc read.ppart Leo Tolstoy, I read ii. *Lav =sam Tolstoja cˇ itala Leo aux Tolstoy.acc read.ppart Leo Tolstoy, I read

These facts appear to establish the conclusion that the syntax must play a determining role in characterizing the initial anchor for second-position clitics in Serbo-Croatian.8 On this picture, the fact that the initial anchor is typically (though not always) a PWord is not relevant. What matters is the fact that this anchor is also a syntactically autonomous unit. Combining these considerations with the evidence for the insufﬁciency of syntactic mechanisms alone and for a role played by the phonology, Boškovi´c concludes that the “Weak Phonology” position is strongly supported for this language. Whatever the correct analysis of Serbo-Croatian may be, we have already seen evidence that an account along these lines (with syntax controlling the order of elements, including clitics, and phonology playing a ﬁltering role to enforce a second-position constraint) cannot be maintained in general. In the Hittite examples in (5.1), for example, there is no warrant for the assumption that the anchors preceding the clitics are syntactic units that can be permuted into this position, since Hittite in general does not allow the kind of reordering characteristic of Serbo-Croatian. Similarly, as we have already seen in section 5.1 above, Chung (2003) argues explicitly that the anchor preceding clitics in Chamorro is deﬁned prosodically (as a PPhrase) and may not be coextensive 8 I will argue below, however, that the import of these facts about the syntactic separability of speciﬁc constructions in Serbo-Croatian is not what Boˇskovi´c claims.

124

Aspects of the Theory of Clitics

with any syntactically unitary constituent. We must conclude for the general case of second-position clitic placement that while the “Weak Phonology” position is an improvement over accounts based either on exclusively syntactic mechanisms or on Prosodic Inversion, it is still not adequate to cover the full range of second-position phenomena. Another class of arguments against the treatment of special clitics as syntactic elements is due to Géraldine Legendre. She shows that special clitic elements in Balkan languages are syntactically “inert.” For instance, in Macedonian (Legendre 1998) full verbs and auxiliaries invert with the subject in questions, but clitic auxiliaries do not, as illustrated in (5.21). (5.21)

a. Ti= ja= dade li Ana vaznata? 2sg-dat it gave-3 q Ana vase-def Did Ana give the vase to you? b. Go= ima li Ivan svrseno? it have-3 q Ivan solved Has Ivan solved it? ja= dade li c. Ke= ti= will 3sg-dat it gave-3 q Will Penka give you the book? d. Kade bi= ja= stavil ti where would it put you Where would you put the book?

Penka knigata? Penka book-def knigata? book-def

Furthermore, Macedonian clitics do not count as heads for the purpose of blocking Head Movement, as shown by the possibility of moving participial heads across them in the sentences in (5.22). (5.22)

a. Izpraznet =ke =e stanot vacated fut be-3 apartment-def The apartment will be vacated b. Proˇcel =sum knigata read have-1 book-def I have read the book

Clitics also do not induce strong crossover effects, as shown by the possibility of examples like (5.23). ˇ (5.23) Covetkot kogoi što goi= vidov [e]i i man-def whom that him saw-1 t the man whom I saw

Theories of Special Clitics

125

Similar arguments (and others) can be given from other Balkan languages (Legendre 1999, 2000a, 2000b, 2000c) to the effect that clitics do not behave in the way we would expect if they were heads (or phrases) in the syntax. This conﬁrms that it is probably incorrect to think of clitics as simply a special class of lexical items, whose location within the sentence is speciﬁed by the mechanisms of the syntax of words and phrases. Accounts that place special clitics in their surface position by purely syntactic means make other predictions that can be tested, as well. If the syntax is doing this work, then in cases where more than one special clitic appears in a sequence, the relative order within the clitic sequence ought to reﬂect some aspect(s) of syntactic structure, such as relative scope of corresponding functional categories. Real examples do not in general conform to this expectation, however. For instance, in French the preverbal clitics appear in the order given in (5.24). (5.24)

me, te, se, nous, vous < le, la, les < lui, leur < y < en

As noted by Perlmutter (1971), one of the ﬁrst discussions of clitic placement within the generative literature, this ordering does not reﬂect syntactic structure, and must be imposed by some other means (for Perlmutter, a template acting as a constraint on Surface Structure). In some languages, the sequence of clitics may show coherence in terms of morphological categories (e.g., ﬁrst person before second person before third person in Ngiyambaa (Donaldson 1980), regardless of grammatical relations) while largely disregarding syntactic roles. Hittite also shows a rigid template for the sequence of second-position clitics, as given in (5.25). (5.25) Sentence connectives < quotative (-wa(r)-) < dative/ accusative plural < 3rd person nominative, accusative singular < 1st, 2nd person dative/ accusative singular, 3rd person dative singular < reﬂexive (-z(a)-) < local, aspectual particles Here a variety of categories (including clitic type, person, number, and case) combine in a template that must be regarded as quite arbitrary from the point of view of the syntax. Yir-Yoront (Alpher 1991) moves even further from a syntactically based principle of clitic ordering. Clusters of pronominal clitics in this language have a ﬁrst-person enclitic ﬁrst (regardless of the grammatical relation to which this corresponds), followed by second- and third-person markers ordered according to a phonological principle, with “lighter” clitics preceding “heavier” ones.

126

Aspects of the Theory of Clitics

Such a principle is taken much further in the elaborate system of secondposition clitics in Sanskrit, if these can be shown (as argued by Stanley Insler in unpublished work) to follow a sequence entirely determined by a small set of purely phonological conditions. The relevant conditions also apply to the sequence of items in Sanskrit compounds, and are similar to conditions studied by Ross, Bolinger and others on preferred orders in English. In the case of Sanskrit clitics, these are chieﬂy “V-initial before C-initial” and “High V before Low V,” as illustrated by the Vedic lines in (5.26). (5.26)

a. dadik´ u carkir¯ ama 4.40.1 ¯rvna ´ıd u n´ ˙ an 10.12.5 b. mitrá´s cid hí s.ma juhur¯an.ó dev´ ¯

Schachter (1973) offers a comparable argument from Tagalog. In this language, monosyllabic clitics always precede polysyllabic ones, a distinction which cannot be related to any syntactically relevant dimension. I will discuss the Tagalog facts in more detail below in section 6.4. A wide variety of considerations thus force us to recognize a role for phonological (speciﬁcally, prosodic) considerations, along with morphological ones (such as person and number, as well as case) in the regularities of special clitic placement. Since we have also seen that purely syntactic accounts of these matters fail in a variety of ways, it is clear that we must look for some alternative—indeed, for a theory of a sort that does not fall entirely within any of the categories contemplated by Boškovi´c. In Chapter 6, I will develop the basic principles of such a theory.

6 An Optimal Theory of Clitic Positioning If the syntax is not the appropriate place to look for the principles of clitic placement, what alternatives are there? Clitics are apparently not just members of a special class of lexical items, elements whose location within the sentence is speciﬁed by the syntax of words and phrases. But if the syntax is not the appropriate place to look for the principles of clitic placement, what alternatives are there? In this chapter, I return to the suggestion of Chapter 4 that there are substantial similarities between special clitics and word-level morphology. Filling out that proposal, I propose an account of the positioning of these elements within the framework of Optimality Theory that brings out the connections between the two domains. I consider some special cases, particularly those that seem to argue for the existence of “endoclitics,” and then illustrate the workings of this theory in analyzing the special clitics of Tagalog.

6.1 Special Clitics as Phrasal Morphology The arguments of Chapter 5 would seem to establish a case for the notion that clitic ordering does not in general follow from syntactic principles. Rather it seems to be a surface, or PF, phenomenon, although a purely phonological account of the ways in which special clitics diverge in their positioning from the predictions of the syntax, based on Prosodic Inversion, does not succeed either. A phonological account would in any event be of no help in avoiding the consequences of arguments suggesting that special clitics are syntactically inert. The phonology can thus be excluded as providing the mechanisms necessary to locate clitics. But clitics do look a lot like morphology, as I suggested in section 4.2. Let us then explore the possibility that clitics are morphology: the morphology of phrases. On that view, the introduction of a clitic into

128

Aspects of the Theory of Clitics

its domain results from a rule of modiﬁcation operating on the phonological realization of that domain. Of course, most of these rules perform simple afﬁxation of ﬁxed phonological material, but in this respect they are entirely parallel to the rules of word-internal morphology, most of which are afﬁxal as well. Let us then accept the argument that clitics and word-level morphology are parallel phenomena, differing primarily in the scale of the unit (words versus phrases) that is affected. A consequence of this parallel is worth noting: treating clitics in this manner reinforces the decision to treat the placement of (non-lexical) inﬂectional afﬁxes as a PF phenomenon too. A rule of afﬁxation (or a rule specifying some non-concatenative modiﬁcation of the phonological form, such as mutation, stress shift, reduplication, etc.)1 speciﬁes what change should be made in the overall phonological representation, but it does not by itself tell us where that change should be made. For any particular clitic or word-level afﬁx, what is at issue is the location within a complete form at which a particular change takes place, regardless of whether that change consists in the introduction of new phonological material or in a modiﬁcation of material already present. An appropriate formalization of the rules of clitic placement remains to be settled on, but whatever mechanism is involved, it must involve a kind of non-syntactic formal modiﬁcation, something distinct from the syntax of independent minimal signs, and a speciﬁcation of where that modiﬁcation occurs.

A Rule-Based Theory One of the basic properties of clitics, as we saw, is the fact that they are introduced within some domain. Let us conﬁne our attention for now to what we will refer to as “inﬂectional” clitics, such as pronominals representing arguments of the lexical verb within a sentential domain, or the Tense/Aspect properties associated with such a domain, or determiner and/or possessor information associated with a nominal (DP) domain. Grossly, we might say that clitics are introduced by rules parallel to those that introduce afﬁxal material into words, reﬂecting the grammatical properties (e.g., case, number, agreement, etc.) of an inﬂected word.

1 In what follows here, I will often omit the fuller qualiﬁcations necessary to provide for nonconcatenative modiﬁcations when talking about “clitics” and “afﬁxes.” This is purely a matter of convenience and brevity of expression, and should not be interpreted as limiting the range of expression of morphosyntactic properties to simple afﬁxes unless this is made explicit.

An Optimal Theory of Clitic Positioning

129

Adopting the assumption that clitics are the overt realization of the grammatical properties—features—of the node dominating some appropriate domain, such as CP, IP, or DP, we can ask how these formal reﬂections of functional content can be made to appear in the correct relative positions in the surface form. If afﬁxes and clitics are introduced by individual rules of modiﬁcation, an obvious possibility is to specify the locus of afﬁxation directly as part of the rule that introduces the material in question, and then to relate the linear sequence of clitics to order of application of these rules. This is the proposal I pursued in Anderson (1992, 1993). There are, however, some formal and mechanical problems in implementing this account. One of these is the fact that sometimes the shape of a clitic depends on that of another clitic, where the most natural description would have the conditioning clitic introduced later in the descriptive sequence than the clitic whose shape it inﬂuences. For instance, in Serbo-Croatian, the femine singular accusative clitic =je is obligatorily replaced by the shape =ju when another clitic with the shape =je occurs after it, as shown in (6.1). (6.1)

a. Oni =su =je/*ju zaboravili they aux her forgotten They forgot her b. On =ju/*je =je zaboravio he her aux forgotten He forgot her

If the clitics are introduced one by one, in an order reﬂecting their leftto-right sequence, the conditioning environment for this change will not be available at the point at which the clitic undergoing it is to be added. Since the change in question is a matter of allomorphic variation in this speciﬁc element, rather than a general phonological rule of the language, it ought properly to be treated as an aspect of the rule that adds fem sg acc =je/ju). This appears to be an instance of the avoidance of two identical phonological shapes in sequence (similar to the replacement of expected si si by ci si in Italian, as Boškovi´c (2001: 103) observes). That intuition is not expressed, however, if the second =je is not present to trigger the change, which must then be conditioned by the set of morphosyntactic properties that will subsequently introduce the auxiliary =je. A similar issue arises with respect to clitic allomorphy in Italian. Here the clitics mi, ti, si are replaced by me, te, se immediately before ne, lo, la, li, le, and fem.dat. le is replaced by glie (Monachesi 1995: 43ff), as illustrated in (6.2).

130

Aspects of the Theory of Clitics

(6.2)

a. Martina te/*ti= lo= spedirà Martina 2sg.dat 3msg.acc will.send Martina will send it to you b. Martina me/*mi= ne spedirà molti Martina 1sg.dat part will.send many Martina will send me many of them c. Glie-/*Le= le= ho date 3fsg.dat 3pf.acc I.have given I have given them to her(/him)

This may well have had a phonological basis at some point in the history of the language, but it is now a matter of idiosyncratic allomorphy that ought to be expressed as a contingency in the choice of inserted material. Such an account is not available, however, if clitics to the right of a given element in the sequence are not introduced until after that element. These problems would disappear if the clitics were introduced in right-toleft order, instead of left-to-right. But Italian also illustrates an allomorphic dependency in the opposite direction. In the literary language, the locative clitic ci= can also be realized as vi=. However, if the locative clitic is preceded by an object clitic, it can only have the form ci=, as shown in (6.3). (6.3)

ci/*vi= porta Mi/Ti/Vi= 1sg.acc/2sg.acc/2pl.acc loc takes He takes me/you(sg)/you(pl) there

Such examples suggest that introduction of clitics one at a time, in a sequence intended to determine their order in the surface form, will run into logistical difﬁculties. Instead, it seems that several clitics forming a unitary cluster need to be introduced simultaneously in a way that would potentially allow each to be sensitive to the properties of the others. A rather different set of problems for this approach to clitic positioning results from the fact that the linear order of a sequence of clitics may remain the same regardless of whether the sequence as a whole precedes or follows some anchor point. Thus, in Italian, clitics precede the ﬁnite verb, but follow imperatives. What is important to note is that the sequence is the same in both cases, despite the fact that the cluster of clitics appears on opposite sides of the anchoring verb. (6.4)

a. Me= lo= dice 1sg.dat 3sg.acc you.tell You tell me it

An Optimal Theory of Clitic Positioning

131

b. Dim=me=lo tell.imp-1sg.dat-3sg.acc Tell me it! In Macedonian, parallel to Italian, clitics appear in the same sequence before ﬁnite verbs and after non-ﬁnite verbs as exempliﬁed in (6.5). (6.5)

a. Ne =bi =mi =go dal neg cond 1sg.dat 3sg.acc gives He wouldn’t give it to me b. Dajte =mi =go give.imper 1sg.dat 3sg.acc Give it to me! ´ c. nosejki =mi =go. . . bring.Pres.Part 1sg.dat 3sg.acc bringing it to me. . .

The facts in Bulgarian are nearly the same as in Macedonian, with the exception that clitics follow a sentence-initial ﬁnite verb as well as all non-ﬁnite verb forms. This is illustrated in (6.6). (6.6)

a. Ivanˇco =mi =go Ivancho 1sg.dat 3sg.acc Ivancho showed it to me b. Pokaza =mi =go showed 1sg.dat 3sg.acc Ivancho showed it to me

pokaza showed Ivanˇco Ivancho

Why does this pose a problem for a theory that introduces special clitics one by one by rule, and derives their linear sequence from the order of application of these rules? The problem is that when the same set of clitics can appear either before or after their anchoring element, we would expect the sequences in the two cases to be mirror images, not the same. If the same set of rules introduces these clitics, differing only in the before/after parameter, the clitic closest to the anchor in each case ought to be the same, yielding reverse orders in the two cases. This situation, which would be expected at least to be the unmarked option, is virtually never found,2 and this requires an explanation. This fact is noted by Kayne (1991), who accounts for the fact that clitics appear in the same order regardless of whether they precede or follow their 2 The most widely cited exception to this is the order of clitics after French imperatives (e.g., Donne-le moi! ‘Give it to me!’), which is the mirror image of the order found before the ﬁnite verb (Tu me le donnes ‘You give it to me’). At the very least, this pattern is extremely rare; but it may not even be relevant, since it is not obvious that the second pronoun (moi) in these imperative forms is a special clitic comparable to the pre-verbal form (me).

132

Aspects of the Theory of Clitics

anchoring verb by not placing them individually with respect to that verb at all. Rather, Kayne attaches the clitics to an empty functional category and then moves that category to a position either before or after the verb, depending on properties such as ﬁniteness. In at least some cases, however, this abstract analysis will be difﬁcult to maintain. Legendre (1998, 2000a) develops arguments for Bulgarian and Macedonian similar to those provided by Boškovi´c for Serbo-Croatian, showing that there is no uniform category to which the clitics could be attached in implementing this analysis. So where do we stand with respect to the placement of second-position clitics? I have argued that we want a theory that can accommodate both clitics and afﬁxes as the phrasal and word-level instantiations of very similar mechanisms, because the positional possibilities are so similar in the two cases. In the case of clitics, we need to be able to deﬁne “second position” with reference either to a prosodic unit (a PWord, PPhrase, etc.) or to a syntactic unit (a phrasal constituent) at the beginning of the relevant domain. Ascribing this placement entirely to syntactic movement is apparently excluded on several grounds, as we have seen. In earlier work (Anderson 1992, 1993), I explored the alternative of introducing clitics by rules of modiﬁcation of the phonological form of the phrase, where the relative ordering of clitics is speciﬁed as relative ordering of the clitic introduction rules. But this approach also has signiﬁcant problems, and so it is necessary to seek an alternative. A Constraint-Based Theory An alternative view is provided by Optimality Theory (“OT”: Prince and Smolensky 1993).3 To see this, let us note that a single “rule” of the sort considered in the previous subsection actually does two things. First, it introduces some change in the shape of the form (typically, the addition of a chunk of phonological material). In addition, the formulation of the rule speciﬁes just where that change is to take place. The primary function of the rule is to provide an exponent for some aspect of linguistic content, and secondarily to determine exactly how a complex form is to be structured. Separating the two yields a potential improvement in the clarity with which each can be expressed. Optimality Theory can provide us with a way to describe the placement of special clitics if we invoke a constraint system to describe the way a complex input should be realized overtly. That input consists of a basic form as supplied As in Chapter 3, I assume that the overall workings of constraint-based theories like OT are familiar, and do not attempt to provide a general introduction here. The reader whose needs are thereby left unmet can consult general works such as Kager (1999), or more speciﬁcally for applications to morphosyntax, the papers in Barbosa, Fox, Hagstrom, McGinnis, and Pesetsky (1998) and Legendre, Vikner, and Grimshaw (2001). 3

An Optimal Theory of Clitic Positioning

133

in a lexically interpreted syntactic representation, together with an indication of one or more changes (phrasal afﬁxes to be introduced or non-segmental alterations to be made) that should affect the form to reﬂect associated morphosyntactic properties. I assume that this input is constructed in one of two ways. First is the case of special clitics representing “grammatical” material, such as pronominals identifying an argument, tense and aspect markers, determiners, etc. Here I assume that the input to rules of phrasal afﬁxation is a pair consisting of the syntactic representation and its phonological shape as this has been developed by other aspects of the grammar (particularly lexical interpretation). The structural description of such a rule identiﬁes features in the complete morphosyntactic representation of the syntactic phrase that is the clitics’ domain. These will include much of what is often represented as the content of separate functional categories, here presumed to be represented as part of a complex symbol. For discussion of these matters at the word level, see the discussion of morphosyntactic representations in Anderson (1992); further elaboration of these notions will be provided in Chapter 8. The effect of the rule is then to specify additional phonological content that is to be associated with the phonological representation of the phrase. The result (possibly involving several such components of added material) is then submitted to the constraint system, which locates the modiﬁcations within the form, and effects any necessary phonological adjustments that may be necessary in light of the additional material. Such adjustments will include the effects of “Stray Adjunction,” for instance, as well as other changes such as assimilations, etc. The second case involves special clitics that do not reﬂect grammatical or functional material, but which rather have semantic and/or pragmatic content of their own. We will see some examples in the discussion of Tagalog below in section 6.4; such elements are often (and unilluminatingly) called “particles” in grammatical descriptions. They constitute the special clitic analog of derivational morphology, while the grammatical clitics mentioned above correspond more or less to inﬂectional morphology. Again, the discussion of Tagalog below will suggest that this parallel is not entirely without force. Special clitic rules of this “derivational” sort take as their input a syntactic structure together with both its phonological and its semantic interpretation. The rule then performs simultaneous modiﬁcations in these. On the one hand, some phonological content is associated with that of the existing phrase, just as in the “inﬂectional” case. On the other hand, a correlated change is also effected in one or both of the syntactic and semantic representations. A special clitic may convert a sentential structure to a nominalized form, for example (see Kaiser 1998 for some examples), by affecting its syntactic categorization.

134

Aspects of the Theory of Clitics

Somewhat more commonly, perhaps, the associated change is semantic, such as the introduction of a negation operator, adverbial material, etc. The output of one or more operations of phrasal afﬁxation then creates a representation consisting of several components, and this is then submitted to the constraint system to determine its phonological realization. It is this constraint system which is responsible for determining the placement, as opposed to the presence, of special clitics, as well as their integration into an overall phonological shape. For present purposes, the immediate advantage of this approach as opposed to the rule-based view comes from the fact that in OT, several of the afﬁxes in a word or the clitics in a phrase can be treated as simultaneously introduced, and thus essentially copresent. Descriptive order can reﬂect the simultaneously evaluated relative ranking of element-speciﬁc constraints, rather than a sequence by which the elements are introduced. Several clitics may all need to be located at the same point in the course of developing a given form, and a single competition among alternative possibilities settles them all simultaneously and with reference to one another. We will see other advantages of the constraint-based view as we proceed to develop it. The possibility of simultaneous evaluation of constraints on the positioning of multiple clitics need not mean that all of the clitics in an utterance, or all of the afﬁxes in a word, are always introduced in a single step. This complete parallelism is, it is true, a common assumption in the OT literature, but it is one that is logically separable from other aspects of a constraint-based approach. Indeed, Kiparsky (2000, forthcoming) has developed a constraintbased version of Lexical Phonology in which serial, cyclically organized derivations play a role, with each stage of phonological adjustment expressed by a system of constraints. I follow essentially the same approach, and will note below where the standard strictly parallel view is not maintained. Nonetheless, in many cases, I assume that multiple adjustments are made to the same structure in a single step. This is true, in particular, where a number of clitics cluster in ways that interact with one another. Word-Level Afﬁxation Let us begin to develop the required theory by discussing the word-level case. How does a constraint-based view describe the placement of afﬁxes? Relative ordering results from the fact that a number of afﬁxes are all constrained to be located in the same position. These constraints are “soft” or violable, however, and the demands of some afﬁxes will outweigh those of others. Suppose a number of afﬁxes are all constrained to be preﬁxes. That means that for each of them, there is a constraint to the effect that it should appear at the left edge of the word. Since these constraints are strictly ranked, however,

An Optimal Theory of Clitic Positioning

135

one of them will outrank the others; and of the remaining constraints, there will again be one that is next highest ranked, etc. In the optimal realization of a given form, the afﬁx corresponding to the constraint with the highest rank will actually be an initial preﬁx; the next highest ranked will occupy a position which violates minimally its requirement of being initial (i.e., as the second preﬁx), etc. Exactly similar considerations apply, mutatis mutandis, for collections of sufﬁxes. To produce the result that preﬁxes or sufﬁxes appear in a particular order with respect to one another (and to the left edge of the word), we invoke members of a family of constraints EdgeMost(e, L/R, D) (where e is some linguistic element such as an afﬁx or clitic, and L/R refers to the left or right edge of the domain D within which e appears). These “EdgeMost” constraints are clearly a subset of what are more commonly referred to in the OT literature as “alignment” constraints. Thus, the interpretation of EdgeMost(Afi , L, D) is the same as a constraint saying that the left edge of the afﬁx Afi should be aligned with the left edge of some including category D, such as the PWord or the Lexical Word. I use a somewhat simpler formulation here simply for convenience, and indeed I will often omit speciﬁcation of the Domain when this is obvious. A constraint-requiring element e to appear at the left edge (of D) will generally be expressed below as LeftMost(e(, D)); each such constraint says that the element e should appear as close to the left edge (of D) as possible. An analogous constraint family RightMost(e(, D)) says that (the right edge of) e should appear at the right edge of D. A given afﬁx Afi is characterized as a preﬁx or as a sufﬁx by whether LeftMost(Afi ,D) dominates RightMost(Afi ,D) or vice versa. The descriptive order of Afi and Afj is determined by the dominance relation that obtains between their corresponding LeftMost (or RightMost) constraints relative to D. Now consider the way inﬁxes are treated in OT, a point on which a substantial literature is in essential agreement. A post-initial inﬁx is generally described by treating it as a type of preﬁx, and thus subject to a high-ranking LeftMost constraint. The location of this afﬁx in initial position, however, is prevented by some other higher-ranking constraint C that keeps it from being absolutely initial. The optimal form resolves this conﬂict by locating the afﬁx as close to the left edge as possible, consistent with the requirements of C . The minimum possible violation of LeftMost that avoids violation of C thus positions the element as a (post-initial) inﬁx. The nature of the higher-ranking constraint C that forces a would-be preﬁx to be located as an inﬁx is clearly crucial to this account. In some wordlevel cases, C may be a phonological requirement on permissible syllable structures. As an example, the Sundanese plural inﬁx –ar–/–al– immediately follows

136

Aspects of the Theory of Clitics

a word-initial consonant. This is arguably due to high-ranking constraints that require syllables to have onsets while deprecating codas. When –ar–/–al– is added to niis ‘cools self ’, the afﬁx is constrained to be LeftMost; but if it were actually to occur at the left edge of the word, the result would be *arniis, whose initial syllable (a) has no onset, and (b) has a coda. Locating the inﬁx after the ﬁrst consonant avoids violations both of Onset and of *Coda, at the cost of violating LeftMost(ar/l) minimally. Given the ranking of the syllable-structure constraints above the alignment of the afﬁx, the optimal result is thus nariis ‘cool selves’. In the Sundanese case, given the structure of the language and of stems, we could attribute the inﬁxation to the operation either of Onset or of *Coda: the positioning of inﬁxes such as –ar/al– and –um– after an initial consonant avoids both types of violations equally while violating LeftMost minimally. In other instances, however, it is possible to be more speciﬁc. Consider the case of Chamorro, where some stems begin with an initial cluster. When inﬁxed, for example, tristi becomes trumisti, trinisti. The fully preﬁxed forms *umtristi, *intristi clearly violate both Onset and *Coda. If we were to position the inﬁx after the ﬁrst consonant, yielding *tumristi, *tinristi, we would have repaired the violations of Onset, but not those of *Coda. The actual forms avoid violating *Coda at the cost of a greater degree of violation of LeftMost, suggesting that it is the coda constraint rather than the requirement of onsets that is responsible for the inﬁxed position of “LeftMost” VC afﬁxes in this language. Both in Chamorro and in Sundanese, the conditions forcing inﬁxation may plausibly be seen as purely phonological. Most of the examples of inﬁxation that have been treated in the recent literature have a similar basis in syllable structure or related phenomena, but in other cases, it is hard to see how the placement of an inﬁx has any effect in improving the syllable structure of the resulting form. A number of Austro-Asiatic languages, for instance, make extensive use of inﬁxes that are placed immediately after the ﬁrst consonant of a word. Thus in KamhmuP4 we ﬁnd an inﬁx /–rn–/ ‘instrument’ in forms such as those of (6.7). When the stem begins with a consonant cluster, the enlarged cluster resulting from inﬁxing /-rn-/ is simpliﬁed or otherwise altered. (6.7)

a. hrniip ‘spoon’; cf. hiip ‘eat with spoon’ b. crnok ‘gouging instrument’; cf. cok ‘to gouge’

4 This is a language of Vietnam, for which the following facts are taken from Merriﬁeld, Naish, Resch, and Story (1965) where they are attributed to William Smalley. Much more information on this language is available in Svantesson (1983). Similar facts can be attested from a variety of related languages, such as Atayal, Semai, Temiar, and Khasi.

An Optimal Theory of Clitic Positioning

137

c. krlam ‘pole carried on the shoulders’; cf. klam ‘carry suspended from a pole’ d. crmool ‘dibble stick’; cf. cmool ‘make a hole with a dibble stick’ e. cndrieh ‘comb’; cf. crieh ‘to comb’). It is clear that the placement of the inﬁx is “following the ﬁrst consonant”. KamhmuP has no vowel-initial stems, so we cannot see what the result would be of adding rn to such a form. Nonetheless, there is no obvious sense in which inﬁxation is preferable to plain preﬁxation in the forms of (6.7). To achieve this, we might assume another constraint family NonInitial(e,D), each of whose members expresses the requirement that the element e should not be initial within the domain D. To characterize an afﬁx Afi as “second position” on this view, we would then say that NonInitial(Afi , D) dominates LeftMost(Afi , D). In the optimal form, then, the afﬁx will appear in a position as far to the left as possible without actually becoming initial: that is, it will be located in second position. This is the way similar facts concerning clitics were treated in Anderson (2000c) and some related papers, but it is certainly worth inquiring whether there is a more motivated way of achieving inﬁxation in such cases. In fact, such an analysis has already been preﬁgured by some of the discussion in Chapter 3. I observed there that some instances of Stray Adjunction seem to operate in a way that suggests a preference within a given language for the left (or potentially, the right) edges of certain constituent types to coincide with lexical, rather than grammatical material. We might invoke the same general principle here, at the word level, if we supposed that in KamhmuP the left edge of the grammatical word is preferentially aligned with lexical rather than grammatical material. A constraint such as (6.8) would have this effect. (6.8) Align(GrWord, L, LexWord, L) If (6.8) is ranked higher than the requirement that rn should be placed at the left edge of the word, the optimal form will be one that violates left alignment of rn by a minimal amount—a single segment. This account has a disadvantage, however, from the point of view of theories of morphology such as that of Anderson (1992) or Stump (2001), since it presumes that the component parts of a morphologically complex word are visible in the grammar. This can be avoided, though, if we replace (6.8) with a faithfulness requirement. The desired effect can be achieved if we rank highly (in the grammar of KamhmuP and similar languages) a constraint (6.9), to the effect that the left edge of the input word must correspond to the left edge of the output word.

138

Aspects of the Theory of Clitics

(6.9)

LeftEdgeFaith(Word): The element at the left edge (of the word) in the output form should correspond to the element at the left edge in the input.

Such a constraint is presumably motivated by the fact that lexical identiﬁcation in many languages is highly sensitive to the beginnings of words. Requiring that the left edge be conserved under afﬁxation can be seen as supporting this preference. And if we rank (6.9) above LeftMost(rn, Word), this afﬁx will once again be placed after exactly a single segment at the beginning of the word in the optimal form. There may be instances of inﬁxation which must be achieved through a constraint such as NonInitial(e, D), but we would like to avoid that where possible. In the case of word-level afﬁxation, that can generally be achieved either by invoking phonological factors or by appeal to (6.9). Phrase-Level Afﬁxation (Clitics) The extension of this apparatus to at least some classes of special clitics is obvious. Phrase-initial and phrase-ﬁnal clitics are directly analogous to preﬁxes and sufﬁxes, respectively. The description of their placement involves constraints from the same family, with clitic elements speciﬁed as being attracted to the left or right edge of an appropriate domain. The placement of second-position clitics, however, requires additional discussion. Recall that second-position clitics were claimed in section 4.2 to be the phrase-level analogs of post-initial inﬁxes at the word level. It is natural, then, to describe them in comparable terms. Analogous to the introduction of morphological afﬁxes on the basis of morphosyntactic properties of a word,5 a rule of special clitic formation speciﬁes material that should be inserted into the phonological form corresponding to the phrasal domain within which the clitics are motivated. The location within this domain where the speciﬁed material appears is governed primarily by a constraint of the LeftMost family, with some other constraint C intervening to prevent them from occurring in absolute phrase-initial position. When we ask what that other constraint might be, the answer is not as immediately obvious as in the word-level case. There are no known examples, for instance, in which the placement of clitics in second position can 5 Derivational morphology is introduced similarly, as afﬁxal material associated with the operation of a Word Formation Rule. The positioning of this material follows from a system of constraints just as in the inﬂectional case. Since many core cases of special clitics are similar in interesting ways to productive inﬂectional morphology, I will discuss the general phenomenon in those terms, for concreteness’ sake, but the same mechanisms apply for more “derivational” material, both word level and phrase level.

An Optimal Theory of Clitic Positioning

139

be motivated by considerations of syllable structure. Nonetheless, there are some instances for which a phonologically driven account has been suggested, and where such an analysis has at least an initial appeal. One of these is the system of second-position clitics in Warlpiri. In this language, a cluster of clitics appears typically in second position within the clause, composed of an auxiliary base (possibly ∅) followed by one or more pronominal clitics. Word order in Warlpiri is generally quite free; the ﬁxed positioning of the clitic cluster is striking in contrast, and suggests that mechanisms other than those of the language’s basic syntax must be at work. As formulated by Hale (1973), auxiliary bases can be either monosyllabic (or null), or bisyllabic, and the clitic cluster can appear either in second position or initially. He proposes a principle governing the placement of the clitic cluster as follows: when the base of the auxiliary is monosyllabic (or ∅), the auxiliary follows the ﬁrst constituent of the sentence. When the base of the auxiliary is bisyllabic, however, the auxiliary can appear either initially or in second position. This difference is illustrated in (6.10). (6.10)

a. Wawiri =ka=rna purra-mi kangaroo PresImpf-1sgSbj cook-NPast I am cooking the kangaroo b. Kapi=rna wawiri purra-mi Fut-1sgSbj kangaroo cook-NPast I will cook the kangaroo

The pronominal clitics appear immediately following the base, wherever that is. The role of phonological form in this principle suggests strongly that the phonology should play a part in any account of it. For the case of null or monosyllabic bases, we can note that the “minimal word” in Warlpiri, as in many languages, is bisyllabic. A monosyllabic element presumably cannot be independently footed, and thus must be treated as prosodically deﬁcient. I assume that all of the pronominal clitic elements in Warlpiri are non-lexical, and thus not analyzed as PWords, even if they contain more than a single syllable (e.g., -pala ‘non-1st dual’). Let us assume further that stray adjunction in Warlpiri always operates so as to incorporate prosodically deﬁcient material leftward, and never to the right into an existing PWord (or other prosodic constituent). In order to be incorporated into prosodic structure, then, a monosyllabic auxiliary base will have to be preceded by some other material. If we say that auxiliary bases are subject to (highly ranked) LeftMost(cli ), the furthest to the

140

Aspects of the Theory of Clitics

left that they can go is after the ﬁrst constituent,6 if they are to be prosodically well-formed. That accounts for their appearance in second position. This account is supported by a remark Simpson (1991) makes, to the effect that even short bases can be initial if the clause is closely linked phonologically with a preceding one. No example is provided in that source, but Jane Simpson (personal communication) has kindly provided one from the electronic ﬁles of Ken Hale’s original texts, which is reproduced here as (6.11). The monosyllabic auxiliary base which appears in clause-initial position is underlined. Ngulanya ka-rnalu ngarri-ni (6.11) Jinarn-kiji-ni-ji. trip-throw-npast-euph. that pres-1pl.ex.sbj call-npast jinarn-kiji-ni-ji, kaji-lpa-npa watiya-rla-rlangu trip-throw-npast-euph, if-past-2sg.sbj log-loc-for.example wanti-yarla. ka jinarn-kiji-ni. fall-irr, pres trip-throw-npast “Jinarn-kijirni” ‘to cause to fall’. We call it “jinarn-kijirni” if you fall over on, say, a piece of wood, it trips one up. (Hale 1966 notes [0125]) A bisyllabic base, unlike shorter ones, contains enough material to constitute a foot (and thus a PWord) on its own. Suppose we say that these bases have two variants: one whose foot is structured as a PWord, and one where it is not a PWord. If we choose the prosodically autonomous form, then LeftMost(cli ) will locate it in initial position. If we choose the deﬁcient alternant, LeftMost(cli ) cannot be fully satisﬁed, since that would yield a prosodically ill-formed structure due to the lack of a host on the left for the stray material. The result will be placement in second position, just as for short bases, unless preceding material from another sentence is interpreted as prosodically “close” enough to constitute a host, as in (6.11). Attractive as such an account is (and essentially this picture has been widely cited in the literature), Legate (forthcoming) shows that it is based on a mis-analysis. The bisyllabic elements such as kapi in (6.10) which Hale treated as Auxiliaries are not members of that class, but rather of the set of complementizers. These latter appear optionally either in ﬁrst or in second position, and their placement is without regard to number of syllables: the complementizer yi ‘for, since’ is monosyllabic, but shows the same (limited) freedom of placement as bisyllabic members of the class. Furthermore, even when occurring in second position, the complementizers show behavior typical of an autonomous PWord, rather than a prosodically deﬁcient clitic, 6 The issue of exactly how much material must constitute the “ﬁrst” element will be addressed below in section 6.2.

An Optimal Theory of Clitic Positioning

141

as observed earlier in section 4.1. Whatever the basis of this optionality, it must be grammatical rather than phonological. While there may be languages for which a phonological account of the non-occurrence of clitics in phrase-initial position is possible, no well supported analyses of this kind have been presented. And in fact it is extremely unlikely that such a prosodically based account will be adequate in general. That is because some special clitics that must be placed post-initially are not prosodically deﬁcient. Tagalog, for instance, has a huge system of clitics, most of which are prosodically autonomous and bear their own stress. There seems no phonological reason why these could not occur initially, and if they do not, that fact must be due to some other constraint. Two possibilities suggest themselves, and these will be examined below. Second-position placement might be treated as an alignment phenomenon, comparable to the analysis of post-initial inﬁxes such as KamhmuP -rn- above. We might, for instance, invoke a preference for the left edge of the syntactic phrase that constitutes the clitic’s domain to correspond to the left edge of the prosodic phrase that realizes it. On the theory being developed here, the clitic is not itself a syntactic constituent of its phrasal domain, but rather a phonological marker of some feature(s) associated with the domain. As a result, if the clitic itself appears in phrase-initial position, it will produce a violation of this alignment requirement. In Bulgarian, for instance, a few sentential clitics occur initially (ne, šti), although most do not. It is not hard to show that Stray Adjunction works in both directions in this language, aligning syntactic and prosodic boundaries where possible. The difference between ne, šti, and other clitics can be ascribed to the fact that while all are subject to LeftMost(cli ), the LeftMost requirements of ne and šti dominate the alignment constraint (which we might treat as LeftEdgeFaith(CP) or the like), while those of the others do not. In the absence of reasons to treat post-initial positioning of clitics as motivated by faithful alignment, we can still achieve the required positioning by treating the clitics in question as subject to a descriptive requirement that, although anchored by the left edge of their phrase, they must nevertheless not occur in absolute initial position. A constraint NonInitial(cli ), that is, dominates the otherwise highly ranked LeftMost(cli ). In such a case, we can say that the placement of the clitics in second position represents a basic descriptive generalization, one not reducible to the effects of some other requirement(s). The constraint family NonInitial(e), where we need to invoke it for the case of clitic positioning, is well established in other areas. In particular, in the analysis of stress and accent, NonInitial(e) and the mirror-image constraint

142

Aspects of the Theory of Clitics

type NonFinal(e) play much the same role in constraint-based analyses that Extrametricality (Hayes 1982) had in earlier, rule-based descriptions. Second-position clitics and post-initial inﬁxes are relatively common in the languages of the world, but we have already noted that “penultimate position” is at best rarely instantiated in the case of pre-ﬁnal inﬁxes and in the corresponding clitic type. Whatever the role of NonInitial(e) and LeftEdgeFaith(P) in the description of morphological and phrasal afﬁxes, their mirror images Non-Final(e) and RightEdgeFaith(P) are much less prominent, if indeed they exist at all.7 It is possible to suggest as a motivation for this asymmetry that the identiﬁcation (and stability) of left edges is important in itself for morphosyntactic parsing, while no corresponding signiﬁcance is attached to right edges. On this picture, the apparent symmetry of the parameters for clitic position that arises on views such as those of Klavans (1985) and Anderson (1992) is less pervasive than it initially appears—a conclusion which seems to be warranted on empirical grounds.

6.2 Second Position: Anchors and Domains At this point, we have the beginnings of a more “morphological” theory of special clitics. Such a theory is based on mechanisms that modify the basic phonological realization of a syntactic phrase by introducing new segmental material (the simplest kind of “clitic”), or by making other modiﬁcations of the sort discussed in section 4.2, as a reﬂection of some of the morphosyntactic content of the phrase. The location of the phonological modiﬁcations thus introduced is determined in a way entirely parallel to the way morphological afﬁxation (whether concatenative or not) is positioned: through a series of constraints, specifying preferred alignment for the elements in question in an interaction with other inﬂuences. There are still many aspects of this theory that remain to be made precise, however. In this section, I address two of these: the speciﬁcation of the “ﬁrst” element with respect to which a clitic may appear in second position, and the speciﬁcation of the domain within which the placement of a given clitic may be constrained. Anchors for Clitics A question that was ﬁnessed in the ﬁrst approximation to a constraint-based theory of clitic placement above is that of how we ought to deﬁne “second position.” That is, what can appear in the ﬁrst position within the same domain? How are we to describe just how much material appears to the left of a 7 We will see below that NonFinal(e) does have a role to play in the description of clitics, at least with respect to prosodically deﬁned domains.

An Optimal Theory of Clitic Positioning

143

second-position clitic? In the case of inﬁxation within the word, the inﬁx either appears in a position determined by prosodic constraints on syllable structure within the word, or following some prosodically deﬁned unit, such as after the ﬁrst segment, syllable, or foot. We certainly do not want clitic positioning to be determined in the same way, since the result would in general be to locate the clitic somewhere within the initial word, and it is virtually never the case that clitics appear in a position where they are properly included within a preexisting PWord.8 To accommodate this observation, let us assume that in the derivation of the form of phrases, there is an overall constraint to the effect that material cannot be inserted inside a PWord. Perhaps new material can be adjoined to a PWord at an edge, but an existing word can not be interrupted. Let us call this constraint Integrity(Word). In virtually all languages, this constraint is undominated. Now consider the case of a clitic cli subject to LeftMost(cli ), where some other constraint prevents cli from appearing in absolute phrase-initial position. As a result of high-ranking Integrity(Word), there will always be at least one phonological word between the second position occupied by cli and the left edge of the phrase, since the only way there could be less would be by violating Integrity(Word); or else by violating the other constraint NonInitial(cli ),9 which (ex hypothesi) dominates LeftMost(cli ). But given LeftMost(cli ), the only way there could be more than one word before the clitic would be if some other higher-ranking constraint required it. This apparatus is all we need to describe the speciﬁc case in which second position is interpreted as “following the ﬁrst word of the phrase,” as in Hittite (and perhaps others of the ancient Indo-European languages, as Wackernagel 1892 assumed) or West Greenlandic. For the somewhat commoner cases in which second position means “after the ﬁrst syntactic phrasal daughter of the domain-deﬁning phrase,” however, something else is clearly necessary to force placement even farther from the left edge than the position after a single PWord. To describe this circumstance, we can posit another related constraint family: Integrity(XP), which prohibits a phrase from (properly) containing elements that are not themselves members of that phrase. Note that this will allow clitics to appear within the largest containing phrase deﬁning the entire domain motivating the introduction of the clitic, 8 The small number of cases of so-called “endo-clitics” that have been discussed in the literature will be addressed below. 9 For convenience, I represent whatever constraint forces second position by NonInitial(cl ), i although (as mentioned above) this is only one of the possibilities that might force such placement. This choice is not intended as a claim about the actual constraint that might be involved.

144

Aspects of the Theory of Clitics

since they reﬂect properties of that phrase. It does, however, mean that they can not appear within another phrase which is itself contained within that domain. Since syntactic phrases come in many varieties, it seems plausible that Integrity(XP) is actually a family of constraints, such that in some languages some phrasal types might be more “permeable” than others. In his careful study of Serbo-Croatian word order, for instance, Boškovi´c (2001), following much earlier literature, notes that while subparts of most phrasal types can be separated, there are a few that cannot. As noted above in section 5.2, these include certain prepositional phrases (at least those with the preposition prema); two-part proper names with overt case marking only on one member; and nominals including a governed genitive expression. Phrases of these types must always remain together as units, despite the general freedom of word order in the language. Boškovi´c takes these facts to be an argument for a theory that derives clitic positioning via syntactic movement. He proposes that the generalization concerning ﬁrst position is that exactly one syntactically autonomous constituent can appear before the second-position clitics. That may be a multi-word constituent, but for many speakers of “eastern” or “Croatian” dialects, it can also be a single word—if and only if that word is capable of “scrambling” in a way independent of other words with which it forms a constituent. Syntactically autonomous movement thus correlates with the possibility of occupying ﬁrst position; and Boškovi´c argues that this correlation goes unexpressed in a theory like the present one, where the set of constituents that must group together in ﬁrst position is described by constraints of the Integrity family, rather than by syntactic movement. This objection does not hold, however. It is true that the set of constituent types which cannot be interrupted by clitics is described, on the view being developed here, by some collection of constraints of the type Integrity(P) for speciﬁed phrasal types P. But in fact, exactly the same constraints can be taken to be responsible for the facts Boškovi´c attributes to the syntax. That is, we can assume that scrambling is (from the point of view of the syntax, though not, presumably, with regard to semantics and discourse structure) a matter of free re-ordering. Some constituents, however, cannot be broken up in the process—and it is precisely because these are subject to highly ranked Integrity(P) that they are impermeable, both to clitic placement and to interruption as a consequence of scrambling. Whatever generalizations there may be about the set of constituent types that must remain intact under scrambling, exactly the same descriptive apparatus that accounts for this will simultaneously account for their behavior with respect to clitic placement. There is thus

An Optimal Theory of Clitic Positioning

145

no lost generalization, and no concomitant argument against the present line of analysis. I assume, then, that XP in Integrity(XP) is parametrically variable. Integrity(Word) and Integrity(XP), obviously, are instances of the same family: constraints to the effect that material cannot be entirely contained within a domain of a certain sort unless it represents a member or element of that domain. Now suppose that the integrity constraints (including Integrity(P) for various values of P, like Integrity(Word)) are generally undominated. In that case, the earliest that “second position” can come is after the ﬁrst phrasal daughter of a type P in the containing phrase, and so that is exactly where second-position clitics will be found. Where Integrity(XP) is lower ranked than various relevant alignment constraints (but Integrity(Word) is still highly ranked), “second position” will mean “after the ﬁrst word.” Variation in the interpretation of “second position,” accordingly, is a matter of variable ranking of parameterized integrity constraints with respect to the rest of the system. Constraint Domains for Clitic Placement On the account being developed here, special clitics are introduced into the phonological representation of a syntactic unit on the basis of properties of that unit (phrase, clause). Alignment constraints inﬂuence the positioning of the clitic within the form by preferring realizations at one edge or the other (LeftMost(e), RightMost(e)), representing the Anchoring parameter of earlier descriptive theories. When the clitic is not positioned at the actual periphery of the form (in particular, when it is located in second position) this is due to some higher-ranking constraint: perhaps a requirement that the left edge of the output form correspond to the left edge of the input; or an alignment constraint specifying that the left edge correspond to lexical material; or simply the requirement that the clitic be NonInitial. These represent the interaction of the Anchoring and Precedence descriptive parameters, while constraints of the Integrity family serve to specify the nature of an initial anchoring element. Up to this point, I have in general said little about the nature of the Domain within which special clitics appear. In most of the cases considered thus far, the domain whose properties they realize is that of the clause or the DP, and the phonological domain within which they are placed is the corresponding prosodic constituent. In fact, however, domain speciﬁcations are not always that simple and direct, and we will have to allow for the possibility that constraints such as those introduced above have additional parameters specifying the domain of their applicability. Thus, a clitic is speciﬁed not simply as LeftMost and/or NonInitial, but LeftMost within some domain D, etc.

146

Aspects of the Theory of Clitics

As I have already suggested, this provides us with a way to limit the descriptive apparatus somewhat. In addition to being anchored by the Initial or Final element of a domain such as the clause, clitics sometimes appear to be anchored by the Head of that domain (in particular, the ﬁnite verb). That possibility was not included in Klavans’s (1985) theory of the parameterization of clitic placement, but has seemed plausible as an account of the location of, e.g., Romance object pronominals and a variety of other clitics. Do we indeed need to posit additional mechanisms in order to describe these cases? Probably not. Where clitics are located with respect to the head of a phrase, we can say that they are actually placed by the same constraints as those just discussed, but within a domain which is circumscribed to contain only the head (e.g., V). Clitics such as the Romance pronominals, and a variety of Bulgarian and Macedonian clitics are all located with respect to the head verb of a clause (ﬁnite or non-ﬁnite, as the case may be). We can accommodate these cases by saying that properties of the larger phrase are inherited (via something like the ‘Head Feature Convention’ of Gazdar, Klein, Pullum, and Sag 1985), and the rules introducing clitics apply within a domain that is circumscribed to that head verb. They are then subject to alignment constraints such as LeftMost(cli ,V) which apply within that domain, ranked in such a way as to yield the observed order. This line of analysis immediately yields a solution to the problem of how to preserve clitic order regardless of whether the actual clitic sequence precedes or follows its anchoring verb. Ceteris paribus, a clitic constrained by high-ranking LeftMost(cli ,V) will appear to the left of the lexical verb within the syntactic domain V, and the order of several clitics within this domain will follow from the relative ranking of their corresponding LeftMost constraints. Suppose we say now that an even higher-ranking constraint requires that inﬁnitives (and such other non-ﬁnite forms as are relevant) must be aligned with the left edge of the same domain. LeftMost(Verb[−Finite],V), that is, outranks all of the speciﬁc LeftMost constraints specifying the positioning of the clitics. In that way, we describe the difference between the ﬁnite and the non-ﬁnite cases as a single additional constraint on the non-ﬁnite form: its LeftMost constraint dominates those of the clitics, while the LeftMost constraint applicable to ﬁnite verbs is itself outranked by those applying to the clitics. But regardless of the position of the verb itself within the head V domain, the relative ordering of its associated clitics remains the same, determined in all cases by the same ranking of clitic-speciﬁc alignment constraints. In Bulgarian, there is an additional wrinkle. Even when linked to a ﬁnite verb, clitics within the V domain follow the verb rather than preceding it

An Optimal Theory of Clitic Positioning

147

when the V itself is sentence initial. To accommodate the facts of V-initial sentences in Bulgarian, we can say that while each clitic is constrained by (ranked) LeftMost(cli ,V), each is also constrained by higher-ranked NonInitial(cli ,CP). The domains of left alignment and of NonInitial, that is, do not coincide. As a result, the clitics precede a ﬁnite verb unless that would put them in sentence-initial position. This particular pattern, in which enclisis versus proclisis depends on whether the associated verb is initial (in some relevant sense) or not is sometimes known as the “Tobler-Mussaﬁa Law,” (Tobler 1875/1912, Mussaﬁa 1898) especially in discussions of Romance languages. In the case of sentence-initial verbs, (as well as with non-ﬁnite verbs) the clitics follow the verb, while they precede it otherwise. Their relative order, which is determined by the ranking of their speciﬁc alignment constraints, remains invariant across these conditions. This eliminates a problem noted in section 6.1 above for the rule-based morphological approach to clitic positioning. In the descriptive theories of Klavans (1985) and Anderson (1992), a single domain is relevant to both of the parameters of clitic placement (Anchoring and Precedence), but on the constraint-based view, since the constraints are separate their relevant domains might be as well, as we have just seen for Bulgarian. A somewhat different example illustrating this theoretical possibility is presented by Richardson (1997) for Czech. Richardson ﬁrst shows that Czech in general has a collection of second-position clitics similar to those of SerboCroatian, as illustrated in (6.12). (6.12)

a. Vˇcera =jsem =se =mu koneˇcne omluvil yesterday past-1sg refl 3sg-dat ﬁnally apologized Yesterday I ﬁnally apologized to him b. Cervené ˇ tulipány =se objednal red tulips refl ordered He did order red tulips

The second of these sentences suggests that what constitutes ﬁrst position in Czech is a phrase, not a single PWord, and we assume that Integrity(XP) is ranked above the LeftMost constraints relevant to the clitics. In embedded sentences, clitics generally come immediately after the complementizer, as in (6.13). ˇ (6.13) Rekl, ˚ =mi =ho mužete ukázat že said-ppl that 1sg-dat 3sg-acc can show He said that you can show him to me

148

Aspects of the Theory of Clitics

From these facts we might conclude that CP (a domain including the clause nucleus together with a complementizer) constitutes the domain of second-position cliticization in Czech. But that is not quite accurate. To illustrate this, Richardson provides the near minimal pair in (6.14). (6.14)

a. Helena rˇ íkala, [CPže [IP=se Petr odstˇehoval]] Helen said that refl Peter moved Helen said that Peter had moved b. Helena rˇ íkala, [CPže Petr [IP=se odstˇehoval]] Helen said that Peter refl moved Helen said of Peter that he had moved

The ﬁrst of these is as we expect. But in the second, the clitic appears to be in third position. Richardson argues that in this example, the subject Petr is not in its normal position within the clause, but rather has been moved out and adjoined to IP as a sort of topic, as shown by the bracketing. What we see then is that the clitics are really constrained differently in two domains: they are drawn to the left edge of IP, but constrained not to be initial in CP. The constraint ranking required to achieve this is that of (6.15). (6.15) NonInitial(cli ,CP) LeftMost(cli ,IP) Since IP and CP generally coincide in main clauses, the clitics are placed correctly in second position when they occur there. In dependent clauses, however, there is material in CP that is not part of IP. As long as that consists of only a single element (the complementizer), the result still looks like simple secondposition placement within CP. It is only when more than one element occurs in CP outside of IP that we can tell the difference, and here we ﬁnd that the position of the clitic is bounded by the left edge of IP no matter how far that may be from the left edge of CP. Another analysis in which LeftMost and NonInitial have distinct domains is offered for European Portuguese by Barbosa (1996). In this language, clitics are attracted to the Verb, but they appear either preceding or following it depending on various factors: (6.16)

a. Ninguém/Alguém o= viu No one/Someone 3sg saw No one/Someone saw him b. O João não a= viu the John not her saw John didn’t see her

An Optimal Theory of Clitic Positioning

149

a= vê c. O João nunca the John immediately her sees John sees her immediately d. Viu =o o João saw him the John John saw him e. Esses livros, dei =os á Maria those books 1sg-gave them to Mary Those books, I gave (them) to Mary f. A Maria viu =o the Mary saw him Mary saw him Barbosa argues that the clitics are attracted to the left edge of a verbal projection (which she identiﬁes as IP, a speciﬁcation that I would replace with that of the ﬁnite V), but that they are prohibited from appearing initially in an Intonational Phrase. This is obvious in the case of V-initial sentences, and plausible when something is clearly dislocated. She also argues that referential subjects are always dislocated, thus establishing a new IntPhrase boundary. This could be represented as the constraint ranking in (6.17). (6.17) NonInitial(cl,IntPhr) LeftMost(cl,Vﬁnite) Here we have a phonological domain (IntPhr) relevant to one constraint, and a syntactic one relevant to the other. Galves and Sandalo (2004) argue that while Barbosa’s analysis is basically correct for Classical Portuguese, the modern language is subtly different. They show that in the language as written in Portugal between the ﬁfteenth and the eighteenth century, proclisis and enclisis are both found in sentences with definite subjects like (6.16f ), and that the variation is correlated with other factors (contrastive focus, constituent length) that are plausibly related to the issue of whether the initial subject constitutes a separate phrase prosodically. They argue that a change has occurred in modern European Portuguese, however. Deﬁnite (and pronoun) subjects now uniformly trigger enclisis under circumstances that cannot be based on prosodic differences, and so the domain of the relevant NonInitial constraint must have shifted from a prosodic category to a grammatical one some time around the end of the eighteenth century. It would take us too far aﬁeld to examine in detail the syntactic assumptions on which their reanalysis is based; I simply note their proposal as potential evidence that such a historical change may be possible.

150

Aspects of the Theory of Clitics

Another instance in which syntactic and prosodic domains interact is provided by the facts noted by Boškovi´c to which I referred in section 5.2 above. Recall that in Serbo-Croatian, clitics do not appear where we expect them when unusual intonation intervenes, as in the examples of (5.17) repeated here as (6.18). (6.18)

a. Sa Petrom Petrovi´cem srela =se samo Milena with Peter Petrovi´c met refl only Milena With Peter Petrovi´c, only Milena met b. Znaˇci da, kao što rekoh, oni =c´ e sutra do´ci means that as said they aux tomorrow arrive It means that, as I said, they will arrive tomorrow c. Ja, tvoja mama, obe´cala =sam =ti sladoled I your mother promised aux 2sg ice cream I, your mother, promised you an ice cream

In each of these cases, we would expect the clitics to come after the initial phrase, but they do not. As Boškovi´c notes, the basic regularity (5.18) is that the clitics appear in second position within an intonational phrase. We might suggest that in Serbo-Croatian, the clitics are attracted to the left edge of the (syntactic) clause, but prevented from occurring initially in the Intonational phrase, as expressed by the ranking in (6.19). (6.19) NonInitial(cli ,IntPhr) LeftMost(cli ,IP) This still does not ensure that the clitics will appear in the right place in the sentences with unusual intonation, however, since it would appear that their optimal location under this ranking would be at the end of the initial IntPhrase, rather than in second position within a later one. We can note with a number of authors that Serbo-Croatian clitics are dispreferred in ﬁnal position. Perhaps, then, another constraint such as (6.20) is also relevant. (6.20) NonFinal(cli ,IntPhr) In that case, putting the clitics after the initial phrase when that is followed by an intonation break would be worse than having them come later in the sentence (after the parenthetical, appositive, etc.). This approach is promising, but there must be more to the matter than this, since Serbo-Croatian clitics sometimes do occur in IntPhrase ﬁnal position in sentences like (6.21). (6.21) Moja sestra, koja =je u Sarajevo, sje´ca =vas =se my sister who 3sg-pres in Sarajevo remembers 2pl refl My sister, who is in Sarajevo, remembers you

An Optimal Theory of Clitic Positioning

151

Location of the clitics in this example after the ﬁrst constituent of either the ﬁrst or the last IntPhrase will violate (6.20). Since the position at the end of Moja sestra is closer to the left edge of the clause, this would appear (counterfactually) to be the optimal place for the sequence “=vas=se” in this sentence. It appears that what is at stake is actually a matter of the domain for clitic insertion. What we need to say to implement Boškovi´c’s insight about these examples is that where a clause is divided into multiple IntPhrases, the clitics are introduced within the IntPhrase that contains the head of the clause (the main verb). It is then constrained to occur as close as possible to the left edge of that phrase, but not initially. On this account, all of the relevant domain parameters refer to prosodically deﬁned units. As a ﬁnal example, let us consider some interesting facts about Bulgarian clitics pointed out by Legendre (2000a). Like those in Portuguese, these clitics are introduced within a projection of V, identiﬁed by Legendre as V . As in European Portuguese, most of the clitics are pre-verbal except when the verb is sentence initial, as shown in (6.22). (6.22)

a. Az =sum ˘ =mu =go dal I aux-1sg him it given I have given it to him b. Dal ˘ =sum =mu =go given aux-1sg him it I have given it to him

When multiple wh-words appear at the left edge of the clause, the clitics may be located in a position within the clause later than second, as illustrated in (6.23). (6.23) Koj kakvo mu= e= dal? who what him aux-3sg given Who gave what to him? This will follow if the domain relevant for clitic placement and for LeftMost(cli , D) is V , but for NonInitial(cli ,D) it is the intonational phrase. Thus far the situation is familiar, but what is interesting is the fact that while some of the Bulgarian clitics behave in this way, others are allowed to appear sentence initially. Ne, šte, li are all integrated into the clitic sequence, but they are not subject to NonInitial(cl,IntPhr). Furthermore, li is subject to an additional constraint requiring its left edge to be aligned with the right edge of a prosodic word. Sentences like (6.24) illustrate these possibilities. (6.24) Šte= go= viždaš =li fut him see-2sg Q Will you see him?

152

Aspects of the Theory of Clitics

Among other things, these facts show us that all of the clitics in a given language may not be treated in exactly the same way. Thus, the LeftMost constraints relevant to ne, šte, li outrank the corresponding NonInitial constraints, while the opposite is true for the rest of the second-position clitics. Nonetheless, the entire complex pattern can be accommodated straightforwardly within the kind of constraint system proposed here.

6.3 “Endoclitics” One property of special clitics which has been widely accepted as general at least since the appearance of Zwicky’s (1977) survey is the claim that these do not appear properly included within a word. That is, while there are both proclitics and enclitics, there are no “mesoclitics” or “endoclitics.” Some examples have been discussed in the literature, however, which appear to controvert this. With the apparatus developed to this point, I can address the three most prominent cases of this sort: Pashto clitics and verbs with preﬁxal stress, European Portuguese pronominal clitics with certain speciﬁc verbal forms, and the clitics of Udi. Pashto The second-position clitics of Pashto were ﬁrst discussed in this connection by Tegey (1977), who noted cases in which they appear to interrupt a monomorphemic word. Kaisse (1981, 1985) argued that the relevant cases should be analyzed in a somewhat less dramatic way, but that the clitics still should be seen as interrupting lexical units. Subsequent discussion in terms of Optimality Theoretic analyses can be found in van der Leeuw (1995a, 1997) and Roberts (1997, 2000). The present discussion largely (but not entirely) follows that of Roberts (1997). Most of the examples below are drawn from that work, which cites them in turn from Tegey (1977). Pashto has a set of clitics that appear in second position, in a sense to be made clear below.10 These include elements of several sorts, as shown in (6.25). (6.25)

Pronominal me 1sg de 2sg ye 3sg, 3pl am 1pl, 2pl mo 1pl, 2pl

modal ba will, might, must, should, may de should, had better, let adverbial xo indeed, really, of course no then

10 There is also a set of clitics associated directly with the verb, which have been largely ignored in the theoretical literature in favor of the second-position elements. That tradition is maintained here.

An Optimal Theory of Clitic Positioning

153

These clitics appear in second position within the clause in both main and subordinate clauses, as illustrated in (6.26). (6.26)

a. xušal =me zyaati ne wah-i Khosal 1sg anymore neg hit-3sgPres Khosal doesn’t hit me anymore b. zyaati =me ne wah-i anymore 1sg neg hit-3sgPres He doesn’t hit me anymore c. ne =me wah-i neg 1sg hit-3sgPres He doesn’t hit me d. wah-i =me hit-3sgPres 1sg He hits me e. z@ GwAó@m cˇ e tor =me wuguri I want comp Tor 1sg see I want Tor to see me

Sentence (6.26d) is particularly interesting, in light of the fairly rigid verb-ﬁnal structure of sentences in Pashto. When the initial constituent of the sentence is a multi-word nominal or a postpositional phrase, the clitics follow the entire phrase rather than its initial word, as shown in (6.27). (6.27)

a. aGa šel kal@na danga aw khAysta peGla =me n@n that 20 year tall and pretty girl 1sg today byA w@lida again saw I saw that twenty-year-old tall and pretty girl again today b. xušal aw patang =ba =ye d@r ta rAwói Khosal and Patang will 3sg you to bring Khosal and Patang will bring it to you c. laylA´ na =de A-xist@ Layla from 2sg buy You were buying it from Layla

On the other hand, an initial VP can be interrupted by a clitic, as in (6.28). (6.28) xar =de n@ rAwali donkey should neg bring He should not bring the donkey

154

Aspects of the Theory of Clitics

From these facts, I conclude that the clitics in (6.25) are each oriented toward the left edge of IP, but this alignment is outranked by the requirement that they not be initial within that domain, and also by the requirement that DPs and PPs not be interrupted. The resulting constraint ranking is given in (6.29). (6.29) Integrity(DP), Integrity(PP), NonInitial(cli , IP) LeftMost(cli , IP) Thus far, there is nothing unusual about these facts in the context of the general phenomenon of second-position clitics. Complications arise, however, when we consider the role of stress in determining the position of clitics. First, sentence-initial stressless constituents are skipped over, even if this results in the clitic’s appearing far from the left edge of the clause, as in (6.30). (6.30) ra ta te rA škAw´@ =de me for from-it here pick 2sg You were picking it for me from it (and bringing it) here The ﬁrst four words of this sentence are unstressed. As prosodically deﬁcient simple clitics, I assume they are adjoined to the following lexical word (the verb) as free clitics (in the sense of section 3.2), forming a PPhrase with that PWord. The fact that second-position clitics do not interrupt this sequence suggests that Integrity(PPhrase) appears with the other highly ranked Integrity constraints in (6.29). The role of prosodic structure in these matters is highlighted by the minimal pair of sentences in (6.31). (6.31)

a. rA sara wí =de me with be let Let it be with me b. rA sará =de wi me with let be Let it be with me

When the normally unstressed PP rA sara ‘with me’ is assigned contrastive or focal stress, the clitic follows it rather than the copula wi. This is presumably because the presence of stress promotes the PP to the status of a separate PPhrase rather than a sequence of free clitics within the PPhrase headed by wi, and the clitic can thus be positioned closer to the left edge of the IP without interrupting a PPhrase. So what has any of this to do with the question of “endoclitics”? The answer emerges when we consider aspectual distinctions in Pashto verbs. The

An Optimal Theory of Clitic Positioning

155

difference between imperfective and perfective forms is often indicated by a shift of the location of stress. In verbs containing a preﬁx, or consisting of a noun, adjective, or adverb plus a “light” verb, the imperfective generally has stress on the second element, but shifts this to the ﬁrst element in the perfective. Thus, úel-wAhé ‘(I) was pushing it,’ but úél-wAhe ‘I pushed it’; porew@st´@ ‘(I) was carrying it across,’ but póre-west@ ‘I carried it across.’ In some monomorphemic verbs, a preﬁx is recruited in the perfect to bear this shifted stress: skund´@la ‘was pinching,’ but w´@-skund@la ‘pinched.’ When one of the internally complex verbs comes at the beginning of IP, clitics are located after the element bearing the stress, as shown in (6.32). (6.32)

a. i. úel wAh´@ =me pfx push 1sg I was pushing it ii. úél =me wAh@ pfx 1sg push I pushed it b. i. pore- west@´ =me across carry 1sg I was carrying it across ii. póre =me west@ across 1sg carry I carried it across c. i. pezand´@ =de recognize 2sg You were recognizing (him) ii w´@ =de pezand@ pfx 2sg recognize You recognized him

If one were inclined to treat the two parts of these verbs as forming a single “word,” the clitics in the (ii) examples of (6.32) would be instances of “endoclitics,” but it seems at least equally reasonable to see each part of these constructions as a distinct unit—let us call it a ‘lexical word’ – in which case the clitics do not interrupt lexical words. There is another (small) class of verbs showing an alternation in the location of stress within a monomorphemic stem, where either possibility occurs in imperfective forms. Crucially, the location of stress in these cases is not relevant to the location of clitics, as shown in (6.33).

156 (6.33)

Aspects of the Theory of Clitics a. p@rebd´@ =me beat 1sg I was beating (him) b. p´@rebd@ =me beat 1sg I was beating (him) c. *p@=me rebd@

These cases conﬁrm that clitic placement cannot in general interrupt a single lexical word. This suggests that Integrity(LexWord) is among the highranking Integrity constraints in Pashto. Word integrity, in various senses of “word,” is of course exactly the prohibition against “endo-clitics.” There is another set of verbs that challenge these assumptions, however. Verbs with initial a that show stress shift do place clitics differently depending on the location of the stress as in (6.34). (6.34)

a. i. axist´@l@ =me buy 1sg I was buying (them) ii. á =me xist@l@ ? 1sg buy I was buying (them) b. i. aGust´@ =me wear 1sg I was wearing (it) ii. á =me Gust@ ? 1sg wear I was wearing (it)

If verbs such as those in (6.34) are genuinely single lexical words, the (ii) sentences here involve real endoclitics, and this is the crux of the issue posed by Pashto for a theory of special clitics. The alternative, of course, is to regard these verbs as internally complex, consisting of a preﬁx a- and a stem, and thus to assimilate them to the behavior of other complex verbs such as those in (6.32). Tegey (1977) argues that this is incorrect: while some of these verbs actually represent the continuation of forms with a distinct preﬁx a-, others (such as ax- ‘buy’) are monomorphemic stems that never contained such a preﬁx. Kaisse (1981, 1985), however, argues (correctly, in my view) that this historical difference is irrelevant, and

An Optimal Theory of Clitic Positioning

157

that there is no reason not to assume that many verbs with initial a- that was not historically a distinct preﬁx have been reanalyzed so that synchronically their structure is like that of the complex verbs in (6.32). In that case, there are no instances of real “endoclitics” to consider in Pashto. The analysis suggested here, then, is as follows. The prosodic structure of Pashto involves assigning PWord status to any lexical element that bears a stress. PPhrases are constructed on the basis of PWords, including preceding unstressed material as free clitics within the PPhrase and ending at the right edge of the head PWord. The second-position clitics in (6.25) are then all subject to LeftMost(cli , IP), itself outranked by NonInitial(cli , IP).11 The place in which the clitics actually lodge is further constrained not to interrupt a lexical word, a (syntactic) DP, or a (prosodic) PPhrase. If we assume that prosodically deﬁcient postpositions are incorporated as afﬁxal clitics into a preceding PWord, we can dispense with the constraint Integrity(PP) appealed to in (6.29). Pashto thus allows clitics to interrupt complex lexical combinations, but not individual lexical words. The sense in which this language instantiates “endoclitics” is thus extremely limited. European Portuguese I argued above that in this language, pronominal clitics are introduced within the domain of V. They are then subject to LeftMost(cli ,V), but these constraints are outranked by NonInitial(cli ,IntPhr). The result is proclisis to the ﬁnite verb unless this is initial within an IntPhrase (or some other constituent, as Galves and Sandalo 2004 argue for Modern Portuguese), in which case enclisis results. This is indeed the normal state of affairs, but under speciﬁc circumstances there are additional complications. In particular, when the verb is a form in the synthetic future or conditional, the clitic appears to be located internal to the form, between the stem and the ending. This apparent case of endocliticization has been discussed in Optimality Theoretic terms by van der Leeuw (1995b, 1997), and the prosodic organization of the relevant forms is clearly established by Vigário (2003). I follow those discussions in part, though the analysis below diverges in some ways from them. The forms involved show the endings in Figure 6.1, added to a stem which is usually the same as the inﬁnitive, except in a small number of (rather frequent) verbs. 11 In place of NonInitial(cl , IP), much of the literature on Pashto in OT terms appeals to an i alignment constraint requiring the clitics to be aligned such that their left edge coincides with the right edge of a PPhrase. The difference between that description and the one assumed here will not be explored further, because it does not bear on the question of possible “endoclitics” with which I am concerned.

158

Aspects of the Theory of Clitics 1sg 2sg 3sg 1pl 2pl 3pl

Future -éi -ás -á -émos -éis -ã´ o

Conditional -ía -ías -ía -íamos -íeis -íam

Figure 6.1. European Portuguese endings in forms with “endoclitics”

When there is no clitic present, the resultant form has a single stress, located on the ending: daríamos ‘(we) would give,’ perceberás ‘(you sg.) will understand,’ dirá ‘(s/he) will tell,’ mostrará ‘(s/he) will show.’ When enclitics are added, however, an additional stress appears on the stem: dár-te-íamos ‘(we) would give to you,’ percebér-me-ás ‘(you sg.) will understand me,’ dír-lhe-ía ‘(s/he) would tell him,’ mostrár-no-los-á ‘(s/he) will show them to us.’ Note also that forms with enclitics following the entire verbal form are quite impossible: *dariamos-te, *perceberas-me, *dira-lhe, *mostrara-no-los. There are two questions to answer, then: (a) when the clitic cannot be proclitic, why does it appear before the ending and not after the entire verb form; and (b) why is there an extra stress on the stem precisely when a clitic intervenes between it and the ending? The history of the forms in question provides some tantalizing hints about the answers. As opposed to all other tense forms in the paradigm of the Portuguese verb, the inﬂected future and conditional result from the relatively recent fusion of a separate auxiliary verb (a form of haver ‘to have’) with the inﬁnitive form of the lexical verb which it formerly governed. This is similar to developments in some other Romance languages (e.g., French); it is also quite consistent with a comparison of the endings for the future and conditional in Figure 6.1 to the present and past forms of haver provided in Figure 6.2.

1sg 2sg 3sg 1pl 2pl 3pl

Present hei hás há havemos haveis hão

Past havia havias havia havíamos havíeis haviam

Figure 6.2. European Portuguese haver ‘to have’

An Optimal Theory of Clitic Positioning

159

If we could regard the future and conditional “endings” and the inﬁnitive base as two separate verbs in the syntax, both questions would be answered. The clitics would appear where they do because they are really proclitic to the auxiliary verb, and the resulting form would show two stresses because there are two distinct words, each of which bears stress. Such an analysis involving formation of the future and conditional in the syntax, rather than the morphology, is argued for by several authors, including Vigário (2003). From a syntactic point of view, however, it faces a number of problems, discussed in part by Crysmann (2000). In addition, it fails to account (except by stipulation) for why two stresses should appear when an “endoclitic” is present, but not otherwise. This last fact suggests that prosodic structure is relevant. An alternative is available, however, that allows us to have our cake and eat it too. This is the proposal that while the future and conditional forms are not internally complex from the point of view of the syntax, they display internal structure with respect to the morphology (and the prosodic phonology). In Anderson (1992), I argued that most forms composed by word formation rules on the basis of other forms do not have an internal structure of the sort standardly assumed, with morpheme divisions represented structurally. The only internal structure, once a morphologically complex form has been constructed is that motivated by the phonology, a position shared with some other views of morphology such as that of Stump (2001). While this is the case for most morphologically complex forms, however, there are others for which it is much less attractive. Compounds are an obvious example, since they appear in many languages to display distinct morphological and phonological domains for each of their elements. I suggested in Anderson (1992) that some noncompound formations in some languages, which I labeled “composites,” also display some internal structure of a non-phonological sort. Icelandic “middle voice” verbs such as kallast ‘be called X, have the name X,’ which derive historically from the fusion of a reﬂexive pronoun with the verb, have a structure such as that in (6.35). (6.35)

V V kalla st

Structure of this sort is not the automatic result of word formation (contrary to the assumption of many “morpheme-based” views of morphology), but can be stipulated as a component of speciﬁc processes, just as the presence of speciﬁc afﬁxes or other phonological effects forms part of the realization of a particular category. We could assume that inﬂected future and conditional forms of Portuguese verbs are also composites, reﬂecting their relatively recent

160

Aspects of the Theory of Clitics

reanalysis from originally complex forms. On this picture, the structure of, e.g., daríamos ‘(we) would give’ is that in (6.36). (6.36)

V LexWord dar iamos

With respect to prosodic structure, I assume that all of the material in (6.36) is organized into a single PWord, whose stress pattern is determined by that of its second element (since stress in these forms always falls on the desinence, rather than the root). Consider what happens when a clitic is added, however, within the syntactic verbal domain which the verb in (6.36) interprets. If the verb is in medial position within an IntPhrase, of course, the clitic is adjoined at the left as a proclitic, and nothing of further interest occurs. If the verb is initial within IntPhrase, however, the clitic cannot be realized in LeftMost position within the verbal domain. It must, however, be positioned as close to the left edge of this domain as possible, consistent with other, higher-ranking constraints. Suppose that in Portuguese lexical words (though not PWords) are subject to high-ranked Integrity. The clitic will then be positioned at the right edge of the lexical word, but preceding the conditional ending. The ranking of Integrity(LexWd), then, in conjunction with what we already know about European Portuguese clitics, accounts for the positioning of the clitics in the future and conditional forms. The additional stress results from a prosodic reorganization which affects them when the stem of the structure in (6.36) acquires an enclitic. Enclitics in Portuguese are structured as afﬁxal clitics together with their hosts, as shown in detail by Vigário (2003) on the basis of a variety of phonological processes. The verbal stem thus becomes the head of an independent PWord, resulting in the emergence of stress on the stem. This, in turn, requires the desinence to be restructured as a PWord in its own right, within which it retains its stress. The resulting prosodic structure, justiﬁed in detail in Vigário (2003), is that of (6.37). (6.37)

PWordMax PWord PWord PWord dar te iamos

The constituent “PWordMax” is the category Vigário assigns to ordinary compounds, as well as these forms with internal clitics. It may or may not be identiﬁable with the PPhrase, a matter which I will not address here.

An Optimal Theory of Clitic Positioning

161

Enclitics associated with future and conditional forms in European Portuguese, then, are not really “endoclitics.” They do not interrupt lexical words, and their position within prosodic structure is not properly included within a PWord, but only within a higher prosodic type associated with internally complex units, compounds and other composites. Their positioning follows from the constraints applicable in the language, as a relatively straightforward variation on the notion of second position, here relative to the category V. Udi Without doubt, the most extensively argued (and most successful) case for genuine endoclitics is that presented by Harris (2000, 2002) for the North-East Caucasian language Udi. The facts here are intricate, and it is quite impossible for me to review them comprehensively here (especially since that has already been done quite thoroughly in Harris (2000), to which the reader is referred for aspects of phonology, morphology, and syntax of Udi not dealt with here). I will therefore focus on the limited set of phenomena directly bearing on the analysis of endoclitics. Udi has a set of person-marking clitics that show agreement with the subject of a clause.12 There are four sets of these, shown in Figure 6.3. The single clitic in the “Question” set simply replaces other 3sg forms in questions. Otherwise, the “Inversion” set is used with a class of experiencer subjects, the “Possession” set with expressions of possession and the subjects of a handful of other verbs, and the “General” set elsewhere. The choice of one of the variants of a given form is entirely based on phonological factors. Under some circumstances, these appear as simple enclitics. Certain speciﬁc categories of Tense and Aspect, for example, require the person-marking

1sg 2sg 3sg 1pl 2pl 3pl

General -zu, -z -nu, -n, -ru, -lu -ne, -le, -re -yan -nan, -ran, -lan -q’un

Inversion -za -va -t’u -ya -va, -va.n -q’o

Possession -bez, -bes -vi -t’a -beš -e.f -q’o

Question

-a

Figure 6.3. Person Markers in Udi 12 The behavior of another clitic -qa- “Subjunctive” falls together with that of the person markers, and will not be discussed separately here.

162

Aspects of the Theory of Clitics

clitics to follow them in their position after the verb. Also, when some constituent of the clause bears focus, the person markers immediately follow it. This includes negative particles and questioned constituents, which are obligatorily focused, as well as any other focused element. Finally, when the clause involves predication by means of a phonetically null copula, the person marker follows the associated predicate nominal. We can note that in all of these cases the clitics are constrained by high-ranking RightMost(cli ) within the domain in which they appear, and nothing else prevents them from appearing as ﬁnal enclitics. The interesting issues arise when no element in the clause takes focus in one of these ways. In that case, the person markers appear together with the verb itself, of which there are two principal varieties to be considered. Some Udi verbs are simple, monomorphemic stems (at least from a synchronic point of view), but many others are internally complex. These consist of an initial element which may be practically anything: nouns, adjective, locative preverbs, historically simplex verb stems, borrowed verbs, inﬁnitives (to form causatives), or opaque elements that do not occur independently. This initial element (which may itself be internally complex) is followed by one of a number of what Harris refers to as “light” verbs. Some of these occur independently with a lexical sense which may or may not be relevant to the meaning of complexes that they head, while others occur only in this use. When the verb is of this type, the person marker is placed immediately before its ﬁnal light verb, as in the examples of (6.38). (6.38)

a. . . . pasˇcaG-un Gar-muG-on lašk’o=q’un-b-esa king-gen boy-pl-erg wedding-3pl-do-pres The king’s sons marry [the girls they had rescued] b. nana-n k’uˇcan äyel-ax ak’-es=ne-d-e mother-erg child-dat see-inf-3sg-caus-aorII puppy.abs The mother showed a puppy to the child c. áyel kala=ne-bak-e child.abs big-3sg-become-aorII The child grew (up)

Let us assume that when the person-marking clitics are associated with the verb (perhaps because the verb takes default focus in the absence of another focused constituent), they are introduced into a domain consisting of exactly the lexical verb. In the examples of (6.38), this is internally complex—a composite, in the sense of Anderson (1992), as discussed above in connection with the future and conditional in Portuguese. Harris provides a wide range of arguments

An Optimal Theory of Clitic Positioning

163

that these complex verbs constitute lexical and syntactic units, but these do not contradict the claim that from a morphological point of view, they are composed of more than one distinct unit. And it is precisely between these pieces that the clitics are located. From the cases in which the person markers are associated with something other than the verb, we know that their positioning is governed by RightMost(cli ,D). If that were all there were to the story, we would expect that when associated with the verb, they would follow it, but instead they immediately precede its ﬁnal element. In the discussion of KamhmuP above in section 6.1, I appealed to a constraint LeftEdgeFaith(Word) (6.9) to ensure that the inﬁx -rn- in that language follows exactly the initial segment of the word in which it appears. What we need in Udi, apparently, is a similar constraint on the right edge of the verb such as (6.39). (6.39) RightEdgeFaith(Verb): The element at the right edge (of the verb, excluding a sufﬁxed Tense marker) in the output form should correspond to the element at the right edge in the input. If RightEdgeFaith(Verb) dominates RightMost(cli ,Verb), the clitics will be located in an inﬁxed position. What we observe is that they precede the ﬁnal structural component of a complex verb, regardless of the size of that element. This suggests that an appropriate constraint from the Integrity family also dominates RightMost(cli ,Verb). For convenience, I will assume that the components of a complex verb are all lexical words, although this speciﬁc choice is not important: there is no doubt that the components of a complex verb are structural units of some kind, and the correct constraint will require that these units not be interrupted. Now, however, we come to the central examples. When the verb is not internally complex, but consists of only a single unit, we ﬁnd the clitics located before the ﬁnal consonant of that unit, as in the example of (6.40).13 (6.40)

a. pasˇcaG-un Gar-en gölö be.=ne-G-sa met’a-laxo king-gen boy-erg much look1 -3sg-look2 -pres this.gen-on The prince looks at this for a long time (root be.G ‘look’) b. kaGuz-ax a=z-q’-e letter-dat receive1 -1sg-receive2 -aorII I received the letter (root aq’ ‘receive’)

13

I follow Harris’s conventions in glossing these sentences, including the indication of an interrupted root by the same gloss with subscripts for its parts, except that I explicitly mark the person markers as clitics.

164

Aspects of the Theory of Clitics c. q’aˇcaG-G-on bez tänginax baš=q’un-q’-e thief-pl-erg my money.dat steal1 -3pl-steal2 -aorII Thieves stole my money (root bašq’ ‘steal’) d. ba=ne-k-sa sa pašˇcaG-k’ena adamar be1 -3sg-be2 -pres one king-like person.abs [Once upon a time, there] is a person like a king (root bak ‘be’)

Here we (ﬁnally) have what appears to be a real case of endoclitics, since (a) Harris (2002) provides a range of arguments that the person markers here really are clitics, and (b) there is no motivation for seeing the material on either side of these clitics as a structurally autonomous unit. We do not really have to add more to the analysis above in order to accommodate these forms, however. True, the constraints already introduced for Udi would be expected to place a clitic, when introduced in the domain of a verb that is not structurally decomposable, at its left edge (the rightmost position that preserves the right edge of the verb and does not interrupt a structural unit). But we can remedy this incorrect prediction by invoking a condition of faithfulness at the left edge of the verb as well. Constraint (6.41) is simply the mirror image of (6.39). (6.41) LeftEdgeFaith(Verb): The element at the left edge (of the verb) in the output form should correspond to the element at the left edge in the input. Assuming that LeftEdgeFaith(Verb) dominates Integrity(LexWd), a single structural unit will be interrupted precisely when necessary to preserve InputOutput Faithfulness at the edges of the verb. In this case, the clitic will necessarily be positioned within the verb, as far to the right as possible without being at the right edge: that is, before the ﬁnal consonant of the root. The overall ranking of constraints that is required is as in (6.42).14 (6.42) LeftEdgeFaith(Verb) RightEdgeFaith(Verb) Integrity(LexWd) RightMost(cli , Verb) So I conclude that endoclitics are indeed a real structural possibility in language—either in the limited sense of clitics introduced between elements of a composite form, as in Pashto and European Portuguese, or in the even more radical sense illustrated by the Udi examples in (6.40). This should not, 14 Left-edge faithfulness is ranked above right-edge faithfulness here in order to deal correctly with verb stems consisting of a single consonant, which are treated in Harris (2002: 127ff) along with some other cases as exceptions to the generalizations above. I do not deal with these exceptional forms here, but they do not appear to present problems for the proposed analysis.

An Optimal Theory of Clitic Positioning

165

however, be a great surprise, on the account offered here. On that picture, the principal barrier against the formation of endoclitic structures is the generally high ranking of constraints of the Integrity family. An essential property of constraints in an OT framework lies in the fact that they may be violated when other conditions require that. The potential existence of a set of requirements that force violations of Integrity(Word) (in various senses of “word”) is thus entirely to be expected.

6.4 Tagalog Second-Position Clitics To illustrate the application and force of the ideas explored in this chapter, I provide here an account of the system of second-position clitics in Tagalog, a language with a particularly interesting and intricate system of clitics. Tagalog (Schachter and Otanes 1972; Schachter 1973) displays clitic elements with two distinct sorts of function, all following the ﬁrst element of the sentence. One set of these, consisting of pronominal elements representing arguments of the clause, is presented in Figure 6.4. Within the set of pronominal clitics, there are two subsets, depending on the grammatical role of the element they represent. Since the analysis of grammatical relations in Tagalog clause structure is not at issue here, and nothing in particular follows from the terminology, I simply follow Schachter in referring to these as the “Topic” set and the “Complement” set. A second set of clitics consists of a set of “particles” ﬁlling a variety of functions, as given in Figure 6.5. All of the elements in Figures 6.4 and 6.5 constitute special clitics in the sense of Chapter 2, because they are all subject to a particular regularity in their placement: all come in second position, following the ﬁrst element of “Topic” “Complement” 1sg ako ko 2sg ka mo 3sg siya niya dual kata nita 1pl (excl) kami namin 1pl (incl) tayo natin 2pl kayo ninyo 3pl sila nila 2sg(T)+1sg(C): ka+ko→kita Figure 6.4. Tagalog Pronominal Clitics

166

Aspects of the Theory of Clitics ba kasi kaya daw din ho lamang man muna

(interrogative) ‘because’ (speculation) (reported speech) ‘too’ (politeness) ‘only’ ‘even’ ‘for a while’

na naman nga pa pala po sana tuloy yata

‘already’ ‘instead’ ‘really’ ‘still’ (surprise) (politeness) (optative) ‘as a result’ (uncertainty)

Figure 6.5. Tagalog Particle Clitics

the sentence. When more than one clitic is present, other regularities govern their sequence relative to one another, and these will be the object of discussion below. Descriptive Regularities The overall framework for the description of clitics in the present book attributes their complex surface properties to the interaction of (individually quite simple) principles from a number of areas of grammar: at least phonology, morphology, and syntax. The analysis of the Tagalog clitics illustrates virtually the full range of considerations that can interact in determining the placement of clitics. The property of appearing in second position characterizes particular clitics in Tagalog, and not the entire language. It is not the case, that is, that all phonologically weak elements appear in this position (as Wackernagel assumed in his 1892 analysis of the early Indo-European languages), since Tagalog also has a set of particles a, e, ha, o (Schachter and Otanes 1972: 461–3) which appear sentence ﬁnally. When more than one of the second-position clitics is present in a sentence, their relative ordering is governed by a set of principles that have been the subject of some subsequent discussion. The basic ordering regularities are those stated in (6.43). (6.43)

a. Monosyllabic members of the set of pronoun clitics in (6.4) always precede particles (the elements listed in Figure 6.5). b. The particles in Figure 6.5 have an internal ordering among themselves. c. Disyllabic pronoun clitics follow all particles.

An Optimal Theory of Clitic Positioning

167

Within each of these three subsets of clitics, there is a ﬁxed relative ordering. In the case of the two subsets of pronominals, this is presumably stipulated by the ranking of corresponding LeftMost constraints, since it does not appear to follow from any more general principles. In the case of the particles, there is also a determinate ordering among the elements. These fall into several subclasses, and I presume their relative order can be made to follow from a combination of semantic scope and language-particular stipulation, much like the relative order of derivational elements in morphology, though I will not attempt to develop the relevant principles here. There is some agreement in the descriptive literature, such as Bloomﬁeld 1917 and Schachter and Otanes 1972, that the clitic elements in ﬁgures 6.4 and 6.5 generally follow the ﬁrst stress-bearing word of the sentence, though there are as usual some complications in this picture. The examples that are most directly consistent with this simple picture involve clitics that appear to interrupt otherwise unitary constituents so as to come after a single sentence-initial word, as in (6.44). (6.44) Ganu =ka =na =ba kakinis? how you already int clever How clever are you now? (Bloomﬁeld 1917: 143) On the other hand, there are various constructions consisting of more than one phonological word that do not get broken up (Schachter and Otanes 1972: 187ff): proper names, numerical expressions, times of day, ages, amounts of money, etc., as in (6.45). (6.45) Bukas ng gabi nang alas otso =siya aalis tomorrow night at eight o’clock he leaves It’s tomorrow night at eight that he’s leaving (Schachter and Otanes 1972: 188) Schachter and Otanes (1972) describe these facts by treating certain construction types, ones they call “obligatory non-pre-enclitics,” as uninterruptable. In the present theory this is expressed by an appropriately parameterized high-ranking Integrity constraint. I have nothing particularly illuminating to say here about why some constructions have this property while others do not. Where a sentence-initial constituent can be interrupted, Integrity(Word) in combination with NonInitial(cli ,D) nonetheless prevents clitics from appearing further to the left than after the initial word in (6.44). One further reﬁnement relates to the domain in which the clitics are placed, and involves the fact that “second position” sometimes has to be within a domain that excludes certain preposed, focused constituents. I will discuss this

168

Aspects of the Theory of Clitics

point in more detail below; for now let us simply assume that the domain within which some element appears in “second position” can be identiﬁed with the one within which it is introduced, and that this follows (in some way) from its grammatical properties. These complications do not affect the major points to be made below: what matters is that “second position” has the same basic sense for all of the elements in Figures 6.4 and 6.5. Components of an Analysis Tagalog clitics pose problems of varying severity for an account based on purely syntactic mechanisms. First, there is the usual difﬁculty of placing clitics “after the ﬁrst word,” a notion that does not have a natural reconstruction in terms of those parts of grammar that normally deal in terms of phrases as basic elements. This is especially true if the relevant notion of “word” is a phonological one, which would result in difﬁculties for a principle of “Phonology-free Syntax” (Pullum and Zwicky 1988). A different (and potentially more serious) puzzle is how to derive the regularities of order among clitics. If these are as stated in (6.43) above, it is not clear that they can be described syntactically at all. This matter is the subject of Schachter (1973): the problem is that the order appears to depend on the phonological property of how many syllables a given clitic has; and syntactic rules, even of the sort that were accepted back in 1973, ought not to have access to the speciﬁcs of the phonological realization of elements they affect. But how can this be avoided? Note in particular the portmanteau element kita. This represents the combination of a 2sg.Topic and a 1sg.Complement, and replaces the expected sequence ka+ko. Each of the latter elements individually is monosyllabic, and thus should precede any particles (by 6.43a), but when they come together, the combination is replaced by a disyllabic portmanteau which must follow particles (by 6.43c). Even a set of rules that place the pronouns (as syntactic elements) one by one, and that treats the length regularities as accidental, will fail in this case. Schachter (1973) evaluated several brute force solutions to these difﬁculties utilizing reordering transformations, and showed that they either fail altogether, or at least fail to capture the generalization about length. His own proposal, following Perlmutter’s (1971) analysis of French and Spanish clitics, is a surface constraint: a template like (6.46), stipulating explicitly that monosyllabic pronouns precede particles, and that these in turn precede longer pronouns. This, he suggests, could serve either to ﬁlter the output of a scrambling rule or else to linearize elements at surface structure. (6.46) Pro1σ < Particles < Pro2σ

An Optimal Theory of Clitic Positioning

169

Such a description would allow us to state the regularity about syllable structure, but note that it does not explain either of the relevant facts: (a) monosyllabic pronouns come ﬁrst; and (b) otherwise, pronouns come after particles. Subsequent work (e.g., Kroeger 1993) similarly leaves this pattern as a matter of stipulation. In Anderson (1992), I noted that an explanation of the fact that at least most of the pronouns follow particles might be derived from the analysis of clitics as sentential morphology. Within words, derivational morphology comes ‘inside of ’ inﬂectional morphology. If we ignore the monosyllabic pronouns for the moment, and think of clitics as sufﬁxes to the ﬁrst word or phrase, then we could say the ordering of particles before pronouns reﬂects the same regularity: particles have various semantic and pragmatic content, and are thus “derivational,” as opposed to pronouns, which are (on this analysis) agreement markers, and thus “inﬂectional.” It would be natural for pronominal elements to be sufﬁxed outside of particles, then, and thus to follow them. What accounts for the extension of the “derivation inside of inﬂection” theorem from morphology to the description of clitics? In the word-internal case, this follows (as argued in Anderson 1992) from the place of lexical insertion in a grammar, given the nature of (productive) inﬂection. What needs to be shown is that in the case of clitics, the corresponding result also follows from the architecture of grammar. Adopting the overall terms of Chomsky (1995), the introduction of semantically contentful particle clitics can be seen as a generalized instance of the operation “Merge.” In the development of complex syntactic structures, the usual step consists of merging a word (taken from the structure’s Numeration) with (some part of) an existing syntactic structure. In the case of contentful clitics, I suggest that the way “Merge” works is to introduce some afﬁx-like material (rather than an autonomous word) into the structure.15 If we suppose, as seems natural, that all instances of this generalized “Merge” operation must be completed prior to the point at which clausal agreement- like material has to be realized phonologically, the particle clitics will all be in place at the point where the structure exists to support the introduction of grammatical (pronominal) clitics. As a result, “grammatical” or “inﬂectional” clitics (such as the Tagalog pronominals) will be introduced “outside of ” semantically and pragmatically contentful (or “derivational”) elements such as the Tagalog particles. 15

Since the mode of introduction of this material is the “morphological” one associated with clitics, it remains syntactically inert and is not visible for syntactic purposes in the way lexical words are.

170

Aspects of the Theory of Clitics

Obviously, this involves serialism in the overall derivation of a sentence— a notion which is not fundamentally at odds with the invocation of principles from OT to describe what happens at a given stage of the derivation, though it is at odds with the common (but incorrect) assumption that constraint systems must always involve a single monolithic collection of constraints applied in parallel. As noted above, much of the non-monolithic architecture of Lexical Phonology is in fact perfectly consistent with the basic insights of a constraintbased approach such as OT. But what about the syllable-length regularity? It is clearly an embarrassment for this account to have some “inﬂectional” clitics coming inside of the “derivational” ones. In Anderson (1992), I proposed to treat this by saying that the monosyllabic clitics are introduced by “head-inﬂection” rules, as opposed to all the others, which are “word inﬂection.” This, however, still misses the fact that the difference depends on prosodic size. In a step backward from Schachter’s analysis, it does not even state this fact, let alone explain it. Suppose that the relation between particles and disyllabic pronouns does indeed follow from an appropriate generalization of the “derivation precedes inﬂection” theorem. That is, all of the derivational clitics are already in place, attached to their host, at the point the pronominal ones are introduced. The difference between principles (6.43a) and (6.43c) must then follow from some other aspect of the grammar of clitics. When we then ask where a difference between the behavior of disyllabic and monosyllabic clitics might lie, the answer is clear: the difference itself is phonological, so the obvious place to look is in the phonology. Let us then suppose that monosyllabic and disyllabic clitics differ in the way they undergo Stray Adjunction. This difference could then interact with the system of constraints determining their behavior as secondposition elements so as to yield the observed placements. The minimal foot in Tagalog is bimoraic, and disyllabic clitics are prosodically rich enough to be structured as feet. Monosyllabic pronominals, on the other hand, are monomoraic, and thus cannot constitute feet on their own: they are simply stray syllables. The difference is represented in (6.47). While differing in their prosodic characterization, both are phonological clitics in the sense of Chapter 2, since they are not PWords, and Stray Adjunction must thus incorporate both types into an adjacent PWord. (6.47)

Disyllabic clitic:

Monosyllabic clitic:

Ft

σ si

σ ya

σ ko

How does Stray Adjunction operate to provide these orphan elements with homes? Suppose we say, in accord with the prosodic hierarchy as discussed in

An Optimal Theory of Clitic Positioning

171

Chapter 3, that a stray foot can (and must) be incorporated into the prosodic word to its left, while a stray syllable is adjoined to a foot on its left. The pronouns are true second-position clitics: that is, they are subject to high-ranking LeftMost and NonInitial constraints while not being allowed to violate a variety of Integrity constraints. In particular, lexical items are Integral in this sense, in addition to various uninterruptable phrasal types as noted above. Stray Adjunction thus presents the two cases in (6.48). The difference between incorporation (in the Foot case, 6.48a) and adjunction (in the case of stray syllables, 6.48b) in these rules follows from the well-known fact that the internal constituency of feet is typically rather narrowly constrained, while phonological words are somewhat more loosely structured and may display an organization that is ﬂatter and internally more homogeneous in comparison with that of feet. (6.48)

a. Stray Foot Adjunction: PWord X Foot b. Stray Syllable Adjunction: Foot Foot

X σ To see how this helps to provide an account of our problem, consider a sentence such as (6.49), with a monosyllabic clitic, a particle, and a disyllabic clitic, in that order. (6.49) nakikita =ka =na =niya sees you (sg.) already he He sees you (sg) now In the construction of this phrase, we begin with a sentence-initial host consisting of a single word nakikita. Within the syntactic system, this comes to be followed by a particle =na as a consequence of the application of “derivational” cliticization. The structure in (6.50a) thus becomes that of (6.50b) at the output of the syntax. (6.50) a.

b.

PWord Ft

σ na

PWord Ft

Ft

σ ki

σ ki

LexWord

σ ta

Ft

σ na

Ft

σ ki

σ ki

LexWord

σ ta

σ =na

172

Aspects of the Theory of Clitics

The pronominal clitics are then introduced into this structure. When the pronoun is disyllabic, it must be parsed as a foot, and thus has to follow the entire lexical word plus particle sequence. The latter is (as a whole) a PWord (perhaps derived by previous instances of Stray Adjunction incorporating particle clitics into the PWord deﬁned by the host). The clitic must follow this entire sequence because (as shown in 6.48a above) the only place where a foot can undergo Stray Adjunction is at the right edge of a PWord. Where the pronominal clitic is monosyllabic, on the other hand, it is only a syllable, not a foot. Potentially it can (and thus, to avoid unnecessary violations of LeftMost, must) be placed further to the left within the word. These clitics must be placed at the right edge of a foot, so that they can undergo Stray Adjunction as in (6.48b). They must also not violate higher-ranking Integrity constraints, including the integrity of lexical words, but they must otherwise be as close as possible to the left edge without being initial. As a result of the interaction of these factors, they will lodge at the right edge of the leftmost foot that is not properly included within a lexical word (or other integral domain). This means, in particular, that they appear to the left of the particles. These two types are both present in (6.51). (6.51)

PrWd Ft Ft Ft

Ft

Ft

σ

σ

σ

σ

σ

σ

σ

σ

na

ki

ki

ta

=ka

=na

=ni

ya

LexWord (6.51) assumes that the added monosyllable =ka is simply adjoined to the innermost foot within the already complex foot [ [ ki ta]—=na]. Depending f f on the details of stress contours in such a conﬁguration, this analysis might well be replaced with one where the two adjoined syllables are restructured to establish a new foot of their own, as in (6.52). (6.52) [

[ na ki][ ki ta ][ ka na][ ni ya]]

PWord Ft

Ft

Ft

Ft

I do not attempt to develop the mechanical details of this reﬁnement of the proposal here, but they do not appear to present conceptual problems. The phonology of Stray Adjunction, then, taken together with the constraints that deﬁne second position, can actually explain why monosyllabic clitics behave differently than disyllabic ones. It is important to note that this

An Optimal Theory of Clitic Positioning

173

result depends crucially on the constraint-based approach adopted here. That is, on the present picture, instead of specifying a unique position in hierarchical (or linear) structure in which clitics are generated or to which they move, the present analysis says that clitics are attracted as far to the left within the relevant domain as possible, subject to the higher-ranking requirement that they not be absolutely initial. The deﬁnition of “second position” that results is a consequence of the interaction of a number of other effects (especially the details of the language’s phonology and of the relevant Integrity constraints). Since those factors play out differently in the cases of monsyllabic and of disyllabic clitics, these elements appear in different positions, in ways that could not be described adequately by syntactic mechanisms per se. The Nature of the NonInitial Effect To this point, we have assumed that the Tagalog clitics—the particles of Figure 6.5 and the pronouns of Figure 6.4 alike—are to be characterized collectively as “second-position” elements in the sense that for each clitic cli there is a constraint NonInitial(cli , D) that outranks an otherwise highly ranked constraint LeftMost(cli , D). We noted in section 6.1, however, that there might be languages in which the “NonInitial” effect results not from an independent constraint, but rather from phonological requirements alone. Could this be the case for the Tagalog clitics? We might be able to examine a possible difference between the effects of NonInitial(clj ) and a potentially phonological effect if we could ﬁnd a circumstance where the domain D in which some clitic is subject to NonInitial(cli , D) phonologically includes preceding material that is part of a distinct syntactic domain D . In that case, we might expect that the clitic would appear initially within the domain D; while if NonInitial(cli , D) in fact outranks LeftMost(cli , D), the clitic should continue to be post-initial within D. It is always difﬁcult to be absolutely certain that the phonological adjacency necessary to establish this kind of situation is in fact met, but there is one construction in Tagalog that appears to provide us with evidence of the type we seek. A range of inversion constructions (described by Schachter and Otanes 1972: 485ff) involve the movement of a topical constituent to an initial position, presumably Spec(CP). When this happens, there may be a pause between the topic phrase and the rest of the sentence (which would, I assume, block stray incorporation of an IP-initial clitic). Such a pause need not be present, though, especially in the case where the topic is marked by the element ay (presumably in C). It would appear that in this case the preceding phrase would provide an appropriate prosodic host for an IP-initial clitic, if the latter element were not subject to NonInitial(cli , IP), parallel to the facilitating role of a preceding

174

Aspects of the Theory of Clitics

connected sentence in Warlpiri examples like (6.11). In such sentences in Tagalog, however, a pronominal clitic continues to appear in second position within IP, as in (6.53) below. (6.53) Ang sulat ay tinanggap =ko kahapon the letter ay received 1sg yesterday I received the letter yesterday (Schachter and Otanes 1972: 486) This is at least weak evidence that the position of pronominal clitics is in fact governed by high-ranking grammatical constraints of the type NonInitial(cli , IP). Furthermore, there is no evidence of the sort suggested for Warlpiri which would support the claim that this phenomenon is purely phonological in origin. When we consider corresponding data involving particle clitics, however, a rather surprising fact emerges. Consider the pair of sentences in (6.54). (6.54)

a. [CP[IP sasayaw =ba =sila ng pandanggo dance Q they fandango bukas ng gabi?]] tomorrow night Will they dance a fandango tomorrow night? =ba ng gabi ’y [IP sasayaw =sila b. [CP bukas tomorrow Q night ay dance they ng pandanggo?]] fandango Will they dance a fandango tomorrow night?

In sentence (6.54a) we see the particle ba preceding the pronoun sila “they” as expected, both coming in second position. In sentence (6.54b), however, where the phrase bukas ng gabi “tomorrow night” appears in fronted topic position, the two clitics are separated. Sila appears as expected in second position within the nucleus of the clause, but ba appears within the topic phrase. We cannot account for the facts in (6.54) by saying that particles, unlike pronouns, are not subject to NonInitial(cli ,IP) and are only constrained to be non-initial within IP by phonological factors. If that were the case, we would expect the particle to appear at the left edge of IP—which it cannot, as shown by the ungrammaticality of (6.55). ng gabi ’y [IP =ba sasayaw =sila (6.55) *[CP bukas tomorrow night ay Q dance they ng pandanggo?]] fandango (Will they dance a fandango tomorrow night?)

An Optimal Theory of Clitic Positioning

175

The fact that the particle appears embedded within the phrase bukas ng gabi (following its ﬁrst word) in (6.54b) shows that it must in fact be located with respect to the left edge of a larger, including constituent (probably CP). In non-topicalized main clause constructions, the left edges of IP and CP will coincide, and so the particles and the pronouns will form a single sequence located in the same “second position.” Where there is a preposed topic phrase, however, the difference between the two domains is revealed and the clitics become separated. One group of particles behaves like ba in (6.54) above, coming inside the topic phrase. Other particles, however, behave like the pronominal clitics in always appearing within IP, such as lamang in the sentences in (6.56). (6.56)

a. [CP[IP sasayaw =lamang =sila ng pandanggo dance just they fandango bukas ng gabi ]] tomorrow night They’ll just dance a fandango tomorrow night ay [IP sasayaw =lamang =sila b. [CP bukas ng gabi tomorrow night ay dance just they ng pandanggo]] fandango They’ll just dance a fandango tomorrow night =lamang ng gabi ay [IP sasayaw =sila c. *[CP bukas tomorrow just night ay dance they ng pandanggo]] fandango (They’ll just dance a fandango tomorrow night)

It seems, then, that the particles fall into (at least) two subclasses, one of which is subject to high-ranking NonInitial(cli ,CP) while the other is subject to NonInitial(cli ,IP). This difference is a property of the particles as grammatical elements, and does not follow simply as a consequence of their phonological form. According to Schachter and Otanes (1972: 429ff), the particles that behave like ba in (6.54) include both monosyllabic elements (ba, man) and disyllabic ones (kasi, kaya). Similarly, the set that are excluded from sentenceinitial topic phrases includes not only disyllabic particles like lamang, muna but also monosyllables (na, pa). A third subset, including both monosyllabic and disyllabic items, can appear in either position, depending (apparently) on considerations of semantic

176

Aspects of the Theory of Clitics

scope.16 If this speculation is indeed correct, it points the way to the essence of the differences we ﬁnd here. Assuming that particles are introduced within a domain that constitutes their scope, as we might wish if the operation Merge is to be semantically compositional, we could say that the relevant LeftMost and NonInitial constraints are relativized to the same domain, whichever it is in a given sentence. The domain of introduction of pronominal elements is presumably uniformly IP, so their placement in second position within IP is consistent with this account. It seems, then, that there is no evidence supporting the notion that the non-initial position of any of the Tagalog clitics follows from phonological considerations alone. Instead, each clitic’s location is governed by the interaction of appropriate LeftMost and NonInitial constraints, relative to the grammatical domain (either IP or CP, depending on the clitic) within which it is introduced. Legendre’s (2000a) suggestion that Tagalog clitics are placed in second position within a phonological constituent (speciﬁcally, a phonological phrase) rather than a grammatical one is thus not conﬁrmed. Tagalog thus presents a particularly interesting and intricate illustration of the interaction of a variety of factors in determining the exact location of “second-position” clitics: (6.57)

a. A Constraint system, including i. “LeftMost” alignment constraints; ii. “Non-Initial” alignment constraints within an appropriate domain (IP or CP); iii. Hierarchical ranking of alignment constraints for speciﬁc clitics; iv. Integrity constraints, which characterize uninterruptible subparts of a domain and thus deﬁne what counts as occupying “ﬁrst” position; b. An architecture of grammar in which “derivational” clitics result from an operation of Merge, while “inﬂectional” ones realize the functional categorial morphology of phrases and clauses; and c. Phonological effects, in particular the details of Stray Adjunction phenomena.

This complex picture has a natural place within a theory that draws on the basic ideas of Optimality Theory, such as the research program developed in the present work. 16 See Condoravdi and Kiparsky (1998) for a discussion of the interaction of scope with alignment constraints in the placement of afﬁxal material in the Tagalog word.

7 Verb Second as Alignment The locus classicus for discussion of second-position phenomena is surely Wackernagel (1892). Most of this paper is dedicated to the special clitics of the early Indo-European languages, and on the basis of the preceding chapters, we now have a theory of the grammatical mechanisms behind this variety of special clitic positioning (among others). But Wackernagel’s paper raises another point, one for which it is rather less often cited. This is the question of whether the notion of “second position” wherever it arises in grammar has a uniform basis. In particular, reference to “second position” comes up in an area that seems quite remote from the analysis of special clitics: the placement of the ﬁnite verb in second position in German (and most other Germanic languages, apart from English). Wackernagel claimed explicitly that “verb second” constructions and second-position clitics were connected, and the goal of the present chapter is to explore the possibility that they do indeed have a common basis. Wackernagel’s own argument was simple and straightforward. Recall that, for him, “clitics” were precisely and by deﬁnition “unstressed words.” He showed that these consistently appeared after the ﬁrst word of the sentence in the earliest known Indo-European texts, a conclusion further conﬁrmed by material from the Anatolian languages that later became available. It seems, then, that unstressed words in proto-Indo-European and its immediate descendants were systematically attracted to second position within the clause. But, he argued, the ﬁnite verb in proto-Indo-European did not bear accent. This conclusion was supported by evidence from several of the early languages, where accent on ﬁnite verbs is either lacking (as in Sanskrit) or clearly a secondary development (as in Greek). From this it should follow that, as unaccented words, ﬁnite verbs should behave as clitics—and thus that they should also be attracted to second position. As a result, he suggested that the location of the verb in second position in modern German main clauses might well be a reﬂex of its earlier link to the broader class of clitics. Elegant as this explanation would be if it could be sustained, it has serious problems. For one thing, the verb-second construction has been argued

178

Aspects of the Theory of Clitics

(Kiparsky 1995) to be a comparatively recent innovation in Germanic, and not a property inherited from earlier Indo-European origins.1 And whatever the historical facts may be, this explanation does not really bear on the synchronic state of affairs in modern German, since (a) the language does not have second-position clitics, and (b) its ﬁnite verbs are not unstressed. Therefore, Wackernagel’s proposed synthesis cannot be literally correct as an analysis of German. But these objections do not allow us simply to dismiss Wackernagel’s point. Is it really just an accident that the notion of “second position” occurs in both of these areas of grammar? Is there in fact a common basis that would suggest that they should be uniﬁed in theoretical terms? Previous chapters have developed a theory of special clitics, within which “second position” is reconstructed in a speciﬁc way. If clitics are introduced into the phonological forms of phrases by a generalization of essentially morphological mechanisms, and ﬁnd their position within the surface form of sentences through a system of hierarchically ranked violable constraints, as proposed here, is there perhaps some relation between second-position clitics and verb second, as Wackernagel proposed? I begin by considering how the constraint-based account of special clitics suggested in this book relates to the standard analysis of verb second as involving movement, as instantiated in modern standard German. I then go on to extend the analysis to other Indo-European languages in which verb second is found, including Icelandic, Kashmiri, Breton, and Surmiran Rumantsch before drawing some general conclusions.

7.1 Verb Movement and Second Position A major theme in recent syntactic discussion is the notion that movement is forced by “morphological” considerations. Theories invoking “morphologically based movement” are not generally very closely related to any theory of morphology, however. The image is that of a scavenger hunt, where word formation has already done its work, the grammar has scattered inﬂectionally relevant features here and there within sentence structure, and it is a constituent’s task to visit all of these places and reassure itself about all of its own features. The actual workings of morphology hardly come into play in these matters, if at all. Perhaps, however, movement in some instances is based on more centrally morphological concerns than the need to check features. 1 It should be noted, though, that others (Eythorsson ´ 1996) have argued that verb second is actually somewhat older than Kiparsky suggests.

Verb Second as Alignment

179

Ignoring for the moment the speciﬁc conditions under which verb second is found in some language or another, the descriptive generalization which deﬁnes the verb-second construction can be stated as in (7.1). (7.1)

Verb Second: The verb which indicates the Tense, Mood and Agreement properties of a clause appears immediately after the ﬁrst constituent of the clause.

Another way to put this is: the formal markers of a clause’s relational properties appear as morphology on a verb which appears immediately after the the ﬁrst unit of the clause’s syntactic analysis. That is not very different from the descriptive regularity governing second-position clitics: these, too, appear immediately after the ﬁrst (non-“permeable”) consituent of the clause. And both ﬁnite main verbs and (many) sentential clitics realize a clause’s inﬂectional features. From this perspective, a verb-second language such as Icelandic differs from a second-position clitic language like Tagalog in that the clause’s grammatical features are realized by the inﬂectional form of the verb in Icelandic, but in part by phrasal afﬁxes (the Agreement clitics) in Tagalog. This point is made by Legendre (2001), who argues along lines very similar to those developed below that a single set of constraints on the location of the functional content of the ﬁnite clause governs the location both of the verb (in verb-second constructions) and of clitics such as those expressing tense and agreement. Taking the analogy seriously, how can we relate the properties of verbsecond structures to the regularity of second-position clitic placement? In the present framework, this is straightforward. We describe second-position clitics as occurring as far to the left within their domain as possible without being at the absolute left edge. We can characterize the position of the verb in a verbsecond structure in parallel fashion by saying that the locus of realization of Tense, Aspect, Agreement, etc., is constrained as in (7.2).2 (7.2)

NonInitial(Vﬁn,S) LeftMost(Vﬁn,S)

This is obviously similar to second-position clitic placement, but what does it mean to think of verb second in terms of constraints? Basically this: a structure in which constraints are met is to be preferred to one where they 2 Legendre (2001) formulates the corresponding constraints as applying to the speciﬁc features of Tense, Aspect, Agreement, etc. When those features are expressed on the verb, the result is verb second; when they are expressed by special clitics, these appear in second position. The difference between that formulation and the present one is not essential to the immediate issue of establishing that such a parallelism exists.

180

Aspects of the Theory of Clitics

are not. When constraints conﬂict, their ranking determines which of them is more important to observe. We thus want the ranked constraints in (7.2) to serve as a mechanism to choose the correct structure from among alternatives provided by the syntax. In the syntactic literature of recent years, a derivation of verb-second clauses as sketched in the Phrase Marker of Figure 7.1 has come to be widely accepted. Let us make the common assumption that German clauses are basically verb ﬁnal, with the inﬂected verb moving from ﬁnal to second position (I omit here the displacement of the Subject from a position internal to the VP into Spec(IP)). Movement of the verb to I in Figure 7.1 simply represents the fact that this verb is the one that bears ﬁnite inﬂection—such movement is not needed on a view where functional content propagates as features rather than constituting a head of its own, on which VP and IP may not be distinct. I include it simply so as to reproduce the standard analysis as closely as possible. Movement of the AdvP gestern ‘yesterday’ to initial (Spec(CP)) position in Figure 7.1 presumably represents an instance of topicalization, and is motivated by the information structure of the utterance. But the crucial question in this analysis has always been the reason for movement of the inﬂected Verb from I to C. Does such movement take place because of some property of the C position itself ? Proposals have been made to force this movement by positing

CP(=S) C

Spec C

IP(=S) I

Spec

I VP I AdvP

V

Gestern yesterday

habe have

I

VP NP

AdvP

NP

ich t eine Zeitung I a newspaper ‘Yesterday I read a newspaper’

V

V

gelesen read

t

Figure 7.1. Movement in the German verb- second construction

t

Verb Second as Alignment

181

some feature-checking relation between the positions Spec(CP) and the verb in C, or by requiring C to be lexically ﬁlled, or in other ways; but these all amount to camouﬂaged language-particular stipulations of the requirement “Move I to C.” The movement remains, accordingly, without independent motivation. The proposal here is that this movement is not motivated by the need to check some abstract feature or the like, but rather by the fact of second position itself. Movement to this position occurs when (a) it is syntactically possible, violating no constraints of the syntactic computational system; and (b) the structures that result violate relevant Alignment constraints such as (7.2) to a lesser degree than structures in which such movement has not taken place, so that the ﬁnite verb would otherwise be farther from the left edge of the sentence than in the preferred structure. The theory of verb second espoused here does not per se deny that verb second involves movement of the Verb into a C position, though it does not require that speciﬁc landing site for movement, either. It simply proposes that the reason for any such movement is to get the verb to a position as far to the left as possible, without being absolutely initial, rather than to check a feature, or to ﬁll the C position, or to meet any other such requirement. The thrust of the proposal concerns the motivation for the movement, and not its mechanics. Its essence is the notion that “second position” is the real generalization about the position of the ﬁnite verb and not an epiphenomenal consequence of movement carried out for some other purpose. The similarity of this analysis of verb second to the treatment of secondposition clitics developed in Chapter 6 is apparent. In the case of special clitics in second position, I argued that what uniﬁes them is not occupancy of a consistent hierarchical position in phrase structure, as syntactic movement accounts predict (or assume), but rather the fact that they are in second position—as far to the left as they can get without being at the actual edge. Once we take seriously the proposal that verb second and second-position clitics result from essentially similar mechanisms, however, another question arises. If verb second results from a “morphological” imperative, akin to that governing clitic placement, and not from a purely syntactic requirement, why do we not ﬁnd languages in which the ﬁnite verb appears (at least as an option) immediately after the ﬁrst PWord of its clause? In every known case, “verb second” treats as its anchor the ﬁrst syntactic daughter within the domain in which the verb appears, rather than the ﬁrst prosodic constituent of some type (such as a PWord). There are nowhere near as many languages with productive verb second to consider as those that instantiate second-position clitics, but it is still striking that a position deﬁned in prosodic terms is apparently not available for verbs while it is for clitics.

182

Aspects of the Theory of Clitics

The answer to this apparent conundrum is that the syntax cannot in general access a position deﬁned in this way. In this respect, the mechanisms that implement the placement of special clitics and of the verb are fundamentally different, although the two may interact with similar constraint systems. Special clitics, on the one hand, are placed by “afﬁxation” rules that modify the phonological shape of the form. It is thus possible for these rules to introduce material anywhere in the structure, subject only to the Integrity constraints that prevent insertion into units of some particular (prosodically or syntactically deﬁned) sort. Verb second, on the other hand, is the result of syntactic movement, and the only structures that the syntax provides for comparative evaluation by the constraint system are ones that instantiate well-formed syntactic operations. Since these depend exclusively on the syntactic analysis of a structure, and not on its prosody, there is no way for a prosodically deﬁned position to play a role directly in syntactic movement. Recall the way in which a constraint-based system works. On such a theory, the computational system involves (1) a subsystem (usually referred to as “gen” in the OT literature) which produces a set of formally possible structures; and (2) a set of constraints on output form that evaluate these candidate structures with respect to a hierarchically organized set of requirements so as to choose the optimal one. It seems reasonable to assume that gen incorporates universal notions of syntactic well-formedness, in that only structures conforming to fundamental principles such as those of X-theory, and involving syntactically permissible displacement relations, are presented for comparative evaluation by the constraint system. Thus, the only structures that are available for evaluation at PF (with respect to constraints such as LeftMost(Vﬁn,S) and NonInitial(Vﬁn,S)) are ones that are syntactically well-formed in terms of these general principles. Syntactic movement of the verb to a position deﬁned as following a sentenceinitial PWord, where that may be internal to a larger containing phrase, will then be disallowed by the general nature of movement. Candidate structures of this sort are not found in the output of gen, and thus no language preferring them could exist. On the other hand, no such limitation applies to the candidates gen provides for evaluation in the case of special clitics. Since these are (on the present theory) instances of phrasal morphology, the input to the system that places them includes both a prosodic and a syntactic analysis, just as word-level morphology operates on the basis of both phonological and morphosyntactic representations of the word. As a result, clitic placement can “see” a position deﬁned prosodically, while syntactic movement cannot.

Verb Second as Alignment

183

Now notice that this provides us with an account of the “asymmetric” nature of verb second in German. As is well known, verb second does not take place in German subordinate clauses just in case they have an overt complementizer, as illustrated in (7.3). (7.3)

a. Hans behauptet, daß er davon nichts gewußt habe Hans maintained that he thereof nothing known have Hans maintained that he knew nothing about it b. Hans behauptet, er habe nichts davon gewußt Hans maintained he have nothing thereof known Hans maintained he knew nothing about it

Suppose that the Verb Movement giving rise to verb second in main clauses is in fact a movement from I to an otherwise empty C, and that this C position is in fact the only possible target of syntactic movement of the verb in German. It then follows that we should fail to ﬁnd verb second exactly where the relevant movement is blocked by general conditions—and of course when the C position is ﬁlled by an overt complementizer like German daß, this will be the case. That is, where C is already ﬁlled, gen does not provide a candidate structure in which the ﬁnite verb is displaced from clause-ﬁnal I position, because there is no well-formed movement available that would reduce the violation of LeftMost(Vﬁn,S). This is essentially the same explanation that is standardly offered for the asymmetry within other syntactic accounts, and it extends quite naturally to the present view. The proposed analysis for verb-second constructions is thus based on the comparative evaluation of well-formed outputs of the underlying syntactic computational system with respect to a constraint system including (7.2). In the remainder of this chapter, I will examine the applicability of this account to verb-second phenomena in a range of other languages. These are all drawn from (distinct subfamilies within) Indo-European, but they appear to be independent of one another, and not simply diverse modern reﬂexes of a single original type.

7.2 Icelandic If the view just suggested is correct, we might expect to ﬁnd cases in which verb second is enforced, even though the structural identity of the position in which the Verb winds up is not uniform (apart from being second within the clause). That seems to be the case in Icelandic, a language that displays verb

184

Aspects of the Theory of Clitics

second symmetrically, in both main and subordinate clauses. Consider the sentences in (7.4). (7.4) a. Jón harmar að þessa bók skuli ég hafa lesið John regrets that this book shall I have read John regrets that I have read this book að það hefur enginn lesið bókina b. Ég veit I know that there has no one read the book I know that no one has read the book The ﬁnite verbs are italicized in these examples, while the initial elements which they follow within the sentential domain are underlined. These instances of verb second suggest that in Icelandic a ﬁnite verb may follow (a) the subject; (b) a preposed, topicalized XP; or (c) an expletive (such as það). There is also a further possibility, illustrated in (7.5). (7.5)

a. Ég hélt að kysst hefðu hana margir stúdentar I thought that kissed had her many students I thought that many students had kissed her b. *Ég hélt að hana kysst hefðu margir stúdentar I thought that her kissed had many students I thought that many students had kissed her

Sentence (7.5a) illustrates “Stylistic Fronting,” a construction with wellestablished and distinctive characteristics (Maling 1980; Jónsson 1991) quite separate from those of Topicalization. It involves the movement to a position before the inﬂected verb of a single word (participle, negative ekki, or certain adverbs). Stylistic Fronting displays two notable characteristics. First of these is the fact that a single word, rather than a phrase is moved. Second, Stylistic Fronting is constrained by a “Subject gap condition”: it is possible exactly when the subject position is not overtly occupied. This is the case in impersonal sentences, which may receive an expletive subject in their surface form if nothing else intervenes, or as a consequence of extraction, postposing, or any other operation rendering the subject position unﬁlled. These various possibilities, and the consequences of ignoring the Subject gap condition in similar structures, are illustrated in (7.6). (7.6) a. i. Honum mætti standa á sama, hvað sagt væri to him would be the same what said would be um hann about him It would be all the same to him, what was said about him

Verb Second as Alignment

185

ii. *Honum mætti standa á sama, hvað sagt hefði Hjördís to him would be the same what said had Hjördís um hann about him It would be all the same to him, what H. had said about him b. i. Hann er sá eini, sem ekki er líklegur til að koma he is the only that not is likely to come He is the only one that is not likely to come ii. *Hún spurði, hvort líklegur væri hann til að koma she asked whether likely would be he to come She asked whether he would be likely to come c. i. Þetta er bærinn, þar sem fæddir eru margir frægir this is the town where born are many famous Íslendingar Icelanders This is the town where many famous Icelanders were born ii. *Þetta er bréﬁð, sem ekki skrifaði Helgi this is the letter that not wrote Helgi This is the letter that Helgi didn’t write Let us assume, as seems uncontroversial, that Icelandic has basic SVO order, and that complete clauses have the structure of (7.7). (7.7) [ (XP) (X) [ (Subject) [ (Verb) . . . ]]] V

The “X” and the “XP” in (7.7) might be “C” and “Spec(CP),” but nothing hinges on that categorial analysis. All we require is that two clause-initial positions are available, one of which can contain a simple head and the other of which can contain a phrasal category. It might be objected that in the sentences of (7.4), the subordinate clauses are already introduced by a complementizer (að). As a result, it is impossible that they should have the structure of (7.7), since the “X” position there is already ﬁlled by að. Precisely this aspect of verb second in Icelandic has led some authors to propose that in languages of the “symmetric V-2” type, the category CP is recursive, making available a new instance of “X” in (7.7). It is not necessary to make that move however. Instead, we can regard að as a marker of subordinate clauses, rather than a complementizer. Speciﬁcally, we can treat að as an initial (special) clitic within subordinate clauses, introduced at the left edge of the structure in (7.7) when this appears in embedded positions, and not occupying the structural position “X” within that structure.

186

Aspects of the Theory of Clitics In these terms, we can describe verb second in Icelandic as in (7.8).

(7.8) Icelandic V-2: Tense/Aspect and Agreement are realized on the main verb, where they appear as its inﬂection. Rank NonInitial(Vﬁn,S) above LeftMost(Vﬁn,S) (and both high), thus preferring a structure in which the ﬁnite Verb appears immediately following the ﬁrst overt syntactic daughter of the clause. There are several ways to satisfy the conditions in (7.8). First, it might be that nothing moves from the core of the clause into the available XP or X positions. In this case everything remains in its “natural” position, and the verb is second by virtue of following exactly the subject DP. Second, it might be the case that some phrase moves into the available XP position (presumably as a topic or for some similar information–structurally motivated reason), and the verb itself moves to the available X position. In this case, the verb is in second position by virtue of following exactly the topic phrase. Note that these two situations are distinct. Sentences with basic SVO constituent order may represent prominently “topical” subjects, but on the other hand, they may be informationally neutral. When something other than the subject appears in sentence-initial position, however, it must be interpreted as topical. As a third possibility, it may be that no phrase moves to the topic XP position, but the subject position is not overtly ﬁlled by lexical material. This might be the case in impersonal sentences, or where the verb comes to precede the subject through an operation of postposing (if this a possibility, as assumed in much of the traditional, descriptively oriented literature). In these cases, an expletive is inserted in the pre-verbal X position, as illustrated in (7.9). (7.9) Það eru margir frægir Íslendingar fæddir í þessum bæ expl are many famous Icelanders born in this town There were many famous Icelanders born in this town Finally, in the case where no phrase moves to topic XP, and there is also a gap in the subject position, Stylistic Fronting can move a participle or other element to the X position, as in the sentences of (7.5) and (7.6). What unites the position of the verb in all of these cases is not that it appears in a consistent conﬁgurational position, but rather the fact that in each instance, the verb is second within its clause. This is exactly the claim of an analysis on which verb second results from a mechanism similar to that governing second-position clitics: compare these facts with the lack of a consistent structural position occupied by second-position clitics in a variety of languages cited by authors such as Boškovi´c (1995) for Serbo-Croatian, and the broader discussion of this matter in Chapter 5.

Verb Second as Alignment

187

This account also provides an immediate way to derive the Subject gap condition, rather than leaving it as a stipulated condition on Stylistic Fronting. Where Stylistic Fronting applies, it has the effect of ﬁlling the X position in (7.7). If an overt subject itself is still in place, however, this results in the verb no longer being second within its clause. The verb cannot move to the X position, since this is already ﬁlled. It also cannot move to the topic (XP) position, since if it did so (even disregarding the problem posed by a head moving to a phrasal position), it would then be initial rather than second. Stylistic Fronting can only take place where it leads to a preferred structure, and thus is only possible when the subject position is unﬁlled. The analysis of verb second as parallel to the positioning of secondposition clitics thus seems motivated for Icelandic, a core instance of a language displaying this property. I now move on to other languages which have been less commonly cited in this regard.

7.3 Kashmiri The literature on the verb-second construction has concentrated largely on Germanic (for the very good reason that nearly all of the languages of this family display it in some form). As I will discuss in subsequent sections of this chapter, verb second is also found in early stages of Romance and in at least one modern Romance language (Rumantsch), and related word-order effects are also found in Breton within Celtic. At least one language in the Indic family also shows a form of verb second that we can compare with the Germanic phenomena, however: Kashmiri.3 Following Hock (1982, 1991), Bhatt (1999) suggests a scenario for the development of verb second in this language that links it to the properties of clitics. Brieﬂy, this goes as follows: Kashmiri (like many other Indic languages) has developed a large number of verbs that are normally conjugated with the aid of an auxiliary element. Middle Indic acch- ‘be’ underwent reduction in Kashmiri to a clitic form with base ch- , resulting in its being placed in second position along with other clitics. As a later development, the positioning of non-clitic ﬁnite verbs was assimilated to this, resulting in verb second. Much of this 3 Hendriksen (1986, 1990) reports that in at least two “Himachali” languages, Kotgarhi and . . Koci, he ﬁnds word-order patterns quite similar to those of Kashmiri. Verb second is reported as the usual order in main clauses and subordinate clauses introduced by “the conjunction /ki/ ‘that’ after verbs such as ‘to say’ ” (Hendriksen 1990: 164), and verb-ﬁnal order in other subordinate clauses. It is quite difﬁcult to determine the details of these patterns on the basis of Hendriksen’s descriptive materials, however, and I will not discuss these languages further. They are spoken in a region of north India near but not adjacent to that in which Kashmiri is the dominant language. Like Kashmiri, they are sometimes said to be part of a “Dardic” subfamily of Indo-Iranian, though modern scholarship treats all of these languages as Indo-Aryan.

188

Aspects of the Theory of Clitics

account remains speculative, but the resemblance to Wackernagel’s (1892) explanation of verb second is not hard to see. The basic verb-second construction in Kashmiri main clauses is illustrated in (7.10).4 (7.10)

a. me dits mohn- as yi k@mi:z ba:gas manz garden in I.erg gave Mohan- dat this shirt I gave Mohan this shirt in the garden b. yi k@mi:z dits me mohn- as ba:gas manz this shirt gave I.erg Mohan- dat garden in I gave Mohan this shirt in the garden dits me yi k@mi:z ba:gas manz c. mohn- as Mohan- dat gave I.erg this shirt garden in I gave Mohan this shirt in the garden yi k@mi:z d. ba:gas manz dits me monh- as garden in gave I.erg Mohan- dat this shirt I gave Mohan this shirt in the garden

As can be seen here, the ﬁnite verb immediately follows the ﬁrst constituent of the main clause. Surprisingly, this is also true for imperatives and yes/no questions, as shown in (7.11). (7.11)

di- t1 kita:b a. ko:r- i girl- dat give- pol.imper book (Please) give the book to the girl! gar1? b. (k’a:) ts1 gatsch- kh- a: paga:h tomorrow home (qpart) you go- 2sg- q Will you go home tomorrow?

When the verb itself is complex, composed of a ﬁnite auxiliary and a nonﬁnite lexical verb, the latter is generally found in sentence-ﬁnal position, as in the sentences of (7.12). (7.12)

a. ma:st.ar- an ch- u [email protected] as kita:b div- a:n teacher aux- m.sg boy- dat book give- nonperf The teacher gives a book to the boy

4 Kashmiri data here are drawn from Wali and Koul (1997) and Bhatt (1999). These two sources use different transcriptions for the language, unfortunately. Examples from Bhatt (1999) have been modiﬁed slightly to conform more closely to Wali and Koul’s transcription. Apparent phonological differences which remain are difﬁcult to resolve since Bhatt does not discuss the phonology of the language.

Verb Second as Alignment

189

b. tem- sund ch- u asi maka:n bad.1 pasand he- of aux- m.sg us.dat house very like We like his house very much Essentially the same facts obtain in subordinate clauses introduced by the complementizer ki ‘that’ as in (7.13). (7.13)

pata: ki yi k@mi:z dits ra:m- an a. me ch- u I aux- m.sg know comp this shirt gave Ram- erg mohn- as Mohan- dat I know that Ram gave this shirt to Mohan ba:sa:n ki su ch- u dili b. me ch- u I aux- m.sg believe comp he aux- m.sg Delhi.loc ro:za:n live.nonperf I believe that he lives in Delhi

On the basis of these facts, we could suggest an analysis (similar to that of Icelandic in section 7.2) along the following lines. From the position of the lexical verb in sentences like (7.12), it appears that Kashmiri has a fundamentally SOV organization. Let us assume that this basic clause structure is embedded in a layer of functional structure similar to that proposed for Icelandic, with an initial Speciﬁer position that can be ﬁlled by a phrasal unit, and an immediately following head position. We can then say that the initial topic phrases in (7.10) occupy the Speciﬁer position in this layer of structure, and the ﬁnite verb is displaced from its underlying ﬁnal position to occupy the corresponding head. The requirement that the initial topic position be ﬁlled when the ﬁnite verb is fronted is apparently so important that “[w]hen nothing in a ﬁnite matrix clause is topicalized, not even the subject which generally acts as the discourse topic by default, dummy yi is inserted” (Bhatt 1999: 97). Unfortunately, neither Bhatt nor Wali and Koul (who suggest the same thing) provides clear examples, although presentational sentences such as (7.14) may represent this phenomenon. (7.14)

yi o:s akh ba:dshah expl was art king (Once upon a time there) was a king

190

Aspects of the Theory of Clitics

Since we ﬁnd the same word order in the embedded clauses of (7.12), it cannot be the case that the functional layer in which topics and the ﬁnite verb appear is that of CP with ki occupying the C (head) position. It seems plausible, though, to take ki (or its alternate form zi) as a phrase-initial special clitic introduced initially in complement clauses, like að in Icelandic. The observed displacement of the ﬁnite verb can then be attributed to the operation of high-ranking LeftMost(Vﬁn) within the including functional domain, with the obligatory presence of an initial topic phrase due to even higher ranked NonInitial(Vﬁn) within the same domain. With the exception that its basic order within the VP is SOV and not SVO, Kashmiri then looks much like Icelandic with regard to the basis for its verb-second construction. A complication to this picture appears to be presented by the structure of wh-questions. In these sentences, the wh-phrase appears immediately before the ﬁnite verb as in (7.15). (7.15)

k’a: d’utnay tse ra:th a. ramesh- an Ramesh- erg what gave you.dat yesterday What did Ramesh give you yesterday? b. tse k’a: d’utnay ramesh- an ra:th you.dat what gave Ramesh- erg yesterday What did Ramesh give you yesterday? tse ra:th c. ? k’a: d’utnay ramesh- an what gave Ramesh- erg you.dat yesterday What did Ramesh give you yesterday?

If (7.15c) were the standard form of wh-questions, we could account for this simply by saying that wh-phrases move to the same position as topicalized elements. Unfortunately, this cannot be the case: while this order is marginally possible, the order of the other sentences in (7.15), with a topic phrase preceding the wh-phrase, is considered much better. Indeed, examples such as (7.16) in which expletive yi ﬁlls this role in questions are provided in both Wali and Koul (1997) and Bhatt (1999). (7.16)

yi kus o:s bar- as expl who was door- at Who was at the door?

Apparently, then, movement of a wh-phrase to the sentence-initial topic position is not the normal way to satisfy the requirement that wh-phrases must

Verb Second as Alignment

191

immediately precede the ﬁnite verb. Furthermore, the presence of a pre-verbal wh-phrase does not by itself appear to eliminate violations of NonInitial(Vﬁn). I propose, then, that instead of moving to a structural speciﬁer position, wh-phrases are adjoined to the left of the ﬁnite verb, resulting in a structure such as Figure 7.2. In multiple-wh questions, more than one wh-phrase can adjoin to the verb in this fashion. Alternatively, one or more (but not all) of the wh-phrases can remain in situ, as in (7.17b). (7.17)

a. rameshan k@m’is k’a: d’ut Ramesh.erg who.dat what gave What did Ramesh give to whom? b. rameshan k’a: d’ut k@m’is Ramesh.erg what gave who.dat What did Ramesh give to whom?

Assuming that the complex V created by adjoining one or more wh-phrases to the ﬁnite verb is the constituent whose position is constrained by NonInitial(Vﬁn), the data from wh-questions fall together with those from declaratives under this analysis. There is still one further set of facts to consider, however. In Kashmiri relative clauses, the ﬁnite verb appears in ﬁnal rather than second position, as in the examples of (7.18). XP X V DP

VP/IP

DP V DP

Adv

ra:j- an k’a: kor [e] ra:th Raj- erg what did yesterday What did Raj do at his home yesterday?

PP

DP

V

pan1ni gari his- at house

[e]

[e]

Figure 7.2. Movement in Kashmiri wh- Questions

192 (7.18)

Aspects of the Theory of Clitics yath kamras manz bat1 [email protected] yus ra:th a. su dem boy rel yesterday dem room in food kh’va:n o:s. . . eat.perf is. . . The boy who was eating food in this room yesterday. . . b. swa kita:b yos ts1 ra:th par- a:n o:suk. . . dem book rel you yesterday read- nonperf was. . . The book which you were reading yesterday. . . ch- a. . . c. yet’i swa bat1 ran- a:n where she food cook- nonperf aux- fsg. . . where she cooks food. . .

Relative clauses can occur as modiﬁers immediately following their associated head noun, as in (7.18), or else in initial position in the “correlative” construction of (7.19). (7.19)

yus no:kar tse ra:th samkhuy su no:kar dra:v rel servant you yesterday met cor servant left vun’ just now The servant who met you yesterday has left just now

Apparently, relative clauses (of both sorts) are based on a full clause structure. Instead of being embedded within the functional structure that supports topicalization and verb second, however, as in the other Kashmiri examples above, relative clauses are apparently embedded within a distinct structure in which the speciﬁer position is occupied by the relative phrase (e.g., yus ‘who’ or yus no:kar ‘which servant’). Perhaps the head of this structure is ﬁlled by an abstract operator corresponding to the anaphoric relation between the relative phrase and a gap in the relative clause; in any event it is apparently not a position to which the verb can move in the way it moves to the head of the topicalization structure in other sentence types. We can tell from these facts that the functional structure in non-relative constructions is not an instance of IP, such that the verb must be displaced to its head in order to inherit ﬁnite inﬂection. Relative clauses show us that the position in which ﬁnite inﬂection is assigned is clause-ﬁnal, and thus that there is no structural difference between VP and IP. Since the head position in a relative construction is not one to which the verb can move, and since there is no alternative position to which displacement of the verb would be well-formed in the syntax, the only candidate

Verb Second as Alignment

193

structures which can be evaluated with respect to the alignment constraints LeftMost(Vﬁn) and NonInitial(Vﬁn) are those in which the ﬁnite verb is ﬁnal within the VP/IP. While this results in violations of the LeftMost constraint, there is no well-formed way to reduce or eliminate those violations. This situation is, of course, exactly parallel to that in German subordinate clauses introduced by the complementizer daß, which also pre-empts the only site to which syntactically well-formed displacement of the verb from ﬁnal position could occur in such clauses. German and Kashmiri are thus both “asymmetric” verb-second languages, in the sense that verb second obtains in some structures and verb ﬁnal in others, although the dimension that deﬁnes the asymmetry is distinct in the two languages. In both cases, the motivation for the movement that takes place to yield verb second where possible can be seen as essentially the same set of Alignment constraints as those that are operative in second-position clitic constructions in other languages. As mentioned at the beginning of this section, the connection with second-position clitics may have historical roots in Kashmiri, though there is no longer a synchronic link between the two in the grammar either of that language or of those in the Germanic family.

7.4 Breton The present section is devoted to an apparent paradox in Breton word order. On the one hand, clause structure seems to be that of a VSO language like other Celtic languages; while on the other hand, most sentences in fact look like what we expect to ﬁnd in a verb-second language. The analysis that I will propose extends the account of section 7.1 to Breton. The results are quite similar to those of Hannahs and Tallerman (2000), and also to Legendre (2001), a paper sharing most of the assumptions and goals of the present chapter. A number of years ago, I made some proposals about Breton clause structure, beginning with those in Anderson and Chung (1977). A later revision of that analysis appeared as Anderson (1981), and served as the basis of some remarks on Breton Agreement in Anderson (1982). A substantial literature on Breton syntax has developed in the meantime, including Borsley (1990); Borsley, Rivero, and Stephens (1996); Borsley and Stephens (1989); Hendrick (1988, 1990a); Ortiz de Urbina (1994); Schafer (1995); Schapansky (2000); Stephens (1982, 1990); and Stump (1989), in addition to several descriptive grammars. Much of the theoretical discussion in that work is (directly or indirectly) critical of one or another proposal in my earlier analyses. In some respects I think these criticisms are well taken, though in others I am less sure of that. In any event, only a limited range of those issues will be relevant here.

194

Aspects of the Theory of Clitics

Breton As a Verb-Second Language In main clauses, Breton ﬁnite verbs are always preceded by another constituent, which may be the subject, an object or adverbial, an adjective, a non-ﬁnite verbal form, etc. In all but one of the sentences of (7.20) below, the ﬁnite form in question is the perfect auxiliary, disregarding some speciﬁc issues raised by the structure of this element (to be mentioned below). (7.20) a. Yann en deus debret e voued er wetur Yann he- has eaten his food in the car Yann has eaten his food in the car b. E voued en deus debret Yann er wetur his food he- has eaten Yann in the car Yann has eaten his food in the car c. Skuizh int da c’hortoz Tired they- are to wait They’re tired of waiting d. Debret en deus Yann e voued er wetur eaten he- has Yann his food in the car Yann has eaten his food in the car Remarkably for a “VSO” language, on the other hand, the verb itself cannot be ﬁrst, as illustrated by the ungrammaticality of (7.21). (7.21) *a/e d/t/zebr Yann e voued er wetur prt eats Yann his food in the car Yann eats his food in the car The element in initial position typically has some particular information value, such as that of a topic. When nothing in the sentence is appropriate for isolation in this way, there are two alternative constructions. One of these involves fronting the lexical verb, in its verbal noun form, with an inﬂected form of ober ‘do’ occupying the position of the verb. The other involves an empty expletive bezañ, homophonous with the verb ‘to be’ but syntactically inert in this usage. These two possibilities are illustrated in (7.22). (7.22)

a. Debriñ a raio Yannig krampouezh hiziv today eat (VN) prt will- do Yannig crêpes Yannig’ll eat crêpes today b. Bez’ e vo pesketaer Yannig a- hed e vuhez [BE] prt- he- will- be ﬁsherman Yannig all his life Yannig will be a ﬁsherman all his life

Verb Second as Alignment

195

c. Bez’ e tebr Mona he boued er gegin [BE] prt- eats Mona her food in- the kitchen Mona eats her food in the kitchen d. Bez’ eo bet labourer- douar Yannig daou vloaz [BE] is been farm worker Yannig two years Yannig’s been a farm worker for two years Breton, like English, has three verbs do: a lexical main verb (as in do the dishes), the pro-form of do so, and an empty auxiliary (as in I don’t like that as much as you do). Compound forms of one or another of these do verbs can also appear (with their ﬁnite part in immediate post-initial position) with a preposed lexical verb, as in (7.23). (7.23)

Debriñ en deus graet Yann e voued er wetur eat he- has done Yann his food in the car Yann has eaten his food in the car

Lexical verbs in initial position with a ﬁnite form of ober following are in what is called either the inﬁnitive or the verbal noun. This difference in terminology is actually quite important. The non-ﬁnite form in question may appear alone, as in (7.23), or with an accompanying object, as in (7.24). (7.24)

a. Debriñ krampouezh a raio Yannig hiziv eat crêpes prt- will- eat Yannig today Yannig will eat crêpes today b. Debriñ e voued en deus graet Yann er wetur eat his food he- has done Yann in the car Yann has eaten his food in the car

Forms in which the object accompanies the verb appear to be instances of topicalization of a VP, and Anderson and Chung (1977) suggested that the construction should be analyzed in that way. On that view, the verbal form is an “inﬁnitive” that heads a VP which may contain a DP complement. In Anderson (1981), however, I argued that what is actually involved is a nominalization with or without its complement. On this account, the verbal form is a “verbal noun” and the complement is an accompanying genitive. Arguments purporting to demonstrate the verbal nature of these phrases have concentrated on the facts of Welsh, and although the corresponding structures in that related language may well be headed by verbs, the arguments of Anderson (1981) to the effect that they are nominals have not been refuted (in my opinion), and I will assume that structure here.

196

Aspects of the Theory of Clitics

In compound tenses, a participle can be fronted by itself, but not together with its object (in either order), as shown in (7.25). (7.25)

a. Kollet am eus ma hent lost I- have my road I have lost my way b. *kollet ma hent am eus lost my road I- have I have lost my way c. *ma hent kollet am eus my road lost I- have I have lost my way d. Ma hent am eus kollet my road I- have lost I have lost my way

Although fronted participles may not take an object complement in that position, they may be accompanied by certain (common, short) adverbs, as in (7.26). (7.26) Debret mat en deus d’ he goan eaten well he- has for his supper He has eaten well for supper There is one further possibility: the ﬁrst part (ne) of the Breton twopart negation can also ﬁll this pre-verbal position. In negative sentences such as (7.27), nothing else can be topicalized. (7.27)

a. N’ en deus ket lennet Tom al levr neg he- has neg read Tom the book Tom has not read the book b. *ne lennet en deus ket Tom al levr neg read he has neg Tom the book Tom has not read the book c. *lennet n’ en deus ket Tom al levr read neg he has neg Tom the book Tom has not read the book d. *beza n’ en deus ket lennet Tom al levr Tom the book [BE] neg he- has neg read Tom has not read the book

Verb Second as Alignment

197

Breton sentences, then, generally have some element preceding the ﬁnite verb. This may be a topicalized phrase, a verbal noun, a participle, an expletive (bezañ), or a negative particle. The generalization appears to be that the inﬂected ﬁnite verb in Breton appears at the left edge of some constitutent—call this IP—and is preceded by exactly one element, which may be either a phrase or a single word (including the ﬁrst part of negation), but not both at the same time. The structural possibilities seem to be those offered in (7.28). (7.28) [ (XP) (X) [ Vﬁn . . . ]] CP

IP

The XP here can be ﬁlled with any maximal projection, including the verbal noun together with an optional complement. The X can be ﬁlled by a head such as a participle or an adjective, or by the ﬁrst element of the negation. What matters is that the ﬁnite verb appears at the very left edge of IP, but second within CP. Superﬁcially at least, this is very similar to the account of Icelandic offered in section 7.2, or other verb-second languages. How are we to describe these facts? Hannahs and Tallerman (2000) show that previous purely syntactic accounts involve either unmotivated diﬀerences in the position occupied by the ﬁnite verb from one sentence type to another, or else an arbitrary laundry list of licensing conditions for ﬁnite verbs. All of the various sentence types, on standard assumptions, involve (at least) movement of the verb. But we need to ask why the verb moves, as well as what else moves (or fails to move) and why. There are really two things to account for. First, we need to know how the position of the verb at the left edge of IP is related to its basic position, and (assuming the two are not identical) what motivates the displacement involved. Second, we need to know what principles govern (and limit) the appearance of pre-verbal material. This whole complex of facts and issues is typical for verb-second languages, although Breton shows some unique features of its own. Breton As a Verb-Initial Language Breton, like other Celtic languages, is typically said to be “basically” a VSO language, a claim that the reader may ﬁnd puzzling on the basis of the evidence supplied thus far. The evidence that is usually cited for this is the order which we ﬁnd in subordinate clauses, as in the clause in bold in (7.29). (7.29) me ’lavar deoc’h e oa ar marc’h-se re gozh I say to-you prt was the horse-dem too old I tell you that horse is too old

198

Aspects of the Theory of Clitics

Let us suppose that the head of the Breton IP precedes its VP complement,5 and that IP is found, in turn, inside the larger domain CP. How then do we diﬀerentiate the order in main clauses from that in embedded clauses? Assuming the structure in (7.28), what we need to do is require that within CP, exactly one of the positions C and SpecCP is ﬁlled. This eﬀect could be obtained by saying that the ﬁnite verb in Breton should be as far to the left as possible within IP (the “Verb-initial” constraint), and also within CP, but should not be absolutely initial within the CP. The relevant constraints and their ranking would then be as in (7.30). (7.30) LeftMost(Vﬁn, IP) NonInitial(Vﬁn, CP) LeftMost(Vﬁn, CP) The constraint LeftMost(Vﬁn, CP) ensures that no more material precedes the Vﬁn within CP than what is necessary to ensure that it is not initial within that constituent, and the high ranking of LeftMost(Vﬁn, IP) ensures that the Spec(IP) position is not ﬁlled. The position to which the verb moves is structurally uniform---I, initial within IP. The generalization about verb second is not a structurally uniform one, however: it requires that the ﬁnite verb not be initial within CP, but this is not due to a requirement that some particular pre-verbal position must be ﬁlled. Rather, exactly one pre-verbal position within CP must be ﬁlled, without regard to which one or what ﬁlls it. Thus far, my analysis basically follows that of Hannahs and Tallerman (2000). The domain within which NonInitial holds still warrants more discussion, though. If embedded clauses are CPs, we must ensure that verb second is not enforced there. Legendre (2001) suggests that the relevant diﬀerence does not depend on main versus subordinate status but rather on initial versus non-initial position within the entire sentence. The domain of NonInitial might then be a prosodic one, consistent with Legendre’s proposal that NonInitial (or “Tobler-Musaﬃa”) eﬀects are always relative to a prosodic rather than a grammatical domain. At a minimum, the category involved cannot be one including the entire sentence, since the second of two coordinated clauses often shows verb second as in (7.31). (7.31)

5

Me a breno ar sae, hogen c’hwi a wisko anezhi I prt will buy the dress but you prt will wear it I’ll buy the dress, but you’ll wear it

I return below to the internal structure of the VP in Breton.

Verb Second as Alignment

199

An alternative account of the (limited) asymmetry of verb-second eﬀects in Breton is to say that complements typically are simply IPs, not CPs, and that there is thus no C or Spec(CP) position available to be ﬁlled. Complements with overt complementizers (apart from the pre-verbal particles, which are themselves clitics and thus do not “count” in this respect) might well be CPs, but since their C position is ﬁlled, the constraints in (7.30) do not require (or permit) further topicalization. Another diﬃculty for the analysis of Hannahs and Tallerman (2000) arises from the facts of negative sentences, and this is less easily eliminated. I have already noted that in negative sentences, nothing need precede the negative particle ne; and in fact a preposed participle or other single head cannot appear in that position, as shown in (7.32). (7.32)

a. N’en deus ket gwellet Yann e vignonez hasn’t seen Yann his girlfriend Yann hasn’t seen his girlfriend b. Gwellet en deus Yann e vignonez seen has Yann his girlfriend Yann has seen his girlfriend c. *Gwellet n’ en deus ket Yann e vignonez seen hasn’t Yann his girlfriend Yann hasn’t seen his girlfriend

This is consistent with the consensus in the literature that the negative particle ne, unlike the other verbal particles a, e, occupies C or some other structural position and thus allows the ﬁnite verb to be non-initial within CP, while precluding the movement of other elements to C. But preposed topical material is found with negatives when that material is phrasal, as in (7.33). (7.33) Gwelout e vignonez ne reas ket Yann seeing his girlfriend didn’t Yann Yann didn’t see his girlfriend Hannahs and Tallerman propose to account for this by saying that another constraint, one requiring that ne appear in C, outranks all others, thus forcing ne to appear in this position even when there is a phrase in topic position. This does not really solve the problem, though. If ne is in the C position, the constraints in (7.30) will still prefer a structure without a topic phrase in SpecCP, since the presence of such a phrase will produce an additional violation of LeftMost(V,CP) that could be avoided by not having such a topicalized constituent. The problem is not that ne must occupy the C position: rather, it is that the

200

Aspects of the Theory of Clitics

associated Spec position should not simultaneously be ﬁlled, although (7.33) suggests that it may be. A clue to the solution of this diﬃculty lies in the facts of agreement in Breton. Verbs show inﬂectional agreement only when the subject is not overtly present, either in VP-internal subject position or in the position of topic. When the subject DP is overt within the CP of (7.28), the verb appears in a uniform non-agreeing form homophonous with the third-person singular. The facts are illustrated in (7.34). The mechanisms of this “dis-agreement” eﬀect are not immediately relevant here, since what matters is its structural characterization. (7.34)

a. Me a lenn al levr I prt read the book I read the book b. Al levr a lennan the book prt I-read I read the book c. Ar wazed a lenn al levr the men prt read the book The men read the book d. Al levr a lenn ar wazed the book prt read the men The men read the book e. Al levr a lennont the book prt 3pl-read They read the book

In negative sentences, however, if an overt subject appears in pre-verbal position, the Verb agrees with it, as in (7.35). (7.35)

a. Ne lennan ket al levr I-don’t-read the book I don’t read the book b. Me ne lennan ket al levr I I-don’t-read the book I don’t read the book c. Ar wazed ne lennont ket al levr the men they-don’t-read the book The men don’t read the book

These facts suggest that the subjects in sentences like (7.35) are not actually in the topic position in (7.28), where they would require null agreement,

Verb Second as Alignment

201

but rather in some position external to the structure of (7.28). If that is the case, though, the presence of this phrase would not produce violations of the verb-second regularity within (7.28). If initial subjects in negative sentences are outside of the clause, where are they? A number of authors have pointed out that the generalization of Anderson and Chung (1977) that only a single “topicalization” is possible in a given sentence is incorrect. Initial topic-like phrases are found even when some other element must be present in C or SpecCP, as in left-dislocated structures such as (7.36). (7.36)

a. Ha me, se en deus bet merket ac’hanon e-barz and I that has been marked of-me in ma vuhez my life And me, that marked me for the rest of my life b. Marhadizion amonenn ha uieu, deu e oent butter-and-eggs merchants two prt 3pl-were Butter and eggs merchants, they were two

It seems plausible to suggest that left-dislocated constituents do not occupy the position of Spec(CP), but instead are left-adjoined to the CP, in a position where they are not relevant to agreement. I suggest that pre-verbal subjects in negative sentences such as (7.35) are actually left-dislocated, and thus occupy this adjoined position rather than that of Spec(CP). In that case, they no longer pose a problem for the analysis of verb second as based on the constraints of (7.30). Two further complications for the analysis of verb-second regularities in Breton have their origins in the incompletely assimilated eﬀects of historical change. First is the behavior of the unusual verb emañ. This serves as a copula in locatives and related expressions; it also combines with a non-ﬁnite verbal form to express progressive aspect. Unlike other verbs in Breton, emañ can (and indeed, prefers to) appear in initial position, as in (7.37). (7.37)

Emañ va breur en ti be-loc my brother in the house My brother is in the house

Another peculiarity of this verb is that when preceded by a nominal topic, it is replaced by (a) zo, as in (7.38). This is the normal form of the copula used in non-situational contexts when preceded by a DP.

202 (7.38)

Aspects of the Theory of Clitics Va breur a zo en ti my brother prt be in house My brother is in the house

From a historical perspective, the anomalous behavior of emañ is straightforward. Its paradigm originates, essentially, as a combination of hemañ ‘here’ plus appropriate forms of the verb ‘to be.’ Presumably, the included element hemañ originally had the eﬀect of satisfying the requirement that the verb appear in second position. When an overt topic appeared, there was no longer a place for the ‘here’ element within the constraints of verb second. The copula would then be immediately preceded by the nominal topic phrase (presumably in SpecCP position), and this in turn would trigger the change in form to (a) zo that is quite general after an initial DP. There is little doubt, though, that emañ should be analyzed as a single verb in modern Breton, and not as a phrasal collocation; and that means that there is no obvious way to implement the generalization above by regarding a subpart of emañ as occupying the C position. This is the kind of puzzle that restructuring in historical change often presents for the goal of a perfectly consistent synchronic grammar. There does not seem to be any more elegant solution in this case than simply to stipulate that emañ itself must appear leftmost within CP, by means of a speciﬁc Alignment constraint (7.39) that outranks the general requirements in (7.30), and in particular the constraint NonInitial(Vﬁn,CP). Given the unique status of this verb, a more principled solution does not seem to be indicated. (7.39)

LeftMost (emañ, CP)

A further historically based peculiarity involves the verb kaout /endevout ‘to have.’ Unlike other verbs in Breton, this verb inﬂects for person not through suﬃxes, but rather by complex internal modiﬁcation. Finite forms (which bear no resemblance to the non-ﬁnite ones) are given in Figure 7.3. A syntactic peculiarity of this verb is the fact that, as shown in (7.40), it displays explicit agreement even with an overt subject, as opposed to the behavior of other verbs as illustrated in (7.34). (7.40) a. Lennet en deus Yann al levr read has (m) Yann the book Yann has read the book b. Lennet he deus Mari al levr read has (f) Marie the book Marie has read the book

Verb Second as Alignment

203

c. Me am eus lennet al levr I have (1sg) read the book I have read the book

1sg 2sg 3sg,m 3sg, f 1pl 2pl 3pl

Present am eus az peus en deus he deus hon eus hoc’h eus o deus

Future am bo az po en devo he devo hor bo ho po o devo

Figure 7.3. Finite forms of Breton kaout/endevout ‘to have’

The historical grounds for this behavior lie in the fact that the forms meaning “have” were originally composed of an inﬂected preposition plus a nonagreeing form of the existential verb. As a result, the position corresponding to subject person marking was that of a prepositional object. This, presumably, stood in a diﬀerent structural relation to the rest of the clause from the agreement material associated with other verbs. Again, however, it is diﬃcult to see how to implement this historically derived insight in the description of modern Breton, since the forms are synchronically those of a verb with an unusual internal form of agreement, and there is no other justiﬁcation for viewing it as a phrasal collocation. Breton VPs and the Structure of IP The basic structure of the clause in Breton is thus plausibly viewed as that in (7.28). But what of the internal structure of the IP in (7.28)? This presumably has a head position (I) at its left edge, to which the verb moves in order to realize the features of Tense and Agreement. An associated preceding SpecIP position could never be ﬁlled, by virtue of the high ranking of the constraint LeftMost(Vﬁn,IP). But what of the rest of the structure internal to IP? I assume here the VP-internal subject hypothesis (Fukui and Speas 1986; Koopman and Sportiche 1991; with roots in much earlier proposals of Fillmore and McCawley), on which the base position of the subject is within VP, the complement of I within IP. As to the internal organization of the VP itself, there is one major piece of evidence in favor of an SVO structure. This comes from the fact that non-ﬁnite clauses in the language, when they contain an overt subject, display SVO order as in (7.41).

204 (7.41)

Aspects of the Theory of Clitics a. Kavet am eus ur bluenn vat din da skrivañ gwelloc’h found I have a good pen for-1sg to write better I have found a good pen so that I can write better b. Pedin a reas Lennaig ar vugale dezho da zebrin invite prt did Lennaig the children for-3pl to eat krampouezh crêpes Lennaig invited the children to eat crêpes c. Da Yann da welout e oant holl aze For Yann to see prt were all there As far as Yann could see, they were all there

If we take this to be the basic structure, the occurrence of VSO will follow from Verb Movement to initial I, while movement of the reverse sort has no clear motivation. This establishes the basic structure of clauses in the language as that of (7.42). (7.42) [ (XP) (X) [ I [ CP

IP

VP

DPSbj [ V (DPObj)]]]] V

The VSO order observed within IP follows from the fact that in order to inherit (and thus realize) the functional content of the IP, the verb must move to this position. It cannot be preceded there by other material, since that would produce (uncompensated) violations of LeftMost(Vﬁn,IP). The verb-second properties of sentences as complete CPs then follow from the remaining constraints of (7.30) with the exception of sentences with the verb emañ, where (7.39) takes precedence and requires that element to appear initially.

7.5 Surmiran Rumantsch Among the modern Romance languages, the clearest example of a verb-second system is found in Rumantsch, one of Switzerland’s (four) oﬃcial languages.6 Whether this forms part of a larger Rhaeto-Romance family together with Friulian and Dolomitic Ladin is disputed (see Haiman and Benincà 1992; Liver 1999), but Rumantsch is clearly distinct within Romance from the various languages referred to as “Italian,” as well as from French, Franco-Provençal, and other surrounding members of the family. Opinions diﬀer as to the origin of its verb-second structure, even within the scope of a single work: one of the authors of Haiman and Benincà (1992), for example, believes that this was 6 The facts discussed here, though only parts of the analysis, are presented in Anderson (2004) together with further information on the language and its speakers.

Verb Second as Alignment

205

borrowed from German, while the other argues that it originated as an independent common northern Italian medieval pattern. Regardless of its origins, however, verb second is a pervasive and thoroughly assimilated aspect of the syntactic structure of Rumantsch, and thus important to examine in comparison with the other languages discussed in this chapter. Most Rumantsch speakers live in the canton of Graubünden, and nearly all are at least bilingual (mostly in German, though Italian is important in some parts of the canton). In the 1990 census, about 66,000 people indicated Rumantsch as the language of which they had the best command or which they most used, and of these 41,000 lived in Graubünden. I focus here on Surmiran, one of ﬁve diﬀerent literary standards that correspond roughly to linguistically distinguishable dialect areas. The amount of variation among these languages is signiﬁcant, but not extreme: most speakers of one can understand the others fairly well. Surselvan is probably the best known and best described, and Surmiran is thus a minority language even within the minority Rumantsch community, with about 3,200 speakers (Simeon 1996). Like other forms of Rumantsch, Surmiran is strongly verb second. The existence of verb-second constructions in the medieval Romance languages has long been recognized (Foulet 1928), and the analysis of these facts in medieval French (Adams 1987; Roberts 1993b; Vance 1997), Spanish (Fontana 1993) and Italian (Benincá 1994) has played a part in discussion of broader theoretical issues. Verb second in the “Rhaeto-Romance” languages of Northern Italy (Benincá 1985, 1994, 1995; Poletto 2000) has also ﬁgured in the syntactic literature, but much less attention has been paid to the facts of (Swiss) Rumantsch. In addition to verb second, Surmiran has a system of pronominal clitics, which will be discussed more fully in Chapter 8. Important to our purposes here, some of these are subject clitics, post-verbal clitic elements which under quite speciﬁc conditions appear in agreement with the subject. These are not simply phonologically reduced forms of full pronoun subjects, since they do not replace but rather may double the subject nominal. They are also not simply verbal morphology: verbs agree with their subjects quite independently of the presence of the clitics, and phonological and other considerations indicate that the two sets of markers are quite distinct. These two aspects of Surmiran syntax are closely interconnected, in that the subject clitics only occur when the full subject is inverted with the verb. Subject clitics and the grammatical mechanisms that implement verb second thus form part of a single complex. A ﬁnal factor in this mix is the impersonal subject element ins. At ﬁrst glance this appears simply to be a pronominal, equivalent to German man or

206

Aspects of the Theory of Clitics

French on. The interaction of ins with verb second, however, suggests that it has the properties of a special clitic rather than those of a simple pronominal. It also helps us establish just what “verb second” means in this language. Some Basic Facts of Surmiran Syntax I shall assume without explicit argument that the basic structure of the clause in Surmiran is SVO, as illustrated in (7.43).7 (7.43)

Ursus discorra stupent rumantsch Ursus speaks.3sg excellently Rumantsch Ursus speaks Rumantsch very well

In both main and subordinate clauses, the major constituents appear consistently in this order and are not subject to signiﬁcant scrambling or other variation, with one exception. In main clauses, a single topical constituent (typically an argument of the verb, or an adverbial) may appear in initial position. When a topical non-subject thus appears initially, the verb and the subject invert, leaving the verb in second position, as illustrated in (7.44). The remainder of the clause preserves the same order as in uninverted sentences. (7.44) a. Rumantsch discorra Ursus stupent Rumantsch speaks.3sg Ursus excellently Ursus speaks Rumantsch very well b. Stupent discorra Ursus rumantsch excellently speaks.3sg Ursus Rumantsch Ursus speaks Rumantsch very well Thus far everything is straightforward, but an additional possibility arises in the case of sentences with subject--verb inversion, such as those in (7.44). Here, the post-verbal subject can be doubled by a clitic element, as in (7.45). Such a clitic is not possible when the subject remains in initial, uninverted position. (7.45) a. Rumantsch discorra=’l Ursus stupent Rumantsch speaks.3sg-3sg.m Ursus excellently Ursus speaks Rumantsch very well 7 Surmiran examples are presented in the standard orthography, based on German and Italian conventions. The digraphs gl and gn represent palatal laterals [L] and nasals [ñ] as in Italian. Tsch represents a laminal alveopalatal affricate [Ù], as in German; tg is a similar but distinct affricate [tC] articulated apically, corresponding to a palatal stop in some other Rumantsch languages. The voiced correspondent of tg is written as g before front vowels. Voiced and voiceless coronal fricatives are distinguished phonologically but not orthographically, as are open and closed e and o (as in standard Italian). Stress is generally on the ﬁnal syllable if this contains a full vowel, or on the penult if the last vowel is schwa.

Verb Second as Alignment

207

b. *Ursus discorra=’l stupent rumantsch Ursus speaks.3sg-3sg.m excellently Rumantsch (Ursus speaks Rumantsch very well) Sentences like (7.45a) where a clitic appears together with an overt subject have an interpretation with mild emphasis on the doubled subject. The presence of the clitic, however, licenses the deletion of the subject (subject to referential recoverability in discourse, naturally) as in (7.46), in which case emphasis is no longer implied. stupent (7.46) Rumantsch discorra=’l Rumantsch speaks.3sg-3sg.m excellently He speaks Rumantsch very well Surmiran is not a pro-drop language in general, and subjects cannot be omitted in the absence of such a clitic, regardless of whether the conditions for inversion are present. These points are illustrated by the ungrammaticality of sentences such as those in (7.47). (7.47)

a. *Discorra rumantsch stupent speaks.3sg Rumantsch excellently (He speaks Rumantsch very well) b. *Rumantsch discorra stupent Rumantsch speaks.3sg excellently (He speaks Rumantsch very well)

These facts are not limited to sentences with third-person nominal subjects---the same pattern shows up in other persons as well, as in (7.48). mal rumantsch (7.48) a. Ia/*∅ discor (I) speak.1sg badly Rumantsch I speak Rumantsch badly b. Rumantsch discor ia/*∅ mal Rumantsch speak.1sg (I) badly I speak Rumantsch badly c. Rumantsch discorr=a mal Rumantsch speak.1sg-1sg badly I speak Rumantsch badly ia mal d. Rumantsch discorr=a Rumantsch speak.1sg-1sg I badly I speak Rumantsch badly

208

Aspects of the Theory of Clitics

The full set of post-verbal subject clitics is given in Figure 7.4. These are overt in all cases except the second-person plural. I will discuss their analysis further in Chapter 8, where I will show that they cannot be treated simply as part of the morphology of the verb. Summarizing the points above, these subject clitics are introduced optionally (and post-lexically) in constructions that call for subject--verb inversion.8 When present, subject clitics sanction the otherwise ungrammatical replacement of overt subjects by phonetically null pronouns. Such omission of the subject is not obligatory, however, and the clitics may be doubled by an overt nominal or pronominal subject phrase. Person/Number(/Gender) 1sg 2sg 3sg m 3sg f 3sg impersonal 1pl 2pl 3pl m/f

Subject clitic =a =t =’l =’la =(i)gl =s(a)9 ∅ =igl

Figure 7.4. Surmiran subject clitics

The Inversion Construction Verb second in Surmiran, as we have seen, involves the inversion of subject and ﬁnite verb and occurs when some other constituent precedes the subject. A single (non-subject) nominal phrase, PP, or adverbial can appear in initial position as in (7.49), yielding an interpretation on which the initial phrase has a mildly topical or focused interpretation.10 (7.49)

a. La steiva ò Ursus nattagea bagn the living room has.3sg Ursus cleaned well Ursus cleaned the living room well

8 Thus far, we have only seen inversion in the presence of a sentence-initial non-subject topic constituent. Other constructions will be discussed below which also call for inversion, and where the subject clitics also appear. 9 The ﬁrst-person plural clitic is =sa following verbs which have ﬁnal stress in the 1pl form, but (non-syllabic) =s after those verbs whose ﬁrst-person plural is (exceptionally) stressed on the penult. This includes (nous) ´ıschan “(we) are,” and is general in tenses other than the present indicative where the 1pl form has penultimate stress, such as the conditional (nous) cant´essan “we would sing.” 10 Examples of inversion below may have either a post-verbal clitic, a full nominal subject following the verb or both, with this focused or emphatic interpretation. Except where necessary, I will not differentiate these cases.

Verb Second as Alignment

209

b. Tar igl gi da Rummy vala igl joker adegna in the game of rummy is.worth.3sg the joker always 25 puncts 25 points In the game of rummy, the joker is always worth 25 points c. Giond ier a spass ò Ursus scuntro Ladina going yesterday for a walk has.3sg Ursus met Ladina While walking yesterday, Ursus met Ladina Only one such topicalized phrase is possible, as the ungrammaticality of (7.50) shows. (7.50)

*Ier la steiva ò Ursus nattagea Yesterday the living room has.3sg Ursus cleaned

Under the conditions of inversion, the (main or auxiliary) ﬁnite verb precedes the subject, accompanied by any clitic elements dependent on it as in (7.51). In addition to clitic object pronouns, this includes the element n(a), a negative particle which occurs together with post-verbal betg ‘not’ and other negatives. This particle is quite often omitted in spoken Surmiran, as ne is in spoken French. Apart from the position of the ﬁnite verb, other elements of the clause remain in the same relative position they occupy in uninverted sentences. (7.51)

a. Cleramaintg n’=ò=’l Ursus betg savia chegl Obviously neg-has.3sg-3sg.m Ursus not known that Obviously Ursus didn’t know that b. Ier seira n’=ans=ò Maria betg telefono Yesterday evening neg-1pl-has.3sg Maria not phoned Yesterday evening Maria didn’t telephone us

As we have seen, no more than a single constituent can precede the verb, whether or not it is inverted. From (7.51) we can see that clitics associated with the verb do not constitute additional constituents in the relevant sense. In compound tenses, the verbal past participle alone can appear at the beginning of the sentence. Although the participle represents the main lexical verb of its clause, and arguably heads a verbal phrase, complements of this verb cannot accompany it in initial position. They remain where they would be in an uninverted clause as shown in (7.52), although as the marginal status of (7.52e) illustrates, frequent adverbs may to some extent appear inverted with the participle.

210 (7.52)

Aspects of the Theory of Clitics a. Maglea va ia en traclo cun caschiel eaten have.1sg I a sandwich with cheese I ate a cheese sandwich b. *Maglea en traclo cun caschiel va ia eaten a sandwich with cheese have.1sg I c. La notg passada ò Gion durmia malamaintg last night has.3sg John slept badly Last night John slept badly d. Durmia ò Gion malamaintg la notg passada slept has.3sg John badly last night John slept badly last night e. ??Durmia malamaintg ò Gion la notg passada slept badly has.3sg John last night

This construction is reminiscent of the Icelandic “Stylistic Fronting” construction, which we saw above in section 7.2. A variety of other diﬀerences between Surmiran and Icelandic, however, make it diﬃcult to establish a clear parallel. It is also possible to emphasize the lexical main verb of the sentence, in which case it appears as an inﬁnitive in initial position, eﬀectively “doubling” the verb of the main clause. As with the participles in (7.52), fronted inﬁnitives cannot be accompanied by complements, as (7.53b) shows. (7.53)

Ursus ena canzung a. Cantar canta=’l to.sing sings.3sg-3sg.m Ursus a song Ursus is singing a song Ursus b. *cantar ena canzung canta=’l to.sing a song sings.3sg-3sg.m Ursus

Again there is a parallel with another language: in Breton, as we saw in section 7.4, the topicalization construction also allows the lexical verb to appear in initial position as an inﬁnitive (or verbal noun). There are important diﬀerences between the Surmiran and Breton constructions, however. First, unlike Surmiran, Breton allows the verbal noun to be accompanied by a complement, as in (7.24). Second, where Surmiran doubles the main verb with the fronted inﬁnitive, Breton replaces the ﬁnite form with a form of ober ‘to do.’ This is impossible in Surmiran, as the ungrammaticality of (7.54) shows. (7.54)

*screiver fatsch ia en codesch to-write do-1sg I a book

Verb Second as Alignment

211

Finally, for some speakers, the Surmiran construction is only possible with synthetic forms of the inﬂected verb and not with periphrastic forms such as the perfect, as illustrated in (7.55). (7.55)

a. Cantar cantava=’l Ursus bagn to.sing sang.3sg.imperf-3sg Ursus well Ursus was singing well canto Ursus bagn b. */?Cantar ò=’l to.sing has3sg-3sg.m sung Ursus well Ursus sang well

On the other hand, Breton shares with Surmiran the possibility of having a verbal participle in sentence-initial position, and also the restriction that such a fronted participle can only marginally be accompanied by a complement, as we saw above in (7.25) and (7.26). The signiﬁcance of these partial parallels among verb-second languages from three distinct families within IndoEuropean must remain a topic for future research. I analyze these basic aspects of Surmiran syntax as follows. The basic organization of the core clause in the language is SVO. In simple declarative clauses with no topic, there is no distinction between IP and VP, and the structure is as in (7.56). (7.56)

[

IP,VP

DPSbj [

I ,V

V (DPObj)]]

When some constituent is to be assigned topic interpretation, however, additional structure is required, and a distinct IP must be constructed within which the VP is embedded. The Speciﬁer position within this phrase is ﬁlled by the topic. The ﬁnite verb then moves to the I position; this might be either because it must be located in the head of IP in order to inherit the Tense and Agreement features of the clause, or in order to remain in second position within IP. On either analysis, it is this movement of the verb to I within an including IP that constitutes “Subject-Verb Inversion” in Surmiran. On this account the structure of sentence (7.44a) above is as shown in Figure 7.5. In terms of this structure, we can now characterize the circumstances under which subject clitics can appear. Speciﬁcally, when the verb in I Ccommands the subject phrase in Spec(VP), a special clitic agreeing with the properties of the subject may be introduced following the verb. When such a clitic appears on the verb (and only then), a null pronominal (pro) is licensed in the corresponding subject position. I leave additional details concerning the operation of this proposal to Chapter 8, but the overall outline should be clear.

212

Aspects of the Theory of Clitics IP I DP

VP V

I V

Rumantsch

discorra

DP

V

DP

Adv

Ursus

[e]

[e]

stupent

Figure 7.5. The Surmiran Inversion Construction

Inversion and Subject Clitics in Other Clause Types Thus far, I have considered only simple declarative main clauses in Surmiran. Languages with verb second can diﬀer in the class of clauses in which these eﬀects can appear: in German, verb second is found only in main clauses (and subordinate daß clauses with no overt complementizer), while most subordinate clauses have the ﬁnite verb in ﬁnal position. Icelandic, in contrast, shows verb-second eﬀects in both subordinate and main clauses. It is worth asking where Surmiran falls with respect to this typology of “symmetric” and “asymmetric” verb-second languages. Inversion in subordinate clauses is diﬃcult to explore, because this construction requires the presence of a topic element, and Topicalization is in general a root phenomenon. It is thus pragmatically unusual to ﬁnd the conditions for inversion in a subordinate clause, but examples such as (7.57) where initial topics are well-formed in embedded clauses do show inversion. (7.57)

a. Cartez tg’igl settember turnan=s believe.2pl that-art September return.sbjnctve.1pl-1pl ainten chel hotel in this hotel Do you think in September we’ll come back to this hotel? b. Ia pains tgi dultschems vegia I think.1sg that sweets have.sbjnctve.3sg Corinna gugent Corinna gladly I think Corinna likes sweets

Verb Second as Alignment

213

Inversion also takes place in interrogative clauses in Surmiran. This occurs with content questions of all sorts when the questioned argument is not the subject, as in (7.58). (7.58)

a. Tge ò=’la (Ladina) cumpro? what has.3sg-3sg.f Ladina bought What did Ladina/she buy? (Ladina) cumpro en auto? b. Cura ò=’la when has.3sg-3sg.f Ladina a car When did Ladina/she buy a car? (Ladina) cumpro? c. Igl auto da tgi ò=’la the car of whom has.3sg-3sg.f Ladina bought Whose car did Ladina/she buy?

We can tell that inversion has taken place here not only from the word order but also from the possibility of a subject clitic. When the questioned element is the subject, the structure is less immediately clear. The word order alone does not suﬃce, and the ungrammaticality of subject clitics in sentences like (7.59) simply shows (on the analysis suggested above) that the verb has not been displaced into a position where it would C-command the (fronted wh-phrase) subject. (7.59)

cumpro en auto? Tgi ò(*=’l/*=’la) who has.3sg(-3sg.m/f) bought a car Who bought a car?

When a non-subject wh-phrase is fronted, inversion occurs along with the possibility of a subject clitic, as in (7.58). We can tell that these eﬀects are associated with the conﬁguration that results from the movement, and not with wh-movement itself, because a wh-phrase fronted out of a complement clause produces them not in the clause from which the movement takes place but rather in the matrix clause, as illustrated in (7.60). Wh-movement causes the wh-phrase to precede the matrix subject, where it is no longer within the complement clause domain at all. (7.60)

Tge manegias te tgi Ladina vegia(*=la) what think.2sg you that Ladina have.sbjnctve.3sg(*-3sg.f) cumpro? bought What do you think that Ladina bought?

214

Aspects of the Theory of Clitics

Sentences such as (7.60) show that the complementizer tgi “that” does not itself trigger inversion in a subordinate clause. In the analyses of Icelandic (section 7.2) and Kashmiri (section 7.3) above, I suggested that the elements að and ki /yi which appear initially in complement clauses in these languages are not complementizers in the sense of occupying the C position in clause structure, but rather initial special clitics introduced as a reﬂection of a CP’s subordinate position. I propose to treat Surmiran tgi= in the same way. Multi-word expressions can also introduce subordinate clauses, without provoking inversion, as in (7.61). (7.61)

Siva tg’els on en unfant, stat el pi savens since that-they have.3pl a child is.3sg he more often a tgesa at home Since they have a child, he is home more often

Rather than treating siva tgi “since” in (7.61) as a complex complementizer occupying a C position, I suggest that this construction involves the independently occurring preposition siva “after, following” governing a clause marked with special clitic tgi= like other embedded clauses. Example (7.61) also demonstrates that an initial subordinate clause can itself occupy the structural Spec(IP) position, triggering movement of the verb to I which constitutes inversion in the main clause of which it is a modiﬁer. Why do we ﬁnd inversion in clauses with an initial wh-phrase? This might be because the wh-phrase itself occupies the same Spec(IP) position as topics, and thus provides an I position outside of the VP to which the verb must move, as in declaratives. This is probably not the best analysis, however, as we shall see. Turning to relative clauses, we ﬁnd that these do not in general show inversion regardless of whether the relativized element is subject or non-subject, as the examples in (7.62) illustrate. (7.62)

aveir a. Igl codesch tgi è sen meisa sas=t the book which is on the table can.2sg-2sg have The book which is on the table you can have b. Igl velo tgi Ursus ò cumpro n’=è betg nov the bike which Ursus has bought neg-is.3sg not new The bike which Ursus bought is not new

Verb Second as Alignment

215

c. Igl gioven agl qual ia va scretg The youngster to.the which I have.1sg written è sto igl mies scolar is.3sg been the my student The youngster to whom I wrote was my student d. La matta dalla qualla te ast survagnia the girl from.the which you have.2sg received en canaster mareida proximamaintg a basket marries.3sg in the near future The girl from whom you got a basket is getting married soon Relative pronoun expressions thus appear to be unlike wh-question phrases in not occupying a structural position that would result in movement of the verb. Another construction in which we do ﬁnd inversion suggests a basis for this structural diﬀerence between relative pronouns and question words. In yes/no questions such as (7.63), inversion takes place with no phrase of any sort preceding the verb in its displaced position. (7.63)

a. È igl viadi sto tger? is.3sg the trip been expensive Was the trip expensive? b. Ast er te gost da neir? have.2sg also you desire to come Do you want to come too? c. Lain=sa nous eir cugl tren? want.1pl-1pl we go with.the train Do we want to take the train? plaschia an Sicilia? d. At=ò=gl 2sg-has.3sg-3impers pleased in Sicily Did you like Sicily?

The possibility of subject clitics in this construction (as with other instances of inversion) suggests that the position of the verb here, as in declaratives with initial topics and content questions with initial non-subject wh-phrases, is one from which it C-commands the original subject position. Inversion in yes/no questions cannot be associated with a requirement that the verb appear in second position, since its eﬀect is precisely to make it initial. Compare the quite diﬀerent facts of Kashmiri, as in (7.11). It also cannot be

216

Aspects of the Theory of Clitics

a consequence of additional functional structure introduced to host an initial phrase, since no such phrase occurs. Something else must be present in these structures to result in movement of the verb. Assume that questions involve an interrogative operator “Q” appearing in I, the head of the clause. In content questions, the questioned wh-phrase must appear as the Speciﬁer of I[+Q]; as a result, where the wh-phrase is a non-subject, such sentences have the core VP structure of the clause embedded within a higher functional IP projection like that of topicalized sentences, with the verb attracted to I[+Q] and the wh-phrase in Speciﬁer position. If the whphrase is the subject, however, no additional structure is motivated beyond the presence of [+Q] on the inﬂected verbal head. In yes-no questions, I[+Q] must C-command the entire clause nucleus in order to establish its scope. It must thus appear on the head of an IP within which the VP is embedded, and this position again attracts the verb. Inversion (or Verb Raising) thus occurs both in yes-no questions and in content questions, except when the questioned phrase is the subject. Relative clauses, in contrast, involve a diﬀerent operator which is unrelated to I and thus does not attract the verb. The relativized phrase appears as the speciﬁer of this operator, and in this position binds a gap inside the VP to which it is related. Relative clauses do not contain topics, and there is no other motivation for Inversion in their structure. “Inversion,” then, consists in the movement of the verb to an I position outside of its basic VP, triggered by the IP structure necessary to support an initial topic phrase, a non-subject wh-phrase, or a position from which I[+Q] can establish appropriate scope in a yes-no question. In exactly these cases, the displaced verb precedes and C-commands the VP-internal subject position, the condition which sanctions the presence of a subject clitic. Inversion and the possibility of these clitics thus receive a uniﬁed analysis as required. The Strange Case of ins One further aspect of the inversion construction in Surmiran remains to be considered: a set of complications that arise in connection with a class of impersonal sentences involving the element ins ‘one.’ This is a modern reﬂex of Latin unus, similar in its interpretation to French on and German man. It indicates an impersonal subject, in sentences like (7.64). Not surprisingly, such sentences show third-person singular agreement. (7.64)

11

Ins pò eir quant spert tg’ins vot ins can-3sg go as fast that-ins wants.3sg sen las autostradas svizras on the freeways Swiss You can go as fast as you want on the Swiss freeways11

This is not true. The author accepts no responsibility for speeding tickets received by readers.

Verb Second as Alignment

217

Like on and man, ins only occurs as a subject, and not as an object or oblique. Sentences like (7.65) are thus excluded. (7.65)

a. *Igls pulizists na pon betg veir ins the policemen neg can.3sg not see ins (The police can’t see one from there)

da lò from there

b. *Mintgign digls guids ò la sia moda each of.the guides has.3sg the his way da trattar cun ins of to-deal with ins (Each of the guides has his way of dealing with one) Despite this distributional limitation, ins seems simply to be an indeﬁnite pronoun. Just like any subject, it precedes the verb, and is otherwise sentence initial. But surprisingly, when a topic phrase precedes ins as in (7.66), the verb does not invert. (7.66)

a. Dalla derivanza digls rets ins so tant scu of.the origin of.the Rhaeti ins know.3sg so-much as navot nothing Of the origins of the Rhaeti12 we know almost nothing. b. D’anviern ins pò eir sur tot igls pass cun auto In winter ins can.3sg go over all the passes with car In the winter you can go over all of the passes by car

Similarly, in questions like (7.67), ins still precedes the verb. (7.67)

a. Ins viagia pi bagn cugl tren u ins travels.3sg more good with-the train or cugl auto sch’ins fò viadis pi lungs? with-the car if-ins makes trips more long Does one travel better by train or by car when making longer trips? b. Tge meis digl onn ins dovra pneus what month of.the year ins needs.3sg tires d’anviern aint igl Grischun? of-winter in the Graubünden What month of the year do you need winter tires in Graubünden?

Unlike all other nominal subjects in Surmiran, then, ins inversion with a following ﬁnite verb. 12

Early indigenous people of the Rumantsch area.

does not undergo

218

Aspects of the Theory of Clitics

This is still not the end of the story, however. Although ins does not itself invert with a following verb, when it represents the subject, the verb may be followed by an additional impersonal subject clitic in inversion contexts only. When “subject” ins is preceded by another element within the clause as in (7.68a), or in questions like (7.68b,c), the verb can be followed by the clitic element =(i)gl. This is otherwise the subject clitic that appears in impersonals (weather it, etc.). (7.68)

a. Ainten chell’ustareia ins na betg magl=igl in this-inn ins neg= eat.3sg-3impers not schi bagn, on=igl detg so well have.3pl-3pl said In this inn you don’t eat so well, they said ﬁmar cò? b. Ins pò=gl ins can.3sg-3impers to-smoke here Can you smoke here? c. Cun tge igl migler tren ins vo=gl with what train ins goes.3sg-3impers the better per eir da Sargans a Wien? to go from Sargans to Vienna Which train is better to go from Sargans to Vienna?

Contemporary Surmiran usage allows the introduction of the impersonal clitic =(i)gl in inversion contexts when the subject is indicated by ins. Apparently, earlier speakers used the personal subject clitic =’l instead in this case. The examples in (7.69) are taken from Grisch (1939: 209) (orthographically adapted from the phonetic transcriptions there); they show the personal clitic in inverted structures instead of the impersonal form. (7.69)

a. Chel ins dei=’l dapertot that ins says.3sg-3sg.m everywhere That they say everywhere b. Ena schi greva lavour ins sto=’l A so hard job ins should.3sg-3sgm betg far tot sulet not do all alone Such a hard job one shouldn’t do all alone

As we expect, Grisch’s examples show no clitic in sentences like (7.70) where inversion would not be motivated.

Verb Second as Alignment (7.70)

219

a. Ins dei dapertot tgi . . . ins says.3sg everywhere that They say everywhere that. . . b. Ins sto betg far tot sulet ena schi greva lavour ins should.3sg not do all alone a so hard job One shouldn’t do such a hard job all alone

Although ins seems like an indeﬁnite subject pronoun, its behavior is more like that of a special clitic. Like the negative element na and the object pronominal clitics, it appears in a ﬁxed position in relation to the ﬁnite verb and does not invert with it. Its presence, however, does satisfy the verb’s need for a subject. This pattern resembles that of impersonal se in Spanish or (perhaps more relevantly) si in Italian, as in the examples of (7.71). (7.71)

Spanish: En México se trabaja mucho in Mexico se works.3sg much In Mexico one works a lot Italian: Si lavora sempre troppo si works.3sg always too much One always works too much

In some other forms of Rumantsch, particularly those spoken in the Engadine, we ﬁnd a much more direct correspondence with the pattern of (7.71) characteristic of other Romance languages. Engadine Rumantsch has adopted the clitic reﬂexive pronoun as for impersonals. For Vallader, Ganzoni (1983: 69) gives the example in (7.72). (7.72)

Passand tras il desert as= chatta qualchevoutas Passing across the desert 3sg.refl ﬁnds.3sg sometimes skelets skeletons Crossing the desert, one sometimes ﬁnds skeletons

In his parallel grammar of Puter, Ganzoni (1977: 69) cites virtually the same example, but also gives (7.73) as a variant: (7.73)

Passand tres il desert chatta ün qualchevoutas passing across the desert ﬁnds.3sg man sometimes skelets skeletons Crossing the desert, one sometimes ﬁnds skeletons

220

Aspects of the Theory of Clitics

The form ün in Puter (like Surmiran ins, a reﬂex of Latin unus) behaves like German man in occupying the subject position and inverting with the verb under the appropriate circumstances. Surselvan also has this construction, for which that language uses the pronoun ins, homophonous with the Surmiran form and derived from the same source but which inverts with the verb when required in sentences like (7.74). (7.74)

Nua ein ins cun la lavur? Ins ei alla ﬁn. where is.3sg ins with the work ins is.3sg at-the end Na, alla ﬁn ein ins mai. no at-the end is.3sg ins never Where are we with the job? We’re ﬁnished. No, we’re never ﬁnished.

The unusual failure of ins to invert with the verb in examples such as (7.66) and (7.67) is not a uniform property of Surmiran usage. Grisch (1939: 209) cites the example (7.75) with inversion, which she describes as typical of usage in Sotses (Casti [Tiefencastel], Alvaschagn). (7.75)

Chel dei ins dapertot That says.3sg ins everywhere That they say everywhere

Speakers in the Surses area of Surmeir (on whose speech the present description is based) also recognize sentences with inverted ins as possible, though in the region roughly between Mon and Marmorera, and also in the normative description of Signorell, Wuethrich-Grisch, and Simeon (1987), uninverted ins is the rule. Speakers who do not invert ins characterize sentences with inversion of this element as imitative of German usage. The structure with inverted ins calls for no particular comment, since that is just what we would expect of an impersonal pronoun occupying the subject position. What is much more remarkable is the possibility of the uninverted structure. Uninverted ins patterns structurally as a clitic, similar to Spanish se, Italian si, or Vallader as, despite origins which would lead us to expect it to be treated as a non-clitic pronoun. How reasonable is it that ins has been reanalyzed from a full indeﬁnite pronoun to a clitic, and what might have led to that development? A glimpse of the relevant history is provided in Grisch (1939), a description of what we might see as an earlier, transitional stage of the language. In that source, the phonetic form in which ins appears is actually [@ns]. This would be homophonous with the ﬁrst-person plural object clitic pronoun ans, which might in turn have led to a conﬂation of the two. Note that in French and

Verb Second as Alignment

221

Italian, the impersonal forms on and si are often used with ﬁrst-person plural reference, as in (7.76). (7.76)

French: Nous, on fait pas ça ici we on does.3sg not that here We don’t do that here Italian: Si è contenti quando =ci scrivono si is.3sg happy.pl when 1pl write.3pl We are happy when they write to us (Burzio 1992)

To the extent ins from unus was pronounced [@ns], it could readily be confused with the ﬁrst-person plural object clitic ans. Despite the fact that ins and ans are clearly distinct elements in contemporary Surmiran, the possibility that they might be confused is apparently still alive. This is suggested by the fact that the normative description of the language contains a remarkable (and quite isolated) warning that “Igl pronom subject ‘ins’ è da disfranztger bagn digl pronom ‘ans’ tgi serva scu pronom reﬂexiv e pronom persunal object dall’amprema persunga plural”13 (Signorell, Wuethrich-Grisch, and Simeon 1987: 120). The distinctness of these two pronouns is supported by a number of facts. First, apart from the diﬀerence in spelling, ins is generally pronounced [ins] while ans is phonetically [@ns]. Second, while ans immediately precedes the verb and is itself preceded by the negative particle na if that is present, ins itself precedes na, as in (7.77). (7.77)

betg cleramaintg Da lò ins n’=ans= vei from there ins neg-1pl sees.3sg not clearly From there one doesn’t see us clearly

Thirdly, ins always precedes the inﬂected ﬁnite verb whose subject is indeﬁnite. In contrast, object pronouns such as ans typically precede the inﬁnitive in modal constructions such as (7.78). (7.78)

a. El vot ans= tarmetter ena factura dumang He wants.3sg 1pl to.send a bill tomorrow He wants to send us a bill tomorrow b. Mintgatant ins stò(=gl) spitgier en po often ins must.3sg(-3impers) wait a bit Often you have to wait a bit

13 “The subject pronoun ins must be clearly distinguished from the pronoun ans which serves as the reﬂexive pronoun and the object personal pronoun of the ﬁrst person plural.”

222

Aspects of the Theory of Clitics ins c. *Mintgatant stò(=gl) often must.3sg(-3impers) ins Often you have to wait a bit

spitgier en po wait a bit

In compound tenses such as the perfect, object pronoun clitics like ans can appear either before the (inﬂected) auxiliary or before the participle, but ins always precedes the auxiliary. For these reasons (among others), ins and ans must be treated as quite distinct grammatical elements. Nonetheless, it appears that at some point in the history of the language, they were at least partially confused, and ins came to be treated as if it were a clitic like ans. This confusion was probably encouraged by the fact that in Italian, a language with which many Surmiran speakers are at least somewhat familiar, the indeﬁnite subject interpretation provided by ins is supplied by a construction with no overt subject and a clitic pronoun si---which can also have ﬁrst-person plural reference. As a result, ins has taken on the behavior of a clitic associated with the ﬁnite verb, rather than that of an autonomous nominal expression. This has the strange consequence that the subject position in ins-sentences is not overtly ﬁlled: strange, because Surmiran normally allows the subjects of ﬁnite verbs to be unexpressed only when they are doubled by a subject clitic in the inversion construction. Other impersonals, weather verbs, etc., all require overt subjects (typically the pronoun igl in these cases), just like French or English. If inversion were really a reordering of the ﬁnite verb and the subject, this ought to mean that in inversion contexts when ins represents the subject, there is nothing to invert with the verb, and we have seen that the word order in these cases remains ﬁxed. We also saw, however, in (7.68), that the subject clitic =(i)gl, normally a diagnostic of inversion, can appear under just these circumstances. In contemporary Surmiran of the variety in which sentences like those of (7.66) and (7.67) occur, a reanalysis appears to have taken place as follows. Originally, ins really was an impersonal pronoun occupying subject position. Under the inﬂuence of the factors discussed above, however, it was reanalyzed as a special clitic associated with the ﬁnite verb. The subject position in impersonal sentences, accordingly, came to be ﬁlled not by ins itself, but rather by a phonetically null pronoun with impersonal reference: the PROArb often assumed to represent the subjects of non-ﬁnite predicates like those of Coming in second is never fun and It is more fun to come in ﬁrst. The special clitic ins= is then introduced at the left edge of a ﬁnite verb whose subject is PROArb. In inversion contexts, this verb will be displaced to a higher head position (I in topicalized structures and questions), but this does not alter its position

Verb Second as Alignment

223

with respect to ins. On the other hand, when the verb is displaced to such a position and thus comes to C-command its original subject, the subject clitic =(i)gl may be introduced as in any other case where a displaced verb C-commands an impersonal subject. The Nature and Origin of Verb Second in Surmiran So what does it mean to say that Surmiran is a “verb second” language? What, that is, is the nature of the generalization concerning the position of the verb? And does “verb second” in the Surmiran sense designate the same property as the one we refer to in describing other languages such as German or Icelandic as “verb second” languages? In Surmiran, unlike the other languages we have surveyed thus far, it appears that the correspondent of verb second involves exactly displacement of the verb from its base position (within VP) to a head position in a higher layer of structure. The displacement, however, appears to be motivated only by the properties of the head itself. In particular, Alignment constraints such as LeftMost and NonInitial play no obvious role in the inversion construction. This is most apparent in impersonal sentences with ins, where we have seen the subject position is phonetically empty, and the verb (together with its clitics, including ins) thus appears at the left edge of the constituent in which it heads. Just the sort of analysis that has commonly been proposed for Germanic languages with verb second thus turns out to be appropriate for Surmiran. Somewhat ironically, however, the evidence that supports that conclusion most strongly comes precisely from the fact that Surmiran is not consistently a verbsecond language. This is quite distinct from the properties of special constructions such as yes/no questions, which must be provided for separately in any case in other verb-second languages like German. In Surmiran, however, it turns out that under the one set of circumstances in which the pre-verbal subject position can be phonetically empty in a neutral declarative clause (in ins-sentences), the verb shows no disposition to avoid the result that it thereby appears initially. What gives Surmiran the appearance of a true verb-second language is the inversion construction, especially as this is associated with initial topic phrases. Surmiran (and Rumantsch more generally) may have borrowed the structural properties of this construction (topic phrase in Spec(IP) and displacement of the verb to I) from Germanic, or it may have developed within the history of Romance, but this historical matter has no bearing on the resultant syntax. We come down to the conclusion that the “verb second” property in Surmiran is not really a requirement that the verb be in second position at all, or

224

Aspects of the Theory of Clitics

indeed in any structurally uniform place. What gives rise to that impression is the combination of two facts: (a) the basic word order in the language is SVO; and (b) when levels of structure above that of the basic VP clause are motivated, properties of their heads may provoke a displacement of the ﬁnite verb to a position immediately following an initial speciﬁer. In this case, at least, Wackernagel’s (1892) proposed uniﬁcation has little to recommend it. Verb second here does not fall together with the mechanisms describing secondposition clitics.

7.6 Conclusions Second-position clitics and verb second have in common the fact that both may reﬂect constraints requiring the grammatical properties (i.e., the functional content) of a phrase or clause to be aligned as closely as possible with the left edge of the inﬂectional domain, without being absolutely initial (in IP, or in some languages CP). They diﬀer in that clitics are introduced by a phonological mechanism of aﬃxation, while verbs that inherit the functional properties of the clause they head are subject to movement by normal syntactic mechanisms. On the other hand, all that glitters is not gold: verb-second constructions in German, Icelandic, Kashmiri, and Breton do appear to involve essentially the same Alignment constraints as those responsible in other languages for second-position special clitics, but the superﬁcially very similar construction in Surmiran does not. To the extent it is valid for some core examples, Wackernagel’s (1892) suggested uniﬁcation thus does not generalize to all instances of verb-second languages. It is also important to note that while second-position clitics are found in a wide range of languages throughout the world, clear instances of verb second languages are diﬃcult (if not impossible) to ﬁnd outside of Indo-European. Furthermore, with the possible exception of Kashmiri, languages that display verb second do not have much to show in the way of second-position clitics. And only in Kashmiri does there even appear to be a plausible historical connection between the two phenomena. If the uniﬁcation we set out to explore at the beginning of this chapter were really pervasive, we would expect these typological matters to work out rather diﬀerently. This does not, however, contradict the claim that the theory of secondposition clitics can be extended to an account of at least a signiﬁcant subset of verb-second phenomena. The two constructions have in common that they

Verb Second as Alignment

225

involve requirements for aligning grammatical material as close to the beginning of a particular structure as possible, while at the same time not allowing it to appear at the absolute beginning of the structure. The main diﬀerence between the two types of phenomena is that the clitics are introduced into the relevant position by a phonological aﬃxation mechanism while the verbs arrive in their position by syntactic displacement. What is especially interesting in this account is the possibility of maintaining a morphological/phonological account of the clitic phenomena and a syntactic account of the verb phenomena, while at the same time, using essentially identical OT constraints to cover both cases.

This page intentionally left blank

8 Pronominal Clitics The canonical examples of special clitics in the minds of many linguists are pronominals such as those found in association with the main verb in Romance languages1 or those found in second position in languages like Warlpiri or Tagalog. Pronominal clitics have a number of interesting properties in their own right. Some of this interest is more or less independent of their special clitic status, and other matters are important to explore in connection with the very notion that these elements can be treated linguistically in the same way as other, non-pronominal special clitics. The purpose of this chapter is to deal with (or at least acknowledge) some of these issues. Given their complexity, and the richness of the existing literature dealing with them, this survey will inevitably have something of the character of a superﬁcial whirlwind tour. Nonetheless, it is important to show that the present framework provides a basis for discussing and analyzing these classic problems. I have been assuming we can call pronominal clitics the functional morphology of phrases, but what does that mean as far as the underlying grammatical structure is concerned? Pronominal clitics are also frequently assimilated in various ways to agreement markers—what connections are there between the two? A key role in understanding these matters is played by Clitic Doubling, a construction whose properties help to clarify the relation between pronominal clitics and the argument positions they are related to. A further topic which has been prominent is the construction known as “Clitic Climbing.” While a relatively traditional analysis of this construction ﬁts well with the general framework adopted here, its extension to cases of Long Distance Agreement is perhaps less familiar. Finally, while attention to pronominal clitics has generally focused on elements corresponding to objects of various types, some 1 Vast amounts have been written about the Romance pronominal clitics, and it is quite impossible for me to cover the complete research history relevant to these elements, although much of the discussion below is devoted to them. A very useful survey of relevant phenomena and analyses, from which I have drawn at several points, is provided by Miller and Monachesi (2003). Many properties of Romance pronominal clitics noted in that paper are not discussed here because these would take us too far aﬁeld from the main point, but I do not think any of these matters pose serious problems for the analysis to be presented.

228

Aspects of the Theory of Clitics

languages also display clitics linked to subjects, which pose particular issues of their own.

8.1 Pronouns and Agreement What is the structural relation between a pronominal clitic and an argument position which it is interpreted as specifying or ﬁlling? In “classical” generative grammar, this was considered a straightforward instance of movement. Such clitics were generated as pronouns of a certain sort, occupying an argument position just as any other nominal expression might. By virtue of some identiﬁable property (e.g., “[+Clitic]”) they then underwent movement to adjoin to the Verb—or wherever else they might appear in a particular language, such as second position. The movement analysis explained one important property of clitics: in many cases, they are mutually exclusive with overt nominal expressions in a given argument position, as illustrated in sentences (8.1) from (standard) French.2 (8.1)

a. Les enfants l’= ont déjà mangé The kids it have already eaten The kids have already eaten it b. *Les enfants l’= ont déjà mangé le gateau The kids it have already eaten the cake (The kids have already eaten the cake)

Unfortunately, this explanation goes a bit too far. In some languages, at least some pronominal clitics are not mutually exclusive with overt nominals, as in the Spanish sentence in (8.2), where the presence of the clitic is consistent with the simultaneous presence of the full phrase al professor. entregué el libro (al professor) (8.2) Le= to- him I gave the book (to the professor) I gave the book to him/the professor In Spanish, dative (and sometimes accusative) clitics can correspond to an overt argument expression. This is called clitic doubling, and it obviously raises problems for the movement analysis. It is hard to see how the argument 2 Sentence (8.1b) is acceptable if the ﬁnal phrase le gateau ‘the cake’ is preceded by a signiﬁcant pause, indicating a right-dislocated structure. This is irrelevant to the point here, which concerns the complementarity of clitics and full nominals specifying the same argument within a clause. It should also be pointed out that some forms of spoken French differ from the “standard” language in permitting much freer use of clitics that “double” a full nominal expression as in (8.1b).

Pronominal Clitics

229

position could remain in place, ﬁlled by a full nominal or prepositional phrase, while also undergoing movement as a clitic.3 On the theory of the present book, of course, pronominal special clitics are not elements that have been moved to where they appear in the surface form from some argument position. Rather, they are the overt reﬂection of properties of that position, construed as part of the functional content of the clause and realized by a principle of phrasal morphology as a modiﬁcation of the phonological form of the clause. In fact, the phenomena surrounding pronominal special clitics are rather similar to what we ﬁnd in verbal agreement (as various authors have noted). I will thus develop a view of these clitics by starting from corresponding facts in that domain.4 Kinds of Agreement In many (perhaps most) languages, agreement as marked on the verb registers certain properties of an argument, and does not supplant the overt expression of the argument. This is true in French, for example, where agreement and a full nominal subject are both present in all ﬁnite clauses. In other cases, however, agreement morphology and overt argument expressions are in complementary distribution. A particularly clear example is provided by the Venezuelan Carib language Pemon, as illustrated by the examples in (8.3).5 (8.3)

a. kamicha ke Antonio- da mure ponte- ’po clothes with Antonio- erg child dress- past Antonio dressed up the child with clothes b. kamicha ke mure ponte- ’po- i- ya clothes with child dress- past-3-erg He dressed up the child with clothes c. kamicha ke i- ponte- ’po Antonio- da clothes with 3- dress- past Antonio- erg Antonio dressed him up with clothes d. kamicha ke i- ponte- ’po- i- ya clothes with 3- dress- past-3-erg He dressed him up with clothes

3 This is true even if one adopts a “copy and delete” view of displacement, since the clitic is not literally a copy of the nominal it doubles. 4 Much of what I propose in this section is quite close to the range of possibilities and theories entertained for agreement within Lexical Functional Grammar. See Bresnan (2001: ch. 8) for an extended discussion in that framework. 5 The data below are from Jos´e Alvarez, unpublished ﬁeld notes. These facts were reported by Alvarez on the Linguist List, vol. 6, no. 574, in 1995.

230

Aspects of the Theory of Clitics

An extreme variant of this situation is proposed by Baker (1995) for Mohawk (and other “Polysynthetic” languages). Baker argues that the Mohawk verb always includes agreement (and/or an incorporated Noun), but that overt nominal argument expressions appear in the position of adjuncts, rather than that of arguments. The presence of agreement is claimed to preclude the presence of a nominal in the argument position, as in Pemon. The difference is that the presence of agreement itself (ignoring for now the case of incorporated nouns, to which I will return in Chapter 9) is obligatory in a language of this type, and thus we do not see the kind of alternation that occurs in (8.3). A third possibility in agreement systems is what we ﬁnd in languages with (perhaps generalized) “pro-drop.” Here also agreement with one or more arguments may always be present. While such agreement is compatible with the simultaneous presence of an overt argument, though, that argument may also be absent from the surface form, in which case the argument it refers to is interpreted in the same way a pronoun would be. An example is provided by Georgian, as in (8.4). The verb momc.era ‘he wrote it to me’ agrees simultaneously with all three of its arguments (although there are only two overt afﬁxal markers of agreement present in the form). Any one of these may be phonologically unrealized in the sentence, and interpreted as pronominal. (8.4) (vano-m) (me) (c.erili) mo- m- c.er- a (Vano- erg) (me) (letter.nom) pfx- 1obj- wrote- 3sg.aor Vano/he wrote a letter/it to me We need to provide, then, for several distinct possibilities. Sometimes agreement marking simply reﬂects the properties of the argument (as for instance in French subject–verb agreement). Sometimes it alternates with a full argument (as an option, as in Pemon, or perhaps obligatorily, as in Baker’s analysis of Mohawk). Finally, its presence sometimes sanctions (while not requiring) a null argument. I propose to regard the case of mere “registration”6 of an argument as involving an operation that copies some relevant morphosyntactic features (typically, person, number, and possibly gender; see Corbett 2003 for a survey) from the argument to the Morphosyntactic Representation7 of the verb, where rules of the morphology may spell them out. Alternative views on which these features are generated independently in the Morphosyntactic Representation or its equivalent, and then checked for the required identity at some point 6 The useful terminological distinction between “registration” and “agreement” proper was introduced in early work by Perlmutter and Postal on a theory of Relational Grammar. Unfortunately, it does not seem to have survived into current discourse, but it is well worth reviving. 7 See Anderson (1992) for this notion.

Pronominal Clitics

231

in the syntax, can be regarded for present purposes as notational variants of this picture. The second type of agreement is supplied by Baker’s (1995) analysis of Mohawk, which I adopt here at least for the purposes of discussion. In this case, we have obligatory agreement coupled with the impossibility of full DPs in argument position. Baker argues at length that the argument positions are in fact present, and occupied by phonetically null (but referential) pro elements whose features are reﬂected in verbal agreement. To describe this situation, we need to establish a link between the particular syntactic conﬁguration posited for Mohawk and something in the morphology of the language. For Baker, that link is provided by the Morphological Visibility Condition (“MVC”). This is a parametric choice in the grammars of individual languages: in some languages, such as Mohawk, the condition holds, while in others, like English, it does not. Where the MVC obtains, it requires that all arguments of a head (such as the verb) must be reﬂected in the morphology of that head. Baker assumes that the markers which appear in this function will then prevent full nominals from receiving Case in argument position. As a result, they must appear (if at all) in some external, adjoined position. This yields the particular clause structure which Baker argues for in Mohawk. A key role in this explanation is played by the assumption that agreement elements in Mohawk absorb the structural case which the verb would otherwise assign to its arguments. This must obviously be interpreted as a fact about Mohawk, at least in part, since agreement is not universally incompatible with the appearance of full nominals in argument positions. Something extra thus needs to be said in the grammar of such a language if the desired consequences are to follow. Mohawk verbs do, of course, subcategorize argument positions, but on Baker’s analysis these can only be ﬁlled by phonetically null pro with some appropriate set of features. In this language, these are (roughly) [±me], [±you], [±sg], [±pl], [±masc], [±fem], [±zoic], where dual number can be represented as [-sg, -pl] and neuter gender as [−masc, −fem, −zoic]. Because of the presumed Case absorption effect, overt nominals appear in clause-external adjunct position, where they form a chain with the presumed pro. An alternative to Baker’s assumption that Mohawk agreement absorbs Case is simply to say that Mohawk verbs do not assign structural Case at all—or rather, that the only principle of structural-case assignment in the language is the one licensing nominals in adjoined position. Notice that since agreement is obligatory, any hypothetical structural Case related to the verb would never, in fact, be assigned on Baker’s assumptions.

232

Aspects of the Theory of Clitics

The exclusion of full DPs from argument positions need not, however, be related to the assignment of abstract Case at all. I will suggest below that this can be derived instead from Binding relations, which would disallow anything but pro in positions with which the verb agrees referentially (as is the case in Mohawk). On that assumption, the assignment of Case is quite orthogonal to the issue of what ﬁlls argument positions. The presence of agreement establishes reference, so let us assume (with Baker) that argument positions are also indexed referentially in the verb by agreement. This relation involves a coindexing relation in addition to the copying (or identity) of features characteristic of agreement in general. Baker believes that what is at work here is the presence in syntactic representations of functional categories (instances of Agr) which bear indices. These are subsequently incorporated (via adjunction) into a single word with the head verb, thus satisfying the MVC. You might notice that the agreement elements thus look like little clitic pronouns. There are (at least) two ways in which Baker’s assumptions differ from those of the present book. First, like much recent work within the Minimalist Program (but for different reasons), I consider it important to avoid positing independent functional categories like Agr as syntactic heads. In that case, however, there is no obvious constituent to bear the referential burden of Baker’s Agr elements. Second, in line with the Lexicalist Hypothesis as articulated in Anderson (1992), the present framework does not allow for the creation of words within the syntax. What is necessary is to coindex something in the representation of the verb with the argument positions to which that verb assigns T-roles. In the present terms, that “something” clearly has to be the verb’s Morphosyntactic Representation. I thus want to say that agreement rules come in two basic ﬂavors, corresponding to the difference between registration and agreement mentioned above. Note that Baker needs to say that Agr absorbs Case in some languages and not in others: the difference to be developed is intended to have the same consequences. The primary function of an agreement rule is to copy some set of features from an argument to the head, as for instance in English or French subject agreement. As a result, the Morphosyntactic Representation of the verb (or other predicate) will contain those features, and they can then trigger the introduction of overt morphological markers in the derivation of the appropriate word form. This kind of agreement—registration—has no further syntactic consequences.

Pronominal Clitics

233

As a second possibility, an agreement rule can also establish a relation of coindexation between the argument position in question and the Morphosyntactic Representation itself, as is arguably the case in Mohawk. Such a relation is obviously very similar to the one Baker assumes, except that the coindexation is not with a “morpheme” internal to the Verb, but rather with a subset of the featural content of the verb’s Morphosyntactic Representation. In Mohawk, then, we could say that agreement is of this second, coindexing sort. Where some proi is coindexed with a layer of agreement features, the two must agree in those features. Now suppose we take seriously the fact that, as a result of agreement, the verb (through its Morphosyntactic Representation) is coindexed with those of its argument positions with which it agrees. Since the verb is the head of the VP, the “Head Feature Convention” (which I borrow from work in GPSG, Generalized Phrase Structure Grammar: see Gazdar et al. 1985) says that features in its Morphosyntactic Representation are transmitted to that of the VP. In case the constituent type IP is distinct in a given language from VP, when the verb moves to I (the head of IP) its features are also transmitted to the Morphosyntactic Representation of the IP node. Note that in many languages, Agreement establishes a relation between the verb and more than one of its arguments. In such a case, the agreement material corresponding to various arguments must in general be kept distinct. The “Layering Convention” of Anderson (1992) covers just this issue for word-level morphology, and extends directly to the case of Morphosyntactic Representations for phrasal categories. Just as the content of agreement may be transmitted from the verb to superordinate phrasal categories which it heads, it may itself inherit other properties from such a phrasal category. Suppose we say that Tense, for example, is a property of IP—part of the Morphosyntactic Representation of that phrasal node. The Head Feature Convention, by asserting identity between the properties of a phrase and of its head, has the effect of transmitting Tense properties to a verb when it occurs in I, the head of IP, and the morphology of the verb may thus reﬂect this phrasal property. In the case of interest to us here, features and/or referential indices assigned to the verb by Agreement are inherited by the Morphosyntactic Representation of the clause which it heads, where they constitute (perhaps a portion of) the “functional content” of the clause. On this view, “functional categories” such as Tense and Agreement are characteristic of a clause, while not projecting additional layers of syntactic structure. When realized as inﬂectional morphology on the ﬁnite verb, this content characterizes that word as well by virtue of the identity established by the Head Feature Convention.

234

Aspects of the Theory of Clitics

Now let us return to the case of genuine agreement: the sort which involves coindexing as well as feature identity. Here a referential index will be present by the mechanisms just described at the level of the clause, coindexed with the agreeing argument position. Now suppose an overt nominal expression were to appear in such an argument position, one with which the verb (and thus the clause node) is coindexed via Agreement. This ought to produce a violation of the Binding Conditions, since the nominal would thus be bound (coindexed with a C-commanding category) within its clause. This would exclude both R-expressions and lexical pronominals from appearing in positions with which the verb stands in a relation of agreement (and not merely registration). It is plausible, however, to assume that phonologically null pronominal elements (pro) are not similarly excluded. In this way, we could derive the exclusion of overt nominal expressions in argument positions from the referential nature of agreement in a language (e.g., Mohawk), without further Case-theoretic stipulation. On this view, phonetically null pronouns are associated with referential (i.e., coindexing) agreement. Assume also that most languages8 do not have null pronouns in their lexicon: the only way a motivated argument position can be phonetically unﬁlled is when lexicalization is blocked by virtue of referential agreement. Thus pro is not really a pronoun, but rather just a kind of empty category which (like other empty categories) must be sanctioned by principles of grammar. In this case, the relevant principle is grounded in the binding relation between referential agreement material and the position pro occupies. Based on these considerations, it is possible to propose a typology of the relations between agreement and the arguments it indexes. English or French represent one extreme, where agreement is always non-coindexing registration. As a result, argument positions in such languages are never required (or allowed) to be empty in the absence of other speciﬁc displacements or deletions. In Georgian, or in Italian and the other classic pro-drop languages, the agreed-with position can optionally remain empty. This can be regarded as an optionality in the operation of Agreement: this always copies the features of the argument, and it may optionally also coindex. Where coindexation appears, we get pro; where it does not, we have a normal pronoun or full lexical nominal expression. In contrast, accepting Baker’s extensively argued proposal for clause structure in Mohawk, Agreement in that language is always coindexing, and thus 8 Exception may need to be made for Chinese, Japanese and similar systems where the syntax of phonologically null DPs is somewhat different and more like that of genuine lexical alternants of full DP expressions.

Pronominal Clitics

235

argument positions are always empty. Mohawk deals with this situation by licensing the formation of chains which relate an empty (but referential) argument position to the content of a nominal in adjunct position. In Pemon, a given argument position can be identiﬁed either by an overt nominal or by verbal agreement, but not both. We can describe this by saying that Pemon agreement is of the referential kind, but Agreement itself is optional. When it occurs, it triggers the associated morphology and precludes an overt nominal; when it does not, there is no morphology, phonologically null pro is not licensed, and the relevant argument position must be ﬁlled with an overt expression. This difference between coindexing “agreement” and non-coindexing “registration” is a way of reconstructing at least one sense of “strong” versus “weak” agreement. It corresponds to the traditional notion that agreement in some languages “identiﬁes” the corresponding position(s). It also comes as close as we are likely to get to formalizing the notion of “strong” agreement, in treating this as a parameter of grammatical variation rather than deriving it from some aspect of the formal nature of agreement itself, such as the precise combination of features represented, or the extent to which various forms are distinct from one another—two tempting but ultimately unproductive approaches that have been taken in the literature to grounding the distinction. I suggest that the present proposal provides a more nearly satisfactory reconstruction. A Complex Example: Finnish Agreement A more intricate agreement system than those considered above is found in Finnish, and it is worth exploring how this is to be characterized in the present framework. Finnish has a set of markers whose status has been the subject of much discussion, appearing on head nouns to indicate properties of a possessor. On the basis of their failure to trigger the well-known process of consonant gradation in a previous syllable, as opposed to clear inﬂectional afﬁxes, some authors (e.g., Nevis 1986) have described these elements as (special) clitics. Despite their phonology, Kanerva (1987) shows clearly that they must be treated as word-level afﬁxes, and not as clitics. They thus constitute markers of agreement within the nominal phrase. They also show interesting similarities to the markers of verbal agreement in the language, and both the nominal and the verbal markers should fall within a theory of agreement relations. The phonology of these afﬁxes in the broader context of Finnish noun inﬂection is analyzed by Kiparsky (2003), and will not concern us here. The account of their syntax below follows in large part that of Toivonen (2000), an analysis in

236

Aspects of the Theory of Clitics

the terms of Lexical Functional Grammar and translated here into the framework of the present book. The facts concerning the nominal agreement system are as follows. When a nominal has a ﬁrst- or second-person possessor, the head of the phrase bears obligatory possessor marking, and an accompanying genitive pronoun may or may not appear for emphasis. This is illustrated in the sentences of (8.5), where (following the descriptive tradition) I gloss the possessive markers as “Px” with the appropriate person. (8.5) a. Poika myi (minun) marsu- ni boy sell.3sg.pst 1sg.gen guinea pig- px1sg The boy sold my guinea pig on sairas b. (Sinun) kissa- si 2sg.gen cat- px2sg is sick Your(sg) cat is sick c. Pekka näkee (meidän) ystävä- mme Pekka see.3sg.pres 1pl.gen friend- px1pl Pekka sees our friend d. (Teidän) kissa- nne on ruma 2pl.gen cat- px2pl is ugly Your(pl) cat is ugly This is straightforward: when the possessor is ﬁrst or second person, Agreement copies its features of person and number onto the head noun, and may optionally coindex that noun’s Morphosyntactic Representation with the possessor. If the coindexation option is taken, the presence of that referential indication prevents an overt possessor phrase, and the position of the possessor is ﬁlled by phonologically null pro. If there is no coindexation, pro is not licensed in that position and a full pronoun must appear. This is the normal state of affairs described above for “pro-drop” languages. In the third person, the situation is more complicated. Here we must distinguish three separate cases as far as the applicability of Agreement is concerned. First, when a human pronoun (hänen ‘his/her’ or heidän ‘their’) appears as possessor, as in (8.6), the head noun must bear the possessive sufﬁx. (8.6) Pojat näkevät heidän ystävä- nsa boy.pl see.3pl.pres 3pl.gen friend- px3 The boysi see theirj,*i friend Agreement here is obligatory and consists only in copying the person feature(s) onto the head noun. Since this does not involve coindexation, an empty

Pronominal Clitics

237

pro possessor is not licensed and the pronoun must appear. From independent considerations of binding, this must always be interpreted as disjoint in reference from the subject of the sentence. When an overt possessor expression appears that is either (a) non-pronominal, or (b) non-human,9 no possessor sufﬁx is possible, as illustrated in (8.7). (8.7) a. Minä pesen Pekan autoa(*- nsa) I wash.1sg.pres Pekka.gen car(*- px3) I am washing Pekka’s car b. Minä annan kissa- lle sen ruokaa(*- nsa) I give.1sg.pres cat- allative it.gen food(*- px3) I give the cat its food With overt possessor phrases other than human genitive pronouns, Agreement does not apply at all. No possessive afﬁx appears on the head noun, and the possessor cannot be replaced by null pro. The third case, that of sentences like (8.8), is particularly interesting. When a third-person possessive marker appears on the head noun, and no overt possessor is present, the nominal is interpreted as possessed by the subject of the clause. Note that there is no requirement that the possessor in this case be human. (8.8) a. Pekka näkee ystävä- nsä Pekka see.3sg.pres friend- px3 Pekkai sees hisi friend b. Se heiluttaa häntää- nsä It wiggle.3sg.pres tail- px3 It wiggles its tail The interpretation of (8.8) shows that a possessor is present, which must be a phonologically null anaphoric element controlled by the subject of the sentence. We might plausibly take this to be PRO, the element that also appears as the subject of certain non-ﬁnite complement clauses. When the possessor in a nominal expression is PRO, then, agreement copies its person and number features onto the head noun along with its referential index. This referential index, in turn, is controlled by the subject of the clause in which the nominal appears. In the third-person case, then, the difference between (non-coindexing) registration, (coindexing) agreement, and no agreement at all is not an option, 9 The usual sort of qualiﬁcation about what counts as “human” must be made here, but does not affect the basic point.

238

Aspects of the Theory of Clitics

as in the ﬁrst and second person. Rather, each of these possibilities corresponds to a speciﬁc set of circumstances. We can now compare this agreement system, as it operates in nominals, with the way Agreement operates in ﬁnite clauses headed by a verb. Here Agreement is always obligatory, and the morphology of the verb reﬂects the person and the number of its subject. When the subject is ﬁrst or second person, it can be phonologically null as shown in (8.9). (8.9) a. (Minä) ammuin karhun 1sg shoot.1sg.pst bear I shot a/the bear b. (Sinä) näit karhut 2sg see.2sg.pst bear.pl You saw the bears Agreement here is exactly parallel to what we saw above in the nominal case. With a ﬁrst-or second-person (possessor or) subject, Agreement copies the person and number features onto the head, and may or may not also introduce coindexing. When coindexing occurs, the possessor or subject expression is replaced by phonologically null pro. In the third person, the verbal and nominal cases diverge, but not entirely. First, verbal Agreement is obligatory regardless of the nature of the subject as human or non-human, pronominal or non-pronominal, as illustrated in (8.10). (8.10)

a. Poika/Hän myi kissa- nsa boy/he sell.3sg.pst cat- px3 The boyi /Hei sold hisi cat b. Kissa- ni kuoli cat- px1sg die.3sg.pst My cat died c. Se oli sairas it was.3sg ill It was ill

In all of these cases, the subject must be present overtly, even if it is pronominal and recoverable from the context. This suggests that third-person agreement, as opposed to ﬁrst or second person, consists only in copying the person and number of the subject to the verb, and cannot involve (even optionally) coindexation.

Pronominal Clitics

239

There is one circumstance in which a third-person subject can be dropped, however. Sentences with third-person singular agreement but no overt subject are interpreted as having an arbitrary or generic subject, as in example (8.11). (8.11)

Voi mennä ulos can.3sg go.inf out (One) can go out

This sort of generic interpretation is often found in other languages with non-ﬁnite clauses where there is no controller for the subject position, such as English To know her is to love her. We saw this in the discussion of ins in Surmiran in section 7.5 of Chapter 7, where I suggested that we should regard such sentences as having the element PRO as their subject, with an “Arbitrary” reading assigned when this appears as the subject of a ﬁnite clause. Similarly, Anderson (1982) argues that Breton ﬁnite clauses with a distinctive “impersonal” verbal inﬂection have PROArb as their subject. I propose that the same is true in Finnish, and that sentences like (8.11) also have PROArb as their subjects. But this, of course, brings the verbal agreement system into a closer relation with the nominal pattern. Recall that in that case, coindexing agreement was limited to the case of controlled anaphoric PRO as possessor. It seems then that the major difference between the verbal and the nominal systems is that in the verbal case, Agreement is always obligatory at least to the extent of copying person and number features. Coindexation is also possible if the argument triggering Agreement is ﬁrst or second person; and obligatory if the trigger is PRO (interpreted as Arbitrary in the absence of a controller, a possibility that is only relevant to the verbal subcase). In the nominal case, no agreement takes place with a non-human or nonpronominal overt third-person possessor, but otherwise the two systems are entirely similar. There are further phenomena to be discussed in connection with Finnish agreement, such as the facts of non-ﬁnite clauses of various types, the exact nature of the control relation relevant to PRO possessors, among others. Nevertheless the present typology of agreement relations seems to provide an appropriate framework for making the distinctions necessary to an adequate analysis.

8.2 Clitics, Agreement, and Doubling Now let us return to the analysis of pronominal clitics. I propose to regard clitic pronominals as in fact a form of agreement, differing from verbal agreement only in whether the functional content is realized as the morphology

240

Aspects of the Theory of Clitics

of a phrase or a word. This is not a novel proposal: Miller and Monachesi (2003) note that it arises fairly naturally within theories such as Lexical Functional Grammar and Generalized (or Head-Driven) Phrase Structure Grammar based on feature systems, and similar suggestions have been entertained within other frameworks, such as the proposals of Franks and King (2000). The overt manifestation of agreement material by pronominal special clitics can appear in various places, as we have already seen. Clitics may appear with reference to the beginning of the clause—in second position, often, as in Warlpiri or Tagalog. The relevant agreement representation, a set of features associated with the S (or IP) node, triggers the introduction of phonological material (the clitic(s)) within this domain. The linear position in which the clitics are found is determined by Alignment constraints in the ways developed in chapter 6. Alternatively, as in Romance, we may ﬁnd the clitics associated with the ﬁnite verb. The functional content of the clause is shared by its Morphosyntactic Representation and that of the Verb node which is its head by virtue of the Head Feature Convention, assuming the relevant features have the status of “Head Features” in the language at issue. In Romance languages, those parts of this feature structure relating to non-subject positions are realized within the restricted syntactic domain of the Verb constituent, while the subject material is realized by inﬂection on the verb itself. In other languages, such as Georgian, the features inherited by the verb are all realized as inﬂectional properties of that word. If pronominal special clitics are closely comparable to verbal agreement, they ought, like agreement, to come in both “weak” and “strong” ﬂavors, corresponding to the distinction drawn above between registration and agreement per se. And indeed, it is not hard to see how to pursue this analogy. The “weak” form is what we ﬁnd in clitic-doubling constructions, where the presence of the clitic merely registers the properties of the corresponding nominal. Consider the Bulgarian examples in (8.12). (8.12)

=ja obiˇcat neja a. Decata children.def her.acc love.3pl her.acc The children love her xrumna edna misâl b. Na Svetozar =mu to Svetozar him.dat dawned.3sg one thought A thought occurred to Svetozar kaža az na tebe koj e= predatel c. Šte =ti will you.dat tell.1sg to you who is traitor I’ll tell you who’s a traitor

Pronominal Clitics

241

d. Uvažavat =go edin uˇcitel zaradi vseotdajnostta =mu respect.3pl him.acc one teacher for devotion.def his A teacher is respected for his devotion

=go uvolnili? e. Kogo kazvaš =sa who.acc say.2sg aux.3pl him.acc ﬁred Who do you say they ﬁred? =e mâˇcno f. Na nego =mu to him him.dat is homesick He is homesick As summarized from a variety of authors by Franks and King (2000), the source from which the sentences in (8.12) are taken, doubling occurs in Bulgarian when an object nominal is topical and speciﬁc, generic, or a wh-expression. It is optional in all of these cases, except for impersonal sentences such as (8.12f ) where it is obligatory. On the other hand, doubling is not possible in other non-speciﬁc contexts, such as sentence (8.13). (8.13)

Târsjat (*=go) nov uˇcitel seek.3pl him.acc new teacher They are looking for a new teacher

The description of this state of affairs is not trivial, but neither are the facts themselves. We say that Agreement with object nominals (which will of course be realized by the introduction of a clitic rather than directly by verbal morphology) is optional, unless the argument is non-referential, as in the intensional reading of (8.13) or an idiom chunk. When it occurs, it is optionally coindexing, except in the case of impersonal sentences with no nominative subject such as (8.12f ), where it is obligatory. As in the case of agreement realized as verbal morphology, where this agreement is referential (or coindexing), it precludes the presence of an overt nominal in the corresponding argument position, and thus the clitic gives the impression of substituting for such an expression. Where the option of coindexation is not taken, simple registration of an overt argument phrase results. A well-known case of a similar sort is supplied by standard Castilian Spanish, in examples such as those of (8.14). (8.14)

a. *(Lo=) veo him.acc see.1sg (I see him) b. (*lo=) veo him.acc see.1sg I see John

a él prep him a Juan prep John

242

Aspects of the Theory of Clitics la mesa c. (*la=) veo her.acc see.1sg the table I see the table d. *(Le=) hablo a él him.dat speak.1sg to him I speak to him hablo a Juan e. (Le=) him.dat speak.1sg to John I speak to John

In Spanish, doubling is possible (but not obligatory) with full indirect object nominals, but not with full direct objects. In contrast, when the object is a pronoun, doubling is obligatory in both cases. As Miller and Monachesi (2003: 87) describe matters, some dialects, particularly Latin-American, double full indirect object expressions whether pronominal or not, and have a strong preference for doubling when the direct object is animate and speciﬁc. As in Bulgarian, the facts are in some ways rather intricate, but the descriptive parameters are clear. We say that where clitics can occur, Agreement (at least in the sense of feature copying) can apply. Where doubling is impossible, we say that this agreement must also be coindexing; and where doubling occurs, we say that coindexation does not. The kind of complications which we see in these cases are not at all peculiar to clitic constructions. Bresnan (2001: 146-7) notes that verbal agreement and clitics may both be sensitive to the same factors, such as pronouns versus full nominals, the character of a nominal as deﬁnite, speciﬁc, animate, etc. We saw above that the nature of possessor agreement in Finnish depends on whether the possessor expression is a human pronoun or not. In general, it is clear that agreement systems (in the broad sense of this notion) can involve the same parameters, regardless of whether the results are realized by verbal morphology or by pronominal special clitics. In a language like (standard) French, the situation is somewhat simpler. The agreement represented (optionally) by an object clitic is of the “strong” (coindexing) sort, and no overt nominal of any sort can appear to double the clitic. In this language, subject agreement is obligatory, weak, and word-based, while object agreement is optional, strong, and phrasally realized. A clitic analog to the agreement pattern of Pemon is furnished by Surmiran, the form of Rumantsch discussed above in section 7.5. In this language, argument positions can be occupied freely by full nominals or by pronouns. Alternatively, non-subject arguments can be referred to by clitics associated with the verb, but there is no doubling: that is, any given argument is

Pronominal Clitics

243

represented either by a nominal expression (including pronouns) or by a clitic, but not both. These points are illustrated by the examples in (8.15). (8.15)

a. Ursus ò purto las bulias a nous Ursus has.3sg bring.pp the mushrooms to us Ursus brought the mushrooms to us b. Ursus ans= ò purto las bulias (*a nous) Ursus us has.3sg bring.pp the mushrooms to us Ursus brought us the mushrooms c. Ursus las= ò purto (*las bulias) a nous Ursus them.fem has.3sg bring.pp the mushrooms to us Ursus brought them to us ans= ò purto (*las bulias) d. Ursus las= Ursus them.fem us has.3sg bring.pp the mushrooms (*a nous) to us Ursus brought us them

Non-subject agreement in Surmiran is thus like French, with one difference. In French, non-subject (“disjunct”) pronouns are not generally available except in special contrastive or other strongly stressed contexts, while in Surmiran, pronouns are freely usable in non-subject positions. Another difference comes from the fact that French freely allows combinations of multiple clitics, while combinations such as that found in (8.15d) are tightly restricted in Surmiran. In particular, the only acceptable combinations consist of a third-person direct object clitic followed by a non-third-person indirect object. Substituting clitics for both nominals in (8.16) thus leads to an ill-formed sentence. (8.16)

a. Tgi dat igl matg a Gelgia? who gives.3sgPres the bouquet to Gelgia Who is giving the bouquet to Gelgia? b. ?*Tgi igl= la dat Who it her gives c. ?*Tgi l’= igl dat Who her it gives

Pronominal special clitics are generally used to index arguments of a predicate, but in some languages, clitics appear that do not correspond to any argument. An example is provided by verbs in French (and many other languages) that require the presence of a reﬂexive clitic without projecting any corresponding argument, as in (8.17). These “lexical reﬂexives” are sometimes known as “pronominal verbs” in the literature.

244 (8.17)

Aspects of the Theory of Clitics a. Marie s’est évanouie ‘Marie fainted’ b. Je me suis comporté comme un fou ‘I behaved like a madman’

Again, there is a parallel with word-level agreement. As discussed in Anderson (1991), verbs in some languages may show agreement with more arguments than they have. For instance, in most Algonquian languages, the conjugation of a verb indicates clearly whether it is treated as transitive (i.e., taking an object) or intransitive; and whether the Theme argument (intransitive subject or transitive object) is grammatically Animate or Inanimate. In some instances, however, verbs are conjugated in a way that indicates the presence of an argument that they do not project. The Menomini examples in (8.18) all involve syntactically intransitive verbs whose inﬂection includes “agreement” with a non-existent Inanimate object. (8.18)

a. noqnonam ‘he swims’ (Transitive Inanimate inﬂection) b. nema;mi;qtEhko;skanan ‘I go bare-legged’ (Transitive Inanimate inﬂection; plural object) c. mo;hkanam ‘he uncovers it’ or ‘it (a heavenly body) rises’ (Transitive Inanimate inﬂection)

Conversely, other Algonquian languages such as Maliseet (Sherwood 1986) may fail to show agreement with a core argument that they do in fact project. For instance, the verb in sentence (8.19) is conjugated as Animate Intransitive, despite the fact that it takes a direct object. (8.19)

cik@ni can pitk@me John pack.ai.indic.3 apple.anim.obv.pl John packed the apples

The complex inﬂectional patterns of Georgian include a number of deviations between syntactic and morphological argument structure. Some of these are illustrated by the sentences in (8.20), all of which include inﬂection (underlined) for an extra argument which is not present in the syntactic representation. (8.20)

a. kari uberavs ‘the wind blows’ (inﬂected for 3sg io) b. sonat.a dauk.ari (ˇcem-tvis) ‘you played a sonata (for me)’ (verb inﬂected for 3sg io) c. mjinavs ‘I’m sleeping’ (3sg do)

As discussed in more detail in Anderson (1991), these facts follow if the agreement representation which characterizes the verb (and thus, in the clitic

Pronominal Clitics

245

case, is inherited by the including phrase) is not necessarily isomorphic to its subcategorization or argument structure. In such cases, the exact collection of features that appears in the Morphosyntactic Representations triggering morphological Agreement and/or the introduction of special clitics is motivated only in part by the syntactic representation. Such a situation has sometimes been asserted to be impossible by those working in heavily syntactic theories of morphology, such as Distributed Morphology, but in fact it is not uncommon in the languages of the world. Sometimes a clitic results from something other than the argument structure of the clause or the lexically determined agreement structure of the verb. Jaeggli (1986) discusses “ethical datives” such as me ‘(on) me’ in example (8.21). (8.21) Juan me le arruinó la vida a esa chica Juan me her ruined the life to that girl Juan ruined that girl’s life on me These datives do not appear to correspond to any argument position present in the syntactic representation, and are apparently limited to ﬁrst and second person. One possible account would make appeal to a “derivational” rule (along lines hinted at in the discussion of Tagalog in section 6.4) which maps the Morphosyntactic Representation of the clause onto a new one with an extra agreement layer corresponding to the ethical dative while also adding the semantic content corresponding to “on me/us/you” to the meaning of the sentence. The extra Morphosyntactic content then triggers the introduction of a clitic, without alteration in the basic syntactic structure of the clause. Much remains to be done to ﬁll out and substantiate this analysis, but it seems promising.

8.3 Clitic Climbing Another construction which has ﬁgured prominently in the literature on pronominal clitics is known as “Clitic Climbing.” In Italian examples such as (8.22), clitics ﬁnd themselves associated with a verb other than the one that subcategorizes for the argument they specify. (8.22)

Mario lo= vuole leggere Mario it wants to.read Mario wants to read it

This arises with a limited set of matrix verbs that take non-ﬁnite complements. Rizzi (1978, 1982) and Burzio (1986), among others, argue that it is

246

Aspects of the Theory of Clitics

associated with a syntactic reanalysis of the embedded structure that is triggered by the matrix verbs in question. While their complements are introduced with the structure of a clause (including some sort of CP layer), the result of this Restructuring (as the reanalysis has come to be called, following Rizzi) is that the matrix and embedded verbs are no longer separated by a clause boundary after it applies. Restructuring is an optional process; when it does not apply, the clitic(s) must appear with the embedded inﬁnitive as in sentence (8.23). (8.23)

Mario vuole legger =lo Mario wants to.read it Mario wants to read it

Let us assume that the result of Restructuring is a structure something like that in (8.24). (8.24)

[ [ vuole] [ [ leggere] pro3sg.m]] VP V

VP V

What we want in this case is for the clitic lo that speciﬁes the phonologically null proi that is the object of leggere to be introduced within the domain of the ﬁnite verb vuole rather than in the domain of the embedded verb. Let us assume that Agreement operates as expected, adding content to the Morphosyntactic Representation of [ leggere] in agreement with the object V argument proi . This content, by the Head Feature Convention, is inherited by the VP that this verb heads. Of course, since Italian has no rule introducing pronominal clitics at the VP level (but only within V), the presence of this material at this level of structure has no direct effect. However, we can then say that features characteristic of an embedded VP are also inherited by an including VP. This requires a notion akin to that of Grimshaw’s (2000) “Extended Projection” to characterize the scope of the relation involved here, but the intuition seems clear. Such inheritance will be blocked by the intervening CP structure when Restructuring has not taken place. Given that the agreement material thus characterizes the matrix VP, it is naturally inherited (again, via the identity required by the Head Feature Convention) by the verb that projects that phrase—in this instance vuole, where it can trigger the introduction of a pronominal special clitic. Where more than one clitic is present, either all of them “climb” or none does, as illustrated in (8.25). (8.25)

=glie =lo a. Mario vuole dar Mario wants to.give him it Mario wants to give it to him

Pronominal Clitics

247

b. Mario glie= lo= vuole dare Mario him it wants to.give Mario wants to give it to him c. *Mario gli= vuole dar=lo d. *Mario lo= vuole dar=gli This is true even in languages where it is not possible to realize both clitics in association with a single verb. Recall that in Surmiran, sequences of two third-person clitics are not allowed: one or both of the arguments of a verb like dar ‘give’ must be represented by a full pronoun in argument position, since only one can be represented only by a clitic. Clitic Climbing in Surmiran is obligatory with causatives (formed with lascher ‘let’ and far ‘make, do’). According to the normative grammar (Signorell, Wuethrich-Grisch, and Simeon 1987), it is not possible with “modal” matrix verbs like leir ‘want’, but many speakers accept sentences with a clitic associated either with the matrix or with the embedded inﬁnitive with such verbs, as well as with the past-tense auxiliary, as illustrated in (8.26). (8.26)

a. Nous lagn la= tarmetter dumang we want.1pl it.fem to.send tomorrow We want to send it tomorrow b. (?) Nous la= lagn tarmetter dumang betg igl= cumpro c. Te n’=ast you.sg neg-have.2sg not it bought You haven’t bought it d. Te n’=igl= ast betg cumpro

When the embedded inﬁnitive has two associated third-person arguments, it is not possible to have one clitic with each verb, even though it is also excluded to have both together. Some of the possibilities are laid out in (8.27). (8.27)

a. Ia vi dar el ad ella I want.1sg to.give it.masc to her I want to give it to her b. Ia igl= vi dar ad ella c. Ia la= vi dar el d. Ia vi igl= dar ad ella e. Ia vi la= dar el f. *Ia vi la= igl= dar

248

Aspects of the Theory of Clitics g. *Ia la= igl= vi dar h. *Ia la= vi igl= dar i. *Ia igl= vi la= dar

In Surmiran, it appears that the Morphosyntactic Representations that would result from applying Agreement to two separate third-person arguments are ill-formed, regardless of where one might try to realize the agreement material as clitics. On the analysis proposed here, pronominal special clitics are a kind of phrase-level agreement phenomenon. If the functional content that triggers the introduction of such clitics can “climb” in the way we have seen, we would expect that the same might be true of functional content triggering word-level agreement in the form of verbal inﬂection. And in fact, when we look for such a parallel, we ﬁnd it. Bobaljik and Wurmbrand (2002) show that the Siberian language Itelmen has a restructuring construction comparable to that of Italian, and that when Restructuring applies, agreement material associated with an embedded verb can show up on the matrix verb. This construction, which Bobaljik and Wurmbrand refer to as “Long Distance Agreement (LDA),” is illustrated in (8.28). (8.28)

a. t’-utu-z-in @lˇcqu- aì- iì 1sg.sbj- unable- pres- 2sg.obj see- fut- inf I am unable to see you b. na @ntxa- Bum=nin kma jeBna- s he forget- 1sg.obj- 3 me meet- inf He forgot to meet me

While the Itelmen facts appear to provide a rather close analogy to the better-known construction in Italian, it must be noted that other instances of apparent “LDA” do not yield to the same analysis. Polinsky and Potsdam (2001) for example discuss a similar construction in the North-East Caucasian language Tsez, and show in detail that it is not based on Restructuring. This construction has no clear analog in the grammar of clitics, and I will not discuss it here. Another construction which should probably be mentioned here is the well-known phenomenon of Complementizer Agreement found (at least) in some dialects of Dutch and Flemish and in some German dialects as well (Haegeman 1992; Hoekstra and Smits 1998). The examples in (8.29) from the Limburg dialect of Dutch are illustrative.

Pronominal Clitics (8.29)

249

a. veurtot- s tiech te bruk zuu- s before- 2sg you.sg the bridge see- 2sg before you see the bridge b. (iech waet neet) boe- t ger zuu- s (I know not) where- 2pl you.pl be.2pl (I don’t know) where you (pl) are

The analysis that suggests itself here assumes that agreement material is inherited by all of the heads in the verb’s “extended projection,” up to the level of a complementizer head of CP. Morphological agreement can then take place not only on the verb itself, but also on the complementizer. The properties of this construction have been the basis of an extensive literature; much more discussion is required to incorporate the results of that literature and resolve its outstanding issues, but the line of attack suggested here does appear to provide a basis for that work. Obviously, there is much more to be said about Clitic Climbing, and the present remarks do not purport to constitute a full theory of the phenomenon. An extensive survey of a broader class of these “Complex Predicate” constructions in the principal Romance languages is provided by Abeillé and Godard (2003), and I have only touched the surface of the intricate array of properties they discuss. Nonetheless, it seems clear that these facts do not in themselves pose special problems with regard to the overall framework proposed here for the grammar of pronominal special clitics. Indeed, that framework appears to offer new possibilities for unifying the analysis of phenomena that have previously been treated in isolation from one another.

8.4 Subject Clitics Most of the attention in the literature dealing with pronominal special clitics is devoted to clitics that correspond to (direct and/or indirect) objects, but these are not the only pronominals that a theory must account for. In at least some languages we also ﬁnd clitics that correspond to subjects, and these present issues of their own. I refer here not simply to the phonologically weak pronouns that may occupy subject position, as in English, but rather to special clitics introduced in association with the verb (or perhaps elsewhere), which may (and in some cases, must) be “doubled” by a full nominal in subject position. Surmiran is one such language, as we have already seen. In section 7.5, we saw that when the verb is “inverted” (that is, displaced to a position outside the VP from which it C-commands the VP-internal subject position), special clitics may be introduced post-verbally in agreement with the subject.

250

Aspects of the Theory of Clitics

These do not necessarily replace, but may rather double the subject nominal. Similar subject clitics are found (under slightly different conditions) in Engadine Rumantsch (Puter and Vallader); their occurrence in Surselvan, the best-known form of Rumantsch, is much more limited and associated with “non-standard” speech. These pronouns are one of the topics of Linder (1987), though the Surmiran data there are quite limited and not drawn from current usage. In some languages of northern Italy (Poletto 1993, 2000) and FrancoProvençal (Roberts 1993a), as well, special clitics associated with subject position have been discussed, and we will turn brieﬂy to those facts after a summary of the Surmiran data and some reasons for calling the elements in question clitics rather than verbal agreement markers. The full set of Surmiran subject clitics was given in Figure 7.4 in section 7.5 of the preceding chapter. Recall that these include overt markers in all cases except the second-person plural. But why should we call these elements (special) clitics, rather than treating them as aspects of the morphology of the verb? They do meet standard criteria for analysis as clitics, as discussed in Chapter 2: for instance, they are not sensitive to lexical (as opposed to purely phonological) properties of their hosts, and they are attached “outside of ” clearly afﬁxal material. But is this sufﬁcient? And if they are clitics, what is their nature? Do they have a referential function, like the object clitics of various Romance languages (including Surmiran), or do they simply register the person, number, and gender of the subject, like the pre-auxiliary subject clitics of FrancoProvençal Valdôtain (Roberts 1993a)? The post-verbal elements in Figure 7.4 cannot simply be morphology registering the properties of the subject, like agreement markers. One piece of evidence in support of this is the fact that they play an essential role in sanctioning the omission of a pronominal subject. As we have seen, (cf. examples (7.47) and (7.48) in Chapter 7), subjects cannot in general be omitted in Surmiran, despite the presence of relatively “rich” agreement on the verb. This shows that normal verbal agreement is not of the strong or coindexing type. It is only the presence of a post-verbal clitic that allows (though it does not require) the subject to be omitted. The agreement that gives rise to the subject clitics, that is, involves optional coindexing with the subject position; and when this occurs, phonologically null pro occupies that position and excludes an overt nominal. At a minimum, then, the operation of Agreement which gives rise to verbal morphology is distinct from the Agreement that gives rise to the subject clitics. The referential nature of the clitics is also supported by the fact that they cannot appear in association with non-referential subjects, such as navot

Pronominal Clitics

251

‘nothing’, nign ‘none, no one’, etc., as in (8.30), although such sentences do show the expected subject agreement on the ﬁnite verb. (8.30)

a. Nign n’am ò(*’l/*’la) anvido no one neg=me has- 3sg(=3sg-m/f) invited No one invited me b. Ossa n’am ò(*’l/*’la) nign anvido today neg=me has- 3sg(=3sg-m/f) no one invited Today no one invited me

Another difference between the clitic elements in Figure 7.4 and verbal agreement is a phonological one, which shows up in the second-person singular form. The regular second-person singular desinences in all tense forms in Surmiran end in -s (e.g., cantas ‘you (sg) sing’). A very general rule of Surmiran phonology replaces /s/ by [S] before all consonants except [l]. We would therefore expect the form cantast ‘sing-2sg=2sg’ to be pronounced *[kánt@St], parallel to scolast [Skól@St] ‘teacher’, but this does not happen: cantast is pronounced [kánt@st], with [s] and not [S]. In this respect, the sequence of 2sg ending followed by =t is comparable to what we ﬁnd in compound forms like (las) clavs-tgesa [kláfstCèz@] ‘(the) house-keys’, not *[kláfStCèz@]. Evidently, the pre-consonantal s→S process (whether this is expressed as a rule or as some set of constraint(s) that require this replacement) is an aspect of the lexical phonology which is no longer active at the point post-verbal clitics are introduced (or when compounds are formed). These facts suggest that the elements in Figure 7.4 are added post-lexically, as would be expected of clitics (though not of inﬂectional agreement markers). I conclude, then, that the elements in Figure 7.4 are clitics and not verbal morphology. They are introduced optionally (and post-lexically) in constructions that call for subject–verb inversion, and when present, sanction the (otherwise ungrammatical) omission of pronominal subjects. Such omission is not obligatory, however, and the clitics can be doubled by an overt nominal or pronominal subject phrase, corresponding to the optionality of the coindexation aspect of this distinct Agreement process. The great bulk of work on clitics linked speciﬁcally to subjects (as opposed to subject clitics forming part of a much more general system, as in Tagalog or Warlpiri) has focused on the languages of Northern Italy, including both those considered “dialects” of Italian and others, such as Friulian and FrancoProvençal. Poletto (2000) provides many useful references and discusses a wide range of data from these systems, which differ in many intricate details from

252

Aspects of the Theory of Clitics

one another. It is quite impossible to cover anything like the whole of that ground here, and in any event the differences between the theoretical assumptions of most of this literature and those of the present book would require even more extensive discussion. All I can realistically hope to accomplish is a rapid sketch of the ways in which some of these facts might be expected to ﬁt into the present framework. Poletto (2000) distinguishes four types of subject clitic. She argues that each of these is associated with a different functional category in clause structure, an account which is at odds with the assumptions of this book and thus not directly available here. Fortunately, that is not a necessary interpretation of the facts, since the required distinctions can be represented in other ways. The ﬁrst of the four types, called “invariable clitics” by Poletto, is a marker that is the same regardless of the person and number of the subject. For instance, the Swiss Lombard dialect of Lugano has an initial particle a which occurs with all subjects as in (8.31). (8.31)

a. A= vegni mi cl come.1sg I I come b. A= ta= vegnat ti cl 2sg come.2sg you.sg You (sg) come luu c. A= vegn cl come.3 he He comes d. A= vegnum cl come.1pl We come e. A= vegnuf cl come.2pl You (pl) come f. A= vegn lur cl come.3 they They come

It is not at all clear why such an element should be called a “subject clitic,” since it does not depend in any way on the properties of the subject. As noted ﬁrst by Benincá (1983), invariable “subject clitics” are found in sentences that convey new information or in exclamative contexts. Since there is no apparent

Pronominal Clitics

253

morphosyntactic property of a clause corresponding to this characterization, it appears that they should be treated along the lines of the Tagalog “particle clitics” discussed in section 6.4. I will therefore assume that they are introduced derivationally at the level of CP, contributing their phonological content together with the semantic/pragmatic interpretive content identiﬁed by Benincá. A second type of “subject clitic” is illustrated by the Friulian dialect of San Michele al Tagliamento. As illustrated in (8.32), an initial element in this language appears as i when the subject is ﬁrst or second person, but as a with third-person subjects. (8.32)

a. I= mangi cl eat.1sg I eat b. I= ti= mangis cl 2sg eat.2sg You (sg) eat mangia c. A= l= cl 3sg.m eat.3sg He eats d. I= mangin cl eat.1pl We eat e. I= mangè cl eat.2pl You (pl) eat f. A= mangin cl eat.3pl They eat

Poletto calls these “deictic clitics.” For reasons to be discussed below, I will treat them as special clitics introduced in the domain of the head of CP, and thus as a type of clitic analogous to the phenomenon of Complementizer Agreement mentioned above. The third and fourth types of subject clitic treated by Poletto involve clitics attached to the verb that reﬂect speciﬁc person and number combinations in the subject. Some of the languages she considers have such clitics speciﬁcally for second-person singular subjects, and some for second- and third-person singular; these she refers to as “person clitics.” Other languages have such clitics for all third-person subjects, distinguishing gender and number, and these

254

Aspects of the Theory of Clitics

are labeled “number clitics.” For reasons that seem primarily theory-internal, these two types are distinguished and assigned to distinct functional categories in Poletto’s (2000) analysis. I will treat them together as “personal subject clitics” and consider their central property to be their orientation with respect to the main verb, as opposed to the head of CP. I assume they are introduced as special clitics within the domain of the head of IP. The northern Italian “subject clitics” thus fall into two major groups, which we can think of as “CP-clitics” and “IP-clitics.” In treating the “person” and “number” clitics as associated with the IP system, while the “invariable” and “deictic” clitics are associated with the CP system, I follow the basic architecture of Poletto’s analysis, though not in detail. Support for this broad division comes from several sources, as Poletto shows. First, we can note that CP clitics and IP clitics can co-occur. In example (8.31b) we see an invariable clitic combined with a 2sg personal clitic, while examples (8.32b,c) show a deictic clitic combined with personal clitics. On the other hand, we do not ﬁnd combinations of two distinct CP clitics or IP clitics. The possibility of combining CP and IP clitics follows from the proposal that agreement material is (potentially) inherited by all of the heads within the extended projection of the agreeing verb, and can potentially be realized in any of several ways: as agreement on the main verb; as a clitic associated with this verb or (in the case of clitic-climbing constructions) as agreement or a clitic on a higher main verb within a complex predicate (or restructuring) structure; or as either agreement or a clitic associated with the head of the dominating CP. Each of these realizations is potentially a rule distinct from the others, and as we have seen, they can co-occur in various combinations. The one kind of combination that apparently is not possible, however, is to have multiple instances of the same realization driven by the same agreement material. Thus we ﬁnd (as in Surmiran) that in clitic-climbing structures the clitic may be realized either on the lower verb or on the higher one, but never on both simultaneously. The possibility of having “subject clitics” of the CP and IP classes co-occur, then, argues that these are two quite distinct types of realization of agreement material.10 Another property distinguishing the two types of clitics, noted by Poletto, is the fact that CP clitics may obligatorily coalesce with the complementizer 10 If “personal subject clitics” are a single category, as assumed here, that accounts for the impossibility of having both “person” and “number” clitics in Poletto’s sense in the same language. Some other factor must be responsible for the fact that “invariable” and “deictic” clitics do not co-occur, though this may in fact be an accidental gap, since the number of languages attesting either is relatively limited.

Pronominal Clitics

255

phonologically, In the Loreo dialect of the Veneto region, an invariable clitic a occurs, which obligatorily coalesces with a vowel-ﬁnal complementizer as shown in examples (8.33). (8.33)

a. Ara ch’=a vegno look that- cl come.1sg Look, I am coming b. *Ara che a vegno c. No so s=a vegno not know.1sg if- cl come.1sg I do not know whether I will come d. *No so se a vegno.

On the other hand, the personal subject clitics in this language coalesce only optionally with a complementizer, as shown in example (8.34). (8.34)

vien a. Ara che el= look that 3sg.m come.3sg Look, he is coming b. Ara ch’el vien

Since CP clitics like Loreo invariable a are introduced in association with the complementizer, it is reasonable that they could trigger allomorphy in that element. Personal subject clitics like el, on the other hand, are introduced in association with the verb, and their combination with a preceding complementizer element is governed by the phonology. In this case, the reduction of such vowel sequences in the post-lexical phonology is evidently an optional process. A ﬁnal difference between subject clitics of the two types is the fact that “invariable” and “deictic” clitics consistently precede “strong” negation markers (in the sense of Zanuttini 1997), while pronominal subject clitics generally follow them. This follows if such negation appears within the IP, typically at its left edge. A CP clitic will be outside this structure, and thus precede the negation, while a clitic introduced within IP in association with the verb will follow it. Poletto’s (2000) study and the literature on which it is based explore a number of additional properties of subject clitics in the languages of northern Italy and adjacent regions, and I cannot address them all. Nevertheless, it seems clear that the basic distinctions to be made in that work are natural ones within the present framework, just as I have argued above for other aspects of the behavior of pronominal special clitics.

This page intentionally left blank

9 Clause Structure and the Grammar of Incorporation

The topic of Noun Incorporation may appear to have little to do with the grammar of clitics.1 Nevertheless, the choice between syntactic and lexical analyses of this construction and the framework developed in this book for the description of special clitics both bear on a set of central issues in the theory of grammar: the nature and scope of “Head Movement” and the role of abstract functional categories in clause structure. I assume that the functional categorial structure of clauses is represented by the features assigned to a limited set of fairly traditional syntactic categories—basically, the standard phrasal types projected by lexical categories, plus I/IP and C/CP; and D/DP for the nominal system. Much of the syntactic literature since the late 1980s, however, presumes that a much wider array of categories, functional rather than lexical in their basis, deﬁne structural constitutent types that stand in a hierarchical relation to one another. If special clitics are not structurally autonomous elements in syntactic representation but instead the overt markers corresponding to the (possibly quite complex) feature structure of syntactic categories, that suggests that the number of such categories that are fundamentally distinct from one another may be quite limited. If this formal possibility exists for the overt expression of functional content, then simply identifying some component of functional organization does not automatically require us to posit new layers of projecting structure. Instead we might assume, in the absence of evidence to the contrary, that such content corresponds to a feature associated with a level of structure whose internal complexity can be established independently. The possibility of analyzing syntactic structure in terms of a vast array of functional categories, each representing essentially a single property but each projecting a full set of head, speciﬁer, complement, and maximal projection, requires that the inﬂected words appearing in the surface forms of sentences 1

Much of this chapter derives from Anderson (2000a) and Anderson (2000b).

258

Aspects of the Theory of Clitics

cumulate the content of a number of layers of structure. This in turn relies on the generality of the mechanism of Head Movement, the operation which eventually conﬂates the presumed functional elements into a more limited degree of arboreal complexity in the surface form. The presumed properties of Head Movement itself, however, are heavily dependent on arguments originally supplied by Baker (1987), founded in turn on the analysis of Incorporation constructions—particularly Noun Incorporation. But if the constructions on which this theory is grounded do not actually involve syntactic movement at all, then the presumed generality of an operation of Head Movement, and with it the apparatus of functional categorial structure that depends on it, is much less self-evident. If Noun Incorporation does not involve Head Movement, the foundations for that apparatus are less secure—or at least different. And if special clitics and related phenomena do not depend on elaborated functional structure, the need for a theory of Head Movement to underpin it is correspondingly reduced. I do not reject all kinds of displacement from one head position to another. In the analysis of verb-second constructions in Chapter 7, for example, I have assumed verbs move from the head of VP to the head of IP and/or CP. However, the extension of this kind of displacement to the full range of constructions that have been attributed to Head Movement does not follow, and indeed doubts about this operation have begun to surface among those pursuing the Minimalist Program. The research strategy of the present book converges with the adoption of a non-syntactic approach to Noun Incorporation in facilitating the limitation of Head Movement to a narrow class of well-motivated cases.

9.1 Introduction: Two Approaches to Noun Incorporation Noun Incorporation has attracted considerable attention in the literature of both syntax and morphology, because it involves the construction of units that are unquestionably words from material that gives the appearance of having been combined within the syntax. If this impression is indeed correct, this operation presents an important prima facie problem for most versions of the Lexicalist Hypothesis. Noun Incorporation is a feature of the grammar of a wide range of languages, including (as a semi-random, quite small sample) Mohawk, Chuckchee, Southern Tiwa, Classical Nahuatl, and many others. Grossly, it involves the combination of a verb stem and a noun stem within a single word, with the noun stem representing an argument of the Verb. Prototypically an incorporated noun corresponds to the direct object of a transitive verb, as in the examples in (9.1) below from Southern Paiute (Sapir 1911: 263).

The Grammar of Incorporation (9.1)

259

a. q¯am;Ú´ yaai- nÚmpÚGa‘ jackrabbit- hunt- usitative- remote past He used to hunt jackrabbits b. cû´q˙uc˙u q¯am;Ú´ vax qa- q;a‘ comp (ss) one jackrabbit- killHaving killed one jackrabbit,. . .

In some languages, the underlying object (derived subject) of an unaccusative Verb can incorporate; in other languages, unergative subjects can incorporate as well. These possibilities are illustrated in the Chuckchee examples in (9.2) below taken from Polinsky (1990).2 (9.2) a. Neyk- @k P@l- @- lg- @- gPi hill- loc snow- evid- melt- evid- aor3sg On the hill, the snow melted b. @tlon Ninqe- et- @- lPet- gPe he child- intens- evid- come- aor3sg He got many children In a few cases, the incorporated Noun appears to be some sort of adjunct, such as an instrumental or locative, as in the examples of (9.3). (9.3) a. (Huauhtla) Nahuatl (Merlan 1976): yaP ki- koˇcillo- tetePki panci 3sg 3sg.it- knife- cut bread He cut the bread with a knife b. Chukchee (Skorik 1948, apud Spencer 1995): walw@N@n g@tg- @lq@t- gPe lake- go- 3sg.subj raven.abs Raven went to the lake Noun Incorporation has long been an important subject of discussion in the theory of grammar. In the early years of the twentieth century, it was asserted by some to constitute a particularly characteristic feature of North American languages. Kroeber (1909), in a somewhat extreme attempt to rebut that notion, argued that in fact there is no such thing as Noun Incorporation, but in a classic paper Sapir (1911) provided the ﬁrst real analysis of the construction across a variety of languages.

2 Polinsky (1990) provides arguments for the “unaccusative” versus “unergative” contrast in the verbs of these examples.

260

Aspects of the Theory of Clitics

There are two basic views of the nature of Noun Incorporation, and both have long histories. One approach, which I will refer to as the “syntactic” analysis, treats Noun Incorporation as a syntactic process by which an argument of the verb (or at least part of that argument) is actually displaced from its syntactic argument position and adjoins to the verb (see, e.g., Mardirussian 1975). In favor of this analysis a priori is the natural account it offers of how the incorporated stem comes to ﬁll the semantic (T) role of a corresponding unincorporated noun. Proponents of this view in the generative literature include Sadock (1980, 1986), who argued, from the apparent need to form words in the syntax by moving a noun into the same word as a verb, that the Lexicalist prohibition against syntactic manipulation of the internal form of words must be wrong. Actually, the construction discussed by Sadock in West Greenlandic (“Eskimo”) is the formation of “denominal verbs,” and several other authors (Sapir 1911; Mithun 1984, 1986; Gerdts 1997) have preferred to distinguish this from Noun Incorporation. From the morphologist’s point of view, classical Noun Incorporation involves combining two stems, each of which is (in principle) a distinct lexical entry, and by deﬁnition, this is an instance of compounding. Denominal verb formation in languages like those of the Eskimo-Aleut or the Wakashan families, in contrast, involves the addition of one of a large inventory of bound afﬁxes to a stem. Morphologically, this is a kind of derivation, not compounding, because the afﬁxes involved never occur independently as stems; and that is what causes authors like Mithun and Gerdts to resist grouping the two constructions together. As Sadock puts the issue, though, both raise the same problem for syntax– morphology interactions, at least for linguists (like Sadock) who accept the possibility that bound morphological material can be syntactically autonomous. Regardless of deﬁnitional issues, the analysis proposed by Sadock for West Greenlandic certainly involves Head Movement, and so it is relevant to my concerns here. A conceptually similar (but technically very different) position is developed by Mark Baker, who has explored it at some length in one book (Baker 1987) and made considerable use of it in another (Baker 1995). Baker also maintains that noun heads move in the syntax to take up their incorporated surface positions inside verbs. But Baker’s theory of the kind of movement involved goes much farther, and in fact is the basis of the rather more general notion of Head Movement in syntax, since he extends the operation founded on the properties of Noun Incorporation to many other sorts of putative movement of heads.

The Grammar of Incorporation

261

The alternative to these syntactic movement analyses is the “lexical” account of Sapir (1911), Mithun (1984), Anderson (1992), and others. On this picture, the noun-plus-verb combinations we refer to as instances of “Noun Incorporation” are built in the lexicon, not in the syntax. The formation of such [ [ stem1 ] stem2 ] combinations, with [ stem1 ] V N

N

interpreted as a speciﬁc argument of [ stem2 ], is surely within the scope of V the type of regularity found in the lexicon. Something quite similar is surely involved in the case of “synthetic” compounds: duck hunting ([ [ duck] N

N

[ [ hunt]ing]]) is built from [ duck] and [ hunt] and has an interpreN V N V tation in which the noun is taken as specifying one of the arguments of the associated verb. Some have argued for a syntactic incorporation analysis of such compounds, but the appeal of such a view is limited. Compounds are evidently structurally parallel to non-compound members of basic lexical categories, and are typically opaque to syntactic processes. The possibility of building all lexical compounds in the syntax was tried out in the early days of generative grammar (Lees 1960), but the context for that proposal was one in which no serious alternative existed: at the time, there really was no theory of the lexicon. Most authors today, though, agree that the formation of compounds takes place in the lexicon and not in the syntax. Compounds like duck hunting involve not only a relation between the noun and the argument structure of the verb, but one which is quite like what we ﬁnd in Noun Incorporation constructions. In both cases the noun most typically corresponds to an argument of the verb that would ﬁll the syntactic direct-object position and the T-role of theme. This is not always the case, however. In other compounds like earthquake, sunrise, landslide, etc., the noun apparently corresponds to the subject of an intransitive verb. In these examples, we might invoke the Unaccusative hypothesis and say that the argument in question represents an underlying direct object. Such an account is less plausible, though, for examples such as crybaby, ﬂashlight, workman, playboy, etc., that seem to involve a noun specifying the agentive argument of an unergative verb. Such verbs, on at least one understanding, are exactly those that cumulate the agent and theme roles and associate both with a single logical variable linked to the verb’s single (subject) argument. What seems constant about all of these cases is the fact that the associated noun typically corresponds to the T-role of theme. Other compounds, though, correspond to other non-thematic Noun Incorporation types: e.g., hand laundry, cottage industry. Once we admit the possibility that all of these verb-argument(/adjunct) relations can be established by a lexical rule, as they

262

Aspects of the Theory of Clitics

must be if we are to deal with true compounds, the initial motivation for a syntactic account of Noun Incorporation becomes much less persuasive. As with the syntactic analysis, the lexical one has some apparent prima facie advantages. One argument that seems to point to a lexical analysis is the fact that the shape of incorporated nominal elements often differs phonologically from a corresponding free form in unpredictable ways. In some instances, such variation is limited to a small number of suppletive forms: e.g., in Mohawk (Mithun 1984: 876) “-nahskw-‘domestic animal’ appears only incorporated, while a semantically equivalent stem -tshenv appears only as an independent N[oun].” In other languages, however, this phenomenon may be quite pervasive. Thus, in the Australian language Tiwi (Osborne 1974: 48), “[i]ncorporated forms and free forms are generally not cognate, [. . . ] as the free forms which were cognate with the existing incorporated forms have long since disappeared, owing to the rapid rate of lexical replacement in Australian languages.” Some representative pairs of incorporated and free forms drawn from Osborne (1974) are given in (9.4). In some cases, a degree of phonological resemblance can be seen (especially in some body part names), but in others there is no apparent similarity. Also, in some cases a single free form corresponds to more than one incorporated form, while in others the reverse is true. (9.4)

incorporated wuôatiuNintamiaw@riaw@rimaratiNa Niliwan ””@ t (N@) paN@laônti-

free - wuôatiNa - j@r@Nintamuôa - wuôara t@muripa t raka ”@ t raka ”@ -”taraNa

gloss forehead nose belly navel (live) wallaby (dead) wallaby buffalo

A language may show semi-systematic relations between free and incorporated forms that do not reﬂect phonological regularities found elsewhere in the language. A particularly interesting instance of this is found in the Munda language Sora and some of its relatives, as documented and analyzed by Zide (1976). In this language, (the free forms of) a number of nouns can be regarded as derived from verbal or other roots by the addition of preﬁxes, sufﬁxes, or inﬁxes. Thus @b-ga ‘feed’ is related to @r-@b-ga ‘food (being fed to an infant)’; koN ‘shave’ is related to k-@n-oN ‘razor’, etc. In some cases, the combining form of such a noun can be formed by simply removing such added morphological material: thus, the combining form of k-@n-oN ‘razor’ is -koN, and that of g tasi ‘play’ is -g si (cf. the verb g si-‘to play’ from which the noun is derived e

e

e

The Grammar of Incorporation

263

by inﬁxation). This relation has been massively overgeneralized and inverted, however. A huge number of polysyllabic free-form nouns have incorporated forms produced by removing non-existent preﬁxes, sufﬁxes, and inﬁxed -VCsequences so as to reduce the form to a canonical -CVC shape. For details, see Zide (1976); the point is simply that the elements appearing in incorporation structures are quite distinct from semantically similar free forms in ways that do not follow from the independent phonology of the language, but rather point to lexical processes. Just as the phonological form of incorporated and non-incorporated elements may require speciﬁcally lexical description, the semantics of an incorporated nominal construction may also differ, again unpredictably, from the interpretation of a corresponding construction in which the nominal is represented by a full DP in argument position. Some Mohawk examples taken from Baker (1995) are given in (9.5), where “#” in (9.5b) means that the sentence is anomalous on the interpretation parallel to (9.5a). (9.5) a. tu- s- a- yu- [a]t- háh- a- hkw- e’ dup-iter-fact-f.sg.subj-srfl- road–pick.up- punc She started her journey (lit.: picked up the road) ne oháha b. #tu- s- a- yú- ([a]te)- hkw- e’ dup-iter-fact-f.sg.subj-srfl- pick.up- punc art road She picked up the road (literal reading only) Further examples from (Huauhtla) Nahuatl (Merlan 1976) are given in (9.6); essentially every language with extensive Noun Incorporation provides instances of this sort of non-compositional interpretation of the construction. (9.6) a. i. ∅- neˇc- maka- ∅- k paPtli 3sg-1sg- give- pst-sg medicine He gave me medicine ii. ∅- neˇc- paP- maka- ∅- k 3sg-1sg- medicine- give- pst-sg He doctored me b. i. tesiwitl weci- ∅- ∅ hail fall- pres-sg (What’s falling?—answer:) Hail is falling ii. tesiwi- weci- ∅- ∅ hail- fall- pres-sg (What’s the weather like?—answer:) It’s hailing Neither the phonological nor the semantic idiosyncrasies just illustrated are to be expected if Noun Incorporation is simply a syntactic process that

264

Aspects of the Theory of Clitics

may (or may not) apply to a fully articulated syntactic structure to move some of its lexical material from one position in the tree to another. Such arguments have not received much attention in the syntactic literature, but this does not mean they are negligible: syntacticians tend to assume that both the phonology and the lexical semantics will take care of themselves in the “interpretive” components of the grammar, and the difﬁculties presented by actually working this out often carry less weight than they should. In fact, though, these aspects of Noun Incorporation structures point away from the syntax in their analysis, and toward the part of the grammar which is classically the locus of item-speciﬁc information, the lexicon. Sapir (1911) was probably the ﬁrst to propose that Noun Incorporation constructions are actually instances of lexical compounding. Since the possibility of a “syntactic” analysis in the modern sense was not really open to him, though, his arguments for this position came mostly from the formal consideration that incorporated structures involve a combination of two stems, like compounds, together with phonological and semantic idiosyncrasies of the sort adduced above.

9.2 Fleshing Out the Syntactic and Lexical Accounts Noun Incorporation, then, is a construction type with some speciﬁc properties: the combination in a single word of verbal and nominal components, where the nominal generally supplies information about one of a small number of thematic possibilities in the argument frame of the verb; and where the whole functions as a verb. Its analysis can be approached from either of two perspectives (syntactic or lexical) each of which seems to have some natural conceptual afﬁnities with the construction’s basic character. The question to be addressed is which of these (if either; or perhaps some combination of the two) is correct. It is quite important for syntactic analyses based extensively on Head Movement operations that syntactic Noun Incorporation be at least possible and common, even if other sorts of “Noun Incorporation” constructions exist as well. For those who maintain that “the syntax neither manipulates nor has access to the internal form of words” (Anderson 1992: 84; see Chapter 2 above), though, as at least many versions of the Lexicalist Hypothesis require, it is quite important not to allow syntactic rules to put words together in this way. The choice of analyses for Noun Incorporation constructions thus has a good deal of importance for the theory of how morphology and syntax are related. And as noted above, Baker’s theory of Head Movement (in support of which Noun Incorporation is the most obvious empirical domain) is central to the

The Grammar of Incorporation

265

articulated-infl account of clause structure common to much current syntax, so any challenge to the syntactic nature of Noun Incorporation undercuts this view, at least indirectly. A choice between the two accounts does not seem to follow from the basic descriptive properties of Noun Incorporation. Both views are able to describe the fact that the noun part supplies content associated with a thematic position in the argument structure of the verb part. As far as the limitations on which positions can be involved, the Lexicalist view accommodates these thematic restrictions rather straightforwardly: lexical rules often refer to the relation of theme, and compounding in particular does so. If Noun Incorporation is simply a form of noun–verb compounding, this is exactly what we would expect. As we have seen, the incorporated noun is not always a theme: it is sometimes a locative or an instrumental, as in (9.3). But this is again quite parallel to the facts about compounds, and supports the view that there is a single regularity at work in the two cases. The syntactic view derives this result in a way that is, at a minimum, rather less direct: some might ﬁnd Baker’s (1995: sect. 7.3) discussion of why only the direct-object position is accessible somewhat tortuous, but for the sake of discussion let us assume that such an account is at least possible. Similarly, I will assume that the phonological and semantic idiosyncrasies referred to above can ﬁnd a home within the syntactic account: idioms, for instance, provide a clear precedent for the assignment of non-compositional interpretations to syntactically complex structures, and it is at least possible that suppletion and other non-phonological variation occurs as a function of syntactically derived environments. If we are to choose between the two theories of Noun Incorporation under consideration, it will probably be on the basis of detailed study, not simply as a consequence of the basic descriptive regularities that deﬁne the construction. Mohawk Noun Incorporation Baker proposes to do just this. His primary evidence that Noun Incorporation is syntactic comes from Mohawk, for which an analysis is developed at length in Baker (1995). On this account, a lexical noun can be generated (as the exhaustive content of a DP) in argument position, and then moved to adjoin to the governing verb. The question naturally arises of why this movement should take place in Mohawk (but not in English). I have already introduced some aspects of Baker’s analysis of Mohawk clause structure, which he supports in detail, in section 8.1 above. According to this account, overt DPs are not licensed in argument positions in Mohawk, but only appear in adjunct positions, where they serve as appositive expressions

266

Aspects of the Theory of Clitics

(a view similar to, but not identical with, that of Jelinek 1984). If such a [ Noun] were generated in an argument position, it would thus be ill-formed. DP In order to get T-marked, such a DP has to be coindexed with an element in the verb (a key subpart of the Morphological Visibility Condition, or MVC, discussed in the preceding chapter). If that element were Agreement, the Agreement would (on Baker’s hypothesis) absorb the case-assigning property of the verb, so the overt DP would still be ill-formed. But if the noun moves so as to adjoin to the verb, the MVC is satisﬁed, the otherwise caseless DP no longer has phonetic content, and no well-formedness conditions are violated. The Morphological Visibility Condition is a parametrically determined characteristic of certain languages (called “Polysynthetic” by Baker, in a usage that deviates somewhat from the traditional sense of this term). We may ask, naturally, which languages have (syntactic) Noun Incorporation of this sort. Baker’s answer is: those that have to. That is, in Mohawk, incorporation is forced, as above. In English, on the other hand, as in most languages, the MVC does not hold, and so movement is not forced. But on Minimalist assumptions, if nothing requires some operation to take place, it is blocked. As a result, Noun Incorporation is impossible in English, since it is not forced. Now in fact many languages have constructions that look like Noun Incorporation (i.e., cases in which noun plus verb together seem to form a single word, and where the noun is interpreted as specifying one of the arguments of the verb). They differ quite a bit from one another, though. For one thing, in most Noun Incorporation languages (but not, e.g., Mohawk), the incorporated noun is always interpreted as indeﬁnite and/or generic. This is comparable to the interpretation of nouns in (English) lexical compounds, as shown in (9.7). (9.7)

a. She’s a truck driver, which is why she has back problems. b. *She’s a truck driver, which is why it’s parked over there.

The sentence She’s a truck driver means that she drives trucks, not (just) some speciﬁc truck. The truck in this compound is not available for anaphoric reference, since it is necessarily generic (or at least referentially underdetermined). In contrast, in Mohawk, an incorporated noun can refer to something that is referentially speciﬁc or deﬁnite, as the short discourse in example (9.8) from Baker (1995: 288) illustrates. (9.8) Thet2´ re’ wa’- ke- nakt- a hnínu- ’ yesterday fact- 1sg.subj- bed- ∅- buy- punc I bought a bed yesterday

The Grammar of Incorporation

267

Í- k- ehr- e’ Uwári 2- ye- núhwe’- ne’ ∅- is.subj- think- impf Mary fut- f.sg.subj- like- punc I think Mary will like it (the bed) This is just what would be expected if the incorporated noun represents a DP in an argument position, since such a DP can perfectly well be speciﬁc or deﬁnite. If the lexical analysis is to remain viable, it must do something to accommodate this possibility. Apparently, the verb-plus-noun structure that constitutes a lexically compound (Noun Incorporating) verb must be able to take an argument in the thematic position that is (also) speciﬁed by the incorporated noun stem, at least to the extent of assigning reference to that argument. The Lexical Analysis The most common view to be found in the literature assumes that when a noun is compounded with a verb, the noun satisﬁes (or saturates) the corresponding argument in the verb’s argument structure, which is also interpreted as generic or indeﬁnite. Assuming that nouns themselves have an external T-role to discharge (the “R-role,” connected with the noun’s possibility of referring), this generic interpretation corresponds to a certain sort of binding. The noun’s own external T-role (R), along with the corresponding variable in the interpretation of the verb, are jointly bound by a generic or indeﬁnite operator Genx. The relevant compounding operation, like many others, introduces this generic operator binding the logical variable corresponding to the noun’s R-role and the verbal T-role identiﬁed with it, thus saturating both and precluding an external referring expression linked to the same argument. To accommodate the facts of languages where an incorporated noun is potentially referential, we need to extend this account. Let us say that in some languages, at least, Noun–Verb Compounding is an operation that “uniﬁes” the semantics of the noun with the argument position of the verb, but without saturating the argument itself. Explicit semantic analyses of such an operation, including its application to Noun Incorporation, are provided (along somewhat different lines) by Bittner (2001) and by Chung and Ladusaw (2004); I will adopt a somewhat less formal account here, similar in the relevant respects to these. I will refer to the operation unifying the semantics of the noun with the verb’s argument position as “Restrict,” employing the terminology of Chung and Ladusaw. On this view, in a language of the relevant type the Noun Incorporating verb ﬁsh-catch is a transitive verb meaning “X catches Y, Y a ﬁsh.” This is essentially what Rosen (1989) has called “Classiﬁer Incorporation,” and which other writers have proposed for at least some incorporation structures

i

i

“Clitics” — 2005/7/26 — 17:39 — page 268 — #278

i

268

i

Aspects of the Theory of Clitics

(e.g., Di Sciullo and Williams 1987, Spencer 1995). On this view, the verb in (9.9a) would have an associated semantics something like (9.9b). (9.9)

a. [ [ fish][ catch]] = “X catches Y, Y a fish.” V N

V

fish catch b. — ⇒ — + catch — , R Agent Theme fish fish-catch — , — catch Agent Theme(,R) fish

This verb is still syntactically transitive; and while its theme argument is specified as something having the properties of a fish, it is not logically or referentially bound. As a result the verb still takes an argument (perhaps just pro, as required in Mohawk) which can bear independent reference. In this picture, Mohawk differs typologically from other languages, including those in which incorporated arguments are always necessarily generic, not in having syntactic (versus lexical) Noun Incorporation, but rather in the fact that its (lexical) Noun Incorporation rule does not saturate (or bind, logically or referentially) the argument position. The semantic operations here are independently required, to deal with lexical operations of compounding, the lexical suffixes of Wakashan and Salish languages, and the denominal verbs of Eskimo-Aleut languages— constructions often conflated with Noun Incorporation. Aleut, for example, has a class of “moveable” suffixes. These can appear on nouns, where they modify the semantics of the noun. They can also appear on verbs, however, and when they do so their semantics applies to one or another of the arguments of the verb, as illustrated in the examples of (9.10) taken from Bergsland (1997). (9.10)

hila- ku- xˆ a. hla- kucha- xˆ boy- little- abs read- fin- 3 The little boy is reading b. hila- kucha- ku- xˆ read- little- fin- 3 He (the little one) is reading c. tayaˆgu- xˆ siida- xˆ saˆga- ku- xˆ man- poor- abs sleep- fin- 3 The poor guy is sleeping d. saˆga- xˆ siida- ku- xˆ hama sleep- poor- fin- 3 dem(up there, invisible) He is sleeping, the poor one

i

i i

i

The Grammar of Incorporation

269

e. kida- kucha- ku- ng help- little- fin- 1sg/3sg I am helping him (the little one) or I (the little one) am helping him When Aleut -kucha, -ˆxsiida, etc., are added to nouns, they are semantically integrated with the meaning of the base noun by means of Restrict operating on the noun’s R T-role. When added to verbs, they are semantically integrated in the same way, with Restrict operating on some one of the T-roles in the verb’s argument structure. Kw akw ’ala similarly has lexical operations that result in combining the semantics of a nominal element with a position in an argument frame, as illustrated by the examples in (9.11) from Anderson (1992) (see also Bach 1998). (9.11)

hoqw @w@ls- ida q’isq’asde- x.a k’@lx- i q’isina a. la- i aux- they go out- dem eat currants- obj raw- dem currants those who have been eating the raw currants go out b. la- i k’@lxk’ax- suP- @m- x.at’- ida q’isina aux- they eat raw- pass- really- also- dem currants raw currants are also eaten c. k’@lxk’axa- Pax.a- ida bak’w @ma- x.a t’@mxw ali Indians- obj gooseberries eat raw- also- dem The Indians also eat raw gooseberries

In this case, the denominal formation takes the form of reduplication: the particular pattern of reduplication seen in (9.11), where the reduplicated sequences are underlined, converts a nominal stem into a verb meaning ‘eat stem’ and it is this derivational process that contributes the argument frame, where the base noun or adjective provides a semantic interpretation to be combined (via Restrict) with the theme argument position. Notice that both in Aleut and in Kw akw ’ala, the Head Movement analysis of the apparent “Incorporation” is not available, since the element that would have to move is sometimes a non-head—or at least not the head of the NP, so that its movement across another head should involve a Head Movement violation. At least on Baker’s account, that would seem to entail the conclusion that some sort of lexical operation is involved rather than Head Movement in the syntax. The proposal that Noun Incorporation involves a lexical operation (Restrict) that uniﬁes the semantics (and perhaps other properties, as we will see below) of a lexical base with those of a thematically characterized position

270

Aspects of the Theory of Clitics

in the argument structure of another base is quite close in spirit to the analyses offered by Malouf (1999) and Bittner (2001) for Greenlandic. While Malouf ’s analysis is presented within the framework of HPSG, and Bittner’s is in terms of Dynamic Semantics, their essential claims are quite parallel (mutatis mutandis) to those of the present account. “Doubling” of Incorporated Nouns Another typological difference among languages with respect to Noun Incorporation derives from the fact that in some languages an incorporated nominal can be doubled by an external phrase whose content also speciﬁes the corresponding argument position. There are two subcases to consider, depending on whether the external phrase consists only of modiﬁer material (I [a new] bed-bought ‘I bought a new bed’), or whether it also contains a head noun (My father [eight bullheads] ﬁsh-bought). The ﬁrst of these subcases is straightforward for the Head-Movement account: the head alone moves, leaving any modiﬁers, determiners, etc. in situ. The second case was much more problematic for the analysis offered in Baker (1987), because it is unclear how the head could have moved if a noun is still present in its original position. On the analysis of Mohawk, presented in Baker (1995) and summarized in section 8.1 above, however, that is no longer a problem. On that view, the doubling nominal is actually an adjunct, related to a separate (phonetically null) DP in argument position, not only in the presence of an incorporated noun but for all overt DPs. Figure 9.1 shows how this works for the relevant parts of a Mohawk sentence taken from Mithun (1984: 870). Dotted lines here indicate anaphoric links, and the arrow indicates a movement relation. The lexical analysis of constructions in which an incorporated noun stem is doubled by all or part of a nominal phrase in argument position is similar in part to the syntactic account. When an incorporated noun is doubled, the analysis is essentially the same as Baker’s. The clause-internal argument position is ﬁlled by a phonologically null pro, and the overt DP appears as an adjunct and forms a chain with that pro. The only structural difference is that the empty category in the argument position is an instance of pro rather than of trace, since it does not result from displacement. Semantically, the incorporated noun uniﬁes with an argument of the verb (by Restrict), which is saturated by its link to the clause-internal argument (pro, forming a chain with the external adjunct expression). Notice that the presence of the empty pro in argument position in Mohawk follows not from anything about Noun Incorporation, but rather from Baker’s

The Grammar of Incorporation

271

IP IP

DP

DP

IP DP

VP DP

8 bullheadsj

proi

V

N

N

V

[e]j

ﬁshj-

bought

my fatheri

Tohka niyohserá:ke tsi nahé several so- it- year- numbers so it- goes sha’té:ku nikú:ti rabahbót wahu- tsy- ahní:nu ki rake’níha this my- father eight of- them bullhead he- ﬁsh- bought [Several years ago,] my father bought eight bullheads (Mithun 1984: 870) Figure 9.1. Mohawk Noun Incorporation with Doubling

independent arguments that overt DPs in Mohawk occur external to the clause as adjuncts. In the absence of evidence for that, we could simply say that the overt nominal phrase is in the expected argument position. Since the incorporated noun did not originate there, there is no syntactic reason why some other expression should not ﬁll this position. In the case of the apparently headless DPs, where it looks as if Head Movement has taken place, we can note that the expressions themselves are wellformed as DPs in the relevant languages. In Mohawk, as in most languages, there is no overt correspondent of English one, so the object phrase in I want [(a) new (one)] consists of just the Adjective new. Such a phrase arises not by movement of its head out of the DP to adjoin to a governing verb, but rather by the selection of pro as head of the NP within the DP, with the semantics of the nominal being supplied from context (or verb-internally via Restrict, in a case like (9.9b)). On this view the headed and headless cases of doubling expressions fall together, at least in principle. With this apparatus in place, we can approach a typology of referentiality and “doubling” within the class of Noun Incorporation languages. As we shall

i

i

“Clitics” — 2005/7/26 — 17:39 — page 272 — #282

i

272

i

Aspects of the Theory of Clitics

see, this is in part similar to what we saw about “doubling” with agreement and with pronominal clitics in Chapter 8, and in part different. Three major classes of construction must be distinguished. The first of these is the case in which the noun is necessarily interpreted as indefinite, nonspecific, or generic, and where no doubling expression can be present. Here I assume that the semantic side of the compounding operation which forms the “noun incorporating” verb identifies the semantic variables corresponding to the noun’s external R T-role and the appropriate verbal argument. It also introduces an appropriate logical operator binding this combined variable, resulting in a semantics like (9.12). The syntactic subcategorization of the resulting verb is reduced by the elimination of the argument position corresponding to this variable. x — (9.12) Genx (catch Agent , Theme ) fish In some languages, the result is that otherwise transitive verb stems subcategorize as intransitives as a result of incorporation. In others, the morphological properties of the derived verb may remain transitive, although no additional argument corresponding to the “incorporated” element can be present, since it could not be bound to the already saturated argument position. I assume that this is an instance of the formal dissociation of morphological and syntactic transitivity, as discussed in Anderson (1991). The second case is that in which the object of a noun-incorporating verb can have independent (definite and/or non-generic) reference, and where this argument can be represented by an overt DP which “doubles” the incorporated noun. This type (“Classifier Incorporation”) is similar to the first. The semantics of forming a noun-incorporating compound in such a language do not involve the introduction of a logical operator binding (and saturating) the relevant argument, but only the identification of the incorporated noun’s R role with that of the verbal argument. As a result, a corresponding semantic variable is (still) available for external specification by an expression in the argument position. Of course, since the semantics of one argument of the verb have been Restricted by the interpretation of the incorporated noun stem, the externally specified expression must be consistent with that: thus, the information provided by the object of (9.9b) must (at minimum) be consistent with the property that what it designates is fish. Finally, we have the case where the incorporated noun can refer freely (even introducing new discourse referents), but where no external doubling expression can be present, as in, e.g., Nahuatl or Southern Tiwa. Clearly, the semantics of the incorporated noun must unify here with a verbal argument

i

i i

i

The Grammar of Incorporation

273

position, but the resulting verb apparently has some property that is incompatible with the presence of an overt DP argument. This does not, however, prevent phonologically empty pro from appearing in the argument position, and this pronominal provides the possibility of independent reference. The question that remains is what the relevant property could be that excludes overt expressions in the argument position corresponding to an incorporated noun in such a language. Baker’s (1995) account of a language like Southern Tiwa, where doubling is excluded but non-generic reference in association with an incorporated noun is allowed, is that the condition licensing adjuncts in association with argument positions only allows them to be licensed by pro, and not by a trace (of Head Movement). Stipulative as this is, it will not sufﬁce if incorporated nouns are introduced lexically and thus correspond to pro in argument position, and not to the trace of head movement. Apparently, overt phrases in argument positions that correspond to an incorporated noun in such a language have some property that is incompatible with the structure of a verb derived by Noun Incorporation. Let us suppose that Noun Incorporation in a language like Mohawk (which permits external doubling phrases) not only uniﬁes the semantics of the noun with that of the corresponding argument within the semantics of the Verb, but also eliminates the Noun’s external R T-role (or merges it with the T-role assigned to that argument). In Southern Tiwa, on the other hand, the incorporated noun’s R T-role is identiﬁed (but not merged) with the T-role of the argument. That means that if the argument position contains an expression headed by a noun that assigns its own R T-role, there is a T-criterion violation. Only a null pro, which has no R role to assign, can occupy such a position. Another aspect of the typology of noun-incorporating languages requires further investigation, but the outlines of a solution are reasonably clear. Languages that permit doubling may differ in which of the following schematic sentence types they admit. (9.13)

a. John ﬁsh-caught three trout. b. John ﬁsh-caught a ﬁsh. c. (*) John trout-caught a ﬁsh.

As far as is known, all languages that permit doubling of an incorporated noun stem allow sentences such as (9.13a), where the doubling expression is more speciﬁc and detailed than the incorporated noun alone. Many (though apparently not all) allow sentences like (9.13b), where the two are essentially synonymous. None, however, appear to allow sentences like (9.13c), where the doubling expression is actually less speciﬁc and detailed than the incorporated noun.

274

Aspects of the Theory of Clitics

These facts appear to result from a requirement (semantic or pragmatic, depending on one’s view of where the line between these aspects of grammar is to be drawn) that overt expressions be at least minimally informative with respect to the content already provided as part of the verb’s semantics. (9.13c) violates this because the information provided by the overt DP is actually a proper subset of that already supplied by the verb, while the overt DP in (9.13a) clearly introduces information above and beyond that present in the verb. Languages apparently differ on the basis of whether they consider an essentially equivalent expression “informative” (perhaps by virtue of the possibility it introduces of independent referentiality) or not. One further point should be noted about the properties of doubling expressions. In many cases, these provide information about the nature of the corresponding argument itself: in a Mohawk sentence like (9.14), for instance, the overt argument expression and the incorporated noun combine to specify a single referent ‘a polka-dotted dress’. (9.14) Kanekwarúnyu wa’- k- akya’tawi’tsher- ú:ni it.dotted.dist past- I- dress- make I made a polka- dotted dress In other cases, however, the external NP may supply information about a possessor of the argument in question, as in the Rembarrnga sentence (9.15) cited in Dixon (1980). paNa- warnta- na- ∅ (9.15) tiNP woman 1sg/3pl- track- see- past I saw the women’s track Note that although it is a single track that is referred to, the agreement is plural, due to the fact that it is the track of more than one woman, even though the DP tiN? ‘women’ is not overtly marked as plural. The overt argument expression, thus, is not simply something with a phonologically null head (as in English I was looking for a track, and I saw [the women’s ∅]). Agreement here is with a third-person plural, showing that ‘women’ is the head of the object DP, and not simply a modiﬁer. In such cases (often referred to as “possessor ascension,” especially in the literature of Relational Grammar; or more theory-neutrally as “external possession” in Payne and Barshi 1999), the semantics must be capable of incorporating the semantics of an external argument expression as a speciﬁcation of the possessor of an (already partly speciﬁed) argument, and not as restricting the argument itself.

The Grammar of Incorporation

275

9.3 Noun Incorporation: Syntax or Lexicon? By and large, up this point, the syntactic and lexical theories of Noun Incorporation are “tied” in that each can be said to account for roughly the same range of phenomena. Baker (1995) considers the lexical account (though not with all of the details supplied here), and argues essentially that this is the case. In fact, he says not only that Noun Incorporation in some languages is lexical (or “morphological”) rather than syntactic, but that even Mohawk has Noun Incorporation of the lexical variety, in addition to the syntactic type. This makes the theory rather close to unprovable, since any fact that appears to argue against the syntactic account can be dealt with by saying that in just such a case, the incorporation is lexical. But Baker also discusses some phenomena which he feels argue for the syntactic account over the lexical one. As the syntactic and lexical views have been elaborated above, they converge to a great extent as far as the representations they assume. But there is one difference: for Baker, the empty category present in argument positions in association with a verb that has undergone syntactic Noun Incorporation is a trace, whereas when the formation is lexical it is pro. As a result, any way in which these two possibilities could be teased apart could provide a way of discriminating between the theories. On this basis, Baker (1995: 314–29) offers three arguments that at least some cases of syntactic Noun Incorporation exist. The ﬁrst of these concerns agreement. He argues that in general, there is no agreement with the position corresponding to the source of Noun Incorporation. In this respect, he rejects Postal’s (1979) analysis of Mohawk, which claimed explicitly that there is agreement with an incorporated noun. For Baker, agreement occurs with a position containing pro, while syntactic incorporation ought to leave not a pro, but a trace and Baker makes the unusual assumption that this should not produce agreement. Absence of agreement with noun-incorporating verbs, as opposed to its presence in association with simple pro, could then argue for the kind of syntactic difference he assumes. One might expect that it would rather easy to decide this issue: just look at Noun Incorporation structures and see whether agreement with the incorporated position is present or not. But things are not so simple as that. Languages tend to be devious and evasive at just the points where the analyst really wants them to commit themselves, and Mohawk is no exception. To evaluate this matter, we must bear in mind some limitations on Mohawk Noun Incorporation. In fact, it is almost exclusively inanimates that are incorporated. This is particularly interesting in light of the fact that in Mohawk, agreement with an inanimate object is formally indistinguishable

i

i

“Clitics” — 2005/7/26 — 17:39 — page 276 — #286

i

276

i

Aspects of the Theory of Clitics

from no agreement at all. That is, when the object is inanimate the agreement markers are the same as in intransitive agreement with the subject alone. Incorporation of animates is generally disfavored, and regarded as pejorative (possibly implying the treatment of a person as an object). Why should this restriction obtain? Baker admits to having no explanation for the avoidance of incorporation of animates. One possibility, though, is that it is a consequence of the fact that it is precisely with inanimate objects that the speaker can “fudge” the issue of whether agreement is or is not present, similar to the use of modal constructions in English in those cases where no particular agreement seems right.

??am/??is/??are coming (9.16) Neither Fred nor I to the party. will come Other examples of the same sort are provided by Pullum and Zwicky (1986). They note, for example, that in German, the conjunction of Verbs taking dative objects with those taking accusative objects is only felicitous when the object NP is ambiguous between the two cases. (9.17)

a. Er findet und hilft Frauen he finds(+ACC) and helps(+DAT) women He finds and helps women b. *Sie findet und hilft Männer(n) she finds(+ACC) and helps(+DAT) men(Dat) She finds and helps men

The tension in Mohawk which is resolved by limiting incorporation to inanimates might be described as follows: where the verb contains an incorporated noun referring to an overt argument expression, that seems like “enough” to a speaker, and so the presence of a separate agreement element (as would be required if the object were animate) seems superfluous. Where the transitive object is inanimate, however, the agreement morphology is ambiguous as to whether it contains a marker referring to the object or not, so no surface discomfort results. In any event, in most cases of Noun Incorporation there is no overt indication of whether agreement is present with the ‘incorporated’ position or not, because of the formal similarity between Sbj/InanimateObj markers and intransitive Sbj markers. In addition, as Baker notes, in sentences like (9.18) where there is no external ‘doubling’ expression we can always say that the lexical operation of incorporation has constructed an intransitive verb (by saturating the argument), so the absence of agreement would follow on either the lexical or the syntactic view.

i

i i

i

The Grammar of Incorporation

277

(9.18) Tu- t- a- yako- kétoht- e’ ts- e- wir- 2háwi dup-cis-fact- f.sg.obj- appear- punc iter-f.sg.subj- baby- carry/stat She appeared carrying a baby. In other sentences, though, such as (9.19), there is clearly an animate noun incorporated. What we ﬁnd is either that there is no agreement with this noun, or else that it is (exceptionally) treated as though it were inanimate. thík2 (owirá’a) (9.19) Ra- wir- a- núhwe’- s m.sg.subj- baby- ∅- like- hab that (baby) He likes that baby This is the sort of incorporation structure that is crucial to Baker’s argument, since (on his analysis) lack of agreement is forced by the fact that the relevant position is occupied, after movement, by a trace. The lexical analysis, on the other hand, says that in such a case, the DP referring to the baby must be being treated as an inanimate. Since (animate) agreement is obligatory in such cases when incorporation has not taken place, Baker concludes that the argument position in a sentence like (9.19) must be ﬁlled with a trace rather than with (inanimate) pro. Baker (1995: 335) rejects the lexical analysis, asserting that “if Mohawks can freely view babies as inanimate entities, then one would expect that they could trigger inanimate agreement even when not incorporated [as in (9.20b) below –SRA], contrary to fact.” (9.20)

(ne owirá’a) a. shako- núhwe’- s m.sg.subj/3obj- like- hab art baby He likes them (babies) b. *ra- nuhwe’- s (ne owirá’a) m.sg.subj- like- hab art baby He likes them (babies) c. ra- wir- a- núhwe’- s m.sg.subj- baby- ∅- like- hab He likes babies

This conclusion does not necessarily follow, however. It might well be the case that the semantic content of the incorporating stem -wir-‘baby’ differs subtly from that of the independent Noun owirá’a, perhaps in being under-speciﬁed for animacy. The notion that incorporating stems are less fully speciﬁed than independent nominals in some respects is entirely consistent with the tendency

278

Aspects of the Theory of Clitics

across languages for such stems to have a somewhat more generic interpretation than full lexical nouns. It seems, however, that agreement is at least optional with animate incorporated objects, as in (9.21). (9.21)

a. Uwári ye(- ruwa)- kst2- hser- 2´ haw- e’ Mary f.sg.subj(/m.sg.obj)- old.person- nom- carry- impf ne rake- ’níha prt my- father Mary is holding my father b. Wa’- ke (- hi)- kst2- hser- áhset- e’ fact- 1sg.subj(/m.sg.obj)- old.person- nom- hide- punc I hid the old person (the old man)

And in fact Baker (1995: 336) observes in a footnote that “when the doubling material makes explicit the gender of the argument in question, the Noun Incorporation plus agreement construction is preferred where possible.” Since agreement is claimed to be impossible in any instance of (syntactic) incorporation via movement, this would appear to be a strike against the syntactic analysis. On the lexical analysis, what is the explanation for the fact that explicit agreement and Noun Incorporation do not comfortably co-occur in Mohawk? The facts are obviously rather complicated. There appears to be a preference for avoiding the appearance within a single verb of both overt agreement material and an incorporated noun referring to the same participant. On the other hand, when a verb (regardless of its internal composition) takes an object, it agrees with that object. One way to resolve this tension is to treat the agreedwith position in a Noun Incorporation construction as if it were inanimate, in which case no overt marker will appear. That provides the basis for the preference for inanimates as incorporated elements. It also underlies the sense that incorporation of animates is somehow pejorative: overt agreement with an animate object can be avoided through the treatment of the DP in question as if it were inanimate, but this is clearly derogatory. We can then say that Noun Incorporation constructions do indeed have morphosyntactic agreement, even though (with objects are—or are treated as—inanimate) this has no overt phonological consequences. There is thus no motivation for seeing their objects as (non-agreeing) traces rather than as pro. Baker notes further that in some languages which are otherwise syntactically similar to Mohawk (Tanoan languages like Southern Tiwa and Gunwinjguan languages like Mayali), overt agreement does appear with

i

i

“Clitics” — 2005/7/26 — 17:39 — page 279 — #289

i

The Grammar of Incorporation

i

279

positions that are also associated with an incorporated noun. In Ainu, which he puts in the class of polysynthetic languages (incorrectly, as shown in Kaiser 1997a), some dialects have agreement with the “incorporated” argument position and some do not. Wherever we find agreement with an incorporated position in a “polysynthetic” language, that presents a problem for Baker’s analysis. On the other hand, the lexical account derives these cases without difficulty, since they represent plain transitive Verbs. Baker’s other two arguments similarly do not provide clear proof of the need for a Head Movement analysis of Noun Incorporation. One of these derives from the fact that in Mohawk the object position associated with a noun-incorporating verb cannot be questioned with a general purpose question word. For Baker, this follows on the syntactic analysis from the fact that the argument position cannot simultaneously contain a noun to be incorporated and a question word, a conclusion that does not follow from the lexical analysis. There are two subcases to this argument. The first is that of who questions like (9.22). *ke ksá- ht- a- ya’k- e’? (9.22) Úhka wa’- khe

*1sS who factchild- nom- ∅- hit- punc 1sS/FsO Who (a child) did I slap? In this case, we see that if agreement is overtly present, the sentence is acceptable; while the absence of agreement leads to ungrammaticality. As Baker points out, the variant with agreement validates the predictions of the lexical account; but he suggests that the impossibility of the alternative without agreement supports the syntactic analysis, since it would only follow (according to him) from the incompatibility of agreement and movement traces. But this does not in fact follow: the ungrammaticality of the sentence without agreement (treated on the lexical view as involving inanimate agreement) is due simply to an evident agreement conflict. That is, the question word ‘who’ makes it explicit that the object is animate, and this conflicts with the apparent inanimate agreement. For what - questions the account is a bit different. In sentence (9.23) there is no animacy conflict between the question word and the incorporated object ‘meat’ but the sentence is still ungrammatical. (9.23) *Nahót2 wa- ha- ’wáhr- a- k- e’? what fact-m.sg.subj- meat- ∅- eat- punc What did he meat- eat?

i

i i

i

280

Aspects of the Theory of Clitics

Baker again concludes that the incompatibility of question words with incorporation in such sentences follows from his syntactic analysis, but there is an alternative that is readily available on the lexical analysis. The ungrammaticality of (9.23) would also follow from the requirement that an overt argument linked to an argument position that is already partially speciﬁed (via Restrict) has to be informative. Since ‘what’ adds nothing to the semantics of ‘meateat’ that would further specify the properties of its object argument, it fails to meet that requirement. The lexical account accommodates these facts without syntactic movement. Baker’s ﬁnal argument is based on “condition C effects.” He notes that in sentences like those in (9.24), coreference between the matrix subject and the argument corresponding to an incorporated noun is at best marginal. (9.24)

tóka 2- ke- ksá- ht- a- ya’k- e’ a. (Sak) ra- tsháni- s Sak MsS- fear- hab maybe fut- 1sS- child- nom- ∅- hit- punc He (Sak)i is afraid that maybe I will slap the childj/??i b. Sak í- hr- ehr- e’ ks- kst2- hser- akéras Sak ∅- m.sg.subj- think- impf n.sg.subj- old.person- nom- stink Saki thinks the old personj/??i smells bad a- ke- kst2- hser- áhset- e’ c. Sak í- hr- ehr- e’ Sak ∅- m.sg.subj- think- impf opt- 1sg.subj- old.person- nomhide- punc Saki wants me to hide the old personj/??i

Baker interprets the verb forms in the embedded clauses in these examples as involving an incorporated object and intransitive agreement. He suggests that the difﬁculty of getting coreferential interpretations would follow if the position corresponding to the incorporated noun contained a trace, rather than pro, and that trace exhibited condition C effects (rather than the condition B effects we would expect of a pro). Another interpretation is quite possible, however. We might say that the agreement with the positions corresponding to incorporated nouns in (9.24) is actually inanimate, rather than missing (as is indeed overt in the case of the embedded intransitive in (9.24b)). In that case, the difﬁculty of getting a coreferential reading for these sentences would follow from an apparent gender conﬂict between the matrix subject (overtly animate and masculine in all cases) and the lower DP, apparently identiﬁed as inanimate. This interpretation is strengthened by Baker’s example given here as (9.25), where appropriate transitive animate agreement with the “incorporated” position appears, and where coreference is perfectly acceptable.

The Grammar of Incorporation

281

tóka 2- hi- ksá- ht- a- ya’k- e’ (9.25) (Sak) ra- tsháni- s Sak m.sg.subj- fear- hab maybe fut- 1sg.subj/m.sg.obj- childnom- ∅- hit- punc He (Sak)i is afraid that maybe I will slap the childi/j Given the uncertain status of condition C effects in the syntactic literature, it would not be wise to put too much weight on this argument in either direction, but it certainly appears that even assuming the disjoint reference principles Baker invokes, it does not necessarily follow that the position corresponding to an incorporated noun is occupied by a trace, rather than by pro. Indeed, Baker explicitly suggests that the relevant DP in (9.25) is pro rather than trace. In summary, the lexical account of Noun Incorporation seems entirely viable, even for the phenomena Baker treats as most centrally syntactic. Since the lexical analysis is clearly motivated in some cases, even within Mohawk, it is incumbent on the advocate of an alternative analysis based on syntactic Head Movement to show that this is the only one available under at least some conditions. There does not appear to be evidence of that sort, however, and so it follows that Mohawk and similar languages do not support the claim that Noun Incorporation ever involves syntactic movement.

9.4 Denominal Verb Formation in West Greenlandic The Mohawk constructions that form the basis of Baker’s analysis are of the type that all analysts agree are core instances of Noun Incorporation. According to Sadock (1980, 1986, 1991, 2003; see also Woodbury and Sadock 1986), the large class of sufﬁxes in “Eskimo” languages like West Greenlandic that derive verbs from noun stems as in (9.26) also involve composition of surface words in the syntax, a view similar to Baker’s and contrary to the Lexicalist Hypothesis. (9.26)

a. Sapanngamik kusanartumik pisivoq beautiful- nom-inst thing- get- indic-3sg bead- inst He bought a beautiful bead b. Kusanartumik sapangarsivoq beautiful- nom-inst bead- get- indic-3sg

These denominal verbs differ in some ways from basic Noun Incorporation constructions, as remarked above. As Sadock shows, though, they pose essentially the same issues for the morphology–syntax interface. That is, in these structures as well we appear to have a syntactic relation between part of a

282

Aspects of the Theory of Clitics

(complex, derived) word and a position in syntactic structure. Sadock (2003) says explicitly that “[s]ome afﬁxes in all classes also count as independent syntactic elements in syntactic representations. This is a highly characteristic feature of W[est] G[reenlandic] grammar that is responsible for the rest of its polysynthetic character. A single word can not only have the meaning of an entire phrase or sentence in more isolating languages, but also the syntax thereof.” Sadock’s usage of “polysynthetic” is much closer to the traditional understanding of this term than Baker’s, but like Baker’s, the relation between a syntactic noun head and the nominal component of a complex denominal verb, at least at ﬁrst glance, appears to be best expressed as displacement. Sadock’s (1991) own approach to this construction is actually somewhat different. Instead of moving a noun to adjoin it to a verbal element, he treats this construction (among others) as a mismatch between syntactic and morphological structures, by which the morphology is allowed to construct words out of pieces that are not co-constituents in the syntactic representation of a sentence. For present purposes, the differences between this view and the Head Movement analysis of Baker (1995) are not signiﬁcant. Both assert that the nominal element in the constructions at issue is to be identiﬁed with material constituting an argument of the verbal element, while the lexical analysis to be proposed here does not make such an identiﬁcation. A number of arguments are cited in Sadock (1986) in support of the notion that parts of these complex words participate in the external syntax of the structures they head. I will address those points in the present section, and suggest that a syntactic analysis of the sort assumed by Sadock is not in fact required. This conclusion is in accord with several other analyses in the literature (e.g., van Geenhoven 1998; Malouf 1999; and Bittner 2001). Sadock’s points are very signiﬁcant, and any account of Noun Incorporation (and indeed the general nature of the relation between syntax and morphology) must come to terms with them. I can not take up all of these matters in as much detail as they deserve, but I will try to provide enough discussion to substantiate my optimism with respect to the lexical account, as it extends to Greenlandic. The ﬁrst of Sadock’s arguments is based on the observation that incorporating verbs alone, as opposed to all others in the language, can have external possessors in the ergative case, as in examples (9.26a,b3 ). Importantly, 3

This sentence is impersonal, with an empty subject. The ergative DP kunggip thus must be interpreted as the possessor of the object ‘(many) daughters,’ and not as the subject of a transitive verb ‘have.’

The Grammar of Incorporation

283

a nominal phrase consisting only of a possessive (in the ergative case) modifying a presumably null head cannot in general appear outside of incorporation structures, as illustrated in (9.27c). (9.27)

a. Puissip neqitorpunga seal- erg meat- consume- indic.1sg I ate seal meat b. Kunngip panippassuaqarpoq king- erg daughter- many- have- indic.3sg There are many princesses c. *Kunngip takuvunga/takuara king- erg see- indic.1sg/1sg.3sg (I saw the king’s)

Elsewhere, nominals without an overt head noun do occur, consisting only of modiﬁers such as a quantiﬁer, and it is important to understand why stranded possessors do not appear to have similar privileges of occurrence. In fact, I propose that the objects in (9.26a,b) do consist of a null head together with a possessor, rather than DPs headed by a lexical noun which becomes part of a complex verb. What needs to be explained is why such DPs can only occur as arguments of noun-incorporating verbs. “Incorporation” only applies to objects4 in Greenlandic, and, the object associated with an “incorporated” noun stem is always and necessarily indefinite. Suppose we take this indeﬁniteness to be a property of the null head, consistent with the inherited semantics of unmodiﬁed basic nouns (which are, of course, the lexical source of the “incorporation” structures). Null-headed DPs do not appear in positions where they should exhibit agreement, because the null head is indeﬁnite, and Greenlandic verbs agree with their objects only when these are deﬁnite. This shows us why the transitive form verbal takuara is excluded in (9.27c). Suppose the verb is inﬂected intransitively, however, as in the variant of (9.27c) with takuvunga. The infelicity of this sentence can then follow from a requirement that the null head is only acceptable where some information is provided to specify some of its semantic properties. Some modiﬁers (e.g., quantiﬁers or adjectives) might be argued to have this effect, but not possessives. The noun base of an “incorporated” form like that in (9.26b) does supply 4

Verbs of being and becoming with a DP complement in the absolutive appear to constitute an exception to this generalization, but as I will suggest below, these should be regarded as “quasitransitive” and any incorporated argument as corresponding to a kind of object.

i

i

“Clitics” — 2005/7/26 — 17:39 — page 284 — #294

i

284

i

Aspects of the Theory of Clitics

such information (via Restrict), and so possessed null-headed nominals are possible as objects of such verbs. The distribution of null-headed nominals in Greenlandic remains to be explored in more detail, but a difference in this regard between possessed expressions and those with other modifiers does not seem implausible. The restriction of the possessed phrases to incorporation structures, where the incorporated stem provides some semantic specification of the referent of the null head, then follows from a more general informativeness requirement that cannot be satisfied by null heads alone. Sadock’s second point, which has become particularly well known in the literature, is that incorporating verbs alone may impose purely formal, semantically unmotivated restrictions on the plurality of an external DP. This is illustrated in example (9.28). (9.28) a. Ataatsinik qamuteqarpog one- inst.pl sled.pl- have- indic.3sg He has one sled b. *Ataatsimik qamuteqarpog one- inst.sg sled.pl- have- indic.3sg (He has one sled) The point here is that the noun qamut ‘sled’ in (9.28) is a pluralia tantem form in Greenlandic, and appears as grammatically plural even when its referent is semantically singular. Even though this noun does not appear in the overt argument expression ataatsinik, this is inflected as a plural in apparent disregard of the semantically singular nature of the sled. This fact would of course follow if the grammatically plural noun qamut ‘sled’ were present in the underlying form of the DP, and displaced to the “incorporated” position after triggering Agreement. I suggest that this behavior follows relatively naturally from the account proposed here, though, without the need to invoke such displacement. The verb qamuteqarpoq ‘he has sled(s)’ is built on the noun qamut, and thus presumably inherits the lexical idiosyncrasies of that word, including its otherwise unmotivated number. (9.29)

qamut

— sled +Plural

–qar + have — , — ⇒ Agent Theme

qamuteqar– ⎡

⎤ — , have Agent — ⎢ ⎥ ⎢ Theme ⎥ ⎣ ⎦ sled +Plural

i

i i

i

The Grammar of Incorporation

285

In this structure, the derived complex verb inherits the grammatical idiosyncrasy of its base noun in the form of a grammatical restriction on the object DP, without assuming that the base noun itself moves in the syntax. To satisfy this restriction, the presumed null head in the external DP must be assigned the formal feature [+Plural], and the modiﬁer agrees with that. Sadock’s third point is that incorporating verbs alone may have an additional absolutive nominal associated with them, which follows the verb and is not understood as a modiﬁer of the subject. Verbs without incorporation, however, cannot be accompanied by such a phrase modifying the absolutive. This is illustrated by the contrast in (9.30). (9.30)

a. Joorut palasinngorpoq tusaamasoq Jørgen priest- become- indic.3sg famous- nom Jørgen became a famous priest b. ?? ??Joorut toquvoq tusaamasoq Jørgen die- indic.3sg famous- nom The famous Jørgen died

I suggest that copulative structures such as those based on ‘be,’ ‘become,’ etc., are syntactically “quasi-transitive,” and thus admit of another argument position. Tusaamasoq ‘famous’ in example (9.30a) thus occupies the position of the second argument, where it further speciﬁes the argument ‘priest’ of the basic verb. The infelicity of (9.30b) follows from a more general problem with having parts of the same DP separated by the verb: in this example there is only a single DP, whose parts are non-contiguous, while in the superﬁcially similar (9.30a) the two parts correspond to two separate phrasal constituents. Predicate nominatives are of course non-referential, as argued long ago by Kuno (1970), and in the context of Greenlandic caseassignment patterns, this accounts for the fact that both DPs in (9.30a) are nominative. Sadock’s fourth and ﬁfth points do not bear directly on the syntactic nature of incorporation. He notes that Greenlandic displays sporadic instances of “the polysynthetic equivalent of Gapping” in which the ﬁrst of two conjuncts consists of an incorporating verb form and the second simply of an absolutive DP with no sufﬁx (see sentence (5.3) in Chapter 5 above for an example). While it might of course be possible to derive the apparent Gapping structures by syntactic deletion, it seems likely that ellipsis of this sort is a matter of semantic interpretation rather than of syntactic deletion or reconstruction. Greenlandic (though not the otherwise similar Yup’ik) also disallows an external occurence of the same nominal that has been “incorporated” with a denominal verbal afﬁx. But this simply shows that Greenlandic does not regard

286

Aspects of the Theory of Clitics

identical nominals as mutually “informative” in the sense suggested earlier, while Yup’ik does. As already noted, Sadock’s own view of the syntax of incorporation constructions does not involve movement of the nominal head per se, but rather a mechanism that allows parts of complex words to be interpreted as syntactically autonomous elements. On this picture, sentence (9.28a) is not formed by moving qamut ‘sled’ out of the DP and into the verb, but rather by allowing the verb-internal noun stem to be treated as the head of the nominal in situ. Malouf (1999: 61) has observed that this presents a particular difﬁculty for the analysis: “Within the noun phrase, the possessor must precede the head noun and any modiﬁers must follow it (Fortescue 1984: 117). Under Sadock’s analysis, examples like [(9.31)], where a nominal modiﬁer precedes an incorporated nominal, violate the linear precedence constraints for noun phrases.” (9.31) kissartu-mik kavvi-sur-put hot-inst coﬀee-drink-3pl.indic They drank hot coﬀee (Fortescue 1984: 83) Sadock must thus allow the word-order constraints within the DP to be relaxed exactly when the head noun is part of an incorporation structure. If we treat the nominal in (9.31) as having a null head, distinct from the stem that appears as part of the (lexically formed) denominal verb, no such proviso is necessary. Of course, an analysis such as Baker’s, on which the nominal head moves into the verb in the syntax, also avoids this diﬃculty, though we have already seen other reasons to doubt the motivation for either of these accounts as opposed to the lexical one.

9.5 Conclusion So where does this survey of incorporation phenomena leave us with respect to a choice between the two primary views of Noun Incorporation? It appears that even Baker agrees that much Noun Incorporation is in fact lexical, not syntactic, even in the language for which he feels the strongest syntactic case can be made (Mohawk). And in fact, the limited sets of data for which the syntactic account is still said to be necessary can also be accommodated within the lexical account, without invoking extraordinary mechanisms. That means that a purely lexical account of Noun Incorporation, without syntactic movement (or its equivalent in Sadock’s system), is almost certainly possible. But that, in turn, means that the best putative support for an operation of syntactic Head

The Grammar of Incorporation

287

Movement may be non-existent---a conclusion with extensive consequences for many areas of contemporary syntax. This is especially true for the “split-Infl” analysis of clause structure, with its proliferation of functional heads. Since the framework oﬀered in the present book also accommodates the grammar of special clitics without recourse to extensive layers of functional structure, these conclusions dovetail to suggest that clause structure in natural language is formally simpler (and thereby closer to surface form) than it is often suggested to be.

This page intentionally left blank

References Abeillé, Anne and Godard, Danièle (2003), “Les prédicats complexes dans les langues romanes”, in Godard (2003), 125–84. Adams, Marianne P. (1987), “Old French, Null Subjects, and Verb Second Phenomena”. Ph.D. thesis, UCLA. Aissen, Judith (1992), “Topic and focus in Mayan”, Language 68: 43–80. Aitchison, Therese M. (2001), “An HPSG analysis of the Tongan deﬁnitive accent”. Master’s thesis, University of Canterbury, Christchurch, New Zealand. Allen, Cynthia (1997), “Investigating the origins of the ‘group genitive’ in English”, Transactions of the Philological Society 95: 111–31. Alpher, Barry (1991), Yir-Yoront lexicon: Sketch and dictionary of an Australian language (Berlin: Mouton de Gruyter). Anderson, Stephen R. (1981), “Topicalization in Breton”, Proceedings of the Annual Meeting of the Berkeley Linguistics Society 7: 27–39. —— (1982), “Where’s morphology?”, Linguistic Inquiry 13: 571–612. —— (1984), “Kwak’wala syntax and the Government-Binding theory”, in E.-D. Cook and D. B. Gerdts (eds.), The syntax of native American languages, Syntax and Semantics 16 (New York: Academic Press), 21–75. —— (1985a), “Inﬂectional morphology”, in T. Shopen (ed.), Language typology and syntactic description, vol. III (Cambridge: Cambridge University Press), 150–201. —— (1985b), Phonology in the twentieth century (Chicago: University of Chicago Press). —— (1991), “Syntactically arbitrary inﬂectional morphology”, Yearbook of Morphology 4: 5–19. —— (1992), A-Morphous morphology (Cambridge: Cambridge University Press). —— (1993), “Wackernagel’s revenge: Clitics, morphology and the syntax of second position”, Language 69: 68–98. —— (2000a), “Lexicalism, incorporated (or incorporation, lexicalized)”, Proceedings of the Chicago Linguistic Society 36: 13–34. —— (2000b), “Some lexicalist remarks on incorporation phenomena”, in B. Stiebels and D. Wunderlich (eds.), Lexicon in focus, Studia Grammatica 45 (Berlin: Akademie Verlag GmbH), 123–42. —— (2000c), “Towards an optimal account of second position phenomena”, in Dekkers, van der Leeuw, and van de Weijer (2000), 302–33. —— (2001), “On some issues of morphological exponence”, Yearbook of Morphology 2000: 1–18. —— (2004), “Subject clitics and verb-second in Surmiran Rumantsch”, MIT Working Papers in Linguistics 47: 1–22. —— and Lightfoot, David W. (2002), The language organ: Linguistics as cognitive physiology (Cambridge: Cambridge University Press).

290

References

Anderson, Stephen R. and Chung, Sandra (1977), “On grammatical relations and clause structure in verb-initial languages”, in P. Cole and J. Sadock (eds.), Grammatical relations, Syntax and Semantics 8 (New York: Academic Press), 1–25. Anderson, Victoria and Otsuka, Yuko (2002), “Phonetic correlates of length, stress and deﬁnite accent in Tongan”. Paper presented at 9th meeting of the Austronesian Formal Linguistics Association, Cornell University, Ithaca, NY. Bach, Emmon (1998), “Semantic relations in and outside of complex words”. Paper presented at Lexicon in Focus Conference, Wuppertal, 19 August, 1998. —— and Howe, Darin (2002), “On the development of North Wakashan”. Unpub. manuscript, University of Massachusetts and University of Calgary. Baker, Mark (1985), “The mirror principle and morphosyntactic explanation”, Linguistic Inquiry 16: 373–415. —— (1987), Incorporation: A theory of grammatical function changing (Chicago: University of Chicago Press). —— (1995), The polysynthesis parameter (New York: Oxford University Press). Barbosa, Pilar (1996), “Clitic placement in European Portuguese and the position of subjects”, in A. L. Halpern and A. Zwicky (eds.), Approaching second: Second position clitics and related phenomena (Center for the Study of Language and Information, Stanford, CA: CSLI Publications), 1–40. —— Fox, Danny, Hagstrom, Paul, McGinnis, Martha, and Pesetsky, David, eds. (1998), Is the best good enough? (Cambridge: MIT Press). Battye, Adrian and Roberts, Ian, eds. (1995), Clause structure and language change (Oxford: Oxford University Press). Belletti, Adriana, ed. (1993), Syntactic theory and the dialects of Italy (Torino: Rosenberg & Seiler). Benincá, Paola (1983), “Il clitico ‘a’ nel dialetto padovano”, Scritti linguistici in onore di Giovan Battista Pellegrini (Pisa: Pacini), 25–32. —— (1985), “L’interferenza sintattica: di un aspetto della sintassi ladina considerato di origine tedesca”, Quaderni Patavini di Linguistica 5: 3–17. —— (1994), Variazione sintattica: studi di dialettologia romanza (Bologna: Il Mulino). —— (1995), “Complement clitics in medieval Romance: The Tobler-Musaﬁa law”, in Battye and Roberts (1995), 325–46. Berendsen, Egon (1986), The phonology of cliticization (Dordrecht: Foris). Bergsland, Knut (1997), Aleut grammar (Fairbanks: Alaska Native Language Center). Beukema, Frits and den Dikken, Marcel, eds. (2000), Clitic phenomena in European languages (Amsterdam: Benjamins). Bhatt, Rakesh Mohan (1999), Verb movement and the syntax of Kashmiri (Dordrecht: Kluwer). Bittner, Maria (2001), “Surface composition as bridging”, Journal of Semantics 18: 127–77. Bloomﬁeld, Leonard (1917), Tagalog texts with grammatical analysis, University of Illinois Studies in Language and Literature, vol. III, nos. 2–4 (Urbana: University of Illinois).

References

291

—— (1946), “Algonquian”, in Cornelius Osgood (ed.), Linguistic structures of Native America, Viking Fund Publications in Anthropology, 6 (New York), 85–129. Boas, Franz (1911), “Kwakiutl”, in F. Boas (ed.), Handbook of American Indian languages, vol. 1 (Washington D.C.: Smithsonian Institution, Bureau of American Ethnology, Bulletin 40.1), 423–557. —— (1928), Bella Bella texts, Columbia University Contributions to Anthroplogy, 5 (New York: Columbia University Press). Reprinted 1969 (New York: AMS Press Inc.). —— (1947), “Kwakiutl grammar, with a glossary of the sufﬁxes”, Transactions of the American Philosophical Society 37(3): 201–377. Bobaljik, Jonathan and Wurmbrand, Susan (2002), “Long distance agreement, restructuring and anti-recontruction”, Proceedings of the North East Linguistics Society 33: 67–86. Booij, Geert (1988), Review article on Nespor and Vogel 1986. Journal of Linguistics 24: 515–25. —— (1995), The phonology of Dutch (Oxford: Oxford University Press). —— (1996), “Cliticization as prosodic integration: The case of Dutch”, The Linguistic Review 13: 219–42. —— (1997), “Non-derivational phonology meets lexical phonology”, in I. Roca (ed.), Derivations and constraints in phonology (Oxford: Clarendon Press), 261–88. —— and Rubach, Jerzy (1987), “Postcyclic versus postlexical rules in lexical phonology”, Linguistic Inquiry 18: 1–44. Borsley, Robert D. (1990), “A GPSG approach to Breton word order”, in Hendrick (1990b), 81–95. —— and Stephens, Janig (1989), “Agreement and the position of subject in Breton”, Natural Language and Linguistic Theory 7: 407–27. —— Rivero, Maria-Luisa, and Stephens, Janig (1996), “Long head movement in Breton”, in R. D. Borsely and I. Roberts (eds.), The syntax of the Celtic languages (Cambridge: Cambridge University Press), 1–52. Boškovi´c, Željko (1995), “Participle movement and second position cliticization in Serbo-Croatian”, Lingua 96: 245–66. —— (2000), “Second position cliticisation: Syntax and/or phonology?”, in Beukema and den Dikken (2000), 71–119. —— (2001), On the nature of the syntax-phonology interface (Oxford: Elsevier Science Ltd.). Boysson-Bardies, Bénédicte de (1999), How language comes to children (Cambridge, MA: MIT Press). Bresnan, Joan (1978), Contraction and the transformational cycle (Bloomington: Indiana University Linguistics Club). —— (2001), Lexical-functional syntax (Oxford: Blackwell). Browne, Wayles (1974), “On the problem of enclitic placement in Serbo-Croatian”, in R. D. Brecht and C. V. Chvany (eds.), Slavic transformational syntax, Michigan Slavic Materials, vol. 10 (Ann Arbor), 36–52.

292

References

Browne, Wayles (1975), “Serbo-Croatian enclitics for English-speaking learners”, Contrastive Analysis of English and Serbo-Croatian 1: 105–34. Burzio, Luigi (1986), Italian syntax: A Government and Binding approach (Dordrecht: Reidel). —— (1992), “On the morphology of reﬂexives and impersonals”, in C. Lauefer and T. Morgan (eds.), Theoretical analyses in Romance Linguistics, LSRL xix (Amsterdam: Benjamins), 399–414. Cardinaletti, Anna (1991), “On pronoun movement: The Italian dative loro”, Probus 3: 127–53. Carstairs, Andrew (1987), “Diachronic evidence and the afﬁx-clitic distinction”, in A. G. Ramat, O. Carruba, and G. Bernini (eds.), Papers from the 7th International Conference on Historical Linguistics (Amsterdam: Benjamins), 151–62. Carstairs-McCarthy, Andrew (1995), “Review of Halpern 1995, Language Quarterly 33: 254–8. Chomsky, Noam (1995), The minimalist program (Cambridge, MA: MIT Press). —— (2001), “Derivation by phase”, in M. Kensotwicz (ed.), Ken Hale: a life in language (Cambridge, MA: MIT Press), 1–52. —— and Halle, Morris (1968), The sound pattern of English (New York: Harper & Row). —— Halle, Morris and Lukoff, Fred (1956), “On accent and juncture in English”, in M. Halle, H. Lunt, and H. MacLean (eds.), For Roman Jakobson (The Hague: Mouton & Co.), 65–80. Chung, Sandra L. (2003), “The syntax and prosody of weak pronouns in Chamorro”, Linguistic Inquiry 34: 547–99. —— and Ladusaw, William A. (2004), Restriction and saturation (Cambridge, MA: MIT Press). Churchward, C. Maxwell (1940), Rotuman grammar and dictionary (Sydney: Australasian Medical Publishing Company). —— (1953), Tongan grammar (London: Oxford University Press). Clark, Ross (1974), “On the origin of the Tongan deﬁnitive accent”, Journal of the Polynesian Society 83: 103–8. Condax, Iovanna D. (1989), “Tongan deﬁnitive accent”, Journal of the Polynesian Society 98: 425–50. Condoravdi, Cleo and Kiparsky, Paul (1998), “Optimal order and scope”. Paper presented at Lexicon in Focus conference, Heinrich-Heine University, Düsseldorf, 19 August, 1998. Corbett, Greville (2003), “Agreement: the range of the phenomenon and the principles of the SMG agreement database”, in D. Brown, G. Corbett, and C. Tiberius (eds.), Agreement: A typological perspective, special number of the Transactions of the Philological Society, vol. 101 (2) (Oxford: Blackwell), 155–202. Crysmann, Berthold (2000), “Clitics and coordination in linear structure”, in Gerlach and Grijzenhout (2000), 121–59. Cysouw, Michael (forthcoming), “Morphology in the wrong place: A survey of preposed enclitics”, in W. U. Dressler (ed.), Morphology and its demarcations (Amsterdam: Benjamins).

References

293

Dekkers, Joost, van der Leeuw, Frank, and van de Weijer, Jeroen, eds. (2000), Optimality Theory: Phonology, syntax and acquisition (Oxford: Oxford University Press). Di Sciullo, Anna-Maria and Williams, Edwin (1987), On the deﬁnition of word (Cambridge, MA: MIT Press). Dixon, Robert M. W. (1980), The languages of Australia (Cambridge: Cambridge University Press). Donaldson, Tamsin (1980), Ngiyambaa: The language of the Wangaaybuwan (Cambridge: Cambridge University Press). Eythórsson, Thórhallur (1996), “Functional categories, cliticization, and verb movement in the early Germanic languages”, in H. Thráinsson, S. D. Epstein and S. Peter (eds.), Studies in comparative Germanic syntax, vol. II (Dordrecht: Kluwer), 109–39. Flagg, Elissa (2003), “Against heterogeneous origins for n’t and not”, Snippets 7: 5–6. Online publication available at . Fontana, Josep M. (1993), “Phrase structure and the history of clitics in the history of Spanish”. Ph.D. thesis, University of Pennsylvania. Fortescue, Michael (1984), West Greenlandic (London: Croom Helm). Foulet, Lucien (1928), Petite syntaxe de l’ancien français, 3rd edn. (Paris: H. Campion). Franks, Steven (2000), “Clitics at the interface”, in Beukema and den Dikken (2000), 1–46. —— and Progovac, Liljana (1999), “Clitic second as verb second”. Paper presented at winter meeting of the Linguistic Society of America, Los Angeles. —— and King, Tracy Holloway (2000), A handbook of slavic clitics (New York: Oxford University Press). Fukui, Naoki and Speas, Margaret (1986), “Speciﬁers and projection”, MIT Working Papers in Linguistics 8: 128–72. Fulmer, Sandra Lee (1990), “Dual-position afﬁxes in Afar: An argument for phonologically driven morphology”, Proceedings of the West Coast Conference on Formal Linguistics 9: 189–203. Galves, Charlotte and Sandalo, Filomena (2004), “Clitic placement in modern and classical European Portuguese”, MIT Working Papers in Linguistics 47: 115–28. Collected Papers on Romance Syntax. Ganzoni, Gian Paul (1977), Grammatica Ladina: Grammatica sistematica dal rumantsch d’Engiadina Ota (Samedan: Lia Romauntscha). —— (1983), Grammatica Ladina: Grammatica sistematica dal rumantsch d’Engiadina Bassa (Samedan: Lia Romantscha). Gazdar, Gerald, Klein, Ewan, Pullum, Geoffrey, and Sag, Ivan (1985), Generalized phrase structure grammar (Cambridge, MA: Harvard University Press). van Geenhoven, Veerle (1998), Semantic incorporation and indeﬁnite descriptions: Semantic and syntactic aspects of noun incorporation in West Greenlandic (Stanford, CA: CSLI Publications). Gerdts, Donna B. (1997), “Incorporation”, in A. Spencer and A. Zwicky (eds.), Handbook of morphology (London: Basil Blackwell), 84–100. Gerlach, Birgit and Grijzenhout, Janet, eds. (2000), Clitics in phonology, morphology and syntax (Amsterdam: Benjamins).

294

References

Godard, Danièle, ed. (2003), Les langues romanes, problèmes de la phrase simple (Paris: CNRS éditions). Goldberg, Adele (1995), Constructions: A construction grammar approach to argument structure (Chicago: University of Chicago Press). Goldsmith, John (1979), Autosegmental phonology (New York: Garland Press). Grimshaw, Jane (2000), “Extended projection and locality”, in P. Coopmans, M. Everaert, and J. Grimshaw. (eds.), Lexical speciﬁcation and insertion (Amsterdam: Benjamins), 115–33. Grisch, Mena (1939), Die Mundart von Surmeir, Romanica Helvetica, vol. 12 (Paris: E. Droz). Haegeman, Liliane (1992), Theory and description in generative syntax: A case study in West Flemish, Cambridge Studies in Linguistics (Supplementary Volume) (Cambridge: Cambridge University Press). Haiman, John and Benincà, Paola (1992), The Rhaeto-Romance languages (London: Routledge). Hale, Kenneth (1973), “Person marking in Walbiri”, in S. R. Anderson and P. Kiparsky (eds.), A Festschrift for Morris Halle (New York: Holt, Rinehart & Winston), 308–44. Hale, Mark and Kissock, Madelyn (1998), “The phonology-syntax interface in Rotuman”, in M. Pearson (ed.), Recent papers in Austronesian linguistics (Los Angeles: UCLA), 115–28. Hall, Alfred J. (1889), “A grammar of the Kwagiutl language”, Transactions of the Royal Society of Canada 6: 59–105. Section II, 1888. Halle, Morris and Marantz, Alec (1993), “Distributed morphology and the pieces of inﬂection”, in K. Hale and S. J. Keyser (eds.), The view from Building 20 (Cambridge, MA: MIT Press), 111–76. Halpern, Aaron (1992a), “On the nature of the Bulgarian deﬁnite article”. Unpub. manuscript, Stanford University, CA. —— (1992b), “Topics in the Placement and Morphology of Clitics”. Ph.D. thesis, Stanford University, CA. —— (1995), On the placement and morphology of clitics (Stanford: CSLI Publications). Hannahs, S. J. and Tallerman, Maggie (2000), “A constraint-based approach to verbsecond syntax in Brythonic Celtic”. Unpub. manuscript, University of Durham. Harris, Alice C. (2000), “Where in the word is the Udi clitic?”, Language 76: 593–616. —— (2002), Endoclitics and the origins of Udi morphosyntax (Oxford: Oxford University Press). Hayes, Bruce (1982), “Extrametricality and English stress”, Linguistic Inquiry 13: 227–76. —— (1989), “The prosodic hierarchy in meter”, in P. Kiparsky and G. Youmans (eds.), Rhythm and meter, Phonetics and Phonology 1 (New York: Academic Press), 201–60. Hendrick, Randall (1988), Anaphora in Celtic and universal grammar (Dordrecht: Reidel). —— (1990a), “Breton pronominals, binding, and barriers”, in Hendrick (1990b), 121–65. Hendrick, Randall, ed. (1990b), The syntax of modern Celtic languages, Syntax & Semantics 23 (New York: Academic Press).

References

295

Hendriksen, Hans (1986), Himachali studies III: Grammar, Historisk-ﬁlosoﬁske Meddelelser 48 (Copenhagen: Det Kongelige Danske Videnskabernes Selskab). —— (1990), “Sentence position of the verb in Himachali”, Acta Linguistica Hafniensia 22: 159–71. Hock, Hans Heinrich (1982), “Aux-cliticization as a motivation for word order change”, Studies in the Linguistic Sciences 12: 91–101. —— (1991), Principles of historical linguistics, 2nd edn. (Berlin: Mouton de Gruyter). Hockett, Charles F. (1948), “Potawatomi (I, II, III, IV)”, International Journal of American Linguistics 14: 1–10, 63–73, 139–49,213–25. Hoekstra, Eric and Smits, Caroline (1998), “Everything you always wanted to know about complementizer agreement”, Proceedings of the Western Conference on Linguistics, 189–200. Holmer, Arthur J. (1996), A parametric grammar of Seediq (Lund: Lund University Press). Inkelas, Sharon (1989), “Prosodic Constituency in the Lexicon”. Doctoral dissertation, Stanford University, CA. Jacobs, Haike (1993), “The phonology of enclisis and proclisis in Gallo-Romance and old French”, in W. J. Ashby, M. Mithun, G. Perissinotto and E. Raposo (eds.), Linguistic perspectives on the Romance languages, Current Issues in Linguistic Theory 103 (Amsterdam: John Benjamins), 149–63. Jaeggli, Osvaldo (1986), “Three issues in the theory of clitics: Case, doubled NPs, and extraction”, in H. Borer (ed.), The syntax of pronominal clitics, Syntax and Semantics 19 (Orlando: Academic Press), 15–42. Jelinek, Eloise (1984), “Empty categories, case and conﬁgurationality”, Natural Language and Linguistic Theory 3: 39–76. Jónsson, Jóhannes Gísli (1991), “Stylistic fronting in Icelandic”, Working Papers in Scandinavian Syntax 48: 1–43. Kager, René (1999), Optimality Theory (Cambridge: Cambridge University Press). Kahn, Daniel (1976), “Syllable-based generalizations in English phonology”. Ph.D. thesis, MIT, distributed by Indiana University Linguistics Club. Kaiser, Lizanne, ed. (1997), Studies in the morphosyntax of clitics, Y.A.L.E. working papers, vol. 1 (Dept. of Linguistics, Yale University). Kaiser, Lizanne (1998a), “The interaction of noun incorporation and applicative formation in Ainu”, Yearbook of Morphology 1997: 157–78. —— (1998b), “The Morphosyntax of Clausal Nominalization Constructions”. Ph.D. thesis, Yale University. Kaisse, Ellen M. (1981), “Separating phonology from syntax: A reanalysis of Pashto cliticization”, Journal of Linguistics 17: 197–208. —— (1985), Connected speech (New York: Academic Press). —— and Shaw, Patricia (1985), “On the theory of lexical phonology”, Phonology 2: 1–30. Kanerva, Jonni (1987), “Morphological integrity and syntax: The evidence from Finnish possessives”, Language 63: 498–521. Kayne, Richard (1975), French syntax: The transformational cycle (Cambridge, MA: MIT Press).

296

References

Kayne, Richard (1991), “Romance clitics, verb movement, and PRO”, Linguistic Inquiry 22: 647–86. Kiparsky, Paul (1995), “Indo-European origins of Germanic syntax”, in Battye and Roberts (1995), 140–69. —— (2000), “Opacity and cyclicity”, The Linguistic Review 17: 351–67. —— (2003), “Finnish noun inﬂection”, in D. Nelson and S. Manninen (eds.), Generative approaches to Finnic and Saami linguistics (Stanford, CA: CSLI Publications), 109–61. —— (forthcoming), Paradigmatic effects (Stanford, CA: CSLI Publications). Klavans, Judith L. (1982), Some problems in a theory of clitics (Bloomington: Indiana University Linguistics Club). —— (1985), “The independence of syntax and phonology in cliticization”, Language 61: 95–120. Klokeid, Terry J. (1976), “Encliticization in Nitinaht”, Working papers for the 11th International Conference on Salishan Languages (Seattle: Department of Anthropology, University of Washington). Koopman, Hilda and Sportiche, Dominique (1991), “The position of subjects”, Lingua 85: 555–88. Kroeber, Alfred (1909), “Noun incorporation in American languages”, Proceedings of the 16th International Congress of Americanists, 569–760. Kroeger, Paul (1993), Phrase structure and grammatical relations in Tagalog (Stanford, CA: CSLI Publications). Kruisinga, E. (1932), A handbook of present-day English. Part II: English accidence and syntax, vol. 2, 5th edn. (Groningen: P. Noordhoff). Kuno, Susumu (1970), “Some properties of non-referential noun phrases”, in R. Jakobson and S. Kawamoto (eds.), Studies in general and oriental linguistics presented to Shiro Hattori on the occasion of his sixtieth birthday (Tokyo: TEC Company Ltd.), 348–73. Kurisu, Kazutaka (2001), “The Phonology of Morpheme Realization”. Ph.D. thesis, University of California, Santa Cruz. —— (2002), “Rotuman incomplete phase as morphologically driven word optimization”. Paper presented at the Linguistics and Phonetics 2002 conference, Meikai University, Urayasu, Japan, September 2–6, 2002. Lahiri, Aditi, Jongman, Allard, and Sereno, Joan (1990), “The pronominal clitic [d@r] in Dutch: A theoretical and experimental approach”, Yearbook of Morphology 1990: 115–28. Laughren, Mary (2002), “Syntactic constraints in a ‘free word order’ language”, in M. Amberber and P. Collins (eds.), Language universals and variation (Westport, CT: Praeger Publishers), 83–130. Lees, Robert B. (1960), The grammar of English nominalizations (The Hague: Mouton & Co.) Legate, Julie Anne (forthcoming), “Second position clitics: The perspective from Warlpiri”. Manuscript, University of Delaware.

References

297

Legendre, Géraldine (1998), “Second position clitics in a V2 language: Conﬂict resolution in Macedonian”, Proceedings of the 1997 ESCOL Meeting (Cornell University: CLC Publications), 139–49. —— (1999), “Morphological and prosodic alignment at work: The case of South-Slavic clitics”, Proceedings of the West Coast Conference on Formal Linguistics XVII, 436–50. —— (2000a), “Morphological and prosodic alignment of Bulgarian clitics”, in Dekkers et al. (2000), 423–62. —— (2000b), “Optimal Romanian clitics: A cross-linguistic perspective”, in V. Motapanyane (ed.), Comparative studies in Romanian syntax, North-Holland Linguistic Series 58 (Oxford: Elsevier), 227–64. —— (2000c), “Positioning Romanian clitics at PF: An Optimality-Theoretic analysis”, in Gerlach and Grijzenhout (2000), 219–54. —— (2001), “Masked V2 effects and the linearization of functional features”, in Legendre et al. (2001), 241–77. Legendre, Géraldine, Sten Vikner, and Jane Grimshaw, eds. (2001), OT syntax (Cambridge, MA: MIT Press). Liberman, Mark Y. (1975), “The Intonational System of English”. Ph.D. thesis, MIT. Published 1979 by Garland Press. —— and Alan S. Prince (1977), “On stress and linguistic rhythm”, Linguistic Inquiry 8: 249–336. Lincoln, Neville J., Rath, John C., and Windsor, Evelyn (1986), “The story of Baxwbakwalanusiwa, as told by Kitlope elder Gordon Robertson”, Working papers for the 21st International Conference on Salishan Languages (Seattle: University of Washington). Linder, Karl Peter (1987), Grammatische Untersuchungen zur Charakteristik des Rätoromanschen in Graubünden (Tübingen: Gunter Narr). Liver, Ricarda (1999), Rätoromanisch: Eine Einführung in das Bündnerromanische (Tübingen: Gunter Narr). Lynge, Hans (1978), Nuuk: Nuumme pisartut erqaamasat (Nuuk: Kalaallit-nunaanni naqiterisitsisarﬁk). Macaulay, Monica (1987), “Cliticization and the morphosyntax of Mixtec”, International Journal of American Linguistics 53: 119–35. Maling, Joan (1980), “Inversion in embedded clauses in modern Icelandic”, Íslenskt Mál 2: 175–93. Malouf, Robert (1999), “West Greenlandic noun incorporation in a monohierarchical theory of grammar”, in G. Webelhuth, A. Kathol, and J.-P. Koenig (eds.), Lexical and constructional aspects of linguistic explanation (Stanford, CA: CSLI Publications), 47– 62. Mardirussian, Galust (1975), “Noun incorporation in universal grammar”, Proceedings of the Chicago Linguistic Society 11: 383–89. McCarthy, John J. (1991), “Synchronic rule inversion”, Proceedings of the Annual Meeting of the Berkeley Linguistics Society 17: 197–207. —— (1995), “Extensions of faithfulness: Rotuman revisited”. Unpub. manuscript, University of Massachusetts; Rutgers Optimality Archive paper ROA-110.

298

References

McCarthy, John J. (2000), “The prosody of phase in Rotuman”, Natural Language and Linguistic Theory 18: 147–97. Merlan, Francesca (1976), “Noun incorporation and discourse reference in modern Nahuatl”, International Journal of American Linguistics 42: 177–91. Merriﬁeld, William R., Naish, Constance M., Resch, Calvin R., and Story, Gillian (1965), Laboratory manual for morphology and syntax (Santa Ana, CA: Summer Institute of Linguistics). Miller, Philip H. and Monachesi, Paola (2003), “Les pronoms clitiques dans les langues romanes”, in Godard (2003), 67–123. Mithun, Marianne (1984), “The evolution of noun incorporation”, Language 60: 847– 94. —— (1986), “On the nature of noun incorporation”, Language 62: 32–7. Monachesi, Paola (1995), “A Grammar of Italian Clitics”. Ph.D. thesis, Tilburg University. Moravcsik, Edith (1977), On rules of inﬁxing (Bloomington: Indiana University Linguistics Club). Mussaﬁa, A. (1898), “Enclisi o proclisi del pronome personale atono qual oggetto”, Romania 27: 145–6. Nespor, Marina (1994), “The phonology of clitic groups”, in L. Hellan and H. van Riemsdijk (eds.), Clitic doubling and clitic groups, vol. 5, ESF-Eurotyp Working Papers, 67–90. —— and Vogel, Irene (1986), Prosodic phonology (Dordrecht: Foris). Nevis, Joel A. (1986), Finnish particle clitics and general clitic theory. Working Papers in Linguistics 33 (Columbus, Ohio: Dept. of Linguistics, The Ohio State University). —— (1990), “Sentential clitics in Finno-Ugric”, Folia Linguistica 24: 349–71. —— Wanner, Dieter, and Zwicky, Arnold M. (1994), Clitics: A comprehensive bibliography 1892–1991 (Amsterdam: Benjamins). —— and Joseph, Brian (1992), “Wackernagel afﬁxes: Evidence from Balto-Slavic”, Yearbook of Morphology 1992: 93–111. Nida, Eugene A. (1949), Morphology: The descriptive analysis of words, 2nd edn. (Ann Arbor: University of Michigan Press). Noyer, Rolf (1994), “Mobile afﬁxes in Huave: Optimality and morphological wellformedness”, Proceedings of the West Coast Conference on Formal Linguistics 12: 67–82. Ortiz de Urbina, Jon (1994), “Verb-initial patters in Basque and Breton”, Lingua 94: 125–53. Osborne, C. R. (1974), The Tiwi language (Canberra: Australian Institute of Aboriginal Studies). Parker, Steve (1999), “On the behavior of deﬁnite articles in Chamicuro”, Language 75: 552–62. Payne, Doris L. and Barshi, Immanuel, eds. (1999), External possession (Amsterdam: Benjamins).

References

299

Payne, Doris L. and Payne, Thomas E. (1990), “Yagua”, in D. C. Derbyshire and G. K. Pullum (eds.), Handbook of Amazonian languages, vol. 2 (Berlin: Mouton de Gruyter), 249–474. Peperkamp, Sharon (1997), Prosodic words (Amsterdam: HIL). Perlmutter, David M. (1971), Deep and surface structure constraints in syntax (New York: Holt, Rinehart & Winston). Poletto, Cecilia (1993), “Subject clitic-verb inversion in north eastern Italian dialects”, in Belletti (1993), 204–51. —— (2000), The higher functional ﬁeld: Evidence from northern Italian dialects (New York: Oxford University Press). Polinsky, Maria S. (1990), “Subject incorporation: Evidence from Chukchee”, in K. Dziwirek, P. Farrell, and E. Mejfas-Bikandi (eds.), Grammatical relations: A crosstheoretical perspective (Stanford, CA: Center for the Study of Language and Information), 349–64. —— and Potsdam, Eric (2001), “Long distance agreement and topic in Tsez”, Natural Language and Linguistic Theory 19: 583–646. Poser, William, (1985), “Cliticization to NP and lexical phonology”, Proceedings of the West Coast Conference on Formal Linguistics 4: 262–72. Postal, Paul (1966), “On so-called pronouns in English”, in F. Dineen (ed.), Report of the 17th Annual Round Table Meeting on Languages and Linguistics (Washington, D.C.: Georgetown University Press), 201–24. —— (1979), Some syntactic rules of Mohawk (New York: Garland). Prince, Alan and Smolensky, Paul (1993), Optimality theory: Constraint interaction in generative grammar. Unpub. manuscript, Rutgers University and University of Colorado. Pullum, Geoffrey K. and Zwicky, Arnold M. (1986), “Phonological resolution of syntactic feature conﬂict”, Language 62: 751–73. —— (1988), “The syntax-phonology interface”, in F. J. Newmeyer (ed.), Linguistics: The Cambridge survey, vol. I (Cambridge: Cambridge University Press), 255–80. —— (1998), “Licensing of prosodic features by syntactic rules: The key to auxiliary reduction”. Paper presented at January Annual Meeting of the Linguistic Society of America. Rath, John C. (1981), A practical Heiltsuk-English dictionary with a grammatical introduction, National Museum of Man Mercury Series, 75 (2 volumes) (Ottawa: National Museums of Canada). Richardson, Matthew (1997), “Czech clitics: 2P or not 2P, that is the question”, in Kaiser (1997), 131–50. Rizzi, Luigi (1978), “A restructuring rule in Italian syntax”, in S. J. Keyser (ed.), Recent transformational studies in European languages (Cambridge, MA: MIT Press), 113–58. —— (1982), Issues in Italian syntax (Dordrecht: Foris). Roberts, Ian (1993a), “The nature of subject clitics in Franco-Provençal Valdôtain”, in Belletti (1993), 319–53. —— (1993b), Verbs and diachronic syntax: A comparative history of English and French (Dordrecht: Kluwer).

300

References

Roberts, Taylor (1997), “The optimal second position in Pashto”, in G. Booij and J. van de Weijer (eds.), Phonology in progress: Progress in phonology, HIL Phonology Papers III (The Hague: Holland Academic Graphics), 367–401. —— (2000), “Clitics and Agreement”. Ph.D. thesis. MIT. Rosen, Sara Thomas (1989), “Two types of noun incorporation: A lexical analysis”, Language 65: 294–317. Sadock, Jerrold (1980), “Noun incorporation in Greenlandic: A case of syntactic word formation”, Language 56: 300–19. —— (1986), “Some notes on noun incorporation”, Language 62: 19–31. —— (1991), Autolexical syntax: A theory of parallel grammatical representations (Chicago: University of Chicago Press). —— (2003), A grammar of Kalaallisut (West Greenlandic Eskimo), Languages of the World/Materials 162 (Munich: Lincom Europa). Salisbury, Mary, (2002), “A Grammar of Pukapukan”. Ph.D. thesis. University of Auckland. Sapir, Edward (1911), “The problem of noun incorporation in American languages”, American Anthropologist (n.s.) 13: 250–82. Schachter, Paul (1973), “Constraints on clitic order in Tagalog”, in A. B. Gonzalez (ed.), Parangal kay Cecilio Lopez (Manila: Linguistic Society of the Philippines), 214–31. —— and Otanes, Fe T. (1972), Tagalog reference grammar (Berkeley: University of California Press). Schafer, Robin (1995), “Negation and verb-second in Breton”, Natural Language and Linguistic Theory 13: 135–72. Schapansky, Nathalie S. M. (2000), Negation, referentiality and boundedness in Gwenedeg Breton (Munich: Lincom Europa). Schmidt, Hans (1999), “Rotuma: Sprache und Geschichte”. Ph.D. thesis, University of Hamburg. Schütze, Carson (1994), “Serbo-Croatian second position clitic placement and the phonology–syntax interface”, MIT Working Papers in Linguistics 21: 373–473. Selkirk, Elizabeth (1980), “Prosodic domains in phonology”, in M. Aronoff and M.-L. Kean (eds.), Juncture (Saratoga, CA: Anma Libri), 107–29. —— (1981), “On prosodic structure and its relation to syntactic structure”, in T. Fretheim (ed.), Nordic prosody II (Trondheim: Tapir), 111–40. —— (1984), Phonology and syntax: The relation between sound and structure (Cambridge, MA: MIT Press). —— (1986), “On derived domains in sentence phonology”, Phonology 3: 371–405. —— (1995), “The prosodic structure of function words”, in J. Beckman, L. W. Dickey, and S. Urbanczyk (eds.), Papers in optimality theory, University Of Massachusetts Occasional Papers in Linguistics 18 (Amherst: Dept. of Linguistics, University of Massachusetts), 439–70. Sherwood, David Fairchild (1986), Maliseet-Passamaquoddy verb morphology, Canadian Ethnology Service, Paper No. 105 (Ottawa: Canadian Museum of Civilization). Signorell, Faust, Wuethrich-Grisch, Mena, and Simeon, Gion Pol (1987), Normas surmiranas (Coira: Tgesa editoura cantunala per stampats e meds d’instrucziun).

References

301

Simeon, Gion Pol (1996), Igl muvimaint rumantsch an Surmeir (Savognin: Uniun rumantscha da Surmeir). Simpson, Jane (1991), Warlpiri morpho-syntax: A lexicalist approach, Studies in Natural Language and Linguistic Theory (Dordrecht: Kluwer Academic). Skorik, Pjotr Ja (1948), Oˇcerk po sintaksisu ˇcukotskogo jazyka: inkorporacija (Leningrad: Uˇcpedgiz). Smith, Ian and Johnson, Steve (2000), “Kugu Nganhcara”, in R. M. W. Dixon and B. Blake (eds.), Handbook of Australian languages, vol. 5 (Oxford: Oxford University Press), 357–507. Spencer, Andrew (1995), “Incorporation in Chukchi”, Language 71: 439–89. Stemberger, Joseph (1981), “Morphological haplology”, Language 57: 791–817. Stephens, Janig (1982), “Word Order in Breton”. Ph.D. thesis, University of London (SOAS). —— (1990), “Non-ﬁnite clauses in Breton”, in M. J. Ball, J. Fife, E. Poppe, and J. Rowland (eds.), Celtic linguistics (Amsterdam: John Benjamin), 151–67. Steriade, Donca (1988), “Greek accent: A case for preserving structure”, Linguistic Inquiry 19: 271–314. Stump, Gregory T. (1989), “Further remarks on Breton agreement”, Natural Language and Linguistic Theory 7: 429–71. —— (2001), Inﬂectional morphology: A theory of paradigm structure (Cambridge: Cambridge University Press). Svantesson, Jan-Olof (1983), Kammu phonology and morphology, Travaux de l’institut de linguistique de Lund, vol. XVIII (Lund: CWK Gleerup). Taumoefolau, Melenaite (2003), “Stress in Tongan”, MIT Working Papers in Linguistics 44: 341–54. Proceedings of AFLA VIII: The Eighth Meeting of the Austronesian Formal Linguistics Association. Tegey, Habibullah (1977), “The Grammar of Clitics: Evidence from Pashto and Other Languages”. Ph.D. thesis, University of Illinois. Tobler, Adolf (1875/1912), “Besprechung von J. Le Coultre, De l’ordre des mots dans Chrétien de Troyes”, Vermischte Beiträge zur französischen Grammatik, vol. 5 (Leipzig: Hirzel), 395–414. Toivonen, Ida (2000), “The morphosyntax of Finnish possessives”, Natural Language and Linguistic Theory 18: 579–609. Truckenbrodt, Hubert (1999), “On the relation between syntactic phrases and phonological phrases”, Linguistic Inquiry 30: 219–55. Vance, Barbara (1997), Syntactic change in medieval French: Verb-second and null subjects (Dordrecht: Kluwer). van der Leeuw, Frank (1995a), “Cliticization as alignment to morphological slots”, Proceedings of the Chicago Linguistic Society 31: 168–80 vol. 2: Proceedings of the Parasession on Clitics. —— (1995b), “Cliticization, stress and phonological words in European Portuguese: An optimal(ity) approach”, Probus 7: 31–68. —— (1997), Clitics: Prosodic studies (The Hague: Holland Academic Graphics).

302

References

Vigário, Marina (2003), The prosodic word in European Portuguese (Berlin: Mouton de Gruyter). Vogel, Irene (1990), “The clitic group in prosodic phonology”, in J. Mascaró and M. Nespor (eds.), Grammar in progress: GLOW studies for Henk van Riemsdijk (Dordrecht: Foris), 447–54. Wackernagel, Jacob (1892), “Über ein Gesetz der indogermanischen Wortstellung”, Indogermanische Forschungen 1: 333–436. Wali, Kashi and Koul, Omkar N. (1997), Kashmiri: A cognitive-descriptive grammar (London: Routledge). Wheeler, Max W., Yates, Alan, and Dols, Nicolau (1999), Catalan: A comprehensive grammar (London: Routledge). Woodbury, Anthony and Sadock, Jerrold (1986), “Afﬁxal verbs in syntax: a reply to Grimshaw and Mester”, Natural Language and Linguistic Theory 4: 229–44. Zanuttini, Raffaella (1997), Negation and clausal structure: A comparative study of Romance languages (Oxford: Oxford University Press). Zec, Draga (1987), “Interactions of prosodic and syntactic constituency”. Unpub. manuscript, Stanford University. —— (1988), “Sonority Constraints on Prosodic Structure”. Ph.D. thesis, Stanford University. —— (1993), “Rule domains and phonological change”, in S. Hargus and E. Kaisse (eds.), Studies in lexical phonology, Phonetics and Phonology 4 (New York: Academic Press), 365–405. —— and Inkelas, Sharon (1990), “Prosodically constrained syntax”, in S. Inkelas and D. Zec (eds.), The phonology–syntax connection (Chicago and Stanford: University of Chicago Press and CSLI Publications), 365–78. —— (1992), “The place of clitics in the prosodic hierarchy”, Proceedings of the West Coast Conference on Formal Linguistics 10: 505–20. Zide, Arlene R. K. (1976), “Nominal combining forms in Sora and Gorum”, Oceanic linguistics special publication 13, Austroasiatic Studies, Part II (Honolulu: University of Hawaii Press), 1259–94. Zwicky, Arnold M. (1977), On clitics (Bloomington: Indiana University Linguistics Club). —— (1985), “Clitics and particles”, Language 61: 283–305. —— (1987), “Suppressing the Zs”, Journal of Linguistics 23: 133–48. —— and Pullum, Geoffrey K. (1983), “Cliticization vs. inﬂection: English n’t”, Language 59: 502–13.

Index of Subjects accent 10, 12, 24 afﬁxation 128, 182, 224 phrasal- and word-level 133–5, 138 afﬁxes 9, 13, 33, 62, 66–7, 83, 85–7, 89, 91–4, 98, 100, 128, 235, 237 see also inﬁxes; preﬁxes; sufﬁxes agreement 161, 169, 186, 228–54, 248–51, 275–80, 283–4 long-distance 227, 248 vs. “registration” 230, 235 Alignment constraints: Align(Cati ,Edger/l ,Catj , Edger/l ) 57, 135, 145, 181 Align(CP,L,PWord,L) 60 Align(GrWord,L,LexWord, L) 137 Align(LexMax ,L,PPhrase, L) 68 Align(LexMax ,R,PPrase, R) 68 Align(LexWord,L,PWord, L) 57, 68 Align(LexWord,R,PWord, R) 57, 68 Align(PPhrase,L,LexWord, L) 59 Align(PPhrase,R,PWord, R) 68 Align(PWord,L,LexWord, L) 64, 68 Align(Pword,R,LexWord, R) 68 Align(XP,L,PPhrase,L) 57, 59 Align(XP,R,PPhrase,R) 57 Align([e],R,PPhrase,L) 70 see also EdgeMost (Alignment) constraints A-Morphous Morphology 3 anchors for clitic placement 79, 80, 85, 115, 123, 142, 147 auxiliaries 16, 25, 48, 140–1 clitics 24, 65–8, 72 contracted 3, 64, 66, 71

in English 25–30, 64–74 reduced 25–30, 64–74 in Warlpiri 140 auxiliary bases in Seediq 76 in Warlpiri 139 Auxiliary Reduction in English 26, 71–2 Binding relation between pro and the verb 232 Case licensing 231 nominal and verbal 238 Case absorption in Mohawk 231 ∗ Clash 53–4 clitic climbing 227–8, 246–249, 254 in French 78, 228 in Italian 245–7 in Serbo-Croatian 117 in Spanish 228 in Surmiran 247 clitic doubling in Bulgarian 240 in Romance 78, 227–8, 240–2 Clitic Groups 42–5 cliticization 13, 17, 25–8, 52, 63, 65, 171 blocked by parentheticals 71 in English 26–30 as “rightward-adjunction” in syntax 28–30 clitics Afﬁxal 46–50 complementizer 105, 254–5

304

Index of Subjects

clitics (Continued) demonstrative 95, 103 Free 46–9, 66 “inﬂectional” 128, 169 Internal 46, 48, 50 Klavans’s typology of 79 phonological 23, 31, 37, 41, 58, 76 pronominal 11, 14, 16–18, 62, 78, 80–2, 152, 227–8 PWord 46, 48 Zwicky’s classiﬁcation of 9–11, 22–6, 31–2, 36, 76, 81 see also enclitics; proclitics; second-position clitics; simple clitics; special clitics; subject clitics ∗ Coda 136 comparative subdeletion 70 Complementizer in Breton 199 in German 77, 87, 105, 140, 148, 183, 193, 249, 254, 255 in Icelandic 185 in Kashimiri 189 in Kw akw, ala 105 in Surmiran 214 Complementizer Phrase (CP) recursion 185 complete/incomplete phase forms in Rotuman 100–1 composites non-compound formation 159 in Portuguese 159 prosodic structure of 160 compounds 261, 264–7 constraints (Optimality Theory) ranking of 55, 58–60, 64, 69, 148–50, 154, 164, 179, 198 correlative constructions in Kashimiri 192 deﬁniteness 95, 99 marked by a stress shift 95

deﬁnitive accent 94–100 in Pukapukan 98–9 in Rotuman 98–100 in Tongan 94, 96, 98–9 demonstratives 95, 101, 103 in Kw akw, ala 102–3 in Tongan 96, 98 Determiner Phrase (DP) headless 271 without NP complements 90–1 Determiners (D) 20–1, 102, 104 in Bulgarian 113 in Kw akw, ala 19–22, 104 in Wakashan languages 19–22 disagreement in Breton 200 Distributed Morphology 245 domains for clitics 79, 80, 82, 112, 115, 147–8, 175 for constraints 145–6, 198 for phonological rules 39 Dynamic Semantics 270 EdgeMost (Alignment) constraints: EdgeMost(e, L/R, D) EdgeMost(Afi , L, D) 135 see also LeftMost (Alignment) constraints; RightMost (Alignment) constraints emañ (Breton) 201–2 enclitics 23–4, 43, 58, 60–1, 81, 85, 158, 161 endoclitics 81, 152, 154, 156, 158–9, 161, 164 Exhaustivity constraints 47–9, 52 Exhaustivity(Ci ) constraint 52 Exhaustivity(PPh) 54 expletives in Breton 194, 197 in Icelandic 186

Index of Subjects Features 90–1, 231, 233, 236–7 [+neg] (English) 35 [D(eterminer)] 91 [Poss] 90–1 Final Devoicing in Polish 40 Foot 38–9, 41, 43, 47, 53 Full Interpretation (at PF) 39, 45–6, 52–3, 58, 67, 69 functional categories proliferation of 287 function words in English 48–9, 69 Gapping 69, 285 gaps, syntactic 29, 65, 73 post-auxiliary 29, 70 Generalized Phrase Structure Grammar (GPSG) 240 Generate (gen) 182 Generative Phonology, Classic 56 Genitive “group genitive” in English 90 Head-Driven Phrase Structure Grammar (HPSG) 95, 240, 270 Headedness constraint 52–3 Head Feature Convention 146, 233, 240, 246 Head-inﬂection (rule) 170 Head Movement 113, 124, 258, 260, 264, 269, 270–1, 279, 282 impersonal constructions in Icelandic 184 in Rumantsch 216, 218–19 incorporation 261, 263, 267, 275, 279–80, 285–6 in Aleut 269 classiﬁer 267, 272 with doubling 270–3, 278 in Kw akw, ala 269 noun 257, 258–9, 260–83, 286 syllable-afﬁx 94

305

syntactic analysis of 260–1, 269, 275–6, 281 of unaccusative verbs 259 Indeﬁniteness Operator Genx 267, 272 inﬁxes 83, 85–6, 128, 134–8, 142, 169 phrasal 83, 91 post-initial 142 Inﬂection node (Inﬂ) 35 split- 287 Inﬂection Phrase (IP) in Breton 203 Input-Output Faithfulness 164 ins (Surmiran Rumantsch) 216–17, 221 Integrity constraints 144–5, 154, 156, 160, 163, 165, 167, 171–3, 176, 182 Integrity(DP) 154 Integrity(LexW(or)d) 156, 160, 164–5 Integrity(P) 144 Integrity(PP) 154, 157 Integrity(PPhrase) 154 Integrity(Word) 143, 145, 165, 167 Integrity(XP) 143–5, 147 interface morphology-syntax 281 Intonational Phrase (IPh/IntPhr(ase)) 39, 47, 116, 121–2, 160 inversion prosodic 111, 117–18 subject-aux 70, 208–9 subject-verb 211 in subordinate clauses 212 in Surmiran 206–14, 216–19 in Tagalog 173, 218

306

Index of Subjects

Layeredness 47, 52–3 Layering Convention 233 Left Dislocation in Breton 201 LeftEdgeFaith constraints LeftEdgeFaith(CP) 141 LeftEdgeFaith(P) 142 LeftEdgeFaith(Verb) 163–5 LeftEdgeFaith(Word) 138, 163 LeftMost (alignment) constraints: LeftMost(e(, D)) 135, 167, 171–2, 176 LeftMost(Afi , D) 137 LeftMost(ar/l) 136 LeftMost(cl, Vﬁnite ) 149 LeftMost(cli ) 139, 140–1, 143 LeftMost(cli , D) 173 LeftMost(cli ,IP) 150, 154, 157 LeftMost(cli ,V) 146, 147, 157 LeftMost(e) 145 LeftMost(emañ, CP) 202 LeftMost(V,CP) 199 LeftMost(Verb[−Finite] , V) 146 LeftMost(Vﬁn ) 190, 193 LeftMost(Vﬁn , CP) 198 LeftMost(Vﬁn , IP) 198, 203–4 LeftMost(Vﬁn , S) 179, 182–3 Lexical Functional Grammar (LFG) 236, 240 Lexicalist Hypothesis 34–5, 258, 264 Lexical Phonology 9, 45, 51–2, 96, 134 Post- 9, 14, 27, 51 LPhCon 68–9 LWdCon 68–9 Max(Pfnc ) 60 Merge (syntactic operation) 169, 176 mesoclitics See endoclitics Minimalist Program for syntax 91, 232 Morphological Visibility Condition (MVC) 231–2, 266

morphology 3, 4, 6, 83–4, 106, 128, 178 derivational elements in 167 inﬂectional 133, 169, 233 of phrases 83–4, 101, 127–8 Morphosyntactic Representation of categories 230, 232–3, 240, 245, 248 movement 119, 128–30, 154, 173, 193, 224 altruistic 114 I-to-C 180–1 verb 178, 180–4, 204, 211, 214–6 see also Head Movement; wh-movement negation contracted forms of 35 in English 11, 35 negative sentences in Breton 96, 196, 200 nominals in Kw akw, ala 19 null-headed 283–4 NonFinal constraints NonFinal (cli ,IntPhr) 150 NonFinal(e) 142 NonInitial constraints: NonInitial(e,D) 137–8, 145, 148–9, 152, 171, 176, 198, 223 NonInitial(Afi , D) 137 NonInitial(cl,IntPhrase) 149 NonInitial(cli ) 141, 143, 173 NonInitial(cli , CP) 147–8, 175 NonInitial(cli , D) 151, 167, 173 NonInitial(cli , IntPhr) 150–1 NonInitial(cli , IP) 154, 157, 173–5 NonInitial(e)141 NonInitial(Vﬁn 190–1, 193 NonInitial(Vﬁn, CP) 198, 202 NonInitial(Vﬁn, S) 179, 182

Index of Subjects Non-recursive constraints 47–9 NonRecursive(Ci ) 52 NonRecursive(PPh) 58 NonRecursive(PW(or)d) 54–5, 60 Numeration 169 Onset 63–4, 136 Optimality Theory (OT) 50, 57, 132, 134, 157, 170, 176, 182, 225 particles grammatical- and discourse markers 36 Phonetic Form (PF) 34 Phonological Phrases (PPh/PPhr(ase) ) 39–41, 47–8, 61, 66, 70, 116, 160 boundaries of 69, 73 and syntactic phrases 67 two-word 77 Phonological representations (non)-hierarchical 37, 38 Phonological Words (PW(or)d) 13, 24, 34, 38–9, 41–5, 47–9, 52–3, 57, 61–2, 64, 68, 70, 76, 95, 114–5, 123, 140, 143, 157, 170, 172, 181 PWordMax 160 Phonology 9, 12, 25, 56 autosegmental 38 lexical 9, 14, 33, 45, 51–2, 96, 134 post-lexical 9, 14, 27, 34, 51 prosodic 45, 56 phrasing groups syntactic and prosodic 56, 72 polysynthetic languages 266, 279, 282 portmanteaux lexicalized 73 possessives in English 89–94 possessors in Finnish 236

307

in Kw akw, ala 102 in West Greenlandic 286 preﬁxes 63, 85, 135 person-marking 86 PRO 237, 239 arbitrary 239 pro 230–1, 234, 246, 268, 271 -drop 230, 234 or a trace 273, 281 proclitics 23, 58, 81 Progressive Devoicing in Polish 40 pronominals 16, 23, 105, 114–6, 227–8 pronouns 11, 12, 17, 48–9, 92, 101, 219, 221–2 clitic- 18, 23, 49, 77, 101, 227 prosodic attachment 17 prosodic categories 112 prosodic constituent types 47 drawn from a universal inventory 47 Prosodic Faithfulness 53–5, 63–4, 66–7 prosodic hierarchy 13, 39, 41–2, 44, 47–8, 74 as a set of constraints 48 prosodic hierarchy condition 41 Prosodic Phonology 45 Prosodic Phrase See Phonological Phrase (PPh/PPhr(ase) ) prosodic structure 45, 57, 66, 69, 73 Prosodic Word See Phonological Word (PW(or)d) PWdCon 68 Readjustment Rules in Sound Pattern of English 56 Reanalysis See Restructuring

308

Index of Subjects

Reduplication in Chamorro 87–8 in Tongan 97 Rejoinder Emphasis 71 Relational Grammar 273–4 Relative constructions in Kashimiri 191–2 in Surmiran 214–6 Restrict (lexical operation) 269–70, 280, 284 Restructuring 246, 248 RightEdgeFaith RightEdgeFaith(Verb) 163 RightMost(Alignment) constraints: RightMost (e (,D)) 135 RightMost(Afi , D) 135 RightMost(cli ) 162 RightMost(cli , D) 163 RightMost(cli ,Verb) 163–4 RightMost(e) 145 second position 61, 75–6, 81, 105, 119–21, 132, 142, 167–8, 173, 177–8, 181, 186 Wackernagel’s idea of 108 second position clitics 12, 81, 83, 104, 108–9, 111, 116–20, 123–4, 126, 138, 142–3, 147–8, 152–7, 165, 176, 179 simple clitics 4, 10, 17, 22, 25, 30, 76 special clitics 10, 26, 31–2, 34, 75–6, 78–9, 82, 84, 89, 101, 105–26, 133, 177, 206, 227, 240, 249, 257 Stray Adjunction 13, 17, 19, 24, 52, 54–5, 58, 60, 70, 74, 82, 112–3, 141, 170–2, 176 Stray Foot Adjunction 171 Stray Syllable Adjunction 171

stress 13, 19, 24, 38, 43–4, 62, 77, 98 Strict Layering Hypothesis 42, 47–50, 52 Struct∗ 58 Stylistic Fronting in Icelandic 184, 186, 187, 210 subject clitics 208, 213, 215, 249–55, 249–255 four types of 252, 253, 254 Subject gap condition in Icelandic 184, 187 subordinate clauses in Breton 197 in German 183 in Icelandic 184–5 in Kashimiri 189 in Kw akw, ala 105 in Pashto 153 in Surmiran 206, 212, 214 sufﬁxes 33, 66, 83, 85, 92, 128, 135 English plural 92 Huave theme vowel 85 Syllable-afﬁx Incorporation 94 syllables 13, 38, 49, 66–7 Syntactic (S-)structure 46, 55 and prosody 45 relevant to phonology 37 Syntax Phonology-free 168 theta-roles (θ-roles) 232, 269 external (R) 267, 269, 273 Tobler-Mussaﬁa effects 147, 198 topicalization in Breton 195, 201 in German 180 in Icelandic 184, 195 in Surmiran 210, 212 topics 19–22, 121, 148, 202, 209 in Breton 194, 199–202 in Bulgarian 214 in Icelandic 186

Index of Subjects in Kashimiri 189–90 in Kw akw, ala 19–22 in Surmiran 206, 211–12, 215–7, 233 in Tagalog 165 Unaccusative Hypothesis 261 underspeciﬁcation of animacy 277 verb initial in Breton 197–8 verbs in Breton 194–5, 197–8, 201–3 incorporating 282 in Kashimiri 188–91 in Udi 162 verb second 177–224

in Breton 193, 199, 201 in German 180–1, 183 in Icelandic 183–6 in Kashimiri 187–193 in Surmiran 208, 212, 223–4 in West Greenlandic 283 Voice Assimilation in Polish 40 wh-movement 69 in Surmiran 213 wh-words/phrases in Bulgarian 151, 241 in Kashimiri 190–1 in Kw akw, ala 17 in Surmiran 213–16 Word Formation Rules (morphology) 34

309

This page intentionally left blank

Index of Names Abeillé, Anne 249 Adams, Marianne P. 205 Aitchison, Therese M. 95 Allen, Cynthia 90 Alpher, Barry 125 Anderson, Stephen R. 3–5, 10, 15–16, 34, 80, 83–4, 86, 90, 94, 97–8, 114, 129, 132–3, 137, 142, 147, 159, 162, 169–70, 193, 195, 201, 232–3, 239, 244, 261, 264, 269, 272 Bach, Emmon 14, 269 Baker, Mark 6, 18, 84, 230–4, 258, 260, 263–6, 269–70, 273, 275–82, 286 Barbosa, Pilar 85, 148, 149 Barshi, Immanuel 274 Benincá, Paola 204–5, 252–3 Bergsland, Knut 268 Bhatt, Rakesh Mohan 187, 189–90 Bittner, Maria 267, 270, 282 Bloomﬁeld, Leonard 87, 167 Boas, Franz 14–16, 19–22, 101 Bobaljik, Jonathan 248 Bolinger, Dwight 126 Booij, Geert 40, 42, 45, 51, 61–3 Borsley, Robert D. 193 Boškovi´c, željko 107, 116–23, 126, 129, 132, 144, 150–1, 186 Bresnan, Joan 29, 71, 242 Browne, Wayles 81, 110–11, 122 Burzio, Luigi 245 Carstairs, Andrew 90 Carstairs-McCarthy, Andrew 93–4 Chomsky, Noam 37, 56, 169 Chung, Sandra L. 87, 114–16, 123, 193, 195, 201, 267 Churchward, C. Maxwell 95, 97–9 Clark, Ross 96 Condax, Iovanna D. 98–9

Corbett, Greville 230 Crysmann, Berthold 159 Cysouw, Michael 82 Di Sciullo, Anna-Maria 268 Dixon, Robert M. W. 274 Dols, Nicolau 3, 59 Donaldson, Tamsin 75, 125 Fillmore, C. J. 203 Fontana, Josep M. 205 Fortescue, Michael 286 Foulet, Lucien 205 Franks, Steven 119, 240–1 Fukui, Naoki 203 Fulmer, Sandra Lee 85 Galves, Charlotte 85, 149, 157 Ganzoni, Gian Paul 219 Gazdar, Gerald 146, 233 Gerdts, Donna B. 260 Godard, Danièle 249 Goldsmith, John 38 Grimshaw, Jane 246 Grisch, Mena 218, 220 Haegeman, Liliane 248 Haiman, John 204 Hale, Kenneth 139–40 Hale, Mark 100 Halle, Morris 37, 56, 86 Halpern, Aaron 111, 120 Hannahs, S. J. 146;193, 197–99 Harris, Alice C. 81, 161–2, 164 Hayes, Bruce 42, 44, 142 Hendrick, Randall 193 Hock, Hans Heinrich 187 Hockett, Charles F. 86 Hoekstra, Eric 248

312

Index of Names

Holmer, Arthur J. 76 Howe, Darin 14

Lukoff, Fred 37 Lynge, Hans 110

Inkelas, Sharon 43–4, 47, 112 Insler, Stanley 126

Macaulay, Monica 11 McCarthy, John J. 49, 99, 100 McCawley, James D. 203 Maling, Joan 184 Malouf, Robert 270, 282, 286 Marantz, Alec 86 Mardirussian, Galust 260 Merlan, Francesca 259, 263 Miller, Phillip H. 240, 242 Mithun, Marianne 260–2, 270–1 Monachesi, Paola 129, 240, 242 Moravcsik, Edith 83 Mussaﬁa, A. 147, 198

Jacobs, Haike 63 Jaeggli, Osvaldo 245 Jelinek, Eloise 266 Johnson, Steve 80, 82 Jónsson, Jóhannes Gísli 184 Joseph, Brian 1, 85–6 Kahn, Daniel 38 Kaiser, Lizanne 133, 279 Kaisse, Ellen M. 4, 9, 26–8, 64–5, 80, 152, 156 Kanerva, Jonni 235 Kayne, Richard 2, 131–2 King, Tracy Holloway 240–1 Kiparsky, Paul 51, 134, 178, 235 Kissock, Madelyn 100 Klavans, Judith L. 4, 60, 63, 79, 81–2, 142, 146–7 Klein, Ewan 146, 233 Klokeid, Terry J. 21 Koopman, Hilda 203 Koul, Omkar N. 189–90 Kroeber, Alfred 259 Kroeger, Paul 169 Kuno, Susumu 285 Kurisu, Kazutaka 100–1 Ladusaw, William A. 267 Lees, Robert B. 261 Legate, Julie Anne 140 Legendre, Géraldine 5, 124–5, 132, 151, 176, 179, 193, 198 Liberman, Mark Y. 24, 38, 39 Liddell, H. G. 1 Lincoln, Neville J. 21 Linder, Karl Peter 250 Liver, Ricarda 204

Nespor, Marina 12, 37, 39, 41–4, 47, 60–1 Nevis, Joel A. 1, 80–1, 85–6, 235 Nida, Eugene 1 Noyer, Rolf 85 Ortiz de Urbina, Jon 193 Osborne, C. R. 262 Otanes, Fe T. 165–7, 173–5 Otsuka, Yuko 98 Parker, Steve 64 Payne, Doris L. 81, 274 Payne, Thomas E. 81 Peperkamp, Sharon 37, 43, 52–3 Perlmutter, David 125, 168 Poletto, Cecilia 205, 250–5 Polinsky, Maria S. 248, 259 Poser, William 96, 97 Postal, Paul 91, 275 Potsdam, Eric 248 Prince, Alan 24, 38, 39, 132 Progovac, Liljana 119 Pullum, Geoffrey K. 11, 28, 32–3, 35, 70–1, 146, 168, 233, 276 Rath, John C. 20–3 Richardson, Matthew 147–8

Index of Names Rivero, Maria-Luisa 193 Rizzi, Luigi 245–6 Roberts, Ian 205, 250 Roberts, Taylor 81, 152 Rosen, Sara Thomas 267 Ross, John Robert 126 Rubach, Jerzy 40 Sadock, Jerrold 6, 109, 110, 260, 281–2, 284–6 Sag, Ivan 146, 233 Salisbury, Mary 98–9 Sandalo, Filomena 85, 149, 157 Sapir, Edward 258–64 Schachter, Paul 126, 165–8, 170, 173–5 Schafer, Robin 193 Schapensky, Nathalie 193 Scott, R. 1 Selkirk, Elizabeth 12, 37, 39, 41, 46–51, 66, 68–9 Shaw, Patricia 9 Sherwood, David Fairchild 244 ShÖtze, Carson 120 Signorell, Faust 220–1, 247 Simeon, Gion Pol 220–1, 247 Simpson, Jane 140 Skorik, Pjotr Ja. 259 Smith, Ian 79, 82 Smits, Caroline 248 Smolensky, Paul 132 Speas, Margaret 203 Spencer, Andrew 259, 268

313

Sportiche, Dominique 203 Stemberger, Joseph 90 Stephens, Janig 193 Stump, Gregory T. 90, 137, 159, 193 Taumoefolau, Melanaite 98 Tegey, Habibullah 81, 152, 156 Tellerman, Maggie 193, 197–9 Tobler, Adolf 147, 198 Toivonen, Ida 235 Vance, Barbara 205 van der Leeuw, Frank 81, 152, 157 van Geenhoven, Veerle 282 Vigário, Marina 37, 157, 159–60 Vogel, Irene 12, 37, 39, 41–4, 47, 60–1 Wackernagel, Jacob 2, 5, 23, 86, 108, 143, 166, 177–8, 188, 224 Wali, Kashi 189–90 Wanner, Dieter 1 Wheeler, Max 3, 59 Williams, Edwin 268 Windsor, Evelyn 21 Wuethrich-Grisch, Mena 220–1, 247 Wurmbrand, Susan 248 Yates, Alan 3, 59 Zec, Draga 43–4, 50 Zide, Arlene R. K. 262–3

This page intentionally left blank

Index of Languages Afar 85 Ainu 279 Albanian 111, 113 Aleut 268–9 Algonquian languages 86, 244 See also Fox; Menomini; Potawatomi Balkan languages 79, 85, 111, 113, 124–5 Balto-Slavic languages 86 Bella Bella See Heiltsuk Breton 5, 178, 193–204, 210–11, 224, 239 Bulgarian 111–13, 131–2, 141, 146–7, 151, 240, 241 Catalan 3, 59, 60 Chamicuro 64 Chamorro 87–8, 105, 114–16, 123, 136 Chuckchee 258–9 Czech 147–8 Ditidaht 21 Dolomitic Ladin 204 Dutch 61–3 Limburg dialect 248–9 English Eastern Massachusetts dialect 49 Middle 35 Old 35 Eskimo See West Greenlandic Eskimo-Aleut languages 260, 268 Finnish 80, 235–9, 242 Flemish 248 Fox 87 Franco-Provençal 250–1 Valdôtain 250

French 10, 31, 73, 75, 78, 125, 158, 168, 209, 216, 220, 228–9, 232, 242–4 Gallo-Romance (Old French) 63–4 medieval 205 Friulian 204, 243, 251 San Michele al Taliamento 253 Gallo-Romance See French Georgian 84, 230, 234, 244 German 5, 177–8, 180, 183, 193, 205, 216, 220, 223–4, 248, 276 Greek 1, 2, 24, 43–4, 60–1, 177–8 Homeric 2, 31, 108 Greenlandic See West Greenlandic Haisla 14, 20–1 Heiltsuk 14, 20, 22–3 Hittite 109, 114, 123, 125, 143 Huave 85 Icelandic 5, 84, 159, 178–9, 183–6, 210, 212, 214–5, 223–4 Indic Middle 187 Indo-European 2, 5, 166, 177–8, 183, 224 Proto 2, 177 Italian 32, 55, 76, 129–31, 205, 219–22, 234, 245–7 “dialects” 6, 44, 52–5, 251–2, 254–5 Loreo dialect (Veneto) 255 Lucanian 44, 54, 76 Lugano 252 Neapolitan 53–4 Itelmen 248

316

Index of Languages

KamhmuP 136–7, 141, 163 Kashimiri 178, 187–93, 214–15, 224 Kugu Nganhcara 79, 82 Kwakiutl See Kw akw, ala Kwakwakawakw See Kw akw, ala Kw akw, ala 3, 5, 14–5, 17–25, 58–9, 63, 79, 81, 89, 101–105, 112–14, 269 Latin 2, 44, 216, 220 early 108 Lithuanian 86 Macedonian 111, 113, 124, 131–2, 146 Madurese 10 Makah 14 Maliseet 244 Mayali 278 Menomini 87, 244 Mixtec 11, 12 Mohawk 18, 230–1, 233–5, 258, 262–8, 270–81, 286 Nahuatl (Huauhtla) 259, 263, 272 Classical 258 Ngiyambaa 75, 125 Nitinaht See Ditidaht Nootka See Nuu-chah-nulth Nuu-chah-nulth 14 Oowiky’ala 14, 20 Oweekeno See Oowiky’ala Owikeno See Oowiky’ala Pashto 81, 152–7, 164 Pemon 229–30. 235, 242 Polish 40–2, 58–9 Portuguese 81, 85, 148, 151, 157–60, 162 Claasical 149 European 85, 148–9, 151–2, 157–8, 160–1, 164 Potawatomi 86–7

Pukapukan 98–9 Puter 219–20, 250 Rembarrnga 274 Rhaeto-Romance 205 Romance languages 3, 58, 80, 84–5, 146–7, 158, 223–4, 240 Romanian 111, 113 Rotuman 99, 100 Rumantsch 204–5, 219, 223, 242, 247, 250 Engadine 219, 250 Swiss 205 See also Puter; Surmiran; Surselvan; Vallader Sanskrit 9, 126, 177 Vedic 108 Seediq 76 Serbo-Croatian 49–50, 81, 110–12, 116–23, 129, 132, 144, 147, 150, 186 Belgrade 50 Eastern Herzegovina 50 eastern (“Serbian”) dialects 111 Neo-Štokavian 49–50 Srem, Maˇcva 50 western (“Croatian”) dialects 111, 144 Sora 262 Southern Paiute 258–9 Southern Tiwa 258, 272–3, 278 Spanish 168, 205, 219, 228–9, 241–2, 245 Castilian 241 Latin American 242 Sundanese 135–6 Surmiran 6, 204–24, 242–3, 247–51, 254 Surselvan 220, 250 Tagalog 32, 76, 126–7, 133, 141, 165–176, 179, 227, 240, 245, 253

Index of Languages Tiwi 262 Tongan 94–99 Tsez 248 Udi 161–4 Uralic languages 79 Vallader 219–20, 250 Vogul Northern 81

317

Wakashan languages 14, 20–1, 260 Warlpiri 77, 84, 139–40, 174, 227, 240 Welsh 85 West Greenlandic 109–10, 143, 174, 260, 270–86 Yagua 81 Yir-Yoront 125

Aspects of the Theory of Clitics (Oxford Studies in Theoretical Linguistics)

Negative Indefinites (Oxford Studies in Theoretical Linguistics)

The Syntax of Sentential Stress (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

A Natural History of Infixation (Oxford Studies in Theoretical Linguistics)

French Dislocation: Interpretation, Syntax, Acquisition (Oxford Studies in Theoretical Linguistics)

Theoretical Aspects of Memory

Tense, Aspect, and Indexicality (Oxford Studies in Theoretical Linguistics)

Lexical Plurals: A Morphosemantic Approach (Oxford Studies in Theoretical Linguistics)

Agreement, Pronominal Clitics and Negation in Tamazight Berber: A Unified Analysis (Continuum Studies In Theoretical Linguistics)

The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface (Oxford Studies in Theoretical Linguistics)

The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface (Oxford Studies in Theoretical Linguistics, 5)

Aspects Of The Theory Of Morphology (Trends in Linguistics. Studies and Monographs)

Theoretical Aspects of Heterogeneous Catalysis

Theoretical aspects of heterogeneous catalysis

The Syntax of Aspect: Deriving Thematic and Aspectual Interpretation (Oxford Studies in Theoretical Linguistics)

About the Speaker: Towards a Syntax of Indexicality (Oxford Studies in Theoretical Linguistics)

Theoretical Aspects of Chemical Reactivity

Finiteness: Theoretical and Empirical Foundations (Oxford Linguistics)

Theoretical Aspects of International Relations

The Oxford Handbook of Linguistic Interfaces (Oxford Handbooks in Linguistics)

Theoretical Linguistics

Referential-Semantic Analysis: Aspects of a Theory of Linguistic Reference (Cambridge Studies in Linguistics (No. 28))

InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces (Oxford Studies in Theoretical Linguistics)

InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces (Oxford Studies in Theoretical Linguistics)

Structural aspects in the theory of probability

Aspects Of Polaron Theory

The Concise Oxford Dictionary of Linguistics

The Concise Oxford Dictionary of Linguistics

Aspects of the Theory of Clitics (Oxford Studies in Theoretical Linguistics)

Negative Indefinites (Oxford Studies in Theoretical Linguistics)

The Syntax of Sentential Stress (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

The Logic of Conventional Implicatures (Oxford Studies in Theoretical Linguistics)

A Natural History of Infixation (Oxford Studies in Theoretical Linguistics)

French Dislocation: Interpretation, Syntax, Acquisition (Oxford Studies in Theoretical Linguistics)

Theoretical Aspects of Memory

Tense, Aspect, and Indexicality (Oxford Studies in Theoretical Linguistics)

Lexical Plurals: A Morphosemantic Approach (Oxford Studies in Theoretical Linguistics)

Agreement, Pronominal Clitics and Negation in Tamazight Berber: A Unified Analysis (Continuum Studies In Theoretical Linguistics)

The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface (Oxford Studies in Theoretical Linguistics)

The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface (Oxford Studies in Theoretical Linguistics, 5)

Aspects Of The Theory Of Morphology (Trends in Linguistics. Studies and Monographs)

Theoretical Aspects of Heterogeneous Catalysis

Theoretical aspects of heterogeneous catalysis

The Syntax of Aspect: Deriving Thematic and Aspectual Interpretation (Oxford Studies in Theoretical Linguistics)

About the Speaker: Towards a Syntax of Indexicality (Oxford Studies in Theoretical Linguistics)

Theoretical Aspects of Chemical Reactivity

Finiteness: Theoretical and Empirical Foundations (Oxford Linguistics)

Theoretical Aspects of International Relations

The Oxford Handbook of Linguistic Interfaces (Oxford Handbooks in Linguistics)

Theoretical Linguistics

Referential-Semantic Analysis: Aspects of a Theory of Linguistic Reference (Cambridge Studies in Linguistics (No. 28))

InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces (Oxford Studies in Theoretical Linguistics)

InterPhases: Phase-Theoretic Investigations of Linguistic Interfaces (Oxford Studies in Theoretical Linguistics)

Structural aspects in the theory of probability

Aspects Of Polaron Theory

The Concise Oxford Dictionary of Linguistics

The Concise Oxford Dictionary of Linguistics

Recommend Documents